xref: /qemu/migration/colo.c (revision 9277d81f)
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 "qapi/error.h"
16 #include "qapi/qapi-commands-migration.h"
17 #include "qemu-file-channel.h"
18 #include "migration.h"
19 #include "qemu-file.h"
20 #include "savevm.h"
21 #include "migration/colo.h"
22 #include "block.h"
23 #include "io/channel-buffer.h"
24 #include "trace.h"
25 #include "qemu/error-report.h"
26 #include "migration/failover.h"
27 #include "replication.h"
28 
29 static bool vmstate_loading;
30 
31 #define COLO_BUFFER_BASE_SIZE (4 * 1024 * 1024)
32 
33 bool migration_in_colo_state(void)
34 {
35     MigrationState *s = migrate_get_current();
36 
37     return (s->state == MIGRATION_STATUS_COLO);
38 }
39 
40 bool migration_incoming_in_colo_state(void)
41 {
42     MigrationIncomingState *mis = migration_incoming_get_current();
43 
44     return mis && (mis->state == MIGRATION_STATUS_COLO);
45 }
46 
47 static bool colo_runstate_is_stopped(void)
48 {
49     return runstate_check(RUN_STATE_COLO) || !runstate_is_running();
50 }
51 
52 static void secondary_vm_do_failover(void)
53 {
54     int old_state;
55     MigrationIncomingState *mis = migration_incoming_get_current();
56 
57     /* Can not do failover during the process of VM's loading VMstate, Or
58      * it will break the secondary VM.
59      */
60     if (vmstate_loading) {
61         old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
62                         FAILOVER_STATUS_RELAUNCH);
63         if (old_state != FAILOVER_STATUS_ACTIVE) {
64             error_report("Unknown error while do failover for secondary VM,"
65                          "old_state: %s", FailoverStatus_str(old_state));
66         }
67         return;
68     }
69 
70     migrate_set_state(&mis->state, MIGRATION_STATUS_COLO,
71                       MIGRATION_STATUS_COMPLETED);
72 
73     if (!autostart) {
74         error_report("\"-S\" qemu option will be ignored in secondary side");
75         /* recover runstate to normal migration finish state */
76         autostart = true;
77     }
78     /*
79      * Make sure COLO incoming thread not block in recv or send,
80      * If mis->from_src_file and mis->to_src_file use the same fd,
81      * The second shutdown() will return -1, we ignore this value,
82      * It is harmless.
83      */
84     if (mis->from_src_file) {
85         qemu_file_shutdown(mis->from_src_file);
86     }
87     if (mis->to_src_file) {
88         qemu_file_shutdown(mis->to_src_file);
89     }
90 
91     old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
92                                    FAILOVER_STATUS_COMPLETED);
93     if (old_state != FAILOVER_STATUS_ACTIVE) {
94         error_report("Incorrect state (%s) while doing failover for "
95                      "secondary VM", FailoverStatus_str(old_state));
96         return;
97     }
98     /* Notify COLO incoming thread that failover work is finished */
99     qemu_sem_post(&mis->colo_incoming_sem);
100     /* For Secondary VM, jump to incoming co */
101     if (mis->migration_incoming_co) {
102         qemu_coroutine_enter(mis->migration_incoming_co);
103     }
104 }
105 
106 static void primary_vm_do_failover(void)
107 {
108     MigrationState *s = migrate_get_current();
109     int old_state;
110 
111     migrate_set_state(&s->state, MIGRATION_STATUS_COLO,
112                       MIGRATION_STATUS_COMPLETED);
113 
114     /*
115      * Wake up COLO thread which may blocked in recv() or send(),
116      * The s->rp_state.from_dst_file and s->to_dst_file may use the
117      * same fd, but we still shutdown the fd for twice, it is harmless.
