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/blocker.h" 21 #include "migration/migration.h" 22 #include "qemu-file-channel.h" 23 #include "migration/qemu-file.h" 24 #include "migration/vmstate.h" 25 #include "sysemu/sysemu.h" 26 #include "block/block.h" 27 #include "qapi/qmp/qerror.h" 28 #include "qapi/util.h" 29 #include "qemu/sockets.h" 30 #include "qemu/rcu.h" 31 #include "migration/block.h" 32 #include "postcopy-ram.h" 33 #include "qemu/thread.h" 34 #include "qmp-commands.h" 35 #include "trace.h" 36 #include "qapi-event.h" 37 #include "qom/cpu.h" 38 #include "exec/memory.h" 39 #include "exec/address-spaces.h" 40 #include "exec/target_page.h" 41 #include "io/channel-buffer.h" 42 #include "io/channel-tls.h" 43 #include "migration/colo.h" 44 45 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */ 46 47 /* Amount of time to allocate to each "chunk" of bandwidth-throttled 48 * data. */ 49 #define BUFFER_DELAY 100 50 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY) 51 52 /* Time in milliseconds we are allowed to stop the source, 53 * for sending the last part */ 54 #define DEFAULT_MIGRATE_SET_DOWNTIME 300 55 56 /* Maximum migrate downtime set to 2000 seconds */ 57 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000 58 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000) 59 60 /* Default compression thread count */ 61 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8 62 /* Default decompression thread count, usually decompression is at 63 * least 4 times as fast as compression.*/ 64 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2 65 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */ 66 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1 67 /* Define default autoconverge cpu throttle migration parameters */ 68 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20 69 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10 70 71 /* Migration XBZRLE default cache size */ 72 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024) 73 74 /* The delay time (in ms) between two COLO checkpoints 75 * Note: Please change this default value to 10000 when we support hybrid mode. 76 */ 77 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200 78 79 static NotifierList migration_state_notifiers = 80 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers); 81 82 static bool deferred_incoming; 83 84 /* When we add fault tolerance, we could have several 85 migrations at once. For now we don't need to add 86 dynamic creation of migration */ 87 88 /* For outgoing */ 89 MigrationState *migrate_get_current(void) 90 { 91 static bool once; 92 static MigrationState current_migration = { 93 .state = MIGRATION_STATUS_NONE, 94 .xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE, 95 .mbps = -1, 96 .parameters = { 97 .compress_level = DEFAULT_MIGRATE_COMPRESS_LEVEL, 98 .compress_threads = DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT, 99 .decompress_threads = DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT, 100 .cpu_throttle_initial = DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL, 101 .cpu_throttle_increment = DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT, 102 .max_bandwidth = MAX_THROTTLE, 103 .downtime_limit = DEFAULT_MIGRATE_SET_DOWNTIME, 104 .x_checkpoint_delay = DEFAULT_MIGRATE_X_CHECKPOINT_DELAY, 105 }, 106 }; 107 108 if (!once) { 109 current_migration.parameters.tls_creds = g_strdup(""); 110 current_migration.parameters.tls_hostname = g_strdup(""); 111 once = true; 112 } 113 return ¤t_migration; 114 } 115 116 MigrationIncomingState *migration_incoming_get_current(void) 117 { 118 static bool once; 119 static MigrationIncomingState mis_current; 120 121 if (!once) { 122 mis_current.state = MIGRATION_STATUS_NONE; 123 memset(&mis_current, 0, sizeof(MigrationIncomingState)); 124 QLIST_INIT(&mis_current.loadvm_handlers); 125 qemu_mutex_init(&mis_current.rp_mutex); 126 qemu_event_init(&mis_current.main_thread_load_event, false); 127 once = true; 128 } 129 return &mis_current; 130 } 131 132 void migration_incoming_state_destroy(void) 133 { 134 struct MigrationIncomingState *mis = migration_incoming_get_current(); 135 136 qemu_event_destroy(&mis->main_thread_load_event); 137 loadvm_free_handlers(mis); 138 } 139 140 141 typedef struct { 142 bool optional; 143 uint32_t size; 144 uint8_t runstate[100]; 145 RunState state; 146 bool received; 147 } GlobalState; 148 149 static GlobalState global_state; 150 151 int global_state_store(void) 152 { 153 if (!runstate_store((char *)global_state.runstate, 154 sizeof(global_state.runstate))) { 155 error_report("runstate name too big: %s", global_state.runstate); 156 trace_migrate_state_too_big(); 157 return -EINVAL; 158 } 159 return 0; 160 } 161 162 void global_state_store_running(void) 163 { 164 const char *state = RunState_lookup[RUN_STATE_RUNNING]; 165 strncpy((char *)global_state.runstate, 166 state, sizeof(global_state.runstate)); 167 } 168 169 static bool global_state_received(void) 170 { 171 return global_state.received; 172 } 173 174 static RunState global_state_get_runstate(void) 175 { 176 return global_state.state; 177 } 178 179 void global_state_set_optional(void) 180 { 181 global_state.optional = true; 182 } 183 184 static bool global_state_needed(void *opaque) 185 { 186 GlobalState *s = opaque; 187 char *runstate = (char *)s->runstate; 188 189 /* If it is not optional, it is mandatory */ 190 191 if (s->optional == false) { 192 return true; 193 } 194 195 /* If state is running or paused, it is not needed */ 196 197 if (strcmp(runstate, "running") == 0 || 198 strcmp(runstate, "paused") == 0) { 199 return false; 200 } 201 202 /* for any other state it is needed */ 203 return true; 204 } 205 206 static int global_state_post_load(void *opaque, int version_id) 207 { 208 GlobalState *s = opaque; 209 Error *local_err = NULL; 210 int r; 211 char *runstate = (char *)s->runstate; 212 213 s->received = true; 214 trace_migrate_global_state_post_load(runstate); 215 216 r = qapi_enum_parse(RunState_lookup, runstate, RUN_STATE__MAX, 217 -1, &local_err); 218 219 if (r == -1) { 220 if (local_err) { 221 error_report_err(local_err); 222 } 223 return -EINVAL; 224 } 225 s->state = r; 226 227 return 0; 228 } 229 230 static void global_state_pre_save(void *opaque) 231 { 232 GlobalState *s = opaque; 233 234 trace_migrate_global_state_pre_save((char *)s->runstate); 235 s->size = strlen((char *)s->runstate) + 1; 236 } 237 238 static const VMStateDescription vmstate_globalstate = { 239 .name = "globalstate", 240 .version_id = 1, 241 .minimum_version_id = 1, 242 .post_load = global_state_post_load, 243 .pre_save = global_state_pre_save, 244 .needed = global_state_needed, 245 .fields = (VMStateField[]) { 246 VMSTATE_UINT32(size, GlobalState), 247 VMSTATE_BUFFER(runstate, GlobalState), 248 VMSTATE_END_OF_LIST() 249 }, 250 }; 251 252 void register_global_state(void) 253 { 254 /* We would use it independently that we receive it */ 255 strcpy((char *)&global_state.runstate, ""); 256 global_state.received = false; 257 vmstate_register(NULL, 0, &vmstate_globalstate, &global_state); 258 } 259 260 static void migrate_generate_event(int new_state) 261 { 262 if (migrate_use_events()) { 263 qapi_event_send_migration(new_state, &error_abort); 264 } 265 } 266 267 /* 268 * Called on -incoming with a defer: uri. 269 * The migration can be started later after any parameters have been 270 * changed. 271 */ 272 static void deferred_incoming_migration(Error **errp) 273 { 274 if (deferred_incoming) { 275 error_setg(errp, "Incoming migration already deferred"); 276 } 277 deferred_incoming = true; 278 } 279 280 /* Request a range of pages from the source VM at the given 281 * start address. 