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 "qemu-file-channel.h"
16 #include "migration.h"
17 #include "qemu-file.h"
19 #include "migration/colo.h"
21 #include "io/channel-buffer.h"
23 #include "qemu/error-report.h"
24 #include "migration/failover.h"
25 #include "replication.h"
26 #include "qmp-commands.h"
28 static bool vmstate_loading
;
30 #define COLO_BUFFER_BASE_SIZE (4 * 1024 * 1024)
32 bool colo_supported(void)
37 bool migration_in_colo_state(void)
39 MigrationState
*s
= migrate_get_current();
41 return (s
->state
== MIGRATION_STATUS_COLO
);
44 bool migration_incoming_in_colo_state(void)
46 MigrationIncomingState
*mis
= migration_incoming_get_current();
48 return mis
&& (mis
->state
== MIGRATION_STATUS_COLO
);
51 static bool colo_runstate_is_stopped(void)
53 return runstate_check(RUN_STATE_COLO
) || !runstate_is_running();
56 static void secondary_vm_do_failover(void)
59 MigrationIncomingState
*mis
= migration_incoming_get_current();
61 /* Can not do failover during the process of VM's loading VMstate, Or
62 * it will break the secondary VM.
64 if (vmstate_loading
) {
65 old_state
= failover_set_state(FAILOVER_STATUS_ACTIVE
,
66 FAILOVER_STATUS_RELAUNCH
);
67 if (old_state
!= FAILOVER_STATUS_ACTIVE
) {
68 error_report("Unknown error while do failover for secondary VM,"
69 "old_state: %s", FailoverStatus_lookup
[old_state
]);
74 migrate_set_state(&mis
->state
, MIGRATION_STATUS_COLO
,
75 MIGRATION_STATUS_COMPLETED
);
78 error_report("\"-S\" qemu option will be ignored in secondary side");
79 /* recover runstate to normal migration finish state */
83 * Make sure COLO incoming thread not block in recv or send,
84 * If mis->from_src_file and mis->to_src_file use the same fd,
85 * The second shutdown() will return -1, we ignore this value,
88 if (mis
->from_src_file
) {
89 qemu_file_shutdown(mis
->from_src_file
);
91 if (mis
->to_src_file
) {
92 qemu_file_shutdown(mis
->to_src_file
);
95 old_state
= failover_set_state(FAILOVER_STATUS_ACTIVE
,
96 FAILOVER_STATUS_COMPLETED
);
97 if (old_state
!= FAILOVER_STATUS_ACTIVE
) {
98 error_report("Incorrect state (%s) while doing failover for "
99 "secondary VM", FailoverStatus_lookup
[old_state
]);
102 /* Notify COLO incoming thread that failover work is finished */
103 qemu_sem_post(&mis
->colo_incoming_sem
);
104 /* For Secondary VM, jump to incoming co */
105 if (mis
->migration_incoming_co
) {
106 qemu_coroutine_enter(mis
->migration_incoming_co
);
110 static void primary_vm_do_failover(void)
112 MigrationState
*s
= migrate_get_current();
115 migrate_set_state(&s
->state
, MIGRATION_STATUS_COLO
,
116 MIGRATION_STATUS_COMPLETED
);
119 * Wake up COLO thread which may blocked in recv() or send(),
120 * The s->rp_state.from_dst_file and s->to_dst_file may use the
121 * same fd, but we still shutdown the fd for twice, it is harmless.
123 if (s
->to_dst_file
) {
124 qemu_file_shutdown(s
->to_dst_file
);
126 if (s
->rp_state
.from_dst_file
) {
127 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
130 old_state
= failover_set_state(FAILOVER_STATUS_ACTIVE
,
131 FAILOVER_STATUS_COMPLETED
);
132 if (old_state
!= FAILOVER_STATUS_ACTIVE
) {
133 error_report("Incorrect state (%s) while doing failover for Primary VM",
134 FailoverStatus_lookup
[old_state
]);
137 /* Notify COLO thread that failover work is finished */
138 qemu_sem_post(&s
->colo_exit_sem
);
141 void colo_do_failover(MigrationState
*s
)
143 /* Make sure VM stopped while failover happened. */
144 if (!colo_runstate_is_stopped()) {
145 vm_stop_force_state(RUN_STATE_COLO
);
148 if (get_colo_mode() == COLO_MODE_PRIMARY
) {
149 primary_vm_do_failover();
151 secondary_vm_do_failover();
155 void qmp_xen_set_replication(bool enable
, bool primary
,
156 bool has_failover
, bool failover
,
159 #ifdef CONFIG_REPLICATION
160 ReplicationMode mode
= primary
?
