xref: /openbmc/qemu/migration/migration.c (revision 795c40b8)
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/blocker.h"
21 #include "migration/migration.h"
22 #include "migration/qemu-file.h"
23 #include "sysemu/sysemu.h"
24 #include "block/block.h"
25 #include "qapi/qmp/qerror.h"
26 #include "qapi/util.h"
27 #include "qemu/sockets.h"
28 #include "qemu/rcu.h"
29 #include "migration/block.h"
30 #include "postcopy-ram.h"
31 #include "qemu/thread.h"
32 #include "qmp-commands.h"
33 #include "trace.h"
34 #include "qapi-event.h"
35 #include "qom/cpu.h"
36 #include "exec/memory.h"
37 #include "exec/address-spaces.h"
38 #include "io/channel-buffer.h"
39 #include "io/channel-tls.h"
40 #include "migration/colo.h"
41 
42 #define MAX_THROTTLE  (32 << 20)      /* Migration transfer speed throttling */
43 
44 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
45  * data. */
46 #define BUFFER_DELAY     100
47 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
48 
49 /* Time in milliseconds we are allowed to stop the source,
50  * for sending the last part */
51 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
52 
53 /* Maximum migrate downtime set to 2000 seconds */
54 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
55 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
56 
57 /* Default compression thread count */
58 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
59 /* Default decompression thread count, usually decompression is at
60  * least 4 times as fast as compression.*/
61 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
62 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
63 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
64 /* Define default autoconverge cpu throttle migration parameters */
65 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
66 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
67 
68 /* Migration XBZRLE default cache size */
69 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
70 
71 /* The delay time (in ms) between two COLO checkpoints
72  * Note: Please change this default value to 10000 when we support hybrid mode.
73  */
74 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
75 
76 static NotifierList migration_state_notifiers =
77     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
78 
79 static bool deferred_incoming;
80 
81 /*
82  * Current state of incoming postcopy; note this is not part of
83  * MigrationIncomingState since it's state is used during cleanup
84  * at the end as MIS is being freed.
85  */
86 static PostcopyState incoming_postcopy_state;
87 
88 /* When we add fault tolerance, we could have several
89    migrations at once.  For now we don't need to add
90    dynamic creation of migration */
91 
92 /* For outgoing */
93 MigrationState *migrate_get_current(void)
94 {
95     static bool once;
96     static MigrationState current_migration = {
97         .state = MIGRATION_STATUS_NONE,
98         .xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE,
99         .mbps = -1,
100         .parameters = {
101             .compress_level = DEFAULT_MIGRATE_COMPRESS_LEVEL,
102             .compress_threads = DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT,
103             .decompress_threads = DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT,
104             .cpu_throttle_initial = DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL,
105             .cpu_throttle_increment = DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT,
106             .max_bandwidth = MAX_THROTTLE,
107             .downtime_limit = DEFAULT_MIGRATE_SET_DOWNTIME,
108             .x_checkpoint_delay = DEFAULT_MIGRATE_X_CHECKPOINT_DELAY,
109         },
110     };
111 
112     if (!once) {
113         current_migration.parameters.tls_creds = g_strdup("");
114         current_migration.parameters.tls_hostname = g_strdup("");
115         once = true;
116     }
117     return &current_migration;
118 }
119 
120 MigrationIncomingState *migration_incoming_get_current(void)
121 {
122     static bool once;
123     static MigrationIncomingState mis_current;
124 
125     if (!once) {
126         mis_current.state = MIGRATION_STATUS_NONE;
127         memset(&mis_current, 0, sizeof(MigrationIncomingState));
128         QLIST_INIT(&mis_current.loadvm_handlers);
129         qemu_mutex_init(&mis_current.rp_mutex);
130         qemu_event_init(&mis_current.main_thread_load_event, false);
131         once = true;
132     }
133     return &mis_current;
134 }
135 
136 void migration_incoming_state_destroy(void)
137 {
138     struct MigrationIncomingState *mis = migration_incoming_get_current();
139 
140     qemu_event_destroy(&mis->main_thread_load_event);
141     loadvm_free_handlers(mis);
142 }
143 
144 
145 typedef struct {
146     bool optional;
147     uint32_t size;
148     uint8_t runstate[100];
149     RunState state;
150     bool received;
151 } GlobalState;
152 
153 static GlobalState global_state;
154 
155 int global_state_store(void)
156 {
157     if (!runstate_store((char *)global_state.runstate,
158                         sizeof(global_state.runstate))) {
159         error_report("runstate name too big: %s", global_state.runstate);
160         trace_migrate_state_too_big();
161         return -EINVAL;
162     }
163     return 0;
164 }
165 
166 void global_state_store_running(void)
167 {
168     const char *state = RunState_lookup[RUN_STATE_RUNNING];
169     strncpy((char *)global_state.runstate,
170            state, sizeof(global_state.runstate));
171 }
172 
173 static bool global_state_received(void)
174 {
175     return global_state.received;
176 }
177 
178 static RunState global_state_get_runstate(void)
179 {
180     return global_state.state;
181 }
182 
183 void global_state_set_optional(void)
184 {
185     global_state.optional = true;
186 }
187 
188 static bool global_state_needed(void *opaque)
189 {
190     GlobalState *s = opaque;
191     char *runstate = (char *)s->runstate;
192 
193     /* If it is not optional, it is mandatory */
