]> git.proxmox.com Git - qemu.git/blob - migration.c
Rename buffered_ to migration_
[qemu.git] / migration.c
1 /*
2 * QEMU live migration
3 *
4 * Copyright IBM, Corp. 2008
5 *
6 * Authors:
7 * Anthony Liguori <aliguori@us.ibm.com>
8 *
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
11 *
12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
14 */
15
16 #include "qemu-common.h"
17 #include "migration/migration.h"
18 #include "monitor/monitor.h"
19 #include "migration/qemu-file.h"
20 #include "sysemu/sysemu.h"
21 #include "block/block.h"
22 #include "qemu/sockets.h"
23 #include "migration/block.h"
24 #include "qemu/thread.h"
25 #include "qmp-commands.h"
26 #include "trace.h"
27
28 //#define DEBUG_MIGRATION
29
30 #ifdef DEBUG_MIGRATION
31 #define DPRINTF(fmt, ...) \
32 do { printf("migration: " fmt, ## __VA_ARGS__); } while (0)
33 #else
34 #define DPRINTF(fmt, ...) \
35 do { } while (0)
36 #endif
37
38 enum {
39 MIG_STATE_ERROR,
40 MIG_STATE_SETUP,
41 MIG_STATE_CANCELLED,
42 MIG_STATE_ACTIVE,
43 MIG_STATE_COMPLETED,
44 };
45
46 #define MAX_THROTTLE (32 << 20) /* Migration speed throttling */
47
48 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
49 * data. */
50 #define BUFFER_DELAY 100
51 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
52
53 /* Migration XBZRLE default cache size */
54 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
55
56 static NotifierList migration_state_notifiers =
57 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
58
59 /* When we add fault tolerance, we could have several
60 migrations at once. For now we don't need to add
61 dynamic creation of migration */
62
63 MigrationState *migrate_get_current(void)
64 {
65 static MigrationState current_migration = {
66 .state = MIG_STATE_SETUP,
67 .bandwidth_limit = MAX_THROTTLE,
68 .xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE,
69 };
70
71 return &current_migration;
72 }
73
74 void qemu_start_incoming_migration(const char *uri, Error **errp)
75 {
76 const char *p;
77
78 if (strstart(uri, "tcp:", &p))
79 tcp_start_incoming_migration(p, errp);
80 #if !defined(WIN32)
81 else if (strstart(uri, "exec:", &p))
82 exec_start_incoming_migration(p, errp);
83 else if (strstart(uri, "unix:", &p))
84 unix_start_incoming_migration(p, errp);
85 else if (strstart(uri, "fd:", &p))
86 fd_start_incoming_migration(p, errp);
87 #endif
88 else {
89 error_setg(errp, "unknown migration protocol: %s", uri);
90 }
91 }
92
93 static void process_incoming_migration_co(void *opaque)
94 {
95 QEMUFile *f = opaque;
96 int ret;
97
98 ret = qemu_loadvm_state(f);
99 qemu_fclose(f);
100 if (ret < 0) {
101 fprintf(stderr, "load of migration failed\n");
102 exit(0);
103 }
104 qemu_announce_self();
105 DPRINTF("successfully loaded vm state\n");
106
107 bdrv_clear_incoming_migration_all();
108 /* Make sure all file formats flush their mutable metadata */
109 bdrv_invalidate_cache_all();
110
111 if (autostart) {
112 vm_start();
113 } else {
114 runstate_set(RUN_STATE_PAUSED);
115 }
116 }
117
118 void process_incoming_migration(QEMUFile *f)
119 {
120 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co);
121 int fd = qemu_get_fd(f);
122
123 assert(fd != -1);
124 socket_set_nonblock(fd);
125 qemu_coroutine_enter(co, f);
126 }
127
128 /* amount of nanoseconds we are willing to wait for migration to be down.
