?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
migration: pause-before-switchover for postcopy
[qemu.git] / migration / migration.c
bloba058f8b46dfbd37fea0727c4ee79e9e587ecfa40
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 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
81 #define DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT 16
82
83 static NotifierList migration_state_notifiers =
84 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
85
86 static bool deferred_incoming;
87
88 /* Messages sent on the return path from destination to source */
89 enum mig_rp_message_type {
90 MIG_RP_MSG_INVALID = 0, /* Must be 0 */
91 MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */
92 MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */
93
94 MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
95 MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */
96
97 MIG_RP_MSG_MAX
98 };
99
100 /* When we add fault tolerance, we could have several
101 migrations at once. For now we don't need to add
102 dynamic creation of migration */
103
104 static MigrationState *current_migration;
105
106 static bool migration_object_check(MigrationState *ms, Error **errp);
107 static int migration_maybe_pause(MigrationState *s,
108 int *current_active_state,
109 int new_state);
110
111 void migration_object_init(void)
112 {
113 MachineState *ms = MACHINE(qdev_get_machine());
114 Error *err = NULL;
115
116 /* This can only be called once. */
117 assert(!current_migration);
118 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
119
120 if (!migration_object_check(current_migration, &err)) {
121 error_report_err(err);
122 exit(1);
123 }
124
125 /*
126 * We cannot really do this in migration_instance_init() since at
127 * that time global properties are not yet applied, then this
128 * value will be definitely replaced by something else.
129 */
130 if (ms->enforce_config_section) {
131 current_migration->send_configuration = true;
132 }
133 }
134
135 /* For outgoing */
136 MigrationState *migrate_get_current(void)
137 {
138 /* This can only be called after the object created. */
139 assert(current_migration);
140 return current_migration;
141 }
142
143 MigrationIncomingState *migration_incoming_get_current(void)
144 {
145 static bool once;
146 static MigrationIncomingState mis_current;
147
148 if (!once) {
149 mis_current.state = MIGRATION_STATUS_NONE;
150 memset(&mis_current, 0, sizeof(MigrationIncomingState));
151 qemu_mutex_init(&mis_current.rp_mutex);
152 qemu_event_init(&mis_current.main_thread_load_event, false);
153 once = true;
154 }
155 return &mis_current;
156 }
157
158 void migration_incoming_state_destroy(void)
159 {
160 struct MigrationIncomingState *mis = migration_incoming_get_current();
161
162 if (mis->to_src_file) {
163 /* Tell source that we are done */
164 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
165 qemu_fclose(mis->to_src_file);
166 mis->to_src_file = NULL;
167 }
168
169 if (mis->from_src_file) {
170 qemu_fclose(mis->from_src_file);
171 mis->from_src_file = NULL;
172 }
173
174 qemu_event_reset(&mis->main_thread_load_event);
175 }
176
177 static void migrate_generate_event(int new_state)
178 {
179 if (migrate_use_events()) {
180 qapi_event_send_migration(new_state, &error_abort);
181 }
182 }
183
184 /*
185 * Called on -incoming with a defer: uri.
186 * The migration can be started later after any parameters have been
187 * changed.
188 */
189 static void deferred_incoming_migration(Error **errp)
190 {
191 if (deferred_incoming) {
192 error_setg(errp, "Incoming migration already deferred");
193 }
194 deferred_incoming = true;
195 }
196
197 /*
198 * Send a message on the return channel back to the source
199 * of the migration.
200 */
201 static void migrate_send_rp_message(MigrationIncomingState *mis,
202 enum mig_rp_message_type message_type,
203 uint16_t len, void *data)
204 {
205 trace_migrate_send_rp_message((int)message_type, len);
206 qemu_mutex_lock(&mis->rp_mutex);
207 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
208 qemu_put_be16(mis->to_src_file, len);
209 qemu_put_buffer(mis->to_src_file, data, len);
210 qemu_fflush(mis->to_src_file);
211 qemu_mutex_unlock(&mis->rp_mutex);
212 }
213
214 /* Request a range of pages from the source VM at the given
215 * start address.
216 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
217 * as the last request (a name must have been given previously)
218 * Start: Address offset within the RB
219 * Len: Length in bytes required - must be a multiple of pagesize
220 */
221 void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
222 ram_addr_t start, size_t len)
223 {
224 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
225 size_t msglen = 12; /* start + len */
226
227 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
228 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
229
230 if (rbname) {
231 int rbname_len = strlen(rbname);
232 assert(rbname_len < 256);
233
234 bufc[msglen++] = rbname_len;
235 memcpy(bufc + msglen, rbname, rbname_len);
236 msglen += rbname_len;
237 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc);
238 } else {
239 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc);
240 }
241 }
242
243 void qemu_start_incoming_migration(const char *uri, Error **errp)
244 {
245 const char *p;
246
247 qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
248 if (!strcmp(uri, "defer")) {
249 deferred_incoming_migration(errp);
250 } else if (strstart(uri, "tcp:", &p)) {
251 tcp_start_incoming_migration(p, errp);
252 #ifdef CONFIG_RDMA
253 } else if (strstart(uri, "rdma:", &p)) {
254 rdma_start_incoming_migration(p, errp);
255 #endif
256 } else if (strstart(uri, "exec:", &p)) {
257 exec_start_incoming_migration(p, errp);
258 } else if (strstart(uri, "unix:", &p)) {
259 unix_start_incoming_migration(p, errp);
260 } else if (strstart(uri, "fd:", &p)) {
261 fd_start_incoming_migration(p, errp);
262 } else {
263 error_setg(errp, "unknown migration protocol: %s", uri);
264 }
265 }
266
267 static void process_incoming_migration_bh(void *opaque)
268 {
269 Error *local_err = NULL;
270 MigrationIncomingState *mis = opaque;
271
272 /* Make sure all file formats flush their mutable metadata.
273 * If we get an error here, just don't restart the VM yet. */
274 bdrv_invalidate_cache_all(&local_err);
275 if (local_err) {
276 error_report_err(local_err);
277 local_err = NULL;
278 autostart = false;
279 }
280
281 /*
282 * This must happen after all error conditions are dealt with and
283 * we're sure the VM is going to be running on this host.
284 */
285 qemu_announce_self();
286
287 if (multifd_load_cleanup(&local_err) != 0) {
288 error_report_err(local_err);
289 autostart = false;
290 }
291 /* If global state section was not received or we are in running
292 state, we need to obey autostart. Any other state is set with
293 runstate_set. */
294
295 if (!global_state_received() ||
296 global_state_get_runstate() == RUN_STATE_RUNNING) {
297 if (autostart) {
298 vm_start();
299 } else {
300 runstate_set(RUN_STATE_PAUSED);
301 }
302 } else {
303 runstate_set(global_state_get_runstate());
304 }
305 /*
306 * This must happen after any state changes since as soon as an external
307 * observer sees this event they might start to prod at the VM assuming
308 * it's ready to use.
309 */
310 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
311 MIGRATION_STATUS_COMPLETED);
312 qemu_bh_delete(mis->bh);
313 migration_incoming_state_destroy();
314 }
315
316 static void process_incoming_migration_co(void *opaque)
317 {
318 MigrationIncomingState *mis = migration_incoming_get_current();
319 PostcopyState ps;
320 int ret;
321
322 assert(mis->from_src_file);
323 mis->largest_page_size = qemu_ram_pagesize_largest();
324 postcopy_state_set(POSTCOPY_INCOMING_NONE);
325 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
326 MIGRATION_STATUS_ACTIVE);
327 ret = qemu_loadvm_state(mis->from_src_file);
328
329 ps = postcopy_state_get();
330 trace_process_incoming_migration_co_end(ret, ps);
331 if (ps != POSTCOPY_INCOMING_NONE) {
332 if (ps == POSTCOPY_INCOMING_ADVISE) {
333 /*
334 * Where a migration had postcopy enabled (and thus went to advise)
335 * but managed to complete within the precopy period, we can use
336 * the normal exit.
337 */
338 postcopy_ram_incoming_cleanup(mis);
339 } else if (ret >= 0) {
340 /*
341 * Postcopy was started, cleanup should happen at the end of the
342 * postcopy thread.
343 */
344 trace_process_incoming_migration_co_postcopy_end_main();
345 return;
346 }
347 /* Else if something went wrong then just fall out of the normal exit */
348 }
349
350 /* we get COLO info, and know if we are in COLO mode */
351 if (!ret && migration_incoming_enable_colo()) {
352 mis->migration_incoming_co = qemu_coroutine_self();
353 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
354 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
355 mis->have_colo_incoming_thread = true;
356 qemu_coroutine_yield();
357
358 /* Wait checkpoint incoming thread exit before free resource */
359 qemu_thread_join(&mis->colo_incoming_thread);
360 }
361
362 if (ret < 0) {
363 Error *local_err = NULL;
364
365 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
366 MIGRATION_STATUS_FAILED);
367 error_report("load of migration failed: %s", strerror(-ret));
368 qemu_fclose(mis->from_src_file);
369 if (multifd_load_cleanup(&local_err) != 0) {
370 error_report_err(local_err);
371 }
372 exit(EXIT_FAILURE);
373 }
374 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
375 qemu_bh_schedule(mis->bh);
376 }
377
378 static void migration_incoming_setup(QEMUFile *f)
379 {
380 MigrationIncomingState *mis = migration_incoming_get_current();
381
382 if (multifd_load_setup() != 0) {
383 /* We haven't been able to create multifd threads
384 nothing better to do */
385 exit(EXIT_FAILURE);
386 }
387
388 if (!mis->from_src_file) {
389 mis->from_src_file = f;
390 }
391 qemu_file_set_blocking(f, false);
392 }
393
394 static void migration_incoming_process(void)
395 {
396 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
397 qemu_coroutine_enter(co);
398 }
399
400 void migration_fd_process_incoming(QEMUFile *f)
401 {
402 migration_incoming_setup(f);
403 migration_incoming_process();
404 }
405
406 void migration_ioc_process_incoming(QIOChannel *ioc)
407 {
408 MigrationIncomingState *mis = migration_incoming_get_current();
409
410 if (!mis->from_src_file) {
411 QEMUFile *f = qemu_fopen_channel_input(ioc);
412 migration_fd_process_incoming(f);
413 }
414 /* We still only have a single channel. Nothing to do here yet */
415 }
416
417 /**
418 * @migration_has_all_channels: We have received all channels that we need
419 *
420 * Returns true when we have got connections to all the channels that
421 * we need for migration.
