xref: /openbmc/qemu/migration/colo.c (revision 864a2178)
1 /*
2  * COarse-grain LOck-stepping Virtual Machines for Non-stop Service (COLO)
3  * (a.k.a. Fault Tolerance or Continuous Replication)
4  *
5  * Copyright (c) 2016 HUAWEI TECHNOLOGIES CO., LTD.
6  * Copyright (c) 2016 FUJITSU LIMITED
7  * Copyright (c) 2016 Intel Corporation
8  *
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.
11  */
12 
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"
18 #include "savevm.h"
19 #include "migration/colo.h"
20 #include "block.h"
21 #include "io/channel-buffer.h"
22 #include "trace.h"
23 #include "qemu/error-report.h"
24 #include "migration/failover.h"
25 #include "replication.h"
26 #include "qmp-commands.h"
27 
28 static bool vmstate_loading;
29 
30 #define COLO_BUFFER_BASE_SIZE (4 * 1024 * 1024)
31 
32 bool colo_supported(void)
33 {
34     return true;
35 }
36 
37 bool migration_in_colo_state(void)
38 {
39     MigrationState *s = migrate_get_current();
40 
41     return (s->state == MIGRATION_STATUS_COLO);
42 }
43 
44 bool migration_incoming_in_colo_state(void)
45 {
46     MigrationIncomingState *mis = migration_incoming_get_current();
47 
48     return mis && (mis->state == MIGRATION_STATUS_COLO);
49 }
50 
51 static bool colo_runstate_is_stopped(void)
52 {
53     return runstate_check(RUN_STATE_COLO) || !runstate_is_running();
54 }
55 
56 static void secondary_vm_do_failover(void)
57 {
58     int old_state;
59     MigrationIncomingState *mis = migration_incoming_get_current();
60 
61     /* Can not do failover during the process of VM's loading VMstate, Or
62      * it will break the secondary VM.
63      */
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]);
70         }
71         return;
72     }
73 
74     migrate_set_state(&mis->state, MIGRATION_STATUS_COLO,
75                       MIGRATION_STATUS_COMPLETED);
76 
77     if (!autostart) {
78         error_report("\"-S\" qemu option will be ignored in secondary side");
79         /* recover runstate to normal migration finish state */
80         autostart = true;
81     }
82     /*
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,
86      * It is harmless.
87      */
88     if (mis->from_src_file) {
89         qemu_file_shutdown(mis->from_src_file);
90     }
91     if (mis->to_src_file) {
92         qemu_file_shutdown(mis->to_src_file);
93     }
94 
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]);
100         return;
101     }
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);
107     }
108 }
109 
110 static void primary_vm_do_failover(void)
111 {
112     MigrationState *s = migrate_get_current();
113     int old_state;
114 
115     migrate_set_state(&s->state, MIGRATION_STATUS_COLO,
116                       MIGRATION_STATUS_COMPLETED);
117 
118     /*
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.
122      */
123     if (s->to_dst_file) {
124         qemu_file_shutdown(s->to_dst_file);
125     }
126     if (s->rp_state.from_dst_file) {
127         qemu_file_shutdown(s->rp_state.from_dst_file);
128     }
129 
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]);
135         return;
136     }
137     /* Notify COLO thread that failover work is finished */
138     qemu_sem_post(&s->colo_exit_sem);
139 }
140 
141 void colo_do_failover(MigrationState *s)
142 {
143     /* Make sure VM stopped while failover happened. */
144     if (!colo_runstate_is_stopped()) {
145         vm_stop_force_state(RUN_STATE_COLO);
146     }
147 
148     if (get_colo_mode() == COLO_MODE_PRIMARY) {
149         primary_vm_do_failover();
150     } else {
151         secondary_vm_do_failover();
