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  */
  13
  14 #include "qemu-common.h"
  15 #include "migration.h"
  16 #include "monitor.h"
  17 #include "buffered_file.h"
  18 #include "sysemu.h"
  19 #include "block.h"
  20 #include "qemu_socket.h"
  21 #include "block-migration.h"
  22 #include "qemu-objects.h"
  23
  24 //#define DEBUG_MIGRATION
  25
  26 #ifdef DEBUG_MIGRATION
  27 #define DPRINTF(fmt, ...) \
  28     do { printf("migration: " fmt, ## __VA_ARGS__); } while (0)
  29 #else
  30 #define DPRINTF(fmt, ...) \
  31     do { } while (0)
  32 #endif
  33
  34 /* Migration speed throttling */
  35 static int64_t max_throttle = (32 << 20);
  36
  37 static MigrationState *current_migration;
  38
  39 static NotifierList migration_state_notifiers =
  40     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
  41
  42 int qemu_start_incoming_migration(const char *uri)
  43 {
  44     const char *p;
  45     int ret;
  46
  47     if (strstart(uri, "tcp:", &p))
  48         ret = tcp_start_incoming_migration(p);
  49 #if !defined(WIN32)
  50     else if (strstart(uri, "exec:", &p))
  51         ret =  exec_start_incoming_migration(p);
  52     else if (strstart(uri, "unix:", &p))
  53         ret = unix_start_incoming_migration(p);
  54     else if (strstart(uri, "fd:", &p))
  55         ret = fd_start_incoming_migration(p);
  56 #endif
  57     else {
  58         fprintf(stderr, "unknown migration protocol: %s\n", uri);
  59         ret = -EPROTONOSUPPORT;
  60     }
  61     return ret;
  62 }
  63
  64 void process_incoming_migration(QEMUFile *f)
  65 {
  66     if (qemu_loadvm_state(f) < 0) {
  67         fprintf(stderr, "load of migration failed\n");
  68         exit(0);
  69     }
  70     qemu_announce_self();
  71     DPRINTF("successfully loaded vm state\n");
  72
  73     incoming_expected = false;
  74
  75     if (autostart) {
  76         vm_start();
  77     } else {
  78         runstate_set(RSTATE_PRE_LAUNCH);
  79     }
  80 }
  81
  82 int do_migrate(Monitor *mon, const QDict *qdict, QObject **ret_data)
  83 {
  84     MigrationState *s = NULL;
  85     const char *p;
  86     int detach = qdict_get_try_bool(qdict, "detach", 0);
  87     int blk = qdict_get_try_bool(qdict, "blk", 0);
  88     int inc = qdict_get_try_bool(qdict, "inc", 0);
  89     const char *uri = qdict_get_str(qdict, "uri");
  90
  91     if (current_migration &&
  92         current_migration->get_status(current_migration) == MIG_STATE_ACTIVE) {
  93         monitor_printf(mon, "migration already in progress\n");
  94         return -1;
  95     }
  96
  97     if (qemu_savevm_state_blocked(mon)) {
  98         return -1;
  99     }
 100
 101     if (strstart(uri, "tcp:", &p)) {
 102         s = tcp_start_outgoing_migration(mon, p, max_throttle, detach,
 103                                          blk, inc);
 104 #if !defined(WIN32)
 105     } else if (strstart(uri, "exec:", &p)) {
 106         s = exec_start_outgoing_migration(mon, p, max_throttle, detach,
 107                                           blk, inc);
 108     } else if (strstart(uri, "unix:", &p)) {
 109         s = unix_start_outgoing_migration(mon, p, max_throttle, detach,
 110                                           blk, inc);
 111     } else if (strstart(uri, "fd:", &p)) {
 112         s = fd_start_outgoing_migration(mon, p, max_throttle, detach,
 113                                         blk, inc);
 114 #endif
 115     } else {
 116         monitor_printf(mon, "unknown migration protocol: %s\n", uri);
 117         return -1;
 118     }
 119
 120     if (s == NULL) {
 121         monitor_printf(mon, "migration failed\n");
 122         return -1;
 123     }
 124
 125     if (current_migration) {
 126         current_migration->release(current_migration);
 127     }
 128
 129     current_migration = s;
 130     notifier_list_notify(&migration_state_notifiers, NULL);
 131     return 0;
 132 }
 133
 134 int do_migrate_cancel(Monitor *mon, const QDict *qdict, QObject **ret_data)
 135 {
 136     MigrationState *s = current_migration;
 137
 138     if (s)
 139         s->cancel(s);
 140
 141     return 0;
 142 }
 143
 144 int do_migrate_set_speed(Monitor *mon, const QDict *qdict, QObject **ret_data)
 145 {
 146     int64_t d;
 147     FdMigrationState *s;
 148
 149     d = qdict_get_int(qdict, "value");
 150     if (d < 0) {
 151         d = 0;
 152     }
 153     max_throttle = d;
 154
 155     s = migrate_to_fms(current_migration);
 156     if (s && s->file) {
 157         qemu_file_set_rate_limit(s->file, max_throttle);
 158     }
 159
 160     return 0;
 161 }
 162
 163 /* amount of nanoseconds we are willing to wait for migration to be down.