282 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same 283 * as the last request (a name must have been given previously) 284 * Start: Address offset within the RB 285 * Len: Length in bytes required - must be a multiple of pagesize 286 */ 287 void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname, 288 ram_addr_t start, size_t len) 289 { 290 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */ 291 size_t msglen = 12; /* start + len */ 292 293 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start); 294 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len); 295 296 if (rbname) { 297 int rbname_len = strlen(rbname); 298 assert(rbname_len < 256); 299 300 bufc[msglen++] = rbname_len; 301 memcpy(bufc + msglen, rbname, rbname_len); 302 msglen += rbname_len; 303 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc); 304 } else { 305 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc); 306 } 307 } 308 309 void qemu_start_incoming_migration(const char *uri, Error **errp) 310 { 311 const char *p; 312 313 qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort); 314 if (!strcmp(uri, "defer")) { 315 deferred_incoming_migration(errp); 316 } else if (strstart(uri, "tcp:", &p)) { 317 tcp_start_incoming_migration(p, errp); 318 #ifdef CONFIG_RDMA 319 } else if (strstart(uri, "rdma:", &p)) { 320 rdma_start_incoming_migration(p, errp); 321 #endif 322 } else if (strstart(uri, "exec:", &p)) { 323 exec_start_incoming_migration(p, errp); 324 } else if (strstart(uri, "unix:", &p)) { 325 unix_start_incoming_migration(p, errp); 326 } else if (strstart(uri, "fd:", &p)) { 327 fd_start_incoming_migration(p, errp); 328 } else { 329 error_setg(errp, "unknown migration protocol: %s", uri); 330 } 331 } 332 333 static void process_incoming_migration_bh(void *opaque) 334 { 335 Error *local_err = NULL; 336 MigrationIncomingState *mis = opaque; 337 338 /* Make sure all file formats flush their mutable metadata. 339 * If we get an error here, just don't restart the VM yet. */ 340 bdrv_invalidate_cache_all(&local_err); 341 if (local_err) { 342 error_report_err(local_err); 343 local_err = NULL; 344 autostart = false; 345 } 346 347 /* 348 * This must happen after all error conditions are dealt with and 349 * we're sure the VM is going to be running on this host. 350 */ 351 qemu_announce_self(); 352 353 /* If global state section was not received or we are in running 354 state, we need to obey autostart. Any other state is set with 355 runstate_set. */ 356 357 if (!global_state_received() || 358 global_state_get_runstate() == RUN_STATE_RUNNING) { 359 if (autostart) { 360 vm_start(); 361 } else { 362 runstate_set(RUN_STATE_PAUSED); 363 } 364 } else { 365 runstate_set(global_state_get_runstate()); 366 } 367 migrate_decompress_threads_join(); 368 /* 369 * This must happen after any state changes since as soon as an external 370 * observer sees this event they might start to prod at the VM assuming 371 * it's ready to use. 372 */ 373 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, 374 MIGRATION_STATUS_COMPLETED); 375 qemu_bh_delete(mis->bh); 376 migration_incoming_state_destroy(); 377 } 378 379 static void process_incoming_migration_co(void *opaque) 380 { 381 QEMUFile *f = opaque; 382 MigrationIncomingState *mis = migration_incoming_get_current(); 383 PostcopyState ps; 384 int ret; 385 386 mis->from_src_file = f; 387 mis->largest_page_size = qemu_ram_pagesize_largest(); 388 postcopy_state_set(POSTCOPY_INCOMING_NONE); 389 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE, 390 MIGRATION_STATUS_ACTIVE); 391 ret = qemu_loadvm_state(f); 392 393 ps = postcopy_state_get(); 394 trace_process_incoming_migration_co_end(ret, ps); 395 if (ps != POSTCOPY_INCOMING_NONE) { 396 if (ps == POSTCOPY_INCOMING_ADVISE) { 397 /* 398 * Where a migration had postcopy enabled (and thus went to advise) 399 * but managed to complete within the precopy period, we can use 400 * the normal exit. 401 */ 402 postcopy_ram_incoming_cleanup(mis); 403 } else if (ret >= 0) { 404 /* 405 * Postcopy was started, cleanup should happen at the end of the 406 * postcopy thread. 407 */ 408 trace_process_incoming_migration_co_postcopy_end_main(); 409 return; 410 } 411 /* Else if something went wrong then just fall out of the normal exit */ 412 } 413 414 /* we get COLO info, and know if we are in COLO mode */ 415 if (!ret && migration_incoming_enable_colo()) { 416 mis->migration_incoming_co = qemu_coroutine_self(); 417 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming", 418 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE); 419 mis->have_colo_incoming_thread = true; 420 qemu_coroutine_yield(); 421 422 /* Wait checkpoint incoming thread exit before free resource */ 423 qemu_thread_join(&mis->colo_incoming_thread); 424 } 425 426 if (ret < 0) { 427 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, 428 MIGRATION_STATUS_FAILED); 429 error_report("load of migration failed: %s", strerror(-ret)); 430 migrate_decompress_threads_join(); 431 exit(EXIT_FAILURE); 432 } 433 434 qemu_fclose(f); 435 free_xbzrle_decoded_buf(); 436 437 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis); 438 qemu_bh_schedule(mis->bh); 439 } 440 441 void migration_fd_process_incoming(QEMUFile *f) 442 { 443 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, f); 444 445 migrate_decompress_threads_create(); 446 qemu_file_set_blocking(f, false); 447 qemu_coroutine_enter(co); 448 } 449 450 /* 451 * Send a message on the return channel back to the source 452 * of the migration. 453 */ 454 void migrate_send_rp_message(MigrationIncomingState *mis, 455 enum mig_rp_message_type message_type, 456 uint16_t len, void *data) 457 { 458 trace_migrate_send_rp_message((int)message_type, len); 459 qemu_mutex_lock(&mis->rp_mutex); 460 qemu_put_be16(mis->to_src_file, (unsigned int)message_type); 461 qemu_put_be16(mis->to_src_file, len); 462 qemu_put_buffer(mis->to_src_file, data, len); 463 qemu_fflush(mis->to_src_file); 464 qemu_mutex_unlock(&mis->rp_mutex); 465 } 466 467 /* 468 * Send a 'SHUT' message on the return channel with the given value 469 * to indicate that we've finished with the RP. Non-0 value indicates 470 * error. 471 */ 472 void migrate_send_rp_shut(MigrationIncomingState *mis, 473 uint32_t value) 474 { 475 uint32_t buf; 476 477 buf = cpu_to_be32(value); 478 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf); 479 } 480 481 /* 482 * Send a 'PONG' message on the return channel with the given value 483 * (normally in response to a 'PING') 484 */ 485 void migrate_send_rp_pong(MigrationIncomingState *mis, 486 uint32_t value) 487 { 488 uint32_t buf; 489 490 buf = cpu_to_be32(value); 491 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf); 492 } 493 494 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp) 495 { 496 MigrationCapabilityStatusList *head = NULL; 497 MigrationCapabilityStatusList *caps; 498 MigrationState *s = migrate_get_current(); 499 int i; 500 501 caps = NULL; /* silence compiler warning */ 502 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { 503 #ifndef CONFIG_LIVE_BLOCK_MIGRATION 504 if (i == MIGRATION_CAPABILITY_BLOCK) { 505 continue; 506 } 507 #endif 508 if (i == MIGRATION_CAPABILITY_X_COLO && !colo_supported()) { 509 continue; 510 } 511 if (head == NULL) { 512 head = g_malloc0(sizeof(*caps)); 513 caps = head; 514 } else { 515 caps->next = g_malloc0(sizeof(*caps)); 516 caps = caps->next; 517 } 518 caps->value = 519 g_malloc(sizeof(*caps->value)); 520 caps->value->capability = i; 521 caps->value->state = s->enabled_capabilities[i]; 522 } 523 524 return head; 525 } 526 527 MigrationParameters *qmp_query_migrate_parameters(Error **errp) 528 { 529 MigrationParameters *params; 530 MigrationState *s = migrate_get_current(); 531 532 params = g_malloc0(sizeof(*params)); 533 params->has_compress_level = true; 534 params->compress_level = s->parameters.compress_level; 535 params->has_compress_threads = true; 536 params->compress_threads = s->parameters.compress_threads; 537 params->has_decompress_threads = true; 538 params->decompress_threads = s->parameters.decompress_threads; 539 params->has_cpu_throttle_initial = true; 540 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial; 541 params->has_cpu_throttle_increment = true; 542 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment; 543 params->has_tls_creds = !!s->parameters.tls_creds; 544 params->tls_creds = g_strdup(s->parameters.tls_creds); 545 params->has_tls_hostname = !!s->parameters.tls_hostname; 546 params->tls_hostname = g_strdup(s->parameters.tls_hostname); 547 params->has_max_bandwidth = true; 548 params->max_bandwidth = s->parameters.max_bandwidth; 549 params->has_downtime_limit = true; 550 params->downtime_limit = s->parameters.downtime_limit; 551 params->has_x_checkpoint_delay = true; 552 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay; 553 params->has_block_incremental = true; 554 params->block_incremental = s->parameters.block_incremental; 555 556 return params; 557 } 558 559 /* 560 * Return true if we're already in the middle of a migration 561 * (i.e. any of the active or setup states) 562 */ 563 static bool migration_is_setup_or_active(int state) 564 { 565 switch (state) { 566 case MIGRATION_STATUS_ACTIVE: 567 case MIGRATION_STATUS_POSTCOPY_ACTIVE: 568 case MIGRATION_STATUS_SETUP: 569 return true; 570 571 default: 572 return false; 573 574 } 575 } 576 577 static void get_xbzrle_cache_stats(MigrationInfo *info) 578 { 579 if (migrate_use_xbzrle()) { 580 info->has_xbzrle_cache = true; 581 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache)); 582 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size(); 583 info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred(); 584 info->xbzrle_cache->pages = xbzrle_mig_pages_transferred(); 585 info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss(); 586 info->xbzrle_cache->cache_miss_rate = xbzrle_mig_cache_miss_rate(); 587 info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow(); 588 } 589 } 590 591 static void populate_ram_info(MigrationInfo *info, MigrationState *s) 592 { 593 info->has_ram = true; 594 info->ram = g_malloc0(sizeof(*info->ram)); 595 info->ram->transferred = ram_bytes_transferred(); 596 info->ram->total = ram_bytes_total(); 597 info->ram->duplicate = dup_mig_pages_transferred(); 598 /* legacy value. It is not used anymore */ 599 info->ram->skipped = 0; 600 info->ram->normal = norm_mig_pages_transferred(); 601 info->ram->normal_bytes = norm_mig_pages_transferred() * 602 qemu_target_page_size(); 603 info->ram->mbps = s->mbps; 604 info->ram->dirty_sync_count = ram_dirty_sync_count(); 605 info->ram->postcopy_requests = ram_postcopy_requests(); 606 info->ram->page_size = qemu_target_page_size(); 607 608 if (s->state != MIGRATION_STATUS_COMPLETED) { 609 info->ram->remaining = ram_bytes_remaining(); 610 info->ram->dirty_pages_rate = ram_dirty_pages_rate(); 611 } 612 } 613 614 MigrationInfo *qmp_query_migrate(Error **errp) 615 { 616 MigrationInfo *info = g_malloc0(sizeof(*info)); 617 MigrationState *s = migrate_get_current(); 618 619 switch (s->state) { 620 case MIGRATION_STATUS_NONE: 621 /* no migration has happened ever */ 622 break; 623 case MIGRATION_STATUS_SETUP: 624 info->has_status = true; 625 info->has_total_time = false; 626 break; 627 case MIGRATION_STATUS_ACTIVE: 628 case MIGRATION_STATUS_CANCELLING: 629 info->has_status = true; 630 info->has_total_time = true; 631 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) 632 - s->total_time; 633 info->has_expected_downtime = true; 634 info->expected_downtime = s->expected_downtime; 635 info->has_setup_time = true; 636 info->setup_time = s->setup_time; 637 638 populate_ram_info(info, s); 639 640 if (blk_mig_active()) { 641 info->has_disk = true; 642 info->disk = g_malloc0(sizeof(*info->disk)); 643 info->disk->transferred = blk_mig_bytes_transferred(); 644 info->disk->remaining = blk_mig_bytes_remaining(); 645 info->disk->total = blk_mig_bytes_total(); 646 } 647 648 if (cpu_throttle_active()) { 649 info->has_cpu_throttle_percentage = true; 650 info->cpu_throttle_percentage = cpu_throttle_get_percentage(); 651 } 652 653 get_xbzrle_cache_stats(info); 654 break; 655 case MIGRATION_STATUS_POSTCOPY_ACTIVE: 656 /* Mostly the same as active; TODO add some postcopy stats */ 657 info->has_status = true; 658 info->has_total_time = true; 659 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) 660 - s->total_time; 661 info->has_expected_downtime = true; 662 info->expected_downtime = s->expected_downtime; 663 info->has_setup_time = true; 664 info->setup_time = s->setup_time; 665 666 populate_ram_info(info, s); 667 668 if (blk_mig_active()) { 669 info->has_disk = true; 670 info->disk = g_malloc0(sizeof(*info->disk)); 671 info->disk->transferred = blk_mig_bytes_transferred(); 672 info->disk->remaining = blk_mig_bytes_remaining(); 673 info->disk->total = blk_mig_bytes_total(); 674 } 675 676 get_xbzrle_cache_stats(info); 677 break; 678 case MIGRATION_STATUS_COLO: 679 info->has_status = true; 680 /* TODO: display COLO specific information (checkpoint info etc.) */ 681 break; 682 case MIGRATION_STATUS_COMPLETED: 683 get_xbzrle_cache_stats(info); 684 685 info->has_status = true; 686 info->has_total_time = true; 687 info->total_time = s->total_time; 688 info->has_downtime = true; 689 info->downtime = s->downtime; 690 info->has_setup_time = true; 691 info->setup_time = s->setup_time; 692 693 populate_ram_info(info, s); 694 break; 695 case MIGRATION_STATUS_FAILED: 696 info->has_status = true; 697 if (s->error) { 698 info->has_error_desc = true; 699 info->error_desc = g_strdup(error_get_pretty(s->error)); 700 } 701 break; 702 case MIGRATION_STATUS_CANCELLED: 703 info->has_status = true; 704 break; 705 } 706 info->status = s->state; 707 708 return info; 709 } 710 711 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params, 712 Error **errp) 713 { 714 MigrationState *s = migrate_get_current(); 715 MigrationCapabilityStatusList *cap; 716 bool old_postcopy_cap = migrate_postcopy_ram(); 717 718 if (migration_is_setup_or_active(s->state)) { 719 error_setg(errp, QERR_MIGRATION_ACTIVE); 720 return; 721 } 722 723 for (cap = params; cap; cap = cap->next) { 724 #ifndef CONFIG_LIVE_BLOCK_MIGRATION 725 if (cap->value->capability == MIGRATION_CAPABILITY_BLOCK 726 && cap->value->state) { 727 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) " 728 "block migration"); 729 error_append_hint(errp, "Use drive_mirror+NBD instead.\n"); 730 continue; 731 } 732 #endif 733 if (cap->value->capability == MIGRATION_CAPABILITY_X_COLO) { 734 if (!colo_supported()) { 735 error_setg(errp, "COLO is not currently supported, please" 736 " configure with --enable-colo option in order to" 737 " support COLO feature"); 738 continue; 739 } 740 } 741 s->enabled_capabilities[cap->value->capability] = cap->value->state; 742 } 743 744 if (migrate_postcopy_ram()) { 745 if (migrate_use_compression()) { 746 /* The decompression threads asynchronously write into RAM 747 * rather than use the atomic copies needed to avoid 748 * userfaulting. It should be possible to fix the decompression 749 * threads for compatibility in future. 750 */ 751 error_report("Postcopy is not currently compatible with " 752 "compression"); 753 s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] = 754 false; 755 } 756 /* This check is reasonably expensive, so only when it's being 757 * set the first time, also it's only the destination that needs 758 * special support. 