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-stats.h" 35 #include "migration/vmstate.h" 36 #include "migration/misc.h" 37 #include "migration/register.h" 38 #include "migration/global_state.h" 39 #include "migration/channel-block.h" 40 #include "ram.h" 41 #include "qemu-file.h" 42 #include "savevm.h" 43 #include "postcopy-ram.h" 44 #include "qapi/error.h" 45 #include "qapi/qapi-commands-migration.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 #include "sysemu/qtest.h" 71 #include "options.h" 72 73 const unsigned int postcopy_ram_discard_version; 74 75 /* Subcommands for QEMU_VM_COMMAND */ 76 enum qemu_vm_cmd { 77 MIG_CMD_INVALID = 0, /* Must be 0 */ 78 MIG_CMD_OPEN_RETURN_PATH, /* Tell the dest to open the Return path */ 79 MIG_CMD_PING, /* Request a PONG on the RP */ 80 81 MIG_CMD_POSTCOPY_ADVISE, /* Prior to any page transfers, just 82 warn we might want to do PC */ 83 MIG_CMD_POSTCOPY_LISTEN, /* Start listening for incoming 84 pages as it's running. */ 85 MIG_CMD_POSTCOPY_RUN, /* Start execution */ 86 87 MIG_CMD_POSTCOPY_RAM_DISCARD, /* A list of pages to discard that 88 were previously sent during 89 precopy but are dirty. */ 90 MIG_CMD_PACKAGED, /* Send a wrapped stream within this stream */ 91 MIG_CMD_ENABLE_COLO, /* Enable COLO */ 92 MIG_CMD_POSTCOPY_RESUME, /* resume postcopy on dest */ 93 MIG_CMD_RECV_BITMAP, /* Request for recved bitmap on dst */ 94 MIG_CMD_MAX 95 }; 96 97 #define MAX_VM_CMD_PACKAGED_SIZE UINT32_MAX 98 static struct mig_cmd_args { 99 ssize_t len; /* -1 = variable */ 100 const char *name; 101 } mig_cmd_args[] = { 102 [MIG_CMD_INVALID] = { .len = -1, .name = "INVALID" }, 103 [MIG_CMD_OPEN_RETURN_PATH] = { .len = 0, .name = "OPEN_RETURN_PATH" }, 104 [MIG_CMD_PING] = { .len = sizeof(uint32_t), .name = "PING" }, 105 [MIG_CMD_POSTCOPY_ADVISE] = { .len = -1, .name = "POSTCOPY_ADVISE" }, 106 [MIG_CMD_POSTCOPY_LISTEN] = { .len = 0, .name = "POSTCOPY_LISTEN" }, 107 [MIG_CMD_POSTCOPY_RUN] = { .len = 0, .name = "POSTCOPY_RUN" }, 108 [MIG_CMD_POSTCOPY_RAM_DISCARD] = { 109 .len = -1, .name = "POSTCOPY_RAM_DISCARD" }, 110 [MIG_CMD_POSTCOPY_RESUME] = { .len = 0, .name = "POSTCOPY_RESUME" }, 111 [MIG_CMD_PACKAGED] = { .len = 4, .name = "PACKAGED" }, 112 [MIG_CMD_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" }, 113 [MIG_CMD_MAX] = { .len = -1, .name = "MAX" }, 114 }; 115 116 /* Note for MIG_CMD_POSTCOPY_ADVISE: 117 * The format of arguments is depending on postcopy mode: 118 * - postcopy RAM only 119 * uint64_t host page size 120 * uint64_t target page size 121 * 122 * - postcopy RAM and postcopy dirty bitmaps 123 * format is the same as for postcopy RAM only 124 * 125 * - postcopy dirty bitmaps only 126 * Nothing. Command length field is 0. 127 * 128 * Be careful: adding a new postcopy entity with some other parameters should 129 * not break format self-description ability. Good way is to introduce some 130 * generic extendable format with an exception for two old entities. 131 */ 132 133 /***********************************************************/ 134 /* savevm/loadvm support */ 135 136 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable) 137 { 138 if (is_writable) { 139 return qemu_file_new_output(QIO_CHANNEL(qio_channel_block_new(bs))); 140 } else { 141 return qemu_file_new_input(QIO_CHANNEL(qio_channel_block_new(bs))); 142 } 143 } 144 145 146 /* QEMUFile timer support. 147 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c 148 */ 149 150 void timer_put(QEMUFile *f, QEMUTimer *ts) 151 { 152 uint64_t expire_time; 153 154 expire_time = timer_expire_time_ns(ts); 155 qemu_put_be64(f, expire_time); 156 } 157 158 void timer_get(QEMUFile *f, QEMUTimer *ts) 159 { 160 uint64_t expire_time; 161 162 expire_time = qemu_get_be64(f); 163 if (expire_time != -1) { 164 timer_mod_ns(ts, expire_time); 165 } else { 166 timer_del(ts); 167 } 168 } 169 170 171 /* VMState timer support. 172 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c 173 */ 174 175 static int get_timer(QEMUFile *f, void *pv, size_t size, 176 const VMStateField *field) 177 { 178 QEMUTimer *v = pv; 179 timer_get(f, v); 180 return 0; 181 } 182 183 static int put_timer(QEMUFile *f, void *pv, size_t size, 184 const VMStateField *field, JSONWriter *vmdesc) 185 { 186 QEMUTimer *v = pv; 187 timer_put(f, v); 188 189 return 0; 190 } 191 192 const VMStateInfo vmstate_info_timer = { 193 .name = "timer", 194 .get = get_timer, 195 .put = put_timer, 196 }; 197 198 199 typedef struct CompatEntry { 200 char idstr[256]; 201 int instance_id; 202 } CompatEntry; 203 204 typedef struct SaveStateEntry { 205 QTAILQ_ENTRY(SaveStateEntry) entry; 206 char idstr[256]; 207 uint32_t instance_id; 208 int alias_id; 209 int version_id; 210 /* version id read from the stream */ 211 int load_version_id; 212 int section_id; 213 /* section id read from the stream */ 214 int load_section_id; 215 const SaveVMHandlers *ops; 216 const VMStateDescription *vmsd; 217 void *opaque; 218 CompatEntry *compat; 219 int is_ram; 220 } SaveStateEntry; 221 222 typedef struct SaveState { 223 QTAILQ_HEAD(, SaveStateEntry) handlers; 224 SaveStateEntry *handler_pri_head[MIG_PRI_MAX + 1]; 225 int global_section_id; 226 uint32_t len; 227 const char *name; 228 uint32_t target_page_bits; 229 uint32_t caps_count; 230 MigrationCapability *capabilities; 231 QemuUUID uuid; 232 } SaveState; 233 234 static SaveState savevm_state = { 235 .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers), 236 .handler_pri_head = { [MIG_PRI_DEFAULT ... MIG_PRI_MAX] = NULL }, 237 .global_section_id = 0, 238 }; 239 240 static SaveStateEntry *find_se(const char *idstr, uint32_t instance_id); 241 242 static bool should_validate_capability(int capability) 243 { 244 assert(capability >= 0 && capability < MIGRATION_CAPABILITY__MAX); 245 /* Validate only new capabilities to keep compatibility. */ 246 switch (capability) { 247 case MIGRATION_CAPABILITY_X_IGNORE_SHARED: 248 case MIGRATION_CAPABILITY_MAPPED_RAM: 249 return true; 250 default: 251 return false; 252 } 253 } 254 255 static uint32_t get_validatable_capabilities_count(void) 256 { 257 MigrationState *s = migrate_get_current(); 258 uint32_t result = 0; 259 int i; 260 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { 261 if (should_validate_capability(i) && s->capabilities[i]) { 262 result++; 263 } 264 } 265 return result; 266 } 267 268 static int configuration_pre_save(void *opaque) 269 { 270 SaveState *state = opaque; 271 const char *current_name = MACHINE_GET_CLASS(current_machine)->name; 272 MigrationState *s = migrate_get_current(); 273 int i, j; 274 275 state->len = strlen(current_name); 276 state->name = current_name; 277 state->target_page_bits = qemu_target_page_bits(); 278 279 state->caps_count = get_validatable_capabilities_count(); 280 state->capabilities = g_renew(MigrationCapability, state->capabilities, 281 state->caps_count); 282 for (i = j = 0; i < MIGRATION_CAPABILITY__MAX; i++) { 283 if (should_validate_capability(i) && s->capabilities[i]) { 284 state->capabilities[j++] = i; 285 } 286 } 287 state->uuid = qemu_uuid; 288 289 return 0; 290 } 291 292 static int configuration_post_save(void *opaque) 293 { 294 SaveState *state = opaque; 295 296 g_free(state->capabilities); 297 state->capabilities = NULL; 298 state->caps_count = 0; 299 return 0; 300 } 301 302 static int configuration_pre_load(void *opaque) 303 { 304 SaveState *state = opaque; 305 306 /* If there is no target-page-bits subsection it means the source 307 * predates the variable-target-page-bits support and is using the 308 * minimum possible value for this CPU. 309 */ 310 state->target_page_bits = qemu_target_page_bits_min(); 311 return 0; 312 } 313 314 static bool configuration_validate_capabilities(SaveState *state) 315 { 316 bool ret = true; 317 MigrationState *s = migrate_get_current(); 318 unsigned long *source_caps_bm; 319 int i; 320 321 source_caps_bm = bitmap_new(MIGRATION_CAPABILITY__MAX); 322 for (i = 0; i < state->caps_count; i++) { 323 MigrationCapability capability = state->capabilities[i]; 324 set_bit(capability, source_caps_bm); 325 } 326 327 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { 328 bool source_state, target_state; 329 if (!should_validate_capability(i)) { 330 continue; 331 } 332 source_state = test_bit(i, source_caps_bm); 333 target_state = s->capabilities[i]; 334 if (source_state != target_state) { 335 error_report("Capability %s is %s, but received capability is %s", 336 MigrationCapability_str(i), 337 target_state ? "on" : "off", 338 source_state ? "on" : "off"); 339 ret = false; 340 /* Don't break here to report all failed capabilities */ 341 } 342 } 343 344 g_free(source_caps_bm); 345 return ret; 346 } 347 348 static int configuration_post_load(void *opaque, int version_id) 349 { 350 SaveState *state = opaque; 351 const char *current_name = MACHINE_GET_CLASS(current_machine)->name; 352 int ret = 0; 353 354 if (strncmp(state->name, current_name, state->len) != 0) { 355 error_report("Machine type received is '%.*s' and local is '%s'", 356 (int) state->len, state->name, current_name); 357 ret = -EINVAL; 358 goto out; 359 } 360 361 if (state->target_page_bits != qemu_target_page_bits()) { 362 error_report("Received TARGET_PAGE_BITS is %d but local is %d", 363 state->target_page_bits, qemu_target_page_bits()); 364 ret = -EINVAL; 365 goto out; 366 } 367 368 if (!configuration_validate_capabilities(state)) { 369 ret = -EINVAL; 370 goto out; 371 } 372 373 out: 374 g_free((void *)state->name); 375 state->name = NULL; 376 state->len = 0; 377 g_free(state->capabilities); 378 state->capabilities = NULL; 379 state->caps_count = 0; 380 381 return ret; 382 } 383 384 static int get_capability(QEMUFile *f, void *pv, size_t size, 385 const VMStateField *field) 386 { 387 MigrationCapability *capability = pv; 388 char capability_str[UINT8_MAX + 1]; 389 uint8_t len; 390 int i; 391 392 len = qemu_get_byte(f); 393 qemu_get_buffer(f, (uint8_t *)capability_str, len); 394 capability_str[len] = '\0'; 395 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { 396 if (!strcmp(MigrationCapability_str(i), capability_str)) { 397 *capability = i; 398 return 0; 399 } 400 } 401 error_report("Received unknown capability %s", capability_str); 402 return -EINVAL; 403 } 404 405 static int put_capability(QEMUFile *f, void *pv, size_t size, 406 const VMStateField *field, JSONWriter *vmdesc) 407 { 408 MigrationCapability *capability = pv; 409 const char *capability_str = MigrationCapability_str(*capability); 410 size_t len = strlen(capability_str); 411 assert(len <= UINT8_MAX); 412 413 qemu_put_byte(f, len); 414 qemu_put_buffer(f, (uint8_t *)capability_str, len); 415 return 0; 416 } 417 418 static const VMStateInfo vmstate_info_capability = { 419 .name = "capability", 420 .get = get_capability, 421 .put = put_capability, 422 }; 423 424 /* The target-page-bits subsection is present only if the 425 * target page size is not the same as the default (ie the 426 * minimum page size for a variable-page-size guest CPU). 