?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
tests: Run filter-redirector and -mirror test only on POSIX systems
[qemu.git] / migration / migration.c
blobc3fe0ed9cafd9a2990aace7ece2c21af1fefdd26
1 /*
2 * QEMU live migration
3 *
4 * Copyright IBM, Corp. 2008
5 *
6 * Authors:
7 * Anthony Liguori <aliguori@us.ibm.com>
8 *
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
11 *
12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
14 */
15
16 #include "qemu/osdep.h"
17 #include "qemu/cutils.h"
18 #include "qemu/error-report.h"
19 #include "migration/blocker.h"
20 #include "exec.h"
21 #include "fd.h"
22 #include "socket.h"
23 #include "rdma.h"
24 #include "ram.h"
25 #include "migration/global_state.h"
26 #include "migration/misc.h"
27 #include "migration.h"
28 #include "savevm.h"
29 #include "qemu-file-channel.h"
30 #include "qemu-file.h"
31 #include "migration/vmstate.h"
32 #include "block/block.h"
33 #include "qapi/qmp/qerror.h"
34 #include "qapi/util.h"
35 #include "qemu/rcu.h"
36 #include "block.h"
37 #include "postcopy-ram.h"
38 #include "qemu/thread.h"
39 #include "qmp-commands.h"
40 #include "trace.h"
41 #include "qapi-event.h"
42 #include "exec/target_page.h"
43 #include "io/channel-buffer.h"
44 #include "migration/colo.h"
45 #include "hw/boards.h"
46 #include "monitor/monitor.h"
47
48 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
49
50 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
51 * data. */
52 #define BUFFER_DELAY 100
53 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
54
55 /* Time in milliseconds we are allowed to stop the source,
56 * for sending the last part */
57 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
58
59 /* Maximum migrate downtime set to 2000 seconds */
60 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
61 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
62
63 /* Default compression thread count */
64 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
65 /* Default decompression thread count, usually decompression is at
66 * least 4 times as fast as compression.*/
67 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
68 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
69 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
70 /* Define default autoconverge cpu throttle migration parameters */
71 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
72 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
73
74 /* Migration XBZRLE default cache size */
75 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
76
77 /* The delay time (in ms) between two COLO checkpoints
78 * Note: Please change this default value to 10000 when we support hybrid mode.
79 */
80 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
81
82 static NotifierList migration_state_notifiers =
83 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
84
85 static bool deferred_incoming;
86
87 /* Messages sent on the return path from destination to source */
88 enum mig_rp_message_type {
89 MIG_RP_MSG_INVALID = 0, /* Must be 0 */
90 MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */
91 MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */
92
93 MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
94 MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */
95
96 MIG_RP_MSG_MAX
97 };
98
99 /* When we add fault tolerance, we could have several
100 migrations at once. For now we don't need to add
101 dynamic creation of migration */
102
103 static MigrationState *current_migration;
104
105 static bool migration_object_check(MigrationState *ms, Error **errp);
106
107 void migration_object_init(void)
108 {
109 MachineState *ms = MACHINE(qdev_get_machine());
110 Error *err = NULL;
111
112 /* This can only be called once. */
113 assert(!current_migration);
114 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
115
116 if (!migration_object_check(current_migration, &err)) {
117 error_report_err(err);
118 exit(1);
119 }
120
121 /*
122 * We cannot really do this in migration_instance_init() since at
123 * that time global properties are not yet applied, then this
124 * value will be definitely replaced by something else.
125 */
126 if (ms->enforce_config_section) {
127 current_migration->send_configuration = true;
128 }
129 }
130
131 /* For outgoing */
132 MigrationState *migrate_get_current(void)
133 {
134 /* This can only be called after the object created. */
135 assert(current_migration);
136 return current_migration;
137 }
138
139 MigrationIncomingState *migration_incoming_get_current(void)
140 {
141 static bool once;
142 static MigrationIncomingState mis_current;
143
144 if (!once) {
145 mis_current.state = MIGRATION_STATUS_NONE;
146 memset(&mis_current, 0, sizeof(MigrationIncomingState));
147 qemu_mutex_init(&mis_current.rp_mutex);
148 qemu_event_init(&mis_current.main_thread_load_event, false);
149 once = true;
150 }
151 return &mis_current;
152 }
153
154 void migration_incoming_state_destroy(void)
155 {
156 struct MigrationIncomingState *mis = migration_incoming_get_current();
157
158 if (mis->to_src_file) {
159 /* Tell source that we are done */
160 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
161 qemu_fclose(mis->to_src_file);
162 mis->to_src_file = NULL;
163 }
164
165 if (mis->from_src_file) {
166 qemu_fclose(mis->from_src_file);
167 mis->from_src_file = NULL;
168 }
169
170 qemu_event_destroy(&mis->main_thread_load_event);
171 }
172
173 static void migrate_generate_event(int new_state)
174 {
175 if (migrate_use_events()) {
176 qapi_event_send_migration(new_state, &error_abort);
177 }
178 }
179
180 /*
181 * Called on -incoming with a defer: uri.
182 * The migration can be started later after any parameters have been
183 * changed.
184 */
185 static void deferred_incoming_migration(Error **errp)
186 {
187 if (deferred_incoming) {
188 error_setg(errp, "Incoming migration already deferred");
189 }
190 deferred_incoming = true;
191 }
192
193 /*
194 * Send a message on the return channel back to the source
195 * of the migration.
196 */
197 static void migrate_send_rp_message(MigrationIncomingState *mis,
198 enum mig_rp_message_type message_type,
199 uint16_t len, void *data)
200 {
201 trace_migrate_send_rp_message((int)message_type, len);
202 qemu_mutex_lock(&mis->rp_mutex);
203 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
204 qemu_put_be16(mis->to_src_file, len);
205 qemu_put_buffer(mis->to_src_file, data, len);
206 qemu_fflush(mis->to_src_file);
207 qemu_mutex_unlock(&mis->rp_mutex);
208 }
209
210 /* Request a range of pages from the source VM at the given
211 * start address.
212 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
213 * as the last request (a name must have been given previously)
214 * Start: Address offset within the RB
215 * Len: Length in bytes required - must be a multiple of pagesize
216 */
217 void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
218 ram_addr_t start, size_t len)
219 {
220 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
221 size_t msglen = 12; /* start + len */
222
223 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
224 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
225
226 if (rbname) {
227 int rbname_len = strlen(rbname);
228 assert(rbname_len < 256);
229
230 bufc[msglen++] = rbname_len;
231 memcpy(bufc + msglen, rbname, rbname_len);
232 msglen += rbname_len;
233 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc);
234 } else {
235 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc);
236 }
237 }
238
239 void qemu_start_incoming_migration(const char *uri, Error **errp)
240 {
241 const char *p;
242
243 qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
244 if (!strcmp(uri, "defer")) {
245 deferred_incoming_migration(errp);
246 } else if (strstart(uri, "tcp:", &p)) {
247 tcp_start_incoming_migration(p, errp);
248 #ifdef CONFIG_RDMA
249 } else if (strstart(uri, "rdma:", &p)) {
250 rdma_start_incoming_migration(p, errp);
251 #endif
252 } else if (strstart(uri, "exec:", &p)) {
253 exec_start_incoming_migration(p, errp);
254 } else if (strstart(uri, "unix:", &p)) {
255 unix_start_incoming_migration(p, errp);
256 } else if (strstart(uri, "fd:", &p)) {
257 fd_start_incoming_migration(p, errp);
258 } else {
259 error_setg(errp, "unknown migration protocol: %s", uri);
260 }
261 }
262
263 static void process_incoming_migration_bh(void *opaque)
264 {
265 Error *local_err = NULL;
266 MigrationIncomingState *mis = opaque;
267
268 /* Make sure all file formats flush their mutable metadata.
269 * If we get an error here, just don't restart the VM yet. */
270 bdrv_invalidate_cache_all(&local_err);
271 if (local_err) {
272 error_report_err(local_err);
273 local_err = NULL;
274 autostart = false;
275 }
276
277 /*
278 * This must happen after all error conditions are dealt with and
279 * we're sure the VM is going to be running on this host.
280 */
281 qemu_announce_self();
282
283 /* If global state section was not received or we are in running
284 state, we need to obey autostart. Any other state is set with
285 runstate_set. */
286
287 if (!global_state_received() ||
288 global_state_get_runstate() == RUN_STATE_RUNNING) {
289 if (autostart) {
290 vm_start();
291 } else {
292 runstate_set(RUN_STATE_PAUSED);
293 }
294 } else {
295 runstate_set(global_state_get_runstate());
296 }
297 /*
298 * This must happen after any state changes since as soon as an external
299 * observer sees this event they might start to prod at the VM assuming
300 * it's ready to use.