129 * the choice of nanoseconds is because it is the maximum resolution that
130 * get_clock() can achieve. It is an internal measure. All user-visible
131 * units must be in seconds */
132 static uint64_t max_downtime = 30000000;
133
134 uint64_t migrate_max_downtime(void)
135 {
136 return max_downtime;
137 }
138
139 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
140 {
141 MigrationCapabilityStatusList *head = NULL;
142 MigrationCapabilityStatusList *caps;
143 MigrationState *s = migrate_get_current();
144 int i;
145
146 for (i = 0; i < MIGRATION_CAPABILITY_MAX; i++) {
147 if (head == NULL) {
148 head = g_malloc0(sizeof(*caps));
149 caps = head;
150 } else {
151 caps->next = g_malloc0(sizeof(*caps));
152 caps = caps->next;
153 }
154 caps->value =
155 g_malloc(sizeof(*caps->value));
156 caps->value->capability = i;
157 caps->value->state = s->enabled_capabilities[i];
158 }
159
160 return head;
161 }
162
163 static void get_xbzrle_cache_stats(MigrationInfo *info)
164 {
165 if (migrate_use_xbzrle()) {
166 info->has_xbzrle_cache = true;
167 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
168 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
169 info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred();
170 info->xbzrle_cache->pages = xbzrle_mig_pages_transferred();
171 info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss();
172 info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow();
173 }
174 }
175
176 MigrationInfo *qmp_query_migrate(Error **errp)
177 {
178 MigrationInfo *info = g_malloc0(sizeof(*info));
179 MigrationState *s = migrate_get_current();
180
181 switch (s->state) {
182 case MIG_STATE_SETUP:
183 /* no migration has happened ever */
184 break;
185 case MIG_STATE_ACTIVE:
186 info->has_status = true;
187 info->status = g_strdup("active");
188 info->has_total_time = true;
189 info->total_time = qemu_get_clock_ms(rt_clock)
190 - s->total_time;
191 info->has_expected_downtime = true;
192 info->expected_downtime = s->expected_downtime;
193
194 info->has_ram = true;
195 info->ram = g_malloc0(sizeof(*info->ram));
196 info->ram->transferred = ram_bytes_transferred();
197 info->ram->remaining = ram_bytes_remaining();
198 info->ram->total = ram_bytes_total();
199 info->ram->duplicate = dup_mig_pages_transferred();
200 info->ram->normal = norm_mig_pages_transferred();
201 info->ram->normal_bytes = norm_mig_bytes_transferred();
202 info->ram->dirty_pages_rate = s->dirty_pages_rate;
203
204
205 if (blk_mig_active()) {
206 info->has_disk = true;
207 info->disk = g_malloc0(sizeof(*info->disk));
208 info->disk->transferred = blk_mig_bytes_transferred();
209 info->disk->remaining = blk_mig_bytes_remaining();
210 info->disk->total = blk_mig_bytes_total();
211 }
212
213 get_xbzrle_cache_stats(info);
214 break;
215 case MIG_STATE_COMPLETED:
216 get_xbzrle_cache_stats(info);
217
218 info->has_status = true;
219 info->status = g_strdup("completed");
220 info->total_time = s->total_time;
221 info->has_downtime = true;
222 info->downtime = s->downtime;
223
224 info->has_ram = true;
225 info->ram = g_malloc0(sizeof(*info->ram));
226 info->ram->transferred = ram_bytes_transferred();
227 info->ram->remaining = 0;
228 info->ram->total = ram_bytes_total();
229 info->ram->duplicate = dup_mig_pages_transferred();
230 info->ram->normal = norm_mig_pages_transferred();
231 info->ram->normal_bytes = norm_mig_bytes_transferred();
232 break;
233 case MIG_STATE_ERROR:
234 info->has_status = true;
235 info->status = g_strdup("failed");
236 break;
237 case MIG_STATE_CANCELLED:
238 info->has_status = true;
239 info->status = g_strdup("cancelled");
240 break;
241 }
242
243 return info;
244 }
245
246 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
247 Error **errp)
248 {
249 MigrationState *s = migrate_get_current();
250 MigrationCapabilityStatusList *cap;
251
252 if (s->state == MIG_STATE_ACTIVE) {
253 error_set(errp, QERR_MIGRATION_ACTIVE);
254 return;
255 }
256
257 for (cap = params; cap; cap = cap->next) {
