migration.c

   1 /*
   2  * QEMU live migration
   3  *
   4  * Copyright IBM, Corp. 2008
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  * Contributions after 2012-01-13 are licensed under the terms of the
  13  * GNU GPL, version 2 or (at your option) any later version.
  14  */
  15
  16 #include "qemu-common.h"
  17 #include "migration.h"
  18 #include "monitor.h"
  19 #include "buffered_file.h"
  20 #include "sysemu.h"
  21 #include "block.h"
  22 #include "qemu_socket.h"
  23 #include "block-migration.h"
  24 #include "qmp-commands.h"
  25
  26 //#define DEBUG_MIGRATION
  27
  28 #ifdef DEBUG_MIGRATION
  29 #define DPRINTF(fmt, ...) \
  30     do { printf("migration: " fmt, ## __VA_ARGS__); } while (0)
  31 #else
  32 #define DPRINTF(fmt, ...) \
  33     do { } while (0)
  34 #endif
  35
  36 enum {
  37     MIG_STATE_ERROR,
  38     MIG_STATE_SETUP,
  39     MIG_STATE_CANCELLED,
  40     MIG_STATE_ACTIVE,
  41     MIG_STATE_COMPLETED,
  42 };
  43
  44 #define MAX_THROTTLE  (32 << 20)      /* Migration speed throttling */
  45
  46 /* Migration XBZRLE default cache size */
  47 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
  48
  49 static NotifierList migration_state_notifiers =
  50     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
  51
  52 /* When we add fault tolerance, we could have several
  53    migrations at once.  For now we don't need to add
  54    dynamic creation of migration */
  55
  56 MigrationState *migrate_get_current(void)
  57 {
  58     static MigrationState current_migration = {
  59         .state = MIG_STATE_SETUP,
  60         .bandwidth_limit = MAX_THROTTLE,
  61         .xbzrle_cache_size = DEFAULT_MIGRATE_CACHE_SIZE,
  62     };
  63
  64     return &current_migration;
  65 }
  66
  67 void qemu_start_incoming_migration(const char *uri, Error **errp)
  68 {
  69     const char *p;
  70
  71     if (strstart(uri, "tcp:", &p))
  72         tcp_start_incoming_migration(p, errp);
  73 #if !defined(WIN32)
  74     else if (strstart(uri, "exec:", &p))
  75         exec_start_incoming_migration(p, errp);
  76     else if (strstart(uri, "unix:", &p))
  77         unix_start_incoming_migration(p, errp);
  78     else if (strstart(uri, "fd:", &p))
  79         fd_start_incoming_migration(p, errp);
  80 #endif
  81     else {
  82         error_setg(errp, "unknown migration protocol: %s\n", uri);
  83     }
  84 }
  85
  86 void process_incoming_migration(QEMUFile *f)
  87 {
  88     if (qemu_loadvm_state(f) < 0) {
  89         fprintf(stderr, "load of migration failed\n");
  90         exit(0);
  91     }
  92     qemu_announce_self();
  93     DPRINTF("successfully loaded vm state\n");
  94
  95     bdrv_clear_incoming_migration_all();
  96     /* Make sure all file formats flush their mutable metadata */
  97     bdrv_invalidate_cache_all();
  98
  99     if (autostart) {
 100         vm_start();
 101     } else {
 102         runstate_set(RUN_STATE_PAUSED);
 103     }
 104 }
 105
 106 /* amount of nanoseconds we are willing to wait for migration to be down.
