1 /* 2 * QEMU live migration 3 * 4 * Copyright IBM, Corp. 2008 5 * 6 * Authors: 7 * Anthony Liguori <aliguori@us.ibm.com> 8 * 9 * This work is licensed under the terms of the GNU GPL, version 2. See 10 * the COPYING file in the top-level directory. 11 * 12 * Contributions after 2012-01-13 are licensed under the terms of the 13 * GNU GPL, version 2 or (at your option) any later version. 14 */ 15 16 #include "qemu/osdep.h" 17 #include "qemu/cutils.h" 18 #include "qemu/error-report.h" 19 #include "migration/blocker.h" 20 #include "exec.h" 21 #include "fd.h" 22 #include "socket.h" 23 #include "rdma.h" 24 #include "ram.h" 25 #include "migration/global_state.h" 26 #include "migration/misc.h" 27 #include "migration.h" 28 #include "savevm.h" 29 #include "qemu-file-channel.h" 30 #include "qemu-file.h" 31 #include "migration/vmstate.h" 32 #include "block/block.h" 33 #include "qapi/qmp/qerror.h" 34 #include "qemu/rcu.h" 35 #include "block.h" 36 #include "postcopy-ram.h" 37 #include "qemu/thread.h" 38 #include "qmp-commands.h" 39 #include "trace.h" 40 #include "qapi-event.h" 41 #include "exec/target_page.h" 42 #include "io/channel-buffer.h" 43 #include "migration/colo.h" 44 #include "hw/boards.h" 45 #include "monitor/monitor.h" 46 47 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */ 48 49 /* Amount of time to allocate to each "chunk" of bandwidth-throttled 50 * data. */ 51 #define BUFFER_DELAY 100 52 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY) 53 54 /* Time in milliseconds we are allowed to stop the source, 55 * for sending the last part */ 56 #define DEFAULT_MIGRATE_SET_DOWNTIME 300 57 58 /* Maximum migrate downtime set to 2000 seconds */ 59 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000 60 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000) 61 62 /* Default compression thread count */ 63 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8 64 /* Default decompression thread count, usually decompression is at 65 * least 4 times as fast as compression.*/ 66 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2 67 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */ 68 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1 69 /* Define default autoconverge cpu throttle migration parameters */ 70 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20 71 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10 72 73 /* Migration XBZRLE default cache size */ 74 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024) 75 76 /* The delay time (in ms) between two COLO checkpoints 77 * Note: Please change this default value to 10000 when we support hybrid mode. 78 */ 79 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200 80 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2 81 #define DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT 16 82 83 static NotifierList migration_state_notifiers = 84 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers); 85 86 static bool deferred_incoming; 87 88 /* Messages sent on the return path from destination to source */ 89 enum mig_rp_message_type { 90 MIG_RP_MSG_INVALID = 0, /* Must be 0 */ 91 MIG_RP_MSG_SHUT, /* sibling will not send any more RP messages */ 92 MIG_RP_MSG_PONG, /* Response to a PING; data (seq: be32 ) */ 93 94 MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */ 95 MIG_RP_MSG_REQ_PAGES, /* data (start: be64, len: be32) */ 96 97 MIG_RP_MSG_MAX 98 }; 99 100 /* When we add fault tolerance, we could have several 101 migrations at once. For now we don't need to add 102 dynamic creation of migration */ 103 104 static MigrationState *current_migration; 105 106 static bool migration_object_check(MigrationState *ms, Error **errp); 107 static int migration_maybe_pause(MigrationState *s, 108 int *current_active_state, 109 int new_state); 110 111 void migration_object_init(void) 112 { 113 MachineState *ms = MACHINE(qdev_get_machine()); 114 Error *err = NULL; 115 116 /* This can only be called once. */ 117 assert(!current_migration); 118 current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION)); 119 120 if (!migration_object_check(current_migration, &err)) { 121 error_report_err(err); 122 exit(1); 123 } 124 125 /* 126 * We cannot really do this in migration_instance_init() since at 127 * that time global properties are not yet applied, then this 128 * value will be definitely replaced by something else. 129 */ 130 if (ms->enforce_config_section) { 131 current_migration->send_configuration = true; 132 } 133 } 134 135 void migration_object_finalize(void) 136 { 137 object_unref(OBJECT(current_migration)); 138 } 139 140 /* For outgoing */ 141 MigrationState *migrate_get_current(void) 142 { 143 /* This can only be called after the object created. */ 144 assert(current_migration); 145 return current_migration; 146 } 147 148 MigrationIncomingState *migration_incoming_get_current(void) 149 { 150 static bool once; 151 static MigrationIncomingState mis_current; 152 153 if (!once) { 154 mis_current.state = MIGRATION_STATUS_NONE; 155 memset(&mis_current, 0, sizeof(MigrationIncomingState)); 156 qemu_mutex_init(&mis_current.rp_mutex); 157 qemu_event_init(&mis_current.main_thread_load_event, false); 158 once = true; 159 } 160 return &mis_current; 161 } 162 163 void migration_incoming_state_destroy(void) 164 { 165 struct MigrationIncomingState *mis = migration_incoming_get_current(); 166 167 if (mis->to_src_file) { 168 /* Tell source that we are done */ 169 migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0); 170 qemu_fclose(mis->to_src_file); 171 mis->to_src_file = NULL; 172 } 173 174 if (mis->from_src_file) { 175 qemu_fclose(mis->from_src_file); 176 mis->from_src_file = NULL; 177 } 178 179 qemu_event_reset(&mis->main_thread_load_event); 180 } 181 182 static void migrate_generate_event(int new_state) 183 { 184 if (migrate_use_events()) { 185 qapi_event_send_migration(new_state, &error_abort); 186 } 187 } 188 189 /* 190 * Called on -incoming with a defer: uri. 191 * The migration can be started later after any parameters have been 192 * changed. 193 */ 194 static void deferred_incoming_migration(Error **errp) 195 { 196 if (deferred_incoming) { 197 error_setg(errp, "Incoming migration already deferred"); 198 } 199 deferred_incoming = true; 200 } 201 202 /* 203 * Send a message on the return channel back to the source 204 * of the migration. 205 */ 206 static void migrate_send_rp_message(MigrationIncomingState *mis, 207 enum mig_rp_message_type message_type, 208 uint16_t len, void *data) 209 { 210 trace_migrate_send_rp_message((int)message_type, len); 211 qemu_mutex_lock(&mis->rp_mutex); 212 qemu_put_be16(mis->to_src_file, (unsigned int)message_type); 213 qemu_put_be16(mis->to_src_file, len); 214 qemu_put_buffer(mis->to_src_file, data, len); 215 qemu_fflush(mis->to_src_file); 216 qemu_mutex_unlock(&mis->rp_mutex); 217 } 218 219 /* Request a range of pages from the source VM at the given 220 * start address. 221 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same 222 * as the last request (a name must have been given previously) 223 * Start: Address offset within the RB 224 * Len: Length in bytes required - must be a multiple of pagesize 225 */ 226 void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname, 227 ram_addr_t start, size_t len) 228 { 229 uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */ 230 size_t msglen = 12; /* start + len */ 231 232 *(uint64_t *)bufc = cpu_to_be64((uint64_t)start); 233 *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len); 234 235 if (rbname) { 236 int rbname_len = strlen(rbname); 237 assert(rbname_len < 256); 238 239 bufc[msglen++] = rbname_len; 240 memcpy(bufc + msglen, rbname, rbname_len); 241 msglen += rbname_len; 242 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc); 243 } else { 244 migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc); 245 } 246 } 247 248 void qemu_start_incoming_migration(const char *uri, Error **errp) 249 { 250 const char *p; 251 252 qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort); 253 if (!strcmp(uri, "defer")) { 254 deferred_incoming_migration(errp); 255 } else if (strstart(uri, "tcp:", &p)) { 256 tcp_start_incoming_migration(p, errp); 257 #ifdef CONFIG_RDMA 258 } else if (strstart(uri, "rdma:", &p)) { 259 rdma_start_incoming_migration(p, errp); 260 #endif 261 } else if (strstart(uri, "exec:", &p)) { 262 exec_start_incoming_migration(p, errp); 263 } else if (strstart(uri, "unix:", &p)) { 264 unix_start_incoming_migration(p, errp); 265 } else if (strstart(uri, "fd:", &p)) { 266 fd_start_incoming_migration(p, errp); 267 } else { 268 error_setg(errp, "unknown migration protocol: %s", uri); 269 } 270 } 271 272 static void process_incoming_migration_bh(void *opaque) 273 { 274 Error *local_err = NULL; 275 MigrationIncomingState *mis = opaque; 276 277 /* Make sure all file formats flush their mutable metadata. 