xref: /openbmc/qemu/migration/migration.c (revision a89f364ae8740dfc31b321eed9ee454e996dc3c1)
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 "qemu/rcu.h"
35 #include "block.h"
36 #include "postcopy-ram.h"
37 #include "qemu/thread.h"
38 #include "qmp-commands.h"
39 #include "trace.h"
40 #include "qapi-event.h"
41 #include "exec/target_page.h"
42 #include "io/channel-buffer.h"
43 #include "migration/colo.h"
44 #include "hw/boards.h"
45 #include "monitor/monitor.h"
46 
47 #define MAX_THROTTLE  (32 << 20)      /* Migration transfer speed throttling */
48 
49 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
50  * data. */
51 #define BUFFER_DELAY     100
52 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
53 
54 /* Time in milliseconds we are allowed to stop the source,
55  * for sending the last part */
56 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
57 
58 /* Maximum migrate downtime set to 2000 seconds */
59 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
60 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
61 
62 /* Default compression thread count */
63 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
64 /* Default decompression thread count, usually decompression is at
65  * least 4 times as fast as compression.*/
66 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
67 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
68 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
69 /* Define default autoconverge cpu throttle migration parameters */
70 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
71 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
72 
73 /* Migration XBZRLE default cache size */
74 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
75 
76 /* The delay time (in ms) between two COLO checkpoints
77  * Note: Please change this default value to 10000 when we support hybrid mode.
78  */
79 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
80 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
81 #define DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT 16
82 
83 static NotifierList migration_state_notifiers =
84     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
85 
86 static bool deferred_incoming;
87 
88 /* Messages sent on the return path from destination to source */
89 enum mig_rp_message_type {
90     MIG_RP_MSG_INVALID = 0,  /* Must be 0 */
91     MIG_RP_MSG_SHUT,         /* sibling will not send any more RP messages */
92     MIG_RP_MSG_PONG,         /* Response to a PING; data (seq: be32 ) */
93 
94     MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
95     MIG_RP_MSG_REQ_PAGES,    /* data (start: be64, len: be32) */
96 
97     MIG_RP_MSG_MAX
98 };
99 
100 /* When we add fault tolerance, we could have several
101    migrations at once.  For now we don't need to add
102    dynamic creation of migration */
103 
104 static MigrationState *current_migration;
105 
106 static bool migration_object_check(MigrationState *ms, Error **errp);
107 static int migration_maybe_pause(MigrationState *s,
108                                  int *current_active_state,
109                                  int new_state);
110 
111 void migration_object_init(void)
112 {
113     MachineState *ms = MACHINE(qdev_get_machine());
114     Error *err = NULL;
115 
116     /* This can only be called once. */
117     assert(!current_migration);
118     current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
119 
120     if (!migration_object_check(current_migration, &err)) {
121         error_report_err(err);
122         exit(1);
123     }
124 
125     /*
126      * We cannot really do this in migration_instance_init() since at
127      * that time global properties are not yet applied, then this
128      * value will be definitely replaced by something else.
129      */
130     if (ms->enforce_config_section) {
131         current_migration->send_configuration = true;
132     }
133 }
134 
135 void migration_object_finalize(void)
136 {
137     object_unref(OBJECT(current_migration));
138 }
139 
140 /* For outgoing */
141 MigrationState *migrate_get_current(void)
142 {
143     /* This can only be called after the object created. */
144     assert(current_migration);
145     return current_migration;
146 }
147 
148 MigrationIncomingState *migration_incoming_get_current(void)
149 {
150     static bool once;
151     static MigrationIncomingState mis_current;
152 
153     if (!once) {
154         mis_current.state = MIGRATION_STATUS_NONE;
155         memset(&mis_current, 0, sizeof(MigrationIncomingState));
156         qemu_mutex_init(&mis_current.rp_mutex);
157         qemu_event_init(&mis_current.main_thread_load_event, false);
158         once = true;
159     }
160     return &mis_current;
161 }
162 
163 void migration_incoming_state_destroy(void)
164 {
165     struct MigrationIncomingState *mis = migration_incoming_get_current();
166 
167     if (mis->to_src_file) {
168         /* Tell source that we are done */
169         migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
170         qemu_fclose(mis->to_src_file);
171         mis->to_src_file = NULL;
172     }
173 
174     if (mis->from_src_file) {
175         qemu_fclose(mis->from_src_file);
176         mis->from_src_file = NULL;
177     }
178 
179     qemu_event_reset(&mis->main_thread_load_event);
180 }
181 
182 static void migrate_generate_event(int new_state)
183 {
184     if (migrate_use_events()) {
185         qapi_event_send_migration(new_state, &error_abort);
186     }
187 }
188 
189 /*
190  * Called on -incoming with a defer: uri.
191  * The migration can be started later after any parameters have been
192  * changed.
193  */
194 static void deferred_incoming_migration(Error **errp)
195 {
196     if (deferred_incoming) {
197         error_setg(errp, "Incoming migration already deferred");
198     }
199     deferred_incoming = true;
200 }
201 
202 /*
203  * Send a message on the return channel back to the source
204  * of the migration.
205  */
206 static void migrate_send_rp_message(MigrationIncomingState *mis,
207                                     enum mig_rp_message_type message_type,
208                                     uint16_t len, void *data)
209 {
210     trace_migrate_send_rp_message((int)message_type, len);
211     qemu_mutex_lock(&mis->rp_mutex);
212     qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
213     qemu_put_be16(mis->to_src_file, len);
214     qemu_put_buffer(mis->to_src_file, data, len);
215     qemu_fflush(mis->to_src_file);
216     qemu_mutex_unlock(&mis->rp_mutex);
217 }
218 
219 /* Request a range of pages from the source VM at the given
220  * start address.
221  *   rbname: Name of the RAMBlock to request the page in, if NULL it's the same
222  *           as the last request (a name must have been given previously)
223  *   Start: Address offset within the RB
224  *   Len: Length in bytes required - must be a multiple of pagesize
225  */
226 void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
227                                ram_addr_t start, size_t len)
228 {
229     uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
230     size_t msglen = 12; /* start + len */
231 
232     *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
233     *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
234 
235     if (rbname) {
236         int rbname_len = strlen(rbname);
237         assert(rbname_len < 256);
238 
239         bufc[msglen++] = rbname_len;
240         memcpy(bufc + msglen, rbname, rbname_len);
241         msglen += rbname_len;
242         migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc);
243     } else {
244         migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc);
245     }
246 }
247 
248 void qemu_start_incoming_migration(const char *uri, Error **errp)
249 {
250     const char *p;
251 
252     qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
253     if (!strcmp(uri, "defer")) {
254         deferred_incoming_migration(errp);
255     } else if (strstart(uri, "tcp:", &p)) {
256         tcp_start_incoming_migration(p, errp);
257 #ifdef CONFIG_RDMA
258     } else if (strstart(uri, "rdma:", &p)) {
259         rdma_start_incoming_migration(p, errp);
260 #endif
261     } else if (strstart(uri, "exec:", &p)) {
262         exec_start_incoming_migration(p, errp);
263     } else if (strstart(uri, "unix:", &p)) {
264         unix_start_incoming_migration(p, errp);
265     } else if (strstart(uri, "fd:", &p)) {
266         fd_start_incoming_migration(p, errp);
267     } else {
268         error_setg(errp, "unknown migration protocol: %s", uri);
269     }
270 }
271 
272 static void process_incoming_migration_bh(void *opaque)
273 {
274     Error *local_err = NULL;
275     MigrationIncomingState *mis = opaque;
276 
277     /* Make sure all file formats flush their mutable metadata.
278      * If we get an error here, just don't restart the VM yet. */
279     bdrv_invalidate_cache_all(&local_err);
280     if (local_err) {
281         error_report_err(local_err);
282         local_err = NULL;
283         autostart = false;
284     }
285 
286     /*
287      * This must happen after all error conditions are dealt with and
288      * we're sure the VM is going to be running on this host.
289      */
290     qemu_announce_self();
291 
292     if (multifd_load_cleanup(&local_err) != 0) {
293         error_report_err(local_err);
294         autostart = false;
295     }
296     /* If global state section was not received or we are in running
297        state, we need to obey autostart. Any other state is set with
298        runstate_set. */
299 
300     if (!global_state_received() ||
301         global_state_get_runstate() == RUN_STATE_RUNNING) {
302         if (autostart) {
303             vm_start();
304         } else {
305             runstate_set(RUN_STATE_PAUSED);
306         }
307     } else {
308         runstate_set(global_state_get_runstate());
309     }
310     /*
311      * This must happen after any state changes since as soon as an external
312      * observer sees this event they might start to prod at the VM assuming
313      * it's ready to use.
314      */
315     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
316                       MIGRATION_STATUS_COMPLETED);
317     qemu_bh_delete(mis->bh);
318     migration_incoming_state_destroy();
319 }
320 
321 static void process_incoming_migration_co(void *opaque)
322 {
323     MigrationIncomingState *mis = migration_incoming_get_current();
324     PostcopyState ps;
325     int ret;
326 
327     assert(mis->from_src_file);
328     mis->largest_page_size = qemu_ram_pagesize_largest();
329     postcopy_state_set(POSTCOPY_INCOMING_NONE);
330     migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
331                       MIGRATION_STATUS_ACTIVE);
332     ret = qemu_loadvm_state(mis->from_src_file);
333 
334     ps = postcopy_state_get();
335     trace_process_incoming_migration_co_end(ret, ps);
336     if (ps != POSTCOPY_INCOMING_NONE) {
337         if (ps == POSTCOPY_INCOMING_ADVISE) {
338             /*
339              * Where a migration had postcopy enabled (and thus went to advise)
340              * but managed to complete within the precopy period, we can use
341              * the normal exit.
342              */
343             postcopy_ram_incoming_cleanup(mis);
344         } else if (ret >= 0) {
345             /*
346              * Postcopy was started, cleanup should happen at the end of the
347              * postcopy thread.