 107  * the choice of nanoseconds is because it is the maximum resolution that
 108  * get_clock() can achieve. It is an internal measure. All user-visible
 109  * units must be in seconds */
 110 static uint64_t max_downtime = 30000000;
 111
 112 uint64_t migrate_max_downtime(void)
 113 {
 114     return max_downtime;
 115 }
 116
 117 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
 118 {
 119     MigrationCapabilityStatusList *head = NULL;
 120     MigrationCapabilityStatusList *caps;
 121     MigrationState *s = migrate_get_current();
 122     int i;
 123
 124     for (i = 0; i < MIGRATION_CAPABILITY_MAX; i++) {
 125         if (head == NULL) {
 126             head = g_malloc0(sizeof(*caps));
 127             caps = head;
 128         } else {
 129             caps->next = g_malloc0(sizeof(*caps));
 130             caps = caps->next;
 131         }
 132         caps->value =
 133             g_malloc(sizeof(*caps->value));
 134         caps->value->capability = i;
 135         caps->value->state = s->enabled_capabilities[i];
 136     }
 137
 138     return head;
 139 }
 140
 141 static void get_xbzrle_cache_stats(MigrationInfo *info)
 142 {
 143     if (migrate_use_xbzrle()) {
 144         info->has_xbzrle_cache = true;
 145         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
 146         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
 147         info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred();
 148         info->xbzrle_cache->pages = xbzrle_mig_pages_transferred();
 149         info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss();
 150         info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow();
 151     }
 152 }
 153
 154 MigrationInfo *qmp_query_migrate(Error **errp)
 155 {
 156     MigrationInfo *info = g_malloc0(sizeof(*info));
 157     MigrationState *s = migrate_get_current();
 158
 159     switch (s->state) {
 160     case MIG_STATE_SETUP:
 161         /* no migration has happened ever */
 162         break;
 163     case MIG_STATE_ACTIVE:
 164         info->has_status = true;
 165         info->status = g_strdup("active");
 166         info->has_total_time = true;
 167         info->total_time = qemu_get_clock_ms(rt_clock)
 168             - s->total_time;
 169         info->has_expected_downtime = true;
 170         info->expected_downtime = s->expected_downtime;
 171
 172         info->has_ram = true;
 173         info->ram = g_malloc0(sizeof(*info->ram));
 174         info->ram->transferred = ram_bytes_transferred();
 175         info->ram->remaining = ram_bytes_remaining();
 176         info->ram->total = ram_bytes_total();
 177         info->ram->duplicate = dup_mig_pages_transferred();
 178         info->ram->normal = norm_mig_pages_transferred();
 179         info->ram->normal_bytes = norm_mig_bytes_transferred();
 180         info->ram->dirty_pages_rate = s->dirty_pages_rate;
 181
 182
 183         if (blk_mig_active()) {
 184             info->has_disk = true;
 185             info->disk = g_malloc0(sizeof(*info->disk));
 186             info->disk->transferred = blk_mig_bytes_transferred();
 187             info->disk->remaining = blk_mig_bytes_remaining();
 188             info->disk->total = blk_mig_bytes_total();
 189         }
 190
 191         get_xbzrle_cache_stats(info);
 192         break;
 193     case MIG_STATE_COMPLETED:
 194         get_xbzrle_cache_stats(info);
 195
 196         info->has_status = true;
 197         info->status = g_strdup("completed");
 198         info->total_time = s->total_time;
 199         info->has_downtime = true;
 200         info->downtime = s->downtime;
 201
 202         info->has_ram = true;
 203         info->ram = g_malloc0(sizeof(*info->ram));
 204         info->ram->transferred = ram_bytes_transferred();
 205         info->ram->remaining = 0;
 206         info->ram->total = ram_bytes_total();
 207         info->ram->duplicate = dup_mig_pages_transferred();
 208         info->ram->normal = norm_mig_pages_transferred();
 209         info->ram->normal_bytes = norm_mig_bytes_transferred();
 210         break;
 211     case MIG_STATE_ERROR:
 212         info->has_status = true;
 213         info->status = g_strdup("failed");
 214         break;
 215     case MIG_STATE_CANCELLED:
 216         info->has_status = true;
 217         info->status = g_strdup("cancelled");
 218         break;
 219     }
 220
 221     return info;
 222 }
 223
 224 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
 225                                   Error **errp)
 226 {
 227     MigrationState *s = migrate_get_current();
 228     MigrationCapabilityStatusList *cap;
 229
 230     if (s->state == MIG_STATE_ACTIVE) {
 231         error_set(errp, QERR_MIGRATION_ACTIVE);
 232         return;
 233     }
 234
 235     for (cap = params; cap; cap = cap->next) {
 236         s->enabled_capabilities[cap->value->capability] = cap->value->state;
 237     }
