xref: /openbmc/qemu/migration/migration.c (revision 0806b30c8dff64e944456aa15bdc6957384e29a8)
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     bdrv_invalidate_cache_all(&local_err);
343     if (local_err) {
344         migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
345                           MIGRATION_STATUS_FAILED);
346         error_report_err(local_err);
347         migrate_decompress_threads_join();
348         exit(EXIT_FAILURE);
349     }
350 
351     /* If we get an error here, just don't restart the VM yet. */
352     blk_resume_after_migration(&local_err);
353     if (local_err) {
354         error_free(local_err);
355         local_err = NULL;
356         autostart = false;
357     }
358 
359     /*
360      * This must happen after all error conditions are dealt with and
361      * we're sure the VM is going to be running on this host.
362      */
363     qemu_announce_self();
364 
365     /* If global state section was not received or we are in running
366        state, we need to obey autostart. Any other state is set with
367        runstate_set. */
368 
369     if (!global_state_received() ||
370         global_state_get_runstate() == RUN_STATE_RUNNING) {
371         if (autostart) {
372             vm_start();
373         } else {
374             runstate_set(RUN_STATE_PAUSED);
375         }
376     } else {
377         runstate_set(global_state_get_runstate());
378     }
379     migrate_decompress_threads_join();
380     /*
381      * This must happen after any state changes since as soon as an external
382      * observer sees this event they might start to prod at the VM assuming
383      * it's ready to use.
384      */
385     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
386                       MIGRATION_STATUS_COMPLETED);
387     qemu_bh_delete(mis->bh);
388     migration_incoming_state_destroy();
389 }
390 
391 static void process_incoming_migration_co(void *opaque)
392 {
393     QEMUFile *f = opaque;
394     MigrationIncomingState *mis = migration_incoming_get_current();
395     PostcopyState ps;
396     int ret;
397 
398     mis->from_src_file = f;
399     mis->largest_page_size = qemu_ram_pagesize_largest();
400     postcopy_state_set(POSTCOPY_INCOMING_NONE);
401     migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
402                       MIGRATION_STATUS_ACTIVE);
403     ret = qemu_loadvm_state(f);
404 
405     ps = postcopy_state_get();
406     trace_process_incoming_migration_co_end(ret, ps);
407     if (ps != POSTCOPY_INCOMING_NONE) {
408         if (ps == POSTCOPY_INCOMING_ADVISE) {
409             /*
410              * Where a migration had postcopy enabled (and thus went to advise)
411              * but managed to complete within the precopy period, we can use
412              * the normal exit.
413              */
414             postcopy_ram_incoming_cleanup(mis);
415         } else if (ret >= 0) {
416             /*
417              * Postcopy was started, cleanup should happen at the end of the
418              * postcopy thread.
419              */
420             trace_process_incoming_migration_co_postcopy_end_main();
421             return;
422         }
423         /* Else if something went wrong then just fall out of the normal exit */
424     }
425 
426     /* we get COLO info, and know if we are in COLO mode */
427     if (!ret && migration_incoming_enable_colo()) {
428         mis->migration_incoming_co = qemu_coroutine_self();
429         qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
430              colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
431         mis->have_colo_incoming_thread = true;
432         qemu_coroutine_yield();
433 
434         /* Wait checkpoint incoming thread exit before free resource */
435         qemu_thread_join(&mis->colo_incoming_thread);
436     }
437 
438     if (ret < 0) {
439         migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
440                           MIGRATION_STATUS_FAILED);
441         error_report("load of migration failed: %s", strerror(-ret));
442         migrate_decompress_threads_join();
443         exit(EXIT_FAILURE);
444     }
445 
446     qemu_fclose(f);
447     free_xbzrle_decoded_buf();
448 
449     mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
450     qemu_bh_schedule(mis->bh);
451 }
452 
453 void migration_fd_process_incoming(QEMUFile *f)
454 {
455     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, f);
456 
457     migrate_decompress_threads_create();
458     qemu_file_set_blocking(f, false);
459     qemu_coroutine_enter(co);
460 }
461 
462 
463 void migration_channel_process_incoming(MigrationState *s,
464                                         QIOChannel *ioc)
465 {
466     trace_migration_set_incoming_channel(
467         ioc, object_get_typename(OBJECT(ioc)));
468 
469     if (s->parameters.tls_creds &&
470         *s->parameters.tls_creds &&
471         !object_dynamic_cast(OBJECT(ioc),
472                              TYPE_QIO_CHANNEL_TLS)) {
473         Error *local_err = NULL;
474         migration_tls_channel_process_incoming(s, ioc, &local_err);
475         if (local_err) {
476             error_report_err(local_err);
477         }
478     } else {
479         QEMUFile *f = qemu_fopen_channel_input(ioc);
480         migration_fd_process_incoming(f);
481     }
482 }
483 
484 
485 void migration_channel_connect(MigrationState *s,
486                                QIOChannel *ioc,
487                                const char *hostname)
488 {
489     trace_migration_set_outgoing_channel(
490         ioc, object_get_typename(OBJECT(ioc)), hostname);
491 
492     if (s->parameters.tls_creds &&
493         *s->parameters.tls_creds &&
494         !object_dynamic_cast(OBJECT(ioc),
495                              TYPE_QIO_CHANNEL_TLS)) {
496         Error *local_err = NULL;
497         migration_tls_channel_connect(s, ioc, hostname, &local_err);
498         if (local_err) {
499             migrate_fd_error(s, local_err);
500             error_free(local_err);
501         }
502     } else {
503         QEMUFile *f = qemu_fopen_channel_output(ioc);
504 
505         s->to_dst_file = f;
506 
507         migrate_fd_connect(s);
508     }
509 }
510 
511 
512 /*
513  * Send a message on the return channel back to the source
514  * of the migration.
515  */
516 void migrate_send_rp_message(MigrationIncomingState *mis,
517                              enum mig_rp_message_type message_type,
518                              uint16_t len, void *data)
519 {
520     trace_migrate_send_rp_message((int)message_type, len);
521     qemu_mutex_lock(&mis->rp_mutex);
522     qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
523     qemu_put_be16(mis->to_src_file, len);
524     qemu_put_buffer(mis->to_src_file, data, len);
525     qemu_fflush(mis->to_src_file);
526     qemu_mutex_unlock(&mis->rp_mutex);
527 }
528 
529 /*
530  * Send a 'SHUT' message on the return channel with the given value
531  * to indicate that we've finished with the RP.  Non-0 value indicates
532  * error.