301 */
302 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
303 MIGRATION_STATUS_COMPLETED);
304 qemu_bh_delete(mis->bh);
305 migration_incoming_state_destroy();
306 }
307
308 static void process_incoming_migration_co(void *opaque)
309 {
310 QEMUFile *f = opaque;
311 MigrationIncomingState *mis = migration_incoming_get_current();
312 PostcopyState ps;
313 int ret;
314
315 mis->from_src_file = f;
316 mis->largest_page_size = qemu_ram_pagesize_largest();
317 postcopy_state_set(POSTCOPY_INCOMING_NONE);
318 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
319 MIGRATION_STATUS_ACTIVE);
320 ret = qemu_loadvm_state(f);
321
322 ps = postcopy_state_get();
323 trace_process_incoming_migration_co_end(ret, ps);
324 if (ps != POSTCOPY_INCOMING_NONE) {
325 if (ps == POSTCOPY_INCOMING_ADVISE) {
326 /*
327 * Where a migration had postcopy enabled (and thus went to advise)
328 * but managed to complete within the precopy period, we can use
329 * the normal exit.
330 */
331 postcopy_ram_incoming_cleanup(mis);
332 } else if (ret >= 0) {
333 /*
334 * Postcopy was started, cleanup should happen at the end of the
335 * postcopy thread.
336 */
337 trace_process_incoming_migration_co_postcopy_end_main();
338 return;
339 }
340 /* Else if something went wrong then just fall out of the normal exit */
341 }
342
343 /* we get COLO info, and know if we are in COLO mode */
344 if (!ret && migration_incoming_enable_colo()) {
345 mis->migration_incoming_co = qemu_coroutine_self();
346 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
347 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
348 mis->have_colo_incoming_thread = true;
349 qemu_coroutine_yield();
350
351 /* Wait checkpoint incoming thread exit before free resource */
352 qemu_thread_join(&mis->colo_incoming_thread);
353 }
354
355 if (ret < 0) {
356 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
357 MIGRATION_STATUS_FAILED);
358 error_report("load of migration failed: %s", strerror(-ret));
359 qemu_fclose(mis->from_src_file);
360 exit(EXIT_FAILURE);
361 }
362 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
363 qemu_bh_schedule(mis->bh);
364 }
365
366 void migration_fd_process_incoming(QEMUFile *f)
367 {
368 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, f);
369
370 qemu_file_set_blocking(f, false);
371 qemu_coroutine_enter(co);
372 }
373
374 /*
375 * Send a 'SHUT' message on the return channel with the given value
376 * to indicate that we've finished with the RP. Non-0 value indicates
377 * error.
378 */
379 void migrate_send_rp_shut(MigrationIncomingState *mis,
380 uint32_t value)
381 {
382 uint32_t buf;
383
384 buf = cpu_to_be32(value);
385 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
386 }
387
388 /*
389 * Send a 'PONG' message on the return channel with the given value
390 * (normally in response to a 'PING')
391 */
392 void migrate_send_rp_pong(MigrationIncomingState *mis,
393 uint32_t value)
394 {
395 uint32_t buf;
396
397 buf = cpu_to_be32(value);
398 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
399 }
400
401 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
402 {
403 MigrationCapabilityStatusList *head = NULL;
404 MigrationCapabilityStatusList *caps;
405 MigrationState *s = migrate_get_current();
406 int i;
407
408 caps = NULL; /* silence compiler warning */
409 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
410 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
411 if (i == MIGRATION_CAPABILITY_BLOCK) {
412 continue;
413 }
414 #endif
415 if (head == NULL) {
416 head = g_malloc0(sizeof(*caps));
417 caps = head;
418 } else {
419 caps->next = g_malloc0(sizeof(*caps));
420 caps = caps->next;
421 }
422 caps->value =
423 g_malloc(sizeof(*caps->value));
424 caps->value->capability = i;
425 caps->value->state = s->enabled_capabilities[i];
426 }
427
428 return head;
429 }
430
431 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
432 {
433 MigrationParameters *params;
434 MigrationState *s = migrate_get_current();
435
436 /* TODO use QAPI_CLONE() instead of duplicating it inline */
437 params = g_malloc0(sizeof(*params));
438 params->has_compress_level = true;
439 params->compress_level = s->parameters.compress_level;
440 params->has_compress_threads = true;
441 params->compress_threads = s->parameters.compress_threads;
442 params->has_decompress_threads = true;
443 params->decompress_threads = s->parameters.decompress_threads;
444 params->has_cpu_throttle_initial = true;
445 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
446 params->has_cpu_throttle_increment = true;
447 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
448 params->has_tls_creds = true;
449 params->tls_creds = g_strdup(s->parameters.tls_creds);
450 params->has_tls_hostname = true;
451 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
452 params->has_max_bandwidth = true;
453 params->max_bandwidth = s->parameters.max_bandwidth;
454 params->has_downtime_limit = true;
455 params->downtime_limit = s->parameters.downtime_limit;
456 params->has_x_checkpoint_delay = true;
457 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
458 params->has_block_incremental = true;
459 params->block_incremental = s->parameters.block_incremental;
460
461 return params;
462 }
463
464 /*
465 * Return true if we're already in the middle of a migration
466 * (i.e. any of the active or setup states)
467 */
468 static bool migration_is_setup_or_active(int state)
469 {
470 switch (state) {
471 case MIGRATION_STATUS_ACTIVE:
472 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
473 case MIGRATION_STATUS_SETUP:
474 return true;
475
476 default:
477 return false;
478
479 }
480 }
481
482 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
483 {
484 info->has_ram = true;
485 info->ram = g_malloc0(sizeof(*info->ram));
486 info->ram->transferred = ram_counters.transferred;
487 info->ram->total = ram_bytes_total();
488 info->ram->duplicate = ram_counters.duplicate;
489 /* legacy value. It is not used anymore */
490 info->ram->skipped = 0;
491 info->ram->normal = ram_counters.normal;
492 info->ram->normal_bytes = ram_counters.normal *
493 qemu_target_page_size();
494 info->ram->mbps = s->mbps;
495 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
496 info->ram->postcopy_requests = ram_counters.postcopy_requests;
497 info->ram->page_size = qemu_target_page_size();
498
499 if (migrate_use_xbzrle()) {
500 info->has_xbzrle_cache = true;
501 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
502 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
503 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
504 info->xbzrle_cache->pages = xbzrle_counters.pages;
505 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
506 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
507 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
508 }
509
510 if (cpu_throttle_active()) {
511 info->has_cpu_throttle_percentage = true;
512 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
513 }
514
515 if (s->state != MIGRATION_STATUS_COMPLETED) {
516 info->ram->remaining = ram_bytes_remaining();
517 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
518 }
519 }
520
521 static void populate_disk_info(MigrationInfo *info)
522 {
523 if (blk_mig_active()) {
524 info->has_disk = true;
525 info->disk = g_malloc0(sizeof(*info->disk));
526 info->disk->transferred = blk_mig_bytes_transferred();
527 info->disk->remaining = blk_mig_bytes_remaining();
528 info->disk->total = blk_mig_bytes_total();
529 }
530 }
531
532 MigrationInfo *qmp_query_migrate(Error **errp)
533 {
534 MigrationInfo *info = g_malloc0(sizeof(*info));
535 MigrationState *s = migrate_get_current();
536
537 switch (s->state) {
538 case MIGRATION_STATUS_NONE:
539 /* no migration has happened ever */
540 break;
541 case MIGRATION_STATUS_SETUP:
542 info->has_status = true;
543 info->has_total_time = false;
544 break;
545 case MIGRATION_STATUS_ACTIVE:
546 case MIGRATION_STATUS_CANCELLING:
547 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
548 /* TODO add some postcopy stats */
549 info->has_status = true;
550 info->has_total_time = true;
551 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
552 - s->total_time;
553 info->has_expected_downtime = true;
554 info->expected_downtime = s->expected_downtime;
555 info->has_setup_time = true;
556 info->setup_time = s->setup_time;
557
558 populate_ram_info(info, s);
559 populate_disk_info(info);
560 break;
561 case MIGRATION_STATUS_COLO:
562 info->has_status = true;
563 /* TODO: display COLO specific information (checkpoint info etc.) */
564 break;
565 case MIGRATION_STATUS_COMPLETED:
566 info->has_status = true;
567 info->has_total_time = true;
568 info->total_time = s->total_time;
569 info->has_downtime = true;
570 info->downtime = s->downtime;
571 info->has_setup_time = true;
572 info->setup_time = s->setup_time;
573
574 populate_ram_info(info, s);
575 break;
576 case MIGRATION_STATUS_FAILED:
577 info->has_status = true;
578 if (s->error) {
579 info->has_error_desc = true;
580 info->error_desc = g_strdup(error_get_pretty(s->error));
581 }
582 break;
583 case MIGRATION_STATUS_CANCELLED:
584 info->has_status = true;
585 break;
586 }
587 info->status = s->state;
588
589 return info;
590 }
591
592 /**
593 * @migration_caps_check - check capability validity
594 *
595 * @cap_list: old capability list, array of bool
596 * @params: new capabilities to be applied soon
597 * @errp: set *errp if the check failed, with reason
598 *
599 * Returns true if check passed, otherwise false.
