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 s
->reason
= COLO_EXIT_REASON_ERROR
;
276 static void colo_send_message(QEMUFile
*f
, COLOMessage msg
,
281 if (msg
>= COLO_MESSAGE__MAX
) {
282 error_setg(errp
, "%s: Invalid message", __func__
);
285 qemu_put_be32(f
, msg
);
288 ret
= qemu_file_get_error(f
);
290 error_setg_errno(errp
, -ret
, "Can't send COLO message");
292 trace_colo_send_message(COLOMessage_str(msg
));
295 static void colo_send_message_value(QEMUFile
*f
, COLOMessage msg
,
296 uint64_t value
, Error
**errp
)
298 Error
*local_err
= NULL
;
301 colo_send_message(f
, msg
, &local_err
);
303 error_propagate(errp
, local_err
);
306 qemu_put_be64(f
, value
);
309 ret
= qemu_file_get_error(f
);
311 error_setg_errno(errp
, -ret
, "Failed to send value for message:%s",
312 COLOMessage_str(msg
));
316 static COLOMessage
colo_receive_message(QEMUFile
*f
, Error
**errp
)
321 msg
= qemu_get_be32(f
);
322 ret
= qemu_file_get_error(f
);
324 error_setg_errno(errp
, -ret
, "Can't receive COLO message");
327 if (msg
>= COLO_MESSAGE__MAX
) {
328 error_setg(errp
, "%s: Invalid message", __func__
);
331 trace_colo_receive_message(COLOMessage_str(msg
));
335 static void colo_receive_check_message(QEMUFile
*f
, COLOMessage expect_msg
,
339 Error
*local_err
= NULL
;
341 msg
= colo_receive_message(f
, &local_err
);
343 error_propagate(errp
, local_err
);
346 if (msg
!= expect_msg
) {
347 error_setg(errp
, "Unexpected COLO message %d, expected %d",
352 static uint64_t colo_receive_message_value(QEMUFile
*f
, uint32_t expect_msg
,
355 Error
*local_err
= NULL
;
359 colo_receive_check_message(f
, expect_msg
, &local_err
);
361 error_propagate(errp
, local_err
);
365 value
= qemu_get_be64(f
);
366 ret
= qemu_file_get_error(f
);
368 error_setg_errno(errp
, -ret
, "Failed to get value for COLO message: %s",
369 COLOMessage_str(expect_msg
));
374 static int colo_do_checkpoint_transaction(MigrationState
*s
,
375 QIOChannelBuffer
*bioc
,
378 Error
*local_err
= NULL
;
381 colo_send_message(s
->to_dst_file
, COLO_MESSAGE_CHECKPOINT_REQUEST
,
387 colo_receive_check_message(s
->rp_state
.from_dst_file
,
388 COLO_MESSAGE_CHECKPOINT_REPLY
, &local_err
);
392 /* Reset channel-buffer directly */
393 qio_channel_io_seek(QIO_CHANNEL(bioc
), 0, 0, NULL
);
396 qemu_mutex_lock_iothread();
397 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
398 qemu_mutex_unlock_iothread();
401 vm_stop_force_state(RUN_STATE_COLO
);
402 qemu_mutex_unlock_iothread();
403 trace_colo_vm_state_change("run", "stop");
405 * Failover request bh could be called after vm_stop_force_state(),
406 * So we need check failover_request_is_active() again.
408 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
412 colo_notify_compares_event(NULL
, COLO_EVENT_CHECKPOINT
, &local_err
);
417 /* Disable block migration */
418 migrate_set_block_enabled(false, &local_err
);
419 qemu_mutex_lock_iothread();
421 #ifdef CONFIG_REPLICATION
422 replication_do_checkpoint_all(&local_err
);
424 qemu_mutex_unlock_iothread();
431 colo_send_message(s
->to_dst_file
, COLO_MESSAGE_VMSTATE_SEND
, &local_err
);
433 qemu_mutex_unlock_iothread();
436 /* Note: device state is saved into buffer */
437 ret
= qemu_save_device_state(fb
);
439 qemu_mutex_unlock_iothread();
444 * Only save VM's live state, which not including device state.
