1 /* 2 * QEMU System Emulator 3 * 4 * Copyright (c) 2003-2008 Fabrice Bellard 5 * Copyright (c) 2009-2015 Red Hat Inc 6 * 7 * Authors: 8 * Juan Quintela <quintela@redhat.com> 9 * 10 * Permission is hereby granted, free of charge, to any person obtaining a copy 11 * of this software and associated documentation files (the "Software"), to deal 12 * in the Software without restriction, including without limitation the rights 13 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 14 * copies of the Software, and to permit persons to whom the Software is 15 * furnished to do so, subject to the following conditions: 16 * 17 * The above copyright notice and this permission notice shall be included in 18 * all copies or substantial portions of the Software. 19 * 20 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 21 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 22 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 23 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 24 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 25 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 26 * THE SOFTWARE. 27 */ 28 29 #include "qemu/osdep.h" 30 #include "hw/boards.h" 31 #include "net/net.h" 32 #include "migration.h" 33 #include "migration/snapshot.h" 34 #include "migration/vmstate.h" 35 #include "migration/misc.h" 36 #include "migration/register.h" 37 #include "migration/global_state.h" 38 #include "migration/channel-block.h" 39 #include "ram.h" 40 #include "qemu-file.h" 41 #include "savevm.h" 42 #include "postcopy-ram.h" 43 #include "qapi/error.h" 44 #include "qapi/qapi-commands-migration.h" 45 #include "qapi/qmp/json-writer.h" 46 #include "qapi/clone-visitor.h" 47 #include "qapi/qapi-builtin-visit.h" 48 #include "qapi/qmp/qerror.h" 49 #include "qemu/error-report.h" 50 #include "sysemu/cpus.h" 51 #include "exec/memory.h" 52 #include "exec/target_page.h" 53 #include "trace.h" 54 #include "qemu/iov.h" 55 #include "qemu/job.h" 56 #include "qemu/main-loop.h" 57 #include "block/snapshot.h" 58 #include "qemu/cutils.h" 59 #include "io/channel-buffer.h" 60 #include "io/channel-file.h" 61 #include "sysemu/replay.h" 62 #include "sysemu/runstate.h" 63 #include "sysemu/sysemu.h" 64 #include "sysemu/xen.h" 65 #include "migration/colo.h" 66 #include "qemu/bitmap.h" 67 #include "net/announce.h" 68 #include "qemu/yank.h" 69 #include "yank_functions.h" 70 71 const unsigned int postcopy_ram_discard_version; 72 73 /* Subcommands for QEMU_VM_COMMAND */ 74 enum qemu_vm_cmd { 75 MIG_CMD_INVALID = 0, /* Must be 0 */ 76 MIG_CMD_OPEN_RETURN_PATH, /* Tell the dest to open the Return path */ 77 MIG_CMD_PING, /* Request a PONG on the RP */ 78 79 MIG_CMD_POSTCOPY_ADVISE, /* Prior to any page transfers, just 80 warn we might want to do PC */ 81 MIG_CMD_POSTCOPY_LISTEN, /* Start listening for incoming 82 pages as it's running. */ 83 MIG_CMD_POSTCOPY_RUN, /* Start execution */ 84 85 MIG_CMD_POSTCOPY_RAM_DISCARD, /* A list of pages to discard that 86 were previously sent during 87 precopy but are dirty. */ 88 MIG_CMD_PACKAGED, /* Send a wrapped stream within this stream */ 89 MIG_CMD_ENABLE_COLO, /* Enable COLO */ 90 MIG_CMD_POSTCOPY_RESUME, /* resume postcopy on dest */ 91 MIG_CMD_RECV_BITMAP, /* Request for recved bitmap on dst */ 92 MIG_CMD_MAX 93 }; 94 95 #define MAX_VM_CMD_PACKAGED_SIZE UINT32_MAX 96 static struct mig_cmd_args { 97 ssize_t len; /* -1 = variable */ 98 const char *name; 99 } mig_cmd_args[] = { 100 [MIG_CMD_INVALID] = { .len = -1, .name = "INVALID" }, 101 [MIG_CMD_OPEN_RETURN_PATH] = { .len = 0, .name = "OPEN_RETURN_PATH" }, 102 [MIG_CMD_PING] = { .len = sizeof(uint32_t), .name = "PING" }, 103 [MIG_CMD_POSTCOPY_ADVISE] = { .len = -1, .name = "POSTCOPY_ADVISE" }, 104 [MIG_CMD_POSTCOPY_LISTEN] = { .len = 0, .name = "POSTCOPY_LISTEN" }, 105 [MIG_CMD_POSTCOPY_RUN] = { .len = 0, .name = "POSTCOPY_RUN" }, 106 [MIG_CMD_POSTCOPY_RAM_DISCARD] = { 107 .len = -1, .name = "POSTCOPY_RAM_DISCARD" }, 108 [MIG_CMD_POSTCOPY_RESUME] = { .len = 0, .name = "POSTCOPY_RESUME" }, 109 [MIG_CMD_PACKAGED] = { .len = 4, .name = "PACKAGED" }, 110 [MIG_CMD_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" }, 111 [MIG_CMD_MAX] = { .len = -1, .name = "MAX" }, 112 }; 113 114 /* Note for MIG_CMD_POSTCOPY_ADVISE: 115 * The format of arguments is depending on postcopy mode: 116 * - postcopy RAM only 117 * uint64_t host page size 118 * uint64_t taget page size 119 * 120 * - postcopy RAM and postcopy dirty bitmaps 121 * format is the same as for postcopy RAM only 122 * 123 * - postcopy dirty bitmaps only 124 * Nothing. Command length field is 0. 125 * 126 * Be careful: adding a new postcopy entity with some other parameters should 127 * not break format self-description ability. Good way is to introduce some 128 * generic extendable format with an exception for two old entities. 129 */ 130 131 /***********************************************************/ 132 /* savevm/loadvm support */ 133 134 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable) 135 { 136 if (is_writable) { 137 return qemu_file_new_output(QIO_CHANNEL(qio_channel_block_new(bs))); 138 } else { 139 return qemu_file_new_input(QIO_CHANNEL(qio_channel_block_new(bs))); 140 } 141 } 142 143 144 /* QEMUFile timer support. 145 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c 146 */ 147 148 void timer_put(QEMUFile *f, QEMUTimer *ts) 149 { 150 uint64_t expire_time; 151 152 expire_time = timer_expire_time_ns(ts); 153 qemu_put_be64(f, expire_time); 154 } 155 156 void timer_get(QEMUFile *f, QEMUTimer *ts) 157 { 158 uint64_t expire_time; 159 160 expire_time = qemu_get_be64(f); 161 if (expire_time != -1) { 162 timer_mod_ns(ts, expire_time); 163 } else { 164 timer_del(ts); 165 } 166 } 167 168 169 /* VMState timer support. 170 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c 171 */ 172 173 static int get_timer(QEMUFile *f, void *pv, size_t size, 174 const VMStateField *field) 175 { 176 QEMUTimer *v = pv; 177 timer_get(f, v); 178 return 0; 179 } 180 181 static int put_timer(QEMUFile *f, void *pv, size_t size, 182 const VMStateField *field, JSONWriter *vmdesc) 183 { 184 QEMUTimer *v = pv; 185 timer_put(f, v); 186 187 return 0; 188 } 189 190 const VMStateInfo vmstate_info_timer = { 191 .name = "timer", 192 .get = get_timer, 193 .put = put_timer, 194 }; 195 196 197 typedef struct CompatEntry { 198 char idstr[256]; 199 int instance_id; 200 } CompatEntry; 201 202 typedef struct SaveStateEntry { 203 QTAILQ_ENTRY(SaveStateEntry) entry; 204 char idstr[256]; 205 uint32_t instance_id; 206 int alias_id; 207 int version_id; 208 /* version id read from the stream */ 209 int load_version_id; 210 int section_id; 211 /* section id read from the stream */ 212 int load_section_id; 213 const SaveVMHandlers *ops; 214 const VMStateDescription *vmsd; 215 void *opaque; 216 CompatEntry *compat; 217 int is_ram; 218 } SaveStateEntry; 219 220 typedef struct SaveState { 221 QTAILQ_HEAD(, SaveStateEntry) handlers; 222 SaveStateEntry *handler_pri_head[MIG_PRI_MAX + 1]; 223 int global_section_id; 224 uint32_t len; 225 const char *name; 226 uint32_t target_page_bits; 227 uint32_t caps_count; 228 MigrationCapability *capabilities; 229 QemuUUID uuid; 230 } SaveState; 231 232 static SaveState savevm_state = { 233 .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers), 234 .handler_pri_head = { [MIG_PRI_DEFAULT ... MIG_PRI_MAX] = NULL }, 235 .global_section_id = 0, 236 }; 237 238 static bool should_validate_capability(int capability) 239 { 240 assert(capability >= 0 && capability < MIGRATION_CAPABILITY__MAX); 241 /* Validate only new capabilities to keep compatibility. */ 242 switch (capability) { 243 case MIGRATION_CAPABILITY_X_IGNORE_SHARED: 244 return true; 245 default: 246 return false; 247 } 248 } 249 250 static uint32_t get_validatable_capabilities_count(void) 251 { 252 MigrationState *s = migrate_get_current(); 253 uint32_t result = 0; 254 int i; 255 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { 256 if (should_validate_capability(i) && s->enabled_capabilities[i]) { 257 result++; 258 } 259 } 260 return result; 261 } 262 263 static int configuration_pre_save(void *opaque) 264 { 265 SaveState *state = opaque; 266 const char *current_name = MACHINE_GET_CLASS(current_machine)->name; 267 MigrationState *s = migrate_get_current(); 268 int i, j; 269 270 state->len = strlen(current_name); 271 state->name = current_name; 272 state->target_page_bits = qemu_target_page_bits(); 273 274 state->caps_count = get_validatable_capabilities_count(); 275 state->capabilities = g_renew(MigrationCapability, state->capabilities, 276 state->caps_count); 277 for (i = j = 0; i < MIGRATION_CAPABILITY__MAX; i++) { 278 if (should_validate_capability(i) && s->enabled_capabilities[i]) { 279 state->capabilities[j++] = i; 280 } 281 } 282 state->uuid = qemu_uuid; 283 284 return 0; 285 } 286 287 static int configuration_post_save(void *opaque) 288 { 289 SaveState *state = opaque; 290 291 g_free(state->capabilities); 292 state->capabilities = NULL; 293 state->caps_count = 0; 294 return 0; 295 } 296 297 static int configuration_pre_load(void *opaque) 298 { 299 SaveState *state = opaque; 300 301 /* If there is no target-page-bits subsection it means the source 302 * predates the variable-target-page-bits support and is using the 303 * minimum possible value for this CPU. 304 */ 305 state->target_page_bits = qemu_target_page_bits_min(); 306 return 0; 307 } 308 309 static bool configuration_validate_capabilities(SaveState *state) 310 { 311 bool ret = true; 312 MigrationState *s = migrate_get_current(); 313 unsigned long *source_caps_bm; 314 int i; 315 316 source_caps_bm = bitmap_new(MIGRATION_CAPABILITY__MAX); 317 for (i = 0; i < state->caps_count; i++) { 318 MigrationCapability capability = state->capabilities[i]; 319 set_bit(capability, source_caps_bm); 320 } 321 322 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { 323 bool source_state, target_state; 324 if (!should_validate_capability(i)) { 325 continue; 326 } 327 source_state = test_bit(i, source_caps_bm); 328 target_state = s->enabled_capabilities[i]; 329 if (source_state != target_state) { 330 error_report("Capability %s is %s, but received capability is %s", 331 MigrationCapability_str(i), 332 target_state ? "on" : "off", 333 source_state ? "on" : "off"); 334 ret = false; 335 /* Don't break here to report all failed capabilities */ 336 } 337 } 338 339 g_free(source_caps_bm); 340 return ret; 341 } 342 343 static int configuration_post_load(void *opaque, int version_id) 344 { 345 SaveState *state = opaque; 346 const char *current_name = MACHINE_GET_CLASS(current_machine)->name; 347 int ret = 0; 348 349 if (strncmp(state->name, current_name, state->len) != 0) { 350 error_report("Machine type received is '%.*s' and local is '%s'", 351 (int) state->len, state->name, current_name); 352 ret = -EINVAL; 353 goto out; 354 } 355 356 if (state->target_page_bits != qemu_target_page_bits()) { 357 error_report("Received TARGET_PAGE_BITS is %d but local is %d", 358 state->target_page_bits, qemu_target_page_bits()); 359 ret = -EINVAL; 360 goto out; 361 } 362 363 if (!configuration_validate_capabilities(state)) { 364 ret = -EINVAL; 365 goto out; 366 } 367 368 out: 369 g_free((void *)state->name); 370 state->name = NULL; 371 state->len = 0; 372 g_free(state->capabilities); 373 state->capabilities = NULL; 374 state->caps_count = 0; 375 376 return ret; 377 } 378 379 static int get_capability(QEMUFile *f, void *pv, size_t size, 380 const VMStateField *field) 381 { 382 MigrationCapability *capability = pv; 383 char capability_str[UINT8_MAX + 1]; 384 uint8_t len; 385 int i; 386 387 len = qemu_get_byte(f); 388 qemu_get_buffer(f, (uint8_t *)capability_str, len); 389 capability_str[len] = '\0'; 390 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { 391 if (!strcmp(MigrationCapability_str(i), capability_str)) { 392 *capability = i; 393 return 0; 394 } 395 } 396 error_report("Received unknown capability %s", capability_str); 397 return -EINVAL; 398 } 399 400 static int put_capability(QEMUFile *f, void *pv, size_t size, 401 const VMStateField *field, JSONWriter *vmdesc) 402 { 403 MigrationCapability *capability = pv; 404 const char *capability_str = MigrationCapability_str(*capability); 405 size_t len = strlen(capability_str); 406 assert(len <= UINT8_MAX); 407 408 qemu_put_byte(f, len); 409 qemu_put_buffer(f, (uint8_t *)capability_str, len); 410 return 0; 411 } 412 413 static const VMStateInfo vmstate_info_capability = { 414 .name = "capability", 415 .get = get_capability, 416 .put = put_capability, 417 }; 418 419 /* The target-page-bits subsection is present only if the 420 * target page size is not the same as the default (ie the 421 * minimum page size for a variable-page-size guest CPU). 