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 MigrationInfo *qmp_query_migrate(Error **errp)
 145 {
 146     MigrationInfo *info = g_malloc0(sizeof(*info));
 147     MigrationState *s = migrate_get_current();
 148
 149     switch (s->state) {
 150     case MIG_STATE_SETUP:
 151         /* no migration has happened ever */
 152         break;
 153     case MIG_STATE_ACTIVE:
 154         info->has_status = true;
 155         info->status = g_strdup("active");
 156
 157         info->has_ram = true;
 158         info->ram = g_malloc0(sizeof(*info->ram));
 159         info->ram->transferred = ram_bytes_transferred();
 160         info->ram->remaining = ram_bytes_remaining();
 161         info->ram->total = ram_bytes_total();
 162         info->ram->total_time = qemu_get_clock_ms(rt_clock)
 163             - s->total_time;
 164         info->ram->duplicate = dup_mig_pages_transferred();
 165         info->ram->normal = norm_mig_pages_transferred();
 166         info->ram->normal_bytes = norm_mig_bytes_transferred();
 167
 168         if (blk_mig_active()) {
 169             info->has_disk = true;
 170             info->disk = g_malloc0(sizeof(*info->disk));
 171             info->disk->transferred = blk_mig_bytes_transferred();
 172             info->disk->remaining = blk_mig_bytes_remaining();
 173             info->disk->total = blk_mig_bytes_total();
 174         }
 175         break;
 176     case MIG_STATE_COMPLETED:
 177         info->has_status = true;
 178         info->status = g_strdup("completed");
 179
 180         info->has_ram = true;
 181         info->ram = g_malloc0(sizeof(*info->ram));
 182         info->ram->transferred = ram_bytes_transferred();
 183         info->ram->remaining = 0;
 184         info->ram->total = ram_bytes_total();
 185         info->ram->total_time = s->total_time;
 186         info->ram->duplicate = dup_mig_pages_transferred();
 187         info->ram->normal = norm_mig_pages_transferred();
 188         info->ram->normal_bytes = norm_mig_bytes_transferred();
 189         break;
 190     case MIG_STATE_ERROR:
 191         info->has_status = true;
 192         info->status = g_strdup("failed");
 193         break;
 194     case MIG_STATE_CANCELLED:
 195         info->has_status = true;
 196         info->status = g_strdup("cancelled");
 197         break;
 198     }
 199
 200     return info;
 201 }
 202
 203 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
 204                                   Error **errp)
 205 {
 206     MigrationState *s = migrate_get_current();
 207     MigrationCapabilityStatusList *cap;
 208
 209     if (s->state == MIG_STATE_ACTIVE) {
 210         error_set(errp, QERR_MIGRATION_ACTIVE);
 211         return;
 212     }
 213
 214     for (cap = params; cap; cap = cap->next) {
 215         s->enabled_capabilities[cap->value->capability] = cap->value->state;
 216     }
 217 }
 218
 219 /* shared migration helpers */
 220
 221 static int migrate_fd_cleanup(MigrationState *s)
 222 {
 223     int ret = 0;
 224
 225     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 226
 227     if (s->file) {
 228         DPRINTF("closing file\n");
 229         ret = qemu_fclose(s->file);
 230         s->file = NULL;
 231     }
 232
 233     if (s->fd != -1) {
 234         close(s->fd);
 235         s->fd = -1;
 236     }
 237
 238     return ret;
 239 }
 240
 241 void migrate_fd_error(MigrationState *s)
 242 {
 243     DPRINTF("setting error state\n");
 244     s->state = MIG_STATE_ERROR;
 245     notifier_list_notify(&migration_state_notifiers, s);
 246     migrate_fd_cleanup(s);
 247 }
 248
 249 static void migrate_fd_completed(MigrationState *s)
 250 {
 251     DPRINTF("setting completed state\n");
 252     if (migrate_fd_cleanup(s) < 0) {
 253         s->state = MIG_STATE_ERROR;
 254     } else {
 255         s->state = MIG_STATE_COMPLETED;
 256         runstate_set(RUN_STATE_POSTMIGRATE);
 257     }
 258     notifier_list_notify(&migration_state_notifiers, s);
 259 }
 260
 261 static void migrate_fd_put_notify(void *opaque)
 262 {
 263     MigrationState *s = opaque;
 264
 265     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 266     qemu_file_put_notify(s->file);
 267     if (s->file && qemu_file_get_error(s->file)) {
 268         migrate_fd_error(s);
 269     }
 270 }
 271
 272 static ssize_t migrate_fd_put_buffer(void *opaque, const void *data,
