xref: /openbmc/qemu/migration/migration.c (revision 7562f907)
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 /* Maximum migrate downtime set to 2000 seconds */
53 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
54 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
55 
56 /* Default compression thread count */
57 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
58 /* Default decompression thread count, usually decompression is at
59  * least 4 times as fast as compression.*/
60 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
61 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
62 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
63 /* Define default autoconverge cpu throttle migration parameters */
64 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
65 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
66 
67 /* Migration XBZRLE default cache size */
68 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
69 
70 /* The delay time (in ms) between two COLO checkpoints
71  * Note: Please change this default value to 10000 when we support hybrid mode.
72  */
73 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
74 
75 static NotifierList migration_state_notifiers =
76     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
77 
78 static bool deferred_incoming;
79 
80 /*
81  * Current state of incoming postcopy; note this is not part of
82  * MigrationIncomingState since it's state is used during cleanup
83  * at the end as MIS is being freed.
84  */
85 static PostcopyState incoming_postcopy_state;
86 
87 /* When we add fault tolerance, we could have several
88    migrations at once.  For now we don't need to add
89    dynamic creation of migration */
90 
91 /* For outgoing */
92 MigrationState *migrate_get_current(void)
93 {
94     static bool once;
95     static MigrationState current_migration = {
96         .state = MIGRATION_STATUS_NONE,
97         .xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE,
98         .mbps = -1,
99         .parameters = {
100             .compress_level = DEFAULT_MIGRATE_COMPRESS_LEVEL,
101             .compress_threads = DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT,
102             .decompress_threads = DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT,
103             .cpu_throttle_initial = DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL,
104             .cpu_throttle_increment = DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT,
105             .max_bandwidth = MAX_THROTTLE,
106             .downtime_limit = DEFAULT_MIGRATE_SET_DOWNTIME,
107             .x_checkpoint_delay = DEFAULT_MIGRATE_X_CHECKPOINT_DELAY,
108         },
109     };
110 
111     if (!once) {
112         qemu_mutex_init(&current_migration.src_page_req_mutex);
113         once = true;
114     }
115     return &current_migration;
116 }
117 
118 MigrationIncomingState *migration_incoming_get_current(void)
119 {
120     static bool once;
121     static MigrationIncomingState mis_current;
122 
123     if (!once) {
124         mis_current.state = MIGRATION_STATUS_NONE;
125         memset(&mis_current, 0, sizeof(MigrationIncomingState));
126         QLIST_INIT(&mis_current.loadvm_handlers);
127         qemu_mutex_init(&mis_current.rp_mutex);
128         qemu_event_init(&mis_current.main_thread_load_event, false);
129         once = true;
130     }
131     return &mis_current;
132 }
133 
134 void migration_incoming_state_destroy(void)
135 {
136     struct MigrationIncomingState *mis = migration_incoming_get_current();
137 
138     qemu_event_destroy(&mis->main_thread_load_event);
139     loadvm_free_handlers(mis);
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 = migration_incoming_get_current();
386     PostcopyState ps;
387     int ret;
388 
389     mis->from_src_file = 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 ||
851          params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
852         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
853                          "the range of 0 to %d milliseconds",
854                          MAX_MIGRATE_DOWNTIME);
855         return;
856     }
857     if (params->has_x_checkpoint_delay && (params->x_checkpoint_delay < 0)) {
858         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
859                     "x_checkpoint_delay",
860                     "is invalid, it should be positive");
861     }
862 
863     if (params->has_compress_level) {
864         s->parameters.compress_level = params->compress_level;
865     }
866     if (params->has_compress_threads) {
867         s->parameters.compress_threads = params->compress_threads;
868     }
869     if (params->has_decompress_threads) {
870         s->parameters.decompress_threads = params->decompress_threads;
871     }
872     if (params->has_cpu_throttle_initial) {
873         s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
874     }
875     if (params->has_cpu_throttle_increment) {
876         s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
877     }
878     if (params->has_tls_creds) {
879         g_free(s->parameters.tls_creds);
880         s->parameters.tls_creds = g_strdup(params->tls_creds);
881     }
882     if (params->has_tls_hostname) {
883         g_free(s->parameters.tls_hostname);
884         s->parameters.tls_hostname = g_strdup(params->tls_hostname);
885     }
886     if (params->has_max_bandwidth) {
887         s->parameters.max_bandwidth = params->max_bandwidth;
888         if (s->to_dst_file) {
889             qemu_file_set_rate_limit(s->to_dst_file,
890                                 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
891         }
892     }
893     if (params->has_downtime_limit) {
894         s->parameters.downtime_limit = params->downtime_limit;
895     }
896 
897     if (params->has_x_checkpoint_delay) {
898         s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
899         if (migration_in_colo_state()) {
900             colo_checkpoint_notify(s);
901         }
902     }
903 }
904 
905 
906 void qmp_migrate_start_postcopy(Error **errp)
907 {
908     MigrationState *s = migrate_get_current();
909 
910     if (!migrate_postcopy_ram()) {
911         error_setg(errp, "Enable postcopy with migrate_set_capability before"
912                          " the start of migration");
913         return;
914     }
915 
916     if (s->state == MIGRATION_STATUS_NONE) {
917         error_setg(errp, "Postcopy must be started after migration has been"
918                          " started");
919         return;
920     }
921     /*
922      * we don't error if migration has finished since that would be racy
923      * with issuing this command.