427 * If it is present then it contains the actual target page 428 * bits for the machine, and migration will fail if the 429 * two ends don't agree about it. 430 */ 431 static bool vmstate_target_page_bits_needed(void *opaque) 432 { 433 return qemu_target_page_bits() 434 > qemu_target_page_bits_min(); 435 } 436 437 static const VMStateDescription vmstate_target_page_bits = { 438 .name = "configuration/target-page-bits", 439 .version_id = 1, 440 .minimum_version_id = 1, 441 .needed = vmstate_target_page_bits_needed, 442 .fields = (const VMStateField[]) { 443 VMSTATE_UINT32(target_page_bits, SaveState), 444 VMSTATE_END_OF_LIST() 445 } 446 }; 447 448 static bool vmstate_capabilites_needed(void *opaque) 449 { 450 return get_validatable_capabilities_count() > 0; 451 } 452 453 static const VMStateDescription vmstate_capabilites = { 454 .name = "configuration/capabilities", 455 .version_id = 1, 456 .minimum_version_id = 1, 457 .needed = vmstate_capabilites_needed, 458 .fields = (const VMStateField[]) { 459 VMSTATE_UINT32_V(caps_count, SaveState, 1), 460 VMSTATE_VARRAY_UINT32_ALLOC(capabilities, SaveState, caps_count, 1, 461 vmstate_info_capability, 462 MigrationCapability), 463 VMSTATE_END_OF_LIST() 464 } 465 }; 466 467 static bool vmstate_uuid_needed(void *opaque) 468 { 469 return qemu_uuid_set && migrate_validate_uuid(); 470 } 471 472 static int vmstate_uuid_post_load(void *opaque, int version_id) 473 { 474 SaveState *state = opaque; 475 char uuid_src[UUID_STR_LEN]; 476 char uuid_dst[UUID_STR_LEN]; 477 478 if (!qemu_uuid_set) { 479 /* 480 * It's warning because user might not know UUID in some cases, 481 * e.g. load an old snapshot 482 */ 483 qemu_uuid_unparse(&state->uuid, uuid_src); 484 warn_report("UUID is received %s, but local uuid isn't set", 485 uuid_src); 486 return 0; 487 } 488 if (!qemu_uuid_is_equal(&state->uuid, &qemu_uuid)) { 489 qemu_uuid_unparse(&state->uuid, uuid_src); 490 qemu_uuid_unparse(&qemu_uuid, uuid_dst); 491 error_report("UUID received is %s and local is %s", uuid_src, uuid_dst); 492 return -EINVAL; 493 } 494 return 0; 495 } 496 497 static const VMStateDescription vmstate_uuid = { 498 .name = "configuration/uuid", 499 .version_id = 1, 500 .minimum_version_id = 1, 501 .needed = vmstate_uuid_needed, 502 .post_load = vmstate_uuid_post_load, 503 .fields = (const VMStateField[]) { 504 VMSTATE_UINT8_ARRAY_V(uuid.data, SaveState, sizeof(QemuUUID), 1), 505 VMSTATE_END_OF_LIST() 506 } 507 }; 508 509 static const VMStateDescription vmstate_configuration = { 510 .name = "configuration", 511 .version_id = 1, 512 .pre_load = configuration_pre_load, 513 .post_load = configuration_post_load, 514 .pre_save = configuration_pre_save, 515 .post_save = configuration_post_save, 516 .fields = (const VMStateField[]) { 517 VMSTATE_UINT32(len, SaveState), 518 VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len), 519 VMSTATE_END_OF_LIST() 520 }, 521 .subsections = (const VMStateDescription * const []) { 522 &vmstate_target_page_bits, 523 &vmstate_capabilites, 524 &vmstate_uuid, 525 NULL 526 } 527 }; 528 529 static void dump_vmstate_vmsd(FILE *out_file, 530 const VMStateDescription *vmsd, int indent, 531 bool is_subsection); 532 533 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field, 534 int indent) 535 { 536 fprintf(out_file, "%*s{\n", indent, ""); 537 indent += 2; 538 fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name); 539 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "", 540 field->version_id); 541 fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "", 542 field->field_exists ? "true" : "false"); 543 if (field->flags & VMS_ARRAY) { 544 fprintf(out_file, "%*s\"num\": %d,\n", indent, "", field->num); 545 } 546 fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size); 547 if (field->vmsd != NULL) { 548 fprintf(out_file, ",\n"); 549 dump_vmstate_vmsd(out_file, field->vmsd, indent, false); 550 } 551 fprintf(out_file, "\n%*s}", indent - 2, ""); 552 } 553 554 static void dump_vmstate_vmss(FILE *out_file, 555 const VMStateDescription *subsection, 556 int indent) 557 { 558 if (subsection != NULL) { 559 dump_vmstate_vmsd(out_file, subsection, indent, true); 560 } 561 } 562 563 static void dump_vmstate_vmsd(FILE *out_file, 564 const VMStateDescription *vmsd, int indent, 565 bool is_subsection) 566 { 567 if (is_subsection) { 568 fprintf(out_file, "%*s{\n", indent, ""); 569 } else { 570 fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description"); 571 } 572 indent += 2; 573 fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name); 574 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "", 575 vmsd->version_id); 576 fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "", 577 vmsd->minimum_version_id); 578 if (vmsd->fields != NULL) { 579 const VMStateField *field = vmsd->fields; 580 bool first; 581 582 fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, ""); 583 first = true; 584 while (field->name != NULL) { 585 if (field->flags & VMS_MUST_EXIST) { 586 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */ 587 field++; 588 continue; 589 } 590 if (!first) { 591 fprintf(out_file, ",\n"); 592 } 593 dump_vmstate_vmsf(out_file, field, indent + 2); 594 field++; 595 first = false; 596 } 597 assert(field->flags == VMS_END); 598 fprintf(out_file, "\n%*s]", indent, ""); 599 } 600 if (vmsd->subsections != NULL) { 601 const VMStateDescription * const *subsection = vmsd->subsections; 602 bool first; 603 604 fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, ""); 605 first = true; 606 while (*subsection != NULL) { 607 if (!first) { 608 fprintf(out_file, ",\n"); 609 } 610 dump_vmstate_vmss(out_file, *subsection, indent + 2); 611 subsection++; 612 first = false; 613 } 614 fprintf(out_file, "\n%*s]", indent, ""); 615 } 616 fprintf(out_file, "\n%*s}", indent - 2, ""); 617 } 618 619 static void dump_machine_type(FILE *out_file) 620 { 621 MachineClass *mc; 622 623 mc = MACHINE_GET_CLASS(current_machine); 624 625 fprintf(out_file, " \"vmschkmachine\": {\n"); 626 fprintf(out_file, " \"Name\": \"%s\"\n", mc->name); 627 fprintf(out_file, " },\n"); 628 } 629 630 void dump_vmstate_json_to_file(FILE *out_file) 631 { 632 GSList *list, *elt; 633 bool first; 634 635 fprintf(out_file, "{\n"); 636 dump_machine_type(out_file); 637 638 first = true; 639 list = object_class_get_list(TYPE_DEVICE, true); 640 for (elt = list; elt; elt = elt->next) { 641 DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data, 642 TYPE_DEVICE); 643 const char *name; 644 int indent = 2; 645 646 if (!dc->vmsd) { 647 continue; 648 } 649 650 if (!first) { 651 fprintf(out_file, ",\n"); 652 } 653 name = object_class_get_name(OBJECT_CLASS(dc)); 654 fprintf(out_file, "%*s\"%s\": {\n", indent, "", name); 655 indent += 2; 656 fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name); 657 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "", 658 dc->vmsd->version_id); 659 fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "", 660 dc->vmsd->minimum_version_id); 661 662 dump_vmstate_vmsd(out_file, dc->vmsd, indent, false); 663 664 fprintf(out_file, "\n%*s}", indent - 2, ""); 665 first = false; 666 } 667 fprintf(out_file, "\n}\n"); 668 fclose(out_file); 669 g_slist_free(list); 670 } 671 672 static uint32_t calculate_new_instance_id(const char *idstr) 673 { 674 SaveStateEntry *se; 675 uint32_t instance_id = 0; 676 677 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 678 if (strcmp(idstr, se->idstr) == 0 679 && instance_id <= se->instance_id) { 680 instance_id = se->instance_id + 1; 681 } 682 } 683 /* Make sure we never loop over without being noticed */ 684 assert(instance_id != VMSTATE_INSTANCE_ID_ANY); 685 return instance_id; 686 } 687 688 static int calculate_compat_instance_id(const char *idstr) 689 { 690 SaveStateEntry *se; 691 int instance_id = 0; 692 693 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 694 if (!se->compat) { 695 continue; 696 } 697 698 if (strcmp(idstr, se->compat->idstr) == 0 699 && instance_id <= se->compat->instance_id) { 700 instance_id = se->compat->instance_id + 1; 701 } 702 } 703 return instance_id; 704 } 705 706 static inline MigrationPriority save_state_priority(SaveStateEntry *se) 707 { 708 if (se->vmsd) { 709 return se->vmsd->priority; 710 } 711 return MIG_PRI_DEFAULT; 712 } 713 714 static void savevm_state_handler_insert(SaveStateEntry *nse) 715 { 716 MigrationPriority priority = save_state_priority(nse); 717 SaveStateEntry *se; 718 int i; 719 720 assert(priority <= MIG_PRI_MAX); 721 722 /* 723 * This should never happen otherwise migration will probably fail 724 * silently somewhere because we can be wrongly applying one 725 * object properties upon another one. Bail out ASAP. 726 */ 727 if (find_se(nse->idstr, nse->instance_id)) { 728 error_report("%s: Detected duplicate SaveStateEntry: " 729 "id=%s, instance_id=0x%"PRIx32, __func__, 730 nse->idstr, nse->instance_id); 731 exit(EXIT_FAILURE); 732 } 733 734 for (i = priority - 1; i >= 0; i--) { 735 se = savevm_state.handler_pri_head[i]; 736 if (se != NULL) { 737 assert(save_state_priority(se) < priority); 738 break; 739 } 740 } 741 742 if (i >= 0) { 743 QTAILQ_INSERT_BEFORE(se, nse, entry); 744 } else { 745 QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry); 746 } 747 748 if (savevm_state.handler_pri_head[priority] == NULL) { 749 savevm_state.handler_pri_head[priority] = nse; 750 } 751 } 752 753 static void savevm_state_handler_remove(SaveStateEntry *se) 754 { 755 SaveStateEntry *next; 756 MigrationPriority priority = save_state_priority(se); 757 758 if (se == savevm_state.handler_pri_head[priority]) { 759 next = QTAILQ_NEXT(se, entry); 760 if (next != NULL && save_state_priority(next) == priority) { 761 savevm_state.handler_pri_head[priority] = next; 762 } else { 763 savevm_state.handler_pri_head[priority] = NULL; 764 } 765 } 766 QTAILQ_REMOVE(&savevm_state.handlers, se, entry); 767 } 768 769 /* TODO: Individual devices generally have very little idea about the rest 770 of the system, so instance_id should be removed/replaced. 771 Meanwhile pass -1 as instance_id if you do not already have a clearly 772 distinguishing id for all instances of your device class. */ 773 int register_savevm_live(const char *idstr, 774 uint32_t instance_id, 775 int version_id, 776 const SaveVMHandlers *ops, 777 void *opaque) 778 { 779 SaveStateEntry *se; 780 781 se = g_new0(SaveStateEntry, 1); 782 se->version_id = version_id; 783 se->section_id = savevm_state.global_section_id++; 784 se->ops = ops; 785 se->opaque = opaque; 786 se->vmsd = NULL; 787 /* if this is a live_savem then set is_ram */ 788 if (ops->save_setup != NULL) { 789 se->is_ram = 1; 790 } 791 792 pstrcat(se->idstr, sizeof(se->idstr), idstr); 793 794 if (instance_id == VMSTATE_INSTANCE_ID_ANY) { 795 se->instance_id = calculate_new_instance_id(se->idstr); 796 } else { 797 se->instance_id = instance_id; 798 } 799 assert(!se->compat || se->instance_id == 0); 800 savevm_state_handler_insert(se); 801 return 0; 802 } 803 804 void unregister_savevm(VMStateIf *obj, const char *idstr, void *opaque) 805 { 806 SaveStateEntry *se, *new_se; 807 char id[256] = ""; 808 809 if (obj) { 810 char *oid = vmstate_if_get_id(obj); 811 if (oid) { 812 pstrcpy(id, sizeof(id), oid); 813 pstrcat(id, sizeof(id), "/"); 814 g_free(oid); 815 } 816 } 817 pstrcat(id, sizeof(id), idstr); 818 819 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) { 820 if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) { 821 savevm_state_handler_remove(se); 822 g_free(se->compat); 823 g_free(se); 824 } 825 } 826 } 827 828 /* 829 * Perform some basic checks on vmsd's at registration 830 * time. 831 */ 832 static void vmstate_check(const VMStateDescription *vmsd) 833 { 834 const VMStateField *field = vmsd->fields; 835 const VMStateDescription * const *subsection = vmsd->subsections; 836 837 if (field) { 838 while (field->name) { 839 if (field->flags & (VMS_STRUCT | VMS_VSTRUCT)) { 840 /* Recurse to sub structures */ 841 vmstate_check(field->vmsd); 842 } 843 /* Carry on */ 844 field++; 845 } 846 /* Check for the end of field list canary */ 847 if (field->flags != VMS_END) { 848 error_report("VMSTATE not ending with VMS_END: %s", vmsd->name); 849 g_assert_not_reached(); 850 } 851 } 852 853 while (subsection && *subsection) { 854 /* 855 * The name of a subsection should start with the name of the 856 * current object. 857 */ 858 assert(!strncmp(vmsd->name, (*subsection)->name, strlen(vmsd->name))); 859 vmstate_check(*subsection); 860 subsection++; 861 } 862 } 863 864 /* 865 * See comment in hw/intc/xics.c:icp_realize() 866 * 867 * This function can be removed when 868 * pre_2_10_vmstate_register_dummy_icp() is removed. 869 */ 870 int vmstate_replace_hack_for_ppc(VMStateIf *obj, int instance_id, 871 const VMStateDescription *vmsd, 872 void *opaque) 873 { 874 SaveStateEntry *se = find_se(vmsd->name, instance_id); 875 876 if (se) { 877 savevm_state_handler_remove(se); 878 } 879 return vmstate_register(obj, instance_id, vmsd, opaque); 880 } 881 882 int vmstate_register_with_alias_id(VMStateIf *obj, uint32_t instance_id, 883 const VMStateDescription *vmsd, 884 void *opaque, int alias_id, 885 int required_for_version, 886 Error **errp) 887 { 888 SaveStateEntry *se; 889 890 /* If this triggers, alias support can be dropped for the vmsd. */ 891 assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id); 892 893 se = g_new0(SaveStateEntry, 1); 894 se->version_id = vmsd->version_id; 895 se->section_id = savevm_state.global_section_id++; 896 se->opaque = opaque; 897 se->vmsd = vmsd; 898 se->alias_id = alias_id; 899 900 if (obj) { 901 char *id = vmstate_if_get_id(obj); 902 if (id) { 903 if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >= 904 sizeof(se->idstr)) { 905 error_setg(errp, "Path too long for VMState (%s)", id); 906 g_free(id); 907 g_free(se); 908 909 return -1; 910 } 911 g_free(id); 912 913 se->compat = g_new0(CompatEntry, 1); 914 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name); 915 se->compat->instance_id = instance_id == VMSTATE_INSTANCE_ID_ANY ? 