xref: /openbmc/qemu/migration/migration.c (revision 9c489ea6)
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 "migration/blocker.h"
20 #include "exec.h"
21 #include "fd.h"
22 #include "socket.h"
23 #include "rdma.h"
24 #include "ram.h"
25 #include "migration/global_state.h"
26 #include "migration/misc.h"
27 #include "migration.h"
28 #include "savevm.h"
29 #include "qemu-file-channel.h"
30 #include "qemu-file.h"
31 #include "migration/vmstate.h"
32 #include "block/block.h"
33 #include "qapi/qmp/qerror.h"
34 #include "qapi/util.h"
35 #include "qemu/rcu.h"
36 #include "block.h"
37 #include "postcopy-ram.h"
38 #include "qemu/thread.h"
39 #include "qmp-commands.h"
40 #include "trace.h"
41 #include "qapi-event.h"
42 #include "exec/target_page.h"
43 #include "io/channel-buffer.h"
44 #include "migration/colo.h"
45 #include "hw/boards.h"
46 #include "monitor/monitor.h"
47 
48 #define MAX_THROTTLE  (32 << 20)      /* Migration transfer speed throttling */
49 
50 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
51  * data. */
52 #define BUFFER_DELAY     100
53 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
54 
55 /* Time in milliseconds we are allowed to stop the source,
56  * for sending the last part */
57 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
58 
59 /* Maximum migrate downtime set to 2000 seconds */
60 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
61 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
62 
63 /* Default compression thread count */
64 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
65 /* Default decompression thread count, usually decompression is at
66  * least 4 times as fast as compression.*/
67 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
68 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
69 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
70 /* Define default autoconverge cpu throttle migration parameters */
71 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
72 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
73 
74 /* Migration XBZRLE default cache size */
75 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
76 
77 /* The delay time (in ms) between two COLO checkpoints
78  * Note: Please change this default value to 10000 when we support hybrid mode.
79  */
80 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
81 
82 static NotifierList migration_state_notifiers =
83     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
84 
85 static bool deferred_incoming;
86 
87 /* Messages sent on the return path from destination to source */
88 enum mig_rp_message_type {
89     MIG_RP_MSG_INVALID = 0,  /* Must be 0 */
90     MIG_RP_MSG_SHUT,         /* sibling will not send any more RP messages */
91     MIG_RP_MSG_PONG,         /* Response to a PING; data (seq: be32 ) */
92 
93     MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
94     MIG_RP_MSG_REQ_PAGES,    /* data (start: be64, len: be32) */
95 
96     MIG_RP_MSG_MAX
97 };
98 
99 /* When we add fault tolerance, we could have several
100    migrations at once.  For now we don't need to add
101    dynamic creation of migration */
102 
103 static MigrationState *current_migration;
104 
105 static bool migration_object_check(MigrationState *ms, Error **errp);
106 
107 void migration_object_init(void)
108 {
109     MachineState *ms = MACHINE(qdev_get_machine());
110     Error *err = NULL;
111 
112     /* This can only be called once. */
113     assert(!current_migration);
114     current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
115 
116     if (!migration_object_check(current_migration, &err)) {
117         error_report_err(err);
118         exit(1);
119     }
120 
121     /*
122      * We cannot really do this in migration_instance_init() since at
123      * that time global properties are not yet applied, then this
124      * value will be definitely replaced by something else.
125      */
126     if (ms->enforce_config_section) {
127         current_migration->send_configuration = true;
128     }
129 }
130 
131 /* For outgoing */
132 MigrationState *migrate_get_current(void)
133 {
134     /* This can only be called after the object created. */
135     assert(current_migration);
136     return current_migration;
137 }
138 
139 MigrationIncomingState *migration_incoming_get_current(void)
140 {
141     static bool once;
142     static MigrationIncomingState mis_current;
143 
144     if (!once) {
145         mis_current.state = MIGRATION_STATUS_NONE;
146         memset(&mis_current, 0, sizeof(MigrationIncomingState));
147         qemu_mutex_init(&mis_current.rp_mutex);
148         qemu_event_init(&mis_current.main_thread_load_event, false);
149         once = true;
150     }
151     return &mis_current;
152 }
153 
154 void migration_incoming_state_destroy(void)
155 {
156     struct MigrationIncomingState *mis = migration_incoming_get_current();
157 
158     if (mis->to_src_file) {
159         /* Tell source that we are done */
160         migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
161         qemu_fclose(mis->to_src_file);
162         mis->to_src_file = NULL;
163     }
164 
165     if (mis->from_src_file) {
166         qemu_fclose(mis->from_src_file);
167         mis->from_src_file = NULL;
168     }
169 
170     qemu_event_destroy(&mis->main_thread_load_event);
171 }
172 
173 static void migrate_generate_event(int new_state)
174 {
175     if (migrate_use_events()) {
176         qapi_event_send_migration(new_state, &error_abort);
177     }
178 }
179 
180 /*
181  * Called on -incoming with a defer: uri.
182  * The migration can be started later after any parameters have been
183  * changed.
184  */
185 static void deferred_incoming_migration(Error **errp)
186 {
187     if (deferred_incoming) {
188         error_setg(errp, "Incoming migration already deferred");
189     }
190     deferred_incoming = true;
191 }
192 
193 /*
194  * Send a message on the return channel back to the source
195  * of the migration.
196  */
197 static void migrate_send_rp_message(MigrationIncomingState *mis,
198                                     enum mig_rp_message_type message_type,
199                                     uint16_t len, void *data)
200 {
201     trace_migrate_send_rp_message((int)message_type, len);
202     qemu_mutex_lock(&mis->rp_mutex);
203     qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
204     qemu_put_be16(mis->to_src_file, len);
205     qemu_put_buffer(mis->to_src_file, data, len);
206     qemu_fflush(mis->to_src_file);
207     qemu_mutex_unlock(&mis->rp_mutex);
208 }
209 
210 /* Request a range of pages from the source VM at the given
211  * start address.
212  *   rbname: Name of the RAMBlock to request the page in, if NULL it's the same
213  *           as the last request (a name must have been given previously)
214  *   Start: Address offset within the RB
215  *   Len: Length in bytes required - must be a multiple of pagesize
216  */
217 void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
218                                ram_addr_t start, size_t len)
219 {
220     uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
221     size_t msglen = 12; /* start + len */
222 
223     *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
224     *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
225 
226     if (rbname) {
227         int rbname_len = strlen(rbname);
228         assert(rbname_len < 256);
229 
230         bufc[msglen++] = rbname_len;
231         memcpy(bufc + msglen, rbname, rbname_len);
232         msglen += rbname_len;
233         migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc);
234     } else {
235         migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc);
236     }
237 }
238 
239 void qemu_start_incoming_migration(const char *uri, Error **errp)
240 {
241     const char *p;
242 
243     qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
244     if (!strcmp(uri, "defer")) {
245         deferred_incoming_migration(errp);
246     } else if (strstart(uri, "tcp:", &p)) {
247         tcp_start_incoming_migration(p, errp);
248 #ifdef CONFIG_RDMA
249     } else if (strstart(uri, "rdma:", &p)) {
250         rdma_start_incoming_migration(p, errp);
251 #endif
252     } else if (strstart(uri, "exec:", &p)) {
253         exec_start_incoming_migration(p, errp);
254     } else if (strstart(uri, "unix:", &p)) {
255         unix_start_incoming_migration(p, errp);
256     } else if (strstart(uri, "fd:", &p)) {
257         fd_start_incoming_migration(p, errp);
258     } else {
259         error_setg(errp, "unknown migration protocol: %s", uri);
260     }
261 }
262 
263 static void process_incoming_migration_bh(void *opaque)
264 {
265     Error *local_err = NULL;
266     MigrationIncomingState *mis = opaque;
267 
268     /* Make sure all file formats flush their mutable metadata.
269      * If we get an error here, just don't restart the VM yet. */
270     bdrv_invalidate_cache_all(&local_err);
271     if (local_err) {
272         error_report_err(local_err);
273         local_err = NULL;
274         autostart = false;
275     }
276 
277     /*
278      * This must happen after all error conditions are dealt with and
279      * we're sure the VM is going to be running on this host.
280      */
281     qemu_announce_self();
282 
283     /* If global state section was not received or we are in running
284        state, we need to obey autostart. Any other state is set with
285        runstate_set. */
286 
287     if (!global_state_received() ||
288         global_state_get_runstate() == RUN_STATE_RUNNING) {
289         if (autostart) {
290             vm_start();
291         } else {
292             runstate_set(RUN_STATE_PAUSED);
293         }
294     } else {
295         runstate_set(global_state_get_runstate());
296     }
297     /*
298      * This must happen after any state changes since as soon as an external
299      * observer sees this event they might start to prod at the VM assuming
300      * it's ready to use.
