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/hw.h" 32 #include "hw/qdev.h" 33 #include "hw/xen/xen.h" 34 #include "net/net.h" 35 #include "sysemu/sysemu.h" 36 #include "qemu/timer.h" 37 #include "migration/migration.h" 38 #include "migration/snapshot.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/qmp/qerror.h" 45 #include "qemu/error-report.h" 46 #include "qemu/queue.h" 47 #include "sysemu/cpus.h" 48 #include "exec/memory.h" 49 #include "exec/target_page.h" 50 #include "qmp-commands.h" 51 #include "trace.h" 52 #include "qemu/bitops.h" 53 #include "qemu/iov.h" 54 #include "block/snapshot.h" 55 #include "qemu/cutils.h" 56 #include "io/channel-buffer.h" 57 #include "io/channel-file.h" 58 59 #ifndef ETH_P_RARP 60 #define ETH_P_RARP 0x8035 61 #endif 62 #define ARP_HTYPE_ETH 0x0001 63 #define ARP_PTYPE_IP 0x0800 64 #define ARP_OP_REQUEST_REV 0x3 65 66 const unsigned int postcopy_ram_discard_version = 0; 67 68 static bool skip_section_footers; 69 70 /* Subcommands for QEMU_VM_COMMAND */ 71 enum qemu_vm_cmd { 72 MIG_CMD_INVALID = 0, /* Must be 0 */ 73 MIG_CMD_OPEN_RETURN_PATH, /* Tell the dest to open the Return path */ 74 MIG_CMD_PING, /* Request a PONG on the RP */ 75 76 MIG_CMD_POSTCOPY_ADVISE, /* Prior to any page transfers, just 77 warn we might want to do PC */ 78 MIG_CMD_POSTCOPY_LISTEN, /* Start listening for incoming 79 pages as it's running. */ 80 MIG_CMD_POSTCOPY_RUN, /* Start execution */ 81 82 MIG_CMD_POSTCOPY_RAM_DISCARD, /* A list of pages to discard that 83 were previously sent during 84 precopy but are dirty. */ 85 MIG_CMD_PACKAGED, /* Send a wrapped stream within this stream */ 86 MIG_CMD_MAX 87 }; 88 89 #define MAX_VM_CMD_PACKAGED_SIZE (1ul << 24) 90 static struct mig_cmd_args { 91 ssize_t len; /* -1 = variable */ 92 const char *name; 93 } mig_cmd_args[] = { 94 [MIG_CMD_INVALID] = { .len = -1, .name = "INVALID" }, 95 [MIG_CMD_OPEN_RETURN_PATH] = { .len = 0, .name = "OPEN_RETURN_PATH" }, 96 [MIG_CMD_PING] = { .len = sizeof(uint32_t), .name = "PING" }, 97 [MIG_CMD_POSTCOPY_ADVISE] = { .len = 16, .name = "POSTCOPY_ADVISE" }, 98 [MIG_CMD_POSTCOPY_LISTEN] = { .len = 0, .name = "POSTCOPY_LISTEN" }, 99 [MIG_CMD_POSTCOPY_RUN] = { .len = 0, .name = "POSTCOPY_RUN" }, 100 [MIG_CMD_POSTCOPY_RAM_DISCARD] = { 101 .len = -1, .name = "POSTCOPY_RAM_DISCARD" }, 102 [MIG_CMD_PACKAGED] = { .len = 4, .name = "PACKAGED" }, 103 [MIG_CMD_MAX] = { .len = -1, .name = "MAX" }, 104 }; 105 106 static int announce_self_create(uint8_t *buf, 107 uint8_t *mac_addr) 108 { 109 /* Ethernet header. */ 110 memset(buf, 0xff, 6); /* destination MAC addr */ 111 memcpy(buf + 6, mac_addr, 6); /* source MAC addr */ 112 *(uint16_t *)(buf + 12) = htons(ETH_P_RARP); /* ethertype */ 113 114 /* RARP header. */ 115 *(uint16_t *)(buf + 14) = htons(ARP_HTYPE_ETH); /* hardware addr space */ 116 *(uint16_t *)(buf + 16) = htons(ARP_PTYPE_IP); /* protocol addr space */ 117 *(buf + 18) = 6; /* hardware addr length (ethernet) */ 118 *(buf + 19) = 4; /* protocol addr length (IPv4) */ 119 *(uint16_t *)(buf + 20) = htons(ARP_OP_REQUEST_REV); /* opcode */ 120 memcpy(buf + 22, mac_addr, 6); /* source hw addr */ 121 memset(buf + 28, 0x00, 4); /* source protocol addr */ 122 memcpy(buf + 32, mac_addr, 6); /* target hw addr */ 123 memset(buf + 38, 0x00, 4); /* target protocol addr */ 124 125 /* Padding to get up to 60 bytes (ethernet min packet size, minus FCS). */ 126 memset(buf + 42, 0x00, 18); 127 128 return 60; /* len (FCS will be added by hardware) */ 129 } 130 131 static void qemu_announce_self_iter(NICState *nic, void *opaque) 132 { 133 uint8_t buf[60]; 134 int len; 135 136 trace_qemu_announce_self_iter(qemu_ether_ntoa(&nic->conf->macaddr)); 137 len = announce_self_create(buf, nic->conf->macaddr.a); 138 139 qemu_send_packet_raw(qemu_get_queue(nic), buf, len); 140 } 141 142 143 static void qemu_announce_self_once(void *opaque) 144 { 145 static int count = SELF_ANNOUNCE_ROUNDS; 146 QEMUTimer *timer = *(QEMUTimer **)opaque; 147 148 qemu_foreach_nic(qemu_announce_self_iter, NULL); 149 150 if (--count) { 151 /* delay 50ms, 150ms, 250ms, ... */ 152 timer_mod(timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) + 153 self_announce_delay(count)); 154 } else { 155 timer_del(timer); 156 timer_free(timer); 157 } 158 } 159 160 void qemu_announce_self(void) 161 { 162 static QEMUTimer *timer; 163 timer = timer_new_ms(QEMU_CLOCK_REALTIME, qemu_announce_self_once, &timer); 164 qemu_announce_self_once(&timer); 165 } 166 167 /***********************************************************/ 168 /* savevm/loadvm support */ 169 170 static ssize_t block_writev_buffer(void *opaque, struct iovec *iov, int iovcnt, 171 int64_t pos) 172 { 173 int ret; 174 QEMUIOVector qiov; 175 176 qemu_iovec_init_external(&qiov, iov, iovcnt); 177 ret = bdrv_writev_vmstate(opaque, &qiov, pos); 178 if (ret < 0) { 179 return ret; 180 } 181 182 return qiov.size; 183 } 184 185 static ssize_t block_get_buffer(void *opaque, uint8_t *buf, int64_t pos, 186 size_t size) 187 { 188 return bdrv_load_vmstate(opaque, buf, pos, size); 189 } 190 191 static int bdrv_fclose(void *opaque) 192 { 193 return bdrv_flush(opaque); 194 } 195 196 static const QEMUFileOps bdrv_read_ops = { 197 .get_buffer = block_get_buffer, 198 .close = bdrv_fclose 199 }; 200 201 static const QEMUFileOps bdrv_write_ops = { 202 .writev_buffer = block_writev_buffer, 203 .close = bdrv_fclose 204 }; 205 206 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable) 207 { 208 if (is_writable) { 209 return qemu_fopen_ops(bs, &bdrv_write_ops); 210 } 211 return qemu_fopen_ops(bs, &bdrv_read_ops); 212 } 213 214 215 /* QEMUFile timer support. 216 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c 217 */ 218 219 void timer_put(QEMUFile *f, QEMUTimer *ts) 220 { 221 uint64_t expire_time; 222 223 expire_time = timer_expire_time_ns(ts); 224 qemu_put_be64(f, expire_time); 225 } 226 227 void timer_get(QEMUFile *f, QEMUTimer *ts) 228 { 229 uint64_t expire_time; 230 231 expire_time = qemu_get_be64(f); 232 if (expire_time != -1) { 233 timer_mod_ns(ts, expire_time); 234 } else { 235 timer_del(ts); 236 } 237 } 238 239 240 /* VMState timer support. 241 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c 242 */ 243 244 static int get_timer(QEMUFile *f, void *pv, size_t size, VMStateField *field) 245 { 246 QEMUTimer *v = pv; 247 timer_get(f, v); 248 return 0; 249 } 250 251 static int put_timer(QEMUFile *f, void *pv, size_t size, VMStateField *field, 252 QJSON *vmdesc) 253 { 254 QEMUTimer *v = pv; 255 timer_put(f, v); 256 257 return 0; 258 } 259 260 const VMStateInfo vmstate_info_timer = { 261 .name = "timer", 262 .get = get_timer, 263 .put = put_timer, 264 }; 265 266 267 typedef struct CompatEntry { 268 char idstr[256]; 269 int instance_id; 270 } CompatEntry; 271 272 typedef struct SaveStateEntry { 273 QTAILQ_ENTRY(SaveStateEntry) entry; 274 char idstr[256]; 275 int instance_id; 276 int alias_id; 277 int version_id; 278 /* version id read from the stream */ 279 int load_version_id; 280 int section_id; 281 /* section id read from the stream */ 282 int load_section_id; 283 SaveVMHandlers *ops; 284 const VMStateDescription *vmsd; 285 void *opaque; 286 CompatEntry *compat; 287 int is_ram; 288 } SaveStateEntry; 289 290 typedef struct SaveState { 291 QTAILQ_HEAD(, SaveStateEntry) handlers; 292 int global_section_id; 293 bool skip_configuration; 294 uint32_t len; 295 const char *name; 296 uint32_t target_page_bits; 297 } SaveState; 298 299 static SaveState savevm_state = { 300 .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers), 301 .global_section_id = 0, 302 .skip_configuration = false, 303 }; 304 305 void savevm_skip_configuration(void) 306 { 307 savevm_state.skip_configuration = true; 308 } 309 310 311 static void configuration_pre_save(void *opaque) 312 { 313 SaveState *state = opaque; 314 const char *current_name = MACHINE_GET_CLASS(current_machine)->name; 315 316 state->len = strlen(current_name); 317 state->name = current_name; 318 state->target_page_bits = qemu_target_page_bits(); 319 } 320 321 static int configuration_pre_load(void *opaque) 322 { 323 SaveState *state = opaque; 324 325 /* If there is no target-page-bits subsection it means the source 326 * predates the variable-target-page-bits support and is using the 327 * minimum possible value for this CPU. 328 */ 329 state->target_page_bits = qemu_target_page_bits_min(); 330 return 0; 331 } 332 333 static int configuration_post_load(void *opaque, int version_id) 334 { 335 SaveState *state = opaque; 336 const char *current_name = MACHINE_GET_CLASS(current_machine)->name; 337 338 if (strncmp(state->name, current_name, state->len) != 0) { 339 error_report("Machine type received is '%.