xref: /openbmc/qemu/migration/migration.c (revision 9c2037d0)
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/osdep.h"
17 #include "qemu/cutils.h"
18 #include "qemu/error-report.h"
19 #include "qemu/main-loop.h"
20 #include "migration/migration.h"
21 #include "migration/qemu-file.h"
22 #include "sysemu/sysemu.h"
23 #include "block/block.h"
24 #include "qapi/qmp/qerror.h"
25 #include "qapi/util.h"
26 #include "qemu/sockets.h"
27 #include "qemu/rcu.h"
28 #include "migration/block.h"
29 #include "migration/postcopy-ram.h"
30 #include "qemu/thread.h"
31 #include "qmp-commands.h"
32 #include "trace.h"
33 #include "qapi-event.h"
34 #include "qom/cpu.h"
35 #include "exec/memory.h"
36 #include "exec/address-spaces.h"
37 #include "io/channel-buffer.h"
38 #include "io/channel-tls.h"
39 #include "migration/colo.h"
40 
41 #define MAX_THROTTLE  (32 << 20)      /* Migration transfer speed throttling */
42 
43 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
44  * data. */
45 #define BUFFER_DELAY     100
46 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
47 
48 /* Time in milliseconds we are allowed to stop the source,
49  * for sending the last part */
50 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
51 
52 /* Default compression thread count */
53 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
54 /* Default decompression thread count, usually decompression is at
55  * least 4 times as fast as compression.*/
56 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
57 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
58 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
59 /* Define default autoconverge cpu throttle migration parameters */
60 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
61 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
62 
63 /* Migration XBZRLE default cache size */
64 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
65 
66 /* The delay time (in ms) between two COLO checkpoints
67  * Note: Please change this default value to 10000 when we support hybrid mode.
68  */
69 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
70 
71 static NotifierList migration_state_notifiers =
72     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
73 
74 static bool deferred_incoming;
75 
76 /*
77  * Current state of incoming postcopy; note this is not part of
78  * MigrationIncomingState since it's state is used during cleanup
79  * at the end as MIS is being freed.
80  */
81 static PostcopyState incoming_postcopy_state;
82 
83 /* When we add fault tolerance, we could have several
84    migrations at once.  For now we don't need to add
85    dynamic creation of migration */
86 
87 /* For outgoing */
88 MigrationState *migrate_get_current(void)
89 {
90     static bool once;
91     static MigrationState current_migration = {
92         .state = MIGRATION_STATUS_NONE,
93         .xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE,
94         .mbps = -1,
95         .parameters = {
96             .compress_level = DEFAULT_MIGRATE_COMPRESS_LEVEL,
97             .compress_threads = DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT,
98             .decompress_threads = DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT,
99             .cpu_throttle_initial = DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL,
100             .cpu_throttle_increment = DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT,
101             .max_bandwidth = MAX_THROTTLE,
102             .downtime_limit = DEFAULT_MIGRATE_SET_DOWNTIME,
103             .x_checkpoint_delay = DEFAULT_MIGRATE_X_CHECKPOINT_DELAY,
104         },
105     };
106 
107     if (!once) {
108         qemu_mutex_init(&current_migration.src_page_req_mutex);
109         once = true;
110     }
111     return &current_migration;
112 }
113 
114 /* For incoming */
115 static MigrationIncomingState *mis_current;
116 
117 MigrationIncomingState *migration_incoming_get_current(void)
118 {
119     return mis_current;
120 }
121 
122 MigrationIncomingState *migration_incoming_state_new(QEMUFile* f)
123 {
124     mis_current = g_new0(MigrationIncomingState, 1);
125     mis_current->from_src_file = f;
126     mis_current->state = MIGRATION_STATUS_NONE;
127     QLIST_INIT(&mis_current->loadvm_handlers);
128     qemu_mutex_init(&mis_current->rp_mutex);
129     qemu_event_init(&mis_current->main_thread_load_event, false);
130 
131     return mis_current;
132 }
133 
134 void migration_incoming_state_destroy(void)
135 {
136     qemu_event_destroy(&mis_current->main_thread_load_event);
137     loadvm_free_handlers(mis_current);
138     g_free(mis_current);
139     mis_current = NULL;
140 }
141 
142 
143 typedef struct {
144     bool optional;
145     uint32_t size;
146     uint8_t runstate[100];
147     RunState state;
148     bool received;
149 } GlobalState;
150 
151 static GlobalState global_state;
152 
153 int global_state_store(void)
154 {
155     if (!runstate_store((char *)global_state.runstate,
156                         sizeof(global_state.runstate))) {
157         error_report("runstate name too big: %s", global_state.runstate);
158         trace_migrate_state_too_big();
159         return -EINVAL;
160     }
161     return 0;
162 }
163 
164 void global_state_store_running(void)
165 {
166     const char *state = RunState_lookup[RUN_STATE_RUNNING];
167     strncpy((char *)global_state.runstate,
168            state, sizeof(global_state.runstate));
169 }
170 
171 static bool global_state_received(void)
172 {
173     return global_state.received;
174 }
175 
176 static RunState global_state_get_runstate(void)
177 {
178     return global_state.state;
179 }
180 
181 void global_state_set_optional(void)
182 {
183     global_state.optional = true;
184 }
185 
186 static bool global_state_needed(void *opaque)
187 {
188     GlobalState *s = opaque;
189     char *runstate = (char *)s->runstate;
190 
191     /* If it is not optional, it is mandatory */
192 
193     if (s->optional == false) {
194         return true;
195     }
196 
197     /* If state is running or paused, it is not needed */
198 
199     if (strcmp(runstate, "running") == 0 ||
200         strcmp(runstate, "paused") == 0) {
201         return false;
202     }
203 
204     /* for any other state it is needed */
205     return true;
206 }
207 
208 static int global_state_post_load(void *opaque, int version_id)
209 {
210     GlobalState *s = opaque;
211     Error *local_err = NULL;
212     int r;
213     char *runstate = (char *)s->runstate;
214 
215     s->received = true;
216     trace_migrate_global_state_post_load(runstate);
217 
218     r = qapi_enum_parse(RunState_lookup, runstate, RUN_STATE__MAX,
219                                 -1, &local_err);
220 
221     if (r == -1) {
222         if (local_err) {
223             error_report_err(local_err);
224         }
225         return -EINVAL;
226     }
227     s->state = r;
228 
229     return 0;
230 }
231 
232 static void global_state_pre_save(void *opaque)
233 {
234     GlobalState *s = opaque;
235 
236     trace_migrate_global_state_pre_save((char *)s->runstate);
237     s->size = strlen((char *)s->runstate) + 1;
238 }
239 
240 static const VMStateDescription vmstate_globalstate = {
241     .name = "globalstate",
242     .version_id = 1,
243     .minimum_version_id = 1,
244     .post_load = global_state_post_load,
245     .pre_save = global_state_pre_save,
246     .needed = global_state_needed,
247     .fields = (VMStateField[]) {
248         VMSTATE_UINT32(size, GlobalState),
249         VMSTATE_BUFFER(runstate, GlobalState),
250         VMSTATE_END_OF_LIST()
251     },
252 };
253 
254 void register_global_state(void)
255 {
256     /* We would use it independently that we receive it */
257     strcpy((char *)&global_state.runstate, "");
258     global_state.received = false;
259     vmstate_register(NULL, 0, &vmstate_globalstate, &global_state);
260 }
261 
262 static void migrate_generate_event(int new_state)
263 {
264     if (migrate_use_events()) {
265         qapi_event_send_migration(new_state, &error_abort);
266     }
267 }
268 
269 /*
270  * Called on -incoming with a defer: uri.
271  * The migration can be started later after any parameters have been
272  * changed.
273  */
274 static void deferred_incoming_migration(Error **errp)
275 {
276     if (deferred_incoming) {
277         error_setg(errp, "Incoming migration already deferred");
278     }
279     deferred_incoming = true;
280 }
281 
282 /* Request a range of pages from the source VM at the given
283  * start address.
