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->file) {
 247         DPRINTF("closing file\n");
 248         ret = qemu_fclose(s->file);
 249         s->file = NULL;
 250     }
 251
 252     migrate_fd_close(s);
 253     return ret;
 254 }
 255
 256 void migrate_fd_error(MigrationState *s)
 257 {
 258     DPRINTF("setting error state\n");
 259     s->state = MIG_STATE_ERROR;
 260     notifier_list_notify(&migration_state_notifiers, s);
 261     migrate_fd_cleanup(s);
 262 }
 263
 264 static void migrate_fd_completed(MigrationState *s)
 265 {
 266     DPRINTF("setting completed state\n");
 267     if (migrate_fd_cleanup(s) < 0) {
 268         s->state = MIG_STATE_ERROR;
 269     } else {
 270         s->state = MIG_STATE_COMPLETED;
 271         runstate_set(RUN_STATE_POSTMIGRATE);
 272     }
 273     notifier_list_notify(&migration_state_notifiers, s);
 274 }
 275
 276 static void migrate_fd_put_notify(void *opaque)
 277 {
 278     MigrationState *s = opaque;
 279     int ret;
 280
 281     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 282     ret = qemu_file_put_notify(s->file);
 283     if (ret) {
 284         migrate_fd_error(s);
 285     }
 286 }
 287
 288 ssize_t migrate_fd_put_buffer(MigrationState *s, const void *data,
 289                               size_t size)
 290 {
 291     ssize_t ret;
 292
 293     if (s->state != MIG_STATE_ACTIVE) {
 294         return -EIO;
 295     }
 296
 297     do {
 298         ret = s->write(s, data, size);
 299     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 300
 301     if (ret == -1)
 302         ret = -(s->get_error(s));
 303
 304     if (ret == -EAGAIN) {
 305         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 306     }
 307
 308     return ret;
 309 }
 310
 311 void migrate_fd_put_ready(MigrationState *s)
 312 {
 313     int ret;
 314
 315     if (s->state != MIG_STATE_ACTIVE) {
 316         DPRINTF("put_ready returning because of non-active state\n");
 317         return;
 318     }
 319
 320     DPRINTF("iterate\n");
 321     ret = qemu_savevm_state_iterate(s->file);
 322     if (ret < 0) {
 323         migrate_fd_error(s);
 324     } else if (ret == 1) {
 325         int old_vm_running = runstate_is_running();
 326         int64_t start_time, end_time;
 327
 328         DPRINTF("done iterating\n");
 329         start_time = qemu_get_clock_ms(rt_clock);
 330         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 331         vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 332
 333         if (qemu_savevm_state_complete(s->file) < 0) {
 334             migrate_fd_error(s);
 335         } else {
 336             migrate_fd_completed(s);
 337         }
 338         end_time = qemu_get_clock_ms(rt_clock);
 339         s->total_time = end_time - s->total_time;
 340         s->downtime = end_time - start_time;
 341         if (s->state != MIG_STATE_COMPLETED) {
 342             if (old_vm_running) {
 343                 vm_start();
 344             }
 345         }
 346     }
 347 }
 348
 349 static void migrate_fd_cancel(MigrationState *s)
 350 {
 351     if (s->state != MIG_STATE_ACTIVE)
 352         return;
 353
 354     DPRINTF("cancelling migration\n");
 355
 356     s->state = MIG_STATE_CANCELLED;
 357     notifier_list_notify(&migration_state_notifiers, s);
 358     qemu_savevm_state_cancel(s->file);
 359
 360     migrate_fd_cleanup(s);
 361 }
 362
 363 int migrate_fd_wait_for_unfreeze(MigrationState *s)
 364 {
 365     int ret;
 366
 367     DPRINTF("wait for unfreeze\n");
 368     if (s->state != MIG_STATE_ACTIVE)
 369         return -EINVAL;
 370
 371     do {
 372         fd_set wfds;
 373
 374         FD_ZERO(&wfds);
 375         FD_SET(s->fd, &wfds);
 376
 377         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 378     } while (ret == -1 && (s->get_error(s)) == EINTR);
 379
 380     if (ret == -1) {
 381         return -s->get_error(s);
 382     }
 383     return 0;
 384 }
 385
 386 int migrate_fd_close(MigrationState *s)
 387 {
 388     int rc = 0;
 389     if (s->fd != -1) {
 390         qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 391         rc = s->close(s);
 392         s->fd = -1;
 393     }
 394     return rc;
 395 }
 396
 397 void add_migration_state_change_notifier(Notifier *notify)
 398 {
 399     notifier_list_add(&migration_state_notifiers, notify);
 400 }
 401
 402 void remove_migration_state_change_notifier(Notifier *notify)
 403 {
 404     notifier_remove(notify);
 405 }
 406
 407 bool migration_is_active(MigrationState *s)
 408 {
 409     return s->state == MIG_STATE_ACTIVE;
 410 }
 411
 412 bool migration_has_finished(MigrationState *s)
 413 {
 414     return s->state == MIG_STATE_COMPLETED;
 415 }
 416
 417 bool migration_has_failed(MigrationState *s)
 418 {
 419     return (s->state == MIG_STATE_CANCELLED ||
