migration/savevm.c

   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 <[email protected]>
   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 "config-host.h"
  30 #include "qemu-common.h"
  31 #include "hw/boards.h"
  32 #include "hw/hw.h"
  33 #include "hw/qdev.h"
  34 #include "net/net.h"
  35 #include "monitor/monitor.h"
  36 #include "sysemu/sysemu.h"
  37 #include "qemu/timer.h"
  38 #include "audio/audio.h"
  39 #include "migration/migration.h"
  40 #include "qapi/qmp/qerror.h"
  41 #include "qemu/error-report.h"
  42 #include "qemu/sockets.h"
  43 #include "qemu/queue.h"
  44 #include "sysemu/cpus.h"
  45 #include "exec/memory.h"
  46 #include "qmp-commands.h"
  47 #include "trace.h"
  48 #include "qemu/iov.h"
  49 #include "block/snapshot.h"
  50 #include "block/qapi.h"
  51
  52
  53 #ifndef ETH_P_RARP
  54 #define ETH_P_RARP 0x8035
  55 #endif
  56 #define ARP_HTYPE_ETH 0x0001
  57 #define ARP_PTYPE_IP 0x0800
  58 #define ARP_OP_REQUEST_REV 0x3
  59
  60 static bool skip_section_footers;
  61
  62 static int announce_self_create(uint8_t *buf,
  63                                 uint8_t *mac_addr)
  64 {
  65     /* Ethernet header. */
  66     memset(buf, 0xff, 6);         /* destination MAC addr */
  67     memcpy(buf + 6, mac_addr, 6); /* source MAC addr */
  68     *(uint16_t *)(buf + 12) = htons(ETH_P_RARP); /* ethertype */
  69
  70     /* RARP header. */
  71     *(uint16_t *)(buf + 14) = htons(ARP_HTYPE_ETH); /* hardware addr space */
  72     *(uint16_t *)(buf + 16) = htons(ARP_PTYPE_IP); /* protocol addr space */
  73     *(buf + 18) = 6; /* hardware addr length (ethernet) */
  74     *(buf + 19) = 4; /* protocol addr length (IPv4) */
  75     *(uint16_t *)(buf + 20) = htons(ARP_OP_REQUEST_REV); /* opcode */
  76     memcpy(buf + 22, mac_addr, 6); /* source hw addr */
  77     memset(buf + 28, 0x00, 4);     /* source protocol addr */
  78     memcpy(buf + 32, mac_addr, 6); /* target hw addr */
  79     memset(buf + 38, 0x00, 4);     /* target protocol addr */
  80
  81     /* Padding to get up to 60 bytes (ethernet min packet size, minus FCS). */
  82     memset(buf + 42, 0x00, 18);
  83
  84     return 60; /* len (FCS will be added by hardware) */
  85 }
  86
  87 static void qemu_announce_self_iter(NICState *nic, void *opaque)
  88 {
  89     uint8_t buf[60];
  90     int len;
  91
  92     trace_qemu_announce_self_iter(qemu_ether_ntoa(&nic->conf->macaddr));
  93     len = announce_self_create(buf, nic->conf->macaddr.a);
  94
  95     qemu_send_packet_raw(qemu_get_queue(nic), buf, len);
  96 }
  97
  98
  99 static void qemu_announce_self_once(void *opaque)
 100 {
 101     static int count = SELF_ANNOUNCE_ROUNDS;
 102     QEMUTimer *timer = *(QEMUTimer **)opaque;
 103
 104     qemu_foreach_nic(qemu_announce_self_iter, NULL);
 105
 106     if (--count) {
 107         /* delay 50ms, 150ms, 250ms, ... */
 108         timer_mod(timer, qemu_clock_get_ms(QEMU_CLOCK_REALTIME) +
 109                   self_announce_delay(count));
 110     } else {
 111             timer_del(timer);
 112             timer_free(timer);
 113     }
 114 }
 115
 116 void qemu_announce_self(void)
 117 {
 118     static QEMUTimer *timer;
 119     timer = timer_new_ms(QEMU_CLOCK_REALTIME, qemu_announce_self_once, &timer);
 120     qemu_announce_self_once(&timer);
 121 }
 122
 123 /***********************************************************/
 124 /* savevm/loadvm support */
 125
 126 static ssize_t block_writev_buffer(void *opaque, struct iovec *iov, int iovcnt,
 127                                    int64_t pos)
 128 {
 129     int ret;
 130     QEMUIOVector qiov;
 131
 132     qemu_iovec_init_external(&qiov, iov, iovcnt);
 133     ret = bdrv_writev_vmstate(opaque, &qiov, pos);
 134     if (ret < 0) {
 135         return ret;
 136     }
 137
 138     return qiov.size;
 139 }
 140
 141 static int block_put_buffer(void *opaque, const uint8_t *buf,
 142                            int64_t pos, int size)
 143 {
 144     bdrv_save_vmstate(opaque, buf, pos, size);
 145     return size;
 146 }
 147
 148 static int block_get_buffer(void *opaque, uint8_t *buf, int64_t pos, int size)
 149 {
 150     return bdrv_load_vmstate(opaque, buf, pos, size);
 151 }
 152
 153 static int bdrv_fclose(void *opaque)
 154 {
 155     return bdrv_flush(opaque);
 156 }
 157
 158 static const QEMUFileOps bdrv_read_ops = {
 159     .get_buffer = block_get_buffer,
 160     .close =      bdrv_fclose
 161 };
 162
 163 static const QEMUFileOps bdrv_write_ops = {
 164     .put_buffer     = block_put_buffer,
 165     .writev_buffer  = block_writev_buffer,
 166     .close          = bdrv_fclose
 167 };
 168
 169 static QEMUFile *qemu_fopen_bdrv(BlockDriverState *bs, int is_writable)
 170 {
 171     if (is_writable) {
 172         return qemu_fopen_ops(bs, &bdrv_write_ops);
 173     }
 174     return qemu_fopen_ops(bs, &bdrv_read_ops);
 175 }
 176
 177
 178 /* QEMUFile timer support.
 179  * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
 180  */
 181
 182 void timer_put(QEMUFile *f, QEMUTimer *ts)
 183 {
 184     uint64_t expire_time;
 185
 186     expire_time = timer_expire_time_ns(ts);
 187     qemu_put_be64(f, expire_time);
 188 }
 189
 190 void timer_get(QEMUFile *f, QEMUTimer *ts)
 191 {
 192     uint64_t expire_time;
 193
 194     expire_time = qemu_get_be64(f);
 195     if (expire_time != -1) {
 196         timer_mod_ns(ts, expire_time);
 197     } else {
 198         timer_del(ts);
 199     }
 200 }
 201
 202
 203 /* VMState timer support.