422 * If it is present then it contains the actual target page 423 * bits for the machine, and migration will fail if the 424 * two ends don't agree about it. 425 */ 426 static bool vmstate_target_page_bits_needed(void *opaque) 427 { 428 return qemu_target_page_bits() 429 > qemu_target_page_bits_min(); 430 } 431 432 static const VMStateDescription vmstate_target_page_bits = { 433 .name = "configuration/target-page-bits", 434 .version_id = 1, 435 .minimum_version_id = 1, 436 .needed = vmstate_target_page_bits_needed, 437 .fields = (VMStateField[]) { 438 VMSTATE_UINT32(target_page_bits, SaveState), 439 VMSTATE_END_OF_LIST() 440 } 441 }; 442 443 static bool vmstate_capabilites_needed(void *opaque) 444 { 445 return get_validatable_capabilities_count() > 0; 446 } 447 448 static const VMStateDescription vmstate_capabilites = { 449 .name = "configuration/capabilities", 450 .version_id = 1, 451 .minimum_version_id = 1, 452 .needed = vmstate_capabilites_needed, 453 .fields = (VMStateField[]) { 454 VMSTATE_UINT32_V(caps_count, SaveState, 1), 455 VMSTATE_VARRAY_UINT32_ALLOC(capabilities, SaveState, caps_count, 1, 456 vmstate_info_capability, 457 MigrationCapability), 458 VMSTATE_END_OF_LIST() 459 } 460 }; 461 462 static bool vmstate_uuid_needed(void *opaque) 463 { 464 return qemu_uuid_set && migrate_validate_uuid(); 465 } 466 467 static int vmstate_uuid_post_load(void *opaque, int version_id) 468 { 469 SaveState *state = opaque; 470 char uuid_src[UUID_FMT_LEN + 1]; 471 char uuid_dst[UUID_FMT_LEN + 1]; 472 473 if (!qemu_uuid_set) { 474 /* 475 * It's warning because user might not know UUID in some cases, 476 * e.g. load an old snapshot 477 */ 478 qemu_uuid_unparse(&state->uuid, uuid_src); 479 warn_report("UUID is received %s, but local uuid isn't set", 480 uuid_src); 481 return 0; 482 } 483 if (!qemu_uuid_is_equal(&state->uuid, &qemu_uuid)) { 484 qemu_uuid_unparse(&state->uuid, uuid_src); 485 qemu_uuid_unparse(&qemu_uuid, uuid_dst); 486 error_report("UUID received is %s and local is %s", uuid_src, uuid_dst); 487 return -EINVAL; 488 } 489 return 0; 490 } 491 492 static const VMStateDescription vmstate_uuid = { 493 .name = "configuration/uuid", 494 .version_id = 1, 495 .minimum_version_id = 1, 496 .needed = vmstate_uuid_needed, 497 .post_load = vmstate_uuid_post_load, 498 .fields = (VMStateField[]) { 499 VMSTATE_UINT8_ARRAY_V(uuid.data, SaveState, sizeof(QemuUUID), 1), 500 VMSTATE_END_OF_LIST() 501 } 502 }; 503 504 static const VMStateDescription vmstate_configuration = { 505 .name = "configuration", 506 .version_id = 1, 507 .pre_load = configuration_pre_load, 508 .post_load = configuration_post_load, 509 .pre_save = configuration_pre_save, 510 .post_save = configuration_post_save, 511 .fields = (VMStateField[]) { 512 VMSTATE_UINT32(len, SaveState), 513 VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len), 514 VMSTATE_END_OF_LIST() 515 }, 516 .subsections = (const VMStateDescription *[]) { 517 &vmstate_target_page_bits, 518 &vmstate_capabilites, 519 &vmstate_uuid, 520 NULL 521 } 522 }; 523 524 static void dump_vmstate_vmsd(FILE *out_file, 525 const VMStateDescription *vmsd, int indent, 526 bool is_subsection); 527 528 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field, 529 int indent) 530 { 531 fprintf(out_file, "%*s{\n", indent, ""); 532 indent += 2; 533 fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name); 534 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "", 535 field->version_id); 536 fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "", 537 field->field_exists ? "true" : "false"); 538 fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size); 539 if (field->vmsd != NULL) { 540 fprintf(out_file, ",\n"); 541 dump_vmstate_vmsd(out_file, field->vmsd, indent, false); 542 } 543 fprintf(out_file, "\n%*s}", indent - 2, ""); 544 } 545 546 static void dump_vmstate_vmss(FILE *out_file, 547 const VMStateDescription **subsection, 548 int indent) 549 { 550 if (*subsection != NULL) { 551 dump_vmstate_vmsd(out_file, *subsection, indent, true); 552 } 553 } 554 555 static void dump_vmstate_vmsd(FILE *out_file, 556 const VMStateDescription *vmsd, int indent, 557 bool is_subsection) 558 { 559 if (is_subsection) { 560 fprintf(out_file, "%*s{\n", indent, ""); 561 } else { 562 fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description"); 563 } 564 indent += 2; 565 fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name); 566 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "", 567 vmsd->version_id); 568 fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "", 569 vmsd->minimum_version_id); 570 if (vmsd->fields != NULL) { 571 const VMStateField *field = vmsd->fields; 572 bool first; 573 574 fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, ""); 575 first = true; 576 while (field->name != NULL) { 577 if (field->flags & VMS_MUST_EXIST) { 578 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */ 579 field++; 580 continue; 581 } 582 if (!first) { 583 fprintf(out_file, ",\n"); 584 } 585 dump_vmstate_vmsf(out_file, field, indent + 2); 586 field++; 587 first = false; 588 } 589 fprintf(out_file, "\n%*s]", indent, ""); 590 } 591 if (vmsd->subsections != NULL) { 592 const VMStateDescription **subsection = vmsd->subsections; 593 bool first; 594 595 fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, ""); 596 first = true; 597 while (*subsection != NULL) { 598 if (!first) { 599 fprintf(out_file, ",\n"); 600 } 601 dump_vmstate_vmss(out_file, subsection, indent + 2); 602 subsection++; 603 first = false; 604 } 605 fprintf(out_file, "\n%*s]", indent, ""); 606 } 607 fprintf(out_file, "\n%*s}", indent - 2, ""); 608 } 609 610 static void dump_machine_type(FILE *out_file) 611 { 612 MachineClass *mc; 613 614 mc = MACHINE_GET_CLASS(current_machine); 615 616 fprintf(out_file, " \"vmschkmachine\": {\n"); 617 fprintf(out_file, " \"Name\": \"%s\"\n", mc->name); 618 fprintf(out_file, " },\n"); 619 } 620 621 void dump_vmstate_json_to_file(FILE *out_file) 622 { 623 GSList *list, *elt; 624 bool first; 625 626 fprintf(out_file, "{\n"); 627 dump_machine_type(out_file); 628 629 first = true; 630 list = object_class_get_list(TYPE_DEVICE, true); 631 for (elt = list; elt; elt = elt->next) { 632 DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data, 633 TYPE_DEVICE); 634 const char *name; 635 int indent = 2; 636 637 if (!dc->vmsd) { 638 continue; 639 } 640 641 if (!first) { 642 fprintf(out_file, ",\n"); 643 } 644 name = object_class_get_name(OBJECT_CLASS(dc)); 645 fprintf(out_file, "%*s\"%s\": {\n", indent, "", name); 646 indent += 2; 647 fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name); 648 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "", 649 dc->vmsd->version_id); 650 fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "", 651 dc->vmsd->minimum_version_id); 652 653 dump_vmstate_vmsd(out_file, dc->vmsd, indent, false); 654 655 fprintf(out_file, "\n%*s}", indent - 2, ""); 656 first = false; 657 } 658 fprintf(out_file, "\n}\n"); 659 fclose(out_file); 660 g_slist_free(list); 661 } 662 663 static uint32_t calculate_new_instance_id(const char *idstr) 664 { 665 SaveStateEntry *se; 666 uint32_t instance_id = 0; 667 668 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 669 if (strcmp(idstr, se->idstr) == 0 670 && instance_id <= se->instance_id) { 671 instance_id = se->instance_id + 1; 672 } 673 } 674 /* Make sure we never loop over without being noticed */ 675 assert(instance_id != VMSTATE_INSTANCE_ID_ANY); 676 return instance_id; 677 } 678 679 static int calculate_compat_instance_id(const char *idstr) 680 { 681 SaveStateEntry *se; 682 int instance_id = 0; 683 684 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 685 if (!se->compat) { 686 continue; 687 } 688 689 if (strcmp(idstr, se->compat->idstr) == 0 690 && instance_id <= se->compat->instance_id) { 691 instance_id = se->compat->instance_id + 1; 692 } 693 } 694 return instance_id; 695 } 696 697 static inline MigrationPriority save_state_priority(SaveStateEntry *se) 698 { 699 if (se->vmsd) { 700 return se->vmsd->priority; 701 } 702 return MIG_PRI_DEFAULT; 703 } 704 705 static void savevm_state_handler_insert(SaveStateEntry *nse) 706 { 707 MigrationPriority priority = save_state_priority(nse); 708 SaveStateEntry *se; 709 int i; 710 711 assert(priority <= MIG_PRI_MAX); 712 713 for (i = priority - 1; i >= 0; i--) { 714 se = savevm_state.handler_pri_head[i]; 715 if (se != NULL) { 716 assert(save_state_priority(se) < priority); 717 break; 718 } 719 } 720 721 if (i >= 0) { 722 QTAILQ_INSERT_BEFORE(se, nse, entry); 723 } else { 724 QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry); 725 } 726 727 if (savevm_state.handler_pri_head[priority] == NULL) { 728 savevm_state.handler_pri_head[priority] = nse; 729 } 730 } 731 732 static void savevm_state_handler_remove(SaveStateEntry *se) 733 { 734 SaveStateEntry *next; 735 MigrationPriority priority = save_state_priority(se); 736 737 if (se == savevm_state.handler_pri_head[priority]) { 738 next = QTAILQ_NEXT(se, entry); 739 if (next != NULL && save_state_priority(next) == priority) { 740 savevm_state.handler_pri_head[priority] = next; 741 } else { 742 savevm_state.handler_pri_head[priority] = NULL; 743 } 744 } 745 QTAILQ_REMOVE(&savevm_state.handlers, se, entry); 746 } 747 748 /* TODO: Individual devices generally have very little idea about the rest 749 of the system, so instance_id should be removed/replaced. 750 Meanwhile pass -1 as instance_id if you do not already have a clearly 751 distinguishing id for all instances of your device class. */ 752 int register_savevm_live(const char *idstr, 753 uint32_t instance_id, 754 int version_id, 755 const SaveVMHandlers *ops, 756 void *opaque) 757 { 758 SaveStateEntry *se; 759 760 se = g_new0(SaveStateEntry, 1); 761 se->version_id = version_id; 762 se->section_id = savevm_state.global_section_id++; 763 se->ops = ops; 764 se->opaque = opaque; 765 se->vmsd = NULL; 766 /* if this is a live_savem then set is_ram */ 767 if (ops->save_setup != NULL) { 768 se->is_ram = 1; 769 } 770 771 pstrcat(se->idstr, sizeof(se->idstr), idstr); 772 773 if (instance_id == VMSTATE_INSTANCE_ID_ANY) { 774 se->instance_id = calculate_new_instance_id(se->idstr); 775 } else { 776 se->instance_id = instance_id; 777 } 778 assert(!se->compat || se->instance_id == 0); 779 savevm_state_handler_insert(se); 780 return 0; 781 } 782 783 void unregister_savevm(VMStateIf *obj, const char *idstr, void *opaque) 784 { 785 SaveStateEntry *se, *new_se; 786 char id[256] = ""; 787 788 if (obj) { 789 char *oid = vmstate_if_get_id(obj); 790 if (oid) { 791 pstrcpy(id, sizeof(id), oid); 792 pstrcat(id, sizeof(id), "/"); 793 g_free(oid); 794 } 795 } 796 pstrcat(id, sizeof(id), idstr); 797 798 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) { 799 if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) { 800 savevm_state_handler_remove(se); 801 g_free(se->compat); 802 g_free(se); 803 } 804 } 805 } 806 807 int vmstate_register_with_alias_id(VMStateIf *obj, uint32_t instance_id, 808 const VMStateDescription *vmsd, 809 void *opaque, int alias_id, 810 int required_for_version, 811 Error **errp) 812 { 813 SaveStateEntry *se; 814 815 /* If this triggers, alias support can be dropped for the vmsd. */ 816 assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id); 817 818 se = g_new0(SaveStateEntry, 1); 819 se->version_id = vmsd->version_id; 820 se->section_id = savevm_state.global_section_id++; 821 se->opaque = opaque; 822 se->vmsd = vmsd; 823 se->alias_id = alias_id; 824 825 if (obj) { 826 char *id = vmstate_if_get_id(obj); 827 if (id) { 828 if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >= 829 sizeof(se->idstr)) { 830 error_setg(errp, "Path too long for VMState (%s)", id); 831 g_free(id); 832 g_free(se); 833 834 return -1; 835 } 836 g_free(id); 837 838 se->compat = g_new0(CompatEntry, 1); 839 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name); 840 se->compat->instance_id = instance_id == VMSTATE_INSTANCE_ID_ANY ? 