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     int ret;
 291
 292     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 293     ret = qemu_file_put_notify(s->file);
 294     if (ret) {
 295         migrate_fd_error(s);
 296     }
 297 }
 298
 299 ssize_t migrate_fd_put_buffer(MigrationState *s, const void *data,
 300                               size_t size)
 301 {
 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 void migrate_fd_put_ready(MigrationState *s)
 323 {
 324     int ret;
 325
 326     if (s->state != MIG_STATE_ACTIVE) {
 327         DPRINTF("put_ready returning because of non-active state\n");
 328         return;
 329     }
 330
 331     DPRINTF("iterate\n");
 332     ret = qemu_savevm_state_iterate(s->file);
 333     if (ret < 0) {
 334         migrate_fd_error(s);
 335     } else if (ret == 1) {
 336         int old_vm_running = runstate_is_running();
 337         int64_t start_time, end_time;
 338
 339         DPRINTF("done iterating\n");
 340         start_time = qemu_get_clock_ms(rt_clock);
 341         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 342         vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 343
 344         if (qemu_savevm_state_complete(s->file) < 0) {
 345             migrate_fd_error(s);
 346         } else {
 347             migrate_fd_completed(s);
 348         }
 349         end_time = qemu_get_clock_ms(rt_clock);
 350         s->total_time = end_time - s->total_time;
 351         s->downtime = end_time - start_time;
 352         if (s->state != MIG_STATE_COMPLETED) {
 353             if (old_vm_running) {
 354                 vm_start();
 355             }
 356         }
 357     }
 358 }
 359
 360 static void migrate_fd_cancel(MigrationState *s)
 361 {
 362     if (s->state != MIG_STATE_ACTIVE)
 363         return;
 364
 365     DPRINTF("cancelling migration\n");
 366
 367     s->state = MIG_STATE_CANCELLED;
 368     notifier_list_notify(&migration_state_notifiers, s);
 369     qemu_savevm_state_cancel(s->file);
 370
 371     migrate_fd_cleanup(s);
 372 }
 373
 374 int migrate_fd_wait_for_unfreeze(MigrationState *s)
 375 {
 376     int ret;
 377
 378     DPRINTF("wait for unfreeze\n");
 379     if (s->state != MIG_STATE_ACTIVE)
 380         return -EINVAL;
 381
 382     do {
 383         fd_set wfds;
 384
 385         FD_ZERO(&wfds);
 386         FD_SET(s->fd, &wfds);
 387
 388         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 389     } while (ret == -1 && (s->get_error(s)) == EINTR);
 390
 391     if (ret == -1) {
 392         return -s->get_error(s);
 393     }
 394     return 0;
 395 }
 396
 397 int migrate_fd_close(MigrationState *s)
 398 {
 399     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 400     return s->close(s);
 401 }
 402
 403 void add_migration_state_change_notifier(Notifier *notify)
 404 {
 405     notifier_list_add(&migration_state_notifiers, notify);
 406 }
 407
 408 void remove_migration_state_change_notifier(Notifier *notify)
 409 {
 410     notifier_remove(notify);
 411 }
 412
 413 bool migration_is_active(MigrationState *s)
 414 {
 415     return s->state == MIG_STATE_ACTIVE;
 416 }
 417
 418 bool migration_has_finished(MigrationState *s)
 419 {
 420     return s->state == MIG_STATE_COMPLETED;
 421 }
 422
 423 bool migration_has_failed(MigrationState *s)
 424 {
 425     return (s->state == MIG_STATE_CANCELLED ||
 426             s->state == MIG_STATE_ERROR);
 427 }
 428
 429 void migrate_fd_connect(MigrationState *s)
 430 {
 431     int ret;
 432
 433     s->state = MIG_STATE_ACTIVE;
 434     s->file = qemu_fopen_ops_buffered(s);
 435
 436     DPRINTF("beginning savevm\n");
 437     ret = qemu_savevm_state_begin(s->file, &s->params);
 438     if (ret < 0) {
 439         DPRINTF("failed, %d\n", ret);
 440         migrate_fd_error(s);
 441         return;
 442     }
 443     migrate_fd_put_ready(s);
 444 }
 445
 446 static MigrationState *migrate_init(const MigrationParams *params)
 447 {
 448     MigrationState *s = migrate_get_current();
 449     int64_t bandwidth_limit = s->bandwidth_limit;
 450     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 451     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 452