 204  * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
 205  */
 206
 207 static int get_timer(QEMUFile *f, void *pv, size_t size)
 208 {
 209     QEMUTimer *v = pv;
 210     timer_get(f, v);
 211     return 0;
 212 }
 213
 214 static void put_timer(QEMUFile *f, void *pv, size_t size)
 215 {
 216     QEMUTimer *v = pv;
 217     timer_put(f, v);
 218 }
 219
 220 const VMStateInfo vmstate_info_timer = {
 221     .name = "timer",
 222     .get  = get_timer,
 223     .put  = put_timer,
 224 };
 225
 226
 227 typedef struct CompatEntry {
 228     char idstr[256];
 229     int instance_id;
 230 } CompatEntry;
 231
 232 typedef struct SaveStateEntry {
 233     QTAILQ_ENTRY(SaveStateEntry) entry;
 234     char idstr[256];
 235     int instance_id;
 236     int alias_id;
 237     int version_id;
 238     int section_id;
 239     SaveVMHandlers *ops;
 240     const VMStateDescription *vmsd;
 241     void *opaque;
 242     CompatEntry *compat;
 243     int is_ram;
 244 } SaveStateEntry;
 245
 246 typedef struct SaveState {
 247     QTAILQ_HEAD(, SaveStateEntry) handlers;
 248     int global_section_id;
 249     bool skip_configuration;
 250     uint32_t len;
 251     const char *name;
 252 } SaveState;
 253
 254 static SaveState savevm_state = {
 255     .handlers = QTAILQ_HEAD_INITIALIZER(savevm_state.handlers),
 256     .global_section_id = 0,
 257     .skip_configuration = false,
 258 };
 259
 260 void savevm_skip_configuration(void)
 261 {
 262     savevm_state.skip_configuration = true;
 263 }
 264
 265
 266 static void configuration_pre_save(void *opaque)
 267 {
 268     SaveState *state = opaque;
 269     const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
 270
 271     state->len = strlen(current_name);
 272     state->name = current_name;
 273 }
 274
 275 static int configuration_post_load(void *opaque, int version_id)
 276 {
 277     SaveState *state = opaque;
 278     const char *current_name = MACHINE_GET_CLASS(current_machine)->name;
 279
 280     if (strncmp(state->name, current_name, state->len) != 0) {
 281         error_report("Machine type received is '%s' and local is '%s'",
 282                      state->name, current_name);
 283         return -EINVAL;
 284     }
 285     return 0;
 286 }
 287
 288 static const VMStateDescription vmstate_configuration = {
 289     .name = "configuration",
 290     .version_id = 1,
 291     .post_load = configuration_post_load,
 292     .pre_save = configuration_pre_save,
 293     .fields = (VMStateField[]) {
 294         VMSTATE_UINT32(len, SaveState),
 295         VMSTATE_VBUFFER_ALLOC_UINT32(name, SaveState, 0, NULL, 0, len),
 296         VMSTATE_END_OF_LIST()
 297     },
 298 };
 299
 300 static void dump_vmstate_vmsd(FILE *out_file,
 301                               const VMStateDescription *vmsd, int indent,
 302                               bool is_subsection);
 303
 304 static void dump_vmstate_vmsf(FILE *out_file, const VMStateField *field,
 305                               int indent)
 306 {
 307     fprintf(out_file, "%*s{\n", indent, "");
 308     indent += 2;
 309     fprintf(out_file, "%*s\"field\": \"%s\",\n", indent, "", field->name);
 310     fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
 311             field->version_id);
 312     fprintf(out_file, "%*s\"field_exists\": %s,\n", indent, "",
 313             field->field_exists ? "true" : "false");
 314     fprintf(out_file, "%*s\"size\": %zu", indent, "", field->size);
 315     if (field->vmsd != NULL) {
 316         fprintf(out_file, ",\n");
 317         dump_vmstate_vmsd(out_file, field->vmsd, indent, false);
 318     }
 319     fprintf(out_file, "\n%*s}", indent - 2, "");
 320 }
 321
 322 static void dump_vmstate_vmss(FILE *out_file,
 323                               const VMStateDescription **subsection,
 324                               int indent)
 325 {
 326     if (*subsection != NULL) {
 327         dump_vmstate_vmsd(out_file, *subsection, indent, true);
 328     }
 329 }
 330
 331 static void dump_vmstate_vmsd(FILE *out_file,
 332                               const VMStateDescription *vmsd, int indent,
 333                               bool is_subsection)
 334 {
 335     if (is_subsection) {
 336         fprintf(out_file, "%*s{\n", indent, "");
 337     } else {
 338         fprintf(out_file, "%*s\"%s\": {\n", indent, "", "Description");
 339     }
 340     indent += 2;
 341     fprintf(out_file, "%*s\"name\": \"%s\",\n", indent, "", vmsd->name);
 342     fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
 343             vmsd->version_id);
 344     fprintf(out_file, "%*s\"minimum_version_id\": %d", indent, "",
 345             vmsd->minimum_version_id);
 346     if (vmsd->fields != NULL) {
 347         const VMStateField *field = vmsd->fields;
 348         bool first;
 349
 350         fprintf(out_file, ",\n%*s\"Fields\": [\n", indent, "");
 351         first = true;
 352         while (field->name != NULL) {
 353             if (field->flags & VMS_MUST_EXIST) {
 354                 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
 355                 field++;
 356                 continue;
 357             }
 358             if (!first) {
 359                 fprintf(out_file, ",\n");
 360             }
 361             dump_vmstate_vmsf(out_file, field, indent + 2);
 362             field++;
 363             first = false;
 364         }
 365         fprintf(out_file, "\n%*s]", indent, "");
 366     }
 367     if (vmsd->subsections != NULL) {
 368         const VMStateDescription **subsection = vmsd->subsections;
 369         bool first;
 370
 371         fprintf(out_file, ",\n%*s\"Subsections\": [\n", indent, "");
 372         first = true;
 373         while (*subsection != NULL) {
 374             if (!first) {
 375                 fprintf(out_file, ",\n");
 376             }
 377             dump_vmstate_vmss(out_file, subsection, indent + 2);
 378             subsection++;
 379             first = false;
 380         }
 381         fprintf(out_file, "\n%*s]", indent, "");
 382     }
 383     fprintf(out_file, "\n%*s}", indent - 2, "");
 384 }
 385
 386 static void dump_machine_type(FILE *out_file)
 387 {
 388     MachineClass *mc;
 389
 390     mc = MACHINE_GET_CLASS(current_machine);
 391
 392     fprintf(out_file, "  \"vmschkmachine\": {\n");
 393     fprintf(out_file, "    \"Name\": \"%s\"\n", mc->name);
 394     fprintf(out_file, "  },\n");
 395 }
 396
 397 void dump_vmstate_json_to_file(FILE *out_file)
 398 {
 399     GSList *list, *elt;
 400     bool first;
 401
 402     fprintf(out_file, "{\n");
 403     dump_machine_type(out_file);
 404
 405     first = true;
 406     list = object_class_get_list(TYPE_DEVICE, true);
