?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
migration: Create migration_ioc_process_incoming()
[qemu/ar7.git] / migration / migration.c
blob2d4c56e612a5b8f07d40dac3d2d89845b98c9470
1 /*
2 * QEMU live migration
3 *
4 * Copyright IBM, Corp. 2008
5 *
6 * Authors:
7 * Anthony Liguori <aliguori@us.ibm.com>
8 *
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
11 *
12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
14 */
15
16 #include "qemu/osdep.h"
17 #include "qemu/cutils.h"
18 #include "qemu/error-report.h"
19 #include "migration/blocker.h"
20 #include "exec.h"
21 #include "fd.h"
22 #include "socket.h"
23 #include "rdma.h"
24 #include "ram.h"
25 #include "migration/global_state.h"
26 #include "migration/misc.h"
27 #include "migration.h"
28 #include "savevm.h"
29 #include "qemu-file-channel.h"
30 #include "qemu-file.h"
31 #include "migration/vmstate.h"
32 #include "block/block.h"
33 #include "qapi/qmp/qerror.h"
34 #include "qemu/rcu.h"
35 #include "block.h"
36 #include "postcopy-ram.h"
37 #include "qemu/thread.h"
38 #include "qmp-commands.h"
39 #include "trace.h"
40 #include "qapi-event.h"
41 #include "exec/target_page.h"
42 #include "io/channel-buffer.h"
43 #include "migration/colo.h"
44 #include "hw/boards.h"
45 #include "monitor/monitor.h"
46
47 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
48
49 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
50 * data. */
51 #define BUFFER_DELAY 100
52 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
53
54 /* Time in milliseconds we are allowed to stop the source,
55 * for sending the last part */
56 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
57
58 /* Maximum migrate downtime set to 2000 seconds */
59 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
60 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
61
62 /* Default compression thread count */
63 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
64 /* Default decompression thread count, usually decompression is at
65 * least 4 times as fast as compression.*/
66 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
67 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
68 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
69 /* Define default autoconverge cpu throttle migration parameters */
70 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
71 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
72
73 /* Migration XBZRLE default cache size */
74 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
75
76 /* The delay time (in ms) between two COLO checkpoints
77 * Note: Please change this default value to 10000 when we support hybrid mode.
78 */
79 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
80
81 static NotifierList migration_state_notifiers =
82 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
83
84 static bool deferred_incoming;
85
86 /* Messages sent on the return path from destination to source */
87 enum mig_rp_message_type {
88 MIG_RP_MSG_INVALID = 0, /* Must be 0 */
89 MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */
90 MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */
91
92 MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
93 MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */
94
95 MIG_RP_MSG_MAX
96 };
97
98 /* When we add fault tolerance, we could have several
99 migrations at once. For now we don't need to add
100 dynamic creation of migration */
101
102 static MigrationState *current_migration;
103
104 static bool migration_object_check(MigrationState *ms, Error **errp);
105
106 void migration_object_init(void)
107 {
108 MachineState *ms = MACHINE(qdev_get_machine());
109 Error *err = NULL;
110
111 /* This can only be called once. */
112 assert(!current_migration);
113 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
114
115 if (!migration_object_check(current_migration, &err)) {
116 error_report_err(err);
117 exit(1);
118 }
119
120 /*
121 * We cannot really do this in migration_instance_init() since at
122 * that time global properties are not yet applied, then this
123 * value will be definitely replaced by something else.
124 */
125 if (ms->enforce_config_section) {
126 current_migration->send_configuration = true;
127 }
128 }
129
130 /* For outgoing */
131 MigrationState *migrate_get_current(void)
132 {
133 /* This can only be called after the object created. */
134 assert(current_migration);
135 return current_migration;
136 }
137
138 MigrationIncomingState *migration_incoming_get_current(void)
139 {
140 static bool once;
141 static MigrationIncomingState mis_current;
142
143 if (!once) {
144 mis_current.state = MIGRATION_STATUS_NONE;
145 memset(&mis_current, 0, sizeof(MigrationIncomingState));
146 qemu_mutex_init(&mis_current.rp_mutex);
147 qemu_event_init(&mis_current.main_thread_load_event, false);
148 once = true;
149 }
150 return &mis_current;
151 }
152
153 void migration_incoming_state_destroy(void)
154 {
155 struct MigrationIncomingState *mis = migration_incoming_get_current();
156
157 if (mis->to_src_file) {
158 /* Tell source that we are done */
159 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
160 qemu_fclose(mis->to_src_file);
161 mis->to_src_file = NULL;
162 }
163
164 if (mis->from_src_file) {
165 qemu_fclose(mis->from_src_file);
166 mis->from_src_file = NULL;
167 }
168
169 qemu_event_reset(&mis->main_thread_load_event);
170 }
171
172 static void migrate_generate_event(int new_state)
173 {
174 if (migrate_use_events()) {
175 qapi_event_send_migration(new_state, &error_abort);
176 }
177 }
178
179 /*
180 * Called on -incoming with a defer: uri.
181 * The migration can be started later after any parameters have been
182 * changed.
183 */
184 static void deferred_incoming_migration(Error **errp)
185 {
186 if (deferred_incoming) {
187 error_setg(errp, "Incoming migration already deferred");
188 }
189 deferred_incoming = true;
190 }
191
192 /*
193 * Send a message on the return channel back to the source
194 * of the migration.
195 */
196 static void migrate_send_rp_message(MigrationIncomingState *mis,
197 enum mig_rp_message_type message_type,
198 uint16_t len, void *data)
199 {
200 trace_migrate_send_rp_message((int)message_type, len);
201 qemu_mutex_lock(&mis->rp_mutex);
202 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
203 qemu_put_be16(mis->to_src_file, len);
204 qemu_put_buffer(mis->to_src_file, data, len);
205 qemu_fflush(mis->to_src_file);
206 qemu_mutex_unlock(&mis->rp_mutex);
207 }
208
209 /* Request a range of pages from the source VM at the given
210 * start address.
211 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
212 * as the last request (a name must have been given previously)
213 * Start: Address offset within the RB
214 * Len: Length in bytes required - must be a multiple of pagesize
215 */
216 void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
217 ram_addr_t start, size_t len)
218 {
219 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
220 size_t msglen = 12; /* start + len */
221
222 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
223 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
224
225 if (rbname) {
226 int rbname_len = strlen(rbname);
227 assert(rbname_len < 256);
228
229 bufc[msglen++] = rbname_len;
230 memcpy(bufc + msglen, rbname, rbname_len);
231 msglen += rbname_len;
232 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc);
233 } else {
234 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc);
235 }
236 }
237
238 void qemu_start_incoming_migration(const char *uri, Error **errp)
239 {
240 const char *p;
241
242 qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
243 if (!strcmp(uri, "defer")) {
244 deferred_incoming_migration(errp);
245 } else if (strstart(uri, "tcp:", &p)) {
246 tcp_start_incoming_migration(p, errp);
247 #ifdef CONFIG_RDMA
248 } else if (strstart(uri, "rdma:", &p)) {
249 rdma_start_incoming_migration(p, errp);
250 #endif
251 } else if (strstart(uri, "exec:", &p)) {
252 exec_start_incoming_migration(p, errp);
253 } else if (strstart(uri, "unix:", &p)) {
254 unix_start_incoming_migration(p, errp);
255 } else if (strstart(uri, "fd:", &p)) {
256 fd_start_incoming_migration(p, errp);
257 } else {
258 error_setg(errp, "unknown migration protocol: %s", uri);
259 }
260 }
261
262 static void process_incoming_migration_bh(void *opaque)
263 {
264 Error *local_err = NULL;
265 MigrationIncomingState *mis = opaque;
266
267 /* Make sure all file formats flush their mutable metadata.
268 * If we get an error here, just don't restart the VM yet. */
269 bdrv_invalidate_cache_all(&local_err);
270 if (local_err) {
271 error_report_err(local_err);
272 local_err = NULL;
273 autostart = false;
274 }
275
276 /*
277 * This must happen after all error conditions are dealt with and
278 * we're sure the VM is going to be running on this host.
279 */
280 qemu_announce_self();
281
282 /* If global state section was not received or we are in running
283 state, we need to obey autostart. Any other state is set with
284 runstate_set. */
285
286 if (!global_state_received() ||
287 global_state_get_runstate() == RUN_STATE_RUNNING) {
288 if (autostart) {
289 vm_start();
290 } else {
291 runstate_set(RUN_STATE_PAUSED);
292 }
293 } else {
294 runstate_set(global_state_get_runstate());
295 }
296 /*
297 * This must happen after any state changes since as soon as an external
298 * observer sees this event they might start to prod at the VM assuming
299 * it's ready to use.
