xref: /openbmc/qemu/migration/migration.c (revision 5d721b78)
1 /*
2  * QEMU live migration
3  *
4  * Copyright IBM, Corp. 2008
5  *
6  * Authors:
7  *  Anthony Liguori   <aliguori@us.ibm.com>
8  *
9  * This work is licensed under the terms of the GNU GPL, version 2.  See
10  * the COPYING file in the top-level directory.
11  *
12  * Contributions after 2012-01-13 are licensed under the terms of the
13  * GNU GPL, version 2 or (at your option) any later version.
14  */
15 
16 #include "qemu/osdep.h"
17 #include "qemu/cutils.h"
18 #include "qemu/error-report.h"
19 #include "migration/blocker.h"
20 #include "exec.h"
21 #include "fd.h"
22 #include "socket.h"
23 #include "rdma.h"
24 #include "ram.h"
25 #include "migration/global_state.h"
26 #include "migration/misc.h"
27 #include "migration.h"
28 #include "savevm.h"
29 #include "qemu-file-channel.h"
30 #include "qemu-file.h"
31 #include "migration/vmstate.h"
32 #include "block/block.h"
33 #include "qapi/qmp/qerror.h"
34 #include "qapi/util.h"
35 #include "qemu/rcu.h"
36 #include "block.h"
37 #include "postcopy-ram.h"
38 #include "qemu/thread.h"
39 #include "qmp-commands.h"
40 #include "trace.h"
41 #include "qapi-event.h"
42 #include "exec/target_page.h"
43 #include "io/channel-buffer.h"
44 #include "migration/colo.h"
45 #include "hw/boards.h"
46 #include "monitor/monitor.h"
47 
48 #define MAX_THROTTLE  (32 << 20)      /* Migration transfer speed throttling */
49 
50 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
51  * data. */
52 #define BUFFER_DELAY     100
53 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
54 
55 /* Time in milliseconds we are allowed to stop the source,
56  * for sending the last part */
57 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
58 
59 /* Maximum migrate downtime set to 2000 seconds */
60 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
61 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
62 
63 /* Default compression thread count */
64 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
65 /* Default decompression thread count, usually decompression is at
66  * least 4 times as fast as compression.*/
67 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
68 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
69 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
70 /* Define default autoconverge cpu throttle migration parameters */
71 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
72 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
73 
74 /* Migration XBZRLE default cache size */
75 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
76 
77 /* The delay time (in ms) between two COLO checkpoints
78  * Note: Please change this default value to 10000 when we support hybrid mode.
79  */
80 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
81 
82 static NotifierList migration_state_notifiers =
83     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
84 
85 static bool deferred_incoming;
86 
87 /* Messages sent on the return path from destination to source */
88 enum mig_rp_message_type {
89     MIG_RP_MSG_INVALID = 0,  /* Must be 0 */
90     MIG_RP_MSG_SHUT,         /* sibling will not send any more RP messages */
91     MIG_RP_MSG_PONG,         /* Response to a PING; data (seq: be32 ) */
92 
93     MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
94     MIG_RP_MSG_REQ_PAGES,    /* data (start: be64, len: be32) */
95 
96     MIG_RP_MSG_MAX
97 };
98 
99 /* When we add fault tolerance, we could have several
100    migrations at once.  For now we don't need to add
101    dynamic creation of migration */
102 
103 static MigrationState *current_migration;
104 
105 void migration_object_init(void)
106 {
107     MachineState *ms = MACHINE(qdev_get_machine());
108 
109     /* This can only be called once. */
110     assert(!current_migration);
111     current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
112 
113     /*
114      * We cannot really do this in migration_instance_init() since at
115      * that time global properties are not yet applied, then this
116      * value will be definitely replaced by something else.
117      */
118     if (ms->enforce_config_section) {
119         current_migration->send_configuration = true;
120     }
121 }
122 
123 /* For outgoing */
124 MigrationState *migrate_get_current(void)
125 {
126     /* This can only be called after the object created. */
127     assert(current_migration);
128     return current_migration;
129 }
130 
131 MigrationIncomingState *migration_incoming_get_current(void)
132 {
133     static bool once;
134     static MigrationIncomingState mis_current;
135 
136     if (!once) {
137         mis_current.state = MIGRATION_STATUS_NONE;
138         memset(&mis_current, 0, sizeof(MigrationIncomingState));
139         qemu_mutex_init(&mis_current.rp_mutex);
140         qemu_event_init(&mis_current.main_thread_load_event, false);
141         once = true;
142     }
143     return &mis_current;
144 }
145 
146 void migration_incoming_state_destroy(void)
147 {
148     struct MigrationIncomingState *mis = migration_incoming_get_current();
149 
150     if (mis->to_src_file) {
151         /* Tell source that we are done */
152         migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
153         qemu_fclose(mis->to_src_file);
154         mis->to_src_file = NULL;
155     }
156 
157     if (mis->from_src_file) {
158         qemu_fclose(mis->from_src_file);
159         mis->from_src_file = NULL;
160     }
161 
162     qemu_event_destroy(&mis->main_thread_load_event);
163 }
164 
165 static void migrate_generate_event(int new_state)
166 {
167     if (migrate_use_events()) {
168         qapi_event_send_migration(new_state, &error_abort);
169     }
170 }
171 
172 /*
173  * Called on -incoming with a defer: uri.
174  * The migration can be started later after any parameters have been
175  * changed.
176  */
177 static void deferred_incoming_migration(Error **errp)
178 {
179     if (deferred_incoming) {
180         error_setg(errp, "Incoming migration already deferred");
181     }
182     deferred_incoming = true;
183 }
184 
185 /*
186  * Send a message on the return channel back to the source
187  * of the migration.
188  */
189 static void migrate_send_rp_message(MigrationIncomingState *mis,
190                                     enum mig_rp_message_type message_type,
191                                     uint16_t len, void *data)
192 {
193     trace_migrate_send_rp_message((int)message_type, len);
194     qemu_mutex_lock(&mis->rp_mutex);
195     qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
196     qemu_put_be16(mis->to_src_file, len);
197     qemu_put_buffer(mis->to_src_file, data, len);
198     qemu_fflush(mis->to_src_file);
199     qemu_mutex_unlock(&mis->rp_mutex);
200 }
201 
202 /* Request a range of pages from the source VM at the given
203  * start address.
204  *   rbname: Name of the RAMBlock to request the page in, if NULL it's the same
205  *           as the last request (a name must have been given previously)
206  *   Start: Address offset within the RB
207  *   Len: Length in bytes required - must be a multiple of pagesize
208  */
209 void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
210                                ram_addr_t start, size_t len)
211 {
212     uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
213     size_t msglen = 12; /* start + len */
214 
215     *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
216     *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
217 
218     if (rbname) {
219         int rbname_len = strlen(rbname);
220         assert(rbname_len < 256);
221 
222         bufc[msglen++] = rbname_len;
223         memcpy(bufc + msglen, rbname, rbname_len);
224         msglen += rbname_len;
225         migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc);
226     } else {
227         migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc);
228     }
229 }
230 
231 void qemu_start_incoming_migration(const char *uri, Error **errp)
232 {
233     const char *p;
234 
235     qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
236     if (!strcmp(uri, "defer")) {
237         deferred_incoming_migration(errp);
238     } else if (strstart(uri, "tcp:", &p)) {
239         tcp_start_incoming_migration(p, errp);
240 #ifdef CONFIG_RDMA
241     } else if (strstart(uri, "rdma:", &p)) {
242         rdma_start_incoming_migration(p, errp);
243 #endif
244     } else if (strstart(uri, "exec:", &p)) {
245         exec_start_incoming_migration(p, errp);
246     } else if (strstart(uri, "unix:", &p)) {
247         unix_start_incoming_migration(p, errp);
248     } else if (strstart(uri, "fd:", &p)) {
249         fd_start_incoming_migration(p, errp);
250     } else {
251         error_setg(errp, "unknown migration protocol: %s", uri);
252     }
253 }
254 
255 static void process_incoming_migration_bh(void *opaque)
256 {
257     Error *local_err = NULL;
258     MigrationIncomingState *mis = opaque;
259 
260     /* Make sure all file formats flush their mutable metadata.
