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 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         info->has_total_time = true;
 170         info->total_time = qemu_get_clock_ms(rt_clock)
 171             - s->total_time;
 172
 173         info->has_ram = true;
 174         info->ram = g_malloc0(sizeof(*info->ram));
 175         info->ram->transferred = ram_bytes_transferred();
 176         info->ram->remaining = ram_bytes_remaining();
 177         info->ram->total = ram_bytes_total();
 178         info->ram->duplicate = dup_mig_pages_transferred();
 179         info->ram->normal = norm_mig_pages_transferred();
 180         info->ram->normal_bytes = norm_mig_bytes_transferred();
 181
 182         if (blk_mig_active()) {
 183             info->has_disk = true;
 184             info->disk = g_malloc0(sizeof(*info->disk));
 185             info->disk->transferred = blk_mig_bytes_transferred();
 186             info->disk->remaining = blk_mig_bytes_remaining();
 187             info->disk->total = blk_mig_bytes_total();
 188         }
 189
 190         get_xbzrle_cache_stats(info);
 191         break;
 192     case MIG_STATE_COMPLETED:
 193         get_xbzrle_cache_stats(info);
 194
 195         info->has_status = true;
 196         info->status = g_strdup("completed");
 197         info->total_time = s->total_time;
 198
 199         info->has_ram = true;
 200         info->ram = g_malloc0(sizeof(*info->ram));
 201         info->ram->transferred = ram_bytes_transferred();
 202         info->ram->remaining = 0;
 203         info->ram->total = ram_bytes_total();
 204         info->ram->duplicate = dup_mig_pages_transferred();
 205         info->ram->normal = norm_mig_pages_transferred();
 206         info->ram->normal_bytes = norm_mig_bytes_transferred();
 207         break;
 208     case MIG_STATE_ERROR:
 209         info->has_status = true;
 210         info->status = g_strdup("failed");
 211         break;
 212     case MIG_STATE_CANCELLED:
 213         info->has_status = true;
 214         info->status = g_strdup("cancelled");
 215         break;
 216     }
 217
 218     return info;
 219 }
 220
 221 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
 222                                   Error **errp)
 223 {
 224     MigrationState *s = migrate_get_current();
 225     MigrationCapabilityStatusList *cap;
 226
 227     if (s->state == MIG_STATE_ACTIVE) {
 228         error_set(errp, QERR_MIGRATION_ACTIVE);
 229         return;
 230     }
 231
 232     for (cap = params; cap; cap = cap->next) {
 233         s->enabled_capabilities[cap->value->capability] = cap->value->state;
 234     }
 235 }
 236
 237 /* shared migration helpers */
 238
 239 static int migrate_fd_cleanup(MigrationState *s)
 240 {
 241     int ret = 0;
 242
 243     if (s->fd != -1) {
 244         qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 245     }
 246
 247     if (s->file) {
 248         DPRINTF("closing file\n");
 249         ret = qemu_fclose(s->file);
 250         s->file = NULL;
 251     }
 252
 253     if (s->fd != -1) {
 254         close(s->fd);
 255         s->fd = -1;
 256     }
 257
 258     return ret;
 259 }
 260
 261 void migrate_fd_error(MigrationState *s)
 262 {
 263     DPRINTF("setting error state\n");
 264     s->state = MIG_STATE_ERROR;
 265     notifier_list_notify(&migration_state_notifiers, s);
 266     migrate_fd_cleanup(s);
 267 }
 268
 269 static void migrate_fd_completed(MigrationState *s)
 270 {
 271     DPRINTF("setting completed state\n");
 272     if (migrate_fd_cleanup(s) < 0) {
 273         s->state = MIG_STATE_ERROR;
 274     } else {
 275         s->state = MIG_STATE_COMPLETED;
 276         runstate_set(RUN_STATE_POSTMIGRATE);
 277     }
 278     notifier_list_notify(&migration_state_notifiers, s);
 279 }
 280
 281 static void migrate_fd_put_notify(void *opaque)
 282 {
 283     MigrationState *s = opaque;
 284
 285     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 286     qemu_file_put_notify(s->file);
 287     if (s->file && qemu_file_get_error(s->file)) {
 288         migrate_fd_error(s);
 289     }
 290 }
 291
 292 static ssize_t migrate_fd_put_buffer(void *opaque, const void *data,
 293                                      size_t size)
 294 {
 295     MigrationState *s = opaque;
 296     ssize_t ret;
