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 enum {
  35     MIG_STATE_ERROR,
  36     MIG_STATE_SETUP,
  37     MIG_STATE_CANCELLED,
  38     MIG_STATE_ACTIVE,
  39     MIG_STATE_COMPLETED,
  40 };
  41
  42 #define MAX_THROTTLE  (32 << 20)      /* Migration speed throttling */
  43
  44 static NotifierList migration_state_notifiers =
  45     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
  46
  47 /* When we add fault tolerance, we could have several
  48    migrations at once.  For now we don't need to add
  49    dynamic creation of migration */
  50
  51 static MigrationState *migrate_get_current(void)
  52 {
  53     static MigrationState current_migration = {
  54         .state = MIG_STATE_SETUP,
  55         .bandwidth_limit = MAX_THROTTLE,
  56     };
  57
  58     return &current_migration;
  59 }
  60
  61 int qemu_start_incoming_migration(const char *uri)
  62 {
  63     const char *p;
  64     int ret;
  65
  66     if (strstart(uri, "tcp:", &p))
  67         ret = tcp_start_incoming_migration(p);
  68 #if !defined(WIN32)
  69     else if (strstart(uri, "exec:", &p))
  70         ret =  exec_start_incoming_migration(p);
  71     else if (strstart(uri, "unix:", &p))
  72         ret = unix_start_incoming_migration(p);
  73     else if (strstart(uri, "fd:", &p))
  74         ret = fd_start_incoming_migration(p);
  75 #endif
  76     else {
  77         fprintf(stderr, "unknown migration protocol: %s\n", uri);
  78         ret = -EPROTONOSUPPORT;
  79     }
  80     return ret;
  81 }
  82
  83 void process_incoming_migration(QEMUFile *f)
  84 {
  85     if (qemu_loadvm_state(f) < 0) {
  86         fprintf(stderr, "load of migration failed\n");
  87         exit(0);
  88     }
  89     qemu_announce_self();
  90     DPRINTF("successfully loaded vm state\n");
  91
  92     if (autostart) {
  93         vm_start();
  94     } else {
  95         runstate_set(RUN_STATE_PRELAUNCH);
  96     }
  97 }
  98
  99 /* amount of nanoseconds we are willing to wait for migration to be down.
 100  * the choice of nanoseconds is because it is the maximum resolution that
 101  * get_clock() can achieve. It is an internal measure. All user-visible
 102  * units must be in seconds */
 103 static uint64_t max_downtime = 30000000;
 104
 105 uint64_t migrate_max_downtime(void)
 106 {
 107     return max_downtime;
 108 }
 109
 110 static void migrate_print_status(Monitor *mon, const char *name,
 111                                  const QDict *status_dict)
 112 {
 113     QDict *qdict;
 114
 115     qdict = qobject_to_qdict(qdict_get(status_dict, name));
 116
 117     monitor_printf(mon, "transferred %s: %" PRIu64 " kbytes\n", name,
 118                         qdict_get_int(qdict, "transferred") >> 10);
 119     monitor_printf(mon, "remaining %s: %" PRIu64 " kbytes\n", name,
 120                         qdict_get_int(qdict, "remaining") >> 10);
 121     monitor_printf(mon, "total %s: %" PRIu64 " kbytes\n", name,
 122                         qdict_get_int(qdict, "total") >> 10);
 123 }
 124
 125 void do_info_migrate_print(Monitor *mon, const QObject *data)
 126 {
 127     QDict *qdict;
 128
 129     qdict = qobject_to_qdict(data);
 130
 131     monitor_printf(mon, "Migration status: %s\n",
 132                    qdict_get_str(qdict, "status"));
 133
 134     if (qdict_haskey(qdict, "ram")) {
 135         migrate_print_status(mon, "ram", qdict);
 136     }
 137
 138     if (qdict_haskey(qdict, "disk")) {
 139         migrate_print_status(mon, "disk", qdict);
 140     }
 141 }
 142