278 * If we get an error here, just don't restart the VM yet. */ 279 bdrv_invalidate_cache_all(&local_err); 280 if (local_err) { 281 error_report_err(local_err); 282 local_err = NULL; 283 autostart = false; 284 } 285 286 /* 287 * This must happen after all error conditions are dealt with and 288 * we're sure the VM is going to be running on this host. 289 */ 290 qemu_announce_self(); 291 292 if (multifd_load_cleanup(&local_err) != 0) { 293 error_report_err(local_err); 294 autostart = false; 295 } 296 /* If global state section was not received or we are in running 297 state, we need to obey autostart. Any other state is set with 298 runstate_set. */ 299 300 if (!global_state_received() || 301 global_state_get_runstate() == RUN_STATE_RUNNING) { 302 if (autostart) { 303 vm_start(); 304 } else { 305 runstate_set(RUN_STATE_PAUSED); 306 } 307 } else { 308 runstate_set(global_state_get_runstate()); 309 } 310 /* 311 * This must happen after any state changes since as soon as an external 312 * observer sees this event they might start to prod at the VM assuming 313 * it's ready to use. 314 */ 315 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, 316 MIGRATION_STATUS_COMPLETED); 317 qemu_bh_delete(mis->bh); 318 migration_incoming_state_destroy(); 319 } 320 321 static void process_incoming_migration_co(void *opaque) 322 { 323 MigrationIncomingState *mis = migration_incoming_get_current(); 324 PostcopyState ps; 325 int ret; 326 327 assert(mis->from_src_file); 328 mis->largest_page_size = qemu_ram_pagesize_largest(); 329 postcopy_state_set(POSTCOPY_INCOMING_NONE); 330 migrate_set_state(&mis->state, MIGRATION_STATUS_NONE, 331 MIGRATION_STATUS_ACTIVE); 332 ret = qemu_loadvm_state(mis->from_src_file); 333 334 ps = postcopy_state_get(); 335 trace_process_incoming_migration_co_end(ret, ps); 336 if (ps != POSTCOPY_INCOMING_NONE) { 337 if (ps == POSTCOPY_INCOMING_ADVISE) { 338 /* 339 * Where a migration had postcopy enabled (and thus went to advise) 340 * but managed to complete within the precopy period, we can use 341 * the normal exit. 342 */ 343 postcopy_ram_incoming_cleanup(mis); 344 } else if (ret >= 0) { 345 /* 346 * Postcopy was started, cleanup should happen at the end of the 347 * postcopy thread. 348 */ 349 trace_process_incoming_migration_co_postcopy_end_main(); 350 return; 351 } 352 /* Else if something went wrong then just fall out of the normal exit */ 353 } 354 355 /* we get COLO info, and know if we are in COLO mode */ 356 if (!ret && migration_incoming_enable_colo()) { 357 mis->migration_incoming_co = qemu_coroutine_self(); 358 qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming", 359 colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE); 360 mis->have_colo_incoming_thread = true; 361 qemu_coroutine_yield(); 362 363 /* Wait checkpoint incoming thread exit before free resource */ 364 qemu_thread_join(&mis->colo_incoming_thread); 365 } 366 367 if (ret < 0) { 368 Error *local_err = NULL; 369 370 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, 371 MIGRATION_STATUS_FAILED); 372 error_report("load of migration failed: %s", strerror(-ret)); 373 qemu_fclose(mis->from_src_file); 374 if (multifd_load_cleanup(&local_err) != 0) { 375 error_report_err(local_err); 376 } 377 exit(EXIT_FAILURE); 378 } 379 mis->bh = qemu_bh_new(process_incoming_migration_bh, mis); 380 qemu_bh_schedule(mis->bh); 381 } 382 383 static void migration_incoming_setup(QEMUFile *f) 384 { 385 MigrationIncomingState *mis = migration_incoming_get_current(); 386 387 if (multifd_load_setup() != 0) { 388 /* We haven't been able to create multifd threads 389 nothing better to do */ 390 exit(EXIT_FAILURE); 391 } 392 393 if (!mis->from_src_file) { 394 mis->from_src_file = f; 395 } 396 qemu_file_set_blocking(f, false); 397 } 398 399 static void migration_incoming_process(void) 400 { 401 Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL); 402 qemu_coroutine_enter(co); 403 } 404 405 void migration_fd_process_incoming(QEMUFile *f) 406 { 407 migration_incoming_setup(f); 408 migration_incoming_process(); 409 } 410 411 void migration_ioc_process_incoming(QIOChannel *ioc) 412 { 413 MigrationIncomingState *mis = migration_incoming_get_current(); 414 415 if (!mis->from_src_file) { 416 QEMUFile *f = qemu_fopen_channel_input(ioc); 417 migration_fd_process_incoming(f); 418 } 419 /* We still only have a single channel. Nothing to do here yet */ 420 } 421 422 /** 423 * @migration_has_all_channels: We have received all channels that we need 424 * 425 * Returns true when we have got connections to all the channels that 426 * we need for migration. 427 */ 428 bool migration_has_all_channels(void) 429 { 430 return true; 431 } 432 433 /* 434 * Send a 'SHUT' message on the return channel with the given value 435 * to indicate that we've finished with the RP. Non-0 value indicates 436 * error. 437 */ 438 void migrate_send_rp_shut(MigrationIncomingState *mis, 439 uint32_t value) 440 { 441 uint32_t buf; 442 443 buf = cpu_to_be32(value); 444 migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf); 445 } 446 447 /* 448 * Send a 'PONG' message on the return channel with the given value 449 * (normally in response to a 'PING') 450 */ 451 void migrate_send_rp_pong(MigrationIncomingState *mis, 452 uint32_t value) 453 { 454 uint32_t buf; 455 456 buf = cpu_to_be32(value); 457 migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf); 458 } 459 460 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp) 461 { 462 MigrationCapabilityStatusList *head = NULL; 463 MigrationCapabilityStatusList *caps; 464 MigrationState *s = migrate_get_current(); 465 int i; 466 467 caps = NULL; /* silence compiler warning */ 468 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { 469 #ifndef CONFIG_LIVE_BLOCK_MIGRATION 470 if (i == MIGRATION_CAPABILITY_BLOCK) { 471 continue; 472 } 473 #endif 474 if (head == NULL) { 475 head = g_malloc0(sizeof(*caps)); 476 caps = head; 477 } else { 478 caps->next = g_malloc0(sizeof(*caps)); 479 caps = caps->next; 480 } 481 caps->value = 482 g_malloc(sizeof(*caps->value)); 483 caps->value->capability = i; 484 caps->value->state = s->enabled_capabilities[i]; 485 } 486 487 return head; 488 } 489 490 MigrationParameters *qmp_query_migrate_parameters(Error **errp) 491 { 492 MigrationParameters *params; 493 MigrationState *s = migrate_get_current(); 494 495 /* TODO use QAPI_CLONE() instead of duplicating it inline */ 496 params = g_malloc0(sizeof(*params)); 497 params->has_compress_level = true; 498 params->compress_level = s->parameters.compress_level; 499 params->has_compress_threads = true; 500 params->compress_threads = s->parameters.compress_threads; 501 params->has_decompress_threads = true; 502 params->decompress_threads = s->parameters.decompress_threads; 503 params->has_cpu_throttle_initial = true; 504 params->cpu_throttle_initial = s->parameters.cpu_throttle_initial; 505 params->has_cpu_throttle_increment = true; 506 params->cpu_throttle_increment = s->parameters.cpu_throttle_increment; 507 params->has_tls_creds = true; 508 params->tls_creds = g_strdup(s->parameters.tls_creds); 509 params->has_tls_hostname = true; 510 params->tls_hostname = g_strdup(s->parameters.tls_hostname); 511 params->has_max_bandwidth = true; 512 params->max_bandwidth = s->parameters.max_bandwidth; 513 params->has_downtime_limit = true; 514 params->downtime_limit = s->parameters.downtime_limit; 515 params->has_x_checkpoint_delay = true; 516 params->x_checkpoint_delay = s->parameters.x_checkpoint_delay; 517 params->has_block_incremental = true; 518 params->block_incremental = s->parameters.block_incremental; 519 params->has_x_multifd_channels = true; 520 params->x_multifd_channels = s->parameters.x_multifd_channels; 521 params->has_x_multifd_page_count = true; 522 params->x_multifd_page_count = s->parameters.x_multifd_page_count; 523 params->has_xbzrle_cache_size = true; 524 params->xbzrle_cache_size = s->parameters.xbzrle_cache_size; 525 526 return params; 527 } 528 529 /* 530 * Return true if we're already in the middle of a migration 531 * (i.e. any of the active or setup states) 532 */ 533 static bool migration_is_setup_or_active(int state) 534 { 535 switch (state) { 536 case MIGRATION_STATUS_ACTIVE: 537 case MIGRATION_STATUS_POSTCOPY_ACTIVE: 538 case MIGRATION_STATUS_SETUP: 539 case MIGRATION_STATUS_PRE_SWITCHOVER: 540 case MIGRATION_STATUS_DEVICE: 541 return true; 542 543 default: 544 return false; 545 546 } 547 } 548 549 static void populate_ram_info(MigrationInfo *info, MigrationState *s) 550 { 551 info->has_ram = true; 552 info->ram = g_malloc0(sizeof(*info->ram)); 553 info->ram->transferred = ram_counters.transferred; 554 info->ram->total = ram_bytes_total(); 555 info->ram->duplicate = ram_counters.duplicate; 556 /* legacy value. It is not used anymore */ 557 info->ram->skipped = 0; 558 info->ram->normal = ram_counters.normal; 559 info->ram->normal_bytes = ram_counters.normal * 560 qemu_target_page_size(); 561 info->ram->mbps = s->mbps; 562 info->ram->dirty_sync_count = ram_counters.dirty_sync_count; 563 info->ram->postcopy_requests = ram_counters.postcopy_requests; 564 info->ram->page_size = qemu_target_page_size(); 565 566 if (migrate_use_xbzrle()) { 567 info->has_xbzrle_cache = true; 568 info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache)); 569 info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size(); 570 info->xbzrle_cache->bytes = xbzrle_counters.bytes; 571 info->xbzrle_cache->pages = xbzrle_counters.pages; 572 info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss; 573 info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate; 574 info->xbzrle_cache->overflow = xbzrle_counters.overflow; 575 } 576 577 if (cpu_throttle_active()) { 578 info->has_cpu_throttle_percentage = true; 579 info->cpu_throttle_percentage = cpu_throttle_get_percentage(); 580 } 581 582 if (s->state != MIGRATION_STATUS_COMPLETED) { 583 info->ram->remaining = ram_bytes_remaining(); 584 info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate; 585 } 586 } 587 588 static void populate_disk_info(MigrationInfo *info) 589 { 590 if (blk_mig_active()) { 591 info->has_disk = true; 592 info->disk = g_malloc0(sizeof(*info->disk)); 593 info->disk->transferred = blk_mig_bytes_transferred(); 594 info->disk->remaining = blk_mig_bytes_remaining(); 595 info->disk->total = blk_mig_bytes_total(); 596 } 597 } 598 599 MigrationInfo *qmp_query_migrate(Error **errp) 600 { 601 MigrationInfo *info = g_malloc0(sizeof(*info)); 602 MigrationState *s = migrate_get_current(); 603 604 switch (s->state) { 605 case MIGRATION_STATUS_NONE: 606 /* no migration has happened ever */ 607 break; 608 case MIGRATION_STATUS_SETUP: 609 info->has_status = true; 610 info->has_total_time = false; 611 break; 612 case MIGRATION_STATUS_ACTIVE: 613 case MIGRATION_STATUS_CANCELLING: 614 case MIGRATION_STATUS_POSTCOPY_ACTIVE: 615 case MIGRATION_STATUS_PRE_SWITCHOVER: 616 case MIGRATION_STATUS_DEVICE: 617 /* TODO add some postcopy stats */ 618 info->has_status = true; 619 info->has_total_time = true; 620 info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) 621 - s->start_time; 622 info->has_expected_downtime = true; 623 info->expected_downtime = s->expected_downtime; 624 info->has_setup_time = true; 625 info->setup_time = s->setup_time; 626 627 populate_ram_info(info, s); 628 populate_disk_info(info); 629 break; 630 case MIGRATION_STATUS_COLO: 631 info->has_status = true; 632 /* TODO: display COLO specific information (checkpoint info etc.) */ 633 break; 634 case MIGRATION_STATUS_COMPLETED: 635 info->has_status = true; 636 info->has_total_time = true; 637 info->total_time = s->total_time; 638 info->has_downtime = true; 639 info->downtime = s->downtime; 640 info->has_setup_time = true; 641 info->setup_time = s->setup_time; 642 643 populate_ram_info(info, s); 644 break; 645 case MIGRATION_STATUS_FAILED: 646 info->has_status = true; 647 if (s->error) { 648 info->has_error_desc = true; 649 info->error_desc = g_strdup(error_get_pretty(s->error)); 650 } 651 break; 652 case MIGRATION_STATUS_CANCELLED: 653 info->has_status = true; 654 break; 655 } 656 info->status = s->state; 657 658 return info; 659 } 660 661 /** 662 * @migration_caps_check - check capability validity 663 * 664 * @cap_list: old capability list, array of bool 665 * @params: new capabilities to be applied soon 666 * @errp: set *errp if the check failed, with reason 667 * 668 * Returns true if check passed, otherwise false. 