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 "qemu/main-loop.h" 20 #include "migration/migration.h" 21 #include "migration/qemu-file.h" 22 #include "sysemu/sysemu.h" 23 #include "block/block.h" 24 #include "qapi/qmp/qerror.h" 25 #include "qapi/util.h" 26 #include "qemu/sockets.h" 27 #include "qemu/rcu.h" 28 #include "migration/block.h" 29 #include "migration/postcopy-ram.h" 30 #include "qemu/thread.h" 31 #include "qmp-commands.h" 32 #include "trace.h" 33 #include "qapi-event.h" 34 #include "qom/cpu.h" 35 #include "exec/memory.h" 36 #include "exec/address-spaces.h" 37 #include "io/channel-buffer.h" 38 #include "io/channel-tls.h" 39 #include "migration/colo.h" 40 41 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */ 42 43 /* Amount of time to allocate to each "chunk" of bandwidth-throttled 44 * data. */ 45 #define BUFFER_DELAY 100 46 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY) 47 48 /* Time in milliseconds we are allowed to stop the source, 49 * for sending the last part */ 50 #define DEFAULT_MIGRATE_SET_DOWNTIME 300 51 52 /* Maximum migrate downtime set to 2000 seconds */ 53 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000 54 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000) 55 56 /* Default compression thread count */ 57 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8 58 /* Default decompression thread count, usually decompression is at 59 * least 4 times as fast as compression.*/ 60 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2 61 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */ 62 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1 63 /* Define default autoconverge cpu throttle migration parameters */ 64 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20 65 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10 66 67 /* Migration XBZRLE default cache size */ 68 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024) 69 70 /* The delay time (in ms) between two COLO checkpoints 71 * Note: Please change this default value to 10000 when we support hybrid mode. 72 */ 73 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200 74 75 static NotifierList migration_state_notifiers = 76 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers); 77 78 static bool deferred_incoming; 79 80 /* 81 * Current state of incoming postcopy; note this is not part of 82 * MigrationIncomingState since it's state is used during cleanup 83 * at the end as MIS is being freed. 84 */ 85 static PostcopyState incoming_postcopy_state; 86 87 /* When we add fault tolerance, we could have several 88 migrations at once. For now we don't need to add 89 dynamic creation of migration */ 90 91 /* For outgoing */ 92 MigrationState *migrate_get_current(void) 93 { 94 static bool once; 95 static MigrationState current_migration = { 96 .state = MIGRATION_STATUS_NONE, 97 .xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE, 98 .mbps = -1, 99 .parameters = { 100 .compress_level = DEFAULT_MIGRATE_COMPRESS_LEVEL, 101 .compress_threads = DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT, 102 .decompress_threads = DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT, 103 .cpu_throttle_initial = DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL, 104 .cpu_throttle_increment = DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT, 105 .max_bandwidth = MAX_THROTTLE, 106 .downtime_limit = DEFAULT_MIGRATE_SET_DOWNTIME, 107 .x_checkpoint_delay = DEFAULT_MIGRATE_X_CHECKPOINT_DELAY, 108 }, 109 }; 110 111 if (!once) { 112 qemu_mutex_init(¤t_migration.src_page_req_mutex); 113 once = true; 114 } 115 return ¤t_migration; 116 } 117 118 MigrationIncomingState *migration_incoming_get_current(void) 119 { 120 static bool once; 121 static MigrationIncomingState mis_current; 122 123 if (!once) { 124 mis_current.state = MIGRATION_STATUS_NONE; 125 memset(&mis_current, 0, sizeof(MigrationIncomingState)); 126 QLIST_INIT(&mis_current.loadvm_handlers); 127 qemu_mutex_init(&mis_current.rp_mutex); 128 qemu_event_init(&mis_current.main_thread_load_event, false); 129 once = true; 130 } 131 return &mis_current; 132 } 133 134 void migration_incoming_state_destroy(void) 135 { 136 struct MigrationIncomingState *mis = migration_incoming_get_current(); 137 138 qemu_event_destroy(&mis->main_thread_load_event); 139 loadvm_free_handlers(mis); 140 } 141 142 143 typedef struct { 144 bool optional; 145 uint32_t size; 146 uint8_t runstate[100]; 147 RunState state; 148 bool received; 149 } GlobalState; 150 151 static GlobalState global_state; 152 153 int global_state_store(void) 154 { 155 if (!runstate_store((char *)global_state.runstate, 156 sizeof(global_state.runstate))) { 157 error_report("runstate name too big: %s", global_state.runstate); 158 trace_migrate_state_too_big(); 159 return -EINVAL; 160 } 161 return 0; 162 } 163 164 void global_state_store_running(void) 165 { 166 const char *state = RunState_lookup[RUN_STATE_RUNNING]; 167 strncpy((char *)global_state.runstate, 168 state, sizeof(global_state.runstate)); 169 } 170 171 static bool global_state_received(void) 172 { 173 return global_state.received; 174 } 175 176 static RunState global_state_get_runstate(void) 177 { 178 return global_state.state; 179 } 180 181 void global_state_set_optional(void) 182 { 183 global_state.optional = true; 184 } 185 186 static bool global_state_needed(void *opaque) 187 { 188 GlobalState *s = opaque; 189 char *runstate = (char *)s->runstate; 190 191 /* If it is not optional, it is mandatory */ 192 193 if (s->optional == false) { 194 return true; 195 } 196 197 /* If state is running or paused, it is not needed */ 198 199 if (strcmp(runstate, "running") == 0 || 200 strcmp(runstate, "paused") == 0) { 201 return false; 202 } 203 204 /* for any other state it is needed */ 205 return true; 206 } 207 208 static int global_state_post_load(void *opaque, int version_id) 209 { 210 GlobalState *s = opaque; 211 Error *local_err = NULL; 212 int r; 213 char *runstate = (char *)s->runstate; 214 215 s->received = true; 216 trace_migrate_global_state_post_load(runstate); 217 218 r = qapi_enum_parse(RunState_lookup, runstate, RUN_STATE__MAX, 219 -1, &local_err); 220 221 if (r == -1) { 222 if (local_err) { 223 error_report_err(local_err); 224 } 225 return -EINVAL; 226 } 227 s->state = r; 228 229 return 0; 230 } 231 232 static void global_state_pre_save(void *opaque) 233 { 234 GlobalState *s = opaque; 235 236 trace_migrate_global_state_pre_save((char *)s->runstate); 237 s->size = strlen((char *)s->runstate) + 1; 238 } 239 240 static const VMStateDescription vmstate_globalstate = { 241 .name = "globalstate", 242 .version_id = 1, 243 .minimum_version_id = 1, 244 .post_load = global_state_post_load, 245 .pre_save = global_state_pre_save, 246 .needed = global_state_needed, 247 .fields = (VMStateField[]) { 248 VMSTATE_UINT32(size, GlobalState), 249 VMSTATE_BUFFER(runstate, GlobalState), 250 VMSTATE_END_OF_LIST() 251 }, 252 }; 253 254 void register_global_state(void) 255 { 256 /* We would use it independently that we receive it */ 257 strcpy((char *)&global_state.runstate, ""); 258 global_state.received = false; 259 vmstate_register(NULL, 0, &vmstate_globalstate, &global_state); 260 } 261 262 static void migrate_generate_event(int new_state) 263 { 264 if (migrate_use_events()) { 265 qapi_event_send_migration(new_state, &error_abort); 266 } 267 } 268 269 /* 270 * Called on -incoming with a defer: uri. 271 * The migration can be started later after any parameters have been 272 * changed. 273 */ 274 static void deferred_incoming_migration(Error **errp) 275 { 276 if (deferred_incoming) { 277 error_setg(errp, "Incoming migration already deferred"); 278 } 279 deferred_incoming = true; 280 } 281 282 /* Request a range of pages from the source VM at the given 283 * start address. 