*s' and local is '%s'", 340 (int) state->len, state->name, current_name); 341 return -EINVAL; 342 } 343 344 if (state->target_page_bits != qemu_target_page_bits()) { 345 error_report("Received TARGET_PAGE_BITS is %d but local is %d", 346 state->target_page_bits, qemu_target_page_bits()); 347 return -EINVAL; 348 } 349 350 return 0; 351 } 352 353 /* The target-page-bits subsection is present only if the 354 * target page size is not the same as the default (ie the 355 * minimum page size for a variable-page-size guest CPU). 356 * If it is present then it contains the actual target page 357 * bits for the machine, and migration will fail if the 358 * two ends don't agree about it. 359 */ 360 static bool vmstate_target_page_bits_needed(void *opaque) 361 { 362 return qemu_target_page_bits() 363 > qemu_target_page_bits_min(); 364 } 365 366 static const VMStateDescription vmstate_target_page_bits = { 367 .name = "configuration/target-page-bits", 368 .version_id = 1, 369 .minimum_version_id = 1, 370 .needed = vmstate_target_page_bits_needed, 371 .fields = (VMStateField[]) { 372 VMSTATE_UINT32(target_page_bits, SaveState), 373 VMSTATE_END_OF_LIST() 374 } 375 }; 376 377 static const VMStateDescription vmstate_configuration = { 378 .name = "configuration", 379 .version_id = 1, 380 .pre_load = configuration_pre_load, 381 .post_load = configuration_post_load, 382 .pre_save = configuration_pre_save, 383 .fields = (VMStateField[]) { 384 VMSTATE_UINT32(len, SaveState), 385 VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, len), 386 VMSTATE_END_OF_LIST() 387 }, 388 .subsections = (const VMStateDescription*[]) { 389 &vmstate_target_page_bits, 390 NULL 391 } 392 }; 393 394 static void dump_vmstate_vmsd(FILE *out_file, 395 const VMStateDescription *vmsd, int indent, 396 bool is_subsection); 397 398 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field, 399 int indent) 400 { 401 fprintf(out_file, "%*s{\n", indent, ""); 402 indent += 2; 403 fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name); 404 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "", 405 field->version_id); 406 fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "", 407 field->field_exists ? "true" : "false"); 408 fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size); 409 if (field->vmsd != NULL) { 410 fprintf(out_file, ",\n"); 411 dump_vmstate_vmsd(out_file, field->vmsd, indent, false); 412 } 413 fprintf(out_file, "\n%*s}", indent - 2, ""); 414 } 415 416 static void dump_vmstate_vmss(FILE *out_file, 417 const VMStateDescription **subsection, 418 int indent) 419 { 420 if (*subsection != NULL) { 421 dump_vmstate_vmsd(out_file, *subsection, indent, true); 422 } 423 } 424 425 static void dump_vmstate_vmsd(FILE *out_file, 426 const VMStateDescription *vmsd, int indent, 427 bool is_subsection) 428 { 429 if (is_subsection) { 430 fprintf(out_file, "%*s{\n", indent, ""); 431 } else { 432 fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description"); 433 } 434 indent += 2; 435 fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name); 436 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "", 437 vmsd->version_id); 438 fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "", 439 vmsd->minimum_version_id); 440 if (vmsd->fields != NULL) { 441 const VMStateField *field = vmsd->fields; 442 bool first; 443 444 fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, ""); 445 first = true; 446 while (field->name != NULL) { 447 if (field->flags & VMS_MUST_EXIST) { 448 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */ 449 field++; 450 continue; 451 } 452 if (!first) { 453 fprintf(out_file, ",\n"); 454 } 455 dump_vmstate_vmsf(out_file, field, indent + 2); 456 field++; 457 first = false; 458 } 459 fprintf(out_file, "\n%*s]", indent, ""); 460 } 461 if (vmsd->subsections != NULL) { 462 const VMStateDescription **subsection = vmsd->subsections; 463 bool first; 464 465 fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, ""); 466 first = true; 467 while (*subsection != NULL) { 468 if (!first) { 469 fprintf(out_file, ",\n"); 470 } 471 dump_vmstate_vmss(out_file, subsection, indent + 2); 472 subsection++; 473 first = false; 474 } 475 fprintf(out_file, "\n%*s]", indent, ""); 476 } 477 fprintf(out_file, "\n%*s}", indent - 2, ""); 478 } 479 480 static void dump_machine_type(FILE *out_file) 481 { 482 MachineClass *mc; 483 484 mc = MACHINE_GET_CLASS(current_machine); 485 486 fprintf(out_file, " \"vmschkmachine\": {\n"); 487 fprintf(out_file, " \"Name\": \"%s\"\n", mc->name); 488 fprintf(out_file, " },\n"); 489 } 490 491 void dump_vmstate_json_to_file(FILE *out_file) 492 { 493 GSList *list, *elt; 494 bool first; 495 496 fprintf(out_file, "{\n"); 497 dump_machine_type(out_file); 498 499 first = true; 500 list = object_class_get_list(TYPE_DEVICE, true); 501 for (elt = list; elt; elt = elt->next) { 502 DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data, 503 TYPE_DEVICE); 504 const char *name; 505 int indent = 2; 506 507 if (!dc->vmsd) { 508 continue; 509 } 510 511 if (!first) { 512 fprintf(out_file, ",\n"); 513 } 514 name = object_class_get_name(OBJECT_CLASS(dc)); 515 fprintf(out_file, "%*s\"%s\": {\n", indent, "", name); 516 indent += 2; 517 fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name); 518 fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "", 519 dc->vmsd->version_id); 520 fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "", 521 dc->vmsd->minimum_version_id); 522 523 dump_vmstate_vmsd(out_file, dc->vmsd, indent, false); 524 525 fprintf(out_file, "\n%*s}", indent - 2, ""); 526 first = false; 527 } 528 fprintf(out_file, "\n}\n"); 529 fclose(out_file); 530 } 531 532 static int calculate_new_instance_id(const char *idstr) 533 { 534 SaveStateEntry *se; 535 int instance_id = 0; 536 537 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 538 if (strcmp(idstr, se->idstr) == 0 539 && instance_id <= se->instance_id) { 540 instance_id = se->instance_id + 1; 541 } 542 } 543 return instance_id; 544 } 545 546 static int calculate_compat_instance_id(const char *idstr) 547 { 548 SaveStateEntry *se; 549 int instance_id = 0; 550 551 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 552 if (!se->compat) { 553 continue; 554 } 555 556 if (strcmp(idstr, se->compat->idstr) == 0 557 && instance_id <= se->compat->instance_id) { 558 instance_id = se->compat->instance_id + 1; 559 } 560 } 561 return instance_id; 562 } 563 564 static inline MigrationPriority save_state_priority(SaveStateEntry *se) 565 { 566 if (se->vmsd) { 567 return se->vmsd->priority; 568 } 569 return MIG_PRI_DEFAULT; 570 } 571 572 static void savevm_state_handler_insert(SaveStateEntry *nse) 573 { 574 MigrationPriority priority = save_state_priority(nse); 575 SaveStateEntry *se; 576 577 assert(priority <= MIG_PRI_MAX); 578 579 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 580 if (save_state_priority(se) < priority) { 581 break; 582 } 583 } 584 585 if (se) { 586 QTAILQ_INSERT_BEFORE(se, nse, entry); 587 } else { 588 QTAILQ_INSERT_TAIL(&savevm_state.handlers, nse, entry); 589 } 590 } 591 592 /* TODO: Individual devices generally have very little idea about the rest 593 of the system, so instance_id should be removed/replaced. 