 164  * the choice of nanoseconds is because it is the maximum resolution that
 165  * get_clock() can achieve. It is an internal measure. All user-visible
 166  * units must be in seconds */
 167 static uint64_t max_downtime = 30000000;
 168
 169 uint64_t migrate_max_downtime(void)
 170 {
 171     return max_downtime;
 172 }
 173
 174 int do_migrate_set_downtime(Monitor *mon, const QDict *qdict,
 175                             QObject **ret_data)
 176 {
 177     double d;
 178
 179     d = qdict_get_double(qdict, "value") * 1e9;
 180     d = MAX(0, MIN(UINT64_MAX, d));
 181     max_downtime = (uint64_t)d;
 182
 183     return 0;
 184 }
 185
 186 static void migrate_print_status(Monitor *mon, const char *name,
 187                                  const QDict *status_dict)
 188 {
 189     QDict *qdict;
 190
 191     qdict = qobject_to_qdict(qdict_get(status_dict, name));
 192
 193     monitor_printf(mon, "transferred %s: %" PRIu64 " kbytes\n", name,
 194                         qdict_get_int(qdict, "transferred") >> 10);
 195     monitor_printf(mon, "remaining %s: %" PRIu64 " kbytes\n", name,
 196                         qdict_get_int(qdict, "remaining") >> 10);
 197     monitor_printf(mon, "total %s: %" PRIu64 " kbytes\n", name,
 198                         qdict_get_int(qdict, "total") >> 10);
 199 }
 200
 201 void do_info_migrate_print(Monitor *mon, const QObject *data)
 202 {
 203     QDict *qdict;
 204
 205     qdict = qobject_to_qdict(data);
 206
 207     monitor_printf(mon, "Migration status: %s\n",
 208                    qdict_get_str(qdict, "status"));
 209
 210     if (qdict_haskey(qdict, "ram")) {
 211         migrate_print_status(mon, "ram", qdict);
 212     }
 213
 214     if (qdict_haskey(qdict, "disk")) {
 215         migrate_print_status(mon, "disk", qdict);
 216     }
 217 }
 218
 219 static void migrate_put_status(QDict *qdict, const char *name,
 220                                uint64_t trans, uint64_t rem, uint64_t total)
 221 {
 222     QObject *obj;
 223
 224     obj = qobject_from_jsonf("{ 'transferred': %" PRId64 ", "
 225                                "'remaining': %" PRId64 ", "
 226                                "'total': %" PRId64 " }", trans, rem, total);
 227     qdict_put_obj(qdict, name, obj);
 228 }
 229
 230 void do_info_migrate(Monitor *mon, QObject **ret_data)
 231 {
 232     QDict *qdict;
 233     MigrationState *s = current_migration;
 234
 235     if (s) {
 236         switch (s->get_status(s)) {
 237         case MIG_STATE_ACTIVE:
 238             qdict = qdict_new();
 239             qdict_put(qdict, "status", qstring_from_str("active"));
 240
 241             migrate_put_status(qdict, "ram", ram_bytes_transferred(),
 242                                ram_bytes_remaining(), ram_bytes_total());
 243
 244             if (blk_mig_active()) {
 245                 migrate_put_status(qdict, "disk", blk_mig_bytes_transferred(),
 246                                    blk_mig_bytes_remaining(),
 247                                    blk_mig_bytes_total());
 248             }
 249
 250             *ret_data = QOBJECT(qdict);
 251             break;
 252         case MIG_STATE_COMPLETED:
 253             *ret_data = qobject_from_jsonf("{ 'status': 'completed' }");
 254             break;
 255         case MIG_STATE_ERROR:
