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 /* User need to know colo mode after COLO failover */
42 static COLOMode last_colo_mode
;
44 #define COLO_BUFFER_BASE_SIZE (4 * 1024 * 1024)
46 bool migration_in_colo_state(void)
48 MigrationState
*s
= migrate_get_current();
50 return (s
->state
== MIGRATION_STATUS_COLO
);
53 bool migration_incoming_in_colo_state(void)
55 MigrationIncomingState
*mis
= migration_incoming_get_current();
57 return mis
&& (mis
->state
== MIGRATION_STATUS_COLO
);
60 static bool colo_runstate_is_stopped(void)
62 return runstate_check(RUN_STATE_COLO
) || !runstate_is_running();
65 static void secondary_vm_do_failover(void)
67 /* COLO needs enable block-replication */
68 #ifdef CONFIG_REPLICATION
70 MigrationIncomingState
*mis
= migration_incoming_get_current();
71 Error
*local_err
= NULL
;
73 /* Can not do failover during the process of VM's loading VMstate, Or
74 * it will break the secondary VM.
76 if (vmstate_loading
) {
77 old_state
= failover_set_state(FAILOVER_STATUS_ACTIVE
,
78 FAILOVER_STATUS_RELAUNCH
);
79 if (old_state
!= FAILOVER_STATUS_ACTIVE
) {
80 error_report("Unknown error while do failover for secondary VM,"
81 "old_state: %s", FailoverStatus_str(old_state
));
86 migrate_set_state(&mis
->state
, MIGRATION_STATUS_COLO
,
87 MIGRATION_STATUS_COMPLETED
);
89 replication_stop_all(true, &local_err
);
91 error_report_err(local_err
);
94 /* Notify all filters of all NIC to do checkpoint */
95 colo_notify_filters_event(COLO_EVENT_FAILOVER
, &local_err
);
97 error_report_err(local_err
);
101 error_report("\"-S\" qemu option will be ignored in secondary side");
102 /* recover runstate to normal migration finish state */
106 * Make sure COLO incoming thread not block in recv or send,
107 * If mis->from_src_file and mis->to_src_file use the same fd,
108 * The second shutdown() will return -1, we ignore this value,
111 if (mis
->from_src_file
) {
112 qemu_file_shutdown(mis
->from_src_file
);
114 if (mis
->to_src_file
) {
115 qemu_file_shutdown(mis
->to_src_file
);
118 old_state
= failover_set_state(FAILOVER_STATUS_ACTIVE
,
119 FAILOVER_STATUS_COMPLETED
);
120 if (old_state
!= FAILOVER_STATUS_ACTIVE
) {
121 error_report("Incorrect state (%s) while doing failover for "
122 "secondary VM", FailoverStatus_str(old_state
));
125 /* Notify COLO incoming thread that failover work is finished */
126 qemu_sem_post(&mis
->colo_incoming_sem
);
128 /* For Secondary VM, jump to incoming co */
129 if (mis
->migration_incoming_co
) {
130 qemu_coroutine_enter(mis
->migration_incoming_co
);
137 static void primary_vm_do_failover(void)
139 #ifdef CONFIG_REPLICATION
140 MigrationState
*s
= migrate_get_current();
142 Error
*local_err
= NULL
;
144 migrate_set_state(&s
->state
, MIGRATION_STATUS_COLO
,
145 MIGRATION_STATUS_COMPLETED
);
147 * kick COLO thread which might wait at
148 * qemu_sem_wait(&s->colo_checkpoint_sem).
150 colo_checkpoint_notify(migrate_get_current());
153 * Wake up COLO thread which may blocked in recv() or send(),
154 * The s->rp_state.from_dst_file and s->to_dst_file may use the
155 * same fd, but we still shutdown the fd for twice, it is harmless.
