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 "hw/xen/xen.h" 32 #include "net/net.h" 33 #include "migration.h" 34 #include "migration/snapshot.h" 35 #include "migration/vmstate.h" 36 #include "migration/misc.h" 37 #include "migration/register.h" 38 #include "migration/global_state.h" 39 #include "ram.h" 40 #include "qemu-file-channel.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/qapi-commands-misc.h" 47 #include "qapi/qmp/qerror.h" 48 #include "qemu/error-report.h" 49 #include "sysemu/cpus.h" 50 #include "exec/memory.h" 51 #include "exec/target_page.h" 52 #include "trace.h" 53 #include "qemu/iov.h" 54 #include "qemu/main-loop.h" 55 #include "block/snapshot.h" 56 #include "qemu/cutils.h" 57 #include "io/channel-buffer.h" 58 #include "io/channel-file.h" 59 #include "sysemu/replay.h" 60 #include "sysemu/runstate.h" 61 #include "sysemu/sysemu.h" 62 #include "qjson.h" 63 #include "migration/colo.h" 64 #include "qemu/bitmap.h" 65 #include "net/announce.h" 66 67 const unsigned int postcopy_ram_discard_version = 0; 68 69 /* Subcommands for QEMU_VM_COMMAND */ 70 enum qemu_vm_cmd { 71 MIG_CMD_INVALID = 0, /* Must be 0 */ 72 MIG_CMD_OPEN_RETURN_PATH, /* Tell the dest to open the Return path */ 73 MIG_CMD_PING, /* Request a PONG on the RP */ 74 75 MIG_CMD_POSTCOPY_ADVISE, /* Prior to any page transfers, just 76 warn we might want to do PC */ 77 MIG_CMD_POSTCOPY_LISTEN, /* Start listening for incoming 78 pages as it's running. */ 79 MIG_CMD_POSTCOPY_RUN, /* Start execution */ 80 81 MIG_CMD_POSTCOPY_RAM_DISCARD, /* A list of pages to discard that 82 were previously sent during 83 precopy but are dirty. */ 84 MIG_CMD_PACKAGED, /* Send a wrapped stream within this stream */ 85 MIG_CMD_ENABLE_COLO, /* Enable COLO */ 86 MIG_CMD_POSTCOPY_RESUME, /* resume postcopy on dest */ 87 MIG_CMD_RECV_BITMAP, /* Request for recved bitmap on dst */ 88 MIG_CMD_MAX 89 }; 90 91 #define MAX_VM_CMD_PACKAGED_SIZE UINT32_MAX 92 static struct mig_cmd_args { 93 ssize_t len; /* -1 = variable */ 94 const char *name; 95 } mig_cmd_args[] = { 96 [MIG_CMD_INVALID] = { .len = -1, .name = "INVALID" }, 97 [MIG_CMD_OPEN_RETURN_PATH] = { .len = 0, .name = "OPEN_RETURN_PATH" }, 98 [MIG_CMD_PING] = { .len = sizeof(uint32_t), .name = "PING" }, 99 [MIG_CMD_POSTCOPY_ADVISE] = { .len = -1, .name = "POSTCOPY_ADVISE" }, 100 [MIG_CMD_POSTCOPY_LISTEN] = { .len = 0, .name = "POSTCOPY_LISTEN" }, 101 [MIG_CMD_POSTCOPY_RUN] = { .len = 0, .name = "POSTCOPY_RUN" }, 102 [MIG_CMD_POSTCOPY_RAM_DISCARD] = { 103 .len = -1, .name = "POSTCOPY_RAM_DISCARD" }, 104 [MIG_CMD_POSTCOPY_RESUME] = { .len = 0, .name = "POSTCOPY_RESUME" }, 105 [MIG_CMD_PACKAGED] = { .len = 4, .name = "PACKAGED" }, 106 [MIG_CMD_RECV_BITMAP] = { .len = -1, .name = "RECV_BITMAP" }, 107 [MIG_CMD_MAX] = { .len = -1, .name = "MAX" }, 108 }; 109 110 /* Note for MIG_CMD_POSTCOPY_ADVISE: 111 * The format of arguments is depending on postcopy mode: 112 * - postcopy RAM only 113 * uint64_t host page size 114 * uint64_t taget page size 115 * 116 * - postcopy RAM and postcopy dirty bitmaps 117 * format is the same as for postcopy RAM only 118 * 119 * - postcopy dirty bitmaps only 120 * Nothing. Command length field is 0. 121 * 122 * Be careful: adding a new postcopy entity with some other parameters should 123 * not break format self-description ability. Good way is to introduce some 124 * generic extendable format with an exception for two old entities. 125 */ 126 127 /***********************************************************/ 128 /* savevm/loadvm support */ 129 130 static ssize_t block_writev_buffer(void *opaque, struct iovec *iov, int iovcnt, 131 int64_t pos, Error **errp) 132 { 133 int ret; 134 QEMUIOVector qiov; 135 136 qemu_iovec_init_external(&qiov, iov, iovcnt); 137 ret = bdrv_writev_vmstate(opaque, &qiov, pos); 138 if (ret < 0) { 139 return ret; 140 } 141 142 return qiov.size; 143 } 144 145 static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos, 146 size_t size, Error **errp) 147 { 148 return bdrv_load_vmstate(opaque, buf, pos, size); 149 } 150 151 static int bdrv_fclose(void *opaque, Error **errp) 152 { 153 return bdrv_flush(opaque); 154 } 155 156 static const QEMUFileOps bdrv_read_ops = { 157 .get_buffer = block_get_buffer, 158 .close = bdrv_fclose 159 }; 160 161 static const QEMUFileOps bdrv_write_ops = { 162 .writev_buffer = block_writev_buffer, 163 .close = bdrv_fclose 164 }; 165 166 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable) 167 { 168 if (is_writable) { 169 return qemu_fopen_ops(bs, &bdrv_write_ops); 170 } 171 return qemu_fopen_ops(bs, &bdrv_read_ops); 172 } 173 174 175 /* QEMUFile timer support. 176 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c 177 */ 178 179 void timer_put(QEMUFile *f, QEMUTimer *ts) 180 { 181 uint64_t expire_time; 182 183 expire_time = timer_expire_time_ns(ts); 184 qemu_put_be64(f, expire_time); 185 } 186 187 void timer_get(QEMUFile *f, QEMUTimer *ts) 188 { 189 uint64_t expire_time; 190 191 expire_time = qemu_get_be64(f); 192 if (expire_time != -1) { 193 timer_mod_ns(ts, expire_time); 194 } else { 195 timer_del(ts); 196 } 197 } 198 199 200 /* VMState timer support. 201 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c 202 */ 203 204 static int get_timer(QEMUFile *f, void *pv, size_t size, 205 const VMStateField *field) 206 { 207 QEMUTimer *v = pv; 208 timer_get(f, v); 209 return 0; 210 } 211 212 static int put_timer(QEMUFile *f, void *pv, size_t size, 213 const VMStateField *field, QJSON *vmdesc) 214 { 215 QEMUTimer *v = pv; 216 timer_put(f, v); 217 218 return 0; 219 } 220 221 const VMStateInfo vmstate_info_timer = { 222 .name = "timer", 223 .get = get_timer, 224 .put = put_timer, 225 }; 226 227 228 typedef struct CompatEntry { 229 char idstr[256]; 230 int instance_id; 231 } CompatEntry; 232 233 typedef struct SaveStateEntry { 234 QTAILQ_ENTRY(SaveStateEntry) entry; 235 char idstr[256]; 236 uint32_t instance_id; 237 int alias_id; 238 int version_id; 239 /* version id read from the stream */ 240 int load_version_id; 241 int section_id; 242 /* section id read from the stream */ 243 int load_section_id; 244 const SaveVMHandlers *ops; 245 const VMStateDescription *vmsd; 246 void *opaque; 247 CompatEntry *compat; 248 int is_ram; 249 } SaveStateEntry; 250 251 typedef struct SaveState { 252 QTAILQ_HEAD(, SaveStateEntry) handlers; 253 SaveStateEntry *handler_pri_head[MIG_PRI_MAX + 1]; 254 int global_section_id; 255 uint32_t len; 256 const char *name; 257 uint32_t target_page_bits; 258 uint32_t caps_count; 259 MigrationCapability *capabilities; 260 QemuUUID uuid; 261 } SaveState; 262 263 static SaveState savevm_state = { 264 .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers), 265 .handler_pri_head = { [MIG_PRI_DEFAULT ... MIG_PRI_MAX] = NULL }, 266 .global_section_id = 0, 267 }; 268 269 static bool should_validate_capability(int capability) 270 { 271 assert(capability >= 0 && capability < MIGRATION_CAPABILITY__MAX); 272 /* Validate only new capabilities to keep compatibility. */ 273 switch (capability) { 274 case MIGRATION_CAPABILITY_X_IGNORE_SHARED: 275 return true; 276 default: 277 return false; 278 } 279 } 280 281 static uint32_t get_validatable_capabilities_count(void) 282 { 283 MigrationState *s = migrate_get_current(); 284 uint32_t result = 0; 285 int i; 286 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { 287 if (should_validate_capability(i) && s->enabled_capabilities[i]) { 288 result++; 289 } 290 } 291 return result; 292 } 293 294 static int configuration_pre_save(void *opaque) 295 { 296 SaveState *state = opaque; 297 const char *current_name = MACHINE_GET_CLASS(current_machine)->name; 298 MigrationState *s = migrate_get_current(); 299 int i, j; 300 301 state->len = strlen(current_name); 302 state->name = current_name; 303 state->target_page_bits = qemu_target_page_bits(); 304 305 state->caps_count = get_validatable_capabilities_count(); 306 state->capabilities = g_renew(MigrationCapability, state->capabilities, 307 state->caps_count); 308 for (i = j = 0; i < MIGRATION_CAPABILITY__MAX; i++) { 309 if (should_validate_capability(i) && s->enabled_capabilities[i]) { 310 state->capabilities[j++] = i; 311 } 312 } 313 state->uuid = qemu_uuid; 314 315 return 0; 316 } 317 318 static int configuration_pre_load(void *opaque) 319 { 320 SaveState *state = opaque; 321 322 /* If there is no target-page-bits subsection it means the source 323 * predates the variable-target-page-bits support and is using the 324 * minimum possible value for this CPU. 325 */ 326 state->target_page_bits = qemu_target_page_bits_min(); 327 return 0; 328 } 329 330 static bool configuration_validate_capabilities(SaveState *state) 331 { 332 bool ret = true; 333 MigrationState *s = migrate_get_current(); 334 unsigned long *source_caps_bm; 335 int i; 336 337 source_caps_bm = bitmap_new(MIGRATION_CAPABILITY__MAX); 338 for (i = 0; i < state->caps_count; i++) { 339 MigrationCapability capability = state->capabilities[i]; 340 set_bit(capability, source_caps_bm); 341 } 342 343 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { 344 bool source_state, target_state; 345 if (!should_validate_capability(i)) { 346 continue; 347 } 348 source_state = test_bit(i, source_caps_bm); 349 target_state = s->enabled_capabilities[i]; 350 if (source_state != target_state) { 351 error_report("Capability %s is %s, but received capability is %s", 352 MigrationCapability_str(i), 353 target_state ? "on" : "off", 354 source_state ? "on" : "off"); 355 ret = false; 356 /* Don't break here to report all failed capabilities */ 357 } 358 } 359 360 g_free(source_caps_bm); 361 return ret; 362 } 363 364 static int configuration_post_load(void *opaque, int version_id) 365 { 366 SaveState *state = opaque; 367 const char *current_name = MACHINE_GET_CLASS(current_machine)->name; 368 369 if (strncmp(state->name, current_name, state->len) != 0) { 370 error_report("Machine type received is '%.