600 */
601 static bool migrate_caps_check(bool *cap_list,
602 MigrationCapabilityStatusList *params,
603 Error **errp)
604 {
605 MigrationCapabilityStatusList *cap;
606 bool old_postcopy_cap;
607
608 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
609
610 for (cap = params; cap; cap = cap->next) {
611 cap_list[cap->value->capability] = cap->value->state;
612 }
613
614 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
615 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
616 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
617 "block migration");
618 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
619 return false;
620 }
621 #endif
622
623 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
624 if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
625 /* The decompression threads asynchronously write into RAM
626 * rather than use the atomic copies needed to avoid
627 * userfaulting. It should be possible to fix the decompression
628 * threads for compatibility in future.
629 */
630 error_setg(errp, "Postcopy is not currently compatible "
631 "with compression");
632 return false;
633 }
634
635 /* This check is reasonably expensive, so only when it's being
636 * set the first time, also it's only the destination that needs
637 * special support.
638 */
639 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
640 !postcopy_ram_supported_by_host()) {
641 /* postcopy_ram_supported_by_host will have emitted a more
642 * detailed message
643 */
644 error_setg(errp, "Postcopy is not supported");
645 return false;
646 }
647 }
648
649 return true;
650 }
651
652 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
653 Error **errp)
654 {
655 MigrationState *s = migrate_get_current();
656 MigrationCapabilityStatusList *cap;
657
658 if (migration_is_setup_or_active(s->state)) {
659 error_setg(errp, QERR_MIGRATION_ACTIVE);
660 return;
661 }
662
663 if (!migrate_caps_check(s->enabled_capabilities, params, errp)) {
664 return;
665 }
666
667 for (cap = params; cap; cap = cap->next) {
668 s->enabled_capabilities[cap->value->capability] = cap->value->state;
669 }
670 }
671
672 /*
673 * Check whether the parameters are valid. Error will be put into errp
674 * (if provided). Return true if valid, otherwise false.
675 */
676 static bool migrate_params_check(MigrationParameters *params, Error **errp)
677 {
678 if (params->has_compress_level &&
679 (params->compress_level < 0 || params->compress_level > 9)) {
680 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
681 "is invalid, it should be in the range of 0 to 9");
682 return false;
683 }
684
685 if (params->has_compress_threads &&
686 (params->compress_threads < 1 || params->compress_threads > 255)) {
687 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
688 "compress_threads",
689 "is invalid, it should be in the range of 1 to 255");
690 return false;
691 }
692
693 if (params->has_decompress_threads &&
694 (params->decompress_threads < 1 || params->decompress_threads > 255)) {
695 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
696 "decompress_threads",
697 "is invalid, it should be in the range of 1 to 255");
698 return false;
699 }
700
701 if (params->has_cpu_throttle_initial &&
702 (params->cpu_throttle_initial < 1 ||
703 params->cpu_throttle_initial > 99)) {
704 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
705 "cpu_throttle_initial",
706 "an integer in the range of 1 to 99");
707 return false;
708 }
709
710 if (params->has_cpu_throttle_increment &&
711 (params->cpu_throttle_increment < 1 ||
712 params->cpu_throttle_increment > 99)) {
713 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
714 "cpu_throttle_increment",
715 "an integer in the range of 1 to 99");
716 return false;
717 }
718
719 if (params->has_max_bandwidth &&
720 (params->max_bandwidth < 0 || params->max_bandwidth > SIZE_MAX)) {
721 error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
722 " range of 0 to %zu bytes/second", SIZE_MAX);
723 return false;
724 }
725
726 if (params->has_downtime_limit &&
727 (params->downtime_limit < 0 ||
728 params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
729 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
730 "the range of 0 to %d milliseconds",
731 MAX_MIGRATE_DOWNTIME);
732 return false;
733 }
734
735 if (params->has_x_checkpoint_delay && (params->x_checkpoint_delay < 0)) {
736 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
737 "x_checkpoint_delay",
738 "is invalid, it should be positive");
739 return false;
740 }
741
742 return true;
743 }
744
745 static void migrate_params_test_apply(MigrateSetParameters *params,
746 MigrationParameters *dest)
747 {
748 *dest = migrate_get_current()->parameters;
749
750 /* TODO use QAPI_CLONE() instead of duplicating it inline */
751
752 if (params->has_compress_level) {
753 dest->compress_level = params->compress_level;
754 }
755
756 if (params->has_compress_threads) {
757 dest->compress_threads = params->compress_threads;
758 }
759
760 if (params->has_decompress_threads) {
761 dest->decompress_threads = params->decompress_threads;
762 }
763
764 if (params->has_cpu_throttle_initial) {
765 dest->cpu_throttle_initial = params->cpu_throttle_initial;
766 }
767
768 if (params->has_cpu_throttle_increment) {
769 dest->cpu_throttle_increment = params->cpu_throttle_increment;
770 }
771
772 if (params->has_tls_creds) {
773 assert(params->tls_creds->type == QTYPE_QSTRING);
774 dest->tls_creds = g_strdup(params->tls_creds->u.s);
775 }
776
777 if (params->has_tls_hostname) {
778 assert(params->tls_hostname->type == QTYPE_QSTRING);
779 dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
780 }
781
782 if (params->has_max_bandwidth) {
783 dest->max_bandwidth = params->max_bandwidth;
784 }
785
786 if (params->has_downtime_limit) {
787 dest->downtime_limit = params->downtime_limit;
788 }
789
790 if (params->has_x_checkpoint_delay) {
791 dest->x_checkpoint_delay = params->x_checkpoint_delay;
792 }
793
794 if (params->has_block_incremental) {
795 dest->block_incremental = params->block_incremental;
796 }
797 }
798
799 static void migrate_params_apply(MigrateSetParameters *params)
800 {
801 MigrationState *s = migrate_get_current();
802
803 /* TODO use QAPI_CLONE() instead of duplicating it inline */
804
805 if (params->has_compress_level) {
806 s->parameters.compress_level = params->compress_level;
807 }
808
809 if (params->has_compress_threads) {
810 s->parameters.compress_threads = params->compress_threads;
811 }
812
813 if (params->has_decompress_threads) {
814 s->parameters.decompress_threads = params->decompress_threads;
815 }
816
817 if (params->has_cpu_throttle_initial) {
818 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
819 }
820
821 if (params->has_cpu_throttle_increment) {
822 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
823 }
824
825 if (params->has_tls_creds) {
826 g_free(s->parameters.tls_creds);
827 assert(params->tls_creds->type == QTYPE_QSTRING);
828 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
829 }
830
831 if (params->has_tls_hostname) {
832 g_free(s->parameters.tls_hostname);
833 assert(params->tls_hostname->type == QTYPE_QSTRING);
834 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
835 }
836
837 if (params->has_max_bandwidth) {
838 s->parameters.max_bandwidth = params->max_bandwidth;
839 if (s->to_dst_file) {
840 qemu_file_set_rate_limit(s->to_dst_file,
841 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
842 }
843 }
844
845 if (params->has_downtime_limit) {
846 s->parameters.downtime_limit = params->downtime_limit;
847 }
848
849 if (params->has_x_checkpoint_delay) {
850 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
851 if (migration_in_colo_state()) {
852 colo_checkpoint_notify(s);
853 }
854 }
855
856 if (params->has_block_incremental) {
857 s->parameters.block_incremental = params->block_incremental;
858 }
859 }
860
861 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
862 {
863 MigrationParameters tmp;
864
865 /* TODO Rewrite "" to null instead */
866 if (params->has_tls_creds
867 && params->tls_creds->type == QTYPE_QNULL) {
868 QDECREF(params->tls_creds->u.n);
869 params->tls_creds->type = QTYPE_QSTRING;
870 params->tls_creds->u.s = strdup("");
871 }
872 /* TODO Rewrite "" to null instead */
873 if (params->has_tls_hostname
874 && params->tls_hostname->type == QTYPE_QNULL) {
875 QDECREF(params->tls_hostname->u.n);
876 params->tls_hostname->type = QTYPE_QSTRING;
877 params->tls_hostname->u.s = strdup("");
878 }
879
880 migrate_params_test_apply(params, &tmp);
881
882 if (!migrate_params_check(&tmp, errp)) {
883 /* Invalid parameter */
884 return;
885 }
886
887 migrate_params_apply(params);
888 }
889
890
891 void qmp_migrate_start_postcopy(Error **errp)
892 {
893 MigrationState *s = migrate_get_current();
894
895 if (!migrate_postcopy_ram()) {
896 error_setg(errp, "Enable postcopy with migrate_set_capability before"
897 " the start of migration");
898 return;
899 }
900
901 if (s->state == MIGRATION_STATUS_NONE) {
902 error_setg(errp, "Postcopy must be started after migration has been"
903 " started");
904 return;
905 }
906 /*
907 * we don't error if migration has finished since that would be racy
908 * with issuing this command.
