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 "migration/colo.h"
17 #include "io/channel-buffer.h"
19 #include "qemu/error-report.h"
20 #include "qapi/error.h"
21 #include "migration/failover.h"
22 #include "replication.h"
23 #include "qmp-commands.h"
25 static bool vmstate_loading
;
27 #define COLO_BUFFER_BASE_SIZE (4 * 1024 * 1024)
29 bool colo_supported(void)
34 bool migration_in_colo_state(void)
36 MigrationState
*s
= migrate_get_current();
38 return (s
->state
== MIGRATION_STATUS_COLO
);
41 bool migration_incoming_in_colo_state(void)
43 MigrationIncomingState
*mis
= migration_incoming_get_current();
45 return mis
&& (mis
->state
== MIGRATION_STATUS_COLO
);
48 static bool colo_runstate_is_stopped(void)
50 return runstate_check(RUN_STATE_COLO
) || !runstate_is_running();
53 static void secondary_vm_do_failover(void)
56 MigrationIncomingState
*mis
= migration_incoming_get_current();
58 /* Can not do failover during the process of VM's loading VMstate, Or
59 * it will break the secondary VM.
61 if (vmstate_loading
) {
62 old_state
= failover_set_state(FAILOVER_STATUS_ACTIVE
,
63 FAILOVER_STATUS_RELAUNCH
);
64 if (old_state
!= FAILOVER_STATUS_ACTIVE
) {
65 error_report("Unknown error while do failover for secondary VM,"
66 "old_state: %s", FailoverStatus_lookup
[old_state
]);
71 migrate_set_state(&mis
->state
, MIGRATION_STATUS_COLO
,
72 MIGRATION_STATUS_COMPLETED
);
75 error_report("\"-S\" qemu option will be ignored in secondary side");
76 /* recover runstate to normal migration finish state */
80 * Make sure COLO incoming thread not block in recv or send,
81 * If mis->from_src_file and mis->to_src_file use the same fd,
82 * The second shutdown() will return -1, we ignore this value,
85 if (mis
->from_src_file
) {
86 qemu_file_shutdown(mis
->from_src_file
);
88 if (mis
->to_src_file
) {
89 qemu_file_shutdown(mis
->to_src_file
);
92 old_state
= failover_set_state(FAILOVER_STATUS_ACTIVE
,
93 FAILOVER_STATUS_COMPLETED
);
94 if (old_state
!= FAILOVER_STATUS_ACTIVE
) {
95 error_report("Incorrect state (%s) while doing failover for "
96 "secondary VM", FailoverStatus_lookup
[old_state
]);
99 /* Notify COLO incoming thread that failover work is finished */
100 qemu_sem_post(&mis
->colo_incoming_sem
);
101 /* For Secondary VM, jump to incoming co */
102 if (mis
->migration_incoming_co
) {
103 qemu_coroutine_enter(mis
->migration_incoming_co
);
107 static void primary_vm_do_failover(void)
109 MigrationState
*s
= migrate_get_current();
112 migrate_set_state(&s
->state
, MIGRATION_STATUS_COLO
,
113 MIGRATION_STATUS_COMPLETED
);
116 * Wake up COLO thread which may blocked in recv() or send(),
117 * The s->rp_state.from_dst_file and s->to_dst_file may use the
118 * same fd, but we still shutdown the fd for twice, it is harmless.
120 if (s
->to_dst_file
) {
121 qemu_file_shutdown(s
->to_dst_file
);
123 if (s
->rp_state
.from_dst_file
) {
124 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
127 old_state
= failover_set_state(FAILOVER_STATUS_ACTIVE
,
128 FAILOVER_STATUS_COMPLETED
);
129 if (old_state
!= FAILOVER_STATUS_ACTIVE
) {
130 error_report("Incorrect state (%s) while doing failover for Primary VM",
131 FailoverStatus_lookup
[old_state
]);
134 /* Notify COLO thread that failover work is finished */
135 qemu_sem_post(&s
->colo_exit_sem
);
138 void colo_do_failover(MigrationState
*s
)
140 /* Make sure VM stopped while failover happened. */
141 if (!colo_runstate_is_stopped()) {
142 vm_stop_force_state(RUN_STATE_COLO
);
145 if (get_colo_mode() == COLO_MODE_PRIMARY
) {
146 primary_vm_do_failover();
148 secondary_vm_do_failover();
152 void qmp_xen_set_replication(bool enable
, bool primary
,
153 bool has_failover
, bool failover
,
156 ReplicationMode mode
= primary
?