 407     for (elt = list; elt; elt = elt->next) {
 408         DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, elt->data,
 409                                              TYPE_DEVICE);
 410         const char *name;
 411         int indent = 2;
 412
 413         if (!dc->vmsd) {
 414             continue;
 415         }
 416
 417         if (!first) {
 418             fprintf(out_file, ",\n");
 419         }
 420         name = object_class_get_name(OBJECT_CLASS(dc));
 421         fprintf(out_file, "%*s\"%s\": {\n", indent, "", name);
 422         indent += 2;
 423         fprintf(out_file, "%*s\"Name\": \"%s\",\n", indent, "", name);
 424         fprintf(out_file, "%*s\"version_id\": %d,\n", indent, "",
 425                 dc->vmsd->version_id);
 426         fprintf(out_file, "%*s\"minimum_version_id\": %d,\n", indent, "",
 427                 dc->vmsd->minimum_version_id);
 428
 429         dump_vmstate_vmsd(out_file, dc->vmsd, indent, false);
 430
 431         fprintf(out_file, "\n%*s}", indent - 2, "");
 432         first = false;
 433     }
 434     fprintf(out_file, "\n}\n");
 435     fclose(out_file);
 436 }
 437
 438 static int calculate_new_instance_id(const char *idstr)
 439 {
 440     SaveStateEntry *se;
 441     int instance_id = 0;
 442
 443     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
 444         if (strcmp(idstr, se->idstr) == 0
 445             && instance_id <= se->instance_id) {
 446             instance_id = se->instance_id + 1;
 447         }
 448     }
 449     return instance_id;
 450 }
 451
 452 static int calculate_compat_instance_id(const char *idstr)
 453 {
 454     SaveStateEntry *se;
 455     int instance_id = 0;
 456
 457     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
 458         if (!se->compat) {
 459             continue;
 460         }
 461
 462         if (strcmp(idstr, se->compat->idstr) == 0
 463             && instance_id <= se->compat->instance_id) {
 464             instance_id = se->compat->instance_id + 1;
 465         }
 466     }
 467     return instance_id;
 468 }
 469
 470 /* TODO: Individual devices generally have very little idea about the rest
 471    of the system, so instance_id should be removed/replaced.
 472    Meanwhile pass -1 as instance_id if you do not already have a clearly
 473    distinguishing id for all instances of your device class. */
 474 int register_savevm_live(DeviceState *dev,
 475                          const char *idstr,
 476                          int instance_id,
 477                          int version_id,
 478                          SaveVMHandlers *ops,
 479                          void *opaque)
 480 {
 481     SaveStateEntry *se;
 482
 483     se = g_malloc0(sizeof(SaveStateEntry));
 484     se->version_id = version_id;
 485     se->section_id = savevm_state.global_section_id++;
 486     se->ops = ops;
 487     se->opaque = opaque;
 488     se->vmsd = NULL;
 489     /* if this is a live_savem then set is_ram */
 490     if (ops->save_live_setup != NULL) {
 491         se->is_ram = 1;
 492     }
 493
 494     if (dev) {
 495         char *id = qdev_get_dev_path(dev);
 496         if (id) {
 497             pstrcpy(se->idstr, sizeof(se->idstr), id);
 498             pstrcat(se->idstr, sizeof(se->idstr), "/");
 499             g_free(id);
 500
 501             se->compat = g_malloc0(sizeof(CompatEntry));
 502             pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), idstr);
 503             se->compat->instance_id = instance_id == -1 ?
 504                          calculate_compat_instance_id(idstr) : instance_id;
 505             instance_id = -1;
 506         }
 507     }
 508     pstrcat(se->idstr, sizeof(se->idstr), idstr);
 509
 510     if (instance_id == -1) {
 511         se->instance_id = calculate_new_instance_id(se->idstr);
 512     } else {
 513         se->instance_id = instance_id;
 514     }
 515     assert(!se->compat || se->instance_id == 0);
 516     /* add at the end of list */
 517     QTAILQ_INSERT_TAIL(&savevm_state.handlers, se, entry);
 518     return 0;
 519 }
 520
 521 int register_savevm(DeviceState *dev,
 522                     const char *idstr,
 523                     int instance_id,
 524                     int version_id,
 525                     SaveStateHandler *save_state,
 526                     LoadStateHandler *load_state,
 527                     void *opaque)
 528 {
 529     SaveVMHandlers *ops = g_malloc0(sizeof(SaveVMHandlers));
 530     ops->save_state = save_state;
 531     ops->load_state = load_state;
 532     return register_savevm_live(dev, idstr, instance_id, version_id,
 533                                 ops, opaque);
 534 }
 535
 536 void unregister_savevm(DeviceState *dev, const char *idstr, void *opaque)
 537 {
 538     SaveStateEntry *se, *new_se;
 539     char id[256] = "";
 540
 541     if (dev) {
 542         char *path = qdev_get_dev_path(dev);
 543         if (path) {
 544             pstrcpy(id, sizeof(id), path);
 545             pstrcat(id, sizeof(id), "/");
 546             g_free(path);
 547         }
 548     }
 549     pstrcat(id, sizeof(id), idstr);
 550
 551     QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
 552         if (strcmp(se->idstr, id) == 0 && se->opaque == opaque) {
 553             QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
 554             if (se->compat) {
 555                 g_free(se->compat);
 556             }
 557             g_free(se->ops);
 558             g_free(se);
 559         }
 560     }
 561 }
 562
 563 int vmstate_register_with_alias_id(DeviceState *dev, int instance_id,
 564                                    const VMStateDescription *vmsd,
 565                                    void *opaque, int alias_id,
 566                                    int required_for_version)
 567 {
 568     SaveStateEntry *se;
 569
 570     /* If this triggers, alias support can be dropped for the vmsd. */
 571     assert(alias_id == -1 || required_for_version >= vmsd->minimum_version_id);
 572
 573     se = g_malloc0(sizeof(SaveStateEntry));
 574     se->version_id = vmsd->version_id;
 575     se->section_id = savevm_state.global_section_id++;
 576     se->opaque = opaque;
 577     se->vmsd = vmsd;
 578     se->alias_id = alias_id;
 579
 580     if (dev) {
 581         char *id = qdev_get_dev_path(dev);
 582         if (id) {
 583             pstrcpy(se->idstr, sizeof(se->idstr), id);
 584             pstrcat(se->idstr, sizeof(se->idstr), "/");
 585             g_free(id);
 586
 587             se->compat = g_malloc0(sizeof(CompatEntry));
 588             pstrcpy(se->compat->idstr, sizeof(se->compat->idstr), vmsd->name);
 589             se->compat->instance_id = instance_id == -1 ?
