migration.c

   1 /*
   2  * QEMU live migration
   3  *
   4  * Copyright IBM, Corp. 2008
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <[email protected]>
   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 static 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
 170         info->has_ram = true;
 171         info->ram = g_malloc0(sizeof(*info->ram));
 172         info->ram->transferred = ram_bytes_transferred();
 173         info->ram->remaining = ram_bytes_remaining();
 174         info->ram->total = ram_bytes_total();
 175         info->ram->total_time = qemu_get_clock_ms(rt_clock)
 176             - s->total_time;
 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
 181         if (blk_mig_active()) {
 182             info->has_disk = true;
 183             info->disk = g_malloc0(sizeof(*info->disk));
 184             info->disk->transferred = blk_mig_bytes_transferred();
 185             info->disk->remaining = blk_mig_bytes_remaining();
 186             info->disk->total = blk_mig_bytes_total();
 187         }
 188
 189         get_xbzrle_cache_stats(info);
 190         break;
 191     case MIG_STATE_COMPLETED:
 192         get_xbzrle_cache_stats(info);
 193
 194         info->has_status = true;
 195         info->status = g_strdup("completed");
 196
 197         info->has_ram = true;
 198         info->ram = g_malloc0(sizeof(*info->ram));
 199         info->ram->transferred = ram_bytes_transferred();
 200         info->ram->remaining = 0;
 201         info->ram->total = ram_bytes_total();
 202         info->ram->total_time = s->total_time;
 203         info->ram->duplicate = dup_mig_pages_transferred();
 204         info->ram->normal = norm_mig_pages_transferred();
 205         info->ram->normal_bytes = norm_mig_bytes_transferred();
 206         break;
 207     case MIG_STATE_ERROR:
 208         info->has_status = true;
 209         info->status = g_strdup("failed");
 210         break;
 211     case MIG_STATE_CANCELLED:
 212         info->has_status = true;
 213         info->status = g_strdup("cancelled");
 214         break;
 215     }
 216
 217     return info;
 218 }
 219
 220 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
 221                                   Error **errp)
 222 {
 223     MigrationState *s = migrate_get_current();
 224     MigrationCapabilityStatusList *cap;
 225
 226     if (s->state == MIG_STATE_ACTIVE) {
 227         error_set(errp, QERR_MIGRATION_ACTIVE);
 228         return;
 229     }
 230
 231     for (cap = params; cap; cap = cap->next) {
 232         s->enabled_capabilities[cap->value->capability] = cap->value->state;
 233     }
 234 }
 235
 236 /* shared migration helpers */
 237
 238 static int migrate_fd_cleanup(MigrationState *s)
 239 {
 240     int ret = 0;
 241
 242     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 243
 244     if (s->file) {
 245         DPRINTF("closing file\n");
 246         ret = qemu_fclose(s->file);
 247         s->file = NULL;
 248     }
 249
 250     if (s->fd != -1) {
 251         close(s->fd);
 252         s->fd = -1;
 253     }
 254
 255     return ret;
 256 }
 257
 258 void migrate_fd_error(MigrationState *s)
 259 {
 260     DPRINTF("setting error state\n");
 261     s->state = MIG_STATE_ERROR;
 262     notifier_list_notify(&migration_state_notifiers, s);
 263     migrate_fd_cleanup(s);
 264 }
 265
 266 static void migrate_fd_completed(MigrationState *s)
 267 {
 268     DPRINTF("setting completed state\n");
 269     if (migrate_fd_cleanup(s) < 0) {
 270         s->state = MIG_STATE_ERROR;
 271     } else {
 272         s->state = MIG_STATE_COMPLETED;
 273         runstate_set(RUN_STATE_POSTMIGRATE);
 274     }
 275     notifier_list_notify(&migration_state_notifiers, s);
 276 }
 277
 278 static void migrate_fd_put_notify(void *opaque)
 279 {
 280     MigrationState *s = opaque;
 281
 282     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 283     qemu_file_put_notify(s->file);
 284     if (s->file && qemu_file_get_error(s->file)) {
 285         migrate_fd_error(s);
 286     }
 287 }
 288
 289 static ssize_t migrate_fd_put_buffer(void *opaque, const void *data,
 290                                      size_t size)
 291 {
