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 "qemu/main-loop.h"
28 #include "migration/failover.h"
29 #ifdef CONFIG_REPLICATION
30 #include "replication.h"
32 #include "net/colo-compare.h"
34 #include "block/block.h"
35 #include "qapi/qapi-events-migration.h"
36 #include "qapi/qmp/qerror.h"
37 #include "sysemu/cpus.h"
38 #include "sysemu/runstate.h"
39 #include "net/filter.h"
41 static bool vmstate_loading
;
42 static Notifier packets_compare_notifier
;
44 /* User need to know colo mode after COLO failover */
45 static COLOMode last_colo_mode
;
47 #define COLO_BUFFER_BASE_SIZE (4 * 1024 * 1024)
49 bool migration_in_colo_state(void)
51 MigrationState
*s
= migrate_get_current();
53 return (s
->state
== MIGRATION_STATUS_COLO
);
56 bool migration_incoming_in_colo_state(void)
58 MigrationIncomingState
*mis
= migration_incoming_get_current();
60 return mis
&& (mis
->state
== MIGRATION_STATUS_COLO
);
63 static bool colo_runstate_is_stopped(void)
65 return runstate_check(RUN_STATE_COLO
) || !runstate_is_running();
68 static void secondary_vm_do_failover(void)
70 /* COLO needs enable block-replication */
71 #ifdef CONFIG_REPLICATION
73 MigrationIncomingState
*mis
= migration_incoming_get_current();
74 Error
*local_err
= NULL
;
76 /* Can not do failover during the process of VM's loading VMstate, Or
77 * it will break the secondary VM.
79 if (vmstate_loading
) {
80 old_state
= failover_set_state(FAILOVER_STATUS_ACTIVE
,
81 FAILOVER_STATUS_RELAUNCH
);
82 if (old_state
!= FAILOVER_STATUS_ACTIVE
) {
83 error_report("Unknown error while do failover for secondary VM,"
84 "old_state: %s", FailoverStatus_str(old_state
));
89 migrate_set_state(&mis
->state
, MIGRATION_STATUS_COLO
,
90 MIGRATION_STATUS_COMPLETED
);
92 replication_stop_all(true, &local_err
);
94 error_report_err(local_err
);
97 /* Notify all filters of all NIC to do checkpoint */
98 colo_notify_filters_event(COLO_EVENT_FAILOVER
, &local_err
);
100 error_report_err(local_err
);
104 error_report("\"-S\" qemu option will be ignored in secondary side");
105 /* recover runstate to normal migration finish state */
109 * Make sure COLO incoming thread not block in recv or send,
110 * If mis->from_src_file and mis->to_src_file use the same fd,
111 * The second shutdown() will return -1, we ignore this value,
114 if (mis
->from_src_file
) {
115 qemu_file_shutdown(mis
->from_src_file
);
117 if (mis
->to_src_file
) {
118 qemu_file_shutdown(mis
->to_src_file
);
121 old_state
= failover_set_state(FAILOVER_STATUS_ACTIVE
,
122 FAILOVER_STATUS_COMPLETED
);
123 if (old_state
!= FAILOVER_STATUS_ACTIVE
) {
124 error_report("Incorrect state (%s) while doing failover for "
125 "secondary VM", FailoverStatus_str(old_state
));
128 /* Notify COLO incoming thread that failover work is finished */
129 qemu_sem_post(&mis
->colo_incoming_sem
);
131 /* For Secondary VM, jump to incoming co */
132 if (mis
->migration_incoming_co
) {
133 qemu_coroutine_enter(mis
->migration_incoming_co
);
140 static void primary_vm_do_failover(void)
142 #ifdef CONFIG_REPLICATION
143 MigrationState
*s
= migrate_get_current();
145 Error
*local_err
= NULL
;
147 migrate_set_state(&s
->state
, MIGRATION_STATUS_COLO
,
148 MIGRATION_STATUS_COMPLETED
);
150 * kick COLO thread which might wait at
151 * qemu_sem_wait(&s->colo_checkpoint_sem).