445 * TODO: We may need a timeout mechanism to prevent COLO process
446 * to be blocked here.
448 qemu_savevm_live_state(s
->to_dst_file
);
453 * We need the size of the VMstate data in Secondary side,
454 * With which we can decide how much data should be read.
456 colo_send_message_value(s
->to_dst_file
, COLO_MESSAGE_VMSTATE_SIZE
,
457 bioc
->usage
, &local_err
);
462 qemu_put_buffer(s
->to_dst_file
, bioc
->data
, bioc
->usage
);
463 qemu_fflush(s
->to_dst_file
);
464 ret
= qemu_file_get_error(s
->to_dst_file
);
469 colo_receive_check_message(s
->rp_state
.from_dst_file
,
470 COLO_MESSAGE_VMSTATE_RECEIVED
, &local_err
);
475 colo_receive_check_message(s
->rp_state
.from_dst_file
,
476 COLO_MESSAGE_VMSTATE_LOADED
, &local_err
);
483 qemu_mutex_lock_iothread();
485 qemu_mutex_unlock_iothread();
486 trace_colo_vm_state_change("stop", "run");
490 error_report_err(local_err
);
495 static void colo_compare_notify_checkpoint(Notifier
*notifier
, void *data
)
497 colo_checkpoint_notify(data
);
500 static void colo_process_checkpoint(MigrationState
*s
)
502 QIOChannelBuffer
*bioc
;
504 int64_t current_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
505 Error
*local_err
= NULL
;
508 failover_init_state();
510 s
->rp_state
.from_dst_file
= qemu_file_get_return_path(s
->to_dst_file
);
511 if (!s
->rp_state
.from_dst_file
) {
512 error_report("Open QEMUFile from_dst_file failed");
516 packets_compare_notifier
.notify
= colo_compare_notify_checkpoint
;
517 colo_compare_register_notifier(&packets_compare_notifier
);
520 * Wait for Secondary finish loading VM states and enter COLO
523 colo_receive_check_message(s
->rp_state
.from_dst_file
,
524 COLO_MESSAGE_CHECKPOINT_READY
, &local_err
);
528 bioc
= qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE
);
529 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
530 object_unref(OBJECT(bioc
));
532 qemu_mutex_lock_iothread();
533 #ifdef CONFIG_REPLICATION
534 replication_start_all(REPLICATION_MODE_PRIMARY
, &local_err
);
536 qemu_mutex_unlock_iothread();
544 qemu_mutex_unlock_iothread();
545 trace_colo_vm_state_change("stop", "run");
547 timer_mod(s
->colo_delay_timer
,
548 current_time
+ s
->parameters
.x_checkpoint_delay
);
550 while (s
->state
== MIGRATION_STATUS_COLO
) {
551 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
552 error_report("failover request");
556 qemu_sem_wait(&s
->colo_checkpoint_sem
);
558 if (s
->state
!= MIGRATION_STATUS_COLO
) {
561 ret
= colo_do_checkpoint_transaction(s
, bioc
, fb
);
568 /* Throw the unreported error message after exited from loop */
570 error_report_err(local_err
);
578 * There are only two reasons we can get here, some error happened
579 * or the user triggered failover.
581 switch (failover_get_state()) {
582 case FAILOVER_STATUS_NONE
:
583 qapi_event_send_colo_exit(COLO_MODE_PRIMARY
,
584 COLO_EXIT_REASON_ERROR
);
586 case FAILOVER_STATUS_COMPLETED
:
587 qapi_event_send_colo_exit(COLO_MODE_PRIMARY
,
588 COLO_EXIT_REASON_REQUEST
);
594 /* Hope this not to be too long to wait here */
595 qemu_sem_wait(&s
->colo_exit_sem
);
596 qemu_sem_destroy(&s
->colo_exit_sem
);
599 * It is safe to unregister notifier after failover finished.