 590                          calculate_compat_instance_id(vmsd->name) : instance_id;
 591             instance_id = -1;
 592         }
 593     }
 594     pstrcat(se->idstr, sizeof(se->idstr), vmsd->name);
 595
 596     if (instance_id == -1) {
 597         se->instance_id = calculate_new_instance_id(se->idstr);
 598     } else {
 599         se->instance_id = instance_id;
 600     }
 601     assert(!se->compat || se->instance_id == 0);
 602     /* add at the end of list */
 603     QTAILQ_INSERT_TAIL(&savevm_state.handlers, se, entry);
 604     return 0;
 605 }
 606
 607 void vmstate_unregister(DeviceState *dev, const VMStateDescription *vmsd,
 608                         void *opaque)
 609 {
 610     SaveStateEntry *se, *new_se;
 611
 612     QTAILQ_FOREACH_SAFE(se, &savevm_state.handlers, entry, new_se) {
 613         if (se->vmsd == vmsd && se->opaque == opaque) {
 614             QTAILQ_REMOVE(&savevm_state.handlers, se, entry);
 615             if (se->compat) {
 616                 g_free(se->compat);
 617             }
 618             g_free(se);
 619         }
 620     }
 621 }
 622
 623 static int vmstate_load(QEMUFile *f, SaveStateEntry *se, int version_id)
 624 {
 625     trace_vmstate_load(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
 626     if (!se->vmsd) {         /* Old style */
 627         return se->ops->load_state(f, se->opaque, version_id);
 628     }
 629     return vmstate_load_state(f, se->vmsd, se->opaque, version_id);
 630 }
 631
 632 static void vmstate_save_old_style(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
 633 {
 634     int64_t old_offset, size;
 635
 636     old_offset = qemu_ftell_fast(f);
 637     se->ops->save_state(f, se->opaque);
 638     size = qemu_ftell_fast(f) - old_offset;
 639
 640     if (vmdesc) {
 641         json_prop_int(vmdesc, "size", size);
 642         json_start_array(vmdesc, "fields");
 643         json_start_object(vmdesc, NULL);
 644         json_prop_str(vmdesc, "name", "data");
 645         json_prop_int(vmdesc, "size", size);
 646         json_prop_str(vmdesc, "type", "buffer");
 647         json_end_object(vmdesc);
 648         json_end_array(vmdesc);
 649     }
 650 }
 651
 652 static void vmstate_save(QEMUFile *f, SaveStateEntry *se, QJSON *vmdesc)
 653 {
 654     trace_vmstate_save(se->idstr, se->vmsd ? se->vmsd->name : "(old)");
 655     if (!se->vmsd) {
 656         vmstate_save_old_style(f, se, vmdesc);
 657         return;
 658     }
 659     vmstate_save_state(f, se->vmsd, se->opaque, vmdesc);
 660 }
 661
 662 void savevm_skip_section_footers(void)
 663 {
 664     skip_section_footers = true;
 665 }
 666
 667 /*
 668  * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
 669  */
 670 static void save_section_header(QEMUFile *f, SaveStateEntry *se,
 671                                 uint8_t section_type)
 672 {
 673     qemu_put_byte(f, section_type);
 674     qemu_put_be32(f, se->section_id);
 675
 676     if (section_type == QEMU_VM_SECTION_FULL ||
 677         section_type == QEMU_VM_SECTION_START) {
 678         /* ID string */
 679         size_t len = strlen(se->idstr);
 680         qemu_put_byte(f, len);
 681         qemu_put_buffer(f, (uint8_t *)se->idstr, len);
 682
 683         qemu_put_be32(f, se->instance_id);
 684         qemu_put_be32(f, se->version_id);
 685     }
 686 }
 687
 688 /*
 689  * Write a footer onto device sections that catches cases misformatted device
 690  * sections.
 691  */
 692 static void save_section_footer(QEMUFile *f, SaveStateEntry *se)
 693 {
 694     if (!skip_section_footers) {
 695         qemu_put_byte(f, QEMU_VM_SECTION_FOOTER);
 696         qemu_put_be32(f, se->section_id);
 697     }
 698 }
 699
 700 bool qemu_savevm_state_blocked(Error **errp)
 701 {
 702     SaveStateEntry *se;
 703
 704     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
 705         if (se->vmsd && se->vmsd->unmigratable) {
 706             error_setg(errp, "State blocked by non-migratable device '%s'",
 707                        se->idstr);
 708             return true;
 709         }
 710     }
 711     return false;
 712 }
 713
 714 void qemu_savevm_state_header(QEMUFile *f)
 715 {
 716     trace_savevm_state_header();
 717     qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
 718     qemu_put_be32(f, QEMU_VM_FILE_VERSION);
 719 }
 720
 721 void qemu_savevm_state_begin(QEMUFile *f,
 722                              const MigrationParams *params)
 723 {
 724     SaveStateEntry *se;
 725     int ret;
 726
 727     trace_savevm_state_begin();
 728     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
 729         if (!se->ops || !se->ops->set_params) {
 730             continue;
 731         }
 732         se->ops->set_params(params, se->opaque);
 733     }
 734
 735     if (!savevm_state.skip_configuration) {
 736         qemu_put_byte(f, QEMU_VM_CONFIGURATION);
 737         vmstate_save_state(f, &vmstate_configuration, &savevm_state, 0);
 738     }
 739
 740     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
 741         if (!se->ops || !se->ops->save_live_setup) {
 742             continue;
 743         }
 744         if (se->ops && se->ops->is_active) {
 745             if (!se->ops->is_active(se->opaque)) {
 746                 continue;
 747             }
 748         }
 749         save_section_header(f, se, QEMU_VM_SECTION_START);
 750
 751         ret = se->ops->save_live_setup(f, se->opaque);
 752         save_section_footer(f, se);
 753         if (ret < 0) {
 754             qemu_file_set_error(f, ret);
 755             break;
 756         }
 757     }
 758 }
 759
 760 /*
 761  * this function has three return values:
 762  *   negative: there was one error, and we have -errno.