841 calculate_compat_instance_id(vmsd->name) : instance_id; 842 instance_id = VMSTATE_INSTANCE_ID_ANY; 843 } 844 } 845 pstrcat(se->idstr, sizeof(se->idstr), vmsd->name); 846 847 if (instance_id == VMSTATE_INSTANCE_ID_ANY) { 848 se->instance_id = calculate_new_instance_id(se->idstr); 849 } else { 850 se->instance_id = instance_id; 851 } 852 assert(!se->compat || se->instance_id == 0); 853 savevm_state_handler_insert(se); 854 return 0; 855 } 856 857 void vmstate_unregister(VMStateIf *obj, const VMStateDescription *vmsd, 858 void *opaque) 859 { 860 SaveStateEntry *se, *new_se; 861 862 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) { 863 if (se->vmsd == vmsd && se->opaque == opaque) { 864 savevm_state_handler_remove(se); 865 g_free(se->compat); 866 g_free(se); 867 } 868 } 869 } 870 871 static int vmstate_load(QEMUFile *f, SaveStateEntry *se) 872 { 873 trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)"); 874 if (!se->vmsd) { /* Old style */ 875 return se->ops->load_state(f, se->opaque, se->load_version_id); 876 } 877 return vmstate_load_state(f, se->vmsd, se->opaque, se->load_version_id); 878 } 879 880 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se, 881 JSONWriter *vmdesc) 882 { 883 int64_t old_offset, size; 884 885 old_offset = qemu_file_total_transferred_fast(f); 886 se->ops->save_state(f, se->opaque); 887 size = qemu_file_total_transferred_fast(f) - old_offset; 888 889 if (vmdesc) { 890 json_writer_int64(vmdesc, "size", size); 891 json_writer_start_array(vmdesc, "fields"); 892 json_writer_start_object(vmdesc, NULL); 893 json_writer_str(vmdesc, "name", "data"); 894 json_writer_int64(vmdesc, "size", size); 895 json_writer_str(vmdesc, "type", "buffer"); 896 json_writer_end_object(vmdesc); 897 json_writer_end_array(vmdesc); 898 } 899 } 900 901 static int vmstate_save(QEMUFile *f, SaveStateEntry *se, 902 JSONWriter *vmdesc) 903 { 904 trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)"); 905 if (!se->vmsd) { 906 vmstate_save_old_style(f, se, vmdesc); 907 return 0; 908 } 909 return vmstate_save_state(f, se->vmsd, se->opaque, vmdesc); 910 } 911 912 /* 913 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL) 914 */ 915 static void save_section_header(QEMUFile *f, SaveStateEntry *se, 916 uint8_t section_type) 917 { 918 qemu_put_byte(f, section_type); 919 qemu_put_be32(f, se->section_id); 920 921 if (section_type == QEMU_VM_SECTION_FULL || 922 section_type == QEMU_VM_SECTION_START) { 923 /* ID string */ 924 size_t len = strlen(se->idstr); 925 qemu_put_byte(f, len); 926 qemu_put_buffer(f, (uint8_t *)se->idstr, len); 927 928 qemu_put_be32(f, se->instance_id); 929 qemu_put_be32(f, se->version_id); 930 } 931 } 932 933 /* 934 * Write a footer onto device sections that catches cases misformatted device 935 * sections. 936 */ 937 static void save_section_footer(QEMUFile *f, SaveStateEntry *se) 938 { 939 if (migrate_get_current()->send_section_footer) { 940 qemu_put_byte(f, QEMU_VM_SECTION_FOOTER); 941 qemu_put_be32(f, se->section_id); 942 } 943 } 944 945 /** 946 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the 947 * command and associated data. 948 * 949 * @f: File to send command on 950 * @command: Command type to send 951 * @len: Length of associated data 952 * @data: Data associated with command. 953 */ 954 static void qemu_savevm_command_send(QEMUFile *f, 955 enum qemu_vm_cmd command, 956 uint16_t len, 957 uint8_t *data) 958 { 959 trace_savevm_command_send(command, len); 960 qemu_put_byte(f, QEMU_VM_COMMAND); 961 qemu_put_be16(f, (uint16_t)command); 962 qemu_put_be16(f, len); 963 qemu_put_buffer(f, data, len); 964 qemu_fflush(f); 965 } 966 967 void qemu_savevm_send_colo_enable(QEMUFile *f) 968 { 969 trace_savevm_send_colo_enable(); 970 qemu_savevm_command_send(f, MIG_CMD_ENABLE_COLO, 0, NULL); 971 } 972 973 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value) 974 { 975 uint32_t buf; 976 977 trace_savevm_send_ping(value); 978 buf = cpu_to_be32(value); 979 qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf); 980 } 981 982 void qemu_savevm_send_open_return_path(QEMUFile *f) 983 { 984 trace_savevm_send_open_return_path(); 985 qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL); 986 } 987 988 /* We have a buffer of data to send; we don't want that all to be loaded 989 * by the command itself, so the command contains just the length of the 990 * extra buffer that we then send straight after it. 991 * TODO: Must be a better way to organise that 992 * 993 * Returns: 994 * 0 on success 995 * -ve on error 996 */ 997 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len) 998 { 999 uint32_t tmp; 1000 1001 if (len > MAX_VM_CMD_PACKAGED_SIZE) { 1002 error_report("%s: Unreasonably large packaged state: %zu", 1003 __func__, len); 1004 return -1; 1005 } 1006 1007 tmp = cpu_to_be32(len); 1008 1009 trace_qemu_savevm_send_packaged(); 1010 qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp); 1011 1012 qemu_put_buffer(f, buf, len); 1013 1014 return 0; 1015 } 1016 1017 /* Send prior to any postcopy transfer */ 1018 void qemu_savevm_send_postcopy_advise(QEMUFile *f) 1019 { 1020 if (migrate_postcopy_ram()) { 1021 uint64_t tmp[2]; 1022 tmp[0] = cpu_to_be64(ram_pagesize_summary()); 1023 tmp[1] = cpu_to_be64(qemu_target_page_size()); 1024 1025 trace_qemu_savevm_send_postcopy_advise(); 1026 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 1027 16, (uint8_t *)tmp); 1028 } else { 1029 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 0, NULL); 1030 } 1031 } 1032 1033 /* Sent prior to starting the destination running in postcopy, discard pages 1034 * that have already been sent but redirtied on the source. 1035 * CMD_POSTCOPY_RAM_DISCARD consist of: 1036 * byte version (0) 1037 * byte Length of name field (not including 0) 1038 * n x byte RAM block name 1039 * byte 0 terminator (just for safety) 1040 * n x Byte ranges within the named RAMBlock 1041 * be64 Start of the range 1042 * be64 Length 1043 * 1044 * name: RAMBlock name that these entries are part of 1045 * len: Number of page entries 1046 * start_list: 'len' addresses 1047 * length_list: 'len' addresses 1048 * 1049 */ 1050 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name, 1051 uint16_t len, 1052 uint64_t *start_list, 1053 uint64_t *length_list) 1054 { 1055 uint8_t *buf; 1056 uint16_t tmplen; 1057 uint16_t t; 1058 size_t name_len = strlen(name); 1059 1060 trace_qemu_savevm_send_postcopy_ram_discard(name, len); 1061 assert(name_len < 256); 1062 buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len); 1063 buf[0] = postcopy_ram_discard_version; 1064 buf[1] = name_len; 1065 memcpy(buf + 2, name, name_len); 1066 tmplen = 2 + name_len; 1067 buf[tmplen++] = '\0'; 1068 1069 for (t = 0; t < len; t++) { 1070 stq_be_p(buf + tmplen, start_list[t]); 1071 tmplen += 8; 1072 stq_be_p(buf + tmplen, length_list[t]); 1073 tmplen += 8; 1074 } 1075 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf); 1076 g_free(buf); 1077 } 1078 1079 /* Get the destination into a state where it can receive postcopy data. */ 1080 void qemu_savevm_send_postcopy_listen(QEMUFile *f) 1081 { 1082 trace_savevm_send_postcopy_listen(); 1083 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL); 1084 } 1085 1086 /* Kick the destination into running */ 1087 void qemu_savevm_send_postcopy_run(QEMUFile *f) 1088 { 1089 trace_savevm_send_postcopy_run(); 1090 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL); 1091 } 1092 1093 void qemu_savevm_send_postcopy_resume(QEMUFile *f) 1094 { 1095 trace_savevm_send_postcopy_resume(); 1096 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RESUME, 0, NULL); 1097 } 1098 1099 void qemu_savevm_send_recv_bitmap(QEMUFile *f, char *block_name) 1100 { 1101 size_t len; 1102 char buf[256]; 1103 1104 trace_savevm_send_recv_bitmap(block_name); 1105 1106 buf[0] = len = strlen(block_name); 1107 memcpy(buf + 1, block_name, len); 1108 1109 qemu_savevm_command_send(f, MIG_CMD_RECV_BITMAP, len + 1, (uint8_t *)buf); 1110 } 1111 1112 bool qemu_savevm_state_blocked(Error **errp) 1113 { 1114 SaveStateEntry *se; 1115 1116 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1117 if (se->vmsd && se->vmsd->unmigratable) { 1118 error_setg(errp, "State blocked by non-migratable device '%s'", 1119 se->idstr); 1120 return true; 1121 } 1122 } 1123 return false; 1124 } 1125 1126 void qemu_savevm_non_migratable_list(strList **reasons) 1127 { 1128 SaveStateEntry *se; 1129 1130 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1131 if (se->vmsd && se->vmsd->unmigratable) { 1132 QAPI_LIST_PREPEND(*reasons, 1133 g_strdup_printf("non-migratable device: %s", 1134 se->idstr)); 1135 } 1136 } 1137 } 1138 1139 void qemu_savevm_state_header(QEMUFile *f) 1140 { 1141 trace_savevm_state_header(); 1142 qemu_put_be32(f, QEMU_VM_FILE_MAGIC); 1143 qemu_put_be32(f, QEMU_VM_FILE_VERSION); 1144 1145 if (migrate_get_current()->send_configuration) { 1146 qemu_put_byte(f, QEMU_VM_CONFIGURATION); 1147 vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0); 1148 } 1149 } 1150 1151 bool qemu_savevm_state_guest_unplug_pending(void) 1152 { 1153 SaveStateEntry *se; 1154 1155 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1156 if (se->vmsd && se->vmsd->dev_unplug_pending && 1157 se->vmsd->dev_unplug_pending(se->opaque)) { 1158 return true; 1159 } 1160 } 1161 1162 return false; 1163 } 1164 1165 void qemu_savevm_state_setup(QEMUFile *f) 1166 { 1167 SaveStateEntry *se; 1168 Error *local_err = NULL; 1169 int ret; 1170 1171 trace_savevm_state_setup(); 1172 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1173 if (!se->ops || !se->ops->save_setup) { 1174 continue; 1175 } 1176 if (se->ops->is_active) { 1177 if (!se->ops->is_active(se->opaque)) { 1178 continue; 1179 } 1180 } 1181 save_section_header(f, se, QEMU_VM_SECTION_START); 1182 1183 ret = se->ops->save_setup(f, se->opaque); 1184 save_section_footer(f, se); 1185 if (ret < 0) { 1186 qemu_file_set_error(f, ret); 1187 break; 1188 } 1189 } 1190 1191 if (precopy_notify(PRECOPY_NOTIFY_SETUP, &local_err)) { 1192 error_report_err(local_err); 1193 } 1194 } 1195 1196 int qemu_savevm_state_resume_prepare(MigrationState *s) 1197 { 1198 SaveStateEntry *se; 1199 int ret; 1200 1201 trace_savevm_state_resume_prepare(); 1202 1203 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1204 if (!se->ops || !se->ops->resume_prepare) { 1205 continue; 1206 } 1207 if (se->ops->is_active) { 1208 if (!se->ops->is_active(se->opaque)) { 1209 continue; 1210 } 1211 } 1212 ret = se->ops->resume_prepare(s, se->opaque); 1213 if (ret < 0) { 1214 return ret; 1215 } 1216 } 1217 1218 return 0; 1219 } 1220 1221 /* 1222 * this function has three return values: 1223 * negative: there was one error, and we have -errno. 