669 */ 670 static bool migrate_caps_check(bool *cap_list, 671 MigrationCapabilityStatusList *params, 672 Error **errp) 673 { 674 MigrationCapabilityStatusList *cap; 675 bool old_postcopy_cap; 676 MigrationIncomingState *mis = migration_incoming_get_current(); 677 678 old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]; 679 680 for (cap = params; cap; cap = cap->next) { 681 cap_list[cap->value->capability] = cap->value->state; 682 } 683 684 #ifndef CONFIG_LIVE_BLOCK_MIGRATION 685 if (cap_list[MIGRATION_CAPABILITY_BLOCK]) { 686 error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) " 687 "block migration"); 688 error_append_hint(errp, "Use drive_mirror+NBD instead.\n"); 689 return false; 690 } 691 #endif 692 693 if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) { 694 if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) { 695 /* The decompression threads asynchronously write into RAM 696 * rather than use the atomic copies needed to avoid 697 * userfaulting. It should be possible to fix the decompression 698 * threads for compatibility in future. 699 */ 700 error_setg(errp, "Postcopy is not currently compatible " 701 "with compression"); 702 return false; 703 } 704 705 /* This check is reasonably expensive, so only when it's being 706 * set the first time, also it's only the destination that needs 707 * special support. 708 */ 709 if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) && 710 !postcopy_ram_supported_by_host(mis)) { 711 /* postcopy_ram_supported_by_host will have emitted a more 712 * detailed message 713 */ 714 error_setg(errp, "Postcopy is not supported"); 715 return false; 716 } 717 } 718 719 return true; 720 } 721 722 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params, 723 Error **errp) 724 { 725 MigrationState *s = migrate_get_current(); 726 MigrationCapabilityStatusList *cap; 727 728 if (migration_is_setup_or_active(s->state)) { 729 error_setg(errp, QERR_MIGRATION_ACTIVE); 730 return; 731 } 732 733 if (!migrate_caps_check(s->enabled_capabilities, params, errp)) { 734 return; 735 } 736 737 for (cap = params; cap; cap = cap->next) { 738 s->enabled_capabilities[cap->value->capability] = cap->value->state; 739 } 740 } 741 742 /* 743 * Check whether the parameters are valid. Error will be put into errp 744 * (if provided). Return true if valid, otherwise false. 745 */ 746 static bool migrate_params_check(MigrationParameters *params, Error **errp) 747 { 748 if (params->has_compress_level && 749 (params->compress_level > 9)) { 750 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level", 751 "is invalid, it should be in the range of 0 to 9"); 752 return false; 753 } 754 755 if (params->has_compress_threads && (params->compress_threads < 1)) { 756 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, 757 "compress_threads", 758 "is invalid, it should be in the range of 1 to 255"); 759 return false; 760 } 761 762 if (params->has_decompress_threads && (params->decompress_threads < 1)) { 763 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, 764 "decompress_threads", 765 "is invalid, it should be in the range of 1 to 255"); 766 return false; 767 } 768 769 if (params->has_cpu_throttle_initial && 770 (params->cpu_throttle_initial < 1 || 771 params->cpu_throttle_initial > 99)) { 772 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, 773 "cpu_throttle_initial", 774 "an integer in the range of 1 to 99"); 775 return false; 776 } 777 778 if (params->has_cpu_throttle_increment && 779 (params->cpu_throttle_increment < 1 || 780 params->cpu_throttle_increment > 99)) { 781 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, 782 "cpu_throttle_increment", 783 "an integer in the range of 1 to 99"); 784 return false; 785 } 786 787 if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) { 788 error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the" 789 " range of 0 to %zu bytes/second", SIZE_MAX); 790 return false; 791 } 792 793 if (params->has_downtime_limit && 794 (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) { 795 error_setg(errp, "Parameter 'downtime_limit' expects an integer in " 796 "the range of 0 to %d milliseconds", 797 MAX_MIGRATE_DOWNTIME); 798 return false; 799 } 800 801 /* x_checkpoint_delay is now always positive */ 802 803 if (params->has_x_multifd_channels && (params->x_multifd_channels < 1)) { 804 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, 805 "multifd_channels", 806 "is invalid, it should be in the range of 1 to 255"); 807 return false; 808 } 809 if (params->has_x_multifd_page_count && 810 (params->x_multifd_page_count < 1 || 811 params->x_multifd_page_count > 10000)) { 812 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, 813 "multifd_page_count", 814 "is invalid, it should be in the range of 1 to 10000"); 815 return false; 816 } 817 818 if (params->has_xbzrle_cache_size && 819 (params->xbzrle_cache_size < qemu_target_page_size() || 820 !is_power_of_2(params->xbzrle_cache_size))) { 821 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, 822 "xbzrle_cache_size", 823 "is invalid, it should be bigger than target page size" 824 " and a power of two"); 825 return false; 826 } 827 828 return true; 829 } 830 831 static void migrate_params_test_apply(MigrateSetParameters *params, 832 MigrationParameters *dest) 833 { 834 *dest = migrate_get_current()->parameters; 835 836 /* TODO use QAPI_CLONE() instead of duplicating it inline */ 837 838 if (params->has_compress_level) { 839 dest->compress_level = params->compress_level; 840 } 841 842 if (params->has_compress_threads) { 843 dest->compress_threads = params->compress_threads; 844 } 845 846 if (params->has_decompress_threads) { 847 dest->decompress_threads = params->decompress_threads; 848 } 849 850 if (params->has_cpu_throttle_initial) { 851 dest->cpu_throttle_initial = params->cpu_throttle_initial; 852 } 853 854 if (params->has_cpu_throttle_increment) { 855 dest->cpu_throttle_increment = params->cpu_throttle_increment; 856 } 857 858 if (params->has_tls_creds) { 859 assert(params->tls_creds->type == QTYPE_QSTRING); 860 dest->tls_creds = g_strdup(params->tls_creds->u.s); 861 } 862 863 if (params->has_tls_hostname) { 864 assert(params->tls_hostname->type == QTYPE_QSTRING); 865 dest->tls_hostname = g_strdup(params->tls_hostname->u.s); 866 } 867 868 if (params->has_max_bandwidth) { 869 dest->max_bandwidth = params->max_bandwidth; 870 } 871 872 if (params->has_downtime_limit) { 873 dest->downtime_limit = params->downtime_limit; 874 } 875 876 if (params->has_x_checkpoint_delay) { 877 dest->x_checkpoint_delay = params->x_checkpoint_delay; 878 } 879 880 if (params->has_block_incremental) { 881 dest->block_incremental = params->block_incremental; 882 } 883 if (params->has_x_multifd_channels) { 884 dest->x_multifd_channels = params->x_multifd_channels; 885 } 886 if (params->has_x_multifd_page_count) { 887 dest->x_multifd_page_count = params->x_multifd_page_count; 888 } 889 if (params->has_xbzrle_cache_size) { 890 dest->xbzrle_cache_size = params->xbzrle_cache_size; 891 } 892 } 893 894 static void migrate_params_apply(MigrateSetParameters *params, Error **errp) 895 { 896 MigrationState *s = migrate_get_current(); 897 898 /* TODO use QAPI_CLONE() instead of duplicating it inline */ 899 900 if (params->has_compress_level) { 901 s->parameters.compress_level = params->compress_level; 902 } 903 904 if (params->has_compress_threads) { 905 s->parameters.compress_threads = params->compress_threads; 906 } 907 908 if (params->has_decompress_threads) { 909 s->parameters.decompress_threads = params->decompress_threads; 910 } 911 912 if (params->has_cpu_throttle_initial) { 913 s->parameters.cpu_throttle_initial = params->cpu_throttle_initial; 914 } 915 916 if (params->has_cpu_throttle_increment) { 917 s->parameters.cpu_throttle_increment = params->cpu_throttle_increment; 918 } 919 920 if (params->has_tls_creds) { 921 g_free(s->parameters.tls_creds); 922 assert(params->tls_creds->type == QTYPE_QSTRING); 923 s->parameters.tls_creds = g_strdup(params->tls_creds->u.s); 924 } 925 926 if (params->has_tls_hostname) { 927 g_free(s->parameters.tls_hostname); 928 assert(params->tls_hostname->type == QTYPE_QSTRING); 929 s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s); 930 } 931 932 if (params->has_max_bandwidth) { 933 s->parameters.max_bandwidth = params->max_bandwidth; 934 if (s->to_dst_file) { 935 qemu_file_set_rate_limit(s->to_dst_file, 936 s->parameters.max_bandwidth / XFER_LIMIT_RATIO); 937 } 938 } 939 940 if (params->has_downtime_limit) { 