759 */ 760 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) && 761 !postcopy_ram_supported_by_host()) { 762 /* postcopy_ram_supported_by_host will have emitted a more 763 * detailed message 764 */ 765 error_report("Postcopy is not supported"); 766 s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM] = 767 false; 768 } 769 } 770 } 771 772 void qmp_migrate_set_parameters(MigrationParameters *params, Error **errp) 773 { 774 MigrationState *s = migrate_get_current(); 775 776 if (params->has_compress_level && 777 (params->compress_level < 0 || params->compress_level > 9)) { 778 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level", 779 "is invalid, it should be in the range of 0 to 9"); 780 return; 781 } 782 if (params->has_compress_threads && 783 (params->compress_threads < 1 || params->compress_threads > 255)) { 784 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, 785 "compress_threads", 786 "is invalid, it should be in the range of 1 to 255"); 787 return; 788 } 789 if (params->has_decompress_threads && 790 (params->decompress_threads < 1 || params->decompress_threads > 255)) { 791 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, 792 "decompress_threads", 793 "is invalid, it should be in the range of 1 to 255"); 794 return; 795 } 796 if (params->has_cpu_throttle_initial && 797 (params->cpu_throttle_initial < 1 || 798 params->cpu_throttle_initial > 99)) { 799 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, 800 "cpu_throttle_initial", 801 "an integer in the range of 1 to 99"); 802 return; 803 } 804 if (params->has_cpu_throttle_increment && 805 (params->cpu_throttle_increment < 1 || 806 params->cpu_throttle_increment > 99)) { 807 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, 808 "cpu_throttle_increment", 809 "an integer in the range of 1 to 99"); 810 return; 811 } 812 if (params->has_max_bandwidth && 813 (params->max_bandwidth < 0 || params->max_bandwidth > SIZE_MAX)) { 814 error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the" 815 " range of 0 to %zu bytes/second", SIZE_MAX); 816 return; 817 } 818 if (params->has_downtime_limit && 819 (params->downtime_limit < 0 || 820 params->downtime_limit > MAX_MIGRATE_DOWNTIME)) { 821 error_setg(errp, "Parameter 'downtime_limit' expects an integer in " 822 "the range of 0 to %d milliseconds", 823 MAX_MIGRATE_DOWNTIME); 824 return; 825 } 826 if (params->has_x_checkpoint_delay && (params->x_checkpoint_delay < 0)) { 827 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, 828 "x_checkpoint_delay", 829 "is invalid, it should be positive"); 830 } 831 832 if (params->has_compress_level) { 833 s->parameters.compress_level = params->compress_level; 834 } 835 if (params->has_compress_threads) { 836 s->parameters.compress_threads = params->compress_threads; 837 } 838 if (params->has_decompress_threads) { 839 s->parameters.decompress_threads = params->decompress_threads; 840 } 841 if (params->has_cpu_throttle_initial) { 842 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial; 843 } 844 if (params->has_cpu_throttle_increment) { 845 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment; 846 } 847 if (params->has_tls_creds) { 848 g_free(s->parameters.tls_creds); 849 s->parameters.tls_creds = g_strdup(params->tls_creds); 850 } 851 if (params->has_tls_hostname) { 852 g_free(s->parameters.tls_hostname); 853 s->parameters.tls_hostname = g_strdup(params->tls_hostname); 854 } 855 if (params->has_max_bandwidth) { 856 s->parameters.max_bandwidth = params->max_bandwidth; 857 if (s->to_dst_file) { 858 qemu_file_set_rate_limit(s->to_dst_file, 859 s->parameters.max_bandwidth / XFER_LIMIT_RATIO); 860 } 861 } 862 if (params->has_downtime_limit) { 863 s->parameters.downtime_limit = params->downtime_limit; 864 } 865 866 if (params->has_x_checkpoint_delay) { 867 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay; 868 if (migration_in_colo_state()) { 869 colo_checkpoint_notify(s); 870 } 871 } 872 if (params->has_block_incremental) { 873 s->parameters.block_incremental = params->block_incremental; 874 } 875 } 876 877 878 void qmp_migrate_start_postcopy(Error **errp) 879 { 880 MigrationState *s = migrate_get_current(); 881 882 if (!migrate_postcopy_ram()) { 883 error_setg(errp, "Enable postcopy with migrate_set_capability before" 884 " the start of migration"); 885 return; 886 } 887 888 if (s->state == MIGRATION_STATUS_NONE) { 889 error_setg(errp, "Postcopy must be started after migration has been" 890 " started"); 891 return; 892 } 893 /* 894 * we don't error if migration has finished since that would be racy 895 * with issuing this command. 896 */ 897 atomic_set(&s->start_postcopy, true); 898 } 899 900 /* shared migration helpers */ 901 902 void migrate_set_state(int *state, int old_state, int new_state) 903 { 904 if (atomic_cmpxchg(state, old_state, new_state) == old_state) { 905 trace_migrate_set_state(new_state); 906 migrate_generate_event(new_state); 907 } 908 } 909 910 void migrate_set_block_enabled(bool value, Error **errp) 911 { 912 MigrationCapabilityStatusList *cap; 913 914 cap = g_new0(MigrationCapabilityStatusList, 1); 915 cap->value = g_new0(MigrationCapabilityStatus, 1); 916 cap->value->capability = MIGRATION_CAPABILITY_BLOCK; 917 cap->value->state = value; 918 qmp_migrate_set_capabilities(cap, errp); 919 qapi_free_MigrationCapabilityStatusList(cap); 920 } 921 922 static void migrate_set_block_incremental(MigrationState *s, bool value) 923 { 924 s->parameters.block_incremental = value; 925 } 926 927 static void block_cleanup_parameters(MigrationState *s) 928 { 929 if (s->must_remove_block_options) { 930 /* setting to false can never fail */ 931 migrate_set_block_enabled(false, &error_abort); 932 migrate_set_block_incremental(s, false); 933 s->must_remove_block_options = false; 934 } 935 } 936 937 static void migrate_fd_cleanup(void *opaque) 938 { 939 MigrationState *s = opaque; 940 941 qemu_bh_delete(s->cleanup_bh); 942 s->cleanup_bh = NULL; 943 944 migration_page_queue_free(); 945 946 if (s->to_dst_file) { 947 trace_migrate_fd_cleanup(); 948 qemu_mutex_unlock_iothread(); 949 if (s->migration_thread_running) { 950 qemu_thread_join(&s->thread); 951 s->migration_thread_running = false; 952 } 953 qemu_mutex_lock_iothread(); 954 955 migrate_compress_threads_join(); 956 qemu_fclose(s->to_dst_file); 957 s->to_dst_file = NULL; 958 } 959 960 assert((s->state != MIGRATION_STATUS_ACTIVE) && 961 (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE)); 962 963 if (s->state == MIGRATION_STATUS_CANCELLING) { 964 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING, 965 MIGRATION_STATUS_CANCELLED); 966 } 967 968 notifier_list_notify(&migration_state_notifiers, s); 969 block_cleanup_parameters(s); 970 } 971 972 void migrate_fd_error(MigrationState *s, const Error *error) 973 { 974 trace_migrate_fd_error(error_get_pretty(error)); 975 assert(s->to_dst_file == NULL); 976 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, 977 MIGRATION_STATUS_FAILED); 978 if (!s->error) { 979 s->error = error_copy(error); 980 } 981 notifier_list_notify(&migration_state_notifiers, s); 982 block_cleanup_parameters(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 block_cleanup_parameters(s); 1025 } 1026 1027 void add_migration_state_change_notifier(Notifier *notify) 1028 { 1029 notifier_list_add(&migration_state_notifiers, notify); 1030 } 1031 1032 void remove_migration_state_change_notifier(Notifier *notify) 1033 { 1034 notifier_remove(notify); 1035 } 1036 1037 bool migration_in_setup(MigrationState *s) 1038 { 1039 return s->state == MIGRATION_STATUS_SETUP; 1040 } 1041 1042 bool migration_has_finished(MigrationState *s) 1043 { 1044 return s->state == MIGRATION_STATUS_COMPLETED; 1045 } 1046 1047 bool migration_has_failed(MigrationState *s) 1048 { 1049 return (s->state == MIGRATION_STATUS_CANCELLED || 1050 s->state == MIGRATION_STATUS_FAILED); 1051 } 1052 1053 bool migration_in_postcopy(void) 1054 { 1055 MigrationState *s = migrate_get_current(); 1056 1057 return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); 1058 } 1059 1060 bool migration_in_postcopy_after_devices(MigrationState *s) 1061 { 1062 return migration_in_postcopy() && s->postcopy_after_devices; 1063 } 1064 1065 bool migration_is_idle(void) 1066 { 1067 MigrationState *s = migrate_get_current(); 1068 1069 switch (s->state) { 1070 case MIGRATION_STATUS_NONE: 1071 case MIGRATION_STATUS_CANCELLED: 1072 case MIGRATION_STATUS_COMPLETED: 1073 case MIGRATION_STATUS_FAILED: 1074 return true; 1075 case MIGRATION_STATUS_SETUP: 1076 case MIGRATION_STATUS_CANCELLING: 1077 case MIGRATION_STATUS_ACTIVE: 1078 case MIGRATION_STATUS_POSTCOPY_ACTIVE: 1079 case MIGRATION_STATUS_COLO: 1080 return false; 1081 case MIGRATION_STATUS__MAX: 1082 g_assert_not_reached(); 1083 } 1084 1085 return false; 1086 } 1087 1088 MigrationState *migrate_init(void) 1089 { 1090 MigrationState *s = migrate_get_current(); 1091 1092 /* 1093 * Reinitialise all migration state, except 1094 * parameters/capabilities that the user set, and 1095 * locks. 