118      */
119     if (s->to_dst_file) {
120         qemu_file_shutdown(s->to_dst_file);
121     }
122     if (s->rp_state.from_dst_file) {
123         qemu_file_shutdown(s->rp_state.from_dst_file);
124     }
125 
126     old_state = failover_set_state(FAILOVER_STATUS_ACTIVE,
127                                    FAILOVER_STATUS_COMPLETED);
128     if (old_state != FAILOVER_STATUS_ACTIVE) {
129         error_report("Incorrect state (%s) while doing failover for Primary VM",
130                      FailoverStatus_str(old_state));
131         return;
132     }
133     /* Notify COLO thread that failover work is finished */
134     qemu_sem_post(&s->colo_exit_sem);
135 }
136 
137 void colo_do_failover(MigrationState *s)
138 {
139     /* Make sure VM stopped while failover happened. */
140     if (!colo_runstate_is_stopped()) {
141         vm_stop_force_state(RUN_STATE_COLO);
142     }
143 
144     if (get_colo_mode() == COLO_MODE_PRIMARY) {
145         primary_vm_do_failover();
146     } else {
147         secondary_vm_do_failover();
148     }
149 }
150 
151 void qmp_xen_set_replication(bool enable, bool primary,
152                              bool has_failover, bool failover,
153                              Error **errp)
154 {
155 #ifdef CONFIG_REPLICATION
156     ReplicationMode mode = primary ?
157                            REPLICATION_MODE_PRIMARY :
158                            REPLICATION_MODE_SECONDARY;
159 
160     if (has_failover && enable) {
161         error_setg(errp, "Parameter 'failover' is only for"
162                    " stopping replication");
163         return;
164     }
165 
166     if (enable) {
167         replication_start_all(mode, errp);
168     } else {
169         if (!has_failover) {
170             failover = NULL;
171         }
172         replication_stop_all(failover, failover ? NULL : errp);
173     }
174 #else
175     abort();
176 #endif
177 }
178 
179 ReplicationStatus *qmp_query_xen_replication_status(Error **errp)
180 {
181 #ifdef CONFIG_REPLICATION
182     Error *err = NULL;
183     ReplicationStatus *s = g_new0(ReplicationStatus, 1);
184 
185     replication_get_error_all(&err);
186     if (err) {
187         s->error = true;
188         s->has_desc = true;
189         s->desc = g_strdup(error_get_pretty(err));
190     } else {
191         s->error = false;
192     }
193 
194     error_free(err);
195     return s;
196 #else
197     abort();
198 #endif
199 }
200 
201 void qmp_xen_colo_do_checkpoint(Error **errp)
202 {
203 #ifdef CONFIG_REPLICATION
204     replication_do_checkpoint_all(errp);
205 #else
206     abort();
207 #endif
208 }
209 
210 static void colo_send_message(QEMUFile *f, COLOMessage msg,
211                               Error **errp)
212 {
213     int ret;
214 
215     if (msg >= COLO_MESSAGE__MAX) {
216         error_setg(errp, "%s: Invalid message", __func__);
217         return;
218     }
219     qemu_put_be32(f, msg);
220     qemu_fflush(f);
221 
222     ret = qemu_file_get_error(f);
223     if (ret < 0) {
224         error_setg_errno(errp, -ret, "Can't send COLO message");
225     }
226     trace_colo_send_message(COLOMessage_str(msg));
227 }
228 
229 static void colo_send_message_value(QEMUFile *f, COLOMessage msg,
230                                     uint64_t value, Error **errp)
231 {
232     Error *local_err = NULL;
233     int ret;
234 
235     colo_send_message(f, msg, &local_err);
236     if (local_err) {
237         error_propagate(errp, local_err);
238         return;
239     }
240     qemu_put_be64(f, value);
241     qemu_fflush(f);
242 
243     ret = qemu_file_get_error(f);
244     if (ret < 0) {
245         error_setg_errno(errp, -ret, "Failed to send value for message:%s",
246                          COLOMessage_str(msg));
247     }
248 }
249 
250 static COLOMessage colo_receive_message(QEMUFile *f, Error **errp)
251 {
252     COLOMessage msg;
253     int ret;
254 
255     msg = qemu_get_be32(f);
256     ret = qemu_file_get_error(f);
257     if (ret < 0) {
258         error_setg_errno(errp, -ret, "Can't receive COLO message");
259         return msg;
260     }
261     if (msg >= COLO_MESSAGE__MAX) {
262         error_setg(errp, "%s: Invalid message", __func__);
263         return msg;
264     }
265     trace_colo_receive_message(COLOMessage_str(msg));
266     return msg;
267 }
268 
269 static void colo_receive_check_message(QEMUFile *f, COLOMessage expect_msg,
270                                        Error **errp)
271 {
272     COLOMessage msg;
273     Error *local_err = NULL;
274 
275     msg = colo_receive_message(f, &local_err);