941 s->parameters.downtime_limit = params->downtime_limit; 942 } 943 944 if (params->has_x_checkpoint_delay) { 945 s->parameters.x_checkpoint_delay = params->x_checkpoint_delay; 946 if (migration_in_colo_state()) { 947 colo_checkpoint_notify(s); 948 } 949 } 950 951 if (params->has_block_incremental) { 952 s->parameters.block_incremental = params->block_incremental; 953 } 954 if (params->has_x_multifd_channels) { 955 s->parameters.x_multifd_channels = params->x_multifd_channels; 956 } 957 if (params->has_x_multifd_page_count) { 958 s->parameters.x_multifd_page_count = params->x_multifd_page_count; 959 } 960 if (params->has_xbzrle_cache_size) { 961 s->parameters.xbzrle_cache_size = params->xbzrle_cache_size; 962 xbzrle_cache_resize(params->xbzrle_cache_size, errp); 963 } 964 } 965 966 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp) 967 { 968 MigrationParameters tmp; 969 970 /* TODO Rewrite "" to null instead */ 971 if (params->has_tls_creds 972 && params->tls_creds->type == QTYPE_QNULL) { 973 QDECREF(params->tls_creds->u.n); 974 params->tls_creds->type = QTYPE_QSTRING; 975 params->tls_creds->u.s = strdup(""); 976 } 977 /* TODO Rewrite "" to null instead */ 978 if (params->has_tls_hostname 979 && params->tls_hostname->type == QTYPE_QNULL) { 980 QDECREF(params->tls_hostname->u.n); 981 params->tls_hostname->type = QTYPE_QSTRING; 982 params->tls_hostname->u.s = strdup(""); 983 } 984 985 migrate_params_test_apply(params, &tmp); 986 987 if (!migrate_params_check(&tmp, errp)) { 988 /* Invalid parameter */ 989 return; 990 } 991 992 migrate_params_apply(params, errp); 993 } 994 995 996 void qmp_migrate_start_postcopy(Error **errp) 997 { 998 MigrationState *s = migrate_get_current(); 999 1000 if (!migrate_postcopy_ram()) { 1001 error_setg(errp, "Enable postcopy with migrate_set_capability before" 1002 " the start of migration"); 1003 return; 1004 } 1005 1006 if (s->state == MIGRATION_STATUS_NONE) { 1007 error_setg(errp, "Postcopy must be started after migration has been" 1008 " started"); 1009 return; 1010 } 1011 /* 1012 * we don't error if migration has finished since that would be racy 1013 * with issuing this command. 1014 */ 1015 atomic_set(&s->start_postcopy, true); 1016 } 1017 1018 /* shared migration helpers */ 1019 1020 void migrate_set_state(int *state, int old_state, int new_state) 1021 { 1022 assert(new_state < MIGRATION_STATUS__MAX); 1023 if (atomic_cmpxchg(state, old_state, new_state) == old_state) { 1024 trace_migrate_set_state(MigrationStatus_str(new_state)); 1025 migrate_generate_event(new_state); 1026 } 1027 } 1028 1029 static MigrationCapabilityStatusList *migrate_cap_add( 1030 MigrationCapabilityStatusList *list, 1031 MigrationCapability index, 1032 bool state) 1033 { 1034 MigrationCapabilityStatusList *cap; 1035 1036 cap = g_new0(MigrationCapabilityStatusList, 1); 1037 cap->value = g_new0(MigrationCapabilityStatus, 1); 1038 cap->value->capability = index; 1039 cap->value->state = state; 1040 cap->next = list; 1041 1042 return cap; 1043 } 1044 1045 void migrate_set_block_enabled(bool value, Error **errp) 1046 { 1047 MigrationCapabilityStatusList *cap; 1048 1049 cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value); 1050 qmp_migrate_set_capabilities(cap, errp); 1051 qapi_free_MigrationCapabilityStatusList(cap); 1052 } 1053 1054 static void migrate_set_block_incremental(MigrationState *s, bool value) 1055 { 1056 s->parameters.block_incremental = value; 1057 } 1058 1059 static void block_cleanup_parameters(MigrationState *s) 1060 { 1061 if (s->must_remove_block_options) { 1062 /* setting to false can never fail */ 1063 migrate_set_block_enabled(false, &error_abort); 1064 migrate_set_block_incremental(s, false); 1065 s->must_remove_block_options = false; 1066 } 1067 } 1068 1069 static void migrate_fd_cleanup(void *opaque) 1070 { 1071 MigrationState *s = opaque; 1072 1073 qemu_bh_delete(s->cleanup_bh); 1074 s->cleanup_bh = NULL; 1075 1076 qemu_savevm_state_cleanup(); 1077 1078 if (s->to_dst_file) { 1079 Error *local_err = NULL; 1080 1081 trace_migrate_fd_cleanup(); 1082 qemu_mutex_unlock_iothread(); 1083 if (s->migration_thread_running) { 1084 qemu_thread_join(&s->thread); 1085 s->migration_thread_running = false; 1086 } 1087 qemu_mutex_lock_iothread(); 1088 1089 if (multifd_save_cleanup(&local_err) != 0) { 1090 error_report_err(local_err); 1091 } 1092 qemu_fclose(s->to_dst_file); 1093 s->to_dst_file = NULL; 1094 } 1095 1096 assert((s->state != MIGRATION_STATUS_ACTIVE) && 1097 (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE)); 1098 1099 if (s->state == MIGRATION_STATUS_CANCELLING) { 1100 migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING, 1101 MIGRATION_STATUS_CANCELLED); 1102 } 1103 1104 if (s->error) { 1105 /* It is used on info migrate. We can't free it */ 1106 error_report_err(error_copy(s->error)); 1107 } 1108 notifier_list_notify(&migration_state_notifiers, s); 1109 block_cleanup_parameters(s); 1110 } 1111 1112 void migrate_set_error(MigrationState *s, const Error *error) 1113 { 1114 qemu_mutex_lock(&s->error_mutex); 1115 if (!s->error) { 1116 s->error = error_copy(error); 1117 } 1118 qemu_mutex_unlock(&s->error_mutex); 1119 } 1120 1121 void migrate_fd_error(MigrationState *s, const Error *error) 1122 { 1123 trace_migrate_fd_error(error_get_pretty(error)); 1124 assert(s->to_dst_file == NULL); 1125 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, 1126 MIGRATION_STATUS_FAILED); 1127 migrate_set_error(s, error); 1128 } 1129 1130 static void migrate_fd_cancel(MigrationState *s) 1131 { 1132 int old_state ; 1133 QEMUFile *f = migrate_get_current()->to_dst_file; 1134 trace_migrate_fd_cancel(); 1135 1136 if (s->rp_state.from_dst_file) { 1137 /* shutdown the rp socket, so causing the rp thread to shutdown */ 1138 qemu_file_shutdown(s->rp_state.from_dst_file); 1139 } 1140 1141 do { 1142 old_state = s->state; 1143 if (!migration_is_setup_or_active(old_state)) { 1144 break; 1145 } 1146 /* If the migration is paused, kick it out of the pause */ 1147 if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) { 1148 qemu_sem_post(&s->pause_sem); 1149 } 1150 migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING); 1151 } while (s->state != MIGRATION_STATUS_CANCELLING); 1152 1153 /* 1154 * If we're unlucky the migration code might be stuck somewhere in a 1155 * send/write while the network has failed and is waiting to timeout; 1156 * if we've got shutdown(2) available then we can force it to quit. 1157 * The outgoing qemu file gets closed in migrate_fd_cleanup that is 1158 * called in a bh, so there is no race against this cancel. 1159 */ 1160 if (s->state == MIGRATION_STATUS_CANCELLING && f) { 1161 qemu_file_shutdown(f); 1162 } 1163 if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) { 1164 Error *local_err = NULL; 1165 1166 bdrv_invalidate_cache_all(&local_err); 1167 if (local_err) { 1168 error_report_err(local_err); 1169 } else { 1170 s->block_inactive = false; 1171 } 1172 } 1173 } 1174 1175 void add_migration_state_change_notifier(Notifier *notify) 1176 { 1177 notifier_list_add(&migration_state_notifiers, notify); 1178 } 1179 1180 void remove_migration_state_change_notifier(Notifier *notify) 1181 { 1182 notifier_remove(notify); 1183 } 1184 1185 bool migration_in_setup(MigrationState *s) 1186 { 1187 return s->state == MIGRATION_STATUS_SETUP; 1188 } 1189 1190 bool migration_has_finished(MigrationState *s) 1191 { 1192 return s->state == MIGRATION_STATUS_COMPLETED; 1193 } 1194 1195 bool migration_has_failed(MigrationState *s) 1196 { 1197 return (s->state == MIGRATION_STATUS_CANCELLED || 1198 s->state == MIGRATION_STATUS_FAILED); 1199 } 1200 1201 bool migration_in_postcopy(void) 1202 { 1203 MigrationState *s = migrate_get_current(); 1204 1205 return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); 1206 } 1207 1208 bool migration_in_postcopy_after_devices(MigrationState *s) 1209 { 1210 return migration_in_postcopy() && s->postcopy_after_devices; 1211 } 1212 1213 bool migration_is_idle(void) 1214 { 1215 MigrationState *s = migrate_get_current(); 1216 1217 switch (s->state) { 1218 case MIGRATION_STATUS_NONE: 1219 case MIGRATION_STATUS_CANCELLED: 1220 case MIGRATION_STATUS_COMPLETED: 1221 case MIGRATION_STATUS_FAILED: 1222 return true; 1223 case MIGRATION_STATUS_SETUP: 1224 case MIGRATION_STATUS_CANCELLING: 1225 case MIGRATION_STATUS_ACTIVE: 1226 case MIGRATION_STATUS_POSTCOPY_ACTIVE: 1227 case MIGRATION_STATUS_COLO: 1228 case MIGRATION_STATUS_PRE_SWITCHOVER: 1229 case MIGRATION_STATUS_DEVICE: 1230 return false; 1231 case MIGRATION_STATUS__MAX: 1232 g_assert_not_reached(); 1233 } 1234 1235 return false; 1236 } 1237 1238 MigrationState *migrate_init(void) 1239 { 1240 MigrationState *s = migrate_get_current(); 1241 1242 /* 1243 * Reinitialise all migration state, except 1244 * parameters/capabilities that the user set, and 1245 * locks. 