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
);
125 /* For Secondary VM, jump to incoming co */
126 if (mis
->migration_incoming_co
) {
127 qemu_coroutine_enter(mis
->migration_incoming_co
);
134 static void primary_vm_do_failover(void)
136 #ifdef CONFIG_REPLICATION
137 MigrationState
*s
= migrate_get_current();
139 Error
*local_err
= NULL
;
141 migrate_set_state(&s
->state
, MIGRATION_STATUS_COLO
,
142 MIGRATION_STATUS_COMPLETED
);
144 * kick COLO thread which might wait at
145 * qemu_sem_wait(&s->colo_checkpoint_sem).
147 colo_checkpoint_notify(migrate_get_current());
150 * Wake up COLO thread which may blocked in recv() or send(),
151 * The s->rp_state.from_dst_file and s->to_dst_file may use the
152 * same fd, but we still shutdown the fd for twice, it is harmless.
154 if (s
->to_dst_file
) {
155 qemu_file_shutdown(s
->to_dst_file
);
157 if (s
->rp_state
.from_dst_file
) {
158 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
161 old_state
= failover_set_state(FAILOVER_STATUS_ACTIVE
,
162 FAILOVER_STATUS_COMPLETED
);
163 if (old_state
!= FAILOVER_STATUS_ACTIVE
) {
164 error_report("Incorrect state (%s) while doing failover for Primary VM",
165 FailoverStatus_str(old_state
));
169 replication_stop_all(true, &local_err
);
171 error_report_err(local_err
);
175 /* Notify COLO thread that failover work is finished */
176 qemu_sem_post(&s
->colo_exit_sem
);
182 COLOMode
get_colo_mode(void)
184 if (migration_in_colo_state()) {
185 return COLO_MODE_PRIMARY
;
186 } else if (migration_incoming_in_colo_state()) {
187 return COLO_MODE_SECONDARY
;
189 return COLO_MODE_NONE
;
193 void colo_do_failover(MigrationState
*s
)
195 /* Make sure VM stopped while failover happened. */
196 if (!colo_runstate_is_stopped()) {
197 vm_stop_force_state(RUN_STATE_COLO
);
200 if (get_colo_mode() == COLO_MODE_PRIMARY
) {
201 primary_vm_do_failover();
203 secondary_vm_do_failover();
207 #ifdef CONFIG_REPLICATION
208 void qmp_xen_set_replication(bool enable
, bool primary
,
209 bool has_failover
, bool failover
,
212 ReplicationMode mode
= primary
?
213 REPLICATION_MODE_PRIMARY
:
214 REPLICATION_MODE_SECONDARY
;
216 if (has_failover
&& enable
) {
217 error_setg(errp
, "Parameter 'failover' is only for"
218 " stopping replication");
223 replication_start_all(mode
, errp
);
228 replication_stop_all(failover
, failover
? NULL
: errp
);
232 ReplicationStatus
*qmp_query_xen_replication_status(Error
**errp
)
235 ReplicationStatus
*s
= g_new0(ReplicationStatus
, 1);
237 replication_get_error_all(&err
);
241 s
->desc
= g_strdup(error_get_pretty(err
));
250 void qmp_xen_colo_do_checkpoint(Error
**errp
)
252 replication_do_checkpoint_all(errp
);
256 COLOStatus
*qmp_query_colo_status(Error
**errp
)
258 COLOStatus
*s
= g_new0(COLOStatus
, 1);
260 s
->mode
= get_colo_mode();
262 switch (failover_get_state()) {
263 case FAILOVER_STATUS_NONE
:
264 s
->reason
= COLO_EXIT_REASON_NONE
;
266 case FAILOVER_STATUS_COMPLETED
:
267 s
->reason
= COLO_EXIT_REASON_REQUEST
;
270 if (migration_in_colo_state()) {
271 s
->reason
= COLO_EXIT_REASON_PROCESSING
;
273 s
->reason
= COLO_EXIT_REASON_ERROR
;
280 static void colo_send_message(QEMUFile
*f
, COLOMessage msg
