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
 184         if (blk_mig_active()) {
 185             info->has_disk = true;
 186             info->disk = g_malloc0(sizeof(*info->disk));
 187             info->disk->transferred = blk_mig_bytes_transferred();
 188             info->disk->remaining = blk_mig_bytes_remaining();
 189             info->disk->total = blk_mig_bytes_total();
 190         }
 191
 192         get_xbzrle_cache_stats(info);
 193         break;
 194     case MIG_STATE_COMPLETED:
 195         get_xbzrle_cache_stats(info);
 196
 197         info->has_status = true;
 198         info->status = g_strdup("completed");
 199         info->total_time = s->total_time;
 200         info->has_downtime = true;
 201         info->downtime = s->downtime;
 202
 203         info->has_ram = true;
 204         info->ram = g_malloc0(sizeof(*info->ram));
 205         info->ram->transferred = ram_bytes_transferred();
 206         info->ram->remaining = 0;
 207         info->ram->total = ram_bytes_total();
 208         info->ram->duplicate = dup_mig_pages_transferred();
 209         info->ram->normal = norm_mig_pages_transferred();
 210         info->ram->normal_bytes = norm_mig_bytes_transferred();
 211         break;
 212     case MIG_STATE_ERROR:
 213         info->has_status = true;
 214         info->status = g_strdup("failed");
 215         break;
 216     case MIG_STATE_CANCELLED:
 217         info->has_status = true;
 218         info->status = g_strdup("cancelled");
 219         break;
 220     }
 221
 222     return info;
 223 }
 224
 225 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
 226                                   Error **errp)
 227 {
 228     MigrationState *s = migrate_get_current();
 229     MigrationCapabilityStatusList *cap;
 230
 231     if (s->state == MIG_STATE_ACTIVE) {
 232         error_set(errp, QERR_MIGRATION_ACTIVE);
 233         return;
 234     }
 235
 236     for (cap = params; cap; cap = cap->next) {
 237         s->enabled_capabilities[cap->value->capability] = cap->value->state;
 238     }
 239 }
 240
 241 /* shared migration helpers */
 242
 243 static int migrate_fd_cleanup(MigrationState *s)
 244 {
 245     int ret = 0;
 246
 247     if (s->fd != -1) {
 248         qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 249     }
 250
 251     if (s->file) {
 252         DPRINTF("closing file\n");
 253         ret = qemu_fclose(s->file);
 254         s->file = NULL;
 255     }
 256
 257     if (s->fd != -1) {
 258         close(s->fd);
 259         s->fd = -1;
 260     }
 261
 262     return ret;
 263 }
 264
 265 void migrate_fd_error(MigrationState *s)
 266 {
 267     DPRINTF("setting error state\n");
 268     s->state = MIG_STATE_ERROR;
 269     notifier_list_notify(&migration_state_notifiers, s);
 270     migrate_fd_cleanup(s);
 271 }
 272
 273 static void migrate_fd_completed(MigrationState *s)
 274 {
 275     DPRINTF("setting completed state\n");
 276     if (migrate_fd_cleanup(s) < 0) {
 277         s->state = MIG_STATE_ERROR;
 278     } else {
 279         s->state = MIG_STATE_COMPLETED;
 280         runstate_set(RUN_STATE_POSTMIGRATE);
 281     }
 282     notifier_list_notify(&migration_state_notifiers, s);
 283 }
 284
 285 static void migrate_fd_put_notify(void *opaque)
 286 {
 287     MigrationState *s = opaque;
 288
 289     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 290     qemu_file_put_notify(s->file);
 291     if (s->file && qemu_file_get_error(s->file)) {
 292         migrate_fd_error(s);
 293     }
 294 }
 295
 296 static ssize_t migrate_fd_put_buffer(void *opaque, const void *data,
 297                                      size_t size)
 298 {
 299     MigrationState *s = opaque;
 300     ssize_t ret;
 301
 302     if (s->state != MIG_STATE_ACTIVE) {
 303         return -EIO;
 304     }
 305
 306     do {
 307         ret = s->write(s, data, size);
 308     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 309
 310     if (ret == -1)
 311         ret = -(s->get_error(s));
 312
 313     if (ret == -EAGAIN) {
 314         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 315     }