284  *   rbname: Name of the RAMBlock to request the page in, if NULL it's the same
285  *           as the last request (a name must have been given previously)
286  *   Start: Address offset within the RB
287  *   Len: Length in bytes required - must be a multiple of pagesize
288  */
289 void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
290                                ram_addr_t start, size_t len)
291 {
292     uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
293     size_t msglen = 12; /* start + len */
294 
295     *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
296     *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
297 
298     if (rbname) {
299         int rbname_len = strlen(rbname);
300         assert(rbname_len < 256);
301 
302         bufc[msglen++] = rbname_len;
303         memcpy(bufc + msglen, rbname, rbname_len);
304         msglen += rbname_len;
305         migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc);
306     } else {
307         migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc);
308     }
309 }
310 
311 void qemu_start_incoming_migration(const char *uri, Error **errp)
312 {
313     const char *p;
314 
315     qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
316     if (!strcmp(uri, "defer")) {
317         deferred_incoming_migration(errp);
318     } else if (strstart(uri, "tcp:", &p)) {
319         tcp_start_incoming_migration(p, errp);
320 #ifdef CONFIG_RDMA
321     } else if (strstart(uri, "rdma:", &p)) {
322         rdma_start_incoming_migration(p, errp);
323 #endif
324     } else if (strstart(uri, "exec:", &p)) {
325         exec_start_incoming_migration(p, errp);
326     } else if (strstart(uri, "unix:", &p)) {
327         unix_start_incoming_migration(p, errp);
328     } else if (strstart(uri, "fd:", &p)) {
329         fd_start_incoming_migration(p, errp);
330     } else {
331         error_setg(errp, "unknown migration protocol: %s", uri);
332     }
333 }
334 
335 static void process_incoming_migration_bh(void *opaque)
336 {
337     Error *local_err = NULL;
338     MigrationIncomingState *mis = opaque;
339 
340     /* Make sure all file formats flush their mutable metadata */
341     bdrv_invalidate_cache_all(&local_err);
342     if (local_err) {
343         migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
344                           MIGRATION_STATUS_FAILED);
345         error_report_err(local_err);
346         migrate_decompress_threads_join();
347         exit(EXIT_FAILURE);
348     }
349 
350     /*
351      * This must happen after all error conditions are dealt with and
352      * we're sure the VM is going to be running on this host.
353      */
354     qemu_announce_self();
355 
356     /* If global state section was not received or we are in running
357        state, we need to obey autostart. Any other state is set with
358        runstate_set. */
359 
360     if (!global_state_received() ||
361         global_state_get_runstate() == RUN_STATE_RUNNING) {
362         if (autostart) {
363             vm_start();
364         } else {
365             runstate_set(RUN_STATE_PAUSED);
366         }
367     } else {
368         runstate_set(global_state_get_runstate());
369     }
370     migrate_decompress_threads_join();
371     /*
372      * This must happen after any state changes since as soon as an external
373      * observer sees this event they might start to prod at the VM assuming
374      * it's ready to use.
375      */
376     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
377                       MIGRATION_STATUS_COMPLETED);
378     qemu_bh_delete(mis->bh);
379     migration_incoming_state_destroy();
380 }
381 
382 static void process_incoming_migration_co(void *opaque)
383 {
384     QEMUFile *f = opaque;
385     MigrationIncomingState *mis;
386     PostcopyState ps;
387     int ret;
388 
389     mis = migration_incoming_state_new(f);
390     postcopy_state_set(POSTCOPY_INCOMING_NONE);
391     migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
392                       MIGRATION_STATUS_ACTIVE);
393     ret = qemu_loadvm_state(f);
394 
395     ps = postcopy_state_get();
396     trace_process_incoming_migration_co_end(ret, ps);
397     if (ps != POSTCOPY_INCOMING_NONE) {
398         if (ps == POSTCOPY_INCOMING_ADVISE) {
399             /*
400              * Where a migration had postcopy enabled (and thus went to advise)
401              * but managed to complete within the precopy period, we can use
402              * the normal exit.
403              */
404             postcopy_ram_incoming_cleanup(mis);
405         } else if (ret >= 0) {
406             /*
407              * Postcopy was started, cleanup should happen at the end of the
408              * postcopy thread.
409              */
410             trace_process_incoming_migration_co_postcopy_end_main();
411             return;
412         }
413         /* Else if something went wrong then just fall out of the normal exit */
414     }
415 
416     /* we get COLO info, and know if we are in COLO mode */
417     if (!ret && migration_incoming_enable_colo()) {
418         mis->migration_incoming_co = qemu_coroutine_self();
419         qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
420              colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
421         mis->have_colo_incoming_thread = true;
422         qemu_coroutine_yield();
423 
424         /* Wait checkpoint incoming thread exit before free resource */
425         qemu_thread_join(&mis->colo_incoming_thread);
426     }
427 
428     qemu_fclose(f);
429     free_xbzrle_decoded_buf();
430 
431     if (ret < 0) {
432         migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
433                           MIGRATION_STATUS_FAILED);
434         error_report("load of migration failed: %s", strerror(-ret));
435         migrate_decompress_threads_join();
436         exit(EXIT_FAILURE);
437     }
438 
439     mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
440     qemu_bh_schedule(mis->bh);
441 }
442 
443 void migration_fd_process_incoming(QEMUFile *f)
444 {
445     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, f);
446 
447     migrate_decompress_threads_create();
448     qemu_file_set_blocking(f, false);
449     qemu_coroutine_enter(co);
450 }
451 
452 
453 void migration_channel_process_incoming(MigrationState *s,
454                                         QIOChannel *ioc)
455 {
456     trace_migration_set_incoming_channel(
457         ioc, object_get_typename(OBJECT(ioc)));
458 
459     if (s->parameters.tls_creds &&
460         !object_dynamic_cast(OBJECT(ioc),
461                              TYPE_QIO_CHANNEL_TLS)) {
462         Error *local_err = NULL;
463         migration_tls_channel_process_incoming(s, ioc, &local_err);
464         if (local_err) {
465             error_report_err(local_err);
466         }
467     } else {
468         QEMUFile *f = qemu_fopen_channel_input(ioc);
469         migration_fd_process_incoming(f);
470     }
471 }
472 
473 
474 void migration_channel_connect(MigrationState *s,
475                                QIOChannel *ioc,
476                                const char *hostname)
477 {
478     trace_migration_set_outgoing_channel(
479         ioc, object_get_typename(OBJECT(ioc)), hostname);
480 
481     if (s->parameters.tls_creds &&
482         !object_dynamic_cast(OBJECT(ioc),
483                              TYPE_QIO_CHANNEL_TLS)) {
484         Error *local_err = NULL;
485         migration_tls_channel_connect(s, ioc, hostname, &local_err);
486         if (local_err) {
487             migrate_fd_error(s, local_err);
488             error_free(local_err);
489         }
490     } else {
491         QEMUFile *f = qemu_fopen_channel_output(ioc);
492 
493         s->to_dst_file = f;
494 
495         migrate_fd_connect(s);
496     }
497 }
498 
499 
500 /*
501  * Send a message on the return channel back to the source
502  * of the migration.
503  */
504 void migrate_send_rp_message(MigrationIncomingState *mis,
505                              enum mig_rp_message_type message_type,
506                              uint16_t len, void *data)
507 {
508     trace_migrate_send_rp_message((int)message_type, len);
509     qemu_mutex_lock(&mis->rp_mutex);
510     qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
511     qemu_put_be16(mis->to_src_file, len);
512     qemu_put_buffer(mis->to_src_file, data, len);
513     qemu_fflush(mis->to_src_file);
514     qemu_mutex_unlock(&mis->rp_mutex);
515 }
516 
517 /*
518  * Send a 'SHUT' message on the return channel with the given value
519  * to indicate that we've finished with the RP.  Non-0 value indicates
520  * error.