533  */
534 void migrate_send_rp_shut(MigrationIncomingState *mis,
535                           uint32_t value)
536 {
537     uint32_t buf;
538 
539     buf = cpu_to_be32(value);
540     migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
541 }
542 
543 /*
544  * Send a 'PONG' message on the return channel with the given value
545  * (normally in response to a 'PING')
546  */
547 void migrate_send_rp_pong(MigrationIncomingState *mis,
548                           uint32_t value)
549 {
550     uint32_t buf;
551 
552     buf = cpu_to_be32(value);
553     migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
554 }
555 
556 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
557 {
558     MigrationCapabilityStatusList *head = NULL;
559     MigrationCapabilityStatusList *caps;
560     MigrationState *s = migrate_get_current();
561     int i;
562 
563     caps = NULL; /* silence compiler warning */
564     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
565         if (i == MIGRATION_CAPABILITY_X_COLO && !colo_supported()) {
566             continue;
567         }
568         if (head == NULL) {
569             head = g_malloc0(sizeof(*caps));
570             caps = head;
571         } else {
572             caps->next = g_malloc0(sizeof(*caps));
573             caps = caps->next;
574         }
575         caps->value =
576             g_malloc(sizeof(*caps->value));
577         caps->value->capability = i;
578         caps->value->state = s->enabled_capabilities[i];
579     }
580 
581     return head;
582 }
583 
584 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
585 {
586     MigrationParameters *params;
587     MigrationState *s = migrate_get_current();
588 
589     params = g_malloc0(sizeof(*params));
590     params->has_compress_level = true;
591     params->compress_level = s->parameters.compress_level;
592     params->has_compress_threads = true;
593     params->compress_threads = s->parameters.compress_threads;
594     params->has_decompress_threads = true;
595     params->decompress_threads = s->parameters.decompress_threads;
596     params->has_cpu_throttle_initial = true;
597     params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
598     params->has_cpu_throttle_increment = true;
599     params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
600     params->has_tls_creds = !!s->parameters.tls_creds;
601     params->tls_creds = g_strdup(s->parameters.tls_creds);
602     params->has_tls_hostname = !!s->parameters.tls_hostname;
603     params->tls_hostname = g_strdup(s->parameters.tls_hostname);
604     params->has_max_bandwidth = true;
605     params->max_bandwidth = s->parameters.max_bandwidth;
606     params->has_downtime_limit = true;
607     params->downtime_limit = s->parameters.downtime_limit;
608     params->has_x_checkpoint_delay = true;
609     params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
610 
611     return params;
612 }
613 
614 /*
615  * Return true if we're already in the middle of a migration
616  * (i.e. any of the active or setup states)
617  */
618 static bool migration_is_setup_or_active(int state)
619 {
620     switch (state) {
621     case MIGRATION_STATUS_ACTIVE:
622     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
623     case MIGRATION_STATUS_SETUP:
624         return true;
625 
626     default:
627         return false;
628 
629     }
630 }
631 
632 static void get_xbzrle_cache_stats(MigrationInfo *info)
633 {
634     if (migrate_use_xbzrle()) {
635         info->has_xbzrle_cache = true;
636         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
637         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
638         info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred();
639         info->xbzrle_cache->pages = xbzrle_mig_pages_transferred();
640         info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss();
641         info->xbzrle_cache->cache_miss_rate = xbzrle_mig_cache_miss_rate();
642         info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow();
643     }
644 }
645 
646 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
647 {
648     info->has_ram = true;
649     info->ram = g_malloc0(sizeof(*info->ram));
650     info->ram->transferred = ram_bytes_transferred();
651     info->ram->total = ram_bytes_total();
652     info->ram->duplicate = dup_mig_pages_transferred();
653     /* legacy value.  It is not used anymore */
654     info->ram->skipped = 0;
655     info->ram->normal = norm_mig_pages_transferred();
656     info->ram->normal_bytes = norm_mig_pages_transferred() *
657         qemu_target_page_size();
658     info->ram->mbps = s->mbps;
659     info->ram->dirty_sync_count = ram_dirty_sync_count();
660     info->ram->postcopy_requests = ram_postcopy_requests();
661     info->ram->page_size = qemu_target_page_size();
662 
663     if (s->state != MIGRATION_STATUS_COMPLETED) {
664         info->ram->remaining = ram_bytes_remaining();
665         info->ram->dirty_pages_rate = ram_dirty_pages_rate();
666     }
667 }
668 
669 MigrationInfo *qmp_query_migrate(Error **errp)
670 {
671     MigrationInfo *info = g_malloc0(sizeof(*info));
672     MigrationState *s = migrate_get_current();
673 
674     switch (s->state) {
675     case MIGRATION_STATUS_NONE:
676         /* no migration has happened ever */
677         break;
678     case MIGRATION_STATUS_SETUP:
679         info->has_status = true;
680         info->has_total_time = false;
681         break;
682     case MIGRATION_STATUS_ACTIVE:
683     case MIGRATION_STATUS_CANCELLING:
684         info->has_status = true;
685         info->has_total_time = true;
686         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
687             - s->total_time;
688         info->has_expected_downtime = true;
689         info->expected_downtime = s->expected_downtime;
690         info->has_setup_time = true;
691         info->setup_time = s->setup_time;
692 
693         populate_ram_info(info, s);
694 
695         if (blk_mig_active()) {
696             info->has_disk = true;
697             info->disk = g_malloc0(sizeof(*info->disk));
698             info->disk->transferred = blk_mig_bytes_transferred();
699             info->disk->remaining = blk_mig_bytes_remaining();
700             info->disk->total = blk_mig_bytes_total();
701         }
702 
703         if (cpu_throttle_active()) {
704             info->has_cpu_throttle_percentage = true;
705             info->cpu_throttle_percentage = cpu_throttle_get_percentage();
706         }
707 
708         get_xbzrle_cache_stats(info);
709         break;
710     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
711         /* Mostly the same as active; TODO add some postcopy stats */
712         info->has_status = true;
713         info->has_total_time = true;
714         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
715             - s->total_time;
716         info->has_expected_downtime = true;
717         info->expected_downtime = s->expected_downtime;
718         info->has_setup_time = true;
719         info->setup_time = s->setup_time;
720 
721         populate_ram_info(info, s);
722 
723         if (blk_mig_active()) {
724             info->has_disk = true;
725             info->disk = g_malloc0(sizeof(*info->disk));
726             info->disk->transferred = blk_mig_bytes_transferred();
727             info->disk->remaining = blk_mig_bytes_remaining();
728             info->disk->total = blk_mig_bytes_total();
729         }
730 
731         get_xbzrle_cache_stats(info);
732         break;
733     case MIGRATION_STATUS_COLO:
734         info->has_status = true;
735         /* TODO: display COLO specific information (checkpoint info etc.) */
736         break;
737     case MIGRATION_STATUS_COMPLETED:
738         get_xbzrle_cache_stats(info);
739 
740         info->has_status = true;
741         info->has_total_time = true;
742         info->total_time = s->total_time;
743         info->has_downtime = true;
744         info->downtime = s->downtime;
745         info->has_setup_time = true;
746         info->setup_time = s->setup_time;
747 
748         populate_ram_info(info, s);
749         break;
750     case MIGRATION_STATUS_FAILED:
751         info->has_status = true;
752         if (s->error) {
753             info->has_error_desc = true;
754             info->error_desc = g_strdup(error_get_pretty(s->error));
755         }
756         break;
757     case MIGRATION_STATUS_CANCELLED:
758         info->has_status = true;
759         break;
760     }
761     info->status = s->state;
762 
763     return info;
764 }
765 
766 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
767                                   Error **errp)
768 {
769     MigrationState *s = migrate_get_current();
770     MigrationCapabilityStatusList *cap;
771     bool old_postcopy_cap = migrate_postcopy_ram();
772 
773     if (migration_is_setup_or_active(s->state)) {
774         error_setg(errp, QERR_MIGRATION_ACTIVE);
775         return;
776     }
777 
778     for (cap = params; cap; cap = cap->next) {
779         if (cap->value->capability == MIGRATION_CAPABILITY_X_COLO) {
780             if (!colo_supported()) {
781                 error_setg(errp, "COLO is not currently supported, please"
782                              " configure with --enable-colo option in order to"
783                              " support COLO feature");
784                 continue;
785             }
786         }
787         s->enabled_capabilities[cap->value->capability] = cap->value->state;
788     }
789 
790     if (migrate_postcopy_ram()) {
791         if (migrate_use_compression()) {
792             /* The decompression threads asynchronously write into RAM
793              * rather than use the atomic copies needed to avoid
794              * userfaulting.  It should be possible to fix the decompression
795              * threads for compatibility in future.
