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 "qmp-commands.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     /* Make sure all file formats flush their mutable metadata */
  93     bdrv_invalidate_cache_all();
  94
  95     if (autostart) {
  96         vm_start();
  97     } else {
  98         runstate_set(RUN_STATE_PRELAUNCH);
  99     }
 100 }
 101
 102 /* amount of nanoseconds we are willing to wait for migration to be down.
 103  * the choice of nanoseconds is because it is the maximum resolution that
 104  * get_clock() can achieve. It is an internal measure. All user-visible
 105  * units must be in seconds */
 106 static uint64_t max_downtime = 30000000;
 107
 108 uint64_t migrate_max_downtime(void)
 109 {
 110     return max_downtime;
 111 }
 112
 113 MigrationInfo *qmp_query_migrate(Error **errp)
 114 {
 115     MigrationInfo *info = g_malloc0(sizeof(*info));
 116     MigrationState *s = migrate_get_current();
 117
 118     switch (s->state) {
 119     case MIG_STATE_SETUP:
 120         /* no migration has happened ever */
 121         break;
 122     case MIG_STATE_ACTIVE:
 123         info->has_status = true;
 124         info->status = g_strdup("active");
 125
 126         info->has_ram = true;
 127         info->ram = g_malloc0(sizeof(*info->ram));
 128         info->ram->transferred = ram_bytes_transferred();
 129         info->ram->remaining = ram_bytes_remaining();
 130         info->ram->total = ram_bytes_total();
 131
 132         if (blk_mig_active()) {
 133             info->has_disk = true;
 134             info->disk = g_malloc0(sizeof(*info->disk));
 135             info->disk->transferred = blk_mig_bytes_transferred();
 136             info->disk->remaining = blk_mig_bytes_remaining();
 137             info->disk->total = blk_mig_bytes_total();
 138         }
 139         break;
 140     case MIG_STATE_COMPLETED:
 141         info->has_status = true;
 142         info->status = g_strdup("completed");
 143         break;
 144     case MIG_STATE_ERROR:
 145         info->has_status = true;
 146         info->status = g_strdup("failed");
 147         break;
 148     case MIG_STATE_CANCELLED:
 149         info->has_status = true;
 150         info->status = g_strdup("cancelled");
 151         break;
 152     }
 153
 154     return info;
 155 }
 156
 157 /* shared migration helpers */
 158
 159 static void migrate_fd_monitor_suspend(MigrationState *s, Monitor *mon)
 160 {
 161     if (monitor_suspend(mon) == 0) {
 162         DPRINTF("suspending monitor\n");
 163     } else {
 164         monitor_printf(mon, "terminal does not allow synchronous "
 165                        "migration, continuing detached\n");
 166     }
 167 }
 168
 169 static int migrate_fd_cleanup(MigrationState *s)
 170 {
 171     int ret = 0;
 172
 173     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 174
 175     if (s->file) {
 176         DPRINTF("closing file\n");
 177         ret = qemu_fclose(s->file);
 178         s->file = NULL;
 179     } else {
 180         if (s->mon) {
 181             monitor_resume(s->mon);
 182         }
 183     }
 184
 185     if (s->fd != -1) {
 186         close(s->fd);
 187         s->fd = -1;
 188     }
 189
 190     return ret;
 191 }
 192
 193 void migrate_fd_error(MigrationState *s)
 194 {
 195     DPRINTF("setting error state\n");
 196     s->state = MIG_STATE_ERROR;
 197     notifier_list_notify(&migration_state_notifiers, s);
 198     migrate_fd_cleanup(s);
 199 }
 200
 201 static void migrate_fd_completed(MigrationState *s)
 202 {
 203     DPRINTF("setting completed state\n");
 204     if (migrate_fd_cleanup(s) < 0) {
 205         s->state = MIG_STATE_ERROR;
 206     } else {
 207         s->state = MIG_STATE_COMPLETED;
 208         runstate_set(RUN_STATE_POSTMIGRATE);
 209     }
 210     notifier_list_notify(&migration_state_notifiers, s);
 211 }
 212
 213 static void migrate_fd_put_notify(void *opaque)
 214 {
 215     MigrationState *s = opaque;
 216
 217     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 218     qemu_file_put_notify(s->file);
 219     if (s->file && qemu_file_get_error(s->file)) {
 220         migrate_fd_error(s);
 221     }
 222 }
 223
 224 static ssize_t migrate_fd_put_buffer(void *opaque, const void *data,
 225                                      size_t size)
 226 {
 227     MigrationState *s = opaque;
 228     ssize_t ret;
 229
 230     if (s->state != MIG_STATE_ACTIVE) {
 231         return -EIO;
 232     }
 233
 234     do {
 235         ret = s->write(s, data, size);
 236     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 237
 238     if (ret == -1)
 239         ret = -(s->get_error(s));
 240
 241     if (ret == -EAGAIN) {
 242         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 243     }
 244
 245     return ret;
 246 }
 247
 248 static void migrate_fd_put_ready(void *opaque)
 249 {
 250     MigrationState *s = opaque;
 251     int ret;
 252
 253     if (s->state != MIG_STATE_ACTIVE) {
 254         DPRINTF("put_ready returning because of non-active state\n");
 255         return;
 256     }
 257
 258     DPRINTF("iterate\n");