*s' and local is '%s'", 371 (int) state->len, state->name, current_name); 372 return -EINVAL; 373 } 374 375 if (state->target_page_bits != qemu_target_page_bits()) { 376 error_report("Received TARGET_PAGE_BITS is %d but local is %d", 377 state->target_page_bits, qemu_target_page_bits()); 378 return -EINVAL; 379 } 380 381 if (!configuration_validate_capabilities(state)) { 382 return -EINVAL; 383 } 384 385 return 0; 386 } 387 388 static int get_capability(QEMUFile *f, void *pv, size_t size, 389 const VMStateField *field) 390 { 391 MigrationCapability *capability = pv; 392 char capability_str[UINT8_MAX + 1]; 393 uint8_t len; 394 int i; 395 396 len = qemu_get_byte(f); 397 qemu_get_buffer(f, (uint8_t *)capability_str, len); 398 capability_str[len] = '\0'; 399 for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) { 400 if (!strcmp(MigrationCapability_str(i), capability_str)) { 401 *capability = i; 402 return 0; 403 } 404 } 405 error_report("Received unknown capability %s", capability_str); 406 return -EINVAL; 407 } 408 409 static int put_capability(QEMUFile *f, void *pv, size_t size, 410 const VMStateField *field, QJSON *vmdesc) 411 { 412 MigrationCapability *capability = pv; 413 const char *capability_str = MigrationCapability_str(*capability); 414 size_t len = strlen(capability_str); 415 assert(len <= UINT8_MAX); 416 417 qemu_put_byte(f, len); 418 qemu_put_buffer(f, (uint8_t *)capability_str, len); 419 return 0; 420 } 421 422 static const VMStateInfo vmstate_info_capability = { 423 .name = "capability", 424 .get = get_capability, 425 .put = put_capability, 426 }; 427 428 /* The target-page-bits subsection is present only if the 429 * target page size is not the same as the default (ie the 430 * minimum page size for a variable-page-size guest CPU). 431 * If it is present then it contains the actual target page 432 * bits for the machine, and migration will fail if the 433 * two ends don't agree about it. 434 */ 435 static bool vmstate_target_page_bits_needed(void *opaque) 436 { 437 return qemu_target_page_bits() 438 > qemu_target_page_bits_min(); 439 } 440 441 static const VMStateDescription vmstate_target_page_bits = { 442 .name = "configuration/target-page-bits", 443 .version_id = 1, 444 .minimum_version_id = 1, 445 .needed = vmstate_target_page_bits_needed, 446 .fields = (VMStateField[]) { 447 VMSTATE_UINT32(target_page_bits, SaveState), 448 VMSTATE_END_OF_LIST() 449 } 450 }; 451 452 static bool vmstate_capabilites_needed(void *opaque) 453 { 454 return get_validatable_capabilities_count() > 0; 455 } 456 457 static const VMStateDescription vmstate_capabilites = { 458 .name = "configuration/capabilities", 459 .version_id = 1, 460 .minimum_version_id = 1, 461 .needed = vmstate_capabilites_needed, 462 .fields = (VMStateField[]) { 463 VMSTATE_UINT32_V(caps_count, SaveState, 1), 464 VMSTATE_VARRAY_UINT32_ALLOC(capabilities, SaveState, caps_count, 1, 465 vmstate_info_capability, 466 MigrationCapability), 467 VMSTATE_END_OF_LIST() 468 } 469 }; 470 471 static bool vmstate_uuid_needed(void *opaque) 472 { 473 return qemu_uuid_set && migrate_validate_uuid(); 474 } 475 476 static int vmstate_uuid_post_load(void *opaque, int version_id) 477 { 478 SaveState *state = opaque; 479 char uuid_src[UUID_FMT_LEN + 1]; 480 char uuid_dst[UUID_FMT_LEN + 1]; 481 482 if (!qemu_uuid_set) { 483 /* 484 * It's warning because user might not know UUID in some cases, 485 * e.g. load an old snapshot 486 */ 487 qemu_uuid_unparse(&state->uuid, uuid_src); 488 warn_report("UUID is received %s, but local uuid isn't set", 489 uuid_src); 490 return 0; 491 } 492 if (!qemu_uuid_is_equal(&state->uuid, &qemu_uuid)) { 493 qemu_uuid_unparse(&state->uuid, uuid_src); 494 qemu_uuid_unparse(&qemu_uuid, uuid_dst); 495 error_report("UUID received is %s and local is %s", uuid_src, uuid_dst); 496 return -EINVAL; 497 } 498 return 0; 499 } 500 501 static const VMStateDescription vmstate_uuid = { 502 .name = "configuration/uuid", 503 .version_id = 1, 504 .minimum_version_id = 1, 505 .needed = vmstate_uuid_needed, 506 .post_load = vmstate_uuid_post_load, 507 .fields = (VMStateField[]) { 508 VMSTATE_UINT8_ARRAY_V(uuid.data, SaveState, sizeof(QemuUUID), 1), 509 VMSTATE_END_OF_LIST() 510 } 511 }; 512 513 static const VMStateDescription vmstate_configuration = { 514 .name = "configuration", 515 .version_id = 1, 516 .pre_load = configuration_pre_load, 517 .post_load = configuration_post_load, 518 .pre_save = configuration_pre_save, 519 .fields = (VMStateField[]) { 520 VMSTATE_UINT32(len, SaveState), 521 VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len), 522 VMSTATE_END_OF_LIST() 523 }, 524 .subsections = (const VMStateDescription*[]) { 525 &vmstate_target_page_bits, 526 &vmstate_capabilites, 527 &vmstate_uuid, 528 NULL 529 } 530 }; 531 532 static void dump_vmstate_vmsd(FILE *out_file, 533 const VMStateDescription *vmsd, int indent, 534 bool is_subsection); 535 536 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field, 537 int indent) 538 { 539 fprintf(out_file, "%*s{\n", indent, ""); 540 indent += 2; 541 fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name); 542 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "", 543 field->version_id); 544 fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "", 545 field->field_exists ? "true" : "false"); 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 fprintf(out_file, "\n%*s]", indent, ""); 598 } 599 if (vmsd->subsections != NULL) { 600 const VMStateDescription **subsection = vmsd->subsections; 601 bool first; 602 603 fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, ""); 604 first = true; 605 while (*subsection != NULL) { 606 if (!first) { 607 fprintf(out_file, ",\n"); 608 } 609 dump_vmstate_vmss(out_file, subsection, indent + 2); 610 subsection++; 611 first = false; 612 } 613 fprintf(out_file, "\n%*s]", indent, ""); 614 } 615 fprintf(out_file, "\n%*s}", indent - 2, ""); 616 } 617 618 static void dump_machine_type(FILE *out_file) 619 { 620 MachineClass *mc; 621 622 mc = MACHINE_GET_CLASS(current_machine); 623 624 fprintf(out_file, " \"vmschkmachine\": {\n"); 625 fprintf(out_file, " \"Name\": \"%s\"\n", mc->name); 626 fprintf(out_file, " },\n"); 627 } 628 629 void dump_vmstate_json_to_file(FILE *out_file) 630 { 631 GSList *list, *elt; 632 bool first; 633 634 fprintf(out_file, "{\n"); 635 dump_machine_type(out_file); 636 637 first = true; 638 list = object_class_get_list(TYPE_DEVICE, true); 639 for (elt = list; elt; elt = elt->next) { 640 DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data, 641 TYPE_DEVICE); 642 const char *name; 643 int indent = 2; 644 645 if (!dc->vmsd) { 646 continue; 647 } 648 649 if (!first) { 650 fprintf(out_file, ",\n"); 651 } 652 name = object_class_get_name(OBJECT_CLASS(dc)); 653 fprintf(out_file, "%*s\"%s\": {\n", indent, "", name); 654 indent += 2; 655 fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name); 656 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "", 657 dc->vmsd->version_id); 658 fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "", 659 dc->vmsd->minimum_version_id); 660 661 dump_vmstate_vmsd(out_file, dc->vmsd, indent, false); 662 663 fprintf(out_file, "\n%*s}", indent - 2, ""); 664 first = false; 665 } 666 fprintf(out_file, "\n}\n"); 667 fclose(out_file); 668 } 669 670 static uint32_t calculate_new_instance_id(const char *idstr) 671 { 672 SaveStateEntry *se; 673 uint32_t instance_id = 0; 674 675 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 676 if (strcmp(idstr, se->idstr) == 0 677 && instance_id <= se->instance_id) { 678 instance_id = se->instance_id + 1; 679 } 680 } 681 /* Make sure we never loop over without being noticed */ 682 assert(instance_id != VMSTATE_INSTANCE_ID_ANY); 683 return instance_id; 684 } 685 686 static int calculate_compat_instance_id(const char *idstr) 687 { 688 SaveStateEntry *se; 689 int instance_id = 0; 690 691 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 692 if (!se->compat) { 693 continue; 694 } 695 696 if (strcmp(idstr, se->compat->idstr) == 0 697 && instance_id <= se->compat->instance_id) { 698 instance_id = se->compat->instance_id + 1; 699 } 700 } 701 return instance_id; 702 } 703 704 static inline MigrationPriority save_state_priority(SaveStateEntry *se) 705 { 706 if (se->vmsd) { 707 return se->vmsd->priority; 708 } 709 return MIG_PRI_DEFAULT; 710 } 711 712 static void savevm_state_handler_insert(SaveStateEntry *nse) 713 { 714 MigrationPriority priority = save_state_priority(nse); 715 SaveStateEntry *se; 716 int i; 717 718 assert(priority <= MIG_PRI_MAX); 719 720 for (i = priority - 1; i >= 0; i--) { 721 se = savevm_state.handler_pri_head[i]; 722 if (se != NULL) { 723 assert(save_state_priority(se) < priority); 724 break; 725 } 726 } 727 728 if (i >= 0) { 729 QTAILQ_INSERT_BEFORE(se, nse, entry); 730 } else { 731 QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry); 732 } 733 734 if (savevm_state.handler_pri_head[priority] == NULL) { 735 savevm_state.handler_pri_head[priority] = nse; 736 } 737 } 738 739 static void savevm_state_handler_remove(SaveStateEntry *se) 740 { 741 SaveStateEntry *next; 742 MigrationPriority priority = save_state_priority(se); 743 744 if (se == savevm_state.handler_pri_head[priority]) { 745 next = QTAILQ_NEXT(se, entry); 746 if (next != NULL && save_state_priority(next) == priority) { 747 savevm_state.handler_pri_head[priority] = next; 748 } else { 749 savevm_state.handler_pri_head[priority] = NULL; 750 } 751 } 752 QTAILQ_REMOVE(&savevm_state.handlers, se, entry); 753 } 754 755 /* TODO: Individual devices generally have very little idea about the rest 756 of the system, so instance_id should be removed/replaced. 