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 int qemu_start_incoming_migration(const char *uri, Error **errp)
  68 {
  69     const char *p;
  70     int ret;
  71
  72     if (strstart(uri, "tcp:", &p))
  73         ret = tcp_start_incoming_migration(p, errp);
  74 #if !defined(WIN32)
  75     else if (strstart(uri, "exec:", &p))
  76         ret =  exec_start_incoming_migration(p);
  77     else if (strstart(uri, "unix:", &p))
  78         ret = unix_start_incoming_migration(p);
  79     else if (strstart(uri, "fd:", &p))
  80         ret = fd_start_incoming_migration(p);
  81 #endif
  82     else {
  83         fprintf(stderr, "unknown migration protocol: %s\n", uri);
  84         ret = -EPROTONOSUPPORT;
  85     }
  86     return ret;
  87 }
  88
  89 void process_incoming_migration(QEMUFile *f)
  90 {
  91     if (qemu_loadvm_state(f) < 0) {
  92         fprintf(stderr, "load of migration failed\n");
  93         exit(0);
  94     }
  95     qemu_announce_self();
  96     DPRINTF("successfully loaded vm state\n");
  97
  98     bdrv_clear_incoming_migration_all();
  99     /* Make sure all file formats flush their mutable metadata */
 100     bdrv_invalidate_cache_all();
 101
 102     if (autostart) {
 103         vm_start();
 104     } else {
 105         runstate_set(RUN_STATE_PRELAUNCH);
 106     }
 107 }
 108
 109 /* amount of nanoseconds we are willing to wait for migration to be down.
 110  * the choice of nanoseconds is because it is the maximum resolution that
 111  * get_clock() can achieve. It is an internal measure. All user-visible
 112  * units must be in seconds */
 113 static uint64_t max_downtime = 30000000;
 114
 115 uint64_t migrate_max_downtime(void)
 116 {
 117     return max_downtime;
 118 }
 119
 120 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
 121 {
 122     MigrationCapabilityStatusList *head = NULL;
 123     MigrationCapabilityStatusList *caps;
 124     MigrationState *s = migrate_get_current();
 125     int i;
 126
 127     for (i = 0; i < MIGRATION_CAPABILITY_MAX; i++) {
 128         if (head == NULL) {
 129             head = g_malloc0(sizeof(*caps));
 130             caps = head;
 131         } else {
 132             caps->next = g_malloc0(sizeof(*caps));
 133             caps = caps->next;
 134         }
 135         caps->value =
 136             g_malloc(sizeof(*caps->value));
 137         caps->value->capability = i;
 138         caps->value->state = s->enabled_capabilities[i];
 139     }
 140
 141     return head;
 142 }
 143
 144 static void get_xbzrle_cache_stats(MigrationInfo *info)
 145 {
 146     if (migrate_use_xbzrle()) {
 147         info->has_xbzrle_cache = true;
 148         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
 149         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
 150         info->xbzrle_cache->bytes = xbzrle_mig_bytes_transferred();
 151         info->xbzrle_cache->pages = xbzrle_mig_pages_transferred();
 152         info->xbzrle_cache->cache_miss = xbzrle_mig_pages_cache_miss();
 153         info->xbzrle_cache->overflow = xbzrle_mig_pages_overflow();
 154     }
 155 }
 156
 157 MigrationInfo *qmp_query_migrate(Error **errp)
 158 {
 159     MigrationInfo *info = g_malloc0(sizeof(*info));
 160     MigrationState *s = migrate_get_current();
 161
 162     switch (s->state) {
 163     case MIG_STATE_SETUP:
 164         /* no migration has happened ever */
 165         break;
 166     case MIG_STATE_ACTIVE:
 167         info->has_status = true;
 168         info->status = g_strdup("active");
 169         info->has_total_time = true;
 170         info->total_time = qemu_get_clock_ms(rt_clock)
 171             - s->total_time;
 172         info->has_expected_downtime = true;
 173         info->expected_downtime = s->expected_downtime;
 174
 175         info->has_ram = true;
 176         info->ram = g_malloc0(sizeof(*info->ram));
 177         info->ram->transferred = ram_bytes_transferred();
 178         info->ram->remaining = ram_bytes_remaining();
 179         info->ram->total = ram_bytes_total();
 180         info->ram->duplicate = dup_mig_pages_transferred();
 181         info->ram->normal = norm_mig_pages_transferred();