796              */
797             error_report("Postcopy is not currently compatible with "
798                          "compression");
799             s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] =
800                 false;
801         }
802         /* This check is reasonably expensive, so only when it's being
803          * set the first time, also it's only the destination that needs
804          * special support.
805          */
806         if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
807             !postcopy_ram_supported_by_host()) {
808             /* postcopy_ram_supported_by_host will have emitted a more
809              * detailed message
810              */
811             error_report("Postcopy is not supported");
812             s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] =
813                 false;
814         }
815     }
816 }
817 
818 void qmp_migrate_set_parameters(MigrationParameters *params, Error **errp)
819 {
820     MigrationState *s = migrate_get_current();
821 
822     if (params->has_compress_level &&
823         (params->compress_level < 0 || params->compress_level > 9)) {
824         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
825                    "is invalid, it should be in the range of 0 to 9");
826         return;
827     }
828     if (params->has_compress_threads &&
829         (params->compress_threads < 1 || params->compress_threads > 255)) {
830         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
831                    "compress_threads",
832                    "is invalid, it should be in the range of 1 to 255");
833         return;
834     }
835     if (params->has_decompress_threads &&
836         (params->decompress_threads < 1 || params->decompress_threads > 255)) {
837         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
838                    "decompress_threads",
839                    "is invalid, it should be in the range of 1 to 255");
840         return;
841     }
842     if (params->has_cpu_throttle_initial &&
843         (params->cpu_throttle_initial < 1 ||
844          params->cpu_throttle_initial > 99)) {
845         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
846                    "cpu_throttle_initial",
847                    "an integer in the range of 1 to 99");
848         return;
849     }
850     if (params->has_cpu_throttle_increment &&
851         (params->cpu_throttle_increment < 1 ||
852          params->cpu_throttle_increment > 99)) {
853         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
854                    "cpu_throttle_increment",
855                    "an integer in the range of 1 to 99");
856         return;
857     }
858     if (params->has_max_bandwidth &&
859         (params->max_bandwidth < 0 || params->max_bandwidth > SIZE_MAX)) {
860         error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
861                          " range of 0 to %zu bytes/second", SIZE_MAX);
862         return;
863     }
864     if (params->has_downtime_limit &&
865         (params->downtime_limit < 0 ||
866          params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
867         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
868                          "the range of 0 to %d milliseconds",
869                          MAX_MIGRATE_DOWNTIME);
870         return;
871     }
872     if (params->has_x_checkpoint_delay && (params->x_checkpoint_delay < 0)) {
873         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
874                     "x_checkpoint_delay",
875                     "is invalid, it should be positive");
876     }
877 
878     if (params->has_compress_level) {
879         s->parameters.compress_level = params->compress_level;
880     }
881     if (params->has_compress_threads) {
882         s->parameters.compress_threads = params->compress_threads;
883     }
884     if (params->has_decompress_threads) {
885         s->parameters.decompress_threads = params->decompress_threads;
886     }
887     if (params->has_cpu_throttle_initial) {
888         s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
889     }
890     if (params->has_cpu_throttle_increment) {
891         s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
892     }
893     if (params->has_tls_creds) {
894         g_free(s->parameters.tls_creds);
895         s->parameters.tls_creds = g_strdup(params->tls_creds);
896     }
897     if (params->has_tls_hostname) {
898         g_free(s->parameters.tls_hostname);
899         s->parameters.tls_hostname = g_strdup(params->tls_hostname);
900     }
901     if (params->has_max_bandwidth) {
902         s->parameters.max_bandwidth = params->max_bandwidth;
903         if (s->to_dst_file) {
904             qemu_file_set_rate_limit(s->to_dst_file,
905                                 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
906         }
907     }
908     if (params->has_downtime_limit) {
909         s->parameters.downtime_limit = params->downtime_limit;
910     }
911 
912     if (params->has_x_checkpoint_delay) {
913         s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
914         if (migration_in_colo_state()) {
915             colo_checkpoint_notify(s);
916         }
917     }
918 }
919 
920 
921 void qmp_migrate_start_postcopy(Error **errp)
922 {
923     MigrationState *s = migrate_get_current();
924 
925     if (!migrate_postcopy_ram()) {
926         error_setg(errp, "Enable postcopy with migrate_set_capability before"
927                          " the start of migration");
928         return;
929     }
930 
931     if (s->state == MIGRATION_STATUS_NONE) {
932         error_setg(errp, "Postcopy must be started after migration has been"
933                          " started");
934         return;
935     }
936     /*
937      * we don't error if migration has finished since that would be racy
938      * with issuing this command.
939      */
940     atomic_set(&s->start_postcopy, true);
941 }
942 
943 /* shared migration helpers */
944 
945 void migrate_set_state(int *state, int old_state, int new_state)
946 {
947     if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
948         trace_migrate_set_state(new_state);
949         migrate_generate_event(new_state);
950     }
951 }
952 
953 static void migrate_fd_cleanup(void *opaque)
954 {
955     MigrationState *s = opaque;
956 
957     qemu_bh_delete(s->cleanup_bh);
958     s->cleanup_bh = NULL;
959 
960     migration_page_queue_free();
961 
962     if (s->to_dst_file) {
963         trace_migrate_fd_cleanup();
964         qemu_mutex_unlock_iothread();
965         if (s->migration_thread_running) {
966             qemu_thread_join(&s->thread);
967             s->migration_thread_running = false;
968         }
969         qemu_mutex_lock_iothread();
970 
971         migrate_compress_threads_join();
972         qemu_fclose(s->to_dst_file);
973         s->to_dst_file = NULL;
974     }
975 
976     assert((s->state != MIGRATION_STATUS_ACTIVE) &&
977            (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
978 
979     if (s->state == MIGRATION_STATUS_CANCELLING) {
980         migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
981                           MIGRATION_STATUS_CANCELLED);
982     }
983 
984     notifier_list_notify(&migration_state_notifiers, s);
985 }
986 
987 void migrate_fd_error(MigrationState *s, const Error *error)
988 {
989     trace_migrate_fd_error(error_get_pretty(error));
990     assert(s->to_dst_file == NULL);
991     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
992                       MIGRATION_STATUS_FAILED);
993     if (!s->error) {
994         s->error = error_copy(error);
995     }
996     notifier_list_notify(&migration_state_notifiers, s);
997 }
998 
999 static void migrate_fd_cancel(MigrationState *s)
1000 {
1001     int old_state ;
1002     QEMUFile *f = migrate_get_current()->to_dst_file;
1003     trace_migrate_fd_cancel();
1004 
1005     if (s->rp_state.from_dst_file) {
1006         /* shutdown the rp socket, so causing the rp thread to shutdown */
1007         qemu_file_shutdown(s->rp_state.from_dst_file);
1008     }
1009 
1010     do {
1011         old_state = s->state;
1012         if (!migration_is_setup_or_active(old_state)) {
1013             break;
1014         }
1015         migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1016     } while (s->state != MIGRATION_STATUS_CANCELLING);
1017 
1018     /*
1019      * If we're unlucky the migration code might be stuck somewhere in a
1020      * send/write while the network has failed and is waiting to timeout;
1021      * if we've got shutdown(2) available then we can force it to quit.