757 Meanwhile pass -1 as instance_id if you do not already have a clearly 758 distinguishing id for all instances of your device class. */ 759 int register_savevm_live(const char *idstr, 760 uint32_t instance_id, 761 int version_id, 762 const SaveVMHandlers *ops, 763 void *opaque) 764 { 765 SaveStateEntry *se; 766 767 se = g_new0(SaveStateEntry, 1); 768 se->version_id = version_id; 769 se->section_id = savevm_state.global_section_id++; 770 se->ops = ops; 771 se->opaque = opaque; 772 se->vmsd = NULL; 773 /* if this is a live_savem then set is_ram */ 774 if (ops->save_setup != NULL) { 775 se->is_ram = 1; 776 } 777 778 pstrcat(se->idstr, sizeof(se->idstr), idstr); 779 780 if (instance_id == VMSTATE_INSTANCE_ID_ANY) { 781 se->instance_id = calculate_new_instance_id(se->idstr); 782 } else { 783 se->instance_id = instance_id; 784 } 785 assert(!se->compat || se->instance_id == 0); 786 savevm_state_handler_insert(se); 787 return 0; 788 } 789 790 void unregister_savevm(VMStateIf *obj, const char *idstr, void *opaque) 791 { 792 SaveStateEntry *se, *new_se; 793 char id[256] = ""; 794 795 if (obj) { 796 char *oid = vmstate_if_get_id(obj); 797 if (oid) { 798 pstrcpy(id, sizeof(id), oid); 799 pstrcat(id, sizeof(id), "/"); 800 g_free(oid); 801 } 802 } 803 pstrcat(id, sizeof(id), idstr); 804 805 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) { 806 if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) { 807 savevm_state_handler_remove(se); 808 g_free(se->compat); 809 g_free(se); 810 } 811 } 812 } 813 814 int vmstate_register_with_alias_id(VMStateIf *obj, uint32_t instance_id, 815 const VMStateDescription *vmsd, 816 void *opaque, int alias_id, 817 int required_for_version, 818 Error **errp) 819 { 820 SaveStateEntry *se; 821 822 /* If this triggers, alias support can be dropped for the vmsd. */ 823 assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id); 824 825 se = g_new0(SaveStateEntry, 1); 826 se->version_id = vmsd->version_id; 827 se->section_id = savevm_state.global_section_id++; 828 se->opaque = opaque; 829 se->vmsd = vmsd; 830 se->alias_id = alias_id; 831 832 if (obj) { 833 char *id = vmstate_if_get_id(obj); 834 if (id) { 835 if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >= 836 sizeof(se->idstr)) { 837 error_setg(errp, "Path too long for VMState (%s)", id); 838 g_free(id); 839 g_free(se); 840 841 return -1; 842 } 843 g_free(id); 844 845 se->compat = g_new0(CompatEntry, 1); 846 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name); 847 se->compat->instance_id = instance_id == VMSTATE_INSTANCE_ID_ANY ? 848 calculate_compat_instance_id(vmsd->name) : instance_id; 849 instance_id = VMSTATE_INSTANCE_ID_ANY; 850 } 851 } 852 pstrcat(se->idstr, sizeof(se->idstr), vmsd->name); 853 854 if (instance_id == VMSTATE_INSTANCE_ID_ANY) { 855 se->instance_id = calculate_new_instance_id(se->idstr); 856 } else { 857 se->instance_id = instance_id; 858 } 859 assert(!se->compat || se->instance_id == 0); 860 savevm_state_handler_insert(se); 861 return 0; 862 } 863 864 void vmstate_unregister(VMStateIf *obj, const VMStateDescription *vmsd, 865 void *opaque) 866 { 867 SaveStateEntry *se, *new_se; 868 869 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) { 870 if (se->vmsd == vmsd && se->opaque == opaque) { 871 savevm_state_handler_remove(se); 872 g_free(se->compat); 873 g_free(se); 874 } 875 } 876 } 877 878 static int vmstate_load(QEMUFile *f, SaveStateEntry *se) 879 { 880 trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)"); 881 if (!se->vmsd) { /* Old style */ 882 return se->ops->load_state(f, se->opaque, se->load_version_id); 883 } 884 return vmstate_load_state(f, se->vmsd, se->opaque, se->load_version_id); 885 } 886 887 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc) 888 { 889 int64_t old_offset, size; 890 891 old_offset = qemu_ftell_fast(f); 892 se->ops->save_state(f, se->opaque); 893 size = qemu_ftell_fast(f) - old_offset; 894 895 if (vmdesc) { 896 json_prop_int(vmdesc, "size", size); 897 json_start_array(vmdesc, "fields"); 898 json_start_object(vmdesc, NULL); 899 json_prop_str(vmdesc, "name", "data"); 900 json_prop_int(vmdesc, "size", size); 901 json_prop_str(vmdesc, "type", "buffer"); 902 json_end_object(vmdesc); 903 json_end_array(vmdesc); 904 } 905 } 906 907 static int vmstate_save(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc) 908 { 909 trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)"); 910 if (!se->vmsd) { 911 vmstate_save_old_style(f, se, vmdesc); 912 return 0; 913 } 914 return vmstate_save_state(f, se->vmsd, se->opaque, vmdesc); 915 } 916 917 /* 918 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL) 919 */ 920 static void save_section_header(QEMUFile *f, SaveStateEntry *se, 921 uint8_t section_type) 922 { 923 qemu_put_byte(f, section_type); 924 qemu_put_be32(f, se->section_id); 925 926 if (section_type == QEMU_VM_SECTION_FULL || 927 section_type == QEMU_VM_SECTION_START) { 928 /* ID string */ 929 size_t len = strlen(se->idstr); 930 qemu_put_byte(f, len); 931 qemu_put_buffer(f, (uint8_t *)se->idstr, len); 932 933 qemu_put_be32(f, se->instance_id); 934 qemu_put_be32(f, se->version_id); 935 } 936 } 937 938 /* 939 * Write a footer onto device sections that catches cases misformatted device 940 * sections. 941 */ 942 static void save_section_footer(QEMUFile *f, SaveStateEntry *se) 943 { 944 if (migrate_get_current()->send_section_footer) { 945 qemu_put_byte(f, QEMU_VM_SECTION_FOOTER); 946 qemu_put_be32(f, se->section_id); 947 } 948 } 949 950 /** 951 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the 952 * command and associated data. 953 * 954 * @f: File to send command on 955 * @command: Command type to send 956 * @len: Length of associated data 957 * @data: Data associated with command. 958 */ 959 static void qemu_savevm_command_send(QEMUFile *f, 960 enum qemu_vm_cmd command, 961 uint16_t len, 962 uint8_t *data) 963 { 964 trace_savevm_command_send(command, len); 965 qemu_put_byte(f, QEMU_VM_COMMAND); 966 qemu_put_be16(f, (uint16_t)command); 967 qemu_put_be16(f, len); 968 qemu_put_buffer(f, data, len); 969 qemu_fflush(f); 970 } 971 972 void qemu_savevm_send_colo_enable(QEMUFile *f) 973 { 974 trace_savevm_send_colo_enable(); 975 qemu_savevm_command_send(f, MIG_CMD_ENABLE_COLO, 0, NULL); 976 } 977 978 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value) 979 { 980 uint32_t buf; 981 982 trace_savevm_send_ping(value); 983 buf = cpu_to_be32(value); 984 qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf); 985 } 986 987 void qemu_savevm_send_open_return_path(QEMUFile *f) 988 { 989 trace_savevm_send_open_return_path(); 990 qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL); 991 } 992 993 /* We have a buffer of data to send; we don't want that all to be loaded 994 * by the command itself, so the command contains just the length of the 995 * extra buffer that we then send straight after it. 996 * TODO: Must be a better way to organise that 997 * 998 * Returns: 999 * 0 on success 1000 * -ve on error 1001 */ 1002 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len) 1003 { 1004 uint32_t tmp; 1005 1006 if (len > MAX_VM_CMD_PACKAGED_SIZE) { 1007 error_report("%s: Unreasonably large packaged state: %zu", 1008 __func__, len); 1009 return -1; 1010 } 1011 1012 tmp = cpu_to_be32(len); 1013 1014 trace_qemu_savevm_send_packaged(); 1015 qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp); 1016 1017 qemu_put_buffer(f, buf, len); 1018 1019 return 0; 1020 } 1021 1022 /* Send prior to any postcopy transfer */ 1023 void qemu_savevm_send_postcopy_advise(QEMUFile *f) 1024 { 1025 if (migrate_postcopy_ram()) { 1026 uint64_t tmp[2]; 1027 tmp[0] = cpu_to_be64(ram_pagesize_summary()); 1028 tmp[1] = cpu_to_be64(qemu_target_page_size()); 1029 1030 trace_qemu_savevm_send_postcopy_advise(); 1031 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 1032 16, (uint8_t *)tmp); 1033 } else { 1034 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 0, NULL); 1035 } 1036 } 1037 1038 /* Sent prior to starting the destination running in postcopy, discard pages 1039 * that have already been sent but redirtied on the source. 1040 * CMD_POSTCOPY_RAM_DISCARD consist of: 1041 * byte version (0) 1042 * byte Length of name field (not including 0) 1043 * n x byte RAM block name 1044 * byte 0 terminator (just for safety) 1045 * n x Byte ranges within the named RAMBlock 1046 * be64 Start of the range 1047 * be64 Length 1048 * 1049 * name: RAMBlock name that these entries are part of 1050 * len: Number of page entries 1051 * start_list: 'len' addresses 1052 * length_list: 'len' addresses 1053 * 1054 */ 1055 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name, 1056 uint16_t len, 1057 uint64_t *start_list, 1058 uint64_t *length_list) 1059 { 1060 uint8_t *buf; 1061 uint16_t tmplen; 1062 uint16_t t; 1063 size_t name_len = strlen(name); 1064 1065 trace_qemu_savevm_send_postcopy_ram_discard(name, len); 1066 assert(name_len < 256); 1067 buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len); 1068 buf[0] = postcopy_ram_discard_version; 1069 buf[1] = name_len; 1070 memcpy(buf + 2, name, name_len); 1071 tmplen = 2 + name_len; 1072 buf[tmplen++] = '\0'; 1073 1074 for (t = 0; t < len; t++) { 1075 stq_be_p(buf + tmplen, start_list[t]); 1076 tmplen += 8; 1077 stq_be_p(buf + tmplen, length_list[t]); 1078 tmplen += 8; 1079 } 1080 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf); 1081 g_free(buf); 1082 } 1083 1084 /* Get the destination into a state where it can receive postcopy data. */ 1085 void qemu_savevm_send_postcopy_listen(QEMUFile *f) 1086 { 1087 trace_savevm_send_postcopy_listen(); 1088 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL); 1089 } 1090 1091 /* Kick the destination into running */ 1092 void qemu_savevm_send_postcopy_run(QEMUFile *f) 1093 { 1094 trace_savevm_send_postcopy_run(); 1095 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL); 1096 } 1097 1098 void qemu_savevm_send_postcopy_resume(QEMUFile *f) 1099 { 1100 trace_savevm_send_postcopy_resume(); 1101 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RESUME, 0, NULL); 1102 } 1103 1104 void qemu_savevm_send_recv_bitmap(QEMUFile *f, char *block_name) 1105 { 1106 size_t len; 1107 char buf[256]; 1108 1109 trace_savevm_send_recv_bitmap(block_name); 1110 1111 buf[0] = len = strlen(block_name); 1112 memcpy(buf + 1, block_name, len); 1113 1114 qemu_savevm_command_send(f, MIG_CMD_RECV_BITMAP, len + 1, (uint8_t *)buf); 1115 } 1116 1117 bool qemu_savevm_state_blocked(Error **errp) 1118 { 1119 SaveStateEntry *se; 1120 1121 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1122 if (se->vmsd && se->vmsd->unmigratable) { 1123 error_setg(errp, "State blocked by non-migratable device '%s'", 1124 se->idstr); 1125 return true; 1126 } 1127 } 1128 return false; 1129 } 1130 1131 void qemu_savevm_state_header(QEMUFile *f) 1132 { 1133 trace_savevm_state_header(); 1134 qemu_put_be32(f, QEMU_VM_FILE_MAGIC); 1135 qemu_put_be32(f, QEMU_VM_FILE_VERSION); 1136 1137 if (migrate_get_current()->send_configuration) { 1138 qemu_put_byte(f, QEMU_VM_CONFIGURATION); 1139 vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0); 1140 } 1141 } 1142 1143 int