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 Error **errp) 1014 { 1015 int ret; 1016 1017 if ((!se->ops || !se->ops->save_state) && !se->vmsd) { 1018 return 0; 1019 } 1020 if (se->vmsd && !vmstate_section_needed(se->vmsd, se->opaque)) { 1021 trace_savevm_section_skip(se->idstr, se->section_id); 1022 return 0; 1023 } 1024 1025 trace_savevm_section_start(se->idstr, se->section_id); 1026 save_section_header(f, se, QEMU_VM_SECTION_FULL); 1027 if (vmdesc) { 1028 json_writer_start_object(vmdesc, NULL); 1029 json_writer_str(vmdesc, "name", se->idstr); 1030 json_writer_int64(vmdesc, "instance_id", se->instance_id); 1031 } 1032 1033 trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)"); 1034 if (!se->vmsd) { 1035 vmstate_save_old_style(f, se, vmdesc); 1036 } else { 1037 ret = vmstate_save_state_with_err(f, se->vmsd, se->opaque, vmdesc, 1038 errp); 1039 if (ret) { 1040 return ret; 1041 } 1042 } 1043 1044 trace_savevm_section_end(se->idstr, se->section_id, 0); 1045 save_section_footer(f, se); 1046 if (vmdesc) { 1047 json_writer_end_object(vmdesc); 1048 } 1049 return 0; 1050 } 1051 /** 1052 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the 1053 * command and associated data. 1054 * 1055 * @f: File to send command on 1056 * @command: Command type to send 1057 * @len: Length of associated data 1058 * @data: Data associated with command. 1059 */ 1060 static void qemu_savevm_command_send(QEMUFile *f, 1061 enum qemu_vm_cmd command, 1062 uint16_t len, 1063 uint8_t *data) 1064 { 1065 trace_savevm_command_send(command, len); 1066 qemu_put_byte(f, QEMU_VM_COMMAND); 1067 qemu_put_be16(f, (uint16_t)command); 1068 qemu_put_be16(f, len); 1069 qemu_put_buffer(f, data, len); 1070 qemu_fflush(f); 1071 } 1072 1073 void qemu_savevm_send_colo_enable(QEMUFile *f) 1074 { 1075 trace_savevm_send_colo_enable(); 1076 qemu_savevm_command_send(f, MIG_CMD_ENABLE_COLO, 0, NULL); 1077 } 1078 1079 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value) 1080 { 1081 uint32_t buf; 1082 1083 trace_savevm_send_ping(value); 1084 buf = cpu_to_be32(value); 1085 qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf); 1086 } 1087 1088 void qemu_savevm_send_open_return_path(QEMUFile *f) 1089 { 1090 trace_savevm_send_open_return_path(); 1091 qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL); 1092 } 1093 1094 /* We have a buffer of data to send; we don't want that all to be loaded 1095 * by the command itself, so the command contains just the length of the 1096 * extra buffer that we then send straight after it. 1097 * TODO: Must be a better way to organise that 1098 * 1099 * Returns: 1100 * 0 on success 1101 * -ve on error 1102 */ 1103 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len) 1104 { 1105 uint32_t tmp; 1106 MigrationState *ms = migrate_get_current(); 1107 Error *local_err = NULL; 1108 1109 if (len > MAX_VM_CMD_PACKAGED_SIZE) { 1110 error_setg(&local_err, "%s: Unreasonably large packaged state: %zu", 1111 __func__, len); 1112 migrate_set_error(ms, local_err); 1113 error_report_err(local_err); 1114 return -1; 1115 } 1116 1117 tmp = cpu_to_be32(len); 1118 1119 trace_qemu_savevm_send_packaged(); 1120 qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp); 1121 1122 qemu_put_buffer(f, buf, len); 1123 1124 return 0; 1125 } 1126 1127 /* Send prior to any postcopy transfer */ 1128 void qemu_savevm_send_postcopy_advise(QEMUFile *f) 1129 { 1130 if (migrate_postcopy_ram()) { 1131 uint64_t tmp[2]; 1132 tmp[0] = cpu_to_be64(ram_pagesize_summary()); 1133 tmp[1] = cpu_to_be64(qemu_target_page_size()); 1134 1135 trace_qemu_savevm_send_postcopy_advise(); 1136 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 1137 16, (uint8_t *)tmp); 1138 } else { 1139 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 0, NULL); 1140 } 1141 } 1142 1143 /* Sent prior to starting the destination running in postcopy, discard pages 1144 * that have already been sent but redirtied on the source. 1145 * CMD_POSTCOPY_RAM_DISCARD consist of: 1146 * byte version (0) 1147 * byte Length of name field (not including 0) 1148 * n x byte RAM block name 1149 * byte 0 terminator (just for safety) 1150 * n x Byte ranges within the named RAMBlock 1151 * be64 Start of the range 1152 * be64 Length 1153 * 1154 * name: RAMBlock name that these entries are part of 1155 * len: Number of page entries 1156 * start_list: 'len' addresses 1157 * length_list: 'len' addresses 1158 * 1159 */ 1160 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name, 1161 uint16_t len, 1162 uint64_t *start_list, 1163 uint64_t *length_list) 1164 { 1165 uint8_t *buf; 1166 uint16_t tmplen; 1167 uint16_t t; 1168 size_t name_len = strlen(name); 1169 1170 trace_qemu_savevm_send_postcopy_ram_discard(name, len); 1171 assert(name_len < 256); 1172 buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len); 1173 buf[0] = postcopy_ram_discard_version; 1174 buf[1] = name_len; 1175 memcpy(buf + 2, name, name_len); 1176 tmplen = 2 + name_len; 1177 buf[tmplen++] = '\0'; 1178 1179 for (t = 0; t < len; t++) { 1180 stq_be_p(buf + tmplen, start_list[t]); 1181 tmplen += 8; 1182 stq_be_p(buf + tmplen, length_list[t]); 1183 tmplen += 8; 1184 } 1185 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf); 1186 g_free(buf); 1187 } 1188 1189 /* Get the destination into a state where it can receive postcopy data. */ 1190 void qemu_savevm_send_postcopy_listen(QEMUFile *f) 1191 { 1192 trace_savevm_send_postcopy_listen(); 1193 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL); 1194 } 1195 1196 /* Kick the destination into running */ 1197 void qemu_savevm_send_postcopy_run(QEMUFile *f) 1198 { 1199 trace_savevm_send_postcopy_run(); 1200 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL); 1201 } 1202 1203 void qemu_savevm_send_postcopy_resume(QEMUFile *f) 1204 { 1205 trace_savevm_send_postcopy_resume(); 1206 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RESUME, 0, NULL); 1207 } 1208 1209 void qemu_savevm_send_recv_bitmap(QEMUFile *f, char *block_name) 1210 { 1211 size_t len; 1212 char buf[256]; 1213 1214 trace_savevm_send_recv_bitmap(block_name); 1215 1216 buf[0] = len = strlen(block_name); 1217 memcpy(buf + 1, block_name, len); 1218 1219 qemu_savevm_command_send(f, MIG_CMD_RECV_BITMAP, len + 1, (uint8_t *)buf); 1220 } 1221 1222 bool qemu_savevm_state_blocked(Error **errp) 1223 { 1224 SaveStateEntry *se; 1225 1226 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1227 if (se->vmsd && se->vmsd->unmigratable) { 1228 error_setg(errp, "State blocked by non-migratable device '%s'", 1229 se->idstr); 1230 return true; 1231 } 1232 } 1233 return false; 1234 } 1235 1236 void qemu_savevm_non_migratable_list(strList **reasons) 1237 { 1238 SaveStateEntry *se; 1239 1240 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1241 if (se->vmsd && se->vmsd->unmigratable) { 1242 QAPI_LIST_PREPEND(*reasons, 1243 g_strdup_printf("non-migratable device: %s", 1244 se->idstr)); 1245 } 1246 } 1247 } 1248 1249 void qemu_savevm_state_header(QEMUFile *f) 1250 { 1251 MigrationState *s = migrate_get_current(); 1252 1253 s->vmdesc = json_writer_new(false); 1254 1255 trace_savevm_state_header(); 1256 qemu_put_be32(f, QEMU_VM_FILE_MAGIC); 1257 qemu_put_be32(f, QEMU_VM_FILE_VERSION); 1258 1259 if (s->send_configuration) { 1260 qemu_put_byte(f, QEMU_VM_CONFIGURATION); 1261 1262 /* 1263 * This starts the main json object and is paired with the 1264 * json_writer_end_object in 1265 * qemu_savevm_state_complete_precopy_non_iterable 1266 */ 1267 json_writer_start_object(s->vmdesc, NULL); 1268 1269 json_writer_start_object(s->vmdesc, "configuration"); 1270 vmstate_save_state(f, &vmstate_configuration, &savevm_state, s->vmdesc); 1271 json_writer_end_object(s->vmdesc); 1272 } 1273 } 1274 1275 bool qemu_savevm_state_guest_unplug_pending(void) 1276 { 1277 SaveStateEntry *se; 1278 1279 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1280 if (se->vmsd && se->vmsd->dev_unplug_pending && 1281 se->vmsd->dev_unplug_pending(se->opaque)) { 1282 return true; 1283 } 1284 } 1285 1286 return false; 1287 } 1288 1289 int qemu_savevm_state_prepare(Error **errp) 1290 { 1291 SaveStateEntry *se; 1292 int ret; 1293 1294 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1295 if (!se->ops || !se->ops->save_prepare) { 1296 continue; 1297 } 1298 if (se->ops->is_active) { 1299 if (!se->ops->is_active(se->opaque)) { 1300 continue; 1301 } 1302 } 1303 1304 ret = se->ops->save_prepare(se->opaque, errp); 1305 if (ret < 0) { 1306 return ret; 1307 } 1308 } 1309 1310 return 0; 1311 } 1312 1313 int qemu_savevm_state_setup(QEMUFile *f, Error **errp) 1314 { 1315 ERRP_GUARD(); 1316 MigrationState *ms = migrate_get_current(); 1317 SaveStateEntry *se; 1318 int ret = 0; 1319 1320 json_writer_int64(ms->vmdesc, "page_size", qemu_target_page_size()); 1321 json_writer_start_array(ms->vmdesc, "devices"); 1322 1323 trace_savevm_state_setup(); 1324 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1325 if (se->vmsd && se->vmsd->early_setup) { 1326 ret = vmstate_save(f, se, ms->vmdesc, errp); 1327 if (ret) { 1328 migrate_set_error(ms, *errp); 1329 qemu_file_set_error(f, ret); 1330 break; 1331 } 1332 continue; 1333 } 1334 1335 if (!se->ops || !se->ops->save_setup) { 1336 continue; 1337 } 1338 if (se->ops->is_active) { 1339 if (!se->ops->is_active(se->opaque)) { 1340 continue; 1341 } 1342 } 1343 save_section_header(f, se, QEMU_VM_SECTION_START); 1344 1345 ret = se->ops->save_setup(f, se->opaque, errp); 1346 save_section_footer(f, se); 1347 if (ret < 0) { 1348 qemu_file_set_error(f, ret); 1349 break; 1350 } 1351 } 1352 1353 if (ret) { 1354 return ret; 1355 } 1356 1357 /* TODO: Should we check that errp is set in case of failure ? */ 1358 return precopy_notify(PRECOPY_NOTIFY_SETUP, errp); 1359 } 1360 1361 int qemu_savevm_state_resume_prepare(MigrationState *s) 1362 { 1363 SaveStateEntry *se; 1364 int ret; 1365 1366 trace_savevm_state_resume_prepare(); 1367 1368 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1369 if (!se->ops || !se->ops->resume_prepare) { 1370 continue; 1371 } 1372 if (se->ops->is_active) { 1373 if (!se->ops->is_active(se->opaque)) { 1374 continue; 1375 } 1376 } 1377 ret = se->ops->resume_prepare(s, se->opaque); 1378 if (ret < 0) { 1379 return ret; 1380 } 1381 } 1382 1383 return 0; 1384 } 1385 1386 /* 1387 * this function has three return values: 1388 * negative: there was one error, and we have -errno. 1389 * 0 : We haven't finished, caller have to go again 1390 * 1 : We have finished, we can go to complete phase 1391 */ 1392 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy) 1393 { 1394 SaveStateEntry *se; 1395 bool all_finished = true; 1396 int ret; 1397 1398 trace_savevm_state_iterate(); 1399 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1400 if (!se->ops || !se->ops->save_live_iterate) { 1401 continue; 1402 } 1403 if (se->ops->is_active && 1404 !se->ops->is_active(se->opaque)) { 1405 continue; 1406 } 1407 if (se->ops->is_active_iterate && 1408 !se->ops->is_active_iterate(se->opaque)) { 1409 continue; 1410 } 1411 /* 1412 * In the postcopy phase, any device that doesn't know how to 1413 * do postcopy should have saved it's state in the _complete 1414 * call that's already run, it might get confused if we call 1415 * iterate afterwards. 1416 */ 1417 if (postcopy && 1418 !(se->ops->has_postcopy && se->ops->has_postcopy(se->opaque))) { 1419 continue; 1420 } 1421 if (migration_rate_exceeded(f)) { 1422 return 0; 1423 } 1424 trace_savevm_section_start(se->idstr, se->section_id); 1425 1426 save_section_header(f, se, QEMU_VM_SECTION_PART); 1427 1428 ret = se->ops->save_live_iterate(f, se->opaque); 1429 trace_savevm_section_end(se->idstr, se->section_id, ret); 1430 save_section_footer(f, se); 1431 1432 if (ret < 0) { 1433 error_report("failed to save SaveStateEntry with id(name): " 1434 "%d(%s): %d", 1435 se->section_id, se->idstr, ret); 1436 qemu_file_set_error(f, ret); 1437 return ret; 1438 } else if (!ret) { 1439 all_finished = false; 1440 } 1441 } 1442 return all_finished; 1443 } 1444 1445 static bool should_send_vmdesc(void) 1446 { 1447 MachineState *machine = MACHINE(qdev_get_machine()); 1448 bool in_postcopy = migration_in_postcopy(); 1449 return !machine->suppress_vmdesc && !in_postcopy; 1450 } 1451 1452 /* 1453 * Calls the save_live_complete_postcopy methods 1454 * causing the last few pages to be sent immediately and doing any associated 1455 * cleanup. 1456 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete 1457 * all the other devices, but that happens at the point we switch to postcopy. 1458 */ 1459 void qemu_savevm_state_complete_postcopy(QEMUFile *f) 1460 { 1461 SaveStateEntry *se; 1462 int ret; 1463 1464 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1465 if (!se->ops || !se->ops->save_live_complete_postcopy) { 1466 continue; 1467 } 1468 if (se->ops->is_active) { 1469 if (!se->ops->is_active(se->opaque)) { 1470 continue; 1471 } 1472 } 1473 trace_savevm_section_start(se->idstr, se->section_id); 1474 /* Section type */ 1475 qemu_put_byte(f, QEMU_VM_SECTION_END); 1476 qemu_put_be32(f, se->section_id); 1477 1478 ret = se->ops->save_live_complete_postcopy(f, se->opaque); 1479 trace_savevm_section_end(se->idstr, se->section_id, ret); 1480 save_section_footer(f, se); 1481 if (ret < 0) { 1482 qemu_file_set_error(f, ret); 1483 return; 1484 } 1485 } 1486 1487 qemu_put_byte(f, QEMU_VM_EOF); 1488 qemu_fflush(f); 1489 } 1490 1491 static 1492 int qemu_savevm_state_complete_precopy_iterable(QEMUFile *f, bool in_postcopy) 1493 { 1494 int64_t start_ts_each, end_ts_each; 1495 SaveStateEntry *se; 1496 int ret; 1497 1498 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1499 if (!se->ops || 1500 (in_postcopy && se->ops->has_postcopy && 1501 se->ops->has_postcopy(se->opaque)) || 1502 !se->ops->save_live_complete_precopy) { 1503 continue; 1504 } 1505 1506 if (se->ops->is_active) { 1507 if (!se->ops->is_active(se->opaque)) { 1508 continue; 1509 } 1510 } 1511 1512 start_ts_each = qemu_clock_get_us(QEMU_CLOCK_REALTIME); 1513 trace_savevm_section_start(se->idstr, se->section_id); 1514 1515 save_section_header(f, se, QEMU_VM_SECTION_END); 1516 1517 ret = se->ops->save_live_complete_precopy(f, se->opaque); 1518 trace_savevm_section_end(se->idstr, se->section_id, ret); 1519 save_section_footer(f, se); 1520 if (ret < 0) { 1521 qemu_file_set_error(f, ret); 1522 return -1; 1523 } 1524 end_ts_each = qemu_clock_get_us(QEMU_CLOCK_REALTIME); 1525 trace_vmstate_downtime_save("iterable", se->idstr, se->instance_id, 1526 end_ts_each - start_ts_each); 1527 } 1528 1529 trace_vmstate_downtime_checkpoint("src-iterable-saved"); 1530 1531 return 0; 1532 } 1533 1534 int qemu_savevm_state_complete_precopy_non_iterable(QEMUFile *f, 1535 bool in_postcopy, 1536 bool inactivate_disks) 1537 { 1538 MigrationState *ms = migrate_get_current(); 1539 int64_t start_ts_each, end_ts_each; 1540 JSONWriter *vmdesc = ms->vmdesc; 1541 int vmdesc_len; 1542 SaveStateEntry *se; 1543 Error *local_err = NULL; 1544 int ret; 1545 1546 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1547 if (se->vmsd && se->vmsd->early_setup) { 1548 /* Already saved during qemu_savevm_state_setup(). */ 1549 continue; 1550 } 1551 1552 start_ts_each = qemu_clock_get_us(QEMU_CLOCK_REALTIME); 1553 1554 ret = vmstate_save(f, se, vmdesc, &local_err); 1555 if (ret) { 1556 migrate_set_error(ms, local_err); 1557 error_report_err(local_err); 1558 qemu_file_set_error(f, ret); 1559 return ret; 1560 } 1561 1562 end_ts_each = qemu_clock_get_us(QEMU_CLOCK_REALTIME); 1563 trace_vmstate_downtime_save("non-iterable", se->idstr, se->instance_id, 1564 end_ts_each - start_ts_each); 1565 } 1566 1567 if (inactivate_disks) { 1568 /* Inactivate before sending QEMU_VM_EOF so that the 1569 * bdrv_activate_all() on the other end won't fail. */ 1570 ret = bdrv_inactivate_all(); 1571 if (ret) { 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, "There's a migration process in progress"); 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 ret = qemu_savevm_state_setup(f, errp); 1727 if (ret) { 1728 goto cleanup; 1729 } 1730 1731 while (qemu_file_get_error(f) == 0) { 1732 if (qemu_savevm_state_iterate(f, false) > 0) { 1733 break; 1734 } 1735 } 1736 1737 ret = qemu_file_get_error(f); 1738 if (ret == 0) { 1739 qemu_savevm_state_complete_precopy(f, false, false); 1740 ret = qemu_file_get_error(f); 1741 } 1742 if (ret != 0) { 1743 error_setg_errno(errp, -ret, "Error while writing VM state"); 1744 } 1745 cleanup: 1746 qemu_savevm_state_cleanup(); 1747 1748 if (ret != 0) { 1749 status = MIGRATION_STATUS_FAILED; 1750 } else { 1751 status = MIGRATION_STATUS_COMPLETED; 1752 } 1753 migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status); 1754 1755 /* f is outer parameter, it should not stay in global migration state after 1756 * this function finished */ 1757 ms->to_dst_file = NULL; 1758 1759 return ret; 1760 } 1761 1762 void qemu_savevm_live_state(QEMUFile *f) 1763 { 1764 /* save QEMU_VM_SECTION_END section */ 1765 qemu_savevm_state_complete_precopy(f, true, false); 1766 qemu_put_byte(f, QEMU_VM_EOF); 1767 } 1768 1769 int qemu_save_device_state(QEMUFile *f) 1770 { 1771 MigrationState *ms = migrate_get_current(); 1772 Error *local_err = NULL; 1773 SaveStateEntry *se; 1774 1775 if (!migration_in_colo_state()) { 1776 qemu_put_be32(f, QEMU_VM_FILE_MAGIC); 1777 qemu_put_be32(f, QEMU_VM_FILE_VERSION); 1778 } 1779 