300 */
301 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
302 MIGRATION_STATUS_COMPLETED);
303 qemu_bh_delete(mis->bh);
304 migration_incoming_state_destroy();
305 }
306
307 static void process_incoming_migration_co(void *opaque)
308 {
309 MigrationIncomingState *mis = migration_incoming_get_current();
310 PostcopyState ps;
311 int ret;
312
313 assert(mis->from_src_file);
314 mis->largest_page_size = qemu_ram_pagesize_largest();
315 postcopy_state_set(POSTCOPY_INCOMING_NONE);
316 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
317 MIGRATION_STATUS_ACTIVE);
318 ret = qemu_loadvm_state(mis->from_src_file);
319
320 ps = postcopy_state_get();
321 trace_process_incoming_migration_co_end(ret, ps);
322 if (ps != POSTCOPY_INCOMING_NONE) {
323 if (ps == POSTCOPY_INCOMING_ADVISE) {
324 /*
325 * Where a migration had postcopy enabled (and thus went to advise)
326 * but managed to complete within the precopy period, we can use
327 * the normal exit.
328 */
329 postcopy_ram_incoming_cleanup(mis);
330 } else if (ret >= 0) {
331 /*
332 * Postcopy was started, cleanup should happen at the end of the
333 * postcopy thread.
334 */
335 trace_process_incoming_migration_co_postcopy_end_main();
336 return;
337 }
338 /* Else if something went wrong then just fall out of the normal exit */
339 }
340
341 /* we get COLO info, and know if we are in COLO mode */
342 if (!ret && migration_incoming_enable_colo()) {
343 mis->migration_incoming_co = qemu_coroutine_self();
344 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
345 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
346 mis->have_colo_incoming_thread = true;
347 qemu_coroutine_yield();
348
349 /* Wait checkpoint incoming thread exit before free resource */
350 qemu_thread_join(&mis->colo_incoming_thread);
351 }
352
353 if (ret < 0) {
354 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
355 MIGRATION_STATUS_FAILED);
356 error_report("load of migration failed: %s", strerror(-ret));
357 qemu_fclose(mis->from_src_file);
358 exit(EXIT_FAILURE);
359 }
360 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
361 qemu_bh_schedule(mis->bh);
362 }
363
364 void migration_fd_process_incoming(QEMUFile *f)
365 {
366 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
367 MigrationIncomingState *mis = migration_incoming_get_current();
368
369 if (!mis->from_src_file) {
370 mis->from_src_file = f;
371 }
372 qemu_file_set_blocking(f, false);
373 qemu_coroutine_enter(co);
374 }
375
376 void migration_ioc_process_incoming(QIOChannel *ioc)
377 {
378 MigrationIncomingState *mis = migration_incoming_get_current();
379
380 if (!mis->from_src_file) {
381 QEMUFile *f = qemu_fopen_channel_input(ioc);
382 migration_fd_process_incoming(f);
383 }
384 /* We still only have a single channel. Nothing to do here yet */
385 }
386
387 /*
388 * Send a 'SHUT' message on the return channel with the given value
389 * to indicate that we've finished with the RP. Non-0 value indicates
390 * error.
391 */
392 void migrate_send_rp_shut(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_SHUT, sizeof(buf), &buf);
399 }
400
401 /*
402 * Send a 'PONG' message on the return channel with the given value
403 * (normally in response to a 'PING')
404 */
405 void migrate_send_rp_pong(MigrationIncomingState *mis,
406 uint32_t value)
407 {
408 uint32_t buf;
409
410 buf = cpu_to_be32(value);
411 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
412 }
413
414 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
415 {
416 MigrationCapabilityStatusList *head = NULL;
417 MigrationCapabilityStatusList *caps;
418 MigrationState *s = migrate_get_current();
419 int i;
420
421 caps = NULL; /* silence compiler warning */
422 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
423 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
424 if (i == MIGRATION_CAPABILITY_BLOCK) {
425 continue;
426 }
427 #endif
428 if (head == NULL) {
429 head = g_malloc0(sizeof(*caps));
430 caps = head;
431 } else {
432 caps->next = g_malloc0(sizeof(*caps));
433 caps = caps->next;
434 }
435 caps->value =
436 g_malloc(sizeof(*caps->value));
437 caps->value->capability = i;
438 caps->value->state = s->enabled_capabilities[i];
439 }
440
441 return head;
442 }
443
444 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
445 {
446 MigrationParameters *params;
447 MigrationState *s = migrate_get_current();
448
449 /* TODO use QAPI_CLONE() instead of duplicating it inline */
450 params = g_malloc0(sizeof(*params));
451 params->has_compress_level = true;
452 params->compress_level = s->parameters.compress_level;
453 params->has_compress_threads = true;
454 params->compress_threads = s->parameters.compress_threads;
455 params->has_decompress_threads = true;
456 params->decompress_threads = s->parameters.decompress_threads;
457 params->has_cpu_throttle_initial = true;
458 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
459 params->has_cpu_throttle_increment = true;
460 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
461 params->has_tls_creds = true;
462 params->tls_creds = g_strdup(s->parameters.tls_creds);
463 params->has_tls_hostname = true;
464 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
465 params->has_max_bandwidth = true;
466 params->max_bandwidth = s->parameters.max_bandwidth;
467 params->has_downtime_limit = true;
468 params->downtime_limit = s->parameters.downtime_limit;
469 params->has_x_checkpoint_delay = true;
470 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
471 params->has_block_incremental = true;
472 params->block_incremental = s->parameters.block_incremental;
473
474 return params;
475 }
476
477 /*
478 * Return true if we're already in the middle of a migration
479 * (i.e. any of the active or setup states)
480 */
481 static bool migration_is_setup_or_active(int state)
482 {
483 switch (state) {
484 case MIGRATION_STATUS_ACTIVE:
485 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
486 case MIGRATION_STATUS_SETUP:
487 return true;
488
489 default:
490 return false;
491
492 }
493 }
494
495 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
496 {
497 info->has_ram = true;
498 info->ram = g_malloc0(sizeof(*info->ram));
499 info->ram->transferred = ram_counters.transferred;
500 info->ram->total = ram_bytes_total();
501 info->ram->duplicate = ram_counters.duplicate;
502 /* legacy value. It is not used anymore */
503 info->ram->skipped = 0;
504 info->ram->normal = ram_counters.normal;
505 info->ram->normal_bytes = ram_counters.normal *
506 qemu_target_page_size();
507 info->ram->mbps = s->mbps;
508 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
509 info->ram->postcopy_requests = ram_counters.postcopy_requests;
510 info->ram->page_size = qemu_target_page_size();
511
512 if (migrate_use_xbzrle()) {
513 info->has_xbzrle_cache = true;
514 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
515 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
516 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
517 info->xbzrle_cache->pages = xbzrle_counters.pages;
518 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
519 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
520 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
521 }
522
523 if (cpu_throttle_active()) {
524 info->has_cpu_throttle_percentage = true;
525 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
526 }
527
528 if (s->state != MIGRATION_STATUS_COMPLETED) {
529 info->ram->remaining = ram_bytes_remaining();
530 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
531 }
532 }
533
534 static void populate_disk_info(MigrationInfo *info)
535 {
536 if (blk_mig_active()) {
537 info->has_disk = true;
538 info->disk = g_malloc0(sizeof(*info->disk));
539 info->disk->transferred = blk_mig_bytes_transferred();
540 info->disk->remaining = blk_mig_bytes_remaining();
541 info->disk->total = blk_mig_bytes_total();
542 }
543 }
544
545 MigrationInfo *qmp_query_migrate(Error **errp)
546 {
547 MigrationInfo *info = g_malloc0(sizeof(*info));
548 MigrationState *s = migrate_get_current();
549
550 switch (s->state) {
551 case MIGRATION_STATUS_NONE:
552 /* no migration has happened ever */
553 break;
554 case MIGRATION_STATUS_SETUP:
555 info->has_status = true;
556 info->has_total_time = false;
557 break;
558 case MIGRATION_STATUS_ACTIVE:
559 case MIGRATION_STATUS_CANCELLING:
560 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
561 /* TODO add some postcopy stats */
562 info->has_status = true;
563 info->has_total_time = true;
564 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
565 - s->total_time;
566 info->has_expected_downtime = true;
567 info->expected_downtime = s->expected_downtime;
568 info->has_setup_time = true;
569 info->setup_time = s->setup_time;
570
571 populate_ram_info(info, s);
572 populate_disk_info(info);
573 break;
574 case MIGRATION_STATUS_COLO:
575 info->has_status = true;
576 /* TODO: display COLO specific information (checkpoint info etc.) */
577 break;
578 case MIGRATION_STATUS_COMPLETED:
579 info->has_status = true;
580 info->has_total_time = true;
581 info->total_time = s->total_time;
582 info->has_downtime = true;
583 info->downtime = s->downtime;
584 info->has_setup_time = true;
585 info->setup_time = s->setup_time;
586
587 populate_ram_info(info, s);
588 break;
589 case MIGRATION_STATUS_FAILED:
590 info->has_status = true;
591 if (s->error) {
592 info->has_error_desc = true;
593 info->error_desc = g_strdup(error_get_pretty(s->error));
594 }
595 break;
596 case MIGRATION_STATUS_CANCELLED:
597 info->has_status = true;
598 break;
599 }
600 info->status = s->state;
601
602 return info;
603 }
604
605 /**
606 * @migration_caps_check - check capability validity
607 *
608 * @cap_list: old capability list, array of bool
609 * @params: new capabilities to be applied soon
610 * @errp: set *errp if the check failed, with reason
611 *
612 * Returns true if check passed, otherwise false.