 182         info->ram->normal_bytes = norm_mig_bytes_transferred();
 183         info->ram->dirty_pages_rate = s->dirty_pages_rate;
 184
 185
 186         if (blk_mig_active()) {
 187             info->has_disk = true;
 188             info->disk = g_malloc0(sizeof(*info->disk));
 189             info->disk->transferred = blk_mig_bytes_transferred();
 190             info->disk->remaining = blk_mig_bytes_remaining();
 191             info->disk->total = blk_mig_bytes_total();
 192         }
 193
 194         get_xbzrle_cache_stats(info);
 195         break;
 196     case MIG_STATE_COMPLETED:
 197         get_xbzrle_cache_stats(info);
 198
 199         info->has_status = true;
 200         info->status = g_strdup("completed");
 201         info->total_time = s->total_time;
 202         info->has_downtime = true;
 203         info->downtime = s->downtime;
 204
 205         info->has_ram = true;
 206         info->ram = g_malloc0(sizeof(*info->ram));
 207         info->ram->transferred = ram_bytes_transferred();
 208         info->ram->remaining = 0;
 209         info->ram->total = ram_bytes_total();
 210         info->ram->duplicate = dup_mig_pages_transferred();
 211         info->ram->normal = norm_mig_pages_transferred();
 212         info->ram->normal_bytes = norm_mig_bytes_transferred();
 213         break;
 214     case MIG_STATE_ERROR:
 215         info->has_status = true;
 216         info->status = g_strdup("failed");
 217         break;
 218     case MIG_STATE_CANCELLED:
 219         info->has_status = true;
 220         info->status = g_strdup("cancelled");
 221         break;
 222     }
 223
 224     return info;
 225 }
 226
 227 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
 228                                   Error **errp)
 229 {
 230     MigrationState *s = migrate_get_current();
 231     MigrationCapabilityStatusList *cap;
 232
 233     if (s->state == MIG_STATE_ACTIVE) {
 234         error_set(errp, QERR_MIGRATION_ACTIVE);
 235         return;
 236     }
 237
 238     for (cap = params; cap; cap = cap->next) {
 239         s->enabled_capabilities[cap->value->capability] = cap->value->state;
 240     }
 241 }
 242
 243 /* shared migration helpers */
 244
 245 static int migrate_fd_cleanup(MigrationState *s)
 246 {
 247     int ret = 0;
 248
 249     if (s->fd != -1) {
 250         qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 251     }
 252
 253     if (s->file) {
 254         DPRINTF("closing file\n");
 255         ret = qemu_fclose(s->file);
 256         s->file = NULL;
 257     }
 258
 259     if (s->fd != -1) {
 260         close(s->fd);
 261         s->fd = -1;
 262     }
 263
 264     return ret;
 265 }
 266
 267 void migrate_fd_error(MigrationState *s)
 268 {
 269     DPRINTF("setting error state\n");
 270     s->state = MIG_STATE_ERROR;
 271     notifier_list_notify(&migration_state_notifiers, s);
 272     migrate_fd_cleanup(s);
 273 }
 274
 275 static void migrate_fd_completed(MigrationState *s)
 276 {
 277     DPRINTF("setting completed state\n");
 278     if (migrate_fd_cleanup(s) < 0) {
 279         s->state = MIG_STATE_ERROR;
 280     } else {
 281         s->state = MIG_STATE_COMPLETED;
 282         runstate_set(RUN_STATE_POSTMIGRATE);
 283     }
 284     notifier_list_notify(&migration_state_notifiers, s);
 285 }
 286
 287 static void migrate_fd_put_notify(void *opaque)
 288 {
 289     MigrationState *s = opaque;
 290
 291     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 292     qemu_file_put_notify(s->file);
 293     if (s->file && qemu_file_get_error(s->file)) {
 294         migrate_fd_error(s);
 295     }
 296 }
 297
 298 static ssize_t migrate_fd_put_buffer(void *opaque, const void *data,
 299                                      size_t size)
 300 {
 301     MigrationState *s = opaque;
 302     ssize_t ret;
 303
 304     if (s->state != MIG_STATE_ACTIVE) {
 305         return -EIO;
 306     }
 307
 308     do {
 309         ret = s->write(s, data, size);
 310     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 311
 312     if (ret == -1)
 313         ret = -(s->get_error(s));
 314
 315     if (ret == -EAGAIN) {
 316         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 317     }
 318