 143 static void migrate_put_status(QDict *qdict, const char *name,
 144                                uint64_t trans, uint64_t rem, uint64_t total)
 145 {
 146     QObject *obj;
 147
 148     obj = qobject_from_jsonf("{ 'transferred': %" PRId64 ", "
 149                                "'remaining': %" PRId64 ", "
 150                                "'total': %" PRId64 " }", trans, rem, total);
 151     qdict_put_obj(qdict, name, obj);
 152 }
 153
 154 void do_info_migrate(Monitor *mon, QObject **ret_data)
 155 {
 156     QDict *qdict;
 157     MigrationState *s = migrate_get_current();
 158
 159     switch (s->state) {
 160     case MIG_STATE_SETUP:
 161         /* no migration has happened ever */
 162         break;
 163     case MIG_STATE_ACTIVE:
 164         qdict = qdict_new();
 165         qdict_put(qdict, "status", qstring_from_str("active"));
 166
 167         migrate_put_status(qdict, "ram", ram_bytes_transferred(),
 168                            ram_bytes_remaining(), ram_bytes_total());
 169
 170         if (blk_mig_active()) {
 171             migrate_put_status(qdict, "disk", blk_mig_bytes_transferred(),
 172                                blk_mig_bytes_remaining(),
 173                                blk_mig_bytes_total());
 174         }
 175
 176         *ret_data = QOBJECT(qdict);
 177         break;
 178     case MIG_STATE_COMPLETED:
 179         *ret_data = qobject_from_jsonf("{ 'status': 'completed' }");
 180         break;
 181     case MIG_STATE_ERROR:
 182         *ret_data = qobject_from_jsonf("{ 'status': 'failed' }");
 183         break;
 184     case MIG_STATE_CANCELLED:
 185         *ret_data = qobject_from_jsonf("{ 'status': 'cancelled' }");
 186         break;
 187     }
 188 }
 189
 190 /* shared migration helpers */
 191
 192 static void migrate_fd_monitor_suspend(MigrationState *s, Monitor *mon)
 193 {
 194     s->mon = mon;
 195     if (monitor_suspend(mon) == 0) {
 196         DPRINTF("suspending monitor\n");
 197     } else {
 198         monitor_printf(mon, "terminal does not allow synchronous "
 199                        "migration, continuing detached\n");
 200     }
 201 }
 202
 203 static int migrate_fd_cleanup(MigrationState *s)
 204 {
 205     int ret = 0;
 206
 207     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 208
 209     if (s->file) {
 210         DPRINTF("closing file\n");
 211         if (qemu_fclose(s->file) != 0) {
 212             ret = -1;
 213         }
 214         s->file = NULL;
 215     } else {
 216         if (s->mon) {
 217             monitor_resume(s->mon);
 218         }
 219     }
 220
 221     if (s->fd != -1) {
 222         close(s->fd);
 223         s->fd = -1;
 224     }
 225
 226     return ret;
 227 }
 228
 229 void migrate_fd_error(MigrationState *s)
 230 {
 231     DPRINTF("setting error state\n");
 232     s->state = MIG_STATE_ERROR;
 233     notifier_list_notify(&migration_state_notifiers, s);
 234     migrate_fd_cleanup(s);
 235 }
 236
 237 static void migrate_fd_completed(MigrationState *s)
 238 {
 239     DPRINTF("setting completed state\n");
 240     if (migrate_fd_cleanup(s) < 0) {
 241         s->state = MIG_STATE_ERROR;
 242     } else {
 243         s->state = MIG_STATE_COMPLETED;
 244         runstate_set(RUN_STATE_POSTMIGRATE);
 245     }
 246     notifier_list_notify(&migration_state_notifiers, s);
 247 }
 248
 249 static void migrate_fd_put_notify(void *opaque)
 250 {
 251     MigrationState *s = opaque;
 252
 253     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 254     qemu_file_put_notify(s->file);