422 */
423 bool migration_has_all_channels(void)
424 {
425 return true;
426 }
427
428 /*
429 * Send a 'SHUT' message on the return channel with the given value
430 * to indicate that we've finished with the RP. Non-0 value indicates
431 * error.
432 */
433 void migrate_send_rp_shut(MigrationIncomingState *mis,
434 uint32_t value)
435 {
436 uint32_t buf;
437
438 buf = cpu_to_be32(value);
439 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
440 }
441
442 /*
443 * Send a 'PONG' message on the return channel with the given value
444 * (normally in response to a 'PING')
445 */
446 void migrate_send_rp_pong(MigrationIncomingState *mis,
447 uint32_t value)
448 {
449 uint32_t buf;
450
451 buf = cpu_to_be32(value);
452 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
453 }
454
455 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
456 {
457 MigrationCapabilityStatusList *head = NULL;
458 MigrationCapabilityStatusList *caps;
459 MigrationState *s = migrate_get_current();
460 int i;
461
462 caps = NULL; /* silence compiler warning */
463 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
464 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
465 if (i == MIGRATION_CAPABILITY_BLOCK) {
466 continue;
467 }
468 #endif
469 if (head == NULL) {
470 head = g_malloc0(sizeof(*caps));
471 caps = head;
472 } else {
473 caps->next = g_malloc0(sizeof(*caps));
474 caps = caps->next;
475 }
476 caps->value =
477 g_malloc(sizeof(*caps->value));
478 caps->value->capability = i;
479 caps->value->state = s->enabled_capabilities[i];
480 }
481
482 return head;
483 }
484
485 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
486 {
487 MigrationParameters *params;
488 MigrationState *s = migrate_get_current();
489
490 /* TODO use QAPI_CLONE() instead of duplicating it inline */
491 params = g_malloc0(sizeof(*params));
492 params->has_compress_level = true;
493 params->compress_level = s->parameters.compress_level;
494 params->has_compress_threads = true;
495 params->compress_threads = s->parameters.compress_threads;
496 params->has_decompress_threads = true;
497 params->decompress_threads = s->parameters.decompress_threads;
498 params->has_cpu_throttle_initial = true;
499 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
500 params->has_cpu_throttle_increment = true;
501 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
502 params->has_tls_creds = true;
503 params->tls_creds = g_strdup(s->parameters.tls_creds);
504 params->has_tls_hostname = true;
505 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
506 params->has_max_bandwidth = true;
507 params->max_bandwidth = s->parameters.max_bandwidth;
508 params->has_downtime_limit = true;
509 params->downtime_limit = s->parameters.downtime_limit;
510 params->has_x_checkpoint_delay = true;
511 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
512 params->has_block_incremental = true;
513 params->block_incremental = s->parameters.block_incremental;
514 params->has_x_multifd_channels = true;
515 params->x_multifd_channels = s->parameters.x_multifd_channels;
516 params->has_x_multifd_page_count = true;
517 params->x_multifd_page_count = s->parameters.x_multifd_page_count;
518
519 return params;
520 }
521
522 /*
523 * Return true if we're already in the middle of a migration
524 * (i.e. any of the active or setup states)
525 */
526 static bool migration_is_setup_or_active(int state)
527 {
528 switch (state) {
529 case MIGRATION_STATUS_ACTIVE:
530 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
531 case MIGRATION_STATUS_SETUP:
532 case MIGRATION_STATUS_PRE_SWITCHOVER:
533 case MIGRATION_STATUS_DEVICE:
534 return true;
535
536 default:
537 return false;
538
539 }
540 }
541
542 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
543 {
544 info->has_ram = true;
545 info->ram = g_malloc0(sizeof(*info->ram));
546 info->ram->transferred = ram_counters.transferred;
547 info->ram->total = ram_bytes_total();
548 info->ram->duplicate = ram_counters.duplicate;
549 /* legacy value. It is not used anymore */
550 info->ram->skipped = 0;
551 info->ram->normal = ram_counters.normal;
552 info->ram->normal_bytes = ram_counters.normal *
553 qemu_target_page_size();
554 info->ram->mbps = s->mbps;
555 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
556 info->ram->postcopy_requests = ram_counters.postcopy_requests;
557 info->ram->page_size = qemu_target_page_size();
558
559 if (migrate_use_xbzrle()) {
560 info->has_xbzrle_cache = true;
561 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
562 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
563 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
564 info->xbzrle_cache->pages = xbzrle_counters.pages;
565 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
566 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
567 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
568 }
569
570 if (cpu_throttle_active()) {
571 info->has_cpu_throttle_percentage = true;
572 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
573 }
574
575 if (s->state != MIGRATION_STATUS_COMPLETED) {
576 info->ram->remaining = ram_bytes_remaining();
577 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
578 }
579 }
580
581 static void populate_disk_info(MigrationInfo *info)
582 {
583 if (blk_mig_active()) {
584 info->has_disk = true;
585 info->disk = g_malloc0(sizeof(*info->disk));
586 info->disk->transferred = blk_mig_bytes_transferred();
587 info->disk->remaining = blk_mig_bytes_remaining();
588 info->disk->total = blk_mig_bytes_total();
589 }
590 }
591
592 MigrationInfo *qmp_query_migrate(Error **errp)
593 {
594 MigrationInfo *info = g_malloc0(sizeof(*info));
595 MigrationState *s = migrate_get_current();
596
597 switch (s->state) {
598 case MIGRATION_STATUS_NONE:
599 /* no migration has happened ever */
600 break;
601 case MIGRATION_STATUS_SETUP:
602 info->has_status = true;
603 info->has_total_time = false;
604 break;
605 case MIGRATION_STATUS_ACTIVE:
606 case MIGRATION_STATUS_CANCELLING:
607 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
608 case MIGRATION_STATUS_PRE_SWITCHOVER:
609 case MIGRATION_STATUS_DEVICE:
610 /* TODO add some postcopy stats */
611 info->has_status = true;
612 info->has_total_time = true;
613 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
614 - s->total_time;
615 info->has_expected_downtime = true;
616 info->expected_downtime = s->expected_downtime;
617 info->has_setup_time = true;
618 info->setup_time = s->setup_time;
619
620 populate_ram_info(info, s);
621 populate_disk_info(info);
622 break;
623 case MIGRATION_STATUS_COLO:
624 info->has_status = true;
625 /* TODO: display COLO specific information (checkpoint info etc.) */
626 break;
627 case MIGRATION_STATUS_COMPLETED:
628 info->has_status = true;
629 info->has_total_time = true;
630 info->total_time = s->total_time;
631 info->has_downtime = true;
632 info->downtime = s->downtime;
633 info->has_setup_time = true;
634 info->setup_time = s->setup_time;
635
636 populate_ram_info(info, s);
637 break;
638 case MIGRATION_STATUS_FAILED:
639 info->has_status = true;
640 if (s->error) {
641 info->has_error_desc = true;
642 info->error_desc = g_strdup(error_get_pretty(s->error));
643 }
644 break;
645 case MIGRATION_STATUS_CANCELLED:
646 info->has_status = true;
647 break;
648 }
649 info->status = s->state;
650
651 return info;
652 }
653
654 /**
655 * @migration_caps_check - check capability validity
656 *
657 * @cap_list: old capability list, array of bool
658 * @params: new capabilities to be applied soon
659 * @errp: set *errp if the check failed, with reason
660 *
661 * Returns true if check passed, otherwise false.
662 */
663 static bool migrate_caps_check(bool *cap_list,
664 MigrationCapabilityStatusList *params,
665 Error **errp)
666 {
667 MigrationCapabilityStatusList *cap;
668 bool old_postcopy_cap;
669 MigrationIncomingState *mis = migration_incoming_get_current();
670
671 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
672
673 for (cap = params; cap; cap = cap->next) {
674 cap_list[cap->value->capability] = cap->value->state;
675 }
676
677 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
678 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
679 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
680 "block migration");
681 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
682 return false;
683 }
684 #endif
685
686 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
687 if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
688 /* The decompression threads asynchronously write into RAM
689 * rather than use the atomic copies needed to avoid
690 * userfaulting. It should be possible to fix the decompression
691 * threads for compatibility in future.
692 */
693 error_setg(errp, "Postcopy is not currently compatible "
694 "with compression");
695 return false;
696 }
697
698 /* This check is reasonably expensive, so only when it's being
699 * set the first time, also it's only the destination that needs
700 * special support.
