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     if (autostart) {
  74         vm_start();
  75     } else {
  76         runstate_set(RUN_STATE_PRELAUNCH);
  77     }
  78 }
  79
  80 static MigrationState *migrate_new(Monitor *mon, int64_t bandwidth_limit,
  81                                    int detach, int blk, int inc);
  82
  83 int do_migrate(Monitor *mon, const QDict *qdict, QObject **ret_data)
  84 {
  85     MigrationState *s = NULL;
  86     const char *p;
  87     int detach = qdict_get_try_bool(qdict, "detach", 0);
  88     int blk = qdict_get_try_bool(qdict, "blk", 0);
  89     int inc = qdict_get_try_bool(qdict, "inc", 0);
  90     const char *uri = qdict_get_str(qdict, "uri");
  91     int ret;
  92
  93     if (current_migration &&
  94         current_migration->get_status(current_migration) == MIG_STATE_ACTIVE) {
  95         monitor_printf(mon, "migration already in progress\n");
  96         return -1;
  97     }
  98
  99     if (qemu_savevm_state_blocked(mon)) {
 100         return -1;
 101     }
 102
 103     s = migrate_new(mon, max_throttle, detach, blk, inc);
 104
 105     if (strstart(uri, "tcp:", &p)) {
 106         ret = tcp_start_outgoing_migration(s, p);
 107 #if !defined(WIN32)
 108     } else if (strstart(uri, "exec:", &p)) {
 109         ret = exec_start_outgoing_migration(s, p);
 110     } else if (strstart(uri, "unix:", &p)) {
 111         ret = unix_start_outgoing_migration(s, p);
 112     } else if (strstart(uri, "fd:", &p)) {
 113         ret = fd_start_outgoing_migration(s, p);
 114 #endif
 115     } else {
 116         monitor_printf(mon, "unknown migration protocol: %s\n", uri);
 117         ret  = -EINVAL;
 118         goto free_migrate_state;
 119     }
 120
 121     if (ret < 0) {
 122         monitor_printf(mon, "migration failed\n");
 123         goto free_migrate_state;
 124     }
 125
 126     if (current_migration) {
 127         current_migration->release(current_migration);
 128     }
 129
 130     current_migration = s;
 131     notifier_list_notify(&migration_state_notifiers, NULL);
 132     return 0;
 133 free_migrate_state:
 134     g_free(s);
 135     return -1;
 136 }
 137
 138 int do_migrate_cancel(Monitor *mon, const QDict *qdict, QObject **ret_data)
 139 {
 140     MigrationState *s = current_migration;
 141
 142     if (s && s->get_status(s) == MIG_STATE_ACTIVE) {
 143         s->cancel(s);
 144     }
 145     return 0;
 146 }
 147
 148 int do_migrate_set_speed(Monitor *mon, const QDict *qdict, QObject **ret_data)
 149 {
 150     int64_t d;
 151     MigrationState *s;
 152
 153     d = qdict_get_int(qdict, "value");
 154     if (d < 0) {
 155         d = 0;
 156     }
 157     max_throttle = d;
 158
 159     s = current_migration;
 160     if (s && s->file) {
 161         qemu_file_set_rate_limit(s->file, max_throttle);
 162     }
 163
 164     return 0;
 165 }
 166
 167 /* amount of nanoseconds we are willing to wait for migration to be down.
 168  * the choice of nanoseconds is because it is the maximum resolution that
 169  * get_clock() can achieve. It is an internal measure. All user-visible
 170  * units must be in seconds */
 171 static uint64_t max_downtime = 30000000;
 172
 173 uint64_t migrate_max_downtime(void)
 174 {
 175     return max_downtime;
 176 }
 177
 178 int do_migrate_set_downtime(Monitor *mon, const QDict *qdict,
 179                             QObject **ret_data)
 180 {
 181     double d;
 182
 183     d = qdict_get_double(qdict, "value") * 1e9;
 184     d = MAX(0, MIN(UINT64_MAX, d));
 185     max_downtime = (uint64_t)d;
 186
 187     return 0;
 188 }
 189
 190 static void migrate_print_status(Monitor *mon, const char *name,
 191                                  const QDict *status_dict)
 192 {
 193     QDict *qdict;
 194
 195     qdict = qobject_to_qdict(qdict_get(status_dict, name));
 196
 197     monitor_printf(mon, "transferred %s: %" PRIu64 " kbytes\n", name,
 198                         qdict_get_int(qdict, "transferred") >> 10);
 199     monitor_printf(mon, "remaining %s: %" PRIu64 " kbytes\n", name,
 200                         qdict_get_int(qdict, "remaining") >> 10);
 201     monitor_printf(mon, "total %s: %" PRIu64 " kbytes\n", name,
 202                         qdict_get_int(qdict, "total") >> 10);
 203 }
 204
 205 void do_info_migrate_print(Monitor *mon, const QObject *data)
 206 {
 207     QDict *qdict;
 208
 209     qdict = qobject_to_qdict(data);
