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
 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         info->has_downtime = true;
 199         info->downtime = s->downtime;
 200
 201         info->has_ram = true;
 202         info->ram = g_malloc0(sizeof(*info->ram));
 203         info->ram->transferred = ram_bytes_transferred();
 204         info->ram->remaining = 0;
 205         info->ram->total = ram_bytes_total();
 206         info->ram->duplicate = dup_mig_pages_transferred();
 207         info->ram->normal = norm_mig_pages_transferred();
 208         info->ram->normal_bytes = norm_mig_bytes_transferred();
 209         break;
 210     case MIG_STATE_ERROR:
 211         info->has_status = true;
 212         info->status = g_strdup("failed");
 213         break;
 214     case MIG_STATE_CANCELLED:
 215         info->has_status = true;
 216         info->status = g_strdup("cancelled");
 217         break;
 218     }
 219
 220     return info;
 221 }
 222
 223 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
 224                                   Error **errp)
 225 {
 226     MigrationState *s = migrate_get_current();
 227     MigrationCapabilityStatusList *cap;
 228
 229     if (s->state == MIG_STATE_ACTIVE) {
 230         error_set(errp, QERR_MIGRATION_ACTIVE);
 231         return;
 232     }
 233
 234     for (cap = params; cap; cap = cap->next) {
 235         s->enabled_capabilities[cap->value->capability] = cap->value->state;
 236     }
 237 }
 238
 239 /* shared migration helpers */
 240
 241 static int migrate_fd_cleanup(MigrationState *s)
 242 {
 243     int ret = 0;
 244
 245     if (s->fd != -1) {
 246         qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 247     }
 248
 249     if (s->file) {
 250         DPRINTF("closing file\n");
 251         ret = qemu_fclose(s->file);
 252         s->file = NULL;
 253     }
 254
 255     if (s->fd != -1) {
 256         close(s->fd);
 257         s->fd = -1;
 258     }
 259
 260     return ret;
 261 }
 262
 263 void migrate_fd_error(MigrationState *s)
 264 {
 265     DPRINTF("setting error state\n");
 266     s->state = MIG_STATE_ERROR;
 267     notifier_list_notify(&migration_state_notifiers, s);
 268     migrate_fd_cleanup(s);
 269 }
 270
 271 static void migrate_fd_completed(MigrationState *s)
 272 {
 273     DPRINTF("setting completed state\n");
 274     if (migrate_fd_cleanup(s) < 0) {
 275         s->state = MIG_STATE_ERROR;
 276     } else {
 277         s->state = MIG_STATE_COMPLETED;
 278         runstate_set(RUN_STATE_POSTMIGRATE);
 279     }
 280     notifier_list_notify(&migration_state_notifiers, s);
 281 }
 282
 283 static void migrate_fd_put_notify(void *opaque)
 284 {
 285     MigrationState *s = opaque;
 286
 287     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 288     qemu_file_put_notify(s->file);
 289     if (s->file && qemu_file_get_error(s->file)) {
 290         migrate_fd_error(s);
 291     }
 292 }
 293
 294 static ssize_t migrate_fd_put_buffer(void *opaque, const void *data,
 295                                      size_t size)
 296 {
 297     MigrationState *s = opaque;
 298     ssize_t ret;
 299
 300     if (s->state != MIG_STATE_ACTIVE) {
 301         return -EIO;
 302     }
 303
 304     do {
 305         ret = s->write(s, data, size);
 306     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 307
 308     if (ret == -1)
 309         ret = -(s->get_error(s));
 310
 311     if (ret == -EAGAIN) {
 312         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 313     }
 314
 315     return ret;
 316 }
 317
 318 static void migrate_fd_put_ready(void *opaque)
 319 {
 320     MigrationState *s = opaque;
 321     int ret;
 322
 323     if (s->state != MIG_STATE_ACTIVE) {
 324         DPRINTF("put_ready returning because of non-active state\n");
 325         return;
 326     }
 327
 328     DPRINTF("iterate\n");
 329     ret = qemu_savevm_state_iterate(s->file);
 330     if (ret < 0) {
 331         migrate_fd_error(s);