258 s->enabled_capabilities[cap->value->capability] = cap->value->state;
259 }
260 }
261
262 /* shared migration helpers */
263
264 static void migrate_fd_cleanup(void *opaque)
265 {
266 MigrationState *s = opaque;
267
268 qemu_bh_delete(s->cleanup_bh);
269 s->cleanup_bh = NULL;
270
271 if (s->file) {
272 DPRINTF("closing file\n");
273 qemu_fclose(s->file);
274 s->file = NULL;
275 }
276
277 assert(s->fd == -1);
278 assert(s->state != MIG_STATE_ACTIVE);
279
280 if (s->state != MIG_STATE_COMPLETED) {
281 qemu_savevm_state_cancel();
282 }
283
284 notifier_list_notify(&migration_state_notifiers, s);
285 }
286
287 static void migrate_finish_set_state(MigrationState *s, int new_state)
288 {
289 if (__sync_val_compare_and_swap(&s->state, MIG_STATE_ACTIVE,
290 new_state) == new_state) {
291 trace_migrate_set_state(new_state);
292 }
293 }
294
295 void migrate_fd_error(MigrationState *s)
296 {
297 DPRINTF("setting error state\n");
298 assert(s->file == NULL);
299 s->state = MIG_STATE_ERROR;
300 trace_migrate_set_state(MIG_STATE_ERROR);
301 notifier_list_notify(&migration_state_notifiers, s);
302 }
303
304 static ssize_t migrate_fd_put_buffer(MigrationState *s, const void *data,
305 size_t size)
306 {
307 ssize_t ret;
308
309 if (s->state != MIG_STATE_ACTIVE) {
310 return -EIO;
311 }
312
313 do {
314 ret = s->write(s, data, size);
315 } while (ret == -1 && ((s->get_error(s)) == EINTR));
316
317 if (ret == -1)
318 ret = -(s->get_error(s));
319
320 return ret;
321 }
322
323 static void migrate_fd_cancel(MigrationState *s)
324 {
325 DPRINTF("cancelling migration\n");
326
327 migrate_finish_set_state(s, MIG_STATE_CANCELLED);
328 }
329
330 int migrate_fd_close(MigrationState *s)
331 {
332 int rc = 0;
333 if (s->fd != -1) {
334 rc = s->close(s);
335 s->fd = -1;
336 }
337 return rc;
338 }
339
340 void add_migration_state_change_notifier(Notifier *notify)
341 {
342 notifier_list_add(&migration_state_notifiers, notify);
343 }
344
345 void remove_migration_state_change_notifier(Notifier *notify)
346 {
347 notifier_remove(notify);
348 }
349
350 bool migration_is_active(MigrationState *s)
351 {
352 return s->state == MIG_STATE_ACTIVE;
353 }
354
355 bool migration_has_finished(MigrationState *s)
356 {
357 return s->state == MIG_STATE_COMPLETED;
358 }
359
360 bool migration_has_failed(MigrationState *s)
361 {
362 return (s->state == MIG_STATE_CANCELLED ||
363 s->state == MIG_STATE_ERROR);
364 }
365
366 static MigrationState *migrate_init(const MigrationParams *params)
367 {
368 MigrationState *s = migrate_get_current();
369 int64_t bandwidth_limit = s->bandwidth_limit;
370 bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
371 int64_t xbzrle_cache_size = s->xbzrle_cache_size;
372
373 memcpy(enabled_capabilities, s->enabled_capabilities,
374 sizeof(enabled_capabilities));
375
376 memset(s, 0, sizeof(*s));
377 s->bandwidth_limit = bandwidth_limit;
378 s->params = *params;
379 memcpy(s->enabled_capabilities, enabled_capabilities,
380 sizeof(enabled_capabilities));
381 s->xbzrle_cache_size = xbzrle_cache_size;
382
383 s->bandwidth_limit = bandwidth_limit;
384 s->state = MIG_STATE_SETUP;
385 trace_migrate_set_state(MIG_STATE_SETUP);
386
387 s->total_time = qemu_get_clock_ms(rt_clock);
388 return s;
389 }
390
391 static GSList *migration_blockers;
392
393 void migrate_add_blocker(Error *reason)
394 {
395 migration_blockers = g_slist_prepend(migration_blockers, reason);
396 }
397
398 void migrate_del_blocker(Error *reason)
399 {
400 migration_blockers = g_slist_remove(migration_blockers, reason);
401 }
402
403 void qmp_migrate(const char *uri, bool has_blk, bool blk,
404 bool has_inc, bool inc, bool has_detach, bool detach,
405 Error **errp)
406 {
407 Error *local_err = NULL;
408 MigrationState *s = migrate_get_current();
409 MigrationParams params;
410 const char *p;
411
412 params.blk = blk;
413 params.shared = inc;
414
415 if (s->state == MIG_STATE_ACTIVE) {
416 error_set(errp, QERR_MIGRATION_ACTIVE);
417 return;
418 }
419
420 if (qemu_savevm_state_blocked(errp)) {
421 return;
422 }
423