261      * If we get an error here, just don't restart the VM yet. */
262     bdrv_invalidate_cache_all(&local_err);
263     if (local_err) {
264         error_report_err(local_err);
265         local_err = NULL;
266         autostart = false;
267     }
268 
269     /*
270      * This must happen after all error conditions are dealt with and
271      * we're sure the VM is going to be running on this host.
272      */
273     qemu_announce_self();
274 
275     /* If global state section was not received or we are in running
276        state, we need to obey autostart. Any other state is set with
277        runstate_set. */
278 
279     if (!global_state_received() ||
280         global_state_get_runstate() == RUN_STATE_RUNNING) {
281         if (autostart) {
282             vm_start();
283         } else {
284             runstate_set(RUN_STATE_PAUSED);
285         }
286     } else {
287         runstate_set(global_state_get_runstate());
288     }
289     /*
290      * This must happen after any state changes since as soon as an external
291      * observer sees this event they might start to prod at the VM assuming
292      * it's ready to use.
293      */
294     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
295                       MIGRATION_STATUS_COMPLETED);
296     qemu_bh_delete(mis->bh);
297     migration_incoming_state_destroy();
298 }
299 
300 static void process_incoming_migration_co(void *opaque)
301 {
302     QEMUFile *f = opaque;
303     MigrationIncomingState *mis = migration_incoming_get_current();
304     PostcopyState ps;
305     int ret;
306 
307     mis->from_src_file = f;
308     mis->largest_page_size = qemu_ram_pagesize_largest();
309     postcopy_state_set(POSTCOPY_INCOMING_NONE);
310     migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
311                       MIGRATION_STATUS_ACTIVE);
312     ret = qemu_loadvm_state(f);
313 
314     ps = postcopy_state_get();
315     trace_process_incoming_migration_co_end(ret, ps);
316     if (ps != POSTCOPY_INCOMING_NONE) {
317         if (ps == POSTCOPY_INCOMING_ADVISE) {
318             /*
319              * Where a migration had postcopy enabled (and thus went to advise)
320              * but managed to complete within the precopy period, we can use
321              * the normal exit.
322              */
323             postcopy_ram_incoming_cleanup(mis);
324         } else if (ret >= 0) {
325             /*
326              * Postcopy was started, cleanup should happen at the end of the
327              * postcopy thread.
328              */
329             trace_process_incoming_migration_co_postcopy_end_main();
330             return;
331         }
332         /* Else if something went wrong then just fall out of the normal exit */
333     }
334 
335     /* we get COLO info, and know if we are in COLO mode */
336     if (!ret && migration_incoming_enable_colo()) {
337         mis->migration_incoming_co = qemu_coroutine_self();
338         qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
339              colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
340         mis->have_colo_incoming_thread = true;
341         qemu_coroutine_yield();
342 
343         /* Wait checkpoint incoming thread exit before free resource */
344         qemu_thread_join(&mis->colo_incoming_thread);
345     }
346 
347     if (ret < 0) {
348         migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
349                           MIGRATION_STATUS_FAILED);
350         error_report("load of migration failed: %s", strerror(-ret));
351         exit(EXIT_FAILURE);
352     }
353     mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
354     qemu_bh_schedule(mis->bh);
355 }
356 
357 void migration_fd_process_incoming(QEMUFile *f)
358 {
359     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, f);
360 
361     qemu_file_set_blocking(f, false);
362     qemu_coroutine_enter(co);
363 }
364 
365 /*
366  * Send a 'SHUT' message on the return channel with the given value
367  * to indicate that we've finished with the RP.  Non-0 value indicates
368  * error.
369  */
370 void migrate_send_rp_shut(MigrationIncomingState *mis,
371                           uint32_t value)
372 {
373     uint32_t buf;
374 
375     buf = cpu_to_be32(value);
376     migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
377 }
378 
379 /*
380  * Send a 'PONG' message on the return channel with the given value
381  * (normally in response to a 'PING')
382  */
383 void migrate_send_rp_pong(MigrationIncomingState *mis,
384                           uint32_t value)
385 {
386     uint32_t buf;
387 
388     buf = cpu_to_be32(value);
389     migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
390 }
391 
392 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
393 {
394     MigrationCapabilityStatusList *head = NULL;
395     MigrationCapabilityStatusList *caps;
396     MigrationState *s = migrate_get_current();
397     int i;
398 
399     caps = NULL; /* silence compiler warning */
400     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
401 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
402         if (i == MIGRATION_CAPABILITY_BLOCK) {
403             continue;
404         }
405 #endif
406         if (i == MIGRATION_CAPABILITY_X_COLO && !colo_supported()) {
407             continue;
408         }
409         if (head == NULL) {
410             head = g_malloc0(sizeof(*caps));
411             caps = head;
412         } else {
413             caps->next = g_malloc0(sizeof(*caps));
414             caps = caps->next;
415         }
416         caps->value =
417             g_malloc(sizeof(*caps->value));
418         caps->value->capability = i;
419         caps->value->state = s->enabled_capabilities[i];
420     }
421 
422     return head;
423 }
424 
425 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
426 {
427     MigrationParameters *params;
428     MigrationState *s = migrate_get_current();
429 
430     params = g_malloc0(sizeof(*params));
431     params->has_compress_level = true;
432     params->compress_level = s->parameters.compress_level;
433     params->has_compress_threads = true;
434     params->compress_threads = s->parameters.compress_threads;
435     params->has_decompress_threads = true;
436     params->decompress_threads = s->parameters.decompress_threads;
437     params->has_cpu_throttle_initial = true;
438     params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
439     params->has_cpu_throttle_increment = true;
440     params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
441     params->has_tls_creds = !!s->parameters.tls_creds;
442     params->tls_creds = g_strdup(s->parameters.tls_creds);
443     params->has_tls_hostname = !!s->parameters.tls_hostname;
444     params->tls_hostname = g_strdup(s->parameters.tls_hostname);
445     params->has_max_bandwidth = true;
446     params->max_bandwidth = s->parameters.max_bandwidth;
447     params->has_downtime_limit = true;
448     params->downtime_limit = s->parameters.downtime_limit;
449     params->has_x_checkpoint_delay = true;
450     params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
451     params->has_block_incremental = true;
452     params->block_incremental = s->parameters.block_incremental;
453 
454     return params;
455 }
456 
457 /*
458  * Return true if we're already in the middle of a migration
459  * (i.e. any of the active or setup states)
460  */
461 static bool migration_is_setup_or_active(int state)
462 {
463     switch (state) {
464     case MIGRATION_STATUS_ACTIVE:
465     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
466     case MIGRATION_STATUS_SETUP:
467         return true;
468 
469     default:
470         return false;
471 
472     }
473 }
474 
475 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
476 {
477     info->has_ram = true;
478     info->ram = g_malloc0(sizeof(*info->ram));
479     info->ram->transferred = ram_counters.transferred;
480     info->ram->total = ram_bytes_total();
481     info->ram->duplicate = ram_counters.duplicate;
482     /* legacy value.  It is not used anymore */
483     info->ram->skipped = 0;
484     info->ram->normal = ram_counters.normal;
485     info->ram->normal_bytes = ram_counters.normal *
486         qemu_target_page_size();
487     info->ram->mbps = s->mbps;
488     info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
489     info->ram->postcopy_requests = ram_counters.postcopy_requests;
490     info->ram->page_size = qemu_target_page_size();
491 
492     if (migrate_use_xbzrle()) {
493         info->has_xbzrle_cache = true;
494         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
495         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
496         info->xbzrle_cache->bytes = xbzrle_counters.bytes;
497         info->xbzrle_cache->pages = xbzrle_counters.pages;
498         info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
499         info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
500         info->xbzrle_cache->overflow = xbzrle_counters.overflow;
501     }
502 
503     if (cpu_throttle_active()) {
504         info->has_cpu_throttle_percentage = true;
505         info->cpu_throttle_percentage = cpu_throttle_get_percentage();
506     }
507 
508     if (s->state != MIGRATION_STATUS_COMPLETED) {
509         info->ram->remaining = ram_bytes_remaining();
510         info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
511     }
512 }
513 
514 static void populate_disk_info(MigrationInfo *info)
515 {
516     if (blk_mig_active()) {
517         info->has_disk = true;
518         info->disk = g_malloc0(sizeof(*info->disk));
519         info->disk->transferred = blk_mig_bytes_transferred();
520         info->disk->remaining = blk_mig_bytes_remaining();
521         info->disk->total = blk_mig_bytes_total();
522     }
523 }
524 