153 colo_checkpoint_notify(migrate_get_current());
156 * Wake up COLO thread which may blocked in recv() or send(),
157 * The s->rp_state.from_dst_file and s->to_dst_file may use the
158 * same fd, but we still shutdown the fd for twice, it is harmless.
160 if (s
->to_dst_file
) {
161 qemu_file_shutdown(s
->to_dst_file
);
163 if (s
->rp_state
.from_dst_file
) {
164 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
167 old_state
= failover_set_state(FAILOVER_STATUS_ACTIVE
,
168 FAILOVER_STATUS_COMPLETED
);
169 if (old_state
!= FAILOVER_STATUS_ACTIVE
) {
170 error_report("Incorrect state (%s) while doing failover for Primary VM",
171 FailoverStatus_str(old_state
));
175 replication_stop_all(true, &local_err
);
177 error_report_err(local_err
);
181 /* Notify COLO thread that failover work is finished */
182 qemu_sem_post(&s
->colo_exit_sem
);
188 COLOMode
get_colo_mode(void)
190 if (migration_in_colo_state()) {
191 return COLO_MODE_PRIMARY
;
192 } else if (migration_incoming_in_colo_state()) {
193 return COLO_MODE_SECONDARY
;
195 return COLO_MODE_NONE
;
199 void colo_do_failover(void)
201 /* Make sure VM stopped while failover happened. */
202 if (!colo_runstate_is_stopped()) {
203 vm_stop_force_state(RUN_STATE_COLO
);
206 switch (get_colo_mode()) {
207 case COLO_MODE_PRIMARY
:
208 primary_vm_do_failover();
210 case COLO_MODE_SECONDARY
:
211 secondary_vm_do_failover();
214 error_report("colo_do_failover failed because the colo mode"
215 " could not be obtained");
219 #ifdef CONFIG_REPLICATION
220 void qmp_xen_set_replication(bool enable
, bool primary
,
221 bool has_failover
, bool failover
,
224 ReplicationMode mode
= primary
?
225 REPLICATION_MODE_PRIMARY
:
226 REPLICATION_MODE_SECONDARY
;
228 if (has_failover
&& enable
) {
229 error_setg(errp
, "Parameter 'failover' is only for"
230 " stopping replication");
235 replication_start_all(mode
, errp
);
240 replication_stop_all(failover
, failover
? NULL
: errp
);
244 ReplicationStatus
*qmp_query_xen_replication_status(Error
**errp
)
247 ReplicationStatus
*s
= g_new0(ReplicationStatus
, 1);
249 replication_get_error_all(&err
);
253 s
->desc
= g_strdup(error_get_pretty(err
));
262 void qmp_xen_colo_do_checkpoint(Error
**errp
)
264 replication_do_checkpoint_all(errp
);
265 /* Notify all filters of all NIC to do checkpoint */
266 colo_notify_filters_event(COLO_EVENT_CHECKPOINT
, errp
);
270 COLOStatus
*qmp_query_colo_status(Error
**errp
)
272 COLOStatus
*s
= g_new0(COLOStatus
, 1);
274 s
->mode
= get_colo_mode();
275 s
->last_mode
= last_colo_mode
;
277 switch (failover_get_state()) {
278 case FAILOVER_STATUS_NONE
:
279 s
->reason
= COLO_EXIT_REASON_NONE
;
281 case FAILOVER_STATUS_COMPLETED
:
282 s
->reason
= COLO_EXIT_REASON_REQUEST
;
285 if (migration_in_colo_state()) {
286 s
->reason
= COLO_EXIT_REASON_PROCESSING
;
288 s
->reason
= COLO_EXIT_REASON_ERROR
;
295 static void colo_send_message(QEMUFile
*f
, COLOMessage msg
,
300 if (msg
>= COLO_MESSAGE__MAX
) {
301 error_setg(errp
, "%s: Invalid message", __func__
);
304 qemu_put_be32(f
