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 FdMigrationState *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     if (autostart) {
  74         vm_start();
  75     } else {
  76         runstate_set(RUN_STATE_PRELAUNCH);
  77     }
  78 }
  79
  80 int do_migrate(Monitor *mon, const QDict *qdict, QObject **ret_data)
  81 {
  82     FdMigrationState *s = NULL;
  83     const char *p;
  84     int detach = qdict_get_try_bool(qdict, "detach", 0);
  85     int blk = qdict_get_try_bool(qdict, "blk", 0);
  86     int inc = qdict_get_try_bool(qdict, "inc", 0);
  87     const char *uri = qdict_get_str(qdict, "uri");
  88
  89     if (current_migration &&
  90         current_migration->mig_state.get_status(current_migration) ==
  91         MIG_STATE_ACTIVE) {
  92         monitor_printf(mon, "migration already in progress\n");
  93         return -1;
  94     }
  95
  96     if (qemu_savevm_state_blocked(mon)) {
  97         return -1;
  98     }
  99
 100     if (strstart(uri, "tcp:", &p)) {
 101         s = tcp_start_outgoing_migration(mon, p, max_throttle, detach,
 102                                          blk, inc);
 103 #if !defined(WIN32)
 104     } else if (strstart(uri, "exec:", &p)) {
 105         s = exec_start_outgoing_migration(mon, p, max_throttle, detach,
 106                                           blk, inc);
 107     } else if (strstart(uri, "unix:", &p)) {
 108         s = unix_start_outgoing_migration(mon, p, max_throttle, detach,
 109                                           blk, inc);
 110     } else if (strstart(uri, "fd:", &p)) {
 111         s = fd_start_outgoing_migration(mon, p, max_throttle, detach,
 112                                         blk, inc);
 113 #endif
 114     } else {
 115         monitor_printf(mon, "unknown migration protocol: %s\n", uri);
 116         return -1;
 117     }
 118
 119     if (s == NULL) {
 120         monitor_printf(mon, "migration failed\n");
 121         return -1;
 122     }
 123
 124     if (current_migration) {
 125         current_migration->mig_state.release(current_migration);
 126     }
 127
 128     current_migration = s;
 129     notifier_list_notify(&migration_state_notifiers, NULL);
 130     return 0;
 131 }
 132
 133 int do_migrate_cancel(Monitor *mon, const QDict *qdict, QObject **ret_data)
 134 {
 135     FdMigrationState *s = current_migration;
 136
 137     if (s && s->mig_state.get_status(s) == MIG_STATE_ACTIVE) {
 138         s->mig_state.cancel(s);
 139     }
 140     return 0;
 141 }
 142
 143 int do_migrate_set_speed(Monitor *mon, const QDict *qdict, QObject **ret_data)
 144 {
 145     int64_t d;
 146     FdMigrationState *s;
 147
 148     d = qdict_get_int(qdict, "value");
 149     if (d < 0) {
 150         d = 0;
 151     }
 152     max_throttle = d;
 153
 154     s = current_migration;
 155     if (s && s->file) {
 156         qemu_file_set_rate_limit(s->file, max_throttle);
 157     }
 158
 159     return 0;
 160 }
 161
 162 /* amount of nanoseconds we are willing to wait for migration to be down.
 163  * the choice of nanoseconds is because it is the maximum resolution that
 164  * get_clock() can achieve. It is an internal measure. All user-visible
 165  * units must be in seconds */
 166 static uint64_t max_downtime = 30000000;
 167
 168 uint64_t migrate_max_downtime(void)
 169 {
 170     return max_downtime;
 171 }
 172
 173 int do_migrate_set_downtime(Monitor *mon, const QDict *qdict,
 174                             QObject **ret_data)
 175 {
 176     double d;
 177
 178     d = qdict_get_double(qdict, "value") * 1e9;
 179     d = MAX(0, MIN(UINT64_MAX, d));
 180     max_downtime = (uint64_t)d;
 181
 182     return 0;
 183 }
 184
 185 static void migrate_print_status(Monitor *mon, const char *name,
 186                                  const QDict *status_dict)
 187 {
 188     QDict *qdict;
 189
 190     qdict = qobject_to_qdict(qdict_get(status_dict, name));
 191
 192     monitor_printf(mon, "transferred %s: %" PRIu64 " kbytes\n", name,
 193                         qdict_get_int(qdict, "transferred") >> 10);
 194     monitor_printf(mon, "remaining %s: %" PRIu64 " kbytes\n", name,
 195                         qdict_get_int(qdict, "remaining") >> 10);
 196     monitor_printf(mon, "total %s: %" PRIu64 " kbytes\n", name,
 197                         qdict_get_int(qdict, "total") >> 10);
 198 }
 199
 200 void do_info_migrate_print(Monitor *mon, const QObject *data)
 201 {
 202     QDict *qdict;
 203
 204     qdict = qobject_to_qdict(data);
 205
 206     monitor_printf(mon, "Migration status: %s\n",
 207                    qdict_get_str(qdict, "status"));
 208
 209     if (qdict_haskey(qdict, "ram")) {
 210         migrate_print_status(mon, "ram", qdict);
 211     }
 212
 213     if (qdict_haskey(qdict, "disk")) {
 214         migrate_print_status(mon, "disk", qdict);
 215     }
 216 }
 217
 218 static void migrate_put_status(QDict *qdict, const char *name,
 219                                uint64_t trans, uint64_t rem, uint64_t total)
 220 {
 221     QObject *obj;
 222
 223     obj = qobject_from_jsonf("{ 'transferred': %" PRId64 ", "
 224                                "'remaining': %" PRId64 ", "
 225                                "'total': %" PRId64 " }", trans, rem, total);
 226     qdict_put_obj(qdict, name, obj);