525 MigrationInfo *qmp_query_migrate(Error **errp)
526 {
527     MigrationInfo *info = g_malloc0(sizeof(*info));
528     MigrationState *s = migrate_get_current();
529 
530     switch (s->state) {
531     case MIGRATION_STATUS_NONE:
532         /* no migration has happened ever */
533         break;
534     case MIGRATION_STATUS_SETUP:
535         info->has_status = true;
536         info->has_total_time = false;
537         break;
538     case MIGRATION_STATUS_ACTIVE:
539     case MIGRATION_STATUS_CANCELLING:
540     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
541          /* TODO add some postcopy stats */
542         info->has_status = true;
543         info->has_total_time = true;
544         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
545             - s->total_time;
546         info->has_expected_downtime = true;
547         info->expected_downtime = s->expected_downtime;
548         info->has_setup_time = true;
549         info->setup_time = s->setup_time;
550 
551         populate_ram_info(info, s);
552         populate_disk_info(info);
553         break;
554     case MIGRATION_STATUS_COLO:
555         info->has_status = true;
556         /* TODO: display COLO specific information (checkpoint info etc.) */
557         break;
558     case MIGRATION_STATUS_COMPLETED:
559         info->has_status = true;
560         info->has_total_time = true;
561         info->total_time = s->total_time;
562         info->has_downtime = true;
563         info->downtime = s->downtime;
564         info->has_setup_time = true;
565         info->setup_time = s->setup_time;
566 
567         populate_ram_info(info, s);
568         break;
569     case MIGRATION_STATUS_FAILED:
570         info->has_status = true;
571         if (s->error) {
572             info->has_error_desc = true;
573             info->error_desc = g_strdup(error_get_pretty(s->error));
574         }
575         break;
576     case MIGRATION_STATUS_CANCELLED:
577         info->has_status = true;
578         break;
579     }
580     info->status = s->state;
581 
582     return info;
583 }
584 
585 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
586                                   Error **errp)
587 {
588     MigrationState *s = migrate_get_current();
589     MigrationCapabilityStatusList *cap;
590     bool old_postcopy_cap = migrate_postcopy_ram();
591 
592     if (migration_is_setup_or_active(s->state)) {
593         error_setg(errp, QERR_MIGRATION_ACTIVE);
594         return;
595     }
596 
597     for (cap = params; cap; cap = cap->next) {
598 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
599         if (cap->value->capability == MIGRATION_CAPABILITY_BLOCK
600             && cap->value->state) {
601             error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
602                        "block migration");
603             error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
604             continue;
605         }
606 #endif
607         if (cap->value->capability == MIGRATION_CAPABILITY_X_COLO) {
608             if (!colo_supported()) {
609                 error_setg(errp, "COLO is not currently supported, please"
610                              " configure with --enable-colo option in order to"
611                              " support COLO feature");
612                 continue;
613             }
614         }
615         s->enabled_capabilities[cap->value->capability] = cap->value->state;
616     }
617 
618     if (migrate_postcopy_ram()) {
619         if (migrate_use_compression()) {
620             /* The decompression threads asynchronously write into RAM
621              * rather than use the atomic copies needed to avoid
622              * userfaulting.  It should be possible to fix the decompression
623              * threads for compatibility in future.
624              */
625             error_report("Postcopy is not currently compatible with "
626                          "compression");
627             s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] =
628                 false;
629         }
630         /* This check is reasonably expensive, so only when it's being
631          * set the first time, also it's only the destination that needs
632          * special support.
633          */
634         if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
635             !postcopy_ram_supported_by_host()) {
636             /* postcopy_ram_supported_by_host will have emitted a more
637              * detailed message
638              */
639             error_report("Postcopy is not supported");
640             s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] =
641                 false;
642         }
643     }
644 }
645 
646 void qmp_migrate_set_parameters(MigrationParameters *params, Error **errp)
647 {
648     MigrationState *s = migrate_get_current();
649 
650     if (params->has_compress_level &&
651         (params->compress_level < 0 || params->compress_level > 9)) {
652         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
653                    "is invalid, it should be in the range of 0 to 9");
654         return;
655     }
656     if (params->has_compress_threads &&
657         (params->compress_threads < 1 || params->compress_threads > 255)) {
658         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
659                    "compress_threads",
660                    "is invalid, it should be in the range of 1 to 255");
661         return;
662     }
663     if (params->has_decompress_threads &&
664         (params->decompress_threads < 1 || params->decompress_threads > 255)) {
665         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
666                    "decompress_threads",
667                    "is invalid, it should be in the range of 1 to 255");
668         return;
669     }
670     if (params->has_cpu_throttle_initial &&
671         (params->cpu_throttle_initial < 1 ||
672          params->cpu_throttle_initial > 99)) {
673         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
674                    "cpu_throttle_initial",
675                    "an integer in the range of 1 to 99");
676         return;
677     }
678     if (params->has_cpu_throttle_increment &&
679         (params->cpu_throttle_increment < 1 ||
680          params->cpu_throttle_increment > 99)) {
681         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
682                    "cpu_throttle_increment",
683                    "an integer in the range of 1 to 99");
684         return;
685     }
686     if (params->has_max_bandwidth &&
687         (params->max_bandwidth < 0 || params->max_bandwidth > SIZE_MAX)) {
688         error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
689                          " range of 0 to %zu bytes/second", SIZE_MAX);
690         return;
691     }
692     if (params->has_downtime_limit &&
693         (params->downtime_limit < 0 ||
694          params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
695         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
696                          "the range of 0 to %d milliseconds",
697                          MAX_MIGRATE_DOWNTIME);
698         return;
699     }
700     if (params->has_x_checkpoint_delay && (params->x_checkpoint_delay < 0)) {
701         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
702                     "x_checkpoint_delay",
703                     "is invalid, it should be positive");
704     }
705 
706     if (params->has_compress_level) {
707         s->parameters.compress_level = params->compress_level;
708     }
709     if (params->has_compress_threads) {
710         s->parameters.compress_threads = params->compress_threads;
711     }
712     if (params->has_decompress_threads) {
713         s->parameters.decompress_threads = params->decompress_threads;
714     }
715     if (params->has_cpu_throttle_initial) {
716         s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
717     }
718     if (params->has_cpu_throttle_increment) {
719         s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
720     }
721     if (params->has_tls_creds) {
722         g_free(s->parameters.tls_creds);
723         s->parameters.tls_creds = g_strdup(params->tls_creds);
724     }
725     if (params->has_tls_hostname) {
726         g_free(s->parameters.tls_hostname);
727         s->parameters.tls_hostname = g_strdup(params->tls_hostname);
728     }
729     if (params->has_max_bandwidth) {
730         s->parameters.max_bandwidth = params->max_bandwidth;
731         if (s->to_dst_file) {
732             qemu_file_set_rate_limit(s->to_dst_file,
733                                 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
734         }
735     }
736     if (params->has_downtime_limit) {
737         s->parameters.downtime_limit = params->downtime_limit;
738     }
739 
740     if (params->has_x_checkpoint_delay) {
741         s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
742         if (migration_in_colo_state()) {
743             colo_checkpoint_notify(s);
744         }
745     }
746     if (params->has_block_incremental) {
747         s->parameters.block_incremental = params->block_incremental;
748     }
749 }
750 
751 
752 void qmp_migrate_start_postcopy(Error **errp)
753 {
754     MigrationState *s = migrate_get_current();
755 
756     if (!migrate_postcopy_ram()) {
757         error_setg(errp, "Enable postcopy with migrate_set_capability before"
758                          " the start of migration");
759         return;
760     }
761 
762     if (s->state == MIGRATION_STATUS_NONE) {
763         error_setg(errp, "Postcopy must be started after migration has been"
764                          " started");
765         return;
766     }
767     /*
768      * we don't error if migration has finished since that would be racy
769      * with issuing this command.