, msg
);
307 ret
= qemu_file_get_error(f
);
309 error_setg_errno(errp
, -ret
, "Can't send COLO message");
311 trace_colo_send_message(COLOMessage_str(msg
));
314 static void colo_send_message_value(QEMUFile
*f
, COLOMessage msg
,
315 uint64_t value
, Error
**errp
)
317 Error
*local_err
= NULL
;
320 colo_send_message(f
, msg
, &local_err
);
322 error_propagate(errp
, local_err
);
325 qemu_put_be64(f
, value
);
328 ret
= qemu_file_get_error(f
);
330 error_setg_errno(errp
, -ret
, "Failed to send value for message:%s",
331 COLOMessage_str(msg
));
335 static COLOMessage
colo_receive_message(QEMUFile
*f
, Error
**errp
)
340 msg
= qemu_get_be32(f
);
341 ret
= qemu_file_get_error(f
);
343 error_setg_errno(errp
, -ret
, "Can't receive COLO message");
346 if (msg
>= COLO_MESSAGE__MAX
) {
347 error_setg(errp
, "%s: Invalid message", __func__
);
350 trace_colo_receive_message(COLOMessage_str(msg
));
354 static void colo_receive_check_message(QEMUFile
*f
, COLOMessage expect_msg
,
358 Error
*local_err
= NULL
;
360 msg
= colo_receive_message(f
, &local_err
);
362 error_propagate(errp
, local_err
);
365 if (msg
!= expect_msg
) {
366 error_setg(errp
, "Unexpected COLO message %d, expected %d",
371 static uint64_t colo_receive_message_value(QEMUFile
*f
, uint32_t expect_msg
,
374 Error
*local_err
= NULL
;
378 colo_receive_check_message(f
, expect_msg
, &local_err
);
380 error_propagate(errp
, local_err
);
384 value
= qemu_get_be64(f
);
385 ret
= qemu_file_get_error(f
);
387 error_setg_errno(errp
, -ret
, "Failed to get value for COLO message: %s",
388 COLOMessage_str(expect_msg
));
393 static int colo_do_checkpoint_transaction(MigrationState
*s
,
394 QIOChannelBuffer
*bioc
,
397 Error
*local_err
= NULL
;
400 colo_send_message(s
->to_dst_file
, COLO_MESSAGE_CHECKPOINT_REQUEST
,
406 colo_receive_check_message(s
->rp_state
.from_dst_file
,
407 COLO_MESSAGE_CHECKPOINT_REPLY
, &local_err
);
411 /* Reset channel-buffer directly */
412 qio_channel_io_seek(QIO_CHANNEL(bioc
), 0, 0, NULL
);
415 qemu_mutex_lock_iothread();
416 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
417 qemu_mutex_unlock_iothread();
420 vm_stop_force_state(RUN_STATE_COLO
);
421 qemu_mutex_unlock_iothread();
422 trace_colo_vm_state_change("run", "stop");
424 * Failover request bh could be called after vm_stop_force_state(),
425 * So we need check failover_request_is_active() again.
427 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
431 colo_notify_compares_event(NULL
, COLO_EVENT_CHECKPOINT
, &local_err
);
436 /* Disable block migration */
437 migrate_set_block_enabled(false, &local_err
);
438 qemu_mutex_lock_iothread();
440 #ifdef CONFIG_REPLICATION
441 replication_do_checkpoint_all(&local_err
);
443 qemu_mutex_unlock_iothread();
450 colo_send_message(s
->to_dst_file
, COLO_MESSAGE_VMSTATE_SEND
, &local_err
);
452 qemu_mutex_unlock_iothread();
455 /* Note: device state is saved into buffer */
456 ret
= qemu_save_device_state(fb
);
458 qemu_mutex_unlock_iothread();
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.