594 Meanwhile pass -1 as instance_id if you do not already have a clearly 595 distinguishing id for all instances of your device class. */ 596 int register_savevm_live(DeviceState *dev, 597 const char *idstr, 598 int instance_id, 599 int version_id, 600 SaveVMHandlers *ops, 601 void *opaque) 602 { 603 SaveStateEntry *se; 604 605 se = g_new0(SaveStateEntry, 1); 606 se->version_id = version_id; 607 se->section_id = savevm_state.global_section_id++; 608 se->ops = ops; 609 se->opaque = opaque; 610 se->vmsd = NULL; 611 /* if this is a live_savem then set is_ram */ 612 if (ops->save_live_setup != NULL) { 613 se->is_ram = 1; 614 } 615 616 if (dev) { 617 char *id = qdev_get_dev_path(dev); 618 if (id) { 619 if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >= 620 sizeof(se->idstr)) { 621 error_report("Path too long for VMState (%s)", id); 622 g_free(id); 623 g_free(se); 624 625 return -1; 626 } 627 g_free(id); 628 629 se->compat = g_new0(CompatEntry, 1); 630 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), idstr); 631 se->compat->instance_id = instance_id == -1 ? 632 calculate_compat_instance_id(idstr) : instance_id; 633 instance_id = -1; 634 } 635 } 636 pstrcat(se->idstr, sizeof(se->idstr), idstr); 637 638 if (instance_id == -1) { 639 se->instance_id = calculate_new_instance_id(se->idstr); 640 } else { 641 se->instance_id = instance_id; 642 } 643 assert(!se->compat || se->instance_id == 0); 644 savevm_state_handler_insert(se); 645 return 0; 646 } 647 648 void unregister_savevm(DeviceState *dev, const char *idstr, void *opaque) 649 { 650 SaveStateEntry *se, *new_se; 651 char id[256] = ""; 652 653 if (dev) { 654 char *path = qdev_get_dev_path(dev); 655 if (path) { 656 pstrcpy(id, sizeof(id), path); 657 pstrcat(id, sizeof(id), "/"); 658 g_free(path); 659 } 660 } 661 pstrcat(id, sizeof(id), idstr); 662 663 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) { 664 if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) { 665 QTAILQ_REMOVE(&savevm_state.handlers, se, entry); 666 g_free(se->compat); 667 g_free(se); 668 } 669 } 670 } 671 672 int vmstate_register_with_alias_id(DeviceState *dev, int instance_id, 673 const VMStateDescription *vmsd, 674 void *opaque, int alias_id, 675 int required_for_version, 676 Error **errp) 677 { 678 SaveStateEntry *se; 679 680 /* If this triggers, alias support can be dropped for the vmsd. */ 681 assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id); 682 683 se = g_new0(SaveStateEntry, 1); 684 se->version_id = vmsd->version_id; 685 se->section_id = savevm_state.global_section_id++; 686 se->opaque = opaque; 687 se->vmsd = vmsd; 688 se->alias_id = alias_id; 689 690 if (dev) { 691 char *id = qdev_get_dev_path(dev); 692 if (id) { 693 if (snprintf(se->idstr, sizeof(se->idstr), "%s/", id) >= 694 sizeof(se->idstr)) { 695 error_setg(errp, "Path too long for VMState (%s)", id); 696 g_free(id); 697 g_free(se); 698 699 return -1; 700 } 701 g_free(id); 702 703 se->compat = g_new0(CompatEntry, 1); 704 pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name); 705 se->compat->instance_id = instance_id == -1 ? 706 calculate_compat_instance_id(vmsd->name) : instance_id; 707 instance_id = -1; 708 } 709 } 710 pstrcat(se->idstr, sizeof(se->idstr), vmsd->name); 711 712 if (instance_id == -1) { 713 se->instance_id = calculate_new_instance_id(se->idstr); 714 } else { 715 se->instance_id = instance_id; 716 } 717 assert(!se->compat || se->instance_id == 0); 718 savevm_state_handler_insert(se); 719 return 0; 720 } 721 722 void vmstate_unregister(DeviceState *dev, const VMStateDescription *vmsd, 723 void *opaque) 724 { 725 SaveStateEntry *se, *new_se; 726 727 QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) { 728 if (se->vmsd == vmsd && se->opaque == opaque) { 729 QTAILQ_REMOVE(&savevm_state.handlers, se, entry); 730 g_free(se->compat); 731 g_free(se); 732 } 733 } 734 } 735 736 static int vmstate_load(QEMUFile *f, SaveStateEntry *se) 737 { 738 trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)"); 739 if (!se->vmsd) { /* Old style */ 740 return se->ops->load_state(f, se->opaque, se->load_version_id); 741 } 742 return vmstate_load_state(f, se->vmsd, se->opaque, se->load_version_id); 743 } 744 745 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc) 746 { 747 int64_t old_offset, size; 748 749 old_offset = qemu_ftell_fast(f); 750 se->ops->save_state(f, se->opaque); 751 size = qemu_ftell_fast(f) - old_offset; 752 753 if (vmdesc) { 754 json_prop_int(vmdesc, "size", size); 755 json_start_array(vmdesc, "fields"); 756 json_start_object(vmdesc, NULL); 757 json_prop_str(vmdesc, "name", "data"); 758 json_prop_int(vmdesc, "size", size); 759 json_prop_str(vmdesc, "type", "buffer"); 760 json_end_object(vmdesc); 761 json_end_array(vmdesc); 762 } 763 } 764 765 static void vmstate_save(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc) 766 { 767 trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)"); 768 if (!se->vmsd) { 769 vmstate_save_old_style(f, se, vmdesc); 770 return; 771 } 772 vmstate_save_state(f, se->vmsd, se->opaque, vmdesc); 773 } 774 775 void savevm_skip_section_footers(void) 776 { 777 skip_section_footers = true; 778 } 779 780 /* 781 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL) 782 */ 783 static void save_section_header(QEMUFile *f, SaveStateEntry *se, 784 uint8_t section_type) 785 { 786 qemu_put_byte(f, section_type); 787 qemu_put_be32(f, se->section_id); 788 789 if (section_type == QEMU_VM_SECTION_FULL || 790 section_type == QEMU_VM_SECTION_START) { 791 /* ID string */ 792 size_t len = strlen(se->idstr); 793 qemu_put_byte(f, len); 794 qemu_put_buffer(f, (uint8_t *)se->idstr, len); 795 796 qemu_put_be32(f, se->instance_id); 797 qemu_put_be32(f, se->version_id); 798 } 799 } 800 801 /* 802 * Write a footer onto device sections that catches cases misformatted device 803 * sections. 804 */ 805 static void save_section_footer(QEMUFile *f, SaveStateEntry *se) 806 { 807 if (!skip_section_footers) { 808 qemu_put_byte(f, QEMU_VM_SECTION_FOOTER); 809 qemu_put_be32(f, se->section_id); 810 } 811 } 812 813 /** 814 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the 815 * command and associated data. 816 * 817 * @f: File to send command on 818 * @command: Command type to send 819 * @len: Length of associated data 820 * @data: Data associated with command. 821 */ 822 static void qemu_savevm_command_send(QEMUFile *f, 823 enum qemu_vm_cmd command, 824 uint16_t len, 825 uint8_t *data) 826 { 827 trace_savevm_command_send(command, len); 828 qemu_put_byte(f, QEMU_VM_COMMAND); 829 qemu_put_be16(f, (uint16_t)command); 830 qemu_put_be16(f, len); 831 qemu_put_buffer(f, data, len); 832 qemu_fflush(f); 833 } 834 835 void qemu_savevm_send_ping(QEMUFile *f, uint32_t value) 836 { 837 uint32_t buf; 838 839 trace_savevm_send_ping(value); 840 buf = cpu_to_be32(value); 841 qemu_savevm_command_send(f, MIG_CMD_PING, sizeof(value), (uint8_t *)&buf); 842 } 843 844 void qemu_savevm_send_open_return_path(QEMUFile *f) 845 { 846 trace_savevm_send_open_return_path(); 847 qemu_savevm_command_send(f, MIG_CMD_OPEN_RETURN_PATH, 0, NULL); 848 } 849 850 /* We have a buffer of data to send; we don't want that all to be loaded 851 * by the command itself, so the command contains just the length of the 852 * extra buffer that we then send straight after it. 853 * TODO: Must be a better way to organise that 854 * 855 * Returns: 856 * 0 on success 857 * -ve on error 858 */ 859 int qemu_savevm_send_packaged(QEMUFile *f, const uint8_t *buf, size_t len) 860 { 861 uint32_t tmp; 862 863 if (len > MAX_VM_CMD_PACKAGED_SIZE) { 864 error_report("%s: Unreasonably large packaged state: %zu", 865 __func__, len); 866 return -1; 867 } 868 869 tmp = cpu_to_be32(len); 870 871 trace_qemu_savevm_send_packaged(); 872 qemu_savevm_command_send(f, MIG_CMD_PACKAGED, 4, (uint8_t *)&tmp); 873 874 qemu_put_buffer(f, buf, len); 875 876 return 0; 877 } 878 879 /* Send prior to any postcopy transfer */ 880 void qemu_savevm_send_postcopy_advise(QEMUFile *f) 881 { 882 uint64_t tmp[2]; 883 tmp[0] = cpu_to_be64(ram_pagesize_summary()); 884 tmp[1] = cpu_to_be64(qemu_target_page_size()); 885 886 trace_qemu_savevm_send_postcopy_advise(); 887 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_ADVISE, 16, (uint8_t *)tmp); 888 } 889 890 /* Sent prior to starting the destination running in postcopy, discard pages 891 * that have already been sent but redirtied on the source. 