 763  *   0 : We haven't finished, caller have to go again
 764  *   1 : We have finished, we can go to complete phase
 765  */
 766 int qemu_savevm_state_iterate(QEMUFile *f)
 767 {
 768     SaveStateEntry *se;
 769     int ret = 1;
 770
 771     trace_savevm_state_iterate();
 772     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
 773         if (!se->ops || !se->ops->save_live_iterate) {
 774             continue;
 775         }
 776         if (se->ops && se->ops->is_active) {
 777             if (!se->ops->is_active(se->opaque)) {
 778                 continue;
 779             }
 780         }
 781         if (qemu_file_rate_limit(f)) {
 782             return 0;
 783         }
 784         trace_savevm_section_start(se->idstr, se->section_id);
 785
 786         save_section_header(f, se, QEMU_VM_SECTION_PART);
 787
 788         ret = se->ops->save_live_iterate(f, se->opaque);
 789         trace_savevm_section_end(se->idstr, se->section_id, ret);
 790         save_section_footer(f, se);
 791
 792         if (ret < 0) {
 793             qemu_file_set_error(f, ret);
 794         }
 795         if (ret <= 0) {
 796             /* Do not proceed to the next vmstate before this one reported
 797                completion of the current stage. This serializes the migration
 798                and reduces the probability that a faster changing state is
 799                synchronized over and over again. */
 800             break;
 801         }
 802     }
 803     return ret;
 804 }
 805
 806 static bool should_send_vmdesc(void)
 807 {
 808     MachineState *machine = MACHINE(qdev_get_machine());
 809     return !machine->suppress_vmdesc;
 810 }
 811
 812 void qemu_savevm_state_complete(QEMUFile *f)
 813 {
 814     QJSON *vmdesc;
 815     int vmdesc_len;
 816     SaveStateEntry *se;
 817     int ret;
 818
 819     trace_savevm_state_complete();
 820
 821     cpu_synchronize_all_states();
 822
 823     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
 824         if (!se->ops || !se->ops->save_live_complete) {
 825             continue;
 826         }
 827         if (se->ops && se->ops->is_active) {
 828             if (!se->ops->is_active(se->opaque)) {
 829                 continue;
 830             }
 831         }
 832         trace_savevm_section_start(se->idstr, se->section_id);
 833
 834         save_section_header(f, se, QEMU_VM_SECTION_END);
 835
 836         ret = se->ops->save_live_complete(f, se->opaque);
 837         trace_savevm_section_end(se->idstr, se->section_id, ret);
 838         save_section_footer(f, se);
 839         if (ret < 0) {
 840             qemu_file_set_error(f, ret);
 841             return;
 842         }
 843     }
 844
 845     vmdesc = qjson_new();
 846     json_prop_int(vmdesc, "page_size", TARGET_PAGE_SIZE);
 847     json_start_array(vmdesc, "devices");
 848     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
 849
 850         if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
 851             continue;
 852         }
 853         if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
 854             trace_savevm_section_skip(se->idstr, se->section_id);
 855             continue;
 856         }
 857
 858         trace_savevm_section_start(se->idstr, se->section_id);
 859
 860         json_start_object(vmdesc, NULL);
 861         json_prop_str(vmdesc, "name", se->idstr);
 862         json_prop_int(vmdesc, "instance_id", se->instance_id);
 863
 864         save_section_header(f, se, QEMU_VM_SECTION_FULL);
 865
 866         vmstate_save(f, se, vmdesc);
 867
 868         json_end_object(vmdesc);
 869         trace_savevm_section_end(se->idstr, se->section_id, 0);
 870         save_section_footer(f, se);
 871     }
 872
 873     qemu_put_byte(f, QEMU_VM_EOF);
 874
 875     json_end_array(vmdesc);
 876     qjson_finish(vmdesc);
 877     vmdesc_len = strlen(qjson_get_str(vmdesc));
 878
 879     if (should_send_vmdesc()) {
 880         qemu_put_byte(f, QEMU_VM_VMDESCRIPTION);
 881         qemu_put_be32(f, vmdesc_len);
 882         qemu_put_buffer(f, (uint8_t *)qjson_get_str(vmdesc), vmdesc_len);
 883     }
 884     object_unref(OBJECT(vmdesc));
 885
 886     qemu_fflush(f);
 887 }
 888
 889 uint64_t qemu_savevm_state_pending(QEMUFile *f, uint64_t max_size)
 890 {
 891     SaveStateEntry *se;
 892     uint64_t ret = 0;
 893
 894     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
 895         if (!se->ops || !se->ops->save_live_pending) {
 896             continue;
 897         }
 898         if (se->ops && se->ops->is_active) {
 899             if (!se->ops->is_active(se->opaque)) {
 900                 continue;
 901             }
 902         }
 903         ret += se->ops->save_live_pending(f, se->opaque, max_size);
 904     }
 905     return ret;
 906 }
 907
 908 void qemu_savevm_state_cancel(void)
 909 {
 910     SaveStateEntry *se;
 911
 912     trace_savevm_state_cancel();
 913     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
 914         if (se->ops && se->ops->cancel) {
 915             se->ops->cancel(se->opaque);
 916         }
 917     }
 918 }
 919
 920 static int qemu_savevm_state(QEMUFile *f, Error **errp)
 921 {
 922     int ret;
 923     MigrationParams params = {
 924         .blk = 0,
 925         .shared = 0
 926     };
 927
 928     if (qemu_savevm_state_blocked(errp)) {
 929         return -EINVAL;
 930     }
 931
 932     qemu_mutex_unlock_iothread();
 933     qemu_savevm_state_header(f);
 934     qemu_savevm_state_begin(f, &params);
 935     qemu_mutex_lock_iothread();
 936
 937     while (qemu_file_get_error(f) == 0) {
 938         if (qemu_savevm_state_iterate(f) > 0) {
 939             break;
 940         }
 941     }
 942
 943     ret = qemu_file_get_error(f);
 944     if (ret == 0) {
 945         qemu_savevm_state_complete(f);
 946         ret = qemu_file_get_error(f);
 947     }
 948     if (ret != 0) {
 949         qemu_savevm_state_cancel();
 950         error_setg_errno(errp, -ret, "Error while writing VM state");
 951     }
 952     return ret;
 953 }
 954
 955 static int qemu_save_device_state(QEMUFile *f)
 956 {
 957     SaveStateEntry *se;
 958
 959     qemu_put_be32(f, QEMU_VM_FILE_MAGIC);
 960     qemu_put_be32(f, QEMU_VM_FILE_VERSION);
 961
 962     cpu_synchronize_all_states();
 963
 964     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
 965         if (se->is_ram) {
 966             continue;
 967         }
 968         if ((!se->ops || !se->ops->save_state) && !se->vmsd) {
 969             continue;
 970         }
 971         if (se->vmsd && !vmstate_save_needed(se->vmsd, se->opaque)) {
 972             continue;
 973         }
 974
 975         save_section_header(f, se, QEMU_VM_SECTION_FULL);
 976
 977         vmstate_save(f, se, NULL);
 978
 979         save_section_footer(f, se);
 980     }
 981
 982     qemu_put_byte(f, QEMU_VM_EOF);
 983
 984     return qemu_file_get_error(f);
 985 }
 986
 987 static SaveStateEntry *find_se(const char *idstr, int instance_id)
 988 {
 989     SaveStateEntry *se;
 990
 991     QTAILQ_FOREACH(se, &savevm_state.handlers, entry) {
 992         if (!strcmp(se->idstr, idstr) &&
 993             (instance_id == se->instance_id ||
