glib: Add compat wrapper for g_poll on old glib
[qemu/pbrook.git] / migration.c
blob11725ae3fc00ca95be6f19d484ec847eb939e8ba
1 /*
2 * QEMU live migration
4 * Copyright IBM, Corp. 2008
6 * Authors:
7 * Anthony Liguori <aliguori@us.ibm.com>
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
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.
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"
27 //#define DEBUG_MIGRATION
29 #ifdef DEBUG_MIGRATION
30 #define DPRINTF(fmt, ...) \
31 do { printf("migration: " fmt, ## __VA_ARGS__); } while (0)
32 #else
33 #define DPRINTF(fmt, ...) \
34 do { } while (0)
35 #endif
37 enum {
38 MIG_STATE_ERROR,
39 MIG_STATE_SETUP,
40 MIG_STATE_CANCELLED,
41 MIG_STATE_ACTIVE,
42 MIG_STATE_COMPLETED,
45 #define MAX_THROTTLE (32 << 20) /* Migration speed throttling */
47 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
48 * data. */
49 #define BUFFER_DELAY 100
50 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
52 /* Migration XBZRLE default cache size */
53 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
55 static NotifierList migration_state_notifiers =
56 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
58 /* When we add fault tolerance, we could have several
59 migrations at once. For now we don't need to add
60 dynamic creation of migration */
62 MigrationState *migrate_get_current(void)
64 static MigrationState current_migration = {
65 .state = MIG_STATE_SETUP,
66 .bandwidth_limit = MAX_THROTTLE,
67 .xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE,
70 return &current_migration;
73 void qemu_start_incoming_migration(const char *uri, Error **errp)
75 const char *p;
77 if (strstart(uri, "tcp:", &p))
78 tcp_start_incoming_migration(p, errp);
79 #if !defined(WIN32)
80 else if (strstart(uri, "exec:", &p))
81 exec_start_incoming_migration(p, errp);
82 else if (strstart(uri, "unix:", &p))
83 unix_start_incoming_migration(p, errp);
84 else if (strstart(uri, "fd:", &p))
85 fd_start_incoming_migration(p, errp);
86 #endif
87 else {
88 error_setg(errp, "unknown migration protocol: %s", uri);
92 static void process_incoming_migration_co(void *opaque)
94 QEMUFile *f = opaque;
95 int ret;
97 ret = qemu_loadvm_state(f);
98 qemu_fclose(f);
99 if (ret < 0) {
100 fprintf(stderr, "load of migration failed\n");
101 exit(0);
103 qemu_announce_self();
104 DPRINTF("successfully loaded vm state\n");
106 bdrv_clear_incoming_migration_all();
107 /* Make sure all file formats flush their mutable metadata */
108 bdrv_invalidate_cache_all();
110 if (autostart) {
111 vm_start();
112 } else {
113 runstate_set(RUN_STATE_PAUSED);
117 void process_incoming_migration(QEMUFile *f)
119 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co);
120 int fd = qemu_get_fd(f);
122 assert(fd != -1);
123 socket_set_nonblock(fd);
124 qemu_coroutine_enter(co, f);
127 /* amount of nanoseconds we are willing to wait for migration to be down.
