?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
migration: Fix migrate_test_apply for multifd parameters
[qemu/ar7.git] / migration / migration.c
blobfb62a639d8042ac7776128a8bd40537064b563ac
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
108 void migration_object_init(void)
109 {
110 MachineState *ms = MACHINE(qdev_get_machine());
111 Error *err = NULL;
112
113 /* This can only be called once. */
114 assert(!current_migration);
115 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
116
117 if (!migration_object_check(current_migration, &err)) {
118 error_report_err(err);
119 exit(1);
120 }
121
122 /*
123 * We cannot really do this in migration_instance_init() since at
124 * that time global properties are not yet applied, then this
125 * value will be definitely replaced by something else.
126 */
127 if (ms->enforce_config_section) {
128 current_migration->send_configuration = true;
129 }
130 }
131
132 /* For outgoing */
133 MigrationState *migrate_get_current(void)
134 {
135 /* This can only be called after the object created. */
136 assert(current_migration);
137 return current_migration;
138 }
139
140 MigrationIncomingState *migration_incoming_get_current(void)
141 {
142 static bool once;
143 static MigrationIncomingState mis_current;
144
145 if (!once) {
146 mis_current.state = MIGRATION_STATUS_NONE;
147 memset(&mis_current, 0, sizeof(MigrationIncomingState));
148 qemu_mutex_init(&mis_current.rp_mutex);
149 qemu_event_init(&mis_current.main_thread_load_event, false);
150 once = true;
151 }
152 return &mis_current;
153 }
154
155 void migration_incoming_state_destroy(void)
156 {
157 struct MigrationIncomingState *mis = migration_incoming_get_current();
158
159 if (mis->to_src_file) {
160 /* Tell source that we are done */
161 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
162 qemu_fclose(mis->to_src_file);
163 mis->to_src_file = NULL;
164 }
165
166 if (mis->from_src_file) {
167 qemu_fclose(mis->from_src_file);
168 mis->from_src_file = NULL;
169 }
170
171 qemu_event_reset(&mis->main_thread_load_event);
172 }
173
174 static void migrate_generate_event(int new_state)
175 {
176 if (migrate_use_events()) {
177 qapi_event_send_migration(new_state, &error_abort);
178 }
179 }
180
181 /*
182 * Called on -incoming with a defer: uri.
183 * The migration can be started later after any parameters have been
184 * changed.
185 */
186 static void deferred_incoming_migration(Error **errp)
187 {
188 if (deferred_incoming) {
189 error_setg(errp, "Incoming migration already deferred");
190 }
191 deferred_incoming = true;
192 }
193
194 /*
195 * Send a message on the return channel back to the source
196 * of the migration.
197 */
198 static void migrate_send_rp_message(MigrationIncomingState *mis,
199 enum mig_rp_message_type message_type,
200 uint16_t len, void *data)
201 {
202 trace_migrate_send_rp_message((int)message_type, len);
203 qemu_mutex_lock(&mis->rp_mutex);
204 qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
205 qemu_put_be16(mis->to_src_file, len);
206 qemu_put_buffer(mis->to_src_file, data, len);
207 qemu_fflush(mis->to_src_file);
208 qemu_mutex_unlock(&mis->rp_mutex);
209 }
210
211 /* Request a range of pages from the source VM at the given
212 * start address.
213 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
214 * as the last request (a name must have been given previously)
215 * Start: Address offset within the RB
216 * Len: Length in bytes required - must be a multiple of pagesize
217 */
218 void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
219 ram_addr_t start, size_t len)
220 {
221 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
222 size_t msglen = 12; /* start + len */
223
224 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
225 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
226
227 if (rbname) {
228 int rbname_len = strlen(rbname);
229 assert(rbname_len < 256);
230
231 bufc[msglen++] = rbname_len;
232 memcpy(bufc + msglen, rbname, rbname_len);
233 msglen += rbname_len;
234 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc);
235 } else {
236 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc);
237 }
238 }
239
240 void qemu_start_incoming_migration(const char *uri, Error **errp)
241 {
242 const char *p;
243
244 qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
245 if (!strcmp(uri, "defer")) {
246 deferred_incoming_migration(errp);
247 } else if (strstart(uri, "tcp:", &p)) {
248 tcp_start_incoming_migration(p, errp);
249 #ifdef CONFIG_RDMA
250 } else if (strstart(uri, "rdma:", &p)) {
251 rdma_start_incoming_migration(p, errp);
252 #endif
253 } else if (strstart(uri, "exec:", &p)) {
254 exec_start_incoming_migration(p, errp);
255 } else if (strstart(uri, "unix:", &p)) {
256 unix_start_incoming_migration(p, errp);
257 } else if (strstart(uri, "fd:", &p)) {
258 fd_start_incoming_migration(p, errp);
259 } else {
260 error_setg(errp, "unknown migration protocol: %s", uri);
261 }
262 }
263
264 static void process_incoming_migration_bh(void *opaque)
265 {
266 Error *local_err = NULL;
267 MigrationIncomingState *mis = opaque;
268
269 /* Make sure all file formats flush their mutable metadata.
270 * If we get an error here, just don't restart the VM yet. */
271 bdrv_invalidate_cache_all(&local_err);
272 if (local_err) {
273 error_report_err(local_err);
274 local_err = NULL;
275 autostart = false;
276 }
277
278 /*
279 * This must happen after all error conditions are dealt with and
280 * we're sure the VM is going to be running on this host.
281 */
282 qemu_announce_self();
283
284 if (multifd_load_cleanup(&local_err) != 0) {
285 error_report_err(local_err);
286 autostart = false;
287 }
288 /* If global state section was not received or we are in running
289 state, we need to obey autostart. Any other state is set with
290 runstate_set. */
291
292 if (!global_state_received() ||
293 global_state_get_runstate() == RUN_STATE_RUNNING) {
294 if (autostart) {
295 vm_start();
296 } else {
297 runstate_set(RUN_STATE_PAUSED);
298 }
299 } else {
300 runstate_set(global_state_get_runstate());
301 }
302 /*
303 * This must happen after any state changes since as soon as an external
304 * observer sees this event they might start to prod at the VM assuming
305 * it's ready to use.
306 */
307 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
308 MIGRATION_STATUS_COMPLETED);
309 qemu_bh_delete(mis->bh);
310 migration_incoming_state_destroy();
311 }
312
313 static void process_incoming_migration_co(void *opaque)
314 {
315 MigrationIncomingState *mis = migration_incoming_get_current();
316 PostcopyState ps;
317 int ret;
318
319 assert(mis->from_src_file);
320 mis->largest_page_size = qemu_ram_pagesize_largest();
321 postcopy_state_set(POSTCOPY_INCOMING_NONE);
322 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
323 MIGRATION_STATUS_ACTIVE);
324 ret = qemu_loadvm_state(mis->from_src_file);
325
326 ps = postcopy_state_get();
327 trace_process_incoming_migration_co_end(ret, ps);
328 if (ps != POSTCOPY_INCOMING_NONE) {
329 if (ps == POSTCOPY_INCOMING_ADVISE) {
330 /*
331 * Where a migration had postcopy enabled (and thus went to advise)
332 * but managed to complete within the precopy period, we can use
333 * the normal exit.
334 */
335 postcopy_ram_incoming_cleanup(mis);
336 } else if (ret >= 0) {
337 /*
338 * Postcopy was started, cleanup should happen at the end of the
339 * postcopy thread.
340 */
341 trace_process_incoming_migration_co_postcopy_end_main();
342 return;
343 }
344 /* Else if something went wrong then just fall out of the normal exit */
345 }
346
347 /* we get COLO info, and know if we are in COLO mode */
348 if (!ret && migration_incoming_enable_colo()) {
349 mis->migration_incoming_co = qemu_coroutine_self();
350 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
351 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
352 mis->have_colo_incoming_thread = true;
353 qemu_coroutine_yield();
354
355 /* Wait checkpoint incoming thread exit before free resource */
356 qemu_thread_join(&mis->colo_incoming_thread);
357 }
358
359 if (ret < 0) {
360 Error *local_err = NULL;
361
362 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
363 MIGRATION_STATUS_FAILED);
364 error_report("load of migration failed: %s", strerror(-ret));
365 qemu_fclose(mis->from_src_file);
366 if (multifd_load_cleanup(&local_err) != 0) {
367 error_report_err(local_err);
368 }
369 exit(EXIT_FAILURE);
370 }
371 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
372 qemu_bh_schedule(mis->bh);
373 }
374
375 static void migration_incoming_setup(QEMUFile *f)
376 {
377 MigrationIncomingState *mis = migration_incoming_get_current();
378
379 if (multifd_load_setup() != 0) {
380 /* We haven't been able to create multifd threads
381 nothing better to do */
382 exit(EXIT_FAILURE);
383 }
384
385 if (!mis->from_src_file) {
386 mis->from_src_file = f;
387 }
388 qemu_file_set_blocking(f, false);
389 }
390
391 static void migration_incoming_process(void)
392 {
393 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
394 qemu_coroutine_enter(co);
395 }
396
397 void migration_fd_process_incoming(QEMUFile *f)
398 {
399 migration_incoming_setup(f);
400 migration_incoming_process();
401 }
402
403 void migration_ioc_process_incoming(QIOChannel *ioc)
404 {
405 MigrationIncomingState *mis = migration_incoming_get_current();
406
407 if (!mis->from_src_file) {
408 QEMUFile *f = qemu_fopen_channel_input(ioc);
409 migration_fd_process_incoming(f);
410 }
411 /* We still only have a single channel. Nothing to do here yet */
412 }
413
414 /**
415 * @migration_has_all_channels: We have received all channels that we need
416 *
417 * Returns true when we have got connections to all the channels that
418 * we need for migration.