 319     return ret;
 320 }
 321
 322 static void migrate_fd_put_ready(void *opaque)
 323 {
 324     MigrationState *s = opaque;
 325     int ret;
 326
 327     if (s->state != MIG_STATE_ACTIVE) {
 328         DPRINTF("put_ready returning because of non-active state\n");
 329         return;
 330     }
 331
 332     DPRINTF("iterate\n");
 333     ret = qemu_savevm_state_iterate(s->file);
 334     if (ret < 0) {
 335         migrate_fd_error(s);
 336     } else if (ret == 1) {
 337         int old_vm_running = runstate_is_running();
 338         int64_t start_time, end_time;
 339
 340         DPRINTF("done iterating\n");
 341         start_time = qemu_get_clock_ms(rt_clock);
 342         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 343         vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 344
 345         if (qemu_savevm_state_complete(s->file) < 0) {
 346             migrate_fd_error(s);
 347         } else {
 348             migrate_fd_completed(s);
 349         }
 350         end_time = qemu_get_clock_ms(rt_clock);
 351         s->total_time = end_time - s->total_time;
 352         s->downtime = end_time - start_time;
 353         if (s->state != MIG_STATE_COMPLETED) {
 354             if (old_vm_running) {
 355                 vm_start();
 356             }
 357         }
 358     }
 359 }
 360
 361 static void migrate_fd_cancel(MigrationState *s)
 362 {
 363     if (s->state != MIG_STATE_ACTIVE)
 364         return;
 365
 366     DPRINTF("cancelling migration\n");
 367
 368     s->state = MIG_STATE_CANCELLED;
 369     notifier_list_notify(&migration_state_notifiers, s);
 370     qemu_savevm_state_cancel(s->file);
 371
 372     migrate_fd_cleanup(s);
 373 }
 374
 375 static void migrate_fd_wait_for_unfreeze(void *opaque)
 376 {
 377     MigrationState *s = opaque;
 378     int ret;
 379
 380     DPRINTF("wait for unfreeze\n");
 381     if (s->state != MIG_STATE_ACTIVE)
 382         return;
 383
 384     do {
 385         fd_set wfds;
 386
 387         FD_ZERO(&wfds);
 388         FD_SET(s->fd, &wfds);
 389
 390         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 391     } while (ret == -1 && (s->get_error(s)) == EINTR);
 392
 393     if (ret == -1) {
 394         qemu_file_set_error(s->file, -s->get_error(s));
 395     }
 396 }
 397
 398 static int migrate_fd_close(void *opaque)
 399 {
 400     MigrationState *s = opaque;
 401
 402     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 403     return s->close(s);
 404 }
 405
 406 void add_migration_state_change_notifier(Notifier *notify)
 407 {
 408     notifier_list_add(&migration_state_notifiers, notify);
 409 }
 410
 411 void remove_migration_state_change_notifier(Notifier *notify)
 412 {
 413     notifier_remove(notify);
 414 }
 415
 416 bool migration_is_active(MigrationState *s)
 417 {
 418     return s->state == MIG_STATE_ACTIVE;
 419 }
 420
 421 bool migration_has_finished(MigrationState *s)
 422 {
 423     return s->state == MIG_STATE_COMPLETED;
 424 }
 425
 426 bool migration_has_failed(MigrationState *s)
 427 {
 428     return (s->state == MIG_STATE_CANCELLED ||
 429             s->state == MIG_STATE_ERROR);
 430 }
 431
 432 void migrate_fd_connect(MigrationState *s)
 433 {
 434     int ret;
 435
 436     s->state = MIG_STATE_ACTIVE;
 437     s->file = qemu_fopen_ops_buffered(s,
 438                                       s->bandwidth_limit,
 439                                       migrate_fd_put_buffer,
 440                                       migrate_fd_put_ready,
 441                                       migrate_fd_wait_for_unfreeze,
 442                                       migrate_fd_close);
 443
 444     DPRINTF("beginning savevm\n");
 445     ret = qemu_savevm_state_begin(s->file, &s->params);
 446     if (ret < 0) {
 447         DPRINTF("failed, %d\n", ret);
 448         migrate_fd_error(s);
 449         return;
 450     }
 451     migrate_fd_put_ready(s);
 452 }
 453
 454 static MigrationState *migrate_init(const MigrationParams *params)
 455 {
 456     MigrationState *s = migrate_get_current();
 457     int64_t bandwidth_limit = s->bandwidth_limit;
 458     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 459     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 460
