xref: /openbmc/qemu/migration/migration.c (revision ba9915e1f87fec742775d64859e881e4ab611429)
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 "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         current_migration.parameters.tls_creds = g_strdup("");
113         current_migration.parameters.tls_hostname = g_strdup("");
114         once = true;
115     }
116     return &current_migration;
117 }
118 
119 MigrationIncomingState *migration_incoming_get_current(void)
120 {
121     static bool once;
122     static MigrationIncomingState mis_current;
123 
124     if (!once) {
125         mis_current.state = MIGRATION_STATUS_NONE;
126         memset(&mis_current, 0, sizeof(MigrationIncomingState));
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         once = true;
131     }
132     return &mis_current;
133 }
134 
135 void migration_incoming_state_destroy(void)
136 {
137     struct MigrationIncomingState *mis = migration_incoming_get_current();
138 
139     qemu_event_destroy(&mis->main_thread_load_event);
140     loadvm_free_handlers(mis);
141 }
142 
143 
144 typedef struct {
145     bool optional;
146     uint32_t size;
147     uint8_t runstate[100];
148     RunState state;
149     bool received;
150 } GlobalState;
151 
152 static GlobalState global_state;
153 
154 int global_state_store(void)
155 {
156     if (!runstate_store((char *)global_state.runstate,
157                         sizeof(global_state.runstate))) {
158         error_report("runstate name too big: %s", global_state.runstate);
159         trace_migrate_state_too_big();
160         return -EINVAL;
161     }
162     return 0;
163 }
164 
165 void global_state_store_running(void)
166 {
167     const char *state = RunState_lookup[RUN_STATE_RUNNING];
168     strncpy((char *)global_state.runstate,
169            state, sizeof(global_state.runstate));
170 }
171 
172 static bool global_state_received(void)
173 {
174     return global_state.received;
175 }
176 
177 static RunState global_state_get_runstate(void)
178 {
179     return global_state.state;
180 }
181 
182 void global_state_set_optional(void)
183 {
184     global_state.optional = true;
185 }
186 
187 static bool global_state_needed(void *opaque)
188 {
189     GlobalState *s = opaque;
190     char *runstate = (char *)s->runstate;
191 
192     /* If it is not optional, it is mandatory */
193 
194     if (s->optional == false) {
195         return true;
196     }
197 
198     /* If state is running or paused, it is not needed */
199 
200     if (strcmp(runstate, "running") == 0 ||
201         strcmp(runstate, "paused") == 0) {
202         return false;
203     }
204 
205     /* for any other state it is needed */
206     return true;
207 }
208 
209 static int global_state_post_load(void *opaque, int version_id)
210 {
211     GlobalState *s = opaque;
212     Error *local_err = NULL;
213     int r;
214     char *runstate = (char *)s->runstate;
215 
216     s->received = true;
217     trace_migrate_global_state_post_load(runstate);
218 
219     r = qapi_enum_parse(RunState_lookup, runstate, RUN_STATE__MAX,
220                                 -1, &local_err);
221 
222     if (r == -1) {
223         if (local_err) {
224             error_report_err(local_err);
225         }
226         return -EINVAL;
227     }
228     s->state = r;
229 
230     return 0;
231 }
232 
233 static void global_state_pre_save(void *opaque)
234 {
235     GlobalState *s = opaque;
236 
237     trace_migrate_global_state_pre_save((char *)s->runstate);
238     s->size = strlen((char *)s->runstate) + 1;
239 }
240 
241 static const VMStateDescription vmstate_globalstate = {
242     .name = "globalstate",
243     .version_id = 1,
244     .minimum_version_id = 1,
245     .post_load = global_state_post_load,
246     .pre_save = global_state_pre_save,
247     .needed = global_state_needed,
248     .fields = (VMStateField[]) {
249         VMSTATE_UINT32(size, GlobalState),
250         VMSTATE_BUFFER(runstate, GlobalState),
251         VMSTATE_END_OF_LIST()
252     },
253 };
254 
255 void register_global_state(void)
256 {
257     /* We would use it independently that we receive it */
258     strcpy((char *)&global_state.runstate, "");
259     global_state.received = false;
260     vmstate_register(NULL, 0, &vmstate_globalstate, &global_state);
261 }
262 
263 static void migrate_generate_event(int new_state)
264 {
265     if (migrate_use_events()) {
266         qapi_event_send_migration(new_state, &error_abort);
267     }
268 }
269 
270 /*
271  * Called on -incoming with a defer: uri.
272  * The migration can be started later after any parameters have been
273  * changed.
274  */
275 static void deferred_incoming_migration(Error **errp)
276 {
277     if (deferred_incoming) {
278         error_setg(errp, "Incoming migration already deferred");
279     }
280     deferred_incoming = true;
281 }
282 
283 /* Request a range of pages from the source VM at the given
284  * start address.
285  *   rbname: Name of the RAMBlock to request the page in, if NULL it's the same
286  *           as the last request (a name must have been given previously)
287  *   Start: Address offset within the RB
288  *   Len: Length in bytes required - must be a multiple of pagesize
289  */
290 void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
291                                ram_addr_t start, size_t len)
292 {
293     uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
294     size_t msglen = 12; /* start + len */
295 
296     *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
297     *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
298 
299     if (rbname) {
300         int rbname_len = strlen(rbname);
301         assert(rbname_len < 256);
302 
303         bufc[msglen++] = rbname_len;
304         memcpy(bufc + msglen, rbname, rbname_len);
305         msglen += rbname_len;
306         migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc);
307     } else {
308         migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc);
309     }
310 }
311 
312 void qemu_start_incoming_migration(const char *uri, Error **errp)
313 {
314     const char *p;
315 
316     qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
317     if (!strcmp(uri, "defer")) {
318         deferred_incoming_migration(errp);
319     } else if (strstart(uri, "tcp:", &p)) {
320         tcp_start_incoming_migration(p, errp);
321 #ifdef CONFIG_RDMA
322     } else if (strstart(uri, "rdma:", &p)) {
323         rdma_start_incoming_migration(p, errp);
324 #endif
325     } else if (strstart(uri, "exec:", &p)) {
326         exec_start_incoming_migration(p, errp);
327     } else if (strstart(uri, "unix:", &p)) {
328         unix_start_incoming_migration(p, errp);
329     } else if (strstart(uri, "fd:", &p)) {
330         fd_start_incoming_migration(p, errp);
331     } else {
332         error_setg(errp, "unknown migration protocol: %s", uri);
333     }
334 }
335 
336 static void process_incoming_migration_bh(void *opaque)
337 {
338     Error *local_err = NULL;
339     MigrationIncomingState *mis = opaque;
340 
341     /* Make sure all file formats flush their mutable metadata.
342      * If we get an error here, just don't restart the VM yet. */
343     bdrv_invalidate_cache_all(&local_err);
344     if (local_err) {
345         error_report_err(local_err);
346         local_err = NULL;
347         autostart = false;
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     mis->largest_page_size = qemu_ram_pagesize_largest();
391     postcopy_state_set(POSTCOPY_INCOMING_NONE);
392     migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
393                       MIGRATION_STATUS_ACTIVE);
394     ret = qemu_loadvm_state(f);
395 
396     ps = postcopy_state_get();
397     trace_process_incoming_migration_co_end(ret, ps);
398     if (ps != POSTCOPY_INCOMING_NONE) {
399         if (ps == POSTCOPY_INCOMING_ADVISE) {
400             /*
401              * Where a migration had postcopy enabled (and thus went to advise)
402              * but managed to complete within the precopy period, we can use
403              * the normal exit.
404              */
405             postcopy_ram_incoming_cleanup(mis);
406         } else if (ret >= 0) {
407             /*
408              * Postcopy was started, cleanup should happen at the end of the
409              * postcopy thread.