348              */
349             trace_process_incoming_migration_co_postcopy_end_main();
350             return;
351         }
352         /* Else if something went wrong then just fall out of the normal exit */
353     }
354 
355     /* we get COLO info, and know if we are in COLO mode */
356     if (!ret && migration_incoming_enable_colo()) {
357         mis->migration_incoming_co = qemu_coroutine_self();
358         qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
359              colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
360         mis->have_colo_incoming_thread = true;
361         qemu_coroutine_yield();
362 
363         /* Wait checkpoint incoming thread exit before free resource */
364         qemu_thread_join(&mis->colo_incoming_thread);
365     }
366 
367     if (ret < 0) {
368         Error *local_err = NULL;
369 
370         migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
371                           MIGRATION_STATUS_FAILED);
372         error_report("load of migration failed: %s", strerror(-ret));
373         qemu_fclose(mis->from_src_file);
374         if (multifd_load_cleanup(&local_err) != 0) {
375             error_report_err(local_err);
376         }
377         exit(EXIT_FAILURE);
378     }
379     mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
380     qemu_bh_schedule(mis->bh);
381 }
382 
383 static void migration_incoming_setup(QEMUFile *f)
384 {
385     MigrationIncomingState *mis = migration_incoming_get_current();
386 
387     if (multifd_load_setup() != 0) {
388         /* We haven't been able to create multifd threads
389            nothing better to do */
390         exit(EXIT_FAILURE);
391     }
392 
393     if (!mis->from_src_file) {
394         mis->from_src_file = f;
395     }
396     qemu_file_set_blocking(f, false);
397 }
398 
399 static void migration_incoming_process(void)
400 {
401     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
402     qemu_coroutine_enter(co);
403 }
404 
405 void migration_fd_process_incoming(QEMUFile *f)
406 {
407     migration_incoming_setup(f);
408     migration_incoming_process();
409 }
410 
411 void migration_ioc_process_incoming(QIOChannel *ioc)
412 {
413     MigrationIncomingState *mis = migration_incoming_get_current();
414 
415     if (!mis->from_src_file) {
416         QEMUFile *f = qemu_fopen_channel_input(ioc);
417         migration_fd_process_incoming(f);
418     }
419     /* We still only have a single channel.  Nothing to do here yet */
420 }
421 
422 /**
423  * @migration_has_all_channels: We have received all channels that we need
424  *
425  * Returns true when we have got connections to all the channels that
426  * we need for migration.
427  */
428 bool migration_has_all_channels(void)
429 {
430     return true;
431 }
432 
433 /*
434  * Send a 'SHUT' message on the return channel with the given value
435  * to indicate that we've finished with the RP.  Non-0 value indicates
436  * error.
437  */
438 void migrate_send_rp_shut(MigrationIncomingState *mis,
439                           uint32_t value)
440 {
441     uint32_t buf;
442 
443     buf = cpu_to_be32(value);
444     migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
445 }
446 
447 /*
448  * Send a 'PONG' message on the return channel with the given value
449  * (normally in response to a 'PING')
450  */
451 void migrate_send_rp_pong(MigrationIncomingState *mis,
452                           uint32_t value)
453 {
454     uint32_t buf;
455 
456     buf = cpu_to_be32(value);
457     migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
458 }
459 
460 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
461 {
462     MigrationCapabilityStatusList *head = NULL;
463     MigrationCapabilityStatusList *caps;
464     MigrationState *s = migrate_get_current();
465     int i;
466 
467     caps = NULL; /* silence compiler warning */
468     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
469 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
470         if (i == MIGRATION_CAPABILITY_BLOCK) {
471             continue;
472         }
473 #endif
474         if (head == NULL) {
475             head = g_malloc0(sizeof(*caps));
476             caps = head;
477         } else {
478             caps->next = g_malloc0(sizeof(*caps));
479             caps = caps->next;
480         }
481         caps->value =
482             g_malloc(sizeof(*caps->value));
483         caps->value->capability = i;
484         caps->value->state = s->enabled_capabilities[i];
485     }
486 
487     return head;
488 }
489 
490 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
491 {
492     MigrationParameters *params;
493     MigrationState *s = migrate_get_current();
494 
495     /* TODO use QAPI_CLONE() instead of duplicating it inline */
496     params = g_malloc0(sizeof(*params));
497     params->has_compress_level = true;
498     params->compress_level = s->parameters.compress_level;
499     params->has_compress_threads = true;
500     params->compress_threads = s->parameters.compress_threads;
501     params->has_decompress_threads = true;
502     params->decompress_threads = s->parameters.decompress_threads;
503     params->has_cpu_throttle_initial = true;
504     params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
505     params->has_cpu_throttle_increment = true;
506     params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
507     params->has_tls_creds = true;
508     params->tls_creds = g_strdup(s->parameters.tls_creds);
509     params->has_tls_hostname = true;
510     params->tls_hostname = g_strdup(s->parameters.tls_hostname);
511     params->has_max_bandwidth = true;
512     params->max_bandwidth = s->parameters.max_bandwidth;
513     params->has_downtime_limit = true;
514     params->downtime_limit = s->parameters.downtime_limit;
515     params->has_x_checkpoint_delay = true;
516     params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
517     params->has_block_incremental = true;
518     params->block_incremental = s->parameters.block_incremental;
519     params->has_x_multifd_channels = true;
520     params->x_multifd_channels = s->parameters.x_multifd_channels;
521     params->has_x_multifd_page_count = true;
522     params->x_multifd_page_count = s->parameters.x_multifd_page_count;
523     params->has_xbzrle_cache_size = true;
524     params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
525 
526     return params;
527 }
528 
529 /*
530  * Return true if we're already in the middle of a migration
531  * (i.e. any of the active or setup states)
532  */
533 static bool migration_is_setup_or_active(int state)
534 {
535     switch (state) {
536     case MIGRATION_STATUS_ACTIVE:
537     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
538     case MIGRATION_STATUS_SETUP:
539     case MIGRATION_STATUS_PRE_SWITCHOVER:
540     case MIGRATION_STATUS_DEVICE:
541         return true;
542 
543     default:
544         return false;
545 
546     }
547 }
548 
549 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
550 {
551     info->has_ram = true;
552     info->ram = g_malloc0(sizeof(*info->ram));
553     info->ram->transferred = ram_counters.transferred;
554     info->ram->total = ram_bytes_total();
555     info->ram->duplicate = ram_counters.duplicate;
556     /* legacy value.  It is not used anymore */
557     info->ram->skipped = 0;
558     info->ram->normal = ram_counters.normal;
559     info->ram->normal_bytes = ram_counters.normal *
560         qemu_target_page_size();
561     info->ram->mbps = s->mbps;
562     info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
563     info->ram->postcopy_requests = ram_counters.postcopy_requests;
564     info->ram->page_size = qemu_target_page_size();
565 
566     if (migrate_use_xbzrle()) {
567         info->has_xbzrle_cache = true;
568         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
569         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
570         info->xbzrle_cache->bytes = xbzrle_counters.bytes;
571         info->xbzrle_cache->pages = xbzrle_counters.pages;
572         info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
573         info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
574         info->xbzrle_cache->overflow = xbzrle_counters.overflow;
575     }
576 
577     if (cpu_throttle_active()) {
578         info->has_cpu_throttle_percentage = true;
579         info->cpu_throttle_percentage = cpu_throttle_get_percentage();
580     }
581 
582     if (s->state != MIGRATION_STATUS_COMPLETED) {
583         info->ram->remaining = ram_bytes_remaining();
584         info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
585     }
586 }
587 
588 static void populate_disk_info(MigrationInfo *info)
589 {
590     if (blk_mig_active()) {
591         info->has_disk = true;
592         info->disk = g_malloc0(sizeof(*info->disk));
593         info->disk->transferred = blk_mig_bytes_transferred();
594         info->disk->remaining = blk_mig_bytes_remaining();
595         info->disk->total = blk_mig_bytes_total();
596     }
597 }
598 
599 static void fill_source_migration_info(MigrationInfo *info)
600 {
601     MigrationState *s = migrate_get_current();
602 
603     switch (s->state) {
604     case MIGRATION_STATUS_NONE:
605         /* no migration has happened ever */
606         /* do not overwrite destination migration status */
607         return;
608         break;
609     case MIGRATION_STATUS_SETUP:
610         info->has_status = true;
611         info->has_total_time = false;
612         break;
613     case MIGRATION_STATUS_ACTIVE:
614     case MIGRATION_STATUS_CANCELLING:
615     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
616     case MIGRATION_STATUS_PRE_SWITCHOVER:
617     case MIGRATION_STATUS_DEVICE:
618          /* TODO add some postcopy stats */
619         info->has_status = true;
620         info->has_total_time = true;
621         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
622             - s->start_time;
623         info->has_expected_downtime = true;
624         info->expected_downtime = s->expected_downtime;
625         info->has_setup_time = true;
626         info->setup_time = s->setup_time;
627 
628         populate_ram_info(info, s);
629         populate_disk_info(info);
630         break;
631     case MIGRATION_STATUS_COLO:
632         info->has_status = true;
633         /* TODO: display COLO specific information (checkpoint info etc.) */
634         break;
635     case MIGRATION_STATUS_COMPLETED:
636         info->has_status = true;
637         info->has_total_time = true;
638         info->total_time = s->total_time;
639         info->has_downtime = true;
640         info->downtime = s->downtime;
641         info->has_setup_time = true;
642         info->setup_time = s->setup_time;
643 
644         populate_ram_info(info, s);
645         break;
646     case MIGRATION_STATUS_FAILED:
647         info->has_status = true;
648         if (s->error) {
649             info->has_error_desc = true;
650             info->error_desc = g_strdup(error_get_pretty(s->error));
651         }
652         break;
653     case MIGRATION_STATUS_CANCELLED:
654         info->has_status = true;
655         break;
656     }
657     info->status = s->state;
658 }
659 
660 /**
661  * @migration_caps_check - check capability validity
662  *
663  * @cap_list: old capability list, array of bool
664  * @params: new capabilities to be applied soon
665  * @errp: set *errp if the check failed, with reason
666  *
667  * Returns true if check passed, otherwise false.
668  */
669 static bool migrate_caps_check(bool *cap_list,
670                                MigrationCapabilityStatusList *params,
671                                Error **errp)
672 {
673     MigrationCapabilityStatusList *cap;
674     bool old_postcopy_cap;
675     MigrationIncomingState *mis = migration_incoming_get_current();
676 
677     old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
678 
679     for (cap = params; cap; cap = cap->next) {
680         cap_list[cap->value->capability] = cap->value->state;
681     }
682 
683 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
684     if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
685         error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
686                    "block migration");
687         error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
688         return false;
689     }
690 #endif
691 
692     if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
693         if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
694             /* The decompression threads asynchronously write into RAM
695              * rather than use the atomic copies needed to avoid
696              * userfaulting.  It should be possible to fix the decompression
697              * threads for compatibility in future.