1096 */ 1097 s->bytes_xfer = 0; 1098 s->xfer_limit = 0; 1099 s->cleanup_bh = 0; 1100 s->to_dst_file = NULL; 1101 s->state = MIGRATION_STATUS_NONE; 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->setup_time = 0; 1108 s->start_postcopy = false; 1109 s->postcopy_after_devices = false; 1110 s->migration_thread_running = false; 1111 error_free(s->error); 1112 s->error = NULL; 1113 1114 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP); 1115 1116 s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 1117 return s; 1118 } 1119 1120 static GSList *migration_blockers; 1121 1122 int migrate_add_blocker(Error *reason, Error **errp) 1123 { 1124 if (only_migratable) { 1125 error_propagate(errp, error_copy(reason)); 1126 error_prepend(errp, "disallowing migration blocker " 1127 "(--only_migratable) for: "); 1128 return -EACCES; 1129 } 1130 1131 if (migration_is_idle()) { 1132 migration_blockers = g_slist_prepend(migration_blockers, reason); 1133 return 0; 1134 } 1135 1136 error_propagate(errp, error_copy(reason)); 1137 error_prepend(errp, "disallowing migration blocker (migration in " 1138 "progress) for: "); 1139 return -EBUSY; 1140 } 1141 1142 void migrate_del_blocker(Error *reason) 1143 { 1144 migration_blockers = g_slist_remove(migration_blockers, reason); 1145 } 1146 1147 void qmp_migrate_incoming(const char *uri, Error **errp) 1148 { 1149 Error *local_err = NULL; 1150 static bool once = true; 1151 1152 if (!deferred_incoming) { 1153 error_setg(errp, "For use with '-incoming defer'"); 1154 return; 1155 } 1156 if (!once) { 1157 error_setg(errp, "The incoming migration has already been started"); 1158 } 1159 1160 qemu_start_incoming_migration(uri, &local_err); 1161 1162 if (local_err) { 1163 error_propagate(errp, local_err); 1164 return; 1165 } 1166 1167 once = false; 1168 } 1169 1170 bool migration_is_blocked(Error **errp) 1171 { 1172 if (qemu_savevm_state_blocked(errp)) { 1173 return true; 1174 } 1175 1176 if (migration_blockers) { 1177 *errp = error_copy(migration_blockers->data); 1178 return true; 1179 } 1180 1181 return false; 1182 } 1183 1184 void qmp_migrate(const char *uri, bool has_blk, bool blk, 1185 bool has_inc, bool inc, bool has_detach, bool detach, 1186 Error **errp) 1187 { 1188 Error *local_err = NULL; 1189 MigrationState *s = migrate_get_current(); 1190 const char *p; 1191 1192 if (migration_is_setup_or_active(s->state) || 1193 s->state == MIGRATION_STATUS_CANCELLING || 1194 s->state == MIGRATION_STATUS_COLO) { 1195 error_setg(errp, QERR_MIGRATION_ACTIVE); 1196 return; 1197 } 1198 if (runstate_check(RUN_STATE_INMIGRATE)) { 1199 error_setg(errp, "Guest is waiting for an incoming migration"); 1200 return; 1201 } 1202 1203 if (migration_is_blocked(errp)) { 1204 return; 1205 } 1206 1207 if ((has_blk && blk) || (has_inc && inc)) { 1208 if (migrate_use_block() || migrate_use_block_incremental()) { 1209 error_setg(errp, "Command options are incompatible with " 1210 "current migration capabilities"); 1211 return; 1212 } 1213 migrate_set_block_enabled(true, &local_err); 1214 if (local_err) { 1215 error_propagate(errp, local_err); 1216 return; 1217 } 1218 s->must_remove_block_options = true; 1219 } 1220 1221 if (has_inc && inc) { 1222 migrate_set_block_incremental(s, true); 1223 } 1224 1225 s = migrate_init(); 1226 1227 if (strstart(uri, "tcp:", &p)) { 1228 tcp_start_outgoing_migration(s, p, &local_err); 1229 #ifdef CONFIG_RDMA 1230 } else if (strstart(uri, "rdma:", &p)) { 1231 rdma_start_outgoing_migration(s, p, &local_err); 1232 #endif 1233 } else if (strstart(uri, "exec:", &p)) { 1234 exec_start_outgoing_migration(s, p, &local_err); 1235 } else if (strstart(uri, "unix:", &p)) { 1236 unix_start_outgoing_migration(s, p, &local_err); 1237 } else if (strstart(uri, "fd:", &p)) { 1238 fd_start_outgoing_migration(s, p, &local_err); 1239 } else { 1240 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri", 1241 "a valid migration protocol"); 1242 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, 1243 MIGRATION_STATUS_FAILED); 1244 return; 1245 } 1246 1247 if (local_err) { 1248 migrate_fd_error(s, local_err); 1249 error_propagate(errp, local_err); 1250 return; 1251 } 1252 } 1253 1254 void qmp_migrate_cancel(Error **errp) 1255 { 1256 migrate_fd_cancel(migrate_get_current()); 1257 } 1258 1259 void qmp_migrate_set_cache_size(int64_t value, Error **errp) 1260 { 1261 MigrationState *s = migrate_get_current(); 1262 int64_t new_size; 1263 1264 /* Check for truncation */ 1265 if (value != (size_t)value) { 1266 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", 1267 "exceeding address space"); 1268 return; 1269 } 1270 1271 /* Cache should not be larger than guest ram size */ 1272 if (value > ram_bytes_total()) { 1273 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", 1274 "exceeds guest ram size "); 1275 return; 1276 } 1277 1278 new_size = xbzrle_cache_resize(value); 1279 if (new_size < 0) { 1280 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "cache size", 1281 "is smaller than page size"); 1282 return; 1283 } 1284 1285 s->xbzrle_cache_size = new_size; 1286 } 1287 1288 int64_t qmp_query_migrate_cache_size(Error **errp) 1289 { 1290 return migrate_xbzrle_cache_size(); 1291 } 1292 1293 void qmp_migrate_set_speed(int64_t value, Error **errp) 1294 { 1295 MigrationParameters p = { 1296 .has_max_bandwidth = true, 1297 .max_bandwidth = value, 1298 }; 1299 1300 qmp_migrate_set_parameters(&p, errp); 1301 } 1302 1303 void qmp_migrate_set_downtime(double value, Error **errp) 1304 { 1305 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) { 1306 error_setg(errp, "Parameter 'downtime_limit' expects an integer in " 1307 "the range of 0 to %d seconds", 1308 MAX_MIGRATE_DOWNTIME_SECONDS); 1309 return; 1310 } 1311 1312 value *= 1000; /* Convert to milliseconds */ 1313 value = MAX(0, MIN(INT64_MAX, value)); 1314 1315 MigrationParameters p = { 1316 .has_downtime_limit = true, 1317 .downtime_limit = value, 1318 }; 1319 1320 qmp_migrate_set_parameters(&p, errp); 1321 } 1322 1323 bool migrate_release_ram(void) 1324 { 1325 MigrationState *s; 1326 1327 s = migrate_get_current(); 1328 1329 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM]; 1330 } 1331 1332 bool migrate_postcopy_ram(void) 1333 { 1334 MigrationState *s; 1335 1336 s = migrate_get_current(); 1337 1338 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM]; 1339 } 1340 1341 bool migrate_auto_converge(void) 1342 { 1343 MigrationState *s; 1344 1345 s = migrate_get_current(); 1346 1347 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE]; 1348 } 1349 1350 bool migrate_zero_blocks(void) 1351 { 1352 MigrationState *s; 1353 1354 s = migrate_get_current(); 1355 1356 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS]; 