,
285 if (msg
>= COLO_MESSAGE__MAX
) {
286 error_setg(errp
, "%s: Invalid message", __func__
);
289 qemu_put_be32(f
, msg
);
292 ret
= qemu_file_get_error(f
);
294 error_setg_errno(errp
, -ret
, "Can't send COLO message");
296 trace_colo_send_message(COLOMessage_str(msg
));
299 static void colo_send_message_value(QEMUFile
*f
, COLOMessage msg
,
300 uint64_t value
, Error
**errp
)
302 Error
*local_err
= NULL
;
305 colo_send_message(f
, msg
, &local_err
);
307 error_propagate(errp
, local_err
);
310 qemu_put_be64(f
, value
);
313 ret
= qemu_file_get_error(f
);
315 error_setg_errno(errp
, -ret
, "Failed to send value for message:%s",
316 COLOMessage_str(msg
));
320 static COLOMessage
colo_receive_message(QEMUFile
*f
, Error
**errp
)
325 msg
= qemu_get_be32(f
);
326 ret
= qemu_file_get_error(f
);
328 error_setg_errno(errp
, -ret
, "Can't receive COLO message");
331 if (msg
>= COLO_MESSAGE__MAX
) {
332 error_setg(errp
, "%s: Invalid message", __func__
);
335 trace_colo_receive_message(COLOMessage_str(msg
));
339 static void colo_receive_check_message(QEMUFile
*f
, COLOMessage expect_msg
,
343 Error
*local_err
= NULL
;
345 msg
= colo_receive_message(f
, &local_err
);
347 error_propagate(errp
, local_err
);
350 if (msg
!= expect_msg
) {
351 error_setg(errp
, "Unexpected COLO message %d, expected %d",
356 static uint64_t colo_receive_message_value(QEMUFile
*f
, uint32_t expect_msg
,
359 Error
*local_err
= NULL
;
363 colo_receive_check_message(f
, expect_msg
, &local_err
);
365 error_propagate(errp
, local_err
);
369 value
= qemu_get_be64(f
);
370 ret
= qemu_file_get_error(f
);
372 error_setg_errno(errp
, -ret
, "Failed to get value for COLO message: %s",
373 COLOMessage_str(expect_msg
));
378 static int colo_do_checkpoint_transaction(MigrationState
*s
,
379 QIOChannelBuffer
*bioc
,
382 Error
*local_err
= NULL
;
385 colo_send_message(s
->to_dst_file
, COLO_MESSAGE_CHECKPOINT_REQUEST
,
391 colo_receive_check_message(s
->rp_state
.from_dst_file
,
392 COLO_MESSAGE_CHECKPOINT_REPLY
, &local_err
);
396 /* Reset channel-buffer directly */
397 qio_channel_io_seek(QIO_CHANNEL(bioc
), 0, 0, NULL
);
400 qemu_mutex_lock_iothread();
401 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
402 qemu_mutex_unlock_iothread();
405 vm_stop_force_state(RUN_STATE_COLO
);
406 qemu_mutex_unlock_iothread();
407 trace_colo_vm_state_change("run", "stop");
409 * Failover request bh could be called after vm_stop_force_state(),
410 * So we need check failover_request_is_active() again.
412 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
416 colo_notify_compares_event(NULL
, COLO_EVENT_CHECKPOINT
, &local_err
);
421 /* Disable block migration */
422 migrate_set_block_enabled(false, &local_err
);
423 qemu_mutex_lock_iothread();
425 #ifdef CONFIG_REPLICATION
426 replication_do_checkpoint_all(&local_err
);
428 qemu_mutex_unlock_iothread();
435 colo_send_message(s
->to_dst_file
, COLO_MESSAGE_VMSTATE_SEND
, &local_err
);
437 qemu_mutex_unlock_iothread();
440 /* Note: device state is saved into buffer */
441 ret
= qemu_save_device_state(fb
);
443 qemu_mutex_unlock_iothread();
448 * Only save VM's live state, which not including device state.