410              */
411             trace_process_incoming_migration_co_postcopy_end_main();
412             return;
413         }
414         /* Else if something went wrong then just fall out of the normal exit */
415     }
416 
417     /* we get COLO info, and know if we are in COLO mode */
418     if (!ret && migration_incoming_enable_colo()) {
419         mis->migration_incoming_co = qemu_coroutine_self();
420         qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
421              colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
422         mis->have_colo_incoming_thread = true;
423         qemu_coroutine_yield();
424 
425         /* Wait checkpoint incoming thread exit before free resource */
426         qemu_thread_join(&mis->colo_incoming_thread);
427     }
428 
429     if (ret < 0) {
430         migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
431                           MIGRATION_STATUS_FAILED);
432         error_report("load of migration failed: %s", strerror(-ret));
433         migrate_decompress_threads_join();
434         exit(EXIT_FAILURE);
435     }
436 
437     qemu_fclose(f);
438     free_xbzrle_decoded_buf();
439 
440     mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
441     qemu_bh_schedule(mis->bh);
442 }
443 
444 void migration_fd_process_incoming(QEMUFile *f)
445 {
446     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, f);
447 
448     migrate_decompress_threads_create();
449     qemu_file_set_blocking(f, false);
450     qemu_coroutine_enter(co);
451 }
452 
453 
454 void migration_channel_process_incoming(MigrationState *s,
455                                         QIOChannel *ioc)
456 {
457     trace_migration_set_incoming_channel(
458         ioc, object_get_typename(OBJECT(ioc)));
459 
460     if (s->parameters.tls_creds &&
461         *s->parameters.tls_creds &&
462         !object_dynamic_cast(OBJECT(ioc),
463                              TYPE_QIO_CHANNEL_TLS)) {
464         Error *local_err = NULL;
465         migration_tls_channel_process_incoming(s, ioc, &local_err);
466         if (local_err) {
467             error_report_err(local_err);
468         }
469     } else {
470         QEMUFile *f = qemu_fopen_channel_input(ioc);
471         migration_fd_process_incoming(f);
472     }
473 }
474 
475 
476 void migration_channel_connect(MigrationState *s,
477                                QIOChannel *ioc,
478                                const char *hostname)
479 {
480     trace_migration_set_outgoing_channel(
481         ioc, object_get_typename(OBJECT(ioc)), hostname);
482 
483     if (s->parameters.tls_creds &&
484         *s->parameters.tls_creds &&
485         !object_dynamic_cast(OBJECT(ioc),
486                              TYPE_QIO_CHANNEL_TLS)) {
487         Error *local_err = NULL;
488         migration_tls_channel_connect(s, ioc, hostname, &local_err);
489         if (local_err) {
490             migrate_fd_error(s, local_err);
491             error_free(local_err);
492         }
493     } else {
494         QEMUFile *f = qemu_fopen_channel_output(ioc);
495 
496         s->to_dst_file = f;
497 
498         migrate_fd_connect(s);
499     }
500 }
501 
502 
503 /*
504  * Send a message on the return channel back to the source
505  * of the migration.
506  */
507 void migrate_send_rp_message(MigrationIncomingState *mis,
508                              enum mig_rp_message_type message_type,
509                              uint16_t len, void *data)
510 {
511     trace_migrate_send_rp_message((int)message_type, len);
512     qemu_mutex_lock(&mis->rp_mutex);
513     qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
514     qemu_put_be16(mis->to_src_file, len);
515     qemu_put_buffer(mis->to_src_file, data, len);
516     qemu_fflush(mis->to_src_file);
517     qemu_mutex_unlock(&mis->rp_mutex);
518 }
519 
520 /*
521  * Send a 'SHUT' message on the return channel with the given value
522  * to indicate that we've finished with the RP.  Non-0 value indicates
523  * error.
524  */
525 void migrate_send_rp_shut(MigrationIncomingState *mis,
526                           uint32_t value)
527 {
528     uint32_t buf;
529 
530     buf = cpu_to_be32(value);
531     migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
532 }
533 
534 /*
535  * Send a 'PONG' message on the return channel with the given value
536  * (normally in response to a 'PING')
537  */
538 void migrate_send_rp_pong(MigrationIncomingState *mis,
539                           uint32_t value)
540 {
541     uint32_t buf;
542 
543     buf = cpu_to_be32(value);
544     migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
545 }
546 
547 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
548 {
549     MigrationCapabilityStatusList *head = NULL;
550     MigrationCapabilityStatusList *caps;
551     MigrationState *s = migrate_get_current();
552     int i;
553 
554     caps = NULL; /* silence compiler warning */
555     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
556         if (i == MIGRATION_CAPABILITY_X_COLO && !colo_supported()) {
557             continue;
558         }
559         if (head == NULL) {
560             head = g_malloc0(sizeof(*caps));
561             caps = head;
562         } else {
563             caps->next = g_malloc0(sizeof(*caps));
564             caps = caps->next;
565         }
566         caps->value =
567             g_malloc(sizeof(*caps->value));
568         caps->value->capability = i;
569         caps->value->state = s->enabled_capabilities[i];
570     }
571 
572     return head;
573 }
574 
575 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
576 {
577     MigrationParameters *params;
578     MigrationState *s = migrate_get_current();
579 
580     params = g_malloc0(sizeof(*params));
581     params->has_compress_level = true;
582     params->compress_level = s->parameters.compress_level;
583     params->has_compress_threads = true;
584     params->compress_threads = s->parameters.compress_threads;
585     params->has_decompress_threads = true;
586     params->decompress_threads = s->parameters.decompress_threads;
587     params->has_cpu_throttle_initial = true;
588     params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
589     params->has_cpu_throttle_increment = true;
590     params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
591     params->has_tls_creds = !!s->parameters.tls_creds;
592     params->tls_creds = g_strdup(s->parameters.tls_creds);
593     params->has_tls_hostname = !!s->parameters.tls_hostname;
594     params->tls_hostname = g_strdup(s->parameters.tls_hostname);
595     params->has_max_bandwidth = true;
596     params->max_bandwidth = s->parameters.max_bandwidth;
597     params->has_downtime_limit = true;
598     params->downtime_limit = s->parameters.downtime_limit;
599     params->has_x_checkpoint_delay = true;
600     params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
601 
602     return params;
603 }
604 
605 /*
606  * Return true if we're already in the middle of a migration
607  * (i.e. any of the active or setup states)
608  */
609 static bool migration_is_setup_or_active(int state)
610 {
611     switch (state) {
612     case MIGRATION_STATUS_ACTIVE:
613     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
614     case MIGRATION_STATUS_SETUP:
615         return true;
616 
617     default:
618         return false;
619 
620     }
621 }
622 
623 static void get_xbzrle_cache_stats(MigrationInfo *info)
624 {
625     if (migrate_use_xbzrle()) {
626         info->has_xbzrle_cache = true;
627         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
628         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
629         info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred();
630         info->xbzrle_cache->pages = xbzrle_mig_pages_transferred();
631         info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss();
632         info->xbzrle_cache->cache_miss_rate = xbzrle_mig_cache_miss_rate();
633         info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow();
634     }
635 }
636 
637 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
638 {
639     info->has_ram = true;
640     info->ram = g_malloc0(sizeof(*info->ram));
641     info->ram->transferred = ram_bytes_transferred();
642     info->ram->total = ram_bytes_total();
643     info->ram->duplicate = dup_mig_pages_transferred();
644     /* legacy value.  It is not used anymore */
645     info->ram->skipped = 0;
646     info->ram->normal = norm_mig_pages_transferred();
647     info->ram->normal_bytes = norm_mig_pages_transferred() *
648         qemu_target_page_size();
649     info->ram->mbps = s->mbps;
650     info->ram->dirty_sync_count = ram_dirty_sync_count();
651     info->ram->postcopy_requests = ram_postcopy_requests();
652     info->ram->page_size = qemu_target_page_size();
653 
654     if (s->state != MIGRATION_STATUS_COMPLETED) {
655         info->ram->remaining = ram_bytes_remaining();
656         info->ram->dirty_pages_rate = ram_dirty_pages_rate();
657     }
658 }
659 
660 MigrationInfo *qmp_query_migrate(Error **errp)
661 {
662     MigrationInfo *info = g_malloc0(sizeof(*info));
663     MigrationState *s = migrate_get_current();
664 
665     switch (s->state) {
666     case MIGRATION_STATUS_NONE:
667         /* no migration has happened ever */
668         break;
669     case MIGRATION_STATUS_SETUP:
670         info->has_status = true;
671         info->has_total_time = false;
672         break;
673     case MIGRATION_STATUS_ACTIVE:
674     case MIGRATION_STATUS_CANCELLING:
675         info->has_status = true;
676         info->has_total_time = true;
677         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
678             - s->total_time;
679         info->has_expected_downtime = true;
680         info->expected_downtime = s->expected_downtime;
681         info->has_setup_time = true;
682         info->setup_time = s->setup_time;