161 REPLICATION_MODE_PRIMARY
:
162 REPLICATION_MODE_SECONDARY
;
164 if (has_failover
&& enable
) {
165 error_setg(errp
, "Parameter 'failover' is only for"
166 " stopping replication");
171 replication_start_all(mode
, errp
);
176 replication_stop_all(failover
, failover
? NULL
: errp
);
183 ReplicationStatus
*qmp_query_xen_replication_status(Error
**errp
)
185 #ifdef CONFIG_REPLICATION
187 ReplicationStatus
*s
= g_new0(ReplicationStatus
, 1);
189 replication_get_error_all(&err
);
193 s
->desc
= g_strdup(error_get_pretty(err
));
205 void qmp_xen_colo_do_checkpoint(Error
**errp
)
207 #ifdef CONFIG_REPLICATION
208 replication_do_checkpoint_all(errp
);
214 static void colo_send_message(QEMUFile
*f
, COLOMessage msg
,
219 if (msg
>= COLO_MESSAGE__MAX
) {
220 error_setg(errp
, "%s: Invalid message", __func__
);
223 qemu_put_be32(f
, msg
);
226 ret
= qemu_file_get_error(f
);
228 error_setg_errno(errp
, -ret
, "Can't send COLO message");
230 trace_colo_send_message(COLOMessage_lookup
[msg
]);
233 static void colo_send_message_value(QEMUFile
*f
, COLOMessage msg
,
234 uint64_t value
, Error
**errp
)
236 Error
*local_err
= NULL
;
239 colo_send_message(f
, msg
, &local_err
);
241 error_propagate(errp
, local_err
);
244 qemu_put_be64(f
, value
);
247 ret
= qemu_file_get_error(f
);
249 error_setg_errno(errp
, -ret
, "Failed to send value for message:%s",
250 COLOMessage_lookup
[msg
]);
254 static COLOMessage
colo_receive_message(QEMUFile
*f
, Error
**errp
)
259 msg
= qemu_get_be32(f
);
260 ret
= qemu_file_get_error(f
);
262 error_setg_errno(errp
, -ret
, "Can't receive COLO message");
265 if (msg
>= COLO_MESSAGE__MAX
) {
266 error_setg(errp
, "%s: Invalid message", __func__
);
269 trace_colo_receive_message(COLOMessage_lookup
[msg
]);
273 static void colo_receive_check_message(QEMUFile
*f
, COLOMessage expect_msg
,
277 Error
*local_err
= NULL
;
279 msg
= colo_receive_message(f
, &local_err
);
281 error_propagate(errp
, local_err
);
284 if (msg
!= expect_msg
) {
285 error_setg(errp
, "Unexpected COLO message %d, expected %d",
290 static uint64_t colo_receive_message_value(QEMUFile
*f
, uint32_t expect_msg
,
293 Error
*local_err
= NULL
;
297 colo_receive_check_message(f
, expect_msg
, &local_err
);
299 error_propagate(errp
, local_err
);
303 value
= qemu_get_be64(f
);
304 ret
= qemu_file_get_error(f
);
306 error_setg_errno(errp
, -ret
, "Failed to get value for COLO message: %s",
307 COLOMessage_lookup
[expect_msg
]);
312 static int colo_do_checkpoint_transaction(MigrationState
*s
,
313 QIOChannelBuffer
*bioc
,
316 Error
*local_err
= NULL
;
319 colo_send_message(s
->to_dst_file
, COLO_MESSAGE_CHECKPOINT_REQUEST
,
325 colo_receive_check_message(s
->rp_state
.from_dst_file
,
326 COLO_MESSAGE_CHECKPOINT_REPLY
, &local_err
);
330 /* Reset channel-buffer directly */
331 qio_channel_io_seek(QIO_CHANNEL(bioc
), 0, 0, NULL
);
334 qemu_mutex_lock_iothread();
335 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
336 qemu_mutex_unlock_iothread();
339 vm_stop_force_state(RUN_STATE_COLO
);
340 qemu_mutex_unlock_iothread();
341 trace_colo_vm_state_change("run", "stop");
343 * Failover request bh could be called after vm_stop_force_state(),
344 * So we need check failover_request_is_active() again.
