xref: /qemu/migration/migration.c (revision bfa3ab61)
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 "qemu/error-report.h"
18 #include "qemu/main-loop.h"
19 #include "migration/migration.h"
20 #include "migration/qemu-file.h"
21 #include "sysemu/sysemu.h"
22 #include "block/block.h"
23 #include "qapi/qmp/qerror.h"
24 #include "qemu/sockets.h"
25 #include "migration/block.h"
26 #include "qemu/thread.h"
27 #include "qmp-commands.h"
28 #include "trace.h"
29 
30 #define MAX_THROTTLE  (32 << 20)      /* Migration speed throttling */
31 
32 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
33  * data. */
34 #define BUFFER_DELAY     100
35 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
36 
37 /* Default compression thread count */
38 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
39 /* Default decompression thread count, usually decompression is at
40  * least 4 times as fast as compression.*/
41 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
42 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
43 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
44 
45 /* Migration XBZRLE default cache size */
46 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
47 
48 static NotifierList migration_state_notifiers =
49     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
50 
51 static bool deferred_incoming;
52 
53 /* When we add fault tolerance, we could have several
54    migrations at once.  For now we don't need to add
55    dynamic creation of migration */
56 
57 /* For outgoing */
58 MigrationState *migrate_get_current(void)
59 {
60     static MigrationState current_migration = {
61         .state = MIGRATION_STATUS_NONE,
62         .bandwidth_limit = MAX_THROTTLE,
63         .xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE,
64         .mbps = -1,
65         .parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] =
66                 DEFAULT_MIGRATE_COMPRESS_LEVEL,
67         .parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] =
68                 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT,
69         .parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
70                 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT,
71     };
72 
73     return &current_migration;
74 }
75 
76 /* For incoming */
77 static MigrationIncomingState *mis_current;
78 
79 MigrationIncomingState *migration_incoming_get_current(void)
80 {
81     return mis_current;
82 }
83 
84 MigrationIncomingState *migration_incoming_state_new(QEMUFile* f)
85 {
86     mis_current = g_malloc0(sizeof(MigrationIncomingState));
87     mis_current->file = f;
88     QLIST_INIT(&mis_current->loadvm_handlers);
89 
90     return mis_current;
91 }
92 
93 void migration_incoming_state_destroy(void)
94 {
95     loadvm_free_handlers(mis_current);
96     g_free(mis_current);
97     mis_current = NULL;
98 }
99 
100 /*
101  * Called on -incoming with a defer: uri.
102  * The migration can be started later after any parameters have been
103  * changed.
104  */
105 static void deferred_incoming_migration(Error **errp)
106 {
107     if (deferred_incoming) {
108         error_setg(errp, "Incoming migration already deferred");
109     }
110     deferred_incoming = true;
111 }
112 
113 void qemu_start_incoming_migration(const char *uri, Error **errp)
114 {
115     const char *p;
116 
117     if (!strcmp(uri, "defer")) {
118         deferred_incoming_migration(errp);
119     } else if (strstart(uri, "tcp:", &p)) {
120         tcp_start_incoming_migration(p, errp);
121 #ifdef CONFIG_RDMA
122     } else if (strstart(uri, "rdma:", &p)) {
123         rdma_start_incoming_migration(p, errp);
124 #endif
125 #if !defined(WIN32)
126     } else if (strstart(uri, "exec:", &p)) {
127         exec_start_incoming_migration(p, errp);
128     } else if (strstart(uri, "unix:", &p)) {
129         unix_start_incoming_migration(p, errp);
130     } else if (strstart(uri, "fd:", &p)) {
