1 /* 2 * Multifd common code 3 * 4 * Copyright (c) 2019-2020 Red Hat Inc 5 * 6 * Authors: 7 * Juan Quintela <quintela@redhat.com> 8 * 9 * This work is licensed under the terms of the GNU GPL, version 2 or later. 10 * See the COPYING file in the top-level directory. 11 */ 12 13 #include "qemu/osdep.h" 14 #include "qemu/cutils.h" 15 #include "qemu/rcu.h" 16 #include "exec/target_page.h" 17 #include "sysemu/sysemu.h" 18 #include "exec/ramblock.h" 19 #include "qemu/error-report.h" 20 #include "qapi/error.h" 21 #include "file.h" 22 #include "migration.h" 23 #include "migration-stats.h" 24 #include "socket.h" 25 #include "tls.h" 26 #include "qemu-file.h" 27 #include "trace.h" 28 #include "multifd.h" 29 #include "threadinfo.h" 30 #include "options.h" 31 #include "qemu/yank.h" 32 #include "io/channel-file.h" 33 #include "io/channel-socket.h" 34 #include "yank_functions.h" 35 36 /* Multiple fd's */ 37 38 #define MULTIFD_MAGIC 0x11223344U 39 #define MULTIFD_VERSION 1 40 41 typedef struct { 42 uint32_t magic; 43 uint32_t version; 44 unsigned char uuid[16]; /* QemuUUID */ 45 uint8_t id; 46 uint8_t unused1[7]; /* Reserved for future use */ 47 uint64_t unused2[4]; /* Reserved for future use */ 48 } __attribute__((packed)) MultiFDInit_t; 49 50 struct { 51 MultiFDSendParams *params; 52 /* 53 * Global number of generated multifd packets. 54 * 55 * Note that we used 'uintptr_t' because it'll naturally support atomic 56 * operations on both 32bit / 64 bits hosts. It means on 32bit systems 57 * multifd will overflow the packet_num easier, but that should be 58 * fine. 59 * 60 * Another option is to use QEMU's Stat64 then it'll be 64 bits on all 61 * hosts, however so far it does not support atomic fetch_add() yet. 62 * Make it easy for now. 63 */ 64 uintptr_t packet_num; 65 /* 66 * Synchronization point past which no more channels will be 67 * created. 68 */ 69 QemuSemaphore channels_created; 70 /* send channels ready */ 71 QemuSemaphore channels_ready; 72 /* 73 * Have we already run terminate threads. There is a race when it 74 * happens that we got one error while we are exiting. 75 * We will use atomic operations. Only valid values are 0 and 1. 76 */ 77 int exiting; 78 /* multifd ops */ 79 const MultiFDMethods *ops; 80 } *multifd_send_state; 81 82 struct { 83 MultiFDRecvParams *params; 84 MultiFDRecvData *data; 85 /* number of created threads */ 86 int count; 87 /* 88 * This is always posted by the recv threads, the migration thread 89 * uses it to wait for recv threads to finish assigned tasks. 90 */ 91 QemuSemaphore sem_sync; 92 /* global number of generated multifd packets */ 93 uint64_t packet_num; 94 int exiting; 95 /* multifd ops */ 96 const MultiFDMethods *ops; 97 } *multifd_recv_state; 98 99 MultiFDSendData *multifd_send_data_alloc(void) 100 { 101 size_t max_payload_size, size_minus_payload; 102 103 /* 104 * MultiFDPages_t has a flexible array at the end, account for it 105 * when allocating MultiFDSendData. Use max() in case other types 106 * added to the union in the future are larger than 107 * (MultiFDPages_t + flex array). 108 */ 109 max_payload_size = MAX(multifd_ram_payload_size(), sizeof(MultiFDPayload)); 110 111 /* 112 * Account for any holes the compiler might insert. We can't pack 113 * the structure because that misaligns the members and triggers 114 * Waddress-of-packed-member. 115 */ 116 size_minus_payload = sizeof(MultiFDSendData) - sizeof(MultiFDPayload); 117 118 return g_malloc0(size_minus_payload + max_payload_size); 119 } 120 121 static bool multifd_use_packets(void) 122 { 123 return !migrate_mapped_ram(); 124 } 125 126 void multifd_send_channel_created(void) 127 { 128 qemu_sem_post(&multifd_send_state->channels_created); 129 } 130 131 static const MultiFDMethods *multifd_ops[MULTIFD_COMPRESSION__MAX] = {}; 132 133 void multifd_register_ops(int method, const MultiFDMethods *ops) 134 { 135 assert(0 <= method && method < MULTIFD_COMPRESSION__MAX); 136 assert(!multifd_ops[method]); 137 multifd_ops[method] = ops; 138 } 139 140 static int multifd_send_initial_packet(MultiFDSendParams *p, Error **errp) 141 { 142 MultiFDInit_t msg = {}; 143 size_t size = sizeof(msg); 144 int ret; 145 146 msg.magic = cpu_to_be32(MULTIFD_MAGIC); 147 msg.version = cpu_to_be32(MULTIFD_VERSION); 148 msg.id = p->id; 149 memcpy(msg.uuid, &qemu_uuid.data, sizeof(msg.uuid)); 150 151 ret = qio_channel_write_all(p->c, (char *)&msg, size, errp); 152 if (ret != 0) { 153 return -1; 154 } 155 stat64_add(&mig_stats.multifd_bytes, size); 156 return 0; 157 } 158 159 static int multifd_recv_initial_packet(QIOChannel *c, Error **errp) 160 { 161 MultiFDInit_t msg; 162 int ret; 163 164 