xref: /openbmc/qemu/migration/migration.c (revision 56e2cd24)
1 /*
2  * QEMU live migration
3  *
4  * Copyright IBM, Corp. 2008
5  *
6  * Authors:
7  *  Anthony Liguori   <aliguori@us.ibm.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  * Contributions after 2012-01-13 are licensed under the terms of the
13  * GNU GPL, version 2 or (at your option) any later version.
14  */
15 
16 #include "qemu/osdep.h"
17 #include "qemu/cutils.h"
18 #include "qemu/error-report.h"
19 #include "qemu/main-loop.h"
20 #include "migration/migration.h"
21 #include "migration/qemu-file.h"
22 #include "sysemu/sysemu.h"
23 #include "block/block.h"
24 #include "qapi/qmp/qerror.h"
25 #include "qapi/util.h"
26 #include "qemu/sockets.h"
27 #include "qemu/rcu.h"
28 #include "migration/block.h"
29 #include "migration/postcopy-ram.h"
30 #include "qemu/thread.h"
31 #include "qmp-commands.h"
32 #include "trace.h"
33 #include "qapi-event.h"
34 #include "qom/cpu.h"
35 #include "exec/memory.h"
36 #include "exec/address-spaces.h"
37 #include "io/channel-buffer.h"
38 #include "io/channel-tls.h"
39 #include "migration/colo.h"
40 
41 #define MAX_THROTTLE  (32 << 20)      /* Migration transfer speed throttling */
42 
43 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
44  * data. */
45 #define BUFFER_DELAY     100
46 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
47 
48 /* Time in milliseconds we are allowed to stop the source,
49  * for sending the last part */
50 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
51 
52 /* Maximum migrate downtime set to 2000 seconds */
53 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
54 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
55 
56 /* Default compression thread count */
57 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
58 /* Default decompression thread count, usually decompression is at
59  * least 4 times as fast as compression.*/
60 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
61 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
62 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
63 /* Define default autoconverge cpu throttle migration parameters */
64 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
65 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
66 
67 /* Migration XBZRLE default cache size */
68 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
69 
70 /* The delay time (in ms) between two COLO checkpoints
71  * Note: Please change this default value to 10000 when we support hybrid mode.
72  */
73 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
74 
75 static NotifierList migration_state_notifiers =
76     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
77 
78 static bool deferred_incoming;
79 
80 /*
81  * Current state of incoming postcopy; note this is not part of
82  * MigrationIncomingState since it's state is used during cleanup
83  * at the end as MIS is being freed.
84  */
85 static PostcopyState incoming_postcopy_state;
86 
87 /* When we add fault tolerance, we could have several
88    migrations at once.  For now we don't need to add
89    dynamic creation of migration */
90 
91 /* For outgoing */
92 MigrationState *migrate_get_current(void)
93 {
94     static bool once;
95     static MigrationState current_migration = {
96         .state = MIGRATION_STATUS_NONE,
97         .xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE,
98         .mbps = -1,
99         .parameters = {
100             .compress_level = DEFAULT_MIGRATE_COMPRESS_LEVEL,
101             .compress_threads = DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT,
102             .decompress_threads = DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT,
103             .cpu_throttle_initial = DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL,
104             .cpu_throttle_increment = DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT,
105             .max_bandwidth = MAX_THROTTLE,
106             .downtime_limit = DEFAULT_MIGRATE_SET_DOWNTIME,
107             .x_checkpoint_delay = DEFAULT_MIGRATE_X_CHECKPOINT_DELAY,
108         },
109     };
110 
111     if (!once) {
112         qemu_mutex_init(&current_migration.src_page_req_mutex);
113         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     bdrv_invalidate_cache_all(&local_err);
344     if (local_err) {
345         migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
346                           MIGRATION_STATUS_FAILED);
347         error_report_err(local_err);
348         migrate_decompress_threads_join();
349         exit(EXIT_FAILURE);
350     }
351 
352     /*
353      * This must happen after all error conditions are dealt with and
354      * we're sure the VM is going to be running on this host.
355      */
356     qemu_announce_self();
357 
358     /* If global state section was not received or we are in running
359        state, we need to obey autostart. Any other state is set with
360        runstate_set. */
361 
362     if (!global_state_received() ||
363         global_state_get_runstate() == RUN_STATE_RUNNING) {
364         if (autostart) {
365             vm_start();
366         } else {
367             runstate_set(RUN_STATE_PAUSED);
368         }
369     } else {
370         runstate_set(global_state_get_runstate());
371     }
372     migrate_decompress_threads_join();
373     /*
374      * This must happen after any state changes since as soon as an external
375      * observer sees this event they might start to prod at the VM assuming
376      * it's ready to use.
377      */
378     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
379                       MIGRATION_STATUS_COMPLETED);
380     qemu_bh_delete(mis->bh);
381     migration_incoming_state_destroy();
382 }
383 
384 static void process_incoming_migration_co(void *opaque)
385 {
386     QEMUFile *f = opaque;
387     MigrationIncomingState *mis = migration_incoming_get_current();
388     PostcopyState ps;
389     int ret;
390 
391     mis->from_src_file = f;
392     mis->largest_page_size = qemu_ram_pagesize_largest();
393     postcopy_state_set(POSTCOPY_INCOMING_NONE);
394     migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
395                       MIGRATION_STATUS_ACTIVE);
396     ret = qemu_loadvm_state(f);
397 
398     ps = postcopy_state_get();
399     trace_process_incoming_migration_co_end(ret, ps);
400     if (ps != POSTCOPY_INCOMING_NONE) {
401         if (ps == POSTCOPY_INCOMING_ADVISE) {
402             /*
403              * Where a migration had postcopy enabled (and thus went to advise)
404              * but managed to complete within the precopy period, we can use
405              * the normal exit.
406              */
407             postcopy_ram_incoming_cleanup(mis);
408         } else if (ret >= 0) {
409             /*
410              * Postcopy was started, cleanup should happen at the end of the
411              * postcopy thread.