157 REPLICATION_MODE_PRIMARY
:
158 REPLICATION_MODE_SECONDARY
;
160 if (has_failover
&& enable
) {
161 error_setg(errp
, "Parameter 'failover' is only for"
162 " stopping replication");
167 replication_start_all(mode
, errp
);
172 replication_stop_all(failover
, failover
? NULL
: errp
);
176 static void colo_send_message(QEMUFile
*f
, COLOMessage msg
,
181 if (msg
>= COLO_MESSAGE__MAX
) {
182 error_setg(errp
, "%s: Invalid message", __func__
);
185 qemu_put_be32(f
, msg
);
188 ret
= qemu_file_get_error(f
);
190 error_setg_errno(errp
, -ret
, "Can't send COLO message");
192 trace_colo_send_message(COLOMessage_lookup
[msg
]);
195 static void colo_send_message_value(QEMUFile
*f
, COLOMessage msg
,
196 uint64_t value
, Error
**errp
)
198 Error
*local_err
= NULL
;
201 colo_send_message(f
, msg
, &local_err
);
203 error_propagate(errp
, local_err
);
206 qemu_put_be64(f
, value
);
209 ret
= qemu_file_get_error(f
);
211 error_setg_errno(errp
, -ret
, "Failed to send value for message:%s",
212 COLOMessage_lookup
[msg
]);
216 static COLOMessage
colo_receive_message(QEMUFile
*f
, Error
**errp
)
221 msg
= qemu_get_be32(f
);
222 ret
= qemu_file_get_error(f
);
224 error_setg_errno(errp
, -ret
, "Can't receive COLO message");
227 if (msg
>= COLO_MESSAGE__MAX
) {
228 error_setg(errp
, "%s: Invalid message", __func__
);
231 trace_colo_receive_message(COLOMessage_lookup
[msg
]);
235 static void colo_receive_check_message(QEMUFile
*f
, COLOMessage expect_msg
,
239 Error
*local_err
= NULL
;
241 msg
= colo_receive_message(f
, &local_err
);
243 error_propagate(errp
, local_err
);
246 if (msg
!= expect_msg
) {
247 error_setg(errp
, "Unexpected COLO message %d, expected %d",
252 static uint64_t colo_receive_message_value(QEMUFile
*f
, uint32_t expect_msg
,
255 Error
*local_err
= NULL
;
259 colo_receive_check_message(f
, expect_msg
, &local_err
);
261 error_propagate(errp
, local_err
);
265 value
= qemu_get_be64(f
);
266 ret
= qemu_file_get_error(f
);
268 error_setg_errno(errp
, -ret
, "Failed to get value for COLO message: %s",
269 COLOMessage_lookup
[expect_msg
]);
274 static int colo_do_checkpoint_transaction(MigrationState
*s
,
275 QIOChannelBuffer
*bioc
,
278 Error
*local_err
= NULL
;
281 colo_send_message(s
->to_dst_file
, COLO_MESSAGE_CHECKPOINT_REQUEST
,
287 colo_receive_check_message(s
->rp_state
.from_dst_file
,
288 COLO_MESSAGE_CHECKPOINT_REPLY
, &local_err
);
292 /* Reset channel-buffer directly */
293 qio_channel_io_seek(QIO_CHANNEL(bioc
), 0, 0, NULL
);
296 qemu_mutex_lock_iothread();
297 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
298 qemu_mutex_unlock_iothread();
301 vm_stop_force_state(RUN_STATE_COLO
);
302 qemu_mutex_unlock_iothread();
303 trace_colo_vm_state_change("run", "stop");
305 * Failover request bh could be called after vm_stop_force_state(),
306 * So we need check failover_request_is_active() again.