cpu_synchronize_all_states(); 1780 1781 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1782 int ret; 1783 1784 if (se->is_ram) { 1785 continue; 1786 } 1787 ret = vmstate_save(f, se, NULL, &local_err); 1788 if (ret) { 1789 migrate_set_error(ms, local_err); 1790 error_report_err(local_err); 1791 return ret; 1792 } 1793 } 1794 1795 qemu_put_byte(f, QEMU_VM_EOF); 1796 1797 return qemu_file_get_error(f); 1798 } 1799 1800 static SaveStateEntry *find_se(const char *idstr, uint32_t instance_id) 1801 { 1802 SaveStateEntry *se; 1803 1804 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1805 if (!strcmp(se->idstr, idstr) && 1806 (instance_id == se->instance_id || 1807 instance_id == se->alias_id)) 1808 return se; 1809 /* Migrating from an older version? */ 1810 if (strstr(se->idstr, idstr) && se->compat) { 1811 if (!strcmp(se->compat->idstr, idstr) && 1812 (instance_id == se->compat->instance_id || 1813 instance_id == se->alias_id)) 1814 return se; 1815 } 1816 } 1817 return NULL; 1818 } 1819 1820 enum LoadVMExitCodes { 1821 /* Allow a command to quit all layers of nested loadvm loops */ 1822 LOADVM_QUIT = 1, 1823 }; 1824 1825 /* ------ incoming postcopy messages ------ */ 1826 /* 'advise' arrives before any transfers just to tell us that a postcopy 1827 * *might* happen - it might be skipped if precopy transferred everything 1828 * quickly. 1829 */ 1830 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis, 1831 uint16_t len) 1832 { 1833 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE); 1834 uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps; 1835 size_t page_size = qemu_target_page_size(); 1836 Error *local_err = NULL; 1837 1838 trace_loadvm_postcopy_handle_advise(); 1839 if (ps != POSTCOPY_INCOMING_NONE) { 1840 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps); 1841 return -1; 1842 } 1843 1844 switch (len) { 1845 case 0: 1846 if (migrate_postcopy_ram()) { 1847 error_report("RAM postcopy is enabled but have 0 byte advise"); 1848 return -EINVAL; 1849 } 1850 return 0; 1851 case 8 + 8: 1852 if (!migrate_postcopy_ram()) { 1853 error_report("RAM postcopy is disabled but have 16 byte advise"); 1854 return -EINVAL; 1855 } 1856 break; 1857 default: 1858 error_report("CMD_POSTCOPY_ADVISE invalid length (%d)", len); 1859 return -EINVAL; 1860 } 1861 1862 if (!postcopy_ram_supported_by_host(mis, &local_err)) { 1863 error_report_err(local_err); 1864 postcopy_state_set(POSTCOPY_INCOMING_NONE); 1865 return -1; 1866 } 1867 1868 remote_pagesize_summary = qemu_get_be64(mis->from_src_file); 1869 local_pagesize_summary = ram_pagesize_summary(); 1870 1871 if (remote_pagesize_summary != local_pagesize_summary) { 1872 /* 1873 * This detects two potential causes of mismatch: 1874 * a) A mismatch in host page sizes 1875 * Some combinations of mismatch are probably possible but it gets 1876 * a bit more complicated. In particular we need to place whole 1877 * host pages on the dest at once, and we need to ensure that we 1878 * handle dirtying to make sure we never end up sending part of 1879 * a hostpage on it's own. 1880 * b) The use of different huge page sizes on source/destination 1881 * a more fine grain test is performed during RAM block migration 1882 * but this test here causes a nice early clear failure, and 1883 * also fails when passed to an older qemu that doesn't 1884 * do huge pages. 1885 */ 1886 error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64 1887 " d=%" PRIx64 ")", 1888 remote_pagesize_summary, local_pagesize_summary); 1889 return -1; 1890 } 1891 1892 remote_tps = qemu_get_be64(mis->from_src_file); 1893 if (remote_tps != page_size) { 1894 /* 1895 * Again, some differences could be dealt with, but for now keep it 1896 * simple. 1897 */ 1898 error_report("Postcopy needs matching target page sizes (s=%d d=%zd)", 1899 (int)remote_tps, page_size); 1900 return -1; 1901 } 1902 1903 if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_ADVISE, &local_err)) { 1904 error_report_err(local_err); 1905 return -1; 1906 } 1907 1908 if (ram_postcopy_incoming_init(mis)) { 1909 return -1; 1910 } 1911 1912 return 0; 1913 } 1914 1915 /* After postcopy we will be told to throw some pages away since they're 1916 * dirty and will have to be demand fetched. Must happen before CPU is 1917 * started. 1918 * There can be 0..many of these messages, each encoding multiple pages. 1919 */ 1920 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis, 1921 uint16_t len) 1922 { 1923 int tmp; 1924 char ramid[256]; 1925 PostcopyState ps = postcopy_state_get(); 1926 1927 trace_loadvm_postcopy_ram_handle_discard(); 1928 1929 switch (ps) { 1930 case POSTCOPY_INCOMING_ADVISE: 1931 /* 1st discard */ 1932 tmp = postcopy_ram_prepare_discard(mis); 1933 if (tmp) { 1934 return tmp; 1935 } 1936 break; 1937 1938 case POSTCOPY_INCOMING_DISCARD: 1939 /* Expected state */ 1940 break; 1941 1942 default: 1943 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)", 1944 ps); 1945 return -1; 1946 } 1947 /* We're expecting a 1948 * Version (0) 1949 * a RAM ID string (length byte, name, 0 term) 1950 * then at least 1 16 byte chunk 1951 */ 1952 if (len < (1 + 1 + 1 + 1 + 2 * 8)) { 1953 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len); 1954 return -1; 1955 } 1956 1957 tmp = qemu_get_byte(mis->from_src_file); 1958 if (tmp != postcopy_ram_discard_version) { 1959 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp); 1960 return -1; 1961 } 1962 1963 if (!qemu_get_counted_string(mis->from_src_file, ramid)) { 1964 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID"); 1965 return -1; 1966 } 1967 tmp = qemu_get_byte(mis->from_src_file); 1968 if (tmp != 0) { 1969 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp); 1970 return -1; 1971 } 1972 1973 len -= 3 + strlen(ramid); 1974 if (len % 16) { 1975 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len); 1976 return -1; 1977 } 1978 trace_loadvm_postcopy_ram_handle_discard_header(ramid, len); 1979 while (len) { 1980 uint64_t start_addr, block_length; 1981 start_addr = qemu_get_be64(mis->from_src_file); 1982 block_length = qemu_get_be64(mis->from_src_file); 1983 1984 len -= 16; 1985 int ret = ram_discard_range(ramid, start_addr, block_length); 1986 if (ret) { 1987 return ret; 1988 } 1989 } 1990 trace_loadvm_postcopy_ram_handle_discard_end(); 1991 1992 return 0; 1993 } 1994 1995 /* 1996 * Triggered by a postcopy_listen command; this thread takes over reading 1997 * the input stream, leaving the main thread free to carry on loading the rest 1998 * of the device state (from RAM). 1999 * (TODO:This could do with being in a postcopy file - but there again it's 2000 * just another input loop, not that postcopy specific) 2001 */ 2002 static void *postcopy_ram_listen_thread(void *opaque) 2003 { 2004 MigrationIncomingState *mis = migration_incoming_get_current(); 2005 QEMUFile *f = mis->from_src_file; 2006 int load_res; 2007 MigrationState *migr = migrate_get_current(); 2008 2009 object_ref(OBJECT(migr)); 2010 2011 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, 2012 MIGRATION_STATUS_POSTCOPY_ACTIVE); 2013 qemu_sem_post(&mis->thread_sync_sem); 2014 trace_postcopy_ram_listen_thread_start(); 2015 2016 rcu_register_thread(); 2017 /* 2018 * Because we're a thread and not a coroutine we can't yield 2019 * in qemu_file, and thus we must be blocking now. 2020 */ 2021 qemu_file_set_blocking(f, true); 2022 load_res = qemu_loadvm_state_main(f, mis); 2023 2024 /* 2025 * This is tricky, but, mis->from_src_file can change after it 2026 * returns, when postcopy recovery happened. In the future, we may 2027 * want a wrapper for the QEMUFile handle. 2028 */ 2029 f = mis->from_src_file; 2030 2031 /* And non-blocking again so we don't block in any cleanup */ 2032 qemu_file_set_blocking(f, false); 2033 2034 trace_postcopy_ram_listen_thread_exit(); 2035 if (load_res < 0) { 2036 qemu_file_set_error(f, load_res); 2037 dirty_bitmap_mig_cancel_incoming(); 2038 if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING && 2039 !migrate_postcopy_ram() && migrate_dirty_bitmaps()) 2040 { 2041 error_report("%s: loadvm failed during postcopy: %d. All states " 2042 "are migrated except dirty bitmaps. Some dirty " 2043 "bitmaps may be lost, and present migrated dirty " 2044 "bitmaps are correctly migrated and valid.", 2045 __func__, load_res); 2046 load_res = 0; /* prevent further exit() */ 2047 } else { 2048 error_report("%s: loadvm failed: %d", __func__, load_res); 2049 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, 2050 MIGRATION_STATUS_FAILED); 2051 } 2052 } 2053 if (load_res >= 0) { 2054 /* 2055 * This looks good, but it's possible that the device loading in the 2056 * main thread hasn't finished yet, and so we might not be in 'RUN' 2057 * state yet; wait for the end of the main thread. 