1357 } 1358 1359 bool migrate_use_compression(void) 1360 { 1361 MigrationState *s; 1362 1363 s = migrate_get_current(); 1364 1365 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS]; 1366 } 1367 1368 int migrate_compress_level(void) 1369 { 1370 MigrationState *s; 1371 1372 s = migrate_get_current(); 1373 1374 return s->parameters.compress_level; 1375 } 1376 1377 int migrate_compress_threads(void) 1378 { 1379 MigrationState *s; 1380 1381 s = migrate_get_current(); 1382 1383 return s->parameters.compress_threads; 1384 } 1385 1386 int migrate_decompress_threads(void) 1387 { 1388 MigrationState *s; 1389 1390 s = migrate_get_current(); 1391 1392 return s->parameters.decompress_threads; 1393 } 1394 1395 bool migrate_use_events(void) 1396 { 1397 MigrationState *s; 1398 1399 s = migrate_get_current(); 1400 1401 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS]; 1402 } 1403 1404 int migrate_use_xbzrle(void) 1405 { 1406 MigrationState *s; 1407 1408 s = migrate_get_current(); 1409 1410 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE]; 1411 } 1412 1413 int64_t migrate_xbzrle_cache_size(void) 1414 { 1415 MigrationState *s; 1416 1417 s = migrate_get_current(); 1418 1419 return s->xbzrle_cache_size; 1420 } 1421 1422 bool migrate_use_block(void) 1423 { 1424 MigrationState *s; 1425 1426 s = migrate_get_current(); 1427 1428 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK]; 1429 } 1430 1431 bool migrate_use_block_incremental(void) 1432 { 1433 MigrationState *s; 1434 1435 s = migrate_get_current(); 1436 1437 return s->parameters.block_incremental; 1438 } 1439 1440 /* migration thread support */ 1441 /* 1442 * Something bad happened to the RP stream, mark an error 1443 * The caller shall print or trace something to indicate why 1444 */ 1445 static void mark_source_rp_bad(MigrationState *s) 1446 { 1447 s->rp_state.error = true; 1448 } 1449 1450 static struct rp_cmd_args { 1451 ssize_t len; /* -1 = variable */ 1452 const char *name; 1453 } rp_cmd_args[] = { 1454 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" }, 1455 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" }, 1456 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" }, 1457 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" }, 1458 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" }, 1459 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" }, 1460 }; 1461 1462 /* 1463 * Process a request for pages received on the return path, 1464 * We're allowed to send more than requested (e.g. to round to our page size) 1465 * and we don't need to send pages that have already been sent. 1466 */ 1467 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname, 1468 ram_addr_t start, size_t len) 1469 { 1470 long our_host_ps = getpagesize(); 1471 1472 trace_migrate_handle_rp_req_pages(rbname, start, len); 1473 1474 /* 1475 * Since we currently insist on matching page sizes, just sanity check 1476 * we're being asked for whole host pages. 1477 */ 1478 if (start & (our_host_ps-1) || 1479 (len & (our_host_ps-1))) { 1480 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT 1481 " len: %zd", __func__, start, len); 1482 mark_source_rp_bad(ms); 1483 return; 1484 } 1485 1486 if (ram_save_queue_pages(rbname, start, len)) { 1487 mark_source_rp_bad(ms); 1488 } 1489 } 1490 1491 /* 1492 * Handles messages sent on the return path towards the source VM 1493 * 1494 */ 1495 static void *source_return_path_thread(void *opaque) 1496 { 1497 MigrationState *ms = opaque; 1498 QEMUFile *rp = ms->rp_state.from_dst_file; 1499 uint16_t header_len, header_type; 1500 uint8_t buf[512]; 1501 uint32_t tmp32, sibling_error; 1502 ram_addr_t start = 0; /* =0 to silence warning */ 1503 size_t len = 0, expected_len; 1504 int res; 1505 1506 trace_source_return_path_thread_entry(); 1507 while (!ms->rp_state.error && !qemu_file_get_error(rp) && 1508 migration_is_setup_or_active(ms->state)) { 1509 trace_source_return_path_thread_loop_top(); 1510 header_type = qemu_get_be16(rp); 1511 header_len = qemu_get_be16(rp); 1512 1513 if (header_type >= MIG_RP_MSG_MAX || 1514 header_type == MIG_RP_MSG_INVALID) { 1515 error_report("RP: Received invalid message 0x%04x length 0x%04x", 1516 header_type, header_len); 1517 mark_source_rp_bad(ms); 1518 goto out; 1519 } 1520 1521 if ((rp_cmd_args[header_type].len != -1 && 1522 header_len != rp_cmd_args[header_type].len) || 1523 header_len > sizeof(buf)) { 1524 error_report("RP: Received '%s' message (0x%04x) with" 1525 "incorrect length %d expecting %zu", 1526 rp_cmd_args[header_type].name, header_type, header_len, 1527 (size_t)rp_cmd_args[header_type].len); 1528 mark_source_rp_bad(ms); 1529 goto out; 1530 } 1531 1532 /* We know we've got a valid header by this point */ 1533 res = qemu_get_buffer(rp, buf, header_len); 1534 if (res != header_len) { 1535 error_report("RP: Failed reading data for message 0x%04x" 1536 " read %d expected %d", 1537 header_type, res, header_len); 1538 mark_source_rp_bad(ms); 1539 goto out; 1540 } 1541 1542 /* OK, we have the message and the data */ 1543 switch (header_type) { 1544 case MIG_RP_MSG_SHUT: 1545 sibling_error = ldl_be_p(buf); 1546 trace_source_return_path_thread_shut(sibling_error); 1547 if (sibling_error) { 1548 error_report("RP: Sibling indicated error %d", sibling_error); 1549 mark_source_rp_bad(ms); 1550 } 1551 /* 1552 * We'll let the main thread deal with closing the RP 1553 * we could do a shutdown(2) on it, but we're the only user 1554 * anyway, so there's nothing gained. 1555 */ 1556 goto out; 1557 1558 case MIG_RP_MSG_PONG: 1559 tmp32 = ldl_be_p(buf); 1560 trace_source_return_path_thread_pong(tmp32); 1561 break; 1562 1563 case MIG_RP_MSG_REQ_PAGES: 1564 start = ldq_be_p(buf); 1565 len = ldl_be_p(buf + 8); 1566 migrate_handle_rp_req_pages(ms, NULL, start, len); 1567 break; 1568 1569 case MIG_RP_MSG_REQ_PAGES_ID: 1570 expected_len = 12 + 1; /* header + termination */ 1571 1572 if (header_len >= expected_len) { 1573 start = ldq_be_p(buf); 1574 len = ldl_be_p(buf + 8); 1575 /* Now we expect an idstr */ 1576 tmp32 = buf[12]; /* Length of the following idstr */ 1577 buf[13 + tmp32] = '\0'; 1578 expected_len += tmp32; 1579 } 1580 if (header_len != expected_len) { 1581 error_report("RP: Req_Page_id with length %d expecting %zd", 1582 header_len, expected_len); 1583 mark_source_rp_bad(ms); 1584 goto out; 1585 } 1586 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len); 1587 break; 1588 1589 default: 1590 break; 1591 } 1592 } 1593 if (qemu_file_get_error(rp)) { 1594 trace_source_return_path_thread_bad_end(); 1595 mark_source_rp_bad(ms); 1596 } 1597 1598 trace_source_return_path_thread_end(); 1599 out: 1600 ms->rp_state.from_dst_file = NULL; 1601 qemu_fclose(rp); 1602 return NULL; 1603 } 1604 1605 static int open_return_path_on_source(MigrationState *ms) 1606 { 1607 1608 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file); 1609 if (!ms->rp_state.from_dst_file) { 1610 return -1; 1611 } 1612 1613 trace_open_return_path_on_source(); 1614 qemu_thread_create(&ms->rp_state.rp_thread, "return path", 1615 source_return_path_thread, ms, QEMU_THREAD_JOINABLE); 1616 1617 trace_open_return_path_on_source_continue(); 1618 1619 return 0; 1620 } 1621 1622 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */ 1623 static int await_return_path_close_on_source(MigrationState *ms) 1624 { 1625 /* 1626 * If this is a normal exit then the destination will send a SHUT and the 1627 * rp_thread will exit, however if there's an error we need to cause 1628 * it to exit. 1629 */ 1630 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) { 1631 /* 1632 * shutdown(2), if we have it, will cause it to unblock if it's stuck 1633 * waiting for the destination. 