521  */
522 void migrate_send_rp_shut(MigrationIncomingState *mis,
523                           uint32_t value)
524 {
525     uint32_t buf;
526 
527     buf = cpu_to_be32(value);
528     migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
529 }
530 
531 /*
532  * Send a 'PONG' message on the return channel with the given value
533  * (normally in response to a 'PING')
534  */
535 void migrate_send_rp_pong(MigrationIncomingState *mis,
536                           uint32_t value)
537 {
538     uint32_t buf;
539 
540     buf = cpu_to_be32(value);
541     migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
542 }
543 
544 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
545 {
546     MigrationCapabilityStatusList *head = NULL;
547     MigrationCapabilityStatusList *caps;
548     MigrationState *s = migrate_get_current();
549     int i;
550 
551     caps = NULL; /* silence compiler warning */
552     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
553         if (i == MIGRATION_CAPABILITY_X_COLO && !colo_supported()) {
554             continue;
555         }
556         if (head == NULL) {
557             head = g_malloc0(sizeof(*caps));
558             caps = head;
559         } else {
560             caps->next = g_malloc0(sizeof(*caps));
561             caps = caps->next;
562         }
563         caps->value =
564             g_malloc(sizeof(*caps->value));
565         caps->value->capability = i;
566         caps->value->state = s->enabled_capabilities[i];
567     }
568 
569     return head;
570 }
571 
572 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
573 {
574     MigrationParameters *params;
575     MigrationState *s = migrate_get_current();
576 
577     params = g_malloc0(sizeof(*params));
578     params->has_compress_level = true;
579     params->compress_level = s->parameters.compress_level;
580     params->has_compress_threads = true;
581     params->compress_threads = s->parameters.compress_threads;
582     params->has_decompress_threads = true;
583     params->decompress_threads = s->parameters.decompress_threads;
584     params->has_cpu_throttle_initial = true;
585     params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
586     params->has_cpu_throttle_increment = true;
587     params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
588     params->has_tls_creds = !!s->parameters.tls_creds;
589     params->tls_creds = g_strdup(s->parameters.tls_creds);
590     params->has_tls_hostname = !!s->parameters.tls_hostname;
591     params->tls_hostname = g_strdup(s->parameters.tls_hostname);
592     params->has_max_bandwidth = true;
593     params->max_bandwidth = s->parameters.max_bandwidth;
594     params->has_downtime_limit = true;
595     params->downtime_limit = s->parameters.downtime_limit;
596     params->has_x_checkpoint_delay = true;
597     params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
598 
599     return params;
600 }
601 
602 /*
603  * Return true if we're already in the middle of a migration
604  * (i.e. any of the active or setup states)
605  */
606 static bool migration_is_setup_or_active(int state)
607 {
608     switch (state) {
609     case MIGRATION_STATUS_ACTIVE:
610     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
611     case MIGRATION_STATUS_SETUP:
612         return true;
613 
614     default:
615         return false;
616 
617     }
618 }
619 
620 static void get_xbzrle_cache_stats(MigrationInfo *info)
621 {
622     if (migrate_use_xbzrle()) {
623         info->has_xbzrle_cache = true;
624         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
625         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
626         info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred();
627         info->xbzrle_cache->pages = xbzrle_mig_pages_transferred();
628         info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss();
629         info->xbzrle_cache->cache_miss_rate = xbzrle_mig_cache_miss_rate();
630         info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow();
631     }
632 }
633 
634 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
635 {
636     info->has_ram = true;
637     info->ram = g_malloc0(sizeof(*info->ram));
638     info->ram->transferred = ram_bytes_transferred();
639     info->ram->total = ram_bytes_total();
640     info->ram->duplicate = dup_mig_pages_transferred();
641     info->ram->skipped = skipped_mig_pages_transferred();
642     info->ram->normal = norm_mig_pages_transferred();
643     info->ram->normal_bytes = norm_mig_bytes_transferred();
644     info->ram->mbps = s->mbps;
645     info->ram->dirty_sync_count = s->dirty_sync_count;
646     info->ram->postcopy_requests = s->postcopy_requests;
647 
648     if (s->state != MIGRATION_STATUS_COMPLETED) {
649         info->ram->remaining = ram_bytes_remaining();
650         info->ram->dirty_pages_rate = s->dirty_pages_rate;
651     }
652 }
653 
654 MigrationInfo *qmp_query_migrate(Error **errp)
655 {
656     MigrationInfo *info = g_malloc0(sizeof(*info));
657     MigrationState *s = migrate_get_current();
658 
659     switch (s->state) {
660     case MIGRATION_STATUS_NONE:
661         /* no migration has happened ever */
662         break;
663     case MIGRATION_STATUS_SETUP:
664         info->has_status = true;
665         info->has_total_time = false;
666         break;
667     case MIGRATION_STATUS_ACTIVE:
668     case MIGRATION_STATUS_CANCELLING:
669         info->has_status = true;
670         info->has_total_time = true;
671         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
672             - s->total_time;
673         info->has_expected_downtime = true;
674         info->expected_downtime = s->expected_downtime;
675         info->has_setup_time = true;
676         info->setup_time = s->setup_time;
677 
678         populate_ram_info(info, s);
679 
680         if (blk_mig_active()) {
681             info->has_disk = true;
682             info->disk = g_malloc0(sizeof(*info->disk));
683             info->disk->transferred = blk_mig_bytes_transferred();
684             info->disk->remaining = blk_mig_bytes_remaining();
685             info->disk->total = blk_mig_bytes_total();
686         }
687 
688         if (cpu_throttle_active()) {
689             info->has_cpu_throttle_percentage = true;
690             info->cpu_throttle_percentage = cpu_throttle_get_percentage();
691         }
692 
693         get_xbzrle_cache_stats(info);
694         break;
695     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
696         /* Mostly the same as active; TODO add some postcopy stats */
697         info->has_status = true;
698         info->has_total_time = true;
699         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
700             - s->total_time;
701         info->has_expected_downtime = true;
702         info->expected_downtime = s->expected_downtime;
703         info->has_setup_time = true;
704         info->setup_time = s->setup_time;
705 
706         populate_ram_info(info, s);
707 
708         if (blk_mig_active()) {
709             info->has_disk = true;
710             info->disk = g_malloc0(sizeof(*info->disk));
711             info->disk->transferred = blk_mig_bytes_transferred();
712             info->disk->remaining = blk_mig_bytes_remaining();
713             info->disk->total = blk_mig_bytes_total();
714         }
715 
716         get_xbzrle_cache_stats(info);
717         break;
718     case MIGRATION_STATUS_COLO:
719         info->has_status = true;
720         /* TODO: display COLO specific information (checkpoint info etc.) */
721         break;
722     case MIGRATION_STATUS_COMPLETED:
723         get_xbzrle_cache_stats(info);
724 
725         info->has_status = true;
726         info->has_total_time = true;
727         info->total_time = s->total_time;
728         info->has_downtime = true;
729         info->downtime = s->downtime;
730         info->has_setup_time = true;
731         info->setup_time = s->setup_time;
732 
733         populate_ram_info(info, s);
734         break;
735     case MIGRATION_STATUS_FAILED:
736         info->has_status = true;
737         if (s->error) {
738             info->has_error_desc = true;
739             info->error_desc = g_strdup(error_get_pretty(s->error));
740         }
741         break;
742     case MIGRATION_STATUS_CANCELLED:
743         info->has_status = true;
744         break;
745     }
746     info->status = s->state;
747 
748     return info;
749 }
750 
751 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
752                                   Error **errp)
753 {
754     MigrationState *s = migrate_get_current();
755     MigrationCapabilityStatusList *cap;
756     bool old_postcopy_cap = migrate_postcopy_ram();
757 
758     if (migration_is_setup_or_active(s->state)) {
759         error_setg(errp, QERR_MIGRATION_ACTIVE);
760         return;
761     }
762 
763     for (cap = params; cap; cap = cap->next) {
764         if (cap->value->capability == MIGRATION_CAPABILITY_X_COLO) {
765             if (!colo_supported()) {
766                 error_setg(errp, "COLO is not currently supported, please"
767                              " configure with --enable-colo option in order to"
768                              " support COLO feature");
769                 continue;
770             }
771         }
772         s->enabled_capabilities[cap->value->capability] = cap->value->state;
773     }
774 
775     if (migrate_postcopy_ram()) {
776         if (migrate_use_compression()) {
777             /* The decompression threads asynchronously write into RAM
778              * rather than use the atomic copies needed to avoid
779              * userfaulting.  It should be possible to fix the decompression
780              * threads for compatibility in future.