157 if (s
->to_dst_file
) {
158 qemu_file_shutdown(s
->to_dst_file
);
160 if (s
->rp_state
.from_dst_file
) {
161 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
164 old_state
= failover_set_state(FAILOVER_STATUS_ACTIVE
,
165 FAILOVER_STATUS_COMPLETED
);
166 if (old_state
!= FAILOVER_STATUS_ACTIVE
) {
167 error_report("Incorrect state (%s) while doing failover for Primary VM",
168 FailoverStatus_str(old_state
));
172 replication_stop_all(true, &local_err
);
174 error_report_err(local_err
);
178 /* Notify COLO thread that failover work is finished */
179 qemu_sem_post(&s
->colo_exit_sem
);
185 COLOMode
get_colo_mode(void)
187 if (migration_in_colo_state()) {
188 return COLO_MODE_PRIMARY
;
189 } else if (migration_incoming_in_colo_state()) {
190 return COLO_MODE_SECONDARY
;
192 return COLO_MODE_NONE
;
196 void colo_do_failover(void)
198 /* Make sure VM stopped while failover happened. */
199 if (!colo_runstate_is_stopped()) {
200 vm_stop_force_state(RUN_STATE_COLO
);
203 switch (get_colo_mode()) {
204 case COLO_MODE_PRIMARY
:
205 primary_vm_do_failover();
207 case COLO_MODE_SECONDARY
:
208 secondary_vm_do_failover();
211 error_report("colo_do_failover failed because the colo mode"
212 " could not be obtained");
216 #ifdef CONFIG_REPLICATION
217 void qmp_xen_set_replication(bool enable
, bool primary
,
218 bool has_failover
, bool failover
,
221 ReplicationMode mode
= primary
?
222 REPLICATION_MODE_PRIMARY
:
223 REPLICATION_MODE_SECONDARY
;
225 if (has_failover
&& enable
) {
226 error_setg(errp
, "Parameter 'failover' is only for"
227 " stopping replication");
232 replication_start_all(mode
, errp
);
237 replication_stop_all(failover
, failover
? NULL
: errp
);
241 ReplicationStatus
*qmp_query_xen_replication_status(Error
**errp
)
244 ReplicationStatus
*s
= g_new0(ReplicationStatus
, 1);
246 replication_get_error_all(&err
);
250 s
->desc
= g_strdup(error_get_pretty(err
));
259 void qmp_xen_colo_do_checkpoint(Error
**errp
)
261 replication_do_checkpoint_all(errp
);
262 /* Notify all filters of all NIC to do checkpoint */
263 colo_notify_filters_event(COLO_EVENT_CHECKPOINT
, errp
);
267 COLOStatus
*qmp_query_colo_status(Error
**errp
)
269 COLOStatus
*s
= g_new0(COLOStatus
, 1);
271 s
->mode
= get_colo_mode();
272 s
->last_mode
= last_colo_mode
;
274 switch (failover_get_state()) {
275 case FAILOVER_STATUS_NONE
:
276 s
->reason
= COLO_EXIT_REASON_NONE
;
278 case FAILOVER_STATUS_COMPLETED
:
279 s
->reason
= COLO_EXIT_REASON_REQUEST
;
282 if (migration_in_colo_state()) {
283 s
->reason
= COLO_EXIT_REASON_PROCESSING
;
285 s
->reason
= COLO_EXIT_REASON_ERROR
;
292 static void colo_send_message(QEMUFile
*f
, COLOMessage msg
,
297 if (msg
>= COLO_MESSAGE__MAX
) {
298 error_setg(errp
, "%s: Invalid message", __func__
);
301 qemu_put_be32(f
, msg
);
304 ret
= qemu_file_get_error(f
);
306 error_setg_errno(errp
, -ret
, "Can't send COLO message");
308 trace_colo_send_message(COLOMessage_str(msg
));
311 static void colo_send_message_value(QEMUFile
*f
, COLOMessage msg
,
312 uint64_t value
, Error
**errp
)
314 Error
*local_err
= NULL
;
317 colo_send_message(f
, msg
, &local_err
);
319 error_propagate(errp
, local_err
);
322 qemu_put_be64(f
, value
);
325 ret
= qemu_file_get_error(f
);
327 error_setg_errno(errp
, -ret
, "Failed to send value for message:%s",
328 COLOMessage_str(msg
));
332 static COLOMessage
colo_receive_message(QEMUFile
*f
, Error
**errp
)
337 msg
= qemu_get_be32(f
);
338 ret
= qemu_file_get_error(f
);
340 error_setg_errno(errp
, -ret
, "Can't receive COLO message");
343 if (msg
>= COLO_MESSAGE__MAX
) {
344 error_setg(errp
, "%s: Invalid message", __func__
);
347 trace_colo_receive_message(COLOMessage_str(msg
));
351 static void colo_receive_check_message(QEMUFile
*f
, COLOMessage expect_msg
,
355 Error
*local_err
= NULL
;
357 msg
= colo_receive_message(f
, &local_err
);
359 error_propagate(errp
, local_err
);
362 if (msg
!= expect_msg
) {
363 error_setg(errp
, "Unexpected COLO message %d, expected %d",
368 static uint64_t colo_receive_message_value(QEMUFile
*f
, uint32_t expect_msg
,
371 Error
*local_err
= NULL
;
375 colo_receive_check_message(f
, expect_msg
, &local_err
);
377 error_propagate(errp
, local_err
);
381 value
= qemu_get_be64(f
);
382 ret
= qemu_file_get_error(f
);
384 error_setg_errno(errp
, -ret
, "Failed to get value for COLO message: %s",
385 COLOMessage_str(expect_msg
));
390 static int colo_do_checkpoint_transaction(MigrationState
*s
,
391 QIOChannelBuffer
*bioc
,
394 Error
*local_err
= NULL
;
397 colo_send_message(s
->to_dst_file
, COLO_MESSAGE_CHECKPOINT_REQUEST
,
403 colo_receive_check_message(s
->rp_state
.from_dst_file
,
404 COLO_MESSAGE_CHECKPOINT_REPLY
, &local_err
);
408 /* Reset channel-buffer directly */
409 qio_channel_io_seek(QIO_CHANNEL(bioc
), 0, 0, NULL
);
412 qemu_mutex_lock_iothread();
413 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
414 qemu_mutex_unlock_iothread();
417 vm_stop_force_state(RUN_STATE_COLO
);
418 qemu_mutex_unlock_iothread();
419 trace_colo_vm_state_change("run", "stop");
421 * Failover request bh could be called after vm_stop_force_state(),
422 * So we need check failover_request_is_active() again.
