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 "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 "qemu/main-loop.h"
27 #include "migration/failover.h"
28 #include "migration/ram.h"
29 #ifdef CONFIG_REPLICATION
30 #include "block/replication.h"
32 #include "net/colo-compare.h"
34 #include "block/block.h"
35 #include "qapi/qapi-events-migration.h"
36 #include "sysemu/cpus.h"
37 #include "sysemu/runstate.h"
38 #include "net/filter.h"
41 static bool vmstate_loading;
42 static Notifier packets_compare_notifier;
44 /* User need to know colo mode after COLO failover */
45 static COLOMode last_colo_mode;
47 #define COLO_BUFFER_BASE_SIZE (4 * 1024 * 1024)
49 bool migration_in_colo_state(void)
51 MigrationState *s = migrate_get_current();
53 return (s->state == MIGRATION_STATUS_COLO);
56 bool migration_incoming_in_colo_state(void)
58 MigrationIncomingState *mis = migration_incoming_get_current();
60 return mis && (mis->state == MIGRATION_STATUS_COLO);
63 static bool colo_runstate_is_stopped(void)
65 return runstate_check(RUN_STATE_COLO) || !runstate_is_running();
68 static void secondary_vm_do_failover(void)
70 /* COLO needs enable block-replication */
71 #ifdef CONFIG_REPLICATION
73 MigrationIncomingState *mis = migration_incoming_get_current();
74 Error *local_err = NULL;
76 /* Can not do failover during the process of VM's loading VMstate, Or
77 * it will break the secondary VM.
79 if (vmstate_loading) {
80 old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
81 FAILOVER_STATUS_RELAUNCH);
82 if (old_state != FAILOVER_STATUS_ACTIVE) {
83 error_report("Unknown error while do failover for secondary VM,"
84 "old_state: %s", FailoverStatus_str(old_state));
89 migrate_set_state(&mis->state, MIGRATION_STATUS_COLO,
90 MIGRATION_STATUS_COMPLETED);
92 replication_stop_all(true, &local_err);
94 error_report_err(local_err);
98 /* Notify all filters of all NIC to do checkpoint */
99 colo_notify_filters_event(COLO_EVENT_FAILOVER, &local_err);
101 error_report_err(local_err);
105 error_report("\"-S\" qemu option will be ignored in secondary side");
106 /* recover runstate to normal migration finish state */
110 * Make sure COLO incoming thread not block in recv or send,
111 * If mis->from_src_file and mis->to_src_file use the same fd,
112 * The second shutdown() will return -1, we ignore this value,
115 if (mis->from_src_file) {
116 qemu_file_shutdown(mis->from_src_file);
118 if (mis->to_src_file) {
119 qemu_file_shutdown(mis->to_src_file);
122 old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
123 FAILOVER_STATUS_COMPLETED);
124 if (old_state != FAILOVER_STATUS_ACTIVE) {
125 error_report("Incorrect state (%s) while doing failover for "
126 "secondary VM", FailoverStatus_str(old_state));
129 /* Notify COLO incoming thread that failover work is finished */
130 qemu_sem_post(&mis->colo_incoming_sem);
132 /* For Secondary VM, jump to incoming co */
133 if (mis->migration_incoming_co) {
134 qemu_coroutine_enter(mis->migration_incoming_co);
141 static void primary_vm_do_failover(void)
143 #ifdef CONFIG_REPLICATION
144 MigrationState *s = migrate_get_current();
146 Error *local_err = NULL;
148 migrate_set_state(&s->state, MIGRATION_STATUS_COLO,
149 MIGRATION_STATUS_COMPLETED);
151 * kick COLO thread which might wait at
152 * qemu_sem_wait(&s->colo_checkpoint_sem).
154 colo_checkpoint_notify(s);
157 * Wake up COLO thread which may blocked in recv() or send(),
158 * The s->rp_state.from_dst_file and s->to_dst_file may use the
159 * same fd, but we still shutdown the fd for twice, it is harmless.