194 
195     if (s->optional == false) {
196         return true;
197     }
198 
199     /* If state is running or paused, it is not needed */
200 
201     if (strcmp(runstate, "running") == 0 ||
202         strcmp(runstate, "paused") == 0) {
203         return false;
204     }
205 
206     /* for any other state it is needed */
207     return true;
208 }
209 
210 static int global_state_post_load(void *opaque, int version_id)
211 {
212     GlobalState *s = opaque;
213     Error *local_err = NULL;
214     int r;
215     char *runstate = (char *)s->runstate;
216 
217     s->received = true;
218     trace_migrate_global_state_post_load(runstate);
219 
220     r = qapi_enum_parse(RunState_lookup, runstate, RUN_STATE__MAX,
221                                 -1, &local_err);
222 
223     if (r == -1) {
224         if (local_err) {
225             error_report_err(local_err);
226         }
227         return -EINVAL;
228     }
229     s->state = r;
230 
231     return 0;
232 }
233 
234 static void global_state_pre_save(void *opaque)
235 {
236     GlobalState *s = opaque;
237 
238     trace_migrate_global_state_pre_save((char *)s->runstate);
239     s->size = strlen((char *)s->runstate) + 1;
240 }
241 
242 static const VMStateDescription vmstate_globalstate = {
243     .name = "globalstate",
244     .version_id = 1,
245     .minimum_version_id = 1,
246     .post_load = global_state_post_load,
247     .pre_save = global_state_pre_save,
248     .needed = global_state_needed,
249     .fields = (VMStateField[]) {
250         VMSTATE_UINT32(size, GlobalState),
251         VMSTATE_BUFFER(runstate, GlobalState),
252         VMSTATE_END_OF_LIST()
253     },
254 };
255 
256 void register_global_state(void)
257 {
258     /* We would use it independently that we receive it */
259     strcpy((char *)&global_state.runstate, "");
260     global_state.received = false;
261     vmstate_register(NULL, 0, &vmstate_globalstate, &global_state);
262 }
263 
264 static void migrate_generate_event(int new_state)
265 {
266     if (migrate_use_events()) {
267         qapi_event_send_migration(new_state, &error_abort);
268     }
269 }
270 
271 /*
272  * Called on -incoming with a defer: uri.
273  * The migration can be started later after any parameters have been
274  * changed.
275  */
276 static void deferred_incoming_migration(Error **errp)
277 {
278     if (deferred_incoming) {
279         error_setg(errp, "Incoming migration already deferred");
280     }
281     deferred_incoming = true;
282 }
283 
284 /* Request a range of pages from the source VM at the given
285  * start address.
286  *   rbname: Name of the RAMBlock to request the page in, if NULL it's the same
287  *           as the last request (a name must have been given previously)
288  *   Start: Address offset within the RB
289  *   Len: Length in bytes required - must be a multiple of pagesize
290  */
291 void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
292                                ram_addr_t start, size_t len)
293 {
294     uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
295     size_t msglen = 12; /* start + len */
296 
297     *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
298     *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
299 
300     if (rbname) {
301         int rbname_len = strlen(rbname);
302         assert(rbname_len < 256);
303 
304         bufc[msglen++] = rbname_len;
305         memcpy(bufc + msglen, rbname, rbname_len);
306         msglen += rbname_len;
307         migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc);
308     } else {
309         migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc);
310     }
311 }
312 
313 void qemu_start_incoming_migration(const char *uri, Error **errp)
314 {
315     const char *p;
316 
317     qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
318     if (!strcmp(uri, "defer")) {
319         deferred_incoming_migration(errp);
320     } else if (strstart(uri, "tcp:", &p)) {
321         tcp_start_incoming_migration(p, errp);
322 #ifdef CONFIG_RDMA
323     } else if (strstart(uri, "rdma:", &p)) {
324         rdma_start_incoming_migration(p, errp);
325 #endif
326     } else if (strstart(uri, "exec:", &p)) {
327         exec_start_incoming_migration(p, errp);
328     } else if (strstart(uri, "unix:", &p)) {
329         unix_start_incoming_migration(p, errp);
330     } else if (strstart(uri, "fd:", &p)) {
331         fd_start_incoming_migration(p, errp);
332     } else {
333         error_setg(errp, "unknown migration protocol: %s", uri);
334     }
335 }
336 
337 static void process_incoming_migration_bh(void *opaque)
338 {
339     Error *local_err = NULL;
340     MigrationIncomingState *mis = opaque;
341 
342     /* Make sure all file formats flush their mutable metadata.
343      * If we get an error here, just don't restart the VM yet. */
344     bdrv_invalidate_cache_all(&local_err);
345     if (local_err) {
346         error_report_err(local_err);
347         local_err = NULL;
348         autostart = false;
349     }
350 
351     /*
352      * This must happen after all error conditions are dealt with and
353      * we're sure the VM is going to be running on this host.
354      */
355     qemu_announce_self();
356 
357     /* If global state section was not received or we are in running
358        state, we need to obey autostart. Any other state is set with
359        runstate_set. */
360 
361     if (!global_state_received() ||
362         global_state_get_runstate() == RUN_STATE_RUNNING) {
363         if (autostart) {
364             vm_start();
365         } else {
366             runstate_set(RUN_STATE_PAUSED);
367         }
368     } else {
369         runstate_set(global_state_get_runstate());
370     }
371     migrate_decompress_threads_join();
372     /*
373      * This must happen after any state changes since as soon as an external
374      * observer sees this event they might start to prod at the VM assuming
375      * it's ready to use.