600 * Besides, colo_delay_timer and colo_checkpoint_sem can't be
601 * released befor unregister notifier, or there will be use-after-free
604 colo_compare_unregister_notifier(&packets_compare_notifier
);
605 timer_del(s
->colo_delay_timer
);
606 timer_free(s
->colo_delay_timer
);
607 qemu_sem_destroy(&s
->colo_checkpoint_sem
);
610 * Must be called after failover BH is completed,
611 * Or the failover BH may shutdown the wrong fd that
612 * re-used by other threads after we release here.
614 if (s
->rp_state
.from_dst_file
) {
615 qemu_fclose(s
->rp_state
.from_dst_file
);
619 void colo_checkpoint_notify(void *opaque
)
621 MigrationState
*s
= opaque
;
622 int64_t next_notify_time
;
624 qemu_sem_post(&s
->colo_checkpoint_sem
);
625 s
->colo_checkpoint_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
626 next_notify_time
= s
->colo_checkpoint_time
+
627 s
->parameters
.x_checkpoint_delay
;
628 timer_mod(s
->colo_delay_timer
, next_notify_time
);
631 void migrate_start_colo_process(MigrationState
*s
)
633 qemu_mutex_unlock_iothread();
634 qemu_sem_init(&s
->colo_checkpoint_sem
, 0);
635 s
->colo_delay_timer
= timer_new_ms(QEMU_CLOCK_HOST
,
636 colo_checkpoint_notify
, s
);
638 qemu_sem_init(&s
->colo_exit_sem
, 0);
639 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
640 MIGRATION_STATUS_COLO
);
641 colo_process_checkpoint(s
);
642 qemu_mutex_lock_iothread();
645 static void colo_wait_handle_message(QEMUFile
*f
, int *checkpoint_request
,
649 Error
*local_err
= NULL
;
651 msg
= colo_receive_message(f
, &local_err
);
653 error_propagate(errp
, local_err
);
658 case COLO_MESSAGE_CHECKPOINT_REQUEST
:
659 *checkpoint_request
= 1;
662 *checkpoint_request
= 0;
663 error_setg(errp
, "Got unknown COLO message: %d", msg
);
668 void *colo_process_incoming_thread(void *opaque
)
670 MigrationIncomingState
*mis
= opaque
;
672 QIOChannelBuffer
*bioc
= NULL
; /* Cache incoming device state */
675 Error
*local_err
= NULL
;
678 rcu_register_thread();
679 qemu_sem_init(&mis
->colo_incoming_sem
, 0);
681 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
682 MIGRATION_STATUS_COLO
);
684 failover_init_state();
686 mis
->to_src_file
= qemu_file_get_return_path(mis
->from_src_file
);
687 if (!mis
->to_src_file
) {
688 error_report("COLO incoming thread: Open QEMUFile to_src_file failed");
692 * Note: the communication between Primary side and Secondary side
693 * should be sequential, we set the fd to unblocked in migration incoming
694 * coroutine, and here we are in the COLO incoming thread, so it is ok to
695 * set the fd back to blocked.