 332     } else if (ret == 1) {
 333         int old_vm_running = runstate_is_running();
 334         int64_t start_time, end_time;
 335
 336         DPRINTF("done iterating\n");
 337         start_time = qemu_get_clock_ms(rt_clock);
 338         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 339         vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 340
 341         if (qemu_savevm_state_complete(s->file) < 0) {
 342             migrate_fd_error(s);
 343         } else {
 344             migrate_fd_completed(s);
 345         }
 346         end_time = qemu_get_clock_ms(rt_clock);
 347         s->total_time = end_time - s->total_time;
 348         s->downtime = end_time - start_time;
 349         if (s->state != MIG_STATE_COMPLETED) {
 350             if (old_vm_running) {
 351                 vm_start();
 352             }
 353         }
 354     }
 355 }
 356
 357 static void migrate_fd_cancel(MigrationState *s)
 358 {
 359     if (s->state != MIG_STATE_ACTIVE)
 360         return;
 361
 362     DPRINTF("cancelling migration\n");
 363
 364     s->state = MIG_STATE_CANCELLED;
 365     notifier_list_notify(&migration_state_notifiers, s);
 366     qemu_savevm_state_cancel(s->file);
 367
 368     migrate_fd_cleanup(s);
 369 }
 370
 371 static void migrate_fd_wait_for_unfreeze(void *opaque)
 372 {
 373     MigrationState *s = opaque;
 374     int ret;
 375
 376     DPRINTF("wait for unfreeze\n");
 377     if (s->state != MIG_STATE_ACTIVE)
 378         return;
 379
 380     do {
 381         fd_set wfds;
 382
 383         FD_ZERO(&wfds);
 384         FD_SET(s->fd, &wfds);
 385
 386         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 387     } while (ret == -1 && (s->get_error(s)) == EINTR);
 388
 389     if (ret == -1) {
 390         qemu_file_set_error(s->file, -s->get_error(s));
 391     }
 392 }
 393
 394 static int migrate_fd_close(void *opaque)
 395 {
 396     MigrationState *s = opaque;
 397
 398     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 399     return s->close(s);
 400 }
 401
 402 void add_migration_state_change_notifier(Notifier *notify)
 403 {
 404     notifier_list_add(&migration_state_notifiers, notify);
 405 }
 406
 407 void remove_migration_state_change_notifier(Notifier *notify)
 408 {
 409     notifier_remove(notify);
 410 }
 411
 412 bool migration_is_active(MigrationState *s)
 413 {
 414     return s->state == MIG_STATE_ACTIVE;
 415 }
 416
 417 bool migration_has_finished(MigrationState *s)
 418 {
 419     return s->state == MIG_STATE_COMPLETED;
 420 }
 421
 422 bool migration_has_failed(MigrationState *s)
 423 {
 424     return (s->state == MIG_STATE_CANCELLED ||
 425             s->state == MIG_STATE_ERROR);
 426 }
 427
 428 void migrate_fd_connect(MigrationState *s)
 429 {
 430     int ret;
 431
 432     s->state = MIG_STATE_ACTIVE;
 433     s->file = qemu_fopen_ops_buffered(s,
 434                                       s->bandwidth_limit,
 435                                       migrate_fd_put_buffer,
 436                                       migrate_fd_put_ready,
 437                                       migrate_fd_wait_for_unfreeze,
 438                                       migrate_fd_close);
 439
 440     DPRINTF("beginning savevm\n");
 441     ret = qemu_savevm_state_begin(s->file, &s->params);
 442     if (ret < 0) {
 443         DPRINTF("failed, %d\n", ret);
 444         migrate_fd_error(s);
 445         return;
 446     }
 447     migrate_fd_put_ready(s);
 448 }
 449
 450 static MigrationState *migrate_init(const MigrationParams *params)
 451 {
 452     MigrationState *s = migrate_get_current();
 453     int64_t bandwidth_limit = s->bandwidth_limit;
 454     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 455     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 456
 457     memcpy(enabled_capabilities, s->enabled_capabilities,
 458            sizeof(enabled_capabilities));
 459
 460     memset(s, 0, sizeof(*s));