346 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
350 /* Disable block migration */
351 migrate_set_block_enabled(false, &local_err
);
352 qemu_savevm_state_header(fb
);
353 qemu_savevm_state_begin(fb
);
354 qemu_mutex_lock_iothread();
355 qemu_savevm_state_complete_precopy(fb
, false, false);
356 qemu_mutex_unlock_iothread();
360 colo_send_message(s
->to_dst_file
, COLO_MESSAGE_VMSTATE_SEND
, &local_err
);
365 * We need the size of the VMstate data in Secondary side,
366 * With which we can decide how much data should be read.
368 colo_send_message_value(s
->to_dst_file
, COLO_MESSAGE_VMSTATE_SIZE
,
369 bioc
->usage
, &local_err
);
374 qemu_put_buffer(s
->to_dst_file
, bioc
->data
, bioc
->usage
);
375 qemu_fflush(s
->to_dst_file
);
376 ret
= qemu_file_get_error(s
->to_dst_file
);
381 colo_receive_check_message(s
->rp_state
.from_dst_file
,
382 COLO_MESSAGE_VMSTATE_RECEIVED
, &local_err
);
387 colo_receive_check_message(s
->rp_state
.from_dst_file
,
388 COLO_MESSAGE_VMSTATE_LOADED
, &local_err
);
395 qemu_mutex_lock_iothread();
397 qemu_mutex_unlock_iothread();
398 trace_colo_vm_state_change("stop", "run");
402 error_report_err(local_err
);
407 static void colo_process_checkpoint(MigrationState
*s
)
409 QIOChannelBuffer
*bioc
;
411 int64_t current_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
412 Error
*local_err
= NULL
;
415 failover_init_state();
417 s
->rp_state
.from_dst_file
= qemu_file_get_return_path(s
->to_dst_file
);
418 if (!s
->rp_state
.from_dst_file
) {
419 error_report("Open QEMUFile from_dst_file failed");
424 * Wait for Secondary finish loading VM states and enter COLO
427 colo_receive_check_message(s
->rp_state
.from_dst_file
,
428 COLO_MESSAGE_CHECKPOINT_READY
, &local_err
);
432 bioc
= qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE
);
433 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
434 object_unref(OBJECT(bioc
));
436 qemu_mutex_lock_iothread();
438 qemu_mutex_unlock_iothread();
439 trace_colo_vm_state_change("stop", "run");
441 timer_mod(s
->colo_delay_timer
,
442 current_time
+ s
->parameters
.x_checkpoint_delay
);
444 while (s
->state
== MIGRATION_STATUS_COLO
) {
445 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
446 error_report("failover request");
450 qemu_sem_wait(&s
->colo_checkpoint_sem
);
452 ret
= colo_do_checkpoint_transaction(s
, bioc
, fb
);
459 /* Throw the unreported error message after exited from loop */
461 error_report_err(local_err
);
468 timer_del(s
->colo_delay_timer
);
470 /* Hope this not to be too long to wait here */
471 qemu_sem_wait(&s
->colo_exit_sem
);
472 qemu_sem_destroy(&s
->colo_exit_sem
);
474 * Must be called after failover BH is completed,
475 * Or the failover BH may shutdown the wrong fd that
476 * re-used by other threads after we release here.