376      */
377     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
378                       MIGRATION_STATUS_COMPLETED);
379     qemu_bh_delete(mis->bh);
380     migration_incoming_state_destroy();
381 }
382 
383 static void process_incoming_migration_co(void *opaque)
384 {
385     QEMUFile *f = opaque;
386     MigrationIncomingState *mis = migration_incoming_get_current();
387     PostcopyState ps;
388     int ret;
389 
390     mis->from_src_file = f;
391     mis->largest_page_size = qemu_ram_pagesize_largest();
392     postcopy_state_set(POSTCOPY_INCOMING_NONE);
393     migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
394                       MIGRATION_STATUS_ACTIVE);
395     ret = qemu_loadvm_state(f);
396 
397     ps = postcopy_state_get();
398     trace_process_incoming_migration_co_end(ret, ps);
399     if (ps != POSTCOPY_INCOMING_NONE) {
400         if (ps == POSTCOPY_INCOMING_ADVISE) {
401             /*
402              * Where a migration had postcopy enabled (and thus went to advise)
403              * but managed to complete within the precopy period, we can use
404              * the normal exit.
405              */
406             postcopy_ram_incoming_cleanup(mis);
407         } else if (ret >= 0) {
408             /*
409              * Postcopy was started, cleanup should happen at the end of the
410              * postcopy thread.
411              */
412             trace_process_incoming_migration_co_postcopy_end_main();
413             return;
414         }
415         /* Else if something went wrong then just fall out of the normal exit */
416     }
417 
418     /* we get COLO info, and know if we are in COLO mode */
419     if (!ret && migration_incoming_enable_colo()) {
420         mis->migration_incoming_co = qemu_coroutine_self();
421         qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
422              colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
423         mis->have_colo_incoming_thread = true;
424         qemu_coroutine_yield();
425 
426         /* Wait checkpoint incoming thread exit before free resource */
427         qemu_thread_join(&mis->colo_incoming_thread);
428     }
429 
430     if (ret < 0) {
431         migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
432                           MIGRATION_STATUS_FAILED);
433         error_report("load of migration failed: %s", strerror(-ret));
434         migrate_decompress_threads_join();
435         exit(EXIT_FAILURE);
436     }
437 
438     qemu_fclose(f);
439     free_xbzrle_decoded_buf();
440 
441     mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
442     qemu_bh_schedule(mis->bh);
443 }
444 
445 void migration_fd_process_incoming(QEMUFile *f)
446 {
447     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, f);
448 
449     migrate_decompress_threads_create();
450     qemu_file_set_blocking(f, false);
451     qemu_coroutine_enter(co);
452 }
453 
454 
455 void migration_channel_process_incoming(MigrationState *s,
456                                         QIOChannel *ioc)
457 {
458     trace_migration_set_incoming_channel(
459         ioc, object_get_typename(OBJECT(ioc)));
460 
461     if (s->parameters.tls_creds &&
462         *s->parameters.tls_creds &&
463         !object_dynamic_cast(OBJECT(ioc),
464                              TYPE_QIO_CHANNEL_TLS)) {
465         Error *local_err = NULL;
466         migration_tls_channel_process_incoming(s, ioc, &local_err);
467         if (local_err) {
468             error_report_err(local_err);
469         }
470     } else {
471         QEMUFile *f = qemu_fopen_channel_input(ioc);
472         migration_fd_process_incoming(f);
473     }
474 }
475 
476 
477 void migration_channel_connect(MigrationState *s,
478                                QIOChannel *ioc,
479                                const char *hostname)
480 {
481     trace_migration_set_outgoing_channel(
482         ioc, object_get_typename(OBJECT(ioc)), hostname);
483 
484     if (s->parameters.tls_creds &&
485         *s->parameters.tls_creds &&
486         !object_dynamic_cast(OBJECT(ioc),
487                              TYPE_QIO_CHANNEL_TLS)) {
488         Error *local_err = NULL;
489         migration_tls_channel_connect(s, ioc, hostname, &local_err);
490         if (local_err) {
491             migrate_fd_error(s, local_err);
492             error_free(local_err);
493         }
494     } else {
495         QEMUFile *f = qemu_fopen_channel_output(ioc);
496 
497         s->to_dst_file = f;
498 
499         migrate_fd_connect(s);
500     }
501 }
502 
503 
504 /*
505  * Send a message on the return channel back to the source
506  * of the migration.
507  */
508 void migrate_send_rp_message(MigrationIncomingState *mis,
509                              enum mig_rp_message_type message_type,
510                              uint16_t len, void *data)
511 {
512     trace_migrate_send_rp_message((int)message_type, len);
513     qemu_mutex_lock(&mis->rp_mutex);
514     qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
515     qemu_put_be16(mis->to_src_file, len);
516     qemu_put_buffer(mis->to_src_file, data, len);
517     qemu_fflush(mis->to_src_file);
518     qemu_mutex_unlock(&mis->rp_mutex);
519 }
520 
521 /*
522  * Send a 'SHUT' message on the return channel with the given value
523  * to indicate that we've finished with the RP.  Non-0 value indicates
524  * error.