449 * TODO: We may need a timeout mechanism to prevent COLO process
450 * to be blocked here.
452 qemu_savevm_live_state(s
->to_dst_file
);
457 * We need the size of the VMstate data in Secondary side,
458 * With which we can decide how much data should be read.
460 colo_send_message_value(s
->to_dst_file
, COLO_MESSAGE_VMSTATE_SIZE
,
461 bioc
->usage
, &local_err
);
466 qemu_put_buffer(s
->to_dst_file
, bioc
->data
, bioc
->usage
);
467 qemu_fflush(s
->to_dst_file
);
468 ret
= qemu_file_get_error(s
->to_dst_file
);
473 colo_receive_check_message(s
->rp_state
.from_dst_file
,
474 COLO_MESSAGE_VMSTATE_RECEIVED
, &local_err
);
479 colo_receive_check_message(s
->rp_state
.from_dst_file
,
480 COLO_MESSAGE_VMSTATE_LOADED
, &local_err
);
487 qemu_mutex_lock_iothread();
489 qemu_mutex_unlock_iothread();
490 trace_colo_vm_state_change("stop", "run");
494 error_report_err(local_err
);
499 static void colo_compare_notify_checkpoint(Notifier
*notifier
, void *data
)
501 colo_checkpoint_notify(data
);
504 static void colo_process_checkpoint(MigrationState
*s
)
506 QIOChannelBuffer
*bioc
;
508 int64_t current_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
509 Error
*local_err
= NULL
;
512 failover_init_state();
514 s
->rp_state
.from_dst_file
= qemu_file_get_return_path(s
->to_dst_file
);
515 if (!s
->rp_state
.from_dst_file
) {
516 error_report("Open QEMUFile from_dst_file failed");
520 packets_compare_notifier
.notify
= colo_compare_notify_checkpoint
;
521 colo_compare_register_notifier(&packets_compare_notifier
);
524 * Wait for Secondary finish loading VM states and enter COLO
527 colo_receive_check_message(s
->rp_state
.from_dst_file
,
528 COLO_MESSAGE_CHECKPOINT_READY
, &local_err
);
532 bioc
= qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE
);
533 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
534 object_unref(OBJECT(bioc
));
536 qemu_mutex_lock_iothread();
537 #ifdef CONFIG_REPLICATION
538 replication_start_all(REPLICATION_MODE_PRIMARY
, &local_err
);
540 qemu_mutex_unlock_iothread();
548 qemu_mutex_unlock_iothread();
549 trace_colo_vm_state_change("stop", "run");
551 timer_mod(s
->colo_delay_timer
,
552 current_time
+ s
->parameters
.x_checkpoint_delay
);
554 while (s
->state
== MIGRATION_STATUS_COLO
) {
555 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
556 error_report("failover request");
560 qemu_sem_wait(&s
->colo_checkpoint_sem
);
562 if (s
->state
!= MIGRATION_STATUS_COLO
) {
565 ret
= colo_do_checkpoint_transaction(s
, bioc
, fb
);
572 /* Throw the unreported error message after exited from loop */
574 error_report_err(local_err
);
582 * There are only two reasons we can get here, some error happened
583 * or the user triggered failover.
585 switch (failover_get_state()) {
586 case FAILOVER_STATUS_COMPLETED
:
587 qapi_event_send_colo_exit(COLO_MODE_PRIMARY
,
588 COLO_EXIT_REASON_REQUEST
);
591 qapi_event_send_colo_exit(COLO_MODE_PRIMARY
,
592 COLO_EXIT_REASON_ERROR
);
595 /* Hope this not to be too long to wait here */
596 qemu_sem_wait(&s
->colo_exit_sem
);
597 qemu_sem_destroy(&s
->colo_exit_sem
);
600 * It is safe to unregister notifier after failover finished.
