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     MigrationState *s;
 146
 147     d = qdict_get_int(qdict, "value");
 148     if (d < 0) {
 149         d = 0;
 150     }
 151     max_throttle = d;
 152
 153     s = 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
 232     if (current_migration) {
 233         MigrationState *s = current_migration;
 234
 235         switch (s->get_status(current_migration)) {
 236         case MIG_STATE_ACTIVE:
 237             qdict = qdict_new();
 238             qdict_put(qdict, "status", qstring_from_str("active"));
 239
 240             migrate_put_status(qdict, "ram", ram_bytes_transferred(),
 241                                ram_bytes_remaining(), ram_bytes_total());
 242
 243             if (blk_mig_active()) {
 244                 migrate_put_status(qdict, "disk", blk_mig_bytes_transferred(),
 245                                    blk_mig_bytes_remaining(),
 246                                    blk_mig_bytes_total());
 247             }
 248
 249             *ret_data = QOBJECT(qdict);
 250             break;
 251         case MIG_STATE_COMPLETED:
 252             *ret_data = qobject_from_jsonf("{ 'status': 'completed' }");
 253             break;
 254         case MIG_STATE_ERROR:
 255             *ret_data = qobject_from_jsonf("{ 'status': 'failed' }");
 256             break;
 257         case MIG_STATE_CANCELLED:
 258             *ret_data = qobject_from_jsonf("{ 'status': 'cancelled' }");
 259             break;
 260         }
 261     }
 262 }
 263
 264 /* shared migration helpers */
 265
 266 static void migrate_fd_monitor_suspend(MigrationState *s, Monitor *mon)
 267 {
 268     s->mon = mon;
 269     if (monitor_suspend(mon) == 0) {
 270         DPRINTF("suspending monitor\n");
 271     } else {
 272         monitor_printf(mon, "terminal does not allow synchronous "
 273                        "migration, continuing detached\n");
 274     }
 275 }
 276
 277 static int migrate_fd_cleanup(MigrationState *s)
 278 {
 279     int ret = 0;
 280
 281     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 282
 283     if (s->file) {
 284         DPRINTF("closing file\n");
 285         if (qemu_fclose(s->file) != 0) {
 286             ret = -1;
 287         }
 288         s->file = NULL;
 289     } else {
 290         if (s->mon) {
 291             monitor_resume(s->mon);
 292         }
 293     }
 294
 295     if (s->fd != -1) {
 296         close(s->fd);
 297         s->fd = -1;
 298     }
 299
 300     return ret;
 301 }
 302
 303 void migrate_fd_error(MigrationState *s)
 304 {
 305     DPRINTF("setting error state\n");
 306     s->state = MIG_STATE_ERROR;
 307     notifier_list_notify(&migration_state_notifiers, NULL);
 308     migrate_fd_cleanup(s);
 309 }
 310
 311 static void migrate_fd_put_notify(void *opaque)
 312 {
 313     MigrationState *s = opaque;
 314
 315     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 316     qemu_file_put_notify(s->file);
 317     if (qemu_file_get_error(s->file)) {
 318         migrate_fd_error(s);
 319     }
 320 }
 321
 322 static ssize_t migrate_fd_put_buffer(void *opaque, const void *data,
 323                                      size_t size)
 324 {
 325     MigrationState *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 static void migrate_fd_put_ready(void *opaque)
 347 {
 348     MigrationState *s = opaque;
 349     int ret;
 350
 351     if (s->state != MIG_STATE_ACTIVE) {
 352         DPRINTF("put_ready returning because of non-active state\n");
 353         return;
 354     }
 355
 356     DPRINTF("iterate\n");
 357     ret = qemu_savevm_state_iterate(s->mon, s->file);
 358     if (ret < 0) {
 359         migrate_fd_error(s);
 360     } else if (ret == 1) {
 361         int old_vm_running = runstate_is_running();
 362
 363         DPRINTF("done iterating\n");
 364         vm_stop(RUN_STATE_FINISH_MIGRATE);
 365
 366         if ((qemu_savevm_state_complete(s->mon, s->file)) < 0) {
 367             if (old_vm_running) {
 368                 vm_start();
 369             }