613 */
614 static bool migrate_caps_check(bool *cap_list,
615 MigrationCapabilityStatusList *params,
616 Error **errp)
617 {
618 MigrationCapabilityStatusList *cap;
619 bool old_postcopy_cap;
620
621 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
622
623 for (cap = params; cap; cap = cap->next) {
624 cap_list[cap->value->capability] = cap->value->state;
625 }
626
627 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
628 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
629 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
630 "block migration");
631 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
632 return false;
633 }
634 #endif
635
636 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
637 if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
638 /* The decompression threads asynchronously write into RAM
639 * rather than use the atomic copies needed to avoid
640 * userfaulting. It should be possible to fix the decompression
641 * threads for compatibility in future.
642 */
643 error_setg(errp, "Postcopy is not currently compatible "
644 "with compression");
645 return false;
646 }
647
648 /* This check is reasonably expensive, so only when it's being
649 * set the first time, also it's only the destination that needs
650 * special support.
651 */
652 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
653 !postcopy_ram_supported_by_host()) {
654 /* postcopy_ram_supported_by_host will have emitted a more
655 * detailed message
656 */
657 error_setg(errp, "Postcopy is not supported");
658 return false;
659 }
660 }
661
662 return true;
663 }
664
665 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
666 Error **errp)
667 {
668 MigrationState *s = migrate_get_current();
669 MigrationCapabilityStatusList *cap;
670
671 if (migration_is_setup_or_active(s->state)) {
672 error_setg(errp, QERR_MIGRATION_ACTIVE);
673 return;
674 }
675
676 if (!migrate_caps_check(s->enabled_capabilities, params, errp)) {
677 return;
678 }
679
680 for (cap = params; cap; cap = cap->next) {
681 s->enabled_capabilities[cap->value->capability] = cap->value->state;
682 }
683 }
684
685 /*
686 * Check whether the parameters are valid. Error will be put into errp
687 * (if provided). Return true if valid, otherwise false.
688 */
689 static bool migrate_params_check(MigrationParameters *params, Error **errp)
690 {
691 if (params->has_compress_level &&
692 (params->compress_level < 0 || params->compress_level > 9)) {
693 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
694 "is invalid, it should be in the range of 0 to 9");
695 return false;
696 }
697
698 if (params->has_compress_threads &&
699 (params->compress_threads < 1 || params->compress_threads > 255)) {
700 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
701 "compress_threads",
702 "is invalid, it should be in the range of 1 to 255");
703 return false;
704 }
705
706 if (params->has_decompress_threads &&
707 (params->decompress_threads < 1 || params->decompress_threads > 255)) {
708 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
709 "decompress_threads",
710 "is invalid, it should be in the range of 1 to 255");
711 return false;
712 }
713
714 if (params->has_cpu_throttle_initial &&
715 (params->cpu_throttle_initial < 1 ||
716 params->cpu_throttle_initial > 99)) {
717 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
718 "cpu_throttle_initial",
719 "an integer in the range of 1 to 99");
720 return false;
721 }
722
723 if (params->has_cpu_throttle_increment &&
724 (params->cpu_throttle_increment < 1 ||
725 params->cpu_throttle_increment > 99)) {
726 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
727 "cpu_throttle_increment",
728 "an integer in the range of 1 to 99");
729 return false;
730 }
731
732 if (params->has_max_bandwidth &&
733 (params->max_bandwidth < 0 || params->max_bandwidth > SIZE_MAX)) {
734 error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
735 " range of 0 to %zu bytes/second", SIZE_MAX);
736 return false;
737 }
738
739 if (params->has_downtime_limit &&
740 (params->downtime_limit < 0 ||
741 params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
742 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
743 "the range of 0 to %d milliseconds",
744 MAX_MIGRATE_DOWNTIME);
745 return false;
746 }
747
748 if (params->has_x_checkpoint_delay && (params->x_checkpoint_delay < 0)) {
749 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
750 "x_checkpoint_delay",
751 "is invalid, it should be positive");
752 return false;
753 }
754
755 return true;
756 }
757
758 static void migrate_params_test_apply(MigrateSetParameters *params,
759 MigrationParameters *dest)
760 {
761 *dest = migrate_get_current()->parameters;
762
763 /* TODO use QAPI_CLONE() instead of duplicating it inline */
764
765 if (params->has_compress_level) {
766 dest->compress_level = params->compress_level;
767 }
768
769 if (params->has_compress_threads) {
770 dest->compress_threads = params->compress_threads;
771 }
772
773 if (params->has_decompress_threads) {
774 dest->decompress_threads = params->decompress_threads;
775 }
776
777 if (params->has_cpu_throttle_initial) {
778 dest->cpu_throttle_initial = params->cpu_throttle_initial;
779 }
780
781 if (params->has_cpu_throttle_increment) {
782 dest->cpu_throttle_increment = params->cpu_throttle_increment;
783 }
784
785 if (params->has_tls_creds) {
786 assert(params->tls_creds->type == QTYPE_QSTRING);
787 dest->tls_creds = g_strdup(params->tls_creds->u.s);
788 }
789
790 if (params->has_tls_hostname) {
791 assert(params->tls_hostname->type == QTYPE_QSTRING);
792 dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
793 }
794
795 if (params->has_max_bandwidth) {
796 dest->max_bandwidth = params->max_bandwidth;
797 }
798
799 if (params->has_downtime_limit) {
800 dest->downtime_limit = params->downtime_limit;
801 }
802
803 if (params->has_x_checkpoint_delay) {
804 dest->x_checkpoint_delay = params->x_checkpoint_delay;
805 }
806
807 if (params->has_block_incremental) {
808 dest->block_incremental = params->block_incremental;
809 }
810 }
811
812 static void migrate_params_apply(MigrateSetParameters *params)
813 {
814 MigrationState *s = migrate_get_current();
815
816 /* TODO use QAPI_CLONE() instead of duplicating it inline */
817
818 if (params->has_compress_level) {
819 s->parameters.compress_level = params->compress_level;
820 }
821
822 if (params->has_compress_threads) {
823 s->parameters.compress_threads = params->compress_threads;
824 }
825
826 if (params->has_decompress_threads) {
827 s->parameters.decompress_threads = params->decompress_threads;
828 }
829
830 if (params->has_cpu_throttle_initial) {
831 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
832 }
833
834 if (params->has_cpu_throttle_increment) {
835 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
836 }
837
838 if (params->has_tls_creds) {
839 g_free(s->parameters.tls_creds);
840 assert(params->tls_creds->type == QTYPE_QSTRING);
841 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
842 }
843
844 if (params->has_tls_hostname) {
845 g_free(s->parameters.tls_hostname);
846 assert(params->tls_hostname->type == QTYPE_QSTRING);
847 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
848 }
849
850 if (params->has_max_bandwidth) {
851 s->parameters.max_bandwidth = params->max_bandwidth;
852 if (s->to_dst_file) {
853 qemu_file_set_rate_limit(s->to_dst_file,
854 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
855 }
856 }
857
858 if (params->has_downtime_limit) {
859 s->parameters.downtime_limit = params->downtime_limit;
860 }
861
862 if (params->has_x_checkpoint_delay) {
863 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
864 if (migration_in_colo_state()) {
865 colo_checkpoint_notify(s);
866 }
867 }
868
869 if (params->has_block_incremental) {
870 s->parameters.block_incremental = params->block_incremental;
871 }
872 }
873
874 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
875 {
876 MigrationParameters tmp;
877
878 /* TODO Rewrite "" to null instead */
879 if (params->has_tls_creds
880 && params->tls_creds->type == QTYPE_QNULL) {
881 QDECREF(params->tls_creds->u.n);
882 params->tls_creds->type = QTYPE_QSTRING;
883 params->tls_creds->u.s = strdup("");
884 }
885 /* TODO Rewrite "" to null instead */
886 if (params->has_tls_hostname
887 && params->tls_hostname->type == QTYPE_QNULL) {
888 QDECREF(params->tls_hostname->u.n);
889 params->tls_hostname->type = QTYPE_QSTRING;
890 params->tls_hostname->u.s = strdup("");
891 }
892
893 migrate_params_test_apply(params, &tmp);
894
895 if (!migrate_params_check(&tmp, errp)) {
896 /* Invalid parameter */
897 return;
898 }
899
900 migrate_params_apply(params);
901 }
902
903
904 void qmp_migrate_start_postcopy(Error **errp)
905 {
906 MigrationState *s = migrate_get_current();
907
908 if (!migrate_postcopy_ram()) {
909 error_setg(errp, "Enable postcopy with migrate_set_capability before"
910 " the start of migration");
911 return;
912 }
913
914 if (s->state == MIGRATION_STATUS_NONE) {
915 error_setg(errp, "Postcopy must be started after migration has been"
916 " started");
917 return;
918 }
919 /*
920 * we don't error if migration has finished since that would be racy
921 * with issuing this command.