 453     memcpy(enabled_capabilities, s->enabled_capabilities,
 454            sizeof(enabled_capabilities));
 455
 456     memset(s, 0, sizeof(*s));
 457     s->bandwidth_limit = bandwidth_limit;
 458     s->params = *params;
 459     memcpy(s->enabled_capabilities, enabled_capabilities,
 460            sizeof(enabled_capabilities));
 461     s->xbzrle_cache_size = xbzrle_cache_size;
 462
 463     s->bandwidth_limit = bandwidth_limit;
 464     s->state = MIG_STATE_SETUP;
 465     s->total_time = qemu_get_clock_ms(rt_clock);
 466
 467     return s;
 468 }
 469
 470 static GSList *migration_blockers;
 471
 472 void migrate_add_blocker(Error *reason)
 473 {
 474     migration_blockers = g_slist_prepend(migration_blockers, reason);
 475 }
 476
 477 void migrate_del_blocker(Error *reason)
 478 {
 479     migration_blockers = g_slist_remove(migration_blockers, reason);
 480 }
 481
 482 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 483                  bool has_inc, bool inc, bool has_detach, bool detach,
 484                  Error **errp)
 485 {
 486     Error *local_err = NULL;
 487     MigrationState *s = migrate_get_current();
 488     MigrationParams params;
 489     const char *p;
 490     int ret;
 491
 492     params.blk = blk;
 493     params.shared = inc;
 494
 495     if (s->state == MIG_STATE_ACTIVE) {
 496         error_set(errp, QERR_MIGRATION_ACTIVE);
 497         return;
 498     }
 499
 500     if (qemu_savevm_state_blocked(errp)) {
 501         return;
 502     }
 503
 504     if (migration_blockers) {
 505         *errp = error_copy(migration_blockers->data);
 506         return;
 507     }
 508
 509     s = migrate_init(&params);
 510
 511     if (strstart(uri, "tcp:", &p)) {
 512         ret = tcp_start_outgoing_migration(s, p, &local_err);
 513 #if !defined(WIN32)
 514     } else if (strstart(uri, "exec:", &p)) {
 515         ret = exec_start_outgoing_migration(s, p);
 516     } else if (strstart(uri, "unix:", &p)) {
 517         ret = unix_start_outgoing_migration(s, p);
 518     } else if (strstart(uri, "fd:", &p)) {
 519         ret = fd_start_outgoing_migration(s, p);
 520 #endif
 521     } else {
 522         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "uri", "a valid migration protocol");
 523         return;
 524     }
 525
 526     if (ret < 0 || local_err) {
 527         migrate_fd_error(s);
 528         if (!local_err) {
 529             error_set_errno(errp, -ret, QERR_UNDEFINED_ERROR);
 530         } else {
 531             error_propagate(errp, local_err);
 532         }
 533         return;
 534     }
 535
 536     notifier_list_notify(&migration_state_notifiers, s);
 537 }
 538
 539 void qmp_migrate_cancel(Error **errp)
 540 {
 541     migrate_fd_cancel(migrate_get_current());
 542 }
 543
 544 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 545 {
 546     MigrationState *s = migrate_get_current();
 547
 548     /* Check for truncation */
 549     if (value != (size_t)value) {
 550         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 551                   "exceeding address space");
 552         return;
 553     }
 554
 555     s->xbzrle_cache_size = xbzrle_cache_resize(value);
 556 }
 557
 558 int64_t qmp_query_migrate_cache_size(Error **errp)
 559 {
 560     return migrate_xbzrle_cache_size();
 561 }
 562
 563 void qmp_migrate_set_speed(int64_t value, Error **errp)
 564 {
 565     MigrationState *s;
 566
 567     if (value < 0) {
 568         value = 0;
 569     }
 570
 571     s = migrate_get_current();
 572     s->bandwidth_limit = value;
 573     qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
 574 }
 575
 576 void qmp_migrate_set_downtime(double value, Error **errp)
 577 {
 578     value *= 1e9;
 579     value = MAX(0, MIN(UINT64_MAX, value));
 580     max_downtime = (uint64_t)value;
 581 }
 582
 583 int migrate_use_xbzrle(void)
 584 {
 585     MigrationState *s;
 586
 587     s = migrate_get_current();
 588
 589     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 590 }
 591
 592 int64_t migrate_xbzrle_cache_size(void)
 593 {
 594     MigrationState *s;
 595
 596     s = migrate_get_current();
 597
 598     return s->xbzrle_cache_size;
 599 }