781              */
782             error_report("Postcopy is not currently compatible with "
783                          "compression");
784             s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] =
785                 false;
786         }
787         /* This check is reasonably expensive, so only when it's being
788          * set the first time, also it's only the destination that needs
789          * special support.
790          */
791         if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
792             !postcopy_ram_supported_by_host()) {
793             /* postcopy_ram_supported_by_host will have emitted a more
794              * detailed message
795              */
796             error_report("Postcopy is not supported");
797             s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] =
798                 false;
799         }
800     }
801 }
802 
803 void qmp_migrate_set_parameters(MigrationParameters *params, Error **errp)
804 {
805     MigrationState *s = migrate_get_current();
806 
807     if (params->has_compress_level &&
808         (params->compress_level < 0 || params->compress_level > 9)) {
809         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
810                    "is invalid, it should be in the range of 0 to 9");
811         return;
812     }
813     if (params->has_compress_threads &&
814         (params->compress_threads < 1 || params->compress_threads > 255)) {
815         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
816                    "compress_threads",
817                    "is invalid, it should be in the range of 1 to 255");
818         return;
819     }
820     if (params->has_decompress_threads &&
821         (params->decompress_threads < 1 || params->decompress_threads > 255)) {
822         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
823                    "decompress_threads",
824                    "is invalid, it should be in the range of 1 to 255");
825         return;
826     }
827     if (params->has_cpu_throttle_initial &&
828         (params->cpu_throttle_initial < 1 ||
829          params->cpu_throttle_initial > 99)) {
830         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
831                    "cpu_throttle_initial",
832                    "an integer in the range of 1 to 99");
833         return;
834     }
835     if (params->has_cpu_throttle_increment &&
836         (params->cpu_throttle_increment < 1 ||
837          params->cpu_throttle_increment > 99)) {
838         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
839                    "cpu_throttle_increment",
840                    "an integer in the range of 1 to 99");
841         return;
842     }
843     if (params->has_max_bandwidth &&
844         (params->max_bandwidth < 0 || params->max_bandwidth > SIZE_MAX)) {
845         error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
846                          " range of 0 to %zu bytes/second", SIZE_MAX);
847         return;
848     }
849     if (params->has_downtime_limit &&
850         (params->downtime_limit < 0 || params->downtime_limit > 2000000)) {
851         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
852                    "downtime_limit",
853                    "an integer in the range of 0 to 2000000 milliseconds");
854         return;
855     }
856     if (params->has_x_checkpoint_delay && (params->x_checkpoint_delay < 0)) {
857         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
858                     "x_checkpoint_delay",
859                     "is invalid, it should be positive");
860     }
861 
862     if (params->has_compress_level) {
863         s->parameters.compress_level = params->compress_level;
864     }
865     if (params->has_compress_threads) {
866         s->parameters.compress_threads = params->compress_threads;
867     }
868     if (params->has_decompress_threads) {
869         s->parameters.decompress_threads = params->decompress_threads;
870     }
871     if (params->has_cpu_throttle_initial) {
872         s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
873     }
874     if (params->has_cpu_throttle_increment) {
875         s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
876     }
877     if (params->has_tls_creds) {
878         g_free(s->parameters.tls_creds);
879         s->parameters.tls_creds = g_strdup(params->tls_creds);
880     }
881     if (params->has_tls_hostname) {
882         g_free(s->parameters.tls_hostname);
883         s->parameters.tls_hostname = g_strdup(params->tls_hostname);
884     }
885     if (params->has_max_bandwidth) {
886         s->parameters.max_bandwidth = params->max_bandwidth;
887         if (s->to_dst_file) {
888             qemu_file_set_rate_limit(s->to_dst_file,
889                                 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
890         }
891     }
892     if (params->has_downtime_limit) {
893         s->parameters.downtime_limit = params->downtime_limit;
894     }
895 
896     if (params->has_x_checkpoint_delay) {
897         s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
898     }
899 }
900 
901 
902 void qmp_migrate_start_postcopy(Error **errp)
903 {
904     MigrationState *s = migrate_get_current();
905 
906     if (!migrate_postcopy_ram()) {
907         error_setg(errp, "Enable postcopy with migrate_set_capability before"
908                          " the start of migration");
909         return;
910     }
911 
912     if (s->state == MIGRATION_STATUS_NONE) {
913         error_setg(errp, "Postcopy must be started after migration has been"
914                          " started");
915         return;
916     }
917     /*
918      * we don't error if migration has finished since that would be racy
919      * with issuing this command.
920      */
921     atomic_set(&s->start_postcopy, true);
922 }
923 
924 /* shared migration helpers */
925 
926 void migrate_set_state(int *state, int old_state, int new_state)
927 {
928     if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
929         trace_migrate_set_state(new_state);
930         migrate_generate_event(new_state);
931     }
932 }
933 
934 static void migrate_fd_cleanup(void *opaque)
935 {
936     MigrationState *s = opaque;
937 
938     qemu_bh_delete(s->cleanup_bh);
939     s->cleanup_bh = NULL;
940 
941     flush_page_queue(s);
942 
943     if (s->to_dst_file) {
944         trace_migrate_fd_cleanup();
945         qemu_mutex_unlock_iothread();
946         if (s->migration_thread_running) {
947             qemu_thread_join(&s->thread);
948             s->migration_thread_running = false;
949         }
950         qemu_mutex_lock_iothread();
951 
952         migrate_compress_threads_join();
953         qemu_fclose(s->to_dst_file);
954         s->to_dst_file = NULL;
955     }
956 
957     assert((s->state != MIGRATION_STATUS_ACTIVE) &&
958            (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
959 
960     if (s->state == MIGRATION_STATUS_CANCELLING) {
961         migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
962                           MIGRATION_STATUS_CANCELLED);
963     }
964 
965     notifier_list_notify(&migration_state_notifiers, s);
966 }
967 
968 void migrate_fd_error(MigrationState *s, const Error *error)
969 {
970     trace_migrate_fd_error(error_get_pretty(error));
971     assert(s->to_dst_file == NULL);
972     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
973                       MIGRATION_STATUS_FAILED);
974     if (!s->error) {
975         s->error = error_copy(error);
976     }
977     notifier_list_notify(&migration_state_notifiers, s);
978 }
979 
980 static void migrate_fd_cancel(MigrationState *s)
981 {
982     int old_state ;
983     QEMUFile *f = migrate_get_current()->to_dst_file;
984     trace_migrate_fd_cancel();
985 
986     if (s->rp_state.from_dst_file) {
987         /* shutdown the rp socket, so causing the rp thread to shutdown */
988         qemu_file_shutdown(s->rp_state.from_dst_file);
989     }
990 
991     do {
992         old_state = s->state;
993         if (!migration_is_setup_or_active(old_state)) {
994             break;
995         }
996         migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
997     } while (s->state != MIGRATION_STATUS_CANCELLING);
998 
999     /*
1000      * If we're unlucky the migration code might be stuck somewhere in a
1001      * send/write while the network has failed and is waiting to timeout;
1002      * if we've got shutdown(2) available then we can force it to quit.
1003      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1004      * called in a bh, so there is no race against this cancel.
