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 "cpu.h" 31 #include "hw/boards.h" 32 #include "hw/hw.h" 33 #include "hw/qdev.h" 34 #include "hw/xen/xen.h" 35 #include "net/net.h" 36 #include "sysemu/sysemu.h" 37 #include "qemu/timer.h" 38 #include "migration/migration.h" 39 #include "postcopy-ram.h" 40 #include "qapi/qmp/qerror.h" 41 #include "qemu/error-report.h" 42 #include "qemu/queue.h" 43 #include "sysemu/cpus.h" 44 #include "exec/memory.h" 45 #include "qmp-commands.h" 46 #include "trace.h" 47 #include "qemu/bitops.h" 48 #include "qemu/iov.h" 49 #include "block/snapshot.h" 50 #include "qemu/cutils.h" 51 #include "io/channel-buffer.h" 52 #include "io/channel-file.h" 53 54 #ifndef ETH_P_RARP 55 #define ETH_P_RARP 0x8035 56 #endif 57 #define ARP_HTYPE_ETH 0x0001 58 #define ARP_PTYPE_IP 0x0800 59 #define ARP_OP_REQUEST_REV 0x3 60 61 const unsigned int postcopy_ram_discard_version = 0; 62 63 static bool skip_section_footers; 64 65 static struct mig_cmd_args { 66 ssize_t len; /* -1 = variable */ 67 const char *name; 68 } mig_cmd_args[] = { 69 [MIG_CMD_INVALID] = { .len = -1, .name = "INVALID" }, 70 [MIG_CMD_OPEN_RETURN_PATH] = { .len = 0, .name = "OPEN_RETURN_PATH" }, 71 [MIG_CMD_PING] = { .len = sizeof(uint32_t), .name = "PING" }, 72 [MIG_CMD_POSTCOPY_ADVISE] = { .len = 16, .name = "POSTCOPY_ADVISE" }, 73 [MIG_CMD_POSTCOPY_LISTEN] = { .len = 0, .name = "POSTCOPY_LISTEN" }, 74 [MIG_CMD_POSTCOPY_RUN] = { .len = 0, .name = "POSTCOPY_RUN" }, 75 [MIG_CMD_POSTCOPY_RAM_DISCARD] = { 76 .len = -1, .name = "POSTCOPY_RAM_DISCARD" }, 77 [MIG_CMD_PACKAGED] = { .len = 4, .name = "PACKAGED" }, 78 [MIG_CMD_MAX] = { .len = -1, .name = "MAX" }, 79 }; 80 81 static int announce_self_create(uint8_t *buf, 82 uint8_t *mac_addr) 83 { 84 /* Ethernet header. */ 85 memset(buf, 0xff, 6); /* destination MAC addr */ 86 memcpy(buf + 6, mac_addr, 6); /* source MAC addr */ 87 *(uint16_t *)(buf + 12) = htons(ETH_P_RARP); /* ethertype */ 88 89 /* RARP header. */ 90 *(uint16_t *)(buf + 14) = htons(ARP_HTYPE_ETH); /* hardware addr space */ 91 *(uint16_t *)(buf + 16) = htons(ARP_PTYPE_IP); /* protocol addr space */ 92 *(buf + 18) = 6; /* hardware addr length (ethernet) */ 93 *(buf + 19) = 4; /* protocol addr length (IPv4) */ 94 *(uint16_t *)(buf + 20) = htons(ARP_OP_REQUEST_REV); /* opcode */ 95 memcpy(buf + 22, mac_addr, 6); /* source hw addr */ 96 memset(buf + 28, 0x00, 4); /* source protocol addr */ 97 memcpy(buf + 32, mac_addr, 6); /* target hw addr */ 98 memset(buf + 38, 0x00, 4); /* target protocol addr */ 99 100 /* Padding to get up to 60 bytes (ethernet min packet size, minus FCS). */ 101 memset(buf + 42, 0x00, 18); 102 103 return 60; /* len (FCS will be added by hardware) */ 104 } 105 106 static void qemu_announce_self_iter(NICState *nic, void *opaque) 107 { 108 uint8_t buf[60]; 109 int len; 110 111 trace_qemu_announce_self_iter(qemu_ether_ntoa(&nic->conf->macaddr)); 112 len = announce_self_create(buf, nic->conf->macaddr.a); 113 114 qemu_send_packet_raw(qemu_get_queue(nic), buf, len); 115 } 116 117 118 static void qemu_announce_self_once(void *opaque) 119 { 120 static int count = SELF_ANNOUNCE_ROUNDS; 121 QEMUTimer *timer = *(QEMUTimer **)opaque; 122 123 qemu_foreach_nic(qemu_announce_self_iter, NULL); 124 125 if (--count) { 126 /* delay 50ms, 150ms, 250ms, ... */ 127 timer_mod(timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + 128 self_announce_delay(count)); 129 } else { 130 timer_del(timer); 131 timer_free(timer); 132 } 133 } 134 135 void qemu_announce_self(void) 136 { 137 static QEMUTimer *timer; 138 timer = timer_new_ms(QEMU_CLOCK_REALTIME, qemu_announce_self_once, &timer); 139 qemu_announce_self_once(&timer); 140 } 141 142 /***********************************************************/ 143 /* savevm/loadvm support */ 144 145 static ssize_t block_writev_buffer(void *opaque, struct iovec *iov, int iovcnt, 146 int64_t pos) 147 { 148 int ret; 149 QEMUIOVector qiov; 150 151 qemu_iovec_init_external(&qiov, iov, iovcnt); 152 ret = bdrv_writev_vmstate(opaque, &qiov, pos); 153 if (ret < 0) { 154 return ret; 155 } 156 157 return qiov.size; 158 } 159 160 static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos, 161 size_t size) 162 { 163 return bdrv_load_vmstate(opaque, buf, pos, size); 164 } 165 166 static int bdrv_fclose(void *opaque) 167 { 168 return bdrv_flush(opaque); 169 } 170 171 static const QEMUFileOps bdrv_read_ops = { 172 .get_buffer = block_get_buffer, 173 .close = bdrv_fclose 174 }; 175 176 static const QEMUFileOps bdrv_write_ops = { 177 .writev_buffer = block_writev_buffer, 178 .close = bdrv_fclose 179 }; 180 181 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable) 182 { 183 if (is_writable) { 184 return qemu_fopen_ops(bs, &bdrv_write_ops); 185 } 186 return qemu_fopen_ops(bs, &bdrv_read_ops); 187 } 188 189 190 /* QEMUFile timer support. 191 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c 192 */ 193 194 void timer_put(QEMUFile *f, QEMUTimer *ts) 195 { 196 uint64_t expire_time; 197 198 expire_time = timer_expire_time_ns(ts); 199 qemu_put_be64(f, expire_time); 200 } 201 202 void timer_get(QEMUFile *f, QEMUTimer *ts) 203 { 204 uint64_t expire_time; 205 206 expire_time = qemu_get_be64(f); 207 if (expire_time != -1) { 208 timer_mod_ns(ts, expire_time); 209 } else { 210 timer_del(ts); 211 } 212 } 213 214 215 /* VMState timer support. 216 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c 217 */ 218 219 static int get_timer(QEMUFile *f, void *pv, size_t size, VMStateField *field) 220 { 221 QEMUTimer *v = pv; 222 timer_get(f, v); 223 return 0; 224 } 225 226 static int put_timer(QEMUFile *f, void *pv, size_t size, VMStateField *field, 227 QJSON *vmdesc) 228 { 229 QEMUTimer *v = pv; 230 timer_put(f, v); 231 232 return 0; 233 } 234 235 const VMStateInfo vmstate_info_timer = { 236 .name = "timer", 237 .get = get_timer, 238 .put = put_timer, 239 }; 240 241 242 typedef struct CompatEntry { 243 char idstr[256]; 244 int instance_id; 245 } CompatEntry; 246 247 typedef struct SaveStateEntry { 248 QTAILQ_ENTRY(SaveStateEntry) entry; 249 char idstr[256]; 250 int instance_id; 251 int alias_id; 252 int version_id; 253 int section_id; 254 SaveVMHandlers *ops; 255 const VMStateDescription *vmsd; 256 void *opaque; 257 CompatEntry *compat; 258 int is_ram; 259 } SaveStateEntry; 260 261 typedef struct SaveState { 262 QTAILQ_HEAD(, SaveStateEntry) handlers; 263 int global_section_id; 264 bool skip_configuration; 265 uint32_t len; 266 const char *name; 267 uint32_t target_page_bits; 268 } SaveState; 269 270 static SaveState savevm_state = { 271 .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers), 272 .global_section_id = 0, 273 .skip_configuration = false, 274 }; 275 276 void savevm_skip_configuration(void) 277 { 278 savevm_state.skip_configuration = true; 279 } 280 281 282 static void configuration_pre_save(void *opaque) 283 { 284 SaveState *state = opaque; 285 const char *current_name = MACHINE_GET_CLASS(current_machine)->name; 286 287 state->len = strlen(current_name); 288 state->name = current_name; 289 state->target_page_bits = TARGET_PAGE_BITS; 290 } 291 292 static int configuration_pre_load(void *opaque) 293 { 294 SaveState *state = opaque; 295 296 /* If there is no target-page-bits subsection it means the source 297 * predates the variable-target-page-bits support and is using the 298 * minimum possible value for this CPU. 299 */ 300 state->target_page_bits = TARGET_PAGE_BITS_MIN; 301 return 0; 302 } 303 304 static int configuration_post_load(void *opaque, int version_id) 305 { 306 SaveState *state = opaque; 307 const char *current_name = MACHINE_GET_CLASS(current_machine)->name; 308 309 if (strncmp(state->name, current_name, state->len) != 0) { 310 error_report("Machine type received is '%.*s' and local is '%s'", 311 (int) state->len, state->name, current_name); 312 return -EINVAL; 313 } 314 315 if (state->target_page_bits != TARGET_PAGE_BITS) { 316 error_report("Received TARGET_PAGE_BITS is %d but local is %d", 317 state->target_page_bits, TARGET_PAGE_BITS); 318 return -EINVAL; 319 } 320 321 return 0; 322 } 323 324 /* The target-page-bits subsection is present only if the 325 * target page size is not the same as the default (ie the 326 * minimum page size for a variable-page-size guest CPU). 