152     }
153 }
154 
155 void qmp_xen_set_replication(bool enable, bool primary,
156                              bool has_failover, bool failover,
157                              Error **errp)
158 {
159 #ifdef CONFIG_REPLICATION
160     ReplicationMode mode = primary ?
161                            REPLICATION_MODE_PRIMARY :
162                            REPLICATION_MODE_SECONDARY;
163 
164     if (has_failover && enable) {
165         error_setg(errp, "Parameter 'failover' is only for"
166                    " stopping replication");
167         return;
168     }
169 
170     if (enable) {
171         replication_start_all(mode, errp);
172     } else {
173         if (!has_failover) {
174             failover = NULL;
175         }
176         replication_stop_all(failover, failover ? NULL : errp);
177     }
178 #else
179     abort();
180 #endif
181 }
182 
183 ReplicationStatus *qmp_query_xen_replication_status(Error **errp)
184 {
185 #ifdef CONFIG_REPLICATION
186     Error *err = NULL;
187     ReplicationStatus *s = g_new0(ReplicationStatus, 1);
188 
189     replication_get_error_all(&err);
190     if (err) {
191         s->error = true;
192         s->has_desc = true;
193         s->desc = g_strdup(error_get_pretty(err));
194     } else {
195         s->error = false;
196     }
197 
198     error_free(err);
199     return s;
200 #else
201     abort();
202 #endif
203 }
204 
205 void qmp_xen_colo_do_checkpoint(Error **errp)
206 {
207 #ifdef CONFIG_REPLICATION
208     replication_do_checkpoint_all(errp);
209 #else
210     abort();
211 #endif
212 }
213 
214 static void colo_send_message(QEMUFile *f, COLOMessage msg,
215                               Error **errp)
216 {
217     int ret;
218 
219     if (msg >= COLO_MESSAGE__MAX) {
220         error_setg(errp, "%s: Invalid message", __func__);
221         return;
222     }
223     qemu_put_be32(f, msg);
224     qemu_fflush(f);
225 
226     ret = qemu_file_get_error(f);
227     if (ret < 0) {
228         error_setg_errno(errp, -ret, "Can't send COLO message");
229     }
230     trace_colo_send_message(COLOMessage_lookup[msg]);
231 }
232 
233 static void colo_send_message_value(QEMUFile *f, COLOMessage msg,
234                                     uint64_t value, Error **errp)
235 {
236     Error *local_err = NULL;
237     int ret;
238 
239     colo_send_message(f, msg, &local_err);
240     if (local_err) {
241         error_propagate(errp, local_err);
242         return;
243     }
244     qemu_put_be64(f, value);
245     qemu_fflush(f);
246 
247     ret = qemu_file_get_error(f);
248     if (ret < 0) {
249         error_setg_errno(errp, -ret, "Failed to send value for message:%s",
250                          COLOMessage_lookup[msg]);
251     }
252 }
253 
254 static COLOMessage colo_receive_message(QEMUFile *f, Error **errp)
255 {
256     COLOMessage msg;
257     int ret;
258 
259     msg = qemu_get_be32(f);
260     ret = qemu_file_get_error(f);
261     if (ret < 0) {
262         error_setg_errno(errp, -ret, "Can't receive COLO message");
263         return msg;
264     }
265     if (msg >= COLO_MESSAGE__MAX) {
266         error_setg(errp, "%s: Invalid message", __func__);
267         return msg;
268     }
269     trace_colo_receive_message(COLOMessage_lookup[msg]);
270     return msg;
271 }
272 
273 static void colo_receive_check_message(QEMUFile *f, COLOMessage expect_msg,
274                                        Error **errp)
275 {
276     COLOMessage msg;
277     Error *local_err = NULL;
278 
279     msg = colo_receive_message(f, &local_err);
280     if (local_err) {
281         error_propagate(errp, local_err);
282         return;
283     }
284     if (msg != expect_msg) {
285         error_setg(errp, "Unexpected COLO message %d, expected %d",
286                           msg, expect_msg);
287     }
288 }
289 
290 static uint64_t colo_receive_message_value(QEMUFile *f, uint32_t expect_msg,
291                                            Error **errp)
292 {
293     Error *local_err = NULL;
294     uint64_t value;
295     int ret;
296 
297     colo_receive_check_message(f, expect_msg, &local_err);
298     if (local_err) {
299         error_propagate(errp, local_err);
300         return 0;
301     }
302 
303     value = qemu_get_be64(f);
304     ret = qemu_file_get_error(f);
305     if (ret < 0) {
306         error_setg_errno(errp, -ret, "Failed to get value for COLO message: %s",
307                          COLOMessage_lookup[expect_msg]);
308     }
309     return value;
310 }
311 
312 static int colo_do_checkpoint_transaction(MigrationState *s,
313                                           QIOChannelBuffer *bioc,
314                                           QEMUFile *fb)
315 {
316     Error *local_err = NULL;
317     int ret = -1;
318 
319     colo_send_message(s->to_dst_file, COLO_MESSAGE_CHECKPOINT_REQUEST,
320                       &local_err);
321     if (local_err) {
322         goto out;
323     }
324 
325     colo_receive_check_message(s->rp_state.from_dst_file,
326                     COLO_MESSAGE_CHECKPOINT_REPLY, &local_err);
327     if (local_err) {
328         goto out;
329     }
330     /* Reset channel-buffer directly */
331     qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
332     bioc->usage = 0;
333 
334     qemu_mutex_lock_iothread();
335     if (failover_get_state() != FAILOVER_STATUS_NONE) {
336         qemu_mutex_unlock_iothread();
337         goto out;
338     }
339     vm_stop_force_state(RUN_STATE_COLO);
340     qemu_mutex_unlock_iothread();
341     trace_colo_vm_state_change("run", "stop");
342     /*
343      * Failover request bh could be called after vm_stop_force_state(),
344      * So we need check failover_request_is_active() again.