424 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
428 colo_notify_compares_event(NULL
, COLO_EVENT_CHECKPOINT
, &local_err
);
433 /* Disable block migration */
434 migrate_set_block_enabled(false, &local_err
);
435 qemu_mutex_lock_iothread();
437 #ifdef CONFIG_REPLICATION
438 replication_do_checkpoint_all(&local_err
);
440 qemu_mutex_unlock_iothread();
447 colo_send_message(s
->to_dst_file
, COLO_MESSAGE_VMSTATE_SEND
, &local_err
);
449 qemu_mutex_unlock_iothread();
452 /* Note: device state is saved into buffer */
453 ret
= qemu_save_device_state(fb
);
455 qemu_mutex_unlock_iothread();
460 * Only save VM's live state, which not including device state.
461 * TODO: We may need a timeout mechanism to prevent COLO process
462 * to be blocked here.
464 qemu_savevm_live_state(s
->to_dst_file
);
469 * We need the size of the VMstate data in Secondary side,
470 * With which we can decide how much data should be read.
472 colo_send_message_value(s
->to_dst_file
, COLO_MESSAGE_VMSTATE_SIZE
,
473 bioc
->usage
, &local_err
);
478 qemu_put_buffer(s
->to_dst_file
, bioc
->data
, bioc
->usage
);
479 qemu_fflush(s
->to_dst_file
);
480 ret
= qemu_file_get_error(s
->to_dst_file
);
485 colo_receive_check_message(s
->rp_state
.from_dst_file
,
486 COLO_MESSAGE_VMSTATE_RECEIVED
, &local_err
);
491 colo_receive_check_message(s
->rp_state
.from_dst_file
,
492 COLO_MESSAGE_VMSTATE_LOADED
, &local_err
);
499 qemu_mutex_lock_iothread();
501 qemu_mutex_unlock_iothread();
502 trace_colo_vm_state_change("stop", "run");
506 error_report_err(local_err
);
511 static void colo_compare_notify_checkpoint(Notifier
*notifier
, void *data
)
513 colo_checkpoint_notify(data
);
516 static void colo_process_checkpoint(MigrationState
*s
)
518 QIOChannelBuffer
*bioc
;
520 int64_t current_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
521 Error
*local_err
= NULL
;
524 last_colo_mode
= get_colo_mode();
525 if (last_colo_mode
!= COLO_MODE_PRIMARY
) {
526 error_report("COLO mode must be COLO_MODE_PRIMARY");
530 failover_init_state();
532 s
->rp_state
.from_dst_file
= qemu_file_get_return_path(s
->to_dst_file
);
533 if (!s
->rp_state
.from_dst_file
) {
534 error_report("Open QEMUFile from_dst_file failed");
538 packets_compare_notifier
.notify
= colo_compare_notify_checkpoint
;
539 colo_compare_register_notifier(&packets_compare_notifier
);
542 * Wait for Secondary finish loading VM states and enter COLO
545 colo_receive_check_message(s
->rp_state
.from_dst_file
,
546 COLO_MESSAGE_CHECKPOINT_READY
, &local_err
);
550 bioc
= qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE
);
551 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
552 object_unref(OBJECT(bioc
));
554 qemu_mutex_lock_iothread();
555 #ifdef CONFIG_REPLICATION
556 replication_start_all(REPLICATION_MODE_PRIMARY
, &local_err
);
558 qemu_mutex_unlock_iothread();
566 qemu_mutex_unlock_iothread();
567 trace_colo_vm_state_change("stop", "run");
569 timer_mod(s
->colo_delay_timer
,
570 current_time
+ s
->parameters
.x_checkpoint_delay
);
572 while (s
->state
== MIGRATION_STATUS_COLO
) {
573 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
574 error_report("failover request");
578 qemu_sem_wait(&s
->colo_checkpoint_sem
);
580 if (s
->state
!= MIGRATION_STATUS_COLO
) {
583 ret
= colo_do_checkpoint_transaction(s
, bioc
, fb
);
590 /* Throw the unreported error message after exited from loop */
592 error_report_err(local_err
);
600 * There are only two reasons we can get here, some error happened
601 * or the user triggered failover.