1246 */ 1247 s->bytes_xfer = 0; 1248 s->xfer_limit = 0; 1249 s->cleanup_bh = 0; 1250 s->to_dst_file = NULL; 1251 s->state = MIGRATION_STATUS_NONE; 1252 s->rp_state.from_dst_file = NULL; 1253 s->rp_state.error = false; 1254 s->mbps = 0.0; 1255 s->downtime = 0; 1256 s->expected_downtime = 0; 1257 s->setup_time = 0; 1258 s->start_postcopy = false; 1259 s->postcopy_after_devices = false; 1260 s->migration_thread_running = false; 1261 error_free(s->error); 1262 s->error = NULL; 1263 1264 migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP); 1265 1266 s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 1267 s->total_time = 0; 1268 s->vm_was_running = false; 1269 s->iteration_initial_bytes = 0; 1270 s->threshold_size = 0; 1271 return s; 1272 } 1273 1274 static GSList *migration_blockers; 1275 1276 int migrate_add_blocker(Error *reason, Error **errp) 1277 { 1278 if (migrate_get_current()->only_migratable) { 1279 error_propagate(errp, error_copy(reason)); 1280 error_prepend(errp, "disallowing migration blocker " 1281 "(--only_migratable) for: "); 1282 return -EACCES; 1283 } 1284 1285 if (migration_is_idle()) { 1286 migration_blockers = g_slist_prepend(migration_blockers, reason); 1287 return 0; 1288 } 1289 1290 error_propagate(errp, error_copy(reason)); 1291 error_prepend(errp, "disallowing migration blocker (migration in " 1292 "progress) for: "); 1293 return -EBUSY; 1294 } 1295 1296 void migrate_del_blocker(Error *reason) 1297 { 1298 migration_blockers = g_slist_remove(migration_blockers, reason); 1299 } 1300 1301 void qmp_migrate_incoming(const char *uri, Error **errp) 1302 { 1303 Error *local_err = NULL; 1304 static bool once = true; 1305 1306 if (!deferred_incoming) { 1307 error_setg(errp, "For use with '-incoming defer'"); 1308 return; 1309 } 1310 if (!once) { 1311 error_setg(errp, "The incoming migration has already been started"); 1312 } 1313 1314 qemu_start_incoming_migration(uri, &local_err); 1315 1316 if (local_err) { 1317 error_propagate(errp, local_err); 1318 return; 1319 } 1320 1321 once = false; 1322 } 1323 1324 bool migration_is_blocked(Error **errp) 1325 { 1326 if (qemu_savevm_state_blocked(errp)) { 1327 return true; 1328 } 1329 1330 if (migration_blockers) { 1331 error_propagate(errp, error_copy(migration_blockers->data)); 1332 return true; 1333 } 1334 1335 return false; 1336 } 1337 1338 void qmp_migrate(const char *uri, bool has_blk, bool blk, 1339 bool has_inc, bool inc, bool has_detach, bool detach, 1340 Error **errp) 1341 { 1342 Error *local_err = NULL; 1343 MigrationState *s = migrate_get_current(); 1344 const char *p; 1345 1346 if (migration_is_setup_or_active(s->state) || 1347 s->state == MIGRATION_STATUS_CANCELLING || 1348 s->state == MIGRATION_STATUS_COLO) { 1349 error_setg(errp, QERR_MIGRATION_ACTIVE); 1350 return; 1351 } 1352 if (runstate_check(RUN_STATE_INMIGRATE)) { 1353 error_setg(errp, "Guest is waiting for an incoming migration"); 1354 return; 1355 } 1356 1357 if (migration_is_blocked(errp)) { 1358 return; 1359 } 1360 1361 if ((has_blk && blk) || (has_inc && inc)) { 1362 if (migrate_use_block() || migrate_use_block_incremental()) { 1363 error_setg(errp, "Command options are incompatible with " 1364 "current migration capabilities"); 1365 return; 1366 } 1367 migrate_set_block_enabled(true, &local_err); 1368 if (local_err) { 1369 error_propagate(errp, local_err); 1370 return; 1371 } 1372 s->must_remove_block_options = true; 1373 } 1374 1375 if (has_inc && inc) { 1376 migrate_set_block_incremental(s, true); 1377 } 1378 1379 s = migrate_init(); 1380 1381 if (strstart(uri, "tcp:", &p)) { 1382 tcp_start_outgoing_migration(s, p, &local_err); 1383 #ifdef CONFIG_RDMA 1384 } else if (strstart(uri, "rdma:", &p)) { 1385 rdma_start_outgoing_migration(s, p, &local_err); 1386 #endif 1387 } else if (strstart(uri, "exec:", &p)) { 1388 exec_start_outgoing_migration(s, p, &local_err); 1389 } else if (strstart(uri, "unix:", &p)) { 1390 unix_start_outgoing_migration(s, p, &local_err); 1391 } else if (strstart(uri, "fd:", &p)) { 1392 fd_start_outgoing_migration(s, p, &local_err); 1393 } else { 1394 error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri", 1395 "a valid migration protocol"); 1396 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, 1397 MIGRATION_STATUS_FAILED); 1398 return; 1399 } 1400 1401 if (local_err) { 1402 migrate_fd_error(s, local_err); 1403 error_propagate(errp, local_err); 1404 return; 1405 } 1406 } 1407 1408 void qmp_migrate_cancel(Error **errp) 1409 { 1410 migrate_fd_cancel(migrate_get_current()); 1411 } 1412 1413 void qmp_migrate_continue(MigrationStatus state, Error **errp) 1414 { 1415 MigrationState *s = migrate_get_current(); 1416 if (s->state != state) { 1417 error_setg(errp, "Migration not in expected state: %s", 1418 MigrationStatus_str(s->state)); 1419 return; 1420 } 1421 qemu_sem_post(&s->pause_sem); 1422 } 1423 1424 void qmp_migrate_set_cache_size(int64_t value, Error **errp) 1425 { 1426 MigrateSetParameters p = { 1427 .has_xbzrle_cache_size = true, 1428 .xbzrle_cache_size = value, 1429 }; 1430 1431 qmp_migrate_set_parameters(&p, errp); 1432 } 1433 1434 int64_t qmp_query_migrate_cache_size(Error **errp) 1435 { 1436 return migrate_xbzrle_cache_size(); 1437 } 1438 1439 void qmp_migrate_set_speed(int64_t value, Error **errp) 1440 { 1441 MigrateSetParameters p = { 1442 .has_max_bandwidth = true, 1443 .max_bandwidth = value, 1444 }; 1445 1446 qmp_migrate_set_parameters(&p, errp); 1447 } 1448 1449 void qmp_migrate_set_downtime(double value, Error **errp) 1450 { 1451 if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) { 1452 error_setg(errp, "Parameter 'downtime_limit' expects an integer in " 1453 "the range of 0 to %d seconds", 1454 MAX_MIGRATE_DOWNTIME_SECONDS); 1455 return; 1456 } 1457 1458 value *= 1000; /* Convert to milliseconds */ 1459 value = MAX(0, MIN(INT64_MAX, value)); 1460 1461 MigrateSetParameters p = { 1462 .has_downtime_limit = true, 1463 .downtime_limit = value, 1464 }; 1465 1466 qmp_migrate_set_parameters(&p, errp); 1467 } 1468 1469 bool migrate_release_ram(void) 1470 { 1471 MigrationState *s; 1472 1473 s = migrate_get_current(); 1474 1475 return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM]; 1476 } 1477 1478 bool migrate_postcopy_ram(void) 1479 { 1480 MigrationState *s; 1481 1482 s = migrate_get_current(); 1483 1484 return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM]; 1485 } 1486 1487 bool migrate_postcopy(void) 1488 { 1489 return migrate_postcopy_ram(); 1490 } 1491 1492 bool migrate_auto_converge(void) 1493 { 1494 MigrationState *s; 1495 1496 s = migrate_get_current(); 1497 1498 return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE]; 1499 } 1500 1501 bool migrate_zero_blocks(void) 1502 { 1503 MigrationState *s; 1504 1505 s = migrate_get_current(); 1506 1507 return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS]; 1508 } 1509 1510 bool migrate_use_compression(void) 1511 { 1512 MigrationState *s; 1513 1514 s = migrate_get_current(); 1515 1516 return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS]; 1517 } 1518 1519 int migrate_compress_level(void) 1520 { 1521 MigrationState *s; 1522 1523 s = migrate_get_current(); 1524 1525 return s->parameters.compress_level; 1526 } 1527 1528 int migrate_compress_threads(void) 1529 { 1530 MigrationState *s; 1531 1532 s = migrate_get_current(); 1533 1534 return s->parameters.compress_threads; 1535 } 1536 1537 int migrate_decompress_threads(void) 1538 { 1539 MigrationState *s; 1540 1541 s = migrate_get_current(); 1542 1543 return s->parameters.decompress_threads; 1544 } 1545 1546 bool migrate_use_events(void) 1547 { 1548 MigrationState *s; 1549 1550 s = migrate_get_current(); 1551 1552 return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS]; 1553 } 1554 1555 bool migrate_use_multifd(void) 1556 { 1557 MigrationState *s; 1558 1559 s = migrate_get_current(); 1560 1561 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_MULTIFD]; 1562 } 1563 1564 bool migrate_pause_before_switchover(void) 1565 { 1566 MigrationState *s; 1567 1568 s = migrate_get_current(); 1569 1570 return s->enabled_capabilities[ 1571 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER]; 1572 } 1573 1574 int migrate_multifd_channels(void) 1575 { 1576 MigrationState *s; 1577 1578 s = migrate_get_current(); 1579 1580 return s->parameters.x_multifd_channels; 1581 } 1582 1583 int migrate_multifd_page_count(void) 1584 { 1585 MigrationState *s; 1586 1587 s = migrate_get_current(); 1588 1589 return s->parameters.x_multifd_page_count; 1590 } 1591 1592 int migrate_use_xbzrle(void) 1593 { 1594 MigrationState *s; 1595 1596 s = migrate_get_current(); 1597 1598 return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE]; 1599 } 1600 1601 int64_t migrate_xbzrle_cache_size(void) 1602 { 1603 MigrationState *s; 1604 1605 s = migrate_get_current(); 1606 1607 return s->parameters.xbzrle_cache_size; 1608 } 1609 1610 bool migrate_use_block(void) 1611 { 1612 MigrationState *s; 1613 1614 s = migrate_get_current(); 1615 1616 return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK]; 1617 } 1618 1619 bool migrate_use_return_path(void) 1620 { 1621 MigrationState *s; 1622 1623 s = migrate_get_current(); 1624 1625 return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH]; 1626 } 1627 1628 bool migrate_use_block_incremental(void) 1629 { 1630 MigrationState *s; 1631 1632 s = migrate_get_current(); 1633 1634 return s->parameters.block_incremental; 1635 } 1636 1637 /* migration thread support */ 1638 /* 1639 * Something bad happened to the RP stream, mark an error 1640 * The caller shall print or trace something to indicate why 1641 */ 1642 static void mark_source_rp_bad(MigrationState *s) 1643 { 1644 s->rp_state.error = true; 1645 } 1646 1647 static struct rp_cmd_args { 1648 ssize_t len; /* -1 = variable */ 1649 const char *name; 1650 } rp_cmd_args[] = { 1651 [MIG_RP_MSG_INVALID] = { .len = -1, .name = "INVALID" }, 1652 [MIG_RP_MSG_SHUT] = { .len = 4, .name = "SHUT" }, 1653 [MIG_RP_MSG_PONG] = { .len = 4, .name = "PONG" }, 1654 [MIG_RP_MSG_REQ_PAGES] = { .len = 12, .name = "REQ_PAGES" }, 1655 [MIG_RP_MSG_REQ_PAGES_ID] = { .len = -1, .name = "REQ_PAGES_ID" }, 1656 [MIG_RP_MSG_MAX] = { .len = -1, .name = "MAX" }, 1657 }; 1658 1659 /* 1660 * Process a request for pages received on the return path, 1661 * We're allowed to send more than requested (e.g. to round to our page size) 1662 * and we don't need to send pages that have already been sent. 1663 */ 1664 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname, 1665 ram_addr_t start, size_t len) 1666 { 1667 long our_host_ps = getpagesize(); 1668 1669 trace_migrate_handle_rp_req_pages(rbname, start, len); 1670 1671 /* 1672 * Since we currently insist on matching page sizes, just sanity check 1673 * we're being asked for whole host pages. 