467 qemu_savevm_live_state(s
->to_dst_file
);
472 * We need the size of the VMstate data in Secondary side,
473 * With which we can decide how much data should be read.
475 colo_send_message_value(s
->to_dst_file
, COLO_MESSAGE_VMSTATE_SIZE
,
476 bioc
->usage
, &local_err
);
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
);
488 colo_receive_check_message(s
->rp_state
.from_dst_file
,
489 COLO_MESSAGE_VMSTATE_RECEIVED
, &local_err
);
494 colo_receive_check_message(s
->rp_state
.from_dst_file
,
495 COLO_MESSAGE_VMSTATE_LOADED
, &local_err
);
502 qemu_mutex_lock_iothread();
504 qemu_mutex_unlock_iothread();
505 trace_colo_vm_state_change("stop", "run");
509 error_report_err(local_err
);
514 static void colo_compare_notify_checkpoint(Notifier
*notifier
, void *data
)
516 colo_checkpoint_notify(data
);
519 static void colo_process_checkpoint(MigrationState
*s
)
521 QIOChannelBuffer
*bioc
;
523 int64_t current_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
524 Error
*local_err
= NULL
;
527 last_colo_mode
= get_colo_mode();
528 if (last_colo_mode
!= COLO_MODE_PRIMARY
) {
529 error_report("COLO mode must be COLO_MODE_PRIMARY");
533 failover_init_state();
535 s
->rp_state
.from_dst_file
= qemu_file_get_return_path(s
->to_dst_file
);
536 if (!s
->rp_state
.from_dst_file
) {
537 error_report("Open QEMUFile from_dst_file failed");
541 packets_compare_notifier
.notify
= colo_compare_notify_checkpoint
;
542 colo_compare_register_notifier(&packets_compare_notifier
);
545 * Wait for Secondary finish loading VM states and enter COLO
548 colo_receive_check_message(s
->rp_state
.from_dst_file
,
549 COLO_MESSAGE_CHECKPOINT_READY
, &local_err
);
553 bioc
= qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE
);
554 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
555 object_unref(OBJECT(bioc
));
557 qemu_mutex_lock_iothread();
558 #ifdef CONFIG_REPLICATION
559 replication_start_all(REPLICATION_MODE_PRIMARY
, &local_err
);
561 qemu_mutex_unlock_iothread();
569 qemu_mutex_unlock_iothread();
570 trace_colo_vm_state_change("stop", "run");
572 timer_mod(s
->colo_delay_timer
,
573 current_time
+ s
->parameters
.x_checkpoint_delay
);
575 while (s
->state
== MIGRATION_STATUS_COLO
) {
576 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
577 error_report("failover request");
581 qemu_sem_wait(&s
->colo_checkpoint_sem
);
583 if (s
->state
!= MIGRATION_STATUS_COLO
) {
586 ret
= colo_do_checkpoint_transaction(s
, bioc
, fb
);
593 /* Throw the unreported error message after exited from loop */
595 error_report_err(local_err
);
603 * There are only two reasons we can get here, some error happened
604 * or the user triggered failover.
606 switch (failover_get_state()) {
607 case FAILOVER_STATUS_COMPLETED
:
608 qapi_event_send_colo_exit(COLO_MODE_PRIMARY
,
609 COLO_EXIT_REASON_REQUEST
);
612 qapi_event_send_colo_exit(COLO_MODE_PRIMARY
,
613 COLO_EXIT_REASON_ERROR
);
616 /* Hope this not to be too long to wait here */
617 qemu_sem_wait(&s
->colo_exit_sem
);
618 qemu_sem_destroy(&s
->colo_exit_sem
);
621 * It is safe to unregister notifier after failover finished.