909 */
910 atomic_set(&s->start_postcopy, true);
911 }
912
913 /* shared migration helpers */
914
915 void migrate_set_state(int *state, int old_state, int new_state)
916 {
917 if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
918 trace_migrate_set_state(new_state);
919 migrate_generate_event(new_state);
920 }
921 }
922
923 static MigrationCapabilityStatusList *migrate_cap_add(
924 MigrationCapabilityStatusList *list,
925 MigrationCapability index,
926 bool state)
927 {
928 MigrationCapabilityStatusList *cap;
929
930 cap = g_new0(MigrationCapabilityStatusList, 1);
931 cap->value = g_new0(MigrationCapabilityStatus, 1);
932 cap->value->capability = index;
933 cap->value->state = state;
934 cap->next = list;
935
936 return cap;
937 }
938
939 void migrate_set_block_enabled(bool value, Error **errp)
940 {
941 MigrationCapabilityStatusList *cap;
942
943 cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
944 qmp_migrate_set_capabilities(cap, errp);
945 qapi_free_MigrationCapabilityStatusList(cap);
946 }
947
948 static void migrate_set_block_incremental(MigrationState *s, bool value)
949 {
950 s->parameters.block_incremental = value;
951 }
952
953 static void block_cleanup_parameters(MigrationState *s)
954 {
955 if (s->must_remove_block_options) {
956 /* setting to false can never fail */
957 migrate_set_block_enabled(false, &error_abort);
958 migrate_set_block_incremental(s, false);
959 s->must_remove_block_options = false;
960 }
961 }
962
963 static void migrate_fd_cleanup(void *opaque)
964 {
965 MigrationState *s = opaque;
966
967 qemu_bh_delete(s->cleanup_bh);
968 s->cleanup_bh = NULL;
969
970 if (s->to_dst_file) {
971 trace_migrate_fd_cleanup();
972 qemu_mutex_unlock_iothread();
973 if (s->migration_thread_running) {
974 qemu_thread_join(&s->thread);
975 s->migration_thread_running = false;
976 }
977 qemu_mutex_lock_iothread();
978
979 qemu_fclose(s->to_dst_file);
980 s->to_dst_file = NULL;
981 }
982
983 assert((s->state != MIGRATION_STATUS_ACTIVE) &&
984 (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
985
986 if (s->state == MIGRATION_STATUS_CANCELLING) {
987 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
988 MIGRATION_STATUS_CANCELLED);
989 }
990
991 notifier_list_notify(&migration_state_notifiers, s);
992 block_cleanup_parameters(s);
993 }
994
995 void migrate_fd_error(MigrationState *s, const Error *error)
996 {
997 trace_migrate_fd_error(error_get_pretty(error));
998 assert(s->to_dst_file == NULL);
999 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1000 MIGRATION_STATUS_FAILED);
1001 if (!s->error) {
1002 s->error = error_copy(error);
1003 }
1004 notifier_list_notify(&migration_state_notifiers, s);
1005 block_cleanup_parameters(s);
1006 }
1007
1008 static void migrate_fd_cancel(MigrationState *s)
1009 {
1010 int old_state ;
1011 QEMUFile *f = migrate_get_current()->to_dst_file;
1012 trace_migrate_fd_cancel();
1013
1014 if (s->rp_state.from_dst_file) {
1015 /* shutdown the rp socket, so causing the rp thread to shutdown */
1016 qemu_file_shutdown(s->rp_state.from_dst_file);
1017 }
1018
1019 do {
1020 old_state = s->state;
1021 if (!migration_is_setup_or_active(old_state)) {
1022 break;
1023 }
1024 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1025 } while (s->state != MIGRATION_STATUS_CANCELLING);
1026
1027 /*
1028 * If we're unlucky the migration code might be stuck somewhere in a
1029 * send/write while the network has failed and is waiting to timeout;
1030 * if we've got shutdown(2) available then we can force it to quit.
1031 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1032 * called in a bh, so there is no race against this cancel.
1033 */
1034 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1035 qemu_file_shutdown(f);
1036 }
1037 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1038 Error *local_err = NULL;
1039
1040 bdrv_invalidate_cache_all(&local_err);
1041 if (local_err) {
1042 error_report_err(local_err);
1043 } else {
1044 s->block_inactive = false;
1045 }
1046 }
1047 block_cleanup_parameters(s);
1048 }
1049
1050 void add_migration_state_change_notifier(Notifier *notify)
1051 {
1052 notifier_list_add(&migration_state_notifiers, notify);
1053 }
1054
1055 void remove_migration_state_change_notifier(Notifier *notify)
1056 {
1057 notifier_remove(notify);
1058 }
1059
1060 bool migration_in_setup(MigrationState *s)
1061 {
1062 return s->state == MIGRATION_STATUS_SETUP;
1063 }
1064
1065 bool migration_has_finished(MigrationState *s)
1066 {
1067 return s->state == MIGRATION_STATUS_COMPLETED;
1068 }
1069
1070 bool migration_has_failed(MigrationState *s)
1071 {
1072 return (s->state == MIGRATION_STATUS_CANCELLED ||
1073 s->state == MIGRATION_STATUS_FAILED);
1074 }
1075
1076 bool migration_in_postcopy(void)
1077 {
1078 MigrationState *s = migrate_get_current();
1079
1080 return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1081 }
1082
1083 bool migration_in_postcopy_after_devices(MigrationState *s)
1084 {
1085 return migration_in_postcopy() && s->postcopy_after_devices;
1086 }
1087
1088 bool migration_is_idle(void)
1089 {
1090 MigrationState *s = migrate_get_current();
1091
1092 switch (s->state) {
1093 case MIGRATION_STATUS_NONE:
1094 case MIGRATION_STATUS_CANCELLED:
1095 case MIGRATION_STATUS_COMPLETED:
1096 case MIGRATION_STATUS_FAILED:
1097 return true;
1098 case MIGRATION_STATUS_SETUP:
1099 case MIGRATION_STATUS_CANCELLING:
1100 case MIGRATION_STATUS_ACTIVE:
1101 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1102 case MIGRATION_STATUS_COLO:
1103 return false;
1104 case MIGRATION_STATUS__MAX:
1105 g_assert_not_reached();
1106 }
1107
1108 return false;
1109 }
1110
1111 MigrationState *migrate_init(void)
1112 {
1113 MigrationState *s = migrate_get_current();
1114
1115 /*
1116 * Reinitialise all migration state, except
1117 * parameters/capabilities that the user set, and
1118 * locks.