1224 * 0 : We haven't finished, caller have to go again 1225 * 1 : We have finished, we can go to complete phase 1226 */ 1227 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy) 1228 { 1229 SaveStateEntry *se; 1230 int ret = 1; 1231 1232 trace_savevm_state_iterate(); 1233 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1234 if (!se->ops || !se->ops->save_live_iterate) { 1235 continue; 1236 } 1237 if (se->ops->is_active && 1238 !se->ops->is_active(se->opaque)) { 1239 continue; 1240 } 1241 if (se->ops->is_active_iterate && 1242 !se->ops->is_active_iterate(se->opaque)) { 1243 continue; 1244 } 1245 /* 1246 * In the postcopy phase, any device that doesn't know how to 1247 * do postcopy should have saved it's state in the _complete 1248 * call that's already run, it might get confused if we call 1249 * iterate afterwards. 1250 */ 1251 if (postcopy && 1252 !(se->ops->has_postcopy && se->ops->has_postcopy(se->opaque))) { 1253 continue; 1254 } 1255 if (qemu_file_rate_limit(f)) { 1256 return 0; 1257 } 1258 trace_savevm_section_start(se->idstr, se->section_id); 1259 1260 save_section_header(f, se, QEMU_VM_SECTION_PART); 1261 1262 ret = se->ops->save_live_iterate(f, se->opaque); 1263 trace_savevm_section_end(se->idstr, se->section_id, ret); 1264 save_section_footer(f, se); 1265 1266 if (ret < 0) { 1267 error_report("failed to save SaveStateEntry with id(name): " 1268 "%d(%s): %d", 1269 se->section_id, se->idstr, ret); 1270 qemu_file_set_error(f, ret); 1271 } 1272 if (ret <= 0) { 1273 /* Do not proceed to the next vmstate before this one reported 1274 completion of the current stage. This serializes the migration 1275 and reduces the probability that a faster changing state is 1276 synchronized over and over again. */ 1277 break; 1278 } 1279 } 1280 return ret; 1281 } 1282 1283 static bool should_send_vmdesc(void) 1284 { 1285 MachineState *machine = MACHINE(qdev_get_machine()); 1286 bool in_postcopy = migration_in_postcopy(); 1287 return !machine->suppress_vmdesc && !in_postcopy; 1288 } 1289 1290 /* 1291 * Calls the save_live_complete_postcopy methods 1292 * causing the last few pages to be sent immediately and doing any associated 1293 * cleanup. 1294 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete 1295 * all the other devices, but that happens at the point we switch to postcopy. 1296 */ 1297 void qemu_savevm_state_complete_postcopy(QEMUFile *f) 1298 { 1299 SaveStateEntry *se; 1300 int ret; 1301 1302 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1303 if (!se->ops || !se->ops->save_live_complete_postcopy) { 1304 continue; 1305 } 1306 if (se->ops->is_active) { 1307 if (!se->ops->is_active(se->opaque)) { 1308 continue; 1309 } 1310 } 1311 trace_savevm_section_start(se->idstr, se->section_id); 1312 /* Section type */ 1313 qemu_put_byte(f, QEMU_VM_SECTION_END); 1314 qemu_put_be32(f, se->section_id); 1315 1316 ret = se->ops->save_live_complete_postcopy(f, se->opaque); 1317 trace_savevm_section_end(se->idstr, se->section_id, ret); 1318 save_section_footer(f, se); 1319 if (ret < 0) { 1320 qemu_file_set_error(f, ret); 1321 return; 1322 } 1323 } 1324 1325 qemu_put_byte(f, QEMU_VM_EOF); 1326 qemu_fflush(f); 1327 } 1328 1329 static 1330 int qemu_savevm_state_complete_precopy_iterable(QEMUFile *f, bool in_postcopy) 1331 { 1332 SaveStateEntry *se; 1333 int ret; 1334 1335 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1336 if (!se->ops || 1337 (in_postcopy && se->ops->has_postcopy && 1338 se->ops->has_postcopy(se->opaque)) || 1339 !se->ops->save_live_complete_precopy) { 1340 continue; 1341 } 1342 1343 if (se->ops->is_active) { 1344 if (!se->ops->is_active(se->opaque)) { 1345 continue; 1346 } 1347 } 1348 trace_savevm_section_start(se->idstr, se->section_id); 1349 1350 save_section_header(f, se, QEMU_VM_SECTION_END); 1351 1352 ret = se->ops->save_live_complete_precopy(f, se->opaque); 1353 trace_savevm_section_end(se->idstr, se->section_id, ret); 1354 save_section_footer(f, se); 1355 if (ret < 0) { 1356 qemu_file_set_error(f, ret); 1357 return -1; 1358 } 1359 } 1360 1361 return 0; 1362 } 1363 1364 int qemu_savevm_state_complete_precopy_non_iterable(QEMUFile *f, 1365 bool in_postcopy, 1366 bool inactivate_disks) 1367 { 1368 g_autoptr(JSONWriter) vmdesc = NULL; 1369 int vmdesc_len; 1370 SaveStateEntry *se; 1371 int ret; 1372 1373 vmdesc = json_writer_new(false); 1374 json_writer_start_object(vmdesc, NULL); 1375 json_writer_int64(vmdesc, "page_size", qemu_target_page_size()); 1376 json_writer_start_array(vmdesc, "devices"); 1377 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1378 1379 if ((!se->ops || !se->ops->save_state) && !se->vmsd) { 1380 continue; 1381 } 1382 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) { 1383 trace_savevm_section_skip(se->idstr, se->section_id); 1384 continue; 1385 } 1386 1387 trace_savevm_section_start(se->idstr, se->section_id); 1388 1389 json_writer_start_object(vmdesc, NULL); 1390 json_writer_str(vmdesc, "name", se->idstr); 1391 json_writer_int64(vmdesc, "instance_id", se->instance_id); 1392 1393 save_section_header(f, se, QEMU_VM_SECTION_FULL); 1394 ret = vmstate_save(f, se, vmdesc); 1395 if (ret) { 1396 qemu_file_set_error(f, ret); 1397 return ret; 1398 } 1399 trace_savevm_section_end(se->idstr, se->section_id, 0); 1400 save_section_footer(f, se); 1401 1402 json_writer_end_object(vmdesc); 1403 } 1404 1405 if (inactivate_disks) { 1406 /* Inactivate before sending QEMU_VM_EOF so that the 1407 * bdrv_activate_all() on the other end won't fail. */ 1408 ret = bdrv_inactivate_all(); 1409 if (ret) { 1410 error_report("%s: bdrv_inactivate_all() failed (%d)", 1411 __func__, ret); 1412 qemu_file_set_error(f, ret); 1413 return ret; 1414 } 1415 } 1416 if (!in_postcopy) { 1417 /* Postcopy stream will still be going */ 1418 qemu_put_byte(f, QEMU_VM_EOF); 1419 } 1420 1421 json_writer_end_array(vmdesc); 1422 json_writer_end_object(vmdesc); 1423 vmdesc_len = strlen(json_writer_get(vmdesc)); 1424 1425 if (should_send_vmdesc()) { 1426 qemu_put_byte(f, QEMU_VM_VMDESCRIPTION); 1427 qemu_put_be32(f, vmdesc_len); 1428 qemu_put_buffer(f, (uint8_t *)json_writer_get(vmdesc), vmdesc_len); 1429 } 1430 1431 return 0; 1432 } 1433 1434 int qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only, 1435 bool inactivate_disks) 1436 { 1437 int ret; 1438 Error *local_err = NULL; 1439 bool in_postcopy = migration_in_postcopy(); 1440 1441 if (precopy_notify(PRECOPY_NOTIFY_COMPLETE, &local_err)) { 1442 error_report_err(local_err); 1443 } 1444 1445 trace_savevm_state_complete_precopy(); 1446 1447 cpu_synchronize_all_states(); 1448 1449 if (!in_postcopy || iterable_only) { 1450 ret = qemu_savevm_state_complete_precopy_iterable(f, in_postcopy); 1451 if (ret) { 1452 return ret; 1453 } 1454 } 1455 1456 if (iterable_only) { 1457 goto flush; 1458 } 1459 1460 ret = qemu_savevm_state_complete_precopy_non_iterable(f, in_postcopy, 1461 inactivate_disks); 1462 if (ret) { 1463 return ret; 1464 } 1465 1466 flush: 1467 qemu_fflush(f); 1468 return 0; 1469 } 1470 1471 /* Give an estimate of the amount left to be transferred, 1472 * the result is split into the amount for units that can and 1473 * for units that can't do postcopy. 1474 */ 1475 void qemu_savevm_state_pending(QEMUFile *f, uint64_t threshold_size, 1476 uint64_t *res_precopy_only, 1477 uint64_t *res_compatible, 1478 uint64_t *res_postcopy_only) 1479 { 1480 SaveStateEntry *se; 1481 1482 *res_precopy_only = 0; 1483 *res_compatible = 0; 1484 *res_postcopy_only = 0; 1485 1486 1487 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1488 if (!se->ops || !se->ops->save_live_pending) { 1489 continue; 1490 } 1491 if (se->ops->is_active) { 1492 if (!se->ops->is_active(se->opaque)) { 1493 continue; 1494 } 1495 } 1496 se->ops->save_live_pending(f, se->opaque, threshold_size, 1497 res_precopy_only, res_compatible, 1498 res_postcopy_only); 1499 } 1500 } 1501 1502 void qemu_savevm_state_cleanup(void) 1503 { 1504 SaveStateEntry *se; 1505 Error *local_err = NULL; 1506 1507 if (precopy_notify(PRECOPY_NOTIFY_CLEANUP, &local_err)) { 1508 error_report_err(local_err); 1509 } 1510 1511 trace_savevm_state_cleanup(); 1512 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1513 if (se->ops && se->ops->save_cleanup) { 1514 se->ops->save_cleanup(se->opaque); 1515 } 1516 } 1517 } 1518 1519 static int qemu_savevm_state(QEMUFile *f, Error **errp) 1520 { 1521 int ret; 1522 MigrationState *ms = migrate_get_current(); 1523 MigrationStatus status; 1524 1525 if (migration_is_running(ms->state)) { 1526 error_setg(errp, QERR_MIGRATION_ACTIVE); 1527 return -EINVAL; 1528 } 1529 1530 if (migrate_use_block()) { 1531 error_setg(errp, "Block migration and snapshots are incompatible"); 1532 return -EINVAL; 1533 } 1534 1535 migrate_init(ms); 1536 memset(&ram_counters, 0, sizeof(ram_counters)); 1537 memset(&compression_counters, 0, sizeof(compression_counters)); 1538 ms->to_dst_file = f; 1539 1540 qemu_mutex_unlock_iothread(); 1541 qemu_savevm_state_header(f); 1542 qemu_savevm_state_setup(f); 1543 qemu_mutex_lock_iothread(); 1544 1545 while (qemu_file_get_error(f) == 0) { 1546 if (qemu_savevm_state_iterate(f, false) > 0) { 1547 break; 1548 } 1549 } 1550 1551 ret = qemu_file_get_error(f); 1552 if (ret == 0) { 1553 qemu_savevm_state_complete_precopy(f, false, false); 1554 ret = qemu_file_get_error(f); 1555 } 1556 qemu_savevm_state_cleanup(); 1557 if (ret != 0) { 1558 error_setg_errno(errp, -ret, "Error while writing VM state"); 1559 } 1560 1561 if (ret != 0) { 1562 status = MIGRATION_STATUS_FAILED; 1563 } else { 1564 status = MIGRATION_STATUS_COMPLETED; 1565 } 1566 migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status); 1567 1568 /* f is outer parameter, it should not stay in global migration state after 1569 * this function finished */ 1570 ms->to_dst_file = NULL; 1571 1572 return ret; 1573 } 1574 1575 void qemu_savevm_live_state(QEMUFile *f) 1576 { 1577 /* save QEMU_VM_SECTION_END section */ 1578 qemu_savevm_state_complete_precopy(f, true, false); 1579 qemu_put_byte(f, QEMU_VM_EOF); 1580 } 1581 1582 int qemu_save_device_state(QEMUFile *f) 1583 { 1584 SaveStateEntry *se; 1585 1586 if (!migration_in_colo_state()) { 1587 qemu_put_be32(f, QEMU_VM_FILE_MAGIC); 1588 qemu_put_be32(f, QEMU_VM_FILE_VERSION); 1589 } 1590 cpu_synchronize_all_states(); 1591 1592 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1593 int ret; 1594 1595 if (se->is_ram) { 1596 continue; 1597 } 1598 if ((!se->ops || !se->ops->save_state) && !se->vmsd) { 1599 continue; 1600 } 1601 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) { 1602 continue; 1603 } 1604 1605 save_section_header(f, se, QEMU_VM_SECTION_FULL); 1606 1607 ret = vmstate_save(f, se, NULL); 1608 if (ret) { 1609 return ret; 1610 } 1611 1612 save_section_footer(f, se); 1613 } 1614 1615 qemu_put_byte(f, QEMU_VM_EOF); 1616 1617 return qemu_file_get_error(f); 1618 } 1619 1620 static SaveStateEntry *find_se(const char *idstr, uint32_t instance_id) 1621 { 1622 SaveStateEntry *se; 1623 1624 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1625 if (!strcmp(se->idstr, idstr) && 1626 (instance_id == se->instance_id || 1627 instance_id == se->alias_id)) 1628 return se; 1629 /* Migrating from an older version? */ 1630 if (strstr(se->idstr, idstr) && se->compat) { 1631 if (!strcmp(se->compat->idstr, idstr) && 1632 (instance_id == se->compat->instance_id || 1633 instance_id == se->alias_id)) 1634 return se; 1635 } 1636 } 1637 return NULL; 1638 } 1639 1640 enum LoadVMExitCodes { 1641 /* Allow a command to quit all layers of nested loadvm loops */ 1642 LOADVM_QUIT = 1, 1643 }; 1644 1645 /* ------ incoming postcopy messages ------ */ 1646 /* 'advise' arrives before any transfers just to tell us that a postcopy 1647 * *might* happen - it might be skipped if precopy transferred everything 1648 * quickly. 