301      */
302     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
303                       MIGRATION_STATUS_COMPLETED);
304     qemu_bh_delete(mis->bh);
305     migration_incoming_state_destroy();
306 }
307 
308 static void process_incoming_migration_co(void *opaque)
309 {
310     QEMUFile *f = opaque;
311     MigrationIncomingState *mis = migration_incoming_get_current();
312     PostcopyState ps;
313     int ret;
314 
315     mis->from_src_file = f;
316     mis->largest_page_size = qemu_ram_pagesize_largest();
317     postcopy_state_set(POSTCOPY_INCOMING_NONE);
318     migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
319                       MIGRATION_STATUS_ACTIVE);
320     ret = qemu_loadvm_state(f);
321 
322     ps = postcopy_state_get();
323     trace_process_incoming_migration_co_end(ret, ps);
324     if (ps != POSTCOPY_INCOMING_NONE) {
325         if (ps == POSTCOPY_INCOMING_ADVISE) {
326             /*
327              * Where a migration had postcopy enabled (and thus went to advise)
328              * but managed to complete within the precopy period, we can use
329              * the normal exit.
330              */
331             postcopy_ram_incoming_cleanup(mis);
332         } else if (ret >= 0) {
333             /*
334              * Postcopy was started, cleanup should happen at the end of the
335              * postcopy thread.
336              */
337             trace_process_incoming_migration_co_postcopy_end_main();
338             return;
339         }
340         /* Else if something went wrong then just fall out of the normal exit */
341     }
342 
343     /* we get COLO info, and know if we are in COLO mode */
344     if (!ret && migration_incoming_enable_colo()) {
345         mis->migration_incoming_co = qemu_coroutine_self();
346         qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
347              colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
348         mis->have_colo_incoming_thread = true;
349         qemu_coroutine_yield();
350 
351         /* Wait checkpoint incoming thread exit before free resource */
352         qemu_thread_join(&mis->colo_incoming_thread);
353     }
354 
355     if (ret < 0) {
356         migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
357                           MIGRATION_STATUS_FAILED);
358         error_report("load of migration failed: %s", strerror(-ret));
359         qemu_fclose(mis->from_src_file);
360         exit(EXIT_FAILURE);
361     }
362     mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
363     qemu_bh_schedule(mis->bh);
364 }
365 
366 void migration_fd_process_incoming(QEMUFile *f)
367 {
368     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, f);
369 
370     qemu_file_set_blocking(f, false);
371     qemu_coroutine_enter(co);
372 }
373 
374 /*
375  * Send a 'SHUT' message on the return channel with the given value
376  * to indicate that we've finished with the RP.  Non-0 value indicates
377  * error.
378  */
379 void migrate_send_rp_shut(MigrationIncomingState *mis,
380                           uint32_t value)
381 {
382     uint32_t buf;
383 
384     buf = cpu_to_be32(value);
385     migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
386 }
387 
388 /*
389  * Send a 'PONG' message on the return channel with the given value
390  * (normally in response to a 'PING')
391  */
392 void migrate_send_rp_pong(MigrationIncomingState *mis,
393                           uint32_t value)
394 {
395     uint32_t buf;
396 
397     buf = cpu_to_be32(value);
398     migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
399 }
400 
401 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
402 {
403     MigrationCapabilityStatusList *head = NULL;
404     MigrationCapabilityStatusList *caps;
405     MigrationState *s = migrate_get_current();
406     int i;
407 
408     caps = NULL; /* silence compiler warning */
409     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
410 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
411         if (i == MIGRATION_CAPABILITY_BLOCK) {
412             continue;
413         }
414 #endif
415         if (head == NULL) {
416             head = g_malloc0(sizeof(*caps));
417             caps = head;
418         } else {
419             caps->next = g_malloc0(sizeof(*caps));
420             caps = caps->next;
421         }
422         caps->value =
423             g_malloc(sizeof(*caps->value));
424         caps->value->capability = i;
425         caps->value->state = s->enabled_capabilities[i];
426     }
427 
428     return head;
429 }
430 
431 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
432 {
433     MigrationParameters *params;
434     MigrationState *s = migrate_get_current();
435 
436     params = g_malloc0(sizeof(*params));
437     params->has_compress_level = true;
438     params->compress_level = s->parameters.compress_level;
439     params->has_compress_threads = true;
440     params->compress_threads = s->parameters.compress_threads;
441     params->has_decompress_threads = true;
442     params->decompress_threads = s->parameters.decompress_threads;
443     params->has_cpu_throttle_initial = true;
444     params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
445     params->has_cpu_throttle_increment = true;
446     params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
447     params->has_tls_creds = !!s->parameters.tls_creds;
448     params->tls_creds = g_strdup(s->parameters.tls_creds);
449     params->has_tls_hostname = !!s->parameters.tls_hostname;
450     params->tls_hostname = g_strdup(s->parameters.tls_hostname);
451     params->has_max_bandwidth = true;
452     params->max_bandwidth = s->parameters.max_bandwidth;
453     params->has_downtime_limit = true;
454     params->downtime_limit = s->parameters.downtime_limit;
455     params->has_x_checkpoint_delay = true;
456     params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
457     params->has_block_incremental = true;
458     params->block_incremental = s->parameters.block_incremental;
459 
460     return params;
461 }
462 
463 /*
464  * Return true if we're already in the middle of a migration
465  * (i.e. any of the active or setup states)
466  */
467 static bool migration_is_setup_or_active(int state)
468 {
469     switch (state) {
470     case MIGRATION_STATUS_ACTIVE:
471     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
472     case MIGRATION_STATUS_SETUP:
473         return true;
474 
475     default:
476         return false;
477 
478     }
479 }
480 
481 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
482 {
483     info->has_ram = true;
484     info->ram = g_malloc0(sizeof(*info->ram));
485     info->ram->transferred = ram_counters.transferred;
486     info->ram->total = ram_bytes_total();
487     info->ram->duplicate = ram_counters.duplicate;
488     /* legacy value.  It is not used anymore */
489     info->ram->skipped = 0;
490     info->ram->normal = ram_counters.normal;
491     info->ram->normal_bytes = ram_counters.normal *
492         qemu_target_page_size();
493     info->ram->mbps = s->mbps;
494     info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
495     info->ram->postcopy_requests = ram_counters.postcopy_requests;
496     info->ram->page_size = qemu_target_page_size();
497 
498     if (migrate_use_xbzrle()) {
499         info->has_xbzrle_cache = true;
500         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
501         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
502         info->xbzrle_cache->bytes = xbzrle_counters.bytes;
503         info->xbzrle_cache->pages = xbzrle_counters.pages;
504         info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
505         info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
506         info->xbzrle_cache->overflow = xbzrle_counters.overflow;
507     }
508 
509     if (cpu_throttle_active()) {
510         info->has_cpu_throttle_percentage = true;
511         info->cpu_throttle_percentage = cpu_throttle_get_percentage();
512     }
513 
514     if (s->state != MIGRATION_STATUS_COMPLETED) {
515         info->ram->remaining = ram_bytes_remaining();
516         info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
517     }
518 }
519 
520 static void populate_disk_info(MigrationInfo *info)
521 {
522     if (blk_mig_active()) {
523         info->has_disk = true;
524         info->disk = g_malloc0(sizeof(*info->disk));
525         info->disk->transferred = blk_mig_bytes_transferred();
526         info->disk->remaining = blk_mig_bytes_remaining();
527         info->disk->total = blk_mig_bytes_total();
528     }
529 }
530 
531 MigrationInfo *qmp_query_migrate(Error **errp)
532 {
533     MigrationInfo *info = g_malloc0(sizeof(*info));
534     MigrationState *s = migrate_get_current();
535 
536     switch (s->state) {
537     case MIGRATION_STATUS_NONE:
538         /* no migration has happened ever */
539         break;
540     case MIGRATION_STATUS_SETUP:
541         info->has_status = true;
542         info->has_total_time = false;
543         break;
544     case MIGRATION_STATUS_ACTIVE:
545     case MIGRATION_STATUS_CANCELLING:
546     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
547          /* TODO add some postcopy stats */
548         info->has_status = true;
549         info->has_total_time = true;
550         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
551             - s->total_time;
552         info->has_expected_downtime = true;
553         info->expected_downtime = s->expected_downtime;
554         info->has_setup_time = true;
555         info->setup_time = s->setup_time;
556 
557         populate_ram_info(info, s);
558         populate_disk_info(info);
559         break;
560     case MIGRATION_STATUS_COLO:
561         info->has_status = true;
562         /* TODO: display COLO specific information (checkpoint info etc.) */
563         break;
564     case MIGRATION_STATUS_COMPLETED:
565         info->has_status = true;
566         info->has_total_time = true;
567         info->total_time = s->total_time;
568         info->has_downtime = true;
569         info->downtime = s->downtime;
570         info->has_setup_time = true;
571         info->setup_time = s->setup_time;
572 
573         populate_ram_info(info, s);
574         break;
575     case MIGRATION_STATUS_FAILED:
576         info->has_status = true;
577         if (s->error) {
578             info->has_error_desc = true;
579             info->error_desc = g_strdup(error_get_pretty(s->error));
580         }
581         break;
582     case MIGRATION_STATUS_CANCELLED:
583         info->has_status = true;
584         break;
585     }
586     info->status = s->state;
587 
588     return info;
589 }
590 
591 /**
592  * @migration_caps_check - check capability validity
593  *
594  * @cap_list: old capability list, array of bool
595  * @params: new capabilities to be applied soon
596  * @errp: set *errp if the check failed, with reason
597  *
598  * Returns true if check passed, otherwise false.