131         fd_start_incoming_migration(p, errp);
132 #endif
133     } else {
134         error_setg(errp, "unknown migration protocol: %s", uri);
135     }
136 }
137 
138 static void process_incoming_migration_co(void *opaque)
139 {
140     QEMUFile *f = opaque;
141     Error *local_err = NULL;
142     int ret;
143 
144     migration_incoming_state_new(f);
145 
146     ret = qemu_loadvm_state(f);
147 
148     qemu_fclose(f);
149     free_xbzrle_decoded_buf();
150     migration_incoming_state_destroy();
151 
152     if (ret < 0) {
153         error_report("load of migration failed: %s", strerror(-ret));
154         migrate_decompress_threads_join();
155         exit(EXIT_FAILURE);
156     }
157     qemu_announce_self();
158 
159     /* Make sure all file formats flush their mutable metadata */
160     bdrv_invalidate_cache_all(&local_err);
161     if (local_err) {
162         error_report_err(local_err);
163         migrate_decompress_threads_join();
164         exit(EXIT_FAILURE);
165     }
166 
167     if (autostart) {
168         vm_start();
169     } else {
170         runstate_set(RUN_STATE_PAUSED);
171     }
172     migrate_decompress_threads_join();
173 }
174 
175 void process_incoming_migration(QEMUFile *f)
176 {
177     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co);
178     int fd = qemu_get_fd(f);
179 
180     assert(fd != -1);
181     migrate_decompress_threads_create();
182     qemu_set_nonblock(fd);
183     qemu_coroutine_enter(co, f);
184 }
185 
186 /* amount of nanoseconds we are willing to wait for migration to be down.
187  * the choice of nanoseconds is because it is the maximum resolution that
188  * get_clock() can achieve. It is an internal measure. All user-visible
189  * units must be in seconds */
190 static uint64_t max_downtime = 300000000;
191 
192 uint64_t migrate_max_downtime(void)
193 {
194     return max_downtime;
195 }
196 
197 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
198 {
199     MigrationCapabilityStatusList *head = NULL;
200     MigrationCapabilityStatusList *caps;
201     MigrationState *s = migrate_get_current();
202     int i;
203 
204     caps = NULL; /* silence compiler warning */
205     for (i = 0; i < MIGRATION_CAPABILITY_MAX; i++) {
206         if (head == NULL) {
207             head = g_malloc0(sizeof(*caps));
208             caps = head;
209         } else {
210             caps->next = g_malloc0(sizeof(*caps));
211             caps = caps->next;
212         }
213         caps->value =
214             g_malloc(sizeof(*caps->value));
215         caps->value->capability = i;
216         caps->value->state = s->enabled_capabilities[i];
217     }
218 
219     return head;
220 }
221 
222 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
223 {
224     MigrationParameters *params;
225     MigrationState *s = migrate_get_current();
226 
227     params = g_malloc0(sizeof(*params));
228     params->compress_level = s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
229     params->compress_threads =
230             s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
231     params->decompress_threads =
232             s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
233 
234     return params;
235 }
236 
237 static void get_xbzrle_cache_stats(MigrationInfo *info)
238 {
239     if (migrate_use_xbzrle()) {
240         info->has_xbzrle_cache = true;
241         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
242         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
243         info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred();
244         info->xbzrle_cache->pages = xbzrle_mig_pages_transferred();
245         info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss();
246         info->xbzrle_cache->cache_miss_rate = xbzrle_mig_cache_miss_rate();
247         info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow();
248     }
249 }
250 
251 MigrationInfo *qmp_query_migrate(Error **errp)
252 {