1022      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1023      * called in a bh, so there is no race against this cancel.
1024      */
1025     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1026         qemu_file_shutdown(f);
1027     }
1028     if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1029         Error *local_err = NULL;
1030 
1031         bdrv_invalidate_cache_all(&local_err);
1032         if (local_err) {
1033             error_report_err(local_err);
1034         } else {
1035             s->block_inactive = false;
1036         }
1037     }
1038 }
1039 
1040 void add_migration_state_change_notifier(Notifier *notify)
1041 {
1042     notifier_list_add(&migration_state_notifiers, notify);
1043 }
1044 
1045 void remove_migration_state_change_notifier(Notifier *notify)
1046 {
1047     notifier_remove(notify);
1048 }
1049 
1050 bool migration_in_setup(MigrationState *s)
1051 {
1052     return s->state == MIGRATION_STATUS_SETUP;
1053 }
1054 
1055 bool migration_has_finished(MigrationState *s)
1056 {
1057     return s->state == MIGRATION_STATUS_COMPLETED;
1058 }
1059 
1060 bool migration_has_failed(MigrationState *s)
1061 {
1062     return (s->state == MIGRATION_STATUS_CANCELLED ||
1063             s->state == MIGRATION_STATUS_FAILED);
1064 }
1065 
1066 bool migration_in_postcopy(void)
1067 {
1068     MigrationState *s = migrate_get_current();
1069 
1070     return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1071 }
1072 
1073 bool migration_in_postcopy_after_devices(MigrationState *s)
1074 {
1075     return migration_in_postcopy() && s->postcopy_after_devices;
1076 }
1077 
1078 bool migration_is_idle(void)
1079 {
1080     MigrationState *s = migrate_get_current();
1081 
1082     switch (s->state) {
1083     case MIGRATION_STATUS_NONE:
1084     case MIGRATION_STATUS_CANCELLED:
1085     case MIGRATION_STATUS_COMPLETED:
1086     case MIGRATION_STATUS_FAILED:
1087         return true;
1088     case MIGRATION_STATUS_SETUP:
1089     case MIGRATION_STATUS_CANCELLING:
1090     case MIGRATION_STATUS_ACTIVE:
1091     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1092     case MIGRATION_STATUS_COLO:
1093         return false;
1094     case MIGRATION_STATUS__MAX:
1095         g_assert_not_reached();
1096     }
1097 
1098     return false;
1099 }
1100 
1101 MigrationState *migrate_init(const MigrationParams *params)
1102 {
1103     MigrationState *s = migrate_get_current();
1104 
1105     /*
1106      * Reinitialise all migration state, except
1107      * parameters/capabilities that the user set, and
1108      * locks.
1109      */
1110     s->bytes_xfer = 0;
1111     s->xfer_limit = 0;
1112     s->cleanup_bh = 0;
1113     s->to_dst_file = NULL;
1114     s->state = MIGRATION_STATUS_NONE;
1115     s->params = *params;
1116     s->rp_state.from_dst_file = NULL;
1117     s->rp_state.error = false;
1118     s->mbps = 0.0;
1119     s->downtime = 0;
1120     s->expected_downtime = 0;
1121     s->setup_time = 0;
1122     s->start_postcopy = false;
1123     s->postcopy_after_devices = false;
1124     s->migration_thread_running = false;
1125     error_free(s->error);
1126     s->error = NULL;
1127 
1128     migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1129 
1130     s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1131     return s;
1132 }
1133 
1134 static GSList *migration_blockers;
1135 
1136 int migrate_add_blocker(Error *reason, Error **errp)
1137 {
1138     if (only_migratable) {
1139         error_propagate(errp, error_copy(reason));
1140         error_prepend(errp, "disallowing migration blocker "
1141                           "(--only_migratable) for: ");
1142         return -EACCES;
1143     }
1144 
1145     if (migration_is_idle()) {
1146         migration_blockers = g_slist_prepend(migration_blockers, reason);
1147         return 0;
1148     }
1149 
1150     error_propagate(errp, error_copy(reason));
1151     error_prepend(errp, "disallowing migration blocker (migration in "
1152                       "progress) for: ");
1153     return -EBUSY;
1154 }
1155 
1156 void migrate_del_blocker(Error *reason)
1157 {
1158     migration_blockers = g_slist_remove(migration_blockers, reason);
1159 }
1160 
1161 int check_migratable(Object *obj, Error **err)
1162 {
1163     DeviceClass *dc = DEVICE_GET_CLASS(obj);
1164     if (only_migratable && dc->vmsd) {
1165         if (dc->vmsd->unmigratable) {
1166             error_setg(err, "Device %s is not migratable, but "
1167                        "--only-migratable was specified",
1168                        object_get_typename(obj));
1169             return -1;
1170         }
1171     }
1172 
1173     return 0;
1174 }
1175 
1176 void qmp_migrate_incoming(const char *uri, Error **errp)
1177 {
1178     Error *local_err = NULL;
1179     static bool once = true;
1180 
1181     if (!deferred_incoming) {
1182         error_setg(errp, "For use with '-incoming defer'");
1183         return;
1184     }
1185     if (!once) {
1186         error_setg(errp, "The incoming migration has already been started");
1187     }
1188 
1189     qemu_start_incoming_migration(uri, &local_err);
1190 
1191     if (local_err) {
1192         error_propagate(errp, local_err);
1193         return;
1194     }
1195 
1196     once = false;
1197 }
1198 
1199 bool migration_is_blocked(Error **errp)
1200 {
1201     if (qemu_savevm_state_blocked(errp)) {
1202         return true;
1203     }
1204 
1205     if (migration_blockers) {
1206         *errp = error_copy(migration_blockers->data);
1207         return true;
1208     }
1209 
1210     return false;
1211 }
1212 
1213 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1214                  bool has_inc, bool inc, bool has_detach, bool detach,
1215                  Error **errp)
1216 {
1217     Error *local_err = NULL;
1218     MigrationState *s = migrate_get_current();
1219     MigrationParams params;
1220     const char *p;
1221 
1222     params.blk = has_blk && blk;
1223     params.shared = has_inc && inc;
1224 
1225     if (migration_is_setup_or_active(s->state) ||
1226         s->state == MIGRATION_STATUS_CANCELLING ||
1227         s->state == MIGRATION_STATUS_COLO) {
1228         error_setg(errp, QERR_MIGRATION_ACTIVE);
1229         return;
1230     }
1231     if (runstate_check(RUN_STATE_INMIGRATE)) {
1232         error_setg(errp, "Guest is waiting for an incoming migration");
1233         return;
1234     }
1235 
1236     if (migration_is_blocked(errp)) {
1237         return;
1238     }
1239 
1240     s = migrate_init(&params);
1241 
1242     if (strstart(uri, "tcp:", &p)) {
1243         tcp_start_outgoing_migration(s, p, &local_err);