qemu_savevm_nr_failover_devices(void) 1144 { 1145 SaveStateEntry *se; 1146 int n = 0; 1147 1148 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1149 if (se->vmsd && se->vmsd->dev_unplug_pending && 1150 se->vmsd->dev_unplug_pending(se->opaque)) { 1151 n++; 1152 } 1153 } 1154 1155 return n; 1156 } 1157 1158 bool qemu_savevm_state_guest_unplug_pending(void) 1159 { 1160 SaveStateEntry *se; 1161 int n = 0; 1162 1163 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1164 if (!se->vmsd || !se->vmsd->dev_unplug_pending) { 1165 continue; 1166 } 1167 if (se->vmsd->dev_unplug_pending(se->opaque)) { 1168 n++; 1169 } 1170 } 1171 1172 return n > 0; 1173 } 1174 1175 void qemu_savevm_state_setup(QEMUFile *f) 1176 { 1177 SaveStateEntry *se; 1178 Error *local_err = NULL; 1179 int ret; 1180 1181 trace_savevm_state_setup(); 1182 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1183 if (!se->ops || !se->ops->save_setup) { 1184 continue; 1185 } 1186 if (se->ops->is_active) { 1187 if (!se->ops->is_active(se->opaque)) { 1188 continue; 1189 } 1190 } 1191 save_section_header(f, se, QEMU_VM_SECTION_START); 1192 1193 ret = se->ops->save_setup(f, se->opaque); 1194 save_section_footer(f, se); 1195 if (ret < 0) { 1196 qemu_file_set_error(f, ret); 1197 break; 1198 } 1199 } 1200 1201 if (precopy_notify(PRECOPY_NOTIFY_SETUP, &local_err)) { 1202 error_report_err(local_err); 1203 } 1204 } 1205 1206 int qemu_savevm_state_resume_prepare(MigrationState *s) 1207 { 1208 SaveStateEntry *se; 1209 int ret; 1210 1211 trace_savevm_state_resume_prepare(); 1212 1213 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1214 if (!se->ops || !se->ops->resume_prepare) { 1215 continue; 1216 } 1217 if (se->ops->is_active) { 1218 if (!se->ops->is_active(se->opaque)) { 1219 continue; 1220 } 1221 } 1222 ret = se->ops->resume_prepare(s, se->opaque); 1223 if (ret < 0) { 1224 return ret; 1225 } 1226 } 1227 1228 return 0; 1229 } 1230 1231 /* 1232 * this function has three return values: 1233 * negative: there was one error, and we have -errno. 1234 * 0 : We haven't finished, caller have to go again 1235 * 1 : We have finished, we can go to complete phase 1236 */ 1237 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy) 1238 { 1239 SaveStateEntry *se; 1240 int ret = 1; 1241 1242 trace_savevm_state_iterate(); 1243 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1244 if (!se->ops || !se->ops->save_live_iterate) { 1245 continue; 1246 } 1247 if (se->ops->is_active && 1248 !se->ops->is_active(se->opaque)) { 1249 continue; 1250 } 1251 if (se->ops->is_active_iterate && 1252 !se->ops->is_active_iterate(se->opaque)) { 1253 continue; 1254 } 1255 /* 1256 * In the postcopy phase, any device that doesn't know how to 1257 * do postcopy should have saved it's state in the _complete 1258 * call that's already run, it might get confused if we call 1259 * iterate afterwards. 1260 */ 1261 if (postcopy && 1262 !(se->ops->has_postcopy && se->ops->has_postcopy(se->opaque))) { 1263 continue; 1264 } 1265 if (qemu_file_rate_limit(f)) { 1266 return 0; 1267 } 1268 trace_savevm_section_start(se->idstr, se->section_id); 1269 1270 save_section_header(f, se, QEMU_VM_SECTION_PART); 1271 1272 ret = se->ops->save_live_iterate(f, se->opaque); 1273 trace_savevm_section_end(se->idstr, se->section_id, ret); 1274 save_section_footer(f, se); 1275 1276 if (ret < 0) { 1277 error_report("failed to save SaveStateEntry with id(name): %d(%s)", 1278 se->section_id, se->idstr); 1279 qemu_file_set_error(f, ret); 1280 } 1281 if (ret <= 0) { 1282 /* Do not proceed to the next vmstate before this one reported 1283 completion of the current stage. This serializes the migration 1284 and reduces the probability that a faster changing state is 1285 synchronized over and over again. */ 1286 break; 1287 } 1288 } 1289 return ret; 1290 } 1291 1292 static bool should_send_vmdesc(void) 1293 { 1294 MachineState *machine = MACHINE(qdev_get_machine()); 1295 bool in_postcopy = migration_in_postcopy(); 1296 return !machine->suppress_vmdesc && !in_postcopy; 1297 } 1298 1299 /* 1300 * Calls the save_live_complete_postcopy methods 1301 * causing the last few pages to be sent immediately and doing any associated 1302 * cleanup. 1303 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete 1304 * all the other devices, but that happens at the point we switch to postcopy. 1305 */ 1306 void qemu_savevm_state_complete_postcopy(QEMUFile *f) 1307 { 1308 SaveStateEntry *se; 1309 int ret; 1310 1311 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1312 if (!se->ops || !se->ops->save_live_complete_postcopy) { 1313 continue; 1314 } 1315 if (se->ops->is_active) { 1316 if (!se->ops->is_active(se->opaque)) { 1317 continue; 1318 } 1319 } 1320 trace_savevm_section_start(se->idstr, se->section_id); 1321 /* Section type */ 1322 qemu_put_byte(f, QEMU_VM_SECTION_END); 1323 qemu_put_be32(f, se->section_id); 1324 1325 ret = se->ops->save_live_complete_postcopy(f, se->opaque); 1326 trace_savevm_section_end(se->idstr, se->section_id, ret); 1327 save_section_footer(f, se); 1328 if (ret < 0) { 1329 qemu_file_set_error(f, ret); 1330 return; 1331 } 1332 } 1333 1334 qemu_put_byte(f, QEMU_VM_EOF); 1335 qemu_fflush(f); 1336 } 1337 1338 static 1339 int qemu_savevm_state_complete_precopy_iterable(QEMUFile *f, bool in_postcopy) 1340 { 1341 SaveStateEntry *se; 1342 int ret; 1343 1344 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1345 if (!se->ops || 1346 (in_postcopy && se->ops->has_postcopy && 1347 se->ops->has_postcopy(se->opaque)) || 1348 !se->ops->save_live_complete_precopy) { 1349 continue; 1350 } 1351 1352 if (se->ops->is_active) { 1353 if (!se->ops->is_active(se->opaque)) { 1354 continue; 1355 } 1356 } 1357 trace_savevm_section_start(se->idstr, se->section_id); 1358 1359 save_section_header(f, se, QEMU_VM_SECTION_END); 1360 1361 ret = se->ops->save_live_complete_precopy(f, se->opaque); 1362 trace_savevm_section_end(se->idstr, se->section_id, ret); 1363 save_section_footer(f, se); 1364 if (ret < 0) { 1365 qemu_file_set_error(f, ret); 1366 return -1; 1367 } 1368 } 1369 1370 return 0; 1371 } 1372 1373 static 1374 int qemu_savevm_state_complete_precopy_non_iterable(QEMUFile *f, 1375 bool in_postcopy, 1376 bool inactivate_disks) 1377 { 1378 g_autoptr(QJSON) vmdesc = NULL; 1379 int vmdesc_len; 1380 SaveStateEntry *se; 1381 int ret; 1382 1383 vmdesc = qjson_new(); 1384 json_prop_int(vmdesc, "page_size", qemu_target_page_size()); 1385 json_start_array(vmdesc, "devices"); 1386 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1387 1388 if ((!se->ops || !se->ops->save_state) && !se->vmsd) { 1389 continue; 1390 } 1391 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) { 1392 trace_savevm_section_skip(se->idstr, se->section_id); 1393 continue; 1394 } 1395 1396 trace_savevm_section_start(se->idstr, se->section_id); 1397 1398 json_start_object(vmdesc, NULL); 1399 json_prop_str(vmdesc, "name", se->idstr); 1400 json_prop_int(vmdesc, "instance_id", se->instance_id); 1401 1402 save_section_header(f, se, QEMU_VM_SECTION_FULL); 1403 ret = vmstate_save(f, se, vmdesc); 1404 if (ret) { 1405 qemu_file_set_error(f, ret); 1406 return ret; 1407 } 1408 trace_savevm_section_end(se->idstr, se->section_id, 0); 1409 save_section_footer(f, se); 1410 1411 json_end_object(vmdesc); 1412 } 1413 1414 if (inactivate_disks) { 1415 /* Inactivate before sending QEMU_VM_EOF so that the 1416 * bdrv_invalidate_cache_all() on the other end won't fail. */ 1417 ret = bdrv_inactivate_all(); 1418 if (ret) { 1419 error_report("%s: bdrv_inactivate_all() failed (%d)", 1420 __func__, ret); 1421 qemu_file_set_error(f, ret); 1422 return ret; 1423 } 1424 } 1425 if (!in_postcopy) { 1426 /* Postcopy stream will still be going */ 1427 qemu_put_byte(f, QEMU_VM_EOF); 1428 } 1429 1430 json_end_array(vmdesc); 1431 qjson_finish(vmdesc); 1432 vmdesc_len = strlen(qjson_get_str(vmdesc)); 1433 1434 if (should_send_vmdesc()) { 1435 qemu_put_byte(f, QEMU_VM_VMDESCRIPTION); 1436 qemu_put_be32(f, vmdesc_len); 1437 qemu_put_buffer(f, (uint8_t *)qjson_get_str(vmdesc), vmdesc_len); 1438 } 1439 1440 return 0; 1441 } 1442 1443 int qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only, 1444 bool inactivate_disks) 1445 { 1446 int ret; 1447 Error *local_err = NULL; 1448 bool in_postcopy = migration_in_postcopy(); 1449 1450 if (precopy_notify(PRECOPY_NOTIFY_COMPLETE, &local_err)) { 1451 error_report_err(local_err); 1452 } 1453 1454 trace_savevm_state_complete_precopy(); 1455 1456 cpu_synchronize_all_states(); 1457 1458 if (!in_postcopy || iterable_only) { 1459 ret = qemu_savevm_state_complete_precopy_iterable(f, in_postcopy); 1460 if (ret) { 1461 return ret; 1462 } 1463 } 1464 1465 if (iterable_only) { 1466 goto flush; 1467 } 1468 1469 ret = qemu_savevm_state_complete_precopy_non_iterable(f, in_postcopy, 1470 inactivate_disks); 1471 if (ret) { 1472 return ret; 1473 } 1474 1475 flush: 1476 qemu_fflush(f); 1477 return 0; 1478 } 1479 1480 /* Give an estimate of the amount left to be transferred, 1481 * the result is split into the amount for units that can and 1482 * for units that can't do postcopy. 