253     MigrationInfo *info = g_malloc0(sizeof(*info));
254     MigrationState *s = migrate_get_current();
255 
256     switch (s->state) {
257     case MIGRATION_STATUS_NONE:
258         /* no migration has happened ever */
259         break;
260     case MIGRATION_STATUS_SETUP:
261         info->has_status = true;
262         info->has_total_time = false;
263         break;
264     case MIGRATION_STATUS_ACTIVE:
265     case MIGRATION_STATUS_CANCELLING:
266         info->has_status = true;
267         info->has_total_time = true;
268         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
269             - s->total_time;
270         info->has_expected_downtime = true;
271         info->expected_downtime = s->expected_downtime;
272         info->has_setup_time = true;
273         info->setup_time = s->setup_time;
274 
275         info->has_ram = true;
276         info->ram = g_malloc0(sizeof(*info->ram));
277         info->ram->transferred = ram_bytes_transferred();
278         info->ram->remaining = ram_bytes_remaining();
279         info->ram->total = ram_bytes_total();
280         info->ram->duplicate = dup_mig_pages_transferred();
281         info->ram->skipped = skipped_mig_pages_transferred();
282         info->ram->normal = norm_mig_pages_transferred();
283         info->ram->normal_bytes = norm_mig_bytes_transferred();
284         info->ram->dirty_pages_rate = s->dirty_pages_rate;
285         info->ram->mbps = s->mbps;
286         info->ram->dirty_sync_count = s->dirty_sync_count;
287 
288         if (blk_mig_active()) {
289             info->has_disk = true;
290             info->disk = g_malloc0(sizeof(*info->disk));
291             info->disk->transferred = blk_mig_bytes_transferred();
292             info->disk->remaining = blk_mig_bytes_remaining();
293             info->disk->total = blk_mig_bytes_total();
294         }
295 
296         get_xbzrle_cache_stats(info);
297         break;
298     case MIGRATION_STATUS_COMPLETED:
299         get_xbzrle_cache_stats(info);
300 
301         info->has_status = true;
302         info->has_total_time = true;
303         info->total_time = s->total_time;
304         info->has_downtime = true;
305         info->downtime = s->downtime;
306         info->has_setup_time = true;
307         info->setup_time = s->setup_time;
308 
309         info->has_ram = true;
310         info->ram = g_malloc0(sizeof(*info->ram));
311         info->ram->transferred = ram_bytes_transferred();
312         info->ram->remaining = 0;
313         info->ram->total = ram_bytes_total();
314         info->ram->duplicate = dup_mig_pages_transferred();
315         info->ram->skipped = skipped_mig_pages_transferred();
316         info->ram->normal = norm_mig_pages_transferred();
317         info->ram->normal_bytes = norm_mig_bytes_transferred();
318         info->ram->mbps = s->mbps;
319         info->ram->dirty_sync_count = s->dirty_sync_count;
320         break;
321     case MIGRATION_STATUS_FAILED:
322         info->has_status = true;
323         break;
324     case MIGRATION_STATUS_CANCELLED:
325         info->has_status = true;
326         break;
327     }
328     info->status = s->state;
329 
330     return info;
331 }
332 
333 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
334                                   Error **errp)
335 {
336     MigrationState *s = migrate_get_current();
337     MigrationCapabilityStatusList *cap;
338 
339     if (s->state == MIGRATION_STATUS_ACTIVE ||
340         s->state == MIGRATION_STATUS_SETUP) {
341         error_setg(errp, QERR_MIGRATION_ACTIVE);
342         return;
343     }
344 
345     for (cap = params; cap; cap = cap->next) {
346         s->enabled_capabilities[cap->value->capability] = cap->value->state;
347     }
348 }
349 
350 void qmp_migrate_set_parameters(bool has_compress_level,
351                                 int64_t compress_level,
352                                 bool has_compress_threads,
353                                 int64_t compress_threads,