1005      */
1006     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1007         qemu_file_shutdown(f);
1008     }
1009     if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1010         Error *local_err = NULL;
1011 
1012         bdrv_invalidate_cache_all(&local_err);
1013         if (local_err) {
1014             error_report_err(local_err);
1015         } else {
1016             s->block_inactive = false;
1017         }
1018     }
1019 }
1020 
1021 void add_migration_state_change_notifier(Notifier *notify)
1022 {
1023     notifier_list_add(&migration_state_notifiers, notify);
1024 }
1025 
1026 void remove_migration_state_change_notifier(Notifier *notify)
1027 {
1028     notifier_remove(notify);
1029 }
1030 
1031 bool migration_in_setup(MigrationState *s)
1032 {
1033     return s->state == MIGRATION_STATUS_SETUP;
1034 }
1035 
1036 bool migration_has_finished(MigrationState *s)
1037 {
1038     return s->state == MIGRATION_STATUS_COMPLETED;
1039 }
1040 
1041 bool migration_has_failed(MigrationState *s)
1042 {
1043     return (s->state == MIGRATION_STATUS_CANCELLED ||
1044             s->state == MIGRATION_STATUS_FAILED);
1045 }
1046 
1047 bool migration_in_postcopy(MigrationState *s)
1048 {
1049     return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1050 }
1051 
1052 bool migration_in_postcopy_after_devices(MigrationState *s)
1053 {
1054     return migration_in_postcopy(s) && s->postcopy_after_devices;
1055 }
1056 
1057 bool migration_is_idle(MigrationState *s)
1058 {
1059     if (!s) {
1060         s = migrate_get_current();
1061     }
1062 
1063     switch (s->state) {
1064     case MIGRATION_STATUS_NONE:
1065     case MIGRATION_STATUS_CANCELLED:
1066     case MIGRATION_STATUS_COMPLETED:
1067     case MIGRATION_STATUS_FAILED:
1068         return true;
1069     case MIGRATION_STATUS_SETUP:
1070     case MIGRATION_STATUS_CANCELLING:
1071     case MIGRATION_STATUS_ACTIVE:
1072     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1073     case MIGRATION_STATUS_COLO:
1074         return false;
1075     case MIGRATION_STATUS__MAX:
1076         g_assert_not_reached();
1077     }
1078 
1079     return false;
1080 }
1081 
1082 MigrationState *migrate_init(const MigrationParams *params)
1083 {
1084     MigrationState *s = migrate_get_current();
1085 
1086     /*
1087      * Reinitialise all migration state, except
1088      * parameters/capabilities that the user set, and
1089      * locks.
1090      */
1091     s->bytes_xfer = 0;
1092     s->xfer_limit = 0;
1093     s->cleanup_bh = 0;
1094     s->to_dst_file = NULL;
1095     s->state = MIGRATION_STATUS_NONE;
1096     s->params = *params;
1097     s->rp_state.from_dst_file = NULL;
1098     s->rp_state.error = false;
1099     s->mbps = 0.0;
1100     s->downtime = 0;
1101     s->expected_downtime = 0;
1102     s->dirty_pages_rate = 0;
1103     s->dirty_bytes_rate = 0;
1104     s->setup_time = 0;
1105     s->dirty_sync_count = 0;
1106     s->start_postcopy = false;
1107     s->postcopy_after_devices = false;
1108     s->postcopy_requests = 0;
1109     s->migration_thread_running = false;
1110     s->last_req_rb = NULL;
1111     error_free(s->error);
1112     s->error = NULL;
1113 
1114     migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1115 
1116     QSIMPLEQ_INIT(&s->src_page_requests);
1117 
1118     s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1119     return s;
1120 }
1121 
1122 static GSList *migration_blockers;
1123 
1124 int migrate_add_blocker(Error *reason, Error **errp)
1125 {
1126     if (only_migratable) {
1127         error_propagate(errp, error_copy(reason));
1128         error_prepend(errp, "disallowing migration blocker "
1129                           "(--only_migratable) for: ");
1130         return -EACCES;
1131     }
1132 
1133     if (migration_is_idle(NULL)) {
1134         migration_blockers = g_slist_prepend(migration_blockers, reason);
1135         return 0;
1136     }
1137 
1138     error_propagate(errp, error_copy(reason));
1139     error_prepend(errp, "disallowing migration blocker (migration in "
1140                       "progress) for: ");
1141     return -EBUSY;
1142 }
1143 
1144 void migrate_del_blocker(Error *reason)
1145 {
1146     migration_blockers = g_slist_remove(migration_blockers, reason);
1147 }
1148 
1149 void qmp_migrate_incoming(const char *uri, Error **errp)
1150 {
1151     Error *local_err = NULL;
1152     static bool once = true;
1153 
1154     if (!deferred_incoming) {
1155         error_setg(errp, "For use with '-incoming defer'");
1156         return;
1157     }
1158     if (!once) {
1159         error_setg(errp, "The incoming migration has already been started");
1160     }
1161 
1162     qemu_start_incoming_migration(uri, &local_err);
1163 
1164     if (local_err) {
1165         error_propagate(errp, local_err);
1166         return;
1167     }
1168 
1169     once = false;
1170 }
1171 
1172 bool migration_is_blocked(Error **errp)
1173 {
1174     if (qemu_savevm_state_blocked(errp)) {
1175         return true;
1176     }
1177 
1178     if (migration_blockers) {
1179         *errp = error_copy(migration_blockers->data);
1180         return true;
1181     }
1182 
1183     return false;
1184 }
1185 
1186 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1187                  bool has_inc, bool inc, bool has_detach, bool detach,
1188                  Error **errp)
1189 {
1190     Error *local_err = NULL;
1191     MigrationState *s = migrate_get_current();
1192     MigrationParams params;
1193     const char *p;
1194 
1195     params.blk = has_blk && blk;
1196     params.shared = has_inc && inc;
1197 
1198     if (migration_is_setup_or_active(s->state) ||
1199         s->state == MIGRATION_STATUS_CANCELLING ||
1200         s->state == MIGRATION_STATUS_COLO) {
1201         error_setg(errp, QERR_MIGRATION_ACTIVE);
1202         return;
1203     }
1204     if (runstate_check(RUN_STATE_INMIGRATE)) {
1205         error_setg(errp, "Guest is waiting for an incoming migration");
1206         return;
1207     }
1208 
1209     if (migration_is_blocked(errp)) {
1210         return;
1211     }
1212 
1213     s = migrate_init(&params);
1214 
1215     if (strstart(uri, "tcp:", &p)) {
1216         tcp_start_outgoing_migration(s, p, &local_err);
1217 #ifdef CONFIG_RDMA
1218     } else if (strstart(uri, "rdma:", &p)) {
1219         rdma_start_outgoing_migration(s, p, &local_err);
1220 #endif
1221     } else if (strstart(uri, "exec:", &p)) {
1222         exec_start_outgoing_migration(s, p, &local_err);
1223     } else if (strstart(uri, "unix:", &p)) {
1224         unix_start_outgoing_migration(s, p, &local_err);
1225     } else if (strstart(uri, "fd:", &p)) {
1226         fd_start_outgoing_migration(s, p, &local_err);
1227     } else {
1228         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1229                    "a valid migration protocol");
1230         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1231                           MIGRATION_STATUS_FAILED);
1232         return;
1233     }
1234 
1235     if (local_err) {
1236         migrate_fd_error(s, local_err);
1237         error_propagate(errp, local_err);
1238         return;
1239     }
1240 }
1241 
1242 void qmp_migrate_cancel(Error **errp)
1243 {
1244     migrate_fd_cancel(migrate_get_current());
1245 }
1246 
1247 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1248 {
1249     MigrationState *s = migrate_get_current();
1250     int64_t new_size;
1251 
1252     /* Check for truncation */
1253     if (value != (size_t)value) {
1254         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1255                    "exceeding address space");
1256         return;
1257     }
1258 
1259     /* Cache should not be larger than guest ram size */
1260     if (value > ram_bytes_total()) {
1261         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1262                    "exceeds guest ram size ");
1263         return;
1264     }
1265 
1266     new_size = xbzrle_cache_resize(value);
1267     if (new_size < 0) {
1268         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1269                    "is smaller than page size");
1270         return;
1271     }
1272 
1273     s->xbzrle_cache_size = new_size;
1274 }
1275 
1276 int64_t qmp_query_migrate_cache_size(Error **errp)
1277 {
1278     return migrate_xbzrle_cache_size();
1279 }
1280 
1281 void qmp_migrate_set_speed(int64_t value, Error **errp)
1282 {
1283     MigrationParameters p = {
1284         .has_max_bandwidth = true,
1285         .max_bandwidth = value,
1286     };
1287 
1288     qmp_migrate_set_parameters(&p, errp);
1289 }
1290 
1291 void qmp_migrate_set_downtime(double value, Error **errp)
1292 {
1293     value *= 1000; /* Convert to milliseconds */
1294     value = MAX(0, MIN(INT64_MAX, value));
1295 
1296     MigrationParameters p = {
1297         .has_downtime_limit = true,
1298         .downtime_limit = value,
1299     };
1300 
1301     qmp_migrate_set_parameters(&p, errp);
1302 }
1303 
1304 bool migrate_postcopy_ram(void)
1305 {
1306     MigrationState *s;