916 calculate_compat_instance_id(vmsd->name) : instance_id; 917 instance_id = VMSTATE_INSTANCE_ID_ANY; 918 } 919 } 920 pstrcat(se->idstr, sizeof(se->idstr), vmsd->name); 921 922 if (instance_id == VMSTATE_INSTANCE_ID_ANY) { 923 se->instance_id = calculate_new_instance_id(se->idstr); 924 } else { 925 se->instance_id = instance_id; 926 } 927 928 /* Perform a recursive sanity check during the test runs */ 929 if (qtest_enabled()) { 930 vmstate_check(vmsd); 931 } 932 assert(!se->compat || se->instance_id == 0); 933 savevm_state_handler_insert(se); 934 return 0; 935 } 936 937 void vmstate_unregister(VMStateIf *obj, const VMStateDescription *vmsd, 938 void *opaque) 939 { 940 SaveStateEntry *se, *new_se; 941 942 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) { 943 if (se->vmsd == vmsd && se->opaque == opaque) { 944 savevm_state_handler_remove(se); 945 g_free(se->compat); 946 g_free(se); 947 } 948 } 949 } 950 951 static int vmstate_load(QEMUFile *f, SaveStateEntry *se) 952 { 953 trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)"); 954 if (!se->vmsd) { /* Old style */ 955 return se->ops->load_state(f, se->opaque, se->load_version_id); 956 } 957 return vmstate_load_state(f, se->vmsd, se->opaque, se->load_version_id); 958 } 959 960 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se, 961 JSONWriter *vmdesc) 962 { 963 uint64_t old_offset = qemu_file_transferred(f); 964 se->ops->save_state(f, se->opaque); 965 uint64_t size = qemu_file_transferred(f) - old_offset; 966 967 if (vmdesc) { 968 json_writer_int64(vmdesc, "size", size); 969 json_writer_start_array(vmdesc, "fields"); 970 json_writer_start_object(vmdesc, NULL); 971 json_writer_str(vmdesc, "name", "data"); 972 json_writer_int64(vmdesc, "size", size); 973 json_writer_str(vmdesc, "type", "buffer"); 974 json_writer_end_object(vmdesc); 975 json_writer_end_array(vmdesc); 976 } 977 } 978 979 /* 980 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL) 981 */ 982 static void save_section_header(QEMUFile *f, SaveStateEntry *se, 983 uint8_t section_type) 984 { 985 qemu_put_byte(f, section_type); 986 qemu_put_be32(f, se->section_id); 987 988 if (section_type == QEMU_VM_SECTION_FULL || 989 section_type == QEMU_VM_SECTION_START) { 990 /* ID string */ 991 size_t len = strlen(se->idstr); 992 qemu_put_byte(f, len); 993 qemu_put_buffer(f, (uint8_t *)se->idstr, len); 994 995 qemu_put_be32(f, se->instance_id); 996 qemu_put_be32(f, se->version_id); 997 } 998 } 999 1000 /* 1001 * Write a footer onto device sections that catches cases misformatted device 1002 * sections. 1003 */ 1004 static void save_section_footer(QEMUFile *f, SaveStateEntry *se) 1005 { 1006 if (migrate_get_current()->send_section_footer) { 1007 qemu_put_byte(f, QEMU_VM_SECTION_FOOTER); 1008 qemu_put_be32(f, se->section_id); 1009 } 1010 } 1011 1012 static int vmstate_save(QEMUFile *f, SaveStateEntry *se, JSONWriter *vmdesc) 1013 { 1014 int ret; 1015 Error *local_err = NULL; 1016 MigrationState *s = migrate_get_current(); 1017 1018 if ((!se->ops || !se->ops->save_state) && !se->vmsd) { 1019 return 0; 1020 } 1021 if (se->vmsd && !vmstate_section_needed(se->vmsd, se->opaque)) { 1022 trace_savevm_section_skip(se->idstr, se->section_id); 1023 return 0; 1024 } 1025 1026 trace_savevm_section_start(se->idstr, se->section_id); 1027 save_section_header(f, se, QEMU_VM_SECTION_FULL); 1028 if (vmdesc) { 1029 json_writer_start_object(vmdesc, NULL); 1030 json_writer_str(vmdesc, "name", se->idstr); 1031 json_writer_int64(vmdesc, "instance_id", se->instance_id); 1032 } 1033 1034 trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)"); 1035 if (!se->vmsd) { 1036 vmstate_save_old_style(f, se, vmdesc); 1037 } else { 1038 ret = vmstate_save_state_with_err(f, se->vmsd, se->opaque, vmdesc, &local_err); 1039 if (ret) { 1040 migrate_set_error(s, local_err); 1041 error_report_err(local_err); 1042 return ret; 1043 } 1044 } 1045 1046 trace_savevm_section_end(se->idstr, se->section_id, 0); 1047 save_section_footer(f, se); 1048 if (vmdesc) { 1049 json_writer_end_object(vmdesc); 1050 } 1051 return 0; 1052 } 1053 /** 1054 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the 1055 * command and associated data. 1056 * 1057 * @f: File to send command on 1058 * @command: Command type to send 1059 * @len: Length of associated data 1060 * @data: Data associated with command. 1061 */ 1062 static void qemu_savevm_command_send(QEMUFile *f, 1063 enum qemu_vm_cmd command, 1064 uint16_t len, 1065 uint8_t *data) 1066 { 1067 trace_savevm_command_send(command, len); 1068 qemu_put_byte(f, QEMU_VM_COMMAND); 1069 qemu_put_be16(f, (uint16_t)command); 1070 qemu_put_be16(f, len); 1071 qemu_put_buffer(f, data, len); 1072 qemu_fflush(f); 1073 } 1074 1075 void qemu_savevm_send_colo_enable(QEMUFile *f) 1076 { 1077 trace_savevm_send_colo_enable(); 1078 qemu_savevm_command_send(f, MIG_CMD_ENABLE_COLO, 0, NULL); 1079 } 1080 1081 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value) 1082 { 1083 uint32_t buf; 1084 1085 trace_savevm_send_ping(value); 1086 buf = cpu_to_be32(value); 1087 qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf); 1088 } 1089 1090 void qemu_savevm_send_open_return_path(QEMUFile *f) 1091 { 1092 trace_savevm_send_open_return_path(); 1093 qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL); 1094 } 1095 1096 /* We have a buffer of data to send; we don't want that all to be loaded 1097 * by the command itself, so the command contains just the length of the 1098 * extra buffer that we then send straight after it. 1099 * TODO: Must be a better way to organise that 1100 * 1101 * Returns: 1102 * 0 on success 1103 * -ve on error 1104 */ 1105 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len) 1106 { 1107 uint32_t tmp; 1108 MigrationState *ms = migrate_get_current(); 1109 Error *local_err = NULL; 1110 1111 if (len > MAX_VM_CMD_PACKAGED_SIZE) { 1112 error_setg(&local_err, "%s: Unreasonably large packaged state: %zu", 1113 __func__, len); 1114 migrate_set_error(ms, local_err); 1115 error_report_err(local_err); 1116 return -1; 1117 } 1118 1119 tmp = cpu_to_be32(len); 1120 1121 trace_qemu_savevm_send_packaged(); 1122 qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp); 1123 1124 qemu_put_buffer(f, buf, len); 1125 1126 return 0; 1127 } 1128 1129 /* Send prior to any postcopy transfer */ 1130 void qemu_savevm_send_postcopy_advise(QEMUFile *f) 1131 { 1132 if (migrate_postcopy_ram()) { 1133 uint64_t tmp[2]; 1134 tmp[0] = cpu_to_be64(ram_pagesize_summary()); 1135 tmp[1] = cpu_to_be64(qemu_target_page_size()); 1136 1137 trace_qemu_savevm_send_postcopy_advise(); 1138 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 1139 16, (uint8_t *)tmp); 1140 } else { 1141 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 0, NULL); 1142 } 1143 } 1144 1145 /* Sent prior to starting the destination running in postcopy, discard pages 1146 * that have already been sent but redirtied on the source. 1147 * CMD_POSTCOPY_RAM_DISCARD consist of: 1148 * byte version (0) 1149 * byte Length of name field (not including 0) 1150 * n x byte RAM block name 1151 * byte 0 terminator (just for safety) 1152 * n x Byte ranges within the named RAMBlock 1153 * be64 Start of the range 1154 * be64 Length 1155 * 1156 * name: RAMBlock name that these entries are part of 1157 * len: Number of page entries 1158 * start_list: 'len' addresses 1159 * length_list: 'len' addresses 1160 * 1161 */ 1162 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name, 1163 uint16_t len, 1164 uint64_t *start_list, 1165 uint64_t *length_list) 1166 { 1167 uint8_t *buf; 1168 uint16_t tmplen; 1169 uint16_t t; 1170 size_t name_len = strlen(name); 1171 1172 trace_qemu_savevm_send_postcopy_ram_discard(name, len); 1173 assert(name_len < 256); 1174 buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len); 1175 buf[0] = postcopy_ram_discard_version; 1176 buf[1] = name_len; 1177 memcpy(buf + 2, name, name_len); 1178 tmplen = 2 + name_len; 1179 buf[tmplen++] = '\0'; 1180 1181 for (t = 0; t < len; t++) { 1182 stq_be_p(buf + tmplen, start_list[t]); 1183 tmplen += 8; 1184 stq_be_p(buf + tmplen, length_list[t]); 1185 tmplen += 8; 1186 } 1187 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf); 1188 g_free(buf); 1189 } 1190 1191 /* Get the destination into a state where it can receive postcopy data. */ 1192 void qemu_savevm_send_postcopy_listen(QEMUFile *f) 1193 { 1194 trace_savevm_send_postcopy_listen(); 1195 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL); 1196 } 1197 1198 /* Kick the destination into running */ 1199 void qemu_savevm_send_postcopy_run(QEMUFile *f) 1200 { 1201 trace_savevm_send_postcopy_run(); 1202 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL); 1203 } 1204 1205 void qemu_savevm_send_postcopy_resume(QEMUFile *f) 1206 { 1207 trace_savevm_send_postcopy_resume(); 1208 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RESUME, 0, NULL); 1209 } 1210 1211 void qemu_savevm_send_recv_bitmap(QEMUFile *f, char *block_name) 1212 { 1213 size_t len; 1214 char buf[256]; 1215 1216 trace_savevm_send_recv_bitmap(block_name); 1217 1218 buf[0] = len = strlen(block_name); 1219 memcpy(buf + 1, block_name, len); 1220 1221 qemu_savevm_command_send(f, MIG_CMD_RECV_BITMAP, len + 1, (uint8_t *)buf); 1222 } 1223 1224 bool qemu_savevm_state_blocked(Error **errp) 1225 { 1226 SaveStateEntry *se; 1227 1228 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1229 if (se->vmsd && se->vmsd->unmigratable) { 1230 error_setg(errp, "State blocked by non-migratable device '%s'", 1231 se->idstr); 1232 return true; 1233 } 1234 } 1235 return false; 1236 } 1237 1238 void qemu_savevm_non_migratable_list(strList **reasons) 1239 { 1240 SaveStateEntry *se; 1241 1242 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1243 if (se->vmsd && se->vmsd->unmigratable) { 1244 QAPI_LIST_PREPEND(*reasons, 1245 g_strdup_printf("non-migratable device: %s", 1246 se->idstr)); 1247 } 1248 } 1249 } 1250 1251 void qemu_savevm_state_header(QEMUFile *f) 1252 { 1253 MigrationState *s = migrate_get_current(); 1254 1255 s->vmdesc = json_writer_new(false); 1256 1257 trace_savevm_state_header(); 1258 qemu_put_be32(f, QEMU_VM_FILE_MAGIC); 1259 qemu_put_be32(f, QEMU_VM_FILE_VERSION); 1260 1261 if (s->send_configuration) { 1262 qemu_put_byte(f, QEMU_VM_CONFIGURATION); 1263 1264 /* 1265 * This starts the main json object and is paired with the 1266 * json_writer_end_object in 1267 * qemu_savevm_state_complete_precopy_non_iterable 1268 */ 1269 json_writer_start_object(s->vmdesc, NULL); 1270 1271 json_writer_start_object(s->vmdesc, "configuration"); 1272 vmstate_save_state(f, &vmstate_configuration, &savevm_state, s->vmdesc); 1273 json_writer_end_object(s->vmdesc); 1274 } 1275 } 1276 1277 bool qemu_savevm_state_guest_unplug_pending(void) 1278 { 1279 SaveStateEntry *se; 1280 1281 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1282 if (se->vmsd && se->vmsd->dev_unplug_pending && 1283 se->vmsd->dev_unplug_pending(se->opaque)) { 1284 return true; 1285 } 1286 } 1287 1288 return false; 1289 } 1290 1291 int qemu_savevm_state_prepare(Error **errp) 1292 { 1293 SaveStateEntry *se; 1294 int ret; 1295 1296 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1297 if (!se->ops || !se->ops->save_prepare) { 1298 continue; 1299 } 1300 if (se->ops->is_active) { 1301 if (!se->ops->is_active(se->opaque)) { 1302 continue; 1303 } 1304 } 1305 1306 ret = se->ops->save_prepare(se->opaque, errp); 1307 if (ret < 0) { 1308 return ret; 1309 } 1310 } 1311 1312 return 0; 1313 } 1314 1315 void qemu_savevm_state_setup(QEMUFile *f) 1316 { 1317 MigrationState *ms = migrate_get_current(); 1318 SaveStateEntry *se; 1319 Error *local_err = NULL; 1320 int ret = 0; 1321 1322 json_writer_int64(ms->vmdesc, "page_size", qemu_target_page_size()); 1323 json_writer_start_array(ms->vmdesc, "devices"); 1324 1325 trace_savevm_state_setup(); 1326 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1327 if (se->vmsd && se->vmsd->early_setup) { 1328 ret = vmstate_save(f, se, ms->vmdesc); 1329 if (ret) { 1330 qemu_file_set_error(f, ret); 1331 break; 1332 } 1333 continue; 1334 } 1335 1336 if (!se->ops || !se->ops->save_setup) { 1337 continue; 1338 } 1339 if (se->ops->is_active) { 1340 if (!se->ops->is_active(se->opaque)) { 1341 continue; 1342 } 1343 } 1344 save_section_header(f, se, QEMU_VM_SECTION_START); 1345 1346 ret = se->ops->save_setup(f, se->opaque); 1347 save_section_footer(f, se); 1348 if (ret < 0) { 1349 qemu_file_set_error(f, ret); 1350 break; 1351 } 1352 } 1353 1354 if (ret) { 1355 return; 1356 } 1357 1358 if (precopy_notify(PRECOPY_NOTIFY_SETUP, &local_err)) { 1359 error_report_err(local_err); 1360 } 1361 } 1362 1363 int qemu_savevm_state_resume_prepare(MigrationState *s) 1364 { 1365 SaveStateEntry *se; 1366 int ret; 1367 1368 trace_savevm_state_resume_prepare(); 1369 1370 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1371 if (!se->ops || !se->ops->resume_prepare) { 1372 continue; 1373 } 1374 if (se->ops->is_active) { 1375 if (!se->ops->is_active(se->opaque)) { 1376 continue; 1377 } 1378 } 1379 ret = se->ops->resume_prepare(s, se->opaque); 1380 if (ret < 0) { 1381 return ret; 1382 } 1383 } 1384 1385 return 0; 1386 } 1387 1388 /* 1389 * this function has three return values: 1390 * negative: there was one error, and we have -errno. 