354                                 bool has_decompress_threads,
355                                 int64_t decompress_threads, Error **errp)
356 {
357     MigrationState *s = migrate_get_current();
358 
359     if (has_compress_level && (compress_level < 0 || compress_level > 9)) {
360         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
361                    "is invalid, it should be in the range of 0 to 9");
362         return;
363     }
364     if (has_compress_threads &&
365             (compress_threads < 1 || compress_threads > 255)) {
366         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
367                    "compress_threads",
368                    "is invalid, it should be in the range of 1 to 255");
369         return;
370     }
371     if (has_decompress_threads &&
372             (decompress_threads < 1 || decompress_threads > 255)) {
373         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
374                    "decompress_threads",
375                    "is invalid, it should be in the range of 1 to 255");
376         return;
377     }
378 
379     if (has_compress_level) {
380         s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] = compress_level;
381     }
382     if (has_compress_threads) {
383         s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] = compress_threads;
384     }
385     if (has_decompress_threads) {
386         s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
387                                                     decompress_threads;
388     }
389 }
390 
391 /* shared migration helpers */
392 
393 static void migrate_set_state(MigrationState *s, int old_state, int new_state)
394 {
395     if (atomic_cmpxchg(&s->state, old_state, new_state) == new_state) {
396         trace_migrate_set_state(new_state);
397     }
398 }
399 
400 static void migrate_fd_cleanup(void *opaque)
401 {
402     MigrationState *s = opaque;
403 
404     qemu_bh_delete(s->cleanup_bh);
405     s->cleanup_bh = NULL;
406 
407     if (s->file) {
408         trace_migrate_fd_cleanup();
409         qemu_mutex_unlock_iothread();
410         qemu_thread_join(&s->thread);
411         qemu_mutex_lock_iothread();
412 
413         migrate_compress_threads_join();
414         qemu_fclose(s->file);
415         s->file = NULL;
416     }
417 
418     assert(s->state != MIGRATION_STATUS_ACTIVE);
419 
420     if (s->state != MIGRATION_STATUS_COMPLETED) {
421         qemu_savevm_state_cancel();
422         if (s->state == MIGRATION_STATUS_CANCELLING) {
423             migrate_set_state(s, MIGRATION_STATUS_CANCELLING,
424                               MIGRATION_STATUS_CANCELLED);
425         }
426     }
427 
428     notifier_list_notify(&migration_state_notifiers, s);
429 }
430 
431 void migrate_fd_error(MigrationState *s)
432 {
433     trace_migrate_fd_error();
434     assert(s->file == NULL);
435     s->state = MIGRATION_STATUS_FAILED;
436     trace_migrate_set_state(MIGRATION_STATUS_FAILED);
437     notifier_list_notify(&migration_state_notifiers, s);
438 }
439 
440 static void migrate_fd_cancel(MigrationState *s)
441 {
442     int old_state ;
443     QEMUFile *f = migrate_get_current()->file;
444     trace_migrate_fd_cancel();
445 
446     do {
447         old_state = s->state;
448         if (old_state != MIGRATION_STATUS_SETUP &&
449             old_state != MIGRATION_STATUS_ACTIVE) {
450             break;
451         }
452         migrate_set_state(s, old_state, MIGRATION_STATUS_CANCELLING);
453     } while (s->state != MIGRATION_STATUS_CANCELLING);
454 
455     /*
456      * If we're unlucky the migration code might be stuck somewhere in a
457      * send/write while the network has failed and is waiting to timeout;
458      * if we've got shutdown(2) available then we can force it to quit.
459      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
460      * called in a bh, so there is no race against this cancel.
