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 "qemu/timer.h"
15 #include "sysemu/sysemu.h"
16 #include "qemu-file-channel.h"
17 #include "migration/migration.h"
18 #include "qemu-file.h"
20 #include "migration/colo.h"
22 #include "io/channel-buffer.h"
24 #include "qemu/error-report.h"
25 #include "qapi/error.h"
26 #include "migration/failover.h"
27 #include "replication.h"
28 #include "qmp-commands.h"
30 static bool vmstate_loading
;
32 #define COLO_BUFFER_BASE_SIZE (4 * 1024 * 1024)
34 bool colo_supported(void)
39 bool migration_in_colo_state(void)
41 MigrationState
*s
= migrate_get_current();
43 return (s
->state
== MIGRATION_STATUS_COLO
);
46 bool migration_incoming_in_colo_state(void)
48 MigrationIncomingState
*mis
= migration_incoming_get_current();
50 return mis
&& (mis
->state
== MIGRATION_STATUS_COLO
);
53 static bool colo_runstate_is_stopped(void)
55 return runstate_check(RUN_STATE_COLO
) || !runstate_is_running();
58 static void secondary_vm_do_failover(void)
61 MigrationIncomingState
*mis
= migration_incoming_get_current();
63 /* Can not do failover during the process of VM's loading VMstate, Or
64 * it will break the secondary VM.
66 if (vmstate_loading
) {
67 old_state
= failover_set_state(FAILOVER_STATUS_ACTIVE
,
68 FAILOVER_STATUS_RELAUNCH
);
69 if (old_state
!= FAILOVER_STATUS_ACTIVE
) {
70 error_report("Unknown error while do failover for secondary VM,"
71 "old_state: %s", FailoverStatus_lookup
[old_state
]);
76 migrate_set_state(&mis
->state
, MIGRATION_STATUS_COLO
,
77 MIGRATION_STATUS_COMPLETED
);
80 error_report("\"-S\" qemu option will be ignored in secondary side");
81 /* recover runstate to normal migration finish state */
85 * Make sure COLO incoming thread not block in recv or send,
86 * If mis->from_src_file and mis->to_src_file use the same fd,
87 * The second shutdown() will return -1, we ignore this value,
90 if (mis
->from_src_file
) {
91 qemu_file_shutdown(mis
->from_src_file
);
93 if (mis
->to_src_file
) {
94 qemu_file_shutdown(mis
->to_src_file
);
97 old_state
= failover_set_state(FAILOVER_STATUS_ACTIVE
,
98 FAILOVER_STATUS_COMPLETED
);
99 if (old_state
!= FAILOVER_STATUS_ACTIVE
) {
100 error_report("Incorrect state (%s) while doing failover for "
101 "secondary VM", FailoverStatus_lookup
[old_state
]);
104 /* Notify COLO incoming thread that failover work is finished */
105 qemu_sem_post(&mis
->colo_incoming_sem
);
106 /* For Secondary VM, jump to incoming co */
107 if (mis
->migration_incoming_co
) {
108 qemu_coroutine_enter(mis
->migration_incoming_co
);
112 static void primary_vm_do_failover(void)
114 MigrationState
*s
= migrate_get_current();
117 migrate_set_state(&s
->state
, MIGRATION_STATUS_COLO
,
118 MIGRATION_STATUS_COMPLETED
);
121 * Wake up COLO thread which may blocked in recv() or send(),
122 * The s->rp_state.from_dst_file and s->to_dst_file may use the
123 * same fd, but we still shutdown the fd for twice, it is harmless.
125 if (s
->to_dst_file
) {
126 qemu_file_shutdown(s
->to_dst_file
);
128 if (s
->rp_state
.from_dst_file
) {
129 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
132 old_state
= failover_set_state(FAILOVER_STATUS_ACTIVE
,
133 FAILOVER_STATUS_COMPLETED
);
134 if (old_state
!= FAILOVER_STATUS_ACTIVE
) {
135 error_report("Incorrect state (%s) while doing failover for Primary VM",
136 FailoverStatus_lookup
[old_state
]);
139 /* Notify COLO thread that failover work is finished */
140 qemu_sem_post(&s
->colo_exit_sem
);
143 void colo_do_failover(MigrationState
*s
)
145 /* Make sure VM stopped while failover happened. */
146 if (!colo_runstate_is_stopped()) {
147 vm_stop_force_state(RUN_STATE_COLO
);
150 if (get_colo_mode() == COLO_MODE_PRIMARY
) {
151 primary_vm_do_failover();
153 secondary_vm_do_failover();
157 void qmp_xen_set_replication(bool enable
, bool primary
,
158 bool has_failover
, bool failover
,
161 #ifdef CONFIG_REPLICATION
162 ReplicationMode mode
= primary
?