1391 * 0 : We haven't finished, caller have to go again 1392 * 1 : We have finished, we can go to complete phase 1393 */ 1394 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy) 1395 { 1396 SaveStateEntry *se; 1397 bool all_finished = true; 1398 int ret; 1399 1400 trace_savevm_state_iterate(); 1401 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1402 if (!se->ops || !se->ops->save_live_iterate) { 1403 continue; 1404 } 1405 if (se->ops->is_active && 1406 !se->ops->is_active(se->opaque)) { 1407 continue; 1408 } 1409 if (se->ops->is_active_iterate && 1410 !se->ops->is_active_iterate(se->opaque)) { 1411 continue; 1412 } 1413 /* 1414 * In the postcopy phase, any device that doesn't know how to 1415 * do postcopy should have saved it's state in the _complete 1416 * call that's already run, it might get confused if we call 1417 * iterate afterwards. 1418 */ 1419 if (postcopy && 1420 !(se->ops->has_postcopy && se->ops->has_postcopy(se->opaque))) { 1421 continue; 1422 } 1423 if (migration_rate_exceeded(f)) { 1424 return 0; 1425 } 1426 trace_savevm_section_start(se->idstr, se->section_id); 1427 1428 save_section_header(f, se, QEMU_VM_SECTION_PART); 1429 1430 ret = se->ops->save_live_iterate(f, se->opaque); 1431 trace_savevm_section_end(se->idstr, se->section_id, ret); 1432 save_section_footer(f, se); 1433 1434 if (ret < 0) { 1435 error_report("failed to save SaveStateEntry with id(name): " 1436 "%d(%s): %d", 1437 se->section_id, se->idstr, ret); 1438 qemu_file_set_error(f, ret); 1439 return ret; 1440 } else if (!ret) { 1441 all_finished = false; 1442 } 1443 } 1444 return all_finished; 1445 } 1446 1447 static bool should_send_vmdesc(void) 1448 { 1449 MachineState *machine = MACHINE(qdev_get_machine()); 1450 bool in_postcopy = migration_in_postcopy(); 1451 return !machine->suppress_vmdesc && !in_postcopy; 1452 } 1453 1454 /* 1455 * Calls the save_live_complete_postcopy methods 1456 * causing the last few pages to be sent immediately and doing any associated 1457 * cleanup. 1458 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete 1459 * all the other devices, but that happens at the point we switch to postcopy. 1460 */ 1461 void qemu_savevm_state_complete_postcopy(QEMUFile *f) 1462 { 1463 SaveStateEntry *se; 1464 int ret; 1465 1466 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1467 if (!se->ops || !se->ops->save_live_complete_postcopy) { 1468 continue; 1469 } 1470 if (se->ops->is_active) { 1471 if (!se->ops->is_active(se->opaque)) { 1472 continue; 1473 } 1474 } 1475 trace_savevm_section_start(se->idstr, se->section_id); 1476 /* Section type */ 1477 qemu_put_byte(f, QEMU_VM_SECTION_END); 1478 qemu_put_be32(f, se->section_id); 1479 1480 ret = se->ops->save_live_complete_postcopy(f, se->opaque); 1481 trace_savevm_section_end(se->idstr, se->section_id, ret); 1482 save_section_footer(f, se); 1483 if (ret < 0) { 1484 qemu_file_set_error(f, ret); 1485 return; 1486 } 1487 } 1488 1489 qemu_put_byte(f, QEMU_VM_EOF); 1490 qemu_fflush(f); 1491 } 1492 1493 static 1494 int qemu_savevm_state_complete_precopy_iterable(QEMUFile *f, bool in_postcopy) 1495 { 1496 int64_t start_ts_each, end_ts_each; 1497 SaveStateEntry *se; 1498 int ret; 1499 1500 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1501 if (!se->ops || 1502 (in_postcopy && se->ops->has_postcopy && 1503 se->ops->has_postcopy(se->opaque)) || 1504 !se->ops->save_live_complete_precopy) { 1505 continue; 1506 } 1507 1508 if (se->ops->is_active) { 1509 if (!se->ops->is_active(se->opaque)) { 1510 continue; 1511 } 1512 } 1513 1514 start_ts_each = qemu_clock_get_us(QEMU_CLOCK_REALTIME); 1515 trace_savevm_section_start(se->idstr, se->section_id); 1516 1517 save_section_header(f, se, QEMU_VM_SECTION_END); 1518 1519 ret = se->ops->save_live_complete_precopy(f, se->opaque); 1520 trace_savevm_section_end(se->idstr, se->section_id, ret); 1521 save_section_footer(f, se); 1522 if (ret < 0) { 1523 qemu_file_set_error(f, ret); 1524 return -1; 1525 } 1526 end_ts_each = qemu_clock_get_us(QEMU_CLOCK_REALTIME); 1527 trace_vmstate_downtime_save("iterable", se->idstr, se->instance_id, 1528 end_ts_each - start_ts_each); 1529 } 1530 1531 trace_vmstate_downtime_checkpoint("src-iterable-saved"); 1532 1533 return 0; 1534 } 1535 1536 int qemu_savevm_state_complete_precopy_non_iterable(QEMUFile *f, 1537 bool in_postcopy, 1538 bool inactivate_disks) 1539 { 1540 MigrationState *ms = migrate_get_current(); 1541 int64_t start_ts_each, end_ts_each; 1542 JSONWriter *vmdesc = ms->vmdesc; 1543 int vmdesc_len; 1544 SaveStateEntry *se; 1545 int ret; 1546 1547 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1548 if (se->vmsd && se->vmsd->early_setup) { 1549 /* Already saved during qemu_savevm_state_setup(). */ 1550 continue; 1551 } 1552 1553 start_ts_each = qemu_clock_get_us(QEMU_CLOCK_REALTIME); 1554 1555 ret = vmstate_save(f, se, vmdesc); 1556 if (ret) { 1557 qemu_file_set_error(f, ret); 1558 return ret; 1559 } 1560 1561 end_ts_each = qemu_clock_get_us(QEMU_CLOCK_REALTIME); 1562 trace_vmstate_downtime_save("non-iterable", se->idstr, se->instance_id, 1563 end_ts_each - start_ts_each); 1564 } 1565 1566 if (inactivate_disks) { 1567 /* Inactivate before sending QEMU_VM_EOF so that the 1568 * bdrv_activate_all() on the other end won't fail. */ 1569 ret = bdrv_inactivate_all(); 1570 if (ret) { 1571 Error *local_err = NULL; 1572 error_setg(&local_err, "%s: bdrv_inactivate_all() failed (%d)", 1573 __func__, ret); 1574 migrate_set_error(ms, local_err); 1575 error_report_err(local_err); 1576 qemu_file_set_error(f, ret); 1577 return ret; 1578 } 1579 } 1580 if (!in_postcopy) { 1581 /* Postcopy stream will still be going */ 1582 qemu_put_byte(f, QEMU_VM_EOF); 1583 } 1584 1585 json_writer_end_array(vmdesc); 1586 json_writer_end_object(vmdesc); 1587 vmdesc_len = strlen(json_writer_get(vmdesc)); 1588 1589 if (should_send_vmdesc()) { 1590 qemu_put_byte(f, QEMU_VM_VMDESCRIPTION); 1591 qemu_put_be32(f, vmdesc_len); 1592 qemu_put_buffer(f, (uint8_t *)json_writer_get(vmdesc), vmdesc_len); 1593 } 1594 1595 /* Free it now to detect any inconsistencies. */ 1596 json_writer_free(vmdesc); 1597 ms->vmdesc = NULL; 1598 1599 trace_vmstate_downtime_checkpoint("src-non-iterable-saved"); 1600 1601 return 0; 1602 } 1603 1604 int qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only, 1605 bool inactivate_disks) 1606 { 1607 int ret; 1608 Error *local_err = NULL; 1609 bool in_postcopy = migration_in_postcopy(); 1610 1611 if (precopy_notify(PRECOPY_NOTIFY_COMPLETE, &local_err)) { 1612 error_report_err(local_err); 1613 } 1614 1615 trace_savevm_state_complete_precopy(); 1616 1617 cpu_synchronize_all_states(); 1618 1619 if (!in_postcopy || iterable_only) { 1620 ret = qemu_savevm_state_complete_precopy_iterable(f, in_postcopy); 1621 if (ret) { 1622 return ret; 1623 } 1624 } 1625 1626 if (iterable_only) { 1627 goto flush; 1628 } 1629 1630 ret = qemu_savevm_state_complete_precopy_non_iterable(f, in_postcopy, 1631 inactivate_disks); 1632 if (ret) { 1633 return ret; 1634 } 1635 1636 flush: 1637 return qemu_fflush(f); 1638 } 1639 1640 /* Give an estimate of the amount left to be transferred, 1641 * the result is split into the amount for units that can and 1642 * for units that can't do postcopy. 1643 */ 1644 void qemu_savevm_state_pending_estimate(uint64_t *must_precopy, 1645 uint64_t *can_postcopy) 1646 { 1647 SaveStateEntry *se; 1648 1649 *must_precopy = 0; 1650 *can_postcopy = 0; 1651 1652 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1653 if (!se->ops || !se->ops->state_pending_estimate) { 1654 continue; 1655 } 1656 if (se->ops->is_active) { 1657 if (!se->ops->is_active(se->opaque)) { 1658 continue; 1659 } 1660 } 1661 se->ops->state_pending_estimate(se->opaque, must_precopy, can_postcopy); 1662 } 1663 } 1664 1665 void qemu_savevm_state_pending_exact(uint64_t *must_precopy, 1666 uint64_t *can_postcopy) 1667 { 1668 SaveStateEntry *se; 1669 1670 *must_precopy = 0; 1671 *can_postcopy = 0; 1672 1673 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1674 if (!se->ops || !se->ops->state_pending_exact) { 1675 continue; 1676 } 1677 if (se->ops->is_active) { 1678 if (!se->ops->is_active(se->opaque)) { 1679 continue; 1680 } 1681 } 1682 se->ops->state_pending_exact(se->opaque, must_precopy, can_postcopy); 1683 } 1684 } 1685 1686 void qemu_savevm_state_cleanup(void) 1687 { 1688 SaveStateEntry *se; 1689 Error *local_err = NULL; 1690 1691 if (precopy_notify(PRECOPY_NOTIFY_CLEANUP, &local_err)) { 1692 error_report_err(local_err); 1693 } 1694 1695 trace_savevm_state_cleanup(); 1696 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1697 if (se->ops && se->ops->save_cleanup) { 1698 se->ops->save_cleanup(se->opaque); 1699 } 1700 } 1701 } 1702 1703 static int qemu_savevm_state(QEMUFile *f, Error **errp) 1704 { 1705 int ret; 1706 MigrationState *ms = migrate_get_current(); 1707 MigrationStatus status; 1708 1709 if (migration_is_running()) { 1710 error_setg(errp, QERR_MIGRATION_ACTIVE); 1711 return -EINVAL; 1712 } 1713 1714 if (migrate_block()) { 1715 error_setg(errp, "Block migration and snapshots are incompatible"); 1716 return -EINVAL; 1717 } 1718 1719 ret = migrate_init(ms, errp); 1720 if (ret) { 1721 return ret; 1722 } 1723 ms->to_dst_file = f; 1724 1725 qemu_savevm_state_header(f); 1726 qemu_savevm_state_setup(f); 1727 1728 while (qemu_file_get_error(f) == 0) { 1729 if (qemu_savevm_state_iterate(f, false) > 0) { 1730 break; 1731 } 1732 } 1733 1734 ret = qemu_file_get_error(f); 1735 if (ret == 0) { 1736 qemu_savevm_state_complete_precopy(f, false, false); 1737 ret = qemu_file_get_error(f); 1738 } 1739 qemu_savevm_state_cleanup(); 1740 if (ret != 0) { 1741 error_setg_errno(errp, -ret, "Error while writing VM state"); 1742 } 1743 1744 if (ret != 0) { 1745 status = MIGRATION_STATUS_FAILED; 1746 } else { 1747 status = MIGRATION_STATUS_COMPLETED; 1748 } 1749 migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status); 1750 1751 /* f is outer parameter, it should not stay in global migration state after 1752 * this function finished */ 1753 ms->to_dst_file = NULL; 1754 1755 return ret; 1756 } 1757 1758 void qemu_savevm_live_state(QEMUFile *f) 1759 { 1760 /* save QEMU_VM_SECTION_END section */ 1761 qemu_savevm_state_complete_precopy(f, true, false); 1762 qemu_put_byte(f, QEMU_VM_EOF); 1763 } 1764 1765 int qemu_save_device_state(QEMUFile *f) 1766 { 1767 SaveStateEntry *se; 1768 1769 if (!migration_in_colo_state()) { 1770 qemu_put_be32(f, QEMU_VM_FILE_MAGIC); 1771 qemu_put_be32(f, QEMU_VM_FILE_VERSION); 1772 } 1773 cpu_synchronize_all_states(); 1774 1775 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1776 int ret; 1777 1778 if (se->is_ram) { 1779 continue; 1780 } 1781 ret = vmstate_save(f, se, NULL); 1782 if (ret) { 1783 return ret; 1784 } 1785 } 1786 1787 qemu_put_byte(f, QEMU_VM_EOF); 1788 1789 return qemu_file_get_error(f); 1790 } 1791 1792 static SaveStateEntry *find_se(const char *idstr, uint32_t instance_id) 1793 { 1794 SaveStateEntry *se; 1795 1796 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1797 if (!strcmp(se->idstr, idstr) && 1798 (instance_id == se->instance_id || 1799 instance_id == se->alias_id)) 1800 return se; 1801 /* Migrating from an older version? */ 1802 if (strstr(se->idstr, idstr) && se->compat) { 1803 if (!strcmp(se->compat->idstr, idstr) && 1804 (instance_id == se->compat->instance_id || 1805 instance_id == se->alias_id)) 1806 return se; 1807 } 1808 } 1809 return NULL; 1810 } 1811 1812 enum LoadVMExitCodes { 1813 /* Allow a command to quit all layers of nested loadvm loops */ 1814 LOADVM_QUIT = 1, 1815 }; 1816 1817 /* ------ incoming postcopy messages ------ */ 1818 /* 'advise' arrives before any transfers just to tell us that a postcopy 1819 * *might* happen - it might be skipped if precopy transferred everything 1820 * quickly. 