419 */
420 bool migration_has_all_channels(void)
421 {
422 return true;
423 }
424
425 /*
426 * Send a 'SHUT' message on the return channel with the given value
427 * to indicate that we've finished with the RP. Non-0 value indicates
428 * error.
429 */
430 void migrate_send_rp_shut(MigrationIncomingState *mis,
431 uint32_t value)
432 {
433 uint32_t buf;
434
435 buf = cpu_to_be32(value);
436 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
437 }
438
439 /*
440 * Send a 'PONG' message on the return channel with the given value
441 * (normally in response to a 'PING')
442 */
443 void migrate_send_rp_pong(MigrationIncomingState *mis,
444 uint32_t value)
445 {
446 uint32_t buf;
447
448 buf = cpu_to_be32(value);
449 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
450 }
451
452 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
453 {
454 MigrationCapabilityStatusList *head = NULL;
455 MigrationCapabilityStatusList *caps;
456 MigrationState *s = migrate_get_current();
457 int i;
458
459 caps = NULL; /* silence compiler warning */
460 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
461 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
462 if (i == MIGRATION_CAPABILITY_BLOCK) {
463 continue;
464 }
465 #endif
466 if (head == NULL) {
467 head = g_malloc0(sizeof(*caps));
468 caps = head;
469 } else {
470 caps->next = g_malloc0(sizeof(*caps));
471 caps = caps->next;
472 }
473 caps->value =
474 g_malloc(sizeof(*caps->value));
475 caps->value->capability = i;
476 caps->value->state = s->enabled_capabilities[i];
477 }
478
479 return head;
480 }
481
482 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
483 {
484 MigrationParameters *params;
485 MigrationState *s = migrate_get_current();
486
487 /* TODO use QAPI_CLONE() instead of duplicating it inline */
488 params = g_malloc0(sizeof(*params));
489 params->has_compress_level = true;
490 params->compress_level = s->parameters.compress_level;
491 params->has_compress_threads = true;
492 params->compress_threads = s->parameters.compress_threads;
493 params->has_decompress_threads = true;
494 params->decompress_threads = s->parameters.decompress_threads;
495 params->has_cpu_throttle_initial = true;
496 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
497 params->has_cpu_throttle_increment = true;
498 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
499 params->has_tls_creds = true;
500 params->tls_creds = g_strdup(s->parameters.tls_creds);
501 params->has_tls_hostname = true;
502 params->tls_hostname = g_strdup(s->parameters.tls_hostname);
503 params->has_max_bandwidth = true;
504 params->max_bandwidth = s->parameters.max_bandwidth;
505 params->has_downtime_limit = true;
506 params->downtime_limit = s->parameters.downtime_limit;
507 params->has_x_checkpoint_delay = true;
508 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
509 params->has_block_incremental = true;
510 params->block_incremental = s->parameters.block_incremental;
511 params->has_x_multifd_channels = true;
512 params->x_multifd_channels = s->parameters.x_multifd_channels;
513 params->has_x_multifd_page_count = true;
514 params->x_multifd_page_count = s->parameters.x_multifd_page_count;
515
516 return params;
517 }
518
519 /*
520 * Return true if we're already in the middle of a migration
521 * (i.e. any of the active or setup states)
522 */
523 static bool migration_is_setup_or_active(int state)
524 {
525 switch (state) {
526 case MIGRATION_STATUS_ACTIVE:
527 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
528 case MIGRATION_STATUS_SETUP:
529 return true;
530
531 default:
532 return false;
533
534 }
535 }
536
537 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
538 {
539 info->has_ram = true;
540 info->ram = g_malloc0(sizeof(*info->ram));
541 info->ram->transferred = ram_counters.transferred;
542 info->ram->total = ram_bytes_total();
543 info->ram->duplicate = ram_counters.duplicate;
544 /* legacy value. It is not used anymore */
545 info->ram->skipped = 0;
546 info->ram->normal = ram_counters.normal;
547 info->ram->normal_bytes = ram_counters.normal *
548 qemu_target_page_size();
549 info->ram->mbps = s->mbps;
550 info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
551 info->ram->postcopy_requests = ram_counters.postcopy_requests;
552 info->ram->page_size = qemu_target_page_size();
553
554 if (migrate_use_xbzrle()) {
555 info->has_xbzrle_cache = true;
556 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
557 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
558 info->xbzrle_cache->bytes = xbzrle_counters.bytes;
559 info->xbzrle_cache->pages = xbzrle_counters.pages;
560 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
561 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
562 info->xbzrle_cache->overflow = xbzrle_counters.overflow;
563 }
564
565 if (cpu_throttle_active()) {
566 info->has_cpu_throttle_percentage = true;
567 info->cpu_throttle_percentage = cpu_throttle_get_percentage();
568 }
569
570 if (s->state != MIGRATION_STATUS_COMPLETED) {
571 info->ram->remaining = ram_bytes_remaining();
572 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
573 }
574 }
575
576 static void populate_disk_info(MigrationInfo *info)
577 {
578 if (blk_mig_active()) {
579 info->has_disk = true;
580 info->disk = g_malloc0(sizeof(*info->disk));
581 info->disk->transferred = blk_mig_bytes_transferred();
582 info->disk->remaining = blk_mig_bytes_remaining();
583 info->disk->total = blk_mig_bytes_total();
584 }
585 }
586
587 MigrationInfo *qmp_query_migrate(Error **errp)
588 {
589 MigrationInfo *info = g_malloc0(sizeof(*info));
590 MigrationState *s = migrate_get_current();
591
592 switch (s->state) {
593 case MIGRATION_STATUS_NONE:
594 /* no migration has happened ever */
595 break;
596 case MIGRATION_STATUS_SETUP:
597 info->has_status = true;
598 info->has_total_time = false;
599 break;
600 case MIGRATION_STATUS_ACTIVE:
601 case MIGRATION_STATUS_CANCELLING:
602 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
603 /* TODO add some postcopy stats */
604 info->has_status = true;
605 info->has_total_time = true;
606 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
607 - s->total_time;
608 info->has_expected_downtime = true;
609 info->expected_downtime = s->expected_downtime;
610 info->has_setup_time = true;
611 info->setup_time = s->setup_time;
612
613 populate_ram_info(info, s);
614 populate_disk_info(info);
615 break;
616 case MIGRATION_STATUS_COLO:
617 info->has_status = true;
618 /* TODO: display COLO specific information (checkpoint info etc.) */
619 break;
620 case MIGRATION_STATUS_COMPLETED:
621 info->has_status = true;
622 info->has_total_time = true;
623 info->total_time = s->total_time;
624 info->has_downtime = true;
625 info->downtime = s->downtime;
626 info->has_setup_time = true;
627 info->setup_time = s->setup_time;
628
629 populate_ram_info(info, s);
630 break;
631 case MIGRATION_STATUS_FAILED:
632 info->has_status = true;
633 if (s->error) {
634 info->has_error_desc = true;
635 info->error_desc = g_strdup(error_get_pretty(s->error));
636 }
637 break;
638 case MIGRATION_STATUS_CANCELLED:
639 info->has_status = true;
640 break;
641 }
642 info->status = s->state;
643
644 return info;
645 }
646
647 /**
648 * @migration_caps_check - check capability validity
649 *
650 * @cap_list: old capability list, array of bool
651 * @params: new capabilities to be applied soon
652 * @errp: set *errp if the check failed, with reason
653 *
654 * Returns true if check passed, otherwise false.
655 */
656 static bool migrate_caps_check(bool *cap_list,
657 MigrationCapabilityStatusList *params,
658 Error **errp)
659 {
660 MigrationCapabilityStatusList *cap;
661 bool old_postcopy_cap;
662 MigrationIncomingState *mis = migration_incoming_get_current();
663
664 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
665
666 for (cap = params; cap; cap = cap->next) {
667 cap_list[cap->value->capability] = cap->value->state;
668 }
669
670 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
671 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
672 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
673 "block migration");
674 error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
675 return false;
676 }
677 #endif
678
679 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
680 if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
681 /* The decompression threads asynchronously write into RAM
682 * rather than use the atomic copies needed to avoid
683 * userfaulting. It should be possible to fix the decompression
684 * threads for compatibility in future.
685 */
686 error_setg(errp, "Postcopy is not currently compatible "
687 "with compression");
688 return false;
689 }
690
691 /* This check is reasonably expensive, so only when it's being
692 * set the first time, also it's only the destination that needs
693 * special support.