327 * If it is present then it contains the actual target page 328 * bits for the machine, and migration will fail if the 329 * two ends don't agree about it. 330 */ 331 static bool vmstate_target_page_bits_needed(void *opaque) 332 { 333 return TARGET_PAGE_BITS > TARGET_PAGE_BITS_MIN; 334 } 335 336 static const VMStateDescription vmstate_target_page_bits = { 337 .name = "configuration/target-page-bits", 338 .version_id = 1, 339 .minimum_version_id = 1, 340 .needed = vmstate_target_page_bits_needed, 341 .fields = (VMStateField[]) { 342 VMSTATE_UINT32(target_page_bits, SaveState), 343 VMSTATE_END_OF_LIST() 344 } 345 }; 346 347 static const VMStateDescription vmstate_configuration = { 348 .name = "configuration", 349 .version_id = 1, 350 .pre_load = configuration_pre_load, 351 .post_load = configuration_post_load, 352 .pre_save = configuration_pre_save, 353 .fields = (VMStateField[]) { 354 VMSTATE_UINT32(len, SaveState), 355 VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len), 356 VMSTATE_END_OF_LIST() 357 }, 358 .subsections = (const VMStateDescription*[]) { 359 &vmstate_target_page_bits, 360 NULL 361 } 362 }; 363 364 static void dump_vmstate_vmsd(FILE *out_file, 365 const VMStateDescription *vmsd, int indent, 366 bool is_subsection); 367 368 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field, 369 int indent) 370 { 371 fprintf(out_file, "%*s{\n", indent, ""); 372 indent += 2; 373 fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name); 374 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "", 375 field->version_id); 376 fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "", 377 field->field_exists ? "true" : "false"); 378 fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size); 379 if (field->vmsd != NULL) { 380 fprintf(out_file, ",\n"); 381 dump_vmstate_vmsd(out_file, field->vmsd, indent, false); 382 } 383 fprintf(out_file, "\n%*s}", indent - 2, ""); 384 } 385 386 static void dump_vmstate_vmss(FILE *out_file, 387 const VMStateDescription **subsection, 388 int indent) 389 { 390 if (*subsection != NULL) { 391 dump_vmstate_vmsd(out_file, *subsection, indent, true); 392 } 393 } 394 395 static void dump_vmstate_vmsd(FILE *out_file, 396 const VMStateDescription *vmsd, int indent, 397 bool is_subsection) 398 { 399 if (is_subsection) { 400 fprintf(out_file, "%*s{\n", indent, ""); 401 } else { 402 fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description"); 403 } 404 indent += 2; 405 fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name); 406 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "", 407 vmsd->version_id); 408 fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "", 409 vmsd->minimum_version_id); 410 if (vmsd->fields != NULL) { 411 const VMStateField *field = vmsd->fields; 412 bool first; 413 414 fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, ""); 415 first = true; 416 while (field->name != NULL) { 417 if (field->flags & VMS_MUST_EXIST) { 418 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */ 419 field++; 420 continue; 421 } 422 if (!first) { 423 fprintf(out_file, ",\n"); 424 } 425 dump_vmstate_vmsf(out_file, field, indent + 2); 426 field++; 427 first = false; 428 } 429 fprintf(out_file, "\n%*s]", indent, ""); 430 } 431 if (vmsd->subsections != NULL) { 432 const VMStateDescription **subsection = vmsd->subsections; 433 bool first; 434 435 fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, ""); 436 first = true; 437 while (*subsection != NULL) { 438 if (!first) { 439 fprintf(out_file, ",\n"); 440 } 441 dump_vmstate_vmss(out_file, subsection, indent + 2); 442 subsection++; 443 first = false; 444 } 445 fprintf(out_file, "\n%*s]", indent, ""); 446 } 447 fprintf(out_file, "\n%*s}", indent - 2, ""); 448 } 449 450 static void dump_machine_type(FILE *out_file) 451 { 452 MachineClass *mc; 453 454 mc = MACHINE_GET_CLASS(current_machine); 455 456 fprintf(out_file, " \"vmschkmachine\": {\n"); 457 fprintf(out_file, " \"Name\": \"%s\"\n", mc->name); 458 fprintf(out_file, " },\n"); 459 } 460 461 void dump_vmstate_json_to_file(FILE *out_file) 462 { 463 GSList *list, *elt; 464 bool first; 465 466 fprintf(out_file, "{\n"); 467 dump_machine_type(out_file); 468 469 first = true; 470 list = object_class_get_list(TYPE_DEVICE, true); 471 for (elt = list; elt; elt = elt->next) { 472 DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data, 473 TYPE_DEVICE); 474 const char *name; 475 int indent = 2; 476 477 if (!dc->vmsd) { 478 continue; 479 } 480 481 if (!first) { 482 fprintf(out_file, ",\n"); 483 } 484 name = object_class_get_name(OBJECT_CLASS(dc)); 485 fprintf(out_file, "%*s\"%s\": {\n", indent, "", name); 486 indent += 2; 487 fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name); 488 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "", 489 dc->vmsd->version_id); 490 fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "", 491 dc->vmsd->minimum_version_id); 492 493 dump_vmstate_vmsd(out_file, dc->vmsd, indent, false); 494 495 fprintf(out_file, "\n%*s}", indent - 2, ""); 496 first = false; 497 } 498 fprintf(out_file, "\n}\n"); 499 fclose(out_file); 500 } 501 502 static int calculate_new_instance_id(const char *idstr) 503 { 504 SaveStateEntry *se; 505 int instance_id = 0; 506 507 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 508 if (strcmp(idstr, se->idstr) == 0 509 && instance_id <= se->instance_id) { 510 instance_id = se->instance_id + 1; 511 } 512 } 513 return instance_id; 514 } 515 516 static int calculate_compat_instance_id(const char *idstr) 517 { 518 SaveStateEntry *se; 519 int instance_id = 0; 520 521 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 522 if (!se->compat) { 523 continue; 524 } 525 526 if (strcmp(idstr, se->compat->idstr) == 0 527 && instance_id <= se->compat->instance_id) { 528 instance_id = se->compat->instance_id + 1; 529 } 530 } 531 return instance_id; 532 } 533 534 static inline MigrationPriority save_state_priority(SaveStateEntry *se) 535 { 536 if (se->vmsd) { 537 return se->vmsd->priority; 538 } 539 return MIG_PRI_DEFAULT; 540 } 541 542 static void savevm_state_handler_insert(SaveStateEntry *nse) 543 { 544 MigrationPriority priority = save_state_priority(nse); 545 SaveStateEntry *se; 546 547 assert(priority <= MIG_PRI_MAX); 548 549 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 550 if (save_state_priority(se) < priority) { 551 break; 552 } 553 } 554 555 if (se) { 556 QTAILQ_INSERT_BEFORE(se, nse, entry); 557 } else { 558 QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry); 559 } 560 } 561 562 /* TODO: Individual devices generally have very little idea about the rest 563 of the system, so instance_id should be removed/replaced. 564 Meanwhile pass -1 as instance_id if you do not already have a clearly 565 distinguishing id for all instances of your device class. */ 566 int register_savevm_live(DeviceState *dev, 567 const char *idstr, 568 int instance_id, 569 int version_id, 570 SaveVMHandlers *ops, 571 void *opaque) 572 { 573 SaveStateEntry *se; 574 575 se = g_new0(SaveStateEntry, 1); 576 se->version_id = version_id; 577 se->section_id = savevm_state.global_section_id++; 578 se->ops = ops; 579 se->opaque = opaque; 580 se->vmsd = NULL; 581 /* if this is a live_savem then set is_ram */ 582 if (ops->save_live_setup != NULL) { 583 se->is_ram = 1; 584 } 585 586 if (dev) { 587 char *id = qdev_get_dev_path(dev); 588 if (id) { 589 if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >= 590 sizeof(se->idstr)) { 591 error_report("Path too long for VMState (%s)", id); 592 g_free(id); 593 g_free(se); 594 595 return -1; 596 } 597 g_free(id); 598 599 se->compat = g_new0(CompatEntry, 1); 600 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), idstr); 601 se->compat->instance_id = instance_id == -1 ? 