599  */
600 static bool migrate_caps_check(bool *cap_list,
601                                MigrationCapabilityStatusList *params,
602                                Error **errp)
603 {
604     MigrationCapabilityStatusList *cap;
605     bool old_postcopy_cap;
606 
607     old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
608 
609     for (cap = params; cap; cap = cap->next) {
610         cap_list[cap->value->capability] = cap->value->state;
611     }
612 
613 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
614     if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
615         error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
616                    "block migration");
617         error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
618         return false;
619     }
620 #endif
621 
622     if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
623         if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
624             /* The decompression threads asynchronously write into RAM
625              * rather than use the atomic copies needed to avoid
626              * userfaulting.  It should be possible to fix the decompression
627              * threads for compatibility in future.
628              */
629             error_setg(errp, "Postcopy is not currently compatible "
630                        "with compression");
631             return false;
632         }
633 
634         /* This check is reasonably expensive, so only when it's being
635          * set the first time, also it's only the destination that needs
636          * special support.
637          */
638         if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
639             !postcopy_ram_supported_by_host()) {
640             /* postcopy_ram_supported_by_host will have emitted a more
641              * detailed message
642              */
643             error_setg(errp, "Postcopy is not supported");
644             return false;
645         }
646     }
647 
648     return true;
649 }
650 
651 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
652                                   Error **errp)
653 {
654     MigrationState *s = migrate_get_current();
655     MigrationCapabilityStatusList *cap;
656 
657     if (migration_is_setup_or_active(s->state)) {
658         error_setg(errp, QERR_MIGRATION_ACTIVE);
659         return;
660     }
661 
662     if (!migrate_caps_check(s->enabled_capabilities, params, errp)) {
663         return;
664     }
665 
666     for (cap = params; cap; cap = cap->next) {
667         s->enabled_capabilities[cap->value->capability] = cap->value->state;
668     }
669 }
670 
671 /*
672  * Check whether the parameters are valid. Error will be put into errp
673  * (if provided). Return true if valid, otherwise false.
674  */
675 static bool migrate_params_check(MigrationParameters *params, Error **errp)
676 {
677     if (params->has_compress_level &&
678         (params->compress_level < 0 || params->compress_level > 9)) {
679         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
680                    "is invalid, it should be in the range of 0 to 9");
681         return false;
682     }
683 
684     if (params->has_compress_threads &&
685         (params->compress_threads < 1 || params->compress_threads > 255)) {
686         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
687                    "compress_threads",
688                    "is invalid, it should be in the range of 1 to 255");
689         return false;
690     }
691 
692     if (params->has_decompress_threads &&
693         (params->decompress_threads < 1 || params->decompress_threads > 255)) {
694         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
695                    "decompress_threads",
696                    "is invalid, it should be in the range of 1 to 255");
697         return false;
698     }
699 
700     if (params->has_cpu_throttle_initial &&
701         (params->cpu_throttle_initial < 1 ||
702          params->cpu_throttle_initial > 99)) {
703         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
704                    "cpu_throttle_initial",
705                    "an integer in the range of 1 to 99");
706         return false;
707     }
708 
709     if (params->has_cpu_throttle_increment &&
710         (params->cpu_throttle_increment < 1 ||
711          params->cpu_throttle_increment > 99)) {
712         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
713                    "cpu_throttle_increment",
714                    "an integer in the range of 1 to 99");
715         return false;
716     }
717 
718     if (params->has_max_bandwidth &&
719         (params->max_bandwidth < 0 || params->max_bandwidth > SIZE_MAX)) {
720         error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
721                          " range of 0 to %zu bytes/second", SIZE_MAX);
722         return false;
723     }
724 
725     if (params->has_downtime_limit &&
726         (params->downtime_limit < 0 ||
727          params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
728         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
729                          "the range of 0 to %d milliseconds",
730                          MAX_MIGRATE_DOWNTIME);
731         return false;
732     }
733 
734     if (params->has_x_checkpoint_delay && (params->x_checkpoint_delay < 0)) {
735         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
736                     "x_checkpoint_delay",
737                     "is invalid, it should be positive");
738         return false;
739     }
740 
741     return true;
742 }
743 
744 static void migrate_params_apply(MigrationParameters *params)
745 {
746     MigrationState *s = migrate_get_current();
747 
748     if (params->has_compress_level) {
749         s->parameters.compress_level = params->compress_level;
750     }
751 
752     if (params->has_compress_threads) {
753         s->parameters.compress_threads = params->compress_threads;
754     }
755 
756     if (params->has_decompress_threads) {
757         s->parameters.decompress_threads = params->decompress_threads;
758     }
759 
760     if (params->has_cpu_throttle_initial) {
761         s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
762     }
763 
764     if (params->has_cpu_throttle_increment) {
765         s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
766     }
767 
768     if (params->has_tls_creds) {
769         g_free(s->parameters.tls_creds);
770         s->parameters.tls_creds = g_strdup(params->tls_creds);
771     }
772 
773     if (params->has_tls_hostname) {
774         g_free(s->parameters.tls_hostname);
775         s->parameters.tls_hostname = g_strdup(params->tls_hostname);
776     }
777 
778     if (params->has_max_bandwidth) {
779         s->parameters.max_bandwidth = params->max_bandwidth;
780         if (s->to_dst_file) {
781             qemu_file_set_rate_limit(s->to_dst_file,
782                                 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
783         }
784     }
785 
786     if (params->has_downtime_limit) {
787         s->parameters.downtime_limit = params->downtime_limit;
788     }
789 
790     if (params->has_x_checkpoint_delay) {
791         s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
792         if (migration_in_colo_state()) {
793             colo_checkpoint_notify(s);
794         }
795     }
796 
797     if (params->has_block_incremental) {
798         s->parameters.block_incremental = params->block_incremental;
799     }
800 }
801 
802 void qmp_migrate_set_parameters(MigrationParameters *params, Error **errp)
803 {
804     if (!migrate_params_check(params, errp)) {
805         /* Invalid parameter */
806         return;
807     }
808 
809     migrate_params_apply(params);
810 }
811 
812 
813 void qmp_migrate_start_postcopy(Error **errp)
814 {
815     MigrationState *s = migrate_get_current();
816 
817     if (!migrate_postcopy_ram()) {
818         error_setg(errp, "Enable postcopy with migrate_set_capability before"
819                          " the start of migration");
820         return;
821     }
822 
823     if (s->state == MIGRATION_STATUS_NONE) {
824         error_setg(errp, "Postcopy must be started after migration has been"
825                          " started");
826         return;
827     }
828     /*
829      * we don't error if migration has finished since that would be racy
830      * with issuing this command.