163 REPLICATION_MODE_PRIMARY
:
164 REPLICATION_MODE_SECONDARY
;
166 if (has_failover
&& enable
) {
167 error_setg(errp
, "Parameter 'failover' is only for"
168 " stopping replication");
173 replication_start_all(mode
, errp
);
178 replication_stop_all(failover
, failover
? NULL
: errp
);
185 ReplicationStatus
*qmp_query_xen_replication_status(Error
**errp
)
187 #ifdef CONFIG_REPLICATION
189 ReplicationStatus
*s
= g_new0(ReplicationStatus
, 1);
191 replication_get_error_all(&err
);
195 s
->desc
= g_strdup(error_get_pretty(err
));
207 void qmp_xen_colo_do_checkpoint(Error
**errp
)
209 #ifdef CONFIG_REPLICATION
210 replication_do_checkpoint_all(errp
);
216 static void colo_send_message(QEMUFile
*f
, COLOMessage msg
,
221 if (msg
>= COLO_MESSAGE__MAX
) {
222 error_setg(errp
, "%s: Invalid message", __func__
);
225 qemu_put_be32(f
, msg
);
228 ret
= qemu_file_get_error(f
);
230 error_setg_errno(errp
, -ret
, "Can't send COLO message");
232 trace_colo_send_message(COLOMessage_lookup
[msg
]);
235 static void colo_send_message_value(QEMUFile
*f
, COLOMessage msg
,
236 uint64_t value
, Error
**errp
)
238 Error
*local_err
= NULL
;
241 colo_send_message(f
, msg
, &local_err
);
243 error_propagate(errp
, local_err
);
246 qemu_put_be64(f
, value
);
249 ret
= qemu_file_get_error(f
);
251 error_setg_errno(errp
, -ret
, "Failed to send value for message:%s",
252 COLOMessage_lookup
[msg
]);
256 static COLOMessage
colo_receive_message(QEMUFile
*f
, Error
**errp
)
261 msg
= qemu_get_be32(f
);
262 ret
= qemu_file_get_error(f
);
264 error_setg_errno(errp
, -ret
, "Can't receive COLO message");
267 if (msg
>= COLO_MESSAGE__MAX
) {
268 error_setg(errp
, "%s: Invalid message", __func__
);
271 trace_colo_receive_message(COLOMessage_lookup
[msg
]);
275 static void colo_receive_check_message(QEMUFile
*f
, COLOMessage expect_msg
,
279 Error
*local_err
= NULL
;
281 msg
= colo_receive_message(f
, &local_err
);
283 error_propagate(errp
, local_err
);
286 if (msg
!= expect_msg
) {
287 error_setg(errp
, "Unexpected COLO message %d, expected %d",
292 static uint64_t colo_receive_message_value(QEMUFile
*f
, uint32_t expect_msg
,
295 Error
*local_err
= NULL
;
299 colo_receive_check_message(f
, expect_msg
, &local_err
);
301 error_propagate(errp
, local_err
);
305 value
= qemu_get_be64(f
);
306 ret
= qemu_file_get_error(f
);
308 error_setg_errno(errp
, -ret
, "Failed to get value for COLO message: %s",
309 COLOMessage_lookup
[expect_msg
]);
314 static int colo_do_checkpoint_transaction(MigrationState
*s
,
315 QIOChannelBuffer
*bioc
,
318 Error
*local_err
= NULL
;
321 colo_send_message(s
->to_dst_file
, COLO_MESSAGE_CHECKPOINT_REQUEST
,
327 colo_receive_check_message(s
->rp_state
.from_dst_file
,
328 COLO_MESSAGE_CHECKPOINT_REPLY
, &local_err
);
332 /* Reset channel-buffer directly */
333 qio_channel_io_seek(QIO_CHANNEL(bioc
), 0, 0, NULL
);
336 qemu_mutex_lock_iothread();
337 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
338 qemu_mutex_unlock_iothread();
341 vm_stop_force_state(RUN_STATE_COLO
);
342 qemu_mutex_unlock_iothread();
343 trace_colo_vm_state_change("run", "stop");
345 * Failover request bh could be called after vm_stop_force_state(),
346 * So we need check failover_request_is_active() again.