412              */
413             trace_process_incoming_migration_co_postcopy_end_main();
414             return;
415         }
416         /* Else if something went wrong then just fall out of the normal exit */
417     }
418 
419     /* we get COLO info, and know if we are in COLO mode */
420     if (!ret && migration_incoming_enable_colo()) {
421         mis->migration_incoming_co = qemu_coroutine_self();
422         qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
423              colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
424         mis->have_colo_incoming_thread = true;
425         qemu_coroutine_yield();
426 
427         /* Wait checkpoint incoming thread exit before free resource */
428         qemu_thread_join(&mis->colo_incoming_thread);
429     }
430 
431     qemu_fclose(f);
432     free_xbzrle_decoded_buf();
433 
434     if (ret < 0) {
435         migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
436                           MIGRATION_STATUS_FAILED);
437         error_report("load of migration failed: %s", strerror(-ret));
438         migrate_decompress_threads_join();
439         exit(EXIT_FAILURE);
440     }
441 
442     mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
443     qemu_bh_schedule(mis->bh);
444 }
445 
446 void migration_fd_process_incoming(QEMUFile *f)
447 {
448     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, f);
449 
450     migrate_decompress_threads_create();
451     qemu_file_set_blocking(f, false);
452     qemu_coroutine_enter(co);
453 }
454 
455 
456 void migration_channel_process_incoming(MigrationState *s,
457                                         QIOChannel *ioc)
458 {
459     trace_migration_set_incoming_channel(
460         ioc, object_get_typename(OBJECT(ioc)));
461 
462     if (s->parameters.tls_creds &&
463         *s->parameters.tls_creds &&
464         !object_dynamic_cast(OBJECT(ioc),
465                              TYPE_QIO_CHANNEL_TLS)) {
466         Error *local_err = NULL;
467         migration_tls_channel_process_incoming(s, ioc, &local_err);
468         if (local_err) {
469             error_report_err(local_err);
470         }
471     } else {
472         QEMUFile *f = qemu_fopen_channel_input(ioc);
473         migration_fd_process_incoming(f);
474     }
475 }
476 
477 
478 void migration_channel_connect(MigrationState *s,
479                                QIOChannel *ioc,
480                                const char *hostname)
481 {
482     trace_migration_set_outgoing_channel(
483         ioc, object_get_typename(OBJECT(ioc)), hostname);
484 
485     if (s->parameters.tls_creds &&
486         *s->parameters.tls_creds &&
487         !object_dynamic_cast(OBJECT(ioc),
488                              TYPE_QIO_CHANNEL_TLS)) {
489         Error *local_err = NULL;
490         migration_tls_channel_connect(s, ioc, hostname, &local_err);
491         if (local_err) {
492             migrate_fd_error(s, local_err);
493             error_free(local_err);
494         }
495     } else {
496         QEMUFile *f = qemu_fopen_channel_output(ioc);
497 
498         s->to_dst_file = f;
499 
500         migrate_fd_connect(s);
501     }
502 }
503 
504 
505 /*
506  * Send a message on the return channel back to the source
507  * of the migration.
508  */
509 void migrate_send_rp_message(MigrationIncomingState *mis,
510                              enum mig_rp_message_type message_type,
511                              uint16_t len, void *data)
512 {
513     trace_migrate_send_rp_message((int)message_type, len);
514     qemu_mutex_lock(&mis->rp_mutex);
515     qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
516     qemu_put_be16(mis->to_src_file, len);
517     qemu_put_buffer(mis->to_src_file, data, len);
518     qemu_fflush(mis->to_src_file);
519     qemu_mutex_unlock(&mis->rp_mutex);
520 }
521 
522 /*
523  * Send a 'SHUT' message on the return channel with the given value
524  * to indicate that we've finished with the RP.  Non-0 value indicates
525  * error.
526  */
527 void migrate_send_rp_shut(MigrationIncomingState *mis,
528                           uint32_t value)
529 {
530     uint32_t buf;
531 
532     buf = cpu_to_be32(value);
533     migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
534 }
535 
536 /*
537  * Send a 'PONG' message on the return channel with the given value
538  * (normally in response to a 'PING')
539  */
540 void migrate_send_rp_pong(MigrationIncomingState *mis,
541                           uint32_t value)
542 {
543     uint32_t buf;
544 
545     buf = cpu_to_be32(value);
546     migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
547 }
548 
549 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
550 {
551     MigrationCapabilityStatusList *head = NULL;
552     MigrationCapabilityStatusList *caps;
553     MigrationState *s = migrate_get_current();
554     int i;
555 
556     caps = NULL; /* silence compiler warning */
557     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
558         if (i == MIGRATION_CAPABILITY_X_COLO && !colo_supported()) {
559             continue;
560         }
561         if (head == NULL) {
562             head = g_malloc0(sizeof(*caps));
563             caps = head;
564         } else {
565             caps->next = g_malloc0(sizeof(*caps));
566             caps = caps->next;
567         }
568         caps->value =
569             g_malloc(sizeof(*caps->value));
570         caps->value->capability = i;
571         caps->value->state = s->enabled_capabilities[i];
572     }
573 
574     return head;
575 }
576 
577 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
578 {
579     MigrationParameters *params;
580     MigrationState *s = migrate_get_current();
581 
582     params = g_malloc0(sizeof(*params));
583     params->has_compress_level = true;
584     params->compress_level = s->parameters.compress_level;
585     params->has_compress_threads = true;
586     params->compress_threads = s->parameters.compress_threads;
587     params->has_decompress_threads = true;
588     params->decompress_threads = s->parameters.decompress_threads;
589     params->has_cpu_throttle_initial = true;
590     params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
591     params->has_cpu_throttle_increment = true;
592     params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
593     params->has_tls_creds = !!s->parameters.tls_creds;
594     params->tls_creds = g_strdup(s->parameters.tls_creds);
595     params->has_tls_hostname = !!s->parameters.tls_hostname;
596     params->tls_hostname = g_strdup(s->parameters.tls_hostname);
597     params->has_max_bandwidth = true;
598     params->max_bandwidth = s->parameters.max_bandwidth;
599     params->has_downtime_limit = true;
600     params->downtime_limit = s->parameters.downtime_limit;
601     params->has_x_checkpoint_delay = true;
602     params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
603 
604     return params;
605 }
606 
607 /*
608  * Return true if we're already in the middle of a migration
609  * (i.e. any of the active or setup states)
610  */
611 static bool migration_is_setup_or_active(int state)
612 {
613     switch (state) {
614     case MIGRATION_STATUS_ACTIVE:
615     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
616     case MIGRATION_STATUS_SETUP:
617         return true;
618 
619     default:
620         return false;
621 
622     }
623 }
624 
625 static void get_xbzrle_cache_stats(MigrationInfo *info)
626 {
627     if (migrate_use_xbzrle()) {
628         info->has_xbzrle_cache = true;
629         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
630         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
631         info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred();
632         info->xbzrle_cache->pages = xbzrle_mig_pages_transferred();
633         info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss();
634         info->xbzrle_cache->cache_miss_rate = xbzrle_mig_cache_miss_rate();
635         info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow();
636     }
637 }
638 
639 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
640 {
641     info->has_ram = true;
642     info->ram = g_malloc0(sizeof(*info->ram));
643     info->ram->transferred = ram_bytes_transferred();
644     info->ram->total = ram_bytes_total();
645     info->ram->duplicate = dup_mig_pages_transferred();
646     info->ram->skipped = skipped_mig_pages_transferred();
647     info->ram->normal = norm_mig_pages_transferred();
648     info->ram->normal_bytes = norm_mig_bytes_transferred();
649     info->ram->mbps = s->mbps;
650     info->ram->dirty_sync_count = s->dirty_sync_count;
651     info->ram->postcopy_requests = s->postcopy_requests;
652 
653     if (s->state != MIGRATION_STATUS_COMPLETED) {
654         info->ram->remaining = ram_bytes_remaining();
655         info->ram->dirty_pages_rate = s->dirty_pages_rate;
656     }
657 }
658 
659 MigrationInfo *qmp_query_migrate(Error **errp)
660 {
661     MigrationInfo *info = g_malloc0(sizeof(*info));
662     MigrationState *s = migrate_get_current();
663 
664     switch (s->state) {
665     case MIGRATION_STATUS_NONE:
666         /* no migration has happened ever */
667         break;
668     case MIGRATION_STATUS_SETUP:
669         info->has_status = true;
670         info->has_total_time = false;
671         break;
672     case MIGRATION_STATUS_ACTIVE:
673     case MIGRATION_STATUS_CANCELLING:
674         info->has_status = true;
675         info->has_total_time = true;
676         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
677             - s->total_time;
678         info->has_expected_downtime = true;
679         info->expected_downtime = s->expected_downtime;
680         info->has_setup_time = true;
681         info->setup_time = s->setup_time;
682 
683         populate_ram_info(info, s);
684 