276     if (local_err) {
277         error_propagate(errp, local_err);
278         return;
279     }
280     if (msg != expect_msg) {
281         error_setg(errp, "Unexpected COLO message %d, expected %d",
282                           msg, expect_msg);
283     }
284 }
285 
286 static uint64_t colo_receive_message_value(QEMUFile *f, uint32_t expect_msg,
287                                            Error **errp)
288 {
289     Error *local_err = NULL;
290     uint64_t value;
291     int ret;
292 
293     colo_receive_check_message(f, expect_msg, &local_err);
294     if (local_err) {
295         error_propagate(errp, local_err);
296         return 0;
297     }
298 
299     value = qemu_get_be64(f);
300     ret = qemu_file_get_error(f);
301     if (ret < 0) {
302         error_setg_errno(errp, -ret, "Failed to get value for COLO message: %s",
303                          COLOMessage_str(expect_msg));
304     }
305     return value;
306 }
307 
308 static int colo_do_checkpoint_transaction(MigrationState *s,
309                                           QIOChannelBuffer *bioc,
310                                           QEMUFile *fb)
311 {
312     Error *local_err = NULL;
313     int ret = -1;
314 
315     colo_send_message(s->to_dst_file, COLO_MESSAGE_CHECKPOINT_REQUEST,
316                       &local_err);
317     if (local_err) {
318         goto out;
319     }
320 
321     colo_receive_check_message(s->rp_state.from_dst_file,
322                     COLO_MESSAGE_CHECKPOINT_REPLY, &local_err);
323     if (local_err) {
324         goto out;
325     }
326     /* Reset channel-buffer directly */
327     qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
328     bioc->usage = 0;
329 
330     qemu_mutex_lock_iothread();
331     if (failover_get_state() != FAILOVER_STATUS_NONE) {
332         qemu_mutex_unlock_iothread();
333         goto out;
334     }
335     vm_stop_force_state(RUN_STATE_COLO);
336     qemu_mutex_unlock_iothread();
337     trace_colo_vm_state_change("run", "stop");
338     /*
339      * Failover request bh could be called after vm_stop_force_state(),
340      * So we need check failover_request_is_active() again.
341      */
342     if (failover_get_state() != FAILOVER_STATUS_NONE) {
343         goto out;
344     }
345 
346     /* Disable block migration */
347     migrate_set_block_enabled(false, &local_err);
348     qemu_savevm_state_header(fb);
349     qemu_savevm_state_setup(fb);
350     qemu_mutex_lock_iothread();
351     qemu_savevm_state_complete_precopy(fb, false, false);
352     qemu_mutex_unlock_iothread();
353 
354     qemu_fflush(fb);
355 
356     colo_send_message(s->to_dst_file, COLO_MESSAGE_VMSTATE_SEND, &local_err);
357     if (local_err) {
358         goto out;
359     }
360     /*
361      * We need the size of the VMstate data in Secondary side,
362      * With which we can decide how much data should be read.
363      */
364     colo_send_message_value(s->to_dst_file, COLO_MESSAGE_VMSTATE_SIZE,
365                             bioc->usage, &local_err);
366     if (local_err) {
367         goto out;
368     }
369 
370     qemu_put_buffer(s->to_dst_file, bioc->data, bioc->usage);
371     qemu_fflush(s->to_dst_file);
372     ret = qemu_file_get_error(s->to_dst_file);
373     if (ret < 0) {
374         goto out;
375     }
376 
377     colo_receive_check_message(s->rp_state.from_dst_file,
378                        COLO_MESSAGE_VMSTATE_RECEIVED, &local_err);
379     if (local_err) {
380         goto out;
381     }
382 
383     colo_receive_check_message(s->rp_state.from_dst_file,
384                        COLO_MESSAGE_VMSTATE_LOADED, &local_err);
385     if (local_err) {
386         goto out;
387     }
388 
389     ret = 0;
390 
391     qemu_mutex_lock_iothread();
392     vm_start();
393     qemu_mutex_unlock_iothread();
394     trace_colo_vm_state_change("stop", "run");
395 
396 out:
397     if (local_err) {
398         error_report_err(local_err);
399     }
400     return ret;
401 }
402 
403 static void colo_process_checkpoint(MigrationState *s)
404 {
405     QIOChannelBuffer *bioc;
406     QEMUFile *fb = NULL;
407     int64_t current_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
408     Error *local_err = NULL;
409     int ret;
410 
411     failover_init_state();
412 
413     s->rp_state.from_dst_file = qemu_file_get_return_path(s->to_dst_file);
414     if (!s->rp_state.from_dst_file) {
415         error_report("Open QEMUFile from_dst_file failed");
416         goto out;
417     }
418 
419     /*
420      * Wait for Secondary finish loading VM states and enter COLO
421      * restore.