 461     s->bandwidth_limit = bandwidth_limit;
 462     s->params = *params;
 463     memcpy(s->enabled_capabilities, enabled_capabilities,
 464            sizeof(enabled_capabilities));
 465     s->xbzrle_cache_size = xbzrle_cache_size;
 466
 467     s->bandwidth_limit = bandwidth_limit;
 468     s->state = MIG_STATE_SETUP;
 469     s->total_time = qemu_get_clock_ms(rt_clock);
 470
 471     return s;
 472 }
 473
 474 static GSList *migration_blockers;
 475
 476 void migrate_add_blocker(Error *reason)
 477 {
 478     migration_blockers = g_slist_prepend(migration_blockers, reason);
 479 }
 480
 481 void migrate_del_blocker(Error *reason)
 482 {
 483     migration_blockers = g_slist_remove(migration_blockers, reason);
 484 }
 485
 486 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 487                  bool has_inc, bool inc, bool has_detach, bool detach,
 488                  Error **errp)
 489 {
 490     MigrationState *s = migrate_get_current();
 491     MigrationParams params;
 492     const char *p;
 493     int ret;
 494
 495     params.blk = blk;
 496     params.shared = inc;
 497
 498     if (s->state == MIG_STATE_ACTIVE) {
 499         error_set(errp, QERR_MIGRATION_ACTIVE);
 500         return;
 501     }
 502
 503     if (qemu_savevm_state_blocked(errp)) {
 504         return;
 505     }
 506
 507     if (migration_blockers) {
 508         *errp = error_copy(migration_blockers->data);
 509         return;
 510     }
 511
 512     s = migrate_init(&params);
 513
 514     if (strstart(uri, "tcp:", &p)) {
 515         ret = tcp_start_outgoing_migration(s, p, errp);
 516 #if !defined(WIN32)
 517     } else if (strstart(uri, "exec:", &p)) {
 518         ret = exec_start_outgoing_migration(s, p);
 519     } else if (strstart(uri, "unix:", &p)) {
 520         ret = unix_start_outgoing_migration(s, p);
 521     } else if (strstart(uri, "fd:", &p)) {
 522         ret = fd_start_outgoing_migration(s, p);
 523 #endif
 524     } else {
 525         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "uri", "a valid migration protocol");
 526         return;
 527     }
 528
 529     if (ret < 0) {
 530         if (!error_is_set(errp)) {
 531             DPRINTF("migration failed: %s\n", strerror(-ret));
 532             /* FIXME: we should return meaningful errors */
 533             error_set(errp, QERR_UNDEFINED_ERROR);
 534         }
 535         return;
 536     }
 537
 538     notifier_list_notify(&migration_state_notifiers, s);
 539 }
 540
 541 void qmp_migrate_cancel(Error **errp)
 542 {
 543     migrate_fd_cancel(migrate_get_current());
 544 }
 545
 546 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 547 {
 548     MigrationState *s = migrate_get_current();
 549
 550     /* Check for truncation */
 551     if (value != (size_t)value) {
 552         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 553                   "exceeding address space");
 554         return;
 555     }
 556
 557     s->xbzrle_cache_size = xbzrle_cache_resize(value);
 558 }
 559
 560 int64_t qmp_query_migrate_cache_size(Error **errp)
 561 {
 562     return migrate_xbzrle_cache_size();
 563 }
 564
 565 void qmp_migrate_set_speed(int64_t value, Error **errp)
 566 {
 567     MigrationState *s;
 568
 569     if (value < 0) {
 570         value = 0;
 571     }
 572
 573     s = migrate_get_current();
 574     s->bandwidth_limit = value;
 575     qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
 576 }
 577
 578 void qmp_migrate_set_downtime(double value, Error **errp)
 579 {
 580     value *= 1e9;
 581     value = MAX(0, MIN(UINT64_MAX, value));
 582     max_downtime = (uint64_t)value;
 583 }
 584
 585 int migrate_use_xbzrle(void)
 586 {
 587     MigrationState *s;
 588
 589     s = migrate_get_current();
 590
 591     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 592 }
 593
 594 int64_t migrate_xbzrle_cache_size(void)
 595 {
 596     MigrationState *s;
 597
 598     s = migrate_get_current();
 599
 600     return s->xbzrle_cache_size;
 601 }