1244 #ifdef CONFIG_RDMA
1245     } else if (strstart(uri, "rdma:", &p)) {
1246         rdma_start_outgoing_migration(s, p, &local_err);
1247 #endif
1248     } else if (strstart(uri, "exec:", &p)) {
1249         exec_start_outgoing_migration(s, p, &local_err);
1250     } else if (strstart(uri, "unix:", &p)) {
1251         unix_start_outgoing_migration(s, p, &local_err);
1252     } else if (strstart(uri, "fd:", &p)) {
1253         fd_start_outgoing_migration(s, p, &local_err);
1254     } else {
1255         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1256                    "a valid migration protocol");
1257         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1258                           MIGRATION_STATUS_FAILED);
1259         return;
1260     }
1261 
1262     if (local_err) {
1263         migrate_fd_error(s, local_err);
1264         error_propagate(errp, local_err);
1265         return;
1266     }
1267 }
1268 
1269 void qmp_migrate_cancel(Error **errp)
1270 {
1271     migrate_fd_cancel(migrate_get_current());
1272 }
1273 
1274 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1275 {
1276     MigrationState *s = migrate_get_current();
1277     int64_t new_size;
1278 
1279     /* Check for truncation */
1280     if (value != (size_t)value) {
1281         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1282                    "exceeding address space");
1283         return;
1284     }
1285 
1286     /* Cache should not be larger than guest ram size */
1287     if (value > ram_bytes_total()) {
1288         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1289                    "exceeds guest ram size ");
1290         return;
1291     }
1292 
1293     new_size = xbzrle_cache_resize(value);
1294     if (new_size < 0) {
1295         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1296                    "is smaller than page size");
1297         return;
1298     }
1299 
1300     s->xbzrle_cache_size = new_size;
1301 }
1302 
1303 int64_t qmp_query_migrate_cache_size(Error **errp)
1304 {
1305     return migrate_xbzrle_cache_size();
1306 }
1307 
1308 void qmp_migrate_set_speed(int64_t value, Error **errp)
1309 {
1310     MigrationParameters p = {
1311         .has_max_bandwidth = true,
1312         .max_bandwidth = value,
1313     };
1314 
1315     qmp_migrate_set_parameters(&p, errp);
1316 }
1317 
1318 void qmp_migrate_set_downtime(double value, Error **errp)
1319 {
1320     if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1321         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1322                          "the range of 0 to %d seconds",
1323                          MAX_MIGRATE_DOWNTIME_SECONDS);
1324         return;
1325     }
1326 
1327     value *= 1000; /* Convert to milliseconds */
1328     value = MAX(0, MIN(INT64_MAX, value));
1329 
1330     MigrationParameters p = {
1331         .has_downtime_limit = true,
1332         .downtime_limit = value,
1333     };
1334 
1335     qmp_migrate_set_parameters(&p, errp);
1336 }
1337 
1338 bool migrate_release_ram(void)
1339 {
1340     MigrationState *s;
1341 
1342     s = migrate_get_current();
1343 
1344     return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1345 }
1346 
1347 bool migrate_postcopy_ram(void)
1348 {
1349     MigrationState *s;
1350 
1351     s = migrate_get_current();
1352 
1353     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1354 }
1355 
1356 bool migrate_auto_converge(void)
1357 {
1358     MigrationState *s;
1359 
1360     s = migrate_get_current();
1361 
1362     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1363 }
1364 
1365 bool migrate_zero_blocks(void)
1366 {
1367     MigrationState *s;
1368 
1369     s = migrate_get_current();
1370 
1371     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1372 }
1373 
1374 bool migrate_use_compression(void)
1375 {
1376     MigrationState *s;
1377 
1378     s = migrate_get_current();
1379 
1380     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
1381 }
1382 
1383 int migrate_compress_level(void)
1384 {
1385     MigrationState *s;
1386 
1387     s = migrate_get_current();
1388 
1389     return s->parameters.compress_level;
1390 }
1391 
1392 int migrate_compress_threads(void)
1393 {
1394     MigrationState *s;
1395 
1396     s = migrate_get_current();
1397 
1398     return s->parameters.compress_threads;
1399 }
1400 
1401 int migrate_decompress_threads(void)
1402 {
1403     MigrationState *s;
1404 
1405     s = migrate_get_current();
1406 
1407     return s->parameters.decompress_threads;
1408 }
1409 
1410 bool migrate_use_events(void)
1411 {
1412     MigrationState *s;
1413 
1414     s = migrate_get_current();
1415 
1416     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1417 }
1418 
1419 int migrate_use_xbzrle(void)
1420 {
1421     MigrationState *s;
1422 
1423     s = migrate_get_current();
1424 
1425     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1426 }
1427 
1428 int64_t migrate_xbzrle_cache_size(void)
1429 {
1430     MigrationState *s;
1431 
1432     s = migrate_get_current();
1433 
1434     return s->xbzrle_cache_size;
1435 }
1436 
1437 /* migration thread support */
1438 /*
1439  * Something bad happened to the RP stream, mark an error
1440  * The caller shall print or trace something to indicate why
1441  */
1442 static void mark_source_rp_bad(MigrationState *s)
1443 {
1444     s->rp_state.error = true;
1445 }
1446 
1447 static struct rp_cmd_args {
1448     ssize_t     len; /* -1 = variable */
1449     const char *name;
1450 } rp_cmd_args[] = {
1451     [MIG_RP_MSG_INVALID]        = { .len = -1, .name = "INVALID" },
1452     [MIG_RP_MSG_SHUT]           = { .len =  4, .name = "SHUT" },
1453     [MIG_RP_MSG_PONG]           = { .len =  4, .name = "PONG" },
1454     [MIG_RP_MSG_REQ_PAGES]      = { .len = 12, .name = "REQ_PAGES" },
1455     [MIG_RP_MSG_REQ_PAGES_ID]   = { .len = -1, .name = "REQ_PAGES_ID" },
1456     [MIG_RP_MSG_MAX]            = { .len = -1, .name = "MAX" },
1457 };
1458 
1459 /*
1460  * Process a request for pages received on the return path,
1461  * We're allowed to send more than requested (e.g. to round to our page size)
1462  * and we don't need to send pages that have already been sent.
1463  */
1464 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1465                                        ram_addr_t start, size_t len)
1466 {
1467     long our_host_ps = getpagesize();
1468 
1469     trace_migrate_handle_rp_req_pages(rbname, start, len);
1470 
1471     /*
1472      * Since we currently insist on matching page sizes, just sanity check
1473      * we're being asked for whole host pages.