2058 */ 2059 qemu_event_wait(&mis->main_thread_load_event); 2060 } 2061 postcopy_ram_incoming_cleanup(mis); 2062 2063 if (load_res < 0) { 2064 /* 2065 * If something went wrong then we have a bad state so exit; 2066 * depending how far we got it might be possible at this point 2067 * to leave the guest running and fire MCEs for pages that never 2068 * arrived as a desperate recovery step. 2069 */ 2070 rcu_unregister_thread(); 2071 exit(EXIT_FAILURE); 2072 } 2073 2074 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, 2075 MIGRATION_STATUS_COMPLETED); 2076 /* 2077 * If everything has worked fine, then the main thread has waited 2078 * for us to start, and we're the last use of the mis. 2079 * (If something broke then qemu will have to exit anyway since it's 2080 * got a bad migration state). 2081 */ 2082 migration_incoming_state_destroy(); 2083 qemu_loadvm_state_cleanup(); 2084 2085 rcu_unregister_thread(); 2086 mis->have_listen_thread = false; 2087 postcopy_state_set(POSTCOPY_INCOMING_END); 2088 2089 object_unref(OBJECT(migr)); 2090 2091 return NULL; 2092 } 2093 2094 /* After this message we must be able to immediately receive postcopy data */ 2095 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis) 2096 { 2097 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING); 2098 Error *local_err = NULL; 2099 2100 trace_loadvm_postcopy_handle_listen("enter"); 2101 2102 if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) { 2103 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps); 2104 return -1; 2105 } 2106 if (ps == POSTCOPY_INCOMING_ADVISE) { 2107 /* 2108 * A rare case, we entered listen without having to do any discards, 2109 * so do the setup that's normally done at the time of the 1st discard. 2110 */ 2111 if (migrate_postcopy_ram()) { 2112 postcopy_ram_prepare_discard(mis); 2113 } 2114 } 2115 2116 trace_loadvm_postcopy_handle_listen("after discard"); 2117 2118 /* 2119 * Sensitise RAM - can now generate requests for blocks that don't exist 2120 * However, at this point the CPU shouldn't be running, and the IO 2121 * shouldn't be doing anything yet so don't actually expect requests 2122 */ 2123 if (migrate_postcopy_ram()) { 2124 if (postcopy_ram_incoming_setup(mis)) { 2125 postcopy_ram_incoming_cleanup(mis); 2126 return -1; 2127 } 2128 } 2129 2130 trace_loadvm_postcopy_handle_listen("after uffd"); 2131 2132 if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_LISTEN, &local_err)) { 2133 error_report_err(local_err); 2134 return -1; 2135 } 2136 2137 mis->have_listen_thread = true; 2138 postcopy_thread_create(mis, &mis->listen_thread, "postcopy/listen", 2139 postcopy_ram_listen_thread, QEMU_THREAD_DETACHED); 2140 trace_loadvm_postcopy_handle_listen("return"); 2141 2142 return 0; 2143 } 2144 2145 static void loadvm_postcopy_handle_run_bh(void *opaque) 2146 { 2147 Error *local_err = NULL; 2148 MigrationIncomingState *mis = opaque; 2149 2150 trace_vmstate_downtime_checkpoint("dst-postcopy-bh-enter"); 2151 2152 /* TODO we should move all of this lot into postcopy_ram.c or a shared code 2153 * in migration.c 2154 */ 2155 cpu_synchronize_all_post_init(); 2156 2157 trace_vmstate_downtime_checkpoint("dst-postcopy-bh-cpu-synced"); 2158 2159 qemu_announce_self(&mis->announce_timer, migrate_announce_params()); 2160 2161 trace_vmstate_downtime_checkpoint("dst-postcopy-bh-announced"); 2162 2163 /* Make sure all file formats throw away their mutable metadata. 2164 * If we get an error here, just don't restart the VM yet. */ 2165 bdrv_activate_all(&local_err); 2166 if (local_err) { 2167 error_report_err(local_err); 2168 local_err = NULL; 2169 autostart = false; 2170 } 2171 2172 trace_vmstate_downtime_checkpoint("dst-postcopy-bh-cache-invalidated"); 2173 2174 dirty_bitmap_mig_before_vm_start(); 2175 2176 if (autostart) { 2177 /* Hold onto your hats, starting the CPU */ 2178 vm_start(); 2179 } else { 2180 /* leave it paused and let management decide when to start the CPU */ 2181 runstate_set(RUN_STATE_PAUSED); 2182 } 2183 2184 trace_vmstate_downtime_checkpoint("dst-postcopy-bh-vm-started"); 2185 } 2186 2187 /* After all discards we can start running and asking for pages */ 2188 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis) 2189 { 2190 PostcopyState ps = postcopy_state_get(); 2191 2192 trace_loadvm_postcopy_handle_run(); 2193 if (ps != POSTCOPY_INCOMING_LISTENING) { 2194 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps); 2195 return -1; 2196 } 2197 2198 postcopy_state_set(POSTCOPY_INCOMING_RUNNING); 2199 migration_bh_schedule(loadvm_postcopy_handle_run_bh, mis); 2200 2201 /* We need to finish reading the stream from the package 2202 * and also stop reading anything more from the stream that loaded the 2203 * package (since it's now being read by the listener thread). 2204 * LOADVM_QUIT will quit all the layers of nested loadvm loops. 2205 */ 2206 return LOADVM_QUIT; 2207 } 2208 2209 /* We must be with page_request_mutex held */ 2210 static gboolean postcopy_sync_page_req(gpointer key, gpointer value, 2211 gpointer data) 2212 { 2213 MigrationIncomingState *mis = data; 2214 void *host_addr = (void *) key; 2215 ram_addr_t rb_offset; 2216 RAMBlock *rb; 2217 int ret; 2218 2219 rb = qemu_ram_block_from_host(host_addr, true, &rb_offset); 2220 if (!rb) { 2221 /* 2222 * This should _never_ happen. However be nice for a migrating VM to 2223 * not crash/assert. Post an error (note: intended to not use *_once 2224 * because we do want to see all the illegal addresses; and this can 2225 * never be triggered by the guest so we're safe) and move on next. 2226 */ 2227 error_report("%s: illegal host addr %p", __func__, host_addr); 2228 /* Try the next entry */ 2229 return FALSE; 2230 } 2231 2232 ret = migrate_send_rp_message_req_pages(mis, rb, rb_offset); 2233 if (ret) { 2234 /* Please refer to above comment. */ 2235 error_report("%s: send rp message failed for addr %p", 2236 __func__, host_addr); 2237 return FALSE; 2238 } 2239 2240 trace_postcopy_page_req_sync(host_addr); 2241 2242 return FALSE; 2243 } 2244 2245 static void migrate_send_rp_req_pages_pending(MigrationIncomingState *mis) 2246 { 2247 WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) { 2248 g_tree_foreach(mis->page_requested, postcopy_sync_page_req, mis); 2249 } 2250 } 2251 2252 static int loadvm_postcopy_handle_resume(MigrationIncomingState *mis) 2253 { 2254 if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) { 2255 error_report("%s: illegal resume received", __func__); 2256 /* Don't fail the load, only for this. */ 2257 return 0; 2258 } 2259 2260 /* 2261 * Reset the last_rb before we resend any page req to source again, since 2262 * the source should have it reset already. 2263 */ 2264 mis->last_rb = NULL; 2265 2266 /* 2267 * This means source VM is ready to resume the postcopy migration. 2268 */ 2269 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_RECOVER, 2270 MIGRATION_STATUS_POSTCOPY_ACTIVE); 2271 2272 trace_loadvm_postcopy_handle_resume(); 2273 2274 /* Tell source that "we are ready" */ 2275 migrate_send_rp_resume_ack(mis, MIGRATION_RESUME_ACK_VALUE); 2276 2277 /* 2278 * After a postcopy recovery, the source should have lost the postcopy 2279 * queue, or potentially the requested pages could have been lost during 2280 * the network down phase. Let's re-sync with the source VM by re-sending 2281 * all the pending pages that we eagerly need, so these threads won't get 2282 * blocked too long due to the recovery. 2283 * 2284 * Without this procedure, the faulted destination VM threads (waiting for 2285 * page requests right before the postcopy is interrupted) can keep hanging 2286 * until the pages are sent by the source during the background copying of 2287 * pages, or another thread faulted on the same address accidentally. 2288 */ 2289 migrate_send_rp_req_pages_pending(mis); 2290 2291 /* 2292 * It's time to switch state and release the fault thread to continue 2293 * service page faults. Note that this should be explicitly after the 2294 * above call to migrate_send_rp_req_pages_pending(). In short: 2295 * migrate_send_rp_message_req_pages() is not thread safe, yet. 2296 */ 2297 qemu_sem_post(&mis->postcopy_pause_sem_fault); 2298 2299 if (migrate_postcopy_preempt()) { 2300 /* 2301 * The preempt channel will be created in async manner, now let's 2302 * wait for it and make sure it's created. 