1307 
1308     s = migrate_get_current();
1309 
1310     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1311 }
1312 
1313 bool migrate_auto_converge(void)
1314 {
1315     MigrationState *s;
1316 
1317     s = migrate_get_current();
1318 
1319     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1320 }
1321 
1322 bool migrate_zero_blocks(void)
1323 {
1324     MigrationState *s;
1325 
1326     s = migrate_get_current();
1327 
1328     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1329 }
1330 
1331 bool migrate_use_compression(void)
1332 {
1333     MigrationState *s;
1334 
1335     s = migrate_get_current();
1336 
1337     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
1338 }
1339 
1340 int migrate_compress_level(void)
1341 {
1342     MigrationState *s;
1343 
1344     s = migrate_get_current();
1345 
1346     return s->parameters.compress_level;
1347 }
1348 
1349 int migrate_compress_threads(void)
1350 {
1351     MigrationState *s;
1352 
1353     s = migrate_get_current();
1354 
1355     return s->parameters.compress_threads;
1356 }
1357 
1358 int migrate_decompress_threads(void)
1359 {
1360     MigrationState *s;
1361 
1362     s = migrate_get_current();
1363 
1364     return s->parameters.decompress_threads;
1365 }
1366 
1367 bool migrate_use_events(void)
1368 {
1369     MigrationState *s;
1370 
1371     s = migrate_get_current();
1372 
1373     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1374 }
1375 
1376 int migrate_use_xbzrle(void)
1377 {
1378     MigrationState *s;
1379 
1380     s = migrate_get_current();
1381 
1382     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1383 }
1384 
1385 int64_t migrate_xbzrle_cache_size(void)
1386 {
1387     MigrationState *s;
1388 
1389     s = migrate_get_current();
1390 
1391     return s->xbzrle_cache_size;
1392 }
1393 
1394 /* migration thread support */
1395 /*
1396  * Something bad happened to the RP stream, mark an error
1397  * The caller shall print or trace something to indicate why
1398  */
1399 static void mark_source_rp_bad(MigrationState *s)
1400 {
1401     s->rp_state.error = true;
1402 }
1403 
1404 static struct rp_cmd_args {
1405     ssize_t     len; /* -1 = variable */
1406     const char *name;
1407 } rp_cmd_args[] = {
1408     [MIG_RP_MSG_INVALID]        = { .len = -1, .name = "INVALID" },
1409     [MIG_RP_MSG_SHUT]           = { .len =  4, .name = "SHUT" },
1410     [MIG_RP_MSG_PONG]           = { .len =  4, .name = "PONG" },
1411     [MIG_RP_MSG_REQ_PAGES]      = { .len = 12, .name = "REQ_PAGES" },
1412     [MIG_RP_MSG_REQ_PAGES_ID]   = { .len = -1, .name = "REQ_PAGES_ID" },
1413     [MIG_RP_MSG_MAX]            = { .len = -1, .name = "MAX" },
1414 };
1415 
1416 /*
1417  * Process a request for pages received on the return path,
1418  * We're allowed to send more than requested (e.g. to round to our page size)
1419  * and we don't need to send pages that have already been sent.
1420  */
1421 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1422                                        ram_addr_t start, size_t len)
1423 {
1424     long our_host_ps = getpagesize();
1425 
1426     trace_migrate_handle_rp_req_pages(rbname, start, len);
1427 
1428     /*
1429      * Since we currently insist on matching page sizes, just sanity check
1430      * we're being asked for whole host pages.
1431      */
1432     if (start & (our_host_ps-1) ||
1433        (len & (our_host_ps-1))) {
1434         error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1435                      " len: %zd", __func__, start, len);
1436         mark_source_rp_bad(ms);
1437         return;
1438     }
1439 
1440     if (ram_save_queue_pages(ms, rbname, start, len)) {
1441         mark_source_rp_bad(ms);
1442     }
1443 }
1444 
1445 /*
1446  * Handles messages sent on the return path towards the source VM
1447  *
1448  */
1449 static void *source_return_path_thread(void *opaque)
1450 {
1451     MigrationState *ms = opaque;
1452     QEMUFile *rp = ms->rp_state.from_dst_file;
1453     uint16_t header_len, header_type;
1454     uint8_t buf[512];
1455     uint32_t tmp32, sibling_error;
1456     ram_addr_t start = 0; /* =0 to silence warning */
1457     size_t  len = 0, expected_len;
1458     int res;
1459 
1460     trace_source_return_path_thread_entry();
1461     while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
1462            migration_is_setup_or_active(ms->state)) {
1463         trace_source_return_path_thread_loop_top();
1464         header_type = qemu_get_be16(rp);
1465         header_len = qemu_get_be16(rp);
1466 
1467         if (header_type >= MIG_RP_MSG_MAX ||
1468             header_type == MIG_RP_MSG_INVALID) {
1469             error_report("RP: Received invalid message 0x%04x length 0x%04x",
1470                     header_type, header_len);
1471             mark_source_rp_bad(ms);
1472             goto out;
1473         }
1474 
1475         if ((rp_cmd_args[header_type].len != -1 &&
1476             header_len != rp_cmd_args[header_type].len) ||
1477             header_len > sizeof(buf)) {
1478             error_report("RP: Received '%s' message (0x%04x) with"
1479                     "incorrect length %d expecting %zu",
1480                     rp_cmd_args[header_type].name, header_type, header_len,
1481                     (size_t)rp_cmd_args[header_type].len);
1482             mark_source_rp_bad(ms);
1483             goto out;
1484         }
1485 
1486         /* We know we've got a valid header by this point */
1487         res = qemu_get_buffer(rp, buf, header_len);
1488         if (res != header_len) {
1489             error_report("RP: Failed reading data for message 0x%04x"
1490                          " read %d expected %d",
1491                          header_type, res, header_len);
1492             mark_source_rp_bad(ms);
1493             goto out;
1494         }
1495 
1496         /* OK, we have the message and the data */
1497         switch (header_type) {
1498         case MIG_RP_MSG_SHUT:
1499             sibling_error = ldl_be_p(buf);
1500             trace_source_return_path_thread_shut(sibling_error);
1501             if (sibling_error) {
1502                 error_report("RP: Sibling indicated error %d", sibling_error);
1503                 mark_source_rp_bad(ms);
1504             }
1505             /*
1506              * We'll let the main thread deal with closing the RP
1507              * we could do a shutdown(2) on it, but we're the only user
1508              * anyway, so there's nothing gained.
1509              */
1510             goto out;
1511 
1512         case MIG_RP_MSG_PONG:
1513             tmp32 = ldl_be_p(buf);
1514             trace_source_return_path_thread_pong(tmp32);
1515             break;
1516 
1517         case MIG_RP_MSG_REQ_PAGES:
1518             start = ldq_be_p(buf);
1519             len = ldl_be_p(buf + 8);
1520             migrate_handle_rp_req_pages(ms, NULL, start, len);
1521             break;
1522 
1523         case MIG_RP_MSG_REQ_PAGES_ID:
1524             expected_len = 12 + 1; /* header + termination */
1525 
1526             if (header_len >= expected_len) {
1527                 start = ldq_be_p(buf);
1528                 len = ldl_be_p(buf + 8);
1529                 /* Now we expect an idstr */
1530                 tmp32 = buf[12]; /* Length of the following idstr */
1531                 buf[13 + tmp32] = '\0';
1532                 expected_len += tmp32;
1533             }
1534             if (header_len != expected_len) {
1535                 error_report("RP: Req_Page_id with length %d expecting %zd",
1536                         header_len, expected_len);
1537                 mark_source_rp_bad(ms);
1538                 goto out;
1539             }
1540             migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
1541             break;
1542 
1543         default:
1544             break;
1545         }
1546     }
1547     if (qemu_file_get_error(rp)) {
1548         trace_source_return_path_thread_bad_end();
1549         mark_source_rp_bad(ms);
1550     }
1551 
1552     trace_source_return_path_thread_end();
1553 out:
1554     ms->rp_state.from_dst_file = NULL;
1555     qemu_fclose(rp);
1556     return NULL;
1557 }
1558 
1559 static int open_return_path_on_source(MigrationState *ms)
1560 {
1561 
1562     ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
1563     if (!ms->rp_state.from_dst_file) {
1564         return -1;
1565     }
1566 
1567     trace_open_return_path_on_source();
1568     qemu_thread_create(&ms->rp_state.rp_thread, "return path",
1569                        source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
1570 
1571     trace_open_return_path_on_source_continue();
1572 
1573     return 0;
1574 }
1575 
1576 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1577 static int await_return_path_close_on_source(MigrationState *ms)
1578 {
1579     /*
1580      * If this is a normal exit then the destination will send a SHUT and the
1581      * rp_thread will exit, however if there's an error we need to cause
1582      * it to exit.