831      */
832     atomic_set(&s->start_postcopy, true);
833 }
834 
835 /* shared migration helpers */
836 
837 void migrate_set_state(int *state, int old_state, int new_state)
838 {
839     if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
840         trace_migrate_set_state(new_state);
841         migrate_generate_event(new_state);
842     }
843 }
844 
845 static MigrationCapabilityStatusList *migrate_cap_add(
846     MigrationCapabilityStatusList *list,
847     MigrationCapability index,
848     bool state)
849 {
850     MigrationCapabilityStatusList *cap;
851 
852     cap = g_new0(MigrationCapabilityStatusList, 1);
853     cap->value = g_new0(MigrationCapabilityStatus, 1);
854     cap->value->capability = index;
855     cap->value->state = state;
856     cap->next = list;
857 
858     return cap;
859 }
860 
861 void migrate_set_block_enabled(bool value, Error **errp)
862 {
863     MigrationCapabilityStatusList *cap;
864 
865     cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
866     qmp_migrate_set_capabilities(cap, errp);
867     qapi_free_MigrationCapabilityStatusList(cap);
868 }
869 
870 static void migrate_set_block_incremental(MigrationState *s, bool value)
871 {
872     s->parameters.block_incremental = value;
873 }
874 
875 static void block_cleanup_parameters(MigrationState *s)
876 {
877     if (s->must_remove_block_options) {
878         /* setting to false can never fail */
879         migrate_set_block_enabled(false, &error_abort);
880         migrate_set_block_incremental(s, false);
881         s->must_remove_block_options = false;
882     }
883 }
884 
885 static void migrate_fd_cleanup(void *opaque)
886 {
887     MigrationState *s = opaque;
888 
889     qemu_bh_delete(s->cleanup_bh);
890     s->cleanup_bh = NULL;
891 
892     if (s->to_dst_file) {
893         trace_migrate_fd_cleanup();
894         qemu_mutex_unlock_iothread();
895         if (s->migration_thread_running) {
896             qemu_thread_join(&s->thread);
897             s->migration_thread_running = false;
898         }
899         qemu_mutex_lock_iothread();
900 
901         qemu_fclose(s->to_dst_file);
902         s->to_dst_file = NULL;
903     }
904 
905     assert((s->state != MIGRATION_STATUS_ACTIVE) &&
906            (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
907 
908     if (s->state == MIGRATION_STATUS_CANCELLING) {
909         migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
910                           MIGRATION_STATUS_CANCELLED);
911     }
912 
913     notifier_list_notify(&migration_state_notifiers, s);
914     block_cleanup_parameters(s);
915 }
916 
917 void migrate_fd_error(MigrationState *s, const Error *error)
918 {
919     trace_migrate_fd_error(error_get_pretty(error));
920     assert(s->to_dst_file == NULL);
921     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
922                       MIGRATION_STATUS_FAILED);
923     if (!s->error) {
924         s->error = error_copy(error);
925     }
926     notifier_list_notify(&migration_state_notifiers, s);
927     block_cleanup_parameters(s);
928 }
929 
930 static void migrate_fd_cancel(MigrationState *s)
931 {
932     int old_state ;
933     QEMUFile *f = migrate_get_current()->to_dst_file;
934     trace_migrate_fd_cancel();
935 
936     if (s->rp_state.from_dst_file) {
937         /* shutdown the rp socket, so causing the rp thread to shutdown */
938         qemu_file_shutdown(s->rp_state.from_dst_file);
939     }
940 
941     do {
942         old_state = s->state;
943         if (!migration_is_setup_or_active(old_state)) {
944             break;
945         }
946         migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
947     } while (s->state != MIGRATION_STATUS_CANCELLING);
948 
949     /*
950      * If we're unlucky the migration code might be stuck somewhere in a
951      * send/write while the network has failed and is waiting to timeout;
952      * if we've got shutdown(2) available then we can force it to quit.
953      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
954      * called in a bh, so there is no race against this cancel.
955      */
956     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
957         qemu_file_shutdown(f);
958     }
959     if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
960         Error *local_err = NULL;
961 
962         bdrv_invalidate_cache_all(&local_err);
963         if (local_err) {
964             error_report_err(local_err);
965         } else {
966             s->block_inactive = false;
967         }
968     }
969     block_cleanup_parameters(s);
970 }
971 
972 void add_migration_state_change_notifier(Notifier *notify)
973 {
974     notifier_list_add(&migration_state_notifiers, notify);
975 }
976 
977 void remove_migration_state_change_notifier(Notifier *notify)
978 {
979     notifier_remove(notify);
980 }
981 
982 bool migration_in_setup(MigrationState *s)
983 {
984     return s->state == MIGRATION_STATUS_SETUP;
985 }
986 
987 bool migration_has_finished(MigrationState *s)
988 {
989     return s->state == MIGRATION_STATUS_COMPLETED;
990 }
991 
992 bool migration_has_failed(MigrationState *s)
993 {
994     return (s->state == MIGRATION_STATUS_CANCELLED ||
995             s->state == MIGRATION_STATUS_FAILED);
996 }
997 
998 bool migration_in_postcopy(void)
999 {
1000     MigrationState *s = migrate_get_current();
1001 
1002     return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1003 }
1004 
1005 bool migration_in_postcopy_after_devices(MigrationState *s)
1006 {
1007     return migration_in_postcopy() && s->postcopy_after_devices;
1008 }
1009 
1010 bool migration_is_idle(void)
1011 {
1012     MigrationState *s = migrate_get_current();
1013 
1014     switch (s->state) {
1015     case MIGRATION_STATUS_NONE:
1016     case MIGRATION_STATUS_CANCELLED:
1017     case MIGRATION_STATUS_COMPLETED:
1018     case MIGRATION_STATUS_FAILED:
1019         return true;
1020     case MIGRATION_STATUS_SETUP:
1021     case MIGRATION_STATUS_CANCELLING:
1022     case MIGRATION_STATUS_ACTIVE:
1023     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1024     case MIGRATION_STATUS_COLO:
1025         return false;
1026     case MIGRATION_STATUS__MAX:
1027         g_assert_not_reached();
1028     }
1029 
1030     return false;
1031 }
1032 
1033 MigrationState *migrate_init(void)
1034 {
1035     MigrationState *s = migrate_get_current();
1036 
1037     /*
1038      * Reinitialise all migration state, except
1039      * parameters/capabilities that the user set, and
1040      * locks.
1041      */
1042     s->bytes_xfer = 0;
1043     s->xfer_limit = 0;
1044     s->cleanup_bh = 0;
1045     s->to_dst_file = NULL;
1046     s->state = MIGRATION_STATUS_NONE;
1047     s->rp_state.from_dst_file = NULL;
1048     s->rp_state.error = false;
1049     s->mbps = 0.0;
1050     s->downtime = 0;
1051     s->expected_downtime = 0;
1052     s->setup_time = 0;
1053     s->start_postcopy = false;
1054     s->postcopy_after_devices = false;
1055     s->migration_thread_running = false;
1056     error_free(s->error);
1057     s->error = NULL;
1058 
1059     migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1060 
1061     s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1062     return s;
1063 }
1064 
1065 static GSList *migration_blockers;
1066 
1067 int migrate_add_blocker(Error *reason, Error **errp)
1068 {
1069     if (migrate_get_current()->only_migratable) {
1070         error_propagate(errp, error_copy(reason));
1071         error_prepend(errp, "disallowing migration blocker "
1072                           "(--only_migratable) for: ");
1073         return -EACCES;
1074     }
1075 
1076     if (migration_is_idle()) {
1077         migration_blockers = g_slist_prepend(migration_blockers, reason);
1078         return 0;
1079     }
1080 
1081     error_propagate(errp, error_copy(reason));
1082     error_prepend(errp, "disallowing migration blocker (migration in "
1083                       "progress) for: ");
1084     return -EBUSY;
1085 }
1086 
1087 void migrate_del_blocker(Error *reason)
1088 {
1089     migration_blockers = g_slist_remove(migration_blockers, reason);
1090 }
1091 
1092 void qmp_migrate_incoming(const char *uri, Error **errp)
1093 {
1094     Error *local_err = NULL;
1095     static bool once = true;
1096 
1097     if (!deferred_incoming) {
1098         error_setg(errp, "For use with '-incoming defer'");
1099         return;
1100     }
1101     if (!once) {
1102         error_setg(errp, "The incoming migration has already been started");
1103     }
1104 
1105     qemu_start_incoming_migration(uri, &local_err);
1106 
1107     if (local_err) {
1108         error_propagate(errp, local_err);
1109         return;
1110     }
1111 
1112     once = false;
1113 }
1114 
1115 bool migration_is_blocked(Error **errp)
1116 {
1117     if (qemu_savevm_state_blocked(errp)) {
1118         return true;
1119     }
1120 
1121     if (migration_blockers) {