2303 */ 2304 qemu_sem_wait(&mis->postcopy_qemufile_dst_done); 2305 assert(mis->postcopy_qemufile_dst); 2306 /* Kick the fast ram load thread too */ 2307 qemu_sem_post(&mis->postcopy_pause_sem_fast_load); 2308 } 2309 2310 return 0; 2311 } 2312 2313 /** 2314 * Immediately following this command is a blob of data containing an embedded 2315 * chunk of migration stream; read it and load it. 2316 * 2317 * @mis: Incoming state 2318 * @length: Length of packaged data to read 2319 * 2320 * Returns: Negative values on error 2321 * 2322 */ 2323 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis) 2324 { 2325 int ret; 2326 size_t length; 2327 QIOChannelBuffer *bioc; 2328 2329 length = qemu_get_be32(mis->from_src_file); 2330 trace_loadvm_handle_cmd_packaged(length); 2331 2332 if (length > MAX_VM_CMD_PACKAGED_SIZE) { 2333 error_report("Unreasonably large packaged state: %zu", length); 2334 return -1; 2335 } 2336 2337 bioc = qio_channel_buffer_new(length); 2338 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer"); 2339 ret = qemu_get_buffer(mis->from_src_file, 2340 bioc->data, 2341 length); 2342 if (ret != length) { 2343 object_unref(OBJECT(bioc)); 2344 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu", 2345 ret, length); 2346 return (ret < 0) ? ret : -EAGAIN; 2347 } 2348 bioc->usage += length; 2349 trace_loadvm_handle_cmd_packaged_received(ret); 2350 2351 QEMUFile *packf = qemu_file_new_input(QIO_CHANNEL(bioc)); 2352 2353 /* 2354 * Before loading the guest states, ensure that the preempt channel has 2355 * been ready to use, as some of the states (e.g. via virtio_load) might 2356 * trigger page faults that will be handled through the preempt channel. 2357 * So yield to the main thread in the case that the channel create event 2358 * hasn't been dispatched. 2359 * 2360 * TODO: if we can move migration loadvm out of main thread, then we 2361 * won't block main thread from polling the accept() fds. We can drop 2362 * this as a whole when that is done. 2363 */ 2364 do { 2365 if (!migrate_postcopy_preempt() || !qemu_in_coroutine() || 2366 mis->postcopy_qemufile_dst) { 2367 break; 2368 } 2369 2370 aio_co_schedule(qemu_get_current_aio_context(), qemu_coroutine_self()); 2371 qemu_coroutine_yield(); 2372 } while (1); 2373 2374 ret = qemu_loadvm_state_main(packf, mis); 2375 trace_loadvm_handle_cmd_packaged_main(ret); 2376 qemu_fclose(packf); 2377 object_unref(OBJECT(bioc)); 2378 2379 return ret; 2380 } 2381 2382 /* 2383 * Handle request that source requests for recved_bitmap on 2384 * destination. Payload format: 2385 * 2386 * len (1 byte) + ramblock_name (<255 bytes) 2387 */ 2388 static int loadvm_handle_recv_bitmap(MigrationIncomingState *mis, 2389 uint16_t len) 2390 { 2391 QEMUFile *file = mis->from_src_file; 2392 RAMBlock *rb; 2393 char block_name[256]; 2394 size_t cnt; 2395 2396 cnt = qemu_get_counted_string(file, block_name); 2397 if (!cnt) { 2398 error_report("%s: failed to read block name", __func__); 2399 return -EINVAL; 2400 } 2401 2402 /* Validate before using the data */ 2403 if (qemu_file_get_error(file)) { 2404 return qemu_file_get_error(file); 2405 } 2406 2407 if (len != cnt + 1) { 2408 error_report("%s: invalid payload length (%d)", __func__, len); 2409 return -EINVAL; 2410 } 2411 2412 rb = qemu_ram_block_by_name(block_name); 2413 if (!rb) { 2414 error_report("%s: block '%s' not found", __func__, block_name); 2415 return -EINVAL; 2416 } 2417 2418 migrate_send_rp_recv_bitmap(mis, block_name); 2419 2420 trace_loadvm_handle_recv_bitmap(block_name); 2421 2422 return 0; 2423 } 2424 2425 static int loadvm_process_enable_colo(MigrationIncomingState *mis) 2426 { 2427 int ret = migration_incoming_enable_colo(); 2428 2429 if (!ret) { 2430 ret = colo_init_ram_cache(); 2431 if (ret) { 2432 migration_incoming_disable_colo(); 2433 } 2434 } 2435 return ret; 2436 } 2437 2438 /* 2439 * Process an incoming 'QEMU_VM_COMMAND' 2440 * 0 just a normal return 2441 * LOADVM_QUIT All good, but exit the loop 2442 * <0 Error 2443 */ 2444 static int loadvm_process_command(QEMUFile *f) 2445 { 2446 MigrationIncomingState *mis = migration_incoming_get_current(); 2447 uint16_t cmd; 2448 uint16_t len; 2449 uint32_t tmp32; 2450 2451 cmd = qemu_get_be16(f); 2452 len = qemu_get_be16(f); 2453 2454 /* Check validity before continue processing of cmds */ 2455 if (qemu_file_get_error(f)) { 2456 return qemu_file_get_error(f); 2457 } 2458 2459 if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) { 2460 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len); 2461 return -EINVAL; 2462 } 2463 2464 trace_loadvm_process_command(mig_cmd_args[cmd].name, len); 2465 2466 if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) { 2467 error_report("%s received with bad length - expecting %zu, got %d", 2468 mig_cmd_args[cmd].name, 2469 (size_t)mig_cmd_args[cmd].len, len); 2470 return -ERANGE; 2471 } 2472 2473 switch (cmd) { 2474 case MIG_CMD_OPEN_RETURN_PATH: 2475 if (mis->to_src_file) { 2476 error_report("CMD_OPEN_RETURN_PATH called when RP already open"); 2477 /* Not really a problem, so don't give up */ 2478 return 0; 2479 } 2480 mis->to_src_file = qemu_file_get_return_path(f); 2481 if (!mis->to_src_file) { 2482 error_report("CMD_OPEN_RETURN_PATH failed"); 2483 return -1; 2484 } 2485 2486 /* 2487 * Switchover ack is enabled but no device uses it, so send an ACK to 2488 * source that it's OK to switchover. Do it here, after return path has 2489 * been created. 2490 */ 2491 if (migrate_switchover_ack() && !mis->switchover_ack_pending_num) { 2492 int ret = migrate_send_rp_switchover_ack(mis); 2493 if (ret) { 2494 error_report( 2495 "Could not send switchover ack RP MSG, err %d (%s)", ret, 2496 strerror(-ret)); 2497 return ret; 2498 } 2499 } 2500 break; 2501 2502 case MIG_CMD_PING: 2503 tmp32 = qemu_get_be32(f); 2504 trace_loadvm_process_command_ping(tmp32); 2505 if (!mis->to_src_file) { 2506 error_report("CMD_PING (0x%x) received with no return path", 2507 tmp32); 2508 return -1; 2509 } 2510 migrate_send_rp_pong(mis, tmp32); 2511 break; 2512 2513 case MIG_CMD_PACKAGED: 2514 return loadvm_handle_cmd_packaged(mis); 2515 2516 case MIG_CMD_POSTCOPY_ADVISE: 2517 return loadvm_postcopy_handle_advise(mis, len); 2518 2519 case MIG_CMD_POSTCOPY_LISTEN: 2520 return loadvm_postcopy_handle_listen(mis); 2521 2522 case MIG_CMD_POSTCOPY_RUN: 2523 return loadvm_postcopy_handle_run(mis); 2524 2525 case MIG_CMD_POSTCOPY_RAM_DISCARD: 2526 return loadvm_postcopy_ram_handle_discard(mis, len); 2527 2528 case MIG_CMD_POSTCOPY_RESUME: 2529 return loadvm_postcopy_handle_resume(mis); 2530 2531 case MIG_CMD_RECV_BITMAP: 2532 return loadvm_handle_recv_bitmap(mis, len); 2533 2534 case MIG_CMD_ENABLE_COLO: 2535 return loadvm_process_enable_colo(mis); 2536 } 2537 2538 return 0; 2539 } 2540 2541 /* 2542 * Read a footer off the wire and check that it matches the expected section 2543 * 2544 * Returns: true if the footer was good 2545 * false if there is a problem (and calls error_report to say why) 2546 */ 2547 static bool check_section_footer(QEMUFile *f, SaveStateEntry *se) 2548 { 2549 int ret; 2550 uint8_t read_mark; 2551 uint32_t read_section_id; 2552 2553 if (!migrate_get_current()->send_section_footer) { 2554 /* No footer to check */ 2555 return true; 2556 } 2557 2558 read_mark = qemu_get_byte(f); 2559 2560 ret = qemu_file_get_error(f); 2561 if (ret) { 2562 error_report("%s: Read section footer failed: %d", 2563 __func__, ret); 2564 return false; 2565 } 2566 2567 if (read_mark != QEMU_VM_SECTION_FOOTER) { 2568 error_report("Missing section footer for %s", se->idstr); 2569 return false; 2570 } 2571 2572 read_section_id = qemu_get_be32(f); 2573 if (read_section_id != se->load_section_id) { 2574 error_report("Mismatched section id in footer for %s -" 2575 " read 0x%x expected 0x%x", 2576 se->idstr, read_section_id, se->load_section_id); 2577 return false; 2578 } 2579 2580 /* All good */ 2581 return true; 2582 } 2583 2584 static int 2585 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis, 2586 uint8_t type) 2587 { 2588 bool trace_downtime = (type == QEMU_VM_SECTION_FULL); 2589 uint32_t instance_id, version_id, section_id; 2590 int64_t start_ts, end_ts; 2591 SaveStateEntry *se; 2592 char idstr[256]; 2593 int ret; 2594 2595 /* Read section start */ 2596 section_id = qemu_get_be32(f); 2597 if (!qemu_get_counted_string(f, idstr)) { 2598 error_report("Unable to read ID string for section %u", 2599 section_id); 2600 return -EINVAL; 2601 } 2602 instance_id = qemu_get_be32(f); 2603 version_id = qemu_get_be32(f); 2604 2605 ret = qemu_file_get_error(f); 2606 if (ret) { 2607 error_report("%s: Failed to read instance/version ID: %d", 2608 __func__, ret); 2609 return ret; 2610 } 2611 2612 trace_qemu_loadvm_state_section_startfull(section_id, idstr, 2613 instance_id, version_id); 2614 /* Find savevm section */ 2615 se = find_se(idstr, instance_id); 2616 if (se == NULL) { 2617 error_report("Unknown savevm section or instance '%s' %"PRIu32". " 2618 "Make sure that your current VM setup matches your " 2619 "saved VM setup, including any hotplugged devices", 2620 idstr, instance_id); 2621 return -EINVAL; 2622 } 2623 2624 /* Validate version */ 2625 if (version_id > se->version_id) { 2626 error_report("savevm: unsupported version %d for '%s' v%d", 2627 version_id, idstr, se->version_id); 2628 return -EINVAL; 2629 } 2630 se->load_version_id = version_id; 2631 se->load_section_id = section_id; 2632 2633 /* Validate if it is a device's state */ 2634 if (xen_enabled() && se->is_ram) { 2635 error_report("loadvm: %s RAM loading not allowed on Xen", idstr); 2636 return -EINVAL; 2637 } 2638 2639 if (trace_downtime) { 2640 start_ts = qemu_clock_get_us(QEMU_CLOCK_REALTIME); 2641 } 2642 2643 ret = vmstate_load(f, se); 2644 if (ret < 0) { 2645 error_report("error while loading state for instance 0x%"PRIx32" of" 2646 " device '%s'", instance_id, idstr); 2647 return ret; 2648 } 2649 2650 if (trace_downtime) { 2651 end_ts = qemu_clock_get_us(QEMU_CLOCK_REALTIME); 2652 trace_vmstate_downtime_load("non-iterable", se->idstr, 2653 se->instance_id, end_ts - start_ts); 2654 } 2655 2656 if (!check_section_footer(f, se)) { 2657 return -EINVAL; 2658 } 2659 2660 return 0; 2661 } 2662 2663 static int 2664 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis, 2665 uint8_t type) 2666 { 2667 bool trace_downtime = (type == QEMU_VM_SECTION_END); 2668 int64_t start_ts, end_ts; 2669 uint32_t section_id; 2670 SaveStateEntry *se; 2671 int ret; 2672 2673 section_id = qemu_get_be32(f); 2674 2675 ret = qemu_file_get_error(f); 2676 if (ret) { 2677 error_report("%s: Failed to read section ID: %d", 2678 __func__, ret); 2679 return ret; 2680 } 2681 2682 trace_qemu_loadvm_state_section_partend(section_id); 2683 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 2684 if (se->load_section_id == section_id) { 2685 break; 2686 } 2687 } 2688 if (se == NULL) { 2689 error_report("Unknown savevm section %d", section_id); 2690 return -EINVAL; 2691 } 2692 2693 if (trace_downtime) { 2694 start_ts = qemu_clock_get_us(QEMU_CLOCK_REALTIME); 2695 } 2696 2697 ret = vmstate_load(f, se); 2698 if (ret < 0) { 2699 error_report("error while loading state section id %d(%s)", 2700 section_id, se->idstr); 2701 return ret; 2702 } 2703 2704 if (trace_downtime) { 2705 end_ts = qemu_clock_get_us(QEMU_CLOCK_REALTIME); 2706 trace_vmstate_downtime_load("iterable", se->idstr, 2707 se->instance_id, end_ts - start_ts); 2708 } 2709 2710 if (!check_section_footer(f, se)) { 2711 return -EINVAL; 2712 } 2713 2714 return 0; 2715 } 2716 2717 static int qemu_loadvm_state_header(QEMUFile *f) 2718 { 2719 unsigned int v; 2720 int ret; 2721 2722 v = qemu_get_be32(f); 2723 if (v != QEMU_VM_FILE_MAGIC) { 2724 error_report("Not a migration stream"); 2725 return -EINVAL; 2726 } 2727 2728 v = qemu_get_be32(f); 2729 if (v == QEMU_VM_FILE_VERSION_COMPAT) { 2730 error_report("SaveVM v2 format is obsolete and don't work anymore"); 2731 return -ENOTSUP; 2732 } 2733 if (v != QEMU_VM_FILE_VERSION) { 2734 error_report("Unsupported migration stream version"); 2735 return -ENOTSUP; 2736 } 2737 2738 if (migrate_get_current()->send_configuration) { 2739 if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) { 2740 error_report("Configuration section missing"); 2741 qemu_loadvm_state_cleanup(); 2742 return -EINVAL; 2743 } 2744 ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0); 2745 2746 if (ret) { 2747 qemu_loadvm_state_cleanup(); 2748 return ret; 2749 } 2750 } 2751 return 0; 2752 } 2753 2754 static void qemu_loadvm_state_switchover_ack_needed(MigrationIncomingState *mis) 2755 { 2756 SaveStateEntry *se; 2757 2758 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 2759 if (!se->ops || !se->ops->switchover_ack_needed) { 2760 continue; 2761 } 2762 2763 if (se->ops->switchover_ack_needed(se->opaque)) { 2764 mis->switchover_ack_pending_num++; 2765 } 2766 } 2767 2768 trace_loadvm_state_switchover_ack_needed(mis->switchover_ack_pending_num); 2769 } 2770 2771 static int qemu_loadvm_state_setup(QEMUFile *f, Error **errp) 2772 { 2773 ERRP_GUARD(); 2774 SaveStateEntry *se; 2775 int ret; 2776 2777 trace_loadvm_state_setup(); 2778 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 2779 if (!se->ops || !se->ops->load_setup) { 2780 continue; 2781 } 2782 if (se->ops->is_active) { 2783 if (!se->ops->is_active(se->opaque)) { 2784 continue; 2785 } 2786 } 2787 2788 ret = se->ops->load_setup(f, se->opaque, errp); 2789 if (ret < 0) { 2790 error_prepend(errp, "Load state of device %s failed: ", 2791 se->idstr); 2792 qemu_file_set_error(f, ret); 2793 return ret; 2794 } 2795 } 2796 return 0; 2797 } 2798 2799 void qemu_loadvm_state_cleanup(void) 2800 { 2801 SaveStateEntry *se; 2802 2803 trace_loadvm_state_cleanup(); 2804 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 2805 if (se->ops && se->ops->load_cleanup) { 2806 se->ops->load_cleanup(se->opaque); 2807 } 2808 } 2809 } 2810 2811 /* Return true if we should continue the migration, or false. */ 2812 static bool postcopy_pause_incoming(MigrationIncomingState *mis) 2813 { 2814 int i; 2815 2816 trace_postcopy_pause_incoming(); 2817 2818 assert(migrate_postcopy_ram()); 2819 2820 /* 2821 * Unregister yank with either from/to src would work, since ioc behind it 2822 * is the same 2823 */ 2824 migration_ioc_unregister_yank_from_file(mis->from_src_file); 2825 2826 assert(mis->from_src_file); 2827 qemu_file_shutdown(mis->from_src_file); 2828 qemu_fclose(mis->from_src_file); 2829 mis->from_src_file = NULL; 2830 2831 assert(mis->to_src_file); 2832 qemu_file_shutdown(mis->to_src_file); 2833 qemu_mutex_lock(&mis->rp_mutex); 2834 qemu_fclose(mis->to_src_file); 2835 mis->to_src_file = NULL; 2836 qemu_mutex_unlock(&mis->rp_mutex); 2837 2838 /* 2839 * NOTE: this must happen before reset the PostcopyTmpPages below, 2840 * otherwise it's racy to reset those fields when the fast load thread 2841 * can be accessing it in parallel. 2842 */ 2843 if (mis->postcopy_qemufile_dst) { 2844 qemu_file_shutdown(mis->postcopy_qemufile_dst); 2845 /* Take the mutex to make sure the fast ram load thread halted */ 2846 qemu_mutex_lock(&mis->postcopy_prio_thread_mutex); 2847 migration_ioc_unregister_yank_from_file(mis->postcopy_qemufile_dst); 2848 qemu_fclose(mis->postcopy_qemufile_dst); 2849 mis->postcopy_qemufile_dst = NULL; 2850 qemu_mutex_unlock(&mis->postcopy_prio_thread_mutex); 2851 } 2852 2853 /* Current state can be either ACTIVE or RECOVER */ 2854 migrate_set_state(&mis->state, mis->state, 2855 MIGRATION_STATUS_POSTCOPY_PAUSED); 2856 2857 /* Notify the fault thread for the invalidated file handle */ 2858 postcopy_fault_thread_notify(mis); 2859 2860 /* 2861 * If network is interrupted, any temp page we received will be useless 2862 * because we didn't mark them as "received" in receivedmap. After a 2863 * proper recovery later (which will sync src dirty bitmap with receivedmap 2864 * on dest) these cached small pages will be resent again. 2865 */ 2866 for (i = 0; i < mis->postcopy_channels; i++) { 2867 postcopy_temp_page_reset(&mis->postcopy_tmp_pages[i]); 2868 } 2869 2870 error_report("Detected IO failure for postcopy. " 2871 "Migration paused."); 2872 2873 while (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { 2874 qemu_sem_wait(&mis->postcopy_pause_sem_dst); 2875 } 2876 2877 trace_postcopy_pause_incoming_continued(); 2878 2879 return true; 2880 } 2881 2882 int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis) 2883 { 2884 uint8_t section_type; 2885 int ret = 0; 2886 2887 retry: 2888 while (true) { 2889 section_type = qemu_get_byte(f); 2890 2891 ret = qemu_file_get_error_obj_any(f, mis->postcopy_qemufile_dst, NULL); 2892 if (ret) { 2893 break; 2894 } 2895 2896 trace_qemu_loadvm_state_section(section_type); 2897 switch (section_type) { 2898 case QEMU_VM_SECTION_START: 2899 case QEMU_VM_SECTION_FULL: 2900 ret = qemu_loadvm_section_start_full(f, mis, section_type); 2901 if (ret < 0) { 2902 goto out; 2903 } 2904 break; 2905 case QEMU_VM_SECTION_PART: 2906 case QEMU_VM_SECTION_END: 2907 ret = qemu_loadvm_section_part_end(f, mis, section_type); 2908 if (ret < 0) { 2909 goto out; 2910 } 2911 break; 2912 case QEMU_VM_COMMAND: 2913 ret = loadvm_process_command(f); 2914 trace_qemu_loadvm_state_section_command(ret); 2915 if ((ret < 0) || (ret == LOADVM_QUIT)) { 2916 goto out; 2917 } 2918 break; 2919 case QEMU_VM_EOF: 2920 /* This is the end of migration */ 2921 goto out; 2922 default: 2923 error_report("Unknown savevm section type %d", section_type); 2924 ret = -EINVAL; 2925 goto out; 2926 } 2927 } 2928 2929 out: 2930 if (ret < 0) { 2931 qemu_file_set_error(f, ret); 2932 2933 /* Cancel bitmaps incoming regardless of recovery */ 2934 dirty_bitmap_mig_cancel_incoming(); 2935 2936 /* 2937 * If we are during an active postcopy, then we pause instead 2938 * of bail out to at least keep the VM's dirty data. Note 2939 * that POSTCOPY_INCOMING_LISTENING stage is still not enough, 2940 * during which we're still receiving device states and we 2941 * still haven't yet started the VM on destination. 