602 calculate_compat_instance_id(idstr) : instance_id; 603 instance_id = -1; 604 } 605 } 606 pstrcat(se->idstr, sizeof(se->idstr), idstr); 607 608 if (instance_id == -1) { 609 se->instance_id = calculate_new_instance_id(se->idstr); 610 } else { 611 se->instance_id = instance_id; 612 } 613 assert(!se->compat || se->instance_id == 0); 614 savevm_state_handler_insert(se); 615 return 0; 616 } 617 618 int register_savevm(DeviceState *dev, 619 const char *idstr, 620 int instance_id, 621 int version_id, 622 SaveStateHandler *save_state, 623 LoadStateHandler *load_state, 624 void *opaque) 625 { 626 SaveVMHandlers *ops = g_new0(SaveVMHandlers, 1); 627 ops->save_state = save_state; 628 ops->load_state = load_state; 629 return register_savevm_live(dev, idstr, instance_id, version_id, 630 ops, opaque); 631 } 632 633 void unregister_savevm(DeviceState *dev, const char *idstr, void *opaque) 634 { 635 SaveStateEntry *se, *new_se; 636 char id[256] = ""; 637 638 if (dev) { 639 char *path = qdev_get_dev_path(dev); 640 if (path) { 641 pstrcpy(id, sizeof(id), path); 642 pstrcat(id, sizeof(id), "/"); 643 g_free(path); 644 } 645 } 646 pstrcat(id, sizeof(id), idstr); 647 648 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) { 649 if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) { 650 QTAILQ_REMOVE(&savevm_state.handlers, se, entry); 651 g_free(se->compat); 652 g_free(se->ops); 653 g_free(se); 654 } 655 } 656 } 657 658 int vmstate_register_with_alias_id(DeviceState *dev, int instance_id, 659 const VMStateDescription *vmsd, 660 void *opaque, int alias_id, 661 int required_for_version, 662 Error **errp) 663 { 664 SaveStateEntry *se; 665 666 /* If this triggers, alias support can be dropped for the vmsd. */ 667 assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id); 668 669 se = g_new0(SaveStateEntry, 1); 670 se->version_id = vmsd->version_id; 671 se->section_id = savevm_state.global_section_id++; 672 se->opaque = opaque; 673 se->vmsd = vmsd; 674 se->alias_id = alias_id; 675 676 if (dev) { 677 char *id = qdev_get_dev_path(dev); 678 if (id) { 679 if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >= 680 sizeof(se->idstr)) { 681 error_setg(errp, "Path too long for VMState (%s)", id); 682 g_free(id); 683 g_free(se); 684 685 return -1; 686 } 687 g_free(id); 688 689 se->compat = g_new0(CompatEntry, 1); 690 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name); 691 se->compat->instance_id = instance_id == -1 ? 692 calculate_compat_instance_id(vmsd->name) : instance_id; 693 instance_id = -1; 694 } 695 } 696 pstrcat(se->idstr, sizeof(se->idstr), vmsd->name); 697 698 if (instance_id == -1) { 699 se->instance_id = calculate_new_instance_id(se->idstr); 700 } else { 701 se->instance_id = instance_id; 702 } 703 assert(!se->compat || se->instance_id == 0); 704 savevm_state_handler_insert(se); 705 return 0; 706 } 707 708 void vmstate_unregister(DeviceState *dev, const VMStateDescription *vmsd, 709 void *opaque) 710 { 711 SaveStateEntry *se, *new_se; 712 713 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) { 714 if (se->vmsd == vmsd && se->opaque == opaque) { 715 QTAILQ_REMOVE(&savevm_state.handlers, se, entry); 716 g_free(se->compat); 717 g_free(se); 718 } 719 } 720 } 721 722 static int vmstate_load(QEMUFile *f, SaveStateEntry *se, int version_id) 723 { 724 trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)"); 725 if (!se->vmsd) { /* Old style */ 726 return se->ops->load_state(f, se->opaque, version_id); 727 } 728 return vmstate_load_state(f, se->vmsd, se->opaque, version_id); 729 } 730 731 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc) 732 { 733 int64_t old_offset, size; 734 735 old_offset = qemu_ftell_fast(f); 736 se->ops->save_state(f, se->opaque); 737 size = qemu_ftell_fast(f) - old_offset; 738 739 if (vmdesc) { 740 json_prop_int(vmdesc, "size", size); 741 json_start_array(vmdesc, "fields"); 742 json_start_object(vmdesc, NULL); 743 json_prop_str(vmdesc, "name", "data"); 744 json_prop_int(vmdesc, "size", size); 745 json_prop_str(vmdesc, "type", "buffer"); 746 json_end_object(vmdesc); 747 json_end_array(vmdesc); 748 } 749 } 750 751 static void vmstate_save(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc) 752 { 753 trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)"); 754 if (!se->vmsd) { 755 vmstate_save_old_style(f, se, vmdesc); 756 return; 757 } 758 vmstate_save_state(f, se->vmsd, se->opaque, vmdesc); 759 } 760 761 void savevm_skip_section_footers(void) 762 { 763 skip_section_footers = true; 764 } 765 766 /* 767 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL) 768 */ 769 static void save_section_header(QEMUFile *f, SaveStateEntry *se, 770 uint8_t section_type) 771 { 772 qemu_put_byte(f, section_type); 773 qemu_put_be32(f, se->section_id); 774 775 if (section_type == QEMU_VM_SECTION_FULL || 776 section_type == QEMU_VM_SECTION_START) { 777 /* ID string */ 778 size_t len = strlen(se->idstr); 779 qemu_put_byte(f, len); 780 qemu_put_buffer(f, (uint8_t *)se->idstr, len); 781 782 qemu_put_be32(f, se->instance_id); 783 qemu_put_be32(f, se->version_id); 784 } 785 } 786 787 /* 788 * Write a footer onto device sections that catches cases misformatted device 789 * sections. 790 */ 791 static void save_section_footer(QEMUFile *f, SaveStateEntry *se) 792 { 793 if (!skip_section_footers) { 794 qemu_put_byte(f, QEMU_VM_SECTION_FOOTER); 795 qemu_put_be32(f, se->section_id); 796 } 797 } 798 799 /** 800 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the 801 * command and associated data. 802 * 803 * @f: File to send command on 804 * @command: Command type to send 805 * @len: Length of associated data 806 * @data: Data associated with command. 807 */ 808 void qemu_savevm_command_send(QEMUFile *f, 809 enum qemu_vm_cmd command, 810 uint16_t len, 811 uint8_t *data) 812 { 813 trace_savevm_command_send(command, len); 814 qemu_put_byte(f, QEMU_VM_COMMAND); 815 qemu_put_be16(f, (uint16_t)command); 816 qemu_put_be16(f, len); 817 qemu_put_buffer(f, data, len); 818 qemu_fflush(f); 819 } 820 821 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value) 822 { 823 uint32_t buf; 824 825 trace_savevm_send_ping(value); 826 buf = cpu_to_be32(value); 827 qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf); 828 } 829 830 void qemu_savevm_send_open_return_path(QEMUFile *f) 831 { 832 trace_savevm_send_open_return_path(); 833 qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL); 834 } 835 836 /* We have a buffer of data to send; we don't want that all to be loaded 837 * by the command itself, so the command contains just the length of the 838 * extra buffer that we then send straight after it. 839 * TODO: Must be a better way to organise that 840 * 841 * Returns: 842 * 0 on success 843 * -ve on error 844 */ 845 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len) 846 { 847 uint32_t tmp; 848 849 if (len > MAX_VM_CMD_PACKAGED_SIZE) { 850 error_report("%s: Unreasonably large packaged state: %zu", 851 __func__, len); 852 return -1; 853 } 854 855 tmp = cpu_to_be32(len); 856 857 trace_qemu_savevm_send_packaged(); 858 qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp); 859 860 qemu_put_buffer(f, buf, len); 861 862 return 0; 863 } 864 865 /* Send prior to any postcopy transfer */ 866 void qemu_savevm_send_postcopy_advise(QEMUFile *f) 867 { 868 uint64_t tmp[2]; 869 tmp[0] = cpu_to_be64(ram_pagesize_summary()); 870 tmp[1] = cpu_to_be64(qemu_target_page_size()); 871 872 trace_qemu_savevm_send_postcopy_advise(); 873 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 16, (uint8_t *)tmp); 874 } 875 876 /* Sent prior to starting the destination running in postcopy, discard pages 877 * that have already been sent but redirtied on the source. 