922 */
923 atomic_set(&s->start_postcopy, true);
924 }
925
926 /* shared migration helpers */
927
928 void migrate_set_state(int *state, int old_state, int new_state)
929 {
930 assert(new_state < MIGRATION_STATUS__MAX);
931 if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
932 trace_migrate_set_state(MigrationStatus_str(new_state));
933 migrate_generate_event(new_state);
934 }
935 }
936
937 static MigrationCapabilityStatusList *migrate_cap_add(
938 MigrationCapabilityStatusList *list,
939 MigrationCapability index,
940 bool state)
941 {
942 MigrationCapabilityStatusList *cap;
943
944 cap = g_new0(MigrationCapabilityStatusList, 1);
945 cap->value = g_new0(MigrationCapabilityStatus, 1);
946 cap->value->capability = index;
947 cap->value->state = state;
948 cap->next = list;
949
950 return cap;
951 }
952
953 void migrate_set_block_enabled(bool value, Error **errp)
954 {
955 MigrationCapabilityStatusList *cap;
956
957 cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
958 qmp_migrate_set_capabilities(cap, errp);
959 qapi_free_MigrationCapabilityStatusList(cap);
960 }
961
962 static void migrate_set_block_incremental(MigrationState *s, bool value)
963 {
964 s->parameters.block_incremental = value;
965 }
966
967 static void block_cleanup_parameters(MigrationState *s)
968 {
969 if (s->must_remove_block_options) {
970 /* setting to false can never fail */
971 migrate_set_block_enabled(false, &error_abort);
972 migrate_set_block_incremental(s, false);
973 s->must_remove_block_options = false;
974 }
975 }
976
977 static void migrate_fd_cleanup(void *opaque)
978 {
979 MigrationState *s = opaque;
980
981 qemu_bh_delete(s->cleanup_bh);
982 s->cleanup_bh = NULL;
983
984 if (s->to_dst_file) {
985 trace_migrate_fd_cleanup();
986 qemu_mutex_unlock_iothread();
987 if (s->migration_thread_running) {
988 qemu_thread_join(&s->thread);
989 s->migration_thread_running = false;
990 }
991 qemu_mutex_lock_iothread();
992
993 qemu_fclose(s->to_dst_file);
994 s->to_dst_file = NULL;
995 }
996
997 assert((s->state != MIGRATION_STATUS_ACTIVE) &&
998 (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
999
1000 if (s->state == MIGRATION_STATUS_CANCELLING) {
1001 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1002 MIGRATION_STATUS_CANCELLED);
1003 }
1004
1005 notifier_list_notify(&migration_state_notifiers, s);
1006 block_cleanup_parameters(s);
1007 }
1008
1009 void migrate_fd_error(MigrationState *s, const Error *error)
1010 {
1011 trace_migrate_fd_error(error_get_pretty(error));
1012 assert(s->to_dst_file == NULL);
1013 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1014 MIGRATION_STATUS_FAILED);
1015 if (!s->error) {
1016 s->error = error_copy(error);
1017 }
1018 notifier_list_notify(&migration_state_notifiers, s);
1019 block_cleanup_parameters(s);
1020 }
1021
1022 static void migrate_fd_cancel(MigrationState *s)
1023 {
1024 int old_state ;
1025 QEMUFile *f = migrate_get_current()->to_dst_file;
1026 trace_migrate_fd_cancel();
1027
1028 if (s->rp_state.from_dst_file) {
1029 /* shutdown the rp socket, so causing the rp thread to shutdown */
1030 qemu_file_shutdown(s->rp_state.from_dst_file);
1031 }
1032
1033 do {
1034 old_state = s->state;
1035 if (!migration_is_setup_or_active(old_state)) {
1036 break;
1037 }
1038 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1039 } while (s->state != MIGRATION_STATUS_CANCELLING);
1040
1041 /*
1042 * If we're unlucky the migration code might be stuck somewhere in a
1043 * send/write while the network has failed and is waiting to timeout;
1044 * if we've got shutdown(2) available then we can force it to quit.
1045 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1046 * called in a bh, so there is no race against this cancel.
1047 */
1048 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1049 qemu_file_shutdown(f);
1050 }
1051 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1052 Error *local_err = NULL;
1053
1054 bdrv_invalidate_cache_all(&local_err);
1055 if (local_err) {
1056 error_report_err(local_err);
1057 } else {
1058 s->block_inactive = false;
1059 }
1060 }
1061 block_cleanup_parameters(s);
1062 }
1063
1064 void add_migration_state_change_notifier(Notifier *notify)
1065 {
1066 notifier_list_add(&migration_state_notifiers, notify);
1067 }
1068
1069 void remove_migration_state_change_notifier(Notifier *notify)
1070 {
1071 notifier_remove(notify);
1072 }
1073
1074 bool migration_in_setup(MigrationState *s)
1075 {
1076 return s->state == MIGRATION_STATUS_SETUP;
1077 }
1078
1079 bool migration_has_finished(MigrationState *s)
1080 {
1081 return s->state == MIGRATION_STATUS_COMPLETED;
1082 }
1083
1084 bool migration_has_failed(MigrationState *s)
1085 {
1086 return (s->state == MIGRATION_STATUS_CANCELLED ||
1087 s->state == MIGRATION_STATUS_FAILED);
1088 }
1089
1090 bool migration_in_postcopy(void)
1091 {
1092 MigrationState *s = migrate_get_current();
1093
1094 return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1095 }
1096
1097 bool migration_in_postcopy_after_devices(MigrationState *s)
1098 {
1099 return migration_in_postcopy() && s->postcopy_after_devices;
1100 }
1101
1102 bool migration_is_idle(void)
1103 {
1104 MigrationState *s = migrate_get_current();
1105
1106 switch (s->state) {
1107 case MIGRATION_STATUS_NONE:
1108 case MIGRATION_STATUS_CANCELLED:
1109 case MIGRATION_STATUS_COMPLETED:
1110 case MIGRATION_STATUS_FAILED:
1111 return true;
1112 case MIGRATION_STATUS_SETUP:
1113 case MIGRATION_STATUS_CANCELLING:
1114 case MIGRATION_STATUS_ACTIVE:
1115 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1116 case MIGRATION_STATUS_COLO:
1117 return false;
1118 case MIGRATION_STATUS__MAX:
1119 g_assert_not_reached();
1120 }
1121
1122 return false;
1123 }
1124
1125 MigrationState *migrate_init(void)
1126 {
1127 MigrationState *s = migrate_get_current();
1128
1129 /*
1130 * Reinitialise all migration state, except
1131 * parameters/capabilities that the user set, and
1132 * locks.