525  */
526 void migrate_send_rp_shut(MigrationIncomingState *mis,
527                           uint32_t value)
528 {
529     uint32_t buf;
530 
531     buf = cpu_to_be32(value);
532     migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
533 }
534 
535 /*
536  * Send a 'PONG' message on the return channel with the given value
537  * (normally in response to a 'PING')
538  */
539 void migrate_send_rp_pong(MigrationIncomingState *mis,
540                           uint32_t value)
541 {
542     uint32_t buf;
543 
544     buf = cpu_to_be32(value);
545     migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
546 }
547 
548 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
549 {
550     MigrationCapabilityStatusList *head = NULL;
551     MigrationCapabilityStatusList *caps;
552     MigrationState *s = migrate_get_current();
553     int i;
554 
555     caps = NULL; /* silence compiler warning */
556     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
557         if (i == MIGRATION_CAPABILITY_X_COLO && !colo_supported()) {
558             continue;
559         }
560         if (head == NULL) {
561             head = g_malloc0(sizeof(*caps));
562             caps = head;
563         } else {
564             caps->next = g_malloc0(sizeof(*caps));
565             caps = caps->next;
566         }
567         caps->value =
568             g_malloc(sizeof(*caps->value));
569         caps->value->capability = i;
570         caps->value->state = s->enabled_capabilities[i];
571     }
572 
573     return head;
574 }
575 
576 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
577 {
578     MigrationParameters *params;
579     MigrationState *s = migrate_get_current();
580 
581     params = g_malloc0(sizeof(*params));
582     params->has_compress_level = true;
583     params->compress_level = s->parameters.compress_level;
584     params->has_compress_threads = true;
585     params->compress_threads = s->parameters.compress_threads;
586     params->has_decompress_threads = true;
587     params->decompress_threads = s->parameters.decompress_threads;
588     params->has_cpu_throttle_initial = true;
589     params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
590     params->has_cpu_throttle_increment = true;
591     params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
592     params->has_tls_creds = !!s->parameters.tls_creds;
593     params->tls_creds = g_strdup(s->parameters.tls_creds);
594     params->has_tls_hostname = !!s->parameters.tls_hostname;
595     params->tls_hostname = g_strdup(s->parameters.tls_hostname);
596     params->has_max_bandwidth = true;
597     params->max_bandwidth = s->parameters.max_bandwidth;
598     params->has_downtime_limit = true;
599     params->downtime_limit = s->parameters.downtime_limit;
600     params->has_x_checkpoint_delay = true;
601     params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
602 
603     return params;
604 }
605 
606 /*
607  * Return true if we're already in the middle of a migration
608  * (i.e. any of the active or setup states)
609  */
610 static bool migration_is_setup_or_active(int state)
611 {
612     switch (state) {
613     case MIGRATION_STATUS_ACTIVE:
614     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
615     case MIGRATION_STATUS_SETUP:
616         return true;
617 
618     default:
619         return false;
620 
621     }
622 }
623 
624 static void get_xbzrle_cache_stats(MigrationInfo *info)
625 {
626     if (migrate_use_xbzrle()) {
627         info->has_xbzrle_cache = true;
628         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
629         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
630         info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred();
631         info->xbzrle_cache->pages = xbzrle_mig_pages_transferred();
632         info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss();
633         info->xbzrle_cache->cache_miss_rate = xbzrle_mig_cache_miss_rate();
634         info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow();
635     }
636 }
637 
638 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
639 {
640     info->has_ram = true;
641     info->ram = g_malloc0(sizeof(*info->ram));
642     info->ram->transferred = ram_bytes_transferred();
643     info->ram->total = ram_bytes_total();
644     info->ram->duplicate = dup_mig_pages_transferred();
645     /* legacy value.  It is not used anymore */
646     info->ram->skipped = 0;
647     info->ram->normal = norm_mig_pages_transferred();
648     info->ram->normal_bytes = norm_mig_pages_transferred() *
649         qemu_target_page_size();
650     info->ram->mbps = s->mbps;
651     info->ram->dirty_sync_count = ram_dirty_sync_count();
652     info->ram->postcopy_requests = ram_postcopy_requests();
653     info->ram->page_size = qemu_target_page_size();
654 
655     if (s->state != MIGRATION_STATUS_COMPLETED) {
656         info->ram->remaining = ram_bytes_remaining();
657         info->ram->dirty_pages_rate = ram_dirty_pages_rate();
658     }
659 }
660 
661 MigrationInfo *qmp_query_migrate(Error **errp)
662 {
663     MigrationInfo *info = g_malloc0(sizeof(*info));
664     MigrationState *s = migrate_get_current();
665 
666     switch (s->state) {
667     case MIGRATION_STATUS_NONE:
668         /* no migration has happened ever */
669         break;
670     case MIGRATION_STATUS_SETUP:
671         info->has_status = true;
672         info->has_total_time = false;
673         break;
674     case MIGRATION_STATUS_ACTIVE:
675     case MIGRATION_STATUS_CANCELLING:
676         info->has_status = true;
677         info->has_total_time = true;
678         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
679             - s->total_time;
680         info->has_expected_downtime = true;
681         info->expected_downtime = s->expected_downtime;
682         info->has_setup_time = true;
683         info->setup_time = s->setup_time;
684 
685         populate_ram_info(info, s);
686 
687         if (blk_mig_active()) {
688             info->has_disk = true;
689             info->disk = g_malloc0(sizeof(*info->disk));
690             info->disk->transferred = blk_mig_bytes_transferred();
691             info->disk->remaining = blk_mig_bytes_remaining();