1674 */ 1675 if (start & (our_host_ps-1) || 1676 (len & (our_host_ps-1))) { 1677 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT 1678 " len: %zd", __func__, start, len); 1679 mark_source_rp_bad(ms); 1680 return; 1681 } 1682 1683 if (ram_save_queue_pages(rbname, start, len)) { 1684 mark_source_rp_bad(ms); 1685 } 1686 } 1687 1688 /* 1689 * Handles messages sent on the return path towards the source VM 1690 * 1691 */ 1692 static void *source_return_path_thread(void *opaque) 1693 { 1694 MigrationState *ms = opaque; 1695 QEMUFile *rp = ms->rp_state.from_dst_file; 1696 uint16_t header_len, header_type; 1697 uint8_t buf[512]; 1698 uint32_t tmp32, sibling_error; 1699 ram_addr_t start = 0; /* =0 to silence warning */ 1700 size_t len = 0, expected_len; 1701 int res; 1702 1703 trace_source_return_path_thread_entry(); 1704 while (!ms->rp_state.error && !qemu_file_get_error(rp) && 1705 migration_is_setup_or_active(ms->state)) { 1706 trace_source_return_path_thread_loop_top(); 1707 header_type = qemu_get_be16(rp); 1708 header_len = qemu_get_be16(rp); 1709 1710 if (header_type >= MIG_RP_MSG_MAX || 1711 header_type == MIG_RP_MSG_INVALID) { 1712 error_report("RP: Received invalid message 0x%04x length 0x%04x", 1713 header_type, header_len); 1714 mark_source_rp_bad(ms); 1715 goto out; 1716 } 1717 1718 if ((rp_cmd_args[header_type].len != -1 && 1719 header_len != rp_cmd_args[header_type].len) || 1720 header_len > sizeof(buf)) { 1721 error_report("RP: Received '%s' message (0x%04x) with" 1722 "incorrect length %d expecting %zu", 1723 rp_cmd_args[header_type].name, header_type, header_len, 1724 (size_t)rp_cmd_args[header_type].len); 1725 mark_source_rp_bad(ms); 1726 goto out; 1727 } 1728 1729 /* We know we've got a valid header by this point */ 1730 res = qemu_get_buffer(rp, buf, header_len); 1731 if (res != header_len) { 1732 error_report("RP: Failed reading data for message 0x%04x" 1733 " read %d expected %d", 1734 header_type, res, header_len); 1735 mark_source_rp_bad(ms); 1736 goto out; 1737 } 1738 1739 /* OK, we have the message and the data */ 1740 switch (header_type) { 1741 case MIG_RP_MSG_SHUT: 1742 sibling_error = ldl_be_p(buf); 1743 trace_source_return_path_thread_shut(sibling_error); 1744 if (sibling_error) { 1745 error_report("RP: Sibling indicated error %d", sibling_error); 1746 mark_source_rp_bad(ms); 1747 } 1748 /* 1749 * We'll let the main thread deal with closing the RP 1750 * we could do a shutdown(2) on it, but we're the only user 1751 * anyway, so there's nothing gained. 1752 */ 1753 goto out; 1754 1755 case MIG_RP_MSG_PONG: 1756 tmp32 = ldl_be_p(buf); 1757 trace_source_return_path_thread_pong(tmp32); 1758 break; 1759 1760 case MIG_RP_MSG_REQ_PAGES: 1761 start = ldq_be_p(buf); 1762 len = ldl_be_p(buf + 8); 1763 migrate_handle_rp_req_pages(ms, NULL, start, len); 1764 break; 1765 1766 case MIG_RP_MSG_REQ_PAGES_ID: 1767 expected_len = 12 + 1; /* header + termination */ 1768 1769 if (header_len >= expected_len) { 1770 start = ldq_be_p(buf); 1771 len = ldl_be_p(buf + 8); 1772 /* Now we expect an idstr */ 1773 tmp32 = buf[12]; /* Length of the following idstr */ 1774 buf[13 + tmp32] = '\0'; 1775 expected_len += tmp32; 1776 } 1777 if (header_len != expected_len) { 1778 error_report("RP: Req_Page_id with length %d expecting %zd", 1779 header_len, expected_len); 1780 mark_source_rp_bad(ms); 1781 goto out; 1782 } 1783 migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len); 1784 break; 1785 1786 default: 1787 break; 1788 } 1789 } 1790 if (qemu_file_get_error(rp)) { 1791 trace_source_return_path_thread_bad_end(); 1792 mark_source_rp_bad(ms); 1793 } 1794 1795 trace_source_return_path_thread_end(); 1796 out: 1797 ms->rp_state.from_dst_file = NULL; 1798 qemu_fclose(rp); 1799 return NULL; 1800 } 1801 1802 static int open_return_path_on_source(MigrationState *ms) 1803 { 1804 1805 ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file); 1806 if (!ms->rp_state.from_dst_file) { 1807 return -1; 1808 } 1809 1810 trace_open_return_path_on_source(); 1811 qemu_thread_create(&ms->rp_state.rp_thread, "return path", 1812 source_return_path_thread, ms, QEMU_THREAD_JOINABLE); 1813 1814 trace_open_return_path_on_source_continue(); 1815 1816 return 0; 1817 } 1818 1819 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */ 1820 static int await_return_path_close_on_source(MigrationState *ms) 1821 { 1822 /* 1823 * If this is a normal exit then the destination will send a SHUT and the 1824 * rp_thread will exit, however if there's an error we need to cause 1825 * it to exit. 1826 */ 1827 if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) { 1828 /* 1829 * shutdown(2), if we have it, will cause it to unblock if it's stuck 1830 * waiting for the destination. 1831 */ 1832 qemu_file_shutdown(ms->rp_state.from_dst_file); 1833 mark_source_rp_bad(ms); 1834 } 1835 trace_await_return_path_close_on_source_joining(); 1836 qemu_thread_join(&ms->rp_state.rp_thread); 1837 trace_await_return_path_close_on_source_close(); 1838 return ms->rp_state.error; 1839 } 1840 1841 /* 1842 * Switch from normal iteration to postcopy 1843 * Returns non-0 on error 1844 */ 1845 static int postcopy_start(MigrationState *ms) 1846 { 1847 int ret; 1848 QIOChannelBuffer *bioc; 1849 QEMUFile *fb; 1850 int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 1851 bool restart_block = false; 1852 int cur_state = MIGRATION_STATUS_ACTIVE; 1853 if (!migrate_pause_before_switchover()) { 1854 migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE, 1855 MIGRATION_STATUS_POSTCOPY_ACTIVE); 1856 } 1857 1858 trace_postcopy_start(); 1859 qemu_mutex_lock_iothread(); 1860 trace_postcopy_start_set_run(); 1861 1862 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER); 1863 global_state_store(); 1864 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); 1865 if (ret < 0) { 1866 goto fail; 1867 } 1868 1869 ret = migration_maybe_pause(ms, &cur_state, 1870 MIGRATION_STATUS_POSTCOPY_ACTIVE); 1871 if (ret < 0) { 1872 goto fail; 1873 } 1874 1875 ret = bdrv_inactivate_all(); 1876 if (ret < 0) { 1877 goto fail; 1878 } 1879 restart_block = true; 1880 1881 /* 1882 * Cause any non-postcopiable, but iterative devices to 1883 * send out their final data. 1884 */ 1885 qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false); 1886 1887 /* 1888 * in Finish migrate and with the io-lock held everything should 1889 * be quiet, but we've potentially still got dirty pages and we 1890 * need to tell the destination to throw any pages it's already received 1891 * that are dirty 1892 */ 1893 if (migrate_postcopy_ram()) { 1894 if (ram_postcopy_send_discard_bitmap(ms)) { 1895 error_report("postcopy send discard bitmap failed"); 1896 goto fail; 1897 } 1898 } 1899 1900 /* 1901 * send rest of state - note things that are doing postcopy 1902 * will notice we're in POSTCOPY_ACTIVE and not actually 1903 * wrap their state up here 1904 */ 1905 qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX); 1906 if (migrate_postcopy_ram()) { 1907 /* Ping just for debugging, helps line traces up */ 1908 qemu_savevm_send_ping(ms->to_dst_file, 2); 1909 } 1910 1911 /* 1912 * While loading the device state we may trigger page transfer 1913 * requests and the fd must be free to process those, and thus 1914 * the destination must read the whole device state off the fd before 1915 * it starts processing it. Unfortunately the ad-hoc migration format 1916 * doesn't allow the destination to know the size to read without fully 1917 * parsing it through each devices load-state code (especially the open 1918 * coded devices that use get/put). 1919 * So we wrap the device state up in a package with a length at the start; 1920 * to do this we use a qemu_buf to hold the whole of the device state. 1921 */ 1922 bioc = qio_channel_buffer_new(4096); 1923 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer"); 1924 fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc)); 1925 object_unref(OBJECT(bioc)); 1926 1927 /* 1928 * Make sure the receiver can get incoming pages before we send the rest 1929 * of the state 1930 */ 1931 qemu_savevm_send_postcopy_listen(fb); 1932 1933 qemu_savevm_state_complete_precopy(fb, false, false); 1934 if (migrate_postcopy_ram()) { 1935 qemu_savevm_send_ping(fb, 3); 1936 } 1937 1938 qemu_savevm_send_postcopy_run(fb); 1939 1940 /* <><> end of stuff going into the package */ 1941 1942 /* Last point of recovery; as soon as we send the package the destination 1943 * can open devices and potentially start running. 1944 * Lets just check again we've not got any errors. 1945 */ 1946 ret = qemu_file_get_error(ms->to_dst_file); 1947 if (ret) { 1948 error_report("postcopy_start: Migration stream errored (pre package)"); 1949 goto fail_closefb; 1950 } 1951 1952 restart_block = false; 1953 1954 /* Now send that blob */ 1955 if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) { 1956 goto fail_closefb; 1957 } 1958 qemu_fclose(fb); 1959 1960 /* Send a notify to give a chance for anything that needs to happen 1961 * at the transition to postcopy and after the device state; in particular 1962 * spice needs to trigger a transition now 1963 */ 1964 ms->postcopy_after_devices = true; 1965 notifier_list_notify(&migration_state_notifiers, ms); 1966 1967 ms->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop; 1968 1969 qemu_mutex_unlock_iothread(); 1970 1971 if (migrate_postcopy_ram()) { 1972 /* 1973 * Although this ping is just for debug, it could potentially be 1974 * used for getting a better measurement of downtime at the source. 1975 */ 1976 qemu_savevm_send_ping(ms->to_dst_file, 4); 1977 } 1978 1979 if (migrate_release_ram()) { 1980 ram_postcopy_migrated_memory_release(ms); 1981 } 1982 1983 ret = qemu_file_get_error(ms->to_dst_file); 1984 if (ret) { 1985 error_report("postcopy_start: Migration stream errored"); 1986 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, 1987 MIGRATION_STATUS_FAILED); 1988 } 1989 1990 return ret; 1991 1992 fail_closefb: 1993 qemu_fclose(fb); 1994 fail: 1995 migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, 1996 MIGRATION_STATUS_FAILED); 1997 if (restart_block) { 1998 /* A failure happened early enough that we know the destination hasn't 1999 * accessed block devices, so we're safe to recover. 