128 * the choice of nanoseconds is because it is the maximum resolution that
129 * get_clock() can achieve. It is an internal measure. All user-visible
130 * units must be in seconds */
131 static uint64_t max_downtime = 30000000;
133 uint64_t migrate_max_downtime(void)
135 return max_downtime;
138 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
140 MigrationCapabilityStatusList *head = NULL;
141 MigrationCapabilityStatusList *caps;
142 MigrationState *s = migrate_get_current();
143 int i;
145 for (i = 0; i < MIGRATION_CAPABILITY_MAX; i++) {
146 if (head == NULL) {
147 head = g_malloc0(sizeof(*caps));
148 caps = head;
149 } else {
150 caps->next = g_malloc0(sizeof(*caps));
151 caps = caps->next;
153 caps->value =
154 g_malloc(sizeof(*caps->value));
155 caps->value->capability = i;
156 caps->value->state = s->enabled_capabilities[i];
159 return head;
162 static void get_xbzrle_cache_stats(MigrationInfo *info)
164 if (migrate_use_xbzrle()) {
165 info->has_xbzrle_cache = true;
166 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
167 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
168 info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred();
169 info->xbzrle_cache->pages = xbzrle_mig_pages_transferred();
170 info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss();
171 info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow();
175 MigrationInfo *qmp_query_migrate(Error **errp)
177 MigrationInfo *info = g_malloc0(sizeof(*info));
178 MigrationState *s = migrate_get_current();
180 switch (s->state) {
181 case MIG_STATE_SETUP:
182 /* no migration has happened ever */
183 break;
184 case MIG_STATE_ACTIVE:
185 info->has_status = true;
186 info->status = g_strdup("active");
187 info->has_total_time = true;
188 info->total_time = qemu_get_clock_ms(rt_clock)
189 - s->total_time;
190 info->has_expected_downtime = true;
191 info->expected_downtime = s->expected_downtime;
193 info->has_ram = true;
194 info->ram = g_malloc0(sizeof(*info->ram));
195 info->ram->transferred = ram_bytes_transferred();
196 info->ram->remaining = ram_bytes_remaining();
197 info->ram->total = ram_bytes_total();
198 info->ram->duplicate = dup_mig_pages_transferred();
199 info->ram->normal = norm_mig_pages_transferred();
200 info->ram->normal_bytes = norm_mig_bytes_transferred();
201 info->ram->dirty_pages_rate = s->dirty_pages_rate;
204 if (blk_mig_active()) {
205 info->has_disk = true;
206 info->disk = g_malloc0(sizeof(*info->disk));
207 info->disk->transferred = blk_mig_bytes_transferred();
208 info->disk->remaining = blk_mig_bytes_remaining();
209 info->disk->total = blk_mig_bytes_total();
212 get_xbzrle_cache_stats(info);
213 break;
214 case MIG_STATE_COMPLETED:
215 get_xbzrle_cache_stats(info);
217 info->has_status = true;
218 info->status = g_strdup("completed");
219 info->total_time = s->total_time;
220 info->has_downtime = true;
221 info->downtime = s->downtime;
223 info->has_ram = true;
224 info->ram = g_malloc0(sizeof(*info->ram));
225 info->ram->transferred = ram_bytes_transferred();
226 info->ram->remaining = 0;
227 info->ram->total = ram_bytes_total();
228 info->ram->duplicate = dup_mig_pages_transferred();
229 info->ram->normal = norm_mig_pages_transferred();
230 info->ram->normal_bytes = norm_mig_bytes_transferred();
231 break;
232 case MIG_STATE_ERROR:
233 info->has_status = true;
234 info->status = g_strdup("failed");
235 break;
236 case MIG_STATE_CANCELLED:
237 info->has_status = true;
238 info->status = g_strdup("cancelled");
239 break;
242 return info;
245 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
246 Error **errp)
248 MigrationState *s = migrate_get_current();
249 MigrationCapabilityStatusList *cap;
251 if (s->state == MIG_STATE_ACTIVE) {
252 error_set(errp, QERR_MIGRATION_ACTIVE);
253 return;
256 for (cap = params; cap; cap = cap->next) {
257 s->enabled_capabilities[cap->value->capability] = cap->value->state;
261 /* shared migration helpers */