692             info->disk->total = blk_mig_bytes_total();
693         }
694 
695         if (cpu_throttle_active()) {
696             info->has_cpu_throttle_percentage = true;
697             info->cpu_throttle_percentage = cpu_throttle_get_percentage();
698         }
699 
700         get_xbzrle_cache_stats(info);
701         break;
702     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
703         /* Mostly the same as active; TODO add some postcopy stats */
704         info->has_status = true;
705         info->has_total_time = true;
706         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
707             - s->total_time;
708         info->has_expected_downtime = true;
709         info->expected_downtime = s->expected_downtime;
710         info->has_setup_time = true;
711         info->setup_time = s->setup_time;
712 
713         populate_ram_info(info, s);
714 
715         if (blk_mig_active()) {
716             info->has_disk = true;
717             info->disk = g_malloc0(sizeof(*info->disk));
718             info->disk->transferred = blk_mig_bytes_transferred();
719             info->disk->remaining = blk_mig_bytes_remaining();
720             info->disk->total = blk_mig_bytes_total();
721         }
722 
723         get_xbzrle_cache_stats(info);
724         break;
725     case MIGRATION_STATUS_COLO:
726         info->has_status = true;
727         /* TODO: display COLO specific information (checkpoint info etc.) */
728         break;
729     case MIGRATION_STATUS_COMPLETED:
730         get_xbzrle_cache_stats(info);
731 
732         info->has_status = true;
733         info->has_total_time = true;
734         info->total_time = s->total_time;
735         info->has_downtime = true;
736         info->downtime = s->downtime;
737         info->has_setup_time = true;
738         info->setup_time = s->setup_time;
739 
740         populate_ram_info(info, s);
741         break;
742     case MIGRATION_STATUS_FAILED:
743         info->has_status = true;
744         if (s->error) {
745             info->has_error_desc = true;
746             info->error_desc = g_strdup(error_get_pretty(s->error));
747         }
748         break;
749     case MIGRATION_STATUS_CANCELLED:
750         info->has_status = true;
751         break;
752     }
753     info->status = s->state;
754 
755     return info;
756 }
757 
758 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
759                                   Error **errp)
760 {
761     MigrationState *s = migrate_get_current();
762     MigrationCapabilityStatusList *cap;
763     bool old_postcopy_cap = migrate_postcopy_ram();
764 
765     if (migration_is_setup_or_active(s->state)) {
766         error_setg(errp, QERR_MIGRATION_ACTIVE);
767         return;
768     }
769 
770     for (cap = params; cap; cap = cap->next) {
771         if (cap->value->capability == MIGRATION_CAPABILITY_X_COLO) {
772             if (!colo_supported()) {
773                 error_setg(errp, "COLO is not currently supported, please"
774                              " configure with --enable-colo option in order to"
775                              " support COLO feature");
776                 continue;
777             }
778         }
779         s->enabled_capabilities[cap->value->capability] = cap->value->state;
780     }
781 
782     if (migrate_postcopy_ram()) {
783         if (migrate_use_compression()) {
784             /* The decompression threads asynchronously write into RAM
785              * rather than use the atomic copies needed to avoid
786              * userfaulting.  It should be possible to fix the decompression
787              * threads for compatibility in future.
788              */
789             error_report("Postcopy is not currently compatible with "
790                          "compression");
791             s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] =
792                 false;
793         }
794         /* This check is reasonably expensive, so only when it's being
795          * set the first time, also it's only the destination that needs
796          * special support.
797          */
798         if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
799             !postcopy_ram_supported_by_host()) {
800             /* postcopy_ram_supported_by_host will have emitted a more
801              * detailed message
802              */
803             error_report("Postcopy is not supported");
804             s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] =
805                 false;
806         }
807     }
808 }
809 
810 void qmp_migrate_set_parameters(MigrationParameters *params, Error **errp)
811 {
812     MigrationState *s = migrate_get_current();
813 
814     if (params->has_compress_level &&
815         (params->compress_level < 0 || params->compress_level > 9)) {
816         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
817                    "is invalid, it should be in the range of 0 to 9");
818         return;
819     }
820     if (params->has_compress_threads &&
821         (params->compress_threads < 1 || params->compress_threads > 255)) {
822         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
823                    "compress_threads",
824                    "is invalid, it should be in the range of 1 to 255");
825         return;
826     }
827     if (params->has_decompress_threads &&
828         (params->decompress_threads < 1 || params->decompress_threads > 255)) {
829         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
830                    "decompress_threads",
831                    "is invalid, it should be in the range of 1 to 255");
832         return;
833     }
834     if (params->has_cpu_throttle_initial &&
835         (params->cpu_throttle_initial < 1 ||
836          params->cpu_throttle_initial > 99)) {
837         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
838                    "cpu_throttle_initial",
839                    "an integer in the range of 1 to 99");
840         return;
841     }
842     if (params->has_cpu_throttle_increment &&
843         (params->cpu_throttle_increment < 1 ||
844          params->cpu_throttle_increment > 99)) {
845         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