878 * CMD_POSTCOPY_RAM_DISCARD consist of: 879 * byte version (0) 880 * byte Length of name field (not including 0) 881 * n x byte RAM block name 882 * byte 0 terminator (just for safety) 883 * n x Byte ranges within the named RAMBlock 884 * be64 Start of the range 885 * be64 Length 886 * 887 * name: RAMBlock name that these entries are part of 888 * len: Number of page entries 889 * start_list: 'len' addresses 890 * length_list: 'len' addresses 891 * 892 */ 893 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name, 894 uint16_t len, 895 uint64_t *start_list, 896 uint64_t *length_list) 897 { 898 uint8_t *buf; 899 uint16_t tmplen; 900 uint16_t t; 901 size_t name_len = strlen(name); 902 903 trace_qemu_savevm_send_postcopy_ram_discard(name, len); 904 assert(name_len < 256); 905 buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len); 906 buf[0] = postcopy_ram_discard_version; 907 buf[1] = name_len; 908 memcpy(buf + 2, name, name_len); 909 tmplen = 2 + name_len; 910 buf[tmplen++] = '\0'; 911 912 for (t = 0; t < len; t++) { 913 stq_be_p(buf + tmplen, start_list[t]); 914 tmplen += 8; 915 stq_be_p(buf + tmplen, length_list[t]); 916 tmplen += 8; 917 } 918 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf); 919 g_free(buf); 920 } 921 922 /* Get the destination into a state where it can receive postcopy data. */ 923 void qemu_savevm_send_postcopy_listen(QEMUFile *f) 924 { 925 trace_savevm_send_postcopy_listen(); 926 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL); 927 } 928 929 /* Kick the destination into running */ 930 void qemu_savevm_send_postcopy_run(QEMUFile *f) 931 { 932 trace_savevm_send_postcopy_run(); 933 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL); 934 } 935 936 bool qemu_savevm_state_blocked(Error **errp) 937 { 938 SaveStateEntry *se; 939 940 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 941 if (se->vmsd && se->vmsd->unmigratable) { 942 error_setg(errp, "State blocked by non-migratable device '%s'", 943 se->idstr); 944 return true; 945 } 946 } 947 return false; 948 } 949 950 static bool enforce_config_section(void) 951 { 952 MachineState *machine = MACHINE(qdev_get_machine()); 953 return machine->enforce_config_section; 954 } 955 956 void qemu_savevm_state_header(QEMUFile *f) 957 { 958 trace_savevm_state_header(); 959 qemu_put_be32(f, QEMU_VM_FILE_MAGIC); 960 qemu_put_be32(f, QEMU_VM_FILE_VERSION); 961 962 if (!savevm_state.skip_configuration || enforce_config_section()) { 963 qemu_put_byte(f, QEMU_VM_CONFIGURATION); 964 vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0); 965 } 966 967 } 968 969 void qemu_savevm_state_begin(QEMUFile *f, 970 const MigrationParams *params) 971 { 972 SaveStateEntry *se; 973 int ret; 974 975 trace_savevm_state_begin(); 976 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 977 if (!se->ops || !se->ops->set_params) { 978 continue; 979 } 980 se->ops->set_params(params, se->opaque); 981 } 982 983 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 984 if (!se->ops || !se->ops->save_live_setup) { 985 continue; 986 } 987 if (se->ops && se->ops->is_active) { 988 if (!se->ops->is_active(se->opaque)) { 989 continue; 990 } 991 } 992 save_section_header(f, se, QEMU_VM_SECTION_START); 993 994 ret = se->ops->save_live_setup(f, se->opaque); 995 save_section_footer(f, se); 996 if (ret < 0) { 997 qemu_file_set_error(f, ret); 998 break; 999 } 1000 } 1001 } 1002 1003 /* 1004 * this function has three return values: 1005 * negative: there was one error, and we have -errno. 1006 * 0 : We haven't finished, caller have to go again 1007 * 1 : We have finished, we can go to complete phase 1008 */ 1009 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy) 1010 { 1011 SaveStateEntry *se; 1012 int ret = 1; 1013 1014 trace_savevm_state_iterate(); 1015 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1016 if (!se->ops || !se->ops->save_live_iterate) { 1017 continue; 1018 } 1019 if (se->ops && se->ops->is_active) { 1020 if (!se->ops->is_active(se->opaque)) { 1021 continue; 1022 } 1023 } 1024 /* 1025 * In the postcopy phase, any device that doesn't know how to 1026 * do postcopy should have saved it's state in the _complete 1027 * call that's already run, it might get confused if we call 1028 * iterate afterwards. 1029 */ 1030 if (postcopy && !se->ops->save_live_complete_postcopy) { 1031 continue; 1032 } 1033 if (qemu_file_rate_limit(f)) { 1034 return 0; 1035 } 1036 trace_savevm_section_start(se->idstr, se->section_id); 1037 1038 save_section_header(f, se, QEMU_VM_SECTION_PART); 1039 1040 ret = se->ops->save_live_iterate(f, se->opaque); 1041 trace_savevm_section_end(se->idstr, se->section_id, ret); 1042 save_section_footer(f, se); 1043 1044 if (ret < 0) { 1045 qemu_file_set_error(f, ret); 1046 } 1047 if (ret <= 0) { 1048 /* Do not proceed to the next vmstate before this one reported 1049 completion of the current stage. This serializes the migration 1050 and reduces the probability that a faster changing state is 1051 synchronized over and over again. */ 1052 break; 1053 } 1054 } 1055 return ret; 1056 } 1057 1058 static bool should_send_vmdesc(void) 1059 { 1060 MachineState *machine = MACHINE(qdev_get_machine()); 1061 bool in_postcopy = migration_in_postcopy(); 1062 return !machine->suppress_vmdesc && !in_postcopy; 1063 } 1064 1065 /* 1066 * Calls the save_live_complete_postcopy methods 1067 * causing the last few pages to be sent immediately and doing any associated 1068 * cleanup. 1069 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete 1070 * all the other devices, but that happens at the point we switch to postcopy. 1071 */ 1072 void qemu_savevm_state_complete_postcopy(QEMUFile *f) 1073 { 1074 SaveStateEntry *se; 1075 int ret; 1076 1077 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1078 if (!se->ops || !se->ops->save_live_complete_postcopy) { 1079 continue; 1080 } 1081 if (se->ops && se->ops->is_active) { 1082 if (!se->ops->is_active(se->opaque)) { 1083 continue; 1084 } 1085 } 1086 trace_savevm_section_start(se->idstr, se->section_id); 1087 /* Section type */ 1088 qemu_put_byte(f, QEMU_VM_SECTION_END); 1089 qemu_put_be32(f, se->section_id); 1090 1091 ret = se->ops->save_live_complete_postcopy(f, se->opaque); 1092 trace_savevm_section_end(se->idstr, se->section_id, ret); 1093 save_section_footer(f, se); 1094 if (ret < 0) { 1095 qemu_file_set_error(f, ret); 1096 return; 1097 } 1098 } 1099 1100 qemu_put_byte(f, QEMU_VM_EOF); 1101 qemu_fflush(f); 1102 } 1103 1104 void qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only) 1105 { 1106 QJSON *vmdesc; 1107 int vmdesc_len; 1108 SaveStateEntry *se; 1109 int ret; 1110 bool in_postcopy = migration_in_postcopy(); 1111 1112 trace_savevm_state_complete_precopy(); 1113 1114 cpu_synchronize_all_states(); 1115 1116 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1117 if (!se->ops || 1118 (in_postcopy && se->ops->save_live_complete_postcopy) || 1119 (in_postcopy && !iterable_only) || 1120 !se->ops->save_live_complete_precopy) { 1121 continue; 1122 } 1123 1124 if (se->ops && se->ops->is_active) { 1125 if (!se->ops->is_active(se->opaque)) { 1126 continue; 1127 } 1128 } 1129 trace_savevm_section_start(se->idstr, se->section_id); 1130 1131 save_section_header(f, se, QEMU_VM_SECTION_END); 1132 1133 ret = se->ops->save_live_complete_precopy(f, se->opaque); 1134 trace_savevm_section_end(se->idstr, se->section_id, ret); 1135 save_section_footer(f, se); 1136 if (ret < 0) { 1137 qemu_file_set_error(f, ret); 1138 return; 1139 } 1140 } 1141 1142 if (iterable_only) { 1143 return; 1144 } 1145 1146 vmdesc = qjson_new(); 1147 json_prop_int(vmdesc, "page_size", TARGET_PAGE_SIZE); 1148 json_start_array(vmdesc, "devices"); 1149 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1150 1151 if ((!se->ops || !se->ops->save_state) && !se->vmsd) { 1152 continue; 1153 } 1154 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) { 1155 trace_savevm_section_skip(se->idstr, se->section_id); 1156 continue; 1157 } 1158 1159 trace_savevm_section_start(se->idstr, se->section_id); 1160 1161 json_start_object(vmdesc, NULL); 1162 json_prop_str(vmdesc, "name", se->idstr); 1163 json_prop_int(vmdesc, "instance_id", se->instance_id); 1164 1165 save_section_header(f, se, QEMU_VM_SECTION_FULL); 1166 vmstate_save(f, se, vmdesc); 1167 trace_savevm_section_end(se->idstr, se->section_id, 0); 1168 save_section_footer(f, se); 1169 1170 json_end_object(vmdesc); 1171 } 1172 1173 if (!in_postcopy) { 1174 /* Postcopy stream will still be going */ 1175 qemu_put_byte(f, QEMU_VM_EOF); 1176 } 1177 1178 json_end_array(vmdesc); 1179 qjson_finish(vmdesc); 1180 vmdesc_len = strlen(qjson_get_str(vmdesc)); 1181 1182 if (should_send_vmdesc()) { 1183 qemu_put_byte(f, QEMU_VM_VMDESCRIPTION); 1184 qemu_put_be32(f, vmdesc_len); 1185 qemu_put_buffer(f, (uint8_t *)qjson_get_str(vmdesc), vmdesc_len); 1186 } 1187 qjson_destroy(vmdesc); 1188 1189 qemu_fflush(f); 1190 } 1191 1192 /* Give an estimate of the amount left to be transferred, 1193 * the result is split into the amount for units that can and 1194 * for units that can't do postcopy. 