 210
 211     monitor_printf(mon, "Migration status: %s\n",
 212                    qdict_get_str(qdict, "status"));
 213
 214     if (qdict_haskey(qdict, "ram")) {
 215         migrate_print_status(mon, "ram", qdict);
 216     }
 217
 218     if (qdict_haskey(qdict, "disk")) {
 219         migrate_print_status(mon, "disk", qdict);
 220     }
 221 }
 222
 223 static void migrate_put_status(QDict *qdict, const char *name,
 224                                uint64_t trans, uint64_t rem, uint64_t total)
 225 {
 226     QObject *obj;
 227
 228     obj = qobject_from_jsonf("{ 'transferred': %" PRId64 ", "
 229                                "'remaining': %" PRId64 ", "
 230                                "'total': %" PRId64 " }", trans, rem, total);
 231     qdict_put_obj(qdict, name, obj);
 232 }
 233
 234 void do_info_migrate(Monitor *mon, QObject **ret_data)
 235 {
 236     QDict *qdict;
 237
 238     if (current_migration) {
 239         MigrationState *s = current_migration;
 240
 241         switch (s->get_status(current_migration)) {
 242         case MIG_STATE_SETUP:
 243             /* no migration has happened ever */
 244             break;
 245         case MIG_STATE_ACTIVE:
 246             qdict = qdict_new();
 247             qdict_put(qdict, "status", qstring_from_str("active"));
 248
 249             migrate_put_status(qdict, "ram", ram_bytes_transferred(),
 250                                ram_bytes_remaining(), ram_bytes_total());
 251
 252             if (blk_mig_active()) {
 253                 migrate_put_status(qdict, "disk", blk_mig_bytes_transferred(),
 254                                    blk_mig_bytes_remaining(),
 255                                    blk_mig_bytes_total());
 256             }
 257
 258             *ret_data = QOBJECT(qdict);
 259             break;
 260         case MIG_STATE_COMPLETED:
 261             *ret_data = qobject_from_jsonf("{ 'status': 'completed' }");
 262             break;
 263         case MIG_STATE_ERROR:
 264             *ret_data = qobject_from_jsonf("{ 'status': 'failed' }");
 265             break;
 266         case MIG_STATE_CANCELLED:
 267             *ret_data = qobject_from_jsonf("{ 'status': 'cancelled' }");
 268             break;
 269         }
 270     }
 271 }
 272
 273 /* shared migration helpers */
 274
 275 static void migrate_fd_monitor_suspend(MigrationState *s, Monitor *mon)
 276 {
 277     s->mon = mon;
 278     if (monitor_suspend(mon) == 0) {
 279         DPRINTF("suspending monitor\n");
 280     } else {
 281         monitor_printf(mon, "terminal does not allow synchronous "
 282                        "migration, continuing detached\n");
 283     }
 284 }
 285
 286 static int migrate_fd_cleanup(MigrationState *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_error(MigrationState *s)
 313 {
 314     DPRINTF("setting error state\n");
 315     s->state = MIG_STATE_ERROR;
 316     notifier_list_notify(&migration_state_notifiers, NULL);
 317     migrate_fd_cleanup(s);
 318 }
 319
 320 static void migrate_fd_put_notify(void *opaque)
 321 {
 322     MigrationState *s = opaque;
 323
 324     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 325     qemu_file_put_notify(s->file);
 326     if (qemu_file_get_error(s->file)) {
 327         migrate_fd_error(s);
 328     }
 329 }
 330
 331 static ssize_t migrate_fd_put_buffer(void *opaque, const void *data,
 332                                      size_t size)
 333 {
 334     MigrationState *s = opaque;
 335     ssize_t ret;
 336
 337     if (s->state != MIG_STATE_ACTIVE) {
 338         return -EIO;
 339     }
 340
 341     do {
 342         ret = s->write(s, data, size);
 343     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 344
 345     if (ret == -1)
 346         ret = -(s->get_error(s));
 347
 348     if (ret == -EAGAIN) {
 349         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 350     }
 351
 352     return ret;
 353 }
 354
 355 static void migrate_fd_put_ready(void *opaque)
 356 {
 357     MigrationState *s = opaque;
 358     int ret;
 359
 360     if (s->state != MIG_STATE_ACTIVE) {
 361         DPRINTF("put_ready returning because of non-active state\n");
 362         return;
 363     }
 364
 365     DPRINTF("iterate\n");
 366     ret = qemu_savevm_state_iterate(s->mon, s->file);
 367     if (ret < 0) {
 368         migrate_fd_error(s);
 369     } else if (ret == 1) {