 256             *ret_data = qobject_from_jsonf("{ 'status': 'failed' }");
 257             break;
 258         case MIG_STATE_CANCELLED:
 259             *ret_data = qobject_from_jsonf("{ 'status': 'cancelled' }");
 260             break;
 261         }
 262     }
 263 }
 264
 265 /* shared migration helpers */
 266
 267 void migrate_fd_monitor_suspend(FdMigrationState *s, Monitor *mon)
 268 {
 269     s->mon = mon;
 270     if (monitor_suspend(mon) == 0) {
 271         DPRINTF("suspending monitor\n");
 272     } else {
 273         monitor_printf(mon, "terminal does not allow synchronous "
 274                        "migration, continuing detached\n");
 275     }
 276 }
 277
 278 void migrate_fd_error(FdMigrationState *s)
 279 {
 280     DPRINTF("setting error state\n");
 281     s->state = MIG_STATE_ERROR;
 282     notifier_list_notify(&migration_state_notifiers, NULL);
 283     migrate_fd_cleanup(s);
 284 }
 285
 286 int migrate_fd_cleanup(FdMigrationState *s)
 287 {
 288     int ret = 0;
 289
 290     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 291
 292     if (s->file) {
 293         DPRINTF("closing file\n");
 294         if (qemu_fclose(s->file) != 0) {
 295             ret = -1;
 296         }
 297         s->file = NULL;
 298     } else {
 299         if (s->mon) {
 300             monitor_resume(s->mon);
 301         }
 302     }
 303
 304     if (s->fd != -1) {
 305         close(s->fd);
 306         s->fd = -1;
 307     }
 308
 309     return ret;
 310 }
 311
 312 void migrate_fd_put_notify(void *opaque)
 313 {
 314     FdMigrationState *s = opaque;
 315
 316     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 317     qemu_file_put_notify(s->file);
 318 }
 319
 320 ssize_t migrate_fd_put_buffer(void *opaque, const void *data, size_t size)
 321 {
 322     FdMigrationState *s = opaque;
 323     ssize_t ret;
 324
 325     do {
 326         ret = s->write(s, data, size);
 327     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 328
 329     if (ret == -1)
 330         ret = -(s->get_error(s));
 331
 332     if (ret == -EAGAIN) {
 333         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 334     } else if (ret < 0) {
 335         s->state = MIG_STATE_ERROR;
 336         notifier_list_notify(&migration_state_notifiers, NULL);
 337     }
 338
 339     return ret;
 340 }
 341
 342 void migrate_fd_connect(FdMigrationState *s)
 343 {
 344     int ret;
 345
 346     s->file = qemu_fopen_ops_buffered(s,
 347                                       s->bandwidth_limit,
 348                                       migrate_fd_put_buffer,
 349                                       migrate_fd_put_ready,
 350                                       migrate_fd_wait_for_unfreeze,
 351                                       migrate_fd_close);
 352
 353     DPRINTF("beginning savevm\n");
 354     ret = qemu_savevm_state_begin(s->mon, s->file, s->mig_state.blk,
 355                                   s->mig_state.shared);
 356     if (ret < 0) {
 357         DPRINTF("failed, %d\n", ret);
 358         migrate_fd_error(s);
 359         return;
 360     }
 361
 362     migrate_fd_put_ready(s);
 363 }
 364
 365 void migrate_fd_put_ready(void *opaque)
 366 {
 367     FdMigrationState *s = opaque;
 368
 369     if (s->state != MIG_STATE_ACTIVE) {