 238 }
 239
 240 /* shared migration helpers */
 241
 242 static int migrate_fd_cleanup(MigrationState *s)
 243 {
 244     int ret = 0;
 245
 246     if (s->fd != -1) {
 247         qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 248     }
 249
 250     if (s->file) {
 251         DPRINTF("closing file\n");
 252         ret = qemu_fclose(s->file);
 253         s->file = NULL;
 254     }
 255
 256     if (s->fd != -1) {
 257         close(s->fd);
 258         s->fd = -1;
 259     }
 260
 261     return ret;
 262 }
 263
 264 void migrate_fd_error(MigrationState *s)
 265 {
 266     DPRINTF("setting error state\n");
 267     s->state = MIG_STATE_ERROR;
 268     notifier_list_notify(&migration_state_notifiers, s);
 269     migrate_fd_cleanup(s);
 270 }
 271
 272 static void migrate_fd_completed(MigrationState *s)
 273 {
 274     DPRINTF("setting completed state\n");
 275     if (migrate_fd_cleanup(s) < 0) {
 276         s->state = MIG_STATE_ERROR;
 277     } else {
 278         s->state = MIG_STATE_COMPLETED;
 279         runstate_set(RUN_STATE_POSTMIGRATE);
 280     }
 281     notifier_list_notify(&migration_state_notifiers, s);
 282 }
 283
 284 static void migrate_fd_put_notify(void *opaque)
 285 {
 286     MigrationState *s = opaque;
 287     int ret;
 288
 289     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 290     ret = qemu_file_put_notify(s->file);
 291     if (ret) {
 292         migrate_fd_error(s);
 293     }
 294 }
 295
 296 ssize_t migrate_fd_put_buffer(MigrationState *s, const void *data,
 297                               size_t size)
 298 {
 299     ssize_t ret;
 300
 301     if (s->state != MIG_STATE_ACTIVE) {
 302         return -EIO;
 303     }
 304
 305     do {
 306         ret = s->write(s, data, size);
 307     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 308
 309     if (ret == -1)
 310         ret = -(s->get_error(s));
 311
 312     if (ret == -EAGAIN) {
 313         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 314     }
 315
 316     return ret;
 317 }
 318
 319 void migrate_fd_put_ready(MigrationState *s)
 320 {
 321     int ret;
 322
 323     if (s->state != MIG_STATE_ACTIVE) {
 324         DPRINTF("put_ready returning because of non-active state\n");
 325         return;
 326     }
 327
 328     DPRINTF("iterate\n");
 329     ret = qemu_savevm_state_iterate(s->file);
 330     if (ret < 0) {
 331         migrate_fd_error(s);
 332     } else if (ret == 1) {
 333         int old_vm_running = runstate_is_running();
 334         int64_t start_time, end_time;
 335
 336         DPRINTF("done iterating\n");
 337         start_time = qemu_get_clock_ms(rt_clock);
 338         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 339         vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 340
 341         if (qemu_savevm_state_complete(s->file) < 0) {
 342             migrate_fd_error(s);
 343         } else {
 344             migrate_fd_completed(s);
 345         }
 346         end_time = qemu_get_clock_ms(rt_clock);
 347         s->total_time = end_time - s->total_time;
 348         s->downtime = end_time - start_time;
 349         if (s->state != MIG_STATE_COMPLETED) {
 350             if (old_vm_running) {
 351                 vm_start();
 352             }
 353         }
 354     }
 355 }
 356
 357 static void migrate_fd_cancel(MigrationState *s)
 358 {
 359     if (s->state != MIG_STATE_ACTIVE)
 360         return;
 361
 362     DPRINTF("cancelling migration\n");
 363
 364     s->state = MIG_STATE_CANCELLED;
 365     notifier_list_notify(&migration_state_notifiers, s);
 366     qemu_savevm_state_cancel(s->file);
 367
 368     migrate_fd_cleanup(s);
 369 }
 370
 371 int migrate_fd_wait_for_unfreeze(MigrationState *s)
 372 {
 373     int ret;
 374
 375     DPRINTF("wait for unfreeze\n");
 376     if (s->state != MIG_STATE_ACTIVE)
 377         return -EINVAL;
 378
 379     do {
 380         fd_set wfds;
 381
 382         FD_ZERO(&wfds);
 383         FD_SET(s->fd, &wfds);
 384
 385         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 386     } while (ret == -1 && (s->get_error(s)) == EINTR);
 387
 388     if (ret == -1) {
 389         return -s->get_error(s);
 390     }
 391     return 0;
 392 }
 393
 394 int migrate_fd_close(MigrationState *s)
 395 {
 396     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 397     return s->close(s);
 398 }
 399
 400 void add_migration_state_change_notifier(Notifier *notify)
 401 {
 402     notifier_list_add(&migration_state_notifiers, notify);
 403 }
 404
 405 void remove_migration_state_change_notifier(Notifier *notify)
 406 {
 407     notifier_remove(notify);
 408 }
 409
 410 bool migration_is_active(MigrationState *s)
 411 {
 412     return s->state == MIG_STATE_ACTIVE;
 413 }
 414
 415 bool migration_has_finished(MigrationState *s)
 416 {
 417     return s->state == MIG_STATE_COMPLETED;
 418 }
 419
 420 bool migration_has_failed(MigrationState *s)