770      */
771     atomic_set(&s->start_postcopy, true);
772 }
773 
774 /* shared migration helpers */
775 
776 void migrate_set_state(int *state, int old_state, int new_state)
777 {
778     if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
779         trace_migrate_set_state(new_state);
780         migrate_generate_event(new_state);
781     }
782 }
783 
784 void migrate_set_block_enabled(bool value, Error **errp)
785 {
786     MigrationCapabilityStatusList *cap;
787 
788     cap = g_new0(MigrationCapabilityStatusList, 1);
789     cap->value = g_new0(MigrationCapabilityStatus, 1);
790     cap->value->capability = MIGRATION_CAPABILITY_BLOCK;
791     cap->value->state = value;
792     qmp_migrate_set_capabilities(cap, errp);
793     qapi_free_MigrationCapabilityStatusList(cap);
794 }
795 
796 static void migrate_set_block_incremental(MigrationState *s, bool value)
797 {
798     s->parameters.block_incremental = value;
799 }
800 
801 static void block_cleanup_parameters(MigrationState *s)
802 {
803     if (s->must_remove_block_options) {
804         /* setting to false can never fail */
805         migrate_set_block_enabled(false, &error_abort);
806         migrate_set_block_incremental(s, false);
807         s->must_remove_block_options = false;
808     }
809 }
810 
811 static void migrate_fd_cleanup(void *opaque)
812 {
813     MigrationState *s = opaque;
814 
815     qemu_bh_delete(s->cleanup_bh);
816     s->cleanup_bh = NULL;
817 
818     if (s->to_dst_file) {
819         trace_migrate_fd_cleanup();
820         qemu_mutex_unlock_iothread();
821         if (s->migration_thread_running) {
822             qemu_thread_join(&s->thread);
823             s->migration_thread_running = false;
824         }
825         qemu_mutex_lock_iothread();
826 
827         qemu_fclose(s->to_dst_file);
828         s->to_dst_file = NULL;
829     }
830 
831     assert((s->state != MIGRATION_STATUS_ACTIVE) &&
832            (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
833 
834     if (s->state == MIGRATION_STATUS_CANCELLING) {
835         migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
836                           MIGRATION_STATUS_CANCELLED);
837     }
838 
839     notifier_list_notify(&migration_state_notifiers, s);
840     block_cleanup_parameters(s);
841 }
842 
843 void migrate_fd_error(MigrationState *s, const Error *error)
844 {
845     trace_migrate_fd_error(error_get_pretty(error));
846     assert(s->to_dst_file == NULL);
847     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
848                       MIGRATION_STATUS_FAILED);
849     if (!s->error) {
850         s->error = error_copy(error);
851     }
852     notifier_list_notify(&migration_state_notifiers, s);
853     block_cleanup_parameters(s);
854 }
855 
856 static void migrate_fd_cancel(MigrationState *s)
857 {
858     int old_state ;
859     QEMUFile *f = migrate_get_current()->to_dst_file;
860     trace_migrate_fd_cancel();
861 
862     if (s->rp_state.from_dst_file) {
863         /* shutdown the rp socket, so causing the rp thread to shutdown */
864         qemu_file_shutdown(s->rp_state.from_dst_file);
865     }
866 
867     do {
868         old_state = s->state;
869         if (!migration_is_setup_or_active(old_state)) {
870             break;
871         }
872         migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
873     } while (s->state != MIGRATION_STATUS_CANCELLING);
874 
875     /*
876      * If we're unlucky the migration code might be stuck somewhere in a
877      * send/write while the network has failed and is waiting to timeout;
878      * if we've got shutdown(2) available then we can force it to quit.
879      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
880      * called in a bh, so there is no race against this cancel.
881      */
882     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
883         qemu_file_shutdown(f);
884     }
885     if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
886         Error *local_err = NULL;
887 
888         bdrv_invalidate_cache_all(&local_err);
889         if (local_err) {
890             error_report_err(local_err);
891         } else {
892             s->block_inactive = false;
893         }
894     }
895     block_cleanup_parameters(s);
896 }
897 
898 void add_migration_state_change_notifier(Notifier *notify)
899 {
900     notifier_list_add(&migration_state_notifiers, notify);
901 }
902 
903 void remove_migration_state_change_notifier(Notifier *notify)
904 {
905     notifier_remove(notify);
906 }
907 
908 bool migration_in_setup(MigrationState *s)
909 {
910     return s->state == MIGRATION_STATUS_SETUP;
911 }
912 
913 bool migration_has_finished(MigrationState *s)
914 {
915     return s->state == MIGRATION_STATUS_COMPLETED;
916 }
917 
918 bool migration_has_failed(MigrationState *s)
919 {
920     return (s->state == MIGRATION_STATUS_CANCELLED ||
921             s->state == MIGRATION_STATUS_FAILED);
922 }
923 
924 bool migration_in_postcopy(void)
925 {
926     MigrationState *s = migrate_get_current();
927 
928     return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
929 }
930 
931 bool migration_in_postcopy_after_devices(MigrationState *s)
932 {
933     return migration_in_postcopy() && s->postcopy_after_devices;
934 }
935 
936 bool migration_is_idle(void)
937 {
938     MigrationState *s = migrate_get_current();
939 
940     switch (s->state) {
941     case MIGRATION_STATUS_NONE:
942     case MIGRATION_STATUS_CANCELLED:
943     case MIGRATION_STATUS_COMPLETED:
944     case MIGRATION_STATUS_FAILED:
945         return true;
946     case MIGRATION_STATUS_SETUP:
947     case MIGRATION_STATUS_CANCELLING:
948     case MIGRATION_STATUS_ACTIVE:
949     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
950     case MIGRATION_STATUS_COLO:
951         return false;
952     case MIGRATION_STATUS__MAX:
953         g_assert_not_reached();
954     }
955 
956     return false;
957 }
958 
959 MigrationState *migrate_init(void)
960 {
961     MigrationState *s = migrate_get_current();
962 
963     /*
964      * Reinitialise all migration state, except
965      * parameters/capabilities that the user set, and
966      * locks.
967      */
968     s->bytes_xfer = 0;
969     s->xfer_limit = 0;
970     s->cleanup_bh = 0;
971     s->to_dst_file = NULL;
972     s->state = MIGRATION_STATUS_NONE;
973     s->rp_state.from_dst_file = NULL;
974     s->rp_state.error = false;
975     s->mbps = 0.0;
976     s->downtime = 0;
977     s->expected_downtime = 0;
978     s->setup_time = 0;
979     s->start_postcopy = false;
980     s->postcopy_after_devices = false;
981     s->migration_thread_running = false;
982     error_free(s->error);
983     s->error = NULL;
984 
985     migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
986 
987     s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
988     return s;
989 }
990 
991 static GSList *migration_blockers;
992 
993 int migrate_add_blocker(Error *reason, Error **errp)
994 {
995     if (migrate_get_current()->only_migratable) {
996         error_propagate(errp, error_copy(reason));
997         error_prepend(errp, "disallowing migration blocker "
998                           "(--only_migratable) for: ");
999         return -EACCES;
1000     }
1001 
1002     if (migration_is_idle()) {
1003         migration_blockers = g_slist_prepend(migration_blockers, reason);
1004         return 0;
1005     }
1006 
1007     error_propagate(errp, error_copy(reason));
1008     error_prepend(errp, "disallowing migration blocker (migration in "
1009                       "progress) for: ");
1010     return -EBUSY;
1011 }
1012 
1013 void migrate_del_blocker(Error *reason)
1014 {
1015     migration_blockers = g_slist_remove(migration_blockers, reason);
1016 }
1017 
1018 void qmp_migrate_incoming(const char *uri, Error **errp)
1019 {
1020     Error *local_err = NULL;
1021     static bool once = true;
1022 
1023     if (!deferred_incoming) {
1024         error_setg(errp, "For use with '-incoming defer'");
1025         return;
1026     }
1027     if (!once) {
1028         error_setg(errp, "The incoming migration has already been started");
1029     }
1030 
1031     qemu_start_incoming_migration(uri, &local_err);
1032 
1033     if (local_err) {
1034         error_propagate(errp, local_err);
1035         return;
1036     }
1037 
1038     once = false;
1039 }
1040 
1041 bool migration_is_blocked(Error **errp)
1042 {
1043     if (qemu_savevm_state_blocked(errp)) {
1044         return true;
1045     }
1046 
1047     if (migration_blockers) {