683 
684         populate_ram_info(info, s);
685 
686         if (blk_mig_active()) {
687             info->has_disk = true;
688             info->disk = g_malloc0(sizeof(*info->disk));
689             info->disk->transferred = blk_mig_bytes_transferred();
690             info->disk->remaining = blk_mig_bytes_remaining();
691             info->disk->total = blk_mig_bytes_total();
692         }
693 
694         if (cpu_throttle_active()) {
695             info->has_cpu_throttle_percentage = true;
696             info->cpu_throttle_percentage = cpu_throttle_get_percentage();
697         }
698 
699         get_xbzrle_cache_stats(info);
700         break;
701     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
702         /* Mostly the same as active; TODO add some postcopy stats */
703         info->has_status = true;
704         info->has_total_time = true;
705         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
706             - s->total_time;
707         info->has_expected_downtime = true;
708         info->expected_downtime = s->expected_downtime;
709         info->has_setup_time = true;
710         info->setup_time = s->setup_time;
711 
712         populate_ram_info(info, s);
713 
714         if (blk_mig_active()) {
715             info->has_disk = true;
716             info->disk = g_malloc0(sizeof(*info->disk));
717             info->disk->transferred = blk_mig_bytes_transferred();
718             info->disk->remaining = blk_mig_bytes_remaining();
719             info->disk->total = blk_mig_bytes_total();
720         }
721 
722         get_xbzrle_cache_stats(info);
723         break;
724     case MIGRATION_STATUS_COLO:
725         info->has_status = true;
726         /* TODO: display COLO specific information (checkpoint info etc.) */
727         break;
728     case MIGRATION_STATUS_COMPLETED:
729         get_xbzrle_cache_stats(info);
730 
731         info->has_status = true;
732         info->has_total_time = true;
733         info->total_time = s->total_time;
734         info->has_downtime = true;
735         info->downtime = s->downtime;
736         info->has_setup_time = true;
737         info->setup_time = s->setup_time;
738 
739         populate_ram_info(info, s);
740         break;
741     case MIGRATION_STATUS_FAILED:
742         info->has_status = true;
743         if (s->error) {
744             info->has_error_desc = true;
745             info->error_desc = g_strdup(error_get_pretty(s->error));
746         }
747         break;
748     case MIGRATION_STATUS_CANCELLED:
749         info->has_status = true;
750         break;
751     }
752     info->status = s->state;
753 
754     return info;
755 }
756 
757 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
758                                   Error **errp)
759 {
760     MigrationState *s = migrate_get_current();
761     MigrationCapabilityStatusList *cap;
762     bool old_postcopy_cap = migrate_postcopy_ram();
763 
764     if (migration_is_setup_or_active(s->state)) {
765         error_setg(errp, QERR_MIGRATION_ACTIVE);
766         return;
767     }
768 
769     for (cap = params; cap; cap = cap->next) {
770         if (cap->value->capability == MIGRATION_CAPABILITY_X_COLO) {
771             if (!colo_supported()) {
772                 error_setg(errp, "COLO is not currently supported, please"
773                              " configure with --enable-colo option in order to"
774                              " support COLO feature");
775                 continue;
776             }
777         }
778         s->enabled_capabilities[cap->value->capability] = cap->value->state;
779     }
780 
781     if (migrate_postcopy_ram()) {
782         if (migrate_use_compression()) {
783             /* The decompression threads asynchronously write into RAM
784              * rather than use the atomic copies needed to avoid
785              * userfaulting.  It should be possible to fix the decompression
786              * threads for compatibility in future.
787              */
788             error_report("Postcopy is not currently compatible with "
789                          "compression");
790             s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] =
791                 false;
792         }
793         /* This check is reasonably expensive, so only when it's being
794          * set the first time, also it's only the destination that needs
795          * special support.
796          */
797         if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
798             !postcopy_ram_supported_by_host()) {
799             /* postcopy_ram_supported_by_host will have emitted a more
800              * detailed message
801              */
802             error_report("Postcopy is not supported");
803             s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] =
804                 false;
805         }
806     }
807 }
808 
809 void qmp_migrate_set_parameters(MigrationParameters *params, Error **errp)
810 {
811     MigrationState *s = migrate_get_current();
812 
813     if (params->has_compress_level &&
814         (params->compress_level < 0 || params->compress_level > 9)) {
815         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
816                    "is invalid, it should be in the range of 0 to 9");
817         return;
818     }
819     if (params->has_compress_threads &&
820         (params->compress_threads < 1 || params->compress_threads > 255)) {
821         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
822                    "compress_threads",
823                    "is invalid, it should be in the range of 1 to 255");
824         return;
825     }
826     if (params->has_decompress_threads &&
827         (params->decompress_threads < 1 || params->decompress_threads > 255)) {
828         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
829                    "decompress_threads",
830                    "is invalid, it should be in the range of 1 to 255");
831         return;
832     }
833     if (params->has_cpu_throttle_initial &&
834         (params->cpu_throttle_initial < 1 ||
835          params->cpu_throttle_initial > 99)) {
836         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
837                    "cpu_throttle_initial",
838                    "an integer in the range of 1 to 99");
839         return;
840     }
841     if (params->has_cpu_throttle_increment &&
842         (params->cpu_throttle_increment < 1 ||
843          params->cpu_throttle_increment > 99)) {
844         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
845                    "cpu_throttle_increment",
846                    "an integer in the range of 1 to 99");
847         return;
848     }
849     if (params->has_max_bandwidth &&
850         (params->max_bandwidth < 0 || params->max_bandwidth > SIZE_MAX)) {
851         error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
852                          " range of 0 to %zu bytes/second", SIZE_MAX);
853         return;
854     }
855     if (params->has_downtime_limit &&
856         (params->downtime_limit < 0 ||
857          params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
858         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
859                          "the range of 0 to %d milliseconds",
860                          MAX_MIGRATE_DOWNTIME);
861         return;
862     }
863     if (params->has_x_checkpoint_delay && (params->x_checkpoint_delay < 0)) {
864         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
865                     "x_checkpoint_delay",
866                     "is invalid, it should be positive");
867     }
868 
869     if (params->has_compress_level) {
870         s->parameters.compress_level = params->compress_level;
871     }
872     if (params->has_compress_threads) {
873         s->parameters.compress_threads = params->compress_threads;
874     }
875     if (params->has_decompress_threads) {
876         s->parameters.decompress_threads = params->decompress_threads;
877     }
878     if (params->has_cpu_throttle_initial) {
879         s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
880     }
881     if (params->has_cpu_throttle_increment) {
882         s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
883     }
884     if (params->has_tls_creds) {
885         g_free(s->parameters.tls_creds);
886         s->parameters.tls_creds = g_strdup(params->tls_creds);
887     }
888     if (params->has_tls_hostname) {
889         g_free(s->parameters.tls_hostname);
890         s->parameters.tls_hostname = g_strdup(params->tls_hostname);
891     }
892     if (params->has_max_bandwidth) {
893         s->parameters.max_bandwidth = params->max_bandwidth;
894         if (s->to_dst_file) {
895             qemu_file_set_rate_limit(s->to_dst_file,
896                                 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
897         }
898     }
899     if (params->has_downtime_limit) {
900         s->parameters.downtime_limit = params->downtime_limit;
901     }
902 
903     if (params->has_x_checkpoint_delay) {
904         s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
905         if (migration_in_colo_state()) {
906             colo_checkpoint_notify(s);
907         }
908     }
909 }
910 
911 
912 void qmp_migrate_start_postcopy(Error **errp)
913 {
914     MigrationState *s = migrate_get_current();
915 
916     if (!migrate_postcopy_ram()) {
917         error_setg(errp, "Enable postcopy with migrate_set_capability before"
918                          " the start of migration");
919         return;
920     }
921 
922     if (s->state == MIGRATION_STATUS_NONE) {
923         error_setg(errp, "Postcopy must be started after migration has been"
924                          " started");
925         return;
926     }
927     /*
928      * we don't error if migration has finished since that would be racy
929      * with issuing this command.