924      */
925     atomic_set(&s->start_postcopy, true);
926 }
927 
928 /* shared migration helpers */
929 
930 void migrate_set_state(int *state, int old_state, int new_state)
931 {
932     if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
933         trace_migrate_set_state(new_state);
934         migrate_generate_event(new_state);
935     }
936 }
937 
938 static void migrate_fd_cleanup(void *opaque)
939 {
940     MigrationState *s = opaque;
941 
942     qemu_bh_delete(s->cleanup_bh);
943     s->cleanup_bh = NULL;
944 
945     flush_page_queue(s);
946 
947     if (s->to_dst_file) {
948         trace_migrate_fd_cleanup();
949         qemu_mutex_unlock_iothread();
950         if (s->migration_thread_running) {
951             qemu_thread_join(&s->thread);
952             s->migration_thread_running = false;
953         }
954         qemu_mutex_lock_iothread();
955 
956         migrate_compress_threads_join();
957         qemu_fclose(s->to_dst_file);
958         s->to_dst_file = NULL;
959     }
960 
961     assert((s->state != MIGRATION_STATUS_ACTIVE) &&
962            (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
963 
964     if (s->state == MIGRATION_STATUS_CANCELLING) {
965         migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
966                           MIGRATION_STATUS_CANCELLED);
967     }
968 
969     notifier_list_notify(&migration_state_notifiers, s);
970 }
971 
972 void migrate_fd_error(MigrationState *s, const Error *error)
973 {
974     trace_migrate_fd_error(error_get_pretty(error));
975     assert(s->to_dst_file == NULL);
976     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
977                       MIGRATION_STATUS_FAILED);
978     if (!s->error) {
979         s->error = error_copy(error);
980     }
981     notifier_list_notify(&migration_state_notifiers, s);
982 }
983 
984 static void migrate_fd_cancel(MigrationState *s)
985 {
986     int old_state ;
987     QEMUFile *f = migrate_get_current()->to_dst_file;
988     trace_migrate_fd_cancel();
989 
990     if (s->rp_state.from_dst_file) {
991         /* shutdown the rp socket, so causing the rp thread to shutdown */
992         qemu_file_shutdown(s->rp_state.from_dst_file);
993     }
994 
995     do {
996         old_state = s->state;
997         if (!migration_is_setup_or_active(old_state)) {
998             break;
999         }
1000         migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1001     } while (s->state != MIGRATION_STATUS_CANCELLING);
1002 
1003     /*
1004      * If we're unlucky the migration code might be stuck somewhere in a
1005      * send/write while the network has failed and is waiting to timeout;
1006      * if we've got shutdown(2) available then we can force it to quit.
1007      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1008      * called in a bh, so there is no race against this cancel.
1009      */
1010     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1011         qemu_file_shutdown(f);
1012     }
1013     if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1014         Error *local_err = NULL;
1015 
1016         bdrv_invalidate_cache_all(&local_err);
1017         if (local_err) {
1018             error_report_err(local_err);
1019         } else {
1020             s->block_inactive = false;
1021         }
1022     }
1023 }
1024 
1025 void add_migration_state_change_notifier(Notifier *notify)
1026 {
1027     notifier_list_add(&migration_state_notifiers, notify);
1028 }
1029 
1030 void remove_migration_state_change_notifier(Notifier *notify)
1031 {
1032     notifier_remove(notify);
1033 }
1034 
1035 bool migration_in_setup(MigrationState *s)
1036 {
1037     return s->state == MIGRATION_STATUS_SETUP;
1038 }
1039 
1040 bool migration_has_finished(MigrationState *s)
1041 {
1042     return s->state == MIGRATION_STATUS_COMPLETED;
1043 }
1044 
1045 bool migration_has_failed(MigrationState *s)
1046 {
1047     return (s->state == MIGRATION_STATUS_CANCELLED ||
1048             s->state == MIGRATION_STATUS_FAILED);
1049 }
1050 
1051 bool migration_in_postcopy(MigrationState *s)
1052 {
1053     return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1054 }
1055 
1056 bool migration_in_postcopy_after_devices(MigrationState *s)
1057 {
1058     return migration_in_postcopy(s) && s->postcopy_after_devices;
1059 }
1060 
1061 bool migration_is_idle(MigrationState *s)
1062 {
1063     if (!s) {
1064         s = migrate_get_current();
1065     }
1066 
1067     switch (s->state) {
1068     case MIGRATION_STATUS_NONE:
1069     case MIGRATION_STATUS_CANCELLED:
1070     case MIGRATION_STATUS_COMPLETED:
1071     case MIGRATION_STATUS_FAILED:
1072         return true;
1073     case MIGRATION_STATUS_SETUP:
1074     case MIGRATION_STATUS_CANCELLING:
1075     case MIGRATION_STATUS_ACTIVE:
1076     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1077     case MIGRATION_STATUS_COLO:
1078         return false;
1079     case MIGRATION_STATUS__MAX:
1080         g_assert_not_reached();
1081     }
1082 
1083     return false;
1084 }
1085 
1086 MigrationState *migrate_init(const MigrationParams *params)
1087 {
1088     MigrationState *s = migrate_get_current();
1089 
1090     /*
1091      * Reinitialise all migration state, except
1092      * parameters/capabilities that the user set, and
1093      * locks.