1133 */
1134 s->bytes_xfer = 0;
1135 s->xfer_limit = 0;
1136 s->cleanup_bh = 0;
1137 s->to_dst_file = NULL;
1138 s->state = MIGRATION_STATUS_NONE;
1139 s->rp_state.from_dst_file = NULL;
1140 s->rp_state.error = false;
1141 s->mbps = 0.0;
1142 s->downtime = 0;
1143 s->expected_downtime = 0;
1144 s->setup_time = 0;
1145 s->start_postcopy = false;
1146 s->postcopy_after_devices = false;
1147 s->migration_thread_running = false;
1148 error_free(s->error);
1149 s->error = NULL;
1150
1151 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1152
1153 s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1154 return s;
1155 }
1156
1157 static GSList *migration_blockers;
1158
1159 int migrate_add_blocker(Error *reason, Error **errp)
1160 {
1161 if (migrate_get_current()->only_migratable) {
1162 error_propagate(errp, error_copy(reason));
1163 error_prepend(errp, "disallowing migration blocker "
1164 "(--only_migratable) for: ");
1165 return -EACCES;
1166 }
1167
1168 if (migration_is_idle()) {
1169 migration_blockers = g_slist_prepend(migration_blockers, reason);
1170 return 0;
1171 }
1172
1173 error_propagate(errp, error_copy(reason));
1174 error_prepend(errp, "disallowing migration blocker (migration in "
1175 "progress) for: ");
1176 return -EBUSY;
1177 }
1178
1179 void migrate_del_blocker(Error *reason)
1180 {
1181 migration_blockers = g_slist_remove(migration_blockers, reason);
1182 }
1183
1184 void qmp_migrate_incoming(const char *uri, Error **errp)
1185 {
1186 Error *local_err = NULL;
1187 static bool once = true;
1188
1189 if (!deferred_incoming) {
1190 error_setg(errp, "For use with '-incoming defer'");
1191 return;
1192 }
1193 if (!once) {
1194 error_setg(errp, "The incoming migration has already been started");
1195 }
1196
1197 qemu_start_incoming_migration(uri, &local_err);
1198
1199 if (local_err) {
1200 error_propagate(errp, local_err);
1201 return;
1202 }
1203
1204 once = false;
1205 }
1206
1207 bool migration_is_blocked(Error **errp)
1208 {
1209 if (qemu_savevm_state_blocked(errp)) {
1210 return true;
1211 }
1212
1213 if (migration_blockers) {
1214 error_propagate(errp, error_copy(migration_blockers->data));
1215 return true;
1216 }
1217
1218 return false;
1219 }
1220
1221 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1222 bool has_inc, bool inc, bool has_detach, bool detach,
1223 Error **errp)
1224 {
1225 Error *local_err = NULL;
1226 MigrationState *s = migrate_get_current();
1227 const char *p;
1228
1229 if (migration_is_setup_or_active(s->state) ||
1230 s->state == MIGRATION_STATUS_CANCELLING ||
1231 s->state == MIGRATION_STATUS_COLO) {
1232 error_setg(errp, QERR_MIGRATION_ACTIVE);
1233 return;
1234 }
1235 if (runstate_check(RUN_STATE_INMIGRATE)) {
1236 error_setg(errp, "Guest is waiting for an incoming migration");
1237 return;
1238 }
1239
1240 if (migration_is_blocked(errp)) {
1241 return;
1242 }
1243
1244 if ((has_blk && blk) || (has_inc && inc)) {
1245 if (migrate_use_block() || migrate_use_block_incremental()) {
1246 error_setg(errp, "Command options are incompatible with "
1247 "current migration capabilities");
1248 return;
1249 }
1250 migrate_set_block_enabled(true, &local_err);
1251 if (local_err) {
1252 error_propagate(errp, local_err);
1253 return;
1254 }
1255 s->must_remove_block_options = true;
1256 }
1257
1258 if (has_inc && inc) {
1259 migrate_set_block_incremental(s, true);
1260 }
1261
1262 s = migrate_init();
1263
1264 if (strstart(uri, "tcp:", &p)) {
1265 tcp_start_outgoing_migration(s, p, &local_err);
1266 #ifdef CONFIG_RDMA
1267 } else if (strstart(uri, "rdma:", &p)) {
1268 rdma_start_outgoing_migration(s, p, &local_err);
1269 #endif
1270 } else if (strstart(uri, "exec:", &p)) {
1271 exec_start_outgoing_migration(s, p, &local_err);
1272 } else if (strstart(uri, "unix:", &p)) {
1273 unix_start_outgoing_migration(s, p, &local_err);
1274 } else if (strstart(uri, "fd:", &p)) {
1275 fd_start_outgoing_migration(s, p, &local_err);
1276 } else {
1277 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1278 "a valid migration protocol");
1279 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1280 MIGRATION_STATUS_FAILED);
1281 return;
1282 }
1283
1284 if (local_err) {
1285 migrate_fd_error(s, local_err);
1286 error_propagate(errp, local_err);
1287 return;
1288 }
1289 }
1290
1291 void qmp_migrate_cancel(Error **errp)
1292 {
1293 migrate_fd_cancel(migrate_get_current());
1294 }
1295
1296 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1297 {
1298 MigrationState *s = migrate_get_current();
1299 int64_t new_size;
1300
1301 /* Check for truncation */
1302 if (value != (size_t)value) {
1303 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1304 "exceeding address space");
1305 return;
1306 }
1307
1308 /* Cache should not be larger than guest ram size */
1309 if (value > ram_bytes_total()) {
1310 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1311 "exceeds guest ram size ");
1312 return;
1313 }
1314
1315 new_size = xbzrle_cache_resize(value);
1316 if (new_size < 0) {
1317 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1318 "is smaller than page size");
1319 return;
1320 }
1321
1322 s->xbzrle_cache_size = new_size;
1323 }
1324
1325 int64_t qmp_query_migrate_cache_size(Error **errp)
1326 {
1327 return migrate_xbzrle_cache_size();
1328 }
1329
1330 void qmp_migrate_set_speed(int64_t value, Error **errp)
1331 {
1332 MigrateSetParameters p = {
1333 .has_max_bandwidth = true,
1334 .max_bandwidth = value,
1335 };
1336
1337 qmp_migrate_set_parameters(&p, errp);
1338 }
1339
1340 void qmp_migrate_set_downtime(double value, Error **errp)
1341 {
1342 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1343 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1344 "the range of 0 to %d seconds",
1345 MAX_MIGRATE_DOWNTIME_SECONDS);
1346 return;
1347 }
1348
1349 value *= 1000; /* Convert to milliseconds */
1350 value = MAX(0, MIN(INT64_MAX, value));
1351
1352 MigrateSetParameters p = {
1353 .has_downtime_limit = true,
1354 .downtime_limit = value,
1355 };
1356
1357 qmp_migrate_set_parameters(&p, errp);
1358 }
1359
1360 bool migrate_release_ram(void)
1361 {
1362 MigrationState *s;
1363
1364 s = migrate_get_current();
1365
1366 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1367 }
1368
1369 bool migrate_postcopy_ram(void)
1370 {
1371 MigrationState *s;
1372
1373 s = migrate_get_current();
1374
1375 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1376 }
1377
1378 bool migrate_auto_converge(void)
1379 {
1380 MigrationState *s;
1381
1382 s = migrate_get_current();
1383
1384 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1385 }
1386
1387 bool migrate_zero_blocks(void)
1388 {
1389 MigrationState *s;
1390
1391 s = migrate_get_current();
1392
1393 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1394 }
1395
1396 bool migrate_use_compression(void)
1397 {
1398 MigrationState *s;
1399
1400 s = migrate_get_current();
1401
1402 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
1403 }
1404
1405 int migrate_compress_level(void)
1406 {
1407 MigrationState *s;
1408
1409 s = migrate_get_current();
1410
1411 return s->parameters.compress_level;
1412 }
1413
1414 int migrate_compress_threads(void)
1415 {
1416 MigrationState *s;
1417
1418 s = migrate_get_current();
1419
1420 return s->parameters.compress_threads;
1421 }
1422
1423 int migrate_decompress_threads(void)
1424 {
1425 MigrationState *s;
1426
1427 s = migrate_get_current();
1428
1429 return s->parameters.decompress_threads;
1430 }
1431
1432 bool migrate_use_events(void)
1433 {
1434 MigrationState *s;
1435
1436 s = migrate_get_current();
1437
1438 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1439 }
1440
1441 int migrate_use_xbzrle(void)
1442 {
1443 MigrationState *s;
1444
1445 s = migrate_get_current();
1446
1447 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1448 }
1449
1450 int64_t migrate_xbzrle_cache_size(void)
1451 {
1452 MigrationState *s;
1453
1454 s = migrate_get_current();
1455
1456 return s->xbzrle_cache_size;
1457 }
1458
1459 bool migrate_use_block(void)
1460 {
1461 MigrationState *s;
1462
1463 s = migrate_get_current();
1464
1465 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
1466 }
1467
1468 bool migrate_use_return_path(void)
1469 {
1470 MigrationState *s;
1471
1472 s = migrate_get_current();
1473
1474 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
1475 }
1476
1477 bool migrate_use_block_incremental(void)
1478 {
1479 MigrationState *s;
1480
1481 s = migrate_get_current();
1482
1483 return s->parameters.block_incremental;
1484 }
1485
1486 /* migration thread support */
1487 /*
1488 * Something bad happened to the RP stream, mark an error
1489 * The caller shall print or trace something to indicate why
1490 */
1491 static void mark_source_rp_bad(MigrationState *s)
1492 {
1493 s->rp_state.error = true;
1494 }
1495
1496 static struct rp_cmd_args {
1497 ssize_t len; /* -1 = variable */
1498 const char *name;
1499 } rp_cmd_args[] = {
1500 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
1501 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
1502 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
1503 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
1504 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
1505 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
1506 };
1507
1508 /*
1509 * Process a request for pages received on the return path,
1510 * We're allowed to send more than requested (e.g. to round to our page size)
1511 * and we don't need to send pages that have already been sent.
1512 */
1513 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1514 ram_addr_t start, size_t len)
1515 {
1516 long our_host_ps = getpagesize();
1517
1518 trace_migrate_handle_rp_req_pages(rbname, start, len);
1519
1520 /*
1521 * Since we currently insist on matching page sizes, just sanity check
1522 * we're being asked for whole host pages.