161 if (s->to_dst_file) {
162 qemu_file_shutdown(s->to_dst_file);
164 if (s->rp_state.from_dst_file) {
165 qemu_file_shutdown(s->rp_state.from_dst_file);
168 old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
169 FAILOVER_STATUS_COMPLETED);
170 if (old_state != FAILOVER_STATUS_ACTIVE) {
171 error_report("Incorrect state (%s) while doing failover for Primary VM",
172 FailoverStatus_str(old_state));
176 replication_stop_all(true, &local_err);
178 error_report_err(local_err);
182 /* Notify COLO thread that failover work is finished */
183 qemu_sem_post(&s->colo_exit_sem);
189 COLOMode get_colo_mode(void)
191 if (migration_in_colo_state()) {
192 return COLO_MODE_PRIMARY;
193 } else if (migration_incoming_in_colo_state()) {
194 return COLO_MODE_SECONDARY;
196 return COLO_MODE_NONE;
200 void colo_do_failover(void)
202 /* Make sure VM stopped while failover happened. */
203 if (!colo_runstate_is_stopped()) {
204 vm_stop_force_state(RUN_STATE_COLO);
207 switch (last_colo_mode = get_colo_mode()) {
208 case COLO_MODE_PRIMARY:
209 primary_vm_do_failover();
211 case COLO_MODE_SECONDARY:
212 secondary_vm_do_failover();
215 error_report("colo_do_failover failed because the colo mode"
216 " could not be obtained");
220 #ifdef CONFIG_REPLICATION
221 void qmp_xen_set_replication(bool enable, bool primary,
222 bool has_failover, bool failover,
225 ReplicationMode mode = primary ?
226 REPLICATION_MODE_PRIMARY :
227 REPLICATION_MODE_SECONDARY;
229 if (has_failover && enable) {
230 error_setg(errp, "Parameter 'failover' is only for"
231 " stopping replication");
236 replication_start_all(mode, errp);
241 replication_stop_all(failover, failover ? NULL : errp);
245 ReplicationStatus *qmp_query_xen_replication_status(Error **errp)
248 ReplicationStatus *s = g_new0(ReplicationStatus, 1);
250 replication_get_error_all(&err);
253 s->desc = g_strdup(error_get_pretty(err));
262 void qmp_xen_colo_do_checkpoint(Error **errp)
266 replication_do_checkpoint_all(&err);
268 error_propagate(errp, err);
271 /* Notify all filters of all NIC to do checkpoint */
272 colo_notify_filters_event(COLO_EVENT_CHECKPOINT, errp);
276 COLOStatus *qmp_query_colo_status(Error **errp)
278 COLOStatus *s = g_new0(COLOStatus, 1);
280 s->mode = get_colo_mode();
281 s->last_mode = last_colo_mode;
283 switch (failover_get_state()) {
284 case FAILOVER_STATUS_NONE:
285 s->reason = COLO_EXIT_REASON_NONE;
287 case FAILOVER_STATUS_COMPLETED:
288 s->reason = COLO_EXIT_REASON_REQUEST;
291 if (migration_in_colo_state()) {
292 s->reason = COLO_EXIT_REASON_PROCESSING;
294 s->reason = COLO_EXIT_REASON_ERROR;
301 static void colo_send_message(QEMUFile *f, COLOMessage msg,
306 if (msg >= COLO_MESSAGE__MAX) {
307 error_setg(errp, "%s: Invalid message", __func__);
310 qemu_put_be32(f, msg);
313 ret = qemu_file_get_error(f);
315 error_setg_errno(errp, -ret, "Can't send COLO message");
317 trace_colo_send_message(COLOMessage_str(msg));
320 static void colo_send_message_value(QEMUFile *f, COLOMessage msg,
321 uint64_t value, Error **errp)
323 Error *local_err = NULL;
326 colo_send_message(f, msg, &local_err);
328 error_propagate(errp, local_err);
331 qemu_put_be64(f, value);
334 ret = qemu_file_get_error(f);
336 error_setg_errno(errp, -ret, "Failed to send value for message:%s",
337 COLOMessage_str(msg));
341 static COLOMessage colo_receive_message(QEMUFile *f, Error **errp)
346 msg = qemu_get_be32(f);
347 ret = qemu_file_get_error(f);
349 error_setg_errno(errp, -ret, "Can't receive COLO message");
352 if (msg >= COLO_MESSAGE__MAX) {
353 error_setg(errp, "%s: Invalid message", __func__);
356 trace_colo_receive_message(COLOMessage_str(msg));
360 static void colo_receive_check_message(QEMUFile *f, COLOMessage expect_msg,
364 Error *local_err = NULL;
366 msg = colo_receive_message(f, &local_err);
368 error_propagate(errp, local_err);
371 if (msg != expect_msg) {
372 error_setg(errp, "Unexpected COLO message %d, expected %d",
377 static uint64_t colo_receive_message_value(QEMUFile *f, uint32_t expect_msg,
380 Error *local_err = NULL;
384 colo_receive_check_message(f, expect_msg, &local_err);
386 error_propagate(errp, local_err);
390 value = qemu_get_be64(f);
391 ret = qemu_file_get_error(f);
393 error_setg_errno(errp, -ret, "Failed to get value for COLO message: %s",
394 COLOMessage_str(expect_msg));
399 static int colo_do_checkpoint_transaction(MigrationState *s,
400 QIOChannelBuffer *bioc,
403 Error *local_err = NULL;
406 colo_send_message(s->to_dst_file, COLO_MESSAGE_CHECKPOINT_REQUEST,
412 colo_receive_check_message(s->rp_state.from_dst_file,
413 COLO_MESSAGE_CHECKPOINT_REPLY, &local_err);
417 /* Reset channel-buffer directly */
418 qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
421 qemu_mutex_lock_iothread();
422 if (failover_get_state() != FAILOVER_STATUS_NONE) {
423 qemu_mutex_unlock_iothread();
426 vm_stop_force_state(RUN_STATE_COLO);
427 qemu_mutex_unlock_iothread();
428 trace_colo_vm_state_change("run", "stop");
430 * Failover request bh could be called after vm_stop_force_state(),
431 * So we need check failover_request_is_active() again.