1094      */
1095     s->bytes_xfer = 0;
1096     s->xfer_limit = 0;
1097     s->cleanup_bh = 0;
1098     s->to_dst_file = NULL;
1099     s->state = MIGRATION_STATUS_NONE;
1100     s->params = *params;
1101     s->rp_state.from_dst_file = NULL;
1102     s->rp_state.error = false;
1103     s->mbps = 0.0;
1104     s->downtime = 0;
1105     s->expected_downtime = 0;
1106     s->dirty_pages_rate = 0;
1107     s->dirty_bytes_rate = 0;
1108     s->setup_time = 0;
1109     s->dirty_sync_count = 0;
1110     s->start_postcopy = false;
1111     s->postcopy_after_devices = false;
1112     s->postcopy_requests = 0;
1113     s->migration_thread_running = false;
1114     s->last_req_rb = NULL;
1115     error_free(s->error);
1116     s->error = NULL;
1117 
1118     migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1119 
1120     QSIMPLEQ_INIT(&s->src_page_requests);
1121 
1122     s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1123     return s;
1124 }
1125 
1126 static GSList *migration_blockers;
1127 
1128 int migrate_add_blocker(Error *reason, Error **errp)
1129 {
1130     if (only_migratable) {
1131         error_propagate(errp, error_copy(reason));
1132         error_prepend(errp, "disallowing migration blocker "
1133                           "(--only_migratable) for: ");
1134         return -EACCES;
1135     }
1136 
1137     if (migration_is_idle(NULL)) {
1138         migration_blockers = g_slist_prepend(migration_blockers, reason);
1139         return 0;
1140     }
1141 
1142     error_propagate(errp, error_copy(reason));
1143     error_prepend(errp, "disallowing migration blocker (migration in "
1144                       "progress) for: ");
1145     return -EBUSY;
1146 }
1147 
1148 void migrate_del_blocker(Error *reason)
1149 {
1150     migration_blockers = g_slist_remove(migration_blockers, reason);
1151 }
1152 
1153 int check_migratable(Object *obj, Error **err)
1154 {
1155     DeviceClass *dc = DEVICE_GET_CLASS(obj);
1156     if (only_migratable && dc->vmsd) {
1157         if (dc->vmsd->unmigratable) {
1158             error_setg(err, "Device %s is not migratable, but "
1159                        "--only-migratable was specified",
1160                        object_get_typename(obj));
1161             return -1;
1162         }
1163     }
1164 
1165     return 0;
1166 }
1167 
1168 void qmp_migrate_incoming(const char *uri, Error **errp)
1169 {
1170     Error *local_err = NULL;
1171     static bool once = true;
1172 
1173     if (!deferred_incoming) {
1174         error_setg(errp, "For use with '-incoming defer'");
1175         return;
1176     }
1177     if (!once) {
1178         error_setg(errp, "The incoming migration has already been started");
1179     }
1180 
1181     qemu_start_incoming_migration(uri, &local_err);
1182 
1183     if (local_err) {
1184         error_propagate(errp, local_err);
1185         return;
1186     }
1187 
1188     once = false;
1189 }
1190 
1191 bool migration_is_blocked(Error **errp)
1192 {
1193     if (qemu_savevm_state_blocked(errp)) {
1194         return true;
1195     }
1196 
1197     if (migration_blockers) {
1198         *errp = error_copy(migration_blockers->data);
1199         return true;
1200     }
1201 
1202     return false;
1203 }
1204 
1205 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1206                  bool has_inc, bool inc, bool has_detach, bool detach,
1207                  Error **errp)
1208 {
1209     Error *local_err = NULL;
1210     MigrationState *s = migrate_get_current();
1211     MigrationParams params;
1212     const char *p;
1213 
1214     params.blk = has_blk && blk;
1215     params.shared = has_inc && inc;
1216 
1217     if (migration_is_setup_or_active(s->state) ||
1218         s->state == MIGRATION_STATUS_CANCELLING ||
1219         s->state == MIGRATION_STATUS_COLO) {
1220         error_setg(errp, QERR_MIGRATION_ACTIVE);
1221         return;
1222     }
1223     if (runstate_check(RUN_STATE_INMIGRATE)) {
1224         error_setg(errp, "Guest is waiting for an incoming migration");
1225         return;
1226     }
1227 
1228     if (migration_is_blocked(errp)) {
1229         return;
1230     }
1231 
1232     s = migrate_init(&params);
1233 
1234     if (strstart(uri, "tcp:", &p)) {
1235         tcp_start_outgoing_migration(s, p, &local_err);
1236 #ifdef CONFIG_RDMA
1237     } else if (strstart(uri, "rdma:", &p)) {
1238         rdma_start_outgoing_migration(s, p, &local_err);
1239 #endif
1240     } else if (strstart(uri, "exec:", &p)) {
1241         exec_start_outgoing_migration(s, p, &local_err);
1242     } else if (strstart(uri, "unix:", &p)) {
1243         unix_start_outgoing_migration(s, p, &local_err);
1244     } else if (strstart(uri, "fd:", &p)) {
1245         fd_start_outgoing_migration(s, p, &local_err);
1246     } else {
1247         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1248                    "a valid migration protocol");
1249         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1250                           MIGRATION_STATUS_FAILED);
1251         return;
1252     }
1253 
1254     if (local_err) {
1255         migrate_fd_error(s, local_err);
1256         error_propagate(errp, local_err);
1257         return;
1258     }
1259 }
1260 
1261 void qmp_migrate_cancel(Error **errp)
1262 {
1263     migrate_fd_cancel(migrate_get_current());
1264 }
1265 
1266 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1267 {
1268     MigrationState *s = migrate_get_current();