1119 */
1120 s->bytes_xfer = 0;
1121 s->xfer_limit = 0;
1122 s->cleanup_bh = 0;
1123 s->to_dst_file = NULL;
1124 s->state = MIGRATION_STATUS_NONE;
1125 s->rp_state.from_dst_file = NULL;
1126 s->rp_state.error = false;
1127 s->mbps = 0.0;
1128 s->downtime = 0;
1129 s->expected_downtime = 0;
1130 s->setup_time = 0;
1131 s->start_postcopy = false;
1132 s->postcopy_after_devices = false;
1133 s->migration_thread_running = false;
1134 error_free(s->error);
1135 s->error = NULL;
1136
1137 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1138
1139 s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1140 return s;
1141 }
1142
1143 static GSList *migration_blockers;
1144
1145 int migrate_add_blocker(Error *reason, Error **errp)
1146 {
1147 if (migrate_get_current()->only_migratable) {
1148 error_propagate(errp, error_copy(reason));
1149 error_prepend(errp, "disallowing migration blocker "
1150 "(--only_migratable) for: ");
1151 return -EACCES;
1152 }
1153
1154 if (migration_is_idle()) {
1155 migration_blockers = g_slist_prepend(migration_blockers, reason);
1156 return 0;
1157 }
1158
1159 error_propagate(errp, error_copy(reason));
1160 error_prepend(errp, "disallowing migration blocker (migration in "
1161 "progress) for: ");
1162 return -EBUSY;
1163 }
1164
1165 void migrate_del_blocker(Error *reason)
1166 {
1167 migration_blockers = g_slist_remove(migration_blockers, reason);
1168 }
1169
1170 void qmp_migrate_incoming(const char *uri, Error **errp)
1171 {
1172 Error *local_err = NULL;
1173 static bool once = true;
1174
1175 if (!deferred_incoming) {
1176 error_setg(errp, "For use with '-incoming defer'");
1177 return;
1178 }
1179 if (!once) {
1180 error_setg(errp, "The incoming migration has already been started");
1181 }
1182
1183 qemu_start_incoming_migration(uri, &local_err);
1184
1185 if (local_err) {
1186 error_propagate(errp, local_err);
1187 return;
1188 }
1189
1190 once = false;
1191 }
1192
1193 bool migration_is_blocked(Error **errp)
1194 {
1195 if (qemu_savevm_state_blocked(errp)) {
1196 return true;
1197 }
1198
1199 if (migration_blockers) {
1200 error_propagate(errp, error_copy(migration_blockers->data));
1201 return true;
1202 }
1203
1204 return false;
1205 }
1206
1207 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1208 bool has_inc, bool inc, bool has_detach, bool detach,
1209 Error **errp)
1210 {
1211 Error *local_err = NULL;
1212 MigrationState *s = migrate_get_current();
1213 const char *p;
1214
1215 if (migration_is_setup_or_active(s->state) ||
1216 s->state == MIGRATION_STATUS_CANCELLING ||
1217 s->state == MIGRATION_STATUS_COLO) {
1218 error_setg(errp, QERR_MIGRATION_ACTIVE);
1219 return;
1220 }
1221 if (runstate_check(RUN_STATE_INMIGRATE)) {
1222 error_setg(errp, "Guest is waiting for an incoming migration");
1223 return;
1224 }
1225
1226 if (migration_is_blocked(errp)) {
1227 return;
1228 }
1229
1230 if ((has_blk && blk) || (has_inc && inc)) {
1231 if (migrate_use_block() || migrate_use_block_incremental()) {
1232 error_setg(errp, "Command options are incompatible with "
1233 "current migration capabilities");
1234 return;
1235 }
1236 migrate_set_block_enabled(true, &local_err);
1237 if (local_err) {
1238 error_propagate(errp, local_err);
1239 return;
1240 }
1241 s->must_remove_block_options = true;
1242 }
1243
1244 if (has_inc && inc) {
1245 migrate_set_block_incremental(s, true);
1246 }
1247
1248 s = migrate_init();
1249
1250 if (strstart(uri, "tcp:", &p)) {
1251 tcp_start_outgoing_migration(s, p, &local_err);
1252 #ifdef CONFIG_RDMA
1253 } else if (strstart(uri, "rdma:", &p)) {
1254 rdma_start_outgoing_migration(s, p, &local_err);
1255 #endif
1256 } else if (strstart(uri, "exec:", &p)) {
1257 exec_start_outgoing_migration(s, p, &local_err);
1258 } else if (strstart(uri, "unix:", &p)) {
1259 unix_start_outgoing_migration(s, p, &local_err);
1260 } else if (strstart(uri, "fd:", &p)) {
1261 fd_start_outgoing_migration(s, p, &local_err);
1262 } else {
1263 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1264 "a valid migration protocol");
1265 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1266 MIGRATION_STATUS_FAILED);
1267 return;
1268 }
1269
1270 if (local_err) {
1271 migrate_fd_error(s, local_err);
1272 error_propagate(errp, local_err);
1273 return;
1274 }
1275 }
1276
1277 void qmp_migrate_cancel(Error **errp)
1278 {
1279 migrate_fd_cancel(migrate_get_current());
1280 }
1281
1282 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1283 {
1284 MigrationState *s = migrate_get_current();
1285 int64_t new_size;
1286
1287 /* Check for truncation */
1288 if (value != (size_t)value) {
1289 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1290 "exceeding address space");
1291 return;
1292 }
1293
1294 /* Cache should not be larger than guest ram size */
1295 if (value > ram_bytes_total()) {
1296 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1297 "exceeds guest ram size ");
1298 return;
1299 }
1300
1301 new_size = xbzrle_cache_resize(value);
1302 if (new_size < 0) {
1303 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1304 "is smaller than page size");
1305 return;
1306 }
1307
1308 s->xbzrle_cache_size = new_size;
1309 }
1310
1311 int64_t qmp_query_migrate_cache_size(Error **errp)
1312 {
1313 return migrate_xbzrle_cache_size();
1314 }
1315
1316 void qmp_migrate_set_speed(int64_t value, Error **errp)
1317 {
1318 MigrateSetParameters p = {
1319 .has_max_bandwidth = true,
1320 .max_bandwidth = value,
1321 };
1322
1323 qmp_migrate_set_parameters(&p, errp);
1324 }
1325
1326 void qmp_migrate_set_downtime(double value, Error **errp)
1327 {
1328 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1329 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1330 "the range of 0 to %d seconds",
1331 MAX_MIGRATE_DOWNTIME_SECONDS);
1332 return;
1333 }
1334
1335 value *= 1000; /* Convert to milliseconds */
1336 value = MAX(0, MIN(INT64_MAX, value));
1337
1338 MigrateSetParameters p = {
1339 .has_downtime_limit = true,
1340 .downtime_limit = value,
1341 };
1342
1343 qmp_migrate_set_parameters(&p, errp);
1344 }
1345
1346 bool migrate_release_ram(void)
1347 {
1348 MigrationState *s;
1349
1350 s = migrate_get_current();
1351
1352 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1353 }
1354
1355 bool migrate_postcopy_ram(void)
1356 {
1357 MigrationState *s;
1358
1359 s = migrate_get_current();
1360
1361 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1362 }
1363
1364 bool migrate_auto_converge(void)
1365 {
1366 MigrationState *s;
1367
1368 s = migrate_get_current();
1369
1370 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1371 }
1372
1373 bool migrate_zero_blocks(void)
1374 {
1375 MigrationState *s;
1376
1377 s = migrate_get_current();
1378
1379 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1380 }
1381
1382 bool migrate_use_compression(void)
1383 {
1384 MigrationState *s;
1385
1386 s = migrate_get_current();
1387
1388 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
1389 }
1390
1391 int migrate_compress_level(void)
1392 {
1393 MigrationState *s;
1394
1395 s = migrate_get_current();
1396
1397 return s->parameters.compress_level;
1398 }
1399
1400 int migrate_compress_threads(void)
1401 {
1402 MigrationState *s;
1403
1404 s = migrate_get_current();
1405
1406 return s->parameters.compress_threads;
1407 }
1408
1409 int migrate_decompress_threads(void)
1410 {
1411 MigrationState *s;
1412
1413 s = migrate_get_current();
1414
1415 return s->parameters.decompress_threads;
1416 }
1417
1418 bool migrate_use_events(void)
1419 {
1420 MigrationState *s;
1421
1422 s = migrate_get_current();
1423
1424 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1425 }
1426
1427 int migrate_use_xbzrle(void)
1428 {
1429 MigrationState *s;
1430
1431 s = migrate_get_current();
1432
1433 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1434 }
1435
1436 int64_t migrate_xbzrle_cache_size(void)
1437 {
1438 MigrationState *s;
1439
1440 s = migrate_get_current();
1441
1442 return s->xbzrle_cache_size;
1443 }
1444
1445 bool migrate_use_block(void)
1446 {
1447 MigrationState *s;
1448
1449 s = migrate_get_current();
1450
1451 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
1452 }
1453
1454 bool migrate_use_return_path(void)
1455 {
1456 MigrationState *s;
1457
1458 s = migrate_get_current();
1459
1460 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
1461 }
1462
1463 bool migrate_use_block_incremental(void)
1464 {
1465 MigrationState *s;
1466
1467 s = migrate_get_current();
1468
1469 return s->parameters.block_incremental;
1470 }
1471
1472 /* migration thread support */
1473 /*
1474 * Something bad happened to the RP stream, mark an error
1475 * The caller shall print or trace something to indicate why
1476 */
1477 static void mark_source_rp_bad(MigrationState *s)
1478 {
1479 s->rp_state.error = true;
1480 }
1481
1482 static struct rp_cmd_args {
1483 ssize_t len; /* -1 = variable */
1484 const char *name;
1485 } rp_cmd_args[] = {
1486 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
1487 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
1488 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
1489 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
1490 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
1491 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
1492 };
1493
1494 /*
1495 * Process a request for pages received on the return path,
1496 * We're allowed to send more than requested (e.g. to round to our page size)
1497 * and we don't need to send pages that have already been sent.
1498 */
1499 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1500 ram_addr_t start, size_t len)
1501 {
1502 long our_host_ps = getpagesize();
1503
1504 trace_migrate_handle_rp_req_pages(rbname, start, len);
1505
1506 /*
1507 * Since we currently insist on matching page sizes, just sanity check
1508 * we're being asked for whole host pages.