433 if (failover_get_state() != FAILOVER_STATUS_NONE) {
436 qemu_mutex_lock_iothread();
438 #ifdef CONFIG_REPLICATION
439 replication_do_checkpoint_all(&local_err);
441 qemu_mutex_unlock_iothread();
448 colo_send_message(s->to_dst_file, COLO_MESSAGE_VMSTATE_SEND, &local_err);
450 qemu_mutex_unlock_iothread();
453 /* Note: device state is saved into buffer */
454 ret = qemu_save_device_state(fb);
456 qemu_mutex_unlock_iothread();
461 if (migrate_auto_converge()) {
462 mig_throttle_counter_reset();
465 * Only save VM's live state, which not including device state.
466 * TODO: We may need a timeout mechanism to prevent COLO process
467 * to be blocked here.
469 qemu_savevm_live_state(s->to_dst_file);
474 * We need the size of the VMstate data in Secondary side,
475 * With which we can decide how much data should be read.
477 colo_send_message_value(s->to_dst_file, COLO_MESSAGE_VMSTATE_SIZE,
478 bioc->usage, &local_err);
483 qemu_put_buffer(s->to_dst_file, bioc->data, bioc->usage);
484 qemu_fflush(s->to_dst_file);
485 ret = qemu_file_get_error(s->to_dst_file);
490 colo_receive_check_message(s->rp_state.from_dst_file,
491 COLO_MESSAGE_VMSTATE_RECEIVED, &local_err);
496 qemu_event_reset(&s->colo_checkpoint_event);
497 colo_notify_compares_event(NULL, COLO_EVENT_CHECKPOINT, &local_err);
502 colo_receive_check_message(s->rp_state.from_dst_file,
503 COLO_MESSAGE_VMSTATE_LOADED, &local_err);
510 qemu_mutex_lock_iothread();
512 qemu_mutex_unlock_iothread();
513 trace_colo_vm_state_change("stop", "run");
517 error_report_err(local_err);
522 static void colo_compare_notify_checkpoint(Notifier *notifier, void *data)
524 colo_checkpoint_notify(data);
527 static void colo_process_checkpoint(MigrationState *s)
529 QIOChannelBuffer *bioc;
531 Error *local_err = NULL;
534 if (get_colo_mode() != COLO_MODE_PRIMARY) {
535 error_report("COLO mode must be COLO_MODE_PRIMARY");
539 failover_init_state();
541 s->rp_state.from_dst_file = qemu_file_get_return_path(s->to_dst_file);
542 if (!s->rp_state.from_dst_file) {
543 error_report("Open QEMUFile from_dst_file failed");
547 packets_compare_notifier.notify = colo_compare_notify_checkpoint;
548 colo_compare_register_notifier(&packets_compare_notifier);
551 * Wait for Secondary finish loading VM states and enter COLO
554 colo_receive_check_message(s->rp_state.from_dst_file,
555 COLO_MESSAGE_CHECKPOINT_READY, &local_err);
559 bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
560 fb = qemu_file_new_output(QIO_CHANNEL(bioc));
561 object_unref(OBJECT(bioc));
563 qemu_mutex_lock_iothread();
564 #ifdef CONFIG_REPLICATION
565 replication_start_all(REPLICATION_MODE_PRIMARY, &local_err);
567 qemu_mutex_unlock_iothread();
575 qemu_mutex_unlock_iothread();
576 trace_colo_vm_state_change("stop", "run");
578 timer_mod(s->colo_delay_timer, qemu_clock_get_ms(QEMU_CLOCK_HOST) +
579 migrate_checkpoint_delay());
581 while (s->state == MIGRATION_STATUS_COLO) {
582 if (failover_get_state() != FAILOVER_STATUS_NONE) {
583 error_report("failover request");
587 qemu_event_wait(&s->colo_checkpoint_event);
589 if (s->state != MIGRATION_STATUS_COLO) {
592 ret = colo_do_checkpoint_transaction(s, bioc, fb);
599 /* Throw the unreported error message after exited from loop */
601 error_report_err(local_err);
609 * There are only two reasons we can get here, some error happened
610 * or the user triggered failover.