601 * Besides, colo_delay_timer and colo_checkpoint_sem can't be
602 * released befor unregister notifier, or there will be use-after-free
605 colo_compare_unregister_notifier(&packets_compare_notifier
);
606 timer_del(s
->colo_delay_timer
);
607 timer_free(s
->colo_delay_timer
);
608 qemu_sem_destroy(&s
->colo_checkpoint_sem
);
611 * Must be called after failover BH is completed,
612 * Or the failover BH may shutdown the wrong fd that
613 * re-used by other threads after we release here.
615 if (s
->rp_state
.from_dst_file
) {
616 qemu_fclose(s
->rp_state
.from_dst_file
);
620 void colo_checkpoint_notify(void *opaque
)
622 MigrationState
*s
= opaque
;
623 int64_t next_notify_time
;
625 qemu_sem_post(&s
->colo_checkpoint_sem
);
626 s
->colo_checkpoint_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
627 next_notify_time
= s
->colo_checkpoint_time
+
628 s
->parameters
.x_checkpoint_delay
;
629 timer_mod(s
->colo_delay_timer
, next_notify_time
);
632 void migrate_start_colo_process(MigrationState
*s
)
634 qemu_mutex_unlock_iothread();
635 qemu_sem_init(&s
->colo_checkpoint_sem
, 0);
636 s
->colo_delay_timer
= timer_new_ms(QEMU_CLOCK_HOST
,
637 colo_checkpoint_notify
, s
);
639 qemu_sem_init(&s
->colo_exit_sem
, 0);
640 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
641 MIGRATION_STATUS_COLO
);
642 colo_process_checkpoint(s
);
643 qemu_mutex_lock_iothread();
646 static void colo_wait_handle_message(QEMUFile
*f
, int *checkpoint_request
,
650 Error
*local_err
= NULL
;
652 msg
= colo_receive_message(f
, &local_err
);
654 error_propagate(errp
, local_err
);
659 case COLO_MESSAGE_CHECKPOINT_REQUEST
:
660 *checkpoint_request
= 1;
663 *checkpoint_request
= 0;
664 error_setg(errp
, "Got unknown COLO message: %d", msg
);
669 void *colo_process_incoming_thread(void *opaque
)
671 MigrationIncomingState
*mis
= opaque
;
673 QIOChannelBuffer
*bioc
= NULL
; /* Cache incoming device state */
676 Error
*local_err
= NULL
;
679 rcu_register_thread();
680 qemu_sem_init(&mis
->colo_incoming_sem
, 0);
682 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
683 MIGRATION_STATUS_COLO
);
685 failover_init_state();
687 mis
->to_src_file
= qemu_file_get_return_path(mis
->from_src_file
);
688 if (!mis
->to_src_file
) {
689 error_report("COLO incoming thread: Open QEMUFile to_src_file failed");
693 * Note: the communication between Primary side and Secondary side
694 * should be sequential, we set the fd to unblocked in migration incoming
695 * coroutine, and here we are in the COLO incoming thread, so it is ok to
696 * set the fd back to blocked.