1122         error_propagate(errp, error_copy(migration_blockers->data));
1123         return true;
1124     }
1125 
1126     return false;
1127 }
1128 
1129 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1130                  bool has_inc, bool inc, bool has_detach, bool detach,
1131                  Error **errp)
1132 {
1133     Error *local_err = NULL;
1134     MigrationState *s = migrate_get_current();
1135     const char *p;
1136 
1137     if (migration_is_setup_or_active(s->state) ||
1138         s->state == MIGRATION_STATUS_CANCELLING ||
1139         s->state == MIGRATION_STATUS_COLO) {
1140         error_setg(errp, QERR_MIGRATION_ACTIVE);
1141         return;
1142     }
1143     if (runstate_check(RUN_STATE_INMIGRATE)) {
1144         error_setg(errp, "Guest is waiting for an incoming migration");
1145         return;
1146     }
1147 
1148     if (migration_is_blocked(errp)) {
1149         return;
1150     }
1151 
1152     if ((has_blk && blk) || (has_inc && inc)) {
1153         if (migrate_use_block() || migrate_use_block_incremental()) {
1154             error_setg(errp, "Command options are incompatible with "
1155                        "current migration capabilities");
1156             return;
1157         }
1158         migrate_set_block_enabled(true, &local_err);
1159         if (local_err) {
1160             error_propagate(errp, local_err);
1161             return;
1162         }
1163         s->must_remove_block_options = true;
1164     }
1165 
1166     if (has_inc && inc) {
1167         migrate_set_block_incremental(s, true);
1168     }
1169 
1170     s = migrate_init();
1171 
1172     if (strstart(uri, "tcp:", &p)) {
1173         tcp_start_outgoing_migration(s, p, &local_err);
1174 #ifdef CONFIG_RDMA
1175     } else if (strstart(uri, "rdma:", &p)) {
1176         rdma_start_outgoing_migration(s, p, &local_err);
1177 #endif
1178     } else if (strstart(uri, "exec:", &p)) {
1179         exec_start_outgoing_migration(s, p, &local_err);
1180     } else if (strstart(uri, "unix:", &p)) {
1181         unix_start_outgoing_migration(s, p, &local_err);
1182     } else if (strstart(uri, "fd:", &p)) {
1183         fd_start_outgoing_migration(s, p, &local_err);
1184     } else {
1185         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1186                    "a valid migration protocol");
1187         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1188                           MIGRATION_STATUS_FAILED);
1189         return;
1190     }
1191 
1192     if (local_err) {
1193         migrate_fd_error(s, local_err);
1194         error_propagate(errp, local_err);
1195         return;
1196     }
1197 }
1198 
1199 void qmp_migrate_cancel(Error **errp)
1200 {
1201     migrate_fd_cancel(migrate_get_current());
1202 }
1203 
1204 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1205 {
1206     MigrationState *s = migrate_get_current();
1207     int64_t new_size;
1208 
1209     /* Check for truncation */
1210     if (value != (size_t)value) {
1211         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1212                    "exceeding address space");
1213         return;
1214     }
1215 
1216     /* Cache should not be larger than guest ram size */
1217     if (value > ram_bytes_total()) {
1218         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1219                    "exceeds guest ram size ");
1220         return;
1221     }
1222 
1223     new_size = xbzrle_cache_resize(value);
1224     if (new_size < 0) {
1225         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1226                    "is smaller than page size");
1227         return;
1228     }
1229 
1230     s->xbzrle_cache_size = new_size;
1231 }
1232 
1233 int64_t qmp_query_migrate_cache_size(Error **errp)
1234 {
1235     return migrate_xbzrle_cache_size();
1236 }
1237 
1238 void qmp_migrate_set_speed(int64_t value, Error **errp)
1239 {
1240     MigrationParameters p = {
1241         .has_max_bandwidth = true,
1242         .max_bandwidth = value,
1243     };
1244 
1245     qmp_migrate_set_parameters(&p, errp);
1246 }
1247 
1248 void qmp_migrate_set_downtime(double value, Error **errp)
1249 {
1250     if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1251         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1252                          "the range of 0 to %d seconds",
1253                          MAX_MIGRATE_DOWNTIME_SECONDS);
1254         return;
1255     }
1256 
1257     value *= 1000; /* Convert to milliseconds */
1258     value = MAX(0, MIN(INT64_MAX, value));
1259 
1260     MigrationParameters p = {
1261         .has_downtime_limit = true,
1262         .downtime_limit = value,
1263     };
1264 
1265     qmp_migrate_set_parameters(&p, errp);
1266 }
1267 
1268 bool migrate_release_ram(void)
1269 {
1270     MigrationState *s;
1271 
1272     s = migrate_get_current();
1273 
1274     return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1275 }
1276 
1277 bool migrate_postcopy_ram(void)
1278 {
1279     MigrationState *s;
1280 
1281     s = migrate_get_current();
1282 
1283     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1284 }
1285 
1286 bool migrate_auto_converge(void)
1287 {
1288     MigrationState *s;
1289 
1290     s = migrate_get_current();
1291 
1292     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1293 }
1294 
1295 bool migrate_zero_blocks(void)
1296 {
1297     MigrationState *s;
1298 
1299     s = migrate_get_current();
1300 
1301     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1302 }
1303 
1304 bool migrate_use_compression(void)
1305 {
1306     MigrationState *s;
1307 
1308     s = migrate_get_current();
1309 
1310     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
1311 }
1312 
1313 int migrate_compress_level(void)
1314 {
1315     MigrationState *s;
1316 
1317     s = migrate_get_current();
1318 
1319     return s->parameters.compress_level;
1320 }
1321 
1322 int migrate_compress_threads(void)
1323 {
1324     MigrationState *s;
1325 
1326     s = migrate_get_current();
1327 
1328     return s->parameters.compress_threads;
1329 }
1330 
1331 int migrate_decompress_threads(void)
1332 {
1333     MigrationState *s;
1334 
1335     s = migrate_get_current();
1336 
1337     return s->parameters.decompress_threads;
1338 }
1339 
1340 bool migrate_use_events(void)
1341 {
1342     MigrationState *s;
1343 
1344     s = migrate_get_current();
1345 
1346     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1347 }
1348 
1349 int migrate_use_xbzrle(void)
1350 {
1351     MigrationState *s;
1352 
1353     s = migrate_get_current();
1354 
1355     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1356 }
1357 
1358 int64_t migrate_xbzrle_cache_size(void)
1359 {
1360     MigrationState *s;
1361 
1362     s = migrate_get_current();
1363 
1364     return s->xbzrle_cache_size;
1365 }
1366 
1367 bool migrate_use_block(void)
1368 {
1369     MigrationState *s;
1370 
1371     s = migrate_get_current();
1372 
1373     return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
1374 }
1375 
1376 bool migrate_use_return_path(void)
1377 {
1378     MigrationState *s;
1379 
1380     s = migrate_get_current();
1381 
1382     return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
1383 }
1384 
1385 bool migrate_use_block_incremental(void)
1386 {
1387     MigrationState *s;
1388 
1389     s = migrate_get_current();
1390 
1391     return s->parameters.block_incremental;
1392 }
1393 
1394 /* migration thread support */
1395 /*
1396  * Something bad happened to the RP stream, mark an error
1397  * The caller shall print or trace something to indicate why
1398  */
1399 static void mark_source_rp_bad(MigrationState *s)
1400 {
1401     s->rp_state.error = true;
1402 }
1403 
1404 static struct rp_cmd_args {
1405     ssize_t     len; /* -1 = variable */
1406     const char *name;
1407 } rp_cmd_args[] = {
1408     [MIG_RP_MSG_INVALID]        = { .len = -1, .name = "INVALID" },
1409     [MIG_RP_MSG_SHUT]           = { .len =  4, .name = "SHUT" },
1410     [MIG_RP_MSG_PONG]           = { .len =  4, .name = "PONG" },
1411     [MIG_RP_MSG_REQ_PAGES]      = { .len = 12, .name = "REQ_PAGES" },
1412     [MIG_RP_MSG_REQ_PAGES_ID]   = { .len = -1, .name = "REQ_PAGES_ID" },
1413     [MIG_RP_MSG_MAX]            = { .len = -1, .name = "MAX" },
1414 };
1415 
1416 /*
1417  * Process a request for pages received on the return path,
1418  * We're allowed to send more than requested (e.g. to round to our page size)
1419  * and we don't need to send pages that have already been sent.
1420  */
1421 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1422                                        ram_addr_t start, size_t len)
1423 {
1424     long our_host_ps = getpagesize();
1425 
1426     trace_migrate_handle_rp_req_pages(rbname, start, len);
1427 
1428     /*
1429      * Since we currently insist on matching page sizes, just sanity check
1430      * we're being asked for whole host pages.