461      */
462     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
463         qemu_file_shutdown(f);
464     }
465 }
466 
467 void add_migration_state_change_notifier(Notifier *notify)
468 {
469     notifier_list_add(&migration_state_notifiers, notify);
470 }
471 
472 void remove_migration_state_change_notifier(Notifier *notify)
473 {
474     notifier_remove(notify);
475 }
476 
477 bool migration_in_setup(MigrationState *s)
478 {
479     return s->state == MIGRATION_STATUS_SETUP;
480 }
481 
482 bool migration_has_finished(MigrationState *s)
483 {
484     return s->state == MIGRATION_STATUS_COMPLETED;
485 }
486 
487 bool migration_has_failed(MigrationState *s)
488 {
489     return (s->state == MIGRATION_STATUS_CANCELLED ||
490             s->state == MIGRATION_STATUS_FAILED);
491 }
492 
493 static MigrationState *migrate_init(const MigrationParams *params)
494 {
495     MigrationState *s = migrate_get_current();
496     int64_t bandwidth_limit = s->bandwidth_limit;
497     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
498     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
499     int compress_level = s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
500     int compress_thread_count =
501             s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
502     int decompress_thread_count =
503             s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
504 
505     memcpy(enabled_capabilities, s->enabled_capabilities,
506            sizeof(enabled_capabilities));
507 
508     memset(s, 0, sizeof(*s));
509     s->params = *params;
510     memcpy(s->enabled_capabilities, enabled_capabilities,
511            sizeof(enabled_capabilities));
512     s->xbzrle_cache_size = xbzrle_cache_size;
513 
514     s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL] = compress_level;
515     s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS] =
516                compress_thread_count;
517     s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS] =
518                decompress_thread_count;
519     s->bandwidth_limit = bandwidth_limit;
520     s->state = MIGRATION_STATUS_SETUP;
521     trace_migrate_set_state(MIGRATION_STATUS_SETUP);
522 
523     s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
524     return s;
525 }
526 
527 static GSList *migration_blockers;
528 
529 void migrate_add_blocker(Error *reason)
530 {
531     migration_blockers = g_slist_prepend(migration_blockers, reason);
532 }
533 
534 void migrate_del_blocker(Error *reason)
535 {
536     migration_blockers = g_slist_remove(migration_blockers, reason);
537 }
538 
539 void qmp_migrate_incoming(const char *uri, Error **errp)
540 {
541     Error *local_err = NULL;
542     static bool once = true;
543 
544     if (!deferred_incoming) {
545         error_setg(errp, "For use with '-incoming defer'");
546         return;
547     }
548     if (!once) {
549         error_setg(errp, "The incoming migration has already been started");
550     }
551 
552     qemu_start_incoming_migration(uri, &local_err);
553 
554     if (local_err) {
555         error_propagate(errp, local_err);
556         return;
557     }
558 
559     once = false;
560 }
561 
562 void qmp_migrate(const char *uri, bool has_blk, bool blk,
563                  bool has_inc, bool inc, bool has_detach, bool detach,
564                  Error **errp)
565 {
566     Error *local_err = NULL;
567     MigrationState *s = migrate_get_current();
568     MigrationParams params;
569     const char *p;
570 
571     params.blk = has_blk && blk;
572     params.shared = has_inc && inc;
573 
574     if (s->state == MIGRATION_STATUS_ACTIVE ||
575         s->state == MIGRATION_STATUS_SETUP ||
576         s->state == MIGRATION_STATUS_CANCELLING) {
577         error_setg(errp, QERR_MIGRATION_ACTIVE);
578         return;
579     }
580 
581     if (runstate_check(RUN_STATE_INMIGRATE)) {
582         error_setg(errp, "Guest is waiting for an incoming migration");
583         return;
584     }
585 
586     if (qemu_savevm_state_blocked(errp)) {
587         return;
588     }
589 
590     if (migration_blockers) {
591         *errp = error_copy(migration_blockers->data);
592         return;
593     }
594 
595     s = migrate_init(&params);
596 
597     if (strstart(uri, "tcp:", &p)) {
598         tcp_start_outgoing_migration(s, p, &local_err);
599 #ifdef CONFIG_RDMA
600     } else if (strstart(uri, "rdma:", &p)) {
601         rdma_start_outgoing_migration(s, p, &local_err);
602 #endif
603 #if !defined(WIN32)
604     } else if (strstart(uri, "exec:", &p)) {