1509 */
1510 if (start & (our_host_ps-1) ||
1511 (len & (our_host_ps-1))) {
1512 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1513 " len: %zd", __func__, start, len);
1514 mark_source_rp_bad(ms);
1515 return;
1516 }
1517
1518 if (ram_save_queue_pages(rbname, start, len)) {
1519 mark_source_rp_bad(ms);
1520 }
1521 }
1522
1523 /*
1524 * Handles messages sent on the return path towards the source VM
1525 *
1526 */
1527 static void *source_return_path_thread(void *opaque)
1528 {
1529 MigrationState *ms = opaque;
1530 QEMUFile *rp = ms->rp_state.from_dst_file;
1531 uint16_t header_len, header_type;
1532 uint8_t buf[512];
1533 uint32_t tmp32, sibling_error;
1534 ram_addr_t start = 0; /* =0 to silence warning */
1535 size_t len = 0, expected_len;
1536 int res;
1537
1538 trace_source_return_path_thread_entry();
1539 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
1540 migration_is_setup_or_active(ms->state)) {
1541 trace_source_return_path_thread_loop_top();
1542 header_type = qemu_get_be16(rp);
1543 header_len = qemu_get_be16(rp);
1544
1545 if (header_type >= MIG_RP_MSG_MAX ||
1546 header_type == MIG_RP_MSG_INVALID) {
1547 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1548 header_type, header_len);
1549 mark_source_rp_bad(ms);
1550 goto out;
1551 }
1552
1553 if ((rp_cmd_args[header_type].len != -1 &&
1554 header_len != rp_cmd_args[header_type].len) ||
1555 header_len > sizeof(buf)) {
1556 error_report("RP: Received '%s' message (0x%04x) with"
1557 "incorrect length %d expecting %zu",
1558 rp_cmd_args[header_type].name, header_type, header_len,
1559 (size_t)rp_cmd_args[header_type].len);
1560 mark_source_rp_bad(ms);
1561 goto out;
1562 }
1563
1564 /* We know we've got a valid header by this point */
1565 res = qemu_get_buffer(rp, buf, header_len);
1566 if (res != header_len) {
1567 error_report("RP: Failed reading data for message 0x%04x"
1568 " read %d expected %d",
1569 header_type, res, header_len);
1570 mark_source_rp_bad(ms);
1571 goto out;
1572 }
1573
1574 /* OK, we have the message and the data */
1575 switch (header_type) {
1576 case MIG_RP_MSG_SHUT:
1577 sibling_error = ldl_be_p(buf);
1578 trace_source_return_path_thread_shut(sibling_error);
1579 if (sibling_error) {
1580 error_report("RP: Sibling indicated error %d", sibling_error);
1581 mark_source_rp_bad(ms);
1582 }
1583 /*
1584 * We'll let the main thread deal with closing the RP
1585 * we could do a shutdown(2) on it, but we're the only user
1586 * anyway, so there's nothing gained.
1587 */
1588 goto out;
1589
1590 case MIG_RP_MSG_PONG:
1591 tmp32 = ldl_be_p(buf);
1592 trace_source_return_path_thread_pong(tmp32);
1593 break;
1594
1595 case MIG_RP_MSG_REQ_PAGES:
1596 start = ldq_be_p(buf);
1597 len = ldl_be_p(buf + 8);
1598 migrate_handle_rp_req_pages(ms, NULL, start, len);
1599 break;
1600
1601 case MIG_RP_MSG_REQ_PAGES_ID:
1602 expected_len = 12 + 1; /* header + termination */
1603
1604 if (header_len >= expected_len) {
1605 start = ldq_be_p(buf);
1606 len = ldl_be_p(buf + 8);
1607 /* Now we expect an idstr */
1608 tmp32 = buf[12]; /* Length of the following idstr */
1609 buf[13 + tmp32] = '\0';
1610 expected_len += tmp32;
1611 }
1612 if (header_len != expected_len) {
1613 error_report("RP: Req_Page_id with length %d expecting %zd",
1614 header_len, expected_len);
1615 mark_source_rp_bad(ms);
1616 goto out;
1617 }
1618 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
1619 break;
1620
1621 default:
1622 break;
1623 }
1624 }
1625 if (qemu_file_get_error(rp)) {
1626 trace_source_return_path_thread_bad_end();
1627 mark_source_rp_bad(ms);
1628 }
1629
1630 trace_source_return_path_thread_end();
1631 out:
1632 ms->rp_state.from_dst_file = NULL;
1633 qemu_fclose(rp);
1634 return NULL;
1635 }
1636
1637 static int open_return_path_on_source(MigrationState *ms)
1638 {
1639
1640 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
1641 if (!ms->rp_state.from_dst_file) {
1642 return -1;
1643 }
1644
1645 trace_open_return_path_on_source();
1646 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
1647 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
1648
1649 trace_open_return_path_on_source_continue();
1650
1651 return 0;
1652 }
1653
1654 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1655 static int await_return_path_close_on_source(MigrationState *ms)
1656 {
1657 /*
1658 * If this is a normal exit then the destination will send a SHUT and the
1659 * rp_thread will exit, however if there's an error we need to cause
1660 * it to exit.
1661 */
1662 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
1663 /*
1664 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1665 * waiting for the destination.
1666 */
1667 qemu_file_shutdown(ms->rp_state.from_dst_file);
1668 mark_source_rp_bad(ms);
1669 }
1670 trace_await_return_path_close_on_source_joining();
1671 qemu_thread_join(&ms->rp_state.rp_thread);
1672 trace_await_return_path_close_on_source_close();
1673 return ms->rp_state.error;
1674 }
1675
1676 /*
1677 * Switch from normal iteration to postcopy
1678 * Returns non-0 on error
1679 */
1680 static int postcopy_start(MigrationState *ms, bool *old_vm_running)
1681 {
1682 int ret;
1683 QIOChannelBuffer *bioc;
1684 QEMUFile *fb;
1685 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1686 bool restart_block = false;
1687 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
1688 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1689
1690 trace_postcopy_start();
1691 qemu_mutex_lock_iothread();
1692 trace_postcopy_start_set_run();
1693
1694 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1695 *old_vm_running = runstate_is_running();
1696 global_state_store();
1697 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1698 if (ret < 0) {
1699 goto fail;
1700 }
1701
1702 ret = bdrv_inactivate_all();
1703 if (ret < 0) {
1704 goto fail;
1705 }
1706 restart_block = true;
1707
1708 /*
1709 * Cause any non-postcopiable, but iterative devices to
1710 * send out their final data.
1711 */
1712 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
1713
1714 /*
1715 * in Finish migrate and with the io-lock held everything should
1716 * be quiet, but we've potentially still got dirty pages and we
1717 * need to tell the destination to throw any pages it's already received
1718 * that are dirty
1719 */
1720 if (ram_postcopy_send_discard_bitmap(ms)) {
1721 error_report("postcopy send discard bitmap failed");
1722 goto fail;
1723 }
1724
1725 /*
1726 * send rest of state - note things that are doing postcopy
1727 * will notice we're in POSTCOPY_ACTIVE and not actually
1728 * wrap their state up here
1729 */
1730 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
1731 /* Ping just for debugging, helps line traces up */
1732 qemu_savevm_send_ping(ms->to_dst_file, 2);
1733
1734 /*
1735 * While loading the device state we may trigger page transfer
1736 * requests and the fd must be free to process those, and thus
1737 * the destination must read the whole device state off the fd before
1738 * it starts processing it. Unfortunately the ad-hoc migration format
1739 * doesn't allow the destination to know the size to read without fully
1740 * parsing it through each devices load-state code (especially the open
1741 * coded devices that use get/put).
1742 * So we wrap the device state up in a package with a length at the start;
1743 * to do this we use a qemu_buf to hold the whole of the device state.
1744 */
1745 bioc = qio_channel_buffer_new(4096);
1746 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
1747 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
1748 object_unref(OBJECT(bioc));
1749
1750 /*
1751 * Make sure the receiver can get incoming pages before we send the rest
1752 * of the state
1753 */
1754 qemu_savevm_send_postcopy_listen(fb);
1755
1756 qemu_savevm_state_complete_precopy(fb, false, false);
1757 qemu_savevm_send_ping(fb, 3);
1758
1759 qemu_savevm_send_postcopy_run(fb);
1760
1761 /* <><> end of stuff going into the package */
1762
1763 /* Last point of recovery; as soon as we send the package the destination
1764 * can open devices and potentially start running.
1765 * Lets just check again we've not got any errors.
1766 */
1767 ret = qemu_file_get_error(ms->to_dst_file);
1768 if (ret) {
1769 error_report("postcopy_start: Migration stream errored (pre package)");
1770 goto fail_closefb;
1771 }
1772
1773 restart_block = false;
1774
1775 /* Now send that blob */
1776 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
1777 goto fail_closefb;
1778 }
1779 qemu_fclose(fb);
1780
1781 /* Send a notify to give a chance for anything that needs to happen
1782 * at the transition to postcopy and after the device state; in particular
1783 * spice needs to trigger a transition now
1784 */
1785 ms->postcopy_after_devices = true;
1786 notifier_list_notify(&migration_state_notifiers, ms);
1787
1788 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
1789
1790 qemu_mutex_unlock_iothread();
1791
1792 /*
1793 * Although this ping is just for debug, it could potentially be
1794 * used for getting a better measurement of downtime at the source.