603 switch (failover_get_state()) {
604 case FAILOVER_STATUS_COMPLETED
:
605 qapi_event_send_colo_exit(COLO_MODE_PRIMARY
,
606 COLO_EXIT_REASON_REQUEST
);
609 qapi_event_send_colo_exit(COLO_MODE_PRIMARY
,
610 COLO_EXIT_REASON_ERROR
);
613 /* Hope this not to be too long to wait here */
614 qemu_sem_wait(&s
->colo_exit_sem
);
615 qemu_sem_destroy(&s
->colo_exit_sem
);
618 * It is safe to unregister notifier after failover finished.
619 * Besides, colo_delay_timer and colo_checkpoint_sem can't be
620 * released befor unregister notifier, or there will be use-after-free
623 colo_compare_unregister_notifier(&packets_compare_notifier
);
624 timer_del(s
->colo_delay_timer
);
625 timer_free(s
->colo_delay_timer
);
626 qemu_sem_destroy(&s
->colo_checkpoint_sem
);
629 * Must be called after failover BH is completed,
630 * Or the failover BH may shutdown the wrong fd that
631 * re-used by other threads after we release here.
633 if (s
->rp_state
.from_dst_file
) {
634 qemu_fclose(s
->rp_state
.from_dst_file
);
638 void colo_checkpoint_notify(void *opaque
)
640 MigrationState
*s
= opaque
;
641 int64_t next_notify_time
;
643 qemu_sem_post(&s
->colo_checkpoint_sem
);
644 s
->colo_checkpoint_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
645 next_notify_time
= s
->colo_checkpoint_time
+
646 s
->parameters
.x_checkpoint_delay
;
647 timer_mod(s
->colo_delay_timer
, next_notify_time
);
650 void migrate_start_colo_process(MigrationState
*s
)
652 qemu_mutex_unlock_iothread();
653 qemu_sem_init(&s
->colo_checkpoint_sem
, 0);
654 s
->colo_delay_timer
= timer_new_ms(QEMU_CLOCK_HOST
,
655 colo_checkpoint_notify
, s
);
657 qemu_sem_init(&s
->colo_exit_sem
, 0);
658 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
659 MIGRATION_STATUS_COLO
);
660 colo_process_checkpoint(s
);
661 qemu_mutex_lock_iothread();
664 static void colo_wait_handle_message(QEMUFile
*f
, int *checkpoint_request
,
668 Error
*local_err
= NULL
;
670 msg
= colo_receive_message(f
, &local_err
);
672 error_propagate(errp
, local_err
);
677 case COLO_MESSAGE_CHECKPOINT_REQUEST
:
678 *checkpoint_request
= 1;
681 *checkpoint_request
= 0;
682 error_setg(errp
, "Got unknown COLO message: %d", msg
);
687 void *colo_process_incoming_thread(void *opaque
)
689 MigrationIncomingState
*mis
= opaque
;
691 QIOChannelBuffer
*bioc
= NULL
; /* Cache incoming device state */
694 Error
*local_err
= NULL
;
697 rcu_register_thread();
698 qemu_sem_init(&mis
->colo_incoming_sem
, 0);
700 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
701 MIGRATION_STATUS_COLO
);
703 last_colo_mode
= get_colo_mode();
704 if (last_colo_mode
!= COLO_MODE_SECONDARY
) {
705 error_report("COLO mode must be COLO_MODE_SECONDARY");
709 failover_init_state();
711 mis
->to_src_file
= qemu_file_get_return_path(mis
->from_src_file
);
712 if (!mis
->to_src_file
) {
713 error_report("COLO incoming thread: Open QEMUFile to_src_file failed");
717 * Note: the communication between Primary side and Secondary side
718 * should be sequential, we set the fd to unblocked in migration incoming
719 * coroutine, and here we are in the COLO incoming thread, so it is ok to
720 * set the fd back to blocked.