345      */
346     if (failover_get_state() != FAILOVER_STATUS_NONE) {
347         goto out;
348     }
349 
350     /* Disable block migration */
351     migrate_set_block_enabled(false, &local_err);
352     qemu_savevm_state_header(fb);
353     qemu_savevm_state_setup(fb);
354     qemu_mutex_lock_iothread();
355     qemu_savevm_state_complete_precopy(fb, false, false);
356     qemu_mutex_unlock_iothread();
357 
358     qemu_fflush(fb);
359 
360     colo_send_message(s->to_dst_file, COLO_MESSAGE_VMSTATE_SEND, &local_err);
361     if (local_err) {
362         goto out;
363     }
364     /*
365      * We need the size of the VMstate data in Secondary side,
366      * With which we can decide how much data should be read.
367      */
368     colo_send_message_value(s->to_dst_file, COLO_MESSAGE_VMSTATE_SIZE,
369                             bioc->usage, &local_err);
370     if (local_err) {
371         goto out;
372     }
373 
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);
377     if (ret < 0) {
378         goto out;
379     }
380 
381     colo_receive_check_message(s->rp_state.from_dst_file,
382                        COLO_MESSAGE_VMSTATE_RECEIVED, &local_err);
383     if (local_err) {
384         goto out;
385     }
386 
387     colo_receive_check_message(s->rp_state.from_dst_file,
388                        COLO_MESSAGE_VMSTATE_LOADED, &local_err);
389     if (local_err) {
390         goto out;
391     }
392 
393     ret = 0;
394 
395     qemu_mutex_lock_iothread();
396     vm_start();
397     qemu_mutex_unlock_iothread();
398     trace_colo_vm_state_change("stop", "run");
399 
400 out:
401     if (local_err) {
402         error_report_err(local_err);
403     }
404     return ret;
405 }
406 
407 static void colo_process_checkpoint(MigrationState *s)
408 {
409     QIOChannelBuffer *bioc;
410     QEMUFile *fb = NULL;
411     int64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
412     Error *local_err = NULL;
413     int ret;
414 
415     failover_init_state();
416 
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");
420         goto out;
421     }
422 
423     /*
424      * Wait for Secondary finish loading VM states and enter COLO
425      * restore.
426      */
427     colo_receive_check_message(s->rp_state.from_dst_file,
428                        COLO_MESSAGE_CHECKPOINT_READY, &local_err);
429     if (local_err) {
430         goto out;
431     }
432     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
433     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
434     object_unref(OBJECT(bioc));
435 
436     qemu_mutex_lock_iothread();
437     vm_start();
438     qemu_mutex_unlock_iothread();
439     trace_colo_vm_state_change("stop", "run");
440 
441     timer_mod(s->colo_delay_timer,
442             current_time + s->parameters.x_checkpoint_delay);
443 
444     while (s->state == MIGRATION_STATUS_COLO) {
445         if (failover_get_state() != FAILOVER_STATUS_NONE) {
446             error_report("failover request");
447             goto out;
448         }
449 
450         qemu_sem_wait(&s->colo_checkpoint_sem);
451 
452         ret = colo_do_checkpoint_transaction(s, bioc, fb);
453         if (ret < 0) {
454             goto out;
455         }
456     }
457 
458 out:
459     /* Throw the unreported error message after exited from loop */
460     if (local_err) {
461         error_report_err(local_err);
462     }
463 
464     if (fb) {
465         qemu_fclose(fb);
466     }
467 
468     timer_del(s->colo_delay_timer);
469 
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);
473     /*
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.