284 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same 285 * as the last request (a name must have been given previously) 286 * Start: Address offset within the RB 287 * Len: Length in bytes required - must be a multiple of pagesize 288 */ 289 void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname, 290 ram_addr_t start, size_t len) 291 { 292 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */ 293 size_t msglen = 12; /* start + len */ 294 295 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start); 296 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len); 297 298 if (rbname) { 299 int rbname_len = strlen(rbname); 300 assert(rbname_len < 256); 301 302 bufc[msglen++] = rbname_len; 303 memcpy(bufc + msglen, rbname, rbname_len); 304 msglen += rbname_len; 305 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc); 306 } else { 307 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc); 308 } 309 } 310 311 void qemu_start_incoming_migration(const char *uri, Error **errp) 312 { 313 const char *p; 314 315 qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort); 316 if (!strcmp(uri, "defer")) { 317 deferred_incoming_migration(errp); 318 } else if (strstart(uri, "tcp:", &p)) { 319 tcp_start_incoming_migration(p, errp); 320 #ifdef CONFIG_RDMA 321 } else if (strstart(uri, "rdma:", &p)) { 322 rdma_start_incoming_migration(p, errp); 323 #endif 324 } else if (strstart(uri, "exec:", &p)) { 325 exec_start_incoming_migration(p, errp); 326 } else if (strstart(uri, "unix:", &p)) { 327 unix_start_incoming_migration(p, errp); 328 } else if (strstart(uri, "fd:", &p)) { 329 fd_start_incoming_migration(p, errp); 330 } else { 331 error_setg(errp, "unknown migration protocol: %s", uri); 332 } 333 } 334 335 static void process_incoming_migration_bh(void *opaque) 336 { 337 Error *local_err = NULL; 338 MigrationIncomingState *mis = opaque; 339 340 /* Make sure all file formats flush their mutable metadata */ 341 bdrv_invalidate_cache_all(&local_err); 342 if (local_err) { 343 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, 344 MIGRATION_STATUS_FAILED); 345 error_report_err(local_err); 346 migrate_decompress_threads_join(); 347 exit(EXIT_FAILURE); 348 } 349 350 /* 351 * This must happen after all error conditions are dealt with and 352 * we're sure the VM is going to be running on this host. 353 */ 354 qemu_announce_self(); 355 356 /* If global state section was not received or we are in running 357 state, we need to obey autostart. Any other state is set with 358 runstate_set. */ 359 360 if (!global_state_received() || 361 global_state_get_runstate() == RUN_STATE_RUNNING) { 362 if (autostart) { 363 vm_start(); 364 } else { 365 runstate_set(RUN_STATE_PAUSED); 366 } 367 } else { 368 runstate_set(global_state_get_runstate()); 369 } 370 migrate_decompress_threads_join(); 371 /* 372 * This must happen after any state changes since as soon as an external 373 * observer sees this event they might start to prod at the VM assuming 374 * it's ready to use. 375 */ 376 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, 377 MIGRATION_STATUS_COMPLETED); 378 qemu_bh_delete(mis->bh); 379 migration_incoming_state_destroy(); 380 } 381 382 static void process_incoming_migration_co(void *opaque) 383 { 384 QEMUFile *f = opaque; 385 MigrationIncomingState *mis = migration_incoming_get_current(); 386 PostcopyState ps; 387 int ret; 388 389 mis->from_src_file = f; 390 mis->largest_page_size = qemu_ram_pagesize_largest(); 391 postcopy_state_set(POSTCOPY_INCOMING_NONE); 392 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE, 393 MIGRATION_STATUS_ACTIVE); 394 ret = qemu_loadvm_state(f); 395 396 ps = postcopy_state_get(); 397 trace_process_incoming_migration_co_end(ret, ps); 398 if (ps != POSTCOPY_INCOMING_NONE) { 399 if (ps == POSTCOPY_INCOMING_ADVISE) { 400 /* 401 * Where a migration had postcopy enabled (and thus went to advise) 402 * but managed to complete within the precopy period, we can use 403 * the normal exit. 404 */ 405 postcopy_ram_incoming_cleanup(mis); 406 } else if (ret >= 0) { 407 /* 408 * Postcopy was started, cleanup should happen at the end of the 409 * postcopy thread. 410 */ 411 trace_process_incoming_migration_co_postcopy_end_main(); 412 return; 413 } 414 /* Else if something went wrong then just fall out of the normal exit */ 415 } 416 417 /* we get COLO info, and know if we are in COLO mode */ 418 if (!ret && migration_incoming_enable_colo()) { 419 mis->migration_incoming_co = qemu_coroutine_self(); 420 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming", 421 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE); 422 mis->have_colo_incoming_thread = true; 423 qemu_coroutine_yield(); 424 425 /* Wait checkpoint incoming thread exit before free resource */ 426 qemu_thread_join(&mis->colo_incoming_thread); 427 } 428 429 qemu_fclose(f); 430 free_xbzrle_decoded_buf(); 431 432 if (ret < 0) { 433 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, 434 MIGRATION_STATUS_FAILED); 435 error_report("load of migration failed: %s", strerror(-ret)); 436 migrate_decompress_threads_join(); 437 exit(EXIT_FAILURE); 438 } 439 440 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis); 441 qemu_bh_schedule(mis->bh); 442 } 443 444 void migration_fd_process_incoming(QEMUFile *f) 445 { 446 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, f); 447 448 migrate_decompress_threads_create(); 449 qemu_file_set_blocking(f, false); 450 qemu_coroutine_enter(co); 451 } 452 453 454 void migration_channel_process_incoming(MigrationState *s, 455 QIOChannel *ioc) 456 { 457 trace_migration_set_incoming_channel( 458 ioc, object_get_typename(OBJECT(ioc))); 459 460 if (s->parameters.tls_creds && 461 !object_dynamic_cast(OBJECT(ioc), 462 TYPE_QIO_CHANNEL_TLS)) { 463 Error *local_err = NULL; 464 migration_tls_channel_process_incoming(s, ioc, &local_err); 465 if (local_err) { 466 error_report_err(local_err); 467 } 468 } else { 469 QEMUFile *f = qemu_fopen_channel_input(ioc); 470 migration_fd_process_incoming(f); 471 } 472 } 473 474 475 void migration_channel_connect(MigrationState *s, 476 QIOChannel *ioc, 477 const char *hostname) 478 { 479 trace_migration_set_outgoing_channel( 480 ioc, object_get_typename(OBJECT(ioc)), hostname); 481 482 if (s->parameters.tls_creds && 483 !object_dynamic_cast(OBJECT(ioc), 484 TYPE_QIO_CHANNEL_TLS)) { 485 Error *local_err = NULL; 486 migration_tls_channel_connect(s, ioc, hostname, &local_err); 487 if (local_err) { 488 migrate_fd_error(s, local_err); 489 error_free(local_err); 490 } 491 } else { 492 QEMUFile *f = qemu_fopen_channel_output(ioc); 493 494 s->to_dst_file = f; 495 496 migrate_fd_connect(s); 497 } 498 } 499 500 501 /* 502 * Send a message on the return channel back to the source 503 * of the migration. 504 */ 505 void migrate_send_rp_message(MigrationIncomingState *mis, 506 enum mig_rp_message_type message_type, 507 uint16_t len, void *data) 508 { 509 trace_migrate_send_rp_message((int)message_type, len); 510 qemu_mutex_lock(&mis->rp_mutex); 511 qemu_put_be16(mis->to_src_file, (unsigned int)message_type); 512 qemu_put_be16(mis->to_src_file, len); 513 qemu_put_buffer(mis->to_src_file, data, len); 514 qemu_fflush(mis->to_src_file); 515 qemu_mutex_unlock(&mis->rp_mutex); 516 } 517 518 /* 519 * Send a 'SHUT' message on the return channel with the given value 520 * to indicate that we've finished with the RP. Non-0 value indicates 521 * error. 