612 switch (failover_get_state()) {
613 case FAILOVER_STATUS_COMPLETED:
614 qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
615 COLO_EXIT_REASON_REQUEST);
618 qapi_event_send_colo_exit(COLO_MODE_PRIMARY,
619 COLO_EXIT_REASON_ERROR);
622 /* Hope this not to be too long to wait here */
623 qemu_sem_wait(&s->colo_exit_sem);
624 qemu_sem_destroy(&s->colo_exit_sem);
627 * It is safe to unregister notifier after failover finished.
628 * Besides, colo_delay_timer and colo_checkpoint_sem can't be
629 * released before unregister notifier, or there will be use-after-free
632 colo_compare_unregister_notifier(&packets_compare_notifier);
633 timer_free(s->colo_delay_timer);
634 qemu_event_destroy(&s->colo_checkpoint_event);
637 * Must be called after failover BH is completed,
638 * Or the failover BH may shutdown the wrong fd that
639 * re-used by other threads after we release here.
641 if (s->rp_state.from_dst_file) {
642 qemu_fclose(s->rp_state.from_dst_file);
643 s->rp_state.from_dst_file = NULL;
647 void colo_checkpoint_notify(void *opaque)
649 MigrationState *s = opaque;
650 int64_t next_notify_time;
652 qemu_event_set(&s->colo_checkpoint_event);
653 s->colo_checkpoint_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
654 next_notify_time = s->colo_checkpoint_time + migrate_checkpoint_delay();
655 timer_mod(s->colo_delay_timer, next_notify_time);
658 void migrate_start_colo_process(MigrationState *s)
660 qemu_mutex_unlock_iothread();
661 qemu_event_init(&s->colo_checkpoint_event, false);
662 s->colo_delay_timer = timer_new_ms(QEMU_CLOCK_HOST,
663 colo_checkpoint_notify, s);
665 qemu_sem_init(&s->colo_exit_sem, 0);
666 colo_process_checkpoint(s);
667 qemu_mutex_lock_iothread();
670 static void colo_incoming_process_checkpoint(MigrationIncomingState *mis,
671 QEMUFile *fb, QIOChannelBuffer *bioc, Error **errp)
675 Error *local_err = NULL;
678 qemu_mutex_lock_iothread();
679 vm_stop_force_state(RUN_STATE_COLO);
680 qemu_mutex_unlock_iothread();
681 trace_colo_vm_state_change("run", "stop");
683 /* FIXME: This is unnecessary for periodic checkpoint mode */
684 colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_REPLY,
687 error_propagate(errp, local_err);
691 colo_receive_check_message(mis->from_src_file,
692 COLO_MESSAGE_VMSTATE_SEND, &local_err);
694 error_propagate(errp, local_err);
698 qemu_mutex_lock_iothread();
699 cpu_synchronize_all_states();
700 ret = qemu_loadvm_state_main(mis->from_src_file, mis);
701 qemu_mutex_unlock_iothread();
704 error_setg(errp, "Load VM's live state (ram) error");
708 value = colo_receive_message_value(mis->from_src_file,
709 COLO_MESSAGE_VMSTATE_SIZE, &local_err);
711 error_propagate(errp, local_err);
716 * Read VM device state data into channel buffer,
717 * It's better to re-use the memory allocated.
718 * Here we need to handle the channel buffer directly.