685         if (blk_mig_active()) {
686             info->has_disk = true;
687             info->disk = g_malloc0(sizeof(*info->disk));
688             info->disk->transferred = blk_mig_bytes_transferred();
689             info->disk->remaining = blk_mig_bytes_remaining();
690             info->disk->total = blk_mig_bytes_total();
691         }
692 
693         if (cpu_throttle_active()) {
694             info->has_cpu_throttle_percentage = true;
695             info->cpu_throttle_percentage = cpu_throttle_get_percentage();
696         }
697 
698         get_xbzrle_cache_stats(info);
699         break;
700     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
701         /* Mostly the same as active; TODO add some postcopy stats */
702         info->has_status = true;
703         info->has_total_time = true;
704         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
705             - s->total_time;
706         info->has_expected_downtime = true;
707         info->expected_downtime = s->expected_downtime;
708         info->has_setup_time = true;
709         info->setup_time = s->setup_time;
710 
711         populate_ram_info(info, s);
712 
713         if (blk_mig_active()) {
714             info->has_disk = true;
715             info->disk = g_malloc0(sizeof(*info->disk));
716             info->disk->transferred = blk_mig_bytes_transferred();
717             info->disk->remaining = blk_mig_bytes_remaining();
718             info->disk->total = blk_mig_bytes_total();
719         }
720 
721         get_xbzrle_cache_stats(info);
722         break;
723     case MIGRATION_STATUS_COLO:
724         info->has_status = true;
725         /* TODO: display COLO specific information (checkpoint info etc.) */
726         break;
727     case MIGRATION_STATUS_COMPLETED:
728         get_xbzrle_cache_stats(info);
729 
730         info->has_status = true;
731         info->has_total_time = true;
732         info->total_time = s->total_time;
733         info->has_downtime = true;
734         info->downtime = s->downtime;
735         info->has_setup_time = true;
736         info->setup_time = s->setup_time;
737 
738         populate_ram_info(info, s);
739         break;
740     case MIGRATION_STATUS_FAILED:
741         info->has_status = true;
742         if (s->error) {
743             info->has_error_desc = true;
744             info->error_desc = g_strdup(error_get_pretty(s->error));
745         }
746         break;
747     case MIGRATION_STATUS_CANCELLED:
748         info->has_status = true;
749         break;
750     }
751     info->status = s->state;
752 
753     return info;
754 }
755 
756 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
757                                   Error **errp)
758 {
759     MigrationState *s = migrate_get_current();
760     MigrationCapabilityStatusList *cap;
761     bool old_postcopy_cap = migrate_postcopy_ram();
762 
763     if (migration_is_setup_or_active(s->state)) {
764         error_setg(errp, QERR_MIGRATION_ACTIVE);
765         return;
766     }
767 
768     for (cap = params; cap; cap = cap->next) {
769         if (cap->value->capability == MIGRATION_CAPABILITY_X_COLO) {
770             if (!colo_supported()) {
771                 error_setg(errp, "COLO is not currently supported, please"
772                              " configure with --enable-colo option in order to"
773                              " support COLO feature");
774                 continue;
775             }
776         }
777         s->enabled_capabilities[cap->value->capability] = cap->value->state;
778     }
779 
780     if (migrate_postcopy_ram()) {
781         if (migrate_use_compression()) {
782             /* The decompression threads asynchronously write into RAM
783              * rather than use the atomic copies needed to avoid
784              * userfaulting.  It should be possible to fix the decompression
785              * threads for compatibility in future.
786              */
787             error_report("Postcopy is not currently compatible with "
788                          "compression");
789             s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] =
790                 false;
791         }
792         /* This check is reasonably expensive, so only when it's being
793          * set the first time, also it's only the destination that needs
794          * special support.
795          */
796         if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
797             !postcopy_ram_supported_by_host()) {
798             /* postcopy_ram_supported_by_host will have emitted a more
799              * detailed message
800              */
801             error_report("Postcopy is not supported");
802             s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] =
803                 false;
804         }
805     }
806 }
807 
808 void qmp_migrate_set_parameters(MigrationParameters *params, Error **errp)
809 {
810     MigrationState *s = migrate_get_current();
811 
812     if (params->has_compress_level &&
813         (params->compress_level < 0 || params->compress_level > 9)) {
814         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
815                    "is invalid, it should be in the range of 0 to 9");
816         return;
817     }
818     if (params->has_compress_threads &&
819         (params->compress_threads < 1 || params->compress_threads > 255)) {
820         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
821                    "compress_threads",
822                    "is invalid, it should be in the range of 1 to 255");
823         return;
824     }
825     if (params->has_decompress_threads &&
826         (params->decompress_threads < 1 || params->decompress_threads > 255)) {
827         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
828                    "decompress_threads",
829                    "is invalid, it should be in the range of 1 to 255");
830         return;
831     }
832     if (params->has_cpu_throttle_initial &&
833         (params->cpu_throttle_initial < 1 ||
834          params->cpu_throttle_initial > 99)) {
835         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
836                    "cpu_throttle_initial",
837                    "an integer in the range of 1 to 99");
838         return;
839     }
840     if (params->has_cpu_throttle_increment &&
841         (params->cpu_throttle_increment < 1 ||
842          params->cpu_throttle_increment > 99)) {
843         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
844                    "cpu_throttle_increment",
845                    "an integer in the range of 1 to 99");
846         return;
847     }
848     if (params->has_max_bandwidth &&
849         (params->max_bandwidth < 0 || params->max_bandwidth > SIZE_MAX)) {
850         error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
851                          " range of 0 to %zu bytes/second", SIZE_MAX);
852         return;
853     }
854     if (params->has_downtime_limit &&
855         (params->downtime_limit < 0 ||
856          params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
857         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
858                          "the range of 0 to %d milliseconds",
859                          MAX_MIGRATE_DOWNTIME);
860         return;
861     }
862     if (params->has_x_checkpoint_delay && (params->x_checkpoint_delay < 0)) {
863         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
864                     "x_checkpoint_delay",
865                     "is invalid, it should be positive");
866     }
867 
868     if (params->has_compress_level) {
869         s->parameters.compress_level = params->compress_level;
870     }
871     if (params->has_compress_threads) {
872         s->parameters.compress_threads = params->compress_threads;
873     }
874     if (params->has_decompress_threads) {
875         s->parameters.decompress_threads = params->decompress_threads;
876     }
877     if (params->has_cpu_throttle_initial) {
878         s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
879     }
880     if (params->has_cpu_throttle_increment) {
881         s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
882     }
883     if (params->has_tls_creds) {
884         g_free(s->parameters.tls_creds);
885         s->parameters.tls_creds = g_strdup(params->tls_creds);
886     }
887     if (params->has_tls_hostname) {
888         g_free(s->parameters.tls_hostname);
889         s->parameters.tls_hostname = g_strdup(params->tls_hostname);
890     }
891     if (params->has_max_bandwidth) {
892         s->parameters.max_bandwidth = params->max_bandwidth;
893         if (s->to_dst_file) {
894             qemu_file_set_rate_limit(s->to_dst_file,
895                                 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
896         }
897     }
898     if (params->has_downtime_limit) {
899         s->parameters.downtime_limit = params->downtime_limit;
900     }
901 
902     if (params->has_x_checkpoint_delay) {
903         s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
904         if (migration_in_colo_state()) {
905             colo_checkpoint_notify(s);
906         }
907     }
908 }
909 
910 
911 void qmp_migrate_start_postcopy(Error **errp)
912 {
913     MigrationState *s = migrate_get_current();
914 
915     if (!migrate_postcopy_ram()) {
916         error_setg(errp, "Enable postcopy with migrate_set_capability before"
917                          " the start of migration");
918         return;
919     }
920 
921     if (s->state == MIGRATION_STATUS_NONE) {
922         error_setg(errp, "Postcopy must be started after migration has been"
923                          " started");
924         return;
925     }
926     /*
927      * we don't error if migration has finished since that would be racy
928      * with issuing this command.