1649 */ 1650 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis, 1651 uint16_t len) 1652 { 1653 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE); 1654 uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps; 1655 size_t page_size = qemu_target_page_size(); 1656 Error *local_err = NULL; 1657 1658 trace_loadvm_postcopy_handle_advise(); 1659 if (ps != POSTCOPY_INCOMING_NONE) { 1660 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps); 1661 return -1; 1662 } 1663 1664 switch (len) { 1665 case 0: 1666 if (migrate_postcopy_ram()) { 1667 error_report("RAM postcopy is enabled but have 0 byte advise"); 1668 return -EINVAL; 1669 } 1670 return 0; 1671 case 8 + 8: 1672 if (!migrate_postcopy_ram()) { 1673 error_report("RAM postcopy is disabled but have 16 byte advise"); 1674 return -EINVAL; 1675 } 1676 break; 1677 default: 1678 error_report("CMD_POSTCOPY_ADVISE invalid length (%d)", len); 1679 return -EINVAL; 1680 } 1681 1682 if (!postcopy_ram_supported_by_host(mis)) { 1683 postcopy_state_set(POSTCOPY_INCOMING_NONE); 1684 return -1; 1685 } 1686 1687 remote_pagesize_summary = qemu_get_be64(mis->from_src_file); 1688 local_pagesize_summary = ram_pagesize_summary(); 1689 1690 if (remote_pagesize_summary != local_pagesize_summary) { 1691 /* 1692 * This detects two potential causes of mismatch: 1693 * a) A mismatch in host page sizes 1694 * Some combinations of mismatch are probably possible but it gets 1695 * a bit more complicated. In particular we need to place whole 1696 * host pages on the dest at once, and we need to ensure that we 1697 * handle dirtying to make sure we never end up sending part of 1698 * a hostpage on it's own. 1699 * b) The use of different huge page sizes on source/destination 1700 * a more fine grain test is performed during RAM block migration 1701 * but this test here causes a nice early clear failure, and 1702 * also fails when passed to an older qemu that doesn't 1703 * do huge pages. 1704 */ 1705 error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64 1706 " d=%" PRIx64 ")", 1707 remote_pagesize_summary, local_pagesize_summary); 1708 return -1; 1709 } 1710 1711 remote_tps = qemu_get_be64(mis->from_src_file); 1712 if (remote_tps != page_size) { 1713 /* 1714 * Again, some differences could be dealt with, but for now keep it 1715 * simple. 1716 */ 1717 error_report("Postcopy needs matching target page sizes (s=%d d=%zd)", 1718 (int)remote_tps, page_size); 1719 return -1; 1720 } 1721 1722 if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_ADVISE, &local_err)) { 1723 error_report_err(local_err); 1724 return -1; 1725 } 1726 1727 if (ram_postcopy_incoming_init(mis)) { 1728 return -1; 1729 } 1730 1731 return 0; 1732 } 1733 1734 /* After postcopy we will be told to throw some pages away since they're 1735 * dirty and will have to be demand fetched. Must happen before CPU is 1736 * started. 1737 * There can be 0..many of these messages, each encoding multiple pages. 1738 */ 1739 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis, 1740 uint16_t len) 1741 { 1742 int tmp; 1743 char ramid[256]; 1744 PostcopyState ps = postcopy_state_get(); 1745 1746 trace_loadvm_postcopy_ram_handle_discard(); 1747 1748 switch (ps) { 1749 case POSTCOPY_INCOMING_ADVISE: 1750 /* 1st discard */ 1751 tmp = postcopy_ram_prepare_discard(mis); 1752 if (tmp) { 1753 return tmp; 1754 } 1755 break; 1756 1757 case POSTCOPY_INCOMING_DISCARD: 1758 /* Expected state */ 1759 break; 1760 1761 default: 1762 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)", 1763 ps); 1764 return -1; 1765 } 1766 /* We're expecting a 1767 * Version (0) 1768 * a RAM ID string (length byte, name, 0 term) 1769 * then at least 1 16 byte chunk 1770 */ 1771 if (len < (1 + 1 + 1 + 1 + 2 * 8)) { 1772 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len); 1773 return -1; 1774 } 1775 1776 tmp = qemu_get_byte(mis->from_src_file); 1777 if (tmp != postcopy_ram_discard_version) { 1778 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp); 1779 return -1; 1780 } 1781 1782 if (!qemu_get_counted_string(mis->from_src_file, ramid)) { 1783 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID"); 1784 return -1; 1785 } 1786 tmp = qemu_get_byte(mis->from_src_file); 1787 if (tmp != 0) { 1788 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp); 1789 return -1; 1790 } 1791 1792 len -= 3 + strlen(ramid); 1793 if (len % 16) { 1794 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len); 1795 return -1; 1796 } 1797 trace_loadvm_postcopy_ram_handle_discard_header(ramid, len); 1798 while (len) { 1799 uint64_t start_addr, block_length; 1800 start_addr = qemu_get_be64(mis->from_src_file); 1801 block_length = qemu_get_be64(mis->from_src_file); 1802 1803 len -= 16; 1804 int ret = ram_discard_range(ramid, start_addr, block_length); 1805 if (ret) { 1806 return ret; 1807 } 1808 } 1809 trace_loadvm_postcopy_ram_handle_discard_end(); 1810 1811 return 0; 1812 } 1813 1814 /* 1815 * Triggered by a postcopy_listen command; this thread takes over reading 1816 * the input stream, leaving the main thread free to carry on loading the rest 1817 * of the device state (from RAM). 1818 * (TODO:This could do with being in a postcopy file - but there again it's 1819 * just another input loop, not that postcopy specific) 1820 */ 1821 static void *postcopy_ram_listen_thread(void *opaque) 1822 { 1823 MigrationIncomingState *mis = migration_incoming_get_current(); 1824 QEMUFile *f = mis->from_src_file; 1825 int load_res; 1826 MigrationState *migr = migrate_get_current(); 1827 1828 object_ref(OBJECT(migr)); 1829 1830 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, 1831 MIGRATION_STATUS_POSTCOPY_ACTIVE); 1832 qemu_sem_post(&mis->thread_sync_sem); 1833 trace_postcopy_ram_listen_thread_start(); 1834 1835 rcu_register_thread(); 1836 /* 1837 * Because we're a thread and not a coroutine we can't yield 1838 * in qemu_file, and thus we must be blocking now. 1839 */ 1840 qemu_file_set_blocking(f, true); 1841 load_res = qemu_loadvm_state_main(f, mis); 1842 1843 /* 1844 * This is tricky, but, mis->from_src_file can change after it 1845 * returns, when postcopy recovery happened. In the future, we may 1846 * want a wrapper for the QEMUFile handle. 1847 */ 1848 f = mis->from_src_file; 1849 1850 /* And non-blocking again so we don't block in any cleanup */ 1851 qemu_file_set_blocking(f, false); 1852 1853 trace_postcopy_ram_listen_thread_exit(); 1854 if (load_res < 0) { 1855 qemu_file_set_error(f, load_res); 1856 dirty_bitmap_mig_cancel_incoming(); 1857 if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING && 1858 !migrate_postcopy_ram() && migrate_dirty_bitmaps()) 1859 { 1860 error_report("%s: loadvm failed during postcopy: %d. All states " 1861 "are migrated except dirty bitmaps. Some dirty " 1862 "bitmaps may be lost, and present migrated dirty " 1863 "bitmaps are correctly migrated and valid.", 1864 __func__, load_res); 1865 load_res = 0; /* prevent further exit() */ 1866 } else { 1867 error_report("%s: loadvm failed: %d", __func__, load_res); 1868 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, 1869 MIGRATION_STATUS_FAILED); 1870 } 1871 } 1872 if (load_res >= 0) { 1873 /* 1874 * This looks good, but it's possible that the device loading in the 1875 * main thread hasn't finished yet, and so we might not be in 'RUN' 1876 * state yet; wait for the end of the main thread. 1877 */ 1878 qemu_event_wait(&mis->main_thread_load_event); 1879 } 1880 postcopy_ram_incoming_cleanup(mis); 1881 1882 if (load_res < 0) { 1883 /* 1884 * If something went wrong then we have a bad state so exit; 1885 * depending how far we got it might be possible at this point 1886 * to leave the guest running and fire MCEs for pages that never 1887 * arrived as a desperate recovery step. 1888 */ 1889 rcu_unregister_thread(); 1890 exit(EXIT_FAILURE); 1891 } 1892 1893 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, 1894 MIGRATION_STATUS_COMPLETED); 1895 /* 1896 * If everything has worked fine, then the main thread has waited 1897 * for us to start, and we're the last use of the mis. 1898 * (If something broke then qemu will have to exit anyway since it's 1899 * got a bad migration state). 1900 */ 1901 migration_incoming_state_destroy(); 1902 qemu_loadvm_state_cleanup(); 1903 1904 rcu_unregister_thread(); 1905 mis->have_listen_thread = false; 1906 postcopy_state_set(POSTCOPY_INCOMING_END); 1907 1908 object_unref(OBJECT(migr)); 1909 1910 return NULL; 1911 } 1912 1913 /* After this message we must be able to immediately receive postcopy data */ 1914 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis) 1915 { 1916 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING); 1917 Error *local_err = NULL; 1918 1919 trace_loadvm_postcopy_handle_listen("enter"); 1920 1921 if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) { 1922 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps); 1923 return -1; 1924 } 1925 if (ps == POSTCOPY_INCOMING_ADVISE) { 1926 /* 1927 * A rare case, we entered listen without having to do any discards, 1928 * so do the setup that's normally done at the time of the 1st discard. 1929 */ 1930 if (migrate_postcopy_ram()) { 1931 postcopy_ram_prepare_discard(mis); 1932 } 1933 } 1934 1935 trace_loadvm_postcopy_handle_listen("after discard"); 1936 1937 /* 1938 * Sensitise RAM - can now generate requests for blocks that don't exist 1939 * However, at this point the CPU shouldn't be running, and the IO 1940 * shouldn't be doing anything yet so don't actually expect requests 1941 */ 1942 if (migrate_postcopy_ram()) { 1943 if (postcopy_ram_incoming_setup(mis)) { 1944 postcopy_ram_incoming_cleanup(mis); 1945 return -1; 1946 } 1947 } 1948 1949 trace_loadvm_postcopy_handle_listen("after uffd"); 1950 1951 if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_LISTEN, &local_err)) { 1952 error_report_err(local_err); 1953 return -1; 1954 } 1955 1956 mis->have_listen_thread = true; 1957 postcopy_thread_create(mis, &mis->listen_thread, "postcopy/listen", 1958 postcopy_ram_listen_thread, QEMU_THREAD_DETACHED); 1959 trace_loadvm_postcopy_handle_listen("return"); 1960 1961 return 0; 1962 } 1963 1964 static void loadvm_postcopy_handle_run_bh(void *opaque) 1965 { 1966 Error *local_err = NULL; 1967 MigrationIncomingState *mis = opaque; 1968 1969 trace_loadvm_postcopy_handle_run_bh("enter"); 1970 1971 /* TODO we should move all of this lot into postcopy_ram.c or a shared code 1972 * in migration.c 1973 */ 1974 cpu_synchronize_all_post_init(); 1975 1976 trace_loadvm_postcopy_handle_run_bh("after cpu sync"); 1977 1978 qemu_announce_self(&mis->announce_timer, migrate_announce_params()); 1979 1980 trace_loadvm_postcopy_handle_run_bh("after announce"); 1981 1982 /* Make sure all file formats throw away their mutable metadata. 1983 * If we get an error here, just don't restart the VM yet. */ 1984 bdrv_activate_all(&local_err); 1985 if (local_err) { 1986 error_report_err(local_err); 1987 local_err = NULL; 1988 autostart = false; 1989 } 1990 1991 trace_loadvm_postcopy_handle_run_bh("after invalidate cache"); 1992 1993 dirty_bitmap_mig_before_vm_start(); 1994 1995 if (autostart) { 1996 /* Hold onto your hats, starting the CPU */ 1997 vm_start(); 1998 } else { 1999 /* leave it paused and let management decide when to start the CPU */ 2000 runstate_set(RUN_STATE_PAUSED); 2001 } 2002 2003 qemu_bh_delete(mis->bh); 2004 2005 trace_loadvm_postcopy_handle_run_bh("return"); 2006 } 2007 2008 /* After all discards we can start running and asking for pages */ 2009 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis) 2010 { 2011 PostcopyState ps = postcopy_state_get(); 2012 2013 trace_loadvm_postcopy_handle_run(); 2014 if (ps != POSTCOPY_INCOMING_LISTENING) { 2015 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps); 2016 return -1; 2017 } 2018 2019 postcopy_state_set(POSTCOPY_INCOMING_RUNNING); 2020 mis->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, mis); 2021 qemu_bh_schedule(mis->bh); 2022 2023 /* We need to finish reading the stream from the package 2024 * and also stop reading anything more from the stream that loaded the 2025 * package (since it's now being read by the listener thread). 2026 * LOADVM_QUIT will quit all the layers of nested loadvm loops. 