2000 */ 2001 Error *local_err = NULL; 2002 2003 bdrv_invalidate_cache_all(&local_err); 2004 if (local_err) { 2005 error_report_err(local_err); 2006 } 2007 } 2008 qemu_mutex_unlock_iothread(); 2009 return -1; 2010 } 2011 2012 /** 2013 * migration_maybe_pause: Pause if required to by 2014 * migrate_pause_before_switchover called with the iothread locked 2015 * Returns: 0 on success 2016 */ 2017 static int migration_maybe_pause(MigrationState *s, 2018 int *current_active_state, 2019 int new_state) 2020 { 2021 if (!migrate_pause_before_switchover()) { 2022 return 0; 2023 } 2024 2025 /* Since leaving this state is not atomic with posting the semaphore 2026 * it's possible that someone could have issued multiple migrate_continue 2027 * and the semaphore is incorrectly positive at this point; 2028 * the docs say it's undefined to reinit a semaphore that's already 2029 * init'd, so use timedwait to eat up any existing posts. 2030 */ 2031 while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) { 2032 /* This block intentionally left blank */ 2033 } 2034 2035 qemu_mutex_unlock_iothread(); 2036 migrate_set_state(&s->state, *current_active_state, 2037 MIGRATION_STATUS_PRE_SWITCHOVER); 2038 qemu_sem_wait(&s->pause_sem); 2039 migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER, 2040 new_state); 2041 *current_active_state = new_state; 2042 qemu_mutex_lock_iothread(); 2043 2044 return s->state == new_state ? 0 : -EINVAL; 2045 } 2046 2047 /** 2048 * migration_completion: Used by migration_thread when there's not much left. 2049 * The caller 'breaks' the loop when this returns. 2050 * 2051 * @s: Current migration state 2052 */ 2053 static void migration_completion(MigrationState *s) 2054 { 2055 int ret; 2056 int current_active_state = s->state; 2057 2058 if (s->state == MIGRATION_STATUS_ACTIVE) { 2059 qemu_mutex_lock_iothread(); 2060 s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 2061 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER); 2062 s->vm_was_running = runstate_is_running(); 2063 ret = global_state_store(); 2064 2065 if (!ret) { 2066 bool inactivate = !migrate_colo_enabled(); 2067 ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE); 2068 if (ret >= 0) { 2069 ret = migration_maybe_pause(s, ¤t_active_state, 2070 MIGRATION_STATUS_DEVICE); 2071 } 2072 if (ret >= 0) { 2073 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX); 2074 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false, 2075 inactivate); 2076 } 2077 if (inactivate && ret >= 0) { 2078 s->block_inactive = true; 2079 } 2080 } 2081 qemu_mutex_unlock_iothread(); 2082 2083 if (ret < 0) { 2084 goto fail; 2085 } 2086 } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { 2087 trace_migration_completion_postcopy_end(); 2088 2089 qemu_savevm_state_complete_postcopy(s->to_dst_file); 2090 trace_migration_completion_postcopy_end_after_complete(); 2091 } 2092 2093 /* 2094 * If rp was opened we must clean up the thread before 2095 * cleaning everything else up (since if there are no failures 2096 * it will wait for the destination to send it's status in 2097 * a SHUT command). 2098 */ 2099 if (s->rp_state.from_dst_file) { 2100 int rp_error; 2101 trace_migration_return_path_end_before(); 2102 rp_error = await_return_path_close_on_source(s); 2103 trace_migration_return_path_end_after(rp_error); 2104 if (rp_error) { 2105 goto fail_invalidate; 2106 } 2107 } 2108 2109 if (qemu_file_get_error(s->to_dst_file)) { 2110 trace_migration_completion_file_err(); 2111 goto fail_invalidate; 2112 } 2113 2114 if (!migrate_colo_enabled()) { 2115 migrate_set_state(&s->state, current_active_state, 2116 MIGRATION_STATUS_COMPLETED); 2117 } 2118 2119 return; 2120 2121 fail_invalidate: 2122 /* If not doing postcopy, vm_start() will be called: let's regain 2123 * control on images. 2124 */ 2125 if (s->state == MIGRATION_STATUS_ACTIVE || 2126 s->state == MIGRATION_STATUS_DEVICE) { 2127 Error *local_err = NULL; 2128 2129 qemu_mutex_lock_iothread(); 2130 bdrv_invalidate_cache_all(&local_err); 2131 if (local_err) { 2132 error_report_err(local_err); 2133 } else { 2134 s->block_inactive = false; 2135 } 2136 qemu_mutex_unlock_iothread(); 2137 } 2138 2139 fail: 2140 migrate_set_state(&s->state, current_active_state, 2141 MIGRATION_STATUS_FAILED); 2142 } 2143 2144 bool migrate_colo_enabled(void) 2145 { 2146 MigrationState *s = migrate_get_current(); 2147 return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO]; 2148 } 2149 2150 static void migration_calculate_complete(MigrationState *s) 2151 { 2152 uint64_t bytes = qemu_ftell(s->to_dst_file); 2153 int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 2154 2155 s->total_time = end_time - s->start_time; 2156 if (!s->downtime) { 2157 /* 2158 * It's still not set, so we are precopy migration. For 2159 * postcopy, downtime is calculated during postcopy_start(). 2160 */ 2161 s->downtime = end_time - s->downtime_start; 2162 } 2163 2164 if (s->total_time) { 2165 s->mbps = ((double) bytes * 8.0) / s->total_time / 1000; 2166 } 2167 } 2168 2169 static void migration_update_counters(MigrationState *s, 2170 int64_t current_time) 2171 { 2172 uint64_t transferred, time_spent; 2173 double bandwidth; 2174 2175 if (current_time < s->iteration_start_time + BUFFER_DELAY) { 2176 return; 2177 } 2178 2179 transferred = qemu_ftell(s->to_dst_file) - s->iteration_initial_bytes; 2180 time_spent = current_time - s->iteration_start_time; 2181 bandwidth = (double)transferred / time_spent; 2182 s->threshold_size = bandwidth * s->parameters.downtime_limit; 2183 2184 s->mbps = (((double) transferred * 8.0) / 2185 ((double) time_spent / 1000.0)) / 1000.0 / 1000.0; 2186 2187 /* 2188 * if we haven't sent anything, we don't want to 2189 * recalculate. 10000 is a small enough number for our purposes 2190 */ 2191 if (ram_counters.dirty_pages_rate && transferred > 10000) { 2192 s->expected_downtime = ram_counters.dirty_pages_rate * 2193 qemu_target_page_size() / bandwidth; 2194 } 2195 2196 qemu_file_reset_rate_limit(s->to_dst_file); 2197 2198 s->iteration_start_time = current_time; 2199 s->iteration_initial_bytes = qemu_ftell(s->to_dst_file); 2200 2201 trace_migrate_transferred(transferred, time_spent, 2202 bandwidth, s->threshold_size); 2203 } 2204 2205 /* Migration thread iteration status */ 2206 typedef enum { 2207 MIG_ITERATE_RESUME, /* Resume current iteration */ 2208 MIG_ITERATE_SKIP, /* Skip current iteration */ 2209 MIG_ITERATE_BREAK, /* Break the loop */ 2210 } MigIterateState; 2211 2212 /* 2213 * Return true if continue to the next iteration directly, false 2214 * otherwise. 2215 */ 2216 static MigIterateState migration_iteration_run(MigrationState *s) 2217 { 2218 uint64_t pending_size, pend_post, pend_nonpost; 2219 bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE; 2220 2221 qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, 2222 &pend_nonpost, &pend_post); 2223 pending_size = pend_nonpost + pend_post; 2224 2225 trace_migrate_pending(pending_size, s->threshold_size, 2226 pend_post, pend_nonpost); 2227 2228 if (pending_size && pending_size >= s->threshold_size) { 2229 /* Still a significant amount to transfer */ 2230 if (migrate_postcopy() && !in_postcopy && 2231 pend_nonpost <= s->threshold_size && 2232 atomic_read(&s->start_postcopy)) { 2233 if (postcopy_start(s)) { 2234 error_report("%s: postcopy failed to start", __func__); 2235 } 2236 return MIG_ITERATE_SKIP; 2237 } 2238 /* Just another iteration step */ 2239 qemu_savevm_state_iterate(s->to_dst_file, 2240 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE); 2241 } else { 2242 trace_migration_thread_low_pending(pending_size); 2243 migration_completion(s); 2244 return MIG_ITERATE_BREAK; 2245 } 2246 2247 return MIG_ITERATE_RESUME; 2248 } 2249 2250 static void migration_iteration_finish(MigrationState *s) 2251 { 2252 /* If we enabled cpu throttling for auto-converge, turn it off. */ 2253 cpu_throttle_stop(); 2254 2255 qemu_mutex_lock_iothread(); 2256 switch (s->state) { 2257 case MIGRATION_STATUS_COMPLETED: 2258 migration_calculate_complete(s); 2259 runstate_set(RUN_STATE_POSTMIGRATE); 2260 break; 2261 2262 case MIGRATION_STATUS_ACTIVE: 2263 /* 2264 * We should really assert here, but since it's during 2265 * migration, let's try to reduce the usage of assertions. 2266 */ 2267 if (!migrate_colo_enabled()) { 2268 error_report("%s: critical error: calling COLO code without " 2269 "COLO enabled", __func__); 2270 } 2271 migrate_start_colo_process(s); 2272 /* 2273 * Fixme: we will run VM in COLO no matter its old running state. 2274 * After exited COLO, we will keep running. 2275 */ 2276 s->vm_was_running = true; 2277 /* Fallthrough */ 2278 case MIGRATION_STATUS_FAILED: 2279 case MIGRATION_STATUS_CANCELLED: 2280 if (s->vm_was_running) { 2281 vm_start(); 2282 } else { 2283 if (runstate_check(RUN_STATE_FINISH_MIGRATE)) { 2284 runstate_set(RUN_STATE_POSTMIGRATE); 2285 } 2286 } 2287 break; 2288 2289 default: 2290 /* Should not reach here, but if so, forgive the VM. */ 2291 error_report("%s: Unknown ending state %d", __func__, s->state); 2292 break; 2293 } 2294 qemu_bh_schedule(s->cleanup_bh); 2295 qemu_mutex_unlock_iothread(); 2296 } 2297 2298 /* 2299 * Master migration thread on the source VM. 2300 * It drives the migration and pumps the data down the outgoing channel. 2301 */ 2302 static void *migration_thread(void *opaque) 2303 { 2304 MigrationState *s = opaque; 2305 int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST); 2306 2307 rcu_register_thread(); 2308 2309 s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 2310 2311 qemu_savevm_state_header(s->to_dst_file); 2312 2313 /* 2314 * If we opened the return path, we need to make sure dst has it 2315 * opened as well. 2316 */ 2317 if (s->rp_state.from_dst_file) { 2318 /* Now tell the dest that it should open its end so it can reply */ 2319 qemu_savevm_send_open_return_path(s->to_dst_file); 2320 2321 /* And do a ping that will make stuff easier to debug */ 2322 qemu_savevm_send_ping(s->to_dst_file, 1); 2323 } 2324 2325 if (migrate_postcopy()) { 2326 /* 2327 * Tell the destination that we *might* want to do postcopy later; 2328 * if the other end can't do postcopy it should fail now, nice and 2329 * early. 