701 */
702 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
703 !postcopy_ram_supported_by_host(mis)) {
704 /* postcopy_ram_supported_by_host will have emitted a more
705 * detailed message
706 */
707 error_setg(errp, "Postcopy is not supported");
708 return false;
709 }
710 }
711
712 return true;
713 }
714
715 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
716 Error **errp)
717 {
718 MigrationState *s = migrate_get_current();
719 MigrationCapabilityStatusList *cap;
720
721 if (migration_is_setup_or_active(s->state)) {
722 error_setg(errp, QERR_MIGRATION_ACTIVE);
723 return;
724 }
725
726 if (!migrate_caps_check(s->enabled_capabilities, params, errp)) {
727 return;
728 }
729
730 for (cap = params; cap; cap = cap->next) {
731 s->enabled_capabilities[cap->value->capability] = cap->value->state;
732 }
733 }
734
735 /*
736 * Check whether the parameters are valid. Error will be put into errp
737 * (if provided). Return true if valid, otherwise false.
738 */
739 static bool migrate_params_check(MigrationParameters *params, Error **errp)
740 {
741 if (params->has_compress_level &&
742 (params->compress_level < 0 || params->compress_level > 9)) {
743 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
744 "is invalid, it should be in the range of 0 to 9");
745 return false;
746 }
747
748 if (params->has_compress_threads &&
749 (params->compress_threads < 1 || params->compress_threads > 255)) {
750 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
751 "compress_threads",
752 "is invalid, it should be in the range of 1 to 255");
753 return false;
754 }
755
756 if (params->has_decompress_threads &&
757 (params->decompress_threads < 1 || params->decompress_threads > 255)) {
758 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
759 "decompress_threads",
760 "is invalid, it should be in the range of 1 to 255");
761 return false;
762 }
763
764 if (params->has_cpu_throttle_initial &&
765 (params->cpu_throttle_initial < 1 ||
766 params->cpu_throttle_initial > 99)) {
767 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
768 "cpu_throttle_initial",
769 "an integer in the range of 1 to 99");
770 return false;
771 }
772
773 if (params->has_cpu_throttle_increment &&
774 (params->cpu_throttle_increment < 1 ||
775 params->cpu_throttle_increment > 99)) {
776 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
777 "cpu_throttle_increment",
778 "an integer in the range of 1 to 99");
779 return false;
780 }
781
782 if (params->has_max_bandwidth &&
783 (params->max_bandwidth < 0 || params->max_bandwidth > SIZE_MAX)) {
784 error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
785 " range of 0 to %zu bytes/second", SIZE_MAX);
786 return false;
787 }
788
789 if (params->has_downtime_limit &&
790 (params->downtime_limit < 0 ||
791 params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
792 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
793 "the range of 0 to %d milliseconds",
794 MAX_MIGRATE_DOWNTIME);
795 return false;
796 }
797
798 if (params->has_x_checkpoint_delay && (params->x_checkpoint_delay < 0)) {
799 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
800 "x_checkpoint_delay",
801 "is invalid, it should be positive");
802 return false;
803 }
804 if (params->has_x_multifd_channels &&
805 (params->x_multifd_channels < 1 || params->x_multifd_channels > 255)) {
806 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
807 "multifd_channels",
808 "is invalid, it should be in the range of 1 to 255");
809 return false;
810 }
811 if (params->has_x_multifd_page_count &&
812 (params->x_multifd_page_count < 1 ||
813 params->x_multifd_page_count > 10000)) {
814 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
815 "multifd_page_count",
816 "is invalid, it should be in the range of 1 to 10000");
817 return false;
818 }
819
820 return true;
821 }
822
823 static void migrate_params_test_apply(MigrateSetParameters *params,
824 MigrationParameters *dest)
825 {
826 *dest = migrate_get_current()->parameters;
827
828 /* TODO use QAPI_CLONE() instead of duplicating it inline */
829
830 if (params->has_compress_level) {
831 dest->compress_level = params->compress_level;
832 }
833
834 if (params->has_compress_threads) {
835 dest->compress_threads = params->compress_threads;
836 }
837
838 if (params->has_decompress_threads) {
839 dest->decompress_threads = params->decompress_threads;
840 }
841
842 if (params->has_cpu_throttle_initial) {
843 dest->cpu_throttle_initial = params->cpu_throttle_initial;
844 }
845
846 if (params->has_cpu_throttle_increment) {
847 dest->cpu_throttle_increment = params->cpu_throttle_increment;
848 }
849
850 if (params->has_tls_creds) {
851 assert(params->tls_creds->type == QTYPE_QSTRING);
852 dest->tls_creds = g_strdup(params->tls_creds->u.s);
853 }
854
855 if (params->has_tls_hostname) {
856 assert(params->tls_hostname->type == QTYPE_QSTRING);
857 dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
858 }
859
860 if (params->has_max_bandwidth) {
861 dest->max_bandwidth = params->max_bandwidth;
862 }
863
864 if (params->has_downtime_limit) {
865 dest->downtime_limit = params->downtime_limit;
866 }
867
868 if (params->has_x_checkpoint_delay) {
869 dest->x_checkpoint_delay = params->x_checkpoint_delay;
870 }
871
872 if (params->has_block_incremental) {
873 dest->block_incremental = params->block_incremental;
874 }
875 if (params->has_x_multifd_channels) {
876 dest->x_multifd_channels = params->x_multifd_channels;
877 }
878 if (params->has_x_multifd_page_count) {
879 dest->x_multifd_page_count = params->x_multifd_page_count;
880 }
881 }
882
883 static void migrate_params_apply(MigrateSetParameters *params)
884 {
885 MigrationState *s = migrate_get_current();
886
887 /* TODO use QAPI_CLONE() instead of duplicating it inline */
888
889 if (params->has_compress_level) {
890 s->parameters.compress_level = params->compress_level;
891 }
892
893 if (params->has_compress_threads) {
894 s->parameters.compress_threads = params->compress_threads;
895 }
896
897 if (params->has_decompress_threads) {
898 s->parameters.decompress_threads = params->decompress_threads;
899 }
900
901 if (params->has_cpu_throttle_initial) {
902 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
903 }
904
905 if (params->has_cpu_throttle_increment) {
906 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
907 }
908
909 if (params->has_tls_creds) {
910 g_free(s->parameters.tls_creds);
911 assert(params->tls_creds->type == QTYPE_QSTRING);
912 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
913 }
914
915 if (params->has_tls_hostname) {
916 g_free(s->parameters.tls_hostname);
917 assert(params->tls_hostname->type == QTYPE_QSTRING);
918 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
919 }
920
921 if (params->has_max_bandwidth) {
922 s->parameters.max_bandwidth = params->max_bandwidth;
923 if (s->to_dst_file) {
924 qemu_file_set_rate_limit(s->to_dst_file,
925 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
926 }
927 }
928
929 if (params->has_downtime_limit) {
930 s->parameters.downtime_limit = params->downtime_limit;
931 }
932
933 if (params->has_x_checkpoint_delay) {
934 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
935 if (migration_in_colo_state()) {
936 colo_checkpoint_notify(s);
937 }
938 }
939
940 if (params->has_block_incremental) {
941 s->parameters.block_incremental = params->block_incremental;
942 }
943 if (params->has_x_multifd_channels) {
944 s->parameters.x_multifd_channels = params->x_multifd_channels;
945 }
946 if (params->has_x_multifd_page_count) {
947 s->parameters.x_multifd_page_count = params->x_multifd_page_count;
948 }
949 }
950
951 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
952 {
953 MigrationParameters tmp;
954
955 /* TODO Rewrite "" to null instead */
956 if (params->has_tls_creds
957 && params->tls_creds->type == QTYPE_QNULL) {
958 QDECREF(params->tls_creds->u.n);
959 params->tls_creds->type = QTYPE_QSTRING;
960 params->tls_creds->u.s = strdup("");
961 }
962 /* TODO Rewrite "" to null instead */
963 if (params->has_tls_hostname
964 && params->tls_hostname->type == QTYPE_QNULL) {
965 QDECREF(params->tls_hostname->u.n);
966 params->tls_hostname->type = QTYPE_QSTRING;
967 params->tls_hostname->u.s = strdup("");
968 }
969
970 migrate_params_test_apply(params, &tmp);
971
972 if (!migrate_params_check(&tmp, errp)) {
973 /* Invalid parameter */
974 return;
975 }
976
977 migrate_params_apply(params);
978 }
979
980
981 void qmp_migrate_start_postcopy(Error **errp)
982 {
983 MigrationState *s = migrate_get_current();
984
985 if (!migrate_postcopy_ram()) {
986 error_setg(errp, "Enable postcopy with migrate_set_capability before"
987 " the start of migration");
988 return;
989 }
990
991 if (s->state == MIGRATION_STATUS_NONE) {
992 error_setg(errp, "Postcopy must be started after migration has been"
993 " started");
994 return;
995 }
996 /*
997 * we don't error if migration has finished since that would be racy
998 * with issuing this command.