698 qemu_file_set_blocking(mis
->from_src_file
, true);
700 bioc
= qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE
);
701 fb
= qemu_fopen_channel_input(QIO_CHANNEL(bioc
));
702 object_unref(OBJECT(bioc
));
704 qemu_mutex_lock_iothread();
705 #ifdef CONFIG_REPLICATION
706 replication_start_all(REPLICATION_MODE_SECONDARY
, &local_err
);
708 qemu_mutex_unlock_iothread();
715 trace_colo_vm_state_change("stop", "run");
716 qemu_mutex_unlock_iothread();
718 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_CHECKPOINT_READY
,
724 while (mis
->state
== MIGRATION_STATUS_COLO
) {
727 colo_wait_handle_message(mis
->from_src_file
, &request
, &local_err
);
732 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
733 error_report("failover request");
737 qemu_mutex_lock_iothread();
738 vm_stop_force_state(RUN_STATE_COLO
);
739 trace_colo_vm_state_change("run", "stop");
740 qemu_mutex_unlock_iothread();
742 /* FIXME: This is unnecessary for periodic checkpoint mode */
743 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_CHECKPOINT_REPLY
,
749 colo_receive_check_message(mis
->from_src_file
,
750 COLO_MESSAGE_VMSTATE_SEND
, &local_err
);
755 qemu_mutex_lock_iothread();
756 cpu_synchronize_all_pre_loadvm();
757 ret
= qemu_loadvm_state_main(mis
->from_src_file
, mis
);
758 qemu_mutex_unlock_iothread();
761 error_report("Load VM's live state (ram) error");
765 value
= colo_receive_message_value(mis
->from_src_file
,
766 COLO_MESSAGE_VMSTATE_SIZE
, &local_err
);
772 * Read VM device state data into channel buffer,
773 * It's better to re-use the memory allocated.
774 * Here we need to handle the channel buffer directly.
776 if (value
> bioc
->capacity
) {
777 bioc
->capacity
= value
;
778 bioc
->data
= g_realloc(bioc
->data
, bioc
->capacity
);
780 total_size
= qemu_get_buffer(mis
->from_src_file
, bioc
->data
, value
);
781 if (total_size
!= value
) {
782 error_report("Got %" PRIu64
" VMState data, less than expected"
783 " %" PRIu64
, total_size
, value
);
786 bioc
->usage
= total_size
;
787 qio_channel_io_seek(QIO_CHANNEL(bioc
), 0, 0, NULL
);
789 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_VMSTATE_RECEIVED
,
795 qemu_mutex_lock_iothread();
796 vmstate_loading
= true;
797 ret
= qemu_load_device_state(fb
);
799 error_report("COLO: load device state failed");
800 qemu_mutex_unlock_iothread();
804 #ifdef CONFIG_REPLICATION
805 replication_get_error_all(&local_err
);
807 qemu_mutex_unlock_iothread();
811 /* discard colo disk buffer */
812 replication_do_checkpoint_all(&local_err
);
814 qemu_mutex_unlock_iothread();
820 /* Notify all filters of all NIC to do checkpoint */
821 colo_notify_filters_event(COLO_EVENT_CHECKPOINT
, &local_err
);
824 qemu_mutex_unlock_iothread();
828 vmstate_loading
= false;
830 trace_colo_vm_state_change("stop", "run");
831 qemu_mutex_unlock_iothread();
833 if (failover_get_state() == FAILOVER_STATUS_RELAUNCH
) {
834 failover_set_state(FAILOVER_STATUS_RELAUNCH
,
835 FAILOVER_STATUS_NONE
);
836 failover_request_active(NULL
);
840 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_VMSTATE_LOADED
,
848 vmstate_loading
= false;
849 /* Throw the unreported error message after exited from loop */
851 error_report_err(local_err
);
855 * There are only two reasons we can get here, some error happened
856 * or the user triggered failover.
858 switch (failover_get_state()) {
859 case FAILOVER_STATUS_COMPLETED
:
860 qapi_event_send_colo_exit(COLO_MODE_SECONDARY
,
861 COLO_EXIT_REASON_REQUEST
);
864 qapi_event_send_colo_exit(COLO_MODE_SECONDARY
,
865 COLO_EXIT_REASON_ERROR
);
872 /* Hope this not to be too long to loop here */
873 qemu_sem_wait(&mis
->colo_incoming_sem
);
874 qemu_sem_destroy(&mis
->colo_incoming_sem
);
875 /* Must be called after failover BH is completed */
876 if (mis
->to_src_file
) {
877 qemu_fclose(mis
->to_src_file
);
878 mis
->to_src_file
= NULL
;
881 rcu_unregister_thread();