698              */
699             error_setg(errp, "Postcopy is not currently compatible "
700                        "with compression");
701             return false;
702         }
703 
704         /* This check is reasonably expensive, so only when it's being
705          * set the first time, also it's only the destination that needs
706          * special support.
707          */
708         if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
709             !postcopy_ram_supported_by_host(mis)) {
710             /* postcopy_ram_supported_by_host will have emitted a more
711              * detailed message
712              */
713             error_setg(errp, "Postcopy is not supported");
714             return false;
715         }
716     }
717 
718     return true;
719 }
720 
721 static void fill_destination_migration_info(MigrationInfo *info)
722 {
723     MigrationIncomingState *mis = migration_incoming_get_current();
724 
725     switch (mis->state) {
726     case MIGRATION_STATUS_NONE:
727         return;
728         break;
729     case MIGRATION_STATUS_SETUP:
730     case MIGRATION_STATUS_CANCELLING:
731     case MIGRATION_STATUS_CANCELLED:
732     case MIGRATION_STATUS_ACTIVE:
733     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
734     case MIGRATION_STATUS_FAILED:
735     case MIGRATION_STATUS_COLO:
736         info->has_status = true;
737         break;
738     case MIGRATION_STATUS_COMPLETED:
739         info->has_status = true;
740         fill_destination_postcopy_migration_info(info);
741         break;
742     }
743     info->status = mis->state;
744 }
745 
746 MigrationInfo *qmp_query_migrate(Error **errp)
747 {
748     MigrationInfo *info = g_malloc0(sizeof(*info));
749 
750     fill_destination_migration_info(info);
751     fill_source_migration_info(info);
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 
762     if (migration_is_setup_or_active(s->state)) {
763         error_setg(errp, QERR_MIGRATION_ACTIVE);
764         return;
765     }
766 
767     if (!migrate_caps_check(s->enabled_capabilities, params, errp)) {
768         return;
769     }
770 
771     for (cap = params; cap; cap = cap->next) {
772         s->enabled_capabilities[cap->value->capability] = cap->value->state;
773     }
774 }
775 
776 /*
777  * Check whether the parameters are valid. Error will be put into errp
778  * (if provided). Return true if valid, otherwise false.
779  */
780 static bool migrate_params_check(MigrationParameters *params, Error **errp)
781 {
782     if (params->has_compress_level &&
783         (params->compress_level > 9)) {
784         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
785                    "is invalid, it should be in the range of 0 to 9");
786         return false;
787     }
788 
789     if (params->has_compress_threads && (params->compress_threads < 1)) {
790         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
791                    "compress_threads",
792                    "is invalid, it should be in the range of 1 to 255");
793         return false;
794     }
795 
796     if (params->has_decompress_threads && (params->decompress_threads < 1)) {
797         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
798                    "decompress_threads",
799                    "is invalid, it should be in the range of 1 to 255");
800         return false;
801     }
802 
803     if (params->has_cpu_throttle_initial &&
804         (params->cpu_throttle_initial < 1 ||
805          params->cpu_throttle_initial > 99)) {
806         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
807                    "cpu_throttle_initial",
808                    "an integer in the range of 1 to 99");
809         return false;
810     }
811 
812     if (params->has_cpu_throttle_increment &&
813         (params->cpu_throttle_increment < 1 ||
814          params->cpu_throttle_increment > 99)) {
815         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
816                    "cpu_throttle_increment",
817                    "an integer in the range of 1 to 99");
818         return false;
819     }
820 
821     if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
822         error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
823                          " range of 0 to %zu bytes/second", SIZE_MAX);
824         return false;
825     }
826 
827     if (params->has_downtime_limit &&
828         (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
829         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
830                          "the range of 0 to %d milliseconds",
831                          MAX_MIGRATE_DOWNTIME);
832         return false;
833     }
834 
835     /* x_checkpoint_delay is now always positive */
836 
837     if (params->has_x_multifd_channels && (params->x_multifd_channels < 1)) {
838         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
839                    "multifd_channels",
840                    "is invalid, it should be in the range of 1 to 255");
841         return false;
842     }
843     if (params->has_x_multifd_page_count &&
844         (params->x_multifd_page_count < 1 ||
845          params->x_multifd_page_count > 10000)) {
846         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
847                    "multifd_page_count",
848                    "is invalid, it should be in the range of 1 to 10000");
849         return false;
850     }
851 
852     if (params->has_xbzrle_cache_size &&
853         (params->xbzrle_cache_size < qemu_target_page_size() ||
854          !is_power_of_2(params->xbzrle_cache_size))) {
855         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
856                    "xbzrle_cache_size",
857                    "is invalid, it should be bigger than target page size"
858                    " and a power of two");
859         return false;
860     }
861 
862     return true;
863 }
864 
865 static void migrate_params_test_apply(MigrateSetParameters *params,
866                                       MigrationParameters *dest)
867 {
868     *dest = migrate_get_current()->parameters;
869 
870     /* TODO use QAPI_CLONE() instead of duplicating it inline */
871 
872     if (params->has_compress_level) {
873         dest->compress_level = params->compress_level;
874     }
875 
876     if (params->has_compress_threads) {
877         dest->compress_threads = params->compress_threads;
878     }
879 
880     if (params->has_decompress_threads) {
881         dest->decompress_threads = params->decompress_threads;
882     }
883 
884     if (params->has_cpu_throttle_initial) {
885         dest->cpu_throttle_initial = params->cpu_throttle_initial;
886     }
887 
888     if (params->has_cpu_throttle_increment) {
889         dest->cpu_throttle_increment = params->cpu_throttle_increment;
890     }
891 
892     if (params->has_tls_creds) {
893         assert(params->tls_creds->type == QTYPE_QSTRING);
894         dest->tls_creds = g_strdup(params->tls_creds->u.s);
895     }
896 
897     if (params->has_tls_hostname) {
898         assert(params->tls_hostname->type == QTYPE_QSTRING);
899         dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
900     }
901 
902     if (params->has_max_bandwidth) {
903         dest->max_bandwidth = params->max_bandwidth;
904     }
905 
906     if (params->has_downtime_limit) {
907         dest->downtime_limit = params->downtime_limit;
908     }
909 
910     if (params->has_x_checkpoint_delay) {
911         dest->x_checkpoint_delay = params->x_checkpoint_delay;
912     }
913 
914     if (params->has_block_incremental) {
915         dest->block_incremental = params->block_incremental;
916     }
917     if (params->has_x_multifd_channels) {
918         dest->x_multifd_channels = params->x_multifd_channels;
919     }
920     if (params->has_x_multifd_page_count) {
921         dest->x_multifd_page_count = params->x_multifd_page_count;
922     }
923     if (params->has_xbzrle_cache_size) {
924         dest->xbzrle_cache_size = params->xbzrle_cache_size;
925     }
926 }
927 
928 static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
929 {
930     MigrationState *s = migrate_get_current();
931 
932     /* TODO use QAPI_CLONE() instead of duplicating it inline */
933 
934     if (params->has_compress_level) {
935         s->parameters.compress_level = params->compress_level;
936     }
937 
938     if (params->has_compress_threads) {
939         s->parameters.compress_threads = params->compress_threads;
940     }
941 
942     if (params->has_decompress_threads) {
943         s->parameters.decompress_threads = params->decompress_threads;
944     }
945 
946     if (params->has_cpu_throttle_initial) {
947         s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
948     }
949 
950     if (params->has_cpu_throttle_increment) {
951         s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
952     }
953 
954     if (params->has_tls_creds) {
955         g_free(s->parameters.tls_creds);
956         assert(params->tls_creds->type == QTYPE_QSTRING);
957         s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
958     }
959 
960     if (params->has_tls_hostname) {
961         g_free(s->parameters.tls_hostname);
962         assert(params->tls_hostname->type == QTYPE_QSTRING);
963         s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
964     }
965 
966     if (params->has_max_bandwidth) {
967         s->parameters.max_bandwidth = params->max_bandwidth;
968         if (s->to_dst_file) {
969             qemu_file_set_rate_limit(s->to_dst_file,
970                                 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
971         }
972     }
973 
974     if (params->has_downtime_limit) {
975         s->parameters.downtime_limit = params->downtime_limit;
976     }
977 
978     if (params->has_x_checkpoint_delay) {
979         s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
980         if (migration_in_colo_state()) {
981             colo_checkpoint_notify(s);
982         }
983     }
984 
985     if (params->has_block_incremental) {
986         s->parameters.block_incremental = params->block_incremental;
987     }
988     if (params->has_x_multifd_channels) {
989         s->parameters.x_multifd_channels = params->x_multifd_channels;
990     }
991     if (params->has_x_multifd_page_count) {
992         s->parameters.x_multifd_page_count = params->x_multifd_page_count;
993     }
994     if (params->has_xbzrle_cache_size) {
995         s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
996         xbzrle_cache_resize(params->xbzrle_cache_size, errp);
997     }
998 }
999 
1000 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
1001 {
1002     MigrationParameters tmp;
1003 
1004     /* TODO Rewrite "" to null instead */
1005     if (params->has_tls_creds
1006         && params->tls_creds->type == QTYPE_QNULL) {
1007         QDECREF(params->tls_creds->u.n);
1008         params->tls_creds->type = QTYPE_QSTRING;
1009         params->tls_creds->u.s = strdup("");
1010     }
1011     /* TODO Rewrite "" to null instead */
1012     if (params->has_tls_hostname
1013         && params->tls_hostname->type == QTYPE_QNULL) {
1014         QDECREF(params->tls_hostname->u.n);
1015         params->tls_hostname->type = QTYPE_QSTRING;
1016         params->tls_hostname->u.s = strdup("");
1017     }
1018 
1019     migrate_params_test_apply(params, &tmp);
1020 
1021     if (!migrate_params_check(&tmp, errp)) {
1022         /* Invalid parameter */
1023         return;
1024     }
1025 
1026     migrate_params_apply(params, errp);
1027 }
1028 
1029 
1030 void qmp_migrate_start_postcopy(Error **errp)
1031 {
1032     MigrationState *s = migrate_get_current();
1033 
1034     if (!migrate_postcopy_ram()) {
1035         error_setg(errp, "Enable postcopy with migrate_set_capability before"
1036                          " the start of migration");
1037         return;
1038     }
1039 
1040     if (s->state == MIGRATION_STATUS_NONE) {
1041         error_setg(errp, "Postcopy must be started after migration has been"
1042                          " started");
1043         return;
1044     }
1045     /*
1046      * we don't error if migration has finished since that would be racy
1047      * with issuing this command.