999 */
1000 atomic_set(&s->start_postcopy, true);
1001 }
1002
1003 /* shared migration helpers */
1004
1005 void migrate_set_state(int *state, int old_state, int new_state)
1006 {
1007 assert(new_state < MIGRATION_STATUS__MAX);
1008 if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
1009 trace_migrate_set_state(MigrationStatus_str(new_state));
1010 migrate_generate_event(new_state);
1011 }
1012 }
1013
1014 static MigrationCapabilityStatusList *migrate_cap_add(
1015 MigrationCapabilityStatusList *list,
1016 MigrationCapability index,
1017 bool state)
1018 {
1019 MigrationCapabilityStatusList *cap;
1020
1021 cap = g_new0(MigrationCapabilityStatusList, 1);
1022 cap->value = g_new0(MigrationCapabilityStatus, 1);
1023 cap->value->capability = index;
1024 cap->value->state = state;
1025 cap->next = list;
1026
1027 return cap;
1028 }
1029
1030 void migrate_set_block_enabled(bool value, Error **errp)
1031 {
1032 MigrationCapabilityStatusList *cap;
1033
1034 cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
1035 qmp_migrate_set_capabilities(cap, errp);
1036 qapi_free_MigrationCapabilityStatusList(cap);
1037 }
1038
1039 static void migrate_set_block_incremental(MigrationState *s, bool value)
1040 {
1041 s->parameters.block_incremental = value;
1042 }
1043
1044 static void block_cleanup_parameters(MigrationState *s)
1045 {
1046 if (s->must_remove_block_options) {
1047 /* setting to false can never fail */
1048 migrate_set_block_enabled(false, &error_abort);
1049 migrate_set_block_incremental(s, false);
1050 s->must_remove_block_options = false;
1051 }
1052 }
1053
1054 static void migrate_fd_cleanup(void *opaque)
1055 {
1056 MigrationState *s = opaque;
1057
1058 qemu_bh_delete(s->cleanup_bh);
1059 s->cleanup_bh = NULL;
1060
1061 if (s->to_dst_file) {
1062 Error *local_err = NULL;
1063
1064 trace_migrate_fd_cleanup();
1065 qemu_mutex_unlock_iothread();
1066 if (s->migration_thread_running) {
1067 qemu_thread_join(&s->thread);
1068 s->migration_thread_running = false;
1069 }
1070 qemu_mutex_lock_iothread();
1071
1072 if (multifd_save_cleanup(&local_err) != 0) {
1073 error_report_err(local_err);
1074 }
1075 qemu_fclose(s->to_dst_file);
1076 s->to_dst_file = NULL;
1077 }
1078
1079 assert((s->state != MIGRATION_STATUS_ACTIVE) &&
1080 (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
1081
1082 if (s->state == MIGRATION_STATUS_CANCELLING) {
1083 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1084 MIGRATION_STATUS_CANCELLED);
1085 }
1086
1087 notifier_list_notify(&migration_state_notifiers, s);
1088 block_cleanup_parameters(s);
1089 }
1090
1091 void migrate_fd_error(MigrationState *s, const Error *error)
1092 {
1093 trace_migrate_fd_error(error_get_pretty(error));
1094 assert(s->to_dst_file == NULL);
1095 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1096 MIGRATION_STATUS_FAILED);
1097 if (!s->error) {
1098 s->error = error_copy(error);
1099 }
1100 notifier_list_notify(&migration_state_notifiers, s);
1101 block_cleanup_parameters(s);
1102 }
1103
1104 static void migrate_fd_cancel(MigrationState *s)
1105 {
1106 int old_state ;
1107 QEMUFile *f = migrate_get_current()->to_dst_file;
1108 trace_migrate_fd_cancel();
1109
1110 if (s->rp_state.from_dst_file) {
1111 /* shutdown the rp socket, so causing the rp thread to shutdown */
1112 qemu_file_shutdown(s->rp_state.from_dst_file);
1113 }
1114
1115 do {
1116 old_state = s->state;
1117 if (!migration_is_setup_or_active(old_state)) {
1118 break;
1119 }
1120 /* If the migration is paused, kick it out of the pause */
1121 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1122 qemu_sem_post(&s->pause_sem);
1123 }
1124 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1125 } while (s->state != MIGRATION_STATUS_CANCELLING);
1126
1127 /*
1128 * If we're unlucky the migration code might be stuck somewhere in a
1129 * send/write while the network has failed and is waiting to timeout;
1130 * if we've got shutdown(2) available then we can force it to quit.
1131 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1132 * called in a bh, so there is no race against this cancel.
1133 */
1134 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1135 qemu_file_shutdown(f);
1136 }
1137 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1138 Error *local_err = NULL;
1139
1140 bdrv_invalidate_cache_all(&local_err);
1141 if (local_err) {
1142 error_report_err(local_err);
1143 } else {
1144 s->block_inactive = false;
1145 }
1146 }
1147 block_cleanup_parameters(s);
1148 }
1149
1150 void add_migration_state_change_notifier(Notifier *notify)
1151 {
1152 notifier_list_add(&migration_state_notifiers, notify);
1153 }
1154
1155 void remove_migration_state_change_notifier(Notifier *notify)
1156 {
1157 notifier_remove(notify);
1158 }
1159
1160 bool migration_in_setup(MigrationState *s)
1161 {
1162 return s->state == MIGRATION_STATUS_SETUP;
1163 }
1164
1165 bool migration_has_finished(MigrationState *s)
1166 {
1167 return s->state == MIGRATION_STATUS_COMPLETED;
1168 }
1169
1170 bool migration_has_failed(MigrationState *s)
1171 {
1172 return (s->state == MIGRATION_STATUS_CANCELLED ||
1173 s->state == MIGRATION_STATUS_FAILED);
1174 }
1175
1176 bool migration_in_postcopy(void)
1177 {
1178 MigrationState *s = migrate_get_current();
1179
1180 return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1181 }
1182
1183 bool migration_in_postcopy_after_devices(MigrationState *s)
1184 {
1185 return migration_in_postcopy() && s->postcopy_after_devices;
1186 }
1187
1188 bool migration_is_idle(void)
1189 {
1190 MigrationState *s = migrate_get_current();
1191
1192 switch (s->state) {
1193 case MIGRATION_STATUS_NONE:
1194 case MIGRATION_STATUS_CANCELLED:
1195 case MIGRATION_STATUS_COMPLETED:
1196 case MIGRATION_STATUS_FAILED:
1197 return true;
1198 case MIGRATION_STATUS_SETUP:
1199 case MIGRATION_STATUS_CANCELLING:
1200 case MIGRATION_STATUS_ACTIVE:
1201 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1202 case MIGRATION_STATUS_COLO:
1203 case MIGRATION_STATUS_PRE_SWITCHOVER:
1204 case MIGRATION_STATUS_DEVICE:
1205 return false;
1206 case MIGRATION_STATUS__MAX:
1207 g_assert_not_reached();
1208 }
1209
1210 return false;
1211 }
1212
1213 MigrationState *migrate_init(void)
1214 {
1215 MigrationState *s = migrate_get_current();
1216
1217 /*
1218 * Reinitialise all migration state, except
1219 * parameters/capabilities that the user set, and
1220 * locks.