 421 {
 422     return (s->state == MIG_STATE_CANCELLED ||
 423             s->state == MIG_STATE_ERROR);
 424 }
 425
 426 void migrate_fd_connect(MigrationState *s)
 427 {
 428     int ret;
 429
 430     s->state = MIG_STATE_ACTIVE;
 431     s->file = qemu_fopen_ops_buffered(s);
 432
 433     DPRINTF("beginning savevm\n");
 434     ret = qemu_savevm_state_begin(s->file, &s->params);
 435     if (ret < 0) {
 436         DPRINTF("failed, %d\n", ret);
 437         migrate_fd_error(s);
 438         return;
 439     }
 440     migrate_fd_put_ready(s);
 441 }
 442
 443 static MigrationState *migrate_init(const MigrationParams *params)
 444 {
 445     MigrationState *s = migrate_get_current();
 446     int64_t bandwidth_limit = s->bandwidth_limit;
 447     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 448     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 449
 450     memcpy(enabled_capabilities, s->enabled_capabilities,
 451            sizeof(enabled_capabilities));
 452
 453     memset(s, 0, sizeof(*s));
 454     s->bandwidth_limit = bandwidth_limit;
 455     s->params = *params;
 456     memcpy(s->enabled_capabilities, enabled_capabilities,
 457            sizeof(enabled_capabilities));
 458     s->xbzrle_cache_size = xbzrle_cache_size;
 459
 460     s->bandwidth_limit = bandwidth_limit;
 461     s->state = MIG_STATE_SETUP;
 462     s->total_time = qemu_get_clock_ms(rt_clock);
 463
 464     return s;
 465 }
 466
 467 static GSList *migration_blockers;
 468
 469 void migrate_add_blocker(Error *reason)
 470 {
 471     migration_blockers = g_slist_prepend(migration_blockers, reason);
 472 }
 473
 474 void migrate_del_blocker(Error *reason)
 475 {
 476     migration_blockers = g_slist_remove(migration_blockers, reason);
 477 }
 478
 479 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 480                  bool has_inc, bool inc, bool has_detach, bool detach,
 481                  Error **errp)
 482 {
 483     Error *local_err = NULL;
 484     MigrationState *s = migrate_get_current();
 485     MigrationParams params;
 486     const char *p;
 487
 488     params.blk = blk;
 489     params.shared = inc;
 490
 491     if (s->state == MIG_STATE_ACTIVE) {
 492         error_set(errp, QERR_MIGRATION_ACTIVE);
 493         return;
 494     }
 495
 496     if (qemu_savevm_state_blocked(errp)) {
 497         return;
 498     }
 499
 500     if (migration_blockers) {
 501         *errp = error_copy(migration_blockers->data);
 502         return;
 503     }
 504
 505     s = migrate_init(&params);
 506
 507     if (strstart(uri, "tcp:", &p)) {
 508         tcp_start_outgoing_migration(s, p, &local_err);
 509 #if !defined(WIN32)
 510     } else if (strstart(uri, "exec:", &p)) {
 511         exec_start_outgoing_migration(s, p, &local_err);
 512     } else if (strstart(uri, "unix:", &p)) {
 513         unix_start_outgoing_migration(s, p, &local_err);
 514     } else if (strstart(uri, "fd:", &p)) {
 515         fd_start_outgoing_migration(s, p, &local_err);
 516 #endif
 517     } else {
 518         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "uri", "a valid migration protocol");
 519         return;
 520     }
 521
 522     if (local_err) {
 523         migrate_fd_error(s);
 524         error_propagate(errp, local_err);
 525         return;
 526     }
 527
 528     notifier_list_notify(&migration_state_notifiers, s);
 529 }
 530
 531 void qmp_migrate_cancel(Error **errp)
 532 {
 533     migrate_fd_cancel(migrate_get_current());
 534 }
 535
 536 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 537 {
 538     MigrationState *s = migrate_get_current();
 539
 540     /* Check for truncation */
 541     if (value != (size_t)value) {
 542         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 543                   "exceeding address space");
 544         return;
 545     }
 546
 547     s->xbzrle_cache_size = xbzrle_cache_resize(value);
 548 }
 549
 550 int64_t qmp_query_migrate_cache_size(Error **errp)
 551 {
 552     return migrate_xbzrle_cache_size();
 553 }
 554
 555 void qmp_migrate_set_speed(int64_t value, Error **errp)
 556 {
 557     MigrationState *s;
 558
 559     if (value < 0) {
 560         value = 0;
 561     }
 562
 563     s = migrate_get_current();
 564     s->bandwidth_limit = value;
 565     qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
 566 }
 567
 568 void qmp_migrate_set_downtime(double value, Error **errp)
 569 {
 570     value *= 1e9;
 571     value = MAX(0, MIN(UINT64_MAX, value));
 572     max_downtime = (uint64_t)value;
 573 }
 574
 575 int migrate_use_xbzrle(void)
 576 {
 577     MigrationState *s;
 578
 579     s = migrate_get_current();
 580
 581     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 582 }
 583
 584 int64_t migrate_xbzrle_cache_size(void)
 585 {
 586     MigrationState *s;
 587
 588     s = migrate_get_current();
 589
 590     return s->xbzrle_cache_size;
 591 }