 420             s->state == MIG_STATE_ERROR);
 421 }
 422
 423 void migrate_fd_connect(MigrationState *s)
 424 {
 425     int ret;
 426
 427     s->state = MIG_STATE_ACTIVE;
 428     s->file = qemu_fopen_ops_buffered(s);
 429
 430     DPRINTF("beginning savevm\n");
 431     ret = qemu_savevm_state_begin(s->file, &s->params);
 432     if (ret < 0) {
 433         DPRINTF("failed, %d\n", ret);
 434         migrate_fd_error(s);
 435         return;
 436     }
 437     migrate_fd_put_ready(s);
 438 }
 439
 440 static MigrationState *migrate_init(const MigrationParams *params)
 441 {
 442     MigrationState *s = migrate_get_current();
 443     int64_t bandwidth_limit = s->bandwidth_limit;
 444     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 445     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 446
 447     memcpy(enabled_capabilities, s->enabled_capabilities,
 448            sizeof(enabled_capabilities));
 449
 450     memset(s, 0, sizeof(*s));
 451     s->bandwidth_limit = bandwidth_limit;
 452     s->params = *params;
 453     memcpy(s->enabled_capabilities, enabled_capabilities,
 454            sizeof(enabled_capabilities));
 455     s->xbzrle_cache_size = xbzrle_cache_size;
 456
 457     s->bandwidth_limit = bandwidth_limit;
 458     s->state = MIG_STATE_SETUP;
 459     s->total_time = qemu_get_clock_ms(rt_clock);
 460
 461     return s;
 462 }
 463
 464 static GSList *migration_blockers;
 465
 466 void migrate_add_blocker(Error *reason)
 467 {
 468     migration_blockers = g_slist_prepend(migration_blockers, reason);
 469 }
 470
 471 void migrate_del_blocker(Error *reason)
 472 {
 473     migration_blockers = g_slist_remove(migration_blockers, reason);
 474 }
 475
 476 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 477                  bool has_inc, bool inc, bool has_detach, bool detach,
 478                  Error **errp)
 479 {
 480     Error *local_err = NULL;
 481     MigrationState *s = migrate_get_current();
 482     MigrationParams params;
 483     const char *p;
 484
 485     params.blk = blk;
 486     params.shared = inc;
 487
 488     if (s->state == MIG_STATE_ACTIVE) {
 489         error_set(errp, QERR_MIGRATION_ACTIVE);
 490         return;
 491     }
 492
 493     if (qemu_savevm_state_blocked(errp)) {
 494         return;
 495     }
 496
 497     if (migration_blockers) {
 498         *errp = error_copy(migration_blockers->data);
 499         return;
 500     }
 501
 502     s = migrate_init(&params);
 503
 504     if (strstart(uri, "tcp:", &p)) {
 505         tcp_start_outgoing_migration(s, p, &local_err);
 506 #if !defined(WIN32)
 507     } else if (strstart(uri, "exec:", &p)) {
 508         exec_start_outgoing_migration(s, p, &local_err);
 509     } else if (strstart(uri, "unix:", &p)) {
 510         unix_start_outgoing_migration(s, p, &local_err);
 511     } else if (strstart(uri, "fd:", &p)) {
 512         fd_start_outgoing_migration(s, p, &local_err);
 513 #endif
 514     } else {
 515         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "uri", "a valid migration protocol");
 516         return;
 517     }
 518
 519     if (local_err) {
 520         migrate_fd_error(s);
 521         error_propagate(errp, local_err);
 522         return;
 523     }
 524
 525     notifier_list_notify(&migration_state_notifiers, s);
 526 }
 527
 528 void qmp_migrate_cancel(Error **errp)
 529 {
 530     migrate_fd_cancel(migrate_get_current());
 531 }
 532
 533 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 534 {
 535     MigrationState *s = migrate_get_current();
 536
 537     /* Check for truncation */
 538     if (value != (size_t)value) {
 539         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 540                   "exceeding address space");
 541         return;
 542     }
 543
 544     s->xbzrle_cache_size = xbzrle_cache_resize(value);
 545 }
 546
 547 int64_t qmp_query_migrate_cache_size(Error **errp)
 548 {
 549     return migrate_xbzrle_cache_size();
 550 }
 551
 552 void qmp_migrate_set_speed(int64_t value, Error **errp)
 553 {
 554     MigrationState *s;
 555
 556     if (value < 0) {
 557         value = 0;
 558     }
 559
 560     s = migrate_get_current();
 561     s->bandwidth_limit = value;
 562     qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
 563 }
 564
 565 void qmp_migrate_set_downtime(double value, Error **errp)
 566 {
 567     value *= 1e9;
 568     value = MAX(0, MIN(UINT64_MAX, value));
 569     max_downtime = (uint64_t)value;
 570 }
 571
 572 int migrate_use_xbzrle(void)
 573 {
 574     MigrationState *s;
 575
 576     s = migrate_get_current();
 577
 578     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 579 }
 580
 581 int64_t migrate_xbzrle_cache_size(void)
 582 {
 583     MigrationState *s;
 584
 585     s = migrate_get_current();
 586
 587     return s->xbzrle_cache_size;
 588 }