1821 */ 1822 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis, 1823 uint16_t len) 1824 { 1825 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE); 1826 uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps; 1827 size_t page_size = qemu_target_page_size(); 1828 Error *local_err = NULL; 1829 1830 trace_loadvm_postcopy_handle_advise(); 1831 if (ps != POSTCOPY_INCOMING_NONE) { 1832 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps); 1833 return -1; 1834 } 1835 1836 switch (len) { 1837 case 0: 1838 if (migrate_postcopy_ram()) { 1839 error_report("RAM postcopy is enabled but have 0 byte advise"); 1840 return -EINVAL; 1841 } 1842 return 0; 1843 case 8 + 8: 1844 if (!migrate_postcopy_ram()) { 1845 error_report("RAM postcopy is disabled but have 16 byte advise"); 1846 return -EINVAL; 1847 } 1848 break; 1849 default: 1850 error_report("CMD_POSTCOPY_ADVISE invalid length (%d)", len); 1851 return -EINVAL; 1852 } 1853 1854 if (!postcopy_ram_supported_by_host(mis, &local_err)) { 1855 error_report_err(local_err); 1856 postcopy_state_set(POSTCOPY_INCOMING_NONE); 1857 return -1; 1858 } 1859 1860 remote_pagesize_summary = qemu_get_be64(mis->from_src_file); 1861 local_pagesize_summary = ram_pagesize_summary(); 1862 1863 if (remote_pagesize_summary != local_pagesize_summary) { 1864 /* 1865 * This detects two potential causes of mismatch: 1866 * a) A mismatch in host page sizes 1867 * Some combinations of mismatch are probably possible but it gets 1868 * a bit more complicated. In particular we need to place whole 1869 * host pages on the dest at once, and we need to ensure that we 1870 * handle dirtying to make sure we never end up sending part of 1871 * a hostpage on it's own. 1872 * b) The use of different huge page sizes on source/destination 1873 * a more fine grain test is performed during RAM block migration 1874 * but this test here causes a nice early clear failure, and 1875 * also fails when passed to an older qemu that doesn't 1876 * do huge pages. 1877 */ 1878 error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64 1879 " d=%" PRIx64 ")", 1880 remote_pagesize_summary, local_pagesize_summary); 1881 return -1; 1882 } 1883 1884 remote_tps = qemu_get_be64(mis->from_src_file); 1885 if (remote_tps != page_size) { 1886 /* 1887 * Again, some differences could be dealt with, but for now keep it 1888 * simple. 1889 */ 1890 error_report("Postcopy needs matching target page sizes (s=%d d=%zd)", 1891 (int)remote_tps, page_size); 1892 return -1; 1893 } 1894 1895 if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_ADVISE, &local_err)) { 1896 error_report_err(local_err); 1897 return -1; 1898 } 1899 1900 if (ram_postcopy_incoming_init(mis)) { 1901 return -1; 1902 } 1903 1904 return 0; 1905 } 1906 1907 /* After postcopy we will be told to throw some pages away since they're 1908 * dirty and will have to be demand fetched. Must happen before CPU is 1909 * started. 1910 * There can be 0..many of these messages, each encoding multiple pages. 1911 */ 1912 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis, 1913 uint16_t len) 1914 { 1915 int tmp; 1916 char ramid[256]; 1917 PostcopyState ps = postcopy_state_get(); 1918 1919 trace_loadvm_postcopy_ram_handle_discard(); 1920 1921 switch (ps) { 1922 case POSTCOPY_INCOMING_ADVISE: 1923 /* 1st discard */ 1924 tmp = postcopy_ram_prepare_discard(mis); 1925 if (tmp) { 1926 return tmp; 1927 } 1928 break; 1929 1930 case POSTCOPY_INCOMING_DISCARD: 1931 /* Expected state */ 1932 break; 1933 1934 default: 1935 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)", 1936 ps); 1937 return -1; 1938 } 1939 /* We're expecting a 1940 * Version (0) 1941 * a RAM ID string (length byte, name, 0 term) 1942 * then at least 1 16 byte chunk 1943 */ 1944 if (len < (1 + 1 + 1 + 1 + 2 * 8)) { 1945 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len); 1946 return -1; 1947 } 1948 1949 tmp = qemu_get_byte(mis->from_src_file); 1950 if (tmp != postcopy_ram_discard_version) { 1951 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp); 1952 return -1; 1953 } 1954 1955 if (!qemu_get_counted_string(mis->from_src_file, ramid)) { 1956 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID"); 1957 return -1; 1958 } 1959 tmp = qemu_get_byte(mis->from_src_file); 1960 if (tmp != 0) { 1961 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp); 1962 return -1; 1963 } 1964 1965 len -= 3 + strlen(ramid); 1966 if (len % 16) { 1967 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len); 1968 return -1; 1969 } 1970 trace_loadvm_postcopy_ram_handle_discard_header(ramid, len); 1971 while (len) { 1972 uint64_t start_addr, block_length; 1973 start_addr = qemu_get_be64(mis->from_src_file); 1974 block_length = qemu_get_be64(mis->from_src_file); 1975 1976 len -= 16; 1977 int ret = ram_discard_range(ramid, start_addr, block_length); 1978 if (ret) { 1979 return ret; 1980 } 1981 } 1982 trace_loadvm_postcopy_ram_handle_discard_end(); 1983 1984 return 0; 1985 } 1986 1987 /* 1988 * Triggered by a postcopy_listen command; this thread takes over reading 1989 * the input stream, leaving the main thread free to carry on loading the rest 1990 * of the device state (from RAM). 1991 * (TODO:This could do with being in a postcopy file - but there again it's 1992 * just another input loop, not that postcopy specific) 1993 */ 1994 static void *postcopy_ram_listen_thread(void *opaque) 1995 { 1996 MigrationIncomingState *mis = migration_incoming_get_current(); 1997 QEMUFile *f = mis->from_src_file; 1998 int load_res; 1999 MigrationState *migr = migrate_get_current(); 2000 2001 object_ref(OBJECT(migr)); 2002 2003 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, 2004 MIGRATION_STATUS_POSTCOPY_ACTIVE); 2005 qemu_sem_post(&mis->thread_sync_sem); 2006 trace_postcopy_ram_listen_thread_start(); 2007 2008 rcu_register_thread(); 2009 /* 2010 * Because we're a thread and not a coroutine we can't yield 2011 * in qemu_file, and thus we must be blocking now. 2012 */ 2013 qemu_file_set_blocking(f, true); 2014 load_res = qemu_loadvm_state_main(f, mis); 2015 2016 /* 2017 * This is tricky, but, mis->from_src_file can change after it 2018 * returns, when postcopy recovery happened. In the future, we may 2019 * want a wrapper for the QEMUFile handle. 2020 */ 2021 f = mis->from_src_file; 2022 2023 /* And non-blocking again so we don't block in any cleanup */ 2024 qemu_file_set_blocking(f, false); 2025 2026 trace_postcopy_ram_listen_thread_exit(); 2027 if (load_res < 0) { 2028 qemu_file_set_error(f, load_res); 2029 dirty_bitmap_mig_cancel_incoming(); 2030 if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING && 2031 !migrate_postcopy_ram() && migrate_dirty_bitmaps()) 2032 { 2033 error_report("%s: loadvm failed during postcopy: %d. All states " 2034 "are migrated except dirty bitmaps. Some dirty " 2035 "bitmaps may be lost, and present migrated dirty " 2036 "bitmaps are correctly migrated and valid.", 2037 __func__, load_res); 2038 load_res = 0; /* prevent further exit() */ 2039 } else { 2040 error_report("%s: loadvm failed: %d", __func__, load_res); 2041 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, 2042 MIGRATION_STATUS_FAILED); 2043 } 2044 } 2045 if (load_res >= 0) { 2046 /* 2047 * This looks good, but it's possible that the device loading in the 2048 * main thread hasn't finished yet, and so we might not be in 'RUN' 2049 * state yet; wait for the end of the main thread. 2050 */ 2051 qemu_event_wait(&mis->main_thread_load_event); 2052 } 2053 postcopy_ram_incoming_cleanup(mis); 2054 2055 if (load_res < 0) { 2056 /* 2057 * If something went wrong then we have a bad state so exit; 2058 * depending how far we got it might be possible at this point 2059 * to leave the guest running and fire MCEs for pages that never 2060 * arrived as a desperate recovery step. 2061 */ 2062 rcu_unregister_thread(); 2063 exit(EXIT_FAILURE); 2064 } 2065 2066 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, 2067 MIGRATION_STATUS_COMPLETED); 2068 /* 2069 * If everything has worked fine, then the main thread has waited 2070 * for us to start, and we're the last use of the mis. 2071 * (If something broke then qemu will have to exit anyway since it's 2072 * got a bad migration state). 2073 */ 2074 migration_incoming_state_destroy(); 2075 qemu_loadvm_state_cleanup(); 2076 2077 rcu_unregister_thread(); 2078 mis->have_listen_thread = false; 2079 postcopy_state_set(POSTCOPY_INCOMING_END); 2080 2081 object_unref(OBJECT(migr)); 2082 2083 return NULL; 2084 } 2085 2086 /* After this message we must be able to immediately receive postcopy data */ 2087 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis) 2088 { 2089 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING); 2090 Error *local_err = NULL; 2091 2092 trace_loadvm_postcopy_handle_listen("enter"); 2093 2094 if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) { 2095 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps); 2096 return -1; 2097 } 2098 if (ps == POSTCOPY_INCOMING_ADVISE) { 2099 /* 2100 * A rare case, we entered listen without having to do any discards, 2101 * so do the setup that's normally done at the time of the 1st discard. 2102 */ 2103 if (migrate_postcopy_ram()) { 2104 postcopy_ram_prepare_discard(mis); 2105 } 2106 } 2107 2108 trace_loadvm_postcopy_handle_listen("after discard"); 2109 2110 /* 2111 * Sensitise RAM - can now generate requests for blocks that don't exist 2112 * However, at this point the CPU shouldn't be running, and the IO 2113 * shouldn't be doing anything yet so don't actually expect requests 2114 */ 2115 if (migrate_postcopy_ram()) { 2116 if (postcopy_ram_incoming_setup(mis)) { 2117 postcopy_ram_incoming_cleanup(mis); 2118 return -1; 2119 } 2120 } 2121 2122 trace_loadvm_postcopy_handle_listen("after uffd"); 2123 2124 if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_LISTEN, &local_err)) { 2125 error_report_err(local_err); 2126 return -1; 2127 } 2128 2129 mis->have_listen_thread = true; 2130 postcopy_thread_create(mis, &mis->listen_thread, "postcopy/listen", 2131 postcopy_ram_listen_thread, QEMU_THREAD_DETACHED); 2132 trace_loadvm_postcopy_handle_listen("return"); 2133 2134 return 0; 2135 } 2136 2137 static void loadvm_postcopy_handle_run_bh(void *opaque) 2138 { 2139 Error *local_err = NULL; 2140 MigrationIncomingState *mis = opaque; 2141 2142 trace_vmstate_downtime_checkpoint("dst-postcopy-bh-enter"); 2143 2144 /* TODO we should move all of this lot into postcopy_ram.c or a shared code 2145 * in migration.c 2146 */ 2147 cpu_synchronize_all_post_init(); 2148 2149 trace_vmstate_downtime_checkpoint("dst-postcopy-bh-cpu-synced"); 2150 2151 qemu_announce_self(&mis->announce_timer, migrate_announce_params()); 2152 2153 trace_vmstate_downtime_checkpoint("dst-postcopy-bh-announced"); 2154 2155 /* Make sure all file formats throw away their mutable metadata. 2156 * If we get an error here, just don't restart the VM yet. */ 2157 bdrv_activate_all(&local_err); 2158 if (local_err) { 2159 error_report_err(local_err); 2160 local_err = NULL; 2161 autostart = false; 2162 } 2163 2164 trace_vmstate_downtime_checkpoint("dst-postcopy-bh-cache-invalidated"); 2165 2166 dirty_bitmap_mig_before_vm_start(); 2167 2168 if (autostart) { 2169 /* Hold onto your hats, starting the CPU */ 2170 vm_start(); 2171 } else { 2172 /* leave it paused and let management decide when to start the CPU */ 2173 runstate_set(RUN_STATE_PAUSED); 2174 } 2175 2176 trace_vmstate_downtime_checkpoint("dst-postcopy-bh-vm-started"); 2177 } 2178 2179 /* After all discards we can start running and asking for pages */ 2180 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis) 2181 { 2182 PostcopyState ps = postcopy_state_get(); 2183 2184 trace_loadvm_postcopy_handle_run(); 2185 if (ps != POSTCOPY_INCOMING_LISTENING) { 2186 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps); 2187 return -1; 2188 } 2189 2190 postcopy_state_set(POSTCOPY_INCOMING_RUNNING); 2191 migration_bh_schedule(loadvm_postcopy_handle_run_bh, mis); 2192 2193 /* We need to finish reading the stream from the package 2194 * and also stop reading anything more from the stream that loaded the 2195 * package (since it's now being read by the listener thread). 2196 * LOADVM_QUIT will quit all the layers of nested loadvm loops. 