1048      */
1049     atomic_set(&s->start_postcopy, true);
1050 }
1051 
1052 /* shared migration helpers */
1053 
1054 void migrate_set_state(int *state, int old_state, int new_state)
1055 {
1056     assert(new_state < MIGRATION_STATUS__MAX);
1057     if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
1058         trace_migrate_set_state(MigrationStatus_str(new_state));
1059         migrate_generate_event(new_state);
1060     }
1061 }
1062 
1063 static MigrationCapabilityStatusList *migrate_cap_add(
1064     MigrationCapabilityStatusList *list,
1065     MigrationCapability index,
1066     bool state)
1067 {
1068     MigrationCapabilityStatusList *cap;
1069 
1070     cap = g_new0(MigrationCapabilityStatusList, 1);
1071     cap->value = g_new0(MigrationCapabilityStatus, 1);
1072     cap->value->capability = index;
1073     cap->value->state = state;
1074     cap->next = list;
1075 
1076     return cap;
1077 }
1078 
1079 void migrate_set_block_enabled(bool value, Error **errp)
1080 {
1081     MigrationCapabilityStatusList *cap;
1082 
1083     cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
1084     qmp_migrate_set_capabilities(cap, errp);
1085     qapi_free_MigrationCapabilityStatusList(cap);
1086 }
1087 
1088 static void migrate_set_block_incremental(MigrationState *s, bool value)
1089 {
1090     s->parameters.block_incremental = value;
1091 }
1092 
1093 static void block_cleanup_parameters(MigrationState *s)
1094 {
1095     if (s->must_remove_block_options) {
1096         /* setting to false can never fail */
1097         migrate_set_block_enabled(false, &error_abort);
1098         migrate_set_block_incremental(s, false);
1099         s->must_remove_block_options = false;
1100     }
1101 }
1102 
1103 static void migrate_fd_cleanup(void *opaque)
1104 {
1105     MigrationState *s = opaque;
1106 
1107     qemu_bh_delete(s->cleanup_bh);
1108     s->cleanup_bh = NULL;
1109 
1110     qemu_savevm_state_cleanup();
1111 
1112     if (s->to_dst_file) {
1113         Error *local_err = NULL;
1114 
1115         trace_migrate_fd_cleanup();
1116         qemu_mutex_unlock_iothread();
1117         if (s->migration_thread_running) {
1118             qemu_thread_join(&s->thread);
1119             s->migration_thread_running = false;
1120         }
1121         qemu_mutex_lock_iothread();
1122 
1123         if (multifd_save_cleanup(&local_err) != 0) {
1124             error_report_err(local_err);
1125         }
1126         qemu_fclose(s->to_dst_file);
1127         s->to_dst_file = NULL;
1128     }
1129 
1130     assert((s->state != MIGRATION_STATUS_ACTIVE) &&
1131            (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
1132 
1133     if (s->state == MIGRATION_STATUS_CANCELLING) {
1134         migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1135                           MIGRATION_STATUS_CANCELLED);
1136     }
1137 
1138     if (s->error) {
1139         /* It is used on info migrate.  We can't free it */
1140         error_report_err(error_copy(s->error));
1141     }
1142     notifier_list_notify(&migration_state_notifiers, s);
1143     block_cleanup_parameters(s);
1144 }
1145 
1146 void migrate_set_error(MigrationState *s, const Error *error)
1147 {
1148     qemu_mutex_lock(&s->error_mutex);
1149     if (!s->error) {
1150         s->error = error_copy(error);
1151     }
1152     qemu_mutex_unlock(&s->error_mutex);
1153 }
1154 
1155 void migrate_fd_error(MigrationState *s, const Error *error)
1156 {
1157     trace_migrate_fd_error(error_get_pretty(error));
1158     assert(s->to_dst_file == NULL);
1159     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1160                       MIGRATION_STATUS_FAILED);
1161     migrate_set_error(s, error);
1162 }
1163 
1164 static void migrate_fd_cancel(MigrationState *s)
1165 {
1166     int old_state ;
1167     QEMUFile *f = migrate_get_current()->to_dst_file;
1168     trace_migrate_fd_cancel();
1169 
1170     if (s->rp_state.from_dst_file) {
1171         /* shutdown the rp socket, so causing the rp thread to shutdown */
1172         qemu_file_shutdown(s->rp_state.from_dst_file);
1173     }
1174 
1175     do {
1176         old_state = s->state;
1177         if (!migration_is_setup_or_active(old_state)) {
1178             break;
1179         }
1180         /* If the migration is paused, kick it out of the pause */
1181         if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1182             qemu_sem_post(&s->pause_sem);
1183         }
1184         migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1185     } while (s->state != MIGRATION_STATUS_CANCELLING);
1186 
1187     /*
1188      * If we're unlucky the migration code might be stuck somewhere in a
1189      * send/write while the network has failed and is waiting to timeout;
1190      * if we've got shutdown(2) available then we can force it to quit.
1191      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1192      * called in a bh, so there is no race against this cancel.
1193      */
1194     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1195         qemu_file_shutdown(f);
1196     }
1197     if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1198         Error *local_err = NULL;
1199 
1200         bdrv_invalidate_cache_all(&local_err);
1201         if (local_err) {
1202             error_report_err(local_err);
1203         } else {
1204             s->block_inactive = false;
1205         }
1206     }
1207 }
1208 
1209 void add_migration_state_change_notifier(Notifier *notify)
1210 {
1211     notifier_list_add(&migration_state_notifiers, notify);
1212 }
1213 
1214 void remove_migration_state_change_notifier(Notifier *notify)
1215 {
1216     notifier_remove(notify);
1217 }
1218 
1219 bool migration_in_setup(MigrationState *s)
1220 {
1221     return s->state == MIGRATION_STATUS_SETUP;
1222 }
1223 
1224 bool migration_has_finished(MigrationState *s)
1225 {
1226     return s->state == MIGRATION_STATUS_COMPLETED;
1227 }
1228 
1229 bool migration_has_failed(MigrationState *s)
1230 {
1231     return (s->state == MIGRATION_STATUS_CANCELLED ||
1232             s->state == MIGRATION_STATUS_FAILED);
1233 }
1234 
1235 bool migration_in_postcopy(void)
1236 {
1237     MigrationState *s = migrate_get_current();
1238 
1239     return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1240 }
1241 
1242 bool migration_in_postcopy_after_devices(MigrationState *s)
1243 {
1244     return migration_in_postcopy() && s->postcopy_after_devices;
1245 }
1246 
1247 bool migration_is_idle(void)
1248 {
1249     MigrationState *s = migrate_get_current();
1250 
1251     switch (s->state) {
1252     case MIGRATION_STATUS_NONE:
1253     case MIGRATION_STATUS_CANCELLED:
1254     case MIGRATION_STATUS_COMPLETED:
1255     case MIGRATION_STATUS_FAILED:
1256         return true;
1257     case MIGRATION_STATUS_SETUP:
1258     case MIGRATION_STATUS_CANCELLING:
1259     case MIGRATION_STATUS_ACTIVE:
1260     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1261     case MIGRATION_STATUS_COLO:
1262     case MIGRATION_STATUS_PRE_SWITCHOVER:
1263     case MIGRATION_STATUS_DEVICE:
1264         return false;
1265     case MIGRATION_STATUS__MAX:
1266         g_assert_not_reached();
1267     }
1268 
1269     return false;
1270 }
1271 
1272 MigrationState *migrate_init(void)
1273 {
1274     MigrationState *s = migrate_get_current();
1275 
1276     /*
1277      * Reinitialise all migration state, except
1278      * parameters/capabilities that the user set, and
1279      * locks.