892 * CMD_POSTCOPY_RAM_DISCARD consist of: 893 * byte version (0) 894 * byte Length of name field (not including 0) 895 * n x byte RAM block name 896 * byte 0 terminator (just for safety) 897 * n x Byte ranges within the named RAMBlock 898 * be64 Start of the range 899 * be64 Length 900 * 901 * name: RAMBlock name that these entries are part of 902 * len: Number of page entries 903 * start_list: 'len' addresses 904 * length_list: 'len' addresses 905 * 906 */ 907 void qemu_savevm_send_postcopy_ram_discard(QEMUFile *f, const char *name, 908 uint16_t len, 909 uint64_t *start_list, 910 uint64_t *length_list) 911 { 912 uint8_t *buf; 913 uint16_t tmplen; 914 uint16_t t; 915 size_t name_len = strlen(name); 916 917 trace_qemu_savevm_send_postcopy_ram_discard(name, len); 918 assert(name_len < 256); 919 buf = g_malloc0(1 + 1 + name_len + 1 + (8 + 8) * len); 920 buf[0] = postcopy_ram_discard_version; 921 buf[1] = name_len; 922 memcpy(buf + 2, name, name_len); 923 tmplen = 2 + name_len; 924 buf[tmplen++] = '\0'; 925 926 for (t = 0; t < len; t++) { 927 stq_be_p(buf + tmplen, start_list[t]); 928 tmplen += 8; 929 stq_be_p(buf + tmplen, length_list[t]); 930 tmplen += 8; 931 } 932 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RAM_DISCARD, tmplen, buf); 933 g_free(buf); 934 } 935 936 /* Get the destination into a state where it can receive postcopy data. */ 937 void qemu_savevm_send_postcopy_listen(QEMUFile *f) 938 { 939 trace_savevm_send_postcopy_listen(); 940 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_LISTEN, 0, NULL); 941 } 942 943 /* Kick the destination into running */ 944 void qemu_savevm_send_postcopy_run(QEMUFile *f) 945 { 946 trace_savevm_send_postcopy_run(); 947 qemu_savevm_command_send(f, MIG_CMD_POSTCOPY_RUN, 0, NULL); 948 } 949 950 bool qemu_savevm_state_blocked(Error **errp) 951 { 952 SaveStateEntry *se; 953 954 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 955 if (se->vmsd && se->vmsd->unmigratable) { 956 error_setg(errp, "State blocked by non-migratable device '%s'", 957 se->idstr); 958 return true; 959 } 960 } 961 return false; 962 } 963 964 static bool enforce_config_section(void) 965 { 966 MachineState *machine = MACHINE(qdev_get_machine()); 967 return machine->enforce_config_section; 968 } 969 970 void qemu_savevm_state_header(QEMUFile *f) 971 { 972 trace_savevm_state_header(); 973 qemu_put_be32(f, QEMU_VM_FILE_MAGIC); 974 qemu_put_be32(f, QEMU_VM_FILE_VERSION); 975 976 if (!savevm_state.skip_configuration || enforce_config_section()) { 977 qemu_put_byte(f, QEMU_VM_CONFIGURATION); 978 vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0); 979 } 980 981 } 982 983 void qemu_savevm_state_begin(QEMUFile *f) 984 { 985 SaveStateEntry *se; 986 int ret; 987 988 trace_savevm_state_begin(); 989 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 990 if (!se->ops || !se->ops->save_live_setup) { 991 continue; 992 } 993 if (se->ops && se->ops->is_active) { 994 if (!se->ops->is_active(se->opaque)) { 995 continue; 996 } 997 } 998 save_section_header(f, se, QEMU_VM_SECTION_START); 999 1000 ret = se->ops->save_live_setup(f, se->opaque); 1001 save_section_footer(f, se); 1002 if (ret < 0) { 1003 qemu_file_set_error(f, ret); 1004 break; 1005 } 1006 } 1007 } 1008 1009 /* 1010 * this function has three return values: 1011 * negative: there was one error, and we have -errno. 1012 * 0 : We haven't finished, caller have to go again 1013 * 1 : We have finished, we can go to complete phase 1014 */ 1015 int qemu_savevm_state_iterate(QEMUFile *f, bool postcopy) 1016 { 1017 SaveStateEntry *se; 1018 int ret = 1; 1019 1020 trace_savevm_state_iterate(); 1021 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1022 if (!se->ops || !se->ops->save_live_iterate) { 1023 continue; 1024 } 1025 if (se->ops && se->ops->is_active) { 1026 if (!se->ops->is_active(se->opaque)) { 1027 continue; 1028 } 1029 } 1030 /* 1031 * In the postcopy phase, any device that doesn't know how to 1032 * do postcopy should have saved it's state in the _complete 1033 * call that's already run, it might get confused if we call 1034 * iterate afterwards. 1035 */ 1036 if (postcopy && !se->ops->save_live_complete_postcopy) { 1037 continue; 1038 } 1039 if (qemu_file_rate_limit(f)) { 1040 return 0; 1041 } 1042 trace_savevm_section_start(se->idstr, se->section_id); 1043 1044 save_section_header(f, se, QEMU_VM_SECTION_PART); 1045 1046 ret = se->ops->save_live_iterate(f, se->opaque); 1047 trace_savevm_section_end(se->idstr, se->section_id, ret); 1048 save_section_footer(f, se); 1049 1050 if (ret < 0) { 1051 qemu_file_set_error(f, ret); 1052 } 1053 if (ret <= 0) { 1054 /* Do not proceed to the next vmstate before this one reported 1055 completion of the current stage. This serializes the migration 1056 and reduces the probability that a faster changing state is 1057 synchronized over and over again. */ 1058 break; 1059 } 1060 } 1061 return ret; 1062 } 1063 1064 static bool should_send_vmdesc(void) 1065 { 1066 MachineState *machine = MACHINE(qdev_get_machine()); 1067 bool in_postcopy = migration_in_postcopy(); 1068 return !machine->suppress_vmdesc && !in_postcopy; 1069 } 1070 1071 /* 1072 * Calls the save_live_complete_postcopy methods 1073 * causing the last few pages to be sent immediately and doing any associated 1074 * cleanup. 1075 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete 1076 * all the other devices, but that happens at the point we switch to postcopy. 1077 */ 1078 void qemu_savevm_state_complete_postcopy(QEMUFile *f) 1079 { 1080 SaveStateEntry *se; 1081 int ret; 1082 1083 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1084 if (!se->ops || !se->ops->save_live_complete_postcopy) { 1085 continue; 1086 } 1087 if (se->ops && se->ops->is_active) { 1088 if (!se->ops->is_active(se->opaque)) { 1089 continue; 1090 } 1091 } 1092 trace_savevm_section_start(se->idstr, se->section_id); 1093 /* Section type */ 1094 qemu_put_byte(f, QEMU_VM_SECTION_END); 1095 qemu_put_be32(f, se->section_id); 1096 1097 ret = se->ops->save_live_complete_postcopy(f, se->opaque); 1098 trace_savevm_section_end(se->idstr, se->section_id, ret); 1099 save_section_footer(f, se); 1100 if (ret < 0) { 1101 qemu_file_set_error(f, ret); 1102 return; 1103 } 1104 } 1105 1106 qemu_put_byte(f, QEMU_VM_EOF); 1107 qemu_fflush(f); 1108 } 1109 1110 void qemu_savevm_state_complete_precopy(QEMUFile *f, bool iterable_only) 1111 { 1112 QJSON *vmdesc; 1113 int vmdesc_len; 1114 SaveStateEntry *se; 1115 int ret; 1116 bool in_postcopy = migration_in_postcopy(); 1117 1118 trace_savevm_state_complete_precopy(); 1119 1120 cpu_synchronize_all_states(); 1121 1122 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1123 if (!se->ops || 1124 (in_postcopy && se->ops->save_live_complete_postcopy) || 1125 (in_postcopy && !iterable_only) || 1126 !se->ops->save_live_complete_precopy) { 1127 continue; 1128 } 1129 1130 if (se->ops && se->ops->is_active) { 1131 if (!se->ops->is_active(se->opaque)) { 1132 continue; 1133 } 1134 } 1135 trace_savevm_section_start(se->idstr, se->section_id); 1136 1137 save_section_header(f, se, QEMU_VM_SECTION_END); 1138 1139 ret = se->ops->save_live_complete_precopy(f, se->opaque); 1140 trace_savevm_section_end(se->idstr, se->section_id, ret); 1141 save_section_footer(f, se); 1142 if (ret < 0) { 1143 qemu_file_set_error(f, ret); 1144 return; 1145 } 1146 } 1147 1148 if (iterable_only) { 1149 return; 1150 } 1151 1152 vmdesc = qjson_new(); 1153 json_prop_int(vmdesc, "page_size", qemu_target_page_size()); 1154 json_start_array(vmdesc, "devices"); 1155 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1156 1157 if ((!se->ops || !se->ops->save_state) && !se->vmsd) { 1158 continue; 1159 } 1160 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) { 1161 trace_savevm_section_skip(se->idstr, se->section_id); 1162 continue; 1163 } 1164 1165 trace_savevm_section_start(se->idstr, se->section_id); 1166 1167 json_start_object(vmdesc, NULL); 1168 json_prop_str(vmdesc, "name", se->idstr); 1169 json_prop_int(vmdesc, "instance_id", se->instance_id); 1170 1171 save_section_header(f, se, QEMU_VM_SECTION_FULL); 1172 vmstate_save(f, se, vmdesc); 1173 trace_savevm_section_end(se->idstr, se->section_id, 0); 1174 save_section_footer(f, se); 1175 1176 json_end_object(vmdesc); 1177 } 1178 1179 if (!in_postcopy) { 1180 /* Postcopy stream will still be going */ 1181 qemu_put_byte(f, QEMU_VM_EOF); 1182 } 1183 1184 json_end_array(vmdesc); 1185 qjson_finish(vmdesc); 1186 vmdesc_len = strlen(qjson_get_str(vmdesc)); 1187 1188 if (should_send_vmdesc()) { 1189 qemu_put_byte(f, QEMU_VM_VMDESCRIPTION); 1190 qemu_put_be32(f, vmdesc_len); 1191 qemu_put_buffer(f, (uint8_t *)qjson_get_str(vmdesc), vmdesc_len); 1192 } 1193 qjson_destroy(vmdesc); 1194 1195 qemu_fflush(f); 1196 } 1197 1198 /* Give an estimate of the amount left to be transferred, 1199 * the result is split into the amount for units that can and 1200 * for units that can't do postcopy. 