846                    "cpu_throttle_increment",
847                    "an integer in the range of 1 to 99");
848         return;
849     }
850     if (params->has_max_bandwidth &&
851         (params->max_bandwidth < 0 || params->max_bandwidth > SIZE_MAX)) {
852         error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
853                          " range of 0 to %zu bytes/second", SIZE_MAX);
854         return;
855     }
856     if (params->has_downtime_limit &&
857         (params->downtime_limit < 0 ||
858          params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
859         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
860                          "the range of 0 to %d milliseconds",
861                          MAX_MIGRATE_DOWNTIME);
862         return;
863     }
864     if (params->has_x_checkpoint_delay && (params->x_checkpoint_delay < 0)) {
865         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
866                     "x_checkpoint_delay",
867                     "is invalid, it should be positive");
868     }
869 
870     if (params->has_compress_level) {
871         s->parameters.compress_level = params->compress_level;
872     }
873     if (params->has_compress_threads) {
874         s->parameters.compress_threads = params->compress_threads;
875     }
876     if (params->has_decompress_threads) {
877         s->parameters.decompress_threads = params->decompress_threads;
878     }
879     if (params->has_cpu_throttle_initial) {
880         s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
881     }
882     if (params->has_cpu_throttle_increment) {
883         s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
884     }
885     if (params->has_tls_creds) {
886         g_free(s->parameters.tls_creds);
887         s->parameters.tls_creds = g_strdup(params->tls_creds);
888     }
889     if (params->has_tls_hostname) {
890         g_free(s->parameters.tls_hostname);
891         s->parameters.tls_hostname = g_strdup(params->tls_hostname);
892     }
893     if (params->has_max_bandwidth) {
894         s->parameters.max_bandwidth = params->max_bandwidth;
895         if (s->to_dst_file) {
896             qemu_file_set_rate_limit(s->to_dst_file,
897                                 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
898         }
899     }
900     if (params->has_downtime_limit) {
901         s->parameters.downtime_limit = params->downtime_limit;
902     }
903 
904     if (params->has_x_checkpoint_delay) {
905         s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
906         if (migration_in_colo_state()) {
907             colo_checkpoint_notify(s);
908         }
909     }
910 }
911 
912 
913 void qmp_migrate_start_postcopy(Error **errp)
914 {
915     MigrationState *s = migrate_get_current();
916 
917     if (!migrate_postcopy_ram()) {
918         error_setg(errp, "Enable postcopy with migrate_set_capability before"
919                          " the start of migration");
920         return;
921     }
922 
923     if (s->state == MIGRATION_STATUS_NONE) {
924         error_setg(errp, "Postcopy must be started after migration has been"
925                          " started");
926         return;
927     }
928     /*
929      * we don't error if migration has finished since that would be racy
930      * with issuing this command.
931      */
932     atomic_set(&s->start_postcopy, true);
933 }
934 
935 /* shared migration helpers */
936 
937 void migrate_set_state(int *state, int old_state, int new_state)
938 {
939     if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
940         trace_migrate_set_state(new_state);
941         migrate_generate_event(new_state);
942     }
943 }
944 
945 static void migrate_fd_cleanup(void *opaque)
946 {
947     MigrationState *s = opaque;
948 
949     qemu_bh_delete(s->cleanup_bh);
950     s->cleanup_bh = NULL;
951 
952     migration_page_queue_free();
953 
954     if (s->to_dst_file) {
955         trace_migrate_fd_cleanup();
956         qemu_mutex_unlock_iothread();
957         if (s->migration_thread_running) {
958             qemu_thread_join(&s->thread);
959             s->migration_thread_running = false;
960         }
961         qemu_mutex_lock_iothread();
962 
963         migrate_compress_threads_join();
964         qemu_fclose(s->to_dst_file);
965         s->to_dst_file = NULL;
966     }
967 
968     assert((s->state != MIGRATION_STATUS_ACTIVE) &&
969            (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
970 
971     if (s->state == MIGRATION_STATUS_CANCELLING) {
972         migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
973                           MIGRATION_STATUS_CANCELLED);
974     }
975 
976     notifier_list_notify(&migration_state_notifiers, s);
977 }
978 
979 void migrate_fd_error(MigrationState *s, const Error *error)
980 {
981     trace_migrate_fd_error(error_get_pretty(error));
982     assert(s->to_dst_file == NULL);
983     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
984                       MIGRATION_STATUS_FAILED);
985     if (!s->error) {
986         s->error = error_copy(error);
987     }
988     notifier_list_notify(&migration_state_notifiers, s);
989 }
990 
991 static void migrate_fd_cancel(MigrationState *s)
992 {
993     int old_state ;
994     QEMUFile *f = migrate_get_current()->to_dst_file;
995     trace_migrate_fd_cancel();
996 
997     if (s->rp_state.from_dst_file) {
998         /* shutdown the rp socket, so causing the rp thread to shutdown */
999         qemu_file_shutdown(s->rp_state.from_dst_file);
1000     }
1001 
1002     do {
1003         old_state = s->state;
1004         if (!migration_is_setup_or_active(old_state)) {
1005             break;
1006         }
1007         migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1008     } while (s->state != MIGRATION_STATUS_CANCELLING);
1009 
1010     /*
1011      * If we're unlucky the migration code might be stuck somewhere in a
1012      * send/write while the network has failed and is waiting to timeout;
1013      * if we've got shutdown(2) available then we can force it to quit.
1014      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1015      * called in a bh, so there is no race against this cancel.