 316
 317     return ret;
 318 }
 319
 320 static void migrate_fd_put_ready(void *opaque)
 321 {
 322     MigrationState *s = opaque;
 323     int ret;
 324
 325     if (s->state != MIG_STATE_ACTIVE) {
 326         DPRINTF("put_ready returning because of non-active state\n");
 327         return;
 328     }
 329
 330     DPRINTF("iterate\n");
 331     ret = qemu_savevm_state_iterate(s->file);
 332     if (ret < 0) {
 333         migrate_fd_error(s);
 334     } else if (ret == 1) {
 335         int old_vm_running = runstate_is_running();
 336         int64_t start_time, end_time;
 337
 338         DPRINTF("done iterating\n");
 339         start_time = qemu_get_clock_ms(rt_clock);
 340         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 341         vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 342
 343         if (qemu_savevm_state_complete(s->file) < 0) {
 344             migrate_fd_error(s);
 345         } else {
 346             migrate_fd_completed(s);
 347         }
 348         end_time = qemu_get_clock_ms(rt_clock);
 349         s->total_time = end_time - s->total_time;
 350         s->downtime = end_time - start_time;
 351         if (s->state != MIG_STATE_COMPLETED) {
 352             if (old_vm_running) {
 353                 vm_start();
 354             }
 355         }
 356     }
 357 }
 358
 359 static void migrate_fd_cancel(MigrationState *s)
 360 {
 361     if (s->state != MIG_STATE_ACTIVE)
 362         return;
 363
 364     DPRINTF("cancelling migration\n");
 365
 366     s->state = MIG_STATE_CANCELLED;
 367     notifier_list_notify(&migration_state_notifiers, s);
 368     qemu_savevm_state_cancel(s->file);
 369
 370     migrate_fd_cleanup(s);
 371 }
 372
 373 static void migrate_fd_wait_for_unfreeze(void *opaque)
 374 {
 375     MigrationState *s = opaque;
 376     int ret;
 377
 378     DPRINTF("wait for unfreeze\n");
 379     if (s->state != MIG_STATE_ACTIVE)
 380         return;
 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         qemu_file_set_error(s->file, -s->get_error(s));
 393     }
 394 }
 395
 396 static int migrate_fd_close(void *opaque)
 397 {
 398     MigrationState *s = opaque;
 399
 400     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 401     return s->close(s);
 402 }
 403
 404 void add_migration_state_change_notifier(Notifier *notify)
 405 {
 406     notifier_list_add(&migration_state_notifiers, notify);
 407 }
 408
 409 void remove_migration_state_change_notifier(Notifier *notify)
 410 {
 411     notifier_remove(notify);
 412 }
 413
 414 bool migration_is_active(MigrationState *s)
 415 {
 416     return s->state == MIG_STATE_ACTIVE;
 417 }
 418
 419 bool migration_has_finished(MigrationState *s)
 420 {
 421     return s->state == MIG_STATE_COMPLETED;
 422 }
 423
 424 bool migration_has_failed(MigrationState *s)
 425 {
 426     return (s->state == MIG_STATE_CANCELLED ||
 427             s->state == MIG_STATE_ERROR);
 428 }
 429
 430 void migrate_fd_connect(MigrationState *s)
 431 {
 432     int ret;
 433
 434     s->state = MIG_STATE_ACTIVE;
 435     s->file = qemu_fopen_ops_buffered(s,
 436                                       s->bandwidth_limit,
 437                                       migrate_fd_put_buffer,
 438                                       migrate_fd_put_ready,
 439                                       migrate_fd_wait_for_unfreeze,
 440                                       migrate_fd_close);
 441
 442     DPRINTF("beginning savevm\n");
 443     ret = qemu_savevm_state_begin(s->file, &s->params);
 444     if (ret < 0) {
 445         DPRINTF("failed, %d\n", ret);
 446         migrate_fd_error(s);
 447         return;
 448     }
 449     migrate_fd_put_ready(s);
 450 }
 451
 452 static MigrationState *migrate_init(const MigrationParams *params)
 453 {
 454     MigrationState *s = migrate_get_current();
 455     int64_t bandwidth_limit = s->bandwidth_limit;
 456     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 457     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 458
 459     memcpy(enabled_capabilities, s->enabled_capabilities,
