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 static NotifierList migration_state_notifiers =
  47     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
  48
  49 /* When we add fault tolerance, we could have several
  50    migrations at once.  For now we don't need to add
  51    dynamic creation of migration */
  52
  53 static MigrationState *migrate_get_current(void)
  54 {
  55     static MigrationState current_migration = {
  56         .state = MIG_STATE_SETUP,
  57         .bandwidth_limit = MAX_THROTTLE,
  58     };
  59
  60     return &current_migration;
  61 }
  62
  63 int qemu_start_incoming_migration(const char *uri)
  64 {
  65     const char *p;
  66     int ret;
  67
  68     if (strstart(uri, "tcp:", &p))
  69         ret = tcp_start_incoming_migration(p);
  70 #if !defined(WIN32)
  71     else if (strstart(uri, "exec:", &p))
  72         ret =  exec_start_incoming_migration(p);
  73     else if (strstart(uri, "unix:", &p))
  74         ret = unix_start_incoming_migration(p);
  75     else if (strstart(uri, "fd:", &p))
  76         ret = fd_start_incoming_migration(p);
  77 #endif
  78     else {
  79         fprintf(stderr, "unknown migration protocol: %s\n", uri);
  80         ret = -EPROTONOSUPPORT;
  81     }
  82     return ret;
  83 }
  84
  85 void process_incoming_migration(QEMUFile *f)
  86 {
  87     if (qemu_loadvm_state(f) < 0) {
  88         fprintf(stderr, "load of migration failed\n");
  89         exit(0);
  90     }
  91     qemu_announce_self();
  92     DPRINTF("successfully loaded vm state\n");
  93
  94     /* Make sure all file formats flush their mutable metadata */
  95     bdrv_invalidate_cache_all();
  96
  97     if (autostart) {
  98         vm_start();
  99     } else {
 100         runstate_set(RUN_STATE_PRELAUNCH);
 101     }
 102 }
 103
 104 /* amount of nanoseconds we are willing to wait for migration to be down.
 105  * the choice of nanoseconds is because it is the maximum resolution that
 106  * get_clock() can achieve. It is an internal measure. All user-visible
 107  * units must be in seconds */
 108 static uint64_t max_downtime = 30000000;
 109
 110 uint64_t migrate_max_downtime(void)
 111 {
 112     return max_downtime;
 113 }
 114
 115 MigrationInfo *qmp_query_migrate(Error **errp)
 116 {
 117     MigrationInfo *info = g_malloc0(sizeof(*info));
 118     MigrationState *s = migrate_get_current();
 119
 120     switch (s->state) {
 121     case MIG_STATE_SETUP:
 122         /* no migration has happened ever */
 123         break;
 124     case MIG_STATE_ACTIVE:
 125         info->has_status = true;
 126         info->status = g_strdup("active");
 127
 128         info->has_ram = true;
 129         info->ram = g_malloc0(sizeof(*info->ram));
 130         info->ram->transferred = ram_bytes_transferred();
 131         info->ram->remaining = ram_bytes_remaining();
 132         info->ram->total = ram_bytes_total();
 133
 134         if (blk_mig_active()) {
 135             info->has_disk = true;
 136             info->disk = g_malloc0(sizeof(*info->disk));
 137             info->disk->transferred = blk_mig_bytes_transferred();
 138             info->disk->remaining = blk_mig_bytes_remaining();
 139             info->disk->total = blk_mig_bytes_total();
 140         }
 141         break;
 142     case MIG_STATE_COMPLETED:
 143         info->has_status = true;
 144         info->status = g_strdup("completed");
 145         break;
 146     case MIG_STATE_ERROR:
 147         info->has_status = true;
 148         info->status = g_strdup("failed");
 149         break;
 150     case MIG_STATE_CANCELLED:
 151         info->has_status = true;
 152         info->status = g_strdup("cancelled");
 153         break;
 154     }
 155
 156     return info;
 157 }
 158
 159 /* shared migration helpers */
 160
 161 static int migrate_fd_cleanup(MigrationState *s)
 162 {
 163     int ret = 0;
 164
 165     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 166
 167     if (s->file) {
 168         DPRINTF("closing file\n");
 169         ret = qemu_fclose(s->file);
 170         s->file = NULL;
 171     }
 172