694 */
695 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
696 !postcopy_ram_supported_by_host(mis)) {
697 /* postcopy_ram_supported_by_host will have emitted a more
698 * detailed message
699 */
700 error_setg(errp, "Postcopy is not supported");
701 return false;
702 }
703 }
704
705 return true;
706 }
707
708 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
709 Error **errp)
710 {
711 MigrationState *s = migrate_get_current();
712 MigrationCapabilityStatusList *cap;
713
714 if (migration_is_setup_or_active(s->state)) {
715 error_setg(errp, QERR_MIGRATION_ACTIVE);
716 return;
717 }
718
719 if (!migrate_caps_check(s->enabled_capabilities, params, errp)) {
720 return;
721 }
722
723 for (cap = params; cap; cap = cap->next) {
724 s->enabled_capabilities[cap->value->capability] = cap->value->state;
725 }
726 }
727
728 /*
729 * Check whether the parameters are valid. Error will be put into errp
730 * (if provided). Return true if valid, otherwise false.
731 */
732 static bool migrate_params_check(MigrationParameters *params, Error **errp)
733 {
734 if (params->has_compress_level &&
735 (params->compress_level < 0 || params->compress_level > 9)) {
736 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
737 "is invalid, it should be in the range of 0 to 9");
738 return false;
739 }
740
741 if (params->has_compress_threads &&
742 (params->compress_threads < 1 || params->compress_threads > 255)) {
743 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
744 "compress_threads",
745 "is invalid, it should be in the range of 1 to 255");
746 return false;
747 }
748
749 if (params->has_decompress_threads &&
750 (params->decompress_threads < 1 || params->decompress_threads > 255)) {
751 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
752 "decompress_threads",
753 "is invalid, it should be in the range of 1 to 255");
754 return false;
755 }
756
757 if (params->has_cpu_throttle_initial &&
758 (params->cpu_throttle_initial < 1 ||
759 params->cpu_throttle_initial > 99)) {
760 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
761 "cpu_throttle_initial",
762 "an integer in the range of 1 to 99");
763 return false;
764 }
765
766 if (params->has_cpu_throttle_increment &&
767 (params->cpu_throttle_increment < 1 ||
768 params->cpu_throttle_increment > 99)) {
769 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
770 "cpu_throttle_increment",
771 "an integer in the range of 1 to 99");
772 return false;
773 }
774
775 if (params->has_max_bandwidth &&
776 (params->max_bandwidth < 0 || params->max_bandwidth > SIZE_MAX)) {
777 error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
778 " range of 0 to %zu bytes/second", SIZE_MAX);
779 return false;
780 }
781
782 if (params->has_downtime_limit &&
783 (params->downtime_limit < 0 ||
784 params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
785 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
786 "the range of 0 to %d milliseconds",
787 MAX_MIGRATE_DOWNTIME);
788 return false;
789 }
790
791 if (params->has_x_checkpoint_delay && (params->x_checkpoint_delay < 0)) {
792 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
793 "x_checkpoint_delay",
794 "is invalid, it should be positive");
795 return false;
796 }
797 if (params->has_x_multifd_channels &&
798 (params->x_multifd_channels < 1 || params->x_multifd_channels > 255)) {
799 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
800 "multifd_channels",
801 "is invalid, it should be in the range of 1 to 255");
802 return false;
803 }
804 if (params->has_x_multifd_page_count &&
805 (params->x_multifd_page_count < 1 ||
806 params->x_multifd_page_count > 10000)) {
807 error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
808 "multifd_page_count",
809 "is invalid, it should be in the range of 1 to 10000");
810 return false;
811 }
812
813 return true;
814 }
815
816 static void migrate_params_test_apply(MigrateSetParameters *params,
817 MigrationParameters *dest)
818 {
819 *dest = migrate_get_current()->parameters;
820
821 /* TODO use QAPI_CLONE() instead of duplicating it inline */
822
823 if (params->has_compress_level) {
824 dest->compress_level = params->compress_level;
825 }
826
827 if (params->has_compress_threads) {
828 dest->compress_threads = params->compress_threads;
829 }
830
831 if (params->has_decompress_threads) {
832 dest->decompress_threads = params->decompress_threads;
833 }
834
835 if (params->has_cpu_throttle_initial) {
836 dest->cpu_throttle_initial = params->cpu_throttle_initial;
837 }
838
839 if (params->has_cpu_throttle_increment) {
840 dest->cpu_throttle_increment = params->cpu_throttle_increment;
841 }
842
843 if (params->has_tls_creds) {
844 assert(params->tls_creds->type == QTYPE_QSTRING);
845 dest->tls_creds = g_strdup(params->tls_creds->u.s);
846 }
847
848 if (params->has_tls_hostname) {
849 assert(params->tls_hostname->type == QTYPE_QSTRING);
850 dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
851 }
852
853 if (params->has_max_bandwidth) {
854 dest->max_bandwidth = params->max_bandwidth;
855 }
856
857 if (params->has_downtime_limit) {
858 dest->downtime_limit = params->downtime_limit;
859 }
860
861 if (params->has_x_checkpoint_delay) {
862 dest->x_checkpoint_delay = params->x_checkpoint_delay;
863 }
864
865 if (params->has_block_incremental) {
866 dest->block_incremental = params->block_incremental;
867 }
868 if (params->has_x_multifd_channels) {
869 dest->x_multifd_channels = params->x_multifd_channels;
870 }
871 if (params->has_x_multifd_page_count) {
872 dest->x_multifd_page_count = params->x_multifd_page_count;
873 }
874 }
875
876 static void migrate_params_apply(MigrateSetParameters *params)
877 {
878 MigrationState *s = migrate_get_current();
879
880 /* TODO use QAPI_CLONE() instead of duplicating it inline */
881
882 if (params->has_compress_level) {
883 s->parameters.compress_level = params->compress_level;
884 }
885
886 if (params->has_compress_threads) {
887 s->parameters.compress_threads = params->compress_threads;
888 }
889
890 if (params->has_decompress_threads) {
891 s->parameters.decompress_threads = params->decompress_threads;
892 }
893
894 if (params->has_cpu_throttle_initial) {
895 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
896 }
897
898 if (params->has_cpu_throttle_increment) {
899 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
900 }
901
902 if (params->has_tls_creds) {
903 g_free(s->parameters.tls_creds);
904 assert(params->tls_creds->type == QTYPE_QSTRING);
905 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
906 }
907
908 if (params->has_tls_hostname) {
909 g_free(s->parameters.tls_hostname);
910 assert(params->tls_hostname->type == QTYPE_QSTRING);
911 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
912 }
913
914 if (params->has_max_bandwidth) {
915 s->parameters.max_bandwidth = params->max_bandwidth;
916 if (s->to_dst_file) {
917 qemu_file_set_rate_limit(s->to_dst_file,
918 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
919 }
920 }
921
922 if (params->has_downtime_limit) {
923 s->parameters.downtime_limit = params->downtime_limit;
924 }
925
926 if (params->has_x_checkpoint_delay) {
927 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
928 if (migration_in_colo_state()) {
929 colo_checkpoint_notify(s);
930 }
931 }
932
933 if (params->has_block_incremental) {
934 s->parameters.block_incremental = params->block_incremental;
935 }
936 if (params->has_x_multifd_channels) {
937 s->parameters.x_multifd_channels = params->x_multifd_channels;
938 }
939 if (params->has_x_multifd_page_count) {
940 s->parameters.x_multifd_page_count = params->x_multifd_page_count;
941 }
942 }
943
944 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
945 {
946 MigrationParameters tmp;
947
948 /* TODO Rewrite "" to null instead */
949 if (params->has_tls_creds
950 && params->tls_creds->type == QTYPE_QNULL) {
951 QDECREF(params->tls_creds->u.n);
952 params->tls_creds->type = QTYPE_QSTRING;
953 params->tls_creds->u.s = strdup("");
954 }
955 /* TODO Rewrite "" to null instead */
956 if (params->has_tls_hostname
957 && params->tls_hostname->type == QTYPE_QNULL) {
958 QDECREF(params->tls_hostname->u.n);
959 params->tls_hostname->type = QTYPE_QSTRING;
960 params->tls_hostname->u.s = strdup("");
961 }
962
963 migrate_params_test_apply(params, &tmp);
964
965 if (!migrate_params_check(&tmp, errp)) {
966 /* Invalid parameter */
967 return;
968 }
969
970 migrate_params_apply(params);
971 }
972
973
974 void qmp_migrate_start_postcopy(Error **errp)
975 {
976 MigrationState *s = migrate_get_current();
977
978 if (!migrate_postcopy_ram()) {
979 error_setg(errp, "Enable postcopy with migrate_set_capability before"
980 " the start of migration");
981 return;
982 }
983
984 if (s->state == MIGRATION_STATUS_NONE) {
985 error_setg(errp, "Postcopy must be started after migration has been"
986 " started");
987 return;
988 }
989 /*
990 * we don't error if migration has finished since that would be racy
991 * with issuing this command.