424 if (migration_blockers) {
425 *errp = error_copy(migration_blockers->data);
426 return;
427 }
428
429 s = migrate_init(&params);
430
431 if (strstart(uri, "tcp:", &p)) {
432 tcp_start_outgoing_migration(s, p, &local_err);
433 #if !defined(WIN32)
434 } else if (strstart(uri, "exec:", &p)) {
435 exec_start_outgoing_migration(s, p, &local_err);
436 } else if (strstart(uri, "unix:", &p)) {
437 unix_start_outgoing_migration(s, p, &local_err);
438 } else if (strstart(uri, "fd:", &p)) {
439 fd_start_outgoing_migration(s, p, &local_err);
440 #endif
441 } else {
442 error_set(errp, QERR_INVALID_PARAMETER_VALUE, "uri", "a valid migration protocol");
443 return;
444 }
445
446 if (local_err) {
447 migrate_fd_error(s);
448 error_propagate(errp, local_err);
449 return;
450 }
451 }
452
453 void qmp_migrate_cancel(Error **errp)
454 {
455 migrate_fd_cancel(migrate_get_current());
456 }
457
458 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
459 {
460 MigrationState *s = migrate_get_current();
461
462 /* Check for truncation */
463 if (value != (size_t)value) {
464 error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
465 "exceeding address space");
466 return;
467 }
468
469 s->xbzrle_cache_size = xbzrle_cache_resize(value);
470 }
471
472 int64_t qmp_query_migrate_cache_size(Error **errp)
473 {
474 return migrate_xbzrle_cache_size();
475 }
476
477 void qmp_migrate_set_speed(int64_t value, Error **errp)
478 {
479 MigrationState *s;
480
481 if (value < 0) {
482 value = 0;
483 }
484
485 s = migrate_get_current();
486 s->bandwidth_limit = value;
487 qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
488 }
489
490 void qmp_migrate_set_downtime(double value, Error **errp)
491 {
492 value *= 1e9;
493 value = MAX(0, MIN(UINT64_MAX, value));
494 max_downtime = (uint64_t)value;
495 }
496
497 int migrate_use_xbzrle(void)
498 {
499 MigrationState *s;
500
501 s = migrate_get_current();
502
503 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
504 }
505
506 int64_t migrate_xbzrle_cache_size(void)
507 {
508 MigrationState *s;
509
510 s = migrate_get_current();
511
512 return s->xbzrle_cache_size;
513 }
514
515 /* migration thread support */
516
517 static int migration_put_buffer(void *opaque, const uint8_t *buf,
518 int64_t pos, int size)
519 {
520 MigrationState *s = opaque;
521 ssize_t ret;
522 size_t sent;
523
524 DPRINTF("putting %d bytes at %" PRId64 "\n", size, pos);
525
526 ret = qemu_file_get_error(s->file);
527 if (ret) {
528 DPRINTF("flush when error, bailing: %s\n", strerror(-ret));
529 return ret;
530 }
531
532 if (size <= 0) {
533 return size;
534 }
535
536 sent = 0;
537 while (size) {
538 ret = migrate_fd_put_buffer(s, buf, size);
539 if (ret <= 0) {
540 DPRINTF("error flushing data, %zd\n", ret);
541 return ret;
542 } else {
543 DPRINTF("flushed %zd byte(s)\n", ret);
544 sent += ret;
545 buf += ret;
546 size -= ret;
547 s->bytes_xfer += ret;
548 }
549 }
550
551 return sent;
552 }
553
554 static int migration_close(void *opaque)
555 {
556 MigrationState *s = opaque;
557
558 DPRINTF("closing\n");
559
560 qemu_mutex_unlock_iothread();
561 qemu_thread_join(&s->thread);
562 qemu_mutex_lock_iothread();
563 assert(s->state != MIG_STATE_ACTIVE);
564
565 return migrate_fd_close(s);
566 }
567
568 static int migration_get_fd(void *opaque)
569 {
570 MigrationState *s = opaque;
571
572 return s->fd;
573 }
574
575 /*
576 * The meaning of the return values is:
577 * 0: We can continue sending
578 * 1: Time to stop
579 * negative: There has been an error
580 */
581 static int migration_rate_limit(void *opaque)
582 {
583 MigrationState *s = opaque;
584 int ret;
585
586 ret = qemu_file_get_error(s->file);
587 if (ret) {
588 return ret;
589 }
590
591 if (s->bytes_xfer >= s->xfer_limit) {
592 return 1;
593 }
594
595 return 0;
596 }
597
598 static int64_t migration_set_rate_limit(void *opaque, int64_t new_rate)
599 {
600 MigrationState *s = opaque;
601 if (qemu_file_get_error(s->file)) {
602 goto out;
603 }
604 if (new_rate > SIZE_MAX) {
605 new_rate = SIZE_MAX;
606 }
607