ret = qio_channel_read_all(c, (char *)&msg, sizeof(msg), errp); 165 if (ret != 0) { 166 return -1; 167 } 168 169 msg.magic = be32_to_cpu(msg.magic); 170 msg.version = be32_to_cpu(msg.version); 171 172 if (msg.magic != MULTIFD_MAGIC) { 173 error_setg(errp, "multifd: received packet magic %x " 174 "expected %x", msg.magic, MULTIFD_MAGIC); 175 return -1; 176 } 177 178 if (msg.version != MULTIFD_VERSION) { 179 error_setg(errp, "multifd: received packet version %u " 180 "expected %u", msg.version, MULTIFD_VERSION); 181 return -1; 182 } 183 184 if (memcmp(msg.uuid, &qemu_uuid, sizeof(qemu_uuid))) { 185 char *uuid = qemu_uuid_unparse_strdup(&qemu_uuid); 186 char *msg_uuid = qemu_uuid_unparse_strdup((const QemuUUID *)msg.uuid); 187 188 error_setg(errp, "multifd: received uuid '%s' and expected " 189 "uuid '%s' for channel %hhd", msg_uuid, uuid, msg.id); 190 g_free(uuid); 191 g_free(msg_uuid); 192 return -1; 193 } 194 195 if (msg.id > migrate_multifd_channels()) { 196 error_setg(errp, "multifd: received channel id %u is greater than " 197 "number of channels %u", msg.id, migrate_multifd_channels()); 198 return -1; 199 } 200 201 return msg.id; 202 } 203 204 void multifd_send_fill_packet(MultiFDSendParams *p) 205 { 206 MultiFDPacket_t *packet = p->packet; 207 uint64_t packet_num; 208 bool sync_packet = p->flags & MULTIFD_FLAG_SYNC; 209 210 memset(packet, 0, p->packet_len); 211 212 packet->magic = cpu_to_be32(MULTIFD_MAGIC); 213 packet->version = cpu_to_be32(MULTIFD_VERSION); 214 215 packet->flags = cpu_to_be32(p->flags); 216 packet->next_packet_size = cpu_to_be32(p->next_packet_size); 217 218 packet_num = qatomic_fetch_inc(&multifd_send_state->packet_num); 219 packet->packet_num = cpu_to_be64(packet_num); 220 221 p->packets_sent++; 222 223 if (!sync_packet) { 224 multifd_ram_fill_packet(p); 225 } 226 227 trace_multifd_send_fill(p->id, packet_num, 228 p->flags, p->next_packet_size); 229 } 230 231 static int multifd_recv_unfill_packet(MultiFDRecvParams *p, Error **errp) 232 { 233 const MultiFDPacket_t *packet = p->packet; 234 uint32_t magic = be32_to_cpu(packet->magic); 235 uint32_t version = be32_to_cpu(packet->version); 236 int ret = 0; 237 238 if (magic != MULTIFD_MAGIC) { 239 error_setg(errp, "multifd: received packet magic %x, expected %x", 240 magic, MULTIFD_MAGIC); 241 return -1; 242 } 243 244 if (version != MULTIFD_VERSION) { 245 error_setg(errp, "multifd: received packet version %u, expected %u", 246 version, MULTIFD_VERSION); 247 return -1; 248 } 249 250 p->flags = be32_to_cpu(packet->flags); 251 p->next_packet_size = be32_to_cpu(packet->next_packet_size); 252 p->packet_num = be64_to_cpu(packet->packet_num); 253 p->packets_recved++; 254 255 if (!(p->flags & MULTIFD_FLAG_SYNC)) { 256 ret = multifd_ram_unfill_packet(p, errp); 257 } 258 259 trace_multifd_recv_unfill(p->id, p->packet_num, p->flags, 260 p->next_packet_size); 261 262 return ret; 263 } 264 265 static bool multifd_send_should_exit(void) 266 { 267 return qatomic_read(&multifd_send_state->exiting); 268 } 269 270 static bool multifd_recv_should_exit(void) 271 { 272 return qatomic_read(&multifd_recv_state->exiting); 273 } 274 275 /* 276 * The migration thread can wait on either of the two semaphores. This 277 * function can be used to kick the main thread out of waiting on either of 278 * them. Should mostly only be called when something wrong happened with 279 * the current multifd send thread. 280 */ 281 static void multifd_send_kick_main(MultiFDSendParams *p) 282 { 283 qemu_sem_post(&p->sem_sync); 284 qemu_sem_post(&multifd_send_state->channels_ready); 285 } 286 287 /* 288 * multifd_send() works by exchanging the MultiFDSendData object 289 * provided by the caller with an unused MultiFDSendData object from 290 * the next channel that is found to be idle. 291 * 292 * The channel owns the data until it finishes transmitting and the 293 * caller owns the empty object until it fills it with data and calls 294 * this function again. No locking necessary. 295 * 296 * Switching is safe because both the migration thread and the channel 297 * thread have barriers in place to serialize access. 298 * 299 * Returns true if succeed, false otherwise. 