930      */
931     atomic_set(&s->start_postcopy, true);
932 }
933 
934 /* shared migration helpers */
935 
936 void migrate_set_state(int *state, int old_state, int new_state)
937 {
938     if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
939         trace_migrate_set_state(new_state);
940         migrate_generate_event(new_state);
941     }
942 }
943 
944 static void migrate_fd_cleanup(void *opaque)
945 {
946     MigrationState *s = opaque;
947 
948     qemu_bh_delete(s->cleanup_bh);
949     s->cleanup_bh = NULL;
950 
951     migration_page_queue_free();
952 
953     if (s->to_dst_file) {
954         trace_migrate_fd_cleanup();
955         qemu_mutex_unlock_iothread();
956         if (s->migration_thread_running) {
957             qemu_thread_join(&s->thread);
958             s->migration_thread_running = false;
959         }
960         qemu_mutex_lock_iothread();
961 
962         migrate_compress_threads_join();
963         qemu_fclose(s->to_dst_file);
964         s->to_dst_file = NULL;
965     }
966 
967     assert((s->state != MIGRATION_STATUS_ACTIVE) &&
968            (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
969 
970     if (s->state == MIGRATION_STATUS_CANCELLING) {
971         migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
972                           MIGRATION_STATUS_CANCELLED);
973     }
974 
975     notifier_list_notify(&migration_state_notifiers, s);
976 }
977 
978 void migrate_fd_error(MigrationState *s, const Error *error)
979 {
980     trace_migrate_fd_error(error_get_pretty(error));
981     assert(s->to_dst_file == NULL);
982     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
983                       MIGRATION_STATUS_FAILED);
984     if (!s->error) {
985         s->error = error_copy(error);
986     }
987     notifier_list_notify(&migration_state_notifiers, s);
988 }
989 
990 static void migrate_fd_cancel(MigrationState *s)
991 {
992     int old_state ;
993     QEMUFile *f = migrate_get_current()->to_dst_file;
994     trace_migrate_fd_cancel();
995 
996     if (s->rp_state.from_dst_file) {
997         /* shutdown the rp socket, so causing the rp thread to shutdown */
998         qemu_file_shutdown(s->rp_state.from_dst_file);
999     }
1000 
1001     do {
1002         old_state = s->state;
1003         if (!migration_is_setup_or_active(old_state)) {
1004             break;
1005         }
1006         migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1007     } while (s->state != MIGRATION_STATUS_CANCELLING);
1008 
1009     /*
1010      * If we're unlucky the migration code might be stuck somewhere in a
1011      * send/write while the network has failed and is waiting to timeout;
1012      * if we've got shutdown(2) available then we can force it to quit.
1013      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1014      * called in a bh, so there is no race against this cancel.
1015      */
1016     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1017         qemu_file_shutdown(f);
1018     }
1019     if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1020         Error *local_err = NULL;
1021 
1022         bdrv_invalidate_cache_all(&local_err);
1023         if (local_err) {
1024             error_report_err(local_err);
1025         } else {
1026             s->block_inactive = false;
1027         }
1028     }
1029 }
1030 
1031 void add_migration_state_change_notifier(Notifier *notify)
1032 {
1033     notifier_list_add(&migration_state_notifiers, notify);
1034 }
1035 
1036 void remove_migration_state_change_notifier(Notifier *notify)
1037 {
1038     notifier_remove(notify);
1039 }
1040 
1041 bool migration_in_setup(MigrationState *s)
1042 {
1043     return s->state == MIGRATION_STATUS_SETUP;
1044 }
1045 
1046 bool migration_has_finished(MigrationState *s)
1047 {
1048     return s->state == MIGRATION_STATUS_COMPLETED;
1049 }
1050 
1051 bool migration_has_failed(MigrationState *s)
1052 {
1053     return (s->state == MIGRATION_STATUS_CANCELLED ||
1054             s->state == MIGRATION_STATUS_FAILED);
1055 }
1056 
1057 bool migration_in_postcopy(void)
1058 {
1059     MigrationState *s = migrate_get_current();
1060 
1061     return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1062 }
1063 
1064 bool migration_in_postcopy_after_devices(MigrationState *s)
1065 {
1066     return migration_in_postcopy() && s->postcopy_after_devices;
1067 }
1068 
1069 bool migration_is_idle(void)
1070 {
1071     MigrationState *s = migrate_get_current();
1072 
1073     switch (s->state) {
1074     case MIGRATION_STATUS_NONE:
1075     case MIGRATION_STATUS_CANCELLED:
1076     case MIGRATION_STATUS_COMPLETED:
1077     case MIGRATION_STATUS_FAILED:
1078         return true;
1079     case MIGRATION_STATUS_SETUP:
1080     case MIGRATION_STATUS_CANCELLING:
1081     case MIGRATION_STATUS_ACTIVE:
1082     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1083     case MIGRATION_STATUS_COLO:
1084         return false;
1085     case MIGRATION_STATUS__MAX:
1086         g_assert_not_reached();
1087     }
1088 
1089     return false;
1090 }
1091 
1092 MigrationState *migrate_init(const MigrationParams *params)
1093 {
1094     MigrationState *s = migrate_get_current();
1095 
1096     /*
1097      * Reinitialise all migration state, except
1098      * parameters/capabilities that the user set, and
1099      * locks.