263 static int migrate_fd_cleanup(MigrationState *s)
265 int ret = 0;
267 if (s->file) {
268 DPRINTF("closing file\n");
269 ret = qemu_fclose(s->file);
270 s->file = NULL;
273 assert(s->fd == -1);
274 return ret;
277 void migrate_fd_error(MigrationState *s)
279 DPRINTF("setting error state\n");
280 s->state = MIG_STATE_ERROR;
281 notifier_list_notify(&migration_state_notifiers, s);
282 migrate_fd_cleanup(s);
285 static void migrate_fd_completed(MigrationState *s)
287 DPRINTF("setting completed state\n");
288 if (migrate_fd_cleanup(s) < 0) {
289 s->state = MIG_STATE_ERROR;
290 } else {
291 s->state = MIG_STATE_COMPLETED;
292 runstate_set(RUN_STATE_POSTMIGRATE);
294 notifier_list_notify(&migration_state_notifiers, s);
297 static ssize_t migrate_fd_put_buffer(MigrationState *s, const void *data,
298 size_t size)
300 ssize_t ret;
302 if (s->state != MIG_STATE_ACTIVE) {
303 return -EIO;
306 do {
307 ret = s->write(s, data, size);
308 } while (ret == -1 && ((s->get_error(s)) == EINTR));
310 if (ret == -1)
311 ret = -(s->get_error(s));
313 return ret;
316 static void migrate_fd_cancel(MigrationState *s)
318 if (s->state != MIG_STATE_ACTIVE)
319 return;
321 DPRINTF("cancelling migration\n");
323 s->state = MIG_STATE_CANCELLED;
324 notifier_list_notify(&migration_state_notifiers, s);
325 qemu_savevm_state_cancel();
327 migrate_fd_cleanup(s);
330 int migrate_fd_close(MigrationState *s)
332 int rc = 0;
333 if (s->fd != -1) {
334 rc = s->close(s);
335 s->fd = -1;
337 return rc;
340 void add_migration_state_change_notifier(Notifier *notify)
342 notifier_list_add(&migration_state_notifiers, notify);
345 void remove_migration_state_change_notifier(Notifier *notify)
347 notifier_remove(notify);
350 bool migration_is_active(MigrationState *s)
352 return s->state == MIG_STATE_ACTIVE;
355 bool migration_has_finished(MigrationState *s)
357 return s->state == MIG_STATE_COMPLETED;
360 bool migration_has_failed(MigrationState *s)
362 return (s->state == MIG_STATE_CANCELLED ||
363 s->state == MIG_STATE_ERROR);
366 static MigrationState *migrate_init(const MigrationParams *params)
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;
373 memcpy(enabled_capabilities, s->enabled_capabilities,
374 sizeof(enabled_capabilities));
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;
383 s->bandwidth_limit = bandwidth_limit;
384 s->state = MIG_STATE_SETUP;
385 s->total_time = qemu_get_clock_ms(rt_clock);
387 return s;
390 static GSList *migration_blockers;
392 void migrate_add_blocker(Error *reason)
394 migration_blockers = g_slist_prepend(migration_blockers, reason);
397 void migrate_del_blocker(Error *reason)
399 migration_blockers = g_slist_remove(migration_blockers, reason);
402 void qmp_migrate(const char *uri, bool has_blk, bool blk,
403 bool has_inc, bool inc, bool has_detach, bool detach,
404 Error **errp)
406 Error *local_err = NULL;
407 MigrationState *s = migrate_get_current();
408 MigrationParams params;
409 const char *p;
411 params.blk = blk;
412 params.shared = inc;
414 if (s->state == MIG_STATE_ACTIVE) {
415 error_set(errp, QERR_MIGRATION_ACTIVE);
416 return;
419 if (qemu_savevm_state_blocked(errp)) {
420 return;
423 if (migration_blockers) {
424 *errp = error_copy(migration_blockers->data);
425 return;
428 s = migrate_init(&params);
430 if (strstart(uri, "tcp:", &p)) {
431 tcp_start_outgoing_migration(s, p, &local_err);
432 #if !defined(WIN32)
433 } else if (strstart(uri, "exec:", &p)) {
434 exec_start_outgoing_migration(s, p, &local_err);
435 } else if (strstart(uri, "unix:", &p)) {
436 unix_start_outgoing_migration(s, p, &local_err);
437 } else if (strstart(uri, "fd:", &p)) {
438 fd_start_outgoing_migration(s, p, &local_err);
439 #endif
440 } else {
441 error_set(errp, QERR_INVALID_PARAMETER_VALUE, "uri", "a valid migration protocol");
442 return;
445 if (local_err) {
446 migrate_fd_error(s);
447 error_propagate(errp, local_err);
448 return;
452 void qmp_migrate_cancel(Error **errp)
454 migrate_fd_cancel(migrate_get_current());