605         exec_start_outgoing_migration(s, p, &local_err);
606     } else if (strstart(uri, "unix:", &p)) {
607         unix_start_outgoing_migration(s, p, &local_err);
608     } else if (strstart(uri, "fd:", &p)) {
609         fd_start_outgoing_migration(s, p, &local_err);
610 #endif
611     } else {
612         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
613                    "a valid migration protocol");
614         s->state = MIGRATION_STATUS_FAILED;
615         return;
616     }
617 
618     if (local_err) {
619         migrate_fd_error(s);
620         error_propagate(errp, local_err);
621         return;
622     }
623 }
624 
625 void qmp_migrate_cancel(Error **errp)
626 {
627     migrate_fd_cancel(migrate_get_current());
628 }
629 
630 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
631 {
632     MigrationState *s = migrate_get_current();
633     int64_t new_size;
634 
635     /* Check for truncation */
636     if (value != (size_t)value) {
637         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
638                    "exceeding address space");
639         return;
640     }
641 
642     /* Cache should not be larger than guest ram size */
643     if (value > ram_bytes_total()) {
644         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
645                    "exceeds guest ram size ");
646         return;
647     }
648 
649     new_size = xbzrle_cache_resize(value);
650     if (new_size < 0) {
651         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
652                    "is smaller than page size");
653         return;
654     }
655 
656     s->xbzrle_cache_size = new_size;
657 }
658 
659 int64_t qmp_query_migrate_cache_size(Error **errp)
660 {
661     return migrate_xbzrle_cache_size();
662 }
663 
664 void qmp_migrate_set_speed(int64_t value, Error **errp)
665 {
666     MigrationState *s;
667 
668     if (value < 0) {
669         value = 0;
670     }
671     if (value > SIZE_MAX) {
672         value = SIZE_MAX;
673     }
674 
675     s = migrate_get_current();
676     s->bandwidth_limit = value;
677     if (s->file) {
678         qemu_file_set_rate_limit(s->file, s->bandwidth_limit / XFER_LIMIT_RATIO);
679     }
680 }
681 
682 void qmp_migrate_set_downtime(double value, Error **errp)
683 {
684     value *= 1e9;
685     value = MAX(0, MIN(UINT64_MAX, value));
686     max_downtime = (uint64_t)value;
687 }
688 
689 bool migrate_auto_converge(void)
690 {
691     MigrationState *s;
692 
693     s = migrate_get_current();
694 
695     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
696 }
697 
698 bool migrate_zero_blocks(void)
699 {
700     MigrationState *s;
701 
702     s = migrate_get_current();
703 
704     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
705 }
706 
707 bool migrate_use_compression(void)
708 {
709     MigrationState *s;
710 
711     s = migrate_get_current();
712 
713     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
714 }
715 
716 int migrate_compress_level(void)
717 {
718     MigrationState *s;
719 
720     s = migrate_get_current();
721 
722     return s->parameters[MIGRATION_PARAMETER_COMPRESS_LEVEL];
723 }
724 
725 int migrate_compress_threads(void)
726 {
727     MigrationState *s;
728 
729     s = migrate_get_current();
730 
731     return s->parameters[MIGRATION_PARAMETER_COMPRESS_THREADS];
732 }
733 
734 int migrate_decompress_threads(void)
735 {
736     MigrationState *s;
737 
738     s = migrate_get_current();
739 
740     return s->parameters[MIGRATION_PARAMETER_DECOMPRESS_THREADS];
741 }
742 
743 int migrate_use_xbzrle(void)
744 {
745     MigrationState *s;
746 
747     s = migrate_get_current();
748 
749     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
750 }
751 
752 int64_t migrate_xbzrle_cache_size(void)
753 {
754     MigrationState *s;
755 
756     s = migrate_get_current();
757 
758     return s->xbzrle_cache_size;
759 }
760 
761 /* migration thread support */
762 
763 static void *migration_thread(void *opaque)
764 {
765     MigrationState *s = opaque;
766     int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
767     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
768     int64_t initial_bytes = 0;
769     int64_t max_size = 0;
770     int64_t start_time = initial_time;
771     bool old_vm_running = false;
772 
773     qemu_savevm_state_header(s->file);
774     qemu_savevm_state_begin(s->file, &s->params);
775 
776     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
777     migrate_set_state(s, MIGRATION_STATUS_SETUP, MIGRATION_STATUS_ACTIVE);
778 
779     while (s->state == MIGRATION_STATUS_ACTIVE) {