1523 */
1524 if (start & (our_host_ps-1) ||
1525 (len & (our_host_ps-1))) {
1526 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1527 " len: %zd", __func__, start, len);
1528 mark_source_rp_bad(ms);
1529 return;
1530 }
1531
1532 if (ram_save_queue_pages(rbname, start, len)) {
1533 mark_source_rp_bad(ms);
1534 }
1535 }
1536
1537 /*
1538 * Handles messages sent on the return path towards the source VM
1539 *
1540 */
1541 static void *source_return_path_thread(void *opaque)
1542 {
1543 MigrationState *ms = opaque;
1544 QEMUFile *rp = ms->rp_state.from_dst_file;
1545 uint16_t header_len, header_type;
1546 uint8_t buf[512];
1547 uint32_t tmp32, sibling_error;
1548 ram_addr_t start = 0; /* =0 to silence warning */
1549 size_t len = 0, expected_len;
1550 int res;
1551
1552 trace_source_return_path_thread_entry();
1553 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
1554 migration_is_setup_or_active(ms->state)) {
1555 trace_source_return_path_thread_loop_top();
1556 header_type = qemu_get_be16(rp);
1557 header_len = qemu_get_be16(rp);
1558
1559 if (header_type >= MIG_RP_MSG_MAX ||
1560 header_type == MIG_RP_MSG_INVALID) {
1561 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1562 header_type, header_len);
1563 mark_source_rp_bad(ms);
1564 goto out;
1565 }
1566
1567 if ((rp_cmd_args[header_type].len != -1 &&
1568 header_len != rp_cmd_args[header_type].len) ||
1569 header_len > sizeof(buf)) {
1570 error_report("RP: Received '%s' message (0x%04x) with"
1571 "incorrect length %d expecting %zu",
1572 rp_cmd_args[header_type].name, header_type, header_len,
1573 (size_t)rp_cmd_args[header_type].len);
1574 mark_source_rp_bad(ms);
1575 goto out;
1576 }
1577
1578 /* We know we've got a valid header by this point */
1579 res = qemu_get_buffer(rp, buf, header_len);
1580 if (res != header_len) {
1581 error_report("RP: Failed reading data for message 0x%04x"
1582 " read %d expected %d",
1583 header_type, res, header_len);
1584 mark_source_rp_bad(ms);
1585 goto out;
1586 }
1587
1588 /* OK, we have the message and the data */
1589 switch (header_type) {
1590 case MIG_RP_MSG_SHUT:
1591 sibling_error = ldl_be_p(buf);
1592 trace_source_return_path_thread_shut(sibling_error);
1593 if (sibling_error) {
1594 error_report("RP: Sibling indicated error %d", sibling_error);
1595 mark_source_rp_bad(ms);
1596 }
1597 /*
1598 * We'll let the main thread deal with closing the RP
1599 * we could do a shutdown(2) on it, but we're the only user
1600 * anyway, so there's nothing gained.
1601 */
1602 goto out;
1603
1604 case MIG_RP_MSG_PONG:
1605 tmp32 = ldl_be_p(buf);
1606 trace_source_return_path_thread_pong(tmp32);
1607 break;
1608
1609 case MIG_RP_MSG_REQ_PAGES:
1610 start = ldq_be_p(buf);
1611 len = ldl_be_p(buf + 8);
1612 migrate_handle_rp_req_pages(ms, NULL, start, len);
1613 break;
1614
1615 case MIG_RP_MSG_REQ_PAGES_ID:
1616 expected_len = 12 + 1; /* header + termination */
1617
1618 if (header_len >= expected_len) {
1619 start = ldq_be_p(buf);
1620 len = ldl_be_p(buf + 8);
1621 /* Now we expect an idstr */
1622 tmp32 = buf[12]; /* Length of the following idstr */
1623 buf[13 + tmp32] = '\0';
1624 expected_len += tmp32;
1625 }
1626 if (header_len != expected_len) {
1627 error_report("RP: Req_Page_id with length %d expecting %zd",
1628 header_len, expected_len);
1629 mark_source_rp_bad(ms);
1630 goto out;
1631 }
1632 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
1633 break;
1634
1635 default:
1636 break;
1637 }
1638 }
1639 if (qemu_file_get_error(rp)) {
1640 trace_source_return_path_thread_bad_end();
1641 mark_source_rp_bad(ms);
1642 }
1643
1644 trace_source_return_path_thread_end();
1645 out:
1646 ms->rp_state.from_dst_file = NULL;
1647 qemu_fclose(rp);
1648 return NULL;
1649 }
1650
1651 static int open_return_path_on_source(MigrationState *ms)
1652 {
1653
1654 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
1655 if (!ms->rp_state.from_dst_file) {
1656 return -1;
1657 }
1658
1659 trace_open_return_path_on_source();
1660 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
1661 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
1662
1663 trace_open_return_path_on_source_continue();
1664
1665 return 0;
1666 }
1667
1668 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1669 static int await_return_path_close_on_source(MigrationState *ms)
1670 {
1671 /*
1672 * If this is a normal exit then the destination will send a SHUT and the
1673 * rp_thread will exit, however if there's an error we need to cause
1674 * it to exit.
1675 */
1676 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
1677 /*
1678 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1679 * waiting for the destination.
1680 */
1681 qemu_file_shutdown(ms->rp_state.from_dst_file);
1682 mark_source_rp_bad(ms);
1683 }
1684 trace_await_return_path_close_on_source_joining();
1685 qemu_thread_join(&ms->rp_state.rp_thread);
1686 trace_await_return_path_close_on_source_close();
1687 return ms->rp_state.error;
1688 }
1689
1690 /*
1691 * Switch from normal iteration to postcopy
1692 * Returns non-0 on error
1693 */
1694 static int postcopy_start(MigrationState *ms, bool *old_vm_running)
1695 {
1696 int ret;
1697 QIOChannelBuffer *bioc;
1698 QEMUFile *fb;
1699 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1700 bool restart_block = false;
1701 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
1702 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1703
1704 trace_postcopy_start();
1705 qemu_mutex_lock_iothread();
1706 trace_postcopy_start_set_run();
1707
1708 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1709 *old_vm_running = runstate_is_running();
1710 global_state_store();
1711 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1712 if (ret < 0) {
1713 goto fail;
1714 }
1715
1716 ret = bdrv_inactivate_all();
1717 if (ret < 0) {
1718 goto fail;
1719 }
1720 restart_block = true;
1721
1722 /*
1723 * Cause any non-postcopiable, but iterative devices to
1724 * send out their final data.
1725 */
1726 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
1727
1728 /*
1729 * in Finish migrate and with the io-lock held everything should
1730 * be quiet, but we've potentially still got dirty pages and we
1731 * need to tell the destination to throw any pages it's already received
1732 * that are dirty
1733 */
1734 if (ram_postcopy_send_discard_bitmap(ms)) {
1735 error_report("postcopy send discard bitmap failed");
1736 goto fail;
1737 }
1738
1739 /*
1740 * send rest of state - note things that are doing postcopy
1741 * will notice we're in POSTCOPY_ACTIVE and not actually
1742 * wrap their state up here
1743 */
1744 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
1745 /* Ping just for debugging, helps line traces up */
1746 qemu_savevm_send_ping(ms->to_dst_file, 2);
1747
1748 /*
1749 * While loading the device state we may trigger page transfer
1750 * requests and the fd must be free to process those, and thus
1751 * the destination must read the whole device state off the fd before
1752 * it starts processing it. Unfortunately the ad-hoc migration format
1753 * doesn't allow the destination to know the size to read without fully
1754 * parsing it through each devices load-state code (especially the open
1755 * coded devices that use get/put).
1756 * So we wrap the device state up in a package with a length at the start;
1757 * to do this we use a qemu_buf to hold the whole of the device state.
1758 */
1759 bioc = qio_channel_buffer_new(4096);
1760 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
1761 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
1762 object_unref(OBJECT(bioc));
1763
1764 /*
1765 * Make sure the receiver can get incoming pages before we send the rest
1766 * of the state
1767 */
1768 qemu_savevm_send_postcopy_listen(fb);
1769
1770 qemu_savevm_state_complete_precopy(fb, false, false);
1771 qemu_savevm_send_ping(fb, 3);
1772
1773 qemu_savevm_send_postcopy_run(fb);
1774
1775 /* <><> end of stuff going into the package */
1776
1777 /* Last point of recovery; as soon as we send the package the destination
1778 * can open devices and potentially start running.
1779 * Lets just check again we've not got any errors.