522 */ 523 void migrate_send_rp_shut(MigrationIncomingState *mis, 524 uint32_t value) 525 { 526 uint32_t buf; 527 528 buf = cpu_to_be32(value); 529 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf); 530 } 531 532 /* 533 * Send a 'PONG' message on the return channel with the given value 534 * (normally in response to a 'PING') 535 */ 536 void migrate_send_rp_pong(MigrationIncomingState *mis, 537 uint32_t value) 538 { 539 uint32_t buf; 540 541 buf = cpu_to_be32(value); 542 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf); 543 } 544 545 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp) 546 { 547 MigrationCapabilityStatusList *head = NULL; 548 MigrationCapabilityStatusList *caps; 549 MigrationState *s = migrate_get_current(); 550 int i; 551 552 caps = NULL; /* silence compiler warning */ 553 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { 554 if (i == MIGRATION_CAPABILITY_X_COLO && !colo_supported()) { 555 continue; 556 } 557 if (head == NULL) { 558 head = g_malloc0(sizeof(*caps)); 559 caps = head; 560 } else { 561 caps->next = g_malloc0(sizeof(*caps)); 562 caps = caps->next; 563 } 564 caps->value = 565 g_malloc(sizeof(*caps->value)); 566 caps->value->capability = i; 567 caps->value->state = s->enabled_capabilities[i]; 568 } 569 570 return head; 571 } 572 573 MigrationParameters *qmp_query_migrate_parameters(Error **errp) 574 { 575 MigrationParameters *params; 576 MigrationState *s = migrate_get_current(); 577 578 params = g_malloc0(sizeof(*params)); 579 params->has_compress_level = true; 580 params->compress_level = s->parameters.compress_level; 581 params->has_compress_threads = true; 582 params->compress_threads = s->parameters.compress_threads; 583 params->has_decompress_threads = true; 584 params->decompress_threads = s->parameters.decompress_threads; 585 params->has_cpu_throttle_initial = true; 586 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial; 587 params->has_cpu_throttle_increment = true; 588 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment; 589 params->has_tls_creds = !!s->parameters.tls_creds; 590 params->tls_creds = g_strdup(s->parameters.tls_creds); 591 params->has_tls_hostname = !!s->parameters.tls_hostname; 592 params->tls_hostname = g_strdup(s->parameters.tls_hostname); 593 params->has_max_bandwidth = true; 594 params->max_bandwidth = s->parameters.max_bandwidth; 595 params->has_downtime_limit = true; 596 params->downtime_limit = s->parameters.downtime_limit; 597 params->has_x_checkpoint_delay = true; 598 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay; 599 600 return params; 601 } 602 603 /* 604 * Return true if we're already in the middle of a migration 605 * (i.e. any of the active or setup states) 606 */ 607 static bool migration_is_setup_or_active(int state) 608 { 609 switch (state) { 610 case MIGRATION_STATUS_ACTIVE: 611 case MIGRATION_STATUS_POSTCOPY_ACTIVE: 612 case MIGRATION_STATUS_SETUP: 613 return true; 614 615 default: 616 return false; 617 618 } 619 } 620 621 static void get_xbzrle_cache_stats(MigrationInfo *info) 622 { 623 if (migrate_use_xbzrle()) { 624 info->has_xbzrle_cache = true; 625 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache)); 626 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size(); 627 info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred(); 628 info->xbzrle_cache->pages = xbzrle_mig_pages_transferred(); 629 info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss(); 630 info->xbzrle_cache->cache_miss_rate = xbzrle_mig_cache_miss_rate(); 631 info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow(); 632 } 633 } 634 635 static void populate_ram_info(MigrationInfo *info, MigrationState *s) 636 { 637 info->has_ram = true; 638 info->ram = g_malloc0(sizeof(*info->ram)); 639 info->ram->transferred = ram_bytes_transferred(); 640 info->ram->total = ram_bytes_total(); 641 info->ram->duplicate = dup_mig_pages_transferred(); 642 info->ram->skipped = skipped_mig_pages_transferred(); 643 info->ram->normal = norm_mig_pages_transferred(); 644 info->ram->normal_bytes = norm_mig_bytes_transferred(); 645 info->ram->mbps = s->mbps; 646 info->ram->dirty_sync_count = s->dirty_sync_count; 647 info->ram->postcopy_requests = s->postcopy_requests; 648 649 if (s->state != MIGRATION_STATUS_COMPLETED) { 650 info->ram->remaining = ram_bytes_remaining(); 651 info->ram->dirty_pages_rate = s->dirty_pages_rate; 652 } 653 } 654 655 MigrationInfo *qmp_query_migrate(Error **errp) 656 { 657 MigrationInfo *info = g_malloc0(sizeof(*info)); 658 MigrationState *s = migrate_get_current(); 659 660 switch (s->state) { 661 case MIGRATION_STATUS_NONE: 662 /* no migration has happened ever */ 663 break; 664 case MIGRATION_STATUS_SETUP: 665 info->has_status = true; 666 info->has_total_time = false; 667 break; 668 case MIGRATION_STATUS_ACTIVE: 669 case MIGRATION_STATUS_CANCELLING: 670 info->has_status = true; 671 info->has_total_time = true; 672 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) 673 - s->total_time; 674 info->has_expected_downtime = true; 675 info->expected_downtime = s->expected_downtime; 676 info->has_setup_time = true; 677 info->setup_time = s->setup_time; 678 679 populate_ram_info(info, s); 680 681 if (blk_mig_active()) { 682 info->has_disk = true; 683 info->disk = g_malloc0(sizeof(*info->disk)); 684 info->disk->transferred = blk_mig_bytes_transferred(); 685 info->disk->remaining = blk_mig_bytes_remaining(); 686 info->disk->total = blk_mig_bytes_total(); 687 } 688 689 if (cpu_throttle_active()) { 690 info->has_cpu_throttle_percentage = true; 691 info->cpu_throttle_percentage = cpu_throttle_get_percentage(); 692 } 693 694 get_xbzrle_cache_stats(info); 695 break; 696 case MIGRATION_STATUS_POSTCOPY_ACTIVE: 697 /* Mostly the same as active; TODO add some postcopy stats */ 698 info->has_status = true; 699 info->has_total_time = true; 700 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) 701 - s->total_time; 702 info->has_expected_downtime = true; 703 info->expected_downtime = s->expected_downtime; 704 info->has_setup_time = true; 705 info->setup_time = s->setup_time; 706 707 populate_ram_info(info, s); 708 709 if (blk_mig_active()) { 710 info->has_disk = true; 711 info->disk = g_malloc0(sizeof(*info->disk)); 712 info->disk->transferred = blk_mig_bytes_transferred(); 713 info->disk->remaining = blk_mig_bytes_remaining(); 714 info->disk->total = blk_mig_bytes_total(); 715 } 716 717 get_xbzrle_cache_stats(info); 718 break; 719 case MIGRATION_STATUS_COLO: 720 info->has_status = true; 721 /* TODO: display COLO specific information (checkpoint info etc.) */ 722 break; 723 case MIGRATION_STATUS_COMPLETED: 724 get_xbzrle_cache_stats(info); 725 726 info->has_status = true; 727 info->has_total_time = true; 728 info->total_time = s->total_time; 729 info->has_downtime = true; 730 info->downtime = s->downtime; 731 info->has_setup_time = true; 732 info->setup_time = s->setup_time; 733 734 populate_ram_info(info, s); 735 break; 736 case MIGRATION_STATUS_FAILED: 737 info->has_status = true; 738 if (s->error) { 739 info->has_error_desc = true; 740 info->error_desc = g_strdup(error_get_pretty(s->error)); 741 } 742 break; 743 case MIGRATION_STATUS_CANCELLED: 744 info->has_status = true; 745 break; 746 } 747 info->status = s->state; 748 749 return info; 750 } 751 752 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params, 753 Error **errp) 754 { 755 MigrationState *s = migrate_get_current(); 756 MigrationCapabilityStatusList *cap; 757 bool old_postcopy_cap = migrate_postcopy_ram(); 758 759 if (migration_is_setup_or_active(s->state)) { 760 error_setg(errp, QERR_MIGRATION_ACTIVE); 761 return; 762 } 763 764 for (cap = params; cap; cap = cap->next) { 765 if (cap->value->capability == MIGRATION_CAPABILITY_X_COLO) { 766 if (!colo_supported()) { 767 error_setg(errp, "COLO is not currently supported, please" 768 " configure with --enable-colo option in order to" 769 " support COLO feature"); 770 continue; 771 } 772 } 773 s->enabled_capabilities[cap->value->capability] = cap->value->state; 774 } 775 776 if (migrate_postcopy_ram()) { 777 if (migrate_use_compression()) { 778 /* The decompression threads asynchronously write into RAM 779 * rather than use the atomic copies needed to avoid 780 * userfaulting. It should be possible to fix the decompression 781 * threads for compatibility in future. 