1221 */
1222 s->bytes_xfer = 0;
1223 s->xfer_limit = 0;
1224 s->cleanup_bh = 0;
1225 s->to_dst_file = NULL;
1226 s->state = MIGRATION_STATUS_NONE;
1227 s->rp_state.from_dst_file = NULL;
1228 s->rp_state.error = false;
1229 s->mbps = 0.0;
1230 s->downtime = 0;
1231 s->expected_downtime = 0;
1232 s->setup_time = 0;
1233 s->start_postcopy = false;
1234 s->postcopy_after_devices = false;
1235 s->migration_thread_running = false;
1236 error_free(s->error);
1237 s->error = NULL;
1238
1239 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1240
1241 s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1242 return s;
1243 }
1244
1245 static GSList *migration_blockers;
1246
1247 int migrate_add_blocker(Error *reason, Error **errp)
1248 {
1249 if (migrate_get_current()->only_migratable) {
1250 error_propagate(errp, error_copy(reason));
1251 error_prepend(errp, "disallowing migration blocker "
1252 "(--only_migratable) for: ");
1253 return -EACCES;
1254 }
1255
1256 if (migration_is_idle()) {
1257 migration_blockers = g_slist_prepend(migration_blockers, reason);
1258 return 0;
1259 }
1260
1261 error_propagate(errp, error_copy(reason));
1262 error_prepend(errp, "disallowing migration blocker (migration in "
1263 "progress) for: ");
1264 return -EBUSY;
1265 }
1266
1267 void migrate_del_blocker(Error *reason)
1268 {
1269 migration_blockers = g_slist_remove(migration_blockers, reason);
1270 }
1271
1272 void qmp_migrate_incoming(const char *uri, Error **errp)
1273 {
1274 Error *local_err = NULL;
1275 static bool once = true;
1276
1277 if (!deferred_incoming) {
1278 error_setg(errp, "For use with '-incoming defer'");
1279 return;
1280 }
1281 if (!once) {
1282 error_setg(errp, "The incoming migration has already been started");
1283 }
1284
1285 qemu_start_incoming_migration(uri, &local_err);
1286
1287 if (local_err) {
1288 error_propagate(errp, local_err);
1289 return;
1290 }
1291
1292 once = false;
1293 }
1294
1295 bool migration_is_blocked(Error **errp)
1296 {
1297 if (qemu_savevm_state_blocked(errp)) {
1298 return true;
1299 }
1300
1301 if (migration_blockers) {
1302 error_propagate(errp, error_copy(migration_blockers->data));
1303 return true;
1304 }
1305
1306 return false;
1307 }
1308
1309 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1310 bool has_inc, bool inc, bool has_detach, bool detach,
1311 Error **errp)
1312 {
1313 Error *local_err = NULL;
1314 MigrationState *s = migrate_get_current();
1315 const char *p;
1316
1317 if (migration_is_setup_or_active(s->state) ||
1318 s->state == MIGRATION_STATUS_CANCELLING ||
1319 s->state == MIGRATION_STATUS_COLO) {
1320 error_setg(errp, QERR_MIGRATION_ACTIVE);
1321 return;
1322 }
1323 if (runstate_check(RUN_STATE_INMIGRATE)) {
1324 error_setg(errp, "Guest is waiting for an incoming migration");
1325 return;
1326 }
1327
1328 if (migration_is_blocked(errp)) {
1329 return;
1330 }
1331
1332 if ((has_blk && blk) || (has_inc && inc)) {
1333 if (migrate_use_block() || migrate_use_block_incremental()) {
1334 error_setg(errp, "Command options are incompatible with "
1335 "current migration capabilities");
1336 return;
1337 }
1338 migrate_set_block_enabled(true, &local_err);
1339 if (local_err) {
1340 error_propagate(errp, local_err);
1341 return;
1342 }
1343 s->must_remove_block_options = true;
1344 }
1345
1346 if (has_inc && inc) {
1347 migrate_set_block_incremental(s, true);
1348 }
1349
1350 s = migrate_init();
1351
1352 if (strstart(uri, "tcp:", &p)) {
1353 tcp_start_outgoing_migration(s, p, &local_err);
1354 #ifdef CONFIG_RDMA
1355 } else if (strstart(uri, "rdma:", &p)) {
1356 rdma_start_outgoing_migration(s, p, &local_err);
1357 #endif
1358 } else if (strstart(uri, "exec:", &p)) {
1359 exec_start_outgoing_migration(s, p, &local_err);
1360 } else if (strstart(uri, "unix:", &p)) {
1361 unix_start_outgoing_migration(s, p, &local_err);
1362 } else if (strstart(uri, "fd:", &p)) {
1363 fd_start_outgoing_migration(s, p, &local_err);
1364 } else {
1365 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1366 "a valid migration protocol");
1367 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1368 MIGRATION_STATUS_FAILED);
1369 return;
1370 }
1371
1372 if (local_err) {
1373 migrate_fd_error(s, local_err);
1374 error_propagate(errp, local_err);
1375 return;
1376 }
1377 }
1378
1379 void qmp_migrate_cancel(Error **errp)
1380 {
1381 migrate_fd_cancel(migrate_get_current());
1382 }
1383
1384 void qmp_migrate_continue(MigrationStatus state, Error **errp)
1385 {
1386 MigrationState *s = migrate_get_current();
1387 if (s->state != state) {
1388 error_setg(errp, "Migration not in expected state: %s",
1389 MigrationStatus_str(s->state));
1390 return;
1391 }
1392 qemu_sem_post(&s->pause_sem);
1393 }
1394
1395 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1396 {
1397 MigrationState *s = migrate_get_current();
1398 int64_t new_size;
1399
1400 new_size = xbzrle_cache_resize(value, errp);
1401 if (new_size < 0) {
1402 return;
1403 }
1404
1405 s->xbzrle_cache_size = new_size;
1406 }
1407
1408 int64_t qmp_query_migrate_cache_size(Error **errp)
1409 {
1410 return migrate_xbzrle_cache_size();
1411 }
1412
1413 void qmp_migrate_set_speed(int64_t value, Error **errp)
1414 {
1415 MigrateSetParameters p = {
1416 .has_max_bandwidth = true,
1417 .max_bandwidth = value,
1418 };
1419
1420 qmp_migrate_set_parameters(&p, errp);
1421 }
1422
1423 void qmp_migrate_set_downtime(double value, Error **errp)
1424 {
1425 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1426 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1427 "the range of 0 to %d seconds",
1428 MAX_MIGRATE_DOWNTIME_SECONDS);
1429 return;
1430 }
1431
1432 value *= 1000; /* Convert to milliseconds */
1433 value = MAX(0, MIN(INT64_MAX, value));
1434
1435 MigrateSetParameters p = {
1436 .has_downtime_limit = true,
1437 .downtime_limit = value,
1438 };
1439
1440 qmp_migrate_set_parameters(&p, errp);
1441 }
1442
1443 bool migrate_release_ram(void)
1444 {
1445 MigrationState *s;
1446
1447 s = migrate_get_current();
1448
1449 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1450 }
1451
1452 bool migrate_postcopy_ram(void)
1453 {
1454 MigrationState *s;
1455
1456 s = migrate_get_current();
1457
1458 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1459 }
1460
1461 bool migrate_postcopy(void)
1462 {
1463 return migrate_postcopy_ram();
1464 }
1465
1466 bool migrate_auto_converge(void)
1467 {
1468 MigrationState *s;
1469
1470 s = migrate_get_current();
1471
1472 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1473 }
1474
1475 bool migrate_zero_blocks(void)
1476 {
1477 MigrationState *s;
1478
1479 s = migrate_get_current();
1480
1481 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1482 }
1483
1484 bool migrate_use_compression(void)
1485 {
1486 MigrationState *s;
1487
1488 s = migrate_get_current();
1489
1490 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
1491 }
1492
1493 int migrate_compress_level(void)
1494 {
1495 MigrationState *s;
1496
1497 s = migrate_get_current();
1498
1499 return s->parameters.compress_level;
1500 }
1501
1502 int migrate_compress_threads(void)
1503 {
1504 MigrationState *s;
1505
1506 s = migrate_get_current();
1507
1508 return s->parameters.compress_threads;
1509 }
1510
1511 int migrate_decompress_threads(void)
1512 {
1513 MigrationState *s;
1514
1515 s = migrate_get_current();
1516
1517 return s->parameters.decompress_threads;
1518 }
1519
1520 bool migrate_use_events(void)
1521 {
1522 MigrationState *s;
1523
1524 s = migrate_get_current();
1525
1526 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1527 }
1528
1529 bool migrate_use_multifd(void)
1530 {
1531 MigrationState *s;
1532
1533 s = migrate_get_current();
1534
1535 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_MULTIFD];
1536 }
1537
1538 bool migrate_pause_before_switchover(void)
1539 {
1540 MigrationState *s;
1541
1542 s = migrate_get_current();
1543
1544 return s->enabled_capabilities[
1545 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
1546 }
1547
1548 int migrate_multifd_channels(void)
1549 {
1550 MigrationState *s;
1551
1552 s = migrate_get_current();
1553
1554 return s->parameters.x_multifd_channels;
1555 }
1556
1557 int migrate_multifd_page_count(void)
1558 {
1559 MigrationState *s;
1560
1561 s = migrate_get_current();
1562
1563 return s->parameters.x_multifd_page_count;
1564 }
1565
1566 int migrate_use_xbzrle(void)
1567 {
1568 MigrationState *s;
1569
1570 s = migrate_get_current();
1571
1572 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1573 }
1574
1575 int64_t migrate_xbzrle_cache_size(void)
1576 {
1577 MigrationState *s;
1578
1579 s = migrate_get_current();
1580
1581 return s->xbzrle_cache_size;
1582 }
1583
1584 bool migrate_use_block(void)
1585 {
1586 MigrationState *s;
1587
1588 s = migrate_get_current();
1589
1590 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
1591 }
1592
1593 bool migrate_use_return_path(void)
1594 {
1595 MigrationState *s;
1596
1597 s = migrate_get_current();
1598
1599 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
1600 }
1601
1602 bool migrate_use_block_incremental(void)
1603 {
1604 MigrationState *s;
1605
1606 s = migrate_get_current();
1607
1608 return s->parameters.block_incremental;
1609 }
1610
1611 /* migration thread support */
1612 /*
1613 * Something bad happened to the RP stream, mark an error
1614 * The caller shall print or trace something to indicate why
1615 */
1616 static void mark_source_rp_bad(MigrationState *s)
1617 {
1618 s->rp_state.error = true;
1619 }
1620
1621 static struct rp_cmd_args {
1622 ssize_t len; /* -1 = variable */
1623 const char *name;
1624 } rp_cmd_args[] = {
1625 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
1626 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
1627 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
1628 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
1629 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
1630 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
1631 };
1632
1633 /*
1634 * Process a request for pages received on the return path,
1635 * We're allowed to send more than requested (e.g. to round to our page size)
1636 * and we don't need to send pages that have already been sent.
1637 */
1638 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1639 ram_addr_t start, size_t len)
1640 {
1641 long our_host_ps = getpagesize();
1642
1643 trace_migrate_handle_rp_req_pages(rbname, start, len);
1644
1645 /*
1646 * Since we currently insist on matching page sizes, just sanity check
1647 * we're being asked for whole host pages.