608 s->xfer_limit = new_rate / XFER_LIMIT_RATIO;
609
610 out:
611 return s->xfer_limit;
612 }
613
614 static int64_t migration_get_rate_limit(void *opaque)
615 {
616 MigrationState *s = opaque;
617
618 return s->xfer_limit;
619 }
620
621 static void *migration_thread(void *opaque)
622 {
623 MigrationState *s = opaque;
624 int64_t initial_time = qemu_get_clock_ms(rt_clock);
625 int64_t sleep_time = 0;
626 int64_t max_size = 0;
627 int64_t start_time = initial_time;
628 bool old_vm_running = false;
629 bool last_round = false;
630
631 DPRINTF("beginning savevm\n");
632 qemu_savevm_state_begin(s->file, &s->params);
633
634 while (s->state == MIG_STATE_ACTIVE) {
635 int64_t current_time;
636 uint64_t pending_size;
637
638 if (s->bytes_xfer < s->xfer_limit) {
639 DPRINTF("iterate\n");
640 pending_size = qemu_savevm_state_pending(s->file, max_size);
641 DPRINTF("pending size %lu max %lu\n", pending_size, max_size);
642 if (pending_size && pending_size >= max_size) {
643 qemu_savevm_state_iterate(s->file);
644 } else {
645 DPRINTF("done iterating\n");
646 qemu_mutex_lock_iothread();
647 start_time = qemu_get_clock_ms(rt_clock);
648 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
649 old_vm_running = runstate_is_running();
650 vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
651 s->xfer_limit = INT_MAX;
652 qemu_savevm_state_complete(s->file);
653 last_round = true;
654 qemu_mutex_unlock_iothread();
655 }
656 }
657
658 current_time = qemu_get_clock_ms(rt_clock);
659 if (current_time >= initial_time + BUFFER_DELAY) {
660 uint64_t transferred_bytes = s->bytes_xfer;
661 uint64_t time_spent = current_time - initial_time - sleep_time;
662 double bandwidth = transferred_bytes / time_spent;
663 max_size = bandwidth * migrate_max_downtime() / 1000000;
664
665 DPRINTF("transferred %" PRIu64 " time_spent %" PRIu64
666 " bandwidth %g max_size %" PRId64 "\n",
667 transferred_bytes, time_spent, bandwidth, max_size);
668 /* if we haven't sent anything, we don't want to recalculate
669 10000 is a small enough number for our purposes */
670 if (s->dirty_bytes_rate && transferred_bytes > 10000) {
671 s->expected_downtime = s->dirty_bytes_rate / bandwidth;
672 }
673
674 s->bytes_xfer = 0;
675 sleep_time = 0;
676 initial_time = current_time;
677 }
678 if (!last_round && (s->bytes_xfer >= s->xfer_limit)) {
679 /* usleep expects microseconds */
680 g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
681 sleep_time += qemu_get_clock_ms(rt_clock) - current_time;
682 }
683 if (qemu_file_get_error(s->file)) {
684 migrate_finish_set_state(s, MIG_STATE_ERROR);
685 } else if (last_round) {
686 migrate_finish_set_state(s, MIG_STATE_COMPLETED);
687 }
688 }
689
690 qemu_mutex_lock_iothread();
691 if (s->state == MIG_STATE_COMPLETED) {
692 int64_t end_time = qemu_get_clock_ms(rt_clock);
693 s->total_time = end_time - s->total_time;
694 s->downtime = end_time - start_time;
695 runstate_set(RUN_STATE_POSTMIGRATE);
696 } else {
697 if (old_vm_running) {
698 assert(last_round);
699 vm_start();
700 }
701 }
702 qemu_bh_schedule(s->cleanup_bh);
703 qemu_mutex_unlock_iothread();
704
705 return NULL;
706 }
707
708 static const QEMUFileOps migration_file_ops = {
709 .get_fd = migration_get_fd,
710 .put_buffer = migration_put_buffer,
711 .close = migration_close,
712 .rate_limit = migration_rate_limit,
713 .get_rate_limit = migration_get_rate_limit,
714 .set_rate_limit = migration_set_rate_limit,
715 };
716
717 void migrate_fd_connect(MigrationState *s)
718 {
719 s->state = MIG_STATE_ACTIVE;
720 trace_migrate_set_state(MIG_STATE_ACTIVE);
721
722 s->bytes_xfer = 0;
723 /* This is a best 1st approximation. ns to ms */
724 s->expected_downtime = max_downtime/1000000;
725
726 s->xfer_limit = s->bandwidth_limit / XFER_LIMIT_RATIO;
727
728 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
729 s->file = qemu_fopen_ops(s, &migration_file_ops);
730
731 qemu_thread_create(&s->thread, migration_thread, s,
732 QEMU_THREAD_JOINABLE);
733 notifier_list_notify(&migration_state_notifiers, s);
734 }