478 if (s
->rp_state
.from_dst_file
) {
479 qemu_fclose(s
->rp_state
.from_dst_file
);
483 void colo_checkpoint_notify(void *opaque
)
485 MigrationState
*s
= opaque
;
486 int64_t next_notify_time
;
488 qemu_sem_post(&s
->colo_checkpoint_sem
);
489 s
->colo_checkpoint_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
490 next_notify_time
= s
->colo_checkpoint_time
+
491 s
->parameters
.x_checkpoint_delay
;
492 timer_mod(s
->colo_delay_timer
, next_notify_time
);
495 void migrate_start_colo_process(MigrationState
*s
)
497 qemu_mutex_unlock_iothread();
498 qemu_sem_init(&s
->colo_checkpoint_sem
, 0);
499 s
->colo_delay_timer
= timer_new_ms(QEMU_CLOCK_HOST
,
500 colo_checkpoint_notify
, s
);
502 qemu_sem_init(&s
->colo_exit_sem
, 0);
503 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
504 MIGRATION_STATUS_COLO
);
505 colo_process_checkpoint(s
);
506 qemu_mutex_lock_iothread();
509 static void colo_wait_handle_message(QEMUFile
*f
, int *checkpoint_request
,
513 Error
*local_err
= NULL
;
515 msg
= colo_receive_message(f
, &local_err
);
517 error_propagate(errp
, local_err
);
522 case COLO_MESSAGE_CHECKPOINT_REQUEST
:
523 *checkpoint_request
= 1;
526 *checkpoint_request
= 0;
527 error_setg(errp
, "Got unknown COLO message: %d", msg
);
532 void *colo_process_incoming_thread(void *opaque
)
534 MigrationIncomingState
*mis
= opaque
;
536 QIOChannelBuffer
*bioc
= NULL
; /* Cache incoming device state */
539 Error
*local_err
= NULL
;
541 qemu_sem_init(&mis
->colo_incoming_sem
, 0);
543 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
544 MIGRATION_STATUS_COLO
);
546 failover_init_state();
548 mis
->to_src_file
= qemu_file_get_return_path(mis
->from_src_file
);
549 if (!mis
->to_src_file
) {
550 error_report("COLO incoming thread: Open QEMUFile to_src_file failed");
554 * Note: the communication between Primary side and Secondary side
555 * should be sequential, we set the fd to unblocked in migration incoming
556 * coroutine, and here we are in the COLO incoming thread, so it is ok to
557 * set the fd back to blocked.
559 qemu_file_set_blocking(mis
->from_src_file
, true);
561 bioc
= qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE
);
562 fb
= qemu_fopen_channel_input(QIO_CHANNEL(bioc
));
563 object_unref(OBJECT(bioc
));
565 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_CHECKPOINT_READY
,
571 while (mis
->state
== MIGRATION_STATUS_COLO
) {
574 colo_wait_handle_message(mis
->from_src_file
, &request
, &local_err
);
579 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
580 error_report("failover request");
584 /* FIXME: This is unnecessary for periodic checkpoint mode */
585 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_CHECKPOINT_REPLY
,
591 colo_receive_check_message(mis
->from_src_file
,
592 COLO_MESSAGE_VMSTATE_SEND
, &local_err
);
597 value
= colo_receive_message_value(mis
->from_src_file
,
598 COLO_MESSAGE_VMSTATE_SIZE
, &local_err
);
604 * Read VM device state data into channel buffer,
605 * It's better to re-use the memory allocated.
606 * Here we need to handle the channel buffer directly.
608 if (value
> bioc
->capacity
) {
609 bioc
->capacity
= value
;
610 bioc
->data
= g_realloc(bioc
->data
, bioc
->capacity
);
612 total_size
= qemu_get_buffer(mis
->from_src_file
, bioc
->data
, value
);
613 if (total_size
!= value
) {
614 error_report("Got %" PRIu64
" VMState data, less than expected"
615 " %" PRIu64
, total_size
, value
);
618 bioc
->usage
= total_size
;
619 qio_channel_io_seek(QIO_CHANNEL(bioc
), 0, 0, NULL
);
621 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_VMSTATE_RECEIVED
,
627 qemu_mutex_lock_iothread();
628 qemu_system_reset(SHUTDOWN_CAUSE_NONE
);
629 vmstate_loading
= true;
630 if (qemu_loadvm_state(fb
) < 0) {
631 error_report("COLO: loadvm failed");
632 qemu_mutex_unlock_iothread();
636 vmstate_loading
= false;
637 qemu_mutex_unlock_iothread();
639 if (failover_get_state() == FAILOVER_STATUS_RELAUNCH
) {
640 failover_set_state(FAILOVER_STATUS_RELAUNCH
,
641 FAILOVER_STATUS_NONE
);
642 failover_request_active(NULL
);
646 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_VMSTATE_LOADED
,
654 vmstate_loading
= false;
655 /* Throw the unreported error message after exited from loop */
657 error_report_err(local_err
);
664 /* Hope this not to be too long to loop here */
665 qemu_sem_wait(&mis
->colo_incoming_sem
);
666 qemu_sem_destroy(&mis
->colo_incoming_sem
);
667 /* Must be called after failover BH is completed */
668 if (mis
->to_src_file
) {
669 qemu_fclose(mis
->to_src_file
);
671 migration_incoming_exit_colo();