1100      */
1101     s->bytes_xfer = 0;
1102     s->xfer_limit = 0;
1103     s->cleanup_bh = 0;
1104     s->to_dst_file = NULL;
1105     s->state = MIGRATION_STATUS_NONE;
1106     s->params = *params;
1107     s->rp_state.from_dst_file = NULL;
1108     s->rp_state.error = false;
1109     s->mbps = 0.0;
1110     s->downtime = 0;
1111     s->expected_downtime = 0;
1112     s->setup_time = 0;
1113     s->start_postcopy = false;
1114     s->postcopy_after_devices = false;
1115     s->migration_thread_running = false;
1116     error_free(s->error);
1117     s->error = NULL;
1118 
1119     migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1120 
1121     s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1122     return s;
1123 }
1124 
1125 static GSList *migration_blockers;
1126 
1127 int migrate_add_blocker(Error *reason, Error **errp)
1128 {
1129     if (only_migratable) {
1130         error_propagate(errp, error_copy(reason));
1131         error_prepend(errp, "disallowing migration blocker "
1132                           "(--only_migratable) for: ");
1133         return -EACCES;
1134     }
1135 
1136     if (migration_is_idle()) {
1137         migration_blockers = g_slist_prepend(migration_blockers, reason);
1138         return 0;
1139     }
1140 
1141     error_propagate(errp, error_copy(reason));
1142     error_prepend(errp, "disallowing migration blocker (migration in "
1143                       "progress) for: ");
1144     return -EBUSY;
1145 }
1146 
1147 void migrate_del_blocker(Error *reason)
1148 {
1149     migration_blockers = g_slist_remove(migration_blockers, reason);
1150 }
1151 
1152 int check_migratable(Object *obj, Error **err)
1153 {
1154     DeviceClass *dc = DEVICE_GET_CLASS(obj);
1155     if (only_migratable && dc->vmsd) {
1156         if (dc->vmsd->unmigratable) {
1157             error_setg(err, "Device %s is not migratable, but "
1158                        "--only-migratable was specified",
1159                        object_get_typename(obj));
1160             return -1;
1161         }
1162     }
1163 
1164     return 0;
1165 }
1166 
1167 void qmp_migrate_incoming(const char *uri, Error **errp)
1168 {
1169     Error *local_err = NULL;
1170     static bool once = true;
1171 
1172     if (!deferred_incoming) {
1173         error_setg(errp, "For use with '-incoming defer'");
1174         return;
1175     }
1176     if (!once) {
1177         error_setg(errp, "The incoming migration has already been started");
1178     }
1179 
1180     qemu_start_incoming_migration(uri, &local_err);
1181 
1182     if (local_err) {
1183         error_propagate(errp, local_err);
1184         return;
1185     }
1186 
1187     once = false;
1188 }
1189 
1190 bool migration_is_blocked(Error **errp)
1191 {
1192     if (qemu_savevm_state_blocked(errp)) {
1193         return true;
1194     }
1195 
1196     if (migration_blockers) {
1197         *errp = error_copy(migration_blockers->data);
1198         return true;
1199     }
1200 
1201     return false;
1202 }
1203 
1204 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1205                  bool has_inc, bool inc, bool has_detach, bool detach,
1206                  Error **errp)
1207 {
1208     Error *local_err = NULL;
1209     MigrationState *s = migrate_get_current();
1210     MigrationParams params;
1211     const char *p;
1212 
1213     params.blk = has_blk && blk;
1214     params.shared = has_inc && inc;
1215 
1216     if (migration_is_setup_or_active(s->state) ||
1217         s->state == MIGRATION_STATUS_CANCELLING ||
1218         s->state == MIGRATION_STATUS_COLO) {
1219         error_setg(errp, QERR_MIGRATION_ACTIVE);
1220         return;
1221     }
1222     if (runstate_check(RUN_STATE_INMIGRATE)) {
1223         error_setg(errp, "Guest is waiting for an incoming migration");
1224         return;
1225     }
1226 
1227     if (migration_is_blocked(errp)) {
1228         return;
1229     }
1230 
1231     s = migrate_init(&params);
1232 
1233     if (strstart(uri, "tcp:", &p)) {
1234         tcp_start_outgoing_migration(s, p, &local_err);
1235 #ifdef CONFIG_RDMA
1236     } else if (strstart(uri, "rdma:", &p)) {
1237         rdma_start_outgoing_migration(s, p, &local_err);
1238 #endif
1239     } else if (strstart(uri, "exec:", &p)) {
1240         exec_start_outgoing_migration(s, p, &local_err);
1241     } else if (strstart(uri, "unix:", &p)) {
1242         unix_start_outgoing_migration(s, p, &local_err);
1243     } else if (strstart(uri, "fd:", &p)) {
1244         fd_start_outgoing_migration(s, p, &local_err);
1245     } else {
1246         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1247                    "a valid migration protocol");
1248         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1249                           MIGRATION_STATUS_FAILED);
1250         return;
1251     }
1252 
1253     if (local_err) {
1254         migrate_fd_error(s, local_err);
1255         error_propagate(errp, local_err);
1256         return;
1257     }
1258 }
1259 
1260 void qmp_migrate_cancel(Error **errp)
1261 {
1262     migrate_fd_cancel(migrate_get_current());
1263 }
1264 
1265 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1266 {
1267     MigrationState *s = migrate_get_current();
1268     int64_t new_size;
1269 
1270     /* Check for truncation */
1271     if (value != (size_t)value) {
1272         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1273                    "exceeding address space");
1274         return;
1275     }
1276 
1277     /* Cache should not be larger than guest ram size */
1278     if (value > ram_bytes_total()) {
1279         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1280                    "exceeds guest ram size ");
1281         return;
1282     }
1283 
1284     new_size = xbzrle_cache_resize(value);
1285     if (new_size < 0) {
1286         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1287                    "is smaller than page size");
1288         return;
1289     }
1290 
1291     s->xbzrle_cache_size = new_size;
1292 }
1293 
1294 int64_t qmp_query_migrate_cache_size(Error **errp)
1295 {
1296     return migrate_xbzrle_cache_size();
1297 }
1298 
1299 void qmp_migrate_set_speed(int64_t value, Error **errp)
1300 {
1301     MigrationParameters p = {
1302         .has_max_bandwidth = true,
1303         .max_bandwidth = value,
1304     };
1305 
1306     qmp_migrate_set_parameters(&p, errp);
1307 }
1308 
1309 void qmp_migrate_set_downtime(double value, Error **errp)
1310 {
1311     if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1312         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1313                          "the range of 0 to %d seconds",
1314                          MAX_MIGRATE_DOWNTIME_SECONDS);
1315         return;
1316     }
1317 
1318     value *= 1000; /* Convert to milliseconds */
1319     value = MAX(0, MIN(INT64_MAX, value));
1320 
1321     MigrationParameters p = {
1322         .has_downtime_limit = true,
1323         .downtime_limit = value,
1324     };
1325 
1326     qmp_migrate_set_parameters(&p, errp);
1327 }
1328 
1329 bool migrate_release_ram(void)
1330 {
1331     MigrationState *s;
1332 
1333     s = migrate_get_current();
1334 
1335     return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1336 }
1337 
1338 bool migrate_postcopy_ram(void)
1339 {
1340     MigrationState *s;
1341 
1342     s = migrate_get_current();
1343 
1344     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1345 }
1346 
1347 bool migrate_auto_converge(void)
1348 {
1349     MigrationState *s;
1350 
1351     s = migrate_get_current();
1352 
1353     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1354 }
1355 
1356 bool migrate_zero_blocks(void)
1357 {
1358     MigrationState *s;
1359 
1360     s = migrate_get_current();
1361 
1362     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1363 }
1364 
1365 bool migrate_use_compression(void)
1366 {
1367     MigrationState *s;
1368 
1369     s = migrate_get_current();
1370 
1371     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
1372 }
1373 
1374 int migrate_compress_level(void)
1375 {
1376     MigrationState *s;
1377 
1378     s = migrate_get_current();
1379 
1380     return s->parameters.compress_level;
1381 }
1382 
1383 int migrate_compress_threads(void)
1384 {
1385     MigrationState *s;
1386 
1387     s = migrate_get_current();
1388 
1389     return s->parameters.compress_threads;
1390 }
1391 
1392 int migrate_decompress_threads(void)
1393 {
1394     MigrationState *s;
1395 
1396     s = migrate_get_current();
1397 
1398     return s->parameters.decompress_threads;
1399 }
1400 
1401 bool migrate_use_events(void)
1402 {
1403     MigrationState *s;
1404 
1405     s = migrate_get_current();
1406 
1407     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1408 }
1409 
1410 int migrate_use_xbzrle(void)
1411 {
1412     MigrationState *s;
1413 
1414     s = migrate_get_current();
1415 
1416     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1417 }
1418 
1419 int64_t migrate_xbzrle_cache_size(void)
1420 {
1421     MigrationState *s;
1422 
1423     s = migrate_get_current();
1424 
1425     return s->xbzrle_cache_size;
1426 }
1427 
1428 /* migration thread support */
1429 /*
1430  * Something bad happened to the RP stream, mark an error
1431  * The caller shall print or trace something to indicate why
1432  */
1433 static void mark_source_rp_bad(MigrationState *s)
1434 {
1435     s->rp_state.error = true;
1436 }
1437 
1438 static struct rp_cmd_args {
1439     ssize_t     len; /* -1 = variable */
1440     const char *name;
1441 } rp_cmd_args[] = {
1442     [MIG_RP_MSG_INVALID]        = { .len = -1, .name = "INVALID" },
1443     [MIG_RP_MSG_SHUT]           = { .len =  4, .name = "SHUT" },
1444     [MIG_RP_MSG_PONG]           = { .len =  4, .name = "PONG" },
1445     [MIG_RP_MSG_REQ_PAGES]      = { .len = 12, .name = "REQ_PAGES" },
1446     [MIG_RP_MSG_REQ_PAGES_ID]   = { .len = -1, .name = "REQ_PAGES_ID" },
1447     [MIG_RP_MSG_MAX]            = { .len = -1, .name = "MAX" },
1448 };
1449 
1450 /*
1451  * Process a request for pages received on the return path,
1452  * We're allowed to send more than requested (e.g. to round to our page size)
1453  * and we don't need to send pages that have already been sent.