1483 */ 1484 void qemu_savevm_state_pending(QEMUFile *f, uint64_t threshold_size, 1485 uint64_t *res_precopy_only, 1486 uint64_t *res_compatible, 1487 uint64_t *res_postcopy_only) 1488 { 1489 SaveStateEntry *se; 1490 1491 *res_precopy_only = 0; 1492 *res_compatible = 0; 1493 *res_postcopy_only = 0; 1494 1495 1496 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1497 if (!se->ops || !se->ops->save_live_pending) { 1498 continue; 1499 } 1500 if (se->ops->is_active) { 1501 if (!se->ops->is_active(se->opaque)) { 1502 continue; 1503 } 1504 } 1505 se->ops->save_live_pending(f, se->opaque, threshold_size, 1506 res_precopy_only, res_compatible, 1507 res_postcopy_only); 1508 } 1509 } 1510 1511 void qemu_savevm_state_cleanup(void) 1512 { 1513 SaveStateEntry *se; 1514 Error *local_err = NULL; 1515 1516 if (precopy_notify(PRECOPY_NOTIFY_CLEANUP, &local_err)) { 1517 error_report_err(local_err); 1518 } 1519 1520 trace_savevm_state_cleanup(); 1521 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1522 if (se->ops && se->ops->save_cleanup) { 1523 se->ops->save_cleanup(se->opaque); 1524 } 1525 } 1526 } 1527 1528 static int qemu_savevm_state(QEMUFile *f, Error **errp) 1529 { 1530 int ret; 1531 MigrationState *ms = migrate_get_current(); 1532 MigrationStatus status; 1533 1534 if (migration_is_setup_or_active(ms->state) || 1535 ms->state == MIGRATION_STATUS_CANCELLING || 1536 ms->state == MIGRATION_STATUS_COLO) { 1537 error_setg(errp, QERR_MIGRATION_ACTIVE); 1538 return -EINVAL; 1539 } 1540 1541 if (migrate_use_block()) { 1542 error_setg(errp, "Block migration and snapshots are incompatible"); 1543 return -EINVAL; 1544 } 1545 1546 migrate_init(ms); 1547 memset(&ram_counters, 0, sizeof(ram_counters)); 1548 ms->to_dst_file = f; 1549 1550 qemu_mutex_unlock_iothread(); 1551 qemu_savevm_state_header(f); 1552 qemu_savevm_state_setup(f); 1553 qemu_mutex_lock_iothread(); 1554 1555 while (qemu_file_get_error(f) == 0) { 1556 if (qemu_savevm_state_iterate(f, false) > 0) { 1557 break; 1558 } 1559 } 1560 1561 ret = qemu_file_get_error(f); 1562 if (ret == 0) { 1563 qemu_savevm_state_complete_precopy(f, false, false); 1564 ret = qemu_file_get_error(f); 1565 } 1566 qemu_savevm_state_cleanup(); 1567 if (ret != 0) { 1568 error_setg_errno(errp, -ret, "Error while writing VM state"); 1569 } 1570 1571 if (ret != 0) { 1572 status = MIGRATION_STATUS_FAILED; 1573 } else { 1574 status = MIGRATION_STATUS_COMPLETED; 1575 } 1576 migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status); 1577 1578 /* f is outer parameter, it should not stay in global migration state after 1579 * this function finished */ 1580 ms->to_dst_file = NULL; 1581 1582 return ret; 1583 } 1584 1585 void qemu_savevm_live_state(QEMUFile *f) 1586 { 1587 /* save QEMU_VM_SECTION_END section */ 1588 qemu_savevm_state_complete_precopy(f, true, false); 1589 qemu_put_byte(f, QEMU_VM_EOF); 1590 } 1591 1592 int qemu_save_device_state(QEMUFile *f) 1593 { 1594 SaveStateEntry *se; 1595 1596 if (!migration_in_colo_state()) { 1597 qemu_put_be32(f, QEMU_VM_FILE_MAGIC); 1598 qemu_put_be32(f, QEMU_VM_FILE_VERSION); 1599 } 1600 cpu_synchronize_all_states(); 1601 1602 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1603 int ret; 1604 1605 if (se->is_ram) { 1606 continue; 1607 } 1608 if ((!se->ops || !se->ops->save_state) && !se->vmsd) { 1609 continue; 1610 } 1611 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) { 1612 continue; 1613 } 1614 1615 save_section_header(f, se, QEMU_VM_SECTION_FULL); 1616 1617 ret = vmstate_save(f, se, NULL); 1618 if (ret) { 1619 return ret; 1620 } 1621 1622 save_section_footer(f, se); 1623 } 1624 1625 qemu_put_byte(f, QEMU_VM_EOF); 1626 1627 return qemu_file_get_error(f); 1628 } 1629 1630 static SaveStateEntry *find_se(const char *idstr, uint32_t instance_id) 1631 { 1632 SaveStateEntry *se; 1633 1634 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1635 if (!strcmp(se->idstr, idstr) && 1636 (instance_id == se->instance_id || 1637 instance_id == se->alias_id)) 1638 return se; 1639 /* Migrating from an older version? */ 1640 if (strstr(se->idstr, idstr) && se->compat) { 1641 if (!strcmp(se->compat->idstr, idstr) && 1642 (instance_id == se->compat->instance_id || 1643 instance_id == se->alias_id)) 1644 return se; 1645 } 1646 } 1647 return NULL; 1648 } 1649 1650 enum LoadVMExitCodes { 1651 /* Allow a command to quit all layers of nested loadvm loops */ 1652 LOADVM_QUIT = 1, 1653 }; 1654 1655 /* ------ incoming postcopy messages ------ */ 1656 /* 'advise' arrives before any transfers just to tell us that a postcopy 1657 * *might* happen - it might be skipped if precopy transferred everything 1658 * quickly. 1659 */ 1660 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis, 1661 uint16_t len) 1662 { 1663 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE); 1664 uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps; 1665 Error *local_err = NULL; 1666 1667 trace_loadvm_postcopy_handle_advise(); 1668 if (ps != POSTCOPY_INCOMING_NONE) { 1669 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps); 1670 return -1; 1671 } 1672 1673 switch (len) { 1674 case 0: 1675 if (migrate_postcopy_ram()) { 1676 error_report("RAM postcopy is enabled but have 0 byte advise"); 1677 return -EINVAL; 1678 } 1679 return 0; 1680 case 8 + 8: 1681 if (!migrate_postcopy_ram()) { 1682 error_report("RAM postcopy is disabled but have 16 byte advise"); 1683 return -EINVAL; 1684 } 1685 break; 1686 default: 1687 error_report("CMD_POSTCOPY_ADVISE invalid length (%d)", len); 1688 return -EINVAL; 1689 } 1690 1691 if (!postcopy_ram_supported_by_host(mis)) { 1692 postcopy_state_set(POSTCOPY_INCOMING_NONE); 1693 return -1; 1694 } 1695 1696 remote_pagesize_summary = qemu_get_be64(mis->from_src_file); 1697 local_pagesize_summary = ram_pagesize_summary(); 1698 1699 if (remote_pagesize_summary != local_pagesize_summary) { 1700 /* 1701 * This detects two potential causes of mismatch: 1702 * a) A mismatch in host page sizes 1703 * Some combinations of mismatch are probably possible but it gets 1704 * a bit more complicated. In particular we need to place whole 1705 * host pages on the dest at once, and we need to ensure that we 1706 * handle dirtying to make sure we never end up sending part of 1707 * a hostpage on it's own. 1708 * b) The use of different huge page sizes on source/destination 1709 * a more fine grain test is performed during RAM block migration 1710 * but this test here causes a nice early clear failure, and 1711 * also fails when passed to an older qemu that doesn't 1712 * do huge pages. 1713 */ 1714 error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64 1715 " d=%" PRIx64 ")", 1716 remote_pagesize_summary, local_pagesize_summary); 1717 return -1; 1718 } 1719 1720 remote_tps = qemu_get_be64(mis->from_src_file); 1721 if (remote_tps != qemu_target_page_size()) { 1722 /* 1723 * Again, some differences could be dealt with, but for now keep it 1724 * simple. 1725 */ 1726 error_report("Postcopy needs matching target page sizes (s=%d d=%zd)", 1727 (int)remote_tps, qemu_target_page_size()); 1728 return -1; 1729 } 1730 1731 if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_ADVISE, &local_err)) { 1732 error_report_err(local_err); 1733 return -1; 1734 } 1735 1736 if (ram_postcopy_incoming_init(mis)) { 1737 return -1; 1738 } 1739 1740 return 0; 1741 } 1742 1743 /* After postcopy we will be told to throw some pages away since they're 1744 * dirty and will have to be demand fetched. Must happen before CPU is 1745 * started. 1746 * There can be 0..many of these messages, each encoding multiple pages. 1747 */ 1748 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis, 1749 uint16_t len) 1750 { 1751 int tmp; 1752 char ramid[256]; 1753 PostcopyState ps = postcopy_state_get(); 1754 1755 trace_loadvm_postcopy_ram_handle_discard(); 1756 1757 switch (ps) { 1758 case POSTCOPY_INCOMING_ADVISE: 1759 /* 1st discard */ 1760 tmp = postcopy_ram_prepare_discard(mis); 1761 if (tmp) { 1762 return tmp; 1763 } 1764 break; 1765 1766 case POSTCOPY_INCOMING_DISCARD: 1767 /* Expected state */ 1768 break; 1769 1770 default: 1771 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)", 1772 ps); 1773 return -1; 1774 } 1775 /* We're expecting a 1776 * Version (0) 1777 * a RAM ID string (length byte, name, 0 term) 1778 * then at least 1 16 byte chunk 1779 */ 1780 if (len < (1 + 1 + 1 + 1 + 2 * 8)) { 1781 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len); 1782 return -1; 1783 } 1784 1785 tmp = qemu_get_byte(mis->from_src_file); 1786 if (tmp != postcopy_ram_discard_version) { 1787 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp); 1788 return -1; 1789 } 1790 1791 if (!qemu_get_counted_string(mis->from_src_file, ramid)) { 1792 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID"); 1793 return -1; 1794 } 1795 tmp = qemu_get_byte(mis->from_src_file); 1796 if (tmp != 0) { 1797 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp); 1798 return -1; 1799 } 1800 1801 len -= 3 + strlen(ramid); 1802 if (len % 16) { 1803 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len); 1804 return -1; 1805 } 1806 trace_loadvm_postcopy_ram_handle_discard_header(ramid, len); 1807 while (len) { 1808 uint64_t start_addr, block_length; 1809 start_addr = qemu_get_be64(mis->from_src_file); 1810 block_length = qemu_get_be64(mis->from_src_file); 1811 1812 len -= 16; 1813 int ret = ram_discard_range(ramid, start_addr, block_length); 1814 if (ret) { 1815 return ret; 1816 } 1817 } 1818 trace_loadvm_postcopy_ram_handle_discard_end(); 1819 1820 return 0; 1821 } 1822 1823 /* 1824 * Triggered by a postcopy_listen command; this thread takes over reading 1825 * the input stream, leaving the main thread free to carry on loading the rest 1826 * of the device state (from RAM). 1827 * (TODO:This could do with being in a postcopy file - but there again it's 1828 * just another input loop, not that postcopy specific) 1829 */ 1830 static void *postcopy_ram_listen_thread(void *opaque) 1831 { 1832 MigrationIncomingState *mis = migration_incoming_get_current(); 1833 QEMUFile *f = mis->from_src_file; 1834 int load_res; 1835 1836 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, 1837 MIGRATION_STATUS_POSTCOPY_ACTIVE); 1838 qemu_sem_post(&mis->listen_thread_sem); 1839 trace_postcopy_ram_listen_thread_start(); 1840 1841 rcu_register_thread(); 1842 /* 1843 * Because we're a thread and not a coroutine we can't yield 1844 * in qemu_file, and thus we must be blocking now. 1845 */ 1846 qemu_file_set_blocking(f, true); 1847 load_res = qemu_loadvm_state_main(f, mis); 1848 1849 /* 1850 * This is tricky, but, mis->from_src_file can change after it 1851 * returns, when postcopy recovery happened. In the future, we may 1852 * want a wrapper for the QEMUFile handle. 1853 */ 1854 f = mis->from_src_file; 1855 1856 /* And non-blocking again so we don't block in any cleanup */ 1857 qemu_file_set_blocking(f, false); 1858 1859 trace_postcopy_ram_listen_thread_exit(); 1860 if (load_res < 0) { 1861 error_report("%s: loadvm failed: %d", __func__, load_res); 1862 qemu_file_set_error(f, load_res); 1863 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, 1864 MIGRATION_STATUS_FAILED); 1865 } else { 1866 /* 1867 * This looks good, but it's possible that the device loading in the 1868 * main thread hasn't finished yet, and so we might not be in 'RUN' 1869 * state yet; wait for the end of the main thread. 