2942 * 2943 * Only RAM postcopy supports recovery. Still, if RAM postcopy is 2944 * enabled, canceled bitmaps postcopy will not affect RAM postcopy 2945 * recovering. 2946 */ 2947 if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING && 2948 migrate_postcopy_ram() && postcopy_pause_incoming(mis)) { 2949 /* Reset f to point to the newly created channel */ 2950 f = mis->from_src_file; 2951 goto retry; 2952 } 2953 } 2954 return ret; 2955 } 2956 2957 int qemu_loadvm_state(QEMUFile *f) 2958 { 2959 MigrationIncomingState *mis = migration_incoming_get_current(); 2960 Error *local_err = NULL; 2961 int ret; 2962 2963 if (qemu_savevm_state_blocked(&local_err)) { 2964 error_report_err(local_err); 2965 return -EINVAL; 2966 } 2967 2968 ret = qemu_loadvm_state_header(f); 2969 if (ret) { 2970 return ret; 2971 } 2972 2973 if (qemu_loadvm_state_setup(f, &local_err) != 0) { 2974 error_report_err(local_err); 2975 return -EINVAL; 2976 } 2977 2978 if (migrate_switchover_ack()) { 2979 qemu_loadvm_state_switchover_ack_needed(mis); 2980 } 2981 2982 cpu_synchronize_all_pre_loadvm(); 2983 2984 ret = qemu_loadvm_state_main(f, mis); 2985 qemu_event_set(&mis->main_thread_load_event); 2986 2987 trace_qemu_loadvm_state_post_main(ret); 2988 2989 if (mis->have_listen_thread) { 2990 /* Listen thread still going, can't clean up yet */ 2991 return ret; 2992 } 2993 2994 if (ret == 0) { 2995 ret = qemu_file_get_error(f); 2996 } 2997 2998 /* 2999 * Try to read in the VMDESC section as well, so that dumping tools that 3000 * intercept our migration stream have the chance to see it. 3001 */ 3002 3003 /* We've got to be careful; if we don't read the data and just shut the fd 3004 * then the sender can error if we close while it's still sending. 3005 * We also mustn't read data that isn't there; some transports (RDMA) 3006 * will stall waiting for that data when the source has already closed. 3007 */ 3008 if (ret == 0 && should_send_vmdesc()) { 3009 uint8_t *buf; 3010 uint32_t size; 3011 uint8_t section_type = qemu_get_byte(f); 3012 3013 if (section_type != QEMU_VM_VMDESCRIPTION) { 3014 error_report("Expected vmdescription section, but got %d", 3015 section_type); 3016 /* 3017 * It doesn't seem worth failing at this point since 3018 * we apparently have an otherwise valid VM state 3019 */ 3020 } else { 3021 buf = g_malloc(0x1000); 3022 size = qemu_get_be32(f); 3023 3024 while (size > 0) { 3025 uint32_t read_chunk = MIN(size, 0x1000); 3026 qemu_get_buffer(f, buf, read_chunk); 3027 size -= read_chunk; 3028 } 3029 g_free(buf); 3030 } 3031 } 3032 3033 qemu_loadvm_state_cleanup(); 3034 cpu_synchronize_all_post_init(); 3035 3036 return ret; 3037 } 3038 3039 int qemu_load_device_state(QEMUFile *f) 3040 { 3041 MigrationIncomingState *mis = migration_incoming_get_current(); 3042 int ret; 3043 3044 /* Load QEMU_VM_SECTION_FULL section */ 3045 ret = qemu_loadvm_state_main(f, mis); 3046 if (ret < 0) { 3047 error_report("Failed to load device state: %d", ret); 3048 return ret; 3049 } 3050 3051 cpu_synchronize_all_post_init(); 3052 return 0; 3053 } 3054 3055 int qemu_loadvm_approve_switchover(void) 3056 { 3057 MigrationIncomingState *mis = migration_incoming_get_current(); 3058 3059 if (!mis->switchover_ack_pending_num) { 3060 return -EINVAL; 3061 } 3062 3063 mis->switchover_ack_pending_num--; 3064 trace_loadvm_approve_switchover(mis->switchover_ack_pending_num); 3065 3066 if (mis->switchover_ack_pending_num) { 3067 return 0; 3068 } 3069 3070 return migrate_send_rp_switchover_ack(mis); 3071 } 3072 3073 bool save_snapshot(const char *name, bool overwrite, const char *vmstate, 3074 bool has_devices, strList *devices, Error **errp) 3075 { 3076 BlockDriverState *bs; 3077 QEMUSnapshotInfo sn1, *sn = &sn1; 3078 int ret = -1, ret2; 3079 QEMUFile *f; 3080 RunState saved_state = runstate_get(); 3081 uint64_t vm_state_size; 3082 g_autoptr(GDateTime) now = g_date_time_new_now_local(); 3083 3084 GLOBAL_STATE_CODE(); 3085 3086 if (migration_is_blocked(errp)) { 3087 return false; 3088 } 3089 3090 if (!replay_can_snapshot()) { 3091 error_setg(errp, "Record/replay does not allow making snapshot " 3092 "right now. Try once more later."); 3093 return false; 3094 } 3095 3096 if (!bdrv_all_can_snapshot(has_devices, devices, errp)) { 3097 return false; 3098 } 3099 3100 /* Delete old snapshots of the same name */ 3101 if (name) { 3102 if (overwrite) { 3103 if (bdrv_all_delete_snapshot(name, has_devices, 3104 devices, errp) < 0) { 3105 return false; 3106 } 3107 } else { 3108 ret2 = bdrv_all_has_snapshot(name, has_devices, devices, errp); 3109 if (ret2 < 0) { 3110 return false; 3111 } 3112 if (ret2 == 1) { 3113 error_setg(errp, 3114 "Snapshot '%s' already exists in one or more devices", 3115 name); 3116 return false; 3117 } 3118 } 3119 } 3120 3121 bs = bdrv_all_find_vmstate_bs(vmstate, has_devices, devices, errp); 3122 if (bs == NULL) { 3123 return false; 3124 } 3125 3126 global_state_store(); 3127 vm_stop(RUN_STATE_SAVE_VM); 3128 3129 bdrv_drain_all_begin(); 3130 3131 memset(sn, 0, sizeof(*sn)); 3132 3133 /* fill auxiliary fields */ 3134 sn->date_sec = g_date_time_to_unix(now); 3135 sn->date_nsec = g_date_time_get_microsecond(now) * 1000; 3136 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL); 3137 if (replay_mode != REPLAY_MODE_NONE) { 3138 sn->icount = replay_get_current_icount(); 3139 } else { 3140 sn->icount = -1ULL; 3141 } 3142 3143 if (name) { 3144 pstrcpy(sn->name, sizeof(sn->name), name); 3145 } else { 3146 g_autofree char *autoname = g_date_time_format(now, "vm-%Y%m%d%H%M%S"); 3147 pstrcpy(sn->name, sizeof(sn->name), autoname); 3148 } 3149 3150 /* save the VM state */ 3151 f = qemu_fopen_bdrv(bs, 1); 3152 if (!f) { 3153 error_setg(errp, "Could not open VM state file"); 3154 goto the_end; 3155 } 3156 ret = qemu_savevm_state(f, errp); 3157 vm_state_size = qemu_file_transferred(f); 3158 ret2 = qemu_fclose(f); 3159 if (ret < 0) { 3160 goto the_end; 3161 } 3162 if (ret2 < 0) { 3163 ret = ret2; 3164 goto the_end; 3165 } 3166 3167 ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, 3168 has_devices, devices, errp); 3169 if (ret < 0) { 3170 bdrv_all_delete_snapshot(sn->name, has_devices, devices, NULL); 3171 goto the_end; 3172 } 3173 3174 ret = 0; 3175 3176 the_end: 3177 bdrv_drain_all_end(); 3178 3179 vm_resume(saved_state); 3180 return ret == 0; 3181 } 3182 3183 void qmp_xen_save_devices_state(const char *filename, bool has_live, bool live, 3184 Error **errp) 3185 { 3186 QEMUFile *f; 3187 QIOChannelFile *ioc; 3188 int saved_vm_running; 3189 int ret; 3190 3191 if (!has_live) { 3192 /* live default to true so old version of Xen tool stack can have a 3193 * successful live migration */ 3194 live = true; 3195 } 3196 3197 saved_vm_running = runstate_is_running(); 3198 vm_stop(RUN_STATE_SAVE_VM); 3199 global_state_store_running(); 3200 3201 ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT | O_TRUNC, 3202 0660, errp); 3203 if (!ioc) { 3204 goto the_end; 3205 } 3206 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state"); 3207 f = qemu_file_new_output(QIO_CHANNEL(ioc)); 3208 object_unref(OBJECT(ioc)); 3209 ret = qemu_save_device_state(f); 3210 if (ret < 0 || qemu_fclose(f) < 0) { 3211 error_setg(errp, QERR_IO_ERROR); 3212 } else { 3213 /* libxl calls the QMP command "stop" before calling 3214 * "xen-save-devices-state" and in case of migration failure, libxl 3215 * would call "cont". 3216 * So call bdrv_inactivate_all (release locks) here to let the other 3217 * side of the migration take control of the images. 