1431      */
1432     if (start & (our_host_ps-1) ||
1433        (len & (our_host_ps-1))) {
1434         error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1435                      " len: %zd", __func__, start, len);
1436         mark_source_rp_bad(ms);
1437         return;
1438     }
1439 
1440     if (ram_save_queue_pages(rbname, start, len)) {
1441         mark_source_rp_bad(ms);
1442     }
1443 }
1444 
1445 /*
1446  * Handles messages sent on the return path towards the source VM
1447  *
1448  */
1449 static void *source_return_path_thread(void *opaque)
1450 {
1451     MigrationState *ms = opaque;
1452     QEMUFile *rp = ms->rp_state.from_dst_file;
1453     uint16_t header_len, header_type;
1454     uint8_t buf[512];
1455     uint32_t tmp32, sibling_error;
1456     ram_addr_t start = 0; /* =0 to silence warning */
1457     size_t  len = 0, expected_len;
1458     int res;
1459 
1460     trace_source_return_path_thread_entry();
1461     while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
1462            migration_is_setup_or_active(ms->state)) {
1463         trace_source_return_path_thread_loop_top();
1464         header_type = qemu_get_be16(rp);
1465         header_len = qemu_get_be16(rp);
1466 
1467         if (header_type >= MIG_RP_MSG_MAX ||
1468             header_type == MIG_RP_MSG_INVALID) {
1469             error_report("RP: Received invalid message 0x%04x length 0x%04x",
1470                     header_type, header_len);
1471             mark_source_rp_bad(ms);
1472             goto out;
1473         }
1474 
1475         if ((rp_cmd_args[header_type].len != -1 &&
1476             header_len != rp_cmd_args[header_type].len) ||
1477             header_len > sizeof(buf)) {
1478             error_report("RP: Received '%s' message (0x%04x) with"
1479                     "incorrect length %d expecting %zu",
1480                     rp_cmd_args[header_type].name, header_type, header_len,
1481                     (size_t)rp_cmd_args[header_type].len);
1482             mark_source_rp_bad(ms);
1483             goto out;
1484         }
1485 
1486         /* We know we've got a valid header by this point */
1487         res = qemu_get_buffer(rp, buf, header_len);
1488         if (res != header_len) {
1489             error_report("RP: Failed reading data for message 0x%04x"
1490                          " read %d expected %d",
1491                          header_type, res, header_len);
1492             mark_source_rp_bad(ms);
1493             goto out;
1494         }
1495 
1496         /* OK, we have the message and the data */
1497         switch (header_type) {
1498         case MIG_RP_MSG_SHUT:
1499             sibling_error = ldl_be_p(buf);
1500             trace_source_return_path_thread_shut(sibling_error);
1501             if (sibling_error) {
1502                 error_report("RP: Sibling indicated error %d", sibling_error);
1503                 mark_source_rp_bad(ms);
1504             }
1505             /*
1506              * We'll let the main thread deal with closing the RP
1507              * we could do a shutdown(2) on it, but we're the only user
1508              * anyway, so there's nothing gained.
1509              */
1510             goto out;
1511 
1512         case MIG_RP_MSG_PONG:
1513             tmp32 = ldl_be_p(buf);
1514             trace_source_return_path_thread_pong(tmp32);
1515             break;
1516 
1517         case MIG_RP_MSG_REQ_PAGES:
1518             start = ldq_be_p(buf);
1519             len = ldl_be_p(buf + 8);
1520             migrate_handle_rp_req_pages(ms, NULL, start, len);
1521             break;
1522 
1523         case MIG_RP_MSG_REQ_PAGES_ID:
1524             expected_len = 12 + 1; /* header + termination */
1525 
1526             if (header_len >= expected_len) {
1527                 start = ldq_be_p(buf);
1528                 len = ldl_be_p(buf + 8);
1529                 /* Now we expect an idstr */
1530                 tmp32 = buf[12]; /* Length of the following idstr */
1531                 buf[13 + tmp32] = '\0';
1532                 expected_len += tmp32;
1533             }
1534             if (header_len != expected_len) {
1535                 error_report("RP: Req_Page_id with length %d expecting %zd",
1536                         header_len, expected_len);
1537                 mark_source_rp_bad(ms);
1538                 goto out;
1539             }
1540             migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
1541             break;
1542 
1543         default:
1544             break;
1545         }
1546     }
1547     if (qemu_file_get_error(rp)) {
1548         trace_source_return_path_thread_bad_end();
1549         mark_source_rp_bad(ms);
1550     }
1551 
1552     trace_source_return_path_thread_end();
1553 out:
1554     ms->rp_state.from_dst_file = NULL;
1555     qemu_fclose(rp);
1556     return NULL;
1557 }
1558 
1559 static int open_return_path_on_source(MigrationState *ms)
1560 {
1561 
1562     ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
1563     if (!ms->rp_state.from_dst_file) {
1564         return -1;
1565     }
1566 
1567     trace_open_return_path_on_source();
1568     qemu_thread_create(&ms->rp_state.rp_thread, "return path",
1569                        source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
1570 
1571     trace_open_return_path_on_source_continue();
1572 
1573     return 0;
1574 }
1575 
1576 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1577 static int await_return_path_close_on_source(MigrationState *ms)
1578 {
1579     /*
1580      * If this is a normal exit then the destination will send a SHUT and the
1581      * rp_thread will exit, however if there's an error we need to cause
1582      * it to exit.
1583      */
1584     if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
1585         /*
1586          * shutdown(2), if we have it, will cause it to unblock if it's stuck
1587          * waiting for the destination.
1588          */
1589         qemu_file_shutdown(ms->rp_state.from_dst_file);
1590         mark_source_rp_bad(ms);
1591     }
1592     trace_await_return_path_close_on_source_joining();
1593     qemu_thread_join(&ms->rp_state.rp_thread);
1594     trace_await_return_path_close_on_source_close();
1595     return ms->rp_state.error;
1596 }
1597 
1598 /*
1599  * Switch from normal iteration to postcopy
1600  * Returns non-0 on error
1601  */
1602 static int postcopy_start(MigrationState *ms, bool *old_vm_running)
1603 {
1604     int ret;
1605     QIOChannelBuffer *bioc;
1606     QEMUFile *fb;
1607     int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1608     bool restart_block = false;
1609     migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
1610                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
1611 
1612     trace_postcopy_start();
1613     qemu_mutex_lock_iothread();
1614     trace_postcopy_start_set_run();
1615 
1616     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1617     *old_vm_running = runstate_is_running();
1618     global_state_store();
1619     ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1620     if (ret < 0) {
1621         goto fail;
1622     }
1623 
1624     ret = bdrv_inactivate_all();
1625     if (ret < 0) {
1626         goto fail;
1627     }
1628     restart_block = true;
1629 
1630     /*
1631      * Cause any non-postcopiable, but iterative devices to
1632      * send out their final data.
1633      */
1634     qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
1635 
1636     /*
1637      * in Finish migrate and with the io-lock held everything should
1638      * be quiet, but we've potentially still got dirty pages and we
1639      * need to tell the destination to throw any pages it's already received
1640      * that are dirty
1641      */
1642     if (ram_postcopy_send_discard_bitmap(ms)) {
1643         error_report("postcopy send discard bitmap failed");
1644         goto fail;
1645     }
1646 
1647     /*
1648      * send rest of state - note things that are doing postcopy
1649      * will notice we're in POSTCOPY_ACTIVE and not actually
1650      * wrap their state up here
1651      */
1652     qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
1653     /* Ping just for debugging, helps line traces up */
1654     qemu_savevm_send_ping(ms->to_dst_file, 2);
1655 
1656     /*
1657      * While loading the device state we may trigger page transfer
1658      * requests and the fd must be free to process those, and thus
1659      * the destination must read the whole device state off the fd before
1660      * it starts processing it.  Unfortunately the ad-hoc migration format
1661      * doesn't allow the destination to know the size to read without fully
1662      * parsing it through each devices load-state code (especially the open
1663      * coded devices that use get/put).
1664      * So we wrap the device state up in a package with a length at the start;
1665      * to do this we use a qemu_buf to hold the whole of the device state.
1666      */
1667     bioc = qio_channel_buffer_new(4096);
1668     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
1669     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
1670     object_unref(OBJECT(bioc));
1671 
1672     /*
1673      * Make sure the receiver can get incoming pages before we send the rest
1674      * of the state
1675      */
1676     qemu_savevm_send_postcopy_listen(fb);
1677 
1678     qemu_savevm_state_complete_precopy(fb, false, false);
1679     qemu_savevm_send_ping(fb, 3);
1680 
1681     qemu_savevm_send_postcopy_run(fb);
1682 
1683     /* <><> end of stuff going into the package */
1684 
1685     /* Last point of recovery; as soon as we send the package the destination
1686      * can open devices and potentially start running.
1687      * Lets just check again we've not got any errors.
1688      */
1689     ret = qemu_file_get_error(ms->to_dst_file);
1690     if (ret) {
1691         error_report("postcopy_start: Migration stream errored (pre package)");
1692         goto fail_closefb;
1693     }
1694 
1695     restart_block = false;
1696 
1697     /* Now send that blob */
1698     if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
1699         goto fail_closefb;
1700     }
1701     qemu_fclose(fb);
1702 
1703     /* Send a notify to give a chance for anything that needs to happen
1704      * at the transition to postcopy and after the device state; in particular
1705      * spice needs to trigger a transition now
1706      */
1707     ms->postcopy_after_devices = true;
1708     notifier_list_notify(&migration_state_notifiers, ms);
1709 
1710     ms->downtime =  qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
1711 
1712     qemu_mutex_unlock_iothread();
1713 
1714     /*
1715      * Although this ping is just for debug, it could potentially be
1716      * used for getting a better measurement of downtime at the source.