1870 */ 1871 qemu_event_wait(&mis->main_thread_load_event); 1872 } 1873 postcopy_ram_incoming_cleanup(mis); 1874 1875 if (load_res < 0) { 1876 /* 1877 * If something went wrong then we have a bad state so exit; 1878 * depending how far we got it might be possible at this point 1879 * to leave the guest running and fire MCEs for pages that never 1880 * arrived as a desperate recovery step. 1881 */ 1882 rcu_unregister_thread(); 1883 exit(EXIT_FAILURE); 1884 } 1885 1886 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, 1887 MIGRATION_STATUS_COMPLETED); 1888 /* 1889 * If everything has worked fine, then the main thread has waited 1890 * for us to start, and we're the last use of the mis. 1891 * (If something broke then qemu will have to exit anyway since it's 1892 * got a bad migration state). 1893 */ 1894 migration_incoming_state_destroy(); 1895 qemu_loadvm_state_cleanup(); 1896 1897 rcu_unregister_thread(); 1898 mis->have_listen_thread = false; 1899 postcopy_state_set(POSTCOPY_INCOMING_END); 1900 1901 return NULL; 1902 } 1903 1904 /* After this message we must be able to immediately receive postcopy data */ 1905 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis) 1906 { 1907 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING); 1908 trace_loadvm_postcopy_handle_listen(); 1909 Error *local_err = NULL; 1910 1911 if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) { 1912 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps); 1913 return -1; 1914 } 1915 if (ps == POSTCOPY_INCOMING_ADVISE) { 1916 /* 1917 * A rare case, we entered listen without having to do any discards, 1918 * so do the setup that's normally done at the time of the 1st discard. 1919 */ 1920 if (migrate_postcopy_ram()) { 1921 postcopy_ram_prepare_discard(mis); 1922 } 1923 } 1924 1925 /* 1926 * Sensitise RAM - can now generate requests for blocks that don't exist 1927 * However, at this point the CPU shouldn't be running, and the IO 1928 * shouldn't be doing anything yet so don't actually expect requests 1929 */ 1930 if (migrate_postcopy_ram()) { 1931 if (postcopy_ram_incoming_setup(mis)) { 1932 postcopy_ram_incoming_cleanup(mis); 1933 return -1; 1934 } 1935 } 1936 1937 if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_LISTEN, &local_err)) { 1938 error_report_err(local_err); 1939 return -1; 1940 } 1941 1942 mis->have_listen_thread = true; 1943 /* Start up the listening thread and wait for it to signal ready */ 1944 qemu_sem_init(&mis->listen_thread_sem, 0); 1945 qemu_thread_create(&mis->listen_thread, "postcopy/listen", 1946 postcopy_ram_listen_thread, NULL, 1947 QEMU_THREAD_DETACHED); 1948 qemu_sem_wait(&mis->listen_thread_sem); 1949 qemu_sem_destroy(&mis->listen_thread_sem); 1950 1951 return 0; 1952 } 1953 1954 static void loadvm_postcopy_handle_run_bh(void *opaque) 1955 { 1956 Error *local_err = NULL; 1957 MigrationIncomingState *mis = opaque; 1958 1959 /* TODO we should move all of this lot into postcopy_ram.c or a shared code 1960 * in migration.c 1961 */ 1962 cpu_synchronize_all_post_init(); 1963 1964 qemu_announce_self(&mis->announce_timer, migrate_announce_params()); 1965 1966 /* Make sure all file formats flush their mutable metadata. 1967 * If we get an error here, just don't restart the VM yet. */ 1968 bdrv_invalidate_cache_all(&local_err); 1969 if (local_err) { 1970 error_report_err(local_err); 1971 local_err = NULL; 1972 autostart = false; 1973 } 1974 1975 trace_loadvm_postcopy_handle_run_cpu_sync(); 1976 1977 trace_loadvm_postcopy_handle_run_vmstart(); 1978 1979 dirty_bitmap_mig_before_vm_start(); 1980 1981 if (autostart) { 1982 /* Hold onto your hats, starting the CPU */ 1983 vm_start(); 1984 } else { 1985 /* leave it paused and let management decide when to start the CPU */ 1986 runstate_set(RUN_STATE_PAUSED); 1987 } 1988 1989 qemu_bh_delete(mis->bh); 1990 } 1991 1992 /* After all discards we can start running and asking for pages */ 1993 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis) 1994 { 1995 PostcopyState ps = postcopy_state_get(); 1996 1997 trace_loadvm_postcopy_handle_run(); 1998 if (ps != POSTCOPY_INCOMING_LISTENING) { 1999 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps); 2000 return -1; 2001 } 2002 2003 postcopy_state_set(POSTCOPY_INCOMING_RUNNING); 2004 mis->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, mis); 2005 qemu_bh_schedule(mis->bh); 2006 2007 /* We need to finish reading the stream from the package 2008 * and also stop reading anything more from the stream that loaded the 2009 * package (since it's now being read by the listener thread). 2010 * LOADVM_QUIT will quit all the layers of nested loadvm loops. 2011 */ 2012 return LOADVM_QUIT; 2013 } 2014 2015 static int loadvm_postcopy_handle_resume(MigrationIncomingState *mis) 2016 { 2017 if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) { 2018 error_report("%s: illegal resume received", __func__); 2019 /* Don't fail the load, only for this. */ 2020 return 0; 2021 } 2022 2023 /* 2024 * This means source VM is ready to resume the postcopy migration. 2025 * It's time to switch state and release the fault thread to 2026 * continue service page faults. 2027 */ 2028 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_RECOVER, 2029 MIGRATION_STATUS_POSTCOPY_ACTIVE); 2030 qemu_sem_post(&mis->postcopy_pause_sem_fault); 2031 2032 trace_loadvm_postcopy_handle_resume(); 2033 2034 /* Tell source that "we are ready" */ 2035 migrate_send_rp_resume_ack(mis, MIGRATION_RESUME_ACK_VALUE); 2036 2037 return 0; 2038 } 2039 2040 /** 2041 * Immediately following this command is a blob of data containing an embedded 2042 * chunk of migration stream; read it and load it. 2043 * 2044 * @mis: Incoming state 2045 * @length: Length of packaged data to read 2046 * 2047 * Returns: Negative values on error 2048 * 2049 */ 2050 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis) 2051 { 2052 int ret; 2053 size_t length; 2054 QIOChannelBuffer *bioc; 2055 2056 length = qemu_get_be32(mis->from_src_file); 2057 trace_loadvm_handle_cmd_packaged(length); 2058 2059 if (length > MAX_VM_CMD_PACKAGED_SIZE) { 2060 error_report("Unreasonably large packaged state: %zu", length); 2061 return -1; 2062 } 2063 2064 bioc = qio_channel_buffer_new(length); 2065 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer"); 2066 ret = qemu_get_buffer(mis->from_src_file, 2067 bioc->data, 2068 length); 2069 if (ret != length) { 2070 object_unref(OBJECT(bioc)); 2071 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu", 2072 ret, length); 2073 return (ret < 0) ? ret : -EAGAIN; 2074 } 2075 bioc->usage += length; 2076 trace_loadvm_handle_cmd_packaged_received(ret); 2077 2078 QEMUFile *packf = qemu_fopen_channel_input(QIO_CHANNEL(bioc)); 2079 2080 ret = qemu_loadvm_state_main(packf, mis); 2081 trace_loadvm_handle_cmd_packaged_main(ret); 2082 qemu_fclose(packf); 2083 object_unref(OBJECT(bioc)); 2084 2085 return ret; 2086 } 2087 2088 /* 2089 * Handle request that source requests for recved_bitmap on 2090 * destination. Payload format: 2091 * 2092 * len (1 byte) + ramblock_name (<255 bytes) 2093 */ 2094 static int loadvm_handle_recv_bitmap(MigrationIncomingState *mis, 2095 uint16_t len) 2096 { 2097 QEMUFile *file = mis->from_src_file; 2098 RAMBlock *rb; 2099 char block_name[256]; 2100 size_t cnt; 2101 2102 cnt = qemu_get_counted_string(file, block_name); 2103 if (!cnt) { 2104 error_report("%s: failed to read block name", __func__); 2105 return -EINVAL; 2106 } 2107 2108 /* Validate before using the data */ 2109 if (qemu_file_get_error(file)) { 2110 return qemu_file_get_error(file); 2111 } 2112 2113 if (len != cnt + 1) { 2114 error_report("%s: invalid payload length (%d)", __func__, len); 2115 return -EINVAL; 2116 } 2117 2118 rb = qemu_ram_block_by_name(block_name); 2119 if (!rb) { 2120 error_report("%s: block '%s' not found", __func__, block_name); 2121 return -EINVAL; 2122 } 2123 2124 migrate_send_rp_recv_bitmap(mis, block_name); 2125 2126 trace_loadvm_handle_recv_bitmap(block_name); 2127 2128 return 0; 2129 } 2130 2131 static int loadvm_process_enable_colo(MigrationIncomingState *mis) 2132 { 2133 migration_incoming_enable_colo(); 2134 return colo_init_ram_cache(); 2135 } 2136 2137 /* 2138 * Process an incoming 'QEMU_VM_COMMAND' 2139 * 0 just a normal return 2140 * LOADVM_QUIT All good, but exit the loop 2141 * <0 Error 2142 */ 2143 static int loadvm_process_command(QEMUFile *f) 2144 { 2145 MigrationIncomingState *mis = migration_incoming_get_current(); 2146 uint16_t cmd; 2147 uint16_t len; 2148 uint32_t tmp32; 2149 2150 cmd = qemu_get_be16(f); 2151 len = qemu_get_be16(f); 2152 2153 /* Check validity before continue processing of cmds */ 2154 if (qemu_file_get_error(f)) { 2155 return qemu_file_get_error(f); 2156 } 2157 2158 trace_loadvm_process_command(cmd, len); 2159 if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) { 2160 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len); 2161 return -EINVAL; 2162 } 2163 2164 if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) { 2165 error_report("%s received with bad length - expecting %zu, got %d", 2166 mig_cmd_args[cmd].name, 2167 (size_t)mig_cmd_args[cmd].len, len); 2168 return -ERANGE; 2169 } 2170 2171 switch (cmd) { 2172 case MIG_CMD_OPEN_RETURN_PATH: 2173 if (mis->to_src_file) { 2174 error_report("CMD_OPEN_RETURN_PATH called when RP already open"); 2175 /* Not really a problem, so don't give up */ 2176 return 0; 2177 } 2178 mis->to_src_file = qemu_file_get_return_path(f); 2179 if (!mis->to_src_file) { 2180 error_report("CMD_OPEN_RETURN_PATH failed"); 2181 return -1; 2182 } 2183 break; 2184 2185 case MIG_CMD_PING: 2186 tmp32 = qemu_get_be32(f); 2187 trace_loadvm_process_command_ping(tmp32); 2188 if (!mis->to_src_file) { 2189 error_report("CMD_PING (0x%x) received with no return path", 2190 tmp32); 2191 return -1; 2192 } 2193 migrate_send_rp_pong(mis, tmp32); 2194 break; 2195 2196 case MIG_CMD_PACKAGED: 2197 return loadvm_handle_cmd_packaged(mis); 2198 2199 case MIG_CMD_POSTCOPY_ADVISE: 2200 return loadvm_postcopy_handle_advise(mis, len); 2201 2202 case MIG_CMD_POSTCOPY_LISTEN: 2203 return loadvm_postcopy_handle_listen(mis); 2204 2205 case MIG_CMD_POSTCOPY_RUN: 2206 return loadvm_postcopy_handle_run(mis); 2207 2208 case MIG_CMD_POSTCOPY_RAM_DISCARD: 