2197 */ 2198 return LOADVM_QUIT; 2199 } 2200 2201 /* We must be with page_request_mutex held */ 2202 static gboolean postcopy_sync_page_req(gpointer key, gpointer value, 2203 gpointer data) 2204 { 2205 MigrationIncomingState *mis = data; 2206 void *host_addr = (void *) key; 2207 ram_addr_t rb_offset; 2208 RAMBlock *rb; 2209 int ret; 2210 2211 rb = qemu_ram_block_from_host(host_addr, true, &rb_offset); 2212 if (!rb) { 2213 /* 2214 * This should _never_ happen. However be nice for a migrating VM to 2215 * not crash/assert. Post an error (note: intended to not use *_once 2216 * because we do want to see all the illegal addresses; and this can 2217 * never be triggered by the guest so we're safe) and move on next. 2218 */ 2219 error_report("%s: illegal host addr %p", __func__, host_addr); 2220 /* Try the next entry */ 2221 return FALSE; 2222 } 2223 2224 ret = migrate_send_rp_message_req_pages(mis, rb, rb_offset); 2225 if (ret) { 2226 /* Please refer to above comment. */ 2227 error_report("%s: send rp message failed for addr %p", 2228 __func__, host_addr); 2229 return FALSE; 2230 } 2231 2232 trace_postcopy_page_req_sync(host_addr); 2233 2234 return FALSE; 2235 } 2236 2237 static void migrate_send_rp_req_pages_pending(MigrationIncomingState *mis) 2238 { 2239 WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) { 2240 g_tree_foreach(mis->page_requested, postcopy_sync_page_req, mis); 2241 } 2242 } 2243 2244 static int loadvm_postcopy_handle_resume(MigrationIncomingState *mis) 2245 { 2246 if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) { 2247 error_report("%s: illegal resume received", __func__); 2248 /* Don't fail the load, only for this. */ 2249 return 0; 2250 } 2251 2252 /* 2253 * Reset the last_rb before we resend any page req to source again, since 2254 * the source should have it reset already. 2255 */ 2256 mis->last_rb = NULL; 2257 2258 /* 2259 * This means source VM is ready to resume the postcopy migration. 2260 */ 2261 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_RECOVER, 2262 MIGRATION_STATUS_POSTCOPY_ACTIVE); 2263 2264 trace_loadvm_postcopy_handle_resume(); 2265 2266 /* Tell source that "we are ready" */ 2267 migrate_send_rp_resume_ack(mis, MIGRATION_RESUME_ACK_VALUE); 2268 2269 /* 2270 * After a postcopy recovery, the source should have lost the postcopy 2271 * queue, or potentially the requested pages could have been lost during 2272 * the network down phase. Let's re-sync with the source VM by re-sending 2273 * all the pending pages that we eagerly need, so these threads won't get 2274 * blocked too long due to the recovery. 2275 * 2276 * Without this procedure, the faulted destination VM threads (waiting for 2277 * page requests right before the postcopy is interrupted) can keep hanging 2278 * until the pages are sent by the source during the background copying of 2279 * pages, or another thread faulted on the same address accidentally. 2280 */ 2281 migrate_send_rp_req_pages_pending(mis); 2282 2283 /* 2284 * It's time to switch state and release the fault thread to continue 2285 * service page faults. Note that this should be explicitly after the 2286 * above call to migrate_send_rp_req_pages_pending(). In short: 2287 * migrate_send_rp_message_req_pages() is not thread safe, yet. 2288 */ 2289 qemu_sem_post(&mis->postcopy_pause_sem_fault); 2290 2291 if (migrate_postcopy_preempt()) { 2292 /* 2293 * The preempt channel will be created in async manner, now let's 2294 * wait for it and make sure it's created. 2295 */ 2296 qemu_sem_wait(&mis->postcopy_qemufile_dst_done); 2297 assert(mis->postcopy_qemufile_dst); 2298 /* Kick the fast ram load thread too */ 2299 qemu_sem_post(&mis->postcopy_pause_sem_fast_load); 2300 } 2301 2302 return 0; 2303 } 2304 2305 /** 2306 * Immediately following this command is a blob of data containing an embedded 2307 * chunk of migration stream; read it and load it. 2308 * 2309 * @mis: Incoming state 2310 * @length: Length of packaged data to read 2311 * 2312 * Returns: Negative values on error 2313 * 2314 */ 2315 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis) 2316 { 2317 int ret; 2318 size_t length; 2319 QIOChannelBuffer *bioc; 2320 2321 length = qemu_get_be32(mis->from_src_file); 2322 trace_loadvm_handle_cmd_packaged(length); 2323 2324 if (length > MAX_VM_CMD_PACKAGED_SIZE) { 2325 error_report("Unreasonably large packaged state: %zu", length); 2326 return -1; 2327 } 2328 2329 bioc = qio_channel_buffer_new(length); 2330 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer"); 2331 ret = qemu_get_buffer(mis->from_src_file, 2332 bioc->data, 2333 length); 2334 if (ret != length) { 2335 object_unref(OBJECT(bioc)); 2336 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu", 2337 ret, length); 2338 return (ret < 0) ? ret : -EAGAIN; 2339 } 2340 bioc->usage += length; 2341 trace_loadvm_handle_cmd_packaged_received(ret); 2342 2343 QEMUFile *packf = qemu_file_new_input(QIO_CHANNEL(bioc)); 2344 2345 ret = qemu_loadvm_state_main(packf, mis); 2346 trace_loadvm_handle_cmd_packaged_main(ret); 2347 qemu_fclose(packf); 2348 object_unref(OBJECT(bioc)); 2349 2350 return ret; 2351 } 2352 2353 /* 2354 * Handle request that source requests for recved_bitmap on 2355 * destination. Payload format: 2356 * 2357 * len (1 byte) + ramblock_name (<255 bytes) 2358 */ 2359 static int loadvm_handle_recv_bitmap(MigrationIncomingState *mis, 2360 uint16_t len) 2361 { 2362 QEMUFile *file = mis->from_src_file; 2363 RAMBlock *rb; 2364 char block_name[256]; 2365 size_t cnt; 2366 2367 cnt = qemu_get_counted_string(file, block_name); 2368 if (!cnt) { 2369 error_report("%s: failed to read block name", __func__); 2370 return -EINVAL; 2371 } 2372 2373 /* Validate before using the data */ 2374 if (qemu_file_get_error(file)) { 2375 return qemu_file_get_error(file); 2376 } 2377 2378 if (len != cnt + 1) { 2379 error_report("%s: invalid payload length (%d)", __func__, len); 2380 return -EINVAL; 2381 } 2382 2383 rb = qemu_ram_block_by_name(block_name); 2384 if (!rb) { 2385 error_report("%s: block '%s' not found", __func__, block_name); 2386 return -EINVAL; 2387 } 2388 2389 migrate_send_rp_recv_bitmap(mis, block_name); 2390 2391 trace_loadvm_handle_recv_bitmap(block_name); 2392 2393 return 0; 2394 } 2395 2396 static int loadvm_process_enable_colo(MigrationIncomingState *mis) 2397 { 2398 int ret = migration_incoming_enable_colo(); 2399 2400 if (!ret) { 2401 ret = colo_init_ram_cache(); 2402 if (ret) { 2403 migration_incoming_disable_colo(); 2404 } 2405 } 2406 return ret; 2407 } 2408 2409 /* 2410 * Process an incoming 'QEMU_VM_COMMAND' 2411 * 0 just a normal return 2412 * LOADVM_QUIT All good, but exit the loop 2413 * <0 Error 2414 */ 2415 static int loadvm_process_command(QEMUFile *f) 2416 { 2417 MigrationIncomingState *mis = migration_incoming_get_current(); 2418 uint16_t cmd; 2419 uint16_t len; 2420 uint32_t tmp32; 2421 2422 cmd = qemu_get_be16(f); 2423 len = qemu_get_be16(f); 2424 2425 /* Check validity before continue processing of cmds */ 2426 if (qemu_file_get_error(f)) { 2427 return qemu_file_get_error(f); 2428 } 2429 2430 if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) { 2431 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len); 2432 return -EINVAL; 2433 } 2434 2435 trace_loadvm_process_command(mig_cmd_args[cmd].name, len); 2436 2437 if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) { 2438 error_report("%s received with bad length - expecting %zu, got %d", 2439 mig_cmd_args[cmd].name, 2440 (size_t)mig_cmd_args[cmd].len, len); 2441 return -ERANGE; 2442 } 2443 2444 switch (cmd) { 2445 case MIG_CMD_OPEN_RETURN_PATH: 2446 if (mis->to_src_file) { 2447 error_report("CMD_OPEN_RETURN_PATH called when RP already open"); 2448 /* Not really a problem, so don't give up */ 2449 return 0; 2450 } 2451 mis->to_src_file = qemu_file_get_return_path(f); 2452 if (!mis->to_src_file) { 2453 error_report("CMD_OPEN_RETURN_PATH failed"); 2454 return -1; 2455 } 2456 2457 /* 2458 * Switchover ack is enabled but no device uses it, so send an ACK to 2459 * source that it's OK to switchover. Do it here, after return path has 2460 * been created. 2461 */ 2462 if (migrate_switchover_ack() && !mis->switchover_ack_pending_num) { 2463 int ret = migrate_send_rp_switchover_ack(mis); 2464 if (ret) { 2465 error_report( 2466 "Could not send switchover ack RP MSG, err %d (%s)", ret, 2467 strerror(-ret)); 2468 return ret; 2469 } 2470 } 2471 break; 2472 2473 case MIG_CMD_PING: 2474 tmp32 = qemu_get_be32(f); 2475 trace_loadvm_process_command_ping(tmp32); 2476 if (!mis->to_src_file) { 2477 error_report("CMD_PING (0x%x) received with no return path", 2478 tmp32); 2479 return -1; 2480 } 2481 migrate_send_rp_pong(mis, tmp32); 2482 break; 2483 2484 case MIG_CMD_PACKAGED: 2485 return loadvm_handle_cmd_packaged(mis); 2486 2487 case MIG_CMD_POSTCOPY_ADVISE: 2488 return loadvm_postcopy_handle_advise(mis, len); 2489 2490 case MIG_CMD_POSTCOPY_LISTEN: 2491 return loadvm_postcopy_handle_listen(mis); 2492 2493 case MIG_CMD_POSTCOPY_RUN: 2494 return loadvm_postcopy_handle_run(mis); 2495 2496 case MIG_CMD_POSTCOPY_RAM_DISCARD: 2497 return loadvm_postcopy_ram_handle_discard(mis, len); 2498 2499 case MIG_CMD_POSTCOPY_RESUME: 2500 return loadvm_postcopy_handle_resume(mis); 2501 2502 case MIG_CMD_RECV_BITMAP: 2503 return loadvm_handle_recv_bitmap(mis, len); 2504 2505 case MIG_CMD_ENABLE_COLO: 2506 return loadvm_process_enable_colo(mis); 2507 } 2508 2509 return 0; 2510 } 2511 2512 /* 2513 * Read a footer off the wire and check that it matches the expected section 2514 * 2515 * Returns: true if the footer was good 2516 * false if there is a problem (and calls error_report to say why) 2517 */ 2518 static bool check_section_footer(QEMUFile *f, SaveStateEntry *se) 2519 { 2520 int ret; 2521 uint8_t read_mark; 2522 uint32_t read_section_id; 2523 2524 if (!migrate_get_current()->send_section_footer) { 2525 /* No footer to check */ 2526 return true; 2527 } 2528 2529 read_mark = qemu_get_byte(f); 2530 2531 ret = qemu_file_get_error(f); 2532 if (ret) { 2533 error_report("%s: Read section footer failed: %d", 2534 __func__, ret); 2535 return false; 2536 } 2537 2538 if (read_mark != QEMU_VM_SECTION_FOOTER) { 2539 error_report("Missing section footer for %s", se->idstr); 2540 return false; 2541 } 2542 2543 read_section_id = qemu_get_be32(f); 2544 if (read_section_id != se->load_section_id) { 2545 error_report("Mismatched section id in footer for %s -" 2546 " read 0x%x expected 0x%x", 2547 se->idstr, read_section_id, se->load_section_id); 2548 return false; 2549 } 2550 2551 /* All good */ 2552 return true; 2553 } 2554 2555 static int 2556 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis, 2557 uint8_t type) 2558 { 2559 bool trace_downtime = (type == QEMU_VM_SECTION_FULL); 2560 uint32_t instance_id, version_id, section_id; 2561 int64_t start_ts, end_ts; 2562 SaveStateEntry *se; 2563 char idstr[256]; 2564 int ret; 2565 2566 /* Read section start */ 2567 section_id = qemu_get_be32(f); 2568 if (!qemu_get_counted_string(f, idstr)) { 2569 error_report("Unable to read ID string for section %u", 2570 section_id); 2571 return -EINVAL; 2572 } 2573 instance_id = qemu_get_be32(f); 2574 version_id = qemu_get_be32(f); 2575 2576 ret = qemu_file_get_error(f); 2577 if (ret) { 2578 error_report("%s: Failed to read instance/version ID: %d", 2579 __func__, ret); 2580 return ret; 2581 } 2582 2583 trace_qemu_loadvm_state_section_startfull(section_id, idstr, 2584 instance_id, version_id); 2585 /* Find savevm section */ 2586 se = find_se(idstr, instance_id); 2587 if (se == NULL) { 2588 error_report("Unknown savevm section or instance '%s' %"PRIu32". " 2589 "Make sure that your current VM setup matches your " 2590 "saved VM setup, including any hotplugged devices", 2591 idstr, instance_id); 2592 return -EINVAL; 2593 } 2594 2595 /* Validate version */ 2596 if (version_id > se->version_id) { 2597 error_report("savevm: unsupported version %d for '%s' v%d", 2598 version_id, idstr, se->version_id); 2599 return -EINVAL; 2600 } 2601 se->load_version_id = version_id; 2602 se->load_section_id = section_id; 2603 2604 /* Validate if it is a device's state */ 2605 if (xen_enabled() && se->is_ram) { 2606 error_report("loadvm: %s RAM loading not allowed on Xen", idstr); 2607 return -EINVAL; 2608 } 2609 2610 if (trace_downtime) { 2611 start_ts = qemu_clock_get_us(QEMU_CLOCK_REALTIME); 2612 } 2613 2614 ret = vmstate_load(f, se); 2615 if (ret < 0) { 2616 error_report("error while loading state for instance 0x%"PRIx32" of" 2617 " device '%s'", instance_id, idstr); 2618 return ret; 2619 } 2620 2621 if (trace_downtime) { 2622 end_ts = qemu_clock_get_us(QEMU_CLOCK_REALTIME); 2623 trace_vmstate_downtime_load("non-iterable", se->idstr, 2624 se->instance_id, end_ts - start_ts); 2625 } 2626 2627 if (!check_section_footer(f, se)) { 2628 return -EINVAL; 2629 } 2630 2631 return 0; 2632 } 2633 2634 static int 2635 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis, 2636 uint8_t type) 2637 { 2638 bool trace_downtime = (type == QEMU_VM_SECTION_END); 2639 int64_t start_ts, end_ts; 2640 uint32_t section_id; 2641 SaveStateEntry *se; 2642 int ret; 2643 2644 section_id = qemu_get_be32(f); 2645 2646 ret = qemu_file_get_error(f); 2647 if (ret) { 2648 error_report("%s: Failed to read section ID: %d", 2649 __func__, ret); 2650 return ret; 2651 } 2652 2653 trace_qemu_loadvm_state_section_partend(section_id); 2654 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 2655 if (se->load_section_id == section_id) { 2656 break; 2657 } 2658 } 2659 if (se == NULL) { 2660 error_report("Unknown savevm section %d", section_id); 2661 return -EINVAL; 2662 } 2663 2664 if (trace_downtime) { 2665 start_ts = qemu_clock_get_us(QEMU_CLOCK_REALTIME); 2666 } 2667 2668 ret = vmstate_load(f, se); 2669 if (ret < 0) { 2670 error_report("error while loading state section id %d(%s)", 2671 section_id, se->idstr); 2672 return ret; 2673 } 2674 2675 if (trace_downtime) { 2676 end_ts = qemu_clock_get_us(QEMU_CLOCK_REALTIME); 2677 trace_vmstate_downtime_load("iterable", se->idstr, 2678 se->instance_id, end_ts - start_ts); 2679 } 2680 2681 if (!check_section_footer(f, se)) { 2682 return -EINVAL; 2683 } 2684 2685 return 0; 2686 } 2687 2688 static int qemu_loadvm_state_header(QEMUFile *f) 2689 { 2690 unsigned int v; 2691 int ret; 2692 2693 v = qemu_get_be32(f); 2694 if (v != QEMU_VM_FILE_MAGIC) { 2695 error_report("Not a migration stream"); 2696 return -EINVAL; 2697 } 2698 2699 v = qemu_get_be32(f); 2700 if (v == QEMU_VM_FILE_VERSION_COMPAT) { 2701 error_report("SaveVM v2 format is obsolete and don't work anymore"); 2702 return -ENOTSUP; 2703 } 2704 if (v != QEMU_VM_FILE_VERSION) { 2705 error_report("Unsupported migration stream version"); 2706 return -ENOTSUP; 2707 } 2708 2709 if (migrate_get_current()->send_configuration) { 2710 if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) { 2711 error_report("Configuration section missing"); 2712 qemu_loadvm_state_cleanup(); 2713 return -EINVAL; 2714 } 2715 ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0); 2716 2717 if (ret) { 2718 qemu_loadvm_state_cleanup(); 2719 return ret; 2720 } 2721 } 2722 return 0; 2723 } 2724 2725 static void qemu_loadvm_state_switchover_ack_needed(MigrationIncomingState *mis) 2726 { 2727 SaveStateEntry *se; 2728 2729 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 2730 if (!se->ops || !se->ops->switchover_ack_needed) { 2731 continue; 2732 } 2733 2734 if (se->ops->switchover_ack_needed(se->opaque)) { 2735 mis->switchover_ack_pending_num++; 2736 } 2737 } 2738 2739 trace_loadvm_state_switchover_ack_needed(mis->switchover_ack_pending_num); 2740 } 2741 2742 static int qemu_loadvm_state_setup(QEMUFile *f) 2743 { 2744 SaveStateEntry *se; 2745 int ret; 2746 2747 trace_loadvm_state_setup(); 2748 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 2749 if (!se->ops || !se->ops->load_setup) { 2750 continue; 2751 } 2752 if (se->ops->is_active) { 2753 if (!se->ops->is_active(se->opaque)) { 2754 continue; 2755 } 2756 } 2757 2758 ret = se->ops->load_setup(f, se->opaque); 2759 if (ret < 0) { 2760 qemu_file_set_error(f, ret); 2761 error_report("Load state of device %s failed", se->idstr); 2762 return ret; 2763 } 2764 } 2765 return 0; 2766 } 2767 2768 void qemu_loadvm_state_cleanup(void) 2769 { 2770 SaveStateEntry *se; 2771 2772 trace_loadvm_state_cleanup(); 2773 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 2774 if (se->ops && se->ops->load_cleanup) { 2775 se->ops->load_cleanup(se->opaque); 2776 } 2777 } 2778 } 2779 2780 /* Return true if we should continue the migration, or false. */ 2781 static bool postcopy_pause_incoming(MigrationIncomingState *mis) 2782 { 2783 int i; 2784 2785 trace_postcopy_pause_incoming(); 2786 2787 assert(migrate_postcopy_ram()); 2788 2789 /* 2790 * Unregister yank with either from/to src would work, since ioc behind it 2791 * is the same 2792 */ 2793 migration_ioc_unregister_yank_from_file(mis->from_src_file); 2794 2795 assert(mis->from_src_file); 2796 qemu_file_shutdown(mis->from_src_file); 2797 qemu_fclose(mis->from_src_file); 2798 mis->from_src_file = NULL; 2799 2800 assert(mis->to_src_file); 2801 qemu_file_shutdown(mis->to_src_file); 2802 qemu_mutex_lock(&mis->rp_mutex); 2803 qemu_fclose(mis->to_src_file); 2804 mis->to_src_file = NULL; 2805 qemu_mutex_unlock(&mis->rp_mutex); 2806 2807 /* 2808 * NOTE: this must happen before reset the PostcopyTmpPages below, 2809 * otherwise it's racy to reset those fields when the fast load thread 2810 * can be accessing it in parallel. 2811 */ 2812 if (mis->postcopy_qemufile_dst) { 2813 qemu_file_shutdown(mis->postcopy_qemufile_dst); 2814 /* Take the mutex to make sure the fast ram load thread halted */ 2815 qemu_mutex_lock(&mis->postcopy_prio_thread_mutex); 2816 migration_ioc_unregister_yank_from_file(mis->postcopy_qemufile_dst); 2817 qemu_fclose(mis->postcopy_qemufile_dst); 2818 mis->postcopy_qemufile_dst = NULL; 2819 qemu_mutex_unlock(&mis->postcopy_prio_thread_mutex); 2820 } 2821 2822 /* Current state can be either ACTIVE or RECOVER */ 2823 migrate_set_state(&mis->state, mis->state, 2824 MIGRATION_STATUS_POSTCOPY_PAUSED); 2825 2826 /* Notify the fault thread for the invalidated file handle */ 2827 postcopy_fault_thread_notify(mis); 2828 2829 /* 2830 * If network is interrupted, any temp page we received will be useless 2831 * because we didn't mark them as "received" in receivedmap. After a 2832 * proper recovery later (which will sync src dirty bitmap with receivedmap 2833 * on dest) these cached small pages will be resent again. 2834 */ 2835 for (i = 0; i < mis->postcopy_channels; i++) { 2836 postcopy_temp_page_reset(&mis->postcopy_tmp_pages[i]); 2837 } 2838 2839 error_report("Detected IO failure for postcopy. " 2840 "Migration paused."); 2841 2842 while (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { 2843 qemu_sem_wait(&mis->postcopy_pause_sem_dst); 2844 } 2845 2846 trace_postcopy_pause_incoming_continued(); 2847 2848 return true; 2849 } 2850 2851 int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis) 2852 { 2853 uint8_t section_type; 2854 int ret = 0; 2855 2856 retry: 2857 while (true) { 2858 section_type = qemu_get_byte(f); 2859 2860 ret = qemu_file_get_error_obj_any(f, mis->postcopy_qemufile_dst, NULL); 2861 if (ret) { 2862 break; 2863 } 2864 2865 trace_qemu_loadvm_state_section(section_type); 2866 switch (section_type) { 2867 case QEMU_VM_SECTION_START: 2868 case QEMU_VM_SECTION_FULL: 2869 ret = qemu_loadvm_section_start_full(f, mis, section_type); 2870 if (ret < 0) { 2871 goto out; 2872 } 2873 break; 2874 case QEMU_VM_SECTION_PART: 2875 case QEMU_VM_SECTION_END: 2876 ret = qemu_loadvm_section_part_end(f, mis, section_type); 2877 if (ret < 0) { 2878 goto out; 2879 } 2880 break; 2881 case QEMU_VM_COMMAND: 2882 ret = loadvm_process_command(f); 2883 trace_qemu_loadvm_state_section_command(ret); 2884 if ((ret < 0) || (ret == LOADVM_QUIT)) { 2885 goto out; 2886 } 2887 break; 2888 case QEMU_VM_EOF: 2889 /* This is the end of migration */ 2890 goto out; 2891 default: 2892 error_report("Unknown savevm section type %d", section_type); 2893 ret = -EINVAL; 2894 goto out; 2895 } 2896 } 2897 2898 out: 2899 if (ret < 0) { 2900 qemu_file_set_error(f, ret); 2901 2902 /* Cancel bitmaps incoming regardless of recovery */ 2903 dirty_bitmap_mig_cancel_incoming(); 2904 2905 /* 2906 * If we are during an active postcopy, then we pause instead 2907 * of bail out to at least keep the VM's dirty data. Note 2908 * that POSTCOPY_INCOMING_LISTENING stage is still not enough, 2909 * during which we're still receiving device states and we 2910 * still haven't yet started the VM on destination. 2911 * 2912 * Only RAM postcopy supports recovery. Still, if RAM postcopy is 2913 * enabled, canceled bitmaps postcopy will not affect RAM postcopy 2914 * recovering. 2915 */ 2916 if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING && 2917 migrate_postcopy_ram() && postcopy_pause_incoming(mis)) { 2918 /* Reset f to point to the newly created channel */ 2919 f = mis->from_src_file; 2920 goto retry; 2921 } 2922 } 2923 return ret; 2924 } 2925 2926 int qemu_loadvm_state(QEMUFile *f) 2927 { 2928 MigrationIncomingState *mis = migration_incoming_get_current(); 2929 Error *local_err = NULL; 2930 int ret; 2931 2932 if (qemu_savevm_state_blocked(&local_err)) { 2933 error_report_err(local_err); 2934 return -EINVAL; 2935 } 2936 2937 ret = qemu_loadvm_state_header(f); 2938 if (ret) { 2939 return ret; 2940 } 2941 2942 if (qemu_loadvm_state_setup(f) != 0) { 2943 return -EINVAL; 2944 } 2945 2946 if (migrate_switchover_ack()) { 2947 qemu_loadvm_state_switchover_ack_needed(mis); 2948 } 2949 2950 cpu_synchronize_all_pre_loadvm(); 2951 2952 ret = qemu_loadvm_state_main(f, mis); 2953 qemu_event_set(&mis->main_thread_load_event); 2954 2955 trace_qemu_loadvm_state_post_main(ret); 2956 2957 if (mis->have_listen_thread) { 2958 /* Listen thread still going, can't clean up yet */ 2959 return ret; 2960 } 2961 2962 if (ret == 0) { 2963 ret = qemu_file_get_error(f); 2964 } 2965 2966 /* 2967 * Try to read in the VMDESC section as well, so that dumping tools that 2968 * intercept our migration stream have the chance to see it. 2969 */ 2970 2971 /* We've got to be careful; if we don't read the data and just shut the fd 2972 * then the sender can error if we close while it's still sending. 2973 * We also mustn't read data that isn't there; some transports (RDMA) 2974 * will stall waiting for that data when the source has already closed. 2975 */ 2976 if (ret == 0 && should_send_vmdesc()) { 2977 uint8_t *buf; 2978 uint32_t size; 2979 uint8_t section_type = qemu_get_byte(f); 2980 2981 if (section_type != QEMU_VM_VMDESCRIPTION) { 2982 error_report("Expected vmdescription section, but got %d", 2983 section_type); 2984 /* 2985 * It doesn't seem worth failing at this point since 2986 * we apparently have an otherwise valid VM state 2987 */ 2988 } else { 2989 buf = g_malloc(0x1000); 2990 size = qemu_get_be32(f); 2991 2992 while (size > 0) { 2993 uint32_t read_chunk = MIN(size, 0x1000); 2994 qemu_get_buffer(f, buf, read_chunk); 2995 size -= read_chunk; 2996 } 2997 g_free(buf); 2998 } 2999 } 3000 3001 qemu_loadvm_state_cleanup(); 3002 cpu_synchronize_all_post_init(); 3003 3004 return ret; 3005 } 3006 3007 int qemu_load_device_state(QEMUFile *f) 3008 { 3009 MigrationIncomingState *mis = migration_incoming_get_current(); 3010 int ret; 3011 3012 /* Load QEMU_VM_SECTION_FULL section */ 3013 ret = qemu_loadvm_state_main(f, mis); 3014 if (ret < 0) { 3015 error_report("Failed to load device state: %d", ret); 3016 return ret; 3017 } 3018 3019 cpu_synchronize_all_post_init(); 3020 return 0; 3021 } 3022 3023 int qemu_loadvm_approve_switchover(void) 3024 { 3025 MigrationIncomingState *mis = migration_incoming_get_current(); 3026 3027 if (!mis->switchover_ack_pending_num) { 3028 return -EINVAL; 3029 } 3030 3031 mis->switchover_ack_pending_num--; 3032 trace_loadvm_approve_switchover(mis->switchover_ack_pending_num); 3033 3034 if (mis->switchover_ack_pending_num) { 3035 return 0; 3036 } 3037 3038 return migrate_send_rp_switchover_ack(mis); 3039 } 3040 3041 bool save_snapshot(const char *name, bool overwrite, const char *vmstate, 3042 bool has_devices, strList *devices, Error **errp) 3043 { 3044 BlockDriverState *bs; 3045 QEMUSnapshotInfo sn1, *sn = &sn1; 3046 int ret = -1, ret2; 3047 QEMUFile *f; 3048 RunState saved_state = runstate_get(); 3049 uint64_t vm_state_size; 3050 g_autoptr(GDateTime) now = g_date_time_new_now_local(); 3051 3052 GLOBAL_STATE_CODE(); 3053 3054 if (migration_is_blocked(errp)) { 3055 return false; 3056 } 3057 3058 if (!replay_can_snapshot()) { 3059 error_setg(errp, "Record/replay does not allow making snapshot " 3060 "right now. Try once more later."); 3061 return false; 3062 } 3063 3064 if (!bdrv_all_can_snapshot(has_devices, devices, errp)) { 3065 return false; 3066 } 3067 3068 /* Delete old snapshots of the same name */ 3069 if (name) { 3070 if (overwrite) { 3071 if (bdrv_all_delete_snapshot(name, has_devices, 3072 devices, errp) < 0) { 3073 return false; 3074 } 3075 } else { 3076 