308 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
312 /* Disable block migration */
314 s
->params
.shared
= 0;
315 qemu_savevm_state_header(fb
);
316 qemu_savevm_state_begin(fb
, &s
->params
);
317 qemu_mutex_lock_iothread();
318 qemu_savevm_state_complete_precopy(fb
, false);
319 qemu_mutex_unlock_iothread();
323 colo_send_message(s
->to_dst_file
, COLO_MESSAGE_VMSTATE_SEND
, &local_err
);
328 * We need the size of the VMstate data in Secondary side,
329 * With which we can decide how much data should be read.
331 colo_send_message_value(s
->to_dst_file
, COLO_MESSAGE_VMSTATE_SIZE
,
332 bioc
->usage
, &local_err
);
337 qemu_put_buffer(s
->to_dst_file
, bioc
->data
, bioc
->usage
);
338 qemu_fflush(s
->to_dst_file
);
339 ret
= qemu_file_get_error(s
->to_dst_file
);
344 colo_receive_check_message(s
->rp_state
.from_dst_file
,
345 COLO_MESSAGE_VMSTATE_RECEIVED
, &local_err
);
350 colo_receive_check_message(s
->rp_state
.from_dst_file
,
351 COLO_MESSAGE_VMSTATE_LOADED
, &local_err
);
358 qemu_mutex_lock_iothread();
360 qemu_mutex_unlock_iothread();
361 trace_colo_vm_state_change("stop", "run");
365 error_report_err(local_err
);
370 static void colo_process_checkpoint(MigrationState
*s
)
372 QIOChannelBuffer
*bioc
;
374 int64_t current_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
375 Error
*local_err
= NULL
;
378 failover_init_state();
380 s
->rp_state
.from_dst_file
= qemu_file_get_return_path(s
->to_dst_file
);
381 if (!s
->rp_state
.from_dst_file
) {
382 error_report("Open QEMUFile from_dst_file failed");
387 * Wait for Secondary finish loading VM states and enter COLO
390 colo_receive_check_message(s
->rp_state
.from_dst_file
,
391 COLO_MESSAGE_CHECKPOINT_READY
, &local_err
);
395 bioc
= qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE
);
396 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
397 object_unref(OBJECT(bioc
));
399 qemu_mutex_lock_iothread();
401 qemu_mutex_unlock_iothread();
402 trace_colo_vm_state_change("stop", "run");
404 timer_mod(s
->colo_delay_timer
,
405 current_time
+ s
->parameters
.x_checkpoint_delay
);
407 while (s
->state
== MIGRATION_STATUS_COLO
) {
408 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
409 error_report("failover request");
413 qemu_sem_wait(&s
->colo_checkpoint_sem
);
415 ret
= colo_do_checkpoint_transaction(s
, bioc
, fb
);
422 /* Throw the unreported error message after exited from loop */
424 error_report_err(local_err
);
431 timer_del(s
->colo_delay_timer
);
433 /* Hope this not to be too long to wait here */
434 qemu_sem_wait(&s
->colo_exit_sem
);
435 qemu_sem_destroy(&s
->colo_exit_sem
);
437 * Must be called after failover BH is completed,
438 * Or the failover BH may shutdown the wrong fd that
439 * re-used by other threads after we release here.
441 if (s
->rp_state
.from_dst_file
) {
442 qemu_fclose(s
->rp_state
.from_dst_file
);
446 void colo_checkpoint_notify(void *opaque
)
448 MigrationState
*s
= opaque
;
449 int64_t next_notify_time
;
451 qemu_sem_post(&s
->colo_checkpoint_sem
);
452 s
->colo_checkpoint_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
453 next_notify_time
= s
->colo_checkpoint_time
+
454 s
->parameters
.x_checkpoint_delay
;
455 timer_mod(s
->colo_delay_timer
, next_notify_time
);
458 void migrate_start_colo_process(MigrationState
*s
)
460 qemu_mutex_unlock_iothread();
461 qemu_sem_init(&s
->colo_checkpoint_sem
, 0);
462 s
->colo_delay_timer
= timer_new_ms(QEMU_CLOCK_HOST
,
463 colo_checkpoint_notify
, s
);
465 qemu_sem_init(&s
->colo_exit_sem
, 0);
466 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
467 MIGRATION_STATUS_COLO
);
468 colo_process_checkpoint(s
);
469 qemu_mutex_lock_iothread();
472 static void colo_wait_handle_message(QEMUFile
*f
, int *checkpoint_request
,
476 Error
*local_err
= NULL
;
478 msg
= colo_receive_message(f
, &local_err
);
480 error_propagate(errp
, local_err
);
485 case COLO_MESSAGE_CHECKPOINT_REQUEST
:
486 *checkpoint_request
= 1;
489 *checkpoint_request
= 0;
490 error_setg(errp
, "Got unknown COLO message: %d", msg
);
495 void *colo_process_incoming_thread(void *opaque
)
497 MigrationIncomingState
*mis
= opaque
;
499 QIOChannelBuffer
*bioc
= NULL
; /* Cache incoming device state */
502 Error
*local_err
= NULL
;
504 qemu_sem_init(&mis
->colo_incoming_sem
, 0);
506 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
507 MIGRATION_STATUS_COLO
);
509 failover_init_state();
511 mis
->to_src_file
= qemu_file_get_return_path(mis
->from_src_file
);
512 if (!mis
->to_src_file
) {
513 error_report("COLO incoming thread: Open QEMUFile to_src_file failed");
517 * Note: the communication between Primary side and Secondary side
518 * should be sequential, we set the fd to unblocked in migration incoming
519 * coroutine, and here we are in the COLO incoming thread, so it is ok to
520 * set the fd back to blocked.