 173     if (s->fd != -1) {
 174         close(s->fd);
 175         s->fd = -1;
 176     }
 177
 178     return ret;
 179 }
 180
 181 void migrate_fd_error(MigrationState *s)
 182 {
 183     DPRINTF("setting error state\n");
 184     s->state = MIG_STATE_ERROR;
 185     notifier_list_notify(&migration_state_notifiers, s);
 186     migrate_fd_cleanup(s);
 187 }
 188
 189 static void migrate_fd_completed(MigrationState *s)
 190 {
 191     DPRINTF("setting completed state\n");
 192     if (migrate_fd_cleanup(s) < 0) {
 193         s->state = MIG_STATE_ERROR;
 194     } else {
 195         s->state = MIG_STATE_COMPLETED;
 196         runstate_set(RUN_STATE_POSTMIGRATE);
 197     }
 198     notifier_list_notify(&migration_state_notifiers, s);
 199 }
 200
 201 static void migrate_fd_put_notify(void *opaque)
 202 {
 203     MigrationState *s = opaque;
 204
 205     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 206     qemu_file_put_notify(s->file);
 207     if (s->file && qemu_file_get_error(s->file)) {
 208         migrate_fd_error(s);
 209     }
 210 }
 211
 212 static ssize_t migrate_fd_put_buffer(void *opaque, const void *data,
 213                                      size_t size)
 214 {
 215     MigrationState *s = opaque;
 216     ssize_t ret;
 217
 218     if (s->state != MIG_STATE_ACTIVE) {
 219         return -EIO;
 220     }
 221
 222     do {
 223         ret = s->write(s, data, size);
 224     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 225
 226     if (ret == -1)
 227         ret = -(s->get_error(s));
 228
 229     if (ret == -EAGAIN) {
 230         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 231     }
 232
 233     return ret;
 234 }
 235
 236 static void migrate_fd_put_ready(void *opaque)
 237 {
 238     MigrationState *s = opaque;
 239     int ret;
 240
 241     if (s->state != MIG_STATE_ACTIVE) {
 242         DPRINTF("put_ready returning because of non-active state\n");
 243         return;
 244     }
 245
 246     DPRINTF("iterate\n");
 247     ret = qemu_savevm_state_iterate(s->file);
 248     if (ret < 0) {
 249         migrate_fd_error(s);
 250     } else if (ret == 1) {
 251         int old_vm_running = runstate_is_running();
 252
 253         DPRINTF("done iterating\n");
 254         vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 255
 256         if (qemu_savevm_state_complete(s->file) < 0) {
 257             migrate_fd_error(s);
 258         } else {
 259             migrate_fd_completed(s);
 260         }
 261         if (s->state != MIG_STATE_COMPLETED) {
 262             if (old_vm_running) {
 263                 vm_start();
 264             }
 265         }
 266     }
 267 }
 268
 269 static void migrate_fd_cancel(MigrationState *s)
 270 {
 271     if (s->state != MIG_STATE_ACTIVE)
 272         return;
 273
 274     DPRINTF("cancelling migration\n");
 275
 276     s->state = MIG_STATE_CANCELLED;
 277     notifier_list_notify(&migration_state_notifiers, s);
 278     qemu_savevm_state_cancel(s->file);
 279
 280     migrate_fd_cleanup(s);
 281 }
 282
 283 static void migrate_fd_wait_for_unfreeze(void *opaque)
 284 {
 285     MigrationState *s = opaque;
 286     int ret;
 287
 288     DPRINTF("wait for unfreeze\n");
 289     if (s->state != MIG_STATE_ACTIVE)
 290         return;
 291
 292     do {
 293         fd_set wfds;
 294
 295         FD_ZERO(&wfds);
 296         FD_SET(s->fd, &wfds);
 297
 298         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 299     } while (ret == -1 && (s->get_error(s)) == EINTR);
 300
 301     if (ret == -1) {
 302         qemu_file_set_error(s->file, -s->get_error(s));
 303     }
 304 }
 305
 306 static int migrate_fd_close(void *opaque)
 307 {
 308     MigrationState *s = opaque;
 309
 310     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 311     return s->close(s);
 312 }
 313
 314 void add_migration_state_change_notifier(Notifier *notify)
 315 {
 316     notifier_list_add(&migration_state_notifiers, notify);
 317 }
 318
 319 void remove_migration_state_change_notifier(Notifier *notify)
 320 {
 321     notifier_remove(notify);
 322 }
 323