 370         DPRINTF("put_ready returning because of non-active state\n");
 371         return;
 372     }
 373
 374     DPRINTF("iterate\n");
 375     if (qemu_savevm_state_iterate(s->mon, s->file) == 1) {
 376         int state;
 377         int old_vm_running = vm_running;
 378
 379         DPRINTF("done iterating\n");
 380         vm_stop(RSTATE_PRE_MIGRATE);
 381
 382         if ((qemu_savevm_state_complete(s->mon, s->file)) < 0) {
 383             if (old_vm_running) {
 384                 vm_start();
 385             }
 386             state = MIG_STATE_ERROR;
 387         } else {
 388             state = MIG_STATE_COMPLETED;
 389         }
 390         if (migrate_fd_cleanup(s) < 0) {
 391             if (old_vm_running) {
 392                 vm_start();
 393             }
 394             state = MIG_STATE_ERROR;
 395         }
 396         if (state == MIG_STATE_COMPLETED) {
 397             runstate_set(RSTATE_POST_MIGRATE);
 398         }
 399         s->state = state;
 400         notifier_list_notify(&migration_state_notifiers, NULL);
 401     }
 402 }
 403
 404 int migrate_fd_get_status(MigrationState *mig_state)
 405 {
 406     FdMigrationState *s = migrate_to_fms(mig_state);
 407     return s->state;
 408 }
 409
 410 void migrate_fd_cancel(MigrationState *mig_state)
 411 {
 412     FdMigrationState *s = migrate_to_fms(mig_state);
 413
 414     if (s->state != MIG_STATE_ACTIVE)
 415         return;
 416
 417     DPRINTF("cancelling migration\n");
 418
 419     s->state = MIG_STATE_CANCELLED;
 420     notifier_list_notify(&migration_state_notifiers, NULL);
 421     qemu_savevm_state_cancel(s->mon, s->file);
 422
 423     migrate_fd_cleanup(s);
 424 }
 425
 426 void migrate_fd_release(MigrationState *mig_state)
 427 {
 428     FdMigrationState *s = migrate_to_fms(mig_state);
 429
 430     DPRINTF("releasing state\n");
 431
 432     if (s->state == MIG_STATE_ACTIVE) {
 433         s->state = MIG_STATE_CANCELLED;
 434         notifier_list_notify(&migration_state_notifiers, NULL);
 435         migrate_fd_cleanup(s);
 436     }
 437     g_free(s);
 438 }
 439
 440 void migrate_fd_wait_for_unfreeze(void *opaque)
 441 {
 442     FdMigrationState *s = opaque;
 443     int ret;
 444
 445     DPRINTF("wait for unfreeze\n");
 446     if (s->state != MIG_STATE_ACTIVE)
 447         return;
 448
 449     do {
 450         fd_set wfds;
 451
 452         FD_ZERO(&wfds);
 453         FD_SET(s->fd, &wfds);
 454
 455         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 456     } while (ret == -1 && (s->get_error(s)) == EINTR);
 457 }
 458
 459 int migrate_fd_close(void *opaque)
 460 {
 461     FdMigrationState *s = opaque;
 462
 463     if (s->mon) {
 464         monitor_resume(s->mon);
 465     }
 466     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 467     return s->close(s);
 468 }
 469
 470 void add_migration_state_change_notifier(Notifier *notify)
 471 {
 472     notifier_list_add(&migration_state_notifiers, notify);
 473 }
 474
 475 void remove_migration_state_change_notifier(Notifier *notify)
 476 {
 477     notifier_list_remove(&migration_state_notifiers, notify);
 478 }
 479
 480 int get_migration_state(void)
 481 {
 482     if (current_migration) {
 483         return migrate_fd_get_status(current_migration);
 484     } else {
 485         return MIG_STATE_ERROR;
 486     }
 487 }