1280      */
1281     s->bytes_xfer = 0;
1282     s->xfer_limit = 0;
1283     s->cleanup_bh = 0;
1284     s->to_dst_file = NULL;
1285     s->state = MIGRATION_STATUS_NONE;
1286     s->rp_state.from_dst_file = NULL;
1287     s->rp_state.error = false;
1288     s->mbps = 0.0;
1289     s->downtime = 0;
1290     s->expected_downtime = 0;
1291     s->setup_time = 0;
1292     s->start_postcopy = false;
1293     s->postcopy_after_devices = false;
1294     s->migration_thread_running = false;
1295     error_free(s->error);
1296     s->error = NULL;
1297 
1298     migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1299 
1300     s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1301     s->total_time = 0;
1302     s->vm_was_running = false;
1303     s->iteration_initial_bytes = 0;
1304     s->threshold_size = 0;
1305     return s;
1306 }
1307 
1308 static GSList *migration_blockers;
1309 
1310 int migrate_add_blocker(Error *reason, Error **errp)
1311 {
1312     if (migrate_get_current()->only_migratable) {
1313         error_propagate(errp, error_copy(reason));
1314         error_prepend(errp, "disallowing migration blocker "
1315                           "(--only_migratable) for: ");
1316         return -EACCES;
1317     }
1318 
1319     if (migration_is_idle()) {
1320         migration_blockers = g_slist_prepend(migration_blockers, reason);
1321         return 0;
1322     }
1323 
1324     error_propagate(errp, error_copy(reason));
1325     error_prepend(errp, "disallowing migration blocker (migration in "
1326                       "progress) for: ");
1327     return -EBUSY;
1328 }
1329 
1330 void migrate_del_blocker(Error *reason)
1331 {
1332     migration_blockers = g_slist_remove(migration_blockers, reason);
1333 }
1334 
1335 void qmp_migrate_incoming(const char *uri, Error **errp)
1336 {
1337     Error *local_err = NULL;
1338     static bool once = true;
1339 
1340     if (!deferred_incoming) {
1341         error_setg(errp, "For use with '-incoming defer'");
1342         return;
1343     }
1344     if (!once) {
1345         error_setg(errp, "The incoming migration has already been started");
1346     }
1347 
1348     qemu_start_incoming_migration(uri, &local_err);
1349 
1350     if (local_err) {
1351         error_propagate(errp, local_err);
1352         return;
1353     }
1354 
1355     once = false;
1356 }
1357 
1358 bool migration_is_blocked(Error **errp)
1359 {
1360     if (qemu_savevm_state_blocked(errp)) {
1361         return true;
1362     }
1363 
1364     if (migration_blockers) {
1365         error_propagate(errp, error_copy(migration_blockers->data));
1366         return true;
1367     }
1368 
1369     return false;
1370 }
1371 
1372 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1373                  bool has_inc, bool inc, bool has_detach, bool detach,
1374                  Error **errp)
1375 {
1376     Error *local_err = NULL;
1377     MigrationState *s = migrate_get_current();
1378     const char *p;
1379 
1380     if (migration_is_setup_or_active(s->state) ||
1381         s->state == MIGRATION_STATUS_CANCELLING ||
1382         s->state == MIGRATION_STATUS_COLO) {
1383         error_setg(errp, QERR_MIGRATION_ACTIVE);
1384         return;
1385     }
1386     if (runstate_check(RUN_STATE_INMIGRATE)) {
1387         error_setg(errp, "Guest is waiting for an incoming migration");
1388         return;
1389     }
1390 
1391     if (migration_is_blocked(errp)) {
1392         return;
1393     }
1394 
1395     if ((has_blk && blk) || (has_inc && inc)) {
1396         if (migrate_use_block() || migrate_use_block_incremental()) {
1397             error_setg(errp, "Command options are incompatible with "
1398                        "current migration capabilities");
1399             return;
1400         }
1401         migrate_set_block_enabled(true, &local_err);
1402         if (local_err) {
1403             error_propagate(errp, local_err);
1404             return;
1405         }
1406         s->must_remove_block_options = true;
1407     }
1408 
1409     if (has_inc && inc) {
1410         migrate_set_block_incremental(s, true);
1411     }
1412 
1413     s = migrate_init();
1414 
1415     if (strstart(uri, "tcp:", &p)) {
1416         tcp_start_outgoing_migration(s, p, &local_err);
1417 #ifdef CONFIG_RDMA
1418     } else if (strstart(uri, "rdma:", &p)) {
1419         rdma_start_outgoing_migration(s, p, &local_err);
1420 #endif
1421     } else if (strstart(uri, "exec:", &p)) {
1422         exec_start_outgoing_migration(s, p, &local_err);
1423     } else if (strstart(uri, "unix:", &p)) {
1424         unix_start_outgoing_migration(s, p, &local_err);
1425     } else if (strstart(uri, "fd:", &p)) {
1426         fd_start_outgoing_migration(s, p, &local_err);
1427     } else {
1428         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1429                    "a valid migration protocol");
1430         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1431                           MIGRATION_STATUS_FAILED);
1432         return;
1433     }
1434 
1435     if (local_err) {
1436         migrate_fd_error(s, local_err);
1437         error_propagate(errp, local_err);
1438         return;
1439     }
1440 }
1441 
1442 void qmp_migrate_cancel(Error **errp)
1443 {
1444     migrate_fd_cancel(migrate_get_current());
1445 }
1446 
1447 void qmp_migrate_continue(MigrationStatus state, Error **errp)
1448 {
1449     MigrationState *s = migrate_get_current();
1450     if (s->state != state) {
1451         error_setg(errp,  "Migration not in expected state: %s",
1452                    MigrationStatus_str(s->state));
1453         return;
1454     }
1455     qemu_sem_post(&s->pause_sem);
1456 }
1457 
1458 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1459 {
1460     MigrateSetParameters p = {
1461         .has_xbzrle_cache_size = true,
1462         .xbzrle_cache_size = value,
1463     };
1464 
1465     qmp_migrate_set_parameters(&p, errp);
1466 }
1467 
1468 int64_t qmp_query_migrate_cache_size(Error **errp)
1469 {
1470     return migrate_xbzrle_cache_size();
1471 }
1472 
1473 void qmp_migrate_set_speed(int64_t value, Error **errp)
1474 {
1475     MigrateSetParameters p = {
1476         .has_max_bandwidth = true,
1477         .max_bandwidth = value,
1478     };
1479 
1480     qmp_migrate_set_parameters(&p, errp);
1481 }
1482 
1483 void qmp_migrate_set_downtime(double value, Error **errp)
1484 {
1485     if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1486         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1487                          "the range of 0 to %d seconds",
1488                          MAX_MIGRATE_DOWNTIME_SECONDS);
1489         return;
1490     }
1491 
1492     value *= 1000; /* Convert to milliseconds */
1493     value = MAX(0, MIN(INT64_MAX, value));
1494 
1495     MigrateSetParameters p = {
1496         .has_downtime_limit = true,
1497         .downtime_limit = value,
1498     };
1499 
1500     qmp_migrate_set_parameters(&p, errp);
1501 }
1502 
1503 bool migrate_release_ram(void)
1504 {
1505     MigrationState *s;
1506 
1507     s = migrate_get_current();
1508 
1509     return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1510 }
1511 
1512 bool migrate_postcopy_ram(void)
1513 {
1514     MigrationState *s;
1515 
1516     s = migrate_get_current();
1517 
1518     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1519 }
1520 
1521 bool migrate_postcopy(void)
1522 {
1523     return migrate_postcopy_ram();
1524 }
1525 
1526 bool migrate_auto_converge(void)
1527 {
1528     MigrationState *s;
1529 
1530     s = migrate_get_current();
1531 
1532     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1533 }
1534 
1535 bool migrate_zero_blocks(void)
1536 {
1537     MigrationState *s;
1538 
1539     s = migrate_get_current();
1540 
1541     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1542 }
1543 
1544 bool migrate_postcopy_blocktime(void)
1545 {
1546     MigrationState *s;
1547 
1548     s = migrate_get_current();
1549 
1550     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
1551 }
1552 
1553 bool migrate_use_compression(void)
1554 {
1555     MigrationState *s;
1556 
1557     s = migrate_get_current();
1558 
1559     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
1560 }
1561 
1562 int migrate_compress_level(void)
1563 {
1564     MigrationState *s;
1565 
1566     s = migrate_get_current();
1567 
1568     return s->parameters.compress_level;
1569 }
1570 
1571 int migrate_compress_threads(void)
1572 {
1573     MigrationState *s;
1574 
1575     s = migrate_get_current();
1576 
1577     return s->parameters.compress_threads;
1578 }
1579 
1580 int migrate_decompress_threads(void)
1581 {
1582     MigrationState *s;
1583 
1584     s = migrate_get_current();
1585 
1586     return s->parameters.decompress_threads;
1587 }
1588 
1589 bool migrate_use_events(void)
1590 {
1591     MigrationState *s;
1592 
1593     s = migrate_get_current();
1594 
1595     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1596 }
1597 
1598 bool migrate_use_multifd(void)
1599 {
1600     MigrationState *s;
1601 
1602     s = migrate_get_current();
1603 
1604     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_MULTIFD];
1605 }
1606 
1607 bool migrate_pause_before_switchover(void)
1608 {
1609     MigrationState *s;
1610 
1611     s = migrate_get_current();
1612 
1613     return s->enabled_capabilities[
1614         MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
1615 }
1616 
1617 int migrate_multifd_channels(void)
1618 {
1619     MigrationState *s;
1620 
1621     s = migrate_get_current();
1622 
1623     return s->parameters.x_multifd_channels;
1624 }
1625 
1626 int migrate_multifd_page_count(void)
1627 {
1628     MigrationState *s;
1629 
1630     s = migrate_get_current();
1631 
1632     return s->parameters.x_multifd_page_count;
1633 }
1634 
1635 int migrate_use_xbzrle(void)
1636 {
1637     MigrationState *s;
1638 
1639     s = migrate_get_current();
1640 
1641     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1642 }
1643 
1644 int64_t migrate_xbzrle_cache_size(void)
1645 {
1646     MigrationState *s;
1647 
1648     s = migrate_get_current();
1649 
1650     return s->parameters.xbzrle_cache_size;
1651 }
1652 
1653 bool migrate_use_block(void)
1654 {
1655     MigrationState *s;
1656 
1657     s = migrate_get_current();
1658 
1659     return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
1660 }
1661 
1662 bool migrate_use_return_path(void)
1663 {
1664     MigrationState *s;
1665 
1666     s = migrate_get_current();
1667 
1668     return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
1669 }
1670 
1671 bool migrate_use_block_incremental(void)
1672 {
1673     MigrationState *s;
1674 
1675     s = migrate_get_current();
1676 
1677     return s->parameters.block_incremental;
1678 }
1679 
1680 /* migration thread support */
1681 /*
1682  * Something bad happened to the RP stream, mark an error
1683  * The caller shall print or trace something to indicate why
1684  */
1685 static void mark_source_rp_bad(MigrationState *s)
1686 {
1687     s->rp_state.error = true;
1688 }
1689 
1690 static struct rp_cmd_args {
1691     ssize_t     len; /* -1 = variable */
1692     const char *name;
1693 } rp_cmd_args[] = {
1694     [MIG_RP_MSG_INVALID]        = { .len = -1, .name = "INVALID" },
1695     [MIG_RP_MSG_SHUT]           = { .len =  4, .name = "SHUT" },
1696     [MIG_RP_MSG_PONG]           = { .len =  4, .name = "PONG" },
1697     [MIG_RP_MSG_REQ_PAGES]      = { .len = 12, .name = "REQ_PAGES" },
1698     [MIG_RP_MSG_REQ_PAGES_ID]   = { .len = -1, .name = "REQ_PAGES_ID" },
1699     [MIG_RP_MSG_MAX]            = { .len = -1, .name = "MAX" },
1700 };
1701 
1702 /*
1703  * Process a request for pages received on the return path,
1704  * We're allowed to send more than requested (e.g. to round to our page size)
1705  * and we don't need to send pages that have already been sent.
1706  */
1707 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1708                                        ram_addr_t start, size_t len)
1709 {
1710     long our_host_ps = getpagesize();
1711 
1712     trace_migrate_handle_rp_req_pages(rbname, start, len);
1713 
1714     /*
1715      * Since we currently insist on matching page sizes, just sanity check
1716      * we're being asked for whole host pages.