 292     MigrationState *s = opaque;
 293     ssize_t ret;
 294
 295     if (s->state != MIG_STATE_ACTIVE) {
 296         return -EIO;
 297     }
 298
 299     do {
 300         ret = s->write(s, data, size);
 301     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 302
 303     if (ret == -1)
 304         ret = -(s->get_error(s));
 305
 306     if (ret == -EAGAIN) {
 307         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 308     }
 309
 310     return ret;
 311 }
 312
 313 static void migrate_fd_put_ready(void *opaque)
 314 {
 315     MigrationState *s = opaque;
 316     int ret;
 317
 318     if (s->state != MIG_STATE_ACTIVE) {
 319         DPRINTF("put_ready returning because of non-active state\n");
 320         return;
 321     }
 322
 323     DPRINTF("iterate\n");
 324     ret = qemu_savevm_state_iterate(s->file);
 325     if (ret < 0) {
 326         migrate_fd_error(s);
 327     } else if (ret == 1) {
 328         int old_vm_running = runstate_is_running();
 329
 330         DPRINTF("done iterating\n");
 331         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 332         vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 333
 334         if (qemu_savevm_state_complete(s->file) < 0) {
 335             migrate_fd_error(s);
 336         } else {
 337             migrate_fd_completed(s);
 338         }
 339         s->total_time = qemu_get_clock_ms(rt_clock) - s->total_time;
 340         if (s->state != MIG_STATE_COMPLETED) {
 341             if (old_vm_running) {
 342                 vm_start();
 343             }
 344         }
 345     }
 346 }
 347
 348 static void migrate_fd_cancel(MigrationState *s)
 349 {
 350     if (s->state != MIG_STATE_ACTIVE)
 351         return;
 352
 353     DPRINTF("cancelling migration\n");
 354
 355     s->state = MIG_STATE_CANCELLED;
 356     notifier_list_notify(&migration_state_notifiers, s);
 357     qemu_savevm_state_cancel(s->file);
 358
 359     migrate_fd_cleanup(s);
 360 }
 361
 362 static void migrate_fd_wait_for_unfreeze(void *opaque)
 363 {
 364     MigrationState *s = opaque;
 365     int ret;
 366
 367     DPRINTF("wait for unfreeze\n");
 368     if (s->state != MIG_STATE_ACTIVE)
 369         return;
 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         qemu_file_set_error(s->file, -s->get_error(s));
 382     }
 383 }
 384
 385 static int migrate_fd_close(void *opaque)
 386 {
 387     MigrationState *s = opaque;
 388
 389     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 390     return s->close(s);
 391 }
 392
 393 void add_migration_state_change_notifier(Notifier *notify)
 394 {
 395     notifier_list_add(&migration_state_notifiers, notify);
 396 }
 397
 398 void remove_migration_state_change_notifier(Notifier *notify)
 399 {
 400     notifier_remove(notify);
 401 }
 402
 403 bool migration_is_active(MigrationState *s)
 404 {
 405     return s->state == MIG_STATE_ACTIVE;
 406 }
 407
 408 bool migration_has_finished(MigrationState *s)
 409 {
 410     return s->state == MIG_STATE_COMPLETED;
 411 }
 412
 413 bool migration_has_failed(MigrationState *s)
 414 {
 415     return (s->state == MIG_STATE_CANCELLED ||
 416             s->state == MIG_STATE_ERROR);
 417 }
 418
 419 void migrate_fd_connect(MigrationState *s)
 420 {
 421     int ret;
 422
 423     s->state = MIG_STATE_ACTIVE;
 424     s->file = qemu_fopen_ops_buffered(s,
 425                                       s->bandwidth_limit,
 426                                       migrate_fd_put_buffer,
 427                                       migrate_fd_put_ready,
 428                                       migrate_fd_wait_for_unfreeze,
 429                                       migrate_fd_close);
 430
 431     DPRINTF("beginning savevm\n");
 432     ret = qemu_savevm_state_begin(s->file, &s->params);
 433     if (ret < 0) {
 434         DPRINTF("failed, %d\n", ret);
 435         migrate_fd_error(s);
 436         return;
 437     }
 438     migrate_fd_put_ready(s);
 439 }
 440
 441 static MigrationState *migrate_init(const MigrationParams *params)
 442 {
 443     MigrationState *s = migrate_get_current();
 444     int64_t bandwidth_limit = s->bandwidth_limit;
 445     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 446     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 447