2027 */ 2028 return LOADVM_QUIT; 2029 } 2030 2031 /* We must be with page_request_mutex held */ 2032 static gboolean postcopy_sync_page_req(gpointer key, gpointer value, 2033 gpointer data) 2034 { 2035 MigrationIncomingState *mis = data; 2036 void *host_addr = (void *) key; 2037 ram_addr_t rb_offset; 2038 RAMBlock *rb; 2039 int ret; 2040 2041 rb = qemu_ram_block_from_host(host_addr, true, &rb_offset); 2042 if (!rb) { 2043 /* 2044 * This should _never_ happen. However be nice for a migrating VM to 2045 * not crash/assert. Post an error (note: intended to not use *_once 2046 * because we do want to see all the illegal addresses; and this can 2047 * never be triggered by the guest so we're safe) and move on next. 2048 */ 2049 error_report("%s: illegal host addr %p", __func__, host_addr); 2050 /* Try the next entry */ 2051 return FALSE; 2052 } 2053 2054 ret = migrate_send_rp_message_req_pages(mis, rb, rb_offset); 2055 if (ret) { 2056 /* Please refer to above comment. */ 2057 error_report("%s: send rp message failed for addr %p", 2058 __func__, host_addr); 2059 return FALSE; 2060 } 2061 2062 trace_postcopy_page_req_sync(host_addr); 2063 2064 return FALSE; 2065 } 2066 2067 static void migrate_send_rp_req_pages_pending(MigrationIncomingState *mis) 2068 { 2069 WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) { 2070 g_tree_foreach(mis->page_requested, postcopy_sync_page_req, mis); 2071 } 2072 } 2073 2074 static int loadvm_postcopy_handle_resume(MigrationIncomingState *mis) 2075 { 2076 if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) { 2077 error_report("%s: illegal resume received", __func__); 2078 /* Don't fail the load, only for this. */ 2079 return 0; 2080 } 2081 2082 /* 2083 * Reset the last_rb before we resend any page req to source again, since 2084 * the source should have it reset already. 2085 */ 2086 mis->last_rb = NULL; 2087 2088 /* 2089 * This means source VM is ready to resume the postcopy migration. 2090 */ 2091 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_RECOVER, 2092 MIGRATION_STATUS_POSTCOPY_ACTIVE); 2093 2094 trace_loadvm_postcopy_handle_resume(); 2095 2096 /* Tell source that "we are ready" */ 2097 migrate_send_rp_resume_ack(mis, MIGRATION_RESUME_ACK_VALUE); 2098 2099 /* 2100 * After a postcopy recovery, the source should have lost the postcopy 2101 * queue, or potentially the requested pages could have been lost during 2102 * the network down phase. Let's re-sync with the source VM by re-sending 2103 * all the pending pages that we eagerly need, so these threads won't get 2104 * blocked too long due to the recovery. 2105 * 2106 * Without this procedure, the faulted destination VM threads (waiting for 2107 * page requests right before the postcopy is interrupted) can keep hanging 2108 * until the pages are sent by the source during the background copying of 2109 * pages, or another thread faulted on the same address accidentally. 2110 */ 2111 migrate_send_rp_req_pages_pending(mis); 2112 2113 /* 2114 * It's time to switch state and release the fault thread to continue 2115 * service page faults. Note that this should be explicitly after the 2116 * above call to migrate_send_rp_req_pages_pending(). In short: 2117 * migrate_send_rp_message_req_pages() is not thread safe, yet. 2118 */ 2119 qemu_sem_post(&mis->postcopy_pause_sem_fault); 2120 2121 if (migrate_postcopy_preempt()) { 2122 /* The channel should already be setup again; make sure of it */ 2123 assert(mis->postcopy_qemufile_dst); 2124 /* Kick the fast ram load thread too */ 2125 qemu_sem_post(&mis->postcopy_pause_sem_fast_load); 2126 } 2127 2128 return 0; 2129 } 2130 2131 /** 2132 * Immediately following this command is a blob of data containing an embedded 2133 * chunk of migration stream; read it and load it. 2134 * 2135 * @mis: Incoming state 2136 * @length: Length of packaged data to read 2137 * 2138 * Returns: Negative values on error 2139 * 2140 */ 2141 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis) 2142 { 2143 int ret; 2144 size_t length; 2145 QIOChannelBuffer *bioc; 2146 2147 length = qemu_get_be32(mis->from_src_file); 2148 trace_loadvm_handle_cmd_packaged(length); 2149 2150 if (length > MAX_VM_CMD_PACKAGED_SIZE) { 2151 error_report("Unreasonably large packaged state: %zu", length); 2152 return -1; 2153 } 2154 2155 bioc = qio_channel_buffer_new(length); 2156 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer"); 2157 ret = qemu_get_buffer(mis->from_src_file, 2158 bioc->data, 2159 length); 2160 if (ret != length) { 2161 object_unref(OBJECT(bioc)); 2162 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu", 2163 ret, length); 2164 return (ret < 0) ? ret : -EAGAIN; 2165 } 2166 bioc->usage += length; 2167 trace_loadvm_handle_cmd_packaged_received(ret); 2168 2169 QEMUFile *packf = qemu_file_new_input(QIO_CHANNEL(bioc)); 2170 2171 ret = qemu_loadvm_state_main(packf, mis); 2172 trace_loadvm_handle_cmd_packaged_main(ret); 2173 qemu_fclose(packf); 2174 object_unref(OBJECT(bioc)); 2175 2176 return ret; 2177 } 2178 2179 /* 2180 * Handle request that source requests for recved_bitmap on 2181 * destination. Payload format: 2182 * 2183 * len (1 byte) + ramblock_name (<255 bytes) 2184 */ 2185 static int loadvm_handle_recv_bitmap(MigrationIncomingState *mis, 2186 uint16_t len) 2187 { 2188 QEMUFile *file = mis->from_src_file; 2189 RAMBlock *rb; 2190 char block_name[256]; 2191 size_t cnt; 2192 2193 cnt = qemu_get_counted_string(file, block_name); 2194 if (!cnt) { 2195 error_report("%s: failed to read block name", __func__); 2196 return -EINVAL; 2197 } 2198 2199 /* Validate before using the data */ 2200 if (qemu_file_get_error(file)) { 2201 return qemu_file_get_error(file); 2202 } 2203 2204 if (len != cnt + 1) { 2205 error_report("%s: invalid payload length (%d)", __func__, len); 2206 return -EINVAL; 2207 } 2208 2209 rb = qemu_ram_block_by_name(block_name); 2210 if (!rb) { 2211 error_report("%s: block '%s' not found", __func__, block_name); 2212 return -EINVAL; 2213 } 2214 2215 migrate_send_rp_recv_bitmap(mis, block_name); 2216 2217 trace_loadvm_handle_recv_bitmap(block_name); 2218 2219 return 0; 2220 } 2221 2222 static int loadvm_process_enable_colo(MigrationIncomingState *mis) 2223 { 2224 int ret = migration_incoming_enable_colo(); 2225 2226 if (!ret) { 2227 ret = colo_init_ram_cache(); 2228 if (ret) { 2229 migration_incoming_disable_colo(); 2230 } 2231 } 2232 return ret; 2233 } 2234 2235 /* 2236 * Process an incoming 'QEMU_VM_COMMAND' 2237 * 0 just a normal return 2238 * LOADVM_QUIT All good, but exit the loop 2239 * <0 Error 2240 */ 2241 static int loadvm_process_command(QEMUFile *f) 2242 { 2243 MigrationIncomingState *mis = migration_incoming_get_current(); 2244 uint16_t cmd; 2245 uint16_t len; 2246 uint32_t tmp32; 2247 2248 cmd = qemu_get_be16(f); 2249 len = qemu_get_be16(f); 2250 2251 /* Check validity before continue processing of cmds */ 2252 if (qemu_file_get_error(f)) { 2253 return qemu_file_get_error(f); 2254 } 2255 2256 if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) { 2257 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len); 2258 return -EINVAL; 2259 } 2260 2261 trace_loadvm_process_command(mig_cmd_args[cmd].name, len); 2262 2263 if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) { 2264 error_report("%s received with bad length - expecting %zu, got %d", 2265 mig_cmd_args[cmd].name, 2266 (size_t)mig_cmd_args[cmd].len, len); 2267 return -ERANGE; 2268 } 2269 2270 switch (cmd) { 2271 case MIG_CMD_OPEN_RETURN_PATH: 2272 if (mis->to_src_file) { 2273 error_report("CMD_OPEN_RETURN_PATH called when RP already open"); 2274 /* Not really a problem, so don't give up */ 2275 return 0; 2276 } 2277 mis->to_src_file = qemu_file_get_return_path(f); 2278 if (!mis->to_src_file) { 2279 error_report("CMD_OPEN_RETURN_PATH failed"); 2280 return -1; 2281 } 2282 break; 2283 2284 case MIG_CMD_PING: 2285 tmp32 = qemu_get_be32(f); 2286 trace_loadvm_process_command_ping(tmp32); 2287 if (!mis->to_src_file) { 2288 error_report("CMD_PING (0x%x) received with no return path", 2289 tmp32); 2290 return -1; 2291 } 2292 migrate_send_rp_pong(mis, tmp32); 2293 break; 2294 2295 case MIG_CMD_PACKAGED: 2296 return loadvm_handle_cmd_packaged(mis); 2297 2298 case MIG_CMD_POSTCOPY_ADVISE: 2299 return loadvm_postcopy_handle_advise(mis, len); 2300 2301 case MIG_CMD_POSTCOPY_LISTEN: 2302 return loadvm_postcopy_handle_listen(mis); 2303 2304 case MIG_CMD_POSTCOPY_RUN: 2305 return loadvm_postcopy_handle_run(mis); 2306 2307 case MIG_CMD_POSTCOPY_RAM_DISCARD: 2308 return loadvm_postcopy_ram_handle_discard(mis, len); 2309 2310 case MIG_CMD_POSTCOPY_RESUME: 2311 return loadvm_postcopy_handle_resume(mis); 2312 2313 case MIG_CMD_RECV_BITMAP: 2314 return loadvm_handle_recv_bitmap(mis, len); 2315 2316 case MIG_CMD_ENABLE_COLO: 2317 return loadvm_process_enable_colo(mis); 2318 } 2319 2320 return 0; 2321 } 2322 2323 /* 2324 * Read a footer off the wire and check that it matches the expected section 2325 * 2326 * Returns: true if the footer was good 2327 * false if there is a problem (and calls error_report to say why) 2328 */ 2329 static bool check_section_footer(QEMUFile *f, SaveStateEntry *se) 2330 { 2331 int ret; 2332 uint8_t read_mark; 2333 uint32_t read_section_id; 2334 2335 if (!migrate_get_current()->send_section_footer) { 2336 /* No footer to check */ 2337 return true; 2338 } 2339 2340 read_mark = qemu_get_byte(f); 2341 2342 ret = qemu_file_get_error(f); 2343 if (ret) { 2344 error_report("%s: Read section footer failed: %d", 2345 __func__, ret); 2346 return false; 2347 } 2348 2349 if (read_mark != QEMU_VM_SECTION_FOOTER) { 2350 error_report("Missing section footer for %s", se->idstr); 2351 return false; 2352 } 2353 2354 read_section_id = qemu_get_be32(f); 2355 if (read_section_id != se->load_section_id) { 2356 error_report("Mismatched section id in footer for %s -" 2357 " read 0x%x expected 0x%x", 2358 se->idstr, read_section_id, se->load_section_id); 2359 return false; 2360 } 2361 2362 /* All good */ 2363 return true; 2364 } 2365 2366 static int 2367 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis) 2368 { 2369 uint32_t instance_id, version_id, section_id; 2370 SaveStateEntry *se; 2371 char idstr[256]; 2372 int ret; 2373 2374 /* Read section start */ 2375 section_id = qemu_get_be32(f); 2376 if (!qemu_get_counted_string(f, idstr)) { 2377 error_report("Unable to read ID string for section %u", 2378 section_id); 2379 return -EINVAL; 2380 } 2381 instance_id = qemu_get_be32(f); 2382 version_id = qemu_get_be32(f); 2383 2384 ret = qemu_file_get_error(f); 2385 if (ret) { 2386 error_report("%s: Failed to read instance/version ID: %d", 2387 __func__, ret); 2388 return ret; 2389 } 2390 2391 trace_qemu_loadvm_state_section_startfull(section_id, idstr, 2392 instance_id, version_id); 2393 /* Find savevm section */ 2394 se = find_se(idstr, instance_id); 2395 if (se == NULL) { 2396 error_report("Unknown savevm section or instance '%s' %"PRIu32". " 2397 "Make sure that your current VM setup matches your " 2398 "saved VM setup, including any hotplugged devices", 2399 idstr, instance_id); 2400 return -EINVAL; 2401 } 2402 2403 /* Validate version */ 2404 if (version_id > se->version_id) { 2405 error_report("savevm: unsupported version %d for '%s' v%d", 2406 version_id, idstr, se->version_id); 2407 return -EINVAL; 2408 } 2409 se->load_version_id = version_id; 2410 se->load_section_id = section_id; 2411 2412 /* Validate if it is a device's state */ 2413 if (xen_enabled() && se->is_ram) { 2414 error_report("loadvm: %s RAM loading not allowed on Xen", idstr); 2415 return -EINVAL; 2416 } 2417 2418 ret = vmstate_load(f, se); 2419 if (ret < 0) { 2420 error_report("error while loading state for instance 0x%"PRIx32" of" 2421 " device '%s'", instance_id, idstr); 2422 return ret; 2423 } 2424 if (!check_section_footer(f, se)) { 2425 return -EINVAL; 2426 } 2427 2428 return 0; 2429 } 2430 2431 static int 2432 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis) 2433 { 2434 uint32_t section_id; 2435 SaveStateEntry *se; 