929      */
930     atomic_set(&s->start_postcopy, true);
931 }
932 
933 /* shared migration helpers */
934 
935 void migrate_set_state(int *state, int old_state, int new_state)
936 {
937     if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
938         trace_migrate_set_state(new_state);
939         migrate_generate_event(new_state);
940     }
941 }
942 
943 static void migrate_fd_cleanup(void *opaque)
944 {
945     MigrationState *s = opaque;
946 
947     qemu_bh_delete(s->cleanup_bh);
948     s->cleanup_bh = NULL;
949 
950     flush_page_queue(s);
951 
952     if (s->to_dst_file) {
953         trace_migrate_fd_cleanup();
954         qemu_mutex_unlock_iothread();
955         if (s->migration_thread_running) {
956             qemu_thread_join(&s->thread);
957             s->migration_thread_running = false;
958         }
959         qemu_mutex_lock_iothread();
960 
961         migrate_compress_threads_join();
962         qemu_fclose(s->to_dst_file);
963         s->to_dst_file = NULL;
964     }
965 
966     assert((s->state != MIGRATION_STATUS_ACTIVE) &&
967            (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
968 
969     if (s->state == MIGRATION_STATUS_CANCELLING) {
970         migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
971                           MIGRATION_STATUS_CANCELLED);
972     }
973 
974     notifier_list_notify(&migration_state_notifiers, s);
975 }
976 
977 void migrate_fd_error(MigrationState *s, const Error *error)
978 {
979     trace_migrate_fd_error(error_get_pretty(error));
980     assert(s->to_dst_file == NULL);
981     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
982                       MIGRATION_STATUS_FAILED);
983     if (!s->error) {
984         s->error = error_copy(error);
985     }
986     notifier_list_notify(&migration_state_notifiers, s);
987 }
988 
989 static void migrate_fd_cancel(MigrationState *s)
990 {
991     int old_state ;
992     QEMUFile *f = migrate_get_current()->to_dst_file;
993     trace_migrate_fd_cancel();
994 
995     if (s->rp_state.from_dst_file) {
996         /* shutdown the rp socket, so causing the rp thread to shutdown */
997         qemu_file_shutdown(s->rp_state.from_dst_file);
998     }
999 
1000     do {
1001         old_state = s->state;
1002         if (!migration_is_setup_or_active(old_state)) {
1003             break;
1004         }
1005         migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1006     } while (s->state != MIGRATION_STATUS_CANCELLING);
1007 
1008     /*
1009      * If we're unlucky the migration code might be stuck somewhere in a
1010      * send/write while the network has failed and is waiting to timeout;
1011      * if we've got shutdown(2) available then we can force it to quit.
1012      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1013      * called in a bh, so there is no race against this cancel.
1014      */
1015     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1016         qemu_file_shutdown(f);
1017     }
1018     if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1019         Error *local_err = NULL;
1020 
1021         bdrv_invalidate_cache_all(&local_err);
1022         if (local_err) {
1023             error_report_err(local_err);
1024         } else {
1025             s->block_inactive = false;
1026         }
1027     }
1028 }
1029 
1030 void add_migration_state_change_notifier(Notifier *notify)
1031 {
1032     notifier_list_add(&migration_state_notifiers, notify);
1033 }
1034 
1035 void remove_migration_state_change_notifier(Notifier *notify)
1036 {
1037     notifier_remove(notify);
1038 }
1039 
1040 bool migration_in_setup(MigrationState *s)
1041 {
1042     return s->state == MIGRATION_STATUS_SETUP;
1043 }
1044 
1045 bool migration_has_finished(MigrationState *s)
1046 {
1047     return s->state == MIGRATION_STATUS_COMPLETED;
1048 }
1049 
1050 bool migration_has_failed(MigrationState *s)
1051 {
1052     return (s->state == MIGRATION_STATUS_CANCELLED ||
1053             s->state == MIGRATION_STATUS_FAILED);
1054 }
1055 
1056 bool migration_in_postcopy(MigrationState *s)
1057 {
1058     return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1059 }
1060 
1061 bool migration_in_postcopy_after_devices(MigrationState *s)
1062 {
1063     return migration_in_postcopy(s) && s->postcopy_after_devices;
1064 }
1065 
1066 bool migration_is_idle(MigrationState *s)
1067 {
1068     if (!s) {
1069         s = migrate_get_current();
1070     }
1071 
1072     switch (s->state) {
1073     case MIGRATION_STATUS_NONE:
1074     case MIGRATION_STATUS_CANCELLED:
1075     case MIGRATION_STATUS_COMPLETED:
1076     case MIGRATION_STATUS_FAILED:
1077         return true;
1078     case MIGRATION_STATUS_SETUP:
1079     case MIGRATION_STATUS_CANCELLING:
1080     case MIGRATION_STATUS_ACTIVE:
1081     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1082     case MIGRATION_STATUS_COLO:
1083         return false;
1084     case MIGRATION_STATUS__MAX:
1085         g_assert_not_reached();
1086     }
1087 
1088     return false;
1089 }
1090 
1091 MigrationState *migrate_init(const MigrationParams *params)
1092 {
1093     MigrationState *s = migrate_get_current();
1094 
1095     /*
1096      * Reinitialise all migration state, except
1097      * parameters/capabilities that the user set, and
1098      * locks.