782 */ 783 error_report("Postcopy is not currently compatible with " 784 "compression"); 785 s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] = 786 false; 787 } 788 /* This check is reasonably expensive, so only when it's being 789 * set the first time, also it's only the destination that needs 790 * special support. 791 */ 792 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) && 793 !postcopy_ram_supported_by_host()) { 794 /* postcopy_ram_supported_by_host will have emitted a more 795 * detailed message 796 */ 797 error_report("Postcopy is not supported"); 798 s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] = 799 false; 800 } 801 } 802 } 803 804 void qmp_migrate_set_parameters(MigrationParameters *params, Error **errp) 805 { 806 MigrationState *s = migrate_get_current(); 807 808 if (params->has_compress_level && 809 (params->compress_level < 0 || params->compress_level > 9)) { 810 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level", 811 "is invalid, it should be in the range of 0 to 9"); 812 return; 813 } 814 if (params->has_compress_threads && 815 (params->compress_threads < 1 || params->compress_threads > 255)) { 816 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, 817 "compress_threads", 818 "is invalid, it should be in the range of 1 to 255"); 819 return; 820 } 821 if (params->has_decompress_threads && 822 (params->decompress_threads < 1 || params->decompress_threads > 255)) { 823 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, 824 "decompress_threads", 825 "is invalid, it should be in the range of 1 to 255"); 826 return; 827 } 828 if (params->has_cpu_throttle_initial && 829 (params->cpu_throttle_initial < 1 || 830 params->cpu_throttle_initial > 99)) { 831 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, 832 "cpu_throttle_initial", 833 "an integer in the range of 1 to 99"); 834 return; 835 } 836 if (params->has_cpu_throttle_increment && 837 (params->cpu_throttle_increment < 1 || 838 params->cpu_throttle_increment > 99)) { 839 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, 840 "cpu_throttle_increment", 841 "an integer in the range of 1 to 99"); 842 return; 843 } 844 if (params->has_max_bandwidth && 845 (params->max_bandwidth < 0 || params->max_bandwidth > SIZE_MAX)) { 846 error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the" 847 " range of 0 to %zu bytes/second", SIZE_MAX); 848 return; 849 } 850 if (params->has_downtime_limit && 851 (params->downtime_limit < 0 || 852 params->downtime_limit > MAX_MIGRATE_DOWNTIME)) { 853 error_setg(errp, "Parameter 'downtime_limit' expects an integer in " 854 "the range of 0 to %d milliseconds", 855 MAX_MIGRATE_DOWNTIME); 856 return; 857 } 858 if (params->has_x_checkpoint_delay && (params->x_checkpoint_delay < 0)) { 859 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, 860 "x_checkpoint_delay", 861 "is invalid, it should be positive"); 862 } 863 864 if (params->has_compress_level) { 865 s->parameters.compress_level = params->compress_level; 866 } 867 if (params->has_compress_threads) { 868 s->parameters.compress_threads = params->compress_threads; 869 } 870 if (params->has_decompress_threads) { 871 s->parameters.decompress_threads = params->decompress_threads; 872 } 873 if (params->has_cpu_throttle_initial) { 874 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial; 875 } 876 if (params->has_cpu_throttle_increment) { 877 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment; 878 } 879 if (params->has_tls_creds) { 880 g_free(s->parameters.tls_creds); 881 s->parameters.tls_creds = g_strdup(params->tls_creds); 882 } 883 if (params->has_tls_hostname) { 884 g_free(s->parameters.tls_hostname); 885 s->parameters.tls_hostname = g_strdup(params->tls_hostname); 886 } 887 if (params->has_max_bandwidth) { 888 s->parameters.max_bandwidth = params->max_bandwidth; 889 if (s->to_dst_file) { 890 qemu_file_set_rate_limit(s->to_dst_file, 891 s->parameters.max_bandwidth / XFER_LIMIT_RATIO); 892 } 893 } 894 if (params->has_downtime_limit) { 895 s->parameters.downtime_limit = params->downtime_limit; 896 } 897 898 if (params->has_x_checkpoint_delay) { 899 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay; 900 if (migration_in_colo_state()) { 901 colo_checkpoint_notify(s); 902 } 903 } 904 } 905 906 907 void qmp_migrate_start_postcopy(Error **errp) 908 { 909 MigrationState *s = migrate_get_current(); 910 911 if (!migrate_postcopy_ram()) { 912 error_setg(errp, "Enable postcopy with migrate_set_capability before" 913 " the start of migration"); 914 return; 915 } 916 917 if (s->state == MIGRATION_STATUS_NONE) { 918 error_setg(errp, "Postcopy must be started after migration has been" 919 " started"); 920 return; 921 } 922 /* 923 * we don't error if migration has finished since that would be racy 924 * with issuing this command. 925 */ 926 atomic_set(&s->start_postcopy, true); 927 } 928 929 /* shared migration helpers */ 930 931 void migrate_set_state(int *state, int old_state, int new_state) 932 { 933 if (atomic_cmpxchg(state, old_state, new_state) == old_state) { 934 trace_migrate_set_state(new_state); 935 migrate_generate_event(new_state); 936 } 937 } 938 939 static void migrate_fd_cleanup(void *opaque) 940 { 941 MigrationState *s = opaque; 942 943 qemu_bh_delete(s->cleanup_bh); 944 s->cleanup_bh = NULL; 945 946 flush_page_queue(s); 947 948 if (s->to_dst_file) { 949 trace_migrate_fd_cleanup(); 950 qemu_mutex_unlock_iothread(); 951 if (s->migration_thread_running) { 952 qemu_thread_join(&s->thread); 953 s->migration_thread_running = false; 954 } 955 qemu_mutex_lock_iothread(); 956 957 migrate_compress_threads_join(); 958 qemu_fclose(s->to_dst_file); 959 s->to_dst_file = NULL; 960 } 961 962 assert((s->state != MIGRATION_STATUS_ACTIVE) && 963 (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE)); 964 965 if (s->state == MIGRATION_STATUS_CANCELLING) { 966 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING, 967 MIGRATION_STATUS_CANCELLED); 968 } 969 970 notifier_list_notify(&migration_state_notifiers, s); 971 } 972 973 void migrate_fd_error(MigrationState *s, const Error *error) 974 { 975 trace_migrate_fd_error(error_get_pretty(error)); 976 assert(s->to_dst_file == NULL); 977 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, 978 MIGRATION_STATUS_FAILED); 979 if (!s->error) { 980 s->error = error_copy(error); 981 } 982 notifier_list_notify(&migration_state_notifiers, s); 983 } 984 985 static void migrate_fd_cancel(MigrationState *s) 986 { 987 int old_state ; 988 QEMUFile *f = migrate_get_current()->to_dst_file; 989 trace_migrate_fd_cancel(); 990 991 if (s->rp_state.from_dst_file) { 992 /* shutdown the rp socket, so causing the rp thread to shutdown */ 993 qemu_file_shutdown(s->rp_state.from_dst_file); 994 } 995 996 do { 997 old_state = s->state; 998 if (!migration_is_setup_or_active(old_state)) { 999 break; 1000 } 1001 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING); 1002 } while (s->state != MIGRATION_STATUS_CANCELLING); 1003 1004 /* 1005 * If we're unlucky the migration code might be stuck somewhere in a 1006 * send/write while the network has failed and is waiting to timeout; 1007 * if we've got shutdown(2) available then we can force it to quit. 1008 * The outgoing qemu file gets closed in migrate_fd_cleanup that is 1009 * called in a bh, so there is no race against this cancel. 