1454  */
1455 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1456                                        ram_addr_t start, size_t len)
1457 {
1458     long our_host_ps = getpagesize();
1459 
1460     trace_migrate_handle_rp_req_pages(rbname, start, len);
1461 
1462     /*
1463      * Since we currently insist on matching page sizes, just sanity check
1464      * we're being asked for whole host pages.
1465      */
1466     if (start & (our_host_ps-1) ||
1467        (len & (our_host_ps-1))) {
1468         error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1469                      " len: %zd", __func__, start, len);
1470         mark_source_rp_bad(ms);
1471         return;
1472     }
1473 
1474     if (ram_save_queue_pages(rbname, start, len)) {
1475         mark_source_rp_bad(ms);
1476     }
1477 }
1478 
1479 /*
1480  * Handles messages sent on the return path towards the source VM
1481  *
1482  */
1483 static void *source_return_path_thread(void *opaque)
1484 {
1485     MigrationState *ms = opaque;
1486     QEMUFile *rp = ms->rp_state.from_dst_file;
1487     uint16_t header_len, header_type;
1488     uint8_t buf[512];
1489     uint32_t tmp32, sibling_error;
1490     ram_addr_t start = 0; /* =0 to silence warning */
1491     size_t  len = 0, expected_len;
1492     int res;
1493 
1494     trace_source_return_path_thread_entry();
1495     while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
1496            migration_is_setup_or_active(ms->state)) {
1497         trace_source_return_path_thread_loop_top();
1498         header_type = qemu_get_be16(rp);
1499         header_len = qemu_get_be16(rp);
1500 
1501         if (header_type >= MIG_RP_MSG_MAX ||
1502             header_type == MIG_RP_MSG_INVALID) {
1503             error_report("RP: Received invalid message 0x%04x length 0x%04x",
1504                     header_type, header_len);
1505             mark_source_rp_bad(ms);
1506             goto out;
1507         }
1508 
1509         if ((rp_cmd_args[header_type].len != -1 &&
1510             header_len != rp_cmd_args[header_type].len) ||
1511             header_len > sizeof(buf)) {
1512             error_report("RP: Received '%s' message (0x%04x) with"
1513                     "incorrect length %d expecting %zu",
1514                     rp_cmd_args[header_type].name, header_type, header_len,
1515                     (size_t)rp_cmd_args[header_type].len);
1516             mark_source_rp_bad(ms);
1517             goto out;
1518         }
1519 
1520         /* We know we've got a valid header by this point */
1521         res = qemu_get_buffer(rp, buf, header_len);
1522         if (res != header_len) {
1523             error_report("RP: Failed reading data for message 0x%04x"
1524                          " read %d expected %d",
1525                          header_type, res, header_len);
1526             mark_source_rp_bad(ms);
1527             goto out;
1528         }
1529 
1530         /* OK, we have the message and the data */
1531         switch (header_type) {
1532         case MIG_RP_MSG_SHUT:
1533             sibling_error = ldl_be_p(buf);
1534             trace_source_return_path_thread_shut(sibling_error);
1535             if (sibling_error) {
1536                 error_report("RP: Sibling indicated error %d", sibling_error);
1537                 mark_source_rp_bad(ms);
1538             }
1539             /*
1540              * We'll let the main thread deal with closing the RP
1541              * we could do a shutdown(2) on it, but we're the only user
1542              * anyway, so there's nothing gained.
1543              */
1544             goto out;
1545 
1546         case MIG_RP_MSG_PONG:
1547             tmp32 = ldl_be_p(buf);
1548             trace_source_return_path_thread_pong(tmp32);
1549             break;
1550 
1551         case MIG_RP_MSG_REQ_PAGES:
1552             start = ldq_be_p(buf);
1553             len = ldl_be_p(buf + 8);
1554             migrate_handle_rp_req_pages(ms, NULL, start, len);
1555             break;
1556 
1557         case MIG_RP_MSG_REQ_PAGES_ID:
1558             expected_len = 12 + 1; /* header + termination */
1559 
1560             if (header_len >= expected_len) {
1561                 start = ldq_be_p(buf);
1562                 len = ldl_be_p(buf + 8);
1563                 /* Now we expect an idstr */
1564                 tmp32 = buf[12]; /* Length of the following idstr */
1565                 buf[13 + tmp32] = '\0';
1566                 expected_len += tmp32;
1567             }
1568             if (header_len != expected_len) {
1569                 error_report("RP: Req_Page_id with length %d expecting %zd",
1570                         header_len, expected_len);
1571                 mark_source_rp_bad(ms);
1572                 goto out;
1573             }
1574             migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
1575             break;
1576 
1577         default:
1578             break;
1579         }
1580     }
1581     if (qemu_file_get_error(rp)) {
1582         trace_source_return_path_thread_bad_end();
1583         mark_source_rp_bad(ms);
1584     }
1585 
1586     trace_source_return_path_thread_end();
1587 out:
1588     ms->rp_state.from_dst_file = NULL;
1589     qemu_fclose(rp);
1590     return NULL;
1591 }
1592 
1593 static int open_return_path_on_source(MigrationState *ms)
1594 {
1595 
1596     ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
1597     if (!ms->rp_state.from_dst_file) {
1598         return -1;
1599     }
1600 
1601     trace_open_return_path_on_source();
1602     qemu_thread_create(&ms->rp_state.rp_thread, "return path",
1603                        source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
1604 
1605     trace_open_return_path_on_source_continue();
1606 
1607     return 0;
1608 }
1609 
1610 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1611 static int await_return_path_close_on_source(MigrationState *ms)
1612 {
1613     /*
1614      * If this is a normal exit then the destination will send a SHUT and the
1615      * rp_thread will exit, however if there's an error we need to cause
1616      * it to exit.
1617      */
1618     if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
1619         /*
1620          * shutdown(2), if we have it, will cause it to unblock if it's stuck
1621          * waiting for the destination.
1622          */
1623         qemu_file_shutdown(ms->rp_state.from_dst_file);
1624         mark_source_rp_bad(ms);
1625     }
1626     trace_await_return_path_close_on_source_joining();
1627     qemu_thread_join(&ms->rp_state.rp_thread);
1628     trace_await_return_path_close_on_source_close();
1629     return ms->rp_state.error;
1630 }
1631 
1632 /*
1633  * Switch from normal iteration to postcopy
1634  * Returns non-0 on error
1635  */
1636 static int postcopy_start(MigrationState *ms, bool *old_vm_running)
1637 {
1638     int ret;
1639     QIOChannelBuffer *bioc;
1640     QEMUFile *fb;
1641     int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1642     bool restart_block = false;
1643     migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
1644                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
1645 
1646     trace_postcopy_start();
1647     qemu_mutex_lock_iothread();
1648     trace_postcopy_start_set_run();
1649 
1650     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1651     *old_vm_running = runstate_is_running();
1652     global_state_store();
1653     ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1654     if (ret < 0) {
1655         goto fail;
1656     }
1657 
1658     ret = bdrv_inactivate_all();
1659     if (ret < 0) {
1660         goto fail;
1661     }
1662     restart_block = true;
1663 
1664     /*
1665      * Cause any non-postcopiable, but iterative devices to
1666      * send out their final data.