422      */
423     colo_receive_check_message(s->rp_state.from_dst_file,
424                        COLO_MESSAGE_CHECKPOINT_READY, &local_err);
425     if (local_err) {
426         goto out;
427     }
428     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
429     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
430     object_unref(OBJECT(bioc));
431 
432     qemu_mutex_lock_iothread();
433     vm_start();
434     qemu_mutex_unlock_iothread();
435     trace_colo_vm_state_change("stop", "run");
436 
437     timer_mod(s->colo_delay_timer,
438             current_time + s->parameters.x_checkpoint_delay);
439 
440     while (s->state == MIGRATION_STATUS_COLO) {
441         if (failover_get_state() != FAILOVER_STATUS_NONE) {
442             error_report("failover request");
443             goto out;
444         }
445 
446         qemu_sem_wait(&s->colo_checkpoint_sem);
447 
448         ret = colo_do_checkpoint_transaction(s, bioc, fb);
449         if (ret < 0) {
450             goto out;
451         }
452     }
453 
454 out:
455     /* Throw the unreported error message after exited from loop */
456     if (local_err) {
457         error_report_err(local_err);
458     }
459 
460     if (fb) {
461         qemu_fclose(fb);
462     }
463 
464     timer_del(s->colo_delay_timer);
465 
466     /* Hope this not to be too long to wait here */
467     qemu_sem_wait(&s->colo_exit_sem);
468     qemu_sem_destroy(&s->colo_exit_sem);
469     /*
470      * Must be called after failover BH is completed,
471      * Or the failover BH may shutdown the wrong fd that
472      * re-used by other threads after we release here.
473      */
474     if (s->rp_state.from_dst_file) {
475         qemu_fclose(s->rp_state.from_dst_file);
476     }
477 }
478 
479 void colo_checkpoint_notify(void *opaque)
480 {
481     MigrationState *s = opaque;
482     int64_t next_notify_time;
483 
484     qemu_sem_post(&s->colo_checkpoint_sem);
485     s->colo_checkpoint_time = qemu_clock_get_ms(QEMU_CLOCK_HOST);
486     next_notify_time = s->colo_checkpoint_time +
487                     s->parameters.x_checkpoint_delay;
488     timer_mod(s->colo_delay_timer, next_notify_time);
489 }
490 
491 void migrate_start_colo_process(MigrationState *s)
492 {
493     qemu_mutex_unlock_iothread();
494     qemu_sem_init(&s->colo_checkpoint_sem, 0);
495     s->colo_delay_timer =  timer_new_ms(QEMU_CLOCK_HOST,
496                                 colo_checkpoint_notify, s);
497 
498     qemu_sem_init(&s->colo_exit_sem, 0);
499     migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE,
500                       MIGRATION_STATUS_COLO);
501     colo_process_checkpoint(s);
502     qemu_mutex_lock_iothread();
503 }
504 
505 static void colo_wait_handle_message(QEMUFile *f, int *checkpoint_request,
506                                      Error **errp)
507 {
508     COLOMessage msg;
509     Error *local_err = NULL;
510 
511     msg = colo_receive_message(f, &local_err);
512     if (local_err) {
513         error_propagate(errp, local_err);
514         return;
515     }
516 
517     switch (msg) {
518     case COLO_MESSAGE_CHECKPOINT_REQUEST:
519         *checkpoint_request = 1;
520         break;
521     default:
522         *checkpoint_request = 0;
523         error_setg(errp, "Got unknown COLO message: %d", msg);
524         break;
525     }
526 }
527 
528 void *colo_process_incoming_thread(void *opaque)
529 {
530     MigrationIncomingState *mis = opaque;
531     QEMUFile *fb = NULL;
532     QIOChannelBuffer *bioc = NULL; /* Cache incoming device state */
533     uint64_t total_size;
534     uint64_t value;
535     Error *local_err = NULL;
536 
537     qemu_sem_init(&mis->colo_incoming_sem, 0);
538 
539     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
540                       MIGRATION_STATUS_COLO);
541 
542     failover_init_state();
543 
544     mis->to_src_file = qemu_file_get_return_path(mis->from_src_file);
545     if (!mis->to_src_file) {
546         error_report("COLO incoming thread: Open QEMUFile to_src_file failed");
547         goto out;
548     }
549     /*
550      * Note: the communication between Primary side and Secondary side
551      * should be sequential, we set the fd to unblocked in migration incoming
552      * coroutine, and here we are in the COLO incoming thread, so it is ok to
553      * set the fd back to blocked.
