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     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 244
 245     if (s->file) {
 246         DPRINTF("closing file\n");
 247         ret = qemu_fclose(s->file);
 248         s->file = NULL;
 249     }
 250
 251     if (s->fd != -1) {
 252         close(s->fd);
 253         s->fd = -1;
 254     }
 255
 256     return ret;
 257 }
 258
 259 void migrate_fd_error(MigrationState *s)
 260 {
 261     DPRINTF("setting error state\n");
 262     s->state = MIG_STATE_ERROR;
 263     notifier_list_notify(&migration_state_notifiers, s);
 264     migrate_fd_cleanup(s);
 265 }
 266
 267 static void migrate_fd_completed(MigrationState *s)
 268 {
 269     DPRINTF("setting completed state\n");
 270     if (migrate_fd_cleanup(s) < 0) {
 271         s->state = MIG_STATE_ERROR;
 272     } else {
 273         s->state = MIG_STATE_COMPLETED;
 274         runstate_set(RUN_STATE_POSTMIGRATE);
 275     }
 276     notifier_list_notify(&migration_state_notifiers, s);
 277 }
 278
 279 static void migrate_fd_put_notify(void *opaque)
 280 {
 281     MigrationState *s = opaque;
 282
 283     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 284     qemu_file_put_notify(s->file);
 285     if (s->file && qemu_file_get_error(s->file)) {
 286         migrate_fd_error(s);
 287     }
 288 }
 289
 290 static ssize_t migrate_fd_put_buffer(void *opaque, const void *data,
 291                                      size_t size)
 292 {
 293     MigrationState *s = opaque;
 294     ssize_t ret;
 295
 296     if (s->state != MIG_STATE_ACTIVE) {
 297         return -EIO;
 298     }
 299
 300     do {
 301         ret = s->write(s, data, size);
 302     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 303
 304     if (ret == -1)
 305         ret = -(s->get_error(s));
 306
 307     if (ret == -EAGAIN) {
 308         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 309     }
 310
 311     return ret;
 312 }
 313
 314 static void migrate_fd_put_ready(void *opaque)
 315 {
 316     MigrationState *s = opaque;
 317     int ret;
 318
 319     if (s->state != MIG_STATE_ACTIVE) {
 320         DPRINTF("put_ready returning because of non-active state\n");
 321         return;
 322     }
 323
 324     DPRINTF("iterate\n");
 325     ret = qemu_savevm_state_iterate(s->file);
 326     if (ret < 0) {
 327         migrate_fd_error(s);
 328     } else if (ret == 1) {
 329         int old_vm_running = runstate_is_running();
 330
 331         DPRINTF("done iterating\n");
 332         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 333         vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 334
 335         if (qemu_savevm_state_complete(s->file) < 0) {
 336             migrate_fd_error(s);
 337         } else {
 338             migrate_fd_completed(s);
 339         }
 340         s->total_time = qemu_get_clock_ms(rt_clock) - s->total_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 static void migrate_fd_wait_for_unfreeze(void *opaque)
 364 {
 365     MigrationState *s = opaque;
 366     int ret;
 367
 368     DPRINTF("wait for unfreeze\n");
 369     if (s->state != MIG_STATE_ACTIVE)
 370         return;
 371
 372     do {
 373         fd_set wfds;
 374
 375         FD_ZERO(&wfds);
 376         FD_SET(s->fd, &wfds);
 377
 378         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 379     } while (ret == -1 && (s->get_error(s)) == EINTR);
 380
 381     if (ret == -1) {
 382         qemu_file_set_error(s->file, -s->get_error(s));
 383     }
 384 }
 385
 386 static int migrate_fd_close(void *opaque)
 387 {
 388     MigrationState *s = opaque;
 389
 390     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 391     return s->close(s);
 392 }
 393
 394 void add_migration_state_change_notifier(Notifier *notify)
 395 {
 396     notifier_list_add(&migration_state_notifiers, notify);
 397 }
 398
 399 void remove_migration_state_change_notifier(Notifier *notify)
 400 {
 401     notifier_remove(notify);
 402 }
 403
 404 bool migration_is_active(MigrationState *s)
 405 {
 406     return s->state == MIG_STATE_ACTIVE;
 407 }
 408
 409 bool migration_has_finished(MigrationState *s)
 410 {
 411     return s->state == MIG_STATE_COMPLETED;
 412 }
 413
 414 bool migration_has_failed(MigrationState *s)
 415 {
 416     return (s->state == MIG_STATE_CANCELLED ||
 417             s->state == MIG_STATE_ERROR);
 418 }
 419
 420 void migrate_fd_connect(MigrationState *s)
 421 {
 422     int ret;
 423
 424     s->state = MIG_STATE_ACTIVE;
 425     s->file = qemu_fopen_ops_buffered(s,
 426                                       s->bandwidth_limit,
 427                                       migrate_fd_put_buffer,
 428                                       migrate_fd_put_ready,
 429                                       migrate_fd_wait_for_unfreeze,