 994              instance_id == se->alias_id))
 995             return se;
 996         /* Migrating from an older version? */
 997         if (strstr(se->idstr, idstr) && se->compat) {
 998             if (!strcmp(se->compat->idstr, idstr) &&
 999                 (instance_id == se->compat->instance_id ||
1000                  instance_id == se->alias_id))
1001                 return se;
1002         }
1003     }
1004     return NULL;
1005 }
1006
1007 struct LoadStateEntry {
1008     QLIST_ENTRY(LoadStateEntry) entry;
1009     SaveStateEntry *se;
1010     int section_id;
1011     int version_id;
1012 };
1013
1014 /*
1015  * Read a footer off the wire and check that it matches the expected section
1016  *
1017  * Returns: true if the footer was good
1018  *          false if there is a problem (and calls error_report to say why)
1019  */
1020 static bool check_section_footer(QEMUFile *f, LoadStateEntry *le)
1021 {
1022     uint8_t read_mark;
1023     uint32_t read_section_id;
1024
1025     if (skip_section_footers) {
1026         /* No footer to check */
1027         return true;
1028     }
1029
1030     read_mark = qemu_get_byte(f);
1031
1032     if (read_mark != QEMU_VM_SECTION_FOOTER) {
1033         error_report("Missing section footer for %s", le->se->idstr);
1034         return false;
1035     }
1036
1037     read_section_id = qemu_get_be32(f);
1038     if (read_section_id != le->section_id) {
1039         error_report("Mismatched section id in footer for %s -"
1040                      " read 0x%x expected 0x%x",
1041                      le->se->idstr, read_section_id, le->section_id);
1042         return false;
1043     }
1044
1045     /* All good */
1046     return true;
1047 }
1048
1049 void loadvm_free_handlers(MigrationIncomingState *mis)
1050 {
1051     LoadStateEntry *le, *new_le;
1052
1053     QLIST_FOREACH_SAFE(le, &mis->loadvm_handlers, entry, new_le) {
1054         QLIST_REMOVE(le, entry);
1055         g_free(le);
1056     }
1057 }
1058
1059 int qemu_loadvm_state(QEMUFile *f)
1060 {
1061     MigrationIncomingState *mis = migration_incoming_get_current();
1062     Error *local_err = NULL;
1063     uint8_t section_type;
1064     unsigned int v;
1065     int ret;
1066     int file_error_after_eof = -1;
1067
1068     if (qemu_savevm_state_blocked(&local_err)) {
1069         error_report_err(local_err);
1070         return -EINVAL;
1071     }
1072
1073     v = qemu_get_be32(f);
1074     if (v != QEMU_VM_FILE_MAGIC) {
1075         error_report("Not a migration stream");
1076         return -EINVAL;
1077     }
1078
1079     v = qemu_get_be32(f);
1080     if (v == QEMU_VM_FILE_VERSION_COMPAT) {
1081         error_report("SaveVM v2 format is obsolete and don't work anymore");
1082         return -ENOTSUP;
1083     }
1084     if (v != QEMU_VM_FILE_VERSION) {
1085         error_report("Unsupported migration stream version");
1086         return -ENOTSUP;
1087     }
1088
1089     if (!savevm_state.skip_configuration) {
1090         if (qemu_get_byte(f) != QEMU_VM_CONFIGURATION) {
1091             error_report("Configuration section missing");
1092             return -EINVAL;
1093         }
1094         ret = vmstate_load_state(f, &vmstate_configuration, &savevm_state, 0);
1095
1096         if (ret) {
1097             return ret;
1098         }
1099     }
1100
1101     while ((section_type = qemu_get_byte(f)) != QEMU_VM_EOF) {
1102         uint32_t instance_id, version_id, section_id;
1103         SaveStateEntry *se;
1104         LoadStateEntry *le;
1105         char idstr[256];
1106
1107         trace_qemu_loadvm_state_section(section_type);
1108         switch (section_type) {
1109         case QEMU_VM_SECTION_START:
1110         case QEMU_VM_SECTION_FULL:
1111             /* Read section start */
1112             section_id = qemu_get_be32(f);
1113             if (!qemu_get_counted_string(f, idstr)) {
1114                 error_report("Unable to read ID string for section %u",
1115                             section_id);
1116                 return -EINVAL;
1117             }
1118             instance_id = qemu_get_be32(f);
1119             version_id = qemu_get_be32(f);
1120
1121             trace_qemu_loadvm_state_section_startfull(section_id, idstr,
1122                                                       instance_id, version_id);
1123             /* Find savevm section */
1124             se = find_se(idstr, instance_id);
1125             if (se == NULL) {
1126                 error_report("Unknown savevm section or instance '%s' %d",
1127                              idstr, instance_id);
1128                 ret = -EINVAL;
1129                 goto out;
1130             }
1131
1132             /* Validate version */
1133             if (version_id > se->version_id) {
1134                 error_report("savevm: unsupported version %d for '%s' v%d",
1135                              version_id, idstr, se->version_id);
1136                 ret = -EINVAL;
1137                 goto out;
1138             }
1139
1140             /* Add entry */
1141             le = g_malloc0(sizeof(*le));
1142
1143             le->se = se;
1144             le->section_id = section_id;
1145             le->version_id = version_id;
1146             QLIST_INSERT_HEAD(&mis->loadvm_handlers, le, entry);
1147
1148             ret = vmstate_load(f, le->se, le->version_id);
1149             if (ret < 0) {
1150                 error_report("error while loading state for instance 0x%x of"
1151                              " device '%s'", instance_id, idstr);
1152                 goto out;
1153             }
1154             if (!check_section_footer(f, le)) {
1155                 ret = -EINVAL;
1156                 goto out;
1157             }
1158             break;
1159         case QEMU_VM_SECTION_PART:
1160         case QEMU_VM_SECTION_END:
1161             section_id = qemu_get_be32(f);
1162
1163             trace_qemu_loadvm_state_section_partend(section_id);
1164             QLIST_FOREACH(le, &mis->loadvm_handlers, entry) {
1165                 if (le->section_id == section_id) {
1166                     break;
1167                 }
1168             }
1169             if (le == NULL) {
1170                 error_report("Unknown savevm section %d", section_id);
1171                 ret = -EINVAL;
1172                 goto out;
1173             }
1174
1175             ret = vmstate_load(f, le->se, le->version_id);
1176             if (ret < 0) {
1177                 error_report("error while loading state section id %d(%s)",
1178                              section_id, le->se->idstr);
1179                 goto out;
1180             }
1181             if (!check_section_footer(f, le)) {
1182                 ret = -EINVAL;
1183                 goto out;
1184             }
1185             break;
1186         default:
1187             error_report("Unknown savevm section type %d", section_type);
1188             ret = -EINVAL;
1189             goto out;
1190         }
1191     }
1192
1193     file_error_after_eof = qemu_file_get_error(f);
1194
1195     /*
1196      * Try to read in the VMDESC section as well, so that dumping tools that
1197      * intercept our migration stream have the chance to see it.