1201 */ 1202 void qemu_savevm_state_pending(QEMUFile *f, uint64_t threshold_size, 1203 uint64_t *res_non_postcopiable, 1204 uint64_t *res_postcopiable) 1205 { 1206 SaveStateEntry *se; 1207 1208 *res_non_postcopiable = 0; 1209 *res_postcopiable = 0; 1210 1211 1212 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1213 if (!se->ops || !se->ops->save_live_pending) { 1214 continue; 1215 } 1216 if (se->ops && se->ops->is_active) { 1217 if (!se->ops->is_active(se->opaque)) { 1218 continue; 1219 } 1220 } 1221 se->ops->save_live_pending(f, se->opaque, threshold_size, 1222 res_non_postcopiable, res_postcopiable); 1223 } 1224 } 1225 1226 void qemu_savevm_state_cleanup(void) 1227 { 1228 SaveStateEntry *se; 1229 1230 trace_savevm_state_cleanup(); 1231 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1232 if (se->ops && se->ops->cleanup) { 1233 se->ops->cleanup(se->opaque); 1234 } 1235 } 1236 } 1237 1238 static int qemu_savevm_state(QEMUFile *f, Error **errp) 1239 { 1240 int ret; 1241 MigrationState *ms = migrate_init(); 1242 MigrationStatus status; 1243 ms->to_dst_file = f; 1244 1245 if (migration_is_blocked(errp)) { 1246 ret = -EINVAL; 1247 goto done; 1248 } 1249 1250 if (migrate_use_block()) { 1251 error_setg(errp, "Block migration and snapshots are incompatible"); 1252 ret = -EINVAL; 1253 goto done; 1254 } 1255 1256 qemu_mutex_unlock_iothread(); 1257 qemu_savevm_state_header(f); 1258 qemu_savevm_state_begin(f); 1259 qemu_mutex_lock_iothread(); 1260 1261 while (qemu_file_get_error(f) == 0) { 1262 if (qemu_savevm_state_iterate(f, false) > 0) { 1263 break; 1264 } 1265 } 1266 1267 ret = qemu_file_get_error(f); 1268 if (ret == 0) { 1269 qemu_savevm_state_complete_precopy(f, false); 1270 ret = qemu_file_get_error(f); 1271 } 1272 qemu_savevm_state_cleanup(); 1273 if (ret != 0) { 1274 error_setg_errno(errp, -ret, "Error while writing VM state"); 1275 } 1276 1277 done: 1278 if (ret != 0) { 1279 status = MIGRATION_STATUS_FAILED; 1280 } else { 1281 status = MIGRATION_STATUS_COMPLETED; 1282 } 1283 migrate_set_state(&ms->state, MIGRATION_STATUS_SETUP, status); 1284 1285 /* f is outer parameter, it should not stay in global migration state after 1286 * this function finished */ 1287 ms->to_dst_file = NULL; 1288 1289 return ret; 1290 } 1291 1292 static int qemu_save_device_state(QEMUFile *f) 1293 { 1294 SaveStateEntry *se; 1295 1296 qemu_put_be32(f, QEMU_VM_FILE_MAGIC); 1297 qemu_put_be32(f, QEMU_VM_FILE_VERSION); 1298 1299 cpu_synchronize_all_states(); 1300 1301 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1302 if (se->is_ram) { 1303 continue; 1304 } 1305 if ((!se->ops || !se->ops->save_state) && !se->vmsd) { 1306 continue; 1307 } 1308 if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) { 1309 continue; 1310 } 1311 1312 save_section_header(f, se, QEMU_VM_SECTION_FULL); 1313 1314 vmstate_save(f, se, NULL); 1315 1316 save_section_footer(f, se); 1317 } 1318 1319 qemu_put_byte(f, QEMU_VM_EOF); 1320 1321 return qemu_file_get_error(f); 1322 } 1323 1324 static SaveStateEntry *find_se(const char *idstr, int instance_id) 1325 { 1326 SaveStateEntry *se; 1327 1328 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1329 if (!strcmp(se->idstr, idstr) && 1330 (instance_id == se->instance_id || 1331 instance_id == se->alias_id)) 1332 return se; 1333 /* Migrating from an older version? */ 1334 if (strstr(se->idstr, idstr) && se->compat) { 1335 if (!strcmp(se->compat->idstr, idstr) && 1336 (instance_id == se->compat->instance_id || 1337 instance_id == se->alias_id)) 1338 return se; 1339 } 1340 } 1341 return NULL; 1342 } 1343 1344 enum LoadVMExitCodes { 1345 /* Allow a command to quit all layers of nested loadvm loops */ 1346 LOADVM_QUIT = 1, 1347 }; 1348 1349 static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis); 1350 1351 /* ------ incoming postcopy messages ------ */ 1352 /* 'advise' arrives before any transfers just to tell us that a postcopy 1353 * *might* happen - it might be skipped if precopy transferred everything 1354 * quickly. 1355 */ 1356 static int loadvm_postcopy_handle_advise(MigrationIncomingState *mis) 1357 { 1358 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_ADVISE); 1359 uint64_t remote_pagesize_summary, local_pagesize_summary, remote_tps; 1360 1361 trace_loadvm_postcopy_handle_advise(); 1362 if (ps != POSTCOPY_INCOMING_NONE) { 1363 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps); 1364 return -1; 1365 } 1366 1367 if (!postcopy_ram_supported_by_host()) { 1368 postcopy_state_set(POSTCOPY_INCOMING_NONE); 1369 return -1; 1370 } 1371 1372 remote_pagesize_summary = qemu_get_be64(mis->from_src_file); 1373 local_pagesize_summary = ram_pagesize_summary(); 1374 1375 if (remote_pagesize_summary != local_pagesize_summary) { 1376 /* 1377 * This detects two potential causes of mismatch: 1378 * a) A mismatch in host page sizes 1379 * Some combinations of mismatch are probably possible but it gets 1380 * a bit more complicated. In particular we need to place whole 1381 * host pages on the dest at once, and we need to ensure that we 1382 * handle dirtying to make sure we never end up sending part of 1383 * a hostpage on it's own. 1384 * b) The use of different huge page sizes on source/destination 1385 * a more fine grain test is performed during RAM block migration 1386 * but this test here causes a nice early clear failure, and 1387 * also fails when passed to an older qemu that doesn't 1388 * do huge pages. 1389 */ 1390 error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64 1391 " d=%" PRIx64 ")", 1392 remote_pagesize_summary, local_pagesize_summary); 1393 return -1; 1394 } 1395 1396 remote_tps = qemu_get_be64(mis->from_src_file); 1397 if (remote_tps != qemu_target_page_size()) { 1398 /* 1399 * Again, some differences could be dealt with, but for now keep it 1400 * simple. 1401 */ 1402 error_report("Postcopy needs matching target page sizes (s=%d d=%zd)", 1403 (int)remote_tps, qemu_target_page_size()); 1404 return -1; 1405 } 1406 1407 if (ram_postcopy_incoming_init(mis)) { 1408 return -1; 1409 } 1410 1411 postcopy_state_set(POSTCOPY_INCOMING_ADVISE); 1412 1413 return 0; 1414 } 1415 1416 /* After postcopy we will be told to throw some pages away since they're 1417 * dirty and will have to be demand fetched. Must happen before CPU is 1418 * started. 1419 * There can be 0..many of these messages, each encoding multiple pages. 1420 */ 1421 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState *mis, 1422 uint16_t len) 1423 { 1424 int tmp; 1425 char ramid[256]; 1426 PostcopyState ps = postcopy_state_get(); 1427 1428 trace_loadvm_postcopy_ram_handle_discard(); 1429 1430 switch (ps) { 1431 case POSTCOPY_INCOMING_ADVISE: 1432 /* 1st discard */ 1433 tmp = postcopy_ram_prepare_discard(mis); 1434 if (tmp) { 1435 return tmp; 1436 } 1437 break; 1438 1439 case POSTCOPY_INCOMING_DISCARD: 1440 /* Expected state */ 1441 break; 1442 1443 default: 1444 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)", 1445 ps); 1446 return -1; 1447 } 1448 /* We're expecting a 1449 * Version (0) 1450 * a RAM ID string (length byte, name, 0 term) 1451 * then at least 1 16 byte chunk 1452 */ 1453 if (len < (1 + 1 + 1 + 1 + 2 * 8)) { 1454 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len); 1455 return -1; 1456 } 1457 1458 tmp = qemu_get_byte(mis->from_src_file); 1459 if (tmp != postcopy_ram_discard_version) { 1460 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp); 1461 return -1; 1462 } 1463 1464 if (!qemu_get_counted_string(mis->from_src_file, ramid)) { 1465 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID"); 1466 return -1; 1467 } 1468 tmp = qemu_get_byte(mis->from_src_file); 1469 if (tmp != 0) { 1470 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp); 1471 return -1; 1472 } 1473 1474 len -= 3 + strlen(ramid); 1475 if (len % 16) { 1476 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len); 1477 return -1; 1478 } 1479 trace_loadvm_postcopy_ram_handle_discard_header(ramid, len); 1480 while (len) { 1481 uint64_t start_addr, block_length; 1482 start_addr = qemu_get_be64(mis->from_src_file); 1483 block_length = qemu_get_be64(mis->from_src_file); 1484 1485 len -= 16; 1486 int ret = ram_discard_range(ramid, start_addr, block_length); 1487 if (ret) { 1488 return ret; 1489 } 1490 } 1491 trace_loadvm_postcopy_ram_handle_discard_end(); 1492 1493 return 0; 1494 } 1495 1496 /* 1497 * Triggered by a postcopy_listen command; this thread takes over reading 1498 * the input stream, leaving the main thread free to carry on loading the rest 1499 * of the device state (from RAM). 