1099      */
1100     s->bytes_xfer = 0;
1101     s->xfer_limit = 0;
1102     s->cleanup_bh = 0;
1103     s->to_dst_file = NULL;
1104     s->state = MIGRATION_STATUS_NONE;
1105     s->params = *params;
1106     s->rp_state.from_dst_file = NULL;
1107     s->rp_state.error = false;
1108     s->mbps = 0.0;
1109     s->downtime = 0;
1110     s->expected_downtime = 0;
1111     s->dirty_pages_rate = 0;
1112     s->dirty_bytes_rate = 0;
1113     s->setup_time = 0;
1114     s->dirty_sync_count = 0;
1115     s->start_postcopy = false;
1116     s->postcopy_after_devices = false;
1117     s->postcopy_requests = 0;
1118     s->migration_thread_running = false;
1119     s->last_req_rb = NULL;
1120     error_free(s->error);
1121     s->error = NULL;
1122 
1123     migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1124 
1125     QSIMPLEQ_INIT(&s->src_page_requests);
1126 
1127     s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1128     return s;
1129 }
1130 
1131 static GSList *migration_blockers;
1132 
1133 int migrate_add_blocker(Error *reason, Error **errp)
1134 {
1135     if (only_migratable) {
1136         error_propagate(errp, error_copy(reason));
1137         error_prepend(errp, "disallowing migration blocker "
1138                           "(--only_migratable) for: ");
1139         return -EACCES;
1140     }
1141 
1142     if (migration_is_idle(NULL)) {
1143         migration_blockers = g_slist_prepend(migration_blockers, reason);
1144         return 0;
1145     }
1146 
1147     error_propagate(errp, error_copy(reason));
1148     error_prepend(errp, "disallowing migration blocker (migration in "
1149                       "progress) for: ");
1150     return -EBUSY;
1151 }
1152 
1153 void migrate_del_blocker(Error *reason)
1154 {
1155     migration_blockers = g_slist_remove(migration_blockers, reason);
1156 }
1157 
1158 int check_migratable(Object *obj, Error **err)
1159 {
1160     DeviceClass *dc = DEVICE_GET_CLASS(obj);
1161     if (only_migratable && dc->vmsd) {
1162         if (dc->vmsd->unmigratable) {
1163             error_setg(err, "Device %s is not migratable, but "
1164                        "--only-migratable was specified",
1165                        object_get_typename(obj));
1166             return -1;
1167         }
1168     }
1169 
1170     return 0;
1171 }
1172 
1173 void qmp_migrate_incoming(const char *uri, Error **errp)
1174 {
1175     Error *local_err = NULL;
1176     static bool once = true;
1177 
1178     if (!deferred_incoming) {
1179         error_setg(errp, "For use with '-incoming defer'");
1180         return;
1181     }
1182     if (!once) {
1183         error_setg(errp, "The incoming migration has already been started");
1184     }
1185 
1186     qemu_start_incoming_migration(uri, &local_err);
1187 
1188     if (local_err) {
1189         error_propagate(errp, local_err);
1190         return;
1191     }
1192 
1193     once = false;
1194 }
1195 
1196 bool migration_is_blocked(Error **errp)
1197 {
1198     if (qemu_savevm_state_blocked(errp)) {
1199         return true;
1200     }
1201 
1202     if (migration_blockers) {
1203         *errp = error_copy(migration_blockers->data);
1204         return true;
1205     }
1206 
1207     return false;
1208 }
1209 
1210 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1211                  bool has_inc, bool inc, bool has_detach, bool detach,
1212                  Error **errp)
1213 {
1214     Error *local_err = NULL;
1215     MigrationState *s = migrate_get_current();
1216     MigrationParams params;
1217     const char *p;
1218 
1219     params.blk = has_blk && blk;
1220     params.shared = has_inc && inc;
1221 
1222     if (migration_is_setup_or_active(s->state) ||
1223         s->state == MIGRATION_STATUS_CANCELLING ||
1224         s->state == MIGRATION_STATUS_COLO) {
1225         error_setg(errp, QERR_MIGRATION_ACTIVE);
1226         return;
1227     }
1228     if (runstate_check(RUN_STATE_INMIGRATE)) {
1229         error_setg(errp, "Guest is waiting for an incoming migration");
1230         return;
1231     }
1232 
1233     if (migration_is_blocked(errp)) {
1234         return;
1235     }
1236 
1237     s = migrate_init(&params);
1238 
1239     if (strstart(uri, "tcp:", &p)) {
1240         tcp_start_outgoing_migration(s, p, &local_err);
1241 #ifdef CONFIG_RDMA
1242     } else if (strstart(uri, "rdma:", &p)) {
1243         rdma_start_outgoing_migration(s, p, &local_err);
1244 #endif
1245     } else if (strstart(uri, "exec:", &p)) {
1246         exec_start_outgoing_migration(s, p, &local_err);
1247     } else if (strstart(uri, "unix:", &p)) {
1248         unix_start_outgoing_migration(s, p, &local_err);
1249     } else if (strstart(uri, "fd:", &p)) {
1250         fd_start_outgoing_migration(s, p, &local_err);
1251     } else {
1252         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1253                    "a valid migration protocol");
1254         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1255                           MIGRATION_STATUS_FAILED);
1256         return;
1257     }
1258 
1259     if (local_err) {
1260         migrate_fd_error(s, local_err);
1261         error_propagate(errp, local_err);
1262         return;
1263     }
1264 }
1265 
1266 void qmp_migrate_cancel(Error **errp)
1267 {
1268     migrate_fd_cancel(migrate_get_current());
1269 }
1270 
1271 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1272 {
1273     MigrationState *s = migrate_get_current();
1274     int64_t new_size;
1275 
1276     /* Check for truncation */
1277     if (value != (size_t)value) {
1278         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1279                    "exceeding address space");
1280         return;
1281     }
1282 
1283     /* Cache should not be larger than guest ram size */
1284     if (value > ram_bytes_total()) {
1285         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1286                    "exceeds guest ram size ");
1287         return;
1288     }
1289 
1290     new_size = xbzrle_cache_resize(value);
1291     if (new_size < 0) {
1292         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1293                    "is smaller than page size");
1294         return;
1295     }
1296 
1297     s->xbzrle_cache_size = new_size;
1298 }
1299 
1300 int64_t qmp_query_migrate_cache_size(Error **errp)
1301 {
1302     return migrate_xbzrle_cache_size();
1303 }
1304 
1305 void qmp_migrate_set_speed(int64_t value, Error **errp)
1306 {
1307     MigrationParameters p = {
1308         .has_max_bandwidth = true,
1309         .max_bandwidth = value,
1310     };
1311 
1312     qmp_migrate_set_parameters(&p, errp);
1313 }
1314 
1315 void qmp_migrate_set_downtime(double value, Error **errp)
1316 {
1317     if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1318         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1319                          "the range of 0 to %d seconds",
1320                          MAX_MIGRATE_DOWNTIME_SECONDS);
1321         return;
1322     }
1323 
1324     value *= 1000; /* Convert to milliseconds */