1048         error_propagate(errp, error_copy(migration_blockers->data));
1049         return true;
1050     }
1051 
1052     return false;
1053 }
1054 
1055 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1056                  bool has_inc, bool inc, bool has_detach, bool detach,
1057                  Error **errp)
1058 {
1059     Error *local_err = NULL;
1060     MigrationState *s = migrate_get_current();
1061     const char *p;
1062 
1063     if (migration_is_setup_or_active(s->state) ||
1064         s->state == MIGRATION_STATUS_CANCELLING ||
1065         s->state == MIGRATION_STATUS_COLO) {
1066         error_setg(errp, QERR_MIGRATION_ACTIVE);
1067         return;
1068     }
1069     if (runstate_check(RUN_STATE_INMIGRATE)) {
1070         error_setg(errp, "Guest is waiting for an incoming migration");
1071         return;
1072     }
1073 
1074     if (migration_is_blocked(errp)) {
1075         return;
1076     }
1077 
1078     if ((has_blk && blk) || (has_inc && inc)) {
1079         if (migrate_use_block() || migrate_use_block_incremental()) {
1080             error_setg(errp, "Command options are incompatible with "
1081                        "current migration capabilities");
1082             return;
1083         }
1084         migrate_set_block_enabled(true, &local_err);
1085         if (local_err) {
1086             error_propagate(errp, local_err);
1087             return;
1088         }
1089         s->must_remove_block_options = true;
1090     }
1091 
1092     if (has_inc && inc) {
1093         migrate_set_block_incremental(s, true);
1094     }
1095 
1096     s = migrate_init();
1097 
1098     if (strstart(uri, "tcp:", &p)) {
1099         tcp_start_outgoing_migration(s, p, &local_err);
1100 #ifdef CONFIG_RDMA
1101     } else if (strstart(uri, "rdma:", &p)) {
1102         rdma_start_outgoing_migration(s, p, &local_err);
1103 #endif
1104     } else if (strstart(uri, "exec:", &p)) {
1105         exec_start_outgoing_migration(s, p, &local_err);
1106     } else if (strstart(uri, "unix:", &p)) {
1107         unix_start_outgoing_migration(s, p, &local_err);
1108     } else if (strstart(uri, "fd:", &p)) {
1109         fd_start_outgoing_migration(s, p, &local_err);
1110     } else {
1111         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1112                    "a valid migration protocol");
1113         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1114                           MIGRATION_STATUS_FAILED);
1115         return;
1116     }
1117 
1118     if (local_err) {
1119         migrate_fd_error(s, local_err);
1120         error_propagate(errp, local_err);
1121         return;
1122     }
1123 }
1124 
1125 void qmp_migrate_cancel(Error **errp)
1126 {
1127     migrate_fd_cancel(migrate_get_current());
1128 }
1129 
1130 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1131 {
1132     MigrationState *s = migrate_get_current();
1133     int64_t new_size;
1134 
1135     /* Check for truncation */
1136     if (value != (size_t)value) {
1137         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1138                    "exceeding address space");
1139         return;
1140     }
1141 
1142     /* Cache should not be larger than guest ram size */
1143     if (value > ram_bytes_total()) {
1144         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1145                    "exceeds guest ram size ");
1146         return;
1147     }
1148 
1149     new_size = xbzrle_cache_resize(value);
1150     if (new_size < 0) {
1151         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1152                    "is smaller than page size");
1153         return;
1154     }
1155 
1156     s->xbzrle_cache_size = new_size;
1157 }
1158 
1159 int64_t qmp_query_migrate_cache_size(Error **errp)
1160 {
1161     return migrate_xbzrle_cache_size();
1162 }
1163 
1164 void qmp_migrate_set_speed(int64_t value, Error **errp)
1165 {
1166     MigrationParameters p = {
1167         .has_max_bandwidth = true,
1168         .max_bandwidth = value,
1169     };
1170 
1171     qmp_migrate_set_parameters(&p, errp);
1172 }
1173 
1174 void qmp_migrate_set_downtime(double value, Error **errp)
1175 {
1176     if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1177         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1178                          "the range of 0 to %d seconds",
1179                          MAX_MIGRATE_DOWNTIME_SECONDS);
1180         return;
1181     }
1182 
1183     value *= 1000; /* Convert to milliseconds */
1184     value = MAX(0, MIN(INT64_MAX, value));
1185 
1186     MigrationParameters p = {
1187         .has_downtime_limit = true,
1188         .downtime_limit = value,
1189     };
1190 
1191     qmp_migrate_set_parameters(&p, errp);
1192 }
1193 
1194 bool migrate_release_ram(void)
1195 {
1196     MigrationState *s;
1197 
1198     s = migrate_get_current();
1199 
1200     return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1201 }
1202 
1203 bool migrate_postcopy_ram(void)
1204 {
1205     MigrationState *s;
1206 
1207     s = migrate_get_current();
1208 
1209     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1210 }
1211 
1212 bool migrate_auto_converge(void)
1213 {
1214     MigrationState *s;
1215 
1216     s = migrate_get_current();
1217 
1218     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1219 }
1220 
1221 bool migrate_zero_blocks(void)
1222 {
1223     MigrationState *s;
1224 
1225     s = migrate_get_current();
1226 
1227     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1228 }
1229 
1230 bool migrate_use_compression(void)
1231 {
1232     MigrationState *s;
1233 
1234     s = migrate_get_current();
1235 
1236     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
1237 }
1238 
1239 int migrate_compress_level(void)
1240 {
1241     MigrationState *s;
1242 
1243     s = migrate_get_current();
1244 
1245     return s->parameters.compress_level;
1246 }
1247 
1248 int migrate_compress_threads(void)
1249 {
1250     MigrationState *s;
1251 
1252     s = migrate_get_current();
1253 
1254     return s->parameters.compress_threads;
1255 }
1256 
1257 int migrate_decompress_threads(void)
1258 {
1259     MigrationState *s;
1260 
1261     s = migrate_get_current();
1262 
1263     return s->parameters.decompress_threads;
1264 }
1265 
1266 bool migrate_use_events(void)
1267 {
1268     MigrationState *s;
1269 
1270     s = migrate_get_current();
1271 
1272     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1273 }
1274 
1275 int migrate_use_xbzrle(void)
1276 {
1277     MigrationState *s;
1278 
1279     s = migrate_get_current();
1280 
1281     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1282 }
1283 
1284 int64_t migrate_xbzrle_cache_size(void)
1285 {
1286     MigrationState *s;
1287 
1288     s = migrate_get_current();
1289 
1290     return s->xbzrle_cache_size;
1291 }
1292 
1293 bool migrate_use_block(void)
1294 {
1295     MigrationState *s;
1296 
1297     s = migrate_get_current();
1298 
1299     return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
1300 }
1301 
1302 bool migrate_use_return_path(void)
1303 {
1304     MigrationState *s;
1305 
1306     s = migrate_get_current();
1307 
1308     return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
1309 }
1310 
1311 bool migrate_use_block_incremental(void)
1312 {
1313     MigrationState *s;
1314 
1315     s = migrate_get_current();
1316 
1317     return s->parameters.block_incremental;
1318 }
1319 
1320 /* migration thread support */
1321 /*
1322  * Something bad happened to the RP stream, mark an error
1323  * The caller shall print or trace something to indicate why
1324  */
1325 static void mark_source_rp_bad(MigrationState *s)
1326 {
1327     s->rp_state.error = true;
1328 }
1329 
1330 static struct rp_cmd_args {
1331     ssize_t     len; /* -1 = variable */
1332     const char *name;
1333 } rp_cmd_args[] = {
1334     [MIG_RP_MSG_INVALID]        = { .len = -1, .name = "INVALID" },
1335     [MIG_RP_MSG_SHUT]           = { .len =  4, .name = "SHUT" },
1336     [MIG_RP_MSG_PONG]           = { .len =  4, .name = "PONG" },
1337     [MIG_RP_MSG_REQ_PAGES]      = { .len = 12, .name = "REQ_PAGES" },
1338     [MIG_RP_MSG_REQ_PAGES_ID]   = { .len = -1, .name = "REQ_PAGES_ID" },
1339     [MIG_RP_MSG_MAX]            = { .len = -1, .name = "MAX" },
1340 };
1341 
1342 /*
1343  * Process a request for pages received on the return path,
1344  * We're allowed to send more than requested (e.g. to round to our page size)
1345  * and we don't need to send pages that have already been sent.