992 */
993 atomic_set(&s->start_postcopy, true);
994 }
995
996 /* shared migration helpers */
997
998 void migrate_set_state(int *state, int old_state, int new_state)
999 {
1000 assert(new_state < MIGRATION_STATUS__MAX);
1001 if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
1002 trace_migrate_set_state(MigrationStatus_str(new_state));
1003 migrate_generate_event(new_state);
1004 }
1005 }
1006
1007 static MigrationCapabilityStatusList *migrate_cap_add(
1008 MigrationCapabilityStatusList *list,
1009 MigrationCapability index,
1010 bool state)
1011 {
1012 MigrationCapabilityStatusList *cap;
1013
1014 cap = g_new0(MigrationCapabilityStatusList, 1);
1015 cap->value = g_new0(MigrationCapabilityStatus, 1);
1016 cap->value->capability = index;
1017 cap->value->state = state;
1018 cap->next = list;
1019
1020 return cap;
1021 }
1022
1023 void migrate_set_block_enabled(bool value, Error **errp)
1024 {
1025 MigrationCapabilityStatusList *cap;
1026
1027 cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
1028 qmp_migrate_set_capabilities(cap, errp);
1029 qapi_free_MigrationCapabilityStatusList(cap);
1030 }
1031
1032 static void migrate_set_block_incremental(MigrationState *s, bool value)
1033 {
1034 s->parameters.block_incremental = value;
1035 }
1036
1037 static void block_cleanup_parameters(MigrationState *s)
1038 {
1039 if (s->must_remove_block_options) {
1040 /* setting to false can never fail */
1041 migrate_set_block_enabled(false, &error_abort);
1042 migrate_set_block_incremental(s, false);
1043 s->must_remove_block_options = false;
1044 }
1045 }
1046
1047 static void migrate_fd_cleanup(void *opaque)
1048 {
1049 MigrationState *s = opaque;
1050
1051 qemu_bh_delete(s->cleanup_bh);
1052 s->cleanup_bh = NULL;
1053
1054 if (s->to_dst_file) {
1055 Error *local_err = NULL;
1056
1057 trace_migrate_fd_cleanup();
1058 qemu_mutex_unlock_iothread();
1059 if (s->migration_thread_running) {
1060 qemu_thread_join(&s->thread);
1061 s->migration_thread_running = false;
1062 }
1063 qemu_mutex_lock_iothread();
1064
1065 if (multifd_save_cleanup(&local_err) != 0) {
1066 error_report_err(local_err);
1067 }
1068 qemu_fclose(s->to_dst_file);
1069 s->to_dst_file = NULL;
1070 }
1071
1072 assert((s->state != MIGRATION_STATUS_ACTIVE) &&
1073 (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
1074
1075 if (s->state == MIGRATION_STATUS_CANCELLING) {
1076 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1077 MIGRATION_STATUS_CANCELLED);
1078 }
1079
1080 notifier_list_notify(&migration_state_notifiers, s);
1081 block_cleanup_parameters(s);
1082 }
1083
1084 void migrate_fd_error(MigrationState *s, const Error *error)
1085 {
1086 trace_migrate_fd_error(error_get_pretty(error));
1087 assert(s->to_dst_file == NULL);
1088 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1089 MIGRATION_STATUS_FAILED);
1090 if (!s->error) {
1091 s->error = error_copy(error);
1092 }
1093 notifier_list_notify(&migration_state_notifiers, s);
1094 block_cleanup_parameters(s);
1095 }
1096
1097 static void migrate_fd_cancel(MigrationState *s)
1098 {
1099 int old_state ;
1100 QEMUFile *f = migrate_get_current()->to_dst_file;
1101 trace_migrate_fd_cancel();
1102
1103 if (s->rp_state.from_dst_file) {
1104 /* shutdown the rp socket, so causing the rp thread to shutdown */
1105 qemu_file_shutdown(s->rp_state.from_dst_file);
1106 }
1107
1108 do {
1109 old_state = s->state;
1110 if (!migration_is_setup_or_active(old_state)) {
1111 break;
1112 }
1113 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1114 } while (s->state != MIGRATION_STATUS_CANCELLING);
1115
1116 /*
1117 * If we're unlucky the migration code might be stuck somewhere in a
1118 * send/write while the network has failed and is waiting to timeout;
1119 * if we've got shutdown(2) available then we can force it to quit.
1120 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1121 * called in a bh, so there is no race against this cancel.
1122 */
1123 if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1124 qemu_file_shutdown(f);
1125 }
1126 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1127 Error *local_err = NULL;
1128
1129 bdrv_invalidate_cache_all(&local_err);
1130 if (local_err) {
1131 error_report_err(local_err);
1132 } else {
1133 s->block_inactive = false;
1134 }
1135 }
1136 block_cleanup_parameters(s);
1137 }
1138
1139 void add_migration_state_change_notifier(Notifier *notify)
1140 {
1141 notifier_list_add(&migration_state_notifiers, notify);
1142 }
1143
1144 void remove_migration_state_change_notifier(Notifier *notify)
1145 {
1146 notifier_remove(notify);
1147 }
1148
1149 bool migration_in_setup(MigrationState *s)
1150 {
1151 return s->state == MIGRATION_STATUS_SETUP;
1152 }
1153
1154 bool migration_has_finished(MigrationState *s)
1155 {
1156 return s->state == MIGRATION_STATUS_COMPLETED;
1157 }
1158
1159 bool migration_has_failed(MigrationState *s)
1160 {
1161 return (s->state == MIGRATION_STATUS_CANCELLED ||
1162 s->state == MIGRATION_STATUS_FAILED);
1163 }
1164
1165 bool migration_in_postcopy(void)
1166 {
1167 MigrationState *s = migrate_get_current();
1168
1169 return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1170 }
1171
1172 bool migration_in_postcopy_after_devices(MigrationState *s)
1173 {
1174 return migration_in_postcopy() && s->postcopy_after_devices;
1175 }
1176
1177 bool migration_is_idle(void)
1178 {
1179 MigrationState *s = migrate_get_current();
1180
1181 switch (s->state) {
1182 case MIGRATION_STATUS_NONE:
1183 case MIGRATION_STATUS_CANCELLED:
1184 case MIGRATION_STATUS_COMPLETED:
1185 case MIGRATION_STATUS_FAILED:
1186 return true;
1187 case MIGRATION_STATUS_SETUP:
1188 case MIGRATION_STATUS_CANCELLING:
1189 case MIGRATION_STATUS_ACTIVE:
1190 case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1191 case MIGRATION_STATUS_COLO:
1192 return false;
1193 case MIGRATION_STATUS__MAX:
1194 g_assert_not_reached();
1195 }
1196
1197 return false;
1198 }
1199
1200 MigrationState *migrate_init(void)
1201 {
1202 MigrationState *s = migrate_get_current();
1203
1204 /*
1205 * Reinitialise all migration state, except
1206 * parameters/capabilities that the user set, and
1207 * locks.