780         int64_t current_time;
781         uint64_t pending_size;
782 
783         if (!qemu_file_rate_limit(s->file)) {
784             pending_size = qemu_savevm_state_pending(s->file, max_size);
785             trace_migrate_pending(pending_size, max_size);
786             if (pending_size && pending_size >= max_size) {
787                 qemu_savevm_state_iterate(s->file);
788             } else {
789                 int ret;
790 
791                 qemu_mutex_lock_iothread();
792                 start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
793                 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
794                 old_vm_running = runstate_is_running();
795 
796                 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
797                 if (ret >= 0) {
798                     qemu_file_set_rate_limit(s->file, INT64_MAX);
799                     qemu_savevm_state_complete(s->file);
800                 }
801                 qemu_mutex_unlock_iothread();
802 
803                 if (ret < 0) {
804                     migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
805                                       MIGRATION_STATUS_FAILED);
806                     break;
807                 }
808 
809                 if (!qemu_file_get_error(s->file)) {
810                     migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
811                                       MIGRATION_STATUS_COMPLETED);
812                     break;
813                 }
814             }
815         }
816 
817         if (qemu_file_get_error(s->file)) {
818             migrate_set_state(s, MIGRATION_STATUS_ACTIVE,
819                               MIGRATION_STATUS_FAILED);
820             break;
821         }
822         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
823         if (current_time >= initial_time + BUFFER_DELAY) {
824             uint64_t transferred_bytes = qemu_ftell(s->file) - initial_bytes;
825             uint64_t time_spent = current_time - initial_time;
826             double bandwidth = transferred_bytes / time_spent;
827             max_size = bandwidth * migrate_max_downtime() / 1000000;
828 
829             s->mbps = time_spent ? (((double) transferred_bytes * 8.0) /
830                     ((double) time_spent / 1000.0)) / 1000.0 / 1000.0 : -1;
831 
832             trace_migrate_transferred(transferred_bytes, time_spent,
833                                       bandwidth, max_size);
834             /* if we haven't sent anything, we don't want to recalculate
835                10000 is a small enough number for our purposes */
836             if (s->dirty_bytes_rate && transferred_bytes > 10000) {
837                 s->expected_downtime = s->dirty_bytes_rate / bandwidth;
838             }
839 
840             qemu_file_reset_rate_limit(s->file);
841             initial_time = current_time;
842             initial_bytes = qemu_ftell(s->file);
843         }
844         if (qemu_file_rate_limit(s->file)) {
845             /* usleep expects microseconds */
846             g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
847         }
848     }
849 
850     qemu_mutex_lock_iothread();
851     if (s->state == MIGRATION_STATUS_COMPLETED) {
852         int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
853         uint64_t transferred_bytes = qemu_ftell(s->file);
854         s->total_time = end_time - s->total_time;
855         s->downtime = end_time - start_time;
856         if (s->total_time) {
857             s->mbps = (((double) transferred_bytes * 8.0) /
858                        ((double) s->total_time)) / 1000;
859         }
860         runstate_set(RUN_STATE_POSTMIGRATE);
861     } else {
862         if (old_vm_running) {
863             vm_start();
864         }
865     }
866     qemu_bh_schedule(s->cleanup_bh);
867     qemu_mutex_unlock_iothread();
868 
869     return NULL;
870 }
871 
872 void migrate_fd_connect(MigrationState *s)
873 {
874     /* This is a best 1st approximation. ns to ms */
875     s->expected_downtime = max_downtime/1000000;
876     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
877 
878     qemu_file_set_rate_limit(s->file,
879                              s->bandwidth_limit / XFER_LIMIT_RATIO);
880 
881     /* Notify before starting migration thread */
882     notifier_list_notify(&migration_state_notifiers, s);
883 
884     migrate_compress_threads_create();
885     qemu_thread_create(&s->thread, "migration", migration_thread, s,
886                        QEMU_THREAD_JOINABLE);
887 }
888