1325     value = MAX(0, MIN(INT64_MAX, value));
1326 
1327     MigrationParameters p = {
1328         .has_downtime_limit = true,
1329         .downtime_limit = value,
1330     };
1331 
1332     qmp_migrate_set_parameters(&p, errp);
1333 }
1334 
1335 bool migrate_release_ram(void)
1336 {
1337     MigrationState *s;
1338 
1339     s = migrate_get_current();
1340 
1341     return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1342 }
1343 
1344 bool migrate_postcopy_ram(void)
1345 {
1346     MigrationState *s;
1347 
1348     s = migrate_get_current();
1349 
1350     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1351 }
1352 
1353 bool migrate_auto_converge(void)
1354 {
1355     MigrationState *s;
1356 
1357     s = migrate_get_current();
1358 
1359     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1360 }
1361 
1362 bool migrate_zero_blocks(void)
1363 {
1364     MigrationState *s;
1365 
1366     s = migrate_get_current();
1367 
1368     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1369 }
1370 
1371 bool migrate_use_compression(void)
1372 {
1373     MigrationState *s;
1374 
1375     s = migrate_get_current();
1376 
1377     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
1378 }
1379 
1380 int migrate_compress_level(void)
1381 {
1382     MigrationState *s;
1383 
1384     s = migrate_get_current();
1385 
1386     return s->parameters.compress_level;
1387 }
1388 
1389 int migrate_compress_threads(void)
1390 {
1391     MigrationState *s;
1392 
1393     s = migrate_get_current();
1394 
1395     return s->parameters.compress_threads;
1396 }
1397 
1398 int migrate_decompress_threads(void)
1399 {
1400     MigrationState *s;
1401 
1402     s = migrate_get_current();
1403 
1404     return s->parameters.decompress_threads;
1405 }
1406 
1407 bool migrate_use_events(void)
1408 {
1409     MigrationState *s;
1410 
1411     s = migrate_get_current();
1412 
1413     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1414 }
1415 
1416 int migrate_use_xbzrle(void)
1417 {
1418     MigrationState *s;
1419 
1420     s = migrate_get_current();
1421 
1422     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1423 }
1424 
1425 int64_t migrate_xbzrle_cache_size(void)
1426 {
1427     MigrationState *s;
1428 
1429     s = migrate_get_current();
1430 
1431     return s->xbzrle_cache_size;
1432 }
1433 
1434 /* migration thread support */
1435 /*
1436  * Something bad happened to the RP stream, mark an error
1437  * The caller shall print or trace something to indicate why
1438  */
1439 static void mark_source_rp_bad(MigrationState *s)
1440 {
1441     s->rp_state.error = true;
1442 }
1443 
1444 static struct rp_cmd_args {
1445     ssize_t     len; /* -1 = variable */
1446     const char *name;
1447 } rp_cmd_args[] = {
1448     [MIG_RP_MSG_INVALID]        = { .len = -1, .name = "INVALID" },
1449     [MIG_RP_MSG_SHUT]           = { .len =  4, .name = "SHUT" },
1450     [MIG_RP_MSG_PONG]           = { .len =  4, .name = "PONG" },
1451     [MIG_RP_MSG_REQ_PAGES]      = { .len = 12, .name = "REQ_PAGES" },
1452     [MIG_RP_MSG_REQ_PAGES_ID]   = { .len = -1, .name = "REQ_PAGES_ID" },
1453     [MIG_RP_MSG_MAX]            = { .len = -1, .name = "MAX" },
1454 };
1455 
1456 /*
1457  * Process a request for pages received on the return path,
1458  * We're allowed to send more than requested (e.g. to round to our page size)
1459  * and we don't need to send pages that have already been sent.
1460  */
1461 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1462                                        ram_addr_t start, size_t len)
1463 {
1464     long our_host_ps = getpagesize();
1465 
1466     trace_migrate_handle_rp_req_pages(rbname, start, len);
1467 
1468     /*
1469      * Since we currently insist on matching page sizes, just sanity check
1470      * we're being asked for whole host pages.
1471      */
1472     if (start & (our_host_ps-1) ||
1473        (len & (our_host_ps-1))) {
1474         error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1475                      " len: %zd", __func__, start, len);
1476         mark_source_rp_bad(ms);
1477         return;
1478     }
1479 
1480     if (ram_save_queue_pages(ms, rbname, start, len)) {
1481         mark_source_rp_bad(ms);
1482     }
1483 }
1484 
1485 /*
1486  * Handles messages sent on the return path towards the source VM
1487  *
1488  */
1489 static void *source_return_path_thread(void *opaque)
1490 {
1491     MigrationState *ms = opaque;
1492     QEMUFile *rp = ms->rp_state.from_dst_file;
1493     uint16_t header_len, header_type;
1494     uint8_t buf[512];
1495     uint32_t tmp32, sibling_error;
1496     ram_addr_t start = 0; /* =0 to silence warning */
1497     size_t  len = 0, expected_len;
1498     int res;
1499 
1500     trace_source_return_path_thread_entry();
1501     while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
1502            migration_is_setup_or_active(ms->state)) {
1503         trace_source_return_path_thread_loop_top();
1504         header_type = qemu_get_be16(rp);
1505         header_len = qemu_get_be16(rp);
1506 
1507         if (header_type >= MIG_RP_MSG_MAX ||
1508             header_type == MIG_RP_MSG_INVALID) {
1509             error_report("RP: Received invalid message 0x%04x length 0x%04x",
1510                     header_type, header_len);
1511             mark_source_rp_bad(ms);
1512             goto out;
1513         }
1514 
1515         if ((rp_cmd_args[header_type].len != -1 &&
1516             header_len != rp_cmd_args[header_type].len) ||
1517             header_len > sizeof(buf)) {
1518             error_report("RP: Received '%s' message (0x%04x) with"
1519                     "incorrect length %d expecting %zu",
1520                     rp_cmd_args[header_type].name, header_type, header_len,
1521                     (size_t)rp_cmd_args[header_type].len);
1522             mark_source_rp_bad(ms);
1523             goto out;
1524         }
1525 
1526         /* We know we've got a valid header by this point */
1527         res = qemu_get_buffer(rp, buf, header_len);
1528         if (res != header_len) {
1529             error_report("RP: Failed reading data for message 0x%04x"
1530                          " read %d expected %d",
1531                          header_type, res, header_len);
1532             mark_source_rp_bad(ms);
1533             goto out;
1534         }
1535 
1536         /* OK, we have the message and the data */
1537         switch (header_type) {
1538         case MIG_RP_MSG_SHUT:
1539             sibling_error = ldl_be_p(buf);
1540             trace_source_return_path_thread_shut(sibling_error);
1541             if (sibling_error) {
1542                 error_report("RP: Sibling indicated error %d", sibling_error);
1543                 mark_source_rp_bad(ms);
1544             }
1545             /*
1546              * We'll let the main thread deal with closing the RP
1547              * we could do a shutdown(2) on it, but we're the only user
1548              * anyway, so there's nothing gained.