1010 */ 1011 if (s->state == MIGRATION_STATUS_CANCELLING && f) { 1012 qemu_file_shutdown(f); 1013 } 1014 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) { 1015 Error *local_err = NULL; 1016 1017 bdrv_invalidate_cache_all(&local_err); 1018 if (local_err) { 1019 error_report_err(local_err); 1020 } else { 1021 s->block_inactive = false; 1022 } 1023 } 1024 } 1025 1026 void add_migration_state_change_notifier(Notifier *notify) 1027 { 1028 notifier_list_add(&migration_state_notifiers, notify); 1029 } 1030 1031 void remove_migration_state_change_notifier(Notifier *notify) 1032 { 1033 notifier_remove(notify); 1034 } 1035 1036 bool migration_in_setup(MigrationState *s) 1037 { 1038 return s->state == MIGRATION_STATUS_SETUP; 1039 } 1040 1041 bool migration_has_finished(MigrationState *s) 1042 { 1043 return s->state == MIGRATION_STATUS_COMPLETED; 1044 } 1045 1046 bool migration_has_failed(MigrationState *s) 1047 { 1048 return (s->state == MIGRATION_STATUS_CANCELLED || 1049 s->state == MIGRATION_STATUS_FAILED); 1050 } 1051 1052 bool migration_in_postcopy(MigrationState *s) 1053 { 1054 return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); 1055 } 1056 1057 bool migration_in_postcopy_after_devices(MigrationState *s) 1058 { 1059 return migration_in_postcopy(s) && s->postcopy_after_devices; 1060 } 1061 1062 bool migration_is_idle(MigrationState *s) 1063 { 1064 if (!s) { 1065 s = migrate_get_current(); 1066 } 1067 1068 switch (s->state) { 1069 case MIGRATION_STATUS_NONE: 1070 case MIGRATION_STATUS_CANCELLED: 1071 case MIGRATION_STATUS_COMPLETED: 1072 case MIGRATION_STATUS_FAILED: 1073 return true; 1074 case MIGRATION_STATUS_SETUP: 1075 case MIGRATION_STATUS_CANCELLING: 1076 case MIGRATION_STATUS_ACTIVE: 1077 case MIGRATION_STATUS_POSTCOPY_ACTIVE: 1078 case MIGRATION_STATUS_COLO: 1079 return false; 1080 case MIGRATION_STATUS__MAX: 1081 g_assert_not_reached(); 1082 } 1083 1084 return false; 1085 } 1086 1087 MigrationState *migrate_init(const MigrationParams *params) 1088 { 1089 MigrationState *s = migrate_get_current(); 1090 1091 /* 1092 * Reinitialise all migration state, except 1093 * parameters/capabilities that the user set, and 1094 * locks. 1095 */ 1096 s->bytes_xfer = 0; 1097 s->xfer_limit = 0; 1098 s->cleanup_bh = 0; 1099 s->to_dst_file = NULL; 1100 s->state = MIGRATION_STATUS_NONE; 1101 s->params = *params; 1102 s->rp_state.from_dst_file = NULL; 1103 s->rp_state.error = false; 1104 s->mbps = 0.0; 1105 s->downtime = 0; 1106 s->expected_downtime = 0; 1107 s->dirty_pages_rate = 0; 1108 s->dirty_bytes_rate = 0; 1109 s->setup_time = 0; 1110 s->dirty_sync_count = 0; 1111 s->start_postcopy = false; 1112 s->postcopy_after_devices = false; 1113 s->postcopy_requests = 0; 1114 s->migration_thread_running = false; 1115 s->last_req_rb = NULL; 1116 error_free(s->error); 1117 s->error = NULL; 1118 1119 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP); 1120 1121 QSIMPLEQ_INIT(&s->src_page_requests); 1122 1123 s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 1124 return s; 1125 } 1126 1127 static GSList *migration_blockers; 1128 1129 int migrate_add_blocker(Error *reason, Error **errp) 1130 { 1131 if (only_migratable) { 1132 error_propagate(errp, error_copy(reason)); 1133 error_prepend(errp, "disallowing migration blocker " 1134 "(--only_migratable) for: "); 1135 return -EACCES; 1136 } 1137 1138 if (migration_is_idle(NULL)) { 1139 migration_blockers = g_slist_prepend(migration_blockers, reason); 1140 return 0; 1141 } 1142 1143 error_propagate(errp, error_copy(reason)); 1144 error_prepend(errp, "disallowing migration blocker (migration in " 1145 "progress) for: "); 1146 return -EBUSY; 1147 } 1148 1149 void migrate_del_blocker(Error *reason) 1150 { 1151 migration_blockers = g_slist_remove(migration_blockers, reason); 1152 } 1153 1154 int check_migratable(Object *obj, Error **err) 1155 { 1156 DeviceClass *dc = DEVICE_GET_CLASS(obj); 1157 if (only_migratable && dc->vmsd) { 1158 if (dc->vmsd->unmigratable) { 1159 error_setg(err, "Device %s is not migratable, but " 1160 "--only-migratable was specified", 1161 object_get_typename(obj)); 1162 return -1; 1163 } 1164 } 1165 1166 return 0; 1167 } 1168 1169 void qmp_migrate_incoming(const char *uri, Error **errp) 1170 { 1171 Error *local_err = NULL; 1172 static bool once = true; 1173 1174 if (!deferred_incoming) { 1175 error_setg(errp, "For use with '-incoming defer'"); 1176 return; 1177 } 1178 if (!once) { 1179 error_setg(errp, "The incoming migration has already been started"); 1180 } 1181 1182 qemu_start_incoming_migration(uri, &local_err); 1183 1184 if (local_err) { 1185 error_propagate(errp, local_err); 1186 return; 1187 } 1188 1189 once = false; 1190 } 1191 1192 bool migration_is_blocked(Error **errp) 1193 { 1194 if (qemu_savevm_state_blocked(errp)) { 1195 return true; 1196 } 1197 1198 if (migration_blockers) { 1199 *errp = error_copy(migration_blockers->data); 1200 return true; 1201 } 1202 1203 return false; 1204 } 1205 1206 void qmp_migrate(const char *uri, bool has_blk, bool blk, 1207 bool has_inc, bool inc, bool has_detach, bool detach, 1208 Error **errp) 1209 { 1210 Error *local_err = NULL; 1211 MigrationState *s = migrate_get_current(); 1212 MigrationParams params; 1213 const char *p; 1214 1215 params.blk = has_blk && blk; 1216 params.shared = has_inc && inc; 1217 1218 if (migration_is_setup_or_active(s->state) || 1219 s->state == MIGRATION_STATUS_CANCELLING || 1220 s->state == MIGRATION_STATUS_COLO) { 1221 error_setg(errp, QERR_MIGRATION_ACTIVE); 1222 return; 1223 } 1224 if (runstate_check(RUN_STATE_INMIGRATE)) { 1225 error_setg(errp, "Guest is waiting for an incoming migration"); 1226 return; 1227 } 1228 1229 if (migration_is_blocked(errp)) { 1230 return; 1231 } 1232 1233 s = migrate_init(¶ms); 1234 1235 if (strstart(uri, "tcp:", &p)) { 1236 tcp_start_outgoing_migration(s, p, &local_err); 1237 #ifdef CONFIG_RDMA 1238 } else if (strstart(uri, "rdma:", &p)) { 1239 rdma_start_outgoing_migration(s, p, &local_err); 1240 #endif 1241 } else if (strstart(uri, "exec:", &p)) { 1242 exec_start_outgoing_migration(s, p, &local_err); 1243 } else if (strstart(uri, "unix:", &p)) { 1244 unix_start_outgoing_migration(s, p, &local_err); 1245 } else if (strstart(uri, "fd:", &p)) { 1246 fd_start_outgoing_migration(s, p, &local_err); 1247 } else { 1248 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri", 1249 "a valid migration protocol"); 1250 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, 1251 MIGRATION_STATUS_FAILED); 1252 return; 1253 } 1254 1255 if (local_err) { 1256 migrate_fd_error(s, local_err); 1257 error_propagate(errp, local_err); 1258 return; 1259 } 1260 } 1261 1262 void qmp_migrate_cancel(Error **errp) 1263 { 1264 migrate_fd_cancel(migrate_get_current()); 1265 } 1266 1267 void qmp_migrate_set_cache_size(int64_t value, Error **errp) 1268 { 1269 MigrationState *s = migrate_get_current(); 1270 int64_t new_size; 1271 1272 /* Check for truncation */ 1273 if (value != (size_t)value) { 1274 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", 1275 "exceeding address space"); 1276 return; 1277 } 1278 1279 /* Cache should not be larger than guest ram size */ 1280 if (value > ram_bytes_total()) { 1281 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", 1282 "exceeds guest ram size "); 1283 return; 1284 } 1285 1286 new_size = xbzrle_cache_resize(value); 1287 if (new_size < 0) { 1288 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", 1289 "is smaller than page size"); 1290 return; 1291 } 1292 1293 s->xbzrle_cache_size = new_size; 1294 } 1295 1296 int64_t qmp_query_migrate_cache_size(Error **errp) 1297 { 1298 return migrate_xbzrle_cache_size(); 1299 } 1300 1301 void qmp_migrate_set_speed(int64_t value, Error **errp) 1302 { 1303 MigrationParameters p = { 1304 .has_max_bandwidth = true, 1305 .max_bandwidth = value, 1306 }; 1307 1308 qmp_migrate_set_parameters(&p, errp); 1309 } 1310 1311 void qmp_migrate_set_downtime(double value, Error **errp) 1312 { 1313 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) { 1314 error_setg(errp, "Parameter 'downtime_limit' expects an integer in " 1315 "the range of 0 to %d seconds", 1316 MAX_MIGRATE_DOWNTIME_SECONDS); 1317 return; 1318 } 1319 1320 value *= 1000; /* Convert to milliseconds */ 1321 value = MAX(0, MIN(INT64_MAX, value)); 1322 1323 MigrationParameters p = { 1324 .has_downtime_limit = true, 1325 .downtime_limit = value, 1326 }; 1327 1328 qmp_migrate_set_parameters(&p, errp); 1329 } 1330 1331 bool migrate_release_ram(void) 1332 { 1333 MigrationState *s; 1334 1335 s = migrate_get_current(); 1336 1337 