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