1795 */
1796 qemu_savevm_send_ping(ms->to_dst_file, 4);
1797
1798 if (migrate_release_ram()) {
1799 ram_postcopy_migrated_memory_release(ms);
1800 }
1801
1802 ret = qemu_file_get_error(ms->to_dst_file);
1803 if (ret) {
1804 error_report("postcopy_start: Migration stream errored");
1805 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1806 MIGRATION_STATUS_FAILED);
1807 }
1808
1809 return ret;
1810
1811 fail_closefb:
1812 qemu_fclose(fb);
1813 fail:
1814 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1815 MIGRATION_STATUS_FAILED);
1816 if (restart_block) {
1817 /* A failure happened early enough that we know the destination hasn't
1818 * accessed block devices, so we're safe to recover.
1819 */
1820 Error *local_err = NULL;
1821
1822 bdrv_invalidate_cache_all(&local_err);
1823 if (local_err) {
1824 error_report_err(local_err);
1825 }
1826 }
1827 qemu_mutex_unlock_iothread();
1828 return -1;
1829 }
1830
1831 /**
1832 * migration_completion: Used by migration_thread when there's not much left.
1833 * The caller 'breaks' the loop when this returns.
1834 *
1835 * @s: Current migration state
1836 * @current_active_state: The migration state we expect to be in
1837 * @*old_vm_running: Pointer to old_vm_running flag
1838 * @*start_time: Pointer to time to update
1839 */
1840 static void migration_completion(MigrationState *s, int current_active_state,
1841 bool *old_vm_running,
1842 int64_t *start_time)
1843 {
1844 int ret;
1845
1846 if (s->state == MIGRATION_STATUS_ACTIVE) {
1847 qemu_mutex_lock_iothread();
1848 *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1849 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1850 *old_vm_running = runstate_is_running();
1851 ret = global_state_store();
1852
1853 if (!ret) {
1854 bool inactivate = !migrate_colo_enabled();
1855 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1856 if (ret >= 0) {
1857 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
1858 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
1859 inactivate);
1860 }
1861 if (inactivate && ret >= 0) {
1862 s->block_inactive = true;
1863 }
1864 }
1865 qemu_mutex_unlock_iothread();
1866
1867 if (ret < 0) {
1868 goto fail;
1869 }
1870 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1871 trace_migration_completion_postcopy_end();
1872
1873 qemu_savevm_state_complete_postcopy(s->to_dst_file);
1874 trace_migration_completion_postcopy_end_after_complete();
1875 }
1876
1877 /*
1878 * If rp was opened we must clean up the thread before
1879 * cleaning everything else up (since if there are no failures
1880 * it will wait for the destination to send it's status in
1881 * a SHUT command).
1882 */
1883 if (s->rp_state.from_dst_file) {
1884 int rp_error;
1885 trace_migration_return_path_end_before();
1886 rp_error = await_return_path_close_on_source(s);
1887 trace_migration_return_path_end_after(rp_error);
1888 if (rp_error) {
1889 goto fail_invalidate;
1890 }
1891 }
1892
1893 if (qemu_file_get_error(s->to_dst_file)) {
1894 trace_migration_completion_file_err();
1895 goto fail_invalidate;
1896 }
1897
1898 if (!migrate_colo_enabled()) {
1899 migrate_set_state(&s->state, current_active_state,
1900 MIGRATION_STATUS_COMPLETED);
1901 }
1902
1903 return;
1904
1905 fail_invalidate:
1906 /* If not doing postcopy, vm_start() will be called: let's regain
1907 * control on images.
1908 */
1909 if (s->state == MIGRATION_STATUS_ACTIVE) {
1910 Error *local_err = NULL;
1911
1912 qemu_mutex_lock_iothread();
1913 bdrv_invalidate_cache_all(&local_err);
1914 if (local_err) {
1915 error_report_err(local_err);
1916 } else {
1917 s->block_inactive = false;
1918 }
1919 qemu_mutex_unlock_iothread();
1920 }
1921
1922 fail:
1923 migrate_set_state(&s->state, current_active_state,
1924 MIGRATION_STATUS_FAILED);
1925 }
1926
1927 bool migrate_colo_enabled(void)
1928 {
1929 MigrationState *s = migrate_get_current();
1930 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
1931 }
1932
1933 /*
1934 * Master migration thread on the source VM.
1935 * It drives the migration and pumps the data down the outgoing channel.
1936 */
1937 static void *migration_thread(void *opaque)
1938 {
1939 MigrationState *s = opaque;
1940 /* Used by the bandwidth calcs, updated later */
1941 int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1942 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
1943 int64_t initial_bytes = 0;
1944 /*
1945 * The final stage happens when the remaining data is smaller than
1946 * this threshold; it's calculated from the requested downtime and
1947 * measured bandwidth
1948 */
1949 int64_t threshold_size = 0;
1950 int64_t start_time = initial_time;
1951 int64_t end_time;
1952 bool old_vm_running = false;
1953 bool entered_postcopy = false;
1954 /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
1955 enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE;
1956 bool enable_colo = migrate_colo_enabled();
1957
1958 rcu_register_thread();
1959
1960 qemu_savevm_state_header(s->to_dst_file);
1961
1962 /*
1963 * If we opened the return path, we need to make sure dst has it
1964 * opened as well.
1965 */
1966 if (s->rp_state.from_dst_file) {
1967 /* Now tell the dest that it should open its end so it can reply */
1968 qemu_savevm_send_open_return_path(s->to_dst_file);
1969
1970 /* And do a ping that will make stuff easier to debug */
1971 qemu_savevm_send_ping(s->to_dst_file, 1);
1972 }
1973
1974 if (migrate_postcopy_ram()) {
1975 /*
1976 * Tell the destination that we *might* want to do postcopy later;
1977 * if the other end can't do postcopy it should fail now, nice and
1978 * early.
1979 */
1980 qemu_savevm_send_postcopy_advise(s->to_dst_file);
1981 }
1982
1983 qemu_savevm_state_setup(s->to_dst_file);
1984
1985 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
1986 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1987 MIGRATION_STATUS_ACTIVE);
1988
1989 trace_migration_thread_setup_complete();
1990
1991 while (s->state == MIGRATION_STATUS_ACTIVE ||
1992 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1993 int64_t current_time;
1994 uint64_t pending_size;
1995
1996 if (!qemu_file_rate_limit(s->to_dst_file)) {
1997 uint64_t pend_post, pend_nonpost;
1998
1999 qemu_savevm_state_pending(s->to_dst_file, threshold_size,
2000 &pend_nonpost, &pend_post);
2001 pending_size = pend_nonpost + pend_post;
2002 trace_migrate_pending(pending_size, threshold_size,
2003 pend_post, pend_nonpost);
2004 if (pending_size && pending_size >= threshold_size) {
2005 /* Still a significant amount to transfer */
2006
2007 if (migrate_postcopy_ram() &&
2008 s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE &&
2009 pend_nonpost <= threshold_size &&
2010 atomic_read(&s->start_postcopy)) {
2011
2012 if (!postcopy_start(s, &old_vm_running)) {
2013 current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE;
2014 entered_postcopy = true;
2015 }
2016
2017 continue;
2018 }
2019 /* Just another iteration step */
2020 qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy);
2021 } else {
2022 trace_migration_thread_low_pending(pending_size);
2023 migration_completion(s, current_active_state,
2024 &old_vm_running, &start_time);
2025 break;
2026 }
2027 }
2028
2029 if (qemu_file_get_error(s->to_dst_file)) {
2030 migrate_set_state(&s->state, current_active_state,
2031 MIGRATION_STATUS_FAILED);
2032 trace_migration_thread_file_err();
2033 break;
2034 }
2035 current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2036 if (current_time >= initial_time + BUFFER_DELAY) {
2037 uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) -
2038 initial_bytes;
2039 uint64_t time_spent = current_time - initial_time;
2040 double bandwidth = (double)transferred_bytes / time_spent;
2041 threshold_size = bandwidth * s->parameters.downtime_limit;
2042
2043 s->mbps = (((double) transferred_bytes * 8.0) /
2044 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
2045
2046 trace_migrate_transferred(transferred_bytes, time_spent,
2047 bandwidth, threshold_size);
2048 /* if we haven't sent anything, we don't want to recalculate
2049 10000 is a small enough number for our purposes */
2050 if (ram_counters.dirty_pages_rate && transferred_bytes > 10000) {
2051 s->expected_downtime = ram_counters.dirty_pages_rate *
2052 qemu_target_page_size() / bandwidth;
2053 }
2054
2055 qemu_file_reset_rate_limit(s->to_dst_file);
2056 initial_time = current_time;
2057 initial_bytes = qemu_ftell(s->to_dst_file);
2058 }
2059 if (qemu_file_rate_limit(s->to_dst_file)) {
2060 /* usleep expects microseconds */
2061 g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
2062 }
2063 }
2064
2065 trace_migration_thread_after_loop();
2066 /* If we enabled cpu throttling for auto-converge, turn it off. */
2067 cpu_throttle_stop();
2068 end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2069
2070 qemu_mutex_lock_iothread();
2071 /*
2072 * The resource has been allocated by migration will be reused in COLO
2073 * process, so don't release them.