457 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
459 MigrationState *s = migrate_get_current();
461 /* Check for truncation */
462 if (value != (size_t)value) {
463 error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
464 "exceeding address space");
465 return;
468 s->xbzrle_cache_size = xbzrle_cache_resize(value);
471 int64_t qmp_query_migrate_cache_size(Error **errp)
473 return migrate_xbzrle_cache_size();
476 void qmp_migrate_set_speed(int64_t value, Error **errp)
478 MigrationState *s;
480 if (value < 0) {
481 value = 0;
484 s = migrate_get_current();
485 s->bandwidth_limit = value;
486 qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
489 void qmp_migrate_set_downtime(double value, Error **errp)
491 value *= 1e9;
492 value = MAX(0, MIN(UINT64_MAX, value));
493 max_downtime = (uint64_t)value;
496 int migrate_use_xbzrle(void)
498 MigrationState *s;
500 s = migrate_get_current();
502 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
505 int64_t migrate_xbzrle_cache_size(void)
507 MigrationState *s;
509 s = migrate_get_current();
511 return s->xbzrle_cache_size;
514 /* migration thread support */
517 static ssize_t buffered_flush(MigrationState *s)
519 size_t offset = 0;
520 ssize_t ret = 0;
522 DPRINTF("flushing %zu byte(s) of data\n", s->buffer_size);
524 while (s->bytes_xfer < s->xfer_limit && offset < s->buffer_size) {
525 size_t to_send = MIN(s->buffer_size - offset, s->xfer_limit - s->bytes_xfer);
526 ret = migrate_fd_put_buffer(s, s->buffer + offset, to_send);
527 if (ret <= 0) {
528 DPRINTF("error flushing data, %zd\n", ret);
529 break;
530 } else {
531 DPRINTF("flushed %zd byte(s)\n", ret);
532 offset += ret;
533 s->bytes_xfer += ret;
537 DPRINTF("flushed %zu of %zu byte(s)\n", offset, s->buffer_size);
538 memmove(s->buffer, s->buffer + offset, s->buffer_size - offset);
539 s->buffer_size -= offset;
541 if (ret < 0) {
542 return ret;
544 return offset;
547 static int buffered_put_buffer(void *opaque, const uint8_t *buf,
548 int64_t pos, int size)
550 MigrationState *s = opaque;
551 ssize_t error;
553 DPRINTF("putting %d bytes at %" PRId64 "\n", size, pos);
555 error = qemu_file_get_error(s->file);
556 if (error) {
557 DPRINTF("flush when error, bailing: %s\n", strerror(-error));
558 return error;
561 if (size <= 0) {
562 return size;
565 if (size > (s->buffer_capacity - s->buffer_size)) {
566 DPRINTF("increasing buffer capacity from %zu by %zu\n",
567 s->buffer_capacity, size + 1024);
569 s->buffer_capacity += size + 1024;
571 s->buffer = g_realloc(s->buffer, s->buffer_capacity);
574 memcpy(s->buffer + s->buffer_size, buf, size);
575 s->buffer_size += size;
577 return size;
580 static int buffered_close(void *opaque)
582 MigrationState *s = opaque;
583 ssize_t ret = 0;
584 int ret2;
586 DPRINTF("closing\n");
588 s->xfer_limit = INT_MAX;
589 while (!qemu_file_get_error(s->file) && s->buffer_size) {
590 ret = buffered_flush(s);
591 if (ret < 0) {
592 break;
596 ret2 = migrate_fd_close(s);
597 if (ret >= 0) {
598 ret = ret2;
600 s->complete = true;
601 return ret;
604 static int buffered_get_fd(void *opaque)
606 MigrationState *s = opaque;
608 return s->fd;
612 * The meaning of the return values is:
613 * 0: We can continue sending
614 * 1: Time to stop
615 * negative: There has been an error
617 static int buffered_rate_limit(void *opaque)
619 MigrationState *s = opaque;
620 int ret;
622 ret = qemu_file_get_error(s->file);
623 if (ret) {
624 return ret;
627 if (s->bytes_xfer >= s->xfer_limit) {
628 return 1;
631 return 0;
634 static int64_t buffered_set_rate_limit(void *opaque, int64_t new_rate)
636 MigrationState *s = opaque;
637 if (qemu_file_get_error(s->file)) {
638 goto out;
640 if (new_rate > SIZE_MAX) {
641 new_rate = SIZE_MAX;
644 s->xfer_limit = new_rate / XFER_LIMIT_RATIO;
646 out:
647 return s->xfer_limit;
650 static int64_t buffered_get_rate_limit(void *opaque)
652 MigrationState *s = opaque;
654 return s->xfer_limit;
657 static void *buffered_file_thread(void *opaque)