300 */ 301 bool multifd_send(MultiFDSendData **send_data) 302 { 303 int i; 304 static int next_channel; 305 MultiFDSendParams *p = NULL; /* make happy gcc */ 306 MultiFDSendData *tmp; 307 308 if (multifd_send_should_exit()) { 309 return false; 310 } 311 312 /* We wait here, until at least one channel is ready */ 313 qemu_sem_wait(&multifd_send_state->channels_ready); 314 315 /* 316 * next_channel can remain from a previous migration that was 317 * using more channels, so ensure it doesn't overflow if the 318 * limit is lower now. 319 */ 320 next_channel %= migrate_multifd_channels(); 321 for (i = next_channel;; i = (i + 1) % migrate_multifd_channels()) { 322 if (multifd_send_should_exit()) { 323 return false; 324 } 325 p = &multifd_send_state->params[i]; 326 /* 327 * Lockless read to p->pending_job is safe, because only multifd 328 * sender thread can clear it. 329 */ 330 if (qatomic_read(&p->pending_job) == false) { 331 next_channel = (i + 1) % migrate_multifd_channels(); 332 break; 333 } 334 } 335 336 /* 337 * Make sure we read p->pending_job before all the rest. Pairs with 338 * qatomic_store_release() in multifd_send_thread(). 339 */ 340 smp_mb_acquire(); 341 342 assert(multifd_payload_empty(p->data)); 343 344 /* 345 * Swap the pointers. The channel gets the client data for 346 * transferring and the client gets back an unused data slot. 347 */ 348 tmp = *send_data; 349 *send_data = p->data; 350 p->data = tmp; 351 352 /* 353 * Making sure p->data is setup before marking pending_job=true. Pairs 354 * with the qatomic_load_acquire() in multifd_send_thread(). 355 */ 356 qatomic_store_release(&p->pending_job, true); 357 qemu_sem_post(&p->sem); 358 359 return true; 360 } 361 362 /* Multifd send side hit an error; remember it and prepare to quit */ 363 static void multifd_send_set_error(Error *err) 364 { 365 /* 366 * We don't want to exit each threads twice. Depending on where 367 * we get the error, or if there are two independent errors in two 368 * threads at the same time, we can end calling this function 369 * twice. 370 */ 371 if (qatomic_xchg(&multifd_send_state->exiting, 1)) { 372 return; 373 } 374 375 if (err) { 376 MigrationState *s = migrate_get_current(); 377 migrate_set_error(s, err); 378 if (s->state == MIGRATION_STATUS_SETUP || 379 s->state == MIGRATION_STATUS_PRE_SWITCHOVER || 380 s->state == MIGRATION_STATUS_DEVICE || 381 s->state == MIGRATION_STATUS_ACTIVE) { 382 migrate_set_state(&s->state, s->state, 383 MIGRATION_STATUS_FAILED); 384 } 385 } 386 } 387 388 static void multifd_send_terminate_threads(void) 389 { 390 int i; 391 392 trace_multifd_send_terminate_threads(); 393 394 /* 395 * Tell everyone we're quitting. No xchg() needed here; we simply 396 * always set it. 397 */ 398 qatomic_set(&multifd_send_state->exiting, 1); 399 400 /* 401 * Firstly, kick all threads out; no matter whether they are just idle, 402 * or blocked in an IO system call. 403 */ 404 for (i = 0; i < migrate_multifd_channels(); i++) { 405 MultiFDSendParams *p = &multifd_send_state->params[i]; 406 407 qemu_sem_post(&p->sem); 408 if (p->c) { 409 qio_channel_shutdown(p->c, QIO_CHANNEL_SHUTDOWN_BOTH, NULL); 410 } 411 } 412 413 /* 414 * Finally recycle all the threads. 415 */ 416 for (i = 0; i < migrate_multifd_channels(); i++) { 417 MultiFDSendParams *p = &multifd_send_state->params[i]; 418 419 if (p->tls_thread_created) { 420 qemu_thread_join(&p->tls_thread); 421 } 422 423 if (p->thread_created) { 424 qemu_thread_join(&p->thread); 425 } 426 } 427 } 428 429 static bool multifd_send_cleanup_channel(MultiFDSendParams *p, Error **errp) 430 { 431 if (p->c) { 432 migration_ioc_unregister_yank(p->c); 433 /* 434 * The object_unref() cannot guarantee the fd will always be 435 * released because finalize() of the iochannel is only 436 * triggered on the last reference and it's not guaranteed 437 * that we always hold the last refcount when reaching here. 438 * 439 * Closing the fd explicitly has the benefit that if there is any 440 * registered I/O handler callbacks on such fd, that will get a 441 * POLLNVAL event and will further trigger the cleanup to finally 442 * release the IOC. 443 * 444 * FIXME: It should logically be guaranteed that all multifd 445 * channels have no I/O handler callback registered when reaching 446 * here, because migration thread will wait for all multifd channel 447 * establishments to complete during setup. Since 448 * migrate_fd_cleanup() will be scheduled in main thread too, all 449 * previous callbacks should guarantee to be completed when 450 * reaching here. See multifd_send_state.channels_created and its 451 * usage. In the future, we could replace this with an assert 452 * making sure we're the last reference, or simply drop it if above 453 * is more clear to be justified. 