 448     memcpy(enabled_capabilities, s->enabled_capabilities,
 449            sizeof(enabled_capabilities));
 450
 451     memset(s, 0, sizeof(*s));
 452     s->bandwidth_limit = bandwidth_limit;
 453     s->params = *params;
 454     memcpy(s->enabled_capabilities, enabled_capabilities,
 455            sizeof(enabled_capabilities));
 456     s->xbzrle_cache_size = xbzrle_cache_size;
 457
 458     s->bandwidth_limit = bandwidth_limit;
 459     s->state = MIG_STATE_SETUP;
 460     s->total_time = qemu_get_clock_ms(rt_clock);
 461
 462     return s;
 463 }
 464
 465 static GSList *migration_blockers;
 466
 467 void migrate_add_blocker(Error *reason)
 468 {
 469     migration_blockers = g_slist_prepend(migration_blockers, reason);
 470 }
 471
 472 void migrate_del_blocker(Error *reason)
 473 {
 474     migration_blockers = g_slist_remove(migration_blockers, reason);
 475 }
 476
 477 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 478                  bool has_inc, bool inc, bool has_detach, bool detach,
 479                  Error **errp)
 480 {
 481     MigrationState *s = migrate_get_current();
 482     MigrationParams params;
 483     const char *p;
 484     int ret;
 485
 486     params.blk = blk;
 487     params.shared = inc;
 488
 489     if (s->state == MIG_STATE_ACTIVE) {
 490         error_set(errp, QERR_MIGRATION_ACTIVE);
 491         return;
 492     }
 493
 494     if (qemu_savevm_state_blocked(errp)) {
 495         return;
 496     }
 497
 498     if (migration_blockers) {
 499         *errp = error_copy(migration_blockers->data);
 500         return;
 501     }
 502
 503     s = migrate_init(&params);
 504
 505     if (strstart(uri, "tcp:", &p)) {
 506         ret = tcp_start_outgoing_migration(s, p, errp);
 507 #if !defined(WIN32)
 508     } else if (strstart(uri, "exec:", &p)) {
 509         ret = exec_start_outgoing_migration(s, p);
 510     } else if (strstart(uri, "unix:", &p)) {
 511         ret = unix_start_outgoing_migration(s, p);
 512     } else if (strstart(uri, "fd:", &p)) {
 513         ret = fd_start_outgoing_migration(s, p);
 514 #endif
 515     } else {
 516         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "uri", "a valid migration protocol");
 517         return;
 518     }
 519
 520     if (ret < 0) {
 521         if (!error_is_set(errp)) {
 522             DPRINTF("migration failed: %s\n", strerror(-ret));
 523             /* FIXME: we should return meaningful errors */
 524             error_set(errp, QERR_UNDEFINED_ERROR);
 525         }
 526         return;
 527     }
 528
 529     notifier_list_notify(&migration_state_notifiers, s);
 530 }
 531
 532 void qmp_migrate_cancel(Error **errp)
 533 {
 534     migrate_fd_cancel(migrate_get_current());
 535 }
 536
 537 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 538 {
 539     MigrationState *s = migrate_get_current();
 540
 541     /* Check for truncation */
 542     if (value != (size_t)value) {
 543         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 544                   "exceeding address space");
 545         return;
 546     }
 547
 548     s->xbzrle_cache_size = xbzrle_cache_resize(value);
 549 }
 550
 551 int64_t qmp_query_migrate_cache_size(Error **errp)
 552 {
 553     return migrate_xbzrle_cache_size();
 554 }
 555
 556 void qmp_migrate_set_speed(int64_t value, Error **errp)
 557 {
 558     MigrationState *s;
 559
 560     if (value < 0) {
 561         value = 0;
 562     }
 563
 564     s = migrate_get_current();
 565     s->bandwidth_limit = value;
 566     qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
 567 }
 568
 569 void qmp_migrate_set_downtime(double value, Error **errp)
 570 {
 571     value *= 1e9;
 572     value = MAX(0, MIN(UINT64_MAX, value));
 573     max_downtime = (uint64_t)value;
 574 }
 575
 576 int migrate_use_xbzrle(void)
 577 {
 578     MigrationState *s;
 579
 580     s = migrate_get_current();
 581
 582     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 583 }
 584
 585 int64_t migrate_xbzrle_cache_size(void)
 586 {
 587     MigrationState *s;
 588
 589     s = migrate_get_current();
 590
 591     return s->xbzrle_cache_size;
 592 }