2209 return loadvm_postcopy_ram_handle_discard(mis, len); 2210 2211 case MIG_CMD_POSTCOPY_RESUME: 2212 return loadvm_postcopy_handle_resume(mis); 2213 2214 case MIG_CMD_RECV_BITMAP: 2215 return loadvm_handle_recv_bitmap(mis, len); 2216 2217 case MIG_CMD_ENABLE_COLO: 2218 return loadvm_process_enable_colo(mis); 2219 } 2220 2221 return 0; 2222 } 2223 2224 /* 2225 * Read a footer off the wire and check that it matches the expected section 2226 * 2227 * Returns: true if the footer was good 2228 * false if there is a problem (and calls error_report to say why) 2229 */ 2230 static bool check_section_footer(QEMUFile *f, SaveStateEntry *se) 2231 { 2232 int ret; 2233 uint8_t read_mark; 2234 uint32_t read_section_id; 2235 2236 if (!migrate_get_current()->send_section_footer) { 2237 /* No footer to check */ 2238 return true; 2239 } 2240 2241 read_mark = qemu_get_byte(f); 2242 2243 ret = qemu_file_get_error(f); 2244 if (ret) { 2245 error_report("%s: Read section footer failed: %d", 2246 __func__, ret); 2247 return false; 2248 } 2249 2250 if (read_mark != QEMU_VM_SECTION_FOOTER) { 2251 error_report("Missing section footer for %s", se->idstr); 2252 return false; 2253 } 2254 2255 read_section_id = qemu_get_be32(f); 2256 if (read_section_id != se->load_section_id) { 2257 error_report("Mismatched section id in footer for %s -" 2258 " read 0x%x expected 0x%x", 2259 se->idstr, read_section_id, se->load_section_id); 2260 return false; 2261 } 2262 2263 /* All good */ 2264 return true; 2265 } 2266 2267 static int 2268 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis) 2269 { 2270 uint32_t instance_id, version_id, section_id; 2271 SaveStateEntry *se; 2272 char idstr[256]; 2273 int ret; 2274 2275 /* Read section start */ 2276 section_id = qemu_get_be32(f); 2277 if (!qemu_get_counted_string(f, idstr)) { 2278 error_report("Unable to read ID string for section %u", 2279 section_id); 2280 return -EINVAL; 2281 } 2282 instance_id = qemu_get_be32(f); 2283 version_id = qemu_get_be32(f); 2284 2285 ret = qemu_file_get_error(f); 2286 if (ret) { 2287 error_report("%s: Failed to read instance/version ID: %d", 2288 __func__, ret); 2289 return ret; 2290 } 2291 2292 trace_qemu_loadvm_state_section_startfull(section_id, idstr, 2293 instance_id, version_id); 2294 /* Find savevm section */ 2295 se = find_se(idstr, instance_id); 2296 if (se == NULL) { 2297 error_report("Unknown savevm section or instance '%s' %"PRIu32". " 2298 "Make sure that your current VM setup matches your " 2299 "saved VM setup, including any hotplugged devices", 2300 idstr, instance_id); 2301 return -EINVAL; 2302 } 2303 2304 /* Validate version */ 2305 if (version_id > se->version_id) { 2306 error_report("savevm: unsupported version %d for '%s' v%d", 2307 version_id, idstr, se->version_id); 2308 return -EINVAL; 2309 } 2310 se->load_version_id = version_id; 2311 se->load_section_id = section_id; 2312 2313 /* Validate if it is a device's state */ 2314 if (xen_enabled() && se->is_ram) { 2315 error_report("loadvm: %s RAM loading not allowed on Xen", idstr); 2316 return -EINVAL; 2317 } 2318 2319 ret = vmstate_load(f, se); 2320 if (ret < 0) { 2321 error_report("error while loading state for instance 0x%"PRIx32" of" 2322 " device '%s'", instance_id, idstr); 2323 return ret; 2324 } 2325 if (!check_section_footer(f, se)) { 2326 return -EINVAL; 2327 } 2328 2329 return 0; 2330 } 2331 2332 static int 2333 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis) 2334 { 2335 uint32_t section_id; 2336 SaveStateEntry *se; 2337 int ret; 2338 2339 section_id = qemu_get_be32(f); 2340 2341 ret = qemu_file_get_error(f); 2342 if (ret) { 2343 error_report("%s: Failed to read section ID: %d", 2344 __func__, ret); 2345 return ret; 2346 } 2347 2348 trace_qemu_loadvm_state_section_partend(section_id); 2349 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 2350 if (se->load_section_id == section_id) { 2351 break; 2352 } 2353 } 2354 if (se == NULL) { 2355 error_report("Unknown savevm section %d", section_id); 2356 return -EINVAL; 2357 } 2358 2359 ret = vmstate_load(f, se); 2360 if (ret < 0) { 2361 error_report("error while loading state section id %d(%s)", 2362 section_id, se->idstr); 2363 return ret; 2364 } 2365 if (!check_section_footer(f, se)) { 2366 return -EINVAL; 2367 } 2368 2369 return 0; 2370 } 2371 2372 static int qemu_loadvm_state_header(QEMUFile *f) 2373 { 2374 unsigned int v; 2375 int ret; 2376 2377 v = qemu_get_be32(f); 2378 if (v != QEMU_VM_FILE_MAGIC) { 2379 error_report("Not a migration stream"); 2380 return -EINVAL; 2381 } 2382 2383 v = qemu_get_be32(f); 2384 if (v == QEMU_VM_FILE_VERSION_COMPAT) { 2385 error_report("SaveVM v2 format is obsolete and don't work anymore"); 2386 return -ENOTSUP; 2387 } 2388 if (v != QEMU_VM_FILE_VERSION) { 2389 error_report("Unsupported migration stream version"); 2390 return -ENOTSUP; 2391 } 2392 2393 if (migrate_get_current()->send_configuration) { 2394 if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) { 2395 error_report("Configuration section missing"); 2396 qemu_loadvm_state_cleanup(); 2397 return -EINVAL; 2398 } 2399 ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0); 2400 2401 if (ret) { 2402 qemu_loadvm_state_cleanup(); 2403 return ret; 2404 } 2405 } 2406 return 0; 2407 } 2408 2409 static int qemu_loadvm_state_setup(QEMUFile *f) 2410 { 2411 SaveStateEntry *se; 2412 int ret; 2413 2414 trace_loadvm_state_setup(); 2415 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 2416 if (!se->ops || !se->ops->load_setup) { 2417 continue; 2418 } 2419 if (se->ops->is_active) { 2420 if (!se->ops->is_active(se->opaque)) { 2421 continue; 2422 } 2423 } 2424 2425 ret = se->ops->load_setup(f, se->opaque); 2426 if (ret < 0) { 2427 qemu_file_set_error(f, ret); 2428 error_report("Load state of device %s failed", se->idstr); 2429 return ret; 2430 } 2431 } 2432 return 0; 2433 } 2434 2435 void qemu_loadvm_state_cleanup(void) 2436 { 2437 SaveStateEntry *se; 2438 2439 trace_loadvm_state_cleanup(); 2440 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 2441 if (se->ops && se->ops->load_cleanup) { 2442 se->ops->load_cleanup(se->opaque); 2443 } 2444 } 2445 } 2446 2447 /* Return true if we should continue the migration, or false. */ 2448 static bool postcopy_pause_incoming(MigrationIncomingState *mis) 2449 { 2450 trace_postcopy_pause_incoming(); 2451 2452 /* Clear the triggered bit to allow one recovery */ 2453 mis->postcopy_recover_triggered = false; 2454 2455 assert(mis->from_src_file); 2456 qemu_file_shutdown(mis->from_src_file); 2457 qemu_fclose(mis->from_src_file); 2458 mis->from_src_file = NULL; 2459 2460 assert(mis->to_src_file); 2461 qemu_file_shutdown(mis->to_src_file); 2462 qemu_mutex_lock(&mis->rp_mutex); 2463 qemu_fclose(mis->to_src_file); 2464 mis->to_src_file = NULL; 2465 qemu_mutex_unlock(&mis->rp_mutex); 2466 2467 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, 2468 MIGRATION_STATUS_POSTCOPY_PAUSED); 2469 2470 /* Notify the fault thread for the invalidated file handle */ 2471 postcopy_fault_thread_notify(mis); 2472 2473 error_report("Detected IO failure for postcopy. " 2474 "Migration paused."); 2475 2476 while (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) { 2477 qemu_sem_wait(&mis->postcopy_pause_sem_dst); 2478 } 2479 2480 trace_postcopy_pause_incoming_continued(); 2481 2482 return true; 2483 } 2484 2485 int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis) 2486 { 2487 uint8_t section_type; 2488 int ret = 0; 2489 2490 retry: 2491 while (true) { 2492 section_type = qemu_get_byte(f); 2493 2494 if (qemu_file_get_error(f)) { 2495 ret = qemu_file_get_error(f); 2496 break; 2497 } 2498 2499 trace_qemu_loadvm_state_section(section_type); 2500 switch (section_type) { 2501 case QEMU_VM_SECTION_START: 2502 case QEMU_VM_SECTION_FULL: 2503 ret = qemu_loadvm_section_start_full(f, mis); 2504 if (ret < 0) { 2505 goto out; 2506 } 2507 break; 2508 case QEMU_VM_SECTION_PART: 2509 case QEMU_VM_SECTION_END: 2510 ret = qemu_loadvm_section_part_end(f, mis); 2511 if (ret < 0) { 2512 goto out; 2513 } 2514 break; 2515 case QEMU_VM_COMMAND: 2516 ret = loadvm_process_command(f); 2517 trace_qemu_loadvm_state_section_command(ret); 2518 if ((ret < 0) || (ret == LOADVM_QUIT)) { 2519 goto out; 2520 } 2521 break; 2522 case QEMU_VM_EOF: 2523 /* This is the end of migration */ 2524 goto out; 2525 default: 2526 error_report("Unknown savevm section type %d", section_type); 2527 ret = -EINVAL; 2528 goto out; 2529 } 2530 } 2531 2532 out: 2533 if (ret < 0) { 2534 qemu_file_set_error(f, ret); 2535 2536 /* 2537 * If we are during an active postcopy, then we pause instead 2538 * of bail out to at least keep the VM's dirty data. Note 2539 * that POSTCOPY_INCOMING_LISTENING stage is still not enough, 2540 * during which we're still receiving device states and we 2541 * still haven't yet started the VM on destination. 2542 */ 2543 if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING && 2544 postcopy_pause_incoming(mis)) { 2545 /* Reset f to point to the newly created channel */ 2546 f = mis->from_src_file; 2547 goto retry; 2548 } 2549 } 2550 return ret; 2551 } 2552 2553 int qemu_loadvm_state(QEMUFile *f) 2554 { 2555 MigrationIncomingState *mis = migration_incoming_get_current(); 2556 Error *local_err = NULL; 2557 int ret; 2558 2559 if (qemu_savevm_state_blocked(&local_err)) { 2560 error_report_err(local_err); 2561 return -EINVAL; 2562 } 2563 2564 ret = qemu_loadvm_state_header(f); 2565 if (ret) { 2566 return ret; 2567 } 2568 2569 if (qemu_loadvm_state_setup(f) != 0) { 2570 return -EINVAL; 2571 } 2572 2573 cpu_synchronize_all_pre_loadvm(); 2574 2575 ret = qemu_loadvm_state_main(f, mis); 2576 qemu_event_set(&mis->main_thread_load_event); 2577 2578 trace_qemu_loadvm_state_post_main(ret); 2579 2580 if (mis->have_listen_thread) { 2581 /* Listen thread still going, can't clean up yet */ 2582 return ret; 2583 } 2584 2585 if (ret == 0) { 2586 ret = qemu_file_get_error(f); 2587 } 2588 2589 /* 2590 * Try to read in the VMDESC section as well, so that dumping tools that 2591 * intercept our migration stream have the chance to see it. 2592 */ 2593 2594 /* We've got to be careful; if we don't read the data and just shut the fd 2595 * then the sender can error if we close while it's still sending. 