1717      */
1718     if (start & (our_host_ps-1) ||
1719        (len & (our_host_ps-1))) {
1720         error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1721                      " len: %zd", __func__, start, len);
1722         mark_source_rp_bad(ms);
1723         return;
1724     }
1725 
1726     if (ram_save_queue_pages(rbname, start, len)) {
1727         mark_source_rp_bad(ms);
1728     }
1729 }
1730 
1731 /*
1732  * Handles messages sent on the return path towards the source VM
1733  *
1734  */
1735 static void *source_return_path_thread(void *opaque)
1736 {
1737     MigrationState *ms = opaque;
1738     QEMUFile *rp = ms->rp_state.from_dst_file;
1739     uint16_t header_len, header_type;
1740     uint8_t buf[512];
1741     uint32_t tmp32, sibling_error;
1742     ram_addr_t start = 0; /* =0 to silence warning */
1743     size_t  len = 0, expected_len;
1744     int res;
1745 
1746     trace_source_return_path_thread_entry();
1747     while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
1748            migration_is_setup_or_active(ms->state)) {
1749         trace_source_return_path_thread_loop_top();
1750         header_type = qemu_get_be16(rp);
1751         header_len = qemu_get_be16(rp);
1752 
1753         if (header_type >= MIG_RP_MSG_MAX ||
1754             header_type == MIG_RP_MSG_INVALID) {
1755             error_report("RP: Received invalid message 0x%04x length 0x%04x",
1756                     header_type, header_len);
1757             mark_source_rp_bad(ms);
1758             goto out;
1759         }
1760 
1761         if ((rp_cmd_args[header_type].len != -1 &&
1762             header_len != rp_cmd_args[header_type].len) ||
1763             header_len > sizeof(buf)) {
1764             error_report("RP: Received '%s' message (0x%04x) with"
1765                     "incorrect length %d expecting %zu",
1766                     rp_cmd_args[header_type].name, header_type, header_len,
1767                     (size_t)rp_cmd_args[header_type].len);
1768             mark_source_rp_bad(ms);
1769             goto out;
1770         }
1771 
1772         /* We know we've got a valid header by this point */
1773         res = qemu_get_buffer(rp, buf, header_len);
1774         if (res != header_len) {
1775             error_report("RP: Failed reading data for message 0x%04x"
1776                          " read %d expected %d",
1777                          header_type, res, header_len);
1778             mark_source_rp_bad(ms);
1779             goto out;
1780         }
1781 
1782         /* OK, we have the message and the data */
1783         switch (header_type) {
1784         case MIG_RP_MSG_SHUT:
1785             sibling_error = ldl_be_p(buf);
1786             trace_source_return_path_thread_shut(sibling_error);
1787             if (sibling_error) {
1788                 error_report("RP: Sibling indicated error %d", sibling_error);
1789                 mark_source_rp_bad(ms);
1790             }
1791             /*
1792              * We'll let the main thread deal with closing the RP
1793              * we could do a shutdown(2) on it, but we're the only user
1794              * anyway, so there's nothing gained.
1795              */
1796             goto out;
1797 
1798         case MIG_RP_MSG_PONG:
1799             tmp32 = ldl_be_p(buf);
1800             trace_source_return_path_thread_pong(tmp32);
1801             break;
1802 
1803         case MIG_RP_MSG_REQ_PAGES:
1804             start = ldq_be_p(buf);
1805             len = ldl_be_p(buf + 8);
1806             migrate_handle_rp_req_pages(ms, NULL, start, len);
1807             break;
1808 
1809         case MIG_RP_MSG_REQ_PAGES_ID:
1810             expected_len = 12 + 1; /* header + termination */
1811 
1812             if (header_len >= expected_len) {
1813                 start = ldq_be_p(buf);
1814                 len = ldl_be_p(buf + 8);
1815                 /* Now we expect an idstr */
1816                 tmp32 = buf[12]; /* Length of the following idstr */
1817                 buf[13 + tmp32] = '\0';
1818                 expected_len += tmp32;
1819             }
1820             if (header_len != expected_len) {
1821                 error_report("RP: Req_Page_id with length %d expecting %zd",
1822                         header_len, expected_len);
1823                 mark_source_rp_bad(ms);
1824                 goto out;
1825             }
1826             migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
1827             break;
1828 
1829         default:
1830             break;
1831         }
1832     }
1833     if (qemu_file_get_error(rp)) {
1834         trace_source_return_path_thread_bad_end();
1835         mark_source_rp_bad(ms);
1836     }
1837 
1838     trace_source_return_path_thread_end();
1839 out:
1840     ms->rp_state.from_dst_file = NULL;
1841     qemu_fclose(rp);
1842     return NULL;
1843 }
1844 
1845 static int open_return_path_on_source(MigrationState *ms)
1846 {
1847 
1848     ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
1849     if (!ms->rp_state.from_dst_file) {
1850         return -1;
1851     }
1852 
1853     trace_open_return_path_on_source();
1854     qemu_thread_create(&ms->rp_state.rp_thread, "return path",
1855                        source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
1856 
1857     trace_open_return_path_on_source_continue();
1858 
1859     return 0;
1860 }
1861 
1862 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1863 static int await_return_path_close_on_source(MigrationState *ms)
1864 {
1865     /*
1866      * If this is a normal exit then the destination will send a SHUT and the
1867      * rp_thread will exit, however if there's an error we need to cause
1868      * it to exit.
1869      */
1870     if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
1871         /*
1872          * shutdown(2), if we have it, will cause it to unblock if it's stuck
1873          * waiting for the destination.
1874          */
1875         qemu_file_shutdown(ms->rp_state.from_dst_file);
1876         mark_source_rp_bad(ms);
1877     }
1878     trace_await_return_path_close_on_source_joining();
1879     qemu_thread_join(&ms->rp_state.rp_thread);
1880     trace_await_return_path_close_on_source_close();
1881     return ms->rp_state.error;
1882 }
1883 
1884 /*
1885  * Switch from normal iteration to postcopy
1886  * Returns non-0 on error
1887  */
1888 static int postcopy_start(MigrationState *ms)
1889 {
1890     int ret;
1891     QIOChannelBuffer *bioc;
1892     QEMUFile *fb;
1893     int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1894     bool restart_block = false;
1895     int cur_state = MIGRATION_STATUS_ACTIVE;
1896     if (!migrate_pause_before_switchover()) {
1897         migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
1898                           MIGRATION_STATUS_POSTCOPY_ACTIVE);
1899     }
1900 
1901     trace_postcopy_start();
1902     qemu_mutex_lock_iothread();
1903     trace_postcopy_start_set_run();
1904 
1905     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1906     global_state_store();
1907     ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1908     if (ret < 0) {
1909         goto fail;
1910     }
1911 
1912     ret = migration_maybe_pause(ms, &cur_state,
1913                                 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1914     if (ret < 0) {
1915         goto fail;
1916     }
1917 
1918     ret = bdrv_inactivate_all();
1919     if (ret < 0) {
1920         goto fail;
1921     }
1922     restart_block = true;
1923 
1924     /*
1925      * Cause any non-postcopiable, but iterative devices to
1926      * send out their final data.
1927      */
1928     qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
1929 
1930     /*
1931      * in Finish migrate and with the io-lock held everything should
1932      * be quiet, but we've potentially still got dirty pages and we
1933      * need to tell the destination to throw any pages it's already received
1934      * that are dirty
1935      */
1936     if (migrate_postcopy_ram()) {
1937         if (ram_postcopy_send_discard_bitmap(ms)) {
1938             error_report("postcopy send discard bitmap failed");
1939             goto fail;
1940         }
1941     }
1942 
1943     /*
1944      * send rest of state - note things that are doing postcopy
1945      * will notice we're in POSTCOPY_ACTIVE and not actually
1946      * wrap their state up here
1947      */
1948     qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
1949     if (migrate_postcopy_ram()) {
1950         /* Ping just for debugging, helps line traces up */
1951         qemu_savevm_send_ping(ms->to_dst_file, 2);
1952     }
1953 
1954     /*
1955      * While loading the device state we may trigger page transfer
1956      * requests and the fd must be free to process those, and thus
1957      * the destination must read the whole device state off the fd before
1958      * it starts processing it.  Unfortunately the ad-hoc migration format
1959      * doesn't allow the destination to know the size to read without fully
1960      * parsing it through each devices load-state code (especially the open
1961      * coded devices that use get/put).
1962      * So we wrap the device state up in a package with a length at the start;
1963      * to do this we use a qemu_buf to hold the whole of the device state.
1964      */
1965     bioc = qio_channel_buffer_new(4096);
1966     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
1967     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
1968     object_unref(OBJECT(bioc));
1969 
1970     /*
1971      * Make sure the receiver can get incoming pages before we send the rest
1972      * of the state
1973      */
1974     qemu_savevm_send_postcopy_listen(fb);
1975 
1976     qemu_savevm_state_complete_precopy(fb, false, false);
1977     if (migrate_postcopy_ram()) {
1978         qemu_savevm_send_ping(fb, 3);
1979     }
1980 
1981     qemu_savevm_send_postcopy_run(fb);
1982 
1983     /* <><> end of stuff going into the package */
1984 
1985     /* Last point of recovery; as soon as we send the package the destination
1986      * can open devices and potentially start running.
1987      * Lets just check again we've not got any errors.
1988      */
1989     ret = qemu_file_get_error(ms->to_dst_file);
1990     if (ret) {
1991         error_report("postcopy_start: Migration stream errored (pre package)");
1992         goto fail_closefb;
1993     }
1994 
1995     restart_block = false;
1996 
1997     /* Now send that blob */
1998     if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
1999         goto fail_closefb;
2000     }
2001     qemu_fclose(fb);
2002 
2003     /* Send a notify to give a chance for anything that needs to happen
2004      * at the transition to postcopy and after the device state; in particular
2005      * spice needs to trigger a transition now
2006      */
2007     ms->postcopy_after_devices = true;
2008     notifier_list_notify(&migration_state_notifiers, ms);
2009 
2010     ms->downtime =  qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
2011 
2012     qemu_mutex_unlock_iothread();
2013 
2014     if (migrate_postcopy_ram()) {
2015         /*
2016          * Although this ping is just for debug, it could potentially be
2017          * used for getting a better measurement of downtime at the source.
2018          */
2019         qemu_savevm_send_ping(ms->to_dst_file, 4);
2020     }
2021 
2022     if (migrate_release_ram()) {
2023         ram_postcopy_migrated_memory_release(ms);
2024     }
2025 
2026     ret = qemu_file_get_error(ms->to_dst_file);
2027     if (ret) {
2028         error_report("postcopy_start: Migration stream errored");
2029         migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2030                               MIGRATION_STATUS_FAILED);
2031     }
2032 
2033     return ret;
2034 
2035 fail_closefb:
2036     qemu_fclose(fb);
2037 fail:
2038     migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2039                           MIGRATION_STATUS_FAILED);
2040     if (restart_block) {
2041         /* A failure happened early enough that we know the destination hasn't
2042          * accessed block devices, so we're safe to recover.