722 qemu_file_set_blocking(mis
->from_src_file
, true);
724 bioc
= qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE
);
725 fb
= qemu_fopen_channel_input(QIO_CHANNEL(bioc
));
726 object_unref(OBJECT(bioc
));
728 qemu_mutex_lock_iothread();
729 #ifdef CONFIG_REPLICATION
730 replication_start_all(REPLICATION_MODE_SECONDARY
, &local_err
);
732 qemu_mutex_unlock_iothread();
739 trace_colo_vm_state_change("stop", "run");
740 qemu_mutex_unlock_iothread();
742 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_CHECKPOINT_READY
,
748 while (mis
->state
== MIGRATION_STATUS_COLO
) {
751 colo_wait_handle_message(mis
->from_src_file
, &request
, &local_err
);
756 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
757 error_report("failover request");
761 qemu_mutex_lock_iothread();
762 vm_stop_force_state(RUN_STATE_COLO
);
763 trace_colo_vm_state_change("run", "stop");
764 qemu_mutex_unlock_iothread();
766 /* FIXME: This is unnecessary for periodic checkpoint mode */
767 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_CHECKPOINT_REPLY
,
773 colo_receive_check_message(mis
->from_src_file
,
774 COLO_MESSAGE_VMSTATE_SEND
, &local_err
);
779 qemu_mutex_lock_iothread();
780 cpu_synchronize_all_pre_loadvm();
781 ret
= qemu_loadvm_state_main(mis
->from_src_file
, mis
);
782 qemu_mutex_unlock_iothread();
785 error_report("Load VM's live state (ram) error");
789 value
= colo_receive_message_value(mis
->from_src_file
,
790 COLO_MESSAGE_VMSTATE_SIZE
, &local_err
);
796 * Read VM device state data into channel buffer,
797 * It's better to re-use the memory allocated.
798 * Here we need to handle the channel buffer directly.
800 if (value
> bioc
->capacity
) {
801 bioc
->capacity
= value
;
802 bioc
->data
= g_realloc(bioc
->data
, bioc
->capacity
);
804 total_size
= qemu_get_buffer(mis
->from_src_file
, bioc
->data
, value
);
805 if (total_size
!= value
) {
806 error_report("Got %" PRIu64
" VMState data, less than expected"
807 " %" PRIu64
, total_size
, value
);
810 bioc
->usage
= total_size
;
811 qio_channel_io_seek(QIO_CHANNEL(bioc
), 0, 0, NULL
);
813 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_VMSTATE_RECEIVED
,
819 qemu_mutex_lock_iothread();
820 vmstate_loading
= true;
821 ret
= qemu_load_device_state(fb
);
823 error_report("COLO: load device state failed");
824 qemu_mutex_unlock_iothread();
828 #ifdef CONFIG_REPLICATION
829 replication_get_error_all(&local_err
);
831 qemu_mutex_unlock_iothread();
835 /* discard colo disk buffer */
836 replication_do_checkpoint_all(&local_err
);
838 qemu_mutex_unlock_iothread();
844 /* Notify all filters of all NIC to do checkpoint */
845 colo_notify_filters_event(COLO_EVENT_CHECKPOINT
, &local_err
);
848 qemu_mutex_unlock_iothread();
852 vmstate_loading
= false;
854 trace_colo_vm_state_change("stop", "run");
855 qemu_mutex_unlock_iothread();
857 if (failover_get_state() == FAILOVER_STATUS_RELAUNCH
) {
858 failover_set_state(FAILOVER_STATUS_RELAUNCH
,
859 FAILOVER_STATUS_NONE
);
860 failover_request_active(NULL
);
864 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_VMSTATE_LOADED
,
872 vmstate_loading
= false;
873 /* Throw the unreported error message after exited from loop */
875 error_report_err(local_err
);
879 * There are only two reasons we can get here, some error happened
880 * or the user triggered failover.
882 switch (failover_get_state()) {
883 case FAILOVER_STATUS_COMPLETED
:
884 qapi_event_send_colo_exit(COLO_MODE_SECONDARY
,
885 COLO_EXIT_REASON_REQUEST
);
888 qapi_event_send_colo_exit(COLO_MODE_SECONDARY
,
889 COLO_EXIT_REASON_ERROR
);
896 /* Hope this not to be too long to loop here */
897 qemu_sem_wait(&mis
->colo_incoming_sem
);
898 qemu_sem_destroy(&mis
->colo_incoming_sem
);
899 /* Must be called after failover BH is completed */
900 if (mis
->to_src_file
) {
901 qemu_fclose(mis
->to_src_file
);
902 mis
->to_src_file
= NULL
;
905 rcu_unregister_thread();