1780 */
1781 ret = qemu_file_get_error(ms->to_dst_file);
1782 if (ret) {
1783 error_report("postcopy_start: Migration stream errored (pre package)");
1784 goto fail_closefb;
1785 }
1786
1787 restart_block = false;
1788
1789 /* Now send that blob */
1790 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
1791 goto fail_closefb;
1792 }
1793 qemu_fclose(fb);
1794
1795 /* Send a notify to give a chance for anything that needs to happen
1796 * at the transition to postcopy and after the device state; in particular
1797 * spice needs to trigger a transition now
1798 */
1799 ms->postcopy_after_devices = true;
1800 notifier_list_notify(&migration_state_notifiers, ms);
1801
1802 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
1803
1804 qemu_mutex_unlock_iothread();
1805
1806 /*
1807 * Although this ping is just for debug, it could potentially be
1808 * used for getting a better measurement of downtime at the source.
1809 */
1810 qemu_savevm_send_ping(ms->to_dst_file, 4);
1811
1812 if (migrate_release_ram()) {
1813 ram_postcopy_migrated_memory_release(ms);
1814 }
1815
1816 ret = qemu_file_get_error(ms->to_dst_file);
1817 if (ret) {
1818 error_report("postcopy_start: Migration stream errored");
1819 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1820 MIGRATION_STATUS_FAILED);
1821 }
1822
1823 return ret;
1824
1825 fail_closefb:
1826 qemu_fclose(fb);
1827 fail:
1828 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1829 MIGRATION_STATUS_FAILED);
1830 if (restart_block) {
1831 /* A failure happened early enough that we know the destination hasn't
1832 * accessed block devices, so we're safe to recover.
1833 */
1834 Error *local_err = NULL;
1835
1836 bdrv_invalidate_cache_all(&local_err);
1837 if (local_err) {
1838 error_report_err(local_err);
1839 }
1840 }
1841 qemu_mutex_unlock_iothread();
1842 return -1;
1843 }
1844
1845 /**
1846 * migration_completion: Used by migration_thread when there's not much left.
1847 * The caller 'breaks' the loop when this returns.
1848 *
1849 * @s: Current migration state
1850 * @current_active_state: The migration state we expect to be in
1851 * @*old_vm_running: Pointer to old_vm_running flag
1852 * @*start_time: Pointer to time to update
1853 */
1854 static void migration_completion(MigrationState *s, int current_active_state,
1855 bool *old_vm_running,
1856 int64_t *start_time)
1857 {
1858 int ret;
1859
1860 if (s->state == MIGRATION_STATUS_ACTIVE) {
1861 qemu_mutex_lock_iothread();
1862 *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1863 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1864 *old_vm_running = runstate_is_running();
1865 ret = global_state_store();
1866
1867 if (!ret) {
1868 bool inactivate = !migrate_colo_enabled();
1869 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1870 if (ret >= 0) {
1871 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
1872 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
1873 inactivate);
1874 }
1875 if (inactivate && ret >= 0) {
1876 s->block_inactive = true;
1877 }
1878 }
1879 qemu_mutex_unlock_iothread();
1880
1881 if (ret < 0) {
1882 goto fail;
1883 }
1884 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1885 trace_migration_completion_postcopy_end();
1886
1887 qemu_savevm_state_complete_postcopy(s->to_dst_file);
1888 trace_migration_completion_postcopy_end_after_complete();
1889 }
1890
1891 /*
1892 * If rp was opened we must clean up the thread before
1893 * cleaning everything else up (since if there are no failures
1894 * it will wait for the destination to send it's status in
1895 * a SHUT command).
1896 */
1897 if (s->rp_state.from_dst_file) {
1898 int rp_error;
1899 trace_migration_return_path_end_before();
1900 rp_error = await_return_path_close_on_source(s);
1901 trace_migration_return_path_end_after(rp_error);
1902 if (rp_error) {
1903 goto fail_invalidate;
1904 }
1905 }
1906
1907 if (qemu_file_get_error(s->to_dst_file)) {
1908 trace_migration_completion_file_err();
1909 goto fail_invalidate;
1910 }
1911
1912 if (!migrate_colo_enabled()) {
1913 migrate_set_state(&s->state, current_active_state,
1914 MIGRATION_STATUS_COMPLETED);
1915 }
1916
1917 return;
1918
1919 fail_invalidate:
1920 /* If not doing postcopy, vm_start() will be called: let's regain
1921 * control on images.
1922 */
1923 if (s->state == MIGRATION_STATUS_ACTIVE) {
1924 Error *local_err = NULL;
1925
1926 qemu_mutex_lock_iothread();
1927 bdrv_invalidate_cache_all(&local_err);
1928 if (local_err) {
1929 error_report_err(local_err);
1930 } else {
1931 s->block_inactive = false;
1932 }
1933 qemu_mutex_unlock_iothread();
1934 }
1935
1936 fail:
1937 migrate_set_state(&s->state, current_active_state,
1938 MIGRATION_STATUS_FAILED);
1939 }
1940
1941 bool migrate_colo_enabled(void)
1942 {
1943 MigrationState *s = migrate_get_current();
1944 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
1945 }
1946
1947 /*
1948 * Master migration thread on the source VM.
1949 * It drives the migration and pumps the data down the outgoing channel.
1950 */
1951 static void *migration_thread(void *opaque)
1952 {
1953 MigrationState *s = opaque;
1954 /* Used by the bandwidth calcs, updated later */
1955 int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1956 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
1957 int64_t initial_bytes = 0;
1958 /*
1959 * The final stage happens when the remaining data is smaller than
1960 * this threshold; it's calculated from the requested downtime and
1961 * measured bandwidth
1962 */
1963 int64_t threshold_size = 0;
1964 int64_t start_time = initial_time;
1965 int64_t end_time;
1966 bool old_vm_running = false;
1967 bool entered_postcopy = false;
1968 /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
1969 enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE;
1970 bool enable_colo = migrate_colo_enabled();
1971
1972 rcu_register_thread();
1973
1974 qemu_savevm_state_header(s->to_dst_file);
1975
1976 /*
1977 * If we opened the return path, we need to make sure dst has it
1978 * opened as well.
1979 */
1980 if (s->rp_state.from_dst_file) {
1981 /* Now tell the dest that it should open its end so it can reply */
1982 qemu_savevm_send_open_return_path(s->to_dst_file);
1983
1984 /* And do a ping that will make stuff easier to debug */
1985 qemu_savevm_send_ping(s->to_dst_file, 1);
1986 }
1987
1988 if (migrate_postcopy_ram()) {
1989 /*
1990 * Tell the destination that we *might* want to do postcopy later;
1991 * if the other end can't do postcopy it should fail now, nice and
1992 * early.
1993 */
1994 qemu_savevm_send_postcopy_advise(s->to_dst_file);
1995 }
1996
1997 qemu_savevm_state_setup(s->to_dst_file);
1998
1999 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
2000 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2001 MIGRATION_STATUS_ACTIVE);
2002
2003 trace_migration_thread_setup_complete();
2004
2005 while (s->state == MIGRATION_STATUS_ACTIVE ||
2006 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2007 int64_t current_time;
2008 uint64_t pending_size;
2009
2010 if (!qemu_file_rate_limit(s->to_dst_file)) {
2011 uint64_t pend_post, pend_nonpost;
2012
2013 qemu_savevm_state_pending(s->to_dst_file, threshold_size,
2014 &pend_nonpost, &pend_post);
2015 pending_size = pend_nonpost + pend_post;
2016 trace_migrate_pending(pending_size, threshold_size,
2017 pend_post, pend_nonpost);
2018 if (pending_size && pending_size >= threshold_size) {
2019 /* Still a significant amount to transfer */
2020
2021 if (migrate_postcopy_ram() &&
2022 s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE &&
2023 pend_nonpost <= threshold_size &&
2024 atomic_read(&s->start_postcopy)) {
2025
2026 if (!postcopy_start(s, &old_vm_running)) {
2027 current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE;
2028 entered_postcopy = true;
2029 }
2030
2031 continue;
2032 }
2033 /* Just another iteration step */
2034 qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy);
2035 } else {
2036 trace_migration_thread_low_pending(pending_size);
2037 migration_completion(s, current_active_state,
2038 &old_vm_running, &start_time);
2039 break;
2040 }
2041 }
2042
2043 if (qemu_file_get_error(s->to_dst_file)) {
2044 migrate_set_state(&s->state, current_active_state,
2045 MIGRATION_STATUS_FAILED);
2046 trace_migration_thread_file_err();
2047 break;
2048 }
2049 current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2050 if (current_time >= initial_time + BUFFER_DELAY) {
2051 uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) -
2052 initial_bytes;
2053 uint64_t time_spent = current_time - initial_time;
2054 double bandwidth = (double)transferred_bytes / time_spent;
2055 threshold_size = bandwidth * s->parameters.downtime_limit;
2056
2057 s->mbps = (((double) transferred_bytes * 8.0) /
2058 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
2059
2060 trace_migrate_transferred(transferred_bytes, time_spent,
2061 bandwidth, threshold_size);
2062 /* if we haven't sent anything, we don't want to recalculate
2063 10000 is a small enough number for our purposes */
2064 if (ram_counters.dirty_pages_rate && transferred_bytes > 10000) {
2065 s->expected_downtime = ram_counters.dirty_pages_rate *
2066 qemu_target_page_size() / bandwidth;
2067 }
2068
2069 qemu_file_reset_rate_limit(s->to_dst_file);
2070 initial_time = current_time;
2071 initial_bytes = qemu_ftell(s->to_dst_file);
2072 }
2073 if (qemu_file_rate_limit(s->to_dst_file)) {
2074 /* usleep expects microseconds */
2075 g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
2076 }
2077 }
2078
2079 trace_migration_thread_after_loop();
2080 /* If we enabled cpu throttling for auto-converge, turn it off. */
2081 cpu_throttle_stop();
2082 end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2083
2084 qemu_mutex_lock_iothread();
2085 /*
2086 * The resource has been allocated by migration will be reused in COLO
2087 * process, so don't release them.