477      */
478     if (s->rp_state.from_dst_file) {
479         qemu_fclose(s->rp_state.from_dst_file);
480     }
481 }
482 
483 void colo_checkpoint_notify(void *opaque)
484 {
485     MigrationState *s = opaque;
486     int64_t next_notify_time;
487 
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);
493 }
494 
495 void migrate_start_colo_process(MigrationState *s)
496 {
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);
501 
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();
507 }
508 
509 static void colo_wait_handle_message(QEMUFile *f, int *checkpoint_request,
510                                      Error **errp)
511 {
512     COLOMessage msg;
513     Error *local_err = NULL;
514 
515     msg = colo_receive_message(f, &local_err);
516     if (local_err) {
517         error_propagate(errp, local_err);
518         return;
519     }
520 
521     switch (msg) {
522     case COLO_MESSAGE_CHECKPOINT_REQUEST:
523         *checkpoint_request = 1;
524         break;
525     default:
526         *checkpoint_request = 0;
527         error_setg(errp, "Got unknown COLO message: %d", msg);
528         break;
529     }
530 }
531 
532 void *colo_process_incoming_thread(void *opaque)
533 {
534     MigrationIncomingState *mis = opaque;
535     QEMUFile *fb = NULL;
536     QIOChannelBuffer *bioc = NULL; /* Cache incoming device state */
537     uint64_t total_size;
538     uint64_t value;
539     Error *local_err = NULL;
540 
541     qemu_sem_init(&mis->colo_incoming_sem, 0);
542 
543     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
544                       MIGRATION_STATUS_COLO);
545 
546     failover_init_state();
547 
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");
551         goto out;
552     }
553     /*
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.
558      */
559     qemu_file_set_blocking(mis->from_src_file, true);
560 
561     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
562     fb = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
563     object_unref(OBJECT(bioc));
564 
565     colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_READY,
566                       &local_err);
567     if (local_err) {
568         goto out;
569     }
570 
571     while (mis->state == MIGRATION_STATUS_COLO) {
572         int request = 0;
573 
574         colo_wait_handle_message(mis->from_src_file, &request, &local_err);
575         if (local_err) {
576             goto out;
577         }
578         assert(request);
579         if (failover_get_state() != FAILOVER_STATUS_NONE) {
580             error_report("failover request");
581             goto out;
582         }
583 
584         /* FIXME: This is unnecessary for periodic checkpoint mode */
585         colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_REPLY,
586                      &local_err);
587         if (local_err) {
588             goto out;
589         }
590 
591         colo_receive_check_message(mis->from_src_file,
592                            COLO_MESSAGE_VMSTATE_SEND, &local_err);
593         if (local_err) {
594             goto out;
595         }
596 
597         value = colo_receive_message_value(mis->from_src_file,
598                                  COLO_MESSAGE_VMSTATE_SIZE, &local_err);
599         if (local_err) {
600             goto out;
601         }
602 
603         /*
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.
607          */
608         if (value > bioc->capacity) {
609             bioc->capacity = value;
610             bioc->data = g_realloc(bioc->data, bioc->capacity);
611         }
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);
616             goto out;
617         }
618         bioc->usage = total_size;
619         qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
620 
621         colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_RECEIVED,
622                      &local_err);
623         if (local_err) {
624             goto out;
625         }
626 
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();
633             goto out;
634         }
635 
636         vmstate_loading = false;
637         qemu_mutex_unlock_iothread();
638 
639         if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
640             failover_set_state(FAILOVER_STATUS_RELAUNCH,
641                             FAILOVER_STATUS_NONE);
642             failover_request_active(NULL);
643             goto out;
644         }
645 
646         colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_LOADED,
647                      &local_err);
648         if (local_err) {
649             goto out;
650         }
651     }
652 
653 out:
654     vmstate_loading = false;
655     /* Throw the unreported error message after exited from loop */
656     if (local_err) {
657         error_report_err(local_err);
658     }
659 
660     if (fb) {
661         qemu_fclose(fb);
662     }
663 
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);
670     }
671     migration_incoming_exit_colo();
672 
673     return NULL;
674 }
675