1634 */ 1635 qemu_file_shutdown(ms->rp_state.from_dst_file); 1636 mark_source_rp_bad(ms); 1637 } 1638 trace_await_return_path_close_on_source_joining(); 1639 qemu_thread_join(&ms->rp_state.rp_thread); 1640 trace_await_return_path_close_on_source_close(); 1641 return ms->rp_state.error; 1642 } 1643 1644 /* 1645 * Switch from normal iteration to postcopy 1646 * Returns non-0 on error 1647 */ 1648 static int postcopy_start(MigrationState *ms, bool *old_vm_running) 1649 { 1650 int ret; 1651 QIOChannelBuffer *bioc; 1652 QEMUFile *fb; 1653 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 1654 bool restart_block = false; 1655 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE, 1656 MIGRATION_STATUS_POSTCOPY_ACTIVE); 1657 1658 trace_postcopy_start(); 1659 qemu_mutex_lock_iothread(); 1660 trace_postcopy_start_set_run(); 1661 1662 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER); 1663 *old_vm_running = runstate_is_running(); 1664 global_state_store(); 1665 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); 1666 if (ret < 0) { 1667 goto fail; 1668 } 1669 1670 ret = bdrv_inactivate_all(); 1671 if (ret < 0) { 1672 goto fail; 1673 } 1674 restart_block = true; 1675 1676 /* 1677 * Cause any non-postcopiable, but iterative devices to 1678 * send out their final data. 1679 */ 1680 qemu_savevm_state_complete_precopy(ms->to_dst_file, true); 1681 1682 /* 1683 * in Finish migrate and with the io-lock held everything should 1684 * be quiet, but we've potentially still got dirty pages and we 1685 * need to tell the destination to throw any pages it's already received 1686 * that are dirty 1687 */ 1688 if (ram_postcopy_send_discard_bitmap(ms)) { 1689 error_report("postcopy send discard bitmap failed"); 1690 goto fail; 1691 } 1692 1693 /* 1694 * send rest of state - note things that are doing postcopy 1695 * will notice we're in POSTCOPY_ACTIVE and not actually 1696 * wrap their state up here 1697 */ 1698 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX); 1699 /* Ping just for debugging, helps line traces up */ 1700 qemu_savevm_send_ping(ms->to_dst_file, 2); 1701 1702 /* 1703 * While loading the device state we may trigger page transfer 1704 * requests and the fd must be free to process those, and thus 1705 * the destination must read the whole device state off the fd before 1706 * it starts processing it. Unfortunately the ad-hoc migration format 1707 * doesn't allow the destination to know the size to read without fully 1708 * parsing it through each devices load-state code (especially the open 1709 * coded devices that use get/put). 1710 * So we wrap the device state up in a package with a length at the start; 1711 * to do this we use a qemu_buf to hold the whole of the device state. 1712 */ 1713 bioc = qio_channel_buffer_new(4096); 1714 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer"); 1715 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc)); 1716 object_unref(OBJECT(bioc)); 1717 1718 /* 1719 * Make sure the receiver can get incoming pages before we send the rest 1720 * of the state 1721 */ 1722 qemu_savevm_send_postcopy_listen(fb); 1723 1724 qemu_savevm_state_complete_precopy(fb, false); 1725 qemu_savevm_send_ping(fb, 3); 1726 1727 qemu_savevm_send_postcopy_run(fb); 1728 1729 /* <><> end of stuff going into the package */ 1730 1731 /* Last point of recovery; as soon as we send the package the destination 1732 * can open devices and potentially start running. 1733 * Lets just check again we've not got any errors. 1734 */ 1735 ret = qemu_file_get_error(ms->to_dst_file); 1736 if (ret) { 1737 error_report("postcopy_start: Migration stream errored (pre package)"); 1738 goto fail_closefb; 1739 } 1740 1741 restart_block = false; 1742 1743 /* Now send that blob */ 1744 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) { 1745 goto fail_closefb; 1746 } 1747 qemu_fclose(fb); 1748 1749 /* Send a notify to give a chance for anything that needs to happen 1750 * at the transition to postcopy and after the device state; in particular 1751 * spice needs to trigger a transition now 1752 */ 1753 ms->postcopy_after_devices = true; 1754 notifier_list_notify(&migration_state_notifiers, ms); 1755 1756 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop; 1757 1758 qemu_mutex_unlock_iothread(); 1759 1760 /* 1761 * Although this ping is just for debug, it could potentially be 1762 * used for getting a better measurement of downtime at the source. 1763 */ 1764 qemu_savevm_send_ping(ms->to_dst_file, 4); 1765 1766 if (migrate_release_ram()) { 1767 ram_postcopy_migrated_memory_release(ms); 1768 } 1769 1770 ret = qemu_file_get_error(ms->to_dst_file); 1771 if (ret) { 1772 error_report("postcopy_start: Migration stream errored"); 1773 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, 1774 MIGRATION_STATUS_FAILED); 1775 } 1776 1777 return ret; 1778 1779 fail_closefb: 1780 qemu_fclose(fb); 1781 fail: 1782 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, 1783 MIGRATION_STATUS_FAILED); 1784 if (restart_block) { 1785 /* A failure happened early enough that we know the destination hasn't 1786 * accessed block devices, so we're safe to recover. 1787 */ 1788 Error *local_err = NULL; 1789 1790 bdrv_invalidate_cache_all(&local_err); 1791 if (local_err) { 1792 error_report_err(local_err); 1793 } 1794 } 1795 qemu_mutex_unlock_iothread(); 1796 return -1; 1797 } 1798 1799 /** 1800 * migration_completion: Used by migration_thread when there's not much left. 1801 * The caller 'breaks' the loop when this returns. 1802 * 1803 * @s: Current migration state 1804 * @current_active_state: The migration state we expect to be in 1805 * @*old_vm_running: Pointer to old_vm_running flag 1806 * @*start_time: Pointer to time to update 1807 */ 1808 static void migration_completion(MigrationState *s, int current_active_state, 1809 bool *old_vm_running, 1810 int64_t *start_time) 1811 { 1812 int ret; 1813 1814 if (s->state == MIGRATION_STATUS_ACTIVE) { 1815 qemu_mutex_lock_iothread(); 1816 *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 1817 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER); 1818 *old_vm_running = runstate_is_running(); 1819 ret = global_state_store(); 1820 1821 if (!ret) { 1822 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); 1823 /* 1824 * Don't mark the image with BDRV_O_INACTIVE flag if 1825 * we will go into COLO stage later. 1826 */ 1827 if (ret >= 0 && !migrate_colo_enabled()) { 1828 ret = bdrv_inactivate_all(); 1829 } 1830 if (ret >= 0) { 1831 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX); 1832 qemu_savevm_state_complete_precopy(s->to_dst_file, false); 1833 s->block_inactive = true; 1834 } 1835 } 1836 qemu_mutex_unlock_iothread(); 1837 1838 if (ret < 0) { 1839 goto fail; 1840 } 1841 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { 1842 trace_migration_completion_postcopy_end(); 1843 1844 qemu_savevm_state_complete_postcopy(s->to_dst_file); 1845 trace_migration_completion_postcopy_end_after_complete(); 1846 } 1847 1848 /* 1849 * If rp was opened we must clean up the thread before 1850 * cleaning everything else up (since if there are no failures 1851 * it will wait for the destination to send it's status in 1852 * a SHUT command). 1853 * Postcopy opens rp if enabled (even if it's not avtivated) 1854 */ 1855 if (migrate_postcopy_ram()) { 1856 int rp_error; 1857 trace_migration_completion_postcopy_end_before_rp(); 1858 rp_error = await_return_path_close_on_source(s); 1859 trace_migration_completion_postcopy_end_after_rp(rp_error); 1860 if (rp_error) { 1861 goto fail_invalidate; 1862 } 1863 } 1864 1865 if (qemu_file_get_error(s->to_dst_file)) { 1866 trace_migration_completion_file_err(); 1867 goto fail_invalidate; 1868 } 1869 1870 if (!migrate_colo_enabled()) { 1871 migrate_set_state(&s->state, current_active_state, 1872 MIGRATION_STATUS_COMPLETED); 1873 } 1874 1875 return; 1876 1877 fail_invalidate: 1878 /* If not doing postcopy, vm_start() will be called: let's regain 1879 * control on images. 