1195 */ 1196 void qemu_savevm_state_pending(QEMUFile *f, uint64_t threshold_size, 1197 uint64_t *res_non_postcopiable, 1198 uint64_t *res_postcopiable) 1199 { 1200 SaveStateEntry *se; 1201 1202 *res_non_postcopiable = 0; 1203 *res_postcopiable = 0; 1204 1205 1206 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1207 if (!se->ops || !se->ops->save_live_pending) { 1208 continue; 1209 } 1210 if (se->ops && se->ops->is_active) { 1211 if (!se->ops->is_active(se->opaque)) { 1212 continue; 1213 } 1214 } 1215 se->ops->save_live_pending(f, se->opaque, threshold_size, 1216 res_non_postcopiable, res_postcopiable); 1217 } 1218 } 1219 1220 void qemu_savevm_state_cleanup(void) 1221 { 1222 SaveStateEntry *se; 1223 1224 trace_savevm_state_cleanup(); 1225 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1226 if (se->ops && se->ops->cleanup) { 1227 se->ops->cleanup(se->opaque); 1228 } 1229 } 1230 } 1231 1232 static int qemu_savevm_state(QEMUFile *f, Error **errp) 1233 { 1234 int ret; 1235 MigrationParams params = { 1236 .blk = 0, 1237 .shared = 0 1238 }; 1239 MigrationState *ms = migrate_init(¶ms); 1240 MigrationStatus status; 1241 ms->to_dst_file = f; 1242 1243 if (migration_is_blocked(errp)) { 1244 ret = -EINVAL; 1245 goto done; 1246 } 1247 1248 qemu_mutex_unlock_iothread(); 1249 qemu_savevm_state_header(f); 1250 qemu_savevm_state_begin(f, ¶ms); 1251 qemu_mutex_lock_iothread(); 1252 1253 while (qemu_file_get_error(f) == 0) { 1254 if (qemu_savevm_state_iterate(f, false) > 0) { 1255 break; 1256 } 1257 } 1258 1259 ret = qemu_file_get_error(f); 1260 if (ret == 0) { 1261 qemu_savevm_state_complete_precopy(f, false); 1262 ret = qemu_file_get_error(f); 1263 } 1264 qemu_savevm_state_cleanup(); 1265 if (ret != 0) { 1266 error_setg_errno(errp, -ret, "Error while writing VM state"); 1267 } 1268 1269 done: 1270 if (ret != 0) { 1271 status = MIGRATION_STATUS_FAILED; 1272 } else { 1273 status = MIGRATION_STATUS_COMPLETED; 1274 } 1275 migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status); 1276 1277 /* f is outer parameter, it should not stay in global migration state after 1278 * this function finished */ 1279 ms->to_dst_file = NULL; 1280 1281 return ret; 1282 } 1283 1284 static int qemu_save_device_state(QEMUFile *f) 1285 { 1286 SaveStateEntry *se; 1287 1288 qemu_put_be32(f, QEMU_VM_FILE_MAGIC); 1289 qemu_put_be32(f, QEMU_VM_FILE_VERSION); 1290 1291 cpu_synchronize_all_states(); 1292 1293 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1294 if (se->is_ram) { 1295 continue; 1296 } 1297 if ((!se->ops || !se->ops->save_state) && !se->vmsd) { 1298 continue; 1299 } 1300 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) { 1301 continue; 1302 } 1303 1304 save_section_header(f, se, QEMU_VM_SECTION_FULL); 1305 1306 vmstate_save(f, se, NULL); 1307 1308 save_section_footer(f, se); 1309 } 1310 1311 qemu_put_byte(f, QEMU_VM_EOF); 1312 1313 return qemu_file_get_error(f); 1314 } 1315 1316 static SaveStateEntry *find_se(const char *idstr, int instance_id) 1317 { 1318 SaveStateEntry *se; 1319 1320 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1321 if (!strcmp(se->idstr, idstr) && 1322 (instance_id == se->instance_id || 1323 instance_id == se->alias_id)) 1324 return se; 1325 /* Migrating from an older version? */ 1326 if (strstr(se->idstr, idstr) && se->compat) { 1327 if (!strcmp(se->compat->idstr, idstr) && 1328 (instance_id == se->compat->instance_id || 1329 instance_id == se->alias_id)) 1330 return se; 1331 } 1332 } 1333 return NULL; 1334 } 1335 1336 enum LoadVMExitCodes { 1337 /* Allow a command to quit all layers of nested loadvm loops */ 1338 LOADVM_QUIT = 1, 1339 }; 1340 1341 static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis); 1342 1343 /* ------ incoming postcopy messages ------ */ 1344 /* 'advise' arrives before any transfers just to tell us that a postcopy 1345 * *might* happen - it might be skipped if precopy transferred everything 1346 * quickly. 1347 */ 1348 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis) 1349 { 1350 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE); 1351 uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps; 1352 1353 trace_loadvm_postcopy_handle_advise(); 1354 if (ps != POSTCOPY_INCOMING_NONE) { 1355 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps); 1356 return -1; 1357 } 1358 1359 if (!postcopy_ram_supported_by_host()) { 1360 postcopy_state_set(POSTCOPY_INCOMING_NONE); 1361 return -1; 1362 } 1363 1364 remote_pagesize_summary = qemu_get_be64(mis->from_src_file); 1365 local_pagesize_summary = ram_pagesize_summary(); 1366 1367 if (remote_pagesize_summary != local_pagesize_summary) { 1368 /* 1369 * This detects two potential causes of mismatch: 1370 * a) A mismatch in host page sizes 1371 * Some combinations of mismatch are probably possible but it gets 1372 * a bit more complicated. In particular we need to place whole 1373 * host pages on the dest at once, and we need to ensure that we 1374 * handle dirtying to make sure we never end up sending part of 1375 * a hostpage on it's own. 1376 * b) The use of different huge page sizes on source/destination 1377 * a more fine grain test is performed during RAM block migration 1378 * but this test here causes a nice early clear failure, and 1379 * also fails when passed to an older qemu that doesn't 1380 * do huge pages. 1381 */ 1382 error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64 1383 " d=%" PRIx64 ")", 1384 remote_pagesize_summary, local_pagesize_summary); 1385 return -1; 1386 } 1387 1388 remote_tps = qemu_get_be64(mis->from_src_file); 1389 if (remote_tps != qemu_target_page_size()) { 1390 /* 1391 * Again, some differences could be dealt with, but for now keep it 1392 * simple. 1393 */ 1394 error_report("Postcopy needs matching target page sizes (s=%d d=%zd)", 1395 (int)remote_tps, qemu_target_page_size()); 1396 return -1; 1397 } 1398 1399 if (ram_postcopy_incoming_init(mis)) { 1400 return -1; 1401 } 1402 1403 postcopy_state_set(POSTCOPY_INCOMING_ADVISE); 1404 1405 return 0; 1406 } 1407 1408 /* After postcopy we will be told to throw some pages away since they're 1409 * dirty and will have to be demand fetched. Must happen before CPU is 1410 * started. 1411 * There can be 0..many of these messages, each encoding multiple pages. 1412 */ 1413 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis, 1414 uint16_t len) 1415 { 1416 int tmp; 1417 char ramid[256]; 1418 PostcopyState ps = postcopy_state_get(); 1419 1420 trace_loadvm_postcopy_ram_handle_discard(); 1421 1422 switch (ps) { 1423 case POSTCOPY_INCOMING_ADVISE: 1424 /* 1st discard */ 1425 tmp = postcopy_ram_prepare_discard(mis); 1426 if (tmp) { 1427 return tmp; 1428 } 1429 break; 1430 1431 case POSTCOPY_INCOMING_DISCARD: 1432 /* Expected state */ 1433 break; 1434 1435 default: 1436 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)", 1437 ps); 1438 return -1; 1439 } 1440 /* We're expecting a 1441 * Version (0) 1442 * a RAM ID string (length byte, name, 0 term) 1443 * then at least 1 16 byte chunk 1444 */ 1445 if (len < (1 + 1 + 1 + 1 + 2 * 8)) { 1446 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len); 1447 return -1; 1448 } 1449 1450 tmp = qemu_get_byte(mis->from_src_file); 1451 if (tmp != postcopy_ram_discard_version) { 1452 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp); 1453 return -1; 1454 } 1455 1456 if (!qemu_get_counted_string(mis->from_src_file, ramid)) { 1457 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID"); 1458 return -1; 1459 } 1460 tmp = qemu_get_byte(mis->from_src_file); 1461 if (tmp != 0) { 1462 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp); 1463 return -1; 1464 } 1465 1466 len -= 3 + strlen(ramid); 1467 if (len % 16) { 1468 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len); 1469 return -1; 1470 } 1471 trace_loadvm_postcopy_ram_handle_discard_header(ramid, len); 1472 while (len) { 1473 uint64_t start_addr, block_length; 1474 start_addr = qemu_get_be64(mis->from_src_file); 1475 block_length = qemu_get_be64(mis->from_src_file); 1476 1477 len -= 16; 1478 int ret = ram_discard_range(ramid, start_addr, block_length); 1479 if (ret) { 1480 return ret; 1481 } 1482 } 1483 trace_loadvm_postcopy_ram_handle_discard_end(); 1484 1485 return 0; 1486 } 1487 1488 /* 1489 * Triggered by a postcopy_listen command; this thread takes over reading 1490 * the input stream, leaving the main thread free to carry on loading the rest 1491 * of the device state (from RAM). 