2436 int ret; 2437 2438 section_id = qemu_get_be32(f); 2439 2440 ret = qemu_file_get_error(f); 2441 if (ret) { 2442 error_report("%s: Failed to read section ID: %d", 2443 __func__, ret); 2444 return ret; 2445 } 2446 2447 trace_qemu_loadvm_state_section_partend(section_id); 2448 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 2449 if (se->load_section_id == section_id) { 2450 break; 2451 } 2452 } 2453 if (se == NULL) { 2454 error_report("Unknown savevm section %d", section_id); 2455 return -EINVAL; 2456 } 2457 2458 ret = vmstate_load(f, se); 2459 if (ret < 0) { 2460 error_report("error while loading state section id %d(%s)", 2461 section_id, se->idstr); 2462 return ret; 2463 } 2464 if (!check_section_footer(f, se)) { 2465 return -EINVAL; 2466 } 2467 2468 return 0; 2469 } 2470 2471 static int qemu_loadvm_state_header(QEMUFile *f) 2472 { 2473 unsigned int v; 2474 int ret; 2475 2476 v = qemu_get_be32(f); 2477 if (v != QEMU_VM_FILE_MAGIC) { 2478 error_report("Not a migration stream"); 2479 return -EINVAL; 2480 } 2481 2482 v = qemu_get_be32(f); 2483 if (v == QEMU_VM_FILE_VERSION_COMPAT) { 2484 error_report("SaveVM v2 format is obsolete and don't work anymore"); 2485 return -ENOTSUP; 2486 } 2487 if (v != QEMU_VM_FILE_VERSION) { 2488 error_report("Unsupported migration stream version"); 2489 return -ENOTSUP; 2490 } 2491 2492 if (migrate_get_current()->send_configuration) { 2493 if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) { 2494 error_report("Configuration section missing"); 2495 qemu_loadvm_state_cleanup(); 2496 return -EINVAL; 2497 } 2498 ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0); 2499 2500 if (ret) { 2501 qemu_loadvm_state_cleanup(); 2502 return ret; 2503 } 2504 } 2505 return 0; 2506 } 2507 2508 static int qemu_loadvm_state_setup(QEMUFile *f) 2509 { 2510 SaveStateEntry *se; 2511 int ret; 2512 2513 trace_loadvm_state_setup(); 2514 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 2515 if (!se->ops || !se->ops->load_setup) { 2516 continue; 2517 } 2518 if (se->ops->is_active) { 2519 if (!se->ops->is_active(se->opaque)) { 2520 continue; 2521 } 2522 } 2523 2524 ret = se->ops->load_setup(f, se->opaque); 2525 if (ret < 0) { 2526 qemu_file_set_error(f, ret); 2527 error_report("Load state of device %s failed", se->idstr); 2528 return ret; 2529 } 2530 } 2531 return 0; 2532 } 2533 2534 void qemu_loadvm_state_cleanup(void) 2535 { 2536 SaveStateEntry *se; 2537 2538 trace_loadvm_state_cleanup(); 2539 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 2540 if (se->ops && se->ops->load_cleanup) { 2541 se->ops->load_cleanup(se->opaque); 2542 } 2543 } 2544 } 2545 2546 /* Return true if we should continue the migration, or false. */ 2547 static bool postcopy_pause_incoming(MigrationIncomingState *mis) 2548 { 2549 int i; 2550 2551 trace_postcopy_pause_incoming(); 2552 2553 assert(migrate_postcopy_ram()); 2554 2555 /* 2556 * Unregister yank with either from/to src would work, since ioc behind it 2557 * is the same 2558 */ 2559 migration_ioc_unregister_yank_from_file(mis->from_src_file); 2560 2561 assert(mis->from_src_file); 2562 qemu_file_shutdown(mis->from_src_file); 2563 qemu_fclose(mis->from_src_file); 2564 mis->from_src_file = NULL; 2565 2566 assert(mis->to_src_file); 2567 qemu_file_shutdown(mis->to_src_file); 2568 qemu_mutex_lock(&mis->rp_mutex); 2569 qemu_fclose(mis->to_src_file); 2570 mis->to_src_file = NULL; 2571 qemu_mutex_unlock(&mis->rp_mutex); 2572 2573 /* 2574 * NOTE: this must happen before reset the PostcopyTmpPages below, 2575 * otherwise it's racy to reset those fields when the fast load thread 2576 * can be accessing it in parallel. 2577 */ 2578 if (mis->postcopy_qemufile_dst) { 2579 qemu_file_shutdown(mis->postcopy_qemufile_dst); 2580 /* Take the mutex to make sure the fast ram load thread halted */ 2581 qemu_mutex_lock(&mis->postcopy_prio_thread_mutex); 2582 migration_ioc_unregister_yank_from_file(mis->postcopy_qemufile_dst); 2583 qemu_fclose(mis->postcopy_qemufile_dst); 2584 mis->postcopy_qemufile_dst = NULL; 2585 qemu_mutex_unlock(&mis->postcopy_prio_thread_mutex); 2586 } 2587 2588 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, 2589 MIGRATION_STATUS_POSTCOPY_PAUSED); 2590 2591 /* Notify the fault thread for the invalidated file handle */ 2592 postcopy_fault_thread_notify(mis); 2593 2594 /* 2595 * If network is interrupted, any temp page we received will be useless 2596 * because we didn't mark them as "received" in receivedmap. After a 2597 * proper recovery later (which will sync src dirty bitmap with receivedmap 2598 * on dest) these cached small pages will be resent again. 2599 */ 2600 for (i = 0; i < mis->postcopy_channels; i++) { 2601 postcopy_temp_page_reset(&mis->postcopy_tmp_pages[i]); 2602 } 2603 2604 error_report("Detected IO failure for postcopy. " 2605 "Migration paused."); 2606 2607 while (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { 2608 qemu_sem_wait(&mis->postcopy_pause_sem_dst); 2609 } 2610 2611 trace_postcopy_pause_incoming_continued(); 2612 2613 return true; 2614 } 2615 2616 int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis) 2617 { 2618 uint8_t section_type; 2619 int ret = 0; 2620 2621 retry: 2622 while (true) { 2623 section_type = qemu_get_byte(f); 2624 2625 ret = qemu_file_get_error_obj_any(f, mis->postcopy_qemufile_dst, NULL); 2626 if (ret) { 2627 break; 2628 } 2629 2630 trace_qemu_loadvm_state_section(section_type); 2631 switch (section_type) { 2632 case QEMU_VM_SECTION_START: 2633 case QEMU_VM_SECTION_FULL: 2634 ret = qemu_loadvm_section_start_full(f, mis); 2635 if (ret < 0) { 2636 goto out; 2637 } 2638 break; 2639 case QEMU_VM_SECTION_PART: 2640 case QEMU_VM_SECTION_END: 2641 ret = qemu_loadvm_section_part_end(f, mis); 2642 if (ret < 0) { 2643 goto out; 2644 } 2645 break; 2646 case QEMU_VM_COMMAND: 2647 ret = loadvm_process_command(f); 2648 trace_qemu_loadvm_state_section_command(ret); 2649 if ((ret < 0) || (ret == LOADVM_QUIT)) { 2650 goto out; 2651 } 2652 break; 2653 case QEMU_VM_EOF: 2654 /* This is the end of migration */ 2655 goto out; 2656 default: 2657 error_report("Unknown savevm section type %d", section_type); 2658 ret = -EINVAL; 2659 goto out; 2660 } 2661 } 2662 2663 out: 2664 if (ret < 0) { 2665 qemu_file_set_error(f, ret); 2666 2667 /* Cancel bitmaps incoming regardless of recovery */ 2668 dirty_bitmap_mig_cancel_incoming(); 2669 2670 /* 2671 * If we are during an active postcopy, then we pause instead 2672 * of bail out to at least keep the VM's dirty data. Note 2673 * that POSTCOPY_INCOMING_LISTENING stage is still not enough, 2674 * during which we're still receiving device states and we 2675 * still haven't yet started the VM on destination. 2676 * 2677 * Only RAM postcopy supports recovery. Still, if RAM postcopy is 2678 * enabled, canceled bitmaps postcopy will not affect RAM postcopy 2679 * recovering. 2680 */ 2681 if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING && 2682 migrate_postcopy_ram() && postcopy_pause_incoming(mis)) { 2683 /* Reset f to point to the newly created channel */ 2684 f = mis->from_src_file; 2685 goto retry; 2686 } 2687 } 2688 return ret; 2689 } 2690 2691 int qemu_loadvm_state(QEMUFile *f) 2692 { 2693 MigrationIncomingState *mis = migration_incoming_get_current(); 2694 Error *local_err = NULL; 2695 int ret; 2696 2697 if (qemu_savevm_state_blocked(&local_err)) { 2698 error_report_err(local_err); 2699 return -EINVAL; 2700 } 2701 2702 ret = qemu_loadvm_state_header(f); 2703 if (ret) { 2704 return ret; 2705 } 2706 2707 if (qemu_loadvm_state_setup(f) != 0) { 2708 return -EINVAL; 2709 } 2710 2711 cpu_synchronize_all_pre_loadvm(); 2712 2713 ret = qemu_loadvm_state_main(f, mis); 2714 qemu_event_set(&mis->main_thread_load_event); 2715 2716 trace_qemu_loadvm_state_post_main(ret); 2717 2718 if (mis->have_listen_thread) { 2719 /* Listen thread still going, can't clean up yet */ 2720 return ret; 2721 } 2722 2723 if (ret == 0) { 2724 ret = qemu_file_get_error(f); 2725 } 2726 2727 /* 2728 * Try to read in the VMDESC section as well, so that dumping tools that 2729 * intercept our migration stream have the chance to see it. 2730 */ 2731 2732 /* We've got to be careful; if we don't read the data and just shut the fd 2733 * then the sender can error if we close while it's still sending. 2734 * We also mustn't read data that isn't there; some transports (RDMA) 2735 * will stall waiting for that data when the source has already closed. 2736 */ 2737 if (ret == 0 && should_send_vmdesc()) { 2738 uint8_t *buf; 2739 uint32_t size; 2740 uint8_t section_type = qemu_get_byte(f); 2741 2742 if (section_type != QEMU_VM_VMDESCRIPTION) { 2743 error_report("Expected vmdescription section, but got %d", 2744 section_type); 2745 /* 2746 * It doesn't seem worth failing at this point since 2747 * we apparently have an otherwise valid VM state 2748 */ 2749 } else { 2750 buf = g_malloc(0x1000); 2751 size = qemu_get_be32(f); 2752 2753 while (size > 0) { 2754 uint32_t read_chunk = MIN(size, 0x1000); 2755 qemu_get_buffer(f, buf, read_chunk); 2756 size -= read_chunk; 2757 } 2758 g_free(buf); 2759 } 2760 } 2761 2762 qemu_loadvm_state_cleanup(); 2763 cpu_synchronize_all_post_init(); 2764 2765 return ret; 2766 } 2767 2768 int qemu_load_device_state(QEMUFile *f) 2769 { 2770 MigrationIncomingState *mis = migration_incoming_get_current(); 2771 int ret; 2772 2773 /* Load QEMU_VM_SECTION_FULL section */ 2774 ret = qemu_loadvm_state_main(f, mis); 2775 if (ret < 0) { 2776 error_report("Failed to load device state: %d", ret); 2777 return ret; 2778 } 2779 2780 cpu_synchronize_all_post_init(); 2781 return 0; 2782 } 2783 2784 bool save_snapshot(const char *name, bool overwrite, const char *vmstate, 2785 bool has_devices, strList *devices, Error **errp) 2786 { 2787 BlockDriverState *bs; 2788 QEMUSnapshotInfo sn1, *sn = &sn1; 2789 int ret = -1, ret2; 2790 QEMUFile *f; 2791 int saved_vm_running; 2792 uint64_t vm_state_size; 2793 g_autoptr(GDateTime) now = g_date_time_new_now_local(); 2794 AioContext *aio_context; 2795 2796 GLOBAL_STATE_CODE(); 2797 2798 if (migration_is_blocked(errp)) { 2799 return false; 2800 } 2801 2802 if (!replay_can_snapshot()) { 2803 error_setg(errp, "Record/replay does not allow making snapshot " 2804 "right now. Try once more later."); 2805 return false; 2806 } 2807 2808 if (!bdrv_all_can_snapshot(has_devices, devices, errp)) { 2809 return false; 2810 } 2811 2812 /* Delete old snapshots of the same name */ 2813 if (name) { 2814 if (overwrite) { 2815 if (bdrv_all_delete_snapshot(name, has_devices, 2816 devices, errp) < 0) { 2817 return false; 2818 } 2819 } else { 2820 ret2 = bdrv_all_has_snapshot(name, has_devices, devices, errp); 2821 if (ret2 < 0) { 2822 return false; 2823 } 2824 if (ret2 == 1) { 2825 error_setg(errp, 2826 "Snapshot '%s' already exists in one or more devices", 2827 name); 2828 return false; 2829 } 2830 } 2831 } 2832 2833 bs = bdrv_all_find_vmstate_bs(vmstate, has_devices, devices, errp); 2834 if (bs == NULL) { 2835 return false; 2836 } 2837 aio_context = bdrv_get_aio_context(bs); 2838 2839 saved_vm_running = runstate_is_running(); 2840 2841 ret = global_state_store(); 2842 if (ret) { 2843 error_setg(errp, "Error saving global state"); 2844 return false; 2845 } 2846 vm_stop(RUN_STATE_SAVE_VM); 2847 2848 bdrv_drain_all_begin(); 2849 2850 aio_context_acquire(aio_context); 2851 2852 memset(sn, 0, sizeof(*sn)); 2853 2854 /* fill auxiliary fields */ 2855 sn->date_sec = g_date_time_to_unix(now); 2856 sn->date_nsec = g_date_time_get_microsecond(now) * 1000; 2857 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL); 2858 if (replay_mode != REPLAY_MODE_NONE) { 2859 sn->icount = replay_get_current_icount(); 2860 } else { 2861 sn->icount = -1ULL; 2862 } 2863 2864 if (name) { 2865 pstrcpy(sn->name, sizeof(sn->name), name); 2866 } else { 2867 g_autofree char *autoname = g_date_time_format(now, "vm-%Y%m%d%H%M%S"); 2868 pstrcpy(sn->name, sizeof(sn->name), autoname); 2869 } 2870 2871 /* save the VM state */ 2872 f = qemu_fopen_bdrv(bs, 1); 2873 if (!f) { 2874 error_setg(errp, "Could not open VM state file"); 2875 goto the_end; 2876 } 2877 ret = qemu_savevm_state(f, errp); 2878 vm_state_size = qemu_file_total_transferred(f); 2879 ret2 = qemu_fclose(f); 2880 if (ret < 0) { 2881 goto the_end; 2882 } 2883 if (ret2 < 0) { 2884 ret = ret2; 2885 goto the_end; 2886 } 2887 2888 /* The bdrv_all_create_snapshot() call that follows acquires the AioContext 2889 * for itself. BDRV_POLL_WHILE() does not support nested locking because 2890 * it only releases the lock once. Therefore synchronous I/O will deadlock 2891 * unless we release the AioContext before bdrv_all_create_snapshot(). 