 461     memcpy(enabled_capabilities, s->enabled_capabilities,
 462            sizeof(enabled_capabilities));
 463
 464     memset(s, 0, sizeof(*s));
 465     s->bandwidth_limit = bandwidth_limit;
 466     s->params = *params;
 467     memcpy(s->enabled_capabilities, enabled_capabilities,
 468            sizeof(enabled_capabilities));
 469     s->xbzrle_cache_size = xbzrle_cache_size;
 470
 471     s->bandwidth_limit = bandwidth_limit;
 472     s->state = MIG_STATE_SETUP;
 473     s->total_time = qemu_get_clock_ms(rt_clock);
 474
 475     return s;
 476 }
 477
 478 static GSList *migration_blockers;
 479
 480 void migrate_add_blocker(Error *reason)
 481 {
 482     migration_blockers = g_slist_prepend(migration_blockers, reason);
 483 }
 484
 485 void migrate_del_blocker(Error *reason)
 486 {
 487     migration_blockers = g_slist_remove(migration_blockers, reason);
 488 }
 489
 490 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 491                  bool has_inc, bool inc, bool has_detach, bool detach,
 492                  Error **errp)
 493 {
 494     MigrationState *s = migrate_get_current();
 495     MigrationParams params;
 496     const char *p;
 497     int ret;
 498
 499     params.blk = blk;
 500     params.shared = inc;
 501
 502     if (s->state == MIG_STATE_ACTIVE) {
 503         error_set(errp, QERR_MIGRATION_ACTIVE);
 504         return;
 505     }
 506
 507     if (qemu_savevm_state_blocked(errp)) {
 508         return;
 509     }
 510
 511     if (migration_blockers) {
 512         *errp = error_copy(migration_blockers->data);
 513         return;
 514     }
 515
 516     s = migrate_init(&params);
 517
 518     if (strstart(uri, "tcp:", &p)) {
 519         ret = tcp_start_outgoing_migration(s, p, errp);
 520 #if !defined(WIN32)
 521     } else if (strstart(uri, "exec:", &p)) {
 522         ret = exec_start_outgoing_migration(s, p);
 523     } else if (strstart(uri, "unix:", &p)) {
 524         ret = unix_start_outgoing_migration(s, p);
 525     } else if (strstart(uri, "fd:", &p)) {
 526         ret = fd_start_outgoing_migration(s, p);
 527 #endif
 528     } else {
 529         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "uri", "a valid migration protocol");
 530         return;
 531     }
 532
 533     if (ret < 0) {
 534         if (!error_is_set(errp)) {
 535             DPRINTF("migration failed: %s\n", strerror(-ret));
 536             /* FIXME: we should return meaningful errors */
 537             error_set(errp, QERR_UNDEFINED_ERROR);
 538         }
 539         return;
 540     }
 541
 542     notifier_list_notify(&migration_state_notifiers, s);
 543 }
 544
 545 void qmp_migrate_cancel(Error **errp)
 546 {
 547     migrate_fd_cancel(migrate_get_current());
 548 }
 549
 550 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 551 {
 552     MigrationState *s = migrate_get_current();
 553
 554     /* Check for truncation */
 555     if (value != (size_t)value) {
 556         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 557                   "exceeding address space");
 558         return;
 559     }
 560
 561     s->xbzrle_cache_size = xbzrle_cache_resize(value);
 562 }
 563
 564 int64_t qmp_query_migrate_cache_size(Error **errp)
 565 {
 566     return migrate_xbzrle_cache_size();
 567 }
 568
 569 void qmp_migrate_set_speed(int64_t value, Error **errp)
 570 {
 571     MigrationState *s;
 572
 573     if (value < 0) {
 574         value = 0;
 575     }
 576
 577     s = migrate_get_current();
 578     s->bandwidth_limit = value;
 579     qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
 580 }
 581
 582 void qmp_migrate_set_downtime(double value, Error **errp)
 583 {
 584     value *= 1e9;
 585     value = MAX(0, MIN(UINT64_MAX, value));
 586     max_downtime = (uint64_t)value;
 587 }
 588
 589 int migrate_use_xbzrle(void)
 590 {
 591     MigrationState *s;
 592
 593     s = migrate_get_current();
 594
 595     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 596 }
 597
 598 int64_t migrate_xbzrle_cache_size(void)
 599 {
 600     MigrationState *s;
 601
 602     s = migrate_get_current();
 603
 604     return s->xbzrle_cache_size;
 605 }