1474      */
1475     if (start & (our_host_ps-1) ||
1476        (len & (our_host_ps-1))) {
1477         error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1478                      " len: %zd", __func__, start, len);
1479         mark_source_rp_bad(ms);
1480         return;
1481     }
1482 
1483     if (ram_save_queue_pages(rbname, start, len)) {
1484         mark_source_rp_bad(ms);
1485     }
1486 }
1487 
1488 /*
1489  * Handles messages sent on the return path towards the source VM
1490  *
1491  */
1492 static void *source_return_path_thread(void *opaque)
1493 {
1494     MigrationState *ms = opaque;
1495     QEMUFile *rp = ms->rp_state.from_dst_file;
1496     uint16_t header_len, header_type;
1497     uint8_t buf[512];
1498     uint32_t tmp32, sibling_error;
1499     ram_addr_t start = 0; /* =0 to silence warning */
1500     size_t  len = 0, expected_len;
1501     int res;
1502 
1503     trace_source_return_path_thread_entry();
1504     while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
1505            migration_is_setup_or_active(ms->state)) {
1506         trace_source_return_path_thread_loop_top();
1507         header_type = qemu_get_be16(rp);
1508         header_len = qemu_get_be16(rp);
1509 
1510         if (header_type >= MIG_RP_MSG_MAX ||
1511             header_type == MIG_RP_MSG_INVALID) {
1512             error_report("RP: Received invalid message 0x%04x length 0x%04x",
1513                     header_type, header_len);
1514             mark_source_rp_bad(ms);
1515             goto out;
1516         }
1517 
1518         if ((rp_cmd_args[header_type].len != -1 &&
1519             header_len != rp_cmd_args[header_type].len) ||
1520             header_len > sizeof(buf)) {
1521             error_report("RP: Received '%s' message (0x%04x) with"
1522                     "incorrect length %d expecting %zu",
1523                     rp_cmd_args[header_type].name, header_type, header_len,
1524                     (size_t)rp_cmd_args[header_type].len);
1525             mark_source_rp_bad(ms);
1526             goto out;
1527         }
1528 
1529         /* We know we've got a valid header by this point */
1530         res = qemu_get_buffer(rp, buf, header_len);
1531         if (res != header_len) {
1532             error_report("RP: Failed reading data for message 0x%04x"
1533                          " read %d expected %d",
1534                          header_type, res, header_len);
1535             mark_source_rp_bad(ms);
1536             goto out;
1537         }
1538 
1539         /* OK, we have the message and the data */
1540         switch (header_type) {
1541         case MIG_RP_MSG_SHUT:
1542             sibling_error = ldl_be_p(buf);
1543             trace_source_return_path_thread_shut(sibling_error);
1544             if (sibling_error) {
1545                 error_report("RP: Sibling indicated error %d", sibling_error);
1546                 mark_source_rp_bad(ms);
1547             }
1548             /*
1549              * We'll let the main thread deal with closing the RP
1550              * we could do a shutdown(2) on it, but we're the only user
1551              * anyway, so there's nothing gained.
1552              */
1553             goto out;
1554 
1555         case MIG_RP_MSG_PONG:
1556             tmp32 = ldl_be_p(buf);
1557             trace_source_return_path_thread_pong(tmp32);
1558             break;
1559 
1560         case MIG_RP_MSG_REQ_PAGES:
1561             start = ldq_be_p(buf);
1562             len = ldl_be_p(buf + 8);
1563             migrate_handle_rp_req_pages(ms, NULL, start, len);
1564             break;
1565 
1566         case MIG_RP_MSG_REQ_PAGES_ID:
1567             expected_len = 12 + 1; /* header + termination */
1568 
1569             if (header_len >= expected_len) {
1570                 start = ldq_be_p(buf);
1571                 len = ldl_be_p(buf + 8);
1572                 /* Now we expect an idstr */
1573                 tmp32 = buf[12]; /* Length of the following idstr */
1574                 buf[13 + tmp32] = '\0';
1575                 expected_len += tmp32;
1576             }
1577             if (header_len != expected_len) {
1578                 error_report("RP: Req_Page_id with length %d expecting %zd",
1579                         header_len, expected_len);
1580                 mark_source_rp_bad(ms);
1581                 goto out;
1582             }
1583             migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
1584             break;
1585 
1586         default:
1587             break;
1588         }
1589     }
1590     if (qemu_file_get_error(rp)) {
1591         trace_source_return_path_thread_bad_end();
1592         mark_source_rp_bad(ms);
1593     }
1594 
1595     trace_source_return_path_thread_end();
1596 out:
1597     ms->rp_state.from_dst_file = NULL;
1598     qemu_fclose(rp);
1599     return NULL;
1600 }
1601 
1602 static int open_return_path_on_source(MigrationState *ms)
1603 {
1604 
1605     ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
1606     if (!ms->rp_state.from_dst_file) {
1607         return -1;
1608     }
1609 
1610     trace_open_return_path_on_source();
1611     qemu_thread_create(&ms->rp_state.rp_thread, "return path",
1612                        source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
1613 
1614     trace_open_return_path_on_source_continue();
1615 
1616     return 0;
1617 }
1618 
1619 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1620 static int await_return_path_close_on_source(MigrationState *ms)
1621 {
1622     /*
1623      * If this is a normal exit then the destination will send a SHUT and the
1624      * rp_thread will exit, however if there's an error we need to cause
1625      * it to exit.
1626      */
1627     if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
1628         /*
1629          * shutdown(2), if we have it, will cause it to unblock if it's stuck
1630          * waiting for the destination.
1631          */
1632         qemu_file_shutdown(ms->rp_state.from_dst_file);
1633         mark_source_rp_bad(ms);
1634     }
1635     trace_await_return_path_close_on_source_joining();
1636     qemu_thread_join(&ms->rp_state.rp_thread);
1637     trace_await_return_path_close_on_source_close();
1638     return ms->rp_state.error;
1639 }
1640 
1641 /*
1642  * Switch from normal iteration to postcopy
1643  * Returns non-0 on error
1644  */
1645 static int postcopy_start(MigrationState *ms, bool *old_vm_running)
1646 {
1647     int ret;
1648     QIOChannelBuffer *bioc;
1649     QEMUFile *fb;
1650     int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1651     bool restart_block = false;
1652     migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
1653                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
1654 
1655     trace_postcopy_start();
1656     qemu_mutex_lock_iothread();
1657     trace_postcopy_start_set_run();
1658 
1659     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1660     *old_vm_running = runstate_is_running();
1661     global_state_store();
1662     ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1663     if (ret < 0) {
1664         goto fail;
1665     }
1666 
1667     ret = bdrv_inactivate_all();
1668     if (ret < 0) {
1669         goto fail;
1670     }
1671     restart_block = true;
1672 
1673     /*
1674      * Cause any non-postcopiable, but iterative devices to
1675      * send out their final data.
1676      */
1677     qemu_savevm_state_complete_precopy(ms->to_dst_file, true);
1678 
1679     /*
1680      * in Finish migrate and with the io-lock held everything should
1681      * be quiet, but we've potentially still got dirty pages and we
1682      * need to tell the destination to throw any pages it's already received
1683      * that are dirty
1684      */
1685     if (ram_postcopy_send_discard_bitmap(ms)) {
1686         error_report("postcopy send discard bitmap failed");
1687         goto fail;
1688     }
1689 
1690     /*
1691      * send rest of state - note things that are doing postcopy
1692      * will notice we're in POSTCOPY_ACTIVE and not actually
1693      * wrap their state up here
1694      */
1695     qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
1696     /* Ping just for debugging, helps line traces up */
1697     qemu_savevm_send_ping(ms->to_dst_file, 2);
1698 
1699     /*
1700      * While loading the device state we may trigger page transfer
1701      * requests and the fd must be free to process those, and thus
1702      * the destination must read the whole device state off the fd before
1703      * it starts processing it.  Unfortunately the ad-hoc migration format
1704      * doesn't allow the destination to know the size to read without fully
1705      * parsing it through each devices load-state code (especially the open
1706      * coded devices that use get/put).