2892 */ 2893 aio_context_release(aio_context); 2894 aio_context = NULL; 2895 2896 ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, 2897 has_devices, devices, errp); 2898 if (ret < 0) { 2899 bdrv_all_delete_snapshot(sn->name, has_devices, devices, NULL); 2900 goto the_end; 2901 } 2902 2903 ret = 0; 2904 2905 the_end: 2906 if (aio_context) { 2907 aio_context_release(aio_context); 2908 } 2909 2910 bdrv_drain_all_end(); 2911 2912 if (saved_vm_running) { 2913 vm_start(); 2914 } 2915 return ret == 0; 2916 } 2917 2918 void qmp_xen_save_devices_state(const char *filename, bool has_live, bool live, 2919 Error **errp) 2920 { 2921 QEMUFile *f; 2922 QIOChannelFile *ioc; 2923 int saved_vm_running; 2924 int ret; 2925 2926 if (!has_live) { 2927 /* live default to true so old version of Xen tool stack can have a 2928 * successful live migration */ 2929 live = true; 2930 } 2931 2932 saved_vm_running = runstate_is_running(); 2933 vm_stop(RUN_STATE_SAVE_VM); 2934 global_state_store_running(); 2935 2936 ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT | O_TRUNC, 2937 0660, errp); 2938 if (!ioc) { 2939 goto the_end; 2940 } 2941 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state"); 2942 f = qemu_file_new_output(QIO_CHANNEL(ioc)); 2943 object_unref(OBJECT(ioc)); 2944 ret = qemu_save_device_state(f); 2945 if (ret < 0 || qemu_fclose(f) < 0) { 2946 error_setg(errp, QERR_IO_ERROR); 2947 } else { 2948 /* libxl calls the QMP command "stop" before calling 2949 * "xen-save-devices-state" and in case of migration failure, libxl 2950 * would call "cont". 2951 * So call bdrv_inactivate_all (release locks) here to let the other 2952 * side of the migration take control of the images. 2953 */ 2954 if (live && !saved_vm_running) { 2955 ret = bdrv_inactivate_all(); 2956 if (ret) { 2957 error_setg(errp, "%s: bdrv_inactivate_all() failed (%d)", 2958 __func__, ret); 2959 } 2960 } 2961 } 2962 2963 the_end: 2964 if (saved_vm_running) { 2965 vm_start(); 2966 } 2967 } 2968 2969 void qmp_xen_load_devices_state(const char *filename, Error **errp) 2970 { 2971 QEMUFile *f; 2972 QIOChannelFile *ioc; 2973 int ret; 2974 2975 /* Guest must be paused before loading the device state; the RAM state 2976 * will already have been loaded by xc 2977 */ 2978 if (runstate_is_running()) { 2979 error_setg(errp, "Cannot update device state while vm is running"); 2980 return; 2981 } 2982 vm_stop(RUN_STATE_RESTORE_VM); 2983 2984 ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp); 2985 if (!ioc) { 2986 return; 2987 } 2988 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state"); 2989 f = qemu_file_new_input(QIO_CHANNEL(ioc)); 2990 object_unref(OBJECT(ioc)); 2991 2992 ret = qemu_loadvm_state(f); 2993 qemu_fclose(f); 2994 if (ret < 0) { 2995 error_setg(errp, QERR_IO_ERROR); 2996 } 2997 migration_incoming_state_destroy(); 2998 } 2999 3000 bool load_snapshot(const char *name, const char *vmstate, 3001 bool has_devices, strList *devices, Error **errp) 3002 { 3003 BlockDriverState *bs_vm_state; 3004 QEMUSnapshotInfo sn; 3005 QEMUFile *f; 3006 int ret; 3007 AioContext *aio_context; 3008 MigrationIncomingState *mis = migration_incoming_get_current(); 3009 3010 if (!bdrv_all_can_snapshot(has_devices, devices, errp)) { 3011 return false; 3012 } 3013 ret = bdrv_all_has_snapshot(name, has_devices, devices, errp); 3014 if (ret < 0) { 3015 return false; 3016 } 3017 if (ret == 0) { 3018 error_setg(errp, "Snapshot '%s' does not exist in one or more devices", 3019 name); 3020 return false; 3021 } 3022 3023 bs_vm_state = bdrv_all_find_vmstate_bs(vmstate, has_devices, devices, errp); 3024 if (!bs_vm_state) { 3025 return false; 3026 } 3027 aio_context = bdrv_get_aio_context(bs_vm_state); 3028 3029 /* Don't even try to load empty VM states */ 3030 aio_context_acquire(aio_context); 3031 ret = bdrv_snapshot_find(bs_vm_state, &sn, name); 3032 aio_context_release(aio_context); 3033 if (ret < 0) { 3034 return false; 3035 } else if (sn.vm_state_size == 0) { 3036 error_setg(errp, "This is a disk-only snapshot. Revert to it " 3037 " offline using qemu-img"); 3038 return false; 3039 } 3040 3041 /* 3042 * Flush the record/replay queue. Now the VM state is going 3043 * to change. Therefore we don't need to preserve its consistency 3044 */ 3045 replay_flush_events(); 3046 3047 /* Flush all IO requests so they don't interfere with the new state. */ 3048 bdrv_drain_all_begin(); 3049 3050 ret = bdrv_all_goto_snapshot(name, has_devices, devices, errp); 3051 if (ret < 0) { 3052 goto err_drain; 3053 } 3054 3055 /* restore the VM state */ 3056 f = qemu_fopen_bdrv(bs_vm_state, 0); 3057 if (!f) { 3058 error_setg(errp, "Could not open VM state file"); 3059 goto err_drain; 3060 } 3061 3062 qemu_system_reset(SHUTDOWN_CAUSE_SNAPSHOT_LOAD); 3063 mis->from_src_file = f; 3064 3065 if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) { 3066 ret = -EINVAL; 3067 goto err_drain; 3068 } 3069 aio_context_acquire(aio_context); 3070 ret = qemu_loadvm_state(f); 3071 migration_incoming_state_destroy(); 3072 aio_context_release(aio_context); 3073 3074 bdrv_drain_all_end(); 3075 3076 if (ret < 0) { 3077 error_setg(errp, "Error %d while loading VM state", ret); 3078 return false; 3079 } 3080 3081 return true; 3082 3083 err_drain: 3084 bdrv_drain_all_end(); 3085 return false; 3086 } 3087 3088 bool delete_snapshot(const char *name, bool has_devices, 3089 strList *devices, Error **errp) 3090 { 3091 if (!bdrv_all_can_snapshot(has_devices, devices, errp)) { 3092 return false; 3093 } 3094 3095 if (bdrv_all_delete_snapshot(name, has_devices, devices, errp) < 0) { 3096 return false; 3097 } 3098 3099 return true; 3100 } 3101 3102 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev) 3103 { 3104 qemu_ram_set_idstr(mr->ram_block, 3105 memory_region_name(mr), dev); 3106 qemu_ram_set_migratable(mr->ram_block); 3107 } 3108 3109 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev) 3110 { 3111 qemu_ram_unset_idstr(mr->ram_block); 3112 qemu_ram_unset_migratable(mr->ram_block); 3113 } 3114 3115 void vmstate_register_ram_global(MemoryRegion *mr) 3116 { 3117 vmstate_register_ram(mr, NULL); 3118 } 3119 3120 bool vmstate_check_only_migratable(const VMStateDescription *vmsd) 3121 { 3122 /* check needed if --only-migratable is specified */ 3123 if (!only_migratable) { 3124 return true; 3125 } 3126 3127 return !(vmsd && vmsd->unmigratable); 3128 } 3129 3130 typedef struct SnapshotJob { 3131 Job common; 3132 char *tag; 3133 char *vmstate; 3134 strList *devices; 3135 Coroutine *co; 3136 Error **errp; 3137 bool ret; 3138 } SnapshotJob; 3139 3140 static void qmp_snapshot_job_free(SnapshotJob *s) 3141 { 3142 g_free(s->tag); 3143 g_free(s->vmstate); 3144 qapi_free_strList(s->devices); 3145 } 3146 3147 3148 static void snapshot_load_job_bh(void *opaque) 3149 { 3150 Job *job = opaque; 3151 SnapshotJob *s = container_of(job, SnapshotJob, common); 3152 int orig_vm_running; 3153 3154 job_progress_set_remaining(&s->common, 1); 3155 3156 orig_vm_running = runstate_is_running(); 3157 vm_stop(RUN_STATE_RESTORE_VM); 3158 3159 s->ret = load_snapshot(s->tag, s->vmstate, true, s->devices, s->errp); 3160 if (s->ret && orig_vm_running) { 3161 vm_start(); 3162 } 3163 3164 job_progress_update(&s->common, 1); 3165 3166 qmp_snapshot_job_free(s); 3167 aio_co_wake(s->co); 3168 } 3169 3170 static void snapshot_save_job_bh(void *opaque) 3171 { 3172 Job *job = opaque; 3173 SnapshotJob *s = container_of(job, SnapshotJob, common); 3174 3175 job_progress_set_remaining(&s->common, 1); 3176 s->ret = save_snapshot(s->tag, false, s->vmstate, 3177 true, s->devices, s->errp); 3178 job_progress_update(&s->common, 1); 3179 3180 qmp_snapshot_job_free(s); 3181 aio_co_wake(s->co); 3182 } 3183 3184 static void snapshot_delete_job_bh(void *opaque) 3185 { 3186 Job *job = opaque; 3187 SnapshotJob *s = container_of(job, SnapshotJob, common); 3188 3189 job_progress_set_remaining(&s->common, 1); 3190 s->ret = delete_snapshot(s->tag, true, s->devices, s->errp); 3191 job_progress_update(&s->common, 1); 3192 3193 qmp_snapshot_job_free(s); 3194 aio_co_wake(s->co); 3195 } 3196 3197 static int coroutine_fn snapshot_save_job_run(Job *job, Error **errp) 3198 { 3199 SnapshotJob *s = container_of(job, SnapshotJob, common); 3200 s->errp = errp; 3201 s->co = qemu_coroutine_self(); 3202 aio_bh_schedule_oneshot(qemu_get_aio_context(), 3203 snapshot_save_job_bh, job); 3204 qemu_coroutine_yield(); 3205 return s->ret ? 0 : -1; 3206 } 3207 3208 static int coroutine_fn snapshot_load_job_run(Job *job, Error **errp) 3209 { 3210 SnapshotJob *s = container_of(job, SnapshotJob, common); 3211 s->errp = errp; 3212 s->co = qemu_coroutine_self(); 3213 aio_bh_schedule_oneshot(qemu_get_aio_context(), 3214 snapshot_load_job_bh, job); 3215 qemu_coroutine_yield(); 3216 return s->ret ? 0 : -1; 3217 } 3218 3219 static int coroutine_fn snapshot_delete_job_run(Job *job, Error **errp) 3220 { 3221 SnapshotJob *s = container_of(job, SnapshotJob, common); 3222 s->errp = errp; 3223 s->co = qemu_coroutine_self(); 3224 aio_bh_schedule_oneshot(qemu_get_aio_context(), 3225 snapshot_delete_job_bh, job); 3226 qemu_coroutine_yield(); 3227 return s->ret ? 0 : -1; 3228 } 3229 3230 3231 static const JobDriver snapshot_load_job_driver = { 3232 .instance_size = sizeof(SnapshotJob), 3233 .job_type = JOB_TYPE_SNAPSHOT_LOAD, 3234 .run = snapshot_load_job_run, 3235 }; 3236 3237 static const JobDriver snapshot_save_job_driver = { 3238 .instance_size = sizeof(SnapshotJob), 3239 .job_type = JOB_TYPE_SNAPSHOT_SAVE, 3240 .run = snapshot_save_job_run, 3241 }; 3242 3243 static const JobDriver snapshot_delete_job_driver = { 3244 .instance_size = sizeof(SnapshotJob), 3245 .job_type = JOB_TYPE_SNAPSHOT_DELETE, 3246 .run = snapshot_delete_job_run, 3247 }; 3248 3249 3250 void qmp_snapshot_save(const char *job_id, 3251 const char *tag, 3252 const char *vmstate, 3253 strList *devices, 3254 Error **errp) 3255 { 3256 SnapshotJob *s; 3257 3258 s = job_create(job_id, &snapshot_save_job_driver, NULL, 3259 qemu_get_aio_context(), JOB_MANUAL_DISMISS, 3260 NULL, NULL, errp); 3261 if (!s) { 3262 return; 3263 } 3264 3265 s->tag = g_strdup(tag); 3266 s->vmstate = g_strdup(vmstate); 3267 s->devices = QAPI_CLONE(strList, devices); 3268 3269 job_start(&s->common); 3270 } 3271 3272 void qmp_snapshot_load(const char *job_id, 3273 const char *tag, 3274 const char *vmstate, 3275 strList *devices, 3276 Error **errp) 3277 { 3278 SnapshotJob *s; 3279 3280 s = job_create(job_id, &snapshot_load_job_driver, NULL, 3281 qemu_get_aio_context(), JOB_MANUAL_DISMISS, 3282 NULL, NULL, errp); 3283 if (!s) { 3284 return; 3285 } 3286 3287 s->tag = g_strdup(tag); 3288 s->vmstate = g_strdup(vmstate); 3289 s->devices = QAPI_CLONE(strList, devices); 3290 3291 job_start(&s->common); 3292 } 3293 3294 void qmp_snapshot_delete(const char *job_id, 3295 const char *tag, 3296 strList *devices, 3297 Error **errp) 3298 { 3299 SnapshotJob *s; 3300 3301 s = job_create(job_id, &snapshot_delete_job_driver, NULL, 3302 qemu_get_aio_context(), JOB_MANUAL_DISMISS, 3303 NULL, NULL, errp); 3304 if (!s) { 3305 return; 3306 } 3307 3308 s->tag = g_strdup(tag); 3309 s->devices = QAPI_CLONE(strList, devices); 3310 3311 job_start(&s->common); 3312 } 3313