522 qemu_file_set_blocking(mis
->from_src_file
, true);
524 bioc
= qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE
);
525 fb
= qemu_fopen_channel_input(QIO_CHANNEL(bioc
));
526 object_unref(OBJECT(bioc
));
528 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_CHECKPOINT_READY
,
534 while (mis
->state
== MIGRATION_STATUS_COLO
) {
537 colo_wait_handle_message(mis
->from_src_file
, &request
, &local_err
);
542 if (failover_get_state() != FAILOVER_STATUS_NONE
) {
543 error_report("failover request");
547 /* FIXME: This is unnecessary for periodic checkpoint mode */
548 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_CHECKPOINT_REPLY
,
554 colo_receive_check_message(mis
->from_src_file
,
555 COLO_MESSAGE_VMSTATE_SEND
, &local_err
);
560 value
= colo_receive_message_value(mis
->from_src_file
,
561 COLO_MESSAGE_VMSTATE_SIZE
, &local_err
);
567 * Read VM device state data into channel buffer,
568 * It's better to re-use the memory allocated.
569 * Here we need to handle the channel buffer directly.
571 if (value
> bioc
->capacity
) {
572 bioc
->capacity
= value
;
573 bioc
->data
= g_realloc(bioc
->data
, bioc
->capacity
);
575 total_size
= qemu_get_buffer(mis
->from_src_file
, bioc
->data
, value
);
576 if (total_size
!= value
) {
577 error_report("Got %" PRIu64
" VMState data, less than expected"
578 " %" PRIu64
, total_size
, value
);
581 bioc
->usage
= total_size
;
582 qio_channel_io_seek(QIO_CHANNEL(bioc
), 0, 0, NULL
);
584 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_VMSTATE_RECEIVED
,
590 qemu_mutex_lock_iothread();
591 qemu_system_reset(VMRESET_SILENT
);
592 vmstate_loading
= true;
593 if (qemu_loadvm_state(fb
) < 0) {
594 error_report("COLO: loadvm failed");
595 qemu_mutex_unlock_iothread();
599 vmstate_loading
= false;
600 qemu_mutex_unlock_iothread();
602 if (failover_get_state() == FAILOVER_STATUS_RELAUNCH
) {
603 failover_set_state(FAILOVER_STATUS_RELAUNCH
,
604 FAILOVER_STATUS_NONE
);
605 failover_request_active(NULL
);
609 colo_send_message(mis
->to_src_file
, COLO_MESSAGE_VMSTATE_LOADED
,
617 vmstate_loading
= false;
618 /* Throw the unreported error message after exited from loop */
620 error_report_err(local_err
);
627 /* Hope this not to be too long to loop here */
628 qemu_sem_wait(&mis
->colo_incoming_sem
);
629 qemu_sem_destroy(&mis
->colo_incoming_sem
);
630 /* Must be called after failover BH is completed */
631 if (mis
->to_src_file
) {
632 qemu_fclose(mis
->to_src_file
);
634 migration_incoming_exit_colo();