 460            sizeof(enabled_capabilities));
 461
 462     memset(s, 0, sizeof(*s));
 463     s->bandwidth_limit = bandwidth_limit;
 464     s->params = *params;
 465     memcpy(s->enabled_capabilities, enabled_capabilities,
 466            sizeof(enabled_capabilities));
 467     s->xbzrle_cache_size = xbzrle_cache_size;
 468
 469     s->bandwidth_limit = bandwidth_limit;
 470     s->state = MIG_STATE_SETUP;
 471     s->total_time = qemu_get_clock_ms(rt_clock);
 472
 473     return s;
 474 }
 475
 476 static GSList *migration_blockers;
 477
 478 void migrate_add_blocker(Error *reason)
 479 {
 480     migration_blockers = g_slist_prepend(migration_blockers, reason);
 481 }
 482
 483 void migrate_del_blocker(Error *reason)
 484 {
 485     migration_blockers = g_slist_remove(migration_blockers, reason);
 486 }
 487
 488 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 489                  bool has_inc, bool inc, bool has_detach, bool detach,
 490                  Error **errp)
 491 {
 492     MigrationState *s = migrate_get_current();
 493     MigrationParams params;
 494     const char *p;
 495     int ret;
 496
 497     params.blk = blk;
 498     params.shared = inc;
 499
 500     if (s->state == MIG_STATE_ACTIVE) {
 501         error_set(errp, QERR_MIGRATION_ACTIVE);
 502         return;
 503     }
 504
 505     if (qemu_savevm_state_blocked(errp)) {
 506         return;
 507     }
 508
 509     if (migration_blockers) {
 510         *errp = error_copy(migration_blockers->data);
 511         return;
 512     }
 513
 514     s = migrate_init(&params);
 515
 516     if (strstart(uri, "tcp:", &p)) {
 517         ret = tcp_start_outgoing_migration(s, p, errp);
 518 #if !defined(WIN32)
 519     } else if (strstart(uri, "exec:", &p)) {
 520         ret = exec_start_outgoing_migration(s, p);
 521     } else if (strstart(uri, "unix:", &p)) {
 522         ret = unix_start_outgoing_migration(s, p);
 523     } else if (strstart(uri, "fd:", &p)) {
 524         ret = fd_start_outgoing_migration(s, p);
 525 #endif
 526     } else {
 527         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "uri", "a valid migration protocol");
 528         return;
 529     }
 530
 531     if (ret < 0) {
 532         if (!error_is_set(errp)) {
 533             DPRINTF("migration failed: %s\n", strerror(-ret));
 534             /* FIXME: we should return meaningful errors */
 535             error_set(errp, QERR_UNDEFINED_ERROR);
 536         }
 537         return;
 538     }
 539
 540     notifier_list_notify(&migration_state_notifiers, s);
 541 }
 542
 543 void qmp_migrate_cancel(Error **errp)
 544 {
 545     migrate_fd_cancel(migrate_get_current());
 546 }
 547
 548 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 549 {
 550     MigrationState *s = migrate_get_current();
 551
 552     /* Check for truncation */
 553     if (value != (size_t)value) {
 554         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 555                   "exceeding address space");
 556         return;
 557     }
 558
 559     s->xbzrle_cache_size = xbzrle_cache_resize(value);
 560 }
 561
 562 int64_t qmp_query_migrate_cache_size(Error **errp)
 563 {
 564     return migrate_xbzrle_cache_size();
 565 }
 566
 567 void qmp_migrate_set_speed(int64_t value, Error **errp)
 568 {
 569     MigrationState *s;
 570
 571     if (value < 0) {
 572         value = 0;
 573     }
 574
 575     s = migrate_get_current();
 576     s->bandwidth_limit = value;
 577     qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
 578 }
 579
 580 void qmp_migrate_set_downtime(double value, Error **errp)
 581 {
 582     value *= 1e9;
 583     value = MAX(0, MIN(UINT64_MAX, value));
 584     max_downtime = (uint64_t)value;
 585 }
 586
 587 int migrate_use_xbzrle(void)
 588 {
 589     MigrationState *s;
 590
 591     s = migrate_get_current();
 592
 593     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 594 }
 595
 596 int64_t migrate_xbzrle_cache_size(void)
 597 {
 598     MigrationState *s;
 599
 600     s = migrate_get_current();
 601
 602     return s->xbzrle_cache_size;
 603 }