1346  */
1347 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1348                                        ram_addr_t start, size_t len)
1349 {
1350     long our_host_ps = getpagesize();
1351 
1352     trace_migrate_handle_rp_req_pages(rbname, start, len);
1353 
1354     /*
1355      * Since we currently insist on matching page sizes, just sanity check
1356      * we're being asked for whole host pages.
1357      */
1358     if (start & (our_host_ps-1) ||
1359        (len & (our_host_ps-1))) {
1360         error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1361                      " len: %zd", __func__, start, len);
1362         mark_source_rp_bad(ms);
1363         return;
1364     }
1365 
1366     if (ram_save_queue_pages(rbname, start, len)) {
1367         mark_source_rp_bad(ms);
1368     }
1369 }
1370 
1371 /*
1372  * Handles messages sent on the return path towards the source VM
1373  *
1374  */
1375 static void *source_return_path_thread(void *opaque)
1376 {
1377     MigrationState *ms = opaque;
1378     QEMUFile *rp = ms->rp_state.from_dst_file;
1379     uint16_t header_len, header_type;
1380     uint8_t buf[512];
1381     uint32_t tmp32, sibling_error;
1382     ram_addr_t start = 0; /* =0 to silence warning */
1383     size_t  len = 0, expected_len;
1384     int res;
1385 
1386     trace_source_return_path_thread_entry();
1387     while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
1388            migration_is_setup_or_active(ms->state)) {
1389         trace_source_return_path_thread_loop_top();
1390         header_type = qemu_get_be16(rp);
1391         header_len = qemu_get_be16(rp);
1392 
1393         if (header_type >= MIG_RP_MSG_MAX ||
1394             header_type == MIG_RP_MSG_INVALID) {
1395             error_report("RP: Received invalid message 0x%04x length 0x%04x",
1396                     header_type, header_len);
1397             mark_source_rp_bad(ms);
1398             goto out;
1399         }
1400 
1401         if ((rp_cmd_args[header_type].len != -1 &&
1402             header_len != rp_cmd_args[header_type].len) ||
1403             header_len > sizeof(buf)) {
1404             error_report("RP: Received '%s' message (0x%04x) with"
1405                     "incorrect length %d expecting %zu",
1406                     rp_cmd_args[header_type].name, header_type, header_len,
1407                     (size_t)rp_cmd_args[header_type].len);
1408             mark_source_rp_bad(ms);
1409             goto out;
1410         }
1411 
1412         /* We know we've got a valid header by this point */
1413         res = qemu_get_buffer(rp, buf, header_len);
1414         if (res != header_len) {
1415             error_report("RP: Failed reading data for message 0x%04x"
1416                          " read %d expected %d",
1417                          header_type, res, header_len);
1418             mark_source_rp_bad(ms);
1419             goto out;
1420         }
1421 
1422         /* OK, we have the message and the data */
1423         switch (header_type) {
1424         case MIG_RP_MSG_SHUT:
1425             sibling_error = ldl_be_p(buf);
1426             trace_source_return_path_thread_shut(sibling_error);
1427             if (sibling_error) {
1428                 error_report("RP: Sibling indicated error %d", sibling_error);
1429                 mark_source_rp_bad(ms);
1430             }
1431             /*
1432              * We'll let the main thread deal with closing the RP
1433              * we could do a shutdown(2) on it, but we're the only user
1434              * anyway, so there's nothing gained.
1435              */
1436             goto out;
1437 
1438         case MIG_RP_MSG_PONG:
1439             tmp32 = ldl_be_p(buf);
1440             trace_source_return_path_thread_pong(tmp32);
1441             break;
1442 
1443         case MIG_RP_MSG_REQ_PAGES:
1444             start = ldq_be_p(buf);
1445             len = ldl_be_p(buf + 8);
1446             migrate_handle_rp_req_pages(ms, NULL, start, len);
1447             break;
1448 
1449         case MIG_RP_MSG_REQ_PAGES_ID:
1450             expected_len = 12 + 1; /* header + termination */
1451 
1452             if (header_len >= expected_len) {
1453                 start = ldq_be_p(buf);
1454                 len = ldl_be_p(buf + 8);
1455                 /* Now we expect an idstr */
1456                 tmp32 = buf[12]; /* Length of the following idstr */
1457                 buf[13 + tmp32] = '\0';
1458                 expected_len += tmp32;
1459             }
1460             if (header_len != expected_len) {
1461                 error_report("RP: Req_Page_id with length %d expecting %zd",
1462                         header_len, expected_len);
1463                 mark_source_rp_bad(ms);
1464                 goto out;
1465             }
1466             migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
1467             break;
1468 
1469         default:
1470             break;
1471         }
1472     }
1473     if (qemu_file_get_error(rp)) {
1474         trace_source_return_path_thread_bad_end();
1475         mark_source_rp_bad(ms);
1476     }
1477 
1478     trace_source_return_path_thread_end();
1479 out:
1480     ms->rp_state.from_dst_file = NULL;
1481     qemu_fclose(rp);
1482     return NULL;
1483 }
1484 
1485 static int open_return_path_on_source(MigrationState *ms)
1486 {
1487 
1488     ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
1489     if (!ms->rp_state.from_dst_file) {
1490         return -1;
1491     }
1492 
1493     trace_open_return_path_on_source();
1494     qemu_thread_create(&ms->rp_state.rp_thread, "return path",
1495                        source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
1496 
1497     trace_open_return_path_on_source_continue();
1498 
1499     return 0;
1500 }
1501 
1502 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1503 static int await_return_path_close_on_source(MigrationState *ms)
1504 {
1505     /*
1506      * If this is a normal exit then the destination will send a SHUT and the
1507      * rp_thread will exit, however if there's an error we need to cause
1508      * it to exit.
1509      */
1510     if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
1511         /*
1512          * shutdown(2), if we have it, will cause it to unblock if it's stuck
1513          * waiting for the destination.
1514          */
1515         qemu_file_shutdown(ms->rp_state.from_dst_file);
1516         mark_source_rp_bad(ms);
1517     }
1518     trace_await_return_path_close_on_source_joining();
1519     qemu_thread_join(&ms->rp_state.rp_thread);
1520     trace_await_return_path_close_on_source_close();
1521     return ms->rp_state.error;
1522 }
1523 
1524 /*
1525  * Switch from normal iteration to postcopy
1526  * Returns non-0 on error
1527  */
1528 static int postcopy_start(MigrationState *ms, bool *old_vm_running)
1529 {
1530     int ret;
1531     QIOChannelBuffer *bioc;
1532     QEMUFile *fb;
1533     int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1534     bool restart_block = false;
1535     migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
1536                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
1537 
1538     trace_postcopy_start();
1539     qemu_mutex_lock_iothread();
1540     trace_postcopy_start_set_run();
1541 
1542     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1543     *old_vm_running = runstate_is_running();
1544     global_state_store();
1545     ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1546     if (ret < 0) {
1547         goto fail;
1548     }
1549 
1550     ret = bdrv_inactivate_all();
1551     if (ret < 0) {
1552         goto fail;
1553     }
1554     restart_block = true;
1555 
1556     /*
1557      * Cause any non-postcopiable, but iterative devices to
1558      * send out their final data.
1559      */
1560     qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
1561 
1562     /*
1563      * in Finish migrate and with the io-lock held everything should
1564      * be quiet, but we've potentially still got dirty pages and we
1565      * need to tell the destination to throw any pages it's already received
1566      * that are dirty
1567      */
1568     if (ram_postcopy_send_discard_bitmap(ms)) {
1569         error_report("postcopy send discard bitmap failed");
1570         goto fail;
1571     }
1572 
1573     /*
1574      * send rest of state - note things that are doing postcopy
1575      * will notice we're in POSTCOPY_ACTIVE and not actually
1576      * wrap their state up here
1577      */
1578     qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
1579     /* Ping just for debugging, helps line traces up */
1580     qemu_savevm_send_ping(ms->to_dst_file, 2);
1581 
1582     /*
1583      * While loading the device state we may trigger page transfer
1584      * requests and the fd must be free to process those, and thus
1585      * the destination must read the whole device state off the fd before
1586      * it starts processing it.  Unfortunately the ad-hoc migration format
1587      * doesn't allow the destination to know the size to read without fully
1588      * parsing it through each devices load-state code (especially the open
1589      * coded devices that use get/put).
1590      * So we wrap the device state up in a package with a length at the start;
1591      * to do this we use a qemu_buf to hold the whole of the device state.