454 */ 455 qio_channel_close(p->c, &error_abort); 456 object_unref(OBJECT(p->c)); 457 p->c = NULL; 458 } 459 qemu_sem_destroy(&p->sem); 460 qemu_sem_destroy(&p->sem_sync); 461 g_free(p->name); 462 p->name = NULL; 463 g_free(p->data); 464 p->data = NULL; 465 p->packet_len = 0; 466 g_free(p->packet); 467 p->packet = NULL; 468 multifd_send_state->ops->send_cleanup(p, errp); 469 assert(!p->iov); 470 471 return *errp == NULL; 472 } 473 474 static void multifd_send_cleanup_state(void) 475 { 476 file_cleanup_outgoing_migration(); 477 socket_cleanup_outgoing_migration(); 478 qemu_sem_destroy(&multifd_send_state->channels_created); 479 qemu_sem_destroy(&multifd_send_state->channels_ready); 480 g_free(multifd_send_state->params); 481 multifd_send_state->params = NULL; 482 g_free(multifd_send_state); 483 multifd_send_state = NULL; 484 } 485 486 void multifd_send_shutdown(void) 487 { 488 int i; 489 490 if (!migrate_multifd()) { 491 return; 492 } 493 494 multifd_send_terminate_threads(); 495 496 for (i = 0; i < migrate_multifd_channels(); i++) { 497 MultiFDSendParams *p = &multifd_send_state->params[i]; 498 Error *local_err = NULL; 499 500 if (!multifd_send_cleanup_channel(p, &local_err)) { 501 migrate_set_error(migrate_get_current(), local_err); 502 error_free(local_err); 503 } 504 } 505 506 multifd_send_cleanup_state(); 507 } 508 509 static int multifd_zero_copy_flush(QIOChannel *c) 510 { 511 int ret; 512 Error *err = NULL; 513 514 ret = qio_channel_flush(c, &err); 515 if (ret < 0) { 516 error_report_err(err); 517 return -1; 518 } 519 if (ret == 1) { 520 stat64_add(&mig_stats.dirty_sync_missed_zero_copy, 1); 521 } 522 523 return ret; 524 } 525 526 int multifd_send_sync_main(void) 527 { 528 int i; 529 bool flush_zero_copy; 530 531 flush_zero_copy = migrate_zero_copy_send(); 532 533 for (i = 0; i < migrate_multifd_channels(); i++) { 534 MultiFDSendParams *p = &multifd_send_state->params[i]; 535 536 if (multifd_send_should_exit()) { 537 return -1; 538 } 539 540 trace_multifd_send_sync_main_signal(p->id); 541 542 /* 543 * We should be the only user so far, so not possible to be set by 544 * others concurrently. 545 */ 546 assert(qatomic_read(&p->pending_sync) == false); 547 qatomic_set(&p->pending_sync, true); 548 qemu_sem_post(&p->sem); 549 } 550 for (i = 0; i < migrate_multifd_channels(); i++) { 551 MultiFDSendParams *p = &multifd_send_state->params[i]; 552 553 if (multifd_send_should_exit()) { 554 return -1; 555 } 556 557 qemu_sem_wait(&multifd_send_state->channels_ready); 558 trace_multifd_send_sync_main_wait(p->id); 559 qemu_sem_wait(&p->sem_sync); 560 561 if (flush_zero_copy && p->c && (multifd_zero_copy_flush(p->c) < 0)) { 562 return -1; 563 } 564 } 565 trace_multifd_send_sync_main(multifd_send_state->packet_num); 566 567 return 0; 568 } 569 570 static void *multifd_send_thread(void *opaque) 571 { 572 MultiFDSendParams *p = opaque; 573 MigrationThread *thread = NULL; 574 Error *local_err = NULL; 575 int ret = 0; 576 bool use_packets = multifd_use_packets(); 577 578 thread = migration_threads_add(p->name, qemu_get_thread_id()); 579 580 trace_multifd_send_thread_start(p->id); 581 rcu_register_thread(); 582 583 if (use_packets) { 584 if (multifd_send_initial_packet(p, &local_err) < 0) { 585 ret = -1; 586 goto out; 587 } 588 } 589 590 while (true) { 591 qemu_sem_post(&multifd_send_state->channels_ready); 592 qemu_sem_wait(&p->sem); 593 594 if (multifd_send_should_exit()) { 595 break; 596 } 597 598 /* 599 * Read pending_job flag before p->data. Pairs with the 600 * qatomic_store_release() in multifd_send(). 601 */ 602 if (qatomic_load_acquire(&p->pending_job)) { 603 p->iovs_num = 0; 604 assert(!multifd_payload_empty(p->data)); 605 606 ret = multifd_send_state->ops->send_prepare(p, &local_err); 607 if (ret != 0) { 608 break; 609 } 610 611 if (migrate_mapped_ram()) { 612 ret = file_write_ramblock_iov(p->c, p->iov, p->iovs_num, 613 &p->data->u.ram, &local_err); 614 } else { 615 ret = qio_channel_writev_full_all(p->c, p->iov, p->iovs_num, 616 NULL, 0, p->write_flags, 617 &local_err); 618 } 619 620 if (ret != 0) { 621 break; 622 } 623 624 stat64_add(&mig_stats.multifd_bytes, 625 p->next_packet_size + p->packet_len); 626 627 p->next_packet_size = 0; 628 multifd_set_payload_type(p->data, MULTIFD_PAYLOAD_NONE); 629 630 /* 631 * Making sure p->data is published before saying "we're 632 * free". Pairs with the smp_mb_acquire() in 633 * multifd_send(). 634 */ 635 qatomic_store_release(&p->pending_job, false); 636 } else { 637 /* 638 * If not a normal job, must be a sync request. Note that 639 * pending_sync is a standalone flag (unlike pending_job), so 640 * it doesn't require explicit memory barriers. 