 255     if (qemu_file_get_error(s->file)) {
 256         migrate_fd_error(s);
 257     }
 258 }
 259
 260 static ssize_t migrate_fd_put_buffer(void *opaque, const void *data,
 261                                      size_t size)
 262 {
 263     MigrationState *s = opaque;
 264     ssize_t ret;
 265
 266     if (s->state != MIG_STATE_ACTIVE) {
 267         return -EIO;
 268     }
 269
 270     do {
 271         ret = s->write(s, data, size);
 272     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 273
 274     if (ret == -1)
 275         ret = -(s->get_error(s));
 276
 277     if (ret == -EAGAIN) {
 278         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 279     }
 280
 281     return ret;
 282 }
 283
 284 static void migrate_fd_put_ready(void *opaque)
 285 {
 286     MigrationState *s = opaque;
 287     int ret;
 288
 289     if (s->state != MIG_STATE_ACTIVE) {
 290         DPRINTF("put_ready returning because of non-active state\n");
 291         return;
 292     }
 293
 294     DPRINTF("iterate\n");
 295     ret = qemu_savevm_state_iterate(s->mon, s->file);
 296     if (ret < 0) {
 297         migrate_fd_error(s);
 298     } else if (ret == 1) {
 299         int old_vm_running = runstate_is_running();
 300
 301         DPRINTF("done iterating\n");
 302         vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 303
 304         if (qemu_savevm_state_complete(s->mon, s->file) < 0) {
 305             migrate_fd_error(s);
 306         } else {
 307             migrate_fd_completed(s);
 308         }
 309         if (s->state != MIG_STATE_COMPLETED) {
 310             if (old_vm_running) {
 311                 vm_start();
 312             }
 313         }
 314     }
 315 }
 316
 317 static void migrate_fd_cancel(MigrationState *s)
 318 {
 319     if (s->state != MIG_STATE_ACTIVE)
 320         return;
 321
 322     DPRINTF("cancelling migration\n");
 323
 324     s->state = MIG_STATE_CANCELLED;
 325     notifier_list_notify(&migration_state_notifiers, s);
 326     qemu_savevm_state_cancel(s->mon, s->file);
 327
 328     migrate_fd_cleanup(s);
 329 }
 330
 331 static void migrate_fd_wait_for_unfreeze(void *opaque)
 332 {
 333     MigrationState *s = opaque;
 334     int ret;
 335
 336     DPRINTF("wait for unfreeze\n");
 337     if (s->state != MIG_STATE_ACTIVE)
 338         return;
 339
 340     do {
 341         fd_set wfds;
 342
 343         FD_ZERO(&wfds);
 344         FD_SET(s->fd, &wfds);
 345
 346         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 347     } while (ret == -1 && (s->get_error(s)) == EINTR);
 348
 349     if (ret == -1) {
 350         qemu_file_set_error(s->file, -s->get_error(s));
 351     }
 352 }
 353
 354 static int migrate_fd_close(void *opaque)
 355 {
 356     MigrationState *s = opaque;
 357
 358     if (s->mon) {
 359         monitor_resume(s->mon);
 360     }
 361     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 362     return s->close(s);
 363 }
 364
 365 void add_migration_state_change_notifier(Notifier *notify)
 366 {
 367     notifier_list_add(&migration_state_notifiers, notify);
 368 }
 369
 370 void remove_migration_state_change_notifier(Notifier *notify)
 371 {
 372     notifier_list_remove(&migration_state_notifiers, notify);
 373 }
 374
 375 bool migration_has_finished(MigrationState *s)
 376 {
 377     return s->state == MIG_STATE_COMPLETED;
 378 }
 379
 380 void migrate_fd_connect(MigrationState *s)
 381 {
 382     int ret;
 383
 384     s->state = MIG_STATE_ACTIVE;
 385     s->file = qemu_fopen_ops_buffered(s,
 386                                       s->bandwidth_limit,