 370             s->state = MIG_STATE_ERROR;
 371         }
 372         if (migrate_fd_cleanup(s) < 0) {
 373             if (old_vm_running) {
 374                 vm_start();
 375             }
 376             s->state = MIG_STATE_ERROR;
 377         }
 378         if (s->state == MIG_STATE_ACTIVE) {
 379             s->state = MIG_STATE_COMPLETED;
 380             runstate_set(RUN_STATE_POSTMIGRATE);
 381         }
 382         notifier_list_notify(&migration_state_notifiers, NULL);
 383     }
 384 }
 385
 386 static int migrate_fd_get_status(MigrationState *s)
 387 {
 388     return s->state;
 389 }
 390
 391 static void migrate_fd_cancel(MigrationState *s)
 392 {
 393     if (s->state != MIG_STATE_ACTIVE)
 394         return;
 395
 396     DPRINTF("cancelling migration\n");
 397
 398     s->state = MIG_STATE_CANCELLED;
 399     notifier_list_notify(&migration_state_notifiers, NULL);
 400     qemu_savevm_state_cancel(s->mon, s->file);
 401
 402     migrate_fd_cleanup(s);
 403 }
 404
 405 static void migrate_fd_release(MigrationState *s)
 406 {
 407
 408     DPRINTF("releasing state\n");
 409
 410     if (s->state == MIG_STATE_ACTIVE) {
 411         s->state = MIG_STATE_CANCELLED;
 412         notifier_list_notify(&migration_state_notifiers, NULL);
 413         migrate_fd_cleanup(s);
 414     }
 415     g_free(s);
 416 }
 417
 418 static void migrate_fd_wait_for_unfreeze(void *opaque)
 419 {
 420     MigrationState *s = opaque;
 421     int ret;
 422
 423     DPRINTF("wait for unfreeze\n");
 424     if (s->state != MIG_STATE_ACTIVE)
 425         return;
 426
 427     do {
 428         fd_set wfds;
 429
 430         FD_ZERO(&wfds);
 431         FD_SET(s->fd, &wfds);
 432
 433         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 434     } while (ret == -1 && (s->get_error(s)) == EINTR);
 435
 436     if (ret == -1) {
 437         qemu_file_set_error(s->file, -s->get_error(s));
 438     }
 439 }
 440
 441 static int migrate_fd_close(void *opaque)
 442 {
 443     MigrationState *s = opaque;
 444
 445     if (s->mon) {
 446         monitor_resume(s->mon);
 447     }
 448     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 449     return s->close(s);
 450 }
 451
 452 void add_migration_state_change_notifier(Notifier *notify)
 453 {
 454     notifier_list_add(&migration_state_notifiers, notify);
 455 }
 456
 457 void remove_migration_state_change_notifier(Notifier *notify)
 458 {
 459     notifier_list_remove(&migration_state_notifiers, notify);
 460 }
 461
 462 int get_migration_state(void)
 463 {
 464     if (current_migration) {
 465         return migrate_fd_get_status(current_migration);
 466     } else {
 467         return MIG_STATE_ERROR;
 468     }
 469 }
 470
 471 void migrate_fd_connect(MigrationState *s)
 472 {
 473     int ret;
 474
 475     s->file = qemu_fopen_ops_buffered(s,
 476                                       s->bandwidth_limit,
 477                                       migrate_fd_put_buffer,
 478                                       migrate_fd_put_ready,
 479                                       migrate_fd_wait_for_unfreeze,
 480                                       migrate_fd_close);
 481
 482     DPRINTF("beginning savevm\n");
 483     ret = qemu_savevm_state_begin(s->mon, s->file, s->blk, s->shared);
 484     if (ret < 0) {
 485         DPRINTF("failed, %d\n", ret);
 486         migrate_fd_error(s);
 487         return;
 488     }
 489     migrate_fd_put_ready(s);
 490 }
 491
 492 MigrationState *migrate_new(Monitor *mon, int64_t bandwidth_limit,
 493                                      int detach, int blk, int inc)
 494 {
 495     MigrationState *s = g_malloc0(sizeof(*s));
 496
 497     s->cancel = migrate_fd_cancel;
 498     s->get_status = migrate_fd_get_status;
 499     s->release = migrate_fd_release;
 500     s->blk = blk;
 501     s->shared = inc;
 502     s->mon = NULL;
 503     s->bandwidth_limit = bandwidth_limit;
 504     s->state = MIG_STATE_ACTIVE;
 505
 506     if (!detach) {
 507         migrate_fd_monitor_suspend(s, mon);
 508     }
 509
 510     return s;
 511 }