3218 */ 3219 if (live && !saved_vm_running) { 3220 ret = bdrv_inactivate_all(); 3221 if (ret) { 3222 error_setg(errp, "%s: bdrv_inactivate_all() failed (%d)", 3223 __func__, ret); 3224 } 3225 } 3226 } 3227 3228 the_end: 3229 if (saved_vm_running) { 3230 vm_start(); 3231 } 3232 } 3233 3234 void qmp_xen_load_devices_state(const char *filename, Error **errp) 3235 { 3236 QEMUFile *f; 3237 QIOChannelFile *ioc; 3238 int ret; 3239 3240 /* Guest must be paused before loading the device state; the RAM state 3241 * will already have been loaded by xc 3242 */ 3243 if (runstate_is_running()) { 3244 error_setg(errp, "Cannot update device state while vm is running"); 3245 return; 3246 } 3247 vm_stop(RUN_STATE_RESTORE_VM); 3248 3249 ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp); 3250 if (!ioc) { 3251 return; 3252 } 3253 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state"); 3254 f = qemu_file_new_input(QIO_CHANNEL(ioc)); 3255 object_unref(OBJECT(ioc)); 3256 3257 ret = qemu_loadvm_state(f); 3258 qemu_fclose(f); 3259 if (ret < 0) { 3260 error_setg(errp, QERR_IO_ERROR); 3261 } 3262 migration_incoming_state_destroy(); 3263 } 3264 3265 bool load_snapshot(const char *name, const char *vmstate, 3266 bool has_devices, strList *devices, Error **errp) 3267 { 3268 BlockDriverState *bs_vm_state; 3269 QEMUSnapshotInfo sn; 3270 QEMUFile *f; 3271 int ret; 3272 MigrationIncomingState *mis = migration_incoming_get_current(); 3273 3274 if (!bdrv_all_can_snapshot(has_devices, devices, errp)) { 3275 return false; 3276 } 3277 ret = bdrv_all_has_snapshot(name, has_devices, devices, errp); 3278 if (ret < 0) { 3279 return false; 3280 } 3281 if (ret == 0) { 3282 error_setg(errp, "Snapshot '%s' does not exist in one or more devices", 3283 name); 3284 return false; 3285 } 3286 3287 bs_vm_state = bdrv_all_find_vmstate_bs(vmstate, has_devices, devices, errp); 3288 if (!bs_vm_state) { 3289 return false; 3290 } 3291 3292 /* Don't even try to load empty VM states */ 3293 ret = bdrv_snapshot_find(bs_vm_state, &sn, name); 3294 if (ret < 0) { 3295 return false; 3296 } else if (sn.vm_state_size == 0) { 3297 error_setg(errp, "This is a disk-only snapshot. Revert to it " 3298 " offline using qemu-img"); 3299 return false; 3300 } 3301 3302 /* 3303 * Flush the record/replay queue. Now the VM state is going 3304 * to change. Therefore we don't need to preserve its consistency 3305 */ 3306 replay_flush_events(); 3307 3308 /* Flush all IO requests so they don't interfere with the new state. */ 3309 bdrv_drain_all_begin(); 3310 3311 ret = bdrv_all_goto_snapshot(name, has_devices, devices, errp); 3312 if (ret < 0) { 3313 goto err_drain; 3314 } 3315 3316 /* restore the VM state */ 3317 f = qemu_fopen_bdrv(bs_vm_state, 0); 3318 if (!f) { 3319 error_setg(errp, "Could not open VM state file"); 3320 goto err_drain; 3321 } 3322 3323 qemu_system_reset(SHUTDOWN_CAUSE_SNAPSHOT_LOAD); 3324 mis->from_src_file = f; 3325 3326 if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) { 3327 ret = -EINVAL; 3328 goto err_drain; 3329 } 3330 ret = qemu_loadvm_state(f); 3331 migration_incoming_state_destroy(); 3332 3333 bdrv_drain_all_end(); 3334 3335 if (ret < 0) { 3336 error_setg(errp, "Error %d while loading VM state", ret); 3337 return false; 3338 } 3339 3340 return true; 3341 3342 err_drain: 3343 bdrv_drain_all_end(); 3344 return false; 3345 } 3346 3347 void load_snapshot_resume(RunState state) 3348 { 3349 vm_resume(state); 3350 if (state == RUN_STATE_RUNNING && runstate_get() == RUN_STATE_SUSPENDED) { 3351 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, &error_abort); 3352 } 3353 } 3354 3355 bool delete_snapshot(const char *name, bool has_devices, 3356 strList *devices, Error **errp) 3357 { 3358 if (!bdrv_all_can_snapshot(has_devices, devices, errp)) { 3359 return false; 3360 } 3361 3362 if (bdrv_all_delete_snapshot(name, has_devices, devices, errp) < 0) { 3363 return false; 3364 } 3365 3366 return true; 3367 } 3368 3369 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev) 3370 { 3371 qemu_ram_set_idstr(mr->ram_block, 3372 memory_region_name(mr), dev); 3373 qemu_ram_set_migratable(mr->ram_block); 3374 } 3375 3376 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev) 3377 { 3378 qemu_ram_unset_idstr(mr->ram_block); 3379 qemu_ram_unset_migratable(mr->ram_block); 3380 } 3381 3382 void vmstate_register_ram_global(MemoryRegion *mr) 3383 { 3384 vmstate_register_ram(mr, NULL); 3385 } 3386 3387 bool vmstate_check_only_migratable(const VMStateDescription *vmsd) 3388 { 3389 /* check needed if --only-migratable is specified */ 3390 if (!only_migratable) { 3391 return true; 3392 } 3393 3394 return !(vmsd && vmsd->unmigratable); 3395 } 3396 3397 typedef struct SnapshotJob { 3398 Job common; 3399 char *tag; 3400 char *vmstate; 3401 strList *devices; 3402 Coroutine *co; 3403 Error **errp; 3404 bool ret; 3405 } SnapshotJob; 3406 3407 static void qmp_snapshot_job_free(SnapshotJob *s) 3408 { 3409 g_free(s->tag); 3410 g_free(s->vmstate); 3411 qapi_free_strList(s->devices); 3412 } 3413 3414 3415 static void snapshot_load_job_bh(void *opaque) 3416 { 3417 Job *job = opaque; 3418 SnapshotJob *s = container_of(job, SnapshotJob, common); 3419 RunState orig_state = runstate_get(); 3420 3421 job_progress_set_remaining(&s->common, 1); 3422 3423 vm_stop(RUN_STATE_RESTORE_VM); 3424 3425 s->ret = load_snapshot(s->tag, s->vmstate, true, s->devices, s->errp); 3426 if (s->ret) { 3427 load_snapshot_resume(orig_state); 3428 } 3429 3430 job_progress_update(&s->common, 1); 3431 3432 qmp_snapshot_job_free(s); 3433 aio_co_wake(s->co); 3434 } 3435 3436 static void snapshot_save_job_bh(void *opaque) 3437 { 3438 Job *job = opaque; 3439 SnapshotJob *s = container_of(job, SnapshotJob, common); 3440 3441 job_progress_set_remaining(&s->common, 1); 3442 s->ret = save_snapshot(s->tag, false, s->vmstate, 3443 true, s->devices, s->errp); 3444 job_progress_update(&s->common, 1); 3445 3446 qmp_snapshot_job_free(s); 3447 aio_co_wake(s->co); 3448 } 3449 3450 static void snapshot_delete_job_bh(void *opaque) 3451 { 3452 Job *job = opaque; 3453 SnapshotJob *s = container_of(job, SnapshotJob, common); 3454 3455 job_progress_set_remaining(&s->common, 1); 3456 s->ret = delete_snapshot(s->tag, true, s->devices, s->errp); 3457 job_progress_update(&s->common, 1); 3458 3459 qmp_snapshot_job_free(s); 3460 aio_co_wake(s->co); 3461 } 3462 3463 static int coroutine_fn snapshot_save_job_run(Job *job, Error **errp) 3464 { 3465 SnapshotJob *s = container_of(job, SnapshotJob, common); 3466 s->errp = errp; 3467 s->co = qemu_coroutine_self(); 3468 aio_bh_schedule_oneshot(qemu_get_aio_context(), 3469 snapshot_save_job_bh, job); 3470 qemu_coroutine_yield(); 3471 return s->ret ? 0 : -1; 3472 } 3473 3474 static int coroutine_fn snapshot_load_job_run(Job *job, Error **errp) 3475 { 3476 SnapshotJob *s = container_of(job, SnapshotJob, common); 3477 s->errp = errp; 3478 s->co = qemu_coroutine_self(); 3479 aio_bh_schedule_oneshot(qemu_get_aio_context(), 3480 snapshot_load_job_bh, job); 3481 qemu_coroutine_yield(); 3482 return s->ret ? 0 : -1; 3483 } 3484 3485 static int coroutine_fn snapshot_delete_job_run(Job *job, Error **errp) 3486 { 3487 SnapshotJob *s = container_of(job, SnapshotJob, common); 3488 s->errp = errp; 3489 s->co = qemu_coroutine_self(); 3490 aio_bh_schedule_oneshot(qemu_get_aio_context(), 3491 snapshot_delete_job_bh, job); 3492 qemu_coroutine_yield(); 3493 return s->ret ? 0 : -1; 3494 } 3495 3496 3497 static const JobDriver snapshot_load_job_driver = { 3498 .instance_size = sizeof(SnapshotJob), 3499 .job_type = JOB_TYPE_SNAPSHOT_LOAD, 3500 .run = snapshot_load_job_run, 3501 }; 3502 3503 static const JobDriver snapshot_save_job_driver = { 3504 .instance_size = sizeof(SnapshotJob), 3505 .job_type = JOB_TYPE_SNAPSHOT_SAVE, 3506 .run = snapshot_save_job_run, 3507 }; 3508 3509 static const JobDriver snapshot_delete_job_driver = { 3510 .instance_size = sizeof(SnapshotJob), 3511 .job_type = JOB_TYPE_SNAPSHOT_DELETE, 3512 .run = snapshot_delete_job_run, 3513 }; 3514 3515 3516 void qmp_snapshot_save(const char *job_id, 3517 const char *tag, 3518 const char *vmstate, 3519 strList *devices, 3520 Error **errp) 3521 { 3522 SnapshotJob *s; 3523 3524 s = job_create(job_id, &snapshot_save_job_driver, NULL, 3525 qemu_get_aio_context(), JOB_MANUAL_DISMISS, 3526 NULL, NULL, errp); 3527 if (!s) { 3528 return; 3529 } 3530 3531 s->tag = g_strdup(tag); 3532 s->vmstate = g_strdup(vmstate); 3533 s->devices = QAPI_CLONE(strList, devices); 3534 3535 job_start(&s->common); 3536 } 3537 3538 void qmp_snapshot_load(const char *job_id, 3539 const char *tag, 3540 const char *vmstate, 3541 strList *devices, 3542 Error **errp) 3543 { 3544 SnapshotJob *s; 3545 3546 s = job_create(job_id, &snapshot_load_job_driver, NULL, 3547 qemu_get_aio_context(), JOB_MANUAL_DISMISS, 3548 NULL, NULL, errp); 3549 if (!s) { 3550 return; 3551 } 3552 3553 s->tag = g_strdup(tag); 3554 s->vmstate = g_strdup(vmstate); 3555 s->devices = QAPI_CLONE(strList, devices); 3556 3557 job_start(&s->common); 3558 } 3559 3560 void qmp_snapshot_delete(const char *job_id, 3561 const char *tag, 3562 strList *devices, 3563 Error **errp) 3564 { 3565 SnapshotJob *s; 3566 3567 s = job_create(job_id, &snapshot_delete_job_driver, NULL, 3568 qemu_get_aio_context(), JOB_MANUAL_DISMISS, 3569 NULL, NULL, errp); 3570 if (!s) { 3571 return; 3572 } 3573 3574 s->tag = g_strdup(tag); 3575 s->devices = QAPI_CLONE(strList, devices); 3576 3577 job_start(&s->common); 3578 } 3579