1549              */
1550             goto out;
1551 
1552         case MIG_RP_MSG_PONG:
1553             tmp32 = ldl_be_p(buf);
1554             trace_source_return_path_thread_pong(tmp32);
1555             break;
1556 
1557         case MIG_RP_MSG_REQ_PAGES:
1558             start = ldq_be_p(buf);
1559             len = ldl_be_p(buf + 8);
1560             migrate_handle_rp_req_pages(ms, NULL, start, len);
1561             break;
1562 
1563         case MIG_RP_MSG_REQ_PAGES_ID:
1564             expected_len = 12 + 1; /* header + termination */
1565 
1566             if (header_len >= expected_len) {
1567                 start = ldq_be_p(buf);
1568                 len = ldl_be_p(buf + 8);
1569                 /* Now we expect an idstr */
1570                 tmp32 = buf[12]; /* Length of the following idstr */
1571                 buf[13 + tmp32] = '\0';
1572                 expected_len += tmp32;
1573             }
1574             if (header_len != expected_len) {
1575                 error_report("RP: Req_Page_id with length %d expecting %zd",
1576                         header_len, expected_len);
1577                 mark_source_rp_bad(ms);
1578                 goto out;
1579             }
1580             migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
1581             break;
1582 
1583         default:
1584             break;
1585         }
1586     }
1587     if (qemu_file_get_error(rp)) {
1588         trace_source_return_path_thread_bad_end();
1589         mark_source_rp_bad(ms);
1590     }
1591 
1592     trace_source_return_path_thread_end();
1593 out:
1594     ms->rp_state.from_dst_file = NULL;
1595     qemu_fclose(rp);
1596     return NULL;
1597 }
1598 
1599 static int open_return_path_on_source(MigrationState *ms)
1600 {
1601 
1602     ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
1603     if (!ms->rp_state.from_dst_file) {
1604         return -1;
1605     }
1606 
1607     trace_open_return_path_on_source();
1608     qemu_thread_create(&ms->rp_state.rp_thread, "return path",
1609                        source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
1610 
1611     trace_open_return_path_on_source_continue();
1612 
1613     return 0;
1614 }
1615 
1616 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1617 static int await_return_path_close_on_source(MigrationState *ms)
1618 {
1619     /*
1620      * If this is a normal exit then the destination will send a SHUT and the
1621      * rp_thread will exit, however if there's an error we need to cause
1622      * it to exit.
1623      */
1624     if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
1625         /*
1626          * shutdown(2), if we have it, will cause it to unblock if it's stuck
1627          * waiting for the destination.
1628          */
1629         qemu_file_shutdown(ms->rp_state.from_dst_file);
1630         mark_source_rp_bad(ms);
1631     }
1632     trace_await_return_path_close_on_source_joining();
1633     qemu_thread_join(&ms->rp_state.rp_thread);
1634     trace_await_return_path_close_on_source_close();
1635     return ms->rp_state.error;
1636 }
1637 
1638 /*
1639  * Switch from normal iteration to postcopy
1640  * Returns non-0 on error
1641  */
1642 static int postcopy_start(MigrationState *ms, bool *old_vm_running)
1643 {
1644     int ret;
1645     QIOChannelBuffer *bioc;
1646     QEMUFile *fb;
1647     int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1648     bool restart_block = false;
1649     migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
1650                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
1651 
1652     trace_postcopy_start();
1653     qemu_mutex_lock_iothread();
1654     trace_postcopy_start_set_run();
1655 
1656     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1657     *old_vm_running = runstate_is_running();
1658     global_state_store();
1659     ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1660     if (ret < 0) {
1661         goto fail;
1662     }
1663 
1664     ret = bdrv_inactivate_all();
1665     if (ret < 0) {
1666         goto fail;
1667     }
1668     restart_block = true;
1669 
1670     /*
1671      * Cause any non-postcopiable, but iterative devices to
1672      * send out their final data.
1673      */
1674     qemu_savevm_state_complete_precopy(ms->to_dst_file, true);
1675 
1676     /*
1677      * in Finish migrate and with the io-lock held everything should
1678      * be quiet, but we've potentially still got dirty pages and we
1679      * need to tell the destination to throw any pages it's already received
1680      * that are dirty
1681      */
1682     if (ram_postcopy_send_discard_bitmap(ms)) {
1683         error_report("postcopy send discard bitmap failed");
1684         goto fail;
1685     }
1686 
1687     /*
1688      * send rest of state - note things that are doing postcopy
1689      * will notice we're in POSTCOPY_ACTIVE and not actually
1690      * wrap their state up here
1691      */
1692     qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
1693     /* Ping just for debugging, helps line traces up */
1694     qemu_savevm_send_ping(ms->to_dst_file, 2);
1695 
1696     /*
1697      * While loading the device state we may trigger page transfer
1698      * requests and the fd must be free to process those, and thus
1699      * the destination must read the whole device state off the fd before
1700      * it starts processing it.  Unfortunately the ad-hoc migration format
1701      * doesn't allow the destination to know the size to read without fully
1702      * parsing it through each devices load-state code (especially the open
1703      * coded devices that use get/put).
1704      * So we wrap the device state up in a package with a length at the start;
1705      * to do this we use a qemu_buf to hold the whole of the device state.
1706      */
1707     bioc = qio_channel_buffer_new(4096);
1708     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
1709     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
1710     object_unref(OBJECT(bioc));
1711 
1712     /*
1713      * Make sure the receiver can get incoming pages before we send the rest
1714      * of the state
1715      */
1716     qemu_savevm_send_postcopy_listen(fb);
1717 
1718     qemu_savevm_state_complete_precopy(fb, false);
1719     qemu_savevm_send_ping(fb, 3);
1720 
1721     qemu_savevm_send_postcopy_run(fb);
1722 
1723     /* <><> end of stuff going into the package */
1724 
1725     /* Last point of recovery; as soon as we send the package the destination
1726      * can open devices and potentially start running.
1727      * Lets just check again we've not got any errors.
1728      */
1729     ret = qemu_file_get_error(ms->to_dst_file);
1730     if (ret) {
1731         error_report("postcopy_start: Migration stream errored (pre package)");
1732         goto fail_closefb;
1733     }
1734 
1735     restart_block = false;
1736 
1737     /* Now send that blob */
1738     if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
1739         goto fail_closefb;
1740     }
1741     qemu_fclose(fb);
1742 
1743     /* Send a notify to give a chance for anything that needs to happen
1744      * at the transition to postcopy and after the device state; in particular
1745      * spice needs to trigger a transition now
1746      */
1747     ms->postcopy_after_devices = true;
1748     notifier_list_notify(&migration_state_notifiers, ms);
1749 
1750     ms->downtime =  qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
1751 
1752     qemu_mutex_unlock_iothread();
1753 
1754     /*
1755      * Although this ping is just for debug, it could potentially be
1756      * used for getting a better measurement of downtime at the source.
1757      */
1758     qemu_savevm_send_ping(ms->to_dst_file, 4);
1759 
1760     if (migrate_release_ram()) {
1761         ram_postcopy_migrated_memory_release(ms);
1762     }
1763 
1764     ret = qemu_file_get_error(ms->to_dst_file);
1765     if (ret) {
1766         error_report("postcopy_start: Migration stream errored");
1767         migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1768                               MIGRATION_STATUS_FAILED);
1769     }
1770 
1771     return ret;
1772 
1773 fail_closefb:
1774     qemu_fclose(fb);
1775 fail:
1776     migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1777                           MIGRATION_STATUS_FAILED);
1778     if (restart_block) {
1779         /* A failure happened early enough that we know the destination hasn't
1780          * accessed block devices, so we're safe to recover.