1016      */
1017     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1018         qemu_file_shutdown(f);
1019     }
1020     if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1021         Error *local_err = NULL;
1022 
1023         bdrv_invalidate_cache_all(&local_err);
1024         if (local_err) {
1025             error_report_err(local_err);
1026         } else {
1027             s->block_inactive = false;
1028         }
1029     }
1030 }
1031 
1032 void add_migration_state_change_notifier(Notifier *notify)
1033 {
1034     notifier_list_add(&migration_state_notifiers, notify);
1035 }
1036 
1037 void remove_migration_state_change_notifier(Notifier *notify)
1038 {
1039     notifier_remove(notify);
1040 }
1041 
1042 bool migration_in_setup(MigrationState *s)
1043 {
1044     return s->state == MIGRATION_STATUS_SETUP;
1045 }
1046 
1047 bool migration_has_finished(MigrationState *s)
1048 {
1049     return s->state == MIGRATION_STATUS_COMPLETED;
1050 }
1051 
1052 bool migration_has_failed(MigrationState *s)
1053 {
1054     return (s->state == MIGRATION_STATUS_CANCELLED ||
1055             s->state == MIGRATION_STATUS_FAILED);
1056 }
1057 
1058 bool migration_in_postcopy(void)
1059 {
1060     MigrationState *s = migrate_get_current();
1061 
1062     return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1063 }
1064 
1065 bool migration_in_postcopy_after_devices(MigrationState *s)
1066 {
1067     return migration_in_postcopy() && s->postcopy_after_devices;
1068 }
1069 
1070 bool migration_is_idle(void)
1071 {
1072     MigrationState *s = migrate_get_current();
1073 
1074     switch (s->state) {
1075     case MIGRATION_STATUS_NONE:
1076     case MIGRATION_STATUS_CANCELLED:
1077     case MIGRATION_STATUS_COMPLETED:
1078     case MIGRATION_STATUS_FAILED:
1079         return true;
1080     case MIGRATION_STATUS_SETUP:
1081     case MIGRATION_STATUS_CANCELLING:
1082     case MIGRATION_STATUS_ACTIVE:
1083     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1084     case MIGRATION_STATUS_COLO:
1085         return false;
1086     case MIGRATION_STATUS__MAX:
1087         g_assert_not_reached();
1088     }
1089 
1090     return false;
1091 }
1092 
1093 MigrationState *migrate_init(const MigrationParams *params)
1094 {
1095     MigrationState *s = migrate_get_current();
1096 
1097     /*
1098      * Reinitialise all migration state, except
1099      * parameters/capabilities that the user set, and
1100      * locks.
1101      */
1102     s->bytes_xfer = 0;
1103     s->xfer_limit = 0;
1104     s->cleanup_bh = 0;
1105     s->to_dst_file = NULL;
1106     s->state = MIGRATION_STATUS_NONE;
1107     s->params = *params;
1108     s->rp_state.from_dst_file = NULL;
1109     s->rp_state.error = false;
1110     s->mbps = 0.0;
1111     s->downtime = 0;
1112     s->expected_downtime = 0;
1113     s->setup_time = 0;
1114     s->start_postcopy = false;
1115     s->postcopy_after_devices = false;
1116     s->migration_thread_running = false;
1117     error_free(s->error);
1118     s->error = NULL;
1119 
1120     migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
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()) {
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(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     /*
1906      * The final stage happens when the remaining data is smaller than
1907      * this threshold; it's calculated from the requested downtime and
1908      * measured bandwidth
1909      */
1910     int64_t threshold_size = 0;
1911     int64_t start_time = initial_time;
1912     int64_t end_time;
1913     bool old_vm_running = false;
1914     bool entered_postcopy = false;
1915     /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
1916     enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE;
1917     bool enable_colo = migrate_colo_enabled();
1918 
1919     rcu_register_thread();
1920 
1921     qemu_savevm_state_header(s->to_dst_file);
1922 
1923     if (migrate_postcopy_ram()) {
1924         /* Now tell the dest that it should open its end so it can reply */
1925         qemu_savevm_send_open_return_path(s->to_dst_file);
1926 
1927         /* And do a ping that will make stuff easier to debug */
1928         qemu_savevm_send_ping(s->to_dst_file, 1);
1929 
1930         /*
1931          * Tell the destination that we *might* want to do postcopy later;
1932          * if the other end can't do postcopy it should fail now, nice and
1933          * early.