 273                                      size_t size)
 274 {
 275     MigrationState *s = opaque;
 276     ssize_t ret;
 277
 278     if (s->state != MIG_STATE_ACTIVE) {
 279         return -EIO;
 280     }
 281
 282     do {
 283         ret = s->write(s, data, size);
 284     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 285
 286     if (ret == -1)
 287         ret = -(s->get_error(s));
 288
 289     if (ret == -EAGAIN) {
 290         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 291     }
 292
 293     return ret;
 294 }
 295
 296 static void migrate_fd_put_ready(void *opaque)
 297 {
 298     MigrationState *s = opaque;
 299     int ret;
 300
 301     if (s->state != MIG_STATE_ACTIVE) {
 302         DPRINTF("put_ready returning because of non-active state\n");
 303         return;
 304     }
 305
 306     DPRINTF("iterate\n");
 307     ret = qemu_savevm_state_iterate(s->file);
 308     if (ret < 0) {
 309         migrate_fd_error(s);
 310     } else if (ret == 1) {
 311         int old_vm_running = runstate_is_running();
 312
 313         DPRINTF("done iterating\n");
 314         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 315         vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 316
 317         if (qemu_savevm_state_complete(s->file) < 0) {
 318             migrate_fd_error(s);
 319         } else {
 320             migrate_fd_completed(s);
 321         }
 322         s->total_time = qemu_get_clock_ms(rt_clock) - s->total_time;
 323         if (s->state != MIG_STATE_COMPLETED) {
 324             if (old_vm_running) {
 325                 vm_start();
 326             }
 327         }
 328     }
 329 }
 330
 331 static void migrate_fd_cancel(MigrationState *s)
 332 {
 333     if (s->state != MIG_STATE_ACTIVE)
 334         return;
 335
 336     DPRINTF("cancelling migration\n");
 337
 338     s->state = MIG_STATE_CANCELLED;
 339     notifier_list_notify(&migration_state_notifiers, s);
 340     qemu_savevm_state_cancel(s->file);
 341
 342     migrate_fd_cleanup(s);
 343 }
 344
 345 static void migrate_fd_wait_for_unfreeze(void *opaque)
 346 {
 347     MigrationState *s = opaque;
 348     int ret;
 349
 350     DPRINTF("wait for unfreeze\n");
 351     if (s->state != MIG_STATE_ACTIVE)
 352         return;
 353
 354     do {
 355         fd_set wfds;
 356
 357         FD_ZERO(&wfds);
 358         FD_SET(s->fd, &wfds);
 359
 360         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 361     } while (ret == -1 && (s->get_error(s)) == EINTR);
 362
 363     if (ret == -1) {
 364         qemu_file_set_error(s->file, -s->get_error(s));
 365     }
 366 }
 367
 368 static int migrate_fd_close(void *opaque)
 369 {
 370     MigrationState *s = opaque;
 371
 372     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 373     return s->close(s);
 374 }
 375
 376 void add_migration_state_change_notifier(Notifier *notify)
 377 {
 378     notifier_list_add(&migration_state_notifiers, notify);
 379 }
 380
 381 void remove_migration_state_change_notifier(Notifier *notify)
 382 {
 383     notifier_remove(notify);
 384 }
 385
 386 bool migration_is_active(MigrationState *s)
 387 {
 388     return s->state == MIG_STATE_ACTIVE;
 389 }
 390
 391 bool migration_has_finished(MigrationState *s)
 392 {
 393     return s->state == MIG_STATE_COMPLETED;
 394 }
 395
 396 bool migration_has_failed(MigrationState *s)
 397 {
 398     return (s->state == MIG_STATE_CANCELLED ||
 399             s->state == MIG_STATE_ERROR);
 400 }
 401
 402 void migrate_fd_connect(MigrationState *s)
 403 {
 404     int ret;
 405
 406     s->state = MIG_STATE_ACTIVE;
 407     s->file = qemu_fopen_ops_buffered(s,
 408                                       s->bandwidth_limit,
 409                                       migrate_fd_put_buffer,
 410                                       migrate_fd_put_ready,
 411                                       migrate_fd_wait_for_unfreeze,
 412                                       migrate_fd_close);
 413
 414     DPRINTF("beginning savevm\n");
 415     ret = qemu_savevm_state_begin(s->file, &s->params);
 416     if (ret < 0) {
 417         DPRINTF("failed, %d\n", ret);
 418         migrate_fd_error(s);
 419         return;
 420     }
 421     migrate_fd_put_ready(s);
 422 }
 423
 424 static MigrationState *migrate_init(const MigrationParams *params)
 425 {