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM]; 1338 } 1339 1340 bool migrate_postcopy_ram(void) 1341 { 1342 MigrationState *s; 1343 1344 s = migrate_get_current(); 1345 1346 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM]; 1347 } 1348 1349 bool migrate_auto_converge(void) 1350 { 1351 MigrationState *s; 1352 1353 s = migrate_get_current(); 1354 1355 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE]; 1356 } 1357 1358 bool migrate_zero_blocks(void) 1359 { 1360 MigrationState *s; 1361 1362 s = migrate_get_current(); 1363 1364 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS]; 1365 } 1366 1367 bool migrate_use_compression(void) 1368 { 1369 MigrationState *s; 1370 1371 s = migrate_get_current(); 1372 1373 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS]; 1374 } 1375 1376 int migrate_compress_level(void) 1377 { 1378 MigrationState *s; 1379 1380 s = migrate_get_current(); 1381 1382 return s->parameters.compress_level; 1383 } 1384 1385 int migrate_compress_threads(void) 1386 { 1387 MigrationState *s; 1388 1389 s = migrate_get_current(); 1390 1391 return s->parameters.compress_threads; 1392 } 1393 1394 int migrate_decompress_threads(void) 1395 { 1396 MigrationState *s; 1397 1398 s = migrate_get_current(); 1399 1400 return s->parameters.decompress_threads; 1401 } 1402 1403 bool migrate_use_events(void) 1404 { 1405 MigrationState *s; 1406 1407 s = migrate_get_current(); 1408 1409 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS]; 1410 } 1411 1412 int migrate_use_xbzrle(void) 1413 { 1414 MigrationState *s; 1415 1416 s = migrate_get_current(); 1417 1418 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE]; 1419 } 1420 1421 int64_t migrate_xbzrle_cache_size(void) 1422 { 1423 MigrationState *s; 1424 1425 s = migrate_get_current(); 1426 1427 return s->xbzrle_cache_size; 1428 } 1429 1430 /* migration thread support */ 1431 /* 1432 * Something bad happened to the RP stream, mark an error 1433 * The caller shall print or trace something to indicate why 1434 */ 1435 static void mark_source_rp_bad(MigrationState *s) 1436 { 1437 s->rp_state.error = true; 1438 } 1439 1440 static struct rp_cmd_args { 1441 ssize_t len; /* -1 = variable */ 1442 const char *name; 1443 } rp_cmd_args[] = { 1444 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" }, 1445 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" }, 1446 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" }, 1447 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" }, 1448 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" }, 1449 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" }, 1450 }; 1451 1452 /* 1453 * Process a request for pages received on the return path, 1454 * We're allowed to send more than requested (e.g. to round to our page size) 1455 * and we don't need to send pages that have already been sent. 1456 */ 1457 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname, 1458 ram_addr_t start, size_t len) 1459 { 1460 long our_host_ps = getpagesize(); 1461 1462 trace_migrate_handle_rp_req_pages(rbname, start, len); 1463 1464 /* 1465 * Since we currently insist on matching page sizes, just sanity check 1466 * we're being asked for whole host pages. 1467 */ 1468 if (start & (our_host_ps-1) || 1469 (len & (our_host_ps-1))) { 1470 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT 1471 " len: %zd", __func__, start, len); 1472 mark_source_rp_bad(ms); 1473 return; 1474 } 1475 1476 if (ram_save_queue_pages(ms, rbname, start, len)) { 1477 mark_source_rp_bad(ms); 1478 } 1479 } 1480 1481 /* 1482 * Handles messages sent on the return path towards the source VM 1483 * 1484 */ 1485 static void *source_return_path_thread(void *opaque) 1486 { 1487 MigrationState *ms = opaque; 1488 QEMUFile *rp = ms->rp_state.from_dst_file; 1489 uint16_t header_len, header_type; 1490 uint8_t buf[512]; 1491 uint32_t tmp32, sibling_error; 1492 ram_addr_t start = 0; /* =0 to silence warning */ 1493 size_t len = 0, expected_len; 1494 int res; 1495 1496 trace_source_return_path_thread_entry(); 1497 while (!ms->rp_state.error && !qemu_file_get_error(rp) && 1498 migration_is_setup_or_active(ms->state)) { 1499 trace_source_return_path_thread_loop_top(); 1500 header_type = qemu_get_be16(rp); 1501 header_len = qemu_get_be16(rp); 1502 1503 if (header_type >= MIG_RP_MSG_MAX || 1504 header_type == MIG_RP_MSG_INVALID) { 1505 error_report("RP: Received invalid message 0x%04x length 0x%04x", 1506 header_type, header_len); 1507 mark_source_rp_bad(ms); 1508 goto out; 1509 } 1510 1511 if ((rp_cmd_args[header_type].len != -1 && 1512 header_len != rp_cmd_args[header_type].len) || 1513 header_len > sizeof(buf)) { 1514 error_report("RP: Received '%s' message (0x%04x) with" 1515 "incorrect length %d expecting %zu", 1516 rp_cmd_args[header_type].name, header_type, header_len, 1517 (size_t)rp_cmd_args[header_type].len); 1518 mark_source_rp_bad(ms); 1519 goto out; 1520 } 1521 1522 /* We know we've got a valid header by this point */ 1523 res = qemu_get_buffer(rp, buf, header_len); 1524 if (res != header_len) { 1525 error_report("RP: Failed reading data for message 0x%04x" 1526 " read %d expected %d", 1527 header_type, res, header_len); 1528 mark_source_rp_bad(ms); 1529 goto out; 1530 } 1531 1532 /* OK, we have the message and the data */ 1533 switch (header_type) { 1534 case MIG_RP_MSG_SHUT: 1535 sibling_error = ldl_be_p(buf); 1536 trace_source_return_path_thread_shut(sibling_error); 1537 if (sibling_error) { 1538 error_report("RP: Sibling indicated error %d", sibling_error); 1539 mark_source_rp_bad(ms); 1540 } 1541 /* 1542 * We'll let the main thread deal with closing the RP 1543 * we could do a shutdown(2) on it, but we're the only user 1544 * anyway, so there's nothing gained. 1545 */ 1546 goto out; 1547 1548 case MIG_RP_MSG_PONG: 1549 tmp32 = ldl_be_p(buf); 1550 trace_source_return_path_thread_pong(tmp32); 1551 break; 1552 1553 case MIG_RP_MSG_REQ_PAGES: 1554 start = ldq_be_p(buf); 1555 len = ldl_be_p(buf + 8); 1556 migrate_handle_rp_req_pages(ms, NULL, start, len); 1557 break; 1558 1559 case MIG_RP_MSG_REQ_PAGES_ID: 1560 expected_len = 12 + 1; /* header + termination */ 1561 1562 if (header_len >= expected_len) { 1563 start = ldq_be_p(buf); 1564 len = ldl_be_p(buf + 8); 1565 /* Now we expect an idstr */ 1566 tmp32 = buf[12]; /* Length of the following idstr */ 1567 buf[13 + tmp32] = '\0'; 1568 expected_len += tmp32; 1569 } 1570 if (header_len != expected_len) { 1571 error_report("RP: Req_Page_id with length %d expecting %zd", 1572 header_len, expected_len); 1573 mark_source_rp_bad(ms); 1574 goto out; 1575 } 1576 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len); 1577 break; 1578 1579 default: 1580 break; 1581 } 1582 } 1583 if (qemu_file_get_error(rp)) { 1584 trace_source_return_path_thread_bad_end(); 1585 mark_source_rp_bad(ms); 1586 } 1587 1588 trace_source_return_path_thread_end(); 1589 out: 1590 ms->rp_state.from_dst_file = NULL; 1591 qemu_fclose(rp); 1592 return NULL; 1593 } 1594 1595 static int open_return_path_on_source(MigrationState *ms) 1596 { 1597 1598 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file); 1599 if (!ms->rp_state.from_dst_file) { 1600 return -1; 1601 } 1602 1603 trace_open_return_path_on_source(); 1604 qemu_thread_create(&ms->rp_state.rp_thread, "return path", 1605 source_return_path_thread, ms, QEMU_THREAD_JOINABLE); 1606 1607 trace_open_return_path_on_source_continue(); 1608 1609 return 0; 1610 } 1611 1612 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */ 1613 static int await_return_path_close_on_source(MigrationState *ms) 1614 { 1615 /* 1616 * If this is a normal exit then the destination will send a SHUT and the 1617 * rp_thread will exit, however if there's an error we need to cause 1618 * it to exit. 1619 */ 1620 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) { 1621 /* 1622 * shutdown(2), if we have it, will cause it to unblock if it's stuck 1623 * waiting for the destination. 