ret2 = bdrv_all_has_snapshot(name, has_devices, devices, errp); 3077 if (ret2 < 0) { 3078 return false; 3079 } 3080 if (ret2 == 1) { 3081 error_setg(errp, 3082 "Snapshot '%s' already exists in one or more devices", 3083 name); 3084 return false; 3085 } 3086 } 3087 } 3088 3089 bs = bdrv_all_find_vmstate_bs(vmstate, has_devices, devices, errp); 3090 if (bs == NULL) { 3091 return false; 3092 } 3093 3094 global_state_store(); 3095 vm_stop(RUN_STATE_SAVE_VM); 3096 3097 bdrv_drain_all_begin(); 3098 3099 memset(sn, 0, sizeof(*sn)); 3100 3101 /* fill auxiliary fields */ 3102 sn->date_sec = g_date_time_to_unix(now); 3103 sn->date_nsec = g_date_time_get_microsecond(now) * 1000; 3104 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL); 3105 if (replay_mode != REPLAY_MODE_NONE) { 3106 sn->icount = replay_get_current_icount(); 3107 } else { 3108 sn->icount = -1ULL; 3109 } 3110 3111 if (name) { 3112 pstrcpy(sn->name, sizeof(sn->name), name); 3113 } else { 3114 g_autofree char *autoname = g_date_time_format(now, "vm-%Y%m%d%H%M%S"); 3115 pstrcpy(sn->name, sizeof(sn->name), autoname); 3116 } 3117 3118 /* save the VM state */ 3119 f = qemu_fopen_bdrv(bs, 1); 3120 if (!f) { 3121 error_setg(errp, "Could not open VM state file"); 3122 goto the_end; 3123 } 3124 ret = qemu_savevm_state(f, errp); 3125 vm_state_size = qemu_file_transferred(f); 3126 ret2 = qemu_fclose(f); 3127 if (ret < 0) { 3128 goto the_end; 3129 } 3130 if (ret2 < 0) { 3131 ret = ret2; 3132 goto the_end; 3133 } 3134 3135 ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, 3136 has_devices, devices, errp); 3137 if (ret < 0) { 3138 bdrv_all_delete_snapshot(sn->name, has_devices, devices, NULL); 3139 goto the_end; 3140 } 3141 3142 ret = 0; 3143 3144 the_end: 3145 bdrv_drain_all_end(); 3146 3147 vm_resume(saved_state); 3148 return ret == 0; 3149 } 3150 3151 void qmp_xen_save_devices_state(const char *filename, bool has_live, bool live, 3152 Error **errp) 3153 { 3154 QEMUFile *f; 3155 QIOChannelFile *ioc; 3156 int saved_vm_running; 3157 int ret; 3158 3159 if (!has_live) { 3160 /* live default to true so old version of Xen tool stack can have a 3161 * successful live migration */ 3162 live = true; 3163 } 3164 3165 saved_vm_running = runstate_is_running(); 3166 vm_stop(RUN_STATE_SAVE_VM); 3167 global_state_store_running(); 3168 3169 ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT | O_TRUNC, 3170 0660, errp); 3171 if (!ioc) { 3172 goto the_end; 3173 } 3174 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state"); 3175 f = qemu_file_new_output(QIO_CHANNEL(ioc)); 3176 object_unref(OBJECT(ioc)); 3177 ret = qemu_save_device_state(f); 3178 if (ret < 0 || qemu_fclose(f) < 0) { 3179 error_setg(errp, QERR_IO_ERROR); 3180 } else { 3181 /* libxl calls the QMP command "stop" before calling 3182 * "xen-save-devices-state" and in case of migration failure, libxl 3183 * would call "cont". 3184 * So call bdrv_inactivate_all (release locks) here to let the other 3185 * side of the migration take control of the images. 3186 */ 3187 if (live && !saved_vm_running) { 3188 ret = bdrv_inactivate_all(); 3189 if (ret) { 3190 error_setg(errp, "%s: bdrv_inactivate_all() failed (%d)", 3191 __func__, ret); 3192 } 3193 } 3194 } 3195 3196 the_end: 3197 if (saved_vm_running) { 3198 vm_start(); 3199 } 3200 } 3201 3202 void qmp_xen_load_devices_state(const char *filename, Error **errp) 3203 { 3204 QEMUFile *f; 3205 QIOChannelFile *ioc; 3206 int ret; 3207 3208 /* Guest must be paused before loading the device state; the RAM state 3209 * will already have been loaded by xc 3210 */ 3211 if (runstate_is_running()) { 3212 error_setg(errp, "Cannot update device state while vm is running"); 3213 return; 3214 } 3215 vm_stop(RUN_STATE_RESTORE_VM); 3216 3217 ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp); 3218 if (!ioc) { 3219 return; 3220 } 3221 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state"); 3222 f = qemu_file_new_input(QIO_CHANNEL(ioc)); 3223 object_unref(OBJECT(ioc)); 3224 3225 ret = qemu_loadvm_state(f); 3226 qemu_fclose(f); 3227 if (ret < 0) { 3228 error_setg(errp, QERR_IO_ERROR); 3229 } 3230 migration_incoming_state_destroy(); 3231 } 3232 3233 bool load_snapshot(const char *name, const char *vmstate, 3234 bool has_devices, strList *devices, Error **errp) 3235 { 3236 BlockDriverState *bs_vm_state; 3237 QEMUSnapshotInfo sn; 3238 QEMUFile *f; 3239 int ret; 3240 MigrationIncomingState *mis = migration_incoming_get_current(); 3241 3242 if (!bdrv_all_can_snapshot(has_devices, devices, errp)) { 3243 return false; 3244 } 3245 ret = bdrv_all_has_snapshot(name, has_devices, devices, errp); 3246 if (ret < 0) { 3247 return false; 3248 } 3249 if (ret == 0) { 3250 error_setg(errp, "Snapshot '%s' does not exist in one or more devices", 3251 name); 3252 return false; 3253 } 3254 3255 bs_vm_state = bdrv_all_find_vmstate_bs(vmstate, has_devices, devices, errp); 3256 if (!bs_vm_state) { 3257 return false; 3258 } 3259 3260 /* Don't even try to load empty VM states */ 3261 ret = bdrv_snapshot_find(bs_vm_state, &sn, name); 3262 if (ret < 0) { 3263 return false; 3264 } else if (sn.vm_state_size == 0) { 3265 error_setg(errp, "This is a disk-only snapshot. Revert to it " 3266 " offline using qemu-img"); 3267 return false; 3268 } 3269 3270 /* 3271 * Flush the record/replay queue. Now the VM state is going 3272 * to change. Therefore we don't need to preserve its consistency 3273 */ 3274 replay_flush_events(); 3275 3276 /* Flush all IO requests so they don't interfere with the new state. */ 3277 bdrv_drain_all_begin(); 3278 3279 ret = bdrv_all_goto_snapshot(name, has_devices, devices, errp); 3280 if (ret < 0) { 3281 goto err_drain; 3282 } 3283 3284 /* restore the VM state */ 3285 f = qemu_fopen_bdrv(bs_vm_state, 0); 3286 if (!f) { 3287 error_setg(errp, "Could not open VM state file"); 3288 goto err_drain; 3289 } 3290 3291 qemu_system_reset(SHUTDOWN_CAUSE_SNAPSHOT_LOAD); 3292 mis->from_src_file = f; 3293 3294 if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) { 3295 ret = -EINVAL; 3296 goto err_drain; 3297 } 3298 ret = qemu_loadvm_state(f); 3299 migration_incoming_state_destroy(); 3300 3301 bdrv_drain_all_end(); 3302 3303 if (ret < 0) { 3304 error_setg(errp, "Error %d while loading VM state", ret); 3305 return false; 3306 } 3307 3308 return true; 3309 3310 err_drain: 3311 bdrv_drain_all_end(); 3312 return false; 3313 } 3314 3315 void load_snapshot_resume(RunState state) 3316 { 3317 vm_resume(state); 3318 if (state == RUN_STATE_RUNNING && runstate_get() == RUN_STATE_SUSPENDED) { 3319 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, &error_abort); 3320 } 3321 } 3322 3323 bool delete_snapshot(const char *name, bool has_devices, 3324 strList *devices, Error **errp) 3325 { 3326 if (!bdrv_all_can_snapshot(has_devices, devices, errp)) { 3327 return false; 3328 } 3329 3330 if (bdrv_all_delete_snapshot(name, has_devices, devices, errp) < 0) { 3331 return false; 3332 } 3333 3334 return true; 3335 } 3336 3337 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev) 3338 { 3339 qemu_ram_set_idstr(mr->ram_block, 3340 memory_region_name(mr), dev); 3341 qemu_ram_set_migratable(mr->ram_block); 3342 } 3343 3344 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev) 3345 { 3346 qemu_ram_unset_idstr(mr->ram_block); 3347 qemu_ram_unset_migratable(mr->ram_block); 3348 } 3349 3350 void vmstate_register_ram_global(MemoryRegion *mr) 3351 { 3352 vmstate_register_ram(mr, NULL); 3353 } 3354 3355 bool vmstate_check_only_migratable(const VMStateDescription *vmsd) 3356 { 3357 /* check needed if --only-migratable is specified */ 3358 if (!only_migratable) { 3359 return true; 3360 } 3361 3362 return !(vmsd && vmsd->unmigratable); 3363 } 3364 3365 typedef struct SnapshotJob { 3366 Job common; 3367 char *tag; 3368 char *vmstate; 3369 strList *devices; 3370 Coroutine *co; 3371 Error **errp; 3372 bool ret; 3373 } SnapshotJob; 3374 3375 static void qmp_snapshot_job_free(SnapshotJob *s) 3376 { 3377 g_free(s->tag); 3378 g_free(s->vmstate); 3379 qapi_free_strList(s->devices); 3380 } 3381 3382 3383 static void snapshot_load_job_bh(void *opaque) 3384 { 3385 Job *job = opaque; 3386 SnapshotJob *s = container_of(job, SnapshotJob, common); 3387 RunState orig_state = runstate_get(); 3388 3389 job_progress_set_remaining(&s->common, 1); 3390 3391 vm_stop(RUN_STATE_RESTORE_VM); 3392 3393 s->ret = load_snapshot(s->tag, s->vmstate, true, s->devices, s->errp); 3394 if (s->ret) { 3395 load_snapshot_resume(orig_state); 3396 } 3397 3398 job_progress_update(&s->common, 1); 3399 3400 qmp_snapshot_job_free(s); 3401 aio_co_wake(s->co); 3402 } 3403 3404 static void snapshot_save_job_bh(void *opaque) 3405 { 3406 Job *job = opaque; 3407 SnapshotJob *s = container_of(job, SnapshotJob, common); 3408 3409 job_progress_set_remaining(&s->common, 1); 3410 s->ret = save_snapshot(s->tag, false, s->vmstate, 3411 true, s->devices, s->errp); 3412 job_progress_update(&s->common, 1); 3413 3414 qmp_snapshot_job_free(s); 3415 aio_co_wake(s->co); 3416 } 3417 3418 static void snapshot_delete_job_bh(void *opaque) 3419 { 3420 Job *job = opaque; 3421 SnapshotJob *s = container_of(job, SnapshotJob, common); 3422 3423 job_progress_set_remaining(&s->common, 1); 3424 s->ret = delete_snapshot(s->tag, true, s->devices, s->errp); 3425 job_progress_update(&s->common, 1); 3426 3427 qmp_snapshot_job_free(s); 3428 aio_co_wake(s->co); 3429 } 3430 3431 static int coroutine_fn snapshot_save_job_run(Job *job, Error **errp) 3432 { 3433 SnapshotJob *s = container_of(job, SnapshotJob, common); 3434 s->errp = errp; 3435 s->co = qemu_coroutine_self(); 3436 aio_bh_schedule_oneshot(qemu_get_aio_context(), 3437 snapshot_save_job_bh, job); 3438 qemu_coroutine_yield(); 3439 return s->ret ? 0 : -1; 3440 } 3441 3442 static int coroutine_fn snapshot_load_job_run(Job *job, Error **errp) 3443 { 3444 SnapshotJob *s = container_of(job, SnapshotJob, common); 3445 s->errp = errp; 3446 s->co = qemu_coroutine_self(); 3447 aio_bh_schedule_oneshot(qemu_get_aio_context(), 3448 snapshot_load_job_bh, job); 3449 qemu_coroutine_yield(); 3450 return s->ret ? 0 : -1; 3451 } 3452 3453 static int coroutine_fn snapshot_delete_job_run(Job *job, Error **errp) 3454 { 3455 SnapshotJob *s = container_of(job, SnapshotJob, common); 3456 s->errp = errp; 3457 s->co = qemu_coroutine_self(); 3458 aio_bh_schedule_oneshot(qemu_get_aio_context(), 3459 snapshot_delete_job_bh, job); 3460 qemu_coroutine_yield(); 3461 return s->ret ? 0 : -1; 3462 } 3463 3464 3465 static const JobDriver snapshot_load_job_driver = { 3466 .instance_size = sizeof(SnapshotJob), 3467 .job_type = JOB_TYPE_SNAPSHOT_LOAD, 3468 .run = snapshot_load_job_run, 3469 }; 3470 3471 static const JobDriver snapshot_save_job_driver = { 3472 .instance_size = sizeof(SnapshotJob), 3473 .job_type = JOB_TYPE_SNAPSHOT_SAVE, 3474 .run = snapshot_save_job_run, 3475 }; 3476 3477 static const JobDriver snapshot_delete_job_driver = { 3478 .instance_size = sizeof(SnapshotJob), 3479 .job_type = JOB_TYPE_SNAPSHOT_DELETE, 3480 .run = snapshot_delete_job_run, 3481 }; 3482 3483 3484 void qmp_snapshot_save(const char *job_id, 3485 const char *tag, 3486 const char *vmstate, 3487 strList *devices, 3488 Error **errp) 3489 { 3490 SnapshotJob *s; 3491 3492 s = job_create(job_id, &snapshot_save_job_driver, NULL, 3493 qemu_get_aio_context(), JOB_MANUAL_DISMISS, 3494 NULL, NULL, errp); 3495 if (!s) { 3496 return; 3497 } 3498 3499 s->tag = g_strdup(tag); 3500 s->vmstate = g_strdup(vmstate); 3501 s->devices = QAPI_CLONE(strList, devices); 3502 3503 job_start(&s->common); 3504 } 3505 3506 void qmp_snapshot_load(const char *job_id, 3507 const char *tag, 3508 const char *vmstate, 3509 strList *devices, 3510 Error **errp) 3511 { 3512 SnapshotJob *s; 3513 3514 s = job_create(job_id, &snapshot_load_job_driver, NULL, 3515 qemu_get_aio_context(), JOB_MANUAL_DISMISS, 3516 NULL, NULL, errp); 3517 if (!s) { 3518 return; 3519 } 3520 3521 s->tag = g_strdup(tag); 3522 s->vmstate = g_strdup(vmstate); 3523 s->devices = QAPI_CLONE(strList, devices); 3524 3525 job_start(&s->common); 3526 } 3527 3528 void qmp_snapshot_delete(const char *job_id, 3529 const char *tag, 3530 strList *devices, 3531 Error **errp) 3532 { 3533 SnapshotJob *s; 3534 3535 s = job_create(job_id, &snapshot_delete_job_driver, NULL, 3536 qemu_get_aio_context(), JOB_MANUAL_DISMISS, 3537 NULL, NULL, errp); 3538 if (!s) { 3539 return; 3540 } 3541 3542 s->tag = g_strdup(tag); 3543 s->devices = QAPI_CLONE(strList, devices); 3544 3545 job_start(&s->common); 3546 } 3547