1492 * (TODO:This could do with being in a postcopy file - but there again it's 1493 * just another input loop, not that postcopy specific) 1494 */ 1495 static void *postcopy_ram_listen_thread(void *opaque) 1496 { 1497 QEMUFile *f = opaque; 1498 MigrationIncomingState *mis = migration_incoming_get_current(); 1499 int load_res; 1500 1501 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, 1502 MIGRATION_STATUS_POSTCOPY_ACTIVE); 1503 qemu_sem_post(&mis->listen_thread_sem); 1504 trace_postcopy_ram_listen_thread_start(); 1505 1506 /* 1507 * Because we're a thread and not a coroutine we can't yield 1508 * in qemu_file, and thus we must be blocking now. 1509 */ 1510 qemu_file_set_blocking(f, true); 1511 load_res = qemu_loadvm_state_main(f, mis); 1512 /* And non-blocking again so we don't block in any cleanup */ 1513 qemu_file_set_blocking(f, false); 1514 1515 trace_postcopy_ram_listen_thread_exit(); 1516 if (load_res < 0) { 1517 error_report("%s: loadvm failed: %d", __func__, load_res); 1518 qemu_file_set_error(f, load_res); 1519 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, 1520 MIGRATION_STATUS_FAILED); 1521 } else { 1522 /* 1523 * This looks good, but it's possible that the device loading in the 1524 * main thread hasn't finished yet, and so we might not be in 'RUN' 1525 * state yet; wait for the end of the main thread. 1526 */ 1527 qemu_event_wait(&mis->main_thread_load_event); 1528 } 1529 postcopy_ram_incoming_cleanup(mis); 1530 1531 if (load_res < 0) { 1532 /* 1533 * If something went wrong then we have a bad state so exit; 1534 * depending how far we got it might be possible at this point 1535 * to leave the guest running and fire MCEs for pages that never 1536 * arrived as a desperate recovery step. 1537 */ 1538 exit(EXIT_FAILURE); 1539 } 1540 1541 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, 1542 MIGRATION_STATUS_COMPLETED); 1543 /* 1544 * If everything has worked fine, then the main thread has waited 1545 * for us to start, and we're the last use of the mis. 1546 * (If something broke then qemu will have to exit anyway since it's 1547 * got a bad migration state). 1548 */ 1549 migration_incoming_state_destroy(); 1550 1551 1552 return NULL; 1553 } 1554 1555 /* After this message we must be able to immediately receive postcopy data */ 1556 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis) 1557 { 1558 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING); 1559 trace_loadvm_postcopy_handle_listen(); 1560 if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) { 1561 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps); 1562 return -1; 1563 } 1564 if (ps == POSTCOPY_INCOMING_ADVISE) { 1565 /* 1566 * A rare case, we entered listen without having to do any discards, 1567 * so do the setup that's normally done at the time of the 1st discard. 1568 */ 1569 postcopy_ram_prepare_discard(mis); 1570 } 1571 1572 /* 1573 * Sensitise RAM - can now generate requests for blocks that don't exist 1574 * However, at this point the CPU shouldn't be running, and the IO 1575 * shouldn't be doing anything yet so don't actually expect requests 1576 */ 1577 if (postcopy_ram_enable_notify(mis)) { 1578 return -1; 1579 } 1580 1581 if (mis->have_listen_thread) { 1582 error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread"); 1583 return -1; 1584 } 1585 1586 mis->have_listen_thread = true; 1587 /* Start up the listening thread and wait for it to signal ready */ 1588 qemu_sem_init(&mis->listen_thread_sem, 0); 1589 qemu_thread_create(&mis->listen_thread, "postcopy/listen", 1590 postcopy_ram_listen_thread, mis->from_src_file, 1591 QEMU_THREAD_DETACHED); 1592 qemu_sem_wait(&mis->listen_thread_sem); 1593 qemu_sem_destroy(&mis->listen_thread_sem); 1594 1595 return 0; 1596 } 1597 1598 1599 typedef struct { 1600 QEMUBH *bh; 1601 } HandleRunBhData; 1602 1603 static void loadvm_postcopy_handle_run_bh(void *opaque) 1604 { 1605 Error *local_err = NULL; 1606 HandleRunBhData *data = opaque; 1607 1608 /* TODO we should move all of this lot into postcopy_ram.c or a shared code 1609 * in migration.c 1610 */ 1611 cpu_synchronize_all_post_init(); 1612 1613 qemu_announce_self(); 1614 1615 /* Make sure all file formats flush their mutable metadata */ 1616 bdrv_invalidate_cache_all(&local_err); 1617 if (local_err) { 1618 error_report_err(local_err); 1619 } 1620 1621 /* If we get an error here, just don't restart the VM yet. */ 1622 blk_resume_after_migration(&local_err); 1623 if (local_err) { 1624 error_free(local_err); 1625 local_err = NULL; 1626 autostart = false; 1627 } 1628 1629 trace_loadvm_postcopy_handle_run_cpu_sync(); 1630 cpu_synchronize_all_post_init(); 1631 1632 trace_loadvm_postcopy_handle_run_vmstart(); 1633 1634 if (autostart) { 1635 /* Hold onto your hats, starting the CPU */ 1636 vm_start(); 1637 } else { 1638 /* leave it paused and let management decide when to start the CPU */ 1639 runstate_set(RUN_STATE_PAUSED); 1640 } 1641 1642 qemu_bh_delete(data->bh); 1643 g_free(data); 1644 } 1645 1646 /* After all discards we can start running and asking for pages */ 1647 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis) 1648 { 1649 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_RUNNING); 1650 HandleRunBhData *data; 1651 1652 trace_loadvm_postcopy_handle_run(); 1653 if (ps != POSTCOPY_INCOMING_LISTENING) { 1654 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps); 1655 return -1; 1656 } 1657 1658 data = g_new(HandleRunBhData, 1); 1659 data->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, data); 1660 qemu_bh_schedule(data->bh); 1661 1662 /* We need to finish reading the stream from the package 1663 * and also stop reading anything more from the stream that loaded the 1664 * package (since it's now being read by the listener thread). 1665 * LOADVM_QUIT will quit all the layers of nested loadvm loops. 1666 */ 1667 return LOADVM_QUIT; 1668 } 1669 1670 /** 1671 * Immediately following this command is a blob of data containing an embedded 1672 * chunk of migration stream; read it and load it. 1673 * 1674 * @mis: Incoming state 1675 * @length: Length of packaged data to read 1676 * 1677 * Returns: Negative values on error 1678 * 1679 */ 1680 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis) 1681 { 1682 int ret; 1683 size_t length; 1684 QIOChannelBuffer *bioc; 1685 1686 length = qemu_get_be32(mis->from_src_file); 1687 trace_loadvm_handle_cmd_packaged(length); 1688 1689 if (length > MAX_VM_CMD_PACKAGED_SIZE) { 1690 error_report("Unreasonably large packaged state: %zu", length); 1691 return -1; 1692 } 1693 1694 bioc = qio_channel_buffer_new(length); 1695 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer"); 1696 ret = qemu_get_buffer(mis->from_src_file, 1697 bioc->data, 1698 length); 1699 if (ret != length) { 1700 object_unref(OBJECT(bioc)); 1701 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu", 1702 ret, length); 1703 return (ret < 0) ? ret : -EAGAIN; 1704 } 1705 bioc->usage += length; 1706 trace_loadvm_handle_cmd_packaged_received(ret); 1707 1708 QEMUFile *packf = qemu_fopen_channel_input(QIO_CHANNEL(bioc)); 1709 1710 ret = qemu_loadvm_state_main(packf, mis); 1711 trace_loadvm_handle_cmd_packaged_main(ret); 1712 qemu_fclose(packf); 1713 object_unref(OBJECT(bioc)); 1714 1715 return ret; 1716 } 1717 1718 /* 1719 * Process an incoming 'QEMU_VM_COMMAND' 1720 * 0 just a normal return 1721 * LOADVM_QUIT All good, but exit the loop 1722 * <0 Error 