2074 */
2075 if (!enable_colo) {
2076 qemu_savevm_state_cleanup();
2077 }
2078 if (s->state == MIGRATION_STATUS_COMPLETED) {
2079 uint64_t transferred_bytes = qemu_ftell(s->to_dst_file);
2080 s->total_time = end_time - s->total_time;
2081 if (!entered_postcopy) {
2082 s->downtime = end_time - start_time;
2083 }
2084 if (s->total_time) {
2085 s->mbps = (((double) transferred_bytes * 8.0) /
2086 ((double) s->total_time)) / 1000;
2087 }
2088 runstate_set(RUN_STATE_POSTMIGRATE);
2089 } else {
2090 if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) {
2091 migrate_start_colo_process(s);
2092 qemu_savevm_state_cleanup();
2093 /*
2094 * Fixme: we will run VM in COLO no matter its old running state.
2095 * After exited COLO, we will keep running.
2096 */
2097 old_vm_running = true;
2098 }
2099 if (old_vm_running && !entered_postcopy) {
2100 vm_start();
2101 } else {
2102 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2103 runstate_set(RUN_STATE_POSTMIGRATE);
2104 }
2105 }
2106 }
2107 qemu_bh_schedule(s->cleanup_bh);
2108 qemu_mutex_unlock_iothread();
2109
2110 rcu_unregister_thread();
2111 return NULL;
2112 }
2113
2114 void migrate_fd_connect(MigrationState *s)
2115 {
2116 s->expected_downtime = s->parameters.downtime_limit;
2117 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
2118
2119 qemu_file_set_blocking(s->to_dst_file, true);
2120 qemu_file_set_rate_limit(s->to_dst_file,
2121 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
2122
2123 /* Notify before starting migration thread */
2124 notifier_list_notify(&migration_state_notifiers, s);
2125
2126 /*
2127 * Open the return path. For postcopy, it is used exclusively. For
2128 * precopy, only if user specified "return-path" capability would
2129 * QEMU uses the return path.
2130 */
2131 if (migrate_postcopy_ram() || migrate_use_return_path()) {
2132 if (open_return_path_on_source(s)) {
2133 error_report("Unable to open return-path for postcopy");
2134 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2135 MIGRATION_STATUS_FAILED);
2136 migrate_fd_cleanup(s);
2137 return;
2138 }
2139 }
2140
2141 qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
2142 QEMU_THREAD_JOINABLE);
2143 s->migration_thread_running = true;
2144 }
2145
2146 void migration_global_dump(Monitor *mon)
2147 {
2148 MigrationState *ms = migrate_get_current();
2149
2150 monitor_printf(mon, "globals: store-global-state=%d, only_migratable=%d, "
2151 "send-configuration=%d, send-section-footer=%d\n",
2152 ms->store_global_state, ms->only_migratable,
2153 ms->send_configuration, ms->send_section_footer);
2154 }
2155
2156 #define DEFINE_PROP_MIG_CAP(name, x) \
2157 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
2158
2159 static Property migration_properties[] = {
2160 DEFINE_PROP_BOOL("store-global-state", MigrationState,
2161 store_global_state, true),
2162 DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false),
2163 DEFINE_PROP_BOOL("send-configuration", MigrationState,
2164 send_configuration, true),
2165 DEFINE_PROP_BOOL("send-section-footer", MigrationState,
2166 send_section_footer, true),
2167
2168 /* Migration parameters */
2169 DEFINE_PROP_INT64("x-compress-level", MigrationState,
2170 parameters.compress_level,
2171 DEFAULT_MIGRATE_COMPRESS_LEVEL),
2172 DEFINE_PROP_INT64("x-compress-threads", MigrationState,
2173 parameters.compress_threads,
2174 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
2175 DEFINE_PROP_INT64("x-decompress-threads", MigrationState,
2176 parameters.decompress_threads,
2177 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
2178 DEFINE_PROP_INT64("x-cpu-throttle-initial", MigrationState,
2179 parameters.cpu_throttle_initial,
2180 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
2181 DEFINE_PROP_INT64("x-cpu-throttle-increment", MigrationState,
2182 parameters.cpu_throttle_increment,
2183 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
2184 DEFINE_PROP_INT64("x-max-bandwidth", MigrationState,
2185 parameters.max_bandwidth, MAX_THROTTLE),
2186 DEFINE_PROP_INT64("x-downtime-limit", MigrationState,
2187 parameters.downtime_limit,
2188 DEFAULT_MIGRATE_SET_DOWNTIME),
2189 DEFINE_PROP_INT64("x-checkpoint-delay", MigrationState,
2190 parameters.x_checkpoint_delay,
2191 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
2192
2193 /* Migration capabilities */
2194 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
2195 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
2196 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
2197 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
2198 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
2199 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
2200 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
2201 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
2202 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
2203 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
2204 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
2205
2206 DEFINE_PROP_END_OF_LIST(),
2207 };
2208
2209 static void migration_class_init(ObjectClass *klass, void *data)
2210 {
2211 DeviceClass *dc = DEVICE_CLASS(klass);
2212
2213 dc->user_creatable = false;
2214 dc->props = migration_properties;
2215 }
2216
2217 static void migration_instance_finalize(Object *obj)
2218 {
2219 MigrationState *ms = MIGRATION_OBJ(obj);
2220 MigrationParameters *params = &ms->parameters;
2221
2222 g_free(params->tls_hostname);
2223 g_free(params->tls_creds);
2224 }
2225
2226 static void migration_instance_init(Object *obj)
2227 {
2228 MigrationState *ms = MIGRATION_OBJ(obj);
2229 MigrationParameters *params = &ms->parameters;
2230
2231 ms->state = MIGRATION_STATUS_NONE;
2232 ms->xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE;
2233 ms->mbps = -1;
2234
2235 params->tls_hostname = g_strdup("");
2236 params->tls_creds = g_strdup("");
2237
2238 /* Set has_* up only for parameter checks */
2239 params->has_compress_level = true;
2240 params->has_compress_threads = true;
2241 params->has_decompress_threads = true;
2242 params->has_cpu_throttle_initial = true;
2243 params->has_cpu_throttle_increment = true;
2244 params->has_max_bandwidth = true;
2245 params->has_downtime_limit = true;
2246 params->has_x_checkpoint_delay = true;
2247 params->has_block_incremental = true;
2248 }
2249
2250 /*
2251 * Return true if check pass, false otherwise. Error will be put
2252 * inside errp if provided.
2253 */
2254 static bool migration_object_check(MigrationState *ms, Error **errp)
2255 {
2256 MigrationCapabilityStatusList *head = NULL;
2257 /* Assuming all off */
2258 bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
2259 int i;
2260
2261 if (!migrate_params_check(&ms->parameters, errp)) {
2262 return false;
2263 }
2264
2265 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
2266 if (ms->enabled_capabilities[i]) {
2267 head = migrate_cap_add(head, i, true);
2268 }
2269 }
2270
2271 ret = migrate_caps_check(cap_list, head, errp);
2272
2273 /* It works with head == NULL */
2274 qapi_free_MigrationCapabilityStatusList(head);
2275
2276 return ret;
2277 }
2278
2279 static const TypeInfo migration_type = {
2280 .name = TYPE_MIGRATION,
2281 /*
2282 * NOTE: TYPE_MIGRATION is not really a device, as the object is
2283 * not created using qdev_create(), it is not attached to the qdev
2284 * device tree, and it is never realized.
2285 *
2286 * TODO: Make this TYPE_OBJECT once QOM provides something like
2287 * TYPE_DEVICE's "-global" properties.
2288 */
2289 .parent = TYPE_DEVICE,
2290 .class_init = migration_class_init,
2291 .class_size = sizeof(MigrationClass),
2292 .instance_size = sizeof(MigrationState),
2293 .instance_init = migration_instance_init,
2294 .instance_finalize = migration_instance_finalize,
2295 };
2296
2297 static void register_migration_types(void)
2298 {
2299 type_register_static(&migration_type);
2300 }
2301
2302 type_init(register_migration_types);
QEmu