348 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
352 /* Disable block migration */
353 migrate_set_block_enabled(false, &local_err
);
354 qemu_savevm_state_header(fb
);
355 qemu_savevm_state_begin(fb
);
356 qemu_mutex_lock_iothread();
357 qemu_savevm_state_complete_precopy(fb
, false);
358 qemu_mutex_unlock_iothread();
362 colo_send_message(s
->to_dst_file
, COLO_MESSAGE_VMSTATE_SEND
, &local_err
);
367 * We need the size of the VMstate data in Secondary side,
368 * With which we can decide how much data should be read.
370 colo_send_message_value(s
->to_dst_file
, COLO_MESSAGE_VMSTATE_SIZE
,
371 bioc
->usage
, &local_err
);
376 qemu_put_buffer(s
->to_dst_file
, bioc
->data
, bioc
->usage
);
377 qemu_fflush(s
->to_dst_file
);
378 ret
= qemu_file_get_error(s
->to_dst_file
);
383 colo_receive_check_message(s
->rp_state
.from_dst_file
,
384 COLO_MESSAGE_VMSTATE_RECEIVED
, &local_err
);
389 colo_receive_check_message(s
->rp_state
.from_dst_file
,
390 COLO_MESSAGE_VMSTATE_LOADED
, &local_err
);
397 qemu_mutex_lock_iothread();
399 qemu_mutex_unlock_iothread();
400 trace_colo_vm_state_change("stop", "run");
404 error_report_err(local_err
);
409 static void colo_process_checkpoint(MigrationState
*s
)
411 QIOChannelBuffer
*bioc
;
413 int64_t current_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
414 Error
*local_err
= NULL
;
417 failover_init_state();
419 s
->rp_state
.from_dst_file
= qemu_file_get_return_path(s
->to_dst_file
);
420 if (!s
->rp_state
.from_dst_file
) {
421 error_report("Open QEMUFile from_dst_file failed");
426 * Wait for Secondary finish loading VM states and enter COLO
429 colo_receive_check_message(s
->rp_state
.from_dst_file
,
430 COLO_MESSAGE_CHECKPOINT_READY
, &local_err
);
434 bioc
= qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE
);
435 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
436 object_unref(OBJECT(bioc
));
438 qemu_mutex_lock_iothread();
440 qemu_mutex_unlock_iothread();
441 trace_colo_vm_state_change("stop", "run");
443 timer_mod(s
->colo_delay_timer
,
444 current_time
+ s
->parameters
.x_checkpoint_delay
);
446 while (s
->state
== MIGRATION_STATUS_COLO
) {
447 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
448 error_report("failover request");
452 qemu_sem_wait(&s
->colo_checkpoint_sem
);
454 ret
= colo_do_checkpoint_transaction(s
, bioc
, fb
);
461 /* Throw the unreported error message after exited from loop */
463 error_report_err(local_err
);
470 timer_del(s
->colo_delay_timer
);
472 /* Hope this not to be too long to wait here */
473 qemu_sem_wait(&s
->colo_exit_sem
);
474 qemu_sem_destroy(&s
->colo_exit_sem
);
476 * Must be called after failover BH is completed,
477 * Or the failover BH may shutdown the wrong fd that
478 * re-used by other threads after we release here.