1269     int64_t new_size;
1270 
1271     /* Check for truncation */
1272     if (value != (size_t)value) {
1273         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1274                    "exceeding address space");
1275         return;
1276     }
1277 
1278     /* Cache should not be larger than guest ram size */
1279     if (value > ram_bytes_total()) {
1280         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1281                    "exceeds guest ram size ");
1282         return;
1283     }
1284 
1285     new_size = xbzrle_cache_resize(value);
1286     if (new_size < 0) {
1287         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1288                    "is smaller than page size");
1289         return;
1290     }
1291 
1292     s->xbzrle_cache_size = new_size;
1293 }
1294 
1295 int64_t qmp_query_migrate_cache_size(Error **errp)
1296 {
1297     return migrate_xbzrle_cache_size();
1298 }
1299 
1300 void qmp_migrate_set_speed(int64_t value, Error **errp)
1301 {
1302     MigrationParameters p = {
1303         .has_max_bandwidth = true,
1304         .max_bandwidth = value,
1305     };
1306 
1307     qmp_migrate_set_parameters(&p, errp);
1308 }
1309 
1310 void qmp_migrate_set_downtime(double value, Error **errp)
1311 {
1312     if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1313         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1314                          "the range of 0 to %d seconds",
1315                          MAX_MIGRATE_DOWNTIME_SECONDS);
1316         return;
1317     }
1318 
1319     value *= 1000; /* Convert to milliseconds */
1320     value = MAX(0, MIN(INT64_MAX, value));
1321 
1322     MigrationParameters p = {
1323         .has_downtime_limit = true,
1324         .downtime_limit = value,
1325     };
1326 
1327     qmp_migrate_set_parameters(&p, errp);
1328 }
1329 
1330 bool migrate_release_ram(void)
1331 {
1332     MigrationState *s;
1333 
1334     s = migrate_get_current();
1335 
1336     return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1337 }
1338 
1339 bool migrate_postcopy_ram(void)
1340 {
1341     MigrationState *s;
1342 
1343     s = migrate_get_current();
1344 
1345     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1346 }
1347 
1348 bool migrate_auto_converge(void)
1349 {
1350     MigrationState *s;
1351 
1352     s = migrate_get_current();
1353 
1354     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1355 }
1356 
1357 bool migrate_zero_blocks(void)
1358 {
1359     MigrationState *s;
1360 
1361     s = migrate_get_current();
1362 
1363     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1364 }
1365 
1366 bool migrate_use_compression(void)
1367 {
1368     MigrationState *s;
1369 
1370     s = migrate_get_current();
1371 
1372     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
1373 }
1374 
1375 int migrate_compress_level(void)
1376 {
1377     MigrationState *s;
1378 
1379     s = migrate_get_current();
1380 
1381     return s->parameters.compress_level;
1382 }
1383 
1384 int migrate_compress_threads(void)
1385 {
1386     MigrationState *s;
1387 
1388     s = migrate_get_current();
1389 
1390     return s->parameters.compress_threads;
1391 }
1392 
1393 int migrate_decompress_threads(void)
1394 {
1395     MigrationState *s;
1396 
1397     s = migrate_get_current();
1398 
1399     return s->parameters.decompress_threads;
1400 }
1401 
1402 bool migrate_use_events(void)
1403 {
1404     MigrationState *s;
1405 
1406     s = migrate_get_current();
1407 
1408     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1409 }
1410 
1411 int migrate_use_xbzrle(void)
1412 {
1413     MigrationState *s;
1414 
1415     s = migrate_get_current();
1416 
1417     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1418 }
1419 
1420 int64_t migrate_xbzrle_cache_size(void)
1421 {
1422     MigrationState *s;
1423 
1424     s = migrate_get_current();
1425 
1426     return s->xbzrle_cache_size;
1427 }
1428 
1429 /* migration thread support */
1430 /*
1431  * Something bad happened to the RP stream, mark an error
1432  * The caller shall print or trace something to indicate why
1433  */
1434 static void mark_source_rp_bad(MigrationState *s)
1435 {
1436     s->rp_state.error = true;
1437 }
1438 
1439 static struct rp_cmd_args {
1440     ssize_t     len; /* -1 = variable */
1441     const char *name;
1442 } rp_cmd_args[] = {
1443     [MIG_RP_MSG_INVALID]        = { .len = -1, .name = "INVALID" },
1444     [MIG_RP_MSG_SHUT]           = { .len =  4, .name = "SHUT" },
1445     [MIG_RP_MSG_PONG]           = { .len =  4, .name = "PONG" },
1446     [MIG_RP_MSG_REQ_PAGES]      = { .len = 12, .name = "REQ_PAGES" },
1447     [MIG_RP_MSG_REQ_PAGES_ID]   = { .len = -1, .name = "REQ_PAGES_ID" },
1448     [MIG_RP_MSG_MAX]            = { .len = -1, .name = "MAX" },
1449 };
1450 
1451 /*
1452  * Process a request for pages received on the return path,
1453  * We're allowed to send more than requested (e.g. to round to our page size)
1454  * and we don't need to send pages that have already been sent.
1455  */
1456 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1457                                        ram_addr_t start, size_t len)
1458 {
1459     long our_host_ps = getpagesize();
1460 
1461     trace_migrate_handle_rp_req_pages(rbname, start, len);
1462 
1463     /*
1464      * Since we currently insist on matching page sizes, just sanity check
1465      * we're being asked for whole host pages.