1667      */
1668     qemu_savevm_state_complete_precopy(ms->to_dst_file, true);
1669 
1670     /*
1671      * in Finish migrate and with the io-lock held everything should
1672      * be quiet, but we've potentially still got dirty pages and we
1673      * need to tell the destination to throw any pages it's already received
1674      * that are dirty
1675      */
1676     if (ram_postcopy_send_discard_bitmap(ms)) {
1677         error_report("postcopy send discard bitmap failed");
1678         goto fail;
1679     }
1680 
1681     /*
1682      * send rest of state - note things that are doing postcopy
1683      * will notice we're in POSTCOPY_ACTIVE and not actually
1684      * wrap their state up here
1685      */
1686     qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
1687     /* Ping just for debugging, helps line traces up */
1688     qemu_savevm_send_ping(ms->to_dst_file, 2);
1689 
1690     /*
1691      * While loading the device state we may trigger page transfer
1692      * requests and the fd must be free to process those, and thus
1693      * the destination must read the whole device state off the fd before
1694      * it starts processing it.  Unfortunately the ad-hoc migration format
1695      * doesn't allow the destination to know the size to read without fully
1696      * parsing it through each devices load-state code (especially the open
1697      * coded devices that use get/put).
1698      * So we wrap the device state up in a package with a length at the start;
1699      * to do this we use a qemu_buf to hold the whole of the device state.
1700      */
1701     bioc = qio_channel_buffer_new(4096);
1702     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
1703     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
1704     object_unref(OBJECT(bioc));
1705 
1706     /*
1707      * Make sure the receiver can get incoming pages before we send the rest
1708      * of the state
1709      */
1710     qemu_savevm_send_postcopy_listen(fb);
1711 
1712     qemu_savevm_state_complete_precopy(fb, false);
1713     qemu_savevm_send_ping(fb, 3);
1714 
1715     qemu_savevm_send_postcopy_run(fb);
1716 
1717     /* <><> end of stuff going into the package */
1718 
1719     /* Last point of recovery; as soon as we send the package the destination
1720      * can open devices and potentially start running.
1721      * Lets just check again we've not got any errors.
1722      */
1723     ret = qemu_file_get_error(ms->to_dst_file);
1724     if (ret) {
1725         error_report("postcopy_start: Migration stream errored (pre package)");
1726         goto fail_closefb;
1727     }
1728 
1729     restart_block = false;
1730 
1731     /* Now send that blob */
1732     if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
1733         goto fail_closefb;
1734     }
1735     qemu_fclose(fb);
1736 
1737     /* Send a notify to give a chance for anything that needs to happen
1738      * at the transition to postcopy and after the device state; in particular
1739      * spice needs to trigger a transition now
1740      */
1741     ms->postcopy_after_devices = true;
1742     notifier_list_notify(&migration_state_notifiers, ms);
1743 
1744     ms->downtime =  qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
1745 
1746     qemu_mutex_unlock_iothread();
1747 
1748     /*
1749      * Although this ping is just for debug, it could potentially be
1750      * used for getting a better measurement of downtime at the source.
1751      */
1752     qemu_savevm_send_ping(ms->to_dst_file, 4);
1753 
1754     if (migrate_release_ram()) {
1755         ram_postcopy_migrated_memory_release(ms);
1756     }
1757 
1758     ret = qemu_file_get_error(ms->to_dst_file);
1759     if (ret) {
1760         error_report("postcopy_start: Migration stream errored");
1761         migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1762                               MIGRATION_STATUS_FAILED);
1763     }
1764 
1765     return ret;
1766 
1767 fail_closefb:
1768     qemu_fclose(fb);
1769 fail:
1770     migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1771                           MIGRATION_STATUS_FAILED);
1772     if (restart_block) {
1773         /* A failure happened early enough that we know the destination hasn't
1774          * accessed block devices, so we're safe to recover.
1775          */
1776         Error *local_err = NULL;
1777 
1778         bdrv_invalidate_cache_all(&local_err);
1779         if (local_err) {
1780             error_report_err(local_err);
1781         }
1782     }
1783     qemu_mutex_unlock_iothread();
1784     return -1;
1785 }
1786 
1787 /**
1788  * migration_completion: Used by migration_thread when there's not much left.
1789  *   The caller 'breaks' the loop when this returns.
1790  *
1791  * @s: Current migration state
1792  * @current_active_state: The migration state we expect to be in
1793  * @*old_vm_running: Pointer to old_vm_running flag
1794  * @*start_time: Pointer to time to update
1795  */
1796 static void migration_completion(MigrationState *s, int current_active_state,
1797                                  bool *old_vm_running,
1798                                  int64_t *start_time)
1799 {
1800     int ret;
1801 
1802     if (s->state == MIGRATION_STATUS_ACTIVE) {
1803         qemu_mutex_lock_iothread();
1804         *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1805         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1806         *old_vm_running = runstate_is_running();
1807         ret = global_state_store();
1808 
1809         if (!ret) {
1810             ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1811             /*
1812              * Don't mark the image with BDRV_O_INACTIVE flag if
1813              * we will go into COLO stage later.
1814              */
1815             if (ret >= 0 && !migrate_colo_enabled()) {
1816                 ret = bdrv_inactivate_all();
1817             }
1818             if (ret >= 0) {
1819                 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
1820                 qemu_savevm_state_complete_precopy(s->to_dst_file, false);
1821                 s->block_inactive = true;
1822             }
1823         }
1824         qemu_mutex_unlock_iothread();
1825 
1826         if (ret < 0) {
1827             goto fail;
1828         }
1829     } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1830         trace_migration_completion_postcopy_end();
1831 
1832         qemu_savevm_state_complete_postcopy(s->to_dst_file);
1833         trace_migration_completion_postcopy_end_after_complete();
1834     }
1835 
1836     /*
1837      * If rp was opened we must clean up the thread before
1838      * cleaning everything else up (since if there are no failures
1839      * it will wait for the destination to send it's status in
1840      * a SHUT command).
1841      * Postcopy opens rp if enabled (even if it's not avtivated)
1842      */
1843     if (migrate_postcopy_ram()) {
1844         int rp_error;
1845         trace_migration_completion_postcopy_end_before_rp();
1846         rp_error = await_return_path_close_on_source(s);
1847         trace_migration_completion_postcopy_end_after_rp(rp_error);
1848         if (rp_error) {
1849             goto fail_invalidate;
1850         }
1851     }
1852 
1853     if (qemu_file_get_error(s->to_dst_file)) {
1854         trace_migration_completion_file_err();
1855         goto fail_invalidate;
1856     }
1857 
1858     if (!migrate_colo_enabled()) {
1859         migrate_set_state(&s->state, current_active_state,
1860                           MIGRATION_STATUS_COMPLETED);
1861     }
1862 
1863     return;
1864 
1865 fail_invalidate:
1866     /* If not doing postcopy, vm_start() will be called: let's regain
1867      * control on images.
1868      */
1869     if (s->state == MIGRATION_STATUS_ACTIVE) {
1870         Error *local_err = NULL;
1871 
1872         qemu_mutex_lock_iothread();
1873         bdrv_invalidate_cache_all(&local_err);
1874         if (local_err) {
1875             error_report_err(local_err);
1876         } else {
1877             s->block_inactive = false;
1878         }
1879         qemu_mutex_unlock_iothread();
1880     }
1881 
1882 fail:
1883     migrate_set_state(&s->state, current_active_state,
1884                       MIGRATION_STATUS_FAILED);
1885 }
1886 
1887 bool migrate_colo_enabled(void)
1888 {
1889     MigrationState *s = migrate_get_current();
1890     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
1891 }
1892 
1893 /*
1894  * Master migration thread on the source VM.