2596 * We also mustn't read data that isn't there; some transports (RDMA) 2597 * will stall waiting for that data when the source has already closed. 2598 */ 2599 if (ret == 0 && should_send_vmdesc()) { 2600 uint8_t *buf; 2601 uint32_t size; 2602 uint8_t section_type = qemu_get_byte(f); 2603 2604 if (section_type != QEMU_VM_VMDESCRIPTION) { 2605 error_report("Expected vmdescription section, but got %d", 2606 section_type); 2607 /* 2608 * It doesn't seem worth failing at this point since 2609 * we apparently have an otherwise valid VM state 2610 */ 2611 } else { 2612 buf = g_malloc(0x1000); 2613 size = qemu_get_be32(f); 2614 2615 while (size > 0) { 2616 uint32_t read_chunk = MIN(size, 0x1000); 2617 qemu_get_buffer(f, buf, read_chunk); 2618 size -= read_chunk; 2619 } 2620 g_free(buf); 2621 } 2622 } 2623 2624 qemu_loadvm_state_cleanup(); 2625 cpu_synchronize_all_post_init(); 2626 2627 return ret; 2628 } 2629 2630 int qemu_load_device_state(QEMUFile *f) 2631 { 2632 MigrationIncomingState *mis = migration_incoming_get_current(); 2633 int ret; 2634 2635 /* Load QEMU_VM_SECTION_FULL section */ 2636 ret = qemu_loadvm_state_main(f, mis); 2637 if (ret < 0) { 2638 error_report("Failed to load device state: %d", ret); 2639 return ret; 2640 } 2641 2642 cpu_synchronize_all_post_init(); 2643 return 0; 2644 } 2645 2646 int save_snapshot(const char *name, Error **errp) 2647 { 2648 BlockDriverState *bs, *bs1; 2649 QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1; 2650 int ret = -1; 2651 QEMUFile *f; 2652 int saved_vm_running; 2653 uint64_t vm_state_size; 2654 qemu_timeval tv; 2655 struct tm tm; 2656 AioContext *aio_context; 2657 2658 if (migration_is_blocked(errp)) { 2659 return ret; 2660 } 2661 2662 if (!replay_can_snapshot()) { 2663 error_setg(errp, "Record/replay does not allow making snapshot " 2664 "right now. Try once more later."); 2665 return ret; 2666 } 2667 2668 if (!bdrv_all_can_snapshot(&bs)) { 2669 error_setg(errp, "Device '%s' is writable but does not support " 2670 "snapshots", bdrv_get_device_name(bs)); 2671 return ret; 2672 } 2673 2674 /* Delete old snapshots of the same name */ 2675 if (name) { 2676 ret = bdrv_all_delete_snapshot(name, &bs1, errp); 2677 if (ret < 0) { 2678 error_prepend(errp, "Error while deleting snapshot on device " 2679 "'%s': ", bdrv_get_device_name(bs1)); 2680 return ret; 2681 } 2682 } 2683 2684 bs = bdrv_all_find_vmstate_bs(); 2685 if (bs == NULL) { 2686 error_setg(errp, "No block device can accept snapshots"); 2687 return ret; 2688 } 2689 aio_context = bdrv_get_aio_context(bs); 2690 2691 saved_vm_running = runstate_is_running(); 2692 2693 ret = global_state_store(); 2694 if (ret) { 2695 error_setg(errp, "Error saving global state"); 2696 return ret; 2697 } 2698 vm_stop(RUN_STATE_SAVE_VM); 2699 2700 bdrv_drain_all_begin(); 2701 2702 aio_context_acquire(aio_context); 2703 2704 memset(sn, 0, sizeof(*sn)); 2705 2706 /* fill auxiliary fields */ 2707 qemu_gettimeofday(&tv); 2708 sn->date_sec = tv.tv_sec; 2709 sn->date_nsec = tv.tv_usec * 1000; 2710 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL); 2711 2712 if (name) { 2713 ret = bdrv_snapshot_find(bs, old_sn, name); 2714 if (ret >= 0) { 2715 pstrcpy(sn->name, sizeof(sn->name), old_sn->name); 2716 pstrcpy(sn->id_str, sizeof(sn->id_str), old_sn->id_str); 2717 } else { 2718 pstrcpy(sn->name, sizeof(sn->name), name); 2719 } 2720 } else { 2721 /* cast below needed for OpenBSD where tv_sec is still 'long' */ 2722 localtime_r((const time_t *)&tv.tv_sec, &tm); 2723 strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm); 2724 } 2725 2726 /* save the VM state */ 2727 f = qemu_fopen_bdrv(bs, 1); 2728 if (!f) { 2729 error_setg(errp, "Could not open VM state file"); 2730 goto the_end; 2731 } 2732 ret = qemu_savevm_state(f, errp); 2733 vm_state_size = qemu_ftell(f); 2734 qemu_fclose(f); 2735 if (ret < 0) { 2736 goto the_end; 2737 } 2738 2739 /* The bdrv_all_create_snapshot() call that follows acquires the AioContext 2740 * for itself. BDRV_POLL_WHILE() does not support nested locking because 2741 * it only releases the lock once. Therefore synchronous I/O will deadlock 2742 * unless we release the AioContext before bdrv_all_create_snapshot(). 2743 */ 2744 aio_context_release(aio_context); 2745 aio_context = NULL; 2746 2747 ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, &bs); 2748 if (ret < 0) { 2749 error_setg(errp, "Error while creating snapshot on '%s'", 2750 bdrv_get_device_name(bs)); 2751 goto the_end; 2752 } 2753 2754 ret = 0; 2755 2756 the_end: 2757 if (aio_context) { 2758 aio_context_release(aio_context); 2759 } 2760 2761 bdrv_drain_all_end(); 2762 2763 if (saved_vm_running) { 2764 vm_start(); 2765 } 2766 return ret; 2767 } 2768 2769 void qmp_xen_save_devices_state(const char *filename, bool has_live, bool live, 2770 Error **errp) 2771 { 2772 QEMUFile *f; 2773 QIOChannelFile *ioc; 2774 int saved_vm_running; 2775 int ret; 2776 2777 if (!has_live) { 2778 /* live default to true so old version of Xen tool stack can have a 2779 * successfull live migration */ 2780 live = true; 2781 } 2782 2783 saved_vm_running = runstate_is_running(); 2784 vm_stop(RUN_STATE_SAVE_VM); 2785 global_state_store_running(); 2786 2787 ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT, 0660, errp); 2788 if (!ioc) { 2789 goto the_end; 2790 } 2791 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state"); 2792 f = qemu_fopen_channel_output(QIO_CHANNEL(ioc)); 2793 object_unref(OBJECT(ioc)); 2794 ret = qemu_save_device_state(f); 2795 if (ret < 0 || qemu_fclose(f) < 0) { 2796 error_setg(errp, QERR_IO_ERROR); 2797 } else { 2798 /* libxl calls the QMP command "stop" before calling 2799 * "xen-save-devices-state" and in case of migration failure, libxl 2800 * would call "cont". 2801 * So call bdrv_inactivate_all (release locks) here to let the other 2802 * side of the migration take controle of the images. 2803 */ 2804 if (live && !saved_vm_running) { 2805 ret = bdrv_inactivate_all(); 2806 if (ret) { 2807 error_setg(errp, "%s: bdrv_inactivate_all() failed (%d)", 2808 __func__, ret); 2809 } 2810 } 2811 } 2812 2813 the_end: 2814 if (saved_vm_running) { 2815 vm_start(); 2816 } 2817 } 2818 2819 void qmp_xen_load_devices_state(const char *filename, Error **errp) 2820 { 2821 QEMUFile *f; 2822 QIOChannelFile *ioc; 2823 int ret; 2824 2825 /* Guest must be paused before loading the device state; the RAM state 2826 * will already have been loaded by xc 2827 */ 2828 if (runstate_is_running()) { 2829 error_setg(errp, "Cannot update device state while vm is running"); 2830 return; 2831 } 2832 vm_stop(RUN_STATE_RESTORE_VM); 2833 2834 ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp); 2835 if (!ioc) { 2836 return; 2837 } 2838 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state"); 2839 f = qemu_fopen_channel_input(QIO_CHANNEL(ioc)); 2840 object_unref(OBJECT(ioc)); 2841 2842 ret = qemu_loadvm_state(f); 2843 qemu_fclose(f); 2844 if (ret < 0) { 2845 error_setg(errp, QERR_IO_ERROR); 2846 } 2847 migration_incoming_state_destroy(); 2848 } 2849 2850 int load_snapshot(const char *name, Error **errp) 2851 { 2852 BlockDriverState *bs, *bs_vm_state; 2853 QEMUSnapshotInfo sn; 2854 QEMUFile *f; 2855 int ret; 2856 AioContext *aio_context; 2857 MigrationIncomingState *mis = migration_incoming_get_current(); 2858 2859 if (!replay_can_snapshot()) { 2860 error_setg(errp, "Record/replay does not allow loading snapshot " 2861 "right now. Try once more later."); 2862 return -EINVAL; 2863 } 2864 2865 if (!bdrv_all_can_snapshot(&bs)) { 2866 error_setg(errp, 2867 "Device '%s' is writable but does not support snapshots", 2868 bdrv_get_device_name(bs)); 2869 return -ENOTSUP; 2870 } 2871 ret = bdrv_all_find_snapshot(name, &bs); 2872 if (ret < 0) { 2873 error_setg(errp, 2874 "Device '%s' does not have the requested snapshot '%s'", 2875 bdrv_get_device_name(bs), name); 2876 return ret; 2877 } 2878 2879 bs_vm_state = bdrv_all_find_vmstate_bs(); 2880 if (!bs_vm_state) { 2881 error_setg(errp, "No block device supports snapshots"); 2882 return -ENOTSUP; 2883 } 2884 aio_context = bdrv_get_aio_context(bs_vm_state); 2885 2886 /* Don't even try to load empty VM states */ 2887 aio_context_acquire(aio_context); 2888 ret = bdrv_snapshot_find(bs_vm_state, &sn, name); 2889 aio_context_release(aio_context); 2890 if (ret < 0) { 2891 return ret; 2892 } else if (sn.vm_state_size == 0) { 2893 error_setg(errp, "This is a disk-only snapshot. Revert to it " 2894 " offline using qemu-img"); 2895 return -EINVAL; 2896 } 2897 2898 /* Flush all IO requests so they don't interfere with the new state. */ 2899 bdrv_drain_all_begin(); 2900 2901 ret = bdrv_all_goto_snapshot(name, &bs, errp); 2902 if (ret < 0) { 2903 error_prepend(errp, "Could not load snapshot '%s' on '%s': ", 2904 name, bdrv_get_device_name(bs)); 2905 goto err_drain; 2906 } 2907 2908 /* restore the VM state */ 2909 f = qemu_fopen_bdrv(bs_vm_state, 0); 2910 if (!f) { 2911 error_setg(errp, "Could not open VM state file"); 2912 ret = -EINVAL; 2913 goto err_drain; 2914 } 2915 2916 qemu_system_reset(SHUTDOWN_CAUSE_NONE); 2917 mis->from_src_file = f; 2918 2919 aio_context_acquire(aio_context); 2920 ret = qemu_loadvm_state(f); 2921 migration_incoming_state_destroy(); 2922 aio_context_release(aio_context); 2923 2924 bdrv_drain_all_end(); 2925 2926 if (ret < 0) { 2927 error_setg(errp, "Error %d while loading VM state", ret); 2928 return ret; 2929 } 2930 2931 return 0; 2932 2933 err_drain: 2934 bdrv_drain_all_end(); 2935 return ret; 2936 } 2937 2938 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev) 2939 { 2940 qemu_ram_set_idstr(mr->ram_block, 2941 memory_region_name(mr), dev); 2942 qemu_ram_set_migratable(mr->ram_block); 2943 } 2944 2945 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev) 2946 { 2947 qemu_ram_unset_idstr(mr->ram_block); 2948 qemu_ram_unset_migratable(mr->ram_block); 2949 } 2950 2951 void vmstate_register_ram_global(MemoryRegion *mr) 2952 { 2953 vmstate_register_ram(mr, NULL); 2954 } 2955 2956 bool vmstate_check_only_migratable(const VMStateDescription *vmsd) 2957 { 2958 /* check needed if --only-migratable is specified */ 2959 if (!only_migratable) { 2960 return true; 2961 } 2962 2963 return !(vmsd && vmsd->unmigratable); 2964 } 2965