1198      */
1199
1200     /* We've got to be careful; if we don't read the data and just shut the fd
1201      * then the sender can error if we close while it's still sending.
1202      * We also mustn't read data that isn't there; some transports (RDMA)
1203      * will stall waiting for that data when the source has already closed.
1204      */
1205     if (should_send_vmdesc()) {
1206         uint8_t *buf;
1207         uint32_t size;
1208         section_type = qemu_get_byte(f);
1209
1210         if (section_type != QEMU_VM_VMDESCRIPTION) {
1211             error_report("Expected vmdescription section, but got %d",
1212                          section_type);
1213             /*
1214              * It doesn't seem worth failing at this point since
1215              * we apparently have an otherwise valid VM state
1216              */
1217         } else {
1218             buf = g_malloc(0x1000);
1219             size = qemu_get_be32(f);
1220
1221             while (size > 0) {
1222                 uint32_t read_chunk = MIN(size, 0x1000);
1223                 qemu_get_buffer(f, buf, read_chunk);
1224                 size -= read_chunk;
1225             }
1226             g_free(buf);
1227         }
1228     }
1229
1230     cpu_synchronize_all_post_init();
1231
1232     ret = 0;
1233
1234 out:
1235     if (ret == 0) {
1236         /* We may not have a VMDESC section, so ignore relative errors */
1237         ret = file_error_after_eof;
1238     }
1239
1240     return ret;
1241 }
1242
1243 static BlockDriverState *find_vmstate_bs(void)
1244 {
1245     BlockDriverState *bs = NULL;
1246     while ((bs = bdrv_next(bs))) {
1247         if (bdrv_can_snapshot(bs)) {
1248             return bs;
1249         }
1250     }
1251     return NULL;
1252 }
1253
1254 /*
1255  * Deletes snapshots of a given name in all opened images.
1256  */
1257 static int del_existing_snapshots(Monitor *mon, const char *name)
1258 {
1259     BlockDriverState *bs;
1260     QEMUSnapshotInfo sn1, *snapshot = &sn1;
1261     Error *err = NULL;
1262
1263     bs = NULL;
1264     while ((bs = bdrv_next(bs))) {
1265         if (bdrv_can_snapshot(bs) &&
1266             bdrv_snapshot_find(bs, snapshot, name) >= 0) {
1267             bdrv_snapshot_delete_by_id_or_name(bs, name, &err);
1268             if (err) {
1269                 monitor_printf(mon,
1270                                "Error while deleting snapshot on device '%s':"
1271                                " %s\n",
1272                                bdrv_get_device_name(bs),
1273                                error_get_pretty(err));
1274                 error_free(err);
1275                 return -1;
1276             }
1277         }
1278     }
1279
1280     return 0;
1281 }
1282
1283 void hmp_savevm(Monitor *mon, const QDict *qdict)
1284 {
1285     BlockDriverState *bs, *bs1;
1286     QEMUSnapshotInfo sn1, *sn = &sn1, old_sn1, *old_sn = &old_sn1;
1287     int ret;
1288     QEMUFile *f;
1289     int saved_vm_running;
1290     uint64_t vm_state_size;
1291     qemu_timeval tv;
1292     struct tm tm;
1293     const char *name = qdict_get_try_str(qdict, "name");
1294     Error *local_err = NULL;
1295
1296     /* Verify if there is a device that doesn't support snapshots and is writable */
1297     bs = NULL;
1298     while ((bs = bdrv_next(bs))) {
1299
1300         if (!bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
1301             continue;
1302         }
1303
1304         if (!bdrv_can_snapshot(bs)) {
1305             monitor_printf(mon, "Device '%s' is writable but does not support snapshots.\n",
1306                                bdrv_get_device_name(bs));
1307             return;
1308         }
1309     }
1310
1311     bs = find_vmstate_bs();
1312     if (!bs) {
1313         monitor_printf(mon, "No block device can accept snapshots\n");
1314         return;
1315     }
1316
1317     saved_vm_running = runstate_is_running();
1318     vm_stop(RUN_STATE_SAVE_VM);
1319
1320     memset(sn, 0, sizeof(*sn));
1321
1322     /* fill auxiliary fields */
1323     qemu_gettimeofday(&tv);
1324     sn->date_sec = tv.tv_sec;
1325     sn->date_nsec = tv.tv_usec * 1000;
1326     sn->vm_clock_nsec = qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL);
1327
1328     if (name) {
1329         ret = bdrv_snapshot_find(bs, old_sn, name);
1330         if (ret >= 0) {
1331             pstrcpy(sn->name, sizeof(sn->name), old_sn->name);
1332             pstrcpy(sn->id_str, sizeof(sn->id_str), old_sn->id_str);
1333         } else {
1334             pstrcpy(sn->name, sizeof(sn->name), name);
1335         }
1336     } else {
1337         /* cast below needed for OpenBSD where tv_sec is still 'long' */
1338         localtime_r((const time_t *)&tv.tv_sec, &tm);
1339         strftime(sn->name, sizeof(sn->name), "vm-%Y%m%d%H%M%S", &tm);
1340     }
1341
1342     /* Delete old snapshots of the same name */
1343     if (name && del_existing_snapshots(mon, name) < 0) {
1344         goto the_end;
1345     }
1346
1347     /* save the VM state */
1348     f = qemu_fopen_bdrv(bs, 1);
1349     if (!f) {
1350         monitor_printf(mon, "Could not open VM state file\n");
1351         goto the_end;
1352     }
1353     ret = qemu_savevm_state(f, &local_err);
1354     vm_state_size = qemu_ftell(f);
1355     qemu_fclose(f);
1356     if (ret < 0) {
1357         monitor_printf(mon, "%s\n", error_get_pretty(local_err));
1358         error_free(local_err);
1359         goto the_end;
1360     }
1361
1362     /* create the snapshots */
1363
1364     bs1 = NULL;
1365     while ((bs1 = bdrv_next(bs1))) {
1366         if (bdrv_can_snapshot(bs1)) {
1367             /* Write VM state size only to the image that contains the state */
1368             sn->vm_state_size = (bs == bs1 ? vm_state_size : 0);
1369             ret = bdrv_snapshot_create(bs1, sn);
1370             if (ret < 0) {
1371                 monitor_printf(mon, "Error while creating snapshot on '%s'\n",
1372                                bdrv_get_device_name(bs1));
1373             }
1374         }
1375     }
1376
1377  the_end:
1378     if (saved_vm_running) {
1379         vm_start();
1380     }
1381 }
1382
1383 void qmp_xen_save_devices_state(const char *filename, Error **errp)
1384 {
1385     QEMUFile *f;
1386     int saved_vm_running;
1387     int ret;
1388
1389     saved_vm_running = runstate_is_running();
1390     vm_stop(RUN_STATE_SAVE_VM);
1391
1392     f = qemu_fopen(filename, "wb");
1393     if (!f) {
1394         error_setg_file_open(errp, errno, filename);