1707      * So we wrap the device state up in a package with a length at the start;
1708      * to do this we use a qemu_buf to hold the whole of the device state.
1709      */
1710     bioc = qio_channel_buffer_new(4096);
1711     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
1712     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
1713     object_unref(OBJECT(bioc));
1714 
1715     /*
1716      * Make sure the receiver can get incoming pages before we send the rest
1717      * of the state
1718      */
1719     qemu_savevm_send_postcopy_listen(fb);
1720 
1721     qemu_savevm_state_complete_precopy(fb, false);
1722     qemu_savevm_send_ping(fb, 3);
1723 
1724     qemu_savevm_send_postcopy_run(fb);
1725 
1726     /* <><> end of stuff going into the package */
1727 
1728     /* Last point of recovery; as soon as we send the package the destination
1729      * can open devices and potentially start running.
1730      * Lets just check again we've not got any errors.
1731      */
1732     ret = qemu_file_get_error(ms->to_dst_file);
1733     if (ret) {
1734         error_report("postcopy_start: Migration stream errored (pre package)");
1735         goto fail_closefb;
1736     }
1737 
1738     restart_block = false;
1739 
1740     /* Now send that blob */
1741     if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
1742         goto fail_closefb;
1743     }
1744     qemu_fclose(fb);
1745 
1746     /* Send a notify to give a chance for anything that needs to happen
1747      * at the transition to postcopy and after the device state; in particular
1748      * spice needs to trigger a transition now
1749      */
1750     ms->postcopy_after_devices = true;
1751     notifier_list_notify(&migration_state_notifiers, ms);
1752 
1753     ms->downtime =  qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
1754 
1755     qemu_mutex_unlock_iothread();
1756 
1757     /*
1758      * Although this ping is just for debug, it could potentially be
1759      * used for getting a better measurement of downtime at the source.
1760      */
1761     qemu_savevm_send_ping(ms->to_dst_file, 4);
1762 
1763     if (migrate_release_ram()) {
1764         ram_postcopy_migrated_memory_release(ms);
1765     }
1766 
1767     ret = qemu_file_get_error(ms->to_dst_file);
1768     if (ret) {
1769         error_report("postcopy_start: Migration stream errored");
1770         migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1771                               MIGRATION_STATUS_FAILED);
1772     }
1773 
1774     return ret;
1775 
1776 fail_closefb:
1777     qemu_fclose(fb);
1778 fail:
1779     migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1780                           MIGRATION_STATUS_FAILED);
1781     if (restart_block) {
1782         /* A failure happened early enough that we know the destination hasn't
1783          * accessed block devices, so we're safe to recover.
1784          */
1785         Error *local_err = NULL;
1786 
1787         bdrv_invalidate_cache_all(&local_err);
1788         if (local_err) {
1789             error_report_err(local_err);
1790         }
1791     }
1792     qemu_mutex_unlock_iothread();
1793     return -1;
1794 }
1795 
1796 /**
1797  * migration_completion: Used by migration_thread when there's not much left.
1798  *   The caller 'breaks' the loop when this returns.
1799  *
1800  * @s: Current migration state
1801  * @current_active_state: The migration state we expect to be in
1802  * @*old_vm_running: Pointer to old_vm_running flag
1803  * @*start_time: Pointer to time to update
1804  */
1805 static void migration_completion(MigrationState *s, int current_active_state,
1806                                  bool *old_vm_running,
1807                                  int64_t *start_time)
1808 {
1809     int ret;
1810 
1811     if (s->state == MIGRATION_STATUS_ACTIVE) {
1812         qemu_mutex_lock_iothread();
1813         *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1814         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1815         *old_vm_running = runstate_is_running();
1816         ret = global_state_store();
1817 
1818         if (!ret) {
1819             ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1820             /*
1821              * Don't mark the image with BDRV_O_INACTIVE flag if
1822              * we will go into COLO stage later.
1823              */
1824             if (ret >= 0 && !migrate_colo_enabled()) {
1825                 ret = bdrv_inactivate_all();
1826             }
1827             if (ret >= 0) {
1828                 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
1829                 qemu_savevm_state_complete_precopy(s->to_dst_file, false);
1830                 s->block_inactive = true;
1831             }
1832         }
1833         qemu_mutex_unlock_iothread();
1834 
1835         if (ret < 0) {
1836             goto fail;
1837         }
1838     } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1839         trace_migration_completion_postcopy_end();
1840 
1841         qemu_savevm_state_complete_postcopy(s->to_dst_file);
1842         trace_migration_completion_postcopy_end_after_complete();
1843     }
1844 
1845     /*
1846      * If rp was opened we must clean up the thread before
1847      * cleaning everything else up (since if there are no failures
1848      * it will wait for the destination to send it's status in
1849      * a SHUT command).
1850      * Postcopy opens rp if enabled (even if it's not avtivated)
1851      */
1852     if (migrate_postcopy_ram()) {
1853         int rp_error;
1854         trace_migration_completion_postcopy_end_before_rp();
1855         rp_error = await_return_path_close_on_source(s);
1856         trace_migration_completion_postcopy_end_after_rp(rp_error);
1857         if (rp_error) {
1858             goto fail_invalidate;
1859         }
1860     }
1861 
1862     if (qemu_file_get_error(s->to_dst_file)) {
1863         trace_migration_completion_file_err();
1864         goto fail_invalidate;
1865     }
1866 
1867     if (!migrate_colo_enabled()) {
1868         migrate_set_state(&s->state, current_active_state,
1869                           MIGRATION_STATUS_COMPLETED);
1870     }
1871 
1872     return;
1873 
1874 fail_invalidate:
1875     /* If not doing postcopy, vm_start() will be called: let's regain
1876      * control on images.
1877      */
1878     if (s->state == MIGRATION_STATUS_ACTIVE) {
1879         Error *local_err = NULL;
1880 
1881         qemu_mutex_lock_iothread();
1882         bdrv_invalidate_cache_all(&local_err);
1883         if (local_err) {
1884             error_report_err(local_err);
1885         } else {
1886             s->block_inactive = false;
1887         }
1888         qemu_mutex_unlock_iothread();
1889     }
1890 
1891 fail:
1892     migrate_set_state(&s->state, current_active_state,
1893                       MIGRATION_STATUS_FAILED);
1894 }
1895 
1896 bool migrate_colo_enabled(void)
1897 {
1898     MigrationState *s = migrate_get_current();
1899     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
1900 }
1901 
1902 /*
1903  * Master migration thread on the source VM.
1904  * It drives the migration and pumps the data down the outgoing channel.