1648 */
1649 if (start & (our_host_ps-1) ||
1650 (len & (our_host_ps-1))) {
1651 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1652 " len: %zd", __func__, start, len);
1653 mark_source_rp_bad(ms);
1654 return;
1655 }
1656
1657 if (ram_save_queue_pages(rbname, start, len)) {
1658 mark_source_rp_bad(ms);
1659 }
1660 }
1661
1662 /*
1663 * Handles messages sent on the return path towards the source VM
1664 *
1665 */
1666 static void *source_return_path_thread(void *opaque)
1667 {
1668 MigrationState *ms = opaque;
1669 QEMUFile *rp = ms->rp_state.from_dst_file;
1670 uint16_t header_len, header_type;
1671 uint8_t buf[512];
1672 uint32_t tmp32, sibling_error;
1673 ram_addr_t start = 0; /* =0 to silence warning */
1674 size_t len = 0, expected_len;
1675 int res;
1676
1677 trace_source_return_path_thread_entry();
1678 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
1679 migration_is_setup_or_active(ms->state)) {
1680 trace_source_return_path_thread_loop_top();
1681 header_type = qemu_get_be16(rp);
1682 header_len = qemu_get_be16(rp);
1683
1684 if (header_type >= MIG_RP_MSG_MAX ||
1685 header_type == MIG_RP_MSG_INVALID) {
1686 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1687 header_type, header_len);
1688 mark_source_rp_bad(ms);
1689 goto out;
1690 }
1691
1692 if ((rp_cmd_args[header_type].len != -1 &&
1693 header_len != rp_cmd_args[header_type].len) ||
1694 header_len > sizeof(buf)) {
1695 error_report("RP: Received '%s' message (0x%04x) with"
1696 "incorrect length %d expecting %zu",
1697 rp_cmd_args[header_type].name, header_type, header_len,
1698 (size_t)rp_cmd_args[header_type].len);
1699 mark_source_rp_bad(ms);
1700 goto out;
1701 }
1702
1703 /* We know we've got a valid header by this point */
1704 res = qemu_get_buffer(rp, buf, header_len);
1705 if (res != header_len) {
1706 error_report("RP: Failed reading data for message 0x%04x"
1707 " read %d expected %d",
1708 header_type, res, header_len);
1709 mark_source_rp_bad(ms);
1710 goto out;
1711 }
1712
1713 /* OK, we have the message and the data */
1714 switch (header_type) {
1715 case MIG_RP_MSG_SHUT:
1716 sibling_error = ldl_be_p(buf);
1717 trace_source_return_path_thread_shut(sibling_error);
1718 if (sibling_error) {
1719 error_report("RP: Sibling indicated error %d", sibling_error);
1720 mark_source_rp_bad(ms);
1721 }
1722 /*
1723 * We'll let the main thread deal with closing the RP
1724 * we could do a shutdown(2) on it, but we're the only user
1725 * anyway, so there's nothing gained.
1726 */
1727 goto out;
1728
1729 case MIG_RP_MSG_PONG:
1730 tmp32 = ldl_be_p(buf);
1731 trace_source_return_path_thread_pong(tmp32);
1732 break;
1733
1734 case MIG_RP_MSG_REQ_PAGES:
1735 start = ldq_be_p(buf);
1736 len = ldl_be_p(buf + 8);
1737 migrate_handle_rp_req_pages(ms, NULL, start, len);
1738 break;
1739
1740 case MIG_RP_MSG_REQ_PAGES_ID:
1741 expected_len = 12 + 1; /* header + termination */
1742
1743 if (header_len >= expected_len) {
1744 start = ldq_be_p(buf);
1745 len = ldl_be_p(buf + 8);
1746 /* Now we expect an idstr */
1747 tmp32 = buf[12]; /* Length of the following idstr */
1748 buf[13 + tmp32] = '\0';
1749 expected_len += tmp32;
1750 }
1751 if (header_len != expected_len) {
1752 error_report("RP: Req_Page_id with length %d expecting %zd",
1753 header_len, expected_len);
1754 mark_source_rp_bad(ms);
1755 goto out;
1756 }
1757 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
1758 break;
1759
1760 default:
1761 break;
1762 }
1763 }
1764 if (qemu_file_get_error(rp)) {
1765 trace_source_return_path_thread_bad_end();
1766 mark_source_rp_bad(ms);
1767 }
1768
1769 trace_source_return_path_thread_end();
1770 out:
1771 ms->rp_state.from_dst_file = NULL;
1772 qemu_fclose(rp);
1773 return NULL;
1774 }
1775
1776 static int open_return_path_on_source(MigrationState *ms)
1777 {
1778
1779 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
1780 if (!ms->rp_state.from_dst_file) {
1781 return -1;
1782 }
1783
1784 trace_open_return_path_on_source();
1785 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
1786 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
1787
1788 trace_open_return_path_on_source_continue();
1789
1790 return 0;
1791 }
1792
1793 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1794 static int await_return_path_close_on_source(MigrationState *ms)
1795 {
1796 /*
1797 * If this is a normal exit then the destination will send a SHUT and the
1798 * rp_thread will exit, however if there's an error we need to cause
1799 * it to exit.
1800 */
1801 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
1802 /*
1803 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1804 * waiting for the destination.
1805 */
1806 qemu_file_shutdown(ms->rp_state.from_dst_file);
1807 mark_source_rp_bad(ms);
1808 }
1809 trace_await_return_path_close_on_source_joining();
1810 qemu_thread_join(&ms->rp_state.rp_thread);
1811 trace_await_return_path_close_on_source_close();
1812 return ms->rp_state.error;
1813 }
1814
1815 /*
1816 * Switch from normal iteration to postcopy
1817 * Returns non-0 on error
1818 */
1819 static int postcopy_start(MigrationState *ms, bool *old_vm_running)
1820 {
1821 int ret;
1822 QIOChannelBuffer *bioc;
1823 QEMUFile *fb;
1824 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1825 bool restart_block = false;
1826 int cur_state = MIGRATION_STATUS_ACTIVE;
1827 if (!migrate_pause_before_switchover()) {
1828 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
1829 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1830 }
1831
1832 trace_postcopy_start();
1833 qemu_mutex_lock_iothread();
1834 trace_postcopy_start_set_run();
1835
1836 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1837 *old_vm_running = runstate_is_running();
1838 global_state_store();
1839 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1840 if (ret < 0) {
1841 goto fail;
1842 }
1843
1844 ret = migration_maybe_pause(ms, &cur_state,
1845 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1846 if (ret < 0) {
1847 goto fail;
1848 }
1849
1850 ret = bdrv_inactivate_all();
1851 if (ret < 0) {
1852 goto fail;
1853 }
1854 restart_block = true;
1855
1856 /*
1857 * Cause any non-postcopiable, but iterative devices to
1858 * send out their final data.
1859 */
1860 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
1861
1862 /*
1863 * in Finish migrate and with the io-lock held everything should
1864 * be quiet, but we've potentially still got dirty pages and we
1865 * need to tell the destination to throw any pages it's already received
1866 * that are dirty
1867 */
1868 if (migrate_postcopy_ram()) {
1869 if (ram_postcopy_send_discard_bitmap(ms)) {
1870 error_report("postcopy send discard bitmap failed");
1871 goto fail;
1872 }
1873 }
1874
1875 /*
1876 * send rest of state - note things that are doing postcopy
1877 * will notice we're in POSTCOPY_ACTIVE and not actually
1878 * wrap their state up here
1879 */
1880 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
1881 if (migrate_postcopy_ram()) {
1882 /* Ping just for debugging, helps line traces up */
1883 qemu_savevm_send_ping(ms->to_dst_file, 2);
1884 }
1885
1886 /*
1887 * While loading the device state we may trigger page transfer
1888 * requests and the fd must be free to process those, and thus
1889 * the destination must read the whole device state off the fd before
1890 * it starts processing it. Unfortunately the ad-hoc migration format
1891 * doesn't allow the destination to know the size to read without fully
1892 * parsing it through each devices load-state code (especially the open
1893 * coded devices that use get/put).
1894 * So we wrap the device state up in a package with a length at the start;
1895 * to do this we use a qemu_buf to hold the whole of the device state.
1896 */
1897 bioc = qio_channel_buffer_new(4096);
1898 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
1899 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
1900 object_unref(OBJECT(bioc));
1901
1902 /*
1903 * Make sure the receiver can get incoming pages before we send the rest
1904 * of the state
1905 */
1906 qemu_savevm_send_postcopy_listen(fb);
1907
1908 qemu_savevm_state_complete_precopy(fb, false, false);
1909 if (migrate_postcopy_ram()) {
1910 qemu_savevm_send_ping(fb, 3);
1911 }
1912
1913 qemu_savevm_send_postcopy_run(fb);
1914
1915 /* <><> end of stuff going into the package */
1916
1917 /* Last point of recovery; as soon as we send the package the destination
1918 * can open devices and potentially start running.
1919 * Lets just check again we've not got any errors.
1920 */
1921 ret = qemu_file_get_error(ms->to_dst_file);
1922 if (ret) {
1923 error_report("postcopy_start: Migration stream errored (pre package)");
1924 goto fail_closefb;
1925 }
1926
1927 restart_block = false;
1928
1929 /* Now send that blob */
1930 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
1931 goto fail_closefb;
1932 }
1933 qemu_fclose(fb);
1934
1935 /* Send a notify to give a chance for anything that needs to happen
1936 * at the transition to postcopy and after the device state; in particular
1937 * spice needs to trigger a transition now
1938 */
1939 ms->postcopy_after_devices = true;
1940 notifier_list_notify(&migration_state_notifiers, ms);
1941
1942 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
1943
1944 qemu_mutex_unlock_iothread();
1945
1946 if (migrate_postcopy_ram()) {
1947 /*
1948 * Although this ping is just for debug, it could potentially be
1949 * used for getting a better measurement of downtime at the source.
1950 */
1951 qemu_savevm_send_ping(ms->to_dst_file, 4);
1952 }
1953
1954 if (migrate_release_ram()) {
1955 ram_postcopy_migrated_memory_release(ms);
1956 }
1957
1958 ret = qemu_file_get_error(ms->to_dst_file);
1959 if (ret) {
1960 error_report("postcopy_start: Migration stream errored");
1961 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1962 MIGRATION_STATUS_FAILED);
1963 }
1964
1965 return ret;
1966
1967 fail_closefb:
1968 qemu_fclose(fb);
1969 fail:
1970 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1971 MIGRATION_STATUS_FAILED);
1972 if (restart_block) {
1973 /* A failure happened early enough that we know the destination hasn't
1974 * accessed block devices, so we're safe to recover.