1583      */
1584     if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
1585         /*
1586          * shutdown(2), if we have it, will cause it to unblock if it's stuck
1587          * waiting for the destination.
1588          */
1589         qemu_file_shutdown(ms->rp_state.from_dst_file);
1590         mark_source_rp_bad(ms);
1591     }
1592     trace_await_return_path_close_on_source_joining();
1593     qemu_thread_join(&ms->rp_state.rp_thread);
1594     trace_await_return_path_close_on_source_close();
1595     return ms->rp_state.error;
1596 }
1597 
1598 /*
1599  * Switch from normal iteration to postcopy
1600  * Returns non-0 on error
1601  */
1602 static int postcopy_start(MigrationState *ms, bool *old_vm_running)
1603 {
1604     int ret;
1605     QIOChannelBuffer *bioc;
1606     QEMUFile *fb;
1607     int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1608     migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
1609                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
1610 
1611     trace_postcopy_start();
1612     qemu_mutex_lock_iothread();
1613     trace_postcopy_start_set_run();
1614 
1615     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1616     *old_vm_running = runstate_is_running();
1617     global_state_store();
1618     ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1619     if (ret < 0) {
1620         goto fail;
1621     }
1622 
1623     ret = bdrv_inactivate_all();
1624     if (ret < 0) {
1625         goto fail;
1626     }
1627 
1628     /*
1629      * Cause any non-postcopiable, but iterative devices to
1630      * send out their final data.
1631      */
1632     qemu_savevm_state_complete_precopy(ms->to_dst_file, true);
1633 
1634     /*
1635      * in Finish migrate and with the io-lock held everything should
1636      * be quiet, but we've potentially still got dirty pages and we
1637      * need to tell the destination to throw any pages it's already received
1638      * that are dirty
1639      */
1640     if (ram_postcopy_send_discard_bitmap(ms)) {
1641         error_report("postcopy send discard bitmap failed");
1642         goto fail;
1643     }
1644 
1645     /*
1646      * send rest of state - note things that are doing postcopy
1647      * will notice we're in POSTCOPY_ACTIVE and not actually
1648      * wrap their state up here
1649      */
1650     qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
1651     /* Ping just for debugging, helps line traces up */
1652     qemu_savevm_send_ping(ms->to_dst_file, 2);
1653 
1654     /*
1655      * While loading the device state we may trigger page transfer
1656      * requests and the fd must be free to process those, and thus
1657      * the destination must read the whole device state off the fd before
1658      * it starts processing it.  Unfortunately the ad-hoc migration format
1659      * doesn't allow the destination to know the size to read without fully
1660      * parsing it through each devices load-state code (especially the open
1661      * coded devices that use get/put).
1662      * So we wrap the device state up in a package with a length at the start;
1663      * to do this we use a qemu_buf to hold the whole of the device state.
1664      */
1665     bioc = qio_channel_buffer_new(4096);
1666     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
1667     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
1668     object_unref(OBJECT(bioc));
1669 
1670     /*
1671      * Make sure the receiver can get incoming pages before we send the rest
1672      * of the state
1673      */
1674     qemu_savevm_send_postcopy_listen(fb);
1675 
1676     qemu_savevm_state_complete_precopy(fb, false);
1677     qemu_savevm_send_ping(fb, 3);
1678 
1679     qemu_savevm_send_postcopy_run(fb);
1680 
1681     /* <><> end of stuff going into the package */
1682 
1683     /* Now send that blob */
1684     if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
1685         goto fail_closefb;
1686     }
1687     qemu_fclose(fb);
1688 
1689     /* Send a notify to give a chance for anything that needs to happen
1690      * at the transition to postcopy and after the device state; in particular
1691      * spice needs to trigger a transition now
1692      */
1693     ms->postcopy_after_devices = true;
1694     notifier_list_notify(&migration_state_notifiers, ms);
1695 
1696     ms->downtime =  qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
1697 
1698     qemu_mutex_unlock_iothread();
1699 
1700     /*
1701      * Although this ping is just for debug, it could potentially be
1702      * used for getting a better measurement of downtime at the source.
1703      */
1704     qemu_savevm_send_ping(ms->to_dst_file, 4);
1705 
1706     ret = qemu_file_get_error(ms->to_dst_file);
1707     if (ret) {
1708         error_report("postcopy_start: Migration stream errored");
1709         migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1710                               MIGRATION_STATUS_FAILED);
1711     }
1712 
1713     return ret;
1714 
1715 fail_closefb:
1716     qemu_fclose(fb);
1717 fail:
1718     migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1719                           MIGRATION_STATUS_FAILED);
1720     qemu_mutex_unlock_iothread();
1721     return -1;
1722 }
1723 
1724 /**
1725  * migration_completion: Used by migration_thread when there's not much left.
1726  *   The caller 'breaks' the loop when this returns.
1727  *
1728  * @s: Current migration state
1729  * @current_active_state: The migration state we expect to be in
1730  * @*old_vm_running: Pointer to old_vm_running flag
1731  * @*start_time: Pointer to time to update
1732  */
1733 static void migration_completion(MigrationState *s, int current_active_state,
1734                                  bool *old_vm_running,
1735                                  int64_t *start_time)
1736 {
1737     int ret;
1738 
1739     if (s->state == MIGRATION_STATUS_ACTIVE) {
1740         qemu_mutex_lock_iothread();
1741         *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1742         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1743         *old_vm_running = runstate_is_running();
1744         ret = global_state_store();
1745 
1746         if (!ret) {
1747             ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1748             /*
1749              * Don't mark the image with BDRV_O_INACTIVE flag if
1750              * we will go into COLO stage later.
1751              */
1752             if (ret >= 0 && !migrate_colo_enabled()) {
1753                 ret = bdrv_inactivate_all();
1754             }
1755             if (ret >= 0) {
1756                 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
1757                 qemu_savevm_state_complete_precopy(s->to_dst_file, false);
1758                 s->block_inactive = true;
1759             }
1760         }
1761         qemu_mutex_unlock_iothread();
1762 
1763         if (ret < 0) {
1764             goto fail;
1765         }
1766     } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1767         trace_migration_completion_postcopy_end();
1768 
1769         qemu_savevm_state_complete_postcopy(s->to_dst_file);
1770         trace_migration_completion_postcopy_end_after_complete();
1771     }
1772 
1773     /*
1774      * If rp was opened we must clean up the thread before
1775      * cleaning everything else up (since if there are no failures
1776      * it will wait for the destination to send it's status in
1777      * a SHUT command).
1778      * Postcopy opens rp if enabled (even if it's not avtivated)
1779      */
1780     if (migrate_postcopy_ram()) {
1781         int rp_error;
1782         trace_migration_completion_postcopy_end_before_rp();
1783         rp_error = await_return_path_close_on_source(s);
1784         trace_migration_completion_postcopy_end_after_rp(rp_error);
1785         if (rp_error) {
1786             goto fail_invalidate;
1787         }
1788     }
1789 
1790     if (qemu_file_get_error(s->to_dst_file)) {
1791         trace_migration_completion_file_err();
1792         goto fail_invalidate;
1793     }
1794 
1795     if (!migrate_colo_enabled()) {
1796         migrate_set_state(&s->state, current_active_state,
1797                           MIGRATION_STATUS_COMPLETED);
1798     }
1799 
1800     return;
1801 
1802 fail_invalidate:
1803     /* If not doing postcopy, vm_start() will be called: let's regain
1804      * control on images.