 426     MigrationState *s = migrate_get_current();
 427     int64_t bandwidth_limit = s->bandwidth_limit;
 428     bool enabled_capabilities[MIGRATION_CAPABILITY_MAX];
 429     int64_t xbzrle_cache_size = s->xbzrle_cache_size;
 430
 431     memcpy(enabled_capabilities, s->enabled_capabilities,
 432            sizeof(enabled_capabilities));
 433
 434     memset(s, 0, sizeof(*s));
 435     s->bandwidth_limit = bandwidth_limit;
 436     s->params = *params;
 437     memcpy(s->enabled_capabilities, enabled_capabilities,
 438            sizeof(enabled_capabilities));
 439     s->xbzrle_cache_size = xbzrle_cache_size;
 440
 441     s->bandwidth_limit = bandwidth_limit;
 442     s->state = MIG_STATE_SETUP;
 443     s->total_time = qemu_get_clock_ms(rt_clock);
 444
 445     return s;
 446 }
 447
 448 static GSList *migration_blockers;
 449
 450 void migrate_add_blocker(Error *reason)
 451 {
 452     migration_blockers = g_slist_prepend(migration_blockers, reason);
 453 }
 454
 455 void migrate_del_blocker(Error *reason)
 456 {
 457     migration_blockers = g_slist_remove(migration_blockers, reason);
 458 }
 459
 460 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 461                  bool has_inc, bool inc, bool has_detach, bool detach,
 462                  Error **errp)
 463 {
 464     MigrationState *s = migrate_get_current();
 465     MigrationParams params;
 466     const char *p;
 467     int ret;
 468
 469     params.blk = blk;
 470     params.shared = inc;
 471
 472     if (s->state == MIG_STATE_ACTIVE) {
 473         error_set(errp, QERR_MIGRATION_ACTIVE);
 474         return;
 475     }
 476
 477     if (qemu_savevm_state_blocked(errp)) {
 478         return;
 479     }
 480
 481     if (migration_blockers) {
 482         *errp = error_copy(migration_blockers->data);
 483         return;
 484     }
 485
 486     s = migrate_init(&params);
 487
 488     if (strstart(uri, "tcp:", &p)) {
 489         ret = tcp_start_outgoing_migration(s, p, errp);
 490 #if !defined(WIN32)
 491     } else if (strstart(uri, "exec:", &p)) {
 492         ret = exec_start_outgoing_migration(s, p);
 493     } else if (strstart(uri, "unix:", &p)) {
 494         ret = unix_start_outgoing_migration(s, p);
 495     } else if (strstart(uri, "fd:", &p)) {
 496         ret = fd_start_outgoing_migration(s, p);
 497 #endif
 498     } else {
 499         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "uri", "a valid migration protocol");
 500         return;
 501     }
 502
 503     if (ret < 0) {
 504         if (!error_is_set(errp)) {
 505             DPRINTF("migration failed: %s\n", strerror(-ret));
 506             /* FIXME: we should return meaningful errors */
 507             error_set(errp, QERR_UNDEFINED_ERROR);
 508         }
 509         return;
 510     }
 511
 512     notifier_list_notify(&migration_state_notifiers, s);
 513 }
 514
 515 void qmp_migrate_cancel(Error **errp)
 516 {
 517     migrate_fd_cancel(migrate_get_current());
 518 }
 519
 520 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
 521 {
 522     MigrationState *s = migrate_get_current();
 523
 524     /* Check for truncation */
 525     if (value != (size_t)value) {
 526         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "cache size",
 527                   "exceeding address space");
 528         return;
 529     }
 530
 531     s->xbzrle_cache_size = xbzrle_cache_resize(value);
 532 }
 533
 534 int64_t qmp_query_migrate_cache_size(Error **errp)
 535 {
 536     return migrate_xbzrle_cache_size();
 537 }
 538
 539 void qmp_migrate_set_speed(int64_t value, Error **errp)
 540 {
 541     MigrationState *s;
 542
 543     if (value < 0) {
 544         value = 0;
 545     }
 546
 547     s = migrate_get_current();
 548     s->bandwidth_limit = value;
 549     qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
 550 }
 551
 552 void qmp_migrate_set_downtime(double value, Error **errp)
 553 {
 554     value *= 1e9;
 555     value = MAX(0, MIN(UINT64_MAX, value));
 556     max_downtime = (uint64_t)value;
 557 }
 558
 559 int migrate_use_xbzrle(void)
 560 {
 561     MigrationState *s;
 562
 563     s = migrate_get_current();
 564
 565     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
 566 }
 567
 568 int64_t migrate_xbzrle_cache_size(void)
 569 {
 570     MigrationState *s;
 571
 572     s = migrate_get_current();
 573
 574     return s->xbzrle_cache_size;
 575 }