1717      */
1718     qemu_savevm_send_ping(ms->to_dst_file, 4);
1719 
1720     if (migrate_release_ram()) {
1721         ram_postcopy_migrated_memory_release(ms);
1722     }
1723 
1724     ret = qemu_file_get_error(ms->to_dst_file);
1725     if (ret) {
1726         error_report("postcopy_start: Migration stream errored");
1727         migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1728                               MIGRATION_STATUS_FAILED);
1729     }
1730 
1731     return ret;
1732 
1733 fail_closefb:
1734     qemu_fclose(fb);
1735 fail:
1736     migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1737                           MIGRATION_STATUS_FAILED);
1738     if (restart_block) {
1739         /* A failure happened early enough that we know the destination hasn't
1740          * accessed block devices, so we're safe to recover.
1741          */
1742         Error *local_err = NULL;
1743 
1744         bdrv_invalidate_cache_all(&local_err);
1745         if (local_err) {
1746             error_report_err(local_err);
1747         }
1748     }
1749     qemu_mutex_unlock_iothread();
1750     return -1;
1751 }
1752 
1753 /**
1754  * migration_completion: Used by migration_thread when there's not much left.
1755  *   The caller 'breaks' the loop when this returns.
1756  *
1757  * @s: Current migration state
1758  * @current_active_state: The migration state we expect to be in
1759  * @*old_vm_running: Pointer to old_vm_running flag
1760  * @*start_time: Pointer to time to update
1761  */
1762 static void migration_completion(MigrationState *s, int current_active_state,
1763                                  bool *old_vm_running,
1764                                  int64_t *start_time)
1765 {
1766     int ret;
1767 
1768     if (s->state == MIGRATION_STATUS_ACTIVE) {
1769         qemu_mutex_lock_iothread();
1770         *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1771         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1772         *old_vm_running = runstate_is_running();
1773         ret = global_state_store();
1774 
1775         if (!ret) {
1776             bool inactivate = !migrate_colo_enabled();
1777             ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1778             if (ret >= 0) {
1779                 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
1780                 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
1781                                                          inactivate);
1782             }
1783             if (inactivate && ret >= 0) {
1784                 s->block_inactive = true;
1785             }
1786         }
1787         qemu_mutex_unlock_iothread();
1788 
1789         if (ret < 0) {
1790             goto fail;
1791         }
1792     } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1793         trace_migration_completion_postcopy_end();
1794 
1795         qemu_savevm_state_complete_postcopy(s->to_dst_file);
1796         trace_migration_completion_postcopy_end_after_complete();
1797     }
1798 
1799     /*
1800      * If rp was opened we must clean up the thread before
1801      * cleaning everything else up (since if there are no failures
1802      * it will wait for the destination to send it's status in
1803      * a SHUT command).
1804      */
1805     if (s->rp_state.from_dst_file) {
1806         int rp_error;
1807         trace_migration_return_path_end_before();
1808         rp_error = await_return_path_close_on_source(s);
1809         trace_migration_return_path_end_after(rp_error);
1810         if (rp_error) {
1811             goto fail_invalidate;
1812         }
1813     }
1814 
1815     if (qemu_file_get_error(s->to_dst_file)) {
1816         trace_migration_completion_file_err();
1817         goto fail_invalidate;
1818     }
1819 
1820     if (!migrate_colo_enabled()) {
1821         migrate_set_state(&s->state, current_active_state,
1822                           MIGRATION_STATUS_COMPLETED);
1823     }
1824 
1825     return;
1826 
1827 fail_invalidate:
1828     /* If not doing postcopy, vm_start() will be called: let's regain
1829      * control on images.
1830      */
1831     if (s->state == MIGRATION_STATUS_ACTIVE) {
1832         Error *local_err = NULL;
1833 
1834         qemu_mutex_lock_iothread();
1835         bdrv_invalidate_cache_all(&local_err);
1836         if (local_err) {
1837             error_report_err(local_err);
1838         } else {
1839             s->block_inactive = false;
1840         }
1841         qemu_mutex_unlock_iothread();
1842     }
1843 
1844 fail:
1845     migrate_set_state(&s->state, current_active_state,
1846                       MIGRATION_STATUS_FAILED);
1847 }
1848 
1849 bool migrate_colo_enabled(void)
1850 {
1851     MigrationState *s = migrate_get_current();
1852     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
1853 }
1854 
1855 /*
1856  * Master migration thread on the source VM.
1857  * It drives the migration and pumps the data down the outgoing channel.
1858  */
1859 static void *migration_thread(void *opaque)
1860 {
1861     MigrationState *s = opaque;
1862     /* Used by the bandwidth calcs, updated later */
1863     int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1864     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
1865     int64_t initial_bytes = 0;
1866     /*
1867      * The final stage happens when the remaining data is smaller than
1868      * this threshold; it's calculated from the requested downtime and
1869      * measured bandwidth
1870      */
1871     int64_t threshold_size = 0;
1872     int64_t start_time = initial_time;
1873     int64_t end_time;
1874     bool old_vm_running = false;
1875     bool entered_postcopy = false;
1876     /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
1877     enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE;
1878     bool enable_colo = migrate_colo_enabled();
1879 
1880     rcu_register_thread();
1881 
1882     qemu_savevm_state_header(s->to_dst_file);
1883 
1884     /*
1885      * If we opened the return path, we need to make sure dst has it
1886      * opened as well.
1887      */
1888     if (s->rp_state.from_dst_file) {
1889         /* Now tell the dest that it should open its end so it can reply */
1890         qemu_savevm_send_open_return_path(s->to_dst_file);
1891 
1892         /* And do a ping that will make stuff easier to debug */
1893         qemu_savevm_send_ping(s->to_dst_file, 1);
1894     }
1895 
1896     if (migrate_postcopy_ram()) {
1897         /*
1898          * Tell the destination that we *might* want to do postcopy later;
1899          * if the other end can't do postcopy it should fail now, nice and
1900          * early.
1901          */
1902         qemu_savevm_send_postcopy_advise(s->to_dst_file);
1903     }
1904 
1905     qemu_savevm_state_setup(s->to_dst_file);
1906 
1907     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
1908     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1909                       MIGRATION_STATUS_ACTIVE);
1910 
1911     trace_migration_thread_setup_complete();
1912 
1913     while (s->state == MIGRATION_STATUS_ACTIVE ||
1914            s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1915         int64_t current_time;
1916         uint64_t pending_size;
1917 
1918         if (!qemu_file_rate_limit(s->to_dst_file)) {
1919             uint64_t pend_post, pend_nonpost;
1920 
1921             qemu_savevm_state_pending(s->to_dst_file, threshold_size,
1922                                       &pend_nonpost, &pend_post);
1923             pending_size = pend_nonpost + pend_post;
1924             trace_migrate_pending(pending_size, threshold_size,
1925                                   pend_post, pend_nonpost);
1926             if (pending_size && pending_size >= threshold_size) {
1927                 /* Still a significant amount to transfer */
1928 
1929                 if (migrate_postcopy_ram() &&
1930                     s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE &&
1931                     pend_nonpost <= threshold_size &&
1932                     atomic_read(&s->start_postcopy)) {
1933 
1934                     if (!postcopy_start(s, &old_vm_running)) {
1935                         current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE;
1936                         entered_postcopy = true;
1937                     }
1938 
1939                     continue;
1940                 }
1941                 /* Just another iteration step */
1942                 qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy);
1943             } else {
1944                 trace_migration_thread_low_pending(pending_size);
1945                 migration_completion(s, current_active_state,
1946                                      &old_vm_running, &start_time);
1947                 break;
1948             }
1949         }
1950 
1951         if (qemu_file_get_error(s->to_dst_file)) {
1952             migrate_set_state(&s->state, current_active_state,
1953                               MIGRATION_STATUS_FAILED);
1954             trace_migration_thread_file_err();
1955             break;
1956         }
1957         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1958         if (current_time >= initial_time + BUFFER_DELAY) {
1959             uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) -
1960                                          initial_bytes;
1961             uint64_t time_spent = current_time - initial_time;
1962             double bandwidth = (double)transferred_bytes / time_spent;
1963             threshold_size = bandwidth * s->parameters.downtime_limit;
1964 
1965             s->mbps = (((double) transferred_bytes * 8.0) /
1966                     ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
1967 
1968             trace_migrate_transferred(transferred_bytes, time_spent,
1969                                       bandwidth, threshold_size);
1970             /* if we haven't sent anything, we don't want to recalculate
1971                10000 is a small enough number for our purposes */
1972             if (ram_counters.dirty_pages_rate && transferred_bytes > 10000) {
1973                 s->expected_downtime = ram_counters.dirty_pages_rate *
1974                     qemu_target_page_size() / bandwidth;
1975             }
1976 
1977             qemu_file_reset_rate_limit(s->to_dst_file);
1978             initial_time = current_time;
1979             initial_bytes = qemu_ftell(s->to_dst_file);
1980         }
1981         if (qemu_file_rate_limit(s->to_dst_file)) {
1982             /* usleep expects microseconds */
1983             g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
1984         }
1985     }
1986 
1987     trace_migration_thread_after_loop();
1988     /* If we enabled cpu throttling for auto-converge, turn it off. */
1989     cpu_throttle_stop();
1990     end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1991 
1992     qemu_mutex_lock_iothread();
1993     /*
1994      * The resource has been allocated by migration will be reused in COLO
1995      * process, so don't release them.