1208 */
1209 s->bytes_xfer = 0;
1210 s->xfer_limit = 0;
1211 s->cleanup_bh = 0;
1212 s->to_dst_file = NULL;
1213 s->state = MIGRATION_STATUS_NONE;
1214 s->rp_state.from_dst_file = NULL;
1215 s->rp_state.error = false;
1216 s->mbps = 0.0;
1217 s->downtime = 0;
1218 s->expected_downtime = 0;
1219 s->setup_time = 0;
1220 s->start_postcopy = false;
1221 s->postcopy_after_devices = false;
1222 s->migration_thread_running = false;
1223 error_free(s->error);
1224 s->error = NULL;
1225
1226 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1227
1228 s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1229 return s;
1230 }
1231
1232 static GSList *migration_blockers;
1233
1234 int migrate_add_blocker(Error *reason, Error **errp)
1235 {
1236 if (migrate_get_current()->only_migratable) {
1237 error_propagate(errp, error_copy(reason));
1238 error_prepend(errp, "disallowing migration blocker "
1239 "(--only_migratable) for: ");
1240 return -EACCES;
1241 }
1242
1243 if (migration_is_idle()) {
1244 migration_blockers = g_slist_prepend(migration_blockers, reason);
1245 return 0;
1246 }
1247
1248 error_propagate(errp, error_copy(reason));
1249 error_prepend(errp, "disallowing migration blocker (migration in "
1250 "progress) for: ");
1251 return -EBUSY;
1252 }
1253
1254 void migrate_del_blocker(Error *reason)
1255 {
1256 migration_blockers = g_slist_remove(migration_blockers, reason);
1257 }
1258
1259 void qmp_migrate_incoming(const char *uri, Error **errp)
1260 {
1261 Error *local_err = NULL;
1262 static bool once = true;
1263
1264 if (!deferred_incoming) {
1265 error_setg(errp, "For use with '-incoming defer'");
1266 return;
1267 }
1268 if (!once) {
1269 error_setg(errp, "The incoming migration has already been started");
1270 }
1271
1272 qemu_start_incoming_migration(uri, &local_err);
1273
1274 if (local_err) {
1275 error_propagate(errp, local_err);
1276 return;
1277 }
1278
1279 once = false;
1280 }
1281
1282 bool migration_is_blocked(Error **errp)
1283 {
1284 if (qemu_savevm_state_blocked(errp)) {
1285 return true;
1286 }
1287
1288 if (migration_blockers) {
1289 error_propagate(errp, error_copy(migration_blockers->data));
1290 return true;
1291 }
1292
1293 return false;
1294 }
1295
1296 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1297 bool has_inc, bool inc, bool has_detach, bool detach,
1298 Error **errp)
1299 {
1300 Error *local_err = NULL;
1301 MigrationState *s = migrate_get_current();
1302 const char *p;
1303
1304 if (migration_is_setup_or_active(s->state) ||
1305 s->state == MIGRATION_STATUS_CANCELLING ||
1306 s->state == MIGRATION_STATUS_COLO) {
1307 error_setg(errp, QERR_MIGRATION_ACTIVE);
1308 return;
1309 }
1310 if (runstate_check(RUN_STATE_INMIGRATE)) {
1311 error_setg(errp, "Guest is waiting for an incoming migration");
1312 return;
1313 }
1314
1315 if (migration_is_blocked(errp)) {
1316 return;
1317 }
1318
1319 if ((has_blk && blk) || (has_inc && inc)) {
1320 if (migrate_use_block() || migrate_use_block_incremental()) {
1321 error_setg(errp, "Command options are incompatible with "
1322 "current migration capabilities");
1323 return;
1324 }
1325 migrate_set_block_enabled(true, &local_err);
1326 if (local_err) {
1327 error_propagate(errp, local_err);
1328 return;
1329 }
1330 s->must_remove_block_options = true;
1331 }
1332
1333 if (has_inc && inc) {
1334 migrate_set_block_incremental(s, true);
1335 }
1336
1337 s = migrate_init();
1338
1339 if (strstart(uri, "tcp:", &p)) {
1340 tcp_start_outgoing_migration(s, p, &local_err);
1341 #ifdef CONFIG_RDMA
1342 } else if (strstart(uri, "rdma:", &p)) {
1343 rdma_start_outgoing_migration(s, p, &local_err);
1344 #endif
1345 } else if (strstart(uri, "exec:", &p)) {
1346 exec_start_outgoing_migration(s, p, &local_err);
1347 } else if (strstart(uri, "unix:", &p)) {
1348 unix_start_outgoing_migration(s, p, &local_err);
1349 } else if (strstart(uri, "fd:", &p)) {
1350 fd_start_outgoing_migration(s, p, &local_err);
1351 } else {
1352 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1353 "a valid migration protocol");
1354 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1355 MIGRATION_STATUS_FAILED);
1356 return;
1357 }
1358
1359 if (local_err) {
1360 migrate_fd_error(s, local_err);
1361 error_propagate(errp, local_err);
1362 return;
1363 }
1364 }
1365
1366 void qmp_migrate_cancel(Error **errp)
1367 {
1368 migrate_fd_cancel(migrate_get_current());
1369 }
1370
1371 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1372 {
1373 MigrationState *s = migrate_get_current();
1374 int64_t new_size;
1375
1376 /* Check for truncation */
1377 if (value != (size_t)value) {
1378 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1379 "exceeding address space");
1380 return;
1381 }
1382
1383 /* Cache should not be larger than guest ram size */
1384 if (value > ram_bytes_total()) {
1385 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1386 "exceeds guest ram size ");
1387 return;
1388 }
1389
1390 new_size = xbzrle_cache_resize(value);
1391 if (new_size < 0) {
1392 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
1393 "is smaller than page size");
1394 return;
1395 }
1396
1397 s->xbzrle_cache_size = new_size;
1398 }
1399
1400 int64_t qmp_query_migrate_cache_size(Error **errp)
1401 {
1402 return migrate_xbzrle_cache_size();
1403 }
1404
1405 void qmp_migrate_set_speed(int64_t value, Error **errp)
1406 {
1407 MigrateSetParameters p = {
1408 .has_max_bandwidth = true,
1409 .max_bandwidth = value,
1410 };
1411
1412 qmp_migrate_set_parameters(&p, errp);
1413 }
1414
1415 void qmp_migrate_set_downtime(double value, Error **errp)
1416 {
1417 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1418 error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1419 "the range of 0 to %d seconds",
1420 MAX_MIGRATE_DOWNTIME_SECONDS);
1421 return;
1422 }
1423
1424 value *= 1000; /* Convert to milliseconds */
1425 value = MAX(0, MIN(INT64_MAX, value));
1426
1427 MigrateSetParameters p = {
1428 .has_downtime_limit = true,
1429 .downtime_limit = value,
1430 };
1431
1432 qmp_migrate_set_parameters(&p, errp);
1433 }
1434
1435 bool migrate_release_ram(void)
1436 {
1437 MigrationState *s;
1438
1439 s = migrate_get_current();
1440
1441 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1442 }
1443
1444 bool migrate_postcopy_ram(void)
1445 {
1446 MigrationState *s;
1447
1448 s = migrate_get_current();
1449
1450 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1451 }
1452
1453 bool migrate_postcopy(void)
1454 {
1455 return migrate_postcopy_ram();
1456 }
1457
1458 bool migrate_auto_converge(void)
1459 {
1460 MigrationState *s;
1461
1462 s = migrate_get_current();
1463
1464 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1465 }
1466
1467 bool migrate_zero_blocks(void)
1468 {
1469 MigrationState *s;
1470
1471 s = migrate_get_current();
1472
1473 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1474 }
1475
1476 bool migrate_use_compression(void)
1477 {
1478 MigrationState *s;
1479
1480 s = migrate_get_current();
1481
1482 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
1483 }
1484
1485 int migrate_compress_level(void)
1486 {
1487 MigrationState *s;
1488
1489 s = migrate_get_current();
1490
1491 return s->parameters.compress_level;
1492 }
1493
1494 int migrate_compress_threads(void)
1495 {
1496 MigrationState *s;
1497
1498 s = migrate_get_current();
1499
1500 return s->parameters.compress_threads;
1501 }
1502
1503 int migrate_decompress_threads(void)
1504 {
1505 MigrationState *s;
1506
1507 s = migrate_get_current();
1508
1509 return s->parameters.decompress_threads;
1510 }
1511
1512 bool migrate_use_events(void)
1513 {
1514 MigrationState *s;
1515
1516 s = migrate_get_current();
1517
1518 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1519 }
1520
1521 bool migrate_use_multifd(void)
1522 {
1523 MigrationState *s;
1524
1525 s = migrate_get_current();
1526
1527 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_MULTIFD];
1528 }
1529
1530 int migrate_multifd_channels(void)
1531 {
1532 MigrationState *s;
1533
1534 s = migrate_get_current();
1535
1536 return s->parameters.x_multifd_channels;
1537 }
1538
1539 int migrate_multifd_page_count(void)
1540 {
1541 MigrationState *s;
1542
1543 s = migrate_get_current();
1544
1545 return s->parameters.x_multifd_page_count;
1546 }
1547
1548 int migrate_use_xbzrle(void)
1549 {
1550 MigrationState *s;
1551
1552 s = migrate_get_current();
1553
1554 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1555 }
1556
1557 int64_t migrate_xbzrle_cache_size(void)
1558 {
1559 MigrationState *s;
1560
1561 s = migrate_get_current();
1562
1563 return s->xbzrle_cache_size;
1564 }
1565
1566 bool migrate_use_block(void)
1567 {
1568 MigrationState *s;
1569
1570 s = migrate_get_current();
1571
1572 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
1573 }
1574
1575 bool migrate_use_return_path(void)
1576 {
1577 MigrationState *s;
1578
1579 s = migrate_get_current();
1580
1581 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
1582 }
1583
1584 bool migrate_use_block_incremental(void)
1585 {
1586 MigrationState *s;
1587
1588 s = migrate_get_current();
1589
1590 return s->parameters.block_incremental;
1591 }
1592
1593 /* migration thread support */
1594 /*
1595 * Something bad happened to the RP stream, mark an error
1596 * The caller shall print or trace something to indicate why
1597 */
1598 static void mark_source_rp_bad(MigrationState *s)
1599 {
1600 s->rp_state.error = true;
1601 }
1602
1603 static struct rp_cmd_args {
1604 ssize_t len; /* -1 = variable */
1605 const char *name;
1606 } rp_cmd_args[] = {
1607 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" },
1608 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" },
1609 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" },
1610 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" },
1611 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" },
1612 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" },
1613 };
1614
1615 /*
1616 * Process a request for pages received on the return path,
1617 * We're allowed to send more than requested (e.g. to round to our page size)
1618 * and we don't need to send pages that have already been sent.
1619 */
1620 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1621 ram_addr_t start, size_t len)
1622 {
1623 long our_host_ps = getpagesize();
1624
1625 trace_migrate_handle_rp_req_pages(rbname, start, len);
1626
1627 /*
1628 * Since we currently insist on matching page sizes, just sanity check
1629 * we're being asked for whole host pages.