554      */
555     qemu_file_set_blocking(mis->from_src_file, true);
556 
557     bioc = qio_channel_buffer_new(COLO_BUFFER_BASE_SIZE);
558     fb = qemu_fopen_channel_input(QIO_CHANNEL(bioc));
559     object_unref(OBJECT(bioc));
560 
561     colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_READY,
562                       &local_err);
563     if (local_err) {
564         goto out;
565     }
566 
567     while (mis->state == MIGRATION_STATUS_COLO) {
568         int request = 0;
569 
570         colo_wait_handle_message(mis->from_src_file, &request, &local_err);
571         if (local_err) {
572             goto out;
573         }
574         assert(request);
575         if (failover_get_state() != FAILOVER_STATUS_NONE) {
576             error_report("failover request");
577             goto out;
578         }
579 
580         /* FIXME: This is unnecessary for periodic checkpoint mode */
581         colo_send_message(mis->to_src_file, COLO_MESSAGE_CHECKPOINT_REPLY,
582                      &local_err);
583         if (local_err) {
584             goto out;
585         }
586 
587         colo_receive_check_message(mis->from_src_file,
588                            COLO_MESSAGE_VMSTATE_SEND, &local_err);
589         if (local_err) {
590             goto out;
591         }
592 
593         value = colo_receive_message_value(mis->from_src_file,
594                                  COLO_MESSAGE_VMSTATE_SIZE, &local_err);
595         if (local_err) {
596             goto out;
597         }
598 
599         /*
600          * Read VM device state data into channel buffer,
601          * It's better to re-use the memory allocated.
602          * Here we need to handle the channel buffer directly.
603          */
604         if (value > bioc->capacity) {
605             bioc->capacity = value;
606             bioc->data = g_realloc(bioc->data, bioc->capacity);
607         }
608         total_size = qemu_get_buffer(mis->from_src_file, bioc->data, value);
609         if (total_size != value) {
610             error_report("Got %" PRIu64 " VMState data, less than expected"
611                         " %" PRIu64, total_size, value);
612             goto out;
613         }
614         bioc->usage = total_size;
615         qio_channel_io_seek(QIO_CHANNEL(bioc), 0, 0, NULL);
616 
617         colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_RECEIVED,
618                      &local_err);
619         if (local_err) {
620             goto out;
621         }
622 
623         qemu_mutex_lock_iothread();
624         qemu_system_reset(SHUTDOWN_CAUSE_NONE);
625         vmstate_loading = true;
626         if (qemu_loadvm_state(fb) < 0) {
627             error_report("COLO: loadvm failed");
628             qemu_mutex_unlock_iothread();
629             goto out;
630         }
631 
632         vmstate_loading = false;
633         qemu_mutex_unlock_iothread();
634 
635         if (failover_get_state() == FAILOVER_STATUS_RELAUNCH) {
636             failover_set_state(FAILOVER_STATUS_RELAUNCH,
637                             FAILOVER_STATUS_NONE);
638             failover_request_active(NULL);
639             goto out;
640         }
641 
642         colo_send_message(mis->to_src_file, COLO_MESSAGE_VMSTATE_LOADED,
643                      &local_err);
644         if (local_err) {
645             goto out;
646         }
647     }
648 
649 out:
650     vmstate_loading = false;
651     /* Throw the unreported error message after exited from loop */
652     if (local_err) {
653         error_report_err(local_err);
654     }
655 
656     if (fb) {
657         qemu_fclose(fb);
658     }
659 
660     /* Hope this not to be too long to loop here */
661     qemu_sem_wait(&mis->colo_incoming_sem);
662     qemu_sem_destroy(&mis->colo_incoming_sem);
663     /* Must be called after failover BH is completed */
664     if (mis->to_src_file) {
665         qemu_fclose(mis->to_src_file);
666     }
667     migration_incoming_exit_colo();
668 
669     return NULL;
670 }
671