1624 */ 1625 qemu_file_shutdown(ms->rp_state.from_dst_file); 1626 mark_source_rp_bad(ms); 1627 } 1628 trace_await_return_path_close_on_source_joining(); 1629 qemu_thread_join(&ms->rp_state.rp_thread); 1630 trace_await_return_path_close_on_source_close(); 1631 return ms->rp_state.error; 1632 } 1633 1634 /* 1635 * Switch from normal iteration to postcopy 1636 * Returns non-0 on error 1637 */ 1638 static int postcopy_start(MigrationState *ms, bool *old_vm_running) 1639 { 1640 int ret; 1641 QIOChannelBuffer *bioc; 1642 QEMUFile *fb; 1643 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 1644 bool restart_block = false; 1645 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE, 1646 MIGRATION_STATUS_POSTCOPY_ACTIVE); 1647 1648 trace_postcopy_start(); 1649 qemu_mutex_lock_iothread(); 1650 trace_postcopy_start_set_run(); 1651 1652 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER); 1653 *old_vm_running = runstate_is_running(); 1654 global_state_store(); 1655 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); 1656 if (ret < 0) { 1657 goto fail; 1658 } 1659 1660 ret = bdrv_inactivate_all(); 1661 if (ret < 0) { 1662 goto fail; 1663 } 1664 restart_block = true; 1665 1666 /* 1667 * Cause any non-postcopiable, but iterative devices to 1668 * send out their final data. 1669 */ 1670 qemu_savevm_state_complete_precopy(ms->to_dst_file, true); 1671 1672 /* 1673 * in Finish migrate and with the io-lock held everything should 1674 * be quiet, but we've potentially still got dirty pages and we 1675 * need to tell the destination to throw any pages it's already received 1676 * that are dirty 1677 */ 1678 if (ram_postcopy_send_discard_bitmap(ms)) { 1679 error_report("postcopy send discard bitmap failed"); 1680 goto fail; 1681 } 1682 1683 /* 1684 * send rest of state - note things that are doing postcopy 1685 * will notice we're in POSTCOPY_ACTIVE and not actually 1686 * wrap their state up here 1687 */ 1688 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX); 1689 /* Ping just for debugging, helps line traces up */ 1690 qemu_savevm_send_ping(ms->to_dst_file, 2); 1691 1692 /* 1693 * While loading the device state we may trigger page transfer 1694 * requests and the fd must be free to process those, and thus 1695 * the destination must read the whole device state off the fd before 1696 * it starts processing it. Unfortunately the ad-hoc migration format 1697 * doesn't allow the destination to know the size to read without fully 1698 * parsing it through each devices load-state code (especially the open 1699 * coded devices that use get/put). 1700 * So we wrap the device state up in a package with a length at the start; 1701 * to do this we use a qemu_buf to hold the whole of the device state. 1702 */ 1703 bioc = qio_channel_buffer_new(4096); 1704 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer"); 1705 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc)); 1706 object_unref(OBJECT(bioc)); 1707 1708 /* 1709 * Make sure the receiver can get incoming pages before we send the rest 1710 * of the state 1711 */ 1712 qemu_savevm_send_postcopy_listen(fb); 1713 1714 qemu_savevm_state_complete_precopy(fb, false); 1715 qemu_savevm_send_ping(fb, 3); 1716 1717 qemu_savevm_send_postcopy_run(fb); 1718 1719 /* <><> end of stuff going into the package */ 1720 1721 /* Last point of recovery; as soon as we send the package the destination 1722 * can open devices and potentially start running. 1723 * Lets just check again we've not got any errors. 1724 */ 1725 ret = qemu_file_get_error(ms->to_dst_file); 1726 if (ret) { 1727 error_report("postcopy_start: Migration stream errored (pre package)"); 1728 goto fail_closefb; 1729 } 1730 1731 restart_block = false; 1732 1733 /* Now send that blob */ 1734 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) { 1735 goto fail_closefb; 1736 } 1737 qemu_fclose(fb); 1738 1739 /* Send a notify to give a chance for anything that needs to happen 1740 * at the transition to postcopy and after the device state; in particular 1741 * spice needs to trigger a transition now 1742 */ 1743 ms->postcopy_after_devices = true; 1744 notifier_list_notify(&migration_state_notifiers, ms); 1745 1746 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop; 1747 1748 qemu_mutex_unlock_iothread(); 1749 1750 /* 1751 * Although this ping is just for debug, it could potentially be 1752 * used for getting a better measurement of downtime at the source. 1753 */ 1754 qemu_savevm_send_ping(ms->to_dst_file, 4); 1755 1756 if (migrate_release_ram()) { 1757 ram_postcopy_migrated_memory_release(ms); 1758 } 1759 1760 ret = qemu_file_get_error(ms->to_dst_file); 1761 if (ret) { 1762 error_report("postcopy_start: Migration stream errored"); 1763 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, 1764 MIGRATION_STATUS_FAILED); 1765 } 1766 1767 return ret; 1768 1769 fail_closefb: 1770 qemu_fclose(fb); 1771 fail: 1772 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, 1773 MIGRATION_STATUS_FAILED); 1774 if (restart_block) { 1775 /* A failure happened early enough that we know the destination hasn't 1776 * accessed block devices, so we're safe to recover. 1777 */ 1778 Error *local_err = NULL; 1779 1780 bdrv_invalidate_cache_all(&local_err); 1781 if (local_err) { 1782 error_report_err(local_err); 1783 } 1784 } 1785 qemu_mutex_unlock_iothread(); 1786 return -1; 1787 } 1788 1789 /** 1790 * migration_completion: Used by migration_thread when there's not much left. 1791 * The caller 'breaks' the loop when this returns. 1792 * 1793 * @s: Current migration state 1794 * @current_active_state: The migration state we expect to be in 1795 * @*old_vm_running: Pointer to old_vm_running flag 1796 * @*start_time: Pointer to time to update 1797 */ 1798 static void migration_completion(MigrationState *s, int current_active_state, 1799 bool *old_vm_running, 1800 int64_t *start_time) 1801 { 1802 int ret; 1803 1804 if (s->state == MIGRATION_STATUS_ACTIVE) { 1805 qemu_mutex_lock_iothread(); 1806 *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 1807 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER); 1808 *old_vm_running = runstate_is_running(); 1809 ret = global_state_store(); 1810 1811 if (!ret) { 1812 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); 1813 /* 1814 * Don't mark the image with BDRV_O_INACTIVE flag if 1815 * we will go into COLO stage later. 1816 */ 1817 if (ret >= 0 && !migrate_colo_enabled()) { 1818 ret = bdrv_inactivate_all(); 1819 } 1820 if (ret >= 0) { 1821 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX); 1822 qemu_savevm_state_complete_precopy(s->to_dst_file, false); 1823 s->block_inactive = true; 1824 } 1825 } 1826 qemu_mutex_unlock_iothread(); 1827 1828 if (ret < 0) { 1829 goto fail; 1830 } 1831 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { 1832 trace_migration_completion_postcopy_end(); 1833 1834 qemu_savevm_state_complete_postcopy(s->to_dst_file); 1835 trace_migration_completion_postcopy_end_after_complete(); 1836 } 1837 1838 /* 1839 * If rp was opened we must clean up the thread before 1840 * cleaning everything else up (since if there are no failures 1841 * it will wait for the destination to send it's status in 1842 * a SHUT command). 1843 * Postcopy opens rp if enabled (even if it's not avtivated) 1844 */ 1845 if (migrate_postcopy_ram()) { 1846 int rp_error; 1847 trace_migration_completion_postcopy_end_before_rp(); 1848 rp_error = await_return_path_close_on_source(s); 1849 trace_migration_completion_postcopy_end_after_rp(rp_error); 1850 if (rp_error) { 1851 goto fail_invalidate; 1852 } 1853 } 1854 1855 if (qemu_file_get_error(s->to_dst_file)) { 1856 trace_migration_completion_file_err(); 1857 goto fail_invalidate; 1858 } 1859 1860 if (!migrate_colo_enabled()) { 1861 migrate_set_state(&s->state, current_active_state, 1862 MIGRATION_STATUS_COMPLETED); 1863 } 1864 1865 return; 1866 1867 fail_invalidate: 1868 /* If not doing postcopy, vm_start() will be called: let's regain 1869 * control on images. 