1630 */
1631 if (start & (our_host_ps-1) ||
1632 (len & (our_host_ps-1))) {
1633 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1634 " len: %zd", __func__, start, len);
1635 mark_source_rp_bad(ms);
1636 return;
1637 }
1638
1639 if (ram_save_queue_pages(rbname, start, len)) {
1640 mark_source_rp_bad(ms);
1641 }
1642 }
1643
1644 /*
1645 * Handles messages sent on the return path towards the source VM
1646 *
1647 */
1648 static void *source_return_path_thread(void *opaque)
1649 {
1650 MigrationState *ms = opaque;
1651 QEMUFile *rp = ms->rp_state.from_dst_file;
1652 uint16_t header_len, header_type;
1653 uint8_t buf[512];
1654 uint32_t tmp32, sibling_error;
1655 ram_addr_t start = 0; /* =0 to silence warning */
1656 size_t len = 0, expected_len;
1657 int res;
1658
1659 trace_source_return_path_thread_entry();
1660 while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
1661 migration_is_setup_or_active(ms->state)) {
1662 trace_source_return_path_thread_loop_top();
1663 header_type = qemu_get_be16(rp);
1664 header_len = qemu_get_be16(rp);
1665
1666 if (header_type >= MIG_RP_MSG_MAX ||
1667 header_type == MIG_RP_MSG_INVALID) {
1668 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1669 header_type, header_len);
1670 mark_source_rp_bad(ms);
1671 goto out;
1672 }
1673
1674 if ((rp_cmd_args[header_type].len != -1 &&
1675 header_len != rp_cmd_args[header_type].len) ||
1676 header_len > sizeof(buf)) {
1677 error_report("RP: Received '%s' message (0x%04x) with"
1678 "incorrect length %d expecting %zu",
1679 rp_cmd_args[header_type].name, header_type, header_len,
1680 (size_t)rp_cmd_args[header_type].len);
1681 mark_source_rp_bad(ms);
1682 goto out;
1683 }
1684
1685 /* We know we've got a valid header by this point */
1686 res = qemu_get_buffer(rp, buf, header_len);
1687 if (res != header_len) {
1688 error_report("RP: Failed reading data for message 0x%04x"
1689 " read %d expected %d",
1690 header_type, res, header_len);
1691 mark_source_rp_bad(ms);
1692 goto out;
1693 }
1694
1695 /* OK, we have the message and the data */
1696 switch (header_type) {
1697 case MIG_RP_MSG_SHUT:
1698 sibling_error = ldl_be_p(buf);
1699 trace_source_return_path_thread_shut(sibling_error);
1700 if (sibling_error) {
1701 error_report("RP: Sibling indicated error %d", sibling_error);
1702 mark_source_rp_bad(ms);
1703 }
1704 /*
1705 * We'll let the main thread deal with closing the RP
1706 * we could do a shutdown(2) on it, but we're the only user
1707 * anyway, so there's nothing gained.
1708 */
1709 goto out;
1710
1711 case MIG_RP_MSG_PONG:
1712 tmp32 = ldl_be_p(buf);
1713 trace_source_return_path_thread_pong(tmp32);
1714 break;
1715
1716 case MIG_RP_MSG_REQ_PAGES:
1717 start = ldq_be_p(buf);
1718 len = ldl_be_p(buf + 8);
1719 migrate_handle_rp_req_pages(ms, NULL, start, len);
1720 break;
1721
1722 case MIG_RP_MSG_REQ_PAGES_ID:
1723 expected_len = 12 + 1; /* header + termination */
1724
1725 if (header_len >= expected_len) {
1726 start = ldq_be_p(buf);
1727 len = ldl_be_p(buf + 8);
1728 /* Now we expect an idstr */
1729 tmp32 = buf[12]; /* Length of the following idstr */
1730 buf[13 + tmp32] = '\0';
1731 expected_len += tmp32;
1732 }
1733 if (header_len != expected_len) {
1734 error_report("RP: Req_Page_id with length %d expecting %zd",
1735 header_len, expected_len);
1736 mark_source_rp_bad(ms);
1737 goto out;
1738 }
1739 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
1740 break;
1741
1742 default:
1743 break;
1744 }
1745 }
1746 if (qemu_file_get_error(rp)) {
1747 trace_source_return_path_thread_bad_end();
1748 mark_source_rp_bad(ms);
1749 }
1750
1751 trace_source_return_path_thread_end();
1752 out:
1753 ms->rp_state.from_dst_file = NULL;
1754 qemu_fclose(rp);
1755 return NULL;
1756 }
1757
1758 static int open_return_path_on_source(MigrationState *ms)
1759 {
1760
1761 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
1762 if (!ms->rp_state.from_dst_file) {
1763 return -1;
1764 }
1765
1766 trace_open_return_path_on_source();
1767 qemu_thread_create(&ms->rp_state.rp_thread, "return path",
1768 source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
1769
1770 trace_open_return_path_on_source_continue();
1771
1772 return 0;
1773 }
1774
1775 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1776 static int await_return_path_close_on_source(MigrationState *ms)
1777 {
1778 /*
1779 * If this is a normal exit then the destination will send a SHUT and the
1780 * rp_thread will exit, however if there's an error we need to cause
1781 * it to exit.
1782 */
1783 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
1784 /*
1785 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1786 * waiting for the destination.
1787 */
1788 qemu_file_shutdown(ms->rp_state.from_dst_file);
1789 mark_source_rp_bad(ms);
1790 }
1791 trace_await_return_path_close_on_source_joining();
1792 qemu_thread_join(&ms->rp_state.rp_thread);
1793 trace_await_return_path_close_on_source_close();
1794 return ms->rp_state.error;
1795 }
1796
1797 /*
1798 * Switch from normal iteration to postcopy
1799 * Returns non-0 on error
1800 */
1801 static int postcopy_start(MigrationState *ms, bool *old_vm_running)
1802 {
1803 int ret;
1804 QIOChannelBuffer *bioc;
1805 QEMUFile *fb;
1806 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1807 bool restart_block = false;
1808 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
1809 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1810
1811 trace_postcopy_start();
1812 qemu_mutex_lock_iothread();
1813 trace_postcopy_start_set_run();
1814
1815 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1816 *old_vm_running = runstate_is_running();
1817 global_state_store();
1818 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1819 if (ret < 0) {
1820 goto fail;
1821 }
1822
1823 ret = bdrv_inactivate_all();
1824 if (ret < 0) {
1825 goto fail;
1826 }
1827 restart_block = true;
1828
1829 /*
1830 * Cause any non-postcopiable, but iterative devices to
1831 * send out their final data.
1832 */
1833 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
1834
1835 /*
1836 * in Finish migrate and with the io-lock held everything should
1837 * be quiet, but we've potentially still got dirty pages and we
1838 * need to tell the destination to throw any pages it's already received
1839 * that are dirty
1840 */
1841 if (migrate_postcopy_ram()) {
1842 if (ram_postcopy_send_discard_bitmap(ms)) {
1843 error_report("postcopy send discard bitmap failed");
1844 goto fail;
1845 }
1846 }
1847
1848 /*
1849 * send rest of state - note things that are doing postcopy
1850 * will notice we're in POSTCOPY_ACTIVE and not actually
1851 * wrap their state up here
1852 */
1853 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
1854 if (migrate_postcopy_ram()) {
1855 /* Ping just for debugging, helps line traces up */
1856 qemu_savevm_send_ping(ms->to_dst_file, 2);
1857 }
1858
1859 /*
1860 * While loading the device state we may trigger page transfer
1861 * requests and the fd must be free to process those, and thus
1862 * the destination must read the whole device state off the fd before
1863 * it starts processing it. Unfortunately the ad-hoc migration format
1864 * doesn't allow the destination to know the size to read without fully
1865 * parsing it through each devices load-state code (especially the open
1866 * coded devices that use get/put).
1867 * So we wrap the device state up in a package with a length at the start;
1868 * to do this we use a qemu_buf to hold the whole of the device state.
1869 */
1870 bioc = qio_channel_buffer_new(4096);
1871 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
1872 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
1873 object_unref(OBJECT(bioc));
1874
1875 /*
1876 * Make sure the receiver can get incoming pages before we send the rest
1877 * of the state
1878 */
1879 qemu_savevm_send_postcopy_listen(fb);
1880
1881 qemu_savevm_state_complete_precopy(fb, false, false);
1882 if (migrate_postcopy_ram()) {
1883 qemu_savevm_send_ping(fb, 3);
1884 }
1885
1886 qemu_savevm_send_postcopy_run(fb);
1887
1888 /* <><> end of stuff going into the package */
1889
1890 /* Last point of recovery; as soon as we send the package the destination
1891 * can open devices and potentially start running.
1892 * Lets just check again we've not got any errors.
1893 */
1894 ret = qemu_file_get_error(ms->to_dst_file);
1895 if (ret) {
1896 error_report("postcopy_start: Migration stream errored (pre package)");
1897 goto fail_closefb;
1898 }
1899
1900 restart_block = false;
1901
1902 /* Now send that blob */
1903 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
1904 goto fail_closefb;
1905 }
1906 qemu_fclose(fb);
1907
1908 /* Send a notify to give a chance for anything that needs to happen
1909 * at the transition to postcopy and after the device state; in particular
1910 * spice needs to trigger a transition now
1911 */
1912 ms->postcopy_after_devices = true;
1913 notifier_list_notify(&migration_state_notifiers, ms);
1914
1915 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
1916
1917 qemu_mutex_unlock_iothread();
1918
1919 if (migrate_postcopy_ram()) {
1920 /*
1921 * Although this ping is just for debug, it could potentially be
1922 * used for getting a better measurement of downtime at the source.