1723 */ 1724 static int loadvm_process_command(QEMUFile *f) 1725 { 1726 MigrationIncomingState *mis = migration_incoming_get_current(); 1727 uint16_t cmd; 1728 uint16_t len; 1729 uint32_t tmp32; 1730 1731 cmd = qemu_get_be16(f); 1732 len = qemu_get_be16(f); 1733 1734 trace_loadvm_process_command(cmd, len); 1735 if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) { 1736 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len); 1737 return -EINVAL; 1738 } 1739 1740 if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) { 1741 error_report("%s received with bad length - expecting %zu, got %d", 1742 mig_cmd_args[cmd].name, 1743 (size_t)mig_cmd_args[cmd].len, len); 1744 return -ERANGE; 1745 } 1746 1747 switch (cmd) { 1748 case MIG_CMD_OPEN_RETURN_PATH: 1749 if (mis->to_src_file) { 1750 error_report("CMD_OPEN_RETURN_PATH called when RP already open"); 1751 /* Not really a problem, so don't give up */ 1752 return 0; 1753 } 1754 mis->to_src_file = qemu_file_get_return_path(f); 1755 if (!mis->to_src_file) { 1756 error_report("CMD_OPEN_RETURN_PATH failed"); 1757 return -1; 1758 } 1759 break; 1760 1761 case MIG_CMD_PING: 1762 tmp32 = qemu_get_be32(f); 1763 trace_loadvm_process_command_ping(tmp32); 1764 if (!mis->to_src_file) { 1765 error_report("CMD_PING (0x%x) received with no return path", 1766 tmp32); 1767 return -1; 1768 } 1769 migrate_send_rp_pong(mis, tmp32); 1770 break; 1771 1772 case MIG_CMD_PACKAGED: 1773 return loadvm_handle_cmd_packaged(mis); 1774 1775 case MIG_CMD_POSTCOPY_ADVISE: 1776 return loadvm_postcopy_handle_advise(mis); 1777 1778 case MIG_CMD_POSTCOPY_LISTEN: 1779 return loadvm_postcopy_handle_listen(mis); 1780 1781 case MIG_CMD_POSTCOPY_RUN: 1782 return loadvm_postcopy_handle_run(mis); 1783 1784 case MIG_CMD_POSTCOPY_RAM_DISCARD: 1785 return loadvm_postcopy_ram_handle_discard(mis, len); 1786 } 1787 1788 return 0; 1789 } 1790 1791 struct LoadStateEntry { 1792 QLIST_ENTRY(LoadStateEntry) entry; 1793 SaveStateEntry *se; 1794 int section_id; 1795 int version_id; 1796 }; 1797 1798 /* 1799 * Read a footer off the wire and check that it matches the expected section 1800 * 1801 * Returns: true if the footer was good 1802 * false if there is a problem (and calls error_report to say why) 1803 */ 1804 static bool check_section_footer(QEMUFile *f, LoadStateEntry *le) 1805 { 1806 uint8_t read_mark; 1807 uint32_t read_section_id; 1808 1809 if (skip_section_footers) { 1810 /* No footer to check */ 1811 return true; 1812 } 1813 1814 read_mark = qemu_get_byte(f); 1815 1816 if (read_mark != QEMU_VM_SECTION_FOOTER) { 1817 error_report("Missing section footer for %s", le->se->idstr); 1818 return false; 1819 } 1820 1821 read_section_id = qemu_get_be32(f); 1822 if (read_section_id != le->section_id) { 1823 error_report("Mismatched section id in footer for %s -" 1824 " read 0x%x expected 0x%x", 1825 le->se->idstr, read_section_id, le->section_id); 1826 return false; 1827 } 1828 1829 /* All good */ 1830 return true; 1831 } 1832 1833 void loadvm_free_handlers(MigrationIncomingState *mis) 1834 { 1835 LoadStateEntry *le, *new_le; 1836 1837 QLIST_FOREACH_SAFE(le, &mis->loadvm_handlers, entry, new_le) { 1838 QLIST_REMOVE(le, entry); 1839 g_free(le); 1840 } 1841 } 1842 1843 static int 1844 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis) 1845 { 1846 uint32_t instance_id, version_id, section_id; 1847 SaveStateEntry *se; 1848 LoadStateEntry *le; 1849 char idstr[256]; 1850 int ret; 1851 1852 /* Read section start */ 1853 section_id = qemu_get_be32(f); 1854 if (!qemu_get_counted_string(f, idstr)) { 1855 error_report("Unable to read ID string for section %u", 1856 section_id); 1857 return -EINVAL; 1858 } 1859 instance_id = qemu_get_be32(f); 1860 version_id = qemu_get_be32(f); 1861 1862 trace_qemu_loadvm_state_section_startfull(section_id, idstr, 1863 instance_id, version_id); 1864 /* Find savevm section */ 1865 se = find_se(idstr, instance_id); 1866 if (se == NULL) { 1867 error_report("Unknown savevm section or instance '%s' %d", 1868 idstr, instance_id); 1869 return -EINVAL; 1870 } 1871 1872 /* Validate version */ 1873 if (version_id > se->version_id) { 1874 error_report("savevm: unsupported version %d for '%s' v%d", 1875 version_id, idstr, se->version_id); 1876 return -EINVAL; 1877 } 1878 1879 /* Validate if it is a device's state */ 1880 if (xen_enabled() && se->is_ram) { 1881 error_report("loadvm: %s RAM loading not allowed on Xen", idstr); 1882 return -EINVAL; 1883 } 1884 1885 /* Add entry */ 1886 le = g_malloc0(sizeof(*le)); 1887 1888 le->se = se; 1889 le->section_id = section_id; 1890 le->version_id = version_id; 1891 QLIST_INSERT_HEAD(&mis->loadvm_handlers, le, entry); 1892 1893 ret = vmstate_load(f, le->se, le->version_id); 1894 if (ret < 0) { 1895 error_report("error while loading state for instance 0x%x of" 1896 " device '%s'", instance_id, idstr); 1897 return ret; 1898 } 1899 if (!check_section_footer(f, le)) { 1900 return -EINVAL; 1901 } 1902 1903 return 0; 1904 } 1905 1906 static int 1907 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis) 1908 { 1909 uint32_t section_id; 1910 LoadStateEntry *le; 1911 int ret; 1912 1913 section_id = qemu_get_be32(f); 1914 1915 trace_qemu_loadvm_state_section_partend(section_id); 1916 QLIST_FOREACH(le, &mis->loadvm_handlers, entry) { 1917 if (le->section_id == section_id) { 1918 break; 1919 } 1920 } 1921 if (le == NULL) { 1922 error_report("Unknown savevm section %d", section_id); 1923 return -EINVAL; 1924 } 1925 1926 ret = vmstate_load(f, le->se, le->version_id); 1927 if (ret < 0) { 1928 error_report("error while loading state section id %d(%s)", 1929 section_id, le->se->idstr); 1930 return ret; 1931 } 1932 if (!check_section_footer(f, le)) { 1933 return -EINVAL; 1934 } 1935 1936 return 0; 1937 } 1938 1939 static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis) 1940 { 1941 uint8_t section_type; 1942 int ret = 0; 1943 1944 while ((section_type = qemu_get_byte(f)) != QEMU_VM_EOF) { 1945 ret = 0; 1946 trace_qemu_loadvm_state_section(section_type); 1947 switch (section_type) { 1948 case QEMU_VM_SECTION_START: 1949 case QEMU_VM_SECTION_FULL: 1950 ret = qemu_loadvm_section_start_full(f, mis); 1951 if (ret < 0) { 1952 goto out; 1953 } 1954 break; 1955 case QEMU_VM_SECTION_PART: 1956 case QEMU_VM_SECTION_END: 1957 ret = qemu_loadvm_section_part_end(f, mis); 1958 if (ret < 0) { 1959 goto out; 1960 } 1961 break; 1962 case QEMU_VM_COMMAND: 1963 ret = loadvm_process_command(f); 1964 trace_qemu_loadvm_state_section_command(ret); 1965 if ((ret < 0) || (ret & LOADVM_QUIT)) { 1966 goto out; 1967 } 1968 break; 1969 default: 1970 error_report("Unknown savevm section type %d", section_type); 1971 ret = -EINVAL; 1972 goto out; 1973 } 1974 } 1975 1976 out: 1977 if (ret < 0) { 1978 qemu_file_set_error(f, ret); 1979 } 1980 return ret; 1981 } 1982 1983 int qemu_loadvm_state(QEMUFile *f) 1984 { 1985 MigrationIncomingState *mis = migration_incoming_get_current(); 1986 Error *local_err = NULL; 1987 unsigned int v; 1988 int ret; 1989 1990 if (qemu_savevm_state_blocked(&local_err)) { 1991 error_report_err(local_err); 1992 return -EINVAL; 1993 } 1994 1995 v = qemu_get_be32(f); 1996 if (v != QEMU_VM_FILE_MAGIC) { 1997 error_report("Not a migration stream"); 1998 return -EINVAL; 1999 } 2000 2001 v = qemu_get_be32(f); 2002 if (v == QEMU_VM_FILE_VERSION_COMPAT) { 2003 error_report("SaveVM v2 format is obsolete and don't work anymore"); 2004 return -ENOTSUP; 2005 } 2006 if (v != QEMU_VM_FILE_VERSION) { 2007 error_report("Unsupported migration stream version"); 2008 return -ENOTSUP; 2009 } 2010 2011 if (!savevm_state.skip_configuration || enforce_config_section()) { 2012 if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) { 2013 error_report("Configuration section missing"); 2014 return -EINVAL; 2015 } 2016 ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0); 2017 2018 if (ret) { 2019 return ret; 2020 } 2021 } 2022 2023 ret = qemu_loadvm_state_main(f, mis); 2024 qemu_event_set(&mis->main_thread_load_event); 2025 2026 trace_qemu_loadvm_state_post_main(ret); 2027 2028 if (mis->have_listen_thread) { 2029 /* Listen thread still going, can't clean up yet */ 2030 return ret; 2031 } 2032 2033 if (ret == 0) { 2034 ret = qemu_file_get_error(f); 2035 } 2036 2037 /* 2038 * Try to read in the VMDESC section as well, so that dumping tools that 2039 * intercept our migration stream have the chance to see it. 2040 */ 2041 2042 /* We've got to be careful; if we don't read the data and just shut the fd 2043 * then the sender can error if we close while it's still sending. 