1500 * (TODO:This could do with being in a postcopy file - but there again it's 1501 * just another input loop, not that postcopy specific) 1502 */ 1503 static void *postcopy_ram_listen_thread(void *opaque) 1504 { 1505 QEMUFile *f = opaque; 1506 MigrationIncomingState *mis = migration_incoming_get_current(); 1507 int load_res; 1508 1509 migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE, 1510 MIGRATION_STATUS_POSTCOPY_ACTIVE); 1511 qemu_sem_post(&mis->listen_thread_sem); 1512 trace_postcopy_ram_listen_thread_start(); 1513 1514 /* 1515 * Because we're a thread and not a coroutine we can't yield 1516 * in qemu_file, and thus we must be blocking now. 1517 */ 1518 qemu_file_set_blocking(f, true); 1519 load_res = qemu_loadvm_state_main(f, mis); 1520 /* And non-blocking again so we don't block in any cleanup */ 1521 qemu_file_set_blocking(f, false); 1522 1523 trace_postcopy_ram_listen_thread_exit(); 1524 if (load_res < 0) { 1525 error_report("%s: loadvm failed: %d", __func__, load_res); 1526 qemu_file_set_error(f, load_res); 1527 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, 1528 MIGRATION_STATUS_FAILED); 1529 } else { 1530 /* 1531 * This looks good, but it's possible that the device loading in the 1532 * main thread hasn't finished yet, and so we might not be in 'RUN' 1533 * state yet; wait for the end of the main thread. 1534 */ 1535 qemu_event_wait(&mis->main_thread_load_event); 1536 } 1537 postcopy_ram_incoming_cleanup(mis); 1538 1539 if (load_res < 0) { 1540 /* 1541 * If something went wrong then we have a bad state so exit; 1542 * depending how far we got it might be possible at this point 1543 * to leave the guest running and fire MCEs for pages that never 1544 * arrived as a desperate recovery step. 1545 */ 1546 exit(EXIT_FAILURE); 1547 } 1548 1549 migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_ACTIVE, 1550 MIGRATION_STATUS_COMPLETED); 1551 /* 1552 * If everything has worked fine, then the main thread has waited 1553 * for us to start, and we're the last use of the mis. 1554 * (If something broke then qemu will have to exit anyway since it's 1555 * got a bad migration state). 1556 */ 1557 migration_incoming_state_destroy(); 1558 1559 1560 return NULL; 1561 } 1562 1563 /* After this message we must be able to immediately receive postcopy data */ 1564 static int loadvm_postcopy_handle_listen(MigrationIncomingState *mis) 1565 { 1566 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_LISTENING); 1567 trace_loadvm_postcopy_handle_listen(); 1568 if (ps != POSTCOPY_INCOMING_ADVISE && ps != POSTCOPY_INCOMING_DISCARD) { 1569 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps); 1570 return -1; 1571 } 1572 if (ps == POSTCOPY_INCOMING_ADVISE) { 1573 /* 1574 * A rare case, we entered listen without having to do any discards, 1575 * so do the setup that's normally done at the time of the 1st discard. 1576 */ 1577 postcopy_ram_prepare_discard(mis); 1578 } 1579 1580 /* 1581 * Sensitise RAM - can now generate requests for blocks that don't exist 1582 * However, at this point the CPU shouldn't be running, and the IO 1583 * shouldn't be doing anything yet so don't actually expect requests 1584 */ 1585 if (postcopy_ram_enable_notify(mis)) { 1586 return -1; 1587 } 1588 1589 if (mis->have_listen_thread) { 1590 error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread"); 1591 return -1; 1592 } 1593 1594 mis->have_listen_thread = true; 1595 /* Start up the listening thread and wait for it to signal ready */ 1596 qemu_sem_init(&mis->listen_thread_sem, 0); 1597 qemu_thread_create(&mis->listen_thread, "postcopy/listen", 1598 postcopy_ram_listen_thread, mis->from_src_file, 1599 QEMU_THREAD_DETACHED); 1600 qemu_sem_wait(&mis->listen_thread_sem); 1601 qemu_sem_destroy(&mis->listen_thread_sem); 1602 1603 return 0; 1604 } 1605 1606 1607 typedef struct { 1608 QEMUBH *bh; 1609 } HandleRunBhData; 1610 1611 static void loadvm_postcopy_handle_run_bh(void *opaque) 1612 { 1613 Error *local_err = NULL; 1614 HandleRunBhData *data = opaque; 1615 1616 /* TODO we should move all of this lot into postcopy_ram.c or a shared code 1617 * in migration.c 1618 */ 1619 cpu_synchronize_all_post_init(); 1620 1621 qemu_announce_self(); 1622 1623 /* Make sure all file formats flush their mutable metadata. 1624 * If we get an error here, just don't restart the VM yet. */ 1625 bdrv_invalidate_cache_all(&local_err); 1626 if (local_err) { 1627 error_report_err(local_err); 1628 local_err = NULL; 1629 autostart = false; 1630 } 1631 1632 trace_loadvm_postcopy_handle_run_cpu_sync(); 1633 cpu_synchronize_all_post_init(); 1634 1635 trace_loadvm_postcopy_handle_run_vmstart(); 1636 1637 if (autostart) { 1638 /* Hold onto your hats, starting the CPU */ 1639 vm_start(); 1640 } else { 1641 /* leave it paused and let management decide when to start the CPU */ 1642 runstate_set(RUN_STATE_PAUSED); 1643 } 1644 1645 qemu_bh_delete(data->bh); 1646 g_free(data); 1647 } 1648 1649 /* After all discards we can start running and asking for pages */ 1650 static int loadvm_postcopy_handle_run(MigrationIncomingState *mis) 1651 { 1652 PostcopyState ps = postcopy_state_set(POSTCOPY_INCOMING_RUNNING); 1653 HandleRunBhData *data; 1654 1655 trace_loadvm_postcopy_handle_run(); 1656 if (ps != POSTCOPY_INCOMING_LISTENING) { 1657 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps); 1658 return -1; 1659 } 1660 1661 data = g_new(HandleRunBhData, 1); 1662 data->bh = qemu_bh_new(loadvm_postcopy_handle_run_bh, data); 1663 qemu_bh_schedule(data->bh); 1664 1665 /* We need to finish reading the stream from the package 1666 * and also stop reading anything more from the stream that loaded the 1667 * package (since it's now being read by the listener thread). 1668 * LOADVM_QUIT will quit all the layers of nested loadvm loops. 1669 */ 1670 return LOADVM_QUIT; 1671 } 1672 1673 /** 1674 * Immediately following this command is a blob of data containing an embedded 1675 * chunk of migration stream; read it and load it. 1676 * 1677 * @mis: Incoming state 1678 * @length: Length of packaged data to read 1679 * 1680 * Returns: Negative values on error 1681 * 1682 */ 1683 static int loadvm_handle_cmd_packaged(MigrationIncomingState *mis) 1684 { 1685 int ret; 1686 size_t length; 1687 QIOChannelBuffer *bioc; 1688 1689 length = qemu_get_be32(mis->from_src_file); 1690 trace_loadvm_handle_cmd_packaged(length); 1691 1692 if (length > MAX_VM_CMD_PACKAGED_SIZE) { 1693 error_report("Unreasonably large packaged state: %zu", length); 1694 return -1; 1695 } 1696 1697 bioc = qio_channel_buffer_new(length); 1698 qio_channel_set_name(QIO_CHANNEL(bioc), "migration-loadvm-buffer"); 1699 ret = qemu_get_buffer(mis->from_src_file, 1700 bioc->data, 1701 length); 1702 if (ret != length) { 1703 object_unref(OBJECT(bioc)); 1704 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu", 1705 ret, length); 1706 return (ret < 0) ? ret : -EAGAIN; 1707 } 1708 bioc->usage += length; 1709 trace_loadvm_handle_cmd_packaged_received(ret); 1710 1711 QEMUFile *packf = qemu_fopen_channel_input(QIO_CHANNEL(bioc)); 1712 1713 ret = qemu_loadvm_state_main(packf, mis); 1714 