622 * Besides, colo_delay_timer and colo_checkpoint_sem can't be
623 * released befor unregister notifier, or there will be use-after-free
626 colo_compare_unregister_notifier(&packets_compare_notifier
);
627 timer_del(s
->colo_delay_timer
);
628 timer_free(s
->colo_delay_timer
);
629 qemu_sem_destroy(&s
->colo_checkpoint_sem
);
632 * Must be called after failover BH is completed,
633 * Or the failover BH may shutdown the wrong fd that
634 * re-used by other threads after we release here.
636 if (s
->rp_state
.from_dst_file
) {
637 qemu_fclose(s
->rp_state
.from_dst_file
);
641 void colo_checkpoint_notify(void *opaque
)
643 MigrationState
*s
= opaque
;
644 int64_t next_notify_time
;
646 qemu_sem_post(&s
->colo_checkpoint_sem
);
647 s
->colo_checkpoint_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
648 next_notify_time
= s
->colo_checkpoint_time
+
649 s
->parameters
.x_checkpoint_delay
;
650 timer_mod(s
->colo_delay_timer
, next_notify_time
);
653 void migrate_start_colo_process(MigrationState
*s
)
655 qemu_mutex_unlock_iothread();
656 qemu_sem_init(&s
->colo_checkpoint_sem
, 0);
657 s
->colo_delay_timer
= timer_new_ms(QEMU_CLOCK_HOST
,
658 colo_checkpoint_notify
, s
);
660 qemu_sem_init(&s
->colo_exit_sem
, 0);
661 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
662 MIGRATION_STATUS_COLO
);
663 colo_process_checkpoint(s
);
664 qemu_mutex_lock_iothread();
667 static void colo_wait_handle_message(QEMUFile
*f
, int *checkpoint_request
,
671 Error
*local_err
= NULL
;
673 msg
= colo_receive_message(f
, &local_err
);
675 error_propagate(errp
, local_err
);
680 case COLO_MESSAGE_CHECKPOINT_REQUEST
:
681 *checkpoint_request
= 1;
684 *checkpoint_request
= 0;
685 error_setg(errp
, "Got unknown COLO message: %d", msg
);
690 void *colo_process_incoming_thread(void *opaque
)
692 MigrationIncomingState
*mis
= opaque
;
694 QIOChannelBuffer
*bioc
= NULL
; /* Cache incoming device state */
697 Error
*local_err
= NULL
;
700 rcu_register_thread();
701 qemu_sem_init(&mis
->colo_incoming_sem
, 0);
703 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
704 MIGRATION_STATUS_COLO
);
706 last_colo_mode
= get_colo_mode();
707 if (last_colo_mode
!= COLO_MODE_SECONDARY
) {
708 error_report("COLO mode must be COLO_MODE_SECONDARY");
712 failover_init_state();
714 mis
->to_src_file
= qemu_file_get_return_path(mis
->from_src_file
);
715 if (!mis
->to_src_file
) {
716 error_report("COLO incoming thread: Open QEMUFile to_src_file failed");
720 * Note: the communication between Primary side and Secondary side
721 * should be sequential, we set the fd to unblocked in migration incoming
722 * coroutine, and here we are in the COLO incoming thread, so it is ok to
723 * set the fd back to blocked.