 297
 298     if (s->state != MIG_STATE_ACTIVE) {
 299         return -EIO;
 300     }
 301
 302     do {
 303         ret = s->write(s, data, size);
 304     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 305
 306     if (ret == -1)
 307         ret = -(s->get_error(s));
 308
 309     if (ret == -EAGAIN) {
 310         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 311     }
 312
 313     return ret;
 314 }
 315
 316 static void migrate_fd_put_ready(void *opaque)
 317 {
 318     MigrationState *s = opaque;
 319     int ret;
 320
 321     if (s->state != MIG_STATE_ACTIVE) {
 322         DPRINTF("put_ready returning because of non-active state\n");
 323         return;
 324     }
 325
 326     DPRINTF("iterate\n");
 327     ret = qemu_savevm_state_iterate(s->file);
 328     if (ret < 0) {
 329         migrate_fd_error(s);
 330     } else if (ret == 1) {
 331         int old_vm_running = runstate_is_running();
 332
 333         DPRINTF("done iterating\n");
 334         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 335         vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 336
 337         if (qemu_savevm_state_complete(s->file) < 0) {
 338             migrate_fd_error(s);
 339         } else {
 340             migrate_fd_completed(s);
 341         }
 342         s->total_time = qemu_get_clock_ms(rt_clock) - s->total_time;
 343         if (s->state != MIG_STATE_COMPLETED) {
 344             if (old_vm_running) {
 345                 vm_start();
 346             }
 347         }
 348     }
 349 }
 350
 351 static void migrate_fd_cancel(MigrationState *s)
 352 {
 353     if (s->state != MIG_STATE_ACTIVE)
 354         return;
 355
 356     DPRINTF("cancelling migration\n");
 357
 358     s->state = MIG_STATE_CANCELLED;
 359     notifier_list_notify(&migration_state_notifiers, s);
 360     qemu_savevm_state_cancel(s->file);
 361
 362     migrate_fd_cleanup(s);
 363 }
 364
 365 static void migrate_fd_wait_for_unfreeze(void *opaque)
 366 {
 367     MigrationState *s = opaque;
 368     int ret;
 369
 370     DPRINTF("wait for unfreeze\n");
 371     if (s->state != MIG_STATE_ACTIVE)
 372         return;
 373
 374     do {
 375         fd_set wfds;
 376
 377         FD_ZERO(&wfds);
 378         FD_SET(s->fd, &wfds);
 379
 380         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 381     } while (ret == -1 && (s->get_error(s)) == EINTR);
 382
 383     if (ret == -1) {
 384         qemu_file_set_error(s->file, -s->get_error(s));
 385     }
 386 }
 387
 388 static int migrate_fd_close(void *opaque)
 389 {
 390     MigrationState *s = opaque;
 391
 392     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 393     return s->close(s);
 394 }
 395
 396 void add_migration_state_change_notifier(Notifier *notify)
 397 {
 398     notifier_list_add(&migration_state_notifiers, notify);
 399 }
 400
 401 void remove_migration_state_change_notifier(Notifier *notify)
 402 {
 403     notifier_remove(notify);
 404 }
 405
 406 bool migration_is_active(MigrationState *s)
 407 {
 408     return s->state == MIG_STATE_ACTIVE;
 409 }
 410
 411 bool migration_has_finished(MigrationState *s)
 412 {
 413     return s->state == MIG_STATE_COMPLETED;
 414 }
 415
 416 bool migration_has_failed(MigrationState *s)
 417 {
 418     return (s->state == MIG_STATE_CANCELLED ||
 419             s->state == MIG_STATE_ERROR);
 420 }
 421
 422 void migrate_fd_connect(MigrationState *s)
 423 {
 424     int ret;
 425
 426     s->state = MIG_STATE_ACTIVE;
 427     s->file = qemu_fopen_ops_buffered(s,
 428                                       s->bandwidth_limit,
 429                                       migrate_fd_put_buffer,
 430                                       migrate_fd_put_ready,
 431                                       migrate_fd_wait_for_unfreeze,
 432                                       migrate_fd_close);
 433
 434     DPRINTF("beginning savevm\n");
 435     ret = qemu_savevm_state_begin(s->file, &s->params);
 436     if (ret < 0) {
 437         DPRINTF("failed, %d\n", ret);
 438         migrate_fd_error(s);
 439         return;
 440     }
 441     migrate_fd_put_ready(s);
 442 }
 443
 444 static MigrationState *migrate_init(const MigrationParams *params)
 445 {
 446     MigrationState *s = migrate_get_current();
 447     int64_t bandwidth_limit = s->bandwidth_limit;