1905  */
1906 static void *migration_thread(void *opaque)
1907 {
1908     MigrationState *s = opaque;
1909     /* Used by the bandwidth calcs, updated later */
1910     int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1911     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
1912     int64_t initial_bytes = 0;
1913     /*
1914      * The final stage happens when the remaining data is smaller than
1915      * this threshold; it's calculated from the requested downtime and
1916      * measured bandwidth
1917      */
1918     int64_t threshold_size = 0;
1919     int64_t start_time = initial_time;
1920     int64_t end_time;
1921     bool old_vm_running = false;
1922     bool entered_postcopy = false;
1923     /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
1924     enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE;
1925     bool enable_colo = migrate_colo_enabled();
1926 
1927     rcu_register_thread();
1928 
1929     qemu_savevm_state_header(s->to_dst_file);
1930 
1931     if (migrate_postcopy_ram()) {
1932         /* Now tell the dest that it should open its end so it can reply */
1933         qemu_savevm_send_open_return_path(s->to_dst_file);
1934 
1935         /* And do a ping that will make stuff easier to debug */
1936         qemu_savevm_send_ping(s->to_dst_file, 1);
1937 
1938         /*
1939          * Tell the destination that we *might* want to do postcopy later;
1940          * if the other end can't do postcopy it should fail now, nice and
1941          * early.
1942          */
1943         qemu_savevm_send_postcopy_advise(s->to_dst_file);
1944     }
1945 
1946     qemu_savevm_state_begin(s->to_dst_file, &s->params);
1947 
1948     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
1949     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1950                       MIGRATION_STATUS_ACTIVE);
1951 
1952     trace_migration_thread_setup_complete();
1953 
1954     while (s->state == MIGRATION_STATUS_ACTIVE ||
1955            s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1956         int64_t current_time;
1957         uint64_t pending_size;
1958 
1959         if (!qemu_file_rate_limit(s->to_dst_file)) {
1960             uint64_t pend_post, pend_nonpost;
1961 
1962             qemu_savevm_state_pending(s->to_dst_file, threshold_size,
1963                                       &pend_nonpost, &pend_post);
1964             pending_size = pend_nonpost + pend_post;
1965             trace_migrate_pending(pending_size, threshold_size,
1966                                   pend_post, pend_nonpost);
1967             if (pending_size && pending_size >= threshold_size) {
1968                 /* Still a significant amount to transfer */
1969 
1970                 if (migrate_postcopy_ram() &&
1971                     s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE &&
1972                     pend_nonpost <= threshold_size &&
1973                     atomic_read(&s->start_postcopy)) {
1974 
1975                     if (!postcopy_start(s, &old_vm_running)) {
1976                         current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE;
1977                         entered_postcopy = true;
1978                     }
1979 
1980                     continue;
1981                 }
1982                 /* Just another iteration step */
1983                 qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy);
1984             } else {
1985                 trace_migration_thread_low_pending(pending_size);
1986                 migration_completion(s, current_active_state,
1987                                      &old_vm_running, &start_time);
1988                 break;
1989             }
1990         }
1991 
1992         if (qemu_file_get_error(s->to_dst_file)) {
1993             migrate_set_state(&s->state, current_active_state,
1994                               MIGRATION_STATUS_FAILED);
1995             trace_migration_thread_file_err();
1996             break;
1997         }
1998         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1999         if (current_time >= initial_time + BUFFER_DELAY) {
2000             uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) -
2001                                          initial_bytes;
2002             uint64_t time_spent = current_time - initial_time;
2003             double bandwidth = (double)transferred_bytes / time_spent;
2004             threshold_size = bandwidth * s->parameters.downtime_limit;
2005 
2006             s->mbps = (((double) transferred_bytes * 8.0) /
2007                     ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
2008 
2009             trace_migrate_transferred(transferred_bytes, time_spent,
2010                                       bandwidth, threshold_size);
2011             /* if we haven't sent anything, we don't want to recalculate
2012                10000 is a small enough number for our purposes */
2013             if (ram_dirty_pages_rate() && transferred_bytes > 10000) {
2014                 s->expected_downtime = ram_dirty_pages_rate() *
2015                     qemu_target_page_size() / bandwidth;
2016             }
2017 
2018             qemu_file_reset_rate_limit(s->to_dst_file);
2019             initial_time = current_time;
2020             initial_bytes = qemu_ftell(s->to_dst_file);
2021         }
2022         if (qemu_file_rate_limit(s->to_dst_file)) {
2023             /* usleep expects microseconds */
2024             g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
2025         }
2026     }
2027 
2028     trace_migration_thread_after_loop();
2029     /* If we enabled cpu throttling for auto-converge, turn it off. */
2030     cpu_throttle_stop();
2031     end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2032 
2033     qemu_mutex_lock_iothread();
2034     /*
2035      * The resource has been allocated by migration will be reused in COLO
2036      * process, so don't release them.
2037      */
2038     if (!enable_colo) {
2039         qemu_savevm_state_cleanup();
2040     }
2041     if (s->state == MIGRATION_STATUS_COMPLETED) {
2042         uint64_t transferred_bytes = qemu_ftell(s->to_dst_file);
2043         s->total_time = end_time - s->total_time;
2044         if (!entered_postcopy) {
2045             s->downtime = end_time - start_time;
2046         }
2047         if (s->total_time) {
2048             s->mbps = (((double) transferred_bytes * 8.0) /
2049                        ((double) s->total_time)) / 1000;
2050         }
2051         runstate_set(RUN_STATE_POSTMIGRATE);
2052     } else {
2053         if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) {
2054             migrate_start_colo_process(s);
2055             qemu_savevm_state_cleanup();
2056             /*
2057             * Fixme: we will run VM in COLO no matter its old running state.
2058             * After exited COLO, we will keep running.
2059             */
2060             old_vm_running = true;
2061         }
2062         if (old_vm_running && !entered_postcopy) {
2063             vm_start();
2064         } else {
2065             if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2066                 runstate_set(RUN_STATE_POSTMIGRATE);
2067             }
2068         }
2069     }
2070     qemu_bh_schedule(s->cleanup_bh);
2071     qemu_mutex_unlock_iothread();
2072 
2073     rcu_unregister_thread();
2074     return NULL;
2075 }
2076 
2077 void migrate_fd_connect(MigrationState *s)
2078 {
2079     s->expected_downtime = s->parameters.downtime_limit;
2080     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
2081 
2082     qemu_file_set_blocking(s->to_dst_file, true);
2083     qemu_file_set_rate_limit(s->to_dst_file,
2084                              s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
2085 
2086     /* Notify before starting migration thread */
2087     notifier_list_notify(&migration_state_notifiers, s);
2088 
2089     /*
2090      * Open the return path; currently for postcopy but other things might
2091      * also want it.
2092      */
2093     if (migrate_postcopy_ram()) {
2094         if (open_return_path_on_source(s)) {
2095             error_report("Unable to open return-path for postcopy");
2096             migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2097                               MIGRATION_STATUS_FAILED);
2098             migrate_fd_cleanup(s);
2099             return;
2100         }
2101     }
2102 
2103     migrate_compress_threads_create();
2104     qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
2105                        QEMU_THREAD_JOINABLE);
2106     s->migration_thread_running = true;
2107 }
2108 
2109 PostcopyState  postcopy_state_get(void)
2110 {
2111     return atomic_mb_read(&incoming_postcopy_state);
2112 }
2113 
2114 /* Set the state and return the old state */
2115 PostcopyState postcopy_state_set(PostcopyState new_state)
2116 {
2117     return atomic_xchg(&incoming_postcopy_state, new_state);
2118 }
2119 
2120