720 if (value > bioc->capacity) {
721 bioc->capacity = value;
722 bioc->data = g_realloc(bioc->data, bioc->capacity);
724 total_size = qemu_get_buffer(mis->from_src_file, bioc->data, value);
725 if (total_size != value) {
726 error_setg(errp, "Got %" PRIu64 " VMState data, less than expected"
727 " %" PRIu64, total_size, value);
730 bioc->usage = total_size;
731 qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
733 colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_RECEIVED,
736 error_propagate(errp, local_err);
740 qemu_mutex_lock_iothread();
741 vmstate_loading = true;
742 colo_flush_ram_cache();
743 ret = qemu_load_device_state(fb);
745 error_setg(errp, "COLO: load device state failed");
746 vmstate_loading = false;
747 qemu_mutex_unlock_iothread();
751 #ifdef CONFIG_REPLICATION
752 replication_get_error_all(&local_err);
754 error_propagate(errp, local_err);
755 vmstate_loading = false;
756 qemu_mutex_unlock_iothread();
760 /* discard colo disk buffer */
761 replication_do_checkpoint_all(&local_err);
763 error_propagate(errp, local_err);
764 vmstate_loading = false;
765 qemu_mutex_unlock_iothread();
771 /* Notify all filters of all NIC to do checkpoint */
772 colo_notify_filters_event(COLO_EVENT_CHECKPOINT, &local_err);
775 error_propagate(errp, local_err);
776 vmstate_loading = false;
777 qemu_mutex_unlock_iothread();
781 vmstate_loading = false;
783 qemu_mutex_unlock_iothread();
784 trace_colo_vm_state_change("stop", "run");
786 if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
790 colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_LOADED,
792 error_propagate(errp, local_err);
795 static void colo_wait_handle_message(MigrationIncomingState *mis,
796 QEMUFile *fb, QIOChannelBuffer *bioc, Error **errp)
799 Error *local_err = NULL;
801 msg = colo_receive_message(mis->from_src_file, &local_err);
803 error_propagate(errp, local_err);
808 case COLO_MESSAGE_CHECKPOINT_REQUEST:
809 colo_incoming_process_checkpoint(mis, fb, bioc, errp);
812 error_setg(errp, "Got unknown COLO message: %d", msg);
817 void colo_shutdown(void)
819 MigrationIncomingState *mis = NULL;
820 MigrationState *s = NULL;
822 switch (get_colo_mode()) {
823 case COLO_MODE_PRIMARY:
824 s = migrate_get_current();
825 qemu_event_set(&s->colo_checkpoint_event);
826 qemu_sem_post(&s->colo_exit_sem);
828 case COLO_MODE_SECONDARY:
829 mis = migration_incoming_get_current();
830 qemu_sem_post(&mis->colo_incoming_sem);
837 void *colo_process_incoming_thread(void *opaque)
839 MigrationIncomingState *mis = opaque;
841 QIOChannelBuffer *bioc = NULL; /* Cache incoming device state */
842 Error *local_err = NULL;
844 rcu_register_thread();
845 qemu_sem_init(&mis->colo_incoming_sem, 0);
847 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
848 MIGRATION_STATUS_COLO);
850 if (get_colo_mode() != COLO_MODE_SECONDARY) {
851 error_report("COLO mode must be COLO_MODE_SECONDARY");
855 failover_init_state();
857 mis->to_src_file = qemu_file_get_return_path(mis->from_src_file);
858 if (!mis->to_src_file) {
859 error_report("COLO incoming thread: Open QEMUFile to_src_file failed");
863 * Note: the communication between Primary side and Secondary side
864 * should be sequential, we set the fd to unblocked in migration incoming
865 * coroutine, and here we are in the COLO incoming thread, so it is ok to
866 * set the fd back to blocked.
868 qemu_file_set_blocking(mis->from_src_file, true);
870 colo_incoming_start_dirty_log();
872 bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
873 fb = qemu_file_new_input(QIO_CHANNEL(bioc));
874 object_unref(OBJECT(bioc));
876 qemu_mutex_lock_iothread();
877 #ifdef CONFIG_REPLICATION
878 replication_start_all(REPLICATION_MODE_SECONDARY, &local_err);
880 qemu_mutex_unlock_iothread();
887 qemu_mutex_unlock_iothread();
888 trace_colo_vm_state_change("stop", "run");
890 colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_READY,
896 while (mis->state == MIGRATION_STATUS_COLO) {
897 colo_wait_handle_message(mis, fb, bioc, &local_err);
899 error_report_err(local_err);
903 if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
904 failover_set_state(FAILOVER_STATUS_RELAUNCH,
905 FAILOVER_STATUS_NONE);
906 failover_request_active(NULL);
910 if (failover_get_state() != FAILOVER_STATUS_NONE) {
911 error_report("failover request");
918 * There are only two reasons we can get here, some error happened
919 * or the user triggered failover.
921 switch (failover_get_state()) {
922 case FAILOVER_STATUS_COMPLETED:
923 qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
924 COLO_EXIT_REASON_REQUEST);
927 qapi_event_send_colo_exit(COLO_MODE_SECONDARY,
928 COLO_EXIT_REASON_ERROR);
935 /* Hope this not to be too long to loop here */
936 qemu_sem_wait(&mis->colo_incoming_sem);
937 qemu_sem_destroy(&mis->colo_incoming_sem);
939 rcu_unregister_thread();