1781          */
1782         Error *local_err = NULL;
1783 
1784         bdrv_invalidate_cache_all(&local_err);
1785         if (local_err) {
1786             error_report_err(local_err);
1787         }
1788     }
1789     qemu_mutex_unlock_iothread();
1790     return -1;
1791 }
1792 
1793 /**
1794  * migration_completion: Used by migration_thread when there's not much left.
1795  *   The caller 'breaks' the loop when this returns.
1796  *
1797  * @s: Current migration state
1798  * @current_active_state: The migration state we expect to be in
1799  * @*old_vm_running: Pointer to old_vm_running flag
1800  * @*start_time: Pointer to time to update
1801  */
1802 static void migration_completion(MigrationState *s, int current_active_state,
1803                                  bool *old_vm_running,
1804                                  int64_t *start_time)
1805 {
1806     int ret;
1807 
1808     if (s->state == MIGRATION_STATUS_ACTIVE) {
1809         qemu_mutex_lock_iothread();
1810         *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1811         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1812         *old_vm_running = runstate_is_running();
1813         ret = global_state_store();
1814 
1815         if (!ret) {
1816             ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1817             /*
1818              * Don't mark the image with BDRV_O_INACTIVE flag if
1819              * we will go into COLO stage later.
1820              */
1821             if (ret >= 0 && !migrate_colo_enabled()) {
1822                 ret = bdrv_inactivate_all();
1823             }
1824             if (ret >= 0) {
1825                 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
1826                 qemu_savevm_state_complete_precopy(s->to_dst_file, false);
1827                 s->block_inactive = true;
1828             }
1829         }
1830         qemu_mutex_unlock_iothread();
1831 
1832         if (ret < 0) {
1833             goto fail;
1834         }
1835     } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1836         trace_migration_completion_postcopy_end();
1837 
1838         qemu_savevm_state_complete_postcopy(s->to_dst_file);
1839         trace_migration_completion_postcopy_end_after_complete();
1840     }
1841 
1842     /*
1843      * If rp was opened we must clean up the thread before
1844      * cleaning everything else up (since if there are no failures
1845      * it will wait for the destination to send it's status in
1846      * a SHUT command).
1847      * Postcopy opens rp if enabled (even if it's not avtivated)
1848      */
1849     if (migrate_postcopy_ram()) {
1850         int rp_error;
1851         trace_migration_completion_postcopy_end_before_rp();
1852         rp_error = await_return_path_close_on_source(s);
1853         trace_migration_completion_postcopy_end_after_rp(rp_error);
1854         if (rp_error) {
1855             goto fail_invalidate;
1856         }
1857     }
1858 
1859     if (qemu_file_get_error(s->to_dst_file)) {
1860         trace_migration_completion_file_err();
1861         goto fail_invalidate;
1862     }
1863 
1864     if (!migrate_colo_enabled()) {
1865         migrate_set_state(&s->state, current_active_state,
1866                           MIGRATION_STATUS_COMPLETED);
1867     }
1868 
1869     return;
1870 
1871 fail_invalidate:
1872     /* If not doing postcopy, vm_start() will be called: let's regain
1873      * control on images.
1874      */
1875     if (s->state == MIGRATION_STATUS_ACTIVE) {
1876         Error *local_err = NULL;
1877 
1878         qemu_mutex_lock_iothread();
1879         bdrv_invalidate_cache_all(&local_err);
1880         if (local_err) {
1881             error_report_err(local_err);
1882         } else {
1883             s->block_inactive = false;
1884         }
1885         qemu_mutex_unlock_iothread();
1886     }
1887 
1888 fail:
1889     migrate_set_state(&s->state, current_active_state,
1890                       MIGRATION_STATUS_FAILED);
1891 }
1892 
1893 bool migrate_colo_enabled(void)
1894 {
1895     MigrationState *s = migrate_get_current();
1896     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
1897 }
1898 
1899 /*
1900  * Master migration thread on the source VM.
1901  * It drives the migration and pumps the data down the outgoing channel.
1902  */
1903 static void *migration_thread(void *opaque)
1904 {
1905     MigrationState *s = opaque;
1906     /* Used by the bandwidth calcs, updated later */
1907     int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1908     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
1909     int64_t initial_bytes = 0;
1910     int64_t max_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     current_active_state = MIGRATION_STATUS_ACTIVE;
1942     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1943                       MIGRATION_STATUS_ACTIVE);
1944 
1945     trace_migration_thread_setup_complete();
1946 
1947     while (s->state == MIGRATION_STATUS_ACTIVE ||
1948            s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1949         int64_t current_time;
1950         uint64_t pending_size;
1951 
1952         if (!qemu_file_rate_limit(s->to_dst_file)) {
1953             uint64_t pend_post, pend_nonpost;
1954 
1955             qemu_savevm_state_pending(s->to_dst_file, max_size, &pend_nonpost,
1956                                       &pend_post);
1957             pending_size = pend_nonpost + pend_post;
1958             trace_migrate_pending(pending_size, max_size,
1959                                   pend_post, pend_nonpost);
1960             if (pending_size && pending_size >= max_size) {
1961                 /* Still a significant amount to transfer */
1962 
1963                 if (migrate_postcopy_ram() &&
1964                     s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE &&
1965                     pend_nonpost <= max_size &&
1966                     atomic_read(&s->start_postcopy)) {
1967 
1968                     if (!postcopy_start(s, &old_vm_running)) {
1969                         current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE;
1970                         entered_postcopy = true;
1971                     }
1972 
1973                     continue;
1974                 }
1975                 /* Just another iteration step */
1976                 qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy);
1977             } else {
1978                 trace_migration_thread_low_pending(pending_size);
1979                 migration_completion(s, current_active_state,
1980                                      &old_vm_running, &start_time);
1981                 break;
1982             }
1983         }
1984 
1985         if (qemu_file_get_error(s->to_dst_file)) {
1986             migrate_set_state(&s->state, current_active_state,
1987                               MIGRATION_STATUS_FAILED);
1988             trace_migration_thread_file_err();
1989             break;
1990         }
1991         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1992         if (current_time >= initial_time + BUFFER_DELAY) {
1993             uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) -
1994                                          initial_bytes;
1995             uint64_t time_spent = current_time - initial_time;
1996             double bandwidth = (double)transferred_bytes / time_spent;
1997             max_size = bandwidth * s->parameters.downtime_limit;
1998 
1999             s->mbps = (((double) transferred_bytes * 8.0) /
2000                     ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
2001 
2002             trace_migrate_transferred(transferred_bytes, time_spent,
2003                                       bandwidth, max_size);
2004             /* if we haven't sent anything, we don't want to recalculate
2005                10000 is a small enough number for our purposes */
2006             if (s->dirty_bytes_rate && transferred_bytes > 10000) {
2007                 s->expected_downtime = s->dirty_bytes_rate / 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