1805      */
1806     if (s->state == MIGRATION_STATUS_ACTIVE) {
1807         Error *local_err = NULL;
1808 
1809         qemu_mutex_lock_iothread();
1810         bdrv_invalidate_cache_all(&local_err);
1811         if (local_err) {
1812             error_report_err(local_err);
1813         } else {
1814             s->block_inactive = false;
1815         }
1816         qemu_mutex_unlock_iothread();
1817     }
1818 
1819 fail:
1820     migrate_set_state(&s->state, current_active_state,
1821                       MIGRATION_STATUS_FAILED);
1822 }
1823 
1824 bool migrate_colo_enabled(void)
1825 {
1826     MigrationState *s = migrate_get_current();
1827     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
1828 }
1829 
1830 /*
1831  * Master migration thread on the source VM.
1832  * It drives the migration and pumps the data down the outgoing channel.
1833  */
1834 static void *migration_thread(void *opaque)
1835 {
1836     MigrationState *s = opaque;
1837     /* Used by the bandwidth calcs, updated later */
1838     int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1839     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
1840     int64_t initial_bytes = 0;
1841     int64_t max_size = 0;
1842     int64_t start_time = initial_time;
1843     int64_t end_time;
1844     bool old_vm_running = false;
1845     bool entered_postcopy = false;
1846     /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
1847     enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE;
1848     bool enable_colo = migrate_colo_enabled();
1849 
1850     rcu_register_thread();
1851 
1852     qemu_savevm_state_header(s->to_dst_file);
1853 
1854     if (migrate_postcopy_ram()) {
1855         /* Now tell the dest that it should open its end so it can reply */
1856         qemu_savevm_send_open_return_path(s->to_dst_file);
1857 
1858         /* And do a ping that will make stuff easier to debug */
1859         qemu_savevm_send_ping(s->to_dst_file, 1);
1860 
1861         /*
1862          * Tell the destination that we *might* want to do postcopy later;
1863          * if the other end can't do postcopy it should fail now, nice and
1864          * early.
1865          */
1866         qemu_savevm_send_postcopy_advise(s->to_dst_file);
1867     }
1868 
1869     qemu_savevm_state_begin(s->to_dst_file, &s->params);
1870 
1871     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
1872     current_active_state = MIGRATION_STATUS_ACTIVE;
1873     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1874                       MIGRATION_STATUS_ACTIVE);
1875 
1876     trace_migration_thread_setup_complete();
1877 
1878     while (s->state == MIGRATION_STATUS_ACTIVE ||
1879            s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1880         int64_t current_time;
1881         uint64_t pending_size;
1882 
1883         if (!qemu_file_rate_limit(s->to_dst_file)) {
1884             uint64_t pend_post, pend_nonpost;
1885 
1886             qemu_savevm_state_pending(s->to_dst_file, max_size, &pend_nonpost,
1887                                       &pend_post);
1888             pending_size = pend_nonpost + pend_post;
1889             trace_migrate_pending(pending_size, max_size,
1890                                   pend_post, pend_nonpost);
1891             if (pending_size && pending_size >= max_size) {
1892                 /* Still a significant amount to transfer */
1893 
1894                 if (migrate_postcopy_ram() &&
1895                     s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE &&
1896                     pend_nonpost <= max_size &&
1897                     atomic_read(&s->start_postcopy)) {
1898 
1899                     if (!postcopy_start(s, &old_vm_running)) {
1900                         current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE;
1901                         entered_postcopy = true;
1902                     }
1903 
1904                     continue;
1905                 }
1906                 /* Just another iteration step */
1907                 qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy);
1908             } else {
1909                 trace_migration_thread_low_pending(pending_size);
1910                 migration_completion(s, current_active_state,
1911                                      &old_vm_running, &start_time);
1912                 break;
1913             }
1914         }
1915 
1916         if (qemu_file_get_error(s->to_dst_file)) {
1917             migrate_set_state(&s->state, current_active_state,
1918                               MIGRATION_STATUS_FAILED);
1919             trace_migration_thread_file_err();
1920             break;
1921         }
1922         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1923         if (current_time >= initial_time + BUFFER_DELAY) {
1924             uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) -
1925                                          initial_bytes;
1926             uint64_t time_spent = current_time - initial_time;
1927             double bandwidth = (double)transferred_bytes / time_spent;
1928             max_size = bandwidth * s->parameters.downtime_limit;
1929 
1930             s->mbps = (((double) transferred_bytes * 8.0) /
1931                     ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
1932 
1933             trace_migrate_transferred(transferred_bytes, time_spent,
1934                                       bandwidth, max_size);
1935             /* if we haven't sent anything, we don't want to recalculate
1936                10000 is a small enough number for our purposes */
1937             if (s->dirty_bytes_rate && transferred_bytes > 10000) {
1938                 s->expected_downtime = s->dirty_bytes_rate / bandwidth;
1939             }
1940 
1941             qemu_file_reset_rate_limit(s->to_dst_file);
1942             initial_time = current_time;
1943             initial_bytes = qemu_ftell(s->to_dst_file);
1944         }
1945         if (qemu_file_rate_limit(s->to_dst_file)) {
1946             /* usleep expects microseconds */
1947             g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
1948         }
1949     }
1950 
1951     trace_migration_thread_after_loop();
1952     /* If we enabled cpu throttling for auto-converge, turn it off. */
1953     cpu_throttle_stop();
1954     end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1955 
1956     qemu_mutex_lock_iothread();
1957     /*
1958      * The resource has been allocated by migration will be reused in COLO
1959      * process, so don't release them.
1960      */
1961     if (!enable_colo) {
1962         qemu_savevm_state_cleanup();
1963     }
1964     if (s->state == MIGRATION_STATUS_COMPLETED) {
1965         uint64_t transferred_bytes = qemu_ftell(s->to_dst_file);
1966         s->total_time = end_time - s->total_time;
1967         if (!entered_postcopy) {
1968             s->downtime = end_time - start_time;
1969         }
1970         if (s->total_time) {
1971             s->mbps = (((double) transferred_bytes * 8.0) /
1972                        ((double) s->total_time)) / 1000;
1973         }
1974         runstate_set(RUN_STATE_POSTMIGRATE);
1975     } else {
1976         if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) {
1977             migrate_start_colo_process(s);
1978             qemu_savevm_state_cleanup();
1979             /*
1980             * Fixme: we will run VM in COLO no matter its old running state.
1981             * After exited COLO, we will keep running.
1982             */
1983             old_vm_running = true;
1984         }
1985         if (old_vm_running && !entered_postcopy) {
1986             vm_start();
1987         } else {
1988             if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
1989                 runstate_set(RUN_STATE_POSTMIGRATE);
1990             }
1991         }
1992     }
1993     qemu_bh_schedule(s->cleanup_bh);
1994     qemu_mutex_unlock_iothread();
1995 
1996     rcu_unregister_thread();
1997     return NULL;
1998 }
1999 
2000 void migrate_fd_connect(MigrationState *s)
2001 {
2002     s->expected_downtime = s->parameters.downtime_limit;
2003     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
2004 
2005     qemu_file_set_blocking(s->to_dst_file, true);
2006     qemu_file_set_rate_limit(s->to_dst_file,
2007                              s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
2008 
2009     /* Notify before starting migration thread */
2010     notifier_list_notify(&migration_state_notifiers, s);
2011 
2012     /*
2013      * Open the return path; currently for postcopy but other things might
2014      * also want it.
2015      */
2016     if (migrate_postcopy_ram()) {
2017         if (open_return_path_on_source(s)) {
2018             error_report("Unable to open return-path for postcopy");
2019             migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2020                               MIGRATION_STATUS_FAILED);
2021             migrate_fd_cleanup(s);
2022             return;
2023         }
2024     }
2025 
2026     migrate_compress_threads_create();
2027     qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
2028                        QEMU_THREAD_JOINABLE);
2029     s->migration_thread_running = true;
2030 }
2031 
2032 PostcopyState  postcopy_state_get(void)
2033 {
2034     return atomic_mb_read(&incoming_postcopy_state);
2035 }
2036 
2037 /* Set the state and return the old state */
2038 PostcopyState postcopy_state_set(PostcopyState new_state)
2039 {
2040     return atomic_xchg(&incoming_postcopy_state, new_state);
2041 }
2042 
2043