1466      */
1467     if (start & (our_host_ps-1) ||
1468        (len & (our_host_ps-1))) {
1469         error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1470                      " len: %zd", __func__, start, len);
1471         mark_source_rp_bad(ms);
1472         return;
1473     }
1474 
1475     if (ram_save_queue_pages(ms, rbname, start, len)) {
1476         mark_source_rp_bad(ms);
1477     }
1478 }
1479 
1480 /*
1481  * Handles messages sent on the return path towards the source VM
1482  *
1483  */
1484 static void *source_return_path_thread(void *opaque)
1485 {
1486     MigrationState *ms = opaque;
1487     QEMUFile *rp = ms->rp_state.from_dst_file;
1488     uint16_t header_len, header_type;
1489     uint8_t buf[512];
1490     uint32_t tmp32, sibling_error;
1491     ram_addr_t start = 0; /* =0 to silence warning */
1492     size_t  len = 0, expected_len;
1493     int res;
1494 
1495     trace_source_return_path_thread_entry();
1496     while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
1497            migration_is_setup_or_active(ms->state)) {
1498         trace_source_return_path_thread_loop_top();
1499         header_type = qemu_get_be16(rp);
1500         header_len = qemu_get_be16(rp);
1501 
1502         if (header_type >= MIG_RP_MSG_MAX ||
1503             header_type == MIG_RP_MSG_INVALID) {
1504             error_report("RP: Received invalid message 0x%04x length 0x%04x",
1505                     header_type, header_len);
1506             mark_source_rp_bad(ms);
1507             goto out;
1508         }
1509 
1510         if ((rp_cmd_args[header_type].len != -1 &&
1511             header_len != rp_cmd_args[header_type].len) ||
1512             header_len > sizeof(buf)) {
1513             error_report("RP: Received '%s' message (0x%04x) with"
1514                     "incorrect length %d expecting %zu",
1515                     rp_cmd_args[header_type].name, header_type, header_len,
1516                     (size_t)rp_cmd_args[header_type].len);
1517             mark_source_rp_bad(ms);
1518             goto out;
1519         }
1520 
1521         /* We know we've got a valid header by this point */
1522         res = qemu_get_buffer(rp, buf, header_len);
1523         if (res != header_len) {
1524             error_report("RP: Failed reading data for message 0x%04x"
1525                          " read %d expected %d",
1526                          header_type, res, header_len);
1527             mark_source_rp_bad(ms);
1528             goto out;
1529         }
1530 
1531         /* OK, we have the message and the data */
1532         switch (header_type) {
1533         case MIG_RP_MSG_SHUT:
1534             sibling_error = ldl_be_p(buf);
1535             trace_source_return_path_thread_shut(sibling_error);
1536             if (sibling_error) {
1537                 error_report("RP: Sibling indicated error %d", sibling_error);
1538                 mark_source_rp_bad(ms);
1539             }
1540             /*
1541              * We'll let the main thread deal with closing the RP
1542              * we could do a shutdown(2) on it, but we're the only user
1543              * anyway, so there's nothing gained.
1544              */
1545             goto out;
1546 
1547         case MIG_RP_MSG_PONG:
1548             tmp32 = ldl_be_p(buf);
1549             trace_source_return_path_thread_pong(tmp32);
1550             break;
1551 
1552         case MIG_RP_MSG_REQ_PAGES:
1553             start = ldq_be_p(buf);
1554             len = ldl_be_p(buf + 8);
1555             migrate_handle_rp_req_pages(ms, NULL, start, len);
1556             break;
1557 
1558         case MIG_RP_MSG_REQ_PAGES_ID:
1559             expected_len = 12 + 1; /* header + termination */
1560 
1561             if (header_len >= expected_len) {
1562                 start = ldq_be_p(buf);
1563                 len = ldl_be_p(buf + 8);
1564                 /* Now we expect an idstr */
1565                 tmp32 = buf[12]; /* Length of the following idstr */
1566                 buf[13 + tmp32] = '\0';
1567                 expected_len += tmp32;
1568             }
1569             if (header_len != expected_len) {
1570                 error_report("RP: Req_Page_id with length %d expecting %zd",
1571                         header_len, expected_len);
1572                 mark_source_rp_bad(ms);
1573                 goto out;
1574             }
1575             migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
1576             break;
1577 
1578         default:
1579             break;
1580         }
1581     }
1582     if (qemu_file_get_error(rp)) {
1583         trace_source_return_path_thread_bad_end();
1584         mark_source_rp_bad(ms);
1585     }
1586 
1587     trace_source_return_path_thread_end();
1588 out:
1589     ms->rp_state.from_dst_file = NULL;
1590     qemu_fclose(rp);
1591     return NULL;
1592 }
1593 
1594 static int open_return_path_on_source(MigrationState *ms)
1595 {
1596 
1597     ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
1598     if (!ms->rp_state.from_dst_file) {
1599         return -1;
1600     }
1601 
1602     trace_open_return_path_on_source();
1603     qemu_thread_create(&ms->rp_state.rp_thread, "return path",
1604                        source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
1605 
1606     trace_open_return_path_on_source_continue();
1607 
1608     return 0;
1609 }
1610 
1611 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1612 static int await_return_path_close_on_source(MigrationState *ms)
1613 {
1614     /*
1615      * If this is a normal exit then the destination will send a SHUT and the
1616      * rp_thread will exit, however if there's an error we need to cause
1617      * it to exit.
1618      */
1619     if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
1620         /*
1621          * shutdown(2), if we have it, will cause it to unblock if it's stuck
1622          * waiting for the destination.