 227 }
 228
 229 void do_info_migrate(Monitor *mon, QObject **ret_data)
 230 {
 231     QDict *qdict;
 232
 233     if (current_migration) {
 234         MigrationState *s = &current_migration->mig_state;
 235
 236         switch (s->get_status(current_migration)) {
 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     if (qemu_file_get_error(s->file)) {
 319         migrate_fd_error(s);
 320     }
 321 }
 322
 323 ssize_t migrate_fd_put_buffer(void *opaque, const void *data, size_t size)
 324 {
 325     FdMigrationState *s = opaque;
 326     ssize_t ret;
 327
 328     if (s->state != MIG_STATE_ACTIVE) {
 329         return -EIO;
 330     }
 331
 332     do {
 333         ret = s->write(s, data, size);
 334     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 335
 336     if (ret == -1)
 337         ret = -(s->get_error(s));
 338
 339     if (ret == -EAGAIN) {
 340         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 341     }
 342
 343     return ret;
 344 }
 345
 346 void migrate_fd_connect(FdMigrationState *s)
 347 {
 348     int ret;
 349
 350     s->file = qemu_fopen_ops_buffered(s,
 351                                       s->bandwidth_limit,
 352                                       migrate_fd_put_buffer,
 353                                       migrate_fd_put_ready,
 354                                       migrate_fd_wait_for_unfreeze,
 355                                       migrate_fd_close);
 356
 357     DPRINTF("beginning savevm\n");
 358     ret = qemu_savevm_state_begin(s->mon, s->file, s->mig_state.blk,
 359                                   s->mig_state.shared);
 360     if (ret < 0) {
 361         DPRINTF("failed, %d\n", ret);
 362         migrate_fd_error(s);
 363         return;
 364     }
 365
 366     migrate_fd_put_ready(s);
 367 }
 368
 369 void migrate_fd_put_ready(void *opaque)
 370 {
 371     FdMigrationState *s = opaque;
 372     int ret;
 373
 374     if (s->state != MIG_STATE_ACTIVE) {
 375         DPRINTF("put_ready returning because of non-active state\n");
 376         return;
 377     }
 378
 379     DPRINTF("iterate\n");
 380     ret = qemu_savevm_state_iterate(s->mon, s->file);
 381     if (ret < 0) {
 382         migrate_fd_error(s);
 383     } else if (ret == 1) {
 384         int old_vm_running = runstate_is_running();
 385
 386         DPRINTF("done iterating\n");
 387         vm_stop(RUN_STATE_FINISH_MIGRATE);
 388
 389         if ((qemu_savevm_state_complete(s->mon, s->file)) < 0) {
 390             if (old_vm_running) {
 391                 vm_start();
 392             }
 393             s->state = MIG_STATE_ERROR;
 394         }
 395         if (migrate_fd_cleanup(s) < 0) {
 396             if (old_vm_running) {
 397                 vm_start();
 398             }
 399             s->state = MIG_STATE_ERROR;
 400         }
 401         if (s->state == MIG_STATE_ACTIVE) {
 402             s->state = MIG_STATE_COMPLETED;
 403             runstate_set(RUN_STATE_POSTMIGRATE);
 404         }
 405         notifier_list_notify(&migration_state_notifiers, NULL);
 406     }
 407 }
 408
 409 int migrate_fd_get_status(FdMigrationState *s)
 410 {
 411     return s->state;
 412 }
 413
 414 void migrate_fd_cancel(FdMigrationState *s)
 415 {
 416     if (s->state != MIG_STATE_ACTIVE)
 417         return;
 418
 419     DPRINTF("cancelling migration\n");
 420
 421     s->state = MIG_STATE_CANCELLED;
 422     notifier_list_notify(&migration_state_notifiers, NULL);
 423     qemu_savevm_state_cancel(s->mon, s->file);
 424
 425     migrate_fd_cleanup(s);
 426 }
 427
 428 void migrate_fd_release(FdMigrationState *s)
 429 {
 430
 431     DPRINTF("releasing state\n");
 432
 433     if (s->state == MIG_STATE_ACTIVE) {
 434         s->state = MIG_STATE_CANCELLED;
 435         notifier_list_notify(&migration_state_notifiers, NULL);
 436         migrate_fd_cleanup(s);
 437     }
 438     g_free(s);
 439 }
 440
 441 void migrate_fd_wait_for_unfreeze(void *opaque)
 442 {
 443     FdMigrationState *s = opaque;
 444     int ret;
 445
 446     DPRINTF("wait for unfreeze\n");
 447     if (s->state != MIG_STATE_ACTIVE)
 448         return;
 449
 450     do {
 451         fd_set wfds;
 452
 453         FD_ZERO(&wfds);
 454         FD_SET(s->fd, &wfds);
 455
 456         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 457     } while (ret == -1 && (s->get_error(s)) == EINTR);
 458
 459     if (ret == -1) {
 460         qemu_file_set_error(s->file, -s->get_error(s));
 461     }
 462 }
 463
 464 int migrate_fd_close(void *opaque)
 465 {
 466     FdMigrationState *s = opaque;
 467
 468     if (s->mon) {
 469         monitor_resume(s->mon);
 470     }
 471     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 472     return s->close(s);
 473 }
 474
 475 void add_migration_state_change_notifier(Notifier *notify)
 476 {
 477     notifier_list_add(&migration_state_notifiers, notify);
 478 }
 479
 480 void remove_migration_state_change_notifier(Notifier *notify)
 481 {
 482     notifier_list_remove(&migration_state_notifiers, notify);
 483 }
 484
 485 int get_migration_state(void)
 486 {
 487     if (current_migration) {
 488         return migrate_fd_get_status(current_migration);
 489     } else {
 490         return MIG_STATE_ERROR;
 491     }
 492 }