641 */ 642 assert(qatomic_read(&p->pending_sync)); 643 644 if (use_packets) { 645 p->flags = MULTIFD_FLAG_SYNC; 646 multifd_send_fill_packet(p); 647 ret = qio_channel_write_all(p->c, (void *)p->packet, 648 p->packet_len, &local_err); 649 if (ret != 0) { 650 break; 651 } 652 /* p->next_packet_size will always be zero for a SYNC packet */ 653 stat64_add(&mig_stats.multifd_bytes, p->packet_len); 654 p->flags = 0; 655 } 656 657 qatomic_set(&p->pending_sync, false); 658 qemu_sem_post(&p->sem_sync); 659 } 660 } 661 662 out: 663 if (ret) { 664 assert(local_err); 665 trace_multifd_send_error(p->id); 666 multifd_send_set_error(local_err); 667 multifd_send_kick_main(p); 668 error_free(local_err); 669 } 670 671 rcu_unregister_thread(); 672 migration_threads_remove(thread); 673 trace_multifd_send_thread_end(p->id, p->packets_sent); 674 675 return NULL; 676 } 677 678 static void multifd_new_send_channel_async(QIOTask *task, gpointer opaque); 679 680 typedef struct { 681 MultiFDSendParams *p; 682 QIOChannelTLS *tioc; 683 } MultiFDTLSThreadArgs; 684 685 static void *multifd_tls_handshake_thread(void *opaque) 686 { 687 MultiFDTLSThreadArgs *args = opaque; 688 689 qio_channel_tls_handshake(args->tioc, 690 multifd_new_send_channel_async, 691 args->p, 692 NULL, 693 NULL); 694 g_free(args); 695 696 return NULL; 697 } 698 699 static bool multifd_tls_channel_connect(MultiFDSendParams *p, 700 QIOChannel *ioc, 701 Error **errp) 702 { 703 MigrationState *s = migrate_get_current(); 704 const char *hostname = s->hostname; 705 MultiFDTLSThreadArgs *args; 706 QIOChannelTLS *tioc; 707 708 tioc = migration_tls_client_create(ioc, hostname, errp); 709 if (!tioc) { 710 return false; 711 } 712 713 /* 714 * Ownership of the socket channel now transfers to the newly 715 * created TLS channel, which has already taken a reference. 716 */ 717 object_unref(OBJECT(ioc)); 718 trace_multifd_tls_outgoing_handshake_start(ioc, tioc, hostname); 719 qio_channel_set_name(QIO_CHANNEL(tioc), "multifd-tls-outgoing"); 720 721 args = g_new0(MultiFDTLSThreadArgs, 1); 722 args->tioc = tioc; 723 args->p = p; 724 725 p->tls_thread_created = true; 726 qemu_thread_create(&p->tls_thread, "mig/src/tls", 727 multifd_tls_handshake_thread, args, 728 QEMU_THREAD_JOINABLE); 729 return true; 730 } 731 732 void multifd_channel_connect(MultiFDSendParams *p, QIOChannel *ioc) 733 { 734 qio_channel_set_delay(ioc, false); 735 736 migration_ioc_register_yank(ioc); 737 /* Setup p->c only if the channel is completely setup */ 738 p->c = ioc; 739 740 p->thread_created = true; 741 qemu_thread_create(&p->thread, p->name, multifd_send_thread, p, 742 QEMU_THREAD_JOINABLE); 743 } 744 745 /* 746 * When TLS is enabled this function is called once to establish the 747 * TLS connection and a second time after the TLS handshake to create 748 * the multifd channel. Without TLS it goes straight into the channel 749 * creation. 750 */ 751 static void multifd_new_send_channel_async(QIOTask *task, gpointer opaque) 752 { 753 MultiFDSendParams *p = opaque; 754 QIOChannel *ioc = QIO_CHANNEL(qio_task_get_source(task)); 755 Error *local_err = NULL; 756 bool ret; 757 758 trace_multifd_new_send_channel_async(p->id); 759 760 if (qio_task_propagate_error(task, &local_err)) { 761 ret = false; 762 goto out; 763 } 764 765 trace_multifd_set_outgoing_channel(ioc, object_get_typename(OBJECT(ioc)), 766 migrate_get_current()->hostname); 767 768 if (migrate_channel_requires_tls_upgrade(ioc)) { 769 ret = multifd_tls_channel_connect(p, ioc, &local_err); 770 if (ret) { 771 return; 772 } 773 } else { 774 multifd_channel_connect(p, ioc); 775 ret = true; 776 } 777 778 out: 779 /* 780 * Here we're not interested whether creation succeeded, only that 781 * it happened at all. 782 */ 783 multifd_send_channel_created(); 784 785 if (ret) { 786 return; 787 } 788 789 trace_multifd_new_send_channel_async_error(p->id, local_err); 790 multifd_send_set_error(local_err); 791 /* 792 * For error cases (TLS or non-TLS), IO channel is always freed here 793 * rather than when cleanup multifd: since p->c is not set, multifd 794 * cleanup code doesn't even know its existence. 795 */ 796 object_unref(OBJECT(ioc)); 797 error_free(local_err); 798 } 799 800 static bool multifd_new_send_channel_create(gpointer opaque, Error **errp) 801 { 802 if (!multifd_use_packets()) { 803 return file_send_channel_create(opaque, errp); 804 } 805 806 socket_send_channel_create(multifd_new_send_channel_async, opaque); 807 return true; 808 } 809 810 bool multifd_send_setup(void) 811 { 812 MigrationState *s = migrate_get_current(); 813 int thread_count, ret = 0; 814 uint32_t page_count = multifd_ram_page_count(); 815 bool use_packets = multifd_use_packets(); 816 uint8_t i; 817 818 if (!migrate_multifd()) { 819 return true; 820 } 821 822 thread_count = migrate_multifd_channels(); 823 multifd_send_state = g_malloc0(sizeof(*multifd_send_state)); 824 multifd_send_state->params = g_new0(MultiFDSendParams, thread_count); 