1623          */
1624         qemu_file_shutdown(ms->rp_state.from_dst_file);
1625         mark_source_rp_bad(ms);
1626     }
1627     trace_await_return_path_close_on_source_joining();
1628     qemu_thread_join(&ms->rp_state.rp_thread);
1629     trace_await_return_path_close_on_source_close();
1630     return ms->rp_state.error;
1631 }
1632 
1633 /*
1634  * Switch from normal iteration to postcopy
1635  * Returns non-0 on error
1636  */
1637 static int postcopy_start(MigrationState *ms, bool *old_vm_running)
1638 {
1639     int ret;
1640     QIOChannelBuffer *bioc;
1641     QEMUFile *fb;
1642     int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1643     bool restart_block = false;
1644     migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
1645                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
1646 
1647     trace_postcopy_start();
1648     qemu_mutex_lock_iothread();
1649     trace_postcopy_start_set_run();
1650 
1651     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1652     *old_vm_running = runstate_is_running();
1653     global_state_store();
1654     ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1655     if (ret < 0) {
1656         goto fail;
1657     }
1658 
1659     ret = bdrv_inactivate_all();
1660     if (ret < 0) {
1661         goto fail;
1662     }
1663     restart_block = true;
1664 
1665     /*
1666      * Cause any non-postcopiable, but iterative devices to
1667      * send out their final data.
1668      */
1669     qemu_savevm_state_complete_precopy(ms->to_dst_file, true);
1670 
1671     /*
1672      * in Finish migrate and with the io-lock held everything should
1673      * be quiet, but we've potentially still got dirty pages and we
1674      * need to tell the destination to throw any pages it's already received
1675      * that are dirty
1676      */
1677     if (ram_postcopy_send_discard_bitmap(ms)) {
1678         error_report("postcopy send discard bitmap failed");
1679         goto fail;
1680     }
1681 
1682     /*
1683      * send rest of state - note things that are doing postcopy
1684      * will notice we're in POSTCOPY_ACTIVE and not actually
1685      * wrap their state up here
1686      */
1687     qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
1688     /* Ping just for debugging, helps line traces up */
1689     qemu_savevm_send_ping(ms->to_dst_file, 2);
1690 
1691     /*
1692      * While loading the device state we may trigger page transfer
1693      * requests and the fd must be free to process those, and thus
1694      * the destination must read the whole device state off the fd before
1695      * it starts processing it.  Unfortunately the ad-hoc migration format
1696      * doesn't allow the destination to know the size to read without fully
1697      * parsing it through each devices load-state code (especially the open
1698      * coded devices that use get/put).
1699      * So we wrap the device state up in a package with a length at the start;
1700      * to do this we use a qemu_buf to hold the whole of the device state.
1701      */
1702     bioc = qio_channel_buffer_new(4096);
1703     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
1704     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
1705     object_unref(OBJECT(bioc));
1706 
1707     /*
1708      * Make sure the receiver can get incoming pages before we send the rest
1709      * of the state
1710      */
1711     qemu_savevm_send_postcopy_listen(fb);
1712 
1713     qemu_savevm_state_complete_precopy(fb, false);
1714     qemu_savevm_send_ping(fb, 3);
1715 
1716     qemu_savevm_send_postcopy_run(fb);
1717 
1718     /* <><> end of stuff going into the package */
1719 
1720     /* Last point of recovery; as soon as we send the package the destination
1721      * can open devices and potentially start running.
1722      * Lets just check again we've not got any errors.
1723      */
1724     ret = qemu_file_get_error(ms->to_dst_file);
1725     if (ret) {
1726         error_report("postcopy_start: Migration stream errored (pre package)");
1727         goto fail_closefb;
1728     }
1729 
1730     restart_block = false;
1731 
1732     /* Now send that blob */
1733     if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
1734         goto fail_closefb;
1735     }
1736     qemu_fclose(fb);
1737 
1738     /* Send a notify to give a chance for anything that needs to happen
1739      * at the transition to postcopy and after the device state; in particular
1740      * spice needs to trigger a transition now
1741      */
1742     ms->postcopy_after_devices = true;
1743     notifier_list_notify(&migration_state_notifiers, ms);
1744 
1745     ms->downtime =  qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
1746 
1747     qemu_mutex_unlock_iothread();
1748 
1749     /*
1750      * Although this ping is just for debug, it could potentially be
1751      * used for getting a better measurement of downtime at the source.
1752      */
1753     qemu_savevm_send_ping(ms->to_dst_file, 4);
1754 
1755     if (migrate_release_ram()) {
1756         ram_postcopy_migrated_memory_release(ms);
1757     }
1758 
1759     ret = qemu_file_get_error(ms->to_dst_file);
1760     if (ret) {
1761         error_report("postcopy_start: Migration stream errored");
1762         migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1763                               MIGRATION_STATUS_FAILED);
1764     }
1765 
1766     return ret;
1767 
1768 fail_closefb:
1769     qemu_fclose(fb);
1770 fail:
1771     migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1772                           MIGRATION_STATUS_FAILED);
1773     if (restart_block) {
1774         /* A failure happened early enough that we know the destination hasn't
1775          * accessed block devices, so we're safe to recover.
1776          */
1777         Error *local_err = NULL;
1778 
1779         bdrv_invalidate_cache_all(&local_err);
1780         if (local_err) {
1781             error_report_err(local_err);
1782         }
1783     }
1784     qemu_mutex_unlock_iothread();
1785     return -1;
1786 }
1787 
1788 /**
1789  * migration_completion: Used by migration_thread when there's not much left.
1790  *   The caller 'breaks' the loop when this returns.