1592      */
1593     bioc = qio_channel_buffer_new(4096);
1594     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
1595     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
1596     object_unref(OBJECT(bioc));
1597 
1598     /*
1599      * Make sure the receiver can get incoming pages before we send the rest
1600      * of the state
1601      */
1602     qemu_savevm_send_postcopy_listen(fb);
1603 
1604     qemu_savevm_state_complete_precopy(fb, false, false);
1605     qemu_savevm_send_ping(fb, 3);
1606 
1607     qemu_savevm_send_postcopy_run(fb);
1608 
1609     /* <><> end of stuff going into the package */
1610 
1611     /* Last point of recovery; as soon as we send the package the destination
1612      * can open devices and potentially start running.
1613      * Lets just check again we've not got any errors.
1614      */
1615     ret = qemu_file_get_error(ms->to_dst_file);
1616     if (ret) {
1617         error_report("postcopy_start: Migration stream errored (pre package)");
1618         goto fail_closefb;
1619     }
1620 
1621     restart_block = false;
1622 
1623     /* Now send that blob */
1624     if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
1625         goto fail_closefb;
1626     }
1627     qemu_fclose(fb);
1628 
1629     /* Send a notify to give a chance for anything that needs to happen
1630      * at the transition to postcopy and after the device state; in particular
1631      * spice needs to trigger a transition now
1632      */
1633     ms->postcopy_after_devices = true;
1634     notifier_list_notify(&migration_state_notifiers, ms);
1635 
1636     ms->downtime =  qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
1637 
1638     qemu_mutex_unlock_iothread();
1639 
1640     /*
1641      * Although this ping is just for debug, it could potentially be
1642      * used for getting a better measurement of downtime at the source.
1643      */
1644     qemu_savevm_send_ping(ms->to_dst_file, 4);
1645 
1646     if (migrate_release_ram()) {
1647         ram_postcopy_migrated_memory_release(ms);
1648     }
1649 
1650     ret = qemu_file_get_error(ms->to_dst_file);
1651     if (ret) {
1652         error_report("postcopy_start: Migration stream errored");
1653         migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1654                               MIGRATION_STATUS_FAILED);
1655     }
1656 
1657     return ret;
1658 
1659 fail_closefb:
1660     qemu_fclose(fb);
1661 fail:
1662     migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1663                           MIGRATION_STATUS_FAILED);
1664     if (restart_block) {
1665         /* A failure happened early enough that we know the destination hasn't
1666          * accessed block devices, so we're safe to recover.
1667          */
1668         Error *local_err = NULL;
1669 
1670         bdrv_invalidate_cache_all(&local_err);
1671         if (local_err) {
1672             error_report_err(local_err);
1673         }
1674     }
1675     qemu_mutex_unlock_iothread();
1676     return -1;
1677 }
1678 
1679 /**
1680  * migration_completion: Used by migration_thread when there's not much left.
1681  *   The caller 'breaks' the loop when this returns.
1682  *
1683  * @s: Current migration state
1684  * @current_active_state: The migration state we expect to be in
1685  * @*old_vm_running: Pointer to old_vm_running flag
1686  * @*start_time: Pointer to time to update
1687  */
1688 static void migration_completion(MigrationState *s, int current_active_state,
1689                                  bool *old_vm_running,
1690                                  int64_t *start_time)
1691 {
1692     int ret;
1693 
1694     if (s->state == MIGRATION_STATUS_ACTIVE) {
1695         qemu_mutex_lock_iothread();
1696         *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1697         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1698         *old_vm_running = runstate_is_running();
1699         ret = global_state_store();
1700 
1701         if (!ret) {
1702             bool inactivate = !migrate_colo_enabled();
1703             ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1704             if (ret >= 0) {
1705                 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
1706                 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
1707                                                          inactivate);
1708             }
1709             if (inactivate && ret >= 0) {
1710                 s->block_inactive = true;
1711             }
1712         }
1713         qemu_mutex_unlock_iothread();
1714 
1715         if (ret < 0) {
1716             goto fail;
1717         }
1718     } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1719         trace_migration_completion_postcopy_end();
1720 
1721         qemu_savevm_state_complete_postcopy(s->to_dst_file);
1722         trace_migration_completion_postcopy_end_after_complete();
1723     }
1724 
1725     /*
1726      * If rp was opened we must clean up the thread before
1727      * cleaning everything else up (since if there are no failures
1728      * it will wait for the destination to send it's status in
1729      * a SHUT command).
1730      */
1731     if (s->rp_state.from_dst_file) {
1732         int rp_error;
1733         trace_migration_return_path_end_before();
1734         rp_error = await_return_path_close_on_source(s);
1735         trace_migration_return_path_end_after(rp_error);
1736         if (rp_error) {
1737             goto fail_invalidate;
1738         }
1739     }
1740 
1741     if (qemu_file_get_error(s->to_dst_file)) {
1742         trace_migration_completion_file_err();
1743         goto fail_invalidate;
1744     }
1745 
1746     if (!migrate_colo_enabled()) {
1747         migrate_set_state(&s->state, current_active_state,
1748                           MIGRATION_STATUS_COMPLETED);
1749     }
1750 
1751     return;
1752 
1753 fail_invalidate:
1754     /* If not doing postcopy, vm_start() will be called: let's regain
1755      * control on images.
1756      */
1757     if (s->state == MIGRATION_STATUS_ACTIVE) {
1758         Error *local_err = NULL;
1759 
1760         qemu_mutex_lock_iothread();
1761         bdrv_invalidate_cache_all(&local_err);
1762         if (local_err) {
1763             error_report_err(local_err);
1764         } else {
1765             s->block_inactive = false;
1766         }
1767         qemu_mutex_unlock_iothread();
1768     }
1769 
1770 fail:
1771     migrate_set_state(&s->state, current_active_state,
1772                       MIGRATION_STATUS_FAILED);
1773 }
1774 
1775 bool migrate_colo_enabled(void)
1776 {
1777     MigrationState *s = migrate_get_current();
1778     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
1779 }
1780 
1781 /*
1782  * Master migration thread on the source VM.
1783  * It drives the migration and pumps the data down the outgoing channel.
1784  */
1785 static void *migration_thread(void *opaque)
1786 {
1787     MigrationState *s = opaque;
1788     /* Used by the bandwidth calcs, updated later */
1789     int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1790     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
1791     int64_t initial_bytes = 0;
1792     /*
1793      * The final stage happens when the remaining data is smaller than
1794      * this threshold; it's calculated from the requested downtime and
1795      * measured bandwidth
1796      */
1797     int64_t threshold_size = 0;
1798     int64_t start_time = initial_time;
1799     int64_t end_time;
1800     bool old_vm_running = false;
1801     bool entered_postcopy = false;
1802     /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
1803     enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE;
1804     bool enable_colo = migrate_colo_enabled();
1805 
1806     rcu_register_thread();
1807 
1808     qemu_savevm_state_header(s->to_dst_file);
1809 
1810     /*
1811      * If we opened the return path, we need to make sure dst has it
1812      * opened as well.
1813      */
1814     if (s->rp_state.from_dst_file) {
1815         /* Now tell the dest that it should open its end so it can reply */
1816         qemu_savevm_send_open_return_path(s->to_dst_file);
1817 
1818         /* And do a ping that will make stuff easier to debug */
1819         qemu_savevm_send_ping(s->to_dst_file, 1);
1820     }
1821 
1822     if (migrate_postcopy_ram()) {
1823         /*
1824          * Tell the destination that we *might* want to do postcopy later;
1825          * if the other end can't do postcopy it should fail now, nice and
1826          * early.