825 qemu_sem_init(&multifd_send_state->channels_created, 0); 826 qemu_sem_init(&multifd_send_state->channels_ready, 0); 827 qatomic_set(&multifd_send_state->exiting, 0); 828 multifd_send_state->ops = multifd_ops[migrate_multifd_compression()]; 829 830 for (i = 0; i < thread_count; i++) { 831 MultiFDSendParams *p = &multifd_send_state->params[i]; 832 Error *local_err = NULL; 833 834 qemu_sem_init(&p->sem, 0); 835 qemu_sem_init(&p->sem_sync, 0); 836 p->id = i; 837 p->data = multifd_send_data_alloc(); 838 839 if (use_packets) { 840 p->packet_len = sizeof(MultiFDPacket_t) 841 + sizeof(uint64_t) * page_count; 842 p->packet = g_malloc0(p->packet_len); 843 } 844 p->name = g_strdup_printf("mig/src/send_%d", i); 845 p->write_flags = 0; 846 847 if (!multifd_new_send_channel_create(p, &local_err)) { 848 migrate_set_error(s, local_err); 849 ret = -1; 850 } 851 } 852 853 /* 854 * Wait until channel creation has started for all channels. The 855 * creation can still fail, but no more channels will be created 856 * past this point. 857 */ 858 for (i = 0; i < thread_count; i++) { 859 qemu_sem_wait(&multifd_send_state->channels_created); 860 } 861 862 if (ret) { 863 goto err; 864 } 865 866 for (i = 0; i < thread_count; i++) { 867 MultiFDSendParams *p = &multifd_send_state->params[i]; 868 Error *local_err = NULL; 869 870 ret = multifd_send_state->ops->send_setup(p, &local_err); 871 if (ret) { 872 migrate_set_error(s, local_err); 873 goto err; 874 } 875 assert(p->iov); 876 } 877 878 return true; 879 880 err: 881 migrate_set_state(&s->state, MIGRATION_STATUS_SETUP, 882 MIGRATION_STATUS_FAILED); 883 return false; 884 } 885 886 bool multifd_recv(void) 887 { 888 int i; 889 static int next_recv_channel; 890 MultiFDRecvParams *p = NULL; 891 MultiFDRecvData *data = multifd_recv_state->data; 892 893 /* 894 * next_channel can remain from a previous migration that was 895 * using more channels, so ensure it doesn't overflow if the 896 * limit is lower now. 897 */ 898 next_recv_channel %= migrate_multifd_channels(); 899 for (i = next_recv_channel;; i = (i + 1) % migrate_multifd_channels()) { 900 if (multifd_recv_should_exit()) { 901 return false; 902 } 903 904 p = &multifd_recv_state->params[i]; 905 906 if (qatomic_read(&p->pending_job) == false) { 907 next_recv_channel = (i + 1) % migrate_multifd_channels(); 908 break; 909 } 910 } 911 912 /* 913 * Order pending_job read before manipulating p->data below. Pairs 914 * with qatomic_store_release() at multifd_recv_thread(). 915 */ 916 smp_mb_acquire(); 917 918 assert(!p->data->size); 919 multifd_recv_state->data = p->data; 920 p->data = data; 921 922 /* 923 * Order p->data update before setting pending_job. Pairs with 924 * qatomic_load_acquire() at multifd_recv_thread(). 925 */ 926 qatomic_store_release(&p->pending_job, true); 927 qemu_sem_post(&p->sem); 928 929 return true; 930 } 931 932 MultiFDRecvData *multifd_get_recv_data(void) 933 { 934 return multifd_recv_state->data; 935 } 936 937 static void multifd_recv_terminate_threads(Error *err) 938 { 939 int i; 940 941 trace_multifd_recv_terminate_threads(err != NULL); 942 943 if (qatomic_xchg(&multifd_recv_state->exiting, 1)) { 944 return; 945 } 946 947 if (err) { 948 MigrationState *s = migrate_get_current(); 949 migrate_set_error(s, err); 950 if (s->state == MIGRATION_STATUS_SETUP || 951 s->state == MIGRATION_STATUS_ACTIVE) { 952 migrate_set_state(&s->state, s->state, 953 MIGRATION_STATUS_FAILED); 954 } 955 } 956 957 for (i = 0; i < migrate_multifd_channels(); i++) { 958 MultiFDRecvParams *p = &multifd_recv_state->params[i]; 959 960 /* 961 * The migration thread and channels interact differently 962 * depending on the presence of packets. 963 */ 964 if (multifd_use_packets()) { 965 /* 966 * The channel receives as long as there are packets. When 967 * packets end (i.e. MULTIFD_FLAG_SYNC is reached), the 968 * channel waits for the migration thread to sync. If the 969 * sync never happens, do it here. 970 */ 971 qemu_sem_post(&p->sem_sync); 972 } else { 973 /* 974 * The channel waits for the migration thread to give it 975 * work. When the migration thread runs out of work, it 976 * releases the channel and waits for any pending work to 977 * finish. If we reach here (e.g. due to error) before the 978 * work runs out, release the channel. 