1791  *
1792  * @s: Current migration state
1793  * @current_active_state: The migration state we expect to be in
1794  * @*old_vm_running: Pointer to old_vm_running flag
1795  * @*start_time: Pointer to time to update
1796  */
1797 static void migration_completion(MigrationState *s, int current_active_state,
1798                                  bool *old_vm_running,
1799                                  int64_t *start_time)
1800 {
1801     int ret;
1802 
1803     if (s->state == MIGRATION_STATUS_ACTIVE) {
1804         qemu_mutex_lock_iothread();
1805         *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1806         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1807         *old_vm_running = runstate_is_running();
1808         ret = global_state_store();
1809 
1810         if (!ret) {
1811             ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1812             /*
1813              * Don't mark the image with BDRV_O_INACTIVE flag if
1814              * we will go into COLO stage later.
1815              */
1816             if (ret >= 0 && !migrate_colo_enabled()) {
1817                 ret = bdrv_inactivate_all();
1818             }
1819             if (ret >= 0) {
1820                 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
1821                 qemu_savevm_state_complete_precopy(s->to_dst_file, false);
1822                 s->block_inactive = true;
1823             }
1824         }
1825         qemu_mutex_unlock_iothread();
1826 
1827         if (ret < 0) {
1828             goto fail;
1829         }
1830     } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1831         trace_migration_completion_postcopy_end();
1832 
1833         qemu_savevm_state_complete_postcopy(s->to_dst_file);
1834         trace_migration_completion_postcopy_end_after_complete();
1835     }
1836 
1837     /*
1838      * If rp was opened we must clean up the thread before
1839      * cleaning everything else up (since if there are no failures
1840      * it will wait for the destination to send it's status in
1841      * a SHUT command).
1842      * Postcopy opens rp if enabled (even if it's not avtivated)
1843      */
1844     if (migrate_postcopy_ram()) {
1845         int rp_error;
1846         trace_migration_completion_postcopy_end_before_rp();
1847         rp_error = await_return_path_close_on_source(s);
1848         trace_migration_completion_postcopy_end_after_rp(rp_error);
1849         if (rp_error) {
1850             goto fail_invalidate;
1851         }
1852     }
1853 
1854     if (qemu_file_get_error(s->to_dst_file)) {
1855         trace_migration_completion_file_err();
1856         goto fail_invalidate;
1857     }
1858 
1859     if (!migrate_colo_enabled()) {
1860         migrate_set_state(&s->state, current_active_state,
1861                           MIGRATION_STATUS_COMPLETED);
1862     }
1863 
1864     return;
1865 
1866 fail_invalidate:
1867     /* If not doing postcopy, vm_start() will be called: let's regain
1868      * control on images.
1869      */
1870     if (s->state == MIGRATION_STATUS_ACTIVE) {
1871         Error *local_err = NULL;
1872 
1873         qemu_mutex_lock_iothread();
1874         bdrv_invalidate_cache_all(&local_err);
1875         if (local_err) {
1876             error_report_err(local_err);
1877         } else {
1878             s->block_inactive = false;
1879         }
1880         qemu_mutex_unlock_iothread();
1881     }
1882 
1883 fail:
1884     migrate_set_state(&s->state, current_active_state,
1885                       MIGRATION_STATUS_FAILED);
1886 }
1887 
1888 bool migrate_colo_enabled(void)
1889 {
1890     MigrationState *s = migrate_get_current();
1891     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
1892 }
1893 
1894 /*
1895  * Master migration thread on the source VM.
1896  * It drives the migration and pumps the data down the outgoing channel.
1897  */
1898 static void *migration_thread(void *opaque)
1899 {
1900     MigrationState *s = opaque;
1901     /* Used by the bandwidth calcs, updated later */
1902     int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1903     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
1904     int64_t initial_bytes = 0;
1905     int64_t max_size = 0;
1906     int64_t start_time = initial_time;
1907     int64_t end_time;
1908     bool old_vm_running = false;
1909     bool entered_postcopy = false;
1910     /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
1911     enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE;
1912     bool enable_colo = migrate_colo_enabled();
1913 
1914     rcu_register_thread();
1915 
1916     qemu_savevm_state_header(s->to_dst_file);
1917 
1918     if (migrate_postcopy_ram()) {
1919         /* Now tell the dest that it should open its end so it can reply */
1920         qemu_savevm_send_open_return_path(s->to_dst_file);
1921 
1922         /* And do a ping that will make stuff easier to debug */
1923         qemu_savevm_send_ping(s->to_dst_file, 1);
1924 
1925         /*
1926          * Tell the destination that we *might* want to do postcopy later;
1927          * if the other end can't do postcopy it should fail now, nice and
1928          * early.