1923 */
1924 qemu_savevm_send_ping(ms->to_dst_file, 4);
1925 }
1926
1927 if (migrate_release_ram()) {
1928 ram_postcopy_migrated_memory_release(ms);
1929 }
1930
1931 ret = qemu_file_get_error(ms->to_dst_file);
1932 if (ret) {
1933 error_report("postcopy_start: Migration stream errored");
1934 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1935 MIGRATION_STATUS_FAILED);
1936 }
1937
1938 return ret;
1939
1940 fail_closefb:
1941 qemu_fclose(fb);
1942 fail:
1943 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1944 MIGRATION_STATUS_FAILED);
1945 if (restart_block) {
1946 /* A failure happened early enough that we know the destination hasn't
1947 * accessed block devices, so we're safe to recover.
1948 */
1949 Error *local_err = NULL;
1950
1951 bdrv_invalidate_cache_all(&local_err);
1952 if (local_err) {
1953 error_report_err(local_err);
1954 }
1955 }
1956 qemu_mutex_unlock_iothread();
1957 return -1;
1958 }
1959
1960 /**
1961 * migration_completion: Used by migration_thread when there's not much left.
1962 * The caller 'breaks' the loop when this returns.
1963 *
1964 * @s: Current migration state
1965 * @current_active_state: The migration state we expect to be in
1966 * @*old_vm_running: Pointer to old_vm_running flag
1967 * @*start_time: Pointer to time to update
1968 */
1969 static void migration_completion(MigrationState *s, int current_active_state,
1970 bool *old_vm_running,
1971 int64_t *start_time)
1972 {
1973 int ret;
1974
1975 if (s->state == MIGRATION_STATUS_ACTIVE) {
1976 qemu_mutex_lock_iothread();
1977 *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1978 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1979 *old_vm_running = runstate_is_running();
1980 ret = global_state_store();
1981
1982 if (!ret) {
1983 bool inactivate = !migrate_colo_enabled();
1984 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1985 if (ret >= 0) {
1986 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
1987 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
1988 inactivate);
1989 }
1990 if (inactivate && ret >= 0) {
1991 s->block_inactive = true;
1992 }
1993 }
1994 qemu_mutex_unlock_iothread();
1995
1996 if (ret < 0) {
1997 goto fail;
1998 }
1999 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2000 trace_migration_completion_postcopy_end();
2001
2002 qemu_savevm_state_complete_postcopy(s->to_dst_file);
2003 trace_migration_completion_postcopy_end_after_complete();
2004 }
2005
2006 /*
2007 * If rp was opened we must clean up the thread before
2008 * cleaning everything else up (since if there are no failures
2009 * it will wait for the destination to send it's status in
2010 * a SHUT command).
2011 */
2012 if (s->rp_state.from_dst_file) {
2013 int rp_error;
2014 trace_migration_return_path_end_before();
2015 rp_error = await_return_path_close_on_source(s);
2016 trace_migration_return_path_end_after(rp_error);
2017 if (rp_error) {
2018 goto fail_invalidate;
2019 }
2020 }
2021
2022 if (qemu_file_get_error(s->to_dst_file)) {
2023 trace_migration_completion_file_err();
2024 goto fail_invalidate;
2025 }
2026
2027 if (!migrate_colo_enabled()) {
2028 migrate_set_state(&s->state, current_active_state,
2029 MIGRATION_STATUS_COMPLETED);
2030 }
2031
2032 return;
2033
2034 fail_invalidate:
2035 /* If not doing postcopy, vm_start() will be called: let's regain
2036 * control on images.
2037 */
2038 if (s->state == MIGRATION_STATUS_ACTIVE) {
2039 Error *local_err = NULL;
2040
2041 qemu_mutex_lock_iothread();
2042 bdrv_invalidate_cache_all(&local_err);
2043 if (local_err) {
2044 error_report_err(local_err);
2045 } else {
2046 s->block_inactive = false;
2047 }
2048 qemu_mutex_unlock_iothread();
2049 }
2050
2051 fail:
2052 migrate_set_state(&s->state, current_active_state,
2053 MIGRATION_STATUS_FAILED);
2054 }
2055
2056 bool migrate_colo_enabled(void)
2057 {
2058 MigrationState *s = migrate_get_current();
2059 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2060 }
2061
2062 /*
2063 * Master migration thread on the source VM.
2064 * It drives the migration and pumps the data down the outgoing channel.
2065 */
2066 static void *migration_thread(void *opaque)
2067 {
2068 MigrationState *s = opaque;
2069 /* Used by the bandwidth calcs, updated later */
2070 int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2071 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
2072 int64_t initial_bytes = 0;
2073 /*
2074 * The final stage happens when the remaining data is smaller than
2075 * this threshold; it's calculated from the requested downtime and
2076 * measured bandwidth
2077 */
2078 int64_t threshold_size = 0;
2079 int64_t start_time = initial_time;
2080 int64_t end_time;
2081 bool old_vm_running = false;
2082 bool entered_postcopy = false;
2083 /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
2084 enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE;
2085 bool enable_colo = migrate_colo_enabled();
2086
2087 rcu_register_thread();
2088
2089 qemu_savevm_state_header(s->to_dst_file);
2090
2091 /*
2092 * If we opened the return path, we need to make sure dst has it
2093 * opened as well.
2094 */
2095 if (s->rp_state.from_dst_file) {
2096 /* Now tell the dest that it should open its end so it can reply */
2097 qemu_savevm_send_open_return_path(s->to_dst_file);
2098
2099 /* And do a ping that will make stuff easier to debug */
2100 qemu_savevm_send_ping(s->to_dst_file, 1);
2101 }
2102
2103 if (migrate_postcopy()) {
2104 /*
2105 * Tell the destination that we *might* want to do postcopy later;
2106 * if the other end can't do postcopy it should fail now, nice and
2107 * early.
2108 */
2109 qemu_savevm_send_postcopy_advise(s->to_dst_file);
2110 }
2111
2112 qemu_savevm_state_setup(s->to_dst_file);
2113
2114 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
2115 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2116 MIGRATION_STATUS_ACTIVE);
2117
2118 trace_migration_thread_setup_complete();
2119
2120 while (s->state == MIGRATION_STATUS_ACTIVE ||
2121 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2122 int64_t current_time;
2123 uint64_t pending_size;
2124
2125 if (!qemu_file_rate_limit(s->to_dst_file)) {
2126 uint64_t pend_post, pend_nonpost;
2127
2128 qemu_savevm_state_pending(s->to_dst_file, threshold_size,
2129 &pend_nonpost, &pend_post);
2130 pending_size = pend_nonpost + pend_post;
2131 trace_migrate_pending(pending_size, threshold_size,
2132 pend_post, pend_nonpost);
2133 if (pending_size && pending_size >= threshold_size) {
2134 /* Still a significant amount to transfer */
2135
2136 if (migrate_postcopy() &&
2137 s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE &&
2138 pend_nonpost <= threshold_size &&
2139 atomic_read(&s->start_postcopy)) {
2140
2141 if (!postcopy_start(s, &old_vm_running)) {
2142 current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE;
2143 entered_postcopy = true;
2144 }
2145
2146 continue;
2147 }
2148 /* Just another iteration step */
2149 qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy);
2150 } else {
2151 trace_migration_thread_low_pending(pending_size);
2152 migration_completion(s, current_active_state,
2153 &old_vm_running, &start_time);
2154 break;
2155 }
2156 }
2157
2158 if (qemu_file_get_error(s->to_dst_file)) {
2159 migrate_set_state(&s->state, current_active_state,
2160 MIGRATION_STATUS_FAILED);
2161 trace_migration_thread_file_err();
2162 break;
2163 }
2164 current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2165 if (current_time >= initial_time + BUFFER_DELAY) {
2166 uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) -
2167 initial_bytes;
2168 uint64_t time_spent = current_time - initial_time;
2169 double bandwidth = (double)transferred_bytes / time_spent;
2170 threshold_size = bandwidth * s->parameters.downtime_limit;
2171
2172 s->mbps = (((double) transferred_bytes * 8.0) /
2173 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
2174
2175 trace_migrate_transferred(transferred_bytes, time_spent,
2176 bandwidth, threshold_size);
2177 /* if we haven't sent anything, we don't want to recalculate
2178 10000 is a small enough number for our purposes */
2179 if (ram_counters.dirty_pages_rate && transferred_bytes > 10000) {
2180 s->expected_downtime = ram_counters.dirty_pages_rate *
2181 qemu_target_page_size() / bandwidth;
2182 }
2183
2184 qemu_file_reset_rate_limit(s->to_dst_file);
2185 initial_time = current_time;
2186 initial_bytes = qemu_ftell(s->to_dst_file);
2187 }
2188 if (qemu_file_rate_limit(s->to_dst_file)) {
2189 /* usleep expects microseconds */
2190 g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
2191 }
2192 }
2193
2194 trace_migration_thread_after_loop();
2195 /* If we enabled cpu throttling for auto-converge, turn it off. */
2196 cpu_throttle_stop();
2197 end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2198
2199 qemu_mutex_lock_iothread();
2200 /*
2201 * The resource has been allocated by migration will be reused in COLO
2202 * process, so don't release them.