725 qemu_file_set_blocking(mis
->from_src_file
, true);
727 bioc
= qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE
);
728 fb
= qemu_fopen_channel_input(QIO_CHANNEL(bioc
));
729 object_unref(OBJECT(bioc
));
731 qemu_mutex_lock_iothread();
732 #ifdef CONFIG_REPLICATION
733 replication_start_all(REPLICATION_MODE_SECONDARY
, &local_err
);
735 qemu_mutex_unlock_iothread();
742 trace_colo_vm_state_change("stop", "run");
743 qemu_mutex_unlock_iothread();
745 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_CHECKPOINT_READY
,
751 while (mis
->state
== MIGRATION_STATUS_COLO
) {
754 colo_wait_handle_message(mis
->from_src_file
, &request
, &local_err
);
759 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
760 error_report("failover request");
764 qemu_mutex_lock_iothread();
765 vm_stop_force_state(RUN_STATE_COLO
);
766 trace_colo_vm_state_change("run", "stop");
767 qemu_mutex_unlock_iothread();
769 /* FIXME: This is unnecessary for periodic checkpoint mode */
770 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_CHECKPOINT_REPLY
,
776 colo_receive_check_message(mis
->from_src_file
,
777 COLO_MESSAGE_VMSTATE_SEND
, &local_err
);
782 qemu_mutex_lock_iothread();
783 cpu_synchronize_all_pre_loadvm();
784 ret
= qemu_loadvm_state_main(mis
->from_src_file
, mis
);
785 qemu_mutex_unlock_iothread();
788 error_report("Load VM's live state (ram) error");
792 value
= colo_receive_message_value(mis
->from_src_file
,
793 COLO_MESSAGE_VMSTATE_SIZE
, &local_err
);
799 * Read VM device state data into channel buffer,
800 * It's better to re-use the memory allocated.
801 * Here we need to handle the channel buffer directly.
803 if (value
> bioc
->capacity
) {
804 bioc
->capacity
= value
;
805 bioc
->data
= g_realloc(bioc
->data
, bioc
->capacity
);
807 total_size
= qemu_get_buffer(mis
->from_src_file
, bioc
->data
, value
);
808 if (total_size
!= value
) {
809 error_report("Got %" PRIu64
" VMState data, less than expected"
810 " %" PRIu64
, total_size
, value
);
813 bioc
->usage
= total_size
;
814 qio_channel_io_seek(QIO_CHANNEL(bioc
), 0, 0, NULL
);
816 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_VMSTATE_RECEIVED
,
822 qemu_mutex_lock_iothread();
823 vmstate_loading
= true;
824 ret
= qemu_load_device_state(fb
);
826 error_report("COLO: load device state failed");
827 qemu_mutex_unlock_iothread();
831 #ifdef CONFIG_REPLICATION
832 replication_get_error_all(&local_err
);
834 qemu_mutex_unlock_iothread();
838 /* discard colo disk buffer */
839 replication_do_checkpoint_all(&local_err
);
841 qemu_mutex_unlock_iothread();
847 /* Notify all filters of all NIC to do checkpoint */
848 colo_notify_filters_event(COLO_EVENT_CHECKPOINT
, &local_err
);
851 qemu_mutex_unlock_iothread();
855 vmstate_loading
= false;
857 trace_colo_vm_state_change("stop", "run");
858 qemu_mutex_unlock_iothread();
860 if (failover_get_state() == FAILOVER_STATUS_RELAUNCH
) {
861 failover_set_state(FAILOVER_STATUS_RELAUNCH
,
862 FAILOVER_STATUS_NONE
);
863 failover_request_active(NULL
);
867 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_VMSTATE_LOADED
,
875 vmstate_loading
= false;
876 /* Throw the unreported error message after exited from loop */
878 error_report_err(local_err
);
882 * There are only two reasons we can get here, some error happened
883 * or the user triggered failover.
885 switch (failover_get_state()) {
886 case FAILOVER_STATUS_COMPLETED
:
887 qapi_event_send_colo_exit(COLO_MODE_SECONDARY
,
888 COLO_EXIT_REASON_REQUEST
);
891 qapi_event_send_colo_exit(COLO_MODE_SECONDARY
,
892 COLO_EXIT_REASON_ERROR
);
899 /* Hope this not to be too long to loop here */
900 qemu_sem_wait(&mis
->colo_incoming_sem
);
901 qemu_sem_destroy(&mis
->colo_incoming_sem
);
902 /* Must be called after failover BH is completed */
903 if (mis
->to_src_file
) {
904 qemu_fclose(mis
->to_src_file
);
905 mis
->to_src_file
= NULL
;
908 rcu_unregister_thread();