979 */ 980 qemu_sem_post(&p->sem); 981 } 982 983 /* 984 * We could arrive here for two reasons: 985 * - normal quit, i.e. everything went fine, just finished 986 * - error quit: We close the channels so the channel threads 987 * finish the qio_channel_read_all_eof() 988 */ 989 if (p->c) { 990 qio_channel_shutdown(p->c, QIO_CHANNEL_SHUTDOWN_BOTH, NULL); 991 } 992 } 993 } 994 995 void multifd_recv_shutdown(void) 996 { 997 if (migrate_multifd()) { 998 multifd_recv_terminate_threads(NULL); 999 } 1000 } 1001 1002 static void multifd_recv_cleanup_channel(MultiFDRecvParams *p) 1003 { 1004 migration_ioc_unregister_yank(p->c); 1005 object_unref(OBJECT(p->c)); 1006 p->c = NULL; 1007 qemu_mutex_destroy(&p->mutex); 1008 qemu_sem_destroy(&p->sem_sync); 1009 qemu_sem_destroy(&p->sem); 1010 g_free(p->data); 1011 p->data = NULL; 1012 g_free(p->name); 1013 p->name = NULL; 1014 p->packet_len = 0; 1015 g_free(p->packet); 1016 p->packet = NULL; 1017 g_free(p->normal); 1018 p->normal = NULL; 1019 g_free(p->zero); 1020 p->zero = NULL; 1021 multifd_recv_state->ops->recv_cleanup(p); 1022 } 1023 1024 static void multifd_recv_cleanup_state(void) 1025 { 1026 qemu_sem_destroy(&multifd_recv_state->sem_sync); 1027 g_free(multifd_recv_state->params); 1028 multifd_recv_state->params = NULL; 1029 g_free(multifd_recv_state->data); 1030 multifd_recv_state->data = NULL; 1031 g_free(multifd_recv_state); 1032 multifd_recv_state = NULL; 1033 } 1034 1035 void multifd_recv_cleanup(void) 1036 { 1037 int i; 1038 1039 if (!migrate_multifd()) { 1040 return; 1041 } 1042 multifd_recv_terminate_threads(NULL); 1043 for (i = 0; i < migrate_multifd_channels(); i++) { 1044 MultiFDRecvParams *p = &multifd_recv_state->params[i]; 1045 1046 if (p->thread_created) { 1047 qemu_thread_join(&p->thread); 1048 } 1049 } 1050 for (i = 0; i < migrate_multifd_channels(); i++) { 1051 multifd_recv_cleanup_channel(&multifd_recv_state->params[i]); 1052 } 1053 multifd_recv_cleanup_state(); 1054 } 1055 1056 void multifd_recv_sync_main(void) 1057 { 1058 int thread_count = migrate_multifd_channels(); 1059 bool file_based = !multifd_use_packets(); 1060 int i; 1061 1062 if (!migrate_multifd()) { 1063 return; 1064 } 1065 1066 /* 1067 * File-based channels don't use packets and therefore need to 1068 * wait for more work. Release them to start the sync. 1069 */ 1070 if (file_based) { 1071 for (i = 0; i < thread_count; i++) { 1072 MultiFDRecvParams *p = &multifd_recv_state->params[i]; 1073 1074 trace_multifd_recv_sync_main_signal(p->id); 1075 qemu_sem_post(&p->sem); 1076 } 1077 } 1078 1079 /* 1080 * Initiate the synchronization by waiting for all channels. 1081 * 1082 * For socket-based migration this means each channel has received 1083 * the SYNC packet on the stream. 1084 * 1085 * For file-based migration this means each channel is done with 1086 * the work (pending_job=false). 1087 */ 1088 for (i = 0; i < thread_count; i++) { 1089 trace_multifd_recv_sync_main_wait(i); 1090 qemu_sem_wait(&multifd_recv_state->sem_sync); 1091 } 1092 1093 if (file_based) { 1094 /* 1095 * For file-based loading is done in one iteration. We're 1096 * done. 1097 */ 1098 return; 1099 } 1100 1101 /* 1102 * Sync done. Release the channels for the next iteration. 1103 */ 1104 for (i = 0; i < thread_count; i++) { 1105 MultiFDRecvParams *p = &multifd_recv_state->params[i]; 1106 1107 WITH_QEMU_LOCK_GUARD(&p->mutex) { 1108 if (multifd_recv_state->packet_num < p->packet_num) { 1109 multifd_recv_state->packet_num = p->packet_num; 1110 } 1111 } 1112 trace_multifd_recv_sync_main_signal(p->id); 1113 qemu_sem_post(&p->sem_sync); 1114 } 1115 trace_multifd_recv_sync_main(multifd_recv_state->packet_num); 1116 } 1117 1118 static void *multifd_recv_thread(void *opaque) 1119 { 1120 MultiFDRecvParams *p = opaque; 1121 Error *local_err = NULL; 1122 bool use_packets = multifd_use_packets(); 1123 int ret; 1124 1125 trace_multifd_recv_thread_start(p->id); 1126 rcu_register_thread(); 1127 1128 while (true) { 1129 uint32_t flags = 0; 1130 bool has_data = false; 1131 p->normal_num = 0; 1132 1133 if (use_packets) { 1134 if (multifd_recv_should_exit()) { 1135 break; 1136 } 1137 1138 ret = qio_channel_read_all_eof(p->c, (void *)p->packet, 1139 p->packet_len, &local_err); 1140 if (ret == 0 || ret == -1) { /* 0: EOF -1: Error */ 1141 break; 1142 } 1143 1144 qemu_mutex_lock(&p->mutex); 1145 ret = multifd_recv_unfill_packet(p, &local_err); 1146 if (ret) { 1147 qemu_mutex_unlock(&p->mutex); 1148 break; 1149 } 1150 1151 flags = p->flags; 1152 /* recv methods don't know how to handle the SYNC flag */ 1153 p->flags &= ~MULTIFD_FLAG_SYNC; 1154 if (!(flags & MULTIFD_FLAG_SYNC)) { 1155 has_data = p->normal_num || p->zero_num; 1156 } 1157 qemu_mutex_unlock(&p->mutex); 1158 } else { 1159 /* 1160 * No packets, so we need to wait for the vmstate code to 1161 * give us work. 1162 */ 1163 qemu_sem_wait(&p->sem); 1164 1165 if (multifd_recv_should_exit()) { 1166 break; 1167 } 1168 1169 /* pairs with qatomic_store_release() at multifd_recv() */ 1170 if (!qatomic_load_acquire(&p->pending_job)) { 1171 /* 1172 * Migration thread did not send work, this is 1173 * equivalent to pending_sync on the sending 1174 * side. Post sem_sync to notify we reached this 1175 * point. 