1870 */ 1871 if (s->state == MIGRATION_STATUS_ACTIVE) { 1872 Error *local_err = NULL; 1873 1874 qemu_mutex_lock_iothread(); 1875 bdrv_invalidate_cache_all(&local_err); 1876 if (local_err) { 1877 error_report_err(local_err); 1878 } else { 1879 s->block_inactive = false; 1880 } 1881 qemu_mutex_unlock_iothread(); 1882 } 1883 1884 fail: 1885 migrate_set_state(&s->state, current_active_state, 1886 MIGRATION_STATUS_FAILED); 1887 } 1888 1889 bool migrate_colo_enabled(void) 1890 { 1891 MigrationState *s = migrate_get_current(); 1892 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO]; 1893 } 1894 1895 /* 1896 * Master migration thread on the source VM. 1897 * It drives the migration and pumps the data down the outgoing channel. 1898 */ 1899 static void *migration_thread(void *opaque) 1900 { 1901 MigrationState *s = opaque; 1902 /* Used by the bandwidth calcs, updated later */ 1903 int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 1904 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); 1905 int64_t initial_bytes = 0; 1906 int64_t max_size = 0; 1907 int64_t start_time = initial_time; 1908 int64_t end_time; 1909 bool old_vm_running = false; 1910 bool entered_postcopy = false; 1911 /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */ 1912 enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE; 1913 bool enable_colo = migrate_colo_enabled(); 1914 1915 rcu_register_thread(); 1916 1917 qemu_savevm_state_header(s->to_dst_file); 1918 1919 if (migrate_postcopy_ram()) { 1920 /* Now tell the dest that it should open its end so it can reply */ 1921 qemu_savevm_send_open_return_path(s->to_dst_file); 1922 1923 /* And do a ping that will make stuff easier to debug */ 1924 qemu_savevm_send_ping(s->to_dst_file, 1); 1925 1926 /* 1927 * Tell the destination that we *might* want to do postcopy later; 1928 * if the other end can't do postcopy it should fail now, nice and 1929 * early. 1930 */ 1931 qemu_savevm_send_postcopy_advise(s->to_dst_file); 1932 } 1933 1934 qemu_savevm_state_begin(s->to_dst_file, &s->params); 1935 1936 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; 1937 current_active_state = MIGRATION_STATUS_ACTIVE; 1938 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, 1939 MIGRATION_STATUS_ACTIVE); 1940 1941 trace_migration_thread_setup_complete(); 1942 1943 while (s->state == MIGRATION_STATUS_ACTIVE || 1944 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { 1945 int64_t current_time; 1946 uint64_t pending_size; 1947 1948 if (!qemu_file_rate_limit(s->to_dst_file)) { 1949 uint64_t pend_post, pend_nonpost; 1950 1951 qemu_savevm_state_pending(s->to_dst_file, max_size, &pend_nonpost, 1952 &pend_post); 1953 pending_size = pend_nonpost + pend_post; 1954 trace_migrate_pending(pending_size, max_size, 1955 pend_post, pend_nonpost); 1956 if (pending_size && pending_size >= max_size) { 1957 /* Still a significant amount to transfer */ 1958 1959 if (migrate_postcopy_ram() && 1960 s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE && 1961 pend_nonpost <= max_size && 1962 atomic_read(&s->start_postcopy)) { 1963 1964 if (!postcopy_start(s, &old_vm_running)) { 1965 current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE; 1966 entered_postcopy = true; 1967 } 1968 1969 continue; 1970 } 1971 /* Just another iteration step */ 1972 qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy); 1973 } else { 1974 trace_migration_thread_low_pending(pending_size); 1975 migration_completion(s, current_active_state, 1976 &old_vm_running, &start_time); 1977 break; 1978 } 1979 } 1980 1981 if (qemu_file_get_error(s->to_dst_file)) { 1982 migrate_set_state(&s->state, current_active_state, 1983 MIGRATION_STATUS_FAILED); 1984 trace_migration_thread_file_err(); 1985 break; 1986 } 1987 current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 1988 if (current_time >= initial_time + BUFFER_DELAY) { 1989 uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) - 1990 initial_bytes; 1991 uint64_t time_spent = current_time - initial_time; 1992 double bandwidth = (double)transferred_bytes / time_spent; 1993 max_size = bandwidth * s->parameters.downtime_limit; 1994 1995 s->mbps = (((double) transferred_bytes * 8.0) / 1996 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0; 1997 1998 trace_migrate_transferred(transferred_bytes, time_spent, 1999 bandwidth, max_size); 2000 /* if we haven't sent anything, we don't want to recalculate 2001 10000 is a small enough number for our purposes */ 2002 if (s->dirty_bytes_rate && transferred_bytes > 10000) { 2003 s->expected_downtime = s->dirty_bytes_rate / bandwidth; 2004 } 2005 2006 qemu_file_reset_rate_limit(s->to_dst_file); 2007 initial_time = current_time; 2008 initial_bytes = qemu_ftell(s->to_dst_file); 2009 } 2010 if (qemu_file_rate_limit(s->to_dst_file)) { 2011 /* usleep expects microseconds */ 2012 g_usleep((initial_time + BUFFER_DELAY - current_time)*1000); 2013 } 2014 } 2015 2016 trace_migration_thread_after_loop(); 2017 /* If we enabled cpu throttling for auto-converge, turn it off. */ 2018 cpu_throttle_stop(); 2019 end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 2020 2021 qemu_mutex_lock_iothread(); 2022 /* 2023 * The resource has been allocated by migration will be reused in COLO 2024 * process, so don't release them. 2025 */ 2026 if (!enable_colo) { 2027 qemu_savevm_state_cleanup(); 2028 } 2029 if (s->state == MIGRATION_STATUS_COMPLETED) { 2030 uint64_t transferred_bytes = qemu_ftell(s->to_dst_file); 2031 s->total_time = end_time - s->total_time; 2032 if (!entered_postcopy) { 2033 s->downtime = end_time - start_time; 2034 } 2035 if (s->total_time) { 2036 s->mbps = (((double) transferred_bytes * 8.0) / 2037 ((double) s->total_time)) / 1000; 2038 } 2039 runstate_set(RUN_STATE_POSTMIGRATE); 2040 } else { 2041 if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) { 2042 migrate_start_colo_process(s); 2043 qemu_savevm_state_cleanup(); 2044 /* 2045 * Fixme: we will run VM in COLO no matter its old running state. 2046 * After exited COLO, we will keep running. 2047 */ 2048 old_vm_running = true; 2049 } 2050 if (old_vm_running && !entered_postcopy) { 2051 vm_start(); 2052 } else { 2053 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) { 2054 runstate_set(RUN_STATE_POSTMIGRATE); 2055 } 2056 } 2057 } 2058 qemu_bh_schedule(s->cleanup_bh); 2059 qemu_mutex_unlock_iothread(); 2060 2061 rcu_unregister_thread(); 2062 return NULL; 2063 } 2064 2065 void migrate_fd_connect(MigrationState *s) 2066 { 2067 s->expected_downtime = s->parameters.downtime_limit; 2068 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s); 2069 2070 qemu_file_set_blocking(s->to_dst_file, true); 2071 qemu_file_set_rate_limit(s->to_dst_file, 2072 s->parameters.max_bandwidth / XFER_LIMIT_RATIO); 2073 2074 /* Notify before starting migration thread */ 2075 notifier_list_notify(&migration_state_notifiers, s); 2076 2077 /* 2078 * Open the return path; currently for postcopy but other things might 2079 * also want it. 2080 */ 2081 if (migrate_postcopy_ram()) { 2082 if (open_return_path_on_source(s)) { 2083 error_report("Unable to open return-path for postcopy"); 2084 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, 2085 MIGRATION_STATUS_FAILED); 2086 migrate_fd_cleanup(s); 2087 return; 2088 } 2089 } 2090 2091 migrate_compress_threads_create(); 2092 qemu_thread_create(&s->thread, "live_migration", migration_thread, s, 2093 QEMU_THREAD_JOINABLE); 2094 s->migration_thread_running = true; 2095 } 2096 2097 PostcopyState postcopy_state_get(void) 2098 { 2099 return atomic_mb_read(&incoming_postcopy_state); 2100 } 2101 2102 /* Set the state and return the old state */ 2103 PostcopyState postcopy_state_set(PostcopyState new_state) 2104 { 2105 return atomic_xchg(&incoming_postcopy_state, new_state); 2106 } 2107 2108