1975 */
1976 Error *local_err = NULL;
1977
1978 bdrv_invalidate_cache_all(&local_err);
1979 if (local_err) {
1980 error_report_err(local_err);
1981 }
1982 }
1983 qemu_mutex_unlock_iothread();
1984 return -1;
1985 }
1986
1987 /**
1988 * migration_maybe_pause: Pause if required to by
1989 * migrate_pause_before_switchover called with the iothread locked
1990 * Returns: 0 on success
1991 */
1992 static int migration_maybe_pause(MigrationState *s,
1993 int *current_active_state,
1994 int new_state)
1995 {
1996 if (!migrate_pause_before_switchover()) {
1997 return 0;
1998 }
1999
2000 /* Since leaving this state is not atomic with posting the semaphore
2001 * it's possible that someone could have issued multiple migrate_continue
2002 * and the semaphore is incorrectly positive at this point;
2003 * the docs say it's undefined to reinit a semaphore that's already
2004 * init'd, so use timedwait to eat up any existing posts.
2005 */
2006 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2007 /* This block intentionally left blank */
2008 }
2009
2010 qemu_mutex_unlock_iothread();
2011 migrate_set_state(&s->state, *current_active_state,
2012 MIGRATION_STATUS_PRE_SWITCHOVER);
2013 qemu_sem_wait(&s->pause_sem);
2014 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
2015 new_state);
2016 *current_active_state = new_state;
2017 qemu_mutex_lock_iothread();
2018
2019 return s->state == new_state ? 0 : -EINVAL;
2020 }
2021
2022 /**
2023 * migration_completion: Used by migration_thread when there's not much left.
2024 * The caller 'breaks' the loop when this returns.
2025 *
2026 * @s: Current migration state
2027 * @current_active_state: The migration state we expect to be in
2028 * @*old_vm_running: Pointer to old_vm_running flag
2029 * @*start_time: Pointer to time to update
2030 */
2031 static void migration_completion(MigrationState *s, int current_active_state,
2032 bool *old_vm_running,
2033 int64_t *start_time)
2034 {
2035 int ret;
2036
2037 if (s->state == MIGRATION_STATUS_ACTIVE) {
2038 qemu_mutex_lock_iothread();
2039 *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2040 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
2041 *old_vm_running = runstate_is_running();
2042 ret = global_state_store();
2043
2044 if (!ret) {
2045 bool inactivate = !migrate_colo_enabled();
2046 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2047 if (ret >= 0) {
2048 ret = migration_maybe_pause(s, &current_active_state,
2049 MIGRATION_STATUS_DEVICE);
2050 }
2051 if (ret >= 0) {
2052 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
2053 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2054 inactivate);
2055 }
2056 if (inactivate && ret >= 0) {
2057 s->block_inactive = true;
2058 }
2059 }
2060 qemu_mutex_unlock_iothread();
2061
2062 if (ret < 0) {
2063 goto fail;
2064 }
2065 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2066 trace_migration_completion_postcopy_end();
2067
2068 qemu_savevm_state_complete_postcopy(s->to_dst_file);
2069 trace_migration_completion_postcopy_end_after_complete();
2070 }
2071
2072 /*
2073 * If rp was opened we must clean up the thread before
2074 * cleaning everything else up (since if there are no failures
2075 * it will wait for the destination to send it's status in
2076 * a SHUT command).
2077 */
2078 if (s->rp_state.from_dst_file) {
2079 int rp_error;
2080 trace_migration_return_path_end_before();
2081 rp_error = await_return_path_close_on_source(s);
2082 trace_migration_return_path_end_after(rp_error);
2083 if (rp_error) {
2084 goto fail_invalidate;
2085 }
2086 }
2087
2088 if (qemu_file_get_error(s->to_dst_file)) {
2089 trace_migration_completion_file_err();
2090 goto fail_invalidate;
2091 }
2092
2093 if (!migrate_colo_enabled()) {
2094 migrate_set_state(&s->state, current_active_state,
2095 MIGRATION_STATUS_COMPLETED);
2096 }
2097
2098 return;
2099
2100 fail_invalidate:
2101 /* If not doing postcopy, vm_start() will be called: let's regain
2102 * control on images.
2103 */
2104 if (s->state == MIGRATION_STATUS_ACTIVE) {
2105 Error *local_err = NULL;
2106
2107 qemu_mutex_lock_iothread();
2108 bdrv_invalidate_cache_all(&local_err);
2109 if (local_err) {
2110 error_report_err(local_err);
2111 } else {
2112 s->block_inactive = false;
2113 }
2114 qemu_mutex_unlock_iothread();
2115 }
2116
2117 fail:
2118 migrate_set_state(&s->state, current_active_state,
2119 MIGRATION_STATUS_FAILED);
2120 }
2121
2122 bool migrate_colo_enabled(void)
2123 {
2124 MigrationState *s = migrate_get_current();
2125 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2126 }
2127
2128 /*
2129 * Master migration thread on the source VM.
2130 * It drives the migration and pumps the data down the outgoing channel.
2131 */
2132 static void *migration_thread(void *opaque)
2133 {
2134 MigrationState *s = opaque;
2135 /* Used by the bandwidth calcs, updated later */
2136 int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2137 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
2138 int64_t initial_bytes = 0;
2139 /*
2140 * The final stage happens when the remaining data is smaller than
2141 * this threshold; it's calculated from the requested downtime and
2142 * measured bandwidth
2143 */
2144 int64_t threshold_size = 0;
2145 int64_t start_time = initial_time;
2146 int64_t end_time;
2147 bool old_vm_running = false;
2148 bool entered_postcopy = false;
2149 /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
2150 enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE;
2151 bool enable_colo = migrate_colo_enabled();
2152
2153 rcu_register_thread();
2154
2155 qemu_savevm_state_header(s->to_dst_file);
2156
2157 /*
2158 * If we opened the return path, we need to make sure dst has it
2159 * opened as well.
2160 */
2161 if (s->rp_state.from_dst_file) {
2162 /* Now tell the dest that it should open its end so it can reply */
2163 qemu_savevm_send_open_return_path(s->to_dst_file);
2164
2165 /* And do a ping that will make stuff easier to debug */
2166 qemu_savevm_send_ping(s->to_dst_file, 1);
2167 }
2168
2169 if (migrate_postcopy()) {
2170 /*
2171 * Tell the destination that we *might* want to do postcopy later;
2172 * if the other end can't do postcopy it should fail now, nice and
2173 * early.
2174 */
2175 qemu_savevm_send_postcopy_advise(s->to_dst_file);
2176 }
2177
2178 qemu_savevm_state_setup(s->to_dst_file);
2179
2180 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
2181 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2182 MIGRATION_STATUS_ACTIVE);
2183
2184 trace_migration_thread_setup_complete();
2185
2186 while (s->state == MIGRATION_STATUS_ACTIVE ||
2187 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2188 int64_t current_time;
2189 uint64_t pending_size;
2190
2191 if (!qemu_file_rate_limit(s->to_dst_file)) {
2192 uint64_t pend_post, pend_nonpost;
2193
2194 qemu_savevm_state_pending(s->to_dst_file, threshold_size,
2195 &pend_nonpost, &pend_post);
2196 pending_size = pend_nonpost + pend_post;
2197 trace_migrate_pending(pending_size, threshold_size,
2198 pend_post, pend_nonpost);
2199 if (pending_size && pending_size >= threshold_size) {
2200 /* Still a significant amount to transfer */
2201
2202 if (migrate_postcopy() &&
2203 s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE &&
2204 pend_nonpost <= threshold_size &&
2205 atomic_read(&s->start_postcopy)) {
2206
2207 if (!postcopy_start(s, &old_vm_running)) {
2208 current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE;
2209 entered_postcopy = true;
2210 }
2211
2212 continue;
2213 }
2214 /* Just another iteration step */
2215 qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy);
2216 } else {
2217 trace_migration_thread_low_pending(pending_size);
2218 migration_completion(s, current_active_state,
2219 &old_vm_running, &start_time);
2220 break;
2221 }
2222 }
2223
2224 if (qemu_file_get_error(s->to_dst_file)) {
2225 migrate_set_state(&s->state, current_active_state,
2226 MIGRATION_STATUS_FAILED);
2227 trace_migration_thread_file_err();
2228 break;
2229 }
2230 current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2231 if (current_time >= initial_time + BUFFER_DELAY) {
2232 uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) -
2233 initial_bytes;
2234 uint64_t time_spent = current_time - initial_time;
2235 double bandwidth = (double)transferred_bytes / time_spent;
2236 threshold_size = bandwidth * s->parameters.downtime_limit;
2237
2238 s->mbps = (((double) transferred_bytes * 8.0) /
2239 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
2240
2241 trace_migrate_transferred(transferred_bytes, time_spent,
2242 bandwidth, threshold_size);
2243 /* if we haven't sent anything, we don't want to recalculate
2244 10000 is a small enough number for our purposes */
2245 if (ram_counters.dirty_pages_rate && transferred_bytes > 10000) {
2246 s->expected_downtime = ram_counters.dirty_pages_rate *
2247 qemu_target_page_size() / bandwidth;
2248 }
2249
2250 qemu_file_reset_rate_limit(s->to_dst_file);
2251 initial_time = current_time;
2252 initial_bytes = qemu_ftell(s->to_dst_file);
2253 }
2254 if (qemu_file_rate_limit(s->to_dst_file)) {
2255 /* usleep expects microseconds */
2256 g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
2257 }
2258 }
2259
2260 trace_migration_thread_after_loop();
2261 /* If we enabled cpu throttling for auto-converge, turn it off. */
2262 cpu_throttle_stop();
2263 end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2264
2265 qemu_mutex_lock_iothread();
2266 /*
2267 * The resource has been allocated by migration will be reused in COLO
2268 * process, so don't release them.