2203 */
2204 if (!enable_colo) {
2205 qemu_savevm_state_cleanup();
2206 }
2207 if (s->state == MIGRATION_STATUS_COMPLETED) {
2208 uint64_t transferred_bytes = qemu_ftell(s->to_dst_file);
2209 s->total_time = end_time - s->total_time;
2210 if (!entered_postcopy) {
2211 s->downtime = end_time - start_time;
2212 }
2213 if (s->total_time) {
2214 s->mbps = (((double) transferred_bytes * 8.0) /
2215 ((double) s->total_time)) / 1000;
2216 }
2217 runstate_set(RUN_STATE_POSTMIGRATE);
2218 } else {
2219 if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) {
2220 migrate_start_colo_process(s);
2221 qemu_savevm_state_cleanup();
2222 /*
2223 * Fixme: we will run VM in COLO no matter its old running state.
2224 * After exited COLO, we will keep running.
2225 */
2226 old_vm_running = true;
2227 }
2228 if (old_vm_running && !entered_postcopy) {
2229 vm_start();
2230 } else {
2231 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2232 runstate_set(RUN_STATE_POSTMIGRATE);
2233 }
2234 }
2235 }
2236 qemu_bh_schedule(s->cleanup_bh);
2237 qemu_mutex_unlock_iothread();
2238
2239 rcu_unregister_thread();
2240 return NULL;
2241 }
2242
2243 void migrate_fd_connect(MigrationState *s)
2244 {
2245 s->expected_downtime = s->parameters.downtime_limit;
2246 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
2247
2248 qemu_file_set_blocking(s->to_dst_file, true);
2249 qemu_file_set_rate_limit(s->to_dst_file,
2250 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
2251
2252 /* Notify before starting migration thread */
2253 notifier_list_notify(&migration_state_notifiers, s);
2254
2255 /*
2256 * Open the return path. For postcopy, it is used exclusively. For
2257 * precopy, only if user specified "return-path" capability would
2258 * QEMU uses the return path.
2259 */
2260 if (migrate_postcopy_ram() || migrate_use_return_path()) {
2261 if (open_return_path_on_source(s)) {
2262 error_report("Unable to open return-path for postcopy");
2263 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2264 MIGRATION_STATUS_FAILED);
2265 migrate_fd_cleanup(s);
2266 return;
2267 }
2268 }
2269
2270 if (multifd_save_setup() != 0) {
2271 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2272 MIGRATION_STATUS_FAILED);
2273 migrate_fd_cleanup(s);
2274 return;
2275 }
2276 qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
2277 QEMU_THREAD_JOINABLE);
2278 s->migration_thread_running = true;
2279 }
2280
2281 void migration_global_dump(Monitor *mon)
2282 {
2283 MigrationState *ms = migrate_get_current();
2284
2285 monitor_printf(mon, "globals: store-global-state=%d, only_migratable=%d, "
2286 "send-configuration=%d, send-section-footer=%d\n",
2287 ms->store_global_state, ms->only_migratable,
2288 ms->send_configuration, ms->send_section_footer);
2289 }
2290
2291 #define DEFINE_PROP_MIG_CAP(name, x) \
2292 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
2293
2294 static Property migration_properties[] = {
2295 DEFINE_PROP_BOOL("store-global-state", MigrationState,
2296 store_global_state, true),
2297 DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false),
2298 DEFINE_PROP_BOOL("send-configuration", MigrationState,
2299 send_configuration, true),
2300 DEFINE_PROP_BOOL("send-section-footer", MigrationState,
2301 send_section_footer, true),
2302
2303 /* Migration parameters */
2304 DEFINE_PROP_INT64("x-compress-level", MigrationState,
2305 parameters.compress_level,
2306 DEFAULT_MIGRATE_COMPRESS_LEVEL),
2307 DEFINE_PROP_INT64("x-compress-threads", MigrationState,
2308 parameters.compress_threads,
2309 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
2310 DEFINE_PROP_INT64("x-decompress-threads", MigrationState,
2311 parameters.decompress_threads,
2312 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
2313 DEFINE_PROP_INT64("x-cpu-throttle-initial", MigrationState,
2314 parameters.cpu_throttle_initial,
2315 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
2316 DEFINE_PROP_INT64("x-cpu-throttle-increment", MigrationState,
2317 parameters.cpu_throttle_increment,
2318 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
2319 DEFINE_PROP_INT64("x-max-bandwidth", MigrationState,
2320 parameters.max_bandwidth, MAX_THROTTLE),
2321 DEFINE_PROP_INT64("x-downtime-limit", MigrationState,
2322 parameters.downtime_limit,
2323 DEFAULT_MIGRATE_SET_DOWNTIME),
2324 DEFINE_PROP_INT64("x-checkpoint-delay", MigrationState,
2325 parameters.x_checkpoint_delay,
2326 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
2327 DEFINE_PROP_INT64("x-multifd-channels", MigrationState,
2328 parameters.x_multifd_channels,
2329 DEFAULT_MIGRATE_MULTIFD_CHANNELS),
2330 DEFINE_PROP_INT64("x-multifd-page-count", MigrationState,
2331 parameters.x_multifd_page_count,
2332 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT),
2333
2334 /* Migration capabilities */
2335 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
2336 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
2337 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
2338 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
2339 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
2340 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
2341 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
2342 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
2343 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
2344 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
2345 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
2346 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD),
2347
2348 DEFINE_PROP_END_OF_LIST(),
2349 };
2350
2351 static void migration_class_init(ObjectClass *klass, void *data)
2352 {
2353 DeviceClass *dc = DEVICE_CLASS(klass);
2354
2355 dc->user_creatable = false;
2356 dc->props = migration_properties;
2357 }
2358
2359 static void migration_instance_finalize(Object *obj)
2360 {
2361 MigrationState *ms = MIGRATION_OBJ(obj);
2362 MigrationParameters *params = &ms->parameters;
2363
2364 g_free(params->tls_hostname);
2365 g_free(params->tls_creds);
2366 }
2367
2368 static void migration_instance_init(Object *obj)
2369 {
2370 MigrationState *ms = MIGRATION_OBJ(obj);
2371 MigrationParameters *params = &ms->parameters;
2372
2373 ms->state = MIGRATION_STATUS_NONE;
2374 ms->xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE;
2375 ms->mbps = -1;
2376
2377 params->tls_hostname = g_strdup("");
2378 params->tls_creds = g_strdup("");
2379
2380 /* Set has_* up only for parameter checks */
2381 params->has_compress_level = true;
2382 params->has_compress_threads = true;
2383 params->has_decompress_threads = true;
2384 params->has_cpu_throttle_initial = true;
2385 params->has_cpu_throttle_increment = true;
2386 params->has_max_bandwidth = true;
2387 params->has_downtime_limit = true;
2388 params->has_x_checkpoint_delay = true;
2389 params->has_block_incremental = true;
2390 params->has_x_multifd_channels = true;
2391 params->has_x_multifd_page_count = true;
2392 }
2393
2394 /*
2395 * Return true if check pass, false otherwise. Error will be put
2396 * inside errp if provided.
2397 */
2398 static bool migration_object_check(MigrationState *ms, Error **errp)
2399 {
2400 MigrationCapabilityStatusList *head = NULL;
2401 /* Assuming all off */
2402 bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
2403 int i;
2404
2405 if (!migrate_params_check(&ms->parameters, errp)) {
2406 return false;
2407 }
2408
2409 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
2410 if (ms->enabled_capabilities[i]) {
2411 head = migrate_cap_add(head, i, true);
2412 }
2413 }
2414
2415 ret = migrate_caps_check(cap_list, head, errp);
2416
2417 /* It works with head == NULL */
2418 qapi_free_MigrationCapabilityStatusList(head);
2419
2420 return ret;
2421 }
2422
2423 static const TypeInfo migration_type = {
2424 .name = TYPE_MIGRATION,
2425 /*
2426 * NOTE: TYPE_MIGRATION is not really a device, as the object is
2427 * not created using qdev_create(), it is not attached to the qdev
2428 * device tree, and it is never realized.
2429 *
2430 * TODO: Make this TYPE_OBJECT once QOM provides something like
2431 * TYPE_DEVICE's "-global" properties.
2432 */
2433 .parent = TYPE_DEVICE,
2434 .class_init = migration_class_init,
2435 .class_size = sizeof(MigrationClass),
2436 .instance_size = sizeof(MigrationState),
2437 .instance_init = migration_instance_init,
2438 .instance_finalize = migration_instance_finalize,
2439 };
2440
2441 static void register_migration_types(void)
2442 {
2443 type_register_static(&migration_type);
2444 }
2445
2446 type_init(register_migration_types);
AR7 emulation for QEMU