1827          */
1828         qemu_savevm_send_postcopy_advise(s->to_dst_file);
1829     }
1830 
1831     qemu_savevm_state_setup(s->to_dst_file);
1832 
1833     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
1834     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1835                       MIGRATION_STATUS_ACTIVE);
1836 
1837     trace_migration_thread_setup_complete();
1838 
1839     while (s->state == MIGRATION_STATUS_ACTIVE ||
1840            s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1841         int64_t current_time;
1842         uint64_t pending_size;
1843 
1844         if (!qemu_file_rate_limit(s->to_dst_file)) {
1845             uint64_t pend_post, pend_nonpost;
1846 
1847             qemu_savevm_state_pending(s->to_dst_file, threshold_size,
1848                                       &pend_nonpost, &pend_post);
1849             pending_size = pend_nonpost + pend_post;
1850             trace_migrate_pending(pending_size, threshold_size,
1851                                   pend_post, pend_nonpost);
1852             if (pending_size && pending_size >= threshold_size) {
1853                 /* Still a significant amount to transfer */
1854 
1855                 if (migrate_postcopy_ram() &&
1856                     s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE &&
1857                     pend_nonpost <= threshold_size &&
1858                     atomic_read(&s->start_postcopy)) {
1859 
1860                     if (!postcopy_start(s, &old_vm_running)) {
1861                         current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE;
1862                         entered_postcopy = true;
1863                     }
1864 
1865                     continue;
1866                 }
1867                 /* Just another iteration step */
1868                 qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy);
1869             } else {
1870                 trace_migration_thread_low_pending(pending_size);
1871                 migration_completion(s, current_active_state,
1872                                      &old_vm_running, &start_time);
1873                 break;
1874             }
1875         }
1876 
1877         if (qemu_file_get_error(s->to_dst_file)) {
1878             migrate_set_state(&s->state, current_active_state,
1879                               MIGRATION_STATUS_FAILED);
1880             trace_migration_thread_file_err();
1881             break;
1882         }
1883         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1884         if (current_time >= initial_time + BUFFER_DELAY) {
1885             uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) -
1886                                          initial_bytes;
1887             uint64_t time_spent = current_time - initial_time;
1888             double bandwidth = (double)transferred_bytes / time_spent;
1889             threshold_size = bandwidth * s->parameters.downtime_limit;
1890 
1891             s->mbps = (((double) transferred_bytes * 8.0) /
1892                     ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
1893 
1894             trace_migrate_transferred(transferred_bytes, time_spent,
1895                                       bandwidth, threshold_size);
1896             /* if we haven't sent anything, we don't want to recalculate
1897                10000 is a small enough number for our purposes */
1898             if (ram_counters.dirty_pages_rate && transferred_bytes > 10000) {
1899                 s->expected_downtime = ram_counters.dirty_pages_rate *
1900                     qemu_target_page_size() / bandwidth;
1901             }
1902 
1903             qemu_file_reset_rate_limit(s->to_dst_file);
1904             initial_time = current_time;
1905             initial_bytes = qemu_ftell(s->to_dst_file);
1906         }
1907         if (qemu_file_rate_limit(s->to_dst_file)) {
1908             /* usleep expects microseconds */
1909             g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
1910         }
1911     }
1912 
1913     trace_migration_thread_after_loop();
1914     /* If we enabled cpu throttling for auto-converge, turn it off. */
1915     cpu_throttle_stop();
1916     end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1917 
1918     qemu_mutex_lock_iothread();
1919     /*
1920      * The resource has been allocated by migration will be reused in COLO
1921      * process, so don't release them.
1922      */
1923     if (!enable_colo) {
1924         qemu_savevm_state_cleanup();
1925     }
1926     if (s->state == MIGRATION_STATUS_COMPLETED) {
1927         uint64_t transferred_bytes = qemu_ftell(s->to_dst_file);
1928         s->total_time = end_time - s->total_time;
1929         if (!entered_postcopy) {
1930             s->downtime = end_time - start_time;
1931         }
1932         if (s->total_time) {
1933             s->mbps = (((double) transferred_bytes * 8.0) /
1934                        ((double) s->total_time)) / 1000;
1935         }
1936         runstate_set(RUN_STATE_POSTMIGRATE);
1937     } else {
1938         if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) {
1939             migrate_start_colo_process(s);
1940             qemu_savevm_state_cleanup();
1941             /*
1942             * Fixme: we will run VM in COLO no matter its old running state.
1943             * After exited COLO, we will keep running.
1944             */
1945             old_vm_running = true;
1946         }
1947         if (old_vm_running && !entered_postcopy) {
1948             vm_start();
1949         } else {
1950             if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
1951                 runstate_set(RUN_STATE_POSTMIGRATE);
1952             }
1953         }
1954     }
1955     qemu_bh_schedule(s->cleanup_bh);
1956     qemu_mutex_unlock_iothread();
1957 
1958     rcu_unregister_thread();
1959     return NULL;
1960 }
1961 
1962 void migrate_fd_connect(MigrationState *s)
1963 {
1964     s->expected_downtime = s->parameters.downtime_limit;
1965     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
1966 
1967     qemu_file_set_blocking(s->to_dst_file, true);
1968     qemu_file_set_rate_limit(s->to_dst_file,
1969                              s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
1970 
1971     /* Notify before starting migration thread */
1972     notifier_list_notify(&migration_state_notifiers, s);
1973 
1974     /*
1975      * Open the return path. For postcopy, it is used exclusively. For
1976      * precopy, only if user specified "return-path" capability would
1977      * QEMU uses the return path.
1978      */
1979     if (migrate_postcopy_ram() || migrate_use_return_path()) {
1980         if (open_return_path_on_source(s)) {
1981             error_report("Unable to open return-path for postcopy");
1982             migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1983                               MIGRATION_STATUS_FAILED);
1984             migrate_fd_cleanup(s);
1985             return;
1986         }
1987     }
1988 
1989     qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
1990                        QEMU_THREAD_JOINABLE);
1991     s->migration_thread_running = true;
1992 }
1993 
1994 void migration_global_dump(Monitor *mon)
1995 {
1996     MigrationState *ms = migrate_get_current();
1997 
1998     monitor_printf(mon, "globals: store-global-state=%d, only_migratable=%d, "
1999                    "send-configuration=%d, send-section-footer=%d\n",
2000                    ms->store_global_state, ms->only_migratable,
2001                    ms->send_configuration, ms->send_section_footer);
2002 }
2003 
2004 static Property migration_properties[] = {
2005     DEFINE_PROP_BOOL("store-global-state", MigrationState,
2006                      store_global_state, true),
2007     DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false),
2008     DEFINE_PROP_BOOL("send-configuration", MigrationState,
2009                      send_configuration, true),
2010     DEFINE_PROP_BOOL("send-section-footer", MigrationState,
2011                      send_section_footer, true),
2012     DEFINE_PROP_END_OF_LIST(),
2013 };
2014 
2015 static void migration_class_init(ObjectClass *klass, void *data)
2016 {
2017     DeviceClass *dc = DEVICE_CLASS(klass);
2018 
2019     dc->user_creatable = false;
2020     dc->props = migration_properties;
2021 }
2022 
2023 static void migration_instance_init(Object *obj)
2024 {
2025     MigrationState *ms = MIGRATION_OBJ(obj);
2026 
2027     ms->state = MIGRATION_STATUS_NONE;
2028     ms->xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE;
2029     ms->mbps = -1;
2030     ms->parameters = (MigrationParameters) {
2031         .compress_level = DEFAULT_MIGRATE_COMPRESS_LEVEL,
2032         .compress_threads = DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT,
2033         .decompress_threads = DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT,
2034         .cpu_throttle_initial = DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL,
2035         .cpu_throttle_increment = DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT,
2036         .max_bandwidth = MAX_THROTTLE,
2037         .downtime_limit = DEFAULT_MIGRATE_SET_DOWNTIME,
2038         .x_checkpoint_delay = DEFAULT_MIGRATE_X_CHECKPOINT_DELAY,
2039     };
2040     ms->parameters.tls_creds = g_strdup("");
2041     ms->parameters.tls_hostname = g_strdup("");
2042 }
2043 
2044 static const TypeInfo migration_type = {
2045     .name = TYPE_MIGRATION,
2046     /*
2047      * NOTE: TYPE_MIGRATION is not really a device, as the object is
2048      * not created using qdev_create(), it is not attached to the qdev
2049      * device tree, and it is never realized.
2050      *
2051      * TODO: Make this TYPE_OBJECT once QOM provides something like
2052      * TYPE_DEVICE's "-global" properties.
2053      */
2054     .parent = TYPE_DEVICE,
2055     .class_init = migration_class_init,
2056     .class_size = sizeof(MigrationClass),
2057     .instance_size = sizeof(MigrationState),
2058     .instance_init = migration_instance_init,
2059 };
2060 
2061 static void register_migration_types(void)
2062 {
2063     type_register_static(&migration_type);
2064 }
2065 
2066 type_init(register_migration_types);
2067