 430                                       migrate_fd_close);
 431
 432     DPRINTF("beginning savevm\n");
 433     ret = qemu_savevm_state_begin(s->file, &s->params);
 434     if (ret < 0) {
 435         DPRINTF("failed, %d\n", ret);
 436         migrate_fd_error(s);
 437         return;
 438     }
 439     migrate_fd_put_ready(s);
 440 }
 441
 442 static MigrationState *migrate_init(const MigrationParams *params)
 443 {
 444     MigrationState *s = migrate_get_current();
 445     int64_t bandwidth_limit = s->bandwidth_limit;
 446     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 447     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 448
 449     memcpy(enabled_capabilities, s->enabled_capabilities,
 450            sizeof(enabled_capabilities));
 451
 452     memset(s, 0, sizeof(*s));
 453     s->bandwidth_limit = bandwidth_limit;
 454     s->params = *params;
 455     memcpy(s->enabled_capabilities, enabled_capabilities,
 456            sizeof(enabled_capabilities));
 457     s->xbzrle_cache_size = xbzrle_cache_size;
 458
 459     s->bandwidth_limit = bandwidth_limit;
 460     s->state = MIG_STATE_SETUP;
 461     s->total_time = qemu_get_clock_ms(rt_clock);
 462
 463     return s;
 464 }
 465
 466 static GSList *migration_blockers;
 467
 468 void migrate_add_blocker(Error *reason)
 469 {
 470     migration_blockers = g_slist_prepend(migration_blockers, reason);
 471 }
 472
 473 void migrate_del_blocker(Error *reason)
 474 {
 475     migration_blockers = g_slist_remove(migration_blockers, reason);
 476 }
 477
 478 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 479                  bool has_inc, bool inc, bool has_detach, bool detach,
 480                  Error **errp)
 481 {
 482     MigrationState *s = migrate_get_current();
 483     MigrationParams params;
 484     const char *p;
 485     int ret;
 486
 487     params.blk = blk;
 488     params.shared = inc;
 489
 490     if (s->state == MIG_STATE_ACTIVE) {
 491         error_set(errp, QERR_MIGRATION_ACTIVE);
 492         return;
 493     }
 494
 495     if (qemu_savevm_state_blocked(errp)) {
 496         return;
 497     }
 498
 499     if (migration_blockers) {
 500         *errp = error_copy(migration_blockers->data);
 501         return;
 502     }
 503
 504     s = migrate_init(&params);
 505
 506     if (strstart(uri, "tcp:", &p)) {
 507         ret = tcp_start_outgoing_migration(s, p, errp);
 508 #if !defined(WIN32)
 509     } else if (strstart(uri, "exec:", &p)) {
 510         ret = exec_start_outgoing_migration(s, p);
 511     } else if (strstart(uri, "unix:", &p)) {
 512         ret = unix_start_outgoing_migration(s, p);
 513     } else if (strstart(uri, "fd:", &p)) {
 514         ret = fd_start_outgoing_migration(s, p);
 515 #endif
 516     } else {
 517         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "uri", "a valid migration protocol");
 518         return;
 519     }
 520
 521     if (ret < 0) {
 522         if (!error_is_set(errp)) {
 523             DPRINTF("migration failed: %s\n", strerror(-ret));
 524             /* FIXME: we should return meaningful errors */
 525             error_set(errp, QERR_UNDEFINED_ERROR);
 526         }
 527         return;
 528     }
 529
 530     notifier_list_notify(&migration_state_notifiers, s);
 531 }
 532
 533 void qmp_migrate_cancel(Error **errp)
 534 {
 535     migrate_fd_cancel(migrate_get_current());
 536 }
 537
 538 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 539 {
 540     MigrationState *s = migrate_get_current();
 541
 542     /* Check for truncation */
 543     if (value != (size_t)value) {
 544         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 545                   "exceeding address space");
 546         return;
 547     }
 548
 549     s->xbzrle_cache_size = xbzrle_cache_resize(value);
 550 }
 551
 552 int64_t qmp_query_migrate_cache_size(Error **errp)
 553 {
 554     return migrate_xbzrle_cache_size();
 555 }
 556
 557 void qmp_migrate_set_speed(int64_t value, Error **errp)
 558 {
 559     MigrationState *s;
 560
 561     if (value < 0) {
 562         value = 0;
 563     }
 564
 565     s = migrate_get_current();
 566     s->bandwidth_limit = value;
 567     qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
 568 }
 569
 570 void qmp_migrate_set_downtime(double value, Error **errp)
 571 {
 572     value *= 1e9;
 573     value = MAX(0, MIN(UINT64_MAX, value));
 574     max_downtime = (uint64_t)value;
 575 }
 576
 577 int migrate_use_xbzrle(void)
 578 {
 579     MigrationState *s;
 580
 581     s = migrate_get_current();
 582
 583     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 584 }
 585
 586 int64_t migrate_xbzrle_cache_size(void)
 587 {
 588     MigrationState *s;
 589
 590     s = migrate_get_current();
 591
 592     return s->xbzrle_cache_size;
 593 }