1895  * It drives the migration and pumps the data down the outgoing channel.
1896  */
1897 static void *migration_thread(void *opaque)
1898 {
1899     MigrationState *s = opaque;
1900     /* Used by the bandwidth calcs, updated later */
1901     int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1902     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
1903     int64_t initial_bytes = 0;
1904     /*
1905      * The final stage happens when the remaining data is smaller than
1906      * this threshold; it's calculated from the requested downtime and
1907      * measured bandwidth
1908      */
1909     int64_t threshold_size = 0;
1910     int64_t start_time = initial_time;
1911     int64_t end_time;
1912     bool old_vm_running = false;
1913     bool entered_postcopy = false;
1914     /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
1915     enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE;
1916     bool enable_colo = migrate_colo_enabled();
1917 
1918     rcu_register_thread();
1919 
1920     qemu_savevm_state_header(s->to_dst_file);
1921 
1922     if (migrate_postcopy_ram()) {
1923         /* Now tell the dest that it should open its end so it can reply */
1924         qemu_savevm_send_open_return_path(s->to_dst_file);
1925 
1926         /* And do a ping that will make stuff easier to debug */
1927         qemu_savevm_send_ping(s->to_dst_file, 1);
1928 
1929         /*
1930          * Tell the destination that we *might* want to do postcopy later;
1931          * if the other end can't do postcopy it should fail now, nice and
1932          * early.
1933          */
1934         qemu_savevm_send_postcopy_advise(s->to_dst_file);
1935     }
1936 
1937     qemu_savevm_state_begin(s->to_dst_file, &s->params);
1938 
1939     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
1940     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1941                       MIGRATION_STATUS_ACTIVE);
1942 
1943     trace_migration_thread_setup_complete();
1944 
1945     while (s->state == MIGRATION_STATUS_ACTIVE ||
1946            s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1947         int64_t current_time;
1948         uint64_t pending_size;
1949 
1950         if (!qemu_file_rate_limit(s->to_dst_file)) {
1951             uint64_t pend_post, pend_nonpost;
1952 
1953             qemu_savevm_state_pending(s->to_dst_file, threshold_size,
1954                                       &pend_nonpost, &pend_post);
1955             pending_size = pend_nonpost + pend_post;
1956             trace_migrate_pending(pending_size, threshold_size,
1957                                   pend_post, pend_nonpost);
1958             if (pending_size && pending_size >= threshold_size) {
1959                 /* Still a significant amount to transfer */
1960 
1961                 if (migrate_postcopy_ram() &&
1962                     s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE &&
1963                     pend_nonpost <= threshold_size &&
1964                     atomic_read(&s->start_postcopy)) {
1965 
1966                     if (!postcopy_start(s, &old_vm_running)) {
1967                         current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE;
1968                         entered_postcopy = true;
1969                     }
1970 
1971                     continue;
1972                 }
1973                 /* Just another iteration step */
1974                 qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy);
1975             } else {
1976                 trace_migration_thread_low_pending(pending_size);
1977                 migration_completion(s, current_active_state,
1978                                      &old_vm_running, &start_time);
1979                 break;
1980             }
1981         }
1982 
1983         if (qemu_file_get_error(s->to_dst_file)) {
1984             migrate_set_state(&s->state, current_active_state,
1985                               MIGRATION_STATUS_FAILED);
1986             trace_migration_thread_file_err();
1987             break;
1988         }
1989         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1990         if (current_time >= initial_time + BUFFER_DELAY) {
1991             uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) -
1992                                          initial_bytes;
1993             uint64_t time_spent = current_time - initial_time;
1994             double bandwidth = (double)transferred_bytes / time_spent;
1995             threshold_size = bandwidth * s->parameters.downtime_limit;
1996 
1997             s->mbps = (((double) transferred_bytes * 8.0) /
1998                     ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
1999 
2000             trace_migrate_transferred(transferred_bytes, time_spent,
2001                                       bandwidth, threshold_size);
2002             /* if we haven't sent anything, we don't want to recalculate
2003                10000 is a small enough number for our purposes */
2004             if (ram_dirty_pages_rate() && transferred_bytes > 10000) {
2005                 s->expected_downtime = ram_dirty_pages_rate() *
2006                     qemu_target_page_size() / bandwidth;
2007             }
2008 
2009             qemu_file_reset_rate_limit(s->to_dst_file);
2010             initial_time = current_time;
2011             initial_bytes = qemu_ftell(s->to_dst_file);
2012         }
2013         if (qemu_file_rate_limit(s->to_dst_file)) {
2014             /* usleep expects microseconds */
2015             g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
2016         }
2017     }
2018 
2019     trace_migration_thread_after_loop();
2020     /* If we enabled cpu throttling for auto-converge, turn it off. */
2021     cpu_throttle_stop();
2022     end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2023 
2024     qemu_mutex_lock_iothread();
2025     /*
2026      * The resource has been allocated by migration will be reused in COLO
2027      * process, so don't release them.
2028      */
2029     if (!enable_colo) {
2030         qemu_savevm_state_cleanup();
2031     }
2032     if (s->state == MIGRATION_STATUS_COMPLETED) {
2033         uint64_t transferred_bytes = qemu_ftell(s->to_dst_file);
2034         s->total_time = end_time - s->total_time;
2035         if (!entered_postcopy) {
2036             s->downtime = end_time - start_time;
2037         }
2038         if (s->total_time) {
2039             s->mbps = (((double) transferred_bytes * 8.0) /
2040                        ((double) s->total_time)) / 1000;
2041         }
2042         runstate_set(RUN_STATE_POSTMIGRATE);
2043     } else {
2044         if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) {
2045             migrate_start_colo_process(s);
2046             qemu_savevm_state_cleanup();
2047             /*
2048             * Fixme: we will run VM in COLO no matter its old running state.
2049             * After exited COLO, we will keep running.
2050             */
2051             old_vm_running = true;
2052         }
2053         if (old_vm_running && !entered_postcopy) {
2054             vm_start();
2055         } else {
2056             if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2057                 runstate_set(RUN_STATE_POSTMIGRATE);
2058             }
2059         }
2060     }
2061     qemu_bh_schedule(s->cleanup_bh);
2062     qemu_mutex_unlock_iothread();
2063 
2064     rcu_unregister_thread();
2065     return NULL;
2066 }
2067 
2068 void migrate_fd_connect(MigrationState *s)
2069 {
2070     s->expected_downtime = s->parameters.downtime_limit;
2071     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
2072 
2073     qemu_file_set_blocking(s->to_dst_file, true);
2074     qemu_file_set_rate_limit(s->to_dst_file,
2075                              s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
2076 
2077     /* Notify before starting migration thread */
2078     notifier_list_notify(&migration_state_notifiers, s);
2079 
2080     /*
2081      * Open the return path; currently for postcopy but other things might
2082      * also want it.
2083      */
2084     if (migrate_postcopy_ram()) {
2085         if (open_return_path_on_source(s)) {
2086             error_report("Unable to open return-path for postcopy");
2087             migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2088                               MIGRATION_STATUS_FAILED);
2089             migrate_fd_cleanup(s);
2090             return;
2091         }
2092     }
2093 
2094     migrate_compress_threads_create();
2095     qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
2096                        QEMU_THREAD_JOINABLE);
2097     s->migration_thread_running = true;
2098 }
2099 
2100 PostcopyState  postcopy_state_get(void)
2101 {
2102     return atomic_mb_read(&incoming_postcopy_state);
2103 }
2104 
2105 /* Set the state and return the old state */
2106 PostcopyState postcopy_state_set(PostcopyState new_state)
2107 {
2108     return atomic_xchg(&incoming_postcopy_state, new_state);
2109 }
2110 
2111