1880 */ 1881 if (s->state == MIGRATION_STATUS_ACTIVE) { 1882 Error *local_err = NULL; 1883 1884 qemu_mutex_lock_iothread(); 1885 bdrv_invalidate_cache_all(&local_err); 1886 if (local_err) { 1887 error_report_err(local_err); 1888 } else { 1889 s->block_inactive = false; 1890 } 1891 qemu_mutex_unlock_iothread(); 1892 } 1893 1894 fail: 1895 migrate_set_state(&s->state, current_active_state, 1896 MIGRATION_STATUS_FAILED); 1897 } 1898 1899 bool migrate_colo_enabled(void) 1900 { 1901 MigrationState *s = migrate_get_current(); 1902 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO]; 1903 } 1904 1905 /* 1906 * Master migration thread on the source VM. 1907 * It drives the migration and pumps the data down the outgoing channel. 1908 */ 1909 static void *migration_thread(void *opaque) 1910 { 1911 MigrationState *s = opaque; 1912 /* Used by the bandwidth calcs, updated later */ 1913 int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 1914 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); 1915 int64_t initial_bytes = 0; 1916 /* 1917 * The final stage happens when the remaining data is smaller than 1918 * this threshold; it's calculated from the requested downtime and 1919 * measured bandwidth 1920 */ 1921 int64_t threshold_size = 0; 1922 int64_t start_time = initial_time; 1923 int64_t end_time; 1924 bool old_vm_running = false; 1925 bool entered_postcopy = false; 1926 /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */ 1927 enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE; 1928 bool enable_colo = migrate_colo_enabled(); 1929 1930 rcu_register_thread(); 1931 1932 qemu_savevm_state_header(s->to_dst_file); 1933 1934 if (migrate_postcopy_ram()) { 1935 /* Now tell the dest that it should open its end so it can reply */ 1936 qemu_savevm_send_open_return_path(s->to_dst_file); 1937 1938 /* And do a ping that will make stuff easier to debug */ 1939 qemu_savevm_send_ping(s->to_dst_file, 1); 1940 1941 /* 1942 * Tell the destination that we *might* want to do postcopy later; 1943 * if the other end can't do postcopy it should fail now, nice and 1944 * early. 1945 */ 1946 qemu_savevm_send_postcopy_advise(s->to_dst_file); 1947 } 1948 1949 qemu_savevm_state_begin(s->to_dst_file); 1950 1951 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; 1952 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, 1953 MIGRATION_STATUS_ACTIVE); 1954 1955 trace_migration_thread_setup_complete(); 1956 1957 while (s->state == MIGRATION_STATUS_ACTIVE || 1958 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { 1959 int64_t current_time; 1960 uint64_t pending_size; 1961 1962 if (!qemu_file_rate_limit(s->to_dst_file)) { 1963 uint64_t pend_post, pend_nonpost; 1964 1965 qemu_savevm_state_pending(s->to_dst_file, threshold_size, 1966 &pend_nonpost, &pend_post); 1967 pending_size = pend_nonpost + pend_post; 1968 trace_migrate_pending(pending_size, threshold_size, 1969 pend_post, pend_nonpost); 1970 if (pending_size && pending_size >= threshold_size) { 1971 /* Still a significant amount to transfer */ 1972 1973 if (migrate_postcopy_ram() && 1974 s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE && 1975 pend_nonpost <= threshold_size && 1976 atomic_read(&s->start_postcopy)) { 1977 1978 if (!postcopy_start(s, &old_vm_running)) { 1979 current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE; 1980 entered_postcopy = true; 1981 } 1982 1983 continue; 1984 } 1985 /* Just another iteration step */ 1986 qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy); 1987 } else { 1988 trace_migration_thread_low_pending(pending_size); 1989 migration_completion(s, current_active_state, 1990 &old_vm_running, &start_time); 1991 break; 1992 } 1993 } 1994 1995 if (qemu_file_get_error(s->to_dst_file)) { 1996 migrate_set_state(&s->state, current_active_state, 1997 MIGRATION_STATUS_FAILED); 1998 trace_migration_thread_file_err(); 1999 break; 2000 } 2001 current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 2002 if (current_time >= initial_time + BUFFER_DELAY) { 2003 uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) - 2004 initial_bytes; 2005 uint64_t time_spent = current_time - initial_time; 2006 double bandwidth = (double)transferred_bytes / time_spent; 2007 threshold_size = bandwidth * s->parameters.downtime_limit; 2008 2009 s->mbps = (((double) transferred_bytes * 8.0) / 2010 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0; 2011 2012 trace_migrate_transferred(transferred_bytes, time_spent, 2013 bandwidth, threshold_size); 2014 /* if we haven't sent anything, we don't want to recalculate 2015 10000 is a small enough number for our purposes */ 2016 if (ram_dirty_pages_rate() && transferred_bytes > 10000) { 2017 s->expected_downtime = ram_dirty_pages_rate() * 2018 qemu_target_page_size() / bandwidth; 2019 } 2020 2021 qemu_file_reset_rate_limit(s->to_dst_file); 2022 initial_time = current_time; 2023 initial_bytes = qemu_ftell(s->to_dst_file); 2024 } 2025 if (qemu_file_rate_limit(s->to_dst_file)) { 2026 /* usleep expects microseconds */ 2027 g_usleep((initial_time + BUFFER_DELAY - current_time)*1000); 2028 } 2029 } 2030 2031 trace_migration_thread_after_loop(); 2032 /* If we enabled cpu throttling for auto-converge, turn it off. */ 2033 cpu_throttle_stop(); 2034 end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 2035 2036 qemu_mutex_lock_iothread(); 2037 /* 2038 * The resource has been allocated by migration will be reused in COLO 2039 * process, so don't release them. 2040 */ 2041 if (!enable_colo) { 2042 qemu_savevm_state_cleanup(); 2043 } 2044 if (s->state == MIGRATION_STATUS_COMPLETED) { 2045 uint64_t transferred_bytes = qemu_ftell(s->to_dst_file); 2046 s->total_time = end_time - s->total_time; 2047 if (!entered_postcopy) { 2048 s->downtime = end_time - start_time; 2049 } 2050 if (s->total_time) { 2051 s->mbps = (((double) transferred_bytes * 8.0) / 2052 ((double) s->total_time)) / 1000; 2053 } 2054 runstate_set(RUN_STATE_POSTMIGRATE); 2055 } else { 2056 if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) { 2057 migrate_start_colo_process(s); 2058 qemu_savevm_state_cleanup(); 2059 /* 2060 * Fixme: we will run VM in COLO no matter its old running state. 2061 * After exited COLO, we will keep running. 2062 */ 2063 old_vm_running = true; 2064 } 2065 if (old_vm_running && !entered_postcopy) { 2066 vm_start(); 2067 } else { 2068 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) { 2069 runstate_set(RUN_STATE_POSTMIGRATE); 2070 } 2071 } 2072 } 2073 qemu_bh_schedule(s->cleanup_bh); 2074 qemu_mutex_unlock_iothread(); 2075 2076 rcu_unregister_thread(); 2077 return NULL; 2078 } 2079 2080 void migrate_fd_connect(MigrationState *s) 2081 { 2082 s->expected_downtime = s->parameters.downtime_limit; 2083 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s); 2084 2085 qemu_file_set_blocking(s->to_dst_file, true); 2086 qemu_file_set_rate_limit(s->to_dst_file, 2087 s->parameters.max_bandwidth / XFER_LIMIT_RATIO); 2088 2089 /* Notify before starting migration thread */ 2090 notifier_list_notify(&migration_state_notifiers, s); 2091 2092 /* 2093 * Open the return path; currently for postcopy but other things might 2094 * also want it. 2095 */ 2096 if (migrate_postcopy_ram()) { 2097 if (open_return_path_on_source(s)) { 2098 error_report("Unable to open return-path for postcopy"); 2099 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, 2100 MIGRATION_STATUS_FAILED); 2101 migrate_fd_cleanup(s); 2102 return; 2103 } 2104 } 2105 2106 migrate_compress_threads_create(); 2107 qemu_thread_create(&s->thread, "live_migration", migration_thread, s, 2108 QEMU_THREAD_JOINABLE); 2109 s->migration_thread_running = true; 2110 } 2111 2112