2043          */
2044         Error *local_err = NULL;
2045 
2046         bdrv_invalidate_cache_all(&local_err);
2047         if (local_err) {
2048             error_report_err(local_err);
2049         }
2050     }
2051     qemu_mutex_unlock_iothread();
2052     return -1;
2053 }
2054 
2055 /**
2056  * migration_maybe_pause: Pause if required to by
2057  * migrate_pause_before_switchover called with the iothread locked
2058  * Returns: 0 on success
2059  */
2060 static int migration_maybe_pause(MigrationState *s,
2061                                  int *current_active_state,
2062                                  int new_state)
2063 {
2064     if (!migrate_pause_before_switchover()) {
2065         return 0;
2066     }
2067 
2068     /* Since leaving this state is not atomic with posting the semaphore
2069      * it's possible that someone could have issued multiple migrate_continue
2070      * and the semaphore is incorrectly positive at this point;
2071      * the docs say it's undefined to reinit a semaphore that's already
2072      * init'd, so use timedwait to eat up any existing posts.
2073      */
2074     while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2075         /* This block intentionally left blank */
2076     }
2077 
2078     qemu_mutex_unlock_iothread();
2079     migrate_set_state(&s->state, *current_active_state,
2080                       MIGRATION_STATUS_PRE_SWITCHOVER);
2081     qemu_sem_wait(&s->pause_sem);
2082     migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
2083                       new_state);
2084     *current_active_state = new_state;
2085     qemu_mutex_lock_iothread();
2086 
2087     return s->state == new_state ? 0 : -EINVAL;
2088 }
2089 
2090 /**
2091  * migration_completion: Used by migration_thread when there's not much left.
2092  *   The caller 'breaks' the loop when this returns.
2093  *
2094  * @s: Current migration state
2095  */
2096 static void migration_completion(MigrationState *s)
2097 {
2098     int ret;
2099     int current_active_state = s->state;
2100 
2101     if (s->state == MIGRATION_STATUS_ACTIVE) {
2102         qemu_mutex_lock_iothread();
2103         s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2104         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
2105         s->vm_was_running = runstate_is_running();
2106         ret = global_state_store();
2107 
2108         if (!ret) {
2109             bool inactivate = !migrate_colo_enabled();
2110             ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2111             if (ret >= 0) {
2112                 ret = migration_maybe_pause(s, &current_active_state,
2113                                             MIGRATION_STATUS_DEVICE);
2114             }
2115             if (ret >= 0) {
2116                 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
2117                 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2118                                                          inactivate);
2119             }
2120             if (inactivate && ret >= 0) {
2121                 s->block_inactive = true;
2122             }
2123         }
2124         qemu_mutex_unlock_iothread();
2125 
2126         if (ret < 0) {
2127             goto fail;
2128         }
2129     } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2130         trace_migration_completion_postcopy_end();
2131 
2132         qemu_savevm_state_complete_postcopy(s->to_dst_file);
2133         trace_migration_completion_postcopy_end_after_complete();
2134     }
2135 
2136     /*
2137      * If rp was opened we must clean up the thread before
2138      * cleaning everything else up (since if there are no failures
2139      * it will wait for the destination to send it's status in
2140      * a SHUT command).
2141      */
2142     if (s->rp_state.from_dst_file) {
2143         int rp_error;
2144         trace_migration_return_path_end_before();
2145         rp_error = await_return_path_close_on_source(s);
2146         trace_migration_return_path_end_after(rp_error);
2147         if (rp_error) {
2148             goto fail_invalidate;
2149         }
2150     }
2151 
2152     if (qemu_file_get_error(s->to_dst_file)) {
2153         trace_migration_completion_file_err();
2154         goto fail_invalidate;
2155     }
2156 
2157     if (!migrate_colo_enabled()) {
2158         migrate_set_state(&s->state, current_active_state,
2159                           MIGRATION_STATUS_COMPLETED);
2160     }
2161 
2162     return;
2163 
2164 fail_invalidate:
2165     /* If not doing postcopy, vm_start() will be called: let's regain
2166      * control on images.
2167      */
2168     if (s->state == MIGRATION_STATUS_ACTIVE) {
2169         Error *local_err = NULL;
2170 
2171         qemu_mutex_lock_iothread();
2172         bdrv_invalidate_cache_all(&local_err);
2173         if (local_err) {
2174             error_report_err(local_err);
2175         } else {
2176             s->block_inactive = false;
2177         }
2178         qemu_mutex_unlock_iothread();
2179     }
2180 
2181 fail:
2182     migrate_set_state(&s->state, current_active_state,
2183                       MIGRATION_STATUS_FAILED);
2184 }
2185 
2186 bool migrate_colo_enabled(void)
2187 {
2188     MigrationState *s = migrate_get_current();
2189     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2190 }
2191 
2192 static void migration_calculate_complete(MigrationState *s)
2193 {
2194     uint64_t bytes = qemu_ftell(s->to_dst_file);
2195     int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2196 
2197     s->total_time = end_time - s->start_time;
2198     if (!s->downtime) {
2199         /*
2200          * It's still not set, so we are precopy migration.  For
2201          * postcopy, downtime is calculated during postcopy_start().
2202          */
2203         s->downtime = end_time - s->downtime_start;
2204     }
2205 
2206     if (s->total_time) {
2207         s->mbps = ((double) bytes * 8.0) / s->total_time / 1000;
2208     }
2209 }
2210 
2211 static void migration_update_counters(MigrationState *s,
2212                                       int64_t current_time)
2213 {
2214     uint64_t transferred, time_spent;
2215     int64_t threshold_size;
2216     double bandwidth;
2217 
2218     if (current_time < s->iteration_start_time + BUFFER_DELAY) {
2219         return;
2220     }
2221 
2222     transferred = qemu_ftell(s->to_dst_file) - s->iteration_initial_bytes;
2223     time_spent = current_time - s->iteration_start_time;
2224     bandwidth = (double)transferred / time_spent;
2225     threshold_size = bandwidth * s->parameters.downtime_limit;
2226 
2227     s->mbps = (((double) transferred * 8.0) /
2228                ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
2229 
2230     /*
2231      * if we haven't sent anything, we don't want to
2232      * recalculate. 10000 is a small enough number for our purposes
2233      */
2234     if (ram_counters.dirty_pages_rate && transferred > 10000) {
2235         s->expected_downtime = ram_counters.dirty_pages_rate *
2236             qemu_target_page_size() / bandwidth;
2237     }
2238 
2239     qemu_file_reset_rate_limit(s->to_dst_file);
2240 
2241     s->iteration_start_time = current_time;
2242     s->iteration_initial_bytes = qemu_ftell(s->to_dst_file);
2243 
2244     trace_migrate_transferred(transferred, time_spent,
2245                               bandwidth, threshold_size);
2246 }
2247 
2248 /* Migration thread iteration status */
2249 typedef enum {
2250     MIG_ITERATE_RESUME,         /* Resume current iteration */
2251     MIG_ITERATE_SKIP,           /* Skip current iteration */
2252     MIG_ITERATE_BREAK,          /* Break the loop */
2253 } MigIterateState;
2254 
2255 /*
2256  * Return true if continue to the next iteration directly, false
2257  * otherwise.
2258  */
2259 static MigIterateState migration_iteration_run(MigrationState *s)
2260 {
2261     uint64_t pending_size, pend_post, pend_nonpost;
2262     bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
2263 
2264     qemu_savevm_state_pending(s->to_dst_file, s->threshold_size,
2265                               &pend_nonpost, &pend_post);
2266     pending_size = pend_nonpost + pend_post;
2267 
2268     trace_migrate_pending(pending_size, s->threshold_size,
2269                           pend_post, pend_nonpost);
2270 
2271     if (pending_size && pending_size >= s->threshold_size) {
2272         /* Still a significant amount to transfer */
2273         if (migrate_postcopy() && !in_postcopy &&
2274             pend_nonpost <= s->threshold_size &&
2275             atomic_read(&s->start_postcopy)) {
2276             if (postcopy_start(s)) {
2277                 error_report("%s: postcopy failed to start", __func__);
2278             }
2279             return MIG_ITERATE_SKIP;
2280         }
2281         /* Just another iteration step */
2282         qemu_savevm_state_iterate(s->to_dst_file,
2283             s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
2284     } else {
2285         trace_migration_thread_low_pending(pending_size);
2286         migration_completion(s);
2287         return MIG_ITERATE_BREAK;
2288     }
2289 
2290     return MIG_ITERATE_RESUME;
2291 }
2292 
2293 static void migration_iteration_finish(MigrationState *s)
2294 {
2295     /* If we enabled cpu throttling for auto-converge, turn it off. */
2296     cpu_throttle_stop();
2297 
2298     qemu_mutex_lock_iothread();
2299     switch (s->state) {
2300     case MIGRATION_STATUS_COMPLETED:
2301         migration_calculate_complete(s);
2302         runstate_set(RUN_STATE_POSTMIGRATE);
2303         break;
2304 
2305     case MIGRATION_STATUS_ACTIVE:
2306         /*
2307          * We should really assert here, but since it's during
2308          * migration, let's try to reduce the usage of assertions.
2309          */
2310         if (!migrate_colo_enabled()) {
2311             error_report("%s: critical error: calling COLO code without "
2312                          "COLO enabled", __func__);
2313         }
2314         migrate_start_colo_process(s);
2315         /*
2316          * Fixme: we will run VM in COLO no matter its old running state.
2317          * After exited COLO, we will keep running.
2318          */
2319         s->vm_was_running = true;
2320         /* Fallthrough */
2321     case MIGRATION_STATUS_FAILED:
2322     case MIGRATION_STATUS_CANCELLED:
2323         if (s->vm_was_running) {
2324             vm_start();
2325         } else {
2326             if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2327                 runstate_set(RUN_STATE_POSTMIGRATE);
2328             }
2329         }
2330         break;
2331 
2332     default:
2333         /* Should not reach here, but if so, forgive the VM. */
2334         error_report("%s: Unknown ending state %d", __func__, s->state);
2335         break;
2336     }
2337     qemu_bh_schedule(s->cleanup_bh);
2338     qemu_mutex_unlock_iothread();
2339 }
2340 
2341 /*
2342  * Master migration thread on the source VM.