1996      */
1997     if (!enable_colo) {
1998         qemu_savevm_state_cleanup();
1999     }
2000     if (s->state == MIGRATION_STATUS_COMPLETED) {
2001         uint64_t transferred_bytes = qemu_ftell(s->to_dst_file);
2002         s->total_time = end_time - s->total_time;
2003         if (!entered_postcopy) {
2004             s->downtime = end_time - start_time;
2005         }
2006         if (s->total_time) {
2007             s->mbps = (((double) transferred_bytes * 8.0) /
2008                        ((double) s->total_time)) / 1000;
2009         }
2010         runstate_set(RUN_STATE_POSTMIGRATE);
2011     } else {
2012         if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) {
2013             migrate_start_colo_process(s);
2014             qemu_savevm_state_cleanup();
2015             /*
2016             * Fixme: we will run VM in COLO no matter its old running state.
2017             * After exited COLO, we will keep running.
2018             */
2019             old_vm_running = true;
2020         }
2021         if (old_vm_running && !entered_postcopy) {
2022             vm_start();
2023         } else {
2024             if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2025                 runstate_set(RUN_STATE_POSTMIGRATE);
2026             }
2027         }
2028     }
2029     qemu_bh_schedule(s->cleanup_bh);
2030     qemu_mutex_unlock_iothread();
2031 
2032     rcu_unregister_thread();
2033     return NULL;
2034 }
2035 
2036 void migrate_fd_connect(MigrationState *s)
2037 {
2038     s->expected_downtime = s->parameters.downtime_limit;
2039     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
2040 
2041     qemu_file_set_blocking(s->to_dst_file, true);
2042     qemu_file_set_rate_limit(s->to_dst_file,
2043                              s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
2044 
2045     /* Notify before starting migration thread */
2046     notifier_list_notify(&migration_state_notifiers, s);
2047 
2048     /*
2049      * Open the return path. For postcopy, it is used exclusively. For
2050      * precopy, only if user specified "return-path" capability would
2051      * QEMU uses the return path.
2052      */
2053     if (migrate_postcopy_ram() || migrate_use_return_path()) {
2054         if (open_return_path_on_source(s)) {
2055             error_report("Unable to open return-path for postcopy");
2056             migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2057                               MIGRATION_STATUS_FAILED);
2058             migrate_fd_cleanup(s);
2059             return;
2060         }
2061     }
2062 
2063     qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
2064                        QEMU_THREAD_JOINABLE);
2065     s->migration_thread_running = true;
2066 }
2067 
2068 void migration_global_dump(Monitor *mon)
2069 {
2070     MigrationState *ms = migrate_get_current();
2071 
2072     monitor_printf(mon, "globals: store-global-state=%d, only_migratable=%d, "
2073                    "send-configuration=%d, send-section-footer=%d\n",
2074                    ms->store_global_state, ms->only_migratable,
2075                    ms->send_configuration, ms->send_section_footer);
2076 }
2077 
2078 #define DEFINE_PROP_MIG_CAP(name, x)             \
2079     DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
2080 
2081 static Property migration_properties[] = {
2082     DEFINE_PROP_BOOL("store-global-state", MigrationState,
2083                      store_global_state, true),
2084     DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false),
2085     DEFINE_PROP_BOOL("send-configuration", MigrationState,
2086                      send_configuration, true),
2087     DEFINE_PROP_BOOL("send-section-footer", MigrationState,
2088                      send_section_footer, true),
2089 
2090     /* Migration parameters */
2091     DEFINE_PROP_INT64("x-compress-level", MigrationState,
2092                       parameters.compress_level,
2093                       DEFAULT_MIGRATE_COMPRESS_LEVEL),
2094     DEFINE_PROP_INT64("x-compress-threads", MigrationState,
2095                       parameters.compress_threads,
2096                       DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
2097     DEFINE_PROP_INT64("x-decompress-threads", MigrationState,
2098                       parameters.decompress_threads,
2099                       DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
2100     DEFINE_PROP_INT64("x-cpu-throttle-initial", MigrationState,
2101                       parameters.cpu_throttle_initial,
2102                       DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
2103     DEFINE_PROP_INT64("x-cpu-throttle-increment", MigrationState,
2104                       parameters.cpu_throttle_increment,
2105                       DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
2106     DEFINE_PROP_INT64("x-max-bandwidth", MigrationState,
2107                       parameters.max_bandwidth, MAX_THROTTLE),
2108     DEFINE_PROP_INT64("x-downtime-limit", MigrationState,
2109                       parameters.downtime_limit,
2110                       DEFAULT_MIGRATE_SET_DOWNTIME),
2111     DEFINE_PROP_INT64("x-checkpoint-delay", MigrationState,
2112                       parameters.x_checkpoint_delay,
2113                       DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
2114 
2115     /* Migration capabilities */
2116     DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
2117     DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
2118     DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
2119     DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
2120     DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
2121     DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
2122     DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
2123     DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
2124     DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
2125     DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
2126     DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
2127 
2128     DEFINE_PROP_END_OF_LIST(),
2129 };
2130 
2131 static void migration_class_init(ObjectClass *klass, void *data)
2132 {
2133     DeviceClass *dc = DEVICE_CLASS(klass);
2134 
2135     dc->user_creatable = false;
2136     dc->props = migration_properties;
2137 }
2138 
2139 static void migration_instance_init(Object *obj)
2140 {
2141     MigrationState *ms = MIGRATION_OBJ(obj);
2142     MigrationParameters *params = &ms->parameters;
2143 
2144     ms->state = MIGRATION_STATUS_NONE;
2145     ms->xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE;
2146     ms->mbps = -1;
2147 
2148     params->tls_hostname = g_strdup("");
2149     params->tls_creds = g_strdup("");
2150 
2151     /* Set has_* up only for parameter checks */
2152     params->has_compress_level = true;
2153     params->has_compress_threads = true;
2154     params->has_decompress_threads = true;
2155     params->has_cpu_throttle_initial = true;
2156     params->has_cpu_throttle_increment = true;
2157     params->has_max_bandwidth = true;
2158     params->has_downtime_limit = true;
2159     params->has_x_checkpoint_delay = true;
2160     params->has_block_incremental = true;
2161 }
2162 
2163 /*
2164  * Return true if check pass, false otherwise. Error will be put
2165  * inside errp if provided.
2166  */
2167 static bool migration_object_check(MigrationState *ms, Error **errp)
2168 {
2169     MigrationCapabilityStatusList *head = NULL;
2170     /* Assuming all off */
2171     bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
2172     int i;
2173 
2174     if (!migrate_params_check(&ms->parameters, errp)) {
2175         return false;
2176     }
2177 
2178     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
2179         if (ms->enabled_capabilities[i]) {
2180             head = migrate_cap_add(head, i, true);
2181         }
2182     }
2183 
2184     ret = migrate_caps_check(cap_list, head, errp);
2185 
2186     /* It works with head == NULL */
2187     qapi_free_MigrationCapabilityStatusList(head);
2188 
2189     return ret;
2190 }
2191 
2192 static const TypeInfo migration_type = {
2193     .name = TYPE_MIGRATION,
2194     /*
2195      * NOTE: TYPE_MIGRATION is not really a device, as the object is
2196      * not created using qdev_create(), it is not attached to the qdev
2197      * device tree, and it is never realized.
2198      *
2199      * TODO: Make this TYPE_OBJECT once QOM provides something like
2200      * TYPE_DEVICE's "-global" properties.
2201      */
2202     .parent = TYPE_DEVICE,
2203     .class_init = migration_class_init,
2204     .class_size = sizeof(MigrationClass),
2205     .instance_size = sizeof(MigrationState),
2206     .instance_init = migration_instance_init,
2207 };
2208 
2209 static void register_migration_types(void)
2210 {
2211     type_register_static(&migration_type);
2212 }
2213 
2214 type_init(register_migration_types);
2215