480 if (s
->rp_state
.from_dst_file
) {
481 qemu_fclose(s
->rp_state
.from_dst_file
);
485 void colo_checkpoint_notify(void *opaque
)
487 MigrationState
*s
= opaque
;
488 int64_t next_notify_time
;
490 qemu_sem_post(&s
->colo_checkpoint_sem
);
491 s
->colo_checkpoint_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
492 next_notify_time
= s
->colo_checkpoint_time
+
493 s
->parameters
.x_checkpoint_delay
;
494 timer_mod(s
->colo_delay_timer
, next_notify_time
);
497 void migrate_start_colo_process(MigrationState
*s
)
499 qemu_mutex_unlock_iothread();
500 qemu_sem_init(&s
->colo_checkpoint_sem
, 0);
501 s
->colo_delay_timer
= timer_new_ms(QEMU_CLOCK_HOST
,
502 colo_checkpoint_notify
, s
);
504 qemu_sem_init(&s
->colo_exit_sem
, 0);
505 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
506 MIGRATION_STATUS_COLO
);
507 colo_process_checkpoint(s
);
508 qemu_mutex_lock_iothread();
511 static void colo_wait_handle_message(QEMUFile
*f
, int *checkpoint_request
,
515 Error
*local_err
= NULL
;
517 msg
= colo_receive_message(f
, &local_err
);
519 error_propagate(errp
, local_err
);
524 case COLO_MESSAGE_CHECKPOINT_REQUEST
:
525 *checkpoint_request
= 1;
528 *checkpoint_request
= 0;
529 error_setg(errp
, "Got unknown COLO message: %d", msg
);
534 void *colo_process_incoming_thread(void *opaque
)
536 MigrationIncomingState
*mis
= opaque
;
538 QIOChannelBuffer
*bioc
= NULL
; /* Cache incoming device state */
541 Error
*local_err
= NULL
;
543 qemu_sem_init(&mis
->colo_incoming_sem
, 0);
545 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
546 MIGRATION_STATUS_COLO
);
548 failover_init_state();
550 mis
->to_src_file
= qemu_file_get_return_path(mis
->from_src_file
);
551 if (!mis
->to_src_file
) {
552 error_report("COLO incoming thread: Open QEMUFile to_src_file failed");
556 * Note: the communication between Primary side and Secondary side
557 * should be sequential, we set the fd to unblocked in migration incoming
558 * coroutine, and here we are in the COLO incoming thread, so it is ok to
559 * set the fd back to blocked.
561 qemu_file_set_blocking(mis
->from_src_file
, true);
563 bioc
= qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE
);
564 fb
= qemu_fopen_channel_input(QIO_CHANNEL(bioc
));
565 object_unref(OBJECT(bioc
));
567 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_CHECKPOINT_READY
,
573 while (mis
->state
== MIGRATION_STATUS_COLO
) {
576 colo_wait_handle_message(mis
->from_src_file
, &request
, &local_err
);
581 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
582 error_report("failover request");
586 /* FIXME: This is unnecessary for periodic checkpoint mode */
587 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_CHECKPOINT_REPLY
,
593 colo_receive_check_message(mis
->from_src_file
,
594 COLO_MESSAGE_VMSTATE_SEND
, &local_err
);
599 value
= colo_receive_message_value(mis
->from_src_file
,
600 COLO_MESSAGE_VMSTATE_SIZE
, &local_err
);
606 * Read VM device state data into channel buffer,
607 * It's better to re-use the memory allocated.
608 * Here we need to handle the channel buffer directly.
610 if (value
> bioc
->capacity
) {
611 bioc
->capacity
= value
;
612 bioc
->data
= g_realloc(bioc
->data
, bioc
->capacity
);
614 total_size
= qemu_get_buffer(mis
->from_src_file
, bioc
->data
, value
);
615 if (total_size
!= value
) {
616 error_report("Got %" PRIu64
" VMState data, less than expected"
617 " %" PRIu64
, total_size
, value
);
620 bioc
->usage
= total_size
;
621 qio_channel_io_seek(QIO_CHANNEL(bioc
), 0, 0, NULL
);
623 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_VMSTATE_RECEIVED
,
629 qemu_mutex_lock_iothread();
630 qemu_system_reset(SHUTDOWN_CAUSE_NONE
);
631 vmstate_loading
= true;
632 if (qemu_loadvm_state(fb
) < 0) {
633 error_report("COLO: loadvm failed");
634 qemu_mutex_unlock_iothread();
638 vmstate_loading
= false;
639 qemu_mutex_unlock_iothread();
641 if (failover_get_state() == FAILOVER_STATUS_RELAUNCH
) {
642 failover_set_state(FAILOVER_STATUS_RELAUNCH
,
643 FAILOVER_STATUS_NONE
);
644 failover_request_active(NULL
);
648 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_VMSTATE_LOADED
,
656 vmstate_loading
= false;
657 /* Throw the unreported error message after exited from loop */
659 error_report_err(local_err
);
666 /* Hope this not to be too long to loop here */
667 qemu_sem_wait(&mis
->colo_incoming_sem
);
668 qemu_sem_destroy(&mis
->colo_incoming_sem
);
669 /* Must be called after failover BH is completed */
670 if (mis
->to_src_file
) {
671 qemu_fclose(mis
->to_src_file
);
673 migration_incoming_exit_colo();