2269 */
2270 if (!enable_colo) {
2271 qemu_savevm_state_cleanup();
2272 }
2273 if (s->state == MIGRATION_STATUS_COMPLETED) {
2274 uint64_t transferred_bytes = qemu_ftell(s->to_dst_file);
2275 s->total_time = end_time - s->total_time;
2276 if (!entered_postcopy) {
2277 s->downtime = end_time - start_time;
2278 }
2279 if (s->total_time) {
2280 s->mbps = (((double) transferred_bytes * 8.0) /
2281 ((double) s->total_time)) / 1000;
2282 }
2283 runstate_set(RUN_STATE_POSTMIGRATE);
2284 } else {
2285 if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) {
2286 migrate_start_colo_process(s);
2287 qemu_savevm_state_cleanup();
2288 /*
2289 * Fixme: we will run VM in COLO no matter its old running state.
2290 * After exited COLO, we will keep running.
2291 */
2292 old_vm_running = true;
2293 }
2294 if (old_vm_running && !entered_postcopy) {
2295 vm_start();
2296 } else {
2297 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2298 runstate_set(RUN_STATE_POSTMIGRATE);
2299 }
2300 }
2301 }
2302 qemu_bh_schedule(s->cleanup_bh);
2303 qemu_mutex_unlock_iothread();
2304
2305 rcu_unregister_thread();
2306 return NULL;
2307 }
2308
2309 void migrate_fd_connect(MigrationState *s)
2310 {
2311 s->expected_downtime = s->parameters.downtime_limit;
2312 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
2313
2314 qemu_file_set_blocking(s->to_dst_file, true);
2315 qemu_file_set_rate_limit(s->to_dst_file,
2316 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
2317
2318 /* Notify before starting migration thread */
2319 notifier_list_notify(&migration_state_notifiers, s);
2320
2321 /*
2322 * Open the return path. For postcopy, it is used exclusively. For
2323 * precopy, only if user specified "return-path" capability would
2324 * QEMU uses the return path.
2325 */
2326 if (migrate_postcopy_ram() || migrate_use_return_path()) {
2327 if (open_return_path_on_source(s)) {
2328 error_report("Unable to open return-path for postcopy");
2329 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2330 MIGRATION_STATUS_FAILED);
2331 migrate_fd_cleanup(s);
2332 return;
2333 }
2334 }
2335
2336 if (multifd_save_setup() != 0) {
2337 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2338 MIGRATION_STATUS_FAILED);
2339 migrate_fd_cleanup(s);
2340 return;
2341 }
2342 qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
2343 QEMU_THREAD_JOINABLE);
2344 s->migration_thread_running = true;
2345 }
2346
2347 void migration_global_dump(Monitor *mon)
2348 {
2349 MigrationState *ms = migrate_get_current();
2350
2351 monitor_printf(mon, "globals: store-global-state=%d, only_migratable=%d, "
2352 "send-configuration=%d, send-section-footer=%d\n",
2353 ms->store_global_state, ms->only_migratable,
2354 ms->send_configuration, ms->send_section_footer);
2355 }
2356
2357 #define DEFINE_PROP_MIG_CAP(name, x) \
2358 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
2359
2360 static Property migration_properties[] = {
2361 DEFINE_PROP_BOOL("store-global-state", MigrationState,
2362 store_global_state, true),
2363 DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false),
2364 DEFINE_PROP_BOOL("send-configuration", MigrationState,
2365 send_configuration, true),
2366 DEFINE_PROP_BOOL("send-section-footer", MigrationState,
2367 send_section_footer, true),
2368
2369 /* Migration parameters */
2370 DEFINE_PROP_INT64("x-compress-level", MigrationState,
2371 parameters.compress_level,
2372 DEFAULT_MIGRATE_COMPRESS_LEVEL),
2373 DEFINE_PROP_INT64("x-compress-threads", MigrationState,
2374 parameters.compress_threads,
2375 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
2376 DEFINE_PROP_INT64("x-decompress-threads", MigrationState,
2377 parameters.decompress_threads,
2378 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
2379 DEFINE_PROP_INT64("x-cpu-throttle-initial", MigrationState,
2380 parameters.cpu_throttle_initial,
2381 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
2382 DEFINE_PROP_INT64("x-cpu-throttle-increment", MigrationState,
2383 parameters.cpu_throttle_increment,
2384 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
2385 DEFINE_PROP_INT64("x-max-bandwidth", MigrationState,
2386 parameters.max_bandwidth, MAX_THROTTLE),
2387 DEFINE_PROP_INT64("x-downtime-limit", MigrationState,
2388 parameters.downtime_limit,
2389 DEFAULT_MIGRATE_SET_DOWNTIME),
2390 DEFINE_PROP_INT64("x-checkpoint-delay", MigrationState,
2391 parameters.x_checkpoint_delay,
2392 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
2393 DEFINE_PROP_INT64("x-multifd-channels", MigrationState,
2394 parameters.x_multifd_channels,
2395 DEFAULT_MIGRATE_MULTIFD_CHANNELS),
2396 DEFINE_PROP_INT64("x-multifd-page-count", MigrationState,
2397 parameters.x_multifd_page_count,
2398 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT),
2399
2400 /* Migration capabilities */
2401 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
2402 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
2403 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
2404 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
2405 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
2406 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
2407 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
2408 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
2409 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
2410 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
2411 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
2412 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD),
2413
2414 DEFINE_PROP_END_OF_LIST(),
2415 };
2416
2417 static void migration_class_init(ObjectClass *klass, void *data)
2418 {
2419 DeviceClass *dc = DEVICE_CLASS(klass);
2420
2421 dc->user_creatable = false;
2422 dc->props = migration_properties;
2423 }
2424
2425 static void migration_instance_finalize(Object *obj)
2426 {
2427 MigrationState *ms = MIGRATION_OBJ(obj);
2428 MigrationParameters *params = &ms->parameters;
2429
2430 g_free(params->tls_hostname);
2431 g_free(params->tls_creds);
2432 qemu_sem_destroy(&ms->pause_sem);
2433 }
2434
2435 static void migration_instance_init(Object *obj)
2436 {
2437 MigrationState *ms = MIGRATION_OBJ(obj);
2438 MigrationParameters *params = &ms->parameters;
2439
2440 ms->state = MIGRATION_STATUS_NONE;
2441 ms->xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE;
2442 ms->mbps = -1;
2443 qemu_sem_init(&ms->pause_sem, 0);
2444
2445 params->tls_hostname = g_strdup("");
2446 params->tls_creds = g_strdup("");
2447
2448 /* Set has_* up only for parameter checks */
2449 params->has_compress_level = true;
2450 params->has_compress_threads = true;
2451 params->has_decompress_threads = true;
2452 params->has_cpu_throttle_initial = true;
2453 params->has_cpu_throttle_increment = true;
2454 params->has_max_bandwidth = true;
2455 params->has_downtime_limit = true;
2456 params->has_x_checkpoint_delay = true;
2457 params->has_block_incremental = true;
2458 params->has_x_multifd_channels = true;
2459 params->has_x_multifd_page_count = true;
2460 }
2461
2462 /*
2463 * Return true if check pass, false otherwise. Error will be put
2464 * inside errp if provided.
2465 */
2466 static bool migration_object_check(MigrationState *ms, Error **errp)
2467 {
2468 MigrationCapabilityStatusList *head = NULL;
2469 /* Assuming all off */
2470 bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
2471 int i;
2472
2473 if (!migrate_params_check(&ms->parameters, errp)) {
2474 return false;
2475 }
2476
2477 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
2478 if (ms->enabled_capabilities[i]) {
2479 head = migrate_cap_add(head, i, true);
2480 }
2481 }
2482
2483 ret = migrate_caps_check(cap_list, head, errp);
2484
2485 /* It works with head == NULL */
2486 qapi_free_MigrationCapabilityStatusList(head);
2487
2488 return ret;
2489 }
2490
2491 static const TypeInfo migration_type = {
2492 .name = TYPE_MIGRATION,
2493 /*
2494 * NOTE: TYPE_MIGRATION is not really a device, as the object is
2495 * not created using qdev_create(), it is not attached to the qdev
2496 * device tree, and it is never realized.
2497 *
2498 * TODO: Make this TYPE_OBJECT once QOM provides something like
2499 * TYPE_DEVICE's "-global" properties.
2500 */
2501 .parent = TYPE_DEVICE,
2502 .class_init = migration_class_init,
2503 .class_size = sizeof(MigrationClass),
2504 .instance_size = sizeof(MigrationState),
2505 .instance_init = migration_instance_init,
2506 .instance_finalize = migration_instance_finalize,
2507 };
2508
2509 static void register_migration_types(void)
2510 {
2511 type_register_static(&migration_type);
2512 }
2513
2514 type_init(register_migration_types);
QEmu