trace_loadvm_handle_cmd_packaged_main(ret); 1715 qemu_fclose(packf); 1716 object_unref(OBJECT(bioc)); 1717 1718 return ret; 1719 } 1720 1721 /* 1722 * Process an incoming 'QEMU_VM_COMMAND' 1723 * 0 just a normal return 1724 * LOADVM_QUIT All good, but exit the loop 1725 * <0 Error 1726 */ 1727 static int loadvm_process_command(QEMUFile *f) 1728 { 1729 MigrationIncomingState *mis = migration_incoming_get_current(); 1730 uint16_t cmd; 1731 uint16_t len; 1732 uint32_t tmp32; 1733 1734 cmd = qemu_get_be16(f); 1735 len = qemu_get_be16(f); 1736 1737 trace_loadvm_process_command(cmd, len); 1738 if (cmd >= MIG_CMD_MAX || cmd == MIG_CMD_INVALID) { 1739 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd, len); 1740 return -EINVAL; 1741 } 1742 1743 if (mig_cmd_args[cmd].len != -1 && mig_cmd_args[cmd].len != len) { 1744 error_report("%s received with bad length - expecting %zu, got %d", 1745 mig_cmd_args[cmd].name, 1746 (size_t)mig_cmd_args[cmd].len, len); 1747 return -ERANGE; 1748 } 1749 1750 switch (cmd) { 1751 case MIG_CMD_OPEN_RETURN_PATH: 1752 if (mis->to_src_file) { 1753 error_report("CMD_OPEN_RETURN_PATH called when RP already open"); 1754 /* Not really a problem, so don't give up */ 1755 return 0; 1756 } 1757 mis->to_src_file = qemu_file_get_return_path(f); 1758 if (!mis->to_src_file) { 1759 error_report("CMD_OPEN_RETURN_PATH failed"); 1760 return -1; 1761 } 1762 break; 1763 1764 case MIG_CMD_PING: 1765 tmp32 = qemu_get_be32(f); 1766 trace_loadvm_process_command_ping(tmp32); 1767 if (!mis->to_src_file) { 1768 error_report("CMD_PING (0x%x) received with no return path", 1769 tmp32); 1770 return -1; 1771 } 1772 migrate_send_rp_pong(mis, tmp32); 1773 break; 1774 1775 case MIG_CMD_PACKAGED: 1776 return loadvm_handle_cmd_packaged(mis); 1777 1778 case MIG_CMD_POSTCOPY_ADVISE: 1779 return loadvm_postcopy_handle_advise(mis); 1780 1781 case MIG_CMD_POSTCOPY_LISTEN: 1782 return loadvm_postcopy_handle_listen(mis); 1783 1784 case MIG_CMD_POSTCOPY_RUN: 1785 return loadvm_postcopy_handle_run(mis); 1786 1787 case MIG_CMD_POSTCOPY_RAM_DISCARD: 1788 return loadvm_postcopy_ram_handle_discard(mis, len); 1789 } 1790 1791 return 0; 1792 } 1793 1794 /* 1795 * Read a footer off the wire and check that it matches the expected section 1796 * 1797 * Returns: true if the footer was good 1798 * false if there is a problem (and calls error_report to say why) 1799 */ 1800 static bool check_section_footer(QEMUFile *f, SaveStateEntry *se) 1801 { 1802 uint8_t read_mark; 1803 uint32_t read_section_id; 1804 1805 if (skip_section_footers) { 1806 /* No footer to check */ 1807 return true; 1808 } 1809 1810 read_mark = qemu_get_byte(f); 1811 1812 if (read_mark != QEMU_VM_SECTION_FOOTER) { 1813 error_report("Missing section footer for %s", se->idstr); 1814 return false; 1815 } 1816 1817 read_section_id = qemu_get_be32(f); 1818 if (read_section_id != se->load_section_id) { 1819 error_report("Mismatched section id in footer for %s -" 1820 " read 0x%x expected 0x%x", 1821 se->idstr, read_section_id, se->load_section_id); 1822 return false; 1823 } 1824 1825 /* All good */ 1826 return true; 1827 } 1828 1829 static int 1830 qemu_loadvm_section_start_full(QEMUFile *f, MigrationIncomingState *mis) 1831 { 1832 uint32_t instance_id, version_id, section_id; 1833 SaveStateEntry *se; 1834 char idstr[256]; 1835 int ret; 1836 1837 /* Read section start */ 1838 section_id = qemu_get_be32(f); 1839 if (!qemu_get_counted_string(f, idstr)) { 1840 error_report("Unable to read ID string for section %u", 1841 section_id); 1842 return -EINVAL; 1843 } 1844 instance_id = qemu_get_be32(f); 1845 version_id = qemu_get_be32(f); 1846 1847 trace_qemu_loadvm_state_section_startfull(section_id, idstr, 1848 instance_id, version_id); 1849 /* Find savevm section */ 1850 se = find_se(idstr, instance_id); 1851 if (se == NULL) { 1852 error_report("Unknown savevm section or instance '%s' %d", 1853 idstr, instance_id); 1854 return -EINVAL; 1855 } 1856 1857 /* Validate version */ 1858 if (version_id > se->version_id) { 1859 error_report("savevm: unsupported version %d for '%s' v%d", 1860 version_id, idstr, se->version_id); 1861 return -EINVAL; 1862 } 1863 se->load_version_id = version_id; 1864 se->load_section_id = section_id; 1865 1866 /* Validate if it is a device's state */ 1867 if (xen_enabled() && se->is_ram) { 1868 error_report("loadvm: %s RAM loading not allowed on Xen", idstr); 1869 return -EINVAL; 1870 } 1871 1872 ret = vmstate_load(f, se); 1873 if (ret < 0) { 1874 error_report("error while loading state for instance 0x%x of" 1875 " device '%s'", instance_id, idstr); 1876 return ret; 1877 } 1878 if (!check_section_footer(f, se)) { 1879 return -EINVAL; 1880 } 1881 1882 return 0; 1883 } 1884 1885 static int 1886 qemu_loadvm_section_part_end(QEMUFile *f, MigrationIncomingState *mis) 1887 { 1888 uint32_t section_id; 1889 SaveStateEntry *se; 1890 int ret; 1891 1892 section_id = qemu_get_be32(f); 1893 1894 trace_qemu_loadvm_state_section_partend(section_id); 1895 QTAILQ_FOREACH(se, &savevm_state.handlers, entry) { 1896 if (se->load_section_id == section_id) { 1897 break; 1898 } 1899 } 1900 if (se == NULL) { 1901 error_report("Unknown savevm section %d", section_id); 1902 return -EINVAL; 1903 } 1904 1905 ret = vmstate_load(f, se); 1906 if (ret < 0) { 1907 error_report("error while loading state section id %d(%s)", 1908 section_id, se->idstr); 1909 return ret; 1910 } 1911 if (!check_section_footer(f, se)) { 1912 return -EINVAL; 1913 } 1914 1915 return 0; 1916 } 1917 1918 static int qemu_loadvm_state_main(QEMUFile *f, MigrationIncomingState *mis) 1919 { 1920 uint8_t section_type; 1921 int ret = 0; 1922 1923 while ((section_type = qemu_get_byte(f)) != QEMU_VM_EOF) { 1924 ret = 0; 1925 trace_qemu_loadvm_state_section(section_type); 1926 switch (section_type) { 1927 case QEMU_VM_SECTION_START: 1928 case QEMU_VM_SECTION_FULL: 1929 ret = qemu_loadvm_section_start_full(f, mis); 1930 if (ret < 0) { 1931 goto out; 1932 } 1933 break; 1934 case QEMU_VM_SECTION_PART: 1935 case QEMU_VM_SECTION_END: 1936 ret = qemu_loadvm_section_part_end(f, mis); 1937 if (ret < 0) { 1938 goto out; 1939 } 1940 break; 1941 case QEMU_VM_COMMAND: 1942 ret = loadvm_process_command(f); 1943 trace_qemu_loadvm_state_section_command(ret); 1944 if ((ret < 0) || (ret & LOADVM_QUIT)) { 1945 goto out; 1946 } 1947 break; 1948 default: 1949 error_report("Unknown savevm section type %d", section_type); 1950 ret = -EINVAL; 1951 goto out; 1952 } 1953 } 1954 1955 out: 1956 if (ret < 0) { 1957 qemu_file_set_error(f, ret); 1958 } 1959 return ret; 1960 } 1961 1962 int qemu_loadvm_state(QEMUFile *f) 1963 { 1964 MigrationIncomingState *mis = migration_incoming_get_current(); 1965 Error *local_err = NULL; 1966 unsigned int v; 1967 int ret; 1968 1969 if (qemu_savevm_state_blocked(&local_err)) { 1970 error_report_err(local_err); 1971 return -EINVAL; 1972 } 1973 1974 v = qemu_get_be32(f); 1975 if (v != QEMU_VM_FILE_MAGIC) { 1976 error_report("Not a migration stream"); 1977 return -EINVAL; 1978 } 1979 1980 v = qemu_get_be32(f); 1981 if (v == QEMU_VM_FILE_VERSION_COMPAT) { 1982 error_report("SaveVM v2 format is obsolete and don't work anymore"); 1983 return -ENOTSUP; 1984 } 1985 if (v != QEMU_VM_FILE_VERSION) { 1986 error_report("Unsupported migration stream version"); 1987 return -ENOTSUP; 1988 } 1989 1990 if (!savevm_state.skip_configuration || enforce_config_section()) { 1991 if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) { 1992 error_report("Configuration section missing"); 1993 return -EINVAL; 1994 } 1995 ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0); 1996 1997 if (ret) { 1998 return ret; 1999 } 2000 } 2001 2002 cpu_synchronize_all_pre_loadvm(); 2003 2004 ret = qemu_loadvm_state_main(f, mis); 2005 qemu_event_set(&mis->main_thread_load_event); 2006 2007 trace_qemu_loadvm_state_post_main(ret); 2008 2009 if (mis->have_listen_thread) { 2010 /* Listen thread still going, can't clean up yet */ 2011 return ret; 2012 } 2013 2014 if (ret == 0) { 2015 ret = qemu_file_get_error(f); 2016 } 2017 2018 /* 2019 * Try to read in the VMDESC section as well, so that dumping tools that 2020 * intercept our migration stream have the chance to see it. 2021 */ 2022 2023 /* We've got to be careful; if we don't read the data and just shut the fd 2024 * then the sender can error if we close while it's still sending. 