2330 */ 2331 qemu_savevm_send_postcopy_advise(s->to_dst_file); 2332 } 2333 2334 qemu_savevm_state_setup(s->to_dst_file); 2335 2336 s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start; 2337 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, 2338 MIGRATION_STATUS_ACTIVE); 2339 2340 trace_migration_thread_setup_complete(); 2341 2342 while (s->state == MIGRATION_STATUS_ACTIVE || 2343 s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) { 2344 int64_t current_time; 2345 2346 if (!qemu_file_rate_limit(s->to_dst_file)) { 2347 MigIterateState iter_state = migration_iteration_run(s); 2348 if (iter_state == MIG_ITERATE_SKIP) { 2349 continue; 2350 } else if (iter_state == MIG_ITERATE_BREAK) { 2351 break; 2352 } 2353 } 2354 2355 if (qemu_file_get_error(s->to_dst_file)) { 2356 if (migration_is_setup_or_active(s->state)) { 2357 migrate_set_state(&s->state, s->state, 2358 MIGRATION_STATUS_FAILED); 2359 } 2360 trace_migration_thread_file_err(); 2361 break; 2362 } 2363 2364 current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME); 2365 2366 migration_update_counters(s, current_time); 2367 2368 if (qemu_file_rate_limit(s->to_dst_file)) { 2369 /* usleep expects microseconds */ 2370 g_usleep((s->iteration_start_time + BUFFER_DELAY - 2371 current_time) * 1000); 2372 } 2373 } 2374 2375 trace_migration_thread_after_loop(); 2376 migration_iteration_finish(s); 2377 rcu_unregister_thread(); 2378 return NULL; 2379 } 2380 2381 void migrate_fd_connect(MigrationState *s, Error *error_in) 2382 { 2383 s->expected_downtime = s->parameters.downtime_limit; 2384 s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s); 2385 if (error_in) { 2386 migrate_fd_error(s, error_in); 2387 migrate_fd_cleanup(s); 2388 return; 2389 } 2390 2391 qemu_file_set_blocking(s->to_dst_file, true); 2392 qemu_file_set_rate_limit(s->to_dst_file, 2393 s->parameters.max_bandwidth / XFER_LIMIT_RATIO); 2394 2395 /* Notify before starting migration thread */ 2396 notifier_list_notify(&migration_state_notifiers, s); 2397 2398 /* 2399 * Open the return path. For postcopy, it is used exclusively. For 2400 * precopy, only if user specified "return-path" capability would 2401 * QEMU uses the return path. 2402 */ 2403 if (migrate_postcopy_ram() || migrate_use_return_path()) { 2404 if (open_return_path_on_source(s)) { 2405 error_report("Unable to open return-path for postcopy"); 2406 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, 2407 MIGRATION_STATUS_FAILED); 2408 migrate_fd_cleanup(s); 2409 return; 2410 } 2411 } 2412 2413 if (multifd_save_setup() != 0) { 2414 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, 2415 MIGRATION_STATUS_FAILED); 2416 migrate_fd_cleanup(s); 2417 return; 2418 } 2419 qemu_thread_create(&s->thread, "live_migration", migration_thread, s, 2420 QEMU_THREAD_JOINABLE); 2421 s->migration_thread_running = true; 2422 } 2423 2424 void migration_global_dump(Monitor *mon) 2425 { 2426 MigrationState *ms = migrate_get_current(); 2427 2428 monitor_printf(mon, "globals:\n"); 2429 monitor_printf(mon, "store-global-state: %s\n", 2430 ms->store_global_state ? "on" : "off"); 2431 monitor_printf(mon, "only-migratable: %s\n", 2432 ms->only_migratable ? "on" : "off"); 2433 monitor_printf(mon, "send-configuration: %s\n", 2434 ms->send_configuration ? "on" : "off"); 2435 monitor_printf(mon, "send-section-footer: %s\n", 2436 ms->send_section_footer ? "on" : "off"); 2437 } 2438 2439 #define DEFINE_PROP_MIG_CAP(name, x) \ 2440 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false) 2441 2442 static Property migration_properties[] = { 2443 DEFINE_PROP_BOOL("store-global-state", MigrationState, 2444 store_global_state, true), 2445 DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false), 2446 DEFINE_PROP_BOOL("send-configuration", MigrationState, 2447 send_configuration, true), 2448 DEFINE_PROP_BOOL("send-section-footer", MigrationState, 2449 send_section_footer, true), 2450 2451 /* Migration parameters */ 2452 DEFINE_PROP_UINT8("x-compress-level", MigrationState, 2453 parameters.compress_level, 2454 DEFAULT_MIGRATE_COMPRESS_LEVEL), 2455 DEFINE_PROP_UINT8("x-compress-threads", MigrationState, 2456 parameters.compress_threads, 2457 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT), 2458 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState, 2459 parameters.decompress_threads, 2460 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT), 2461 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState, 2462 parameters.cpu_throttle_initial, 2463 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL), 2464 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState, 2465 parameters.cpu_throttle_increment, 2466 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT), 2467 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState, 2468 parameters.max_bandwidth, MAX_THROTTLE), 2469 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState, 2470 parameters.downtime_limit, 2471 DEFAULT_MIGRATE_SET_DOWNTIME), 2472 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState, 2473 parameters.x_checkpoint_delay, 2474 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY), 2475 DEFINE_PROP_UINT8("x-multifd-channels", MigrationState, 2476 parameters.x_multifd_channels, 2477 DEFAULT_MIGRATE_MULTIFD_CHANNELS), 2478 DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState, 2479 parameters.x_multifd_page_count, 2480 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT), 2481 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState, 2482 parameters.xbzrle_cache_size, 2483 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE), 2484 2485 /* Migration capabilities */ 2486 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE), 2487 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL), 2488 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE), 2489 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS), 2490 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS), 2491 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS), 2492 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM), 2493 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO), 2494 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM), 2495 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK), 2496 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH), 2497 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD), 2498 2499 DEFINE_PROP_END_OF_LIST(), 2500 }; 2501 2502 static void migration_class_init(ObjectClass *klass, void *data) 2503 { 2504 DeviceClass *dc = DEVICE_CLASS(klass); 2505 2506 dc->user_creatable = false; 2507 dc->props = migration_properties; 2508 } 2509 2510 static void migration_instance_finalize(Object *obj) 2511 { 2512 MigrationState *ms = MIGRATION_OBJ(obj); 2513 MigrationParameters *params = &ms->parameters; 2514 2515 qemu_mutex_destroy(&ms->error_mutex); 2516 g_free(params->tls_hostname); 2517 g_free(params->tls_creds); 2518 qemu_sem_destroy(&ms->pause_sem); 2519 } 2520 2521 static void migration_instance_init(Object *obj) 2522 { 2523 MigrationState *ms = MIGRATION_OBJ(obj); 2524 MigrationParameters *params = &ms->parameters; 2525 2526 ms->state = MIGRATION_STATUS_NONE; 2527 ms->mbps = -1; 2528 qemu_sem_init(&ms->pause_sem, 0); 2529 qemu_mutex_init(&ms->error_mutex); 2530 2531 params->tls_hostname = g_strdup(""); 2532 params->tls_creds = g_strdup(""); 2533 2534 /* Set has_* up only for parameter checks */ 2535 params->has_compress_level = true; 2536 params->has_compress_threads = true; 2537 params->has_decompress_threads = true; 2538 params->has_cpu_throttle_initial = true; 2539 params->has_cpu_throttle_increment = true; 2540 params->has_max_bandwidth = true; 2541 params->has_downtime_limit = true; 2542 params->has_x_checkpoint_delay = true; 2543 params->has_block_incremental = true; 2544 params->has_x_multifd_channels = true; 2545 params->has_x_multifd_page_count = true; 2546 params->has_xbzrle_cache_size = true; 2547 } 2548 2549 /* 2550 * Return true if check pass, false otherwise. Error will be put 2551 * inside errp if provided. 2552 */ 2553 static bool migration_object_check(MigrationState *ms, Error **errp) 2554 { 2555 MigrationCapabilityStatusList *head = NULL; 2556 /* Assuming all off */ 2557 bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret; 2558 int i; 2559 2560 if (!migrate_params_check(&ms->parameters, errp)) { 2561 return false; 2562 } 2563 2564 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { 2565 if (ms->enabled_capabilities[i]) { 2566 head = migrate_cap_add(head, i, true); 2567 } 2568 } 2569 2570 ret = migrate_caps_check(cap_list, head, errp); 2571 2572 /* It works with head == NULL */ 2573 qapi_free_MigrationCapabilityStatusList(head); 2574 2575 return ret; 2576 } 2577 2578 static const TypeInfo migration_type = { 2579 .name = TYPE_MIGRATION, 2580 /* 2581 * NOTE: TYPE_MIGRATION is not really a device, as the object is 2582 * not created using qdev_create(), it is not attached to the qdev 2583 * device tree, and it is never realized. 2584 * 2585 * TODO: Make this TYPE_OBJECT once QOM provides something like 2586 * TYPE_DEVICE's "-global" properties. 2587 */ 2588 .parent = TYPE_DEVICE, 2589 .class_init = migration_class_init, 2590 .class_size = sizeof(MigrationClass), 2591 .instance_size = sizeof(MigrationState), 2592 .instance_init = migration_instance_init, 2593 .instance_finalize = migration_instance_finalize, 2594 }; 2595 2596 static void register_migration_types(void) 2597 { 2598 type_register_static(&migration_type); 2599 } 2600 2601 type_init(register_migration_types); 2602