1934          */
1935         qemu_savevm_send_postcopy_advise(s->to_dst_file);
1936     }
1937 
1938     qemu_savevm_state_begin(s->to_dst_file, &s->params);
1939 
1940     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
1941     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1942                       MIGRATION_STATUS_ACTIVE);
1943 
1944     trace_migration_thread_setup_complete();
1945 
1946     while (s->state == MIGRATION_STATUS_ACTIVE ||
1947            s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1948         int64_t current_time;
1949         uint64_t pending_size;
1950 
1951         if (!qemu_file_rate_limit(s->to_dst_file)) {
1952             uint64_t pend_post, pend_nonpost;
1953 
1954             qemu_savevm_state_pending(s->to_dst_file, threshold_size,
1955                                       &pend_nonpost, &pend_post);
1956             pending_size = pend_nonpost + pend_post;
1957             trace_migrate_pending(pending_size, threshold_size,
1958                                   pend_post, pend_nonpost);
1959             if (pending_size && pending_size >= threshold_size) {
1960                 /* Still a significant amount to transfer */
1961 
1962                 if (migrate_postcopy_ram() &&
1963                     s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE &&
1964                     pend_nonpost <= threshold_size &&
1965                     atomic_read(&s->start_postcopy)) {
1966 
1967                     if (!postcopy_start(s, &old_vm_running)) {
1968                         current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE;
1969                         entered_postcopy = true;
1970                     }
1971 
1972                     continue;
1973                 }
1974                 /* Just another iteration step */
1975                 qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy);
1976             } else {
1977                 trace_migration_thread_low_pending(pending_size);
1978                 migration_completion(s, current_active_state,
1979                                      &old_vm_running, &start_time);
1980                 break;
1981             }
1982         }
1983 
1984         if (qemu_file_get_error(s->to_dst_file)) {
1985             migrate_set_state(&s->state, current_active_state,
1986                               MIGRATION_STATUS_FAILED);
1987             trace_migration_thread_file_err();
1988             break;
1989         }
1990         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1991         if (current_time >= initial_time + BUFFER_DELAY) {
1992             uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) -
1993                                          initial_bytes;
1994             uint64_t time_spent = current_time - initial_time;
1995             double bandwidth = (double)transferred_bytes / time_spent;
1996             threshold_size = bandwidth * s->parameters.downtime_limit;
1997 
1998             s->mbps = (((double) transferred_bytes * 8.0) /
1999                     ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
2000 
2001             trace_migrate_transferred(transferred_bytes, time_spent,
2002                                       bandwidth, threshold_size);
2003             /* if we haven't sent anything, we don't want to recalculate
2004                10000 is a small enough number for our purposes */
2005             if (ram_dirty_pages_rate() && transferred_bytes > 10000) {
2006                 s->expected_downtime = ram_dirty_pages_rate() *
2007                     qemu_target_page_size() / bandwidth;
2008             }
2009 
2010             qemu_file_reset_rate_limit(s->to_dst_file);
2011             initial_time = current_time;
2012             initial_bytes = qemu_ftell(s->to_dst_file);
2013         }
2014         if (qemu_file_rate_limit(s->to_dst_file)) {
2015             /* usleep expects microseconds */
2016             g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
2017         }
2018     }
2019 
2020     trace_migration_thread_after_loop();
2021     /* If we enabled cpu throttling for auto-converge, turn it off. */
2022     cpu_throttle_stop();
2023     end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2024 
2025     qemu_mutex_lock_iothread();
2026     /*
2027      * The resource has been allocated by migration will be reused in COLO
2028      * process, so don't release them.
2029      */
2030     if (!enable_colo) {
2031         qemu_savevm_state_cleanup();
2032     }
2033     if (s->state == MIGRATION_STATUS_COMPLETED) {
2034         uint64_t transferred_bytes = qemu_ftell(s->to_dst_file);
2035         s->total_time = end_time - s->total_time;
2036         if (!entered_postcopy) {
2037             s->downtime = end_time - start_time;
2038         }
2039         if (s->total_time) {
2040             s->mbps = (((double) transferred_bytes * 8.0) /
2041                        ((double) s->total_time)) / 1000;
2042         }
2043         runstate_set(RUN_STATE_POSTMIGRATE);
2044     } else {
2045         if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) {
2046             migrate_start_colo_process(s);
2047             qemu_savevm_state_cleanup();
2048             /*
2049             * Fixme: we will run VM in COLO no matter its old running state.
2050             * After exited COLO, we will keep running.
2051             */
2052             old_vm_running = true;
2053         }
2054         if (old_vm_running && !entered_postcopy) {
2055             vm_start();
2056         } else {
2057             if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2058                 runstate_set(RUN_STATE_POSTMIGRATE);
2059             }
2060         }
2061     }
2062     qemu_bh_schedule(s->cleanup_bh);
2063     qemu_mutex_unlock_iothread();
2064 
2065     rcu_unregister_thread();
2066     return NULL;
2067 }
2068 
2069 void migrate_fd_connect(MigrationState *s)
2070 {
2071     s->expected_downtime = s->parameters.downtime_limit;
2072     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
2073 
2074     qemu_file_set_blocking(s->to_dst_file, true);
2075     qemu_file_set_rate_limit(s->to_dst_file,
2076                              s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
2077 
2078     /* Notify before starting migration thread */
2079     notifier_list_notify(&migration_state_notifiers, s);
2080 
2081     /*
2082      * Open the return path; currently for postcopy but other things might
2083      * also want it.
2084      */
2085     if (migrate_postcopy_ram()) {
2086         if (open_return_path_on_source(s)) {
2087             error_report("Unable to open return-path for postcopy");
2088             migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2089                               MIGRATION_STATUS_FAILED);
2090             migrate_fd_cleanup(s);
2091             return;
2092         }
2093     }
2094 
2095     migrate_compress_threads_create();
2096     qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
2097                        QEMU_THREAD_JOINABLE);
2098     s->migration_thread_running = true;
2099 }
2100 
2101 PostcopyState  postcopy_state_get(void)
2102 {
2103     return atomic_mb_read(&incoming_postcopy_state);
2104 }
2105 
2106 /* Set the state and return the old state */
2107 PostcopyState postcopy_state_set(PostcopyState new_state)
2108 {
2109     return atomic_xchg(&incoming_postcopy_state, new_state);
2110 }
2111 
2112