2025 * We also mustn't read data that isn't there; some transports (RDMA) 2026 * will stall waiting for that data when the source has already closed. 2027 */ 2028 if (ret == 0 && should_send_vmdesc()) { 2029 uint8_t *buf; 2030 uint32_t size; 2031 uint8_t section_type = qemu_get_byte(f); 2032 2033 if (section_type != QEMU_VM_VMDESCRIPTION) { 2034 error_report("Expected vmdescription section, but got %d", 2035 section_type); 2036 /* 2037 * It doesn't seem worth failing at this point since 2038 * we apparently have an otherwise valid VM state 2039 */ 2040 } else { 2041 buf = g_malloc(0x1000); 2042 size = qemu_get_be32(f); 2043 2044 while (size > 0) { 2045 uint32_t read_chunk = MIN(size, 0x1000); 2046 qemu_get_buffer(f, buf, read_chunk); 2047 size -= read_chunk; 2048 } 2049 g_free(buf); 2050 } 2051 } 2052 2053 cpu_synchronize_all_post_init(); 2054 2055 return ret; 2056 } 2057 2058 int save_snapshot(const char *name, Error **errp) 2059 { 2060 BlockDriverState *bs, *bs1; 2061 QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1; 2062 int ret = -1; 2063 QEMUFile *f; 2064 int saved_vm_running; 2065 uint64_t vm_state_size; 2066 qemu_timeval tv; 2067 struct tm tm; 2068 AioContext *aio_context; 2069 2070 if (!bdrv_all_can_snapshot(&bs)) { 2071 error_setg(errp, "Device '%s' is writable but does not support " 2072 "snapshots", bdrv_get_device_name(bs)); 2073 return ret; 2074 } 2075 2076 /* Delete old snapshots of the same name */ 2077 if (name) { 2078 ret = bdrv_all_delete_snapshot(name, &bs1, errp); 2079 if (ret < 0) { 2080 error_prepend(errp, "Error while deleting snapshot on device " 2081 "'%s': ", bdrv_get_device_name(bs1)); 2082 return ret; 2083 } 2084 } 2085 2086 bs = bdrv_all_find_vmstate_bs(); 2087 if (bs == NULL) { 2088 error_setg(errp, "No block device can accept snapshots"); 2089 return ret; 2090 } 2091 aio_context = bdrv_get_aio_context(bs); 2092 2093 saved_vm_running = runstate_is_running(); 2094 2095 ret = global_state_store(); 2096 if (ret) { 2097 error_setg(errp, "Error saving global state"); 2098 return ret; 2099 } 2100 vm_stop(RUN_STATE_SAVE_VM); 2101 2102 aio_context_acquire(aio_context); 2103 2104 memset(sn, 0, sizeof(*sn)); 2105 2106 /* fill auxiliary fields */ 2107 qemu_gettimeofday(&tv); 2108 sn->date_sec = tv.tv_sec; 2109 sn->date_nsec = tv.tv_usec * 1000; 2110 sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL); 2111 2112 if (name) { 2113 ret = bdrv_snapshot_find(bs, old_sn, name); 2114 if (ret >= 0) { 2115 pstrcpy(sn->name, sizeof(sn->name), old_sn->name); 2116 pstrcpy(sn->id_str, sizeof(sn->id_str), old_sn->id_str); 2117 } else { 2118 pstrcpy(sn->name, sizeof(sn->name), name); 2119 } 2120 } else { 2121 /* cast below needed for OpenBSD where tv_sec is still 'long' */ 2122 localtime_r((const time_t *)&tv.tv_sec, &tm); 2123 strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm); 2124 } 2125 2126 /* save the VM state */ 2127 f = qemu_fopen_bdrv(bs, 1); 2128 if (!f) { 2129 error_setg(errp, "Could not open VM state file"); 2130 goto the_end; 2131 } 2132 ret = qemu_savevm_state(f, errp); 2133 vm_state_size = qemu_ftell(f); 2134 qemu_fclose(f); 2135 if (ret < 0) { 2136 goto the_end; 2137 } 2138 2139 ret = bdrv_all_create_snapshot(sn, bs, vm_state_size, &bs); 2140 if (ret < 0) { 2141 error_setg(errp, "Error while creating snapshot on '%s'", 2142 bdrv_get_device_name(bs)); 2143 goto the_end; 2144 } 2145 2146 ret = 0; 2147 2148 the_end: 2149 aio_context_release(aio_context); 2150 if (saved_vm_running) { 2151 vm_start(); 2152 } 2153 return ret; 2154 } 2155 2156 void qmp_xen_save_devices_state(const char *filename, Error **errp) 2157 { 2158 QEMUFile *f; 2159 QIOChannelFile *ioc; 2160 int saved_vm_running; 2161 int ret; 2162 2163 saved_vm_running = runstate_is_running(); 2164 vm_stop(RUN_STATE_SAVE_VM); 2165 global_state_store_running(); 2166 2167 ioc = qio_channel_file_new_path(filename, O_WRONLY | O_CREAT, 0660, errp); 2168 if (!ioc) { 2169 goto the_end; 2170 } 2171 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-save-state"); 2172 f = qemu_fopen_channel_output(QIO_CHANNEL(ioc)); 2173 ret = qemu_save_device_state(f); 2174 qemu_fclose(f); 2175 if (ret < 0) { 2176 error_setg(errp, QERR_IO_ERROR); 2177 } 2178 2179 the_end: 2180 if (saved_vm_running) { 2181 vm_start(); 2182 } 2183 } 2184 2185 void qmp_xen_load_devices_state(const char *filename, Error **errp) 2186 { 2187 QEMUFile *f; 2188 QIOChannelFile *ioc; 2189 int ret; 2190 2191 /* Guest must be paused before loading the device state; the RAM state 2192 * will already have been loaded by xc 2193 */ 2194 if (runstate_is_running()) { 2195 error_setg(errp, "Cannot update device state while vm is running"); 2196 return; 2197 } 2198 vm_stop(RUN_STATE_RESTORE_VM); 2199 2200 ioc = qio_channel_file_new_path(filename, O_RDONLY | O_BINARY, 0, errp); 2201 if (!ioc) { 2202 return; 2203 } 2204 qio_channel_set_name(QIO_CHANNEL(ioc), "migration-xen-load-state"); 2205 f = qemu_fopen_channel_input(QIO_CHANNEL(ioc)); 2206 2207 ret = qemu_loadvm_state(f); 2208 qemu_fclose(f); 2209 if (ret < 0) { 2210 error_setg(errp, QERR_IO_ERROR); 2211 } 2212 migration_incoming_state_destroy(); 2213 } 2214 2215 int load_snapshot(const char *name, Error **errp) 2216 { 2217 BlockDriverState *bs, *bs_vm_state; 2218 QEMUSnapshotInfo sn; 2219 QEMUFile *f; 2220 int ret; 2221 AioContext *aio_context; 2222 MigrationIncomingState *mis = migration_incoming_get_current(); 2223 2224 if (!bdrv_all_can_snapshot(&bs)) { 2225 error_setg(errp, 2226 "Device '%s' is writable but does not support snapshots", 2227 bdrv_get_device_name(bs)); 2228 return -ENOTSUP; 2229 } 2230 ret = bdrv_all_find_snapshot(name, &bs); 2231 if (ret < 0) { 2232 error_setg(errp, 2233 "Device '%s' does not have the requested snapshot '%s'", 2234 bdrv_get_device_name(bs), name); 2235 return ret; 2236 } 2237 2238 bs_vm_state = bdrv_all_find_vmstate_bs(); 2239 if (!bs_vm_state) { 2240 error_setg(errp, "No block device supports snapshots"); 2241 return -ENOTSUP; 2242 } 2243 aio_context = bdrv_get_aio_context(bs_vm_state); 2244 2245 /* Don't even try to load empty VM states */ 2246 aio_context_acquire(aio_context); 2247 ret = bdrv_snapshot_find(bs_vm_state, &sn, name); 2248 aio_context_release(aio_context); 2249 if (ret < 0) { 2250 return ret; 2251 } else if (sn.vm_state_size == 0) { 2252 error_setg(errp, "This is a disk-only snapshot. Revert to it " 2253 " offline using qemu-img"); 2254 return -EINVAL; 2255 } 2256 2257 /* Flush all IO requests so they don't interfere with the new state. */ 2258 bdrv_drain_all(); 2259 2260 ret = bdrv_all_goto_snapshot(name, &bs); 2261 if (ret < 0) { 2262 error_setg(errp, "Error %d while activating snapshot '%s' on '%s'", 2263 ret, name, bdrv_get_device_name(bs)); 2264 return ret; 2265 } 2266 2267 /* restore the VM state */ 2268 f = qemu_fopen_bdrv(bs_vm_state, 0); 2269 if (!f) { 2270 error_setg(errp, "Could not open VM state file"); 2271 return -EINVAL; 2272 } 2273 2274 qemu_system_reset(SHUTDOWN_CAUSE_NONE); 2275 mis->from_src_file = f; 2276 2277 aio_context_acquire(aio_context); 2278 ret = qemu_loadvm_state(f); 2279 qemu_fclose(f); 2280 aio_context_release(aio_context); 2281 2282 migration_incoming_state_destroy(); 2283 if (ret < 0) { 2284 error_setg(errp, "Error %d while loading VM state", ret); 2285 return ret; 2286 } 2287 2288 return 0; 2289 } 2290 2291 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev) 2292 { 2293 qemu_ram_set_idstr(mr->ram_block, 2294 memory_region_name(mr), dev); 2295 } 2296 2297 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev) 2298 { 2299 qemu_ram_unset_idstr(mr->ram_block); 2300 } 2301 2302 void vmstate_register_ram_global(MemoryRegion *mr) 2303 { 2304 vmstate_register_ram(mr, NULL); 2305 } 2306 2307 bool vmstate_check_only_migratable(const VMStateDescription *vmsd) 2308 { 2309 /* check needed if --only-migratable is specified */ 2310 if (!only_migratable) { 2311 return true; 2312 } 2313 2314 return !(vmsd && vmsd->unmigratable); 2315 } 2316