1929          */
1930         qemu_savevm_send_postcopy_advise(s->to_dst_file);
1931     }
1932 
1933     qemu_savevm_state_begin(s->to_dst_file, &s->params);
1934 
1935     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
1936     current_active_state = MIGRATION_STATUS_ACTIVE;
1937     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1938                       MIGRATION_STATUS_ACTIVE);
1939 
1940     trace_migration_thread_setup_complete();
1941 
1942     while (s->state == MIGRATION_STATUS_ACTIVE ||
1943            s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1944         int64_t current_time;
1945         uint64_t pending_size;
1946 
1947         if (!qemu_file_rate_limit(s->to_dst_file)) {
1948             uint64_t pend_post, pend_nonpost;
1949 
1950             qemu_savevm_state_pending(s->to_dst_file, max_size, &pend_nonpost,
1951                                       &pend_post);
1952             pending_size = pend_nonpost + pend_post;
1953             trace_migrate_pending(pending_size, max_size,
1954                                   pend_post, pend_nonpost);
1955             if (pending_size && pending_size >= max_size) {
1956                 /* Still a significant amount to transfer */
1957 
1958                 if (migrate_postcopy_ram() &&
1959                     s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE &&
1960                     pend_nonpost <= max_size &&
1961                     atomic_read(&s->start_postcopy)) {
1962 
1963                     if (!postcopy_start(s, &old_vm_running)) {
1964                         current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE;
1965                         entered_postcopy = true;
1966                     }
1967 
1968                     continue;
1969                 }
1970                 /* Just another iteration step */
1971                 qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy);
1972             } else {
1973                 trace_migration_thread_low_pending(pending_size);
1974                 migration_completion(s, current_active_state,
1975                                      &old_vm_running, &start_time);
1976                 break;
1977             }
1978         }
1979 
1980         if (qemu_file_get_error(s->to_dst_file)) {
1981             migrate_set_state(&s->state, current_active_state,
1982                               MIGRATION_STATUS_FAILED);
1983             trace_migration_thread_file_err();
1984             break;
1985         }
1986         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1987         if (current_time >= initial_time + BUFFER_DELAY) {
1988             uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) -
1989                                          initial_bytes;
1990             uint64_t time_spent = current_time - initial_time;
1991             double bandwidth = (double)transferred_bytes / time_spent;
1992             max_size = bandwidth * s->parameters.downtime_limit;
1993 
1994             s->mbps = (((double) transferred_bytes * 8.0) /
1995                     ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
1996 
1997             trace_migrate_transferred(transferred_bytes, time_spent,
1998                                       bandwidth, max_size);
1999             /* if we haven't sent anything, we don't want to recalculate
2000                10000 is a small enough number for our purposes */
2001             if (s->dirty_bytes_rate && transferred_bytes > 10000) {
2002                 s->expected_downtime = s->dirty_bytes_rate / bandwidth;
2003             }
2004 
2005             qemu_file_reset_rate_limit(s->to_dst_file);
2006             initial_time = current_time;
2007             initial_bytes = qemu_ftell(s->to_dst_file);
2008         }
2009         if (qemu_file_rate_limit(s->to_dst_file)) {
2010             /* usleep expects microseconds */
2011             g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
2012         }
2013     }
2014 
2015     trace_migration_thread_after_loop();
2016     /* If we enabled cpu throttling for auto-converge, turn it off. */
2017     cpu_throttle_stop();
2018     end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2019 
2020     qemu_mutex_lock_iothread();
2021     /*
2022      * The resource has been allocated by migration will be reused in COLO
2023      * process, so don't release them.
2024      */
2025     if (!enable_colo) {
2026         qemu_savevm_state_cleanup();
2027     }
2028     if (s->state == MIGRATION_STATUS_COMPLETED) {
2029         uint64_t transferred_bytes = qemu_ftell(s->to_dst_file);
2030         s->total_time = end_time - s->total_time;
2031         if (!entered_postcopy) {
2032             s->downtime = end_time - start_time;
2033         }
2034         if (s->total_time) {
2035             s->mbps = (((double) transferred_bytes * 8.0) /
2036                        ((double) s->total_time)) / 1000;
2037         }
2038         runstate_set(RUN_STATE_POSTMIGRATE);
2039     } else {
2040         if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) {
2041             migrate_start_colo_process(s);
2042             qemu_savevm_state_cleanup();
2043             /*
2044             * Fixme: we will run VM in COLO no matter its old running state.
2045             * After exited COLO, we will keep running.
2046             */
2047             old_vm_running = true;
2048         }
2049         if (old_vm_running && !entered_postcopy) {
2050             vm_start();
2051         } else {
2052             if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2053                 runstate_set(RUN_STATE_POSTMIGRATE);
2054             }
2055         }
2056     }
2057     qemu_bh_schedule(s->cleanup_bh);
2058     qemu_mutex_unlock_iothread();
2059 
2060     rcu_unregister_thread();
2061     return NULL;
2062 }
2063 
2064 void migrate_fd_connect(MigrationState *s)
2065 {
2066     s->expected_downtime = s->parameters.downtime_limit;
2067     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
2068 
2069     qemu_file_set_blocking(s->to_dst_file, true);
2070     qemu_file_set_rate_limit(s->to_dst_file,
2071                              s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
2072 
2073     /* Notify before starting migration thread */
2074     notifier_list_notify(&migration_state_notifiers, s);
2075 
2076     /*
2077      * Open the return path; currently for postcopy but other things might
2078      * also want it.
2079      */
2080     if (migrate_postcopy_ram()) {
2081         if (open_return_path_on_source(s)) {
2082             error_report("Unable to open return-path for postcopy");
2083             migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2084                               MIGRATION_STATUS_FAILED);
2085             migrate_fd_cleanup(s);
2086             return;
2087         }
2088     }
2089 
2090     migrate_compress_threads_create();
2091     qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
2092                        QEMU_THREAD_JOINABLE);
2093     s->migration_thread_running = true;
2094 }
2095 
2096 PostcopyState  postcopy_state_get(void)
2097 {
2098     return atomic_mb_read(&incoming_postcopy_state);
2099 }
2100 
2101 /* Set the state and return the old state */
2102 PostcopyState postcopy_state_set(PostcopyState new_state)
2103 {
2104     return atomic_xchg(&incoming_postcopy_state, new_state);
2105 }
2106 
2107