 324 bool migration_is_active(MigrationState *s)
 325 {
 326     return s->state == MIG_STATE_ACTIVE;
 327 }
 328
 329 bool migration_has_finished(MigrationState *s)
 330 {
 331     return s->state == MIG_STATE_COMPLETED;
 332 }
 333
 334 bool migration_has_failed(MigrationState *s)
 335 {
 336     return (s->state == MIG_STATE_CANCELLED ||
 337             s->state == MIG_STATE_ERROR);
 338 }
 339
 340 void migrate_fd_connect(MigrationState *s)
 341 {
 342     int ret;
 343
 344     s->state = MIG_STATE_ACTIVE;
 345     s->file = qemu_fopen_ops_buffered(s,
 346                                       s->bandwidth_limit,
 347                                       migrate_fd_put_buffer,
 348                                       migrate_fd_put_ready,
 349                                       migrate_fd_wait_for_unfreeze,
 350                                       migrate_fd_close);
 351
 352     DPRINTF("beginning savevm\n");
 353     ret = qemu_savevm_state_begin(s->file, s->blk, s->shared);
 354     if (ret < 0) {
 355         DPRINTF("failed, %d\n", ret);
 356         migrate_fd_error(s);
 357         return;
 358     }
 359     migrate_fd_put_ready(s);
 360 }
 361
 362 static MigrationState *migrate_init(int blk, int inc)
 363 {
 364     MigrationState *s = migrate_get_current();
 365     int64_t bandwidth_limit = s->bandwidth_limit;
 366
 367     memset(s, 0, sizeof(*s));
 368     s->bandwidth_limit = bandwidth_limit;
 369     s->blk = blk;
 370     s->shared = inc;
 371
 372     s->bandwidth_limit = bandwidth_limit;
 373     s->state = MIG_STATE_SETUP;
 374
 375     return s;
 376 }
 377
 378 static GSList *migration_blockers;
 379
 380 void migrate_add_blocker(Error *reason)
 381 {
 382     migration_blockers = g_slist_prepend(migration_blockers, reason);
 383 }
 384
 385 void migrate_del_blocker(Error *reason)
 386 {
 387     migration_blockers = g_slist_remove(migration_blockers, reason);
 388 }
 389
 390 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 391                  bool has_inc, bool inc, bool has_detach, bool detach,
 392                  Error **errp)
 393 {
 394     MigrationState *s = migrate_get_current();
 395     const char *p;
 396     int ret;
 397
 398     if (s->state == MIG_STATE_ACTIVE) {
 399         error_set(errp, QERR_MIGRATION_ACTIVE);
 400         return;
 401     }
 402
 403     if (qemu_savevm_state_blocked(errp)) {
 404         return;
 405     }
 406
 407     if (migration_blockers) {
 408         *errp = error_copy(migration_blockers->data);
 409         return;
 410     }
 411
 412     s = migrate_init(blk, inc);
 413
 414     if (strstart(uri, "tcp:", &p)) {
 415         ret = tcp_start_outgoing_migration(s, p);
 416 #if !defined(WIN32)
 417     } else if (strstart(uri, "exec:", &p)) {
 418         ret = exec_start_outgoing_migration(s, p);
 419     } else if (strstart(uri, "unix:", &p)) {
 420         ret = unix_start_outgoing_migration(s, p);
 421     } else if (strstart(uri, "fd:", &p)) {
 422         ret = fd_start_outgoing_migration(s, p);
 423 #endif
 424     } else {
 425         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "uri", "a valid migration protocol");
 426         return;
 427     }
 428
 429     if (ret < 0) {
 430         DPRINTF("migration failed: %s\n", strerror(-ret));
 431         /* FIXME: we should return meaningful errors */
 432         error_set(errp, QERR_UNDEFINED_ERROR);
 433         return;
 434     }
 435
 436     notifier_list_notify(&migration_state_notifiers, s);
 437 }
 438
 439 void qmp_migrate_cancel(Error **errp)
 440 {
 441     migrate_fd_cancel(migrate_get_current());
 442 }
 443
 444 void qmp_migrate_set_speed(int64_t value, Error **errp)
 445 {
 446     MigrationState *s;
 447
 448     if (value < 0) {
 449         value = 0;
 450     }
 451
 452     s = migrate_get_current();
 453     s->bandwidth_limit = value;
 454     qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
 455 }
 456
 457 void qmp_migrate_set_downtime(double value, Error **errp)
 458 {
 459     value *= 1e9;
 460     value = MAX(0, MIN(UINT64_MAX, value));
 461     max_downtime = (uint64_t)value;
 462 }