 387                                       migrate_fd_put_buffer,
 388                                       migrate_fd_put_ready,
 389                                       migrate_fd_wait_for_unfreeze,
 390                                       migrate_fd_close);
 391
 392     DPRINTF("beginning savevm\n");
 393     ret = qemu_savevm_state_begin(s->mon, s->file, s->blk, s->shared);
 394     if (ret < 0) {
 395         DPRINTF("failed, %d\n", ret);
 396         migrate_fd_error(s);
 397         return;
 398     }
 399     migrate_fd_put_ready(s);
 400 }
 401
 402 static MigrationState *migrate_init(Monitor *mon, int detach, int blk, int inc)
 403 {
 404     MigrationState *s = migrate_get_current();
 405     int64_t bandwidth_limit = s->bandwidth_limit;
 406
 407     memset(s, 0, sizeof(*s));
 408     s->bandwidth_limit = bandwidth_limit;
 409     s->blk = blk;
 410     s->shared = inc;
 411     s->mon = NULL;
 412     s->bandwidth_limit = bandwidth_limit;
 413     s->state = MIG_STATE_SETUP;
 414
 415     if (!detach) {
 416         migrate_fd_monitor_suspend(s, mon);
 417     }
 418
 419     return s;
 420 }
 421
 422 int do_migrate(Monitor *mon, const QDict *qdict, QObject **ret_data)
 423 {
 424     MigrationState *s = migrate_get_current();
 425     const char *p;
 426     int detach = qdict_get_try_bool(qdict, "detach", 0);
 427     int blk = qdict_get_try_bool(qdict, "blk", 0);
 428     int inc = qdict_get_try_bool(qdict, "inc", 0);
 429     const char *uri = qdict_get_str(qdict, "uri");
 430     int ret;
 431
 432     if (s->state == MIG_STATE_ACTIVE) {
 433         monitor_printf(mon, "migration already in progress\n");
 434         return -1;
 435     }
 436
 437     if (qemu_savevm_state_blocked(mon)) {
 438         return -1;
 439     }
 440
 441     s = migrate_init(mon, detach, blk, inc);
 442
 443     if (strstart(uri, "tcp:", &p)) {
 444         ret = tcp_start_outgoing_migration(s, p);
 445 #if !defined(WIN32)
 446     } else if (strstart(uri, "exec:", &p)) {
 447         ret = exec_start_outgoing_migration(s, p);
 448     } else if (strstart(uri, "unix:", &p)) {
 449         ret = unix_start_outgoing_migration(s, p);
 450     } else if (strstart(uri, "fd:", &p)) {
 451         ret = fd_start_outgoing_migration(s, p);
 452 #endif
 453     } else {
 454         monitor_printf(mon, "unknown migration protocol: %s\n", uri);
 455         ret  = -EINVAL;
 456     }
 457
 458     if (ret < 0) {
 459         monitor_printf(mon, "migration failed: %s\n", strerror(-ret));
 460         return ret;
 461     }
 462
 463     notifier_list_notify(&migration_state_notifiers, s);
 464     return 0;
 465 }
 466
 467 int do_migrate_cancel(Monitor *mon, const QDict *qdict, QObject **ret_data)
 468 {
 469     migrate_fd_cancel(migrate_get_current());
 470     return 0;
 471 }
 472
 473 int do_migrate_set_speed(Monitor *mon, const QDict *qdict, QObject **ret_data)
 474 {
 475     int64_t d;
 476     MigrationState *s;
 477
 478     d = qdict_get_int(qdict, "value");
 479     if (d < 0) {
 480         d = 0;
 481     }
 482
 483     s = migrate_get_current();
 484     s->bandwidth_limit = d;
 485     qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
 486
 487     return 0;
 488 }
 489
 490 int do_migrate_set_downtime(Monitor *mon, const QDict *qdict,
 491                             QObject **ret_data)
 492 {
 493     double d;
 494
 495     d = qdict_get_double(qdict, "value") * 1e9;
 496     d = MAX(0, MIN(UINT64_MAX, d));
 497     max_downtime = (uint64_t)d;
 498
 499     return 0;
 500 }