1395         goto the_end;
1396     }
1397     ret = qemu_save_device_state(f);
1398     qemu_fclose(f);
1399     if (ret < 0) {
1400         error_setg(errp, QERR_IO_ERROR);
1401     }
1402
1403  the_end:
1404     if (saved_vm_running) {
1405         vm_start();
1406     }
1407 }
1408
1409 int load_vmstate(const char *name)
1410 {
1411     BlockDriverState *bs, *bs_vm_state;
1412     QEMUSnapshotInfo sn;
1413     QEMUFile *f;
1414     int ret;
1415
1416     bs_vm_state = find_vmstate_bs();
1417     if (!bs_vm_state) {
1418         error_report("No block device supports snapshots");
1419         return -ENOTSUP;
1420     }
1421
1422     /* Don't even try to load empty VM states */
1423     ret = bdrv_snapshot_find(bs_vm_state, &sn, name);
1424     if (ret < 0) {
1425         return ret;
1426     } else if (sn.vm_state_size == 0) {
1427         error_report("This is a disk-only snapshot. Revert to it offline "
1428             "using qemu-img.");
1429         return -EINVAL;
1430     }
1431
1432     /* Verify if there is any device that doesn't support snapshots and is
1433     writable and check if the requested snapshot is available too. */
1434     bs = NULL;
1435     while ((bs = bdrv_next(bs))) {
1436
1437         if (!bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) {
1438             continue;
1439         }
1440
1441         if (!bdrv_can_snapshot(bs)) {
1442             error_report("Device '%s' is writable but does not support snapshots.",
1443                                bdrv_get_device_name(bs));
1444             return -ENOTSUP;
1445         }
1446
1447         ret = bdrv_snapshot_find(bs, &sn, name);
1448         if (ret < 0) {
1449             error_report("Device '%s' does not have the requested snapshot '%s'",
1450                            bdrv_get_device_name(bs), name);
1451             return ret;
1452         }
1453     }
1454
1455     /* Flush all IO requests so they don't interfere with the new state.  */
1456     bdrv_drain_all();
1457
1458     bs = NULL;
1459     while ((bs = bdrv_next(bs))) {
1460         if (bdrv_can_snapshot(bs)) {
1461             ret = bdrv_snapshot_goto(bs, name);
1462             if (ret < 0) {
1463                 error_report("Error %d while activating snapshot '%s' on '%s'",
1464                              ret, name, bdrv_get_device_name(bs));
1465                 return ret;
1466             }
1467         }
1468     }
1469
1470     /* restore the VM state */
1471     f = qemu_fopen_bdrv(bs_vm_state, 0);
1472     if (!f) {
1473         error_report("Could not open VM state file");
1474         return -EINVAL;
1475     }
1476
1477     qemu_system_reset(VMRESET_SILENT);
1478     migration_incoming_state_new(f);
1479     ret = qemu_loadvm_state(f);
1480
1481     qemu_fclose(f);
1482     migration_incoming_state_destroy();
1483     if (ret < 0) {
1484         error_report("Error %d while loading VM state", ret);
1485         return ret;
1486     }
1487
1488     return 0;
1489 }
1490
1491 void hmp_delvm(Monitor *mon, const QDict *qdict)
1492 {
1493     BlockDriverState *bs;
1494     Error *err;
1495     const char *name = qdict_get_str(qdict, "name");
1496
1497     if (!find_vmstate_bs()) {
1498         monitor_printf(mon, "No block device supports snapshots\n");
1499         return;
1500     }
1501
1502     bs = NULL;
1503     while ((bs = bdrv_next(bs))) {
1504         if (bdrv_can_snapshot(bs)) {
1505             err = NULL;
1506             bdrv_snapshot_delete_by_id_or_name(bs, name, &err);
1507             if (err) {
1508                 monitor_printf(mon,
1509                                "Error while deleting snapshot on device '%s':"
1510                                " %s\n",
1511                                bdrv_get_device_name(bs),
1512                                error_get_pretty(err));
1513                 error_free(err);
1514             }
1515         }
1516     }
1517 }
1518
1519 void hmp_info_snapshots(Monitor *mon, const QDict *qdict)
1520 {
1521     BlockDriverState *bs, *bs1;
1522     QEMUSnapshotInfo *sn_tab, *sn, s, *sn_info = &s;
1523     int nb_sns, i, ret, available;
1524     int total;
1525     int *available_snapshots;
1526
1527     bs = find_vmstate_bs();
1528     if (!bs) {
1529         monitor_printf(mon, "No available block device supports snapshots\n");
1530         return;
1531     }
1532
1533     nb_sns = bdrv_snapshot_list(bs, &sn_tab);
1534     if (nb_sns < 0) {
1535         monitor_printf(mon, "bdrv_snapshot_list: error %d\n", nb_sns);
1536         return;
1537     }
1538
1539     if (nb_sns == 0) {
1540         monitor_printf(mon, "There is no snapshot available.\n");
1541         return;
1542     }
1543
1544     available_snapshots = g_malloc0(sizeof(int) * nb_sns);
1545     total = 0;
1546     for (i = 0; i < nb_sns; i++) {
1547         sn = &sn_tab[i];
1548         available = 1;
1549         bs1 = NULL;
1550
1551         while ((bs1 = bdrv_next(bs1))) {
1552             if (bdrv_can_snapshot(bs1) && bs1 != bs) {
1553                 ret = bdrv_snapshot_find(bs1, sn_info, sn->id_str);
1554                 if (ret < 0) {
1555                     available = 0;
1556                     break;
1557                 }
1558             }
1559         }
1560
1561         if (available) {
1562             available_snapshots[total] = i;
1563             total++;
1564         }
1565     }
1566
1567     if (total > 0) {
1568         bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, NULL);
1569         monitor_printf(mon, "\n");
1570         for (i = 0; i < total; i++) {
1571             sn = &sn_tab[available_snapshots[i]];
1572             bdrv_snapshot_dump((fprintf_function)monitor_printf, mon, sn);
1573             monitor_printf(mon, "\n");
1574         }
1575     } else {
1576         monitor_printf(mon, "There is no suitable snapshot available\n");
1577     }
1578
1579     g_free(sn_tab);
1580     g_free(available_snapshots);
1581
1582 }
1583
1584 void vmstate_register_ram(MemoryRegion *mr, DeviceState *dev)
1585 {
1586     qemu_ram_set_idstr(memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK,
1587                        memory_region_name(mr), dev);
1588 }
1589
1590 void vmstate_unregister_ram(MemoryRegion *mr, DeviceState *dev)
1591 {
1592     qemu_ram_unset_idstr(memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK);
1593 }
1594
1595 void vmstate_register_ram_global(MemoryRegion *mr)
1596 {
1597     vmstate_register_ram(mr, NULL);
1598 }