 370         int old_vm_running = runstate_is_running();
 371
 372         DPRINTF("done iterating\n");
 373         vm_stop(RUN_STATE_FINISH_MIGRATE);
 374
 375         if (qemu_savevm_state_complete(s->mon, s->file) < 0) {
 376             migrate_fd_error(s);
 377         } else {
 378             if (migrate_fd_cleanup(s) < 0) {
 379                 migrate_fd_error(s);
 380             } else {
 381                 s->state = MIG_STATE_COMPLETED;
 382                 runstate_set(RUN_STATE_POSTMIGRATE);
 383                 notifier_list_notify(&migration_state_notifiers, NULL);
 384             }
 385         }
 386         if (s->get_status(s) != MIG_STATE_COMPLETED) {
 387             if (old_vm_running) {
 388                 vm_start();
 389             }
 390         }
 391     }
 392 }
 393
 394 static int migrate_fd_get_status(MigrationState *s)
 395 {
 396     return s->state;
 397 }
 398
 399 static void migrate_fd_cancel(MigrationState *s)
 400 {
 401     if (s->state != MIG_STATE_ACTIVE)
 402         return;
 403
 404     DPRINTF("cancelling migration\n");
 405
 406     s->state = MIG_STATE_CANCELLED;
 407     notifier_list_notify(&migration_state_notifiers, NULL);
 408     qemu_savevm_state_cancel(s->mon, s->file);
 409
 410     migrate_fd_cleanup(s);
 411 }
 412
 413 static void migrate_fd_release(MigrationState *s)
 414 {
 415
 416     DPRINTF("releasing state\n");
 417
 418     if (s->state == MIG_STATE_ACTIVE) {
 419         s->state = MIG_STATE_CANCELLED;
 420         notifier_list_notify(&migration_state_notifiers, NULL);
 421         migrate_fd_cleanup(s);
 422     }
 423     g_free(s);
 424 }
 425
 426 static void migrate_fd_wait_for_unfreeze(void *opaque)
 427 {
 428     MigrationState *s = opaque;
 429     int ret;
 430
 431     DPRINTF("wait for unfreeze\n");
 432     if (s->state != MIG_STATE_ACTIVE)
 433         return;
 434
 435     do {
 436         fd_set wfds;
 437
 438         FD_ZERO(&wfds);
 439         FD_SET(s->fd, &wfds);
 440
 441         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 442     } while (ret == -1 && (s->get_error(s)) == EINTR);
 443
 444     if (ret == -1) {
 445         qemu_file_set_error(s->file, -s->get_error(s));
 446     }
 447 }
 448
 449 static int migrate_fd_close(void *opaque)
 450 {
 451     MigrationState *s = opaque;
 452
 453     if (s->mon) {
 454         monitor_resume(s->mon);
 455     }
 456     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 457     return s->close(s);
 458 }
 459
 460 void add_migration_state_change_notifier(Notifier *notify)
 461 {
 462     notifier_list_add(&migration_state_notifiers, notify);
 463 }
 464
 465 void remove_migration_state_change_notifier(Notifier *notify)
 466 {
 467     notifier_list_remove(&migration_state_notifiers, notify);
 468 }
 469
 470 int get_migration_state(void)
 471 {
 472     if (current_migration) {
 473         return migrate_fd_get_status(current_migration);
 474     } else {
 475         return MIG_STATE_ERROR;
 476     }
 477 }
 478
 479 void migrate_fd_connect(MigrationState *s)
 480 {
 481     int ret;
 482
 483     s->state = MIG_STATE_ACTIVE;
 484     s->file = qemu_fopen_ops_buffered(s,
 485                                       s->bandwidth_limit,
 486                                       migrate_fd_put_buffer,
 487                                       migrate_fd_put_ready,
 488                                       migrate_fd_wait_for_unfreeze,
 489                                       migrate_fd_close);
 490
 491     DPRINTF("beginning savevm\n");
 492     ret = qemu_savevm_state_begin(s->mon, s->file, s->blk, s->shared);
 493     if (ret < 0) {
 494         DPRINTF("failed, %d\n", ret);
 495         migrate_fd_error(s);
 496         return;
 497     }
 498     migrate_fd_put_ready(s);
 499 }
 500
 501 static MigrationState *migrate_new(Monitor *mon, int64_t bandwidth_limit,
 502                                    int detach, int blk, int inc)
 503 {
 504     MigrationState *s = g_malloc0(sizeof(*s));
 505
 506     s->cancel = migrate_fd_cancel;
 507     s->get_status = migrate_fd_get_status;
 508     s->release = migrate_fd_release;
 509     s->blk = blk;
 510     s->shared = inc;
 511     s->mon = NULL;
 512     s->bandwidth_limit = bandwidth_limit;
 513     s->state = MIG_STATE_SETUP;
 514
 515     if (!detach) {
 516         migrate_fd_monitor_suspend(s, mon);
 517     }
 518
 519     return s;
 520 }