2343  * It drives the migration and pumps the data down the outgoing channel.
2344  */
2345 static void *migration_thread(void *opaque)
2346 {
2347     MigrationState *s = opaque;
2348     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
2349 
2350     rcu_register_thread();
2351 
2352     s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2353 
2354     qemu_savevm_state_header(s->to_dst_file);
2355 
2356     /*
2357      * If we opened the return path, we need to make sure dst has it
2358      * opened as well.
2359      */
2360     if (s->rp_state.from_dst_file) {
2361         /* Now tell the dest that it should open its end so it can reply */
2362         qemu_savevm_send_open_return_path(s->to_dst_file);
2363 
2364         /* And do a ping that will make stuff easier to debug */
2365         qemu_savevm_send_ping(s->to_dst_file, 1);
2366     }
2367 
2368     if (migrate_postcopy()) {
2369         /*
2370          * Tell the destination that we *might* want to do postcopy later;
2371          * if the other end can't do postcopy it should fail now, nice and
2372          * early.
2373          */
2374         qemu_savevm_send_postcopy_advise(s->to_dst_file);
2375     }
2376 
2377     qemu_savevm_state_setup(s->to_dst_file);
2378 
2379     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
2380     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2381                       MIGRATION_STATUS_ACTIVE);
2382 
2383     trace_migration_thread_setup_complete();
2384 
2385     while (s->state == MIGRATION_STATUS_ACTIVE ||
2386            s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2387         int64_t current_time;
2388 
2389         if (!qemu_file_rate_limit(s->to_dst_file)) {
2390             MigIterateState iter_state = migration_iteration_run(s);
2391             if (iter_state == MIG_ITERATE_SKIP) {
2392                 continue;
2393             } else if (iter_state == MIG_ITERATE_BREAK) {
2394                 break;
2395             }
2396         }
2397 
2398         if (qemu_file_get_error(s->to_dst_file)) {
2399             if (migration_is_setup_or_active(s->state)) {
2400                 migrate_set_state(&s->state, s->state,
2401                                   MIGRATION_STATUS_FAILED);
2402             }
2403             trace_migration_thread_file_err();
2404             break;
2405         }
2406 
2407         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2408 
2409         migration_update_counters(s, current_time);
2410 
2411         if (qemu_file_rate_limit(s->to_dst_file)) {
2412             /* usleep expects microseconds */
2413             g_usleep((s->iteration_start_time + BUFFER_DELAY -
2414                       current_time) * 1000);
2415         }
2416     }
2417 
2418     trace_migration_thread_after_loop();
2419     migration_iteration_finish(s);
2420     rcu_unregister_thread();
2421     return NULL;
2422 }
2423 
2424 void migrate_fd_connect(MigrationState *s)
2425 {
2426     s->expected_downtime = s->parameters.downtime_limit;
2427     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
2428 
2429     qemu_file_set_blocking(s->to_dst_file, true);
2430     qemu_file_set_rate_limit(s->to_dst_file,
2431                              s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
2432 
2433     /* Notify before starting migration thread */
2434     notifier_list_notify(&migration_state_notifiers, s);
2435 
2436     /*
2437      * Open the return path. For postcopy, it is used exclusively. For
2438      * precopy, only if user specified "return-path" capability would
2439      * QEMU uses the return path.
2440      */
2441     if (migrate_postcopy_ram() || migrate_use_return_path()) {
2442         if (open_return_path_on_source(s)) {
2443             error_report("Unable to open return-path for postcopy");
2444             migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2445                               MIGRATION_STATUS_FAILED);
2446             migrate_fd_cleanup(s);
2447             return;
2448         }
2449     }
2450 
2451     if (multifd_save_setup() != 0) {
2452         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2453                           MIGRATION_STATUS_FAILED);
2454         migrate_fd_cleanup(s);
2455         return;
2456     }
2457     qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
2458                        QEMU_THREAD_JOINABLE);
2459     s->migration_thread_running = true;
2460 }
2461 
2462 void migration_global_dump(Monitor *mon)
2463 {
2464     MigrationState *ms = migrate_get_current();
2465 
2466     monitor_printf(mon, "globals:\n");
2467     monitor_printf(mon, "store-global-state: %s\n",
2468                    ms->store_global_state ? "on" : "off");
2469     monitor_printf(mon, "only-migratable: %s\n",
2470                    ms->only_migratable ? "on" : "off");
2471     monitor_printf(mon, "send-configuration: %s\n",
2472                    ms->send_configuration ? "on" : "off");
2473     monitor_printf(mon, "send-section-footer: %s\n",
2474                    ms->send_section_footer ? "on" : "off");
2475 }
2476 
2477 #define DEFINE_PROP_MIG_CAP(name, x)             \
2478     DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
2479 
2480 static Property migration_properties[] = {
2481     DEFINE_PROP_BOOL("store-global-state", MigrationState,
2482                      store_global_state, true),
2483     DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false),
2484     DEFINE_PROP_BOOL("send-configuration", MigrationState,
2485                      send_configuration, true),
2486     DEFINE_PROP_BOOL("send-section-footer", MigrationState,
2487                      send_section_footer, true),
2488 
2489     /* Migration parameters */
2490     DEFINE_PROP_UINT8("x-compress-level", MigrationState,
2491                       parameters.compress_level,
2492                       DEFAULT_MIGRATE_COMPRESS_LEVEL),
2493     DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
2494                       parameters.compress_threads,
2495                       DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
2496     DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
2497                       parameters.decompress_threads,
2498                       DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
2499     DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
2500                       parameters.cpu_throttle_initial,
2501                       DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
2502     DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
2503                       parameters.cpu_throttle_increment,
2504                       DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
2505     DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
2506                       parameters.max_bandwidth, MAX_THROTTLE),
2507     DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
2508                       parameters.downtime_limit,
2509                       DEFAULT_MIGRATE_SET_DOWNTIME),
2510     DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
2511                       parameters.x_checkpoint_delay,
2512                       DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
2513     DEFINE_PROP_UINT8("x-multifd-channels", MigrationState,
2514                       parameters.x_multifd_channels,
2515                       DEFAULT_MIGRATE_MULTIFD_CHANNELS),
2516     DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState,
2517                       parameters.x_multifd_page_count,
2518                       DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT),
2519     DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
2520                       parameters.xbzrle_cache_size,
2521                       DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
2522 
2523     /* Migration capabilities */
2524     DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
2525     DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
2526     DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
2527     DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
2528     DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
2529     DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
2530     DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
2531     DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
2532     DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
2533     DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
2534     DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
2535     DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD),
2536 
2537     DEFINE_PROP_END_OF_LIST(),
2538 };
2539 
2540 static void migration_class_init(ObjectClass *klass, void *data)
2541 {
2542     DeviceClass *dc = DEVICE_CLASS(klass);
2543 
2544     dc->user_creatable = false;
2545     dc->props = migration_properties;
2546 }
2547 
2548 static void migration_instance_finalize(Object *obj)
2549 {
2550     MigrationState *ms = MIGRATION_OBJ(obj);
2551     MigrationParameters *params = &ms->parameters;
2552 
2553     qemu_mutex_destroy(&ms->error_mutex);
2554     g_free(params->tls_hostname);
2555     g_free(params->tls_creds);
2556     qemu_sem_destroy(&ms->pause_sem);
2557 }
2558 
2559 static void migration_instance_init(Object *obj)
2560 {
2561     MigrationState *ms = MIGRATION_OBJ(obj);
2562     MigrationParameters *params = &ms->parameters;
2563 
2564     ms->state = MIGRATION_STATUS_NONE;
2565     ms->mbps = -1;
2566     qemu_sem_init(&ms->pause_sem, 0);
2567     qemu_mutex_init(&ms->error_mutex);
2568 
2569     params->tls_hostname = g_strdup("");
2570     params->tls_creds = g_strdup("");
2571 
2572     /* Set has_* up only for parameter checks */
2573     params->has_compress_level = true;
2574     params->has_compress_threads = true;
2575     params->has_decompress_threads = true;
2576     params->has_cpu_throttle_initial = true;
2577     params->has_cpu_throttle_increment = true;
2578     params->has_max_bandwidth = true;
2579     params->has_downtime_limit = true;
2580     params->has_x_checkpoint_delay = true;
2581     params->has_block_incremental = true;
2582     params->has_x_multifd_channels = true;
2583     params->has_x_multifd_page_count = true;
2584     params->has_xbzrle_cache_size = true;
2585 }
2586 
2587 /*
2588  * Return true if check pass, false otherwise. Error will be put
2589  * inside errp if provided.
2590  */
2591 static bool migration_object_check(MigrationState *ms, Error **errp)
2592 {
2593     MigrationCapabilityStatusList *head = NULL;
2594     /* Assuming all off */
2595     bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
2596     int i;
2597 
2598     if (!migrate_params_check(&ms->parameters, errp)) {
2599         return false;
2600     }
2601 
2602     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
2603         if (ms->enabled_capabilities[i]) {
2604             head = migrate_cap_add(head, i, true);
2605         }
2606     }
2607 
2608     ret = migrate_caps_check(cap_list, head, errp);
2609 
2610     /* It works with head == NULL */
2611     qapi_free_MigrationCapabilityStatusList(head);
2612 
2613     return ret;
2614 }
2615 
2616 static const TypeInfo migration_type = {
2617     .name = TYPE_MIGRATION,
2618     /*
2619      * NOTE: TYPE_MIGRATION is not really a device, as the object is
2620      * not created using qdev_create(), it is not attached to the qdev
2621      * device tree, and it is never realized.
2622      *
2623      * TODO: Make this TYPE_OBJECT once QOM provides something like
2624      * TYPE_DEVICE's "-global" properties.
2625      */
2626     .parent = TYPE_DEVICE,
2627     .class_init = migration_class_init,
2628     .class_size = sizeof(MigrationClass),
2629     .instance_size = sizeof(MigrationState),
2630     .instance_init = migration_instance_init,
2631     .instance_finalize = migration_instance_finalize,
2632 };
2633 
2634 static void register_migration_types(void)
2635 {
2636     type_register_static(&migration_type);
2637 }
2638 
2639 type_init(register_migration_types);
2640