2088 */
2089 if (!enable_colo) {
2090 qemu_savevm_state_cleanup();
2091 }
2092 if (s->state == MIGRATION_STATUS_COMPLETED) {
2093 uint64_t transferred_bytes = qemu_ftell(s->to_dst_file);
2094 s->total_time = end_time - s->total_time;
2095 if (!entered_postcopy) {
2096 s->downtime = end_time - start_time;
2097 }
2098 if (s->total_time) {
2099 s->mbps = (((double) transferred_bytes * 8.0) /
2100 ((double) s->total_time)) / 1000;
2101 }
2102 runstate_set(RUN_STATE_POSTMIGRATE);
2103 } else {
2104 if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) {
2105 migrate_start_colo_process(s);
2106 qemu_savevm_state_cleanup();
2107 /*
2108 * Fixme: we will run VM in COLO no matter its old running state.
2109 * After exited COLO, we will keep running.
2110 */
2111 old_vm_running = true;
2112 }
2113 if (old_vm_running && !entered_postcopy) {
2114 vm_start();
2115 } else {
2116 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2117 runstate_set(RUN_STATE_POSTMIGRATE);
2118 }
2119 }
2120 }
2121 qemu_bh_schedule(s->cleanup_bh);
2122 qemu_mutex_unlock_iothread();
2123
2124 rcu_unregister_thread();
2125 return NULL;
2126 }
2127
2128 void migrate_fd_connect(MigrationState *s)
2129 {
2130 s->expected_downtime = s->parameters.downtime_limit;
2131 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
2132
2133 qemu_file_set_blocking(s->to_dst_file, true);
2134 qemu_file_set_rate_limit(s->to_dst_file,
2135 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
2136
2137 /* Notify before starting migration thread */
2138 notifier_list_notify(&migration_state_notifiers, s);
2139
2140 /*
2141 * Open the return path. For postcopy, it is used exclusively. For
2142 * precopy, only if user specified "return-path" capability would
2143 * QEMU uses the return path.
2144 */
2145 if (migrate_postcopy_ram() || migrate_use_return_path()) {
2146 if (open_return_path_on_source(s)) {
2147 error_report("Unable to open return-path for postcopy");
2148 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2149 MIGRATION_STATUS_FAILED);
2150 migrate_fd_cleanup(s);
2151 return;
2152 }
2153 }
2154
2155 qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
2156 QEMU_THREAD_JOINABLE);
2157 s->migration_thread_running = true;
2158 }
2159
2160 void migration_global_dump(Monitor *mon)
2161 {
2162 MigrationState *ms = migrate_get_current();
2163
2164 monitor_printf(mon, "globals: store-global-state=%d, only_migratable=%d, "
2165 "send-configuration=%d, send-section-footer=%d\n",
2166 ms->store_global_state, ms->only_migratable,
2167 ms->send_configuration, ms->send_section_footer);
2168 }
2169
2170 #define DEFINE_PROP_MIG_CAP(name, x) \
2171 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
2172
2173 static Property migration_properties[] = {
2174 DEFINE_PROP_BOOL("store-global-state", MigrationState,
2175 store_global_state, true),
2176 DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false),
2177 DEFINE_PROP_BOOL("send-configuration", MigrationState,
2178 send_configuration, true),
2179 DEFINE_PROP_BOOL("send-section-footer", MigrationState,
2180 send_section_footer, true),
2181
2182 /* Migration parameters */
2183 DEFINE_PROP_INT64("x-compress-level", MigrationState,
2184 parameters.compress_level,
2185 DEFAULT_MIGRATE_COMPRESS_LEVEL),
2186 DEFINE_PROP_INT64("x-compress-threads", MigrationState,
2187 parameters.compress_threads,
2188 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
2189 DEFINE_PROP_INT64("x-decompress-threads", MigrationState,
2190 parameters.decompress_threads,
2191 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
2192 DEFINE_PROP_INT64("x-cpu-throttle-initial", MigrationState,
2193 parameters.cpu_throttle_initial,
2194 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
2195 DEFINE_PROP_INT64("x-cpu-throttle-increment", MigrationState,
2196 parameters.cpu_throttle_increment,
2197 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
2198 DEFINE_PROP_INT64("x-max-bandwidth", MigrationState,
2199 parameters.max_bandwidth, MAX_THROTTLE),
2200 DEFINE_PROP_INT64("x-downtime-limit", MigrationState,
2201 parameters.downtime_limit,
2202 DEFAULT_MIGRATE_SET_DOWNTIME),
2203 DEFINE_PROP_INT64("x-checkpoint-delay", MigrationState,
2204 parameters.x_checkpoint_delay,
2205 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
2206
2207 /* Migration capabilities */
2208 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
2209 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
2210 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
2211 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
2212 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
2213 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
2214 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
2215 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
2216 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
2217 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
2218 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
2219
2220 DEFINE_PROP_END_OF_LIST(),
2221 };
2222
2223 static void migration_class_init(ObjectClass *klass, void *data)
2224 {
2225 DeviceClass *dc = DEVICE_CLASS(klass);
2226
2227 dc->user_creatable = false;
2228 dc->props = migration_properties;
2229 }
2230
2231 static void migration_instance_finalize(Object *obj)
2232 {
2233 MigrationState *ms = MIGRATION_OBJ(obj);
2234 MigrationParameters *params = &ms->parameters;
2235
2236 g_free(params->tls_hostname);
2237 g_free(params->tls_creds);
2238 }
2239
2240 static void migration_instance_init(Object *obj)
2241 {
2242 MigrationState *ms = MIGRATION_OBJ(obj);
2243 MigrationParameters *params = &ms->parameters;
2244
2245 ms->state = MIGRATION_STATUS_NONE;
2246 ms->xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE;
2247 ms->mbps = -1;
2248
2249 params->tls_hostname = g_strdup("");
2250 params->tls_creds = g_strdup("");
2251
2252 /* Set has_* up only for parameter checks */
2253 params->has_compress_level = true;
2254 params->has_compress_threads = true;
2255 params->has_decompress_threads = true;
2256 params->has_cpu_throttle_initial = true;
2257 params->has_cpu_throttle_increment = true;
2258 params->has_max_bandwidth = true;
2259 params->has_downtime_limit = true;
2260 params->has_x_checkpoint_delay = true;
2261 params->has_block_incremental = true;
2262 }
2263
2264 /*
2265 * Return true if check pass, false otherwise. Error will be put
2266 * inside errp if provided.
2267 */
2268 static bool migration_object_check(MigrationState *ms, Error **errp)
2269 {
2270 MigrationCapabilityStatusList *head = NULL;
2271 /* Assuming all off */
2272 bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
2273 int i;
2274
2275 if (!migrate_params_check(&ms->parameters, errp)) {
2276 return false;
2277 }
2278
2279 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
2280 if (ms->enabled_capabilities[i]) {
2281 head = migrate_cap_add(head, i, true);
2282 }
2283 }
2284
2285 ret = migrate_caps_check(cap_list, head, errp);
2286
2287 /* It works with head == NULL */
2288 qapi_free_MigrationCapabilityStatusList(head);
2289
2290 return ret;
2291 }
2292
2293 static const TypeInfo migration_type = {
2294 .name = TYPE_MIGRATION,
2295 /*
2296 * NOTE: TYPE_MIGRATION is not really a device, as the object is
2297 * not created using qdev_create(), it is not attached to the qdev
2298 * device tree, and it is never realized.
2299 *
2300 * TODO: Make this TYPE_OBJECT once QOM provides something like
2301 * TYPE_DEVICE's "-global" properties.
2302 */
2303 .parent = TYPE_DEVICE,
2304 .class_init = migration_class_init,
2305 .class_size = sizeof(MigrationClass),
2306 .instance_size = sizeof(MigrationState),
2307 .instance_init = migration_instance_init,
2308 .instance_finalize = migration_instance_finalize,
2309 };
2310
2311 static void register_migration_types(void)
2312 {
2313 type_register_static(&migration_type);
2314 }
2315
2316 type_init(register_migration_types);
AR7 emulation for QEMU