 448     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 449     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 450
 451     memcpy(enabled_capabilities, s->enabled_capabilities,
 452            sizeof(enabled_capabilities));
 453
 454     memset(s, 0, sizeof(*s));
 455     s->bandwidth_limit = bandwidth_limit;
 456     s->params = *params;
 457     memcpy(s->enabled_capabilities, enabled_capabilities,
 458            sizeof(enabled_capabilities));
 459     s->xbzrle_cache_size = xbzrle_cache_size;
 460
 461     s->bandwidth_limit = bandwidth_limit;
 462     s->state = MIG_STATE_SETUP;
 463     s->total_time = qemu_get_clock_ms(rt_clock);
 464
 465     return s;
 466 }
 467
 468 static GSList *migration_blockers;
 469
 470 void migrate_add_blocker(Error *reason)
 471 {
 472     migration_blockers = g_slist_prepend(migration_blockers, reason);
 473 }
 474
 475 void migrate_del_blocker(Error *reason)
 476 {
 477     migration_blockers = g_slist_remove(migration_blockers, reason);
 478 }
 479
 480 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 481                  bool has_inc, bool inc, bool has_detach, bool detach,
 482                  Error **errp)
 483 {
 484     MigrationState *s = migrate_get_current();
 485     MigrationParams params;
 486     const char *p;
 487     int ret;
 488
 489     params.blk = blk;
 490     params.shared = inc;
 491
 492     if (s->state == MIG_STATE_ACTIVE) {
 493         error_set(errp, QERR_MIGRATION_ACTIVE);
 494         return;
 495     }
 496
 497     if (qemu_savevm_state_blocked(errp)) {
 498         return;
 499     }
 500
 501     if (migration_blockers) {
 502         *errp = error_copy(migration_blockers->data);
 503         return;
 504     }
 505
 506     s = migrate_init(&params);
 507
 508     if (strstart(uri, "tcp:", &p)) {
 509         ret = tcp_start_outgoing_migration(s, p, errp);
 510 #if !defined(WIN32)
 511     } else if (strstart(uri, "exec:", &p)) {
 512         ret = exec_start_outgoing_migration(s, p);
 513     } else if (strstart(uri, "unix:", &p)) {
 514         ret = unix_start_outgoing_migration(s, p);
 515     } else if (strstart(uri, "fd:", &p)) {
 516         ret = fd_start_outgoing_migration(s, p);
 517 #endif
 518     } else {
 519         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "uri", "a valid migration protocol");
 520         return;
 521     }
 522
 523     if (ret < 0) {
 524         if (!error_is_set(errp)) {
 525             DPRINTF("migration failed: %s\n", strerror(-ret));
 526             /* FIXME: we should return meaningful errors */
 527             error_set(errp, QERR_UNDEFINED_ERROR);
 528         }
 529         return;
 530     }
 531
 532     notifier_list_notify(&migration_state_notifiers, s);
 533 }
 534
 535 void qmp_migrate_cancel(Error **errp)
 536 {
 537     migrate_fd_cancel(migrate_get_current());
 538 }
 539
 540 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 541 {
 542     MigrationState *s = migrate_get_current();
 543
 544     /* Check for truncation */
 545     if (value != (size_t)value) {
 546         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 547                   "exceeding address space");
 548         return;
 549     }
 550
 551     s->xbzrle_cache_size = xbzrle_cache_resize(value);
 552 }
 553
 554 int64_t qmp_query_migrate_cache_size(Error **errp)
 555 {
 556     return migrate_xbzrle_cache_size();
 557 }
 558
 559 void qmp_migrate_set_speed(int64_t value, Error **errp)
 560 {
 561     MigrationState *s;
 562
 563     if (value < 0) {
 564         value = 0;
 565     }
 566
 567     s = migrate_get_current();
 568     s->bandwidth_limit = value;
 569     qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
 570 }
 571
 572 void qmp_migrate_set_downtime(double value, Error **errp)
 573 {
 574     value *= 1e9;
 575     value = MAX(0, MIN(UINT64_MAX, value));
 576     max_downtime = (uint64_t)value;
 577 }
 578
 579 int migrate_use_xbzrle(void)
 580 {
 581     MigrationState *s;
 582
 583     s = migrate_get_current();
 584
 585     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 586 }
 587
 588 int64_t migrate_xbzrle_cache_size(void)
 589 {
 590     MigrationState *s;
 591
 592     s = migrate_get_current();
 593
 594     return s->xbzrle_cache_size;
 595 }