697 qemu_file_set_blocking(mis
->from_src_file
, true);
699 bioc
= qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE
);
700 fb
= qemu_fopen_channel_input(QIO_CHANNEL(bioc
));
701 object_unref(OBJECT(bioc
));
703 qemu_mutex_lock_iothread();
704 #ifdef CONFIG_REPLICATION
705 replication_start_all(REPLICATION_MODE_SECONDARY
, &local_err
);
707 qemu_mutex_unlock_iothread();
714 trace_colo_vm_state_change("stop", "run");
715 qemu_mutex_unlock_iothread();
717 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_CHECKPOINT_READY
,
723 while (mis
->state
== MIGRATION_STATUS_COLO
) {
726 colo_wait_handle_message(mis
->from_src_file
, &request
, &local_err
);
731 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
732 error_report("failover request");
736 qemu_mutex_lock_iothread();
737 vm_stop_force_state(RUN_STATE_COLO
);
738 trace_colo_vm_state_change("run", "stop");
739 qemu_mutex_unlock_iothread();
741 /* FIXME: This is unnecessary for periodic checkpoint mode */
742 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_CHECKPOINT_REPLY
,
748 colo_receive_check_message(mis
->from_src_file
,
749 COLO_MESSAGE_VMSTATE_SEND
, &local_err
);
754 qemu_mutex_lock_iothread();
755 cpu_synchronize_all_pre_loadvm();
756 ret
= qemu_loadvm_state_main(mis
->from_src_file
, mis
);
757 qemu_mutex_unlock_iothread();
760 error_report("Load VM's live state (ram) error");
764 value
= colo_receive_message_value(mis
->from_src_file
,
765 COLO_MESSAGE_VMSTATE_SIZE
, &local_err
);
771 * Read VM device state data into channel buffer,
772 * It's better to re-use the memory allocated.
773 * Here we need to handle the channel buffer directly.
775 if (value
> bioc
->capacity
) {
776 bioc
->capacity
= value
;
777 bioc
->data
= g_realloc(bioc
->data
, bioc
->capacity
);
779 total_size
= qemu_get_buffer(mis
->from_src_file
, bioc
->data
, value
);
780 if (total_size
!= value
) {
781 error_report("Got %" PRIu64
" VMState data, less than expected"
782 " %" PRIu64
, total_size
, value
);
785 bioc
->usage
= total_size
;
786 qio_channel_io_seek(QIO_CHANNEL(bioc
), 0, 0, NULL
);
788 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_VMSTATE_RECEIVED
,
794 qemu_mutex_lock_iothread();
795 vmstate_loading
= true;
796 ret
= qemu_load_device_state(fb
);
798 error_report("COLO: load device state failed");
799 qemu_mutex_unlock_iothread();
803 #ifdef CONFIG_REPLICATION
804 replication_get_error_all(&local_err
);
806 qemu_mutex_unlock_iothread();
810 /* discard colo disk buffer */
811 replication_do_checkpoint_all(&local_err
);
813 qemu_mutex_unlock_iothread();
819 /* Notify all filters of all NIC to do checkpoint */
820 colo_notify_filters_event(COLO_EVENT_CHECKPOINT
, &local_err
);
823 qemu_mutex_unlock_iothread();
827 vmstate_loading
= false;
829 trace_colo_vm_state_change("stop", "run");
830 qemu_mutex_unlock_iothread();
832 if (failover_get_state() == FAILOVER_STATUS_RELAUNCH
) {
833 failover_set_state(FAILOVER_STATUS_RELAUNCH
,
834 FAILOVER_STATUS_NONE
);
835 failover_request_active(NULL
);
839 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_VMSTATE_LOADED
,
847 vmstate_loading
= false;
848 /* Throw the unreported error message after exited from loop */
850 error_report_err(local_err
);
853 switch (failover_get_state()) {
854 case FAILOVER_STATUS_NONE
:
855 qapi_event_send_colo_exit(COLO_MODE_SECONDARY
,
856 COLO_EXIT_REASON_ERROR
);
858 case FAILOVER_STATUS_COMPLETED
:
859 qapi_event_send_colo_exit(COLO_MODE_SECONDARY
,
860 COLO_EXIT_REASON_REQUEST
);
870 /* Hope this not to be too long to loop here */
871 qemu_sem_wait(&mis
->colo_incoming_sem
);
872 qemu_sem_destroy(&mis
->colo_incoming_sem
);
873 /* Must be called after failover BH is completed */
874 if (mis
->to_src_file
) {
875 qemu_fclose(mis
->to_src_file
);
876 mis
->to_src_file
= NULL
;
879 rcu_unregister_thread();