659 MigrationState *s = opaque;
660 int64_t initial_time = qemu_get_clock_ms(rt_clock);
661 int64_t sleep_time = 0;
662 int64_t max_size = 0;
663 bool last_round = false;
664 int ret;
666 qemu_mutex_lock_iothread();
667 DPRINTF("beginning savevm\n");
668 ret = qemu_savevm_state_begin(s->file, &s->params);
669 if (ret < 0) {
670 DPRINTF("failed, %d\n", ret);
671 qemu_mutex_unlock_iothread();
672 goto out;
674 qemu_mutex_unlock_iothread();
676 while (true) {
677 int64_t current_time;
678 uint64_t pending_size;
680 qemu_mutex_lock_iothread();
681 if (s->state != MIG_STATE_ACTIVE) {
682 DPRINTF("put_ready returning because of non-active state\n");
683 qemu_mutex_unlock_iothread();
684 break;
686 if (s->complete) {
687 qemu_mutex_unlock_iothread();
688 break;
690 if (s->bytes_xfer < s->xfer_limit) {
691 DPRINTF("iterate\n");
692 pending_size = qemu_savevm_state_pending(s->file, max_size);
693 DPRINTF("pending size %lu max %lu\n", pending_size, max_size);
694 if (pending_size && pending_size >= max_size) {
695 ret = qemu_savevm_state_iterate(s->file);
696 if (ret < 0) {
697 qemu_mutex_unlock_iothread();
698 break;
700 } else {
701 int old_vm_running = runstate_is_running();
702 int64_t start_time, end_time;
704 DPRINTF("done iterating\n");
705 start_time = qemu_get_clock_ms(rt_clock);
706 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
707 if (old_vm_running) {
708 vm_stop(RUN_STATE_FINISH_MIGRATE);
709 } else {
710 vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
712 ret = qemu_savevm_state_complete(s->file);
713 if (ret < 0) {
714 qemu_mutex_unlock_iothread();
715 break;
716 } else {
717 migrate_fd_completed(s);
719 end_time = qemu_get_clock_ms(rt_clock);
720 s->total_time = end_time - s->total_time;
721 s->downtime = end_time - start_time;
722 if (s->state != MIG_STATE_COMPLETED) {
723 if (old_vm_running) {
724 vm_start();
727 last_round = true;
730 qemu_mutex_unlock_iothread();
731 current_time = qemu_get_clock_ms(rt_clock);
732 if (current_time >= initial_time + BUFFER_DELAY) {
733 uint64_t transferred_bytes = s->bytes_xfer;
734 uint64_t time_spent = current_time - initial_time - sleep_time;
735 double bandwidth = transferred_bytes / time_spent;
736 max_size = bandwidth * migrate_max_downtime() / 1000000;
738 DPRINTF("transferred %" PRIu64 " time_spent %" PRIu64
739 " bandwidth %g max_size %" PRId64 "\n",
740 transferred_bytes, time_spent, bandwidth, max_size);
741 /* if we haven't sent anything, we don't want to recalculate
742 10000 is a small enough number for our purposes */
743 if (s->dirty_bytes_rate && transferred_bytes > 10000) {
744 s->expected_downtime = s->dirty_bytes_rate / bandwidth;
747 s->bytes_xfer = 0;
748 sleep_time = 0;
749 initial_time = current_time;
751 if (!last_round && (s->bytes_xfer >= s->xfer_limit)) {
752 /* usleep expects microseconds */
753 g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
754 sleep_time += qemu_get_clock_ms(rt_clock) - current_time;
756 ret = buffered_flush(s);
757 if (ret < 0) {
758 break;
762 out:
763 if (ret < 0) {
764 migrate_fd_error(s);
766 g_free(s->buffer);
767 return NULL;
770 static const QEMUFileOps buffered_file_ops = {
771 .get_fd = buffered_get_fd,
772 .put_buffer = buffered_put_buffer,
773 .close = buffered_close,
774 .rate_limit = buffered_rate_limit,
775 .get_rate_limit = buffered_get_rate_limit,
776 .set_rate_limit = buffered_set_rate_limit,
779 void migrate_fd_connect(MigrationState *s)
781 s->state = MIG_STATE_ACTIVE;
782 s->bytes_xfer = 0;
783 s->buffer = NULL;
784 s->buffer_size = 0;
785 s->buffer_capacity = 0;
786 /* This is a best 1st approximation. ns to ms */
787 s->expected_downtime = max_downtime/1000000;
789 s->xfer_limit = s->bandwidth_limit / XFER_LIMIT_RATIO;
790 s->complete = false;
792 s->file = qemu_fopen_ops(s, &buffered_file_ops);
794 qemu_thread_create(&s->thread, buffered_file_thread, s,
795 QEMU_THREAD_DETACHED);
796 notifier_list_notify(&migration_state_notifiers, s);