1176 */ 1177 qemu_sem_post(&multifd_recv_state->sem_sync); 1178 continue; 1179 } 1180 1181 has_data = !!p->data->size; 1182 } 1183 1184 if (has_data) { 1185 ret = multifd_recv_state->ops->recv(p, &local_err); 1186 if (ret != 0) { 1187 break; 1188 } 1189 } 1190 1191 if (use_packets) { 1192 if (flags & MULTIFD_FLAG_SYNC) { 1193 qemu_sem_post(&multifd_recv_state->sem_sync); 1194 qemu_sem_wait(&p->sem_sync); 1195 } 1196 } else { 1197 p->data->size = 0; 1198 /* 1199 * Order data->size update before clearing 1200 * pending_job. Pairs with smp_mb_acquire() at 1201 * multifd_recv(). 1202 */ 1203 qatomic_store_release(&p->pending_job, false); 1204 } 1205 } 1206 1207 if (local_err) { 1208 multifd_recv_terminate_threads(local_err); 1209 error_free(local_err); 1210 } 1211 1212 rcu_unregister_thread(); 1213 trace_multifd_recv_thread_end(p->id, p->packets_recved); 1214 1215 return NULL; 1216 } 1217 1218 int multifd_recv_setup(Error **errp) 1219 { 1220 int thread_count; 1221 uint32_t page_count = multifd_ram_page_count(); 1222 bool use_packets = multifd_use_packets(); 1223 uint8_t i; 1224 1225 /* 1226 * Return successfully if multiFD recv state is already initialised 1227 * or multiFD is not enabled. 1228 */ 1229 if (multifd_recv_state || !migrate_multifd()) { 1230 return 0; 1231 } 1232 1233 thread_count = migrate_multifd_channels(); 1234 multifd_recv_state = g_malloc0(sizeof(*multifd_recv_state)); 1235 multifd_recv_state->params = g_new0(MultiFDRecvParams, thread_count); 1236 1237 multifd_recv_state->data = g_new0(MultiFDRecvData, 1); 1238 multifd_recv_state->data->size = 0; 1239 1240 qatomic_set(&multifd_recv_state->count, 0); 1241 qatomic_set(&multifd_recv_state->exiting, 0); 1242 qemu_sem_init(&multifd_recv_state->sem_sync, 0); 1243 multifd_recv_state->ops = multifd_ops[migrate_multifd_compression()]; 1244 1245 for (i = 0; i < thread_count; i++) { 1246 MultiFDRecvParams *p = &multifd_recv_state->params[i]; 1247 1248 qemu_mutex_init(&p->mutex); 1249 qemu_sem_init(&p->sem_sync, 0); 1250 qemu_sem_init(&p->sem, 0); 1251 p->pending_job = false; 1252 p->id = i; 1253 1254 p->data = g_new0(MultiFDRecvData, 1); 1255 p->data->size = 0; 1256 1257 if (use_packets) { 1258 p->packet_len = sizeof(MultiFDPacket_t) 1259 + sizeof(uint64_t) * page_count; 1260 p->packet = g_malloc0(p->packet_len); 1261 } 1262 p->name = g_strdup_printf("mig/dst/recv_%d", i); 1263 p->normal = g_new0(ram_addr_t, page_count); 1264 p->zero = g_new0(ram_addr_t, page_count); 1265 } 1266 1267 for (i = 0; i < thread_count; i++) { 1268 MultiFDRecvParams *p = &multifd_recv_state->params[i]; 1269 int ret; 1270 1271 ret = multifd_recv_state->ops->recv_setup(p, errp); 1272 if (ret) { 1273 return ret; 1274 } 1275 } 1276 return 0; 1277 } 1278 1279 bool multifd_recv_all_channels_created(void) 1280 { 1281 int thread_count = migrate_multifd_channels(); 1282 1283 if (!migrate_multifd()) { 1284 return true; 1285 } 1286 1287 if (!multifd_recv_state) { 1288 /* Called before any connections created */ 1289 return false; 1290 } 1291 1292 return thread_count == qatomic_read(&multifd_recv_state->count); 1293 } 1294 1295 /* 1296 * Try to receive all multifd channels to get ready for the migration. 1297 * Sets @errp when failing to receive the current channel. 1298 */ 1299 void multifd_recv_new_channel(QIOChannel *ioc, Error **errp) 1300 { 1301 MultiFDRecvParams *p; 1302 Error *local_err = NULL; 1303 bool use_packets = multifd_use_packets(); 1304 int id; 1305 1306 if (use_packets) { 1307 id = multifd_recv_initial_packet(ioc, &local_err); 1308 if (id < 0) { 1309 multifd_recv_terminate_threads(local_err); 1310 error_propagate_prepend(errp, local_err, 1311 "failed to receive packet" 1312 " via multifd channel %d: ", 1313 qatomic_read(&multifd_recv_state->count)); 1314 return; 1315 } 1316 trace_multifd_recv_new_channel(id); 1317 } else { 1318 id = qatomic_read(&multifd_recv_state->count); 1319 } 1320 1321 p = &multifd_recv_state->params[id]; 1322 if (p->c != NULL) { 1323 error_setg(&local_err, "multifd: received id '%d' already setup'", 1324 id); 1325 multifd_recv_terminate_threads(local_err); 1326 error_propagate(errp, local_err); 1327 return; 1328 } 1329 p->c = ioc; 1330 object_ref(OBJECT(ioc)); 1331 1332 p->thread_created = true; 1333 qemu_thread_create(&p->thread, p->name, multifd_recv_thread, p, 1334 QEMU_THREAD_JOINABLE); 1335 qatomic_inc(&multifd_recv_state->count); 1336 } 1337