2044 * We also mustn't read data that isn't there; some transports (RDMA) 2045 * will stall waiting for that data when the source has already closed. 2046 */ 2047 if (ret == 0 && should_send_vmdesc()) { 2048 uint8_t *buf; 2049 uint32_t size; 2050 uint8_t section_type = qemu_get_byte(f); 2051 2052 if (section_type != QEMU_VM_VMDESCRIPTION) { 2053 error_report("Expected vmdescription section, but got %d", 2054 section_type); 2055 /* 2056 * It doesn't seem worth failing at this point since 2057 * we apparently have an otherwise valid VM state 2058 */ 2059 } else { 2060 buf = g_malloc(0x1000); 2061 size = qemu_get_be32(f); 2062 2063 while (size > 0) { 2064 uint32_t read_chunk = MIN(size, 0x1000); 2065 qemu_get_buffer(f, buf, read_chunk); 2066 size -= read_chunk; 2067 } 2068 g_free(buf); 2069 } 2070 } 2071 2072 cpu_synchronize_all_post_init(); 2073 2074 return ret; 2075 } 2076 2077 int save_vmstate(const char *name) 2078 { 2079 BlockDriverState *bs, *bs1; 2080 QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1; 2081 int ret = -1; 2082 QEMUFile *f; 2083 int saved_vm_running; 2084 uint64_t vm_state_size; 2085 qemu_timeval tv; 2086 struct tm tm; 2087 Error *local_err = NULL; 2088 AioContext *aio_context; 2089 2090 if (!bdrv_all_can_snapshot(&bs)) { 2091 error_report("Device '%s' is writable but does not support snapshots", 2092 bdrv_get_device_name(bs)); 2093 return ret; 2094 } 2095 2096 /* Delete old snapshots of the same name */ 2097 if (name) { 2098 ret = bdrv_all_delete_snapshot(name, &bs1, &local_err); 2099 if (ret < 0) { 2100 error_reportf_err(local_err, 2101 "Error while deleting snapshot on device '%s': ", 2102 bdrv_get_device_name(bs1)); 2103 return ret; 2104 } 2105 } 2106 2107 bs = bdrv_all_find_vmstate_bs(); 2108 if (bs == NULL) { 2109 error_report("No block device can accept snapshots"); 2110 return ret; 2111 } 2112 aio_context = bdrv_get_aio_context(bs); 2113 2114 saved_vm_running = runstate_is_running(); 2115 2116 ret = global_state_store(); 2117 if (ret) { 2118 error_report("Error saving global state"); 2119 return ret; 2120 } 2121 vm_stop(RUN_STATE_SAVE_VM); 2122 2123 aio_context_acquire(aio_context); 2124 2125 memset(sn, 0, sizeof(*sn)); 2126 2127 /* fill auxiliary fields */ 2128 qemu_gettimeofday(&tv); 2129 sn->date_sec = tv.tv_sec; 2130 sn->date_nsec = tv.tv_usec * 1000; 2131 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL); 2132 2133 if (name) { 2134 ret = bdrv_snapshot_find(bs, old_sn, name); 2135 if (ret >= 0) { 2136 pstrcpy(sn->name, sizeof(sn->name), old_sn->name); 2137 pstrcpy(sn->id_str, sizeof(sn->id_str), old_sn->id_str); 2138 } else { 2139 pstrcpy(sn->name, sizeof(sn->name), name); 2140 } 2141 } else { 2142 /* cast below needed for OpenBSD where tv_sec is still 'long' */ 2143 localtime_r((const time_t *)&tv.tv_sec, &tm); 2144 strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm); 2145 } 2146 2147 /* save the VM state */ 2148 f = qemu_fopen_bdrv(bs, 1); 2149 if (!f) { 2150 error_report("Could not open VM state file"); 2151 goto the_end; 2152 } 2153 ret = qemu_savevm_state(f, &local_err); 2154 vm_state_size = qemu_ftell(f); 2155 qemu_fclose(f); 2156 if (ret < 0) { 2157 error_report_err(local_err); 2158 goto the_end; 2159 } 2160 2161 ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, &bs); 2162 if (ret < 0) { 2163 error_report("Error while creating snapshot on '%s'", 2164 bdrv_get_device_name(bs)); 2165 goto the_end; 2166 } 2167 2168 ret = 0; 2169 2170 the_end: 2171 aio_context_release(aio_context); 2172 if (saved_vm_running) { 2173 vm_start(); 2174 } 2175 return ret; 2176 } 2177 2178 void qmp_xen_save_devices_state(const char *filename, Error **errp) 2179 { 2180 QEMUFile *f; 2181 QIOChannelFile *ioc; 2182 int saved_vm_running; 2183 int ret; 2184 2185 saved_vm_running = runstate_is_running(); 2186 vm_stop(RUN_STATE_SAVE_VM); 2187 global_state_store_running(); 2188 2189 ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT, 0660, errp); 2190 if (!ioc) { 2191 goto the_end; 2192 } 2193 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state"); 2194 f = qemu_fopen_channel_output(QIO_CHANNEL(ioc)); 2195 ret = qemu_save_device_state(f); 2196 qemu_fclose(f); 2197 if (ret < 0) { 2198 error_setg(errp, QERR_IO_ERROR); 2199 } 2200 2201 the_end: 2202 if (saved_vm_running) { 2203 vm_start(); 2204 } 2205 } 2206 2207 void qmp_xen_load_devices_state(const char *filename, Error **errp) 2208 { 2209 QEMUFile *f; 2210 QIOChannelFile *ioc; 2211 int ret; 2212 2213 /* Guest must be paused before loading the device state; the RAM state 2214 * will already have been loaded by xc 2215 */ 2216 if (runstate_is_running()) { 2217 error_setg(errp, "Cannot update device state while vm is running"); 2218 return; 2219 } 2220 vm_stop(RUN_STATE_RESTORE_VM); 2221 2222 ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp); 2223 if (!ioc) { 2224 return; 2225 } 2226 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state"); 2227 f = qemu_fopen_channel_input(QIO_CHANNEL(ioc)); 2228 2229 ret = qemu_loadvm_state(f); 2230 qemu_fclose(f); 2231 if (ret < 0) { 2232 error_setg(errp, QERR_IO_ERROR); 2233 } 2234 migration_incoming_state_destroy(); 2235 } 2236 2237 int load_vmstate(const char *name) 2238 { 2239 BlockDriverState *bs, *bs_vm_state; 2240 QEMUSnapshotInfo sn; 2241 QEMUFile *f; 2242 int ret; 2243 AioContext *aio_context; 2244 MigrationIncomingState *mis = migration_incoming_get_current(); 2245 2246 if (!bdrv_all_can_snapshot(&bs)) { 2247 error_report("Device '%s' is writable but does not support snapshots", 2248 bdrv_get_device_name(bs)); 2249 return -ENOTSUP; 2250 } 2251 ret = bdrv_all_find_snapshot(name, &bs); 2252 if (ret < 0) { 2253 error_report("Device '%s' does not have the requested snapshot '%s'", 2254 bdrv_get_device_name(bs), name); 2255 return ret; 2256 } 2257 2258 bs_vm_state = bdrv_all_find_vmstate_bs(); 2259 if (!bs_vm_state) { 2260 error_report("No block device supports snapshots"); 2261 return -ENOTSUP; 2262 } 2263 aio_context = bdrv_get_aio_context(bs_vm_state); 2264 2265 /* Don't even try to load empty VM states */ 2266 aio_context_acquire(aio_context); 2267 ret = bdrv_snapshot_find(bs_vm_state, &sn, name); 2268 aio_context_release(aio_context); 2269 if (ret < 0) { 2270 return ret; 2271 } else if (sn.vm_state_size == 0) { 2272 error_report("This is a disk-only snapshot. Revert to it offline " 2273 "using qemu-img."); 2274 return -EINVAL; 2275 } 2276 2277 /* Flush all IO requests so they don't interfere with the new state. */ 2278 bdrv_drain_all(); 2279 2280 ret = bdrv_all_goto_snapshot(name, &bs); 2281 if (ret < 0) { 2282 error_report("Error %d while activating snapshot '%s' on '%s'", 2283 ret, name, bdrv_get_device_name(bs)); 2284 return ret; 2285 } 2286 2287 /* restore the VM state */ 2288 f = qemu_fopen_bdrv(bs_vm_state, 0); 2289 if (!f) { 2290 error_report("Could not open VM state file"); 2291 return -EINVAL; 2292 } 2293 2294 qemu_system_reset(VMRESET_SILENT); 2295 mis->from_src_file = f; 2296 2297 aio_context_acquire(aio_context); 2298 ret = qemu_loadvm_state(f); 2299 qemu_fclose(f); 2300 aio_context_release(aio_context); 2301 2302 migration_incoming_state_destroy(); 2303 if (ret < 0) { 2304 error_report("Error %d while loading VM state", ret); 2305 return ret; 2306 } 2307 2308 return 0; 2309 } 2310 2311 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev) 2312 { 2313 qemu_ram_set_idstr(mr->ram_block, 2314 memory_region_name(mr), dev); 2315 } 2316 2317 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev) 2318 { 2319 qemu_ram_unset_idstr(mr->ram_block); 2320 } 2321 2322 void vmstate_register_ram_global(MemoryRegion *mr) 2323 { 2324 vmstate_register_ram(mr, NULL); 2325 } 2326