 259     ret = qemu_savevm_state_iterate(s->mon, s->file);
 260     if (ret < 0) {
 261         migrate_fd_error(s);
 262     } else if (ret == 1) {
 263         int old_vm_running = runstate_is_running();
 264
 265         DPRINTF("done iterating\n");
 266         vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 267
 268         if (qemu_savevm_state_complete(s->mon, s->file) < 0) {
 269             migrate_fd_error(s);
 270         } else {
 271             migrate_fd_completed(s);
 272         }
 273         if (s->state != MIG_STATE_COMPLETED) {
 274             if (old_vm_running) {
 275                 vm_start();
 276             }
 277         }
 278     }
 279 }
 280
 281 static void migrate_fd_cancel(MigrationState *s)
 282 {
 283     if (s->state != MIG_STATE_ACTIVE)
 284         return;
 285
 286     DPRINTF("cancelling migration\n");
 287
 288     s->state = MIG_STATE_CANCELLED;
 289     notifier_list_notify(&migration_state_notifiers, s);
 290     qemu_savevm_state_cancel(s->mon, s->file);
 291
 292     migrate_fd_cleanup(s);
 293 }
 294
 295 static void migrate_fd_wait_for_unfreeze(void *opaque)
 296 {
 297     MigrationState *s = opaque;
 298     int ret;
 299
 300     DPRINTF("wait for unfreeze\n");
 301     if (s->state != MIG_STATE_ACTIVE)
 302         return;
 303
 304     do {
 305         fd_set wfds;
 306
 307         FD_ZERO(&wfds);
 308         FD_SET(s->fd, &wfds);
 309
 310         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 311     } while (ret == -1 && (s->get_error(s)) == EINTR);
 312
 313     if (ret == -1) {
 314         qemu_file_set_error(s->file, -s->get_error(s));
 315     }
 316 }
 317
 318 static int migrate_fd_close(void *opaque)
 319 {
 320     MigrationState *s = opaque;
 321
 322     if (s->mon) {
 323         monitor_resume(s->mon);
 324     }
 325     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 326     return s->close(s);
 327 }
 328
 329 void add_migration_state_change_notifier(Notifier *notify)
 330 {
 331     notifier_list_add(&migration_state_notifiers, notify);
 332 }
 333
 334 void remove_migration_state_change_notifier(Notifier *notify)
 335 {
 336     notifier_list_remove(&migration_state_notifiers, notify);
 337 }
 338
 339 bool migration_is_active(MigrationState *s)
 340 {
 341     return s->state == MIG_STATE_ACTIVE;
 342 }
 343
 344 bool migration_has_finished(MigrationState *s)
 345 {
 346     return s->state == MIG_STATE_COMPLETED;
 347 }
 348
 349 bool migration_has_failed(MigrationState *s)
 350 {
 351     return (s->state == MIG_STATE_CANCELLED ||
 352             s->state == MIG_STATE_ERROR);
 353 }
 354
 355 void migrate_fd_connect(MigrationState *s)
 356 {
 357     int ret;
 358
 359     s->state = MIG_STATE_ACTIVE;
 360     s->file = qemu_fopen_ops_buffered(s,
 361                                       s->bandwidth_limit,
 362                                       migrate_fd_put_buffer,
 363                                       migrate_fd_put_ready,
 364                                       migrate_fd_wait_for_unfreeze,
 365                                       migrate_fd_close);
 366
 367     DPRINTF("beginning savevm\n");
 368     ret = qemu_savevm_state_begin(s->mon, s->file, s->blk, s->shared);
 369     if (ret < 0) {
 370         DPRINTF("failed, %d\n", ret);
 371         migrate_fd_error(s);
 372         return;
 373     }
 374     migrate_fd_put_ready(s);
 375 }
 376
 377 static MigrationState *migrate_init(Monitor *mon, int detach, int blk, int inc)
 378 {
 379     MigrationState *s = migrate_get_current();
 380     int64_t bandwidth_limit = s->bandwidth_limit;
 381
 382     memset(s, 0, sizeof(*s));
 383     s->bandwidth_limit = bandwidth_limit;
 384     s->blk = blk;
 385     s->shared = inc;
 386
 387     /* s->mon is used for two things:
 388        - pass fd in fd migration
 389        - suspend/resume monitor for not detached migration
 390     */
 391     s->mon = mon;
 392     s->bandwidth_limit = bandwidth_limit;
 393     s->state = MIG_STATE_SETUP;
 394
 395     if (!detach) {
 396         migrate_fd_monitor_suspend(s, mon);
 397     }
 398
 399     return s;
 400 }
 401
 402 static GSList *migration_blockers;
 403
 404 void migrate_add_blocker(Error *reason)
 405 {
 406     migration_blockers = g_slist_prepend(migration_blockers, reason);
 407 }
 408
 409 void migrate_del_blocker(Error *reason)
 410 {
 411     migration_blockers = g_slist_remove(migration_blockers, reason);
 412 }
 413
 414 int do_migrate(Monitor *mon, const QDict *qdict, QObject **ret_data)
 415 {
 416     MigrationState *s = migrate_get_current();
 417     const char *p;
 418     int detach = qdict_get_try_bool(qdict, "detach", 0);
 419     int blk = qdict_get_try_bool(qdict, "blk", 0);
 420     int inc = qdict_get_try_bool(qdict, "inc", 0);
 421     const char *uri = qdict_get_str(qdict, "uri");
 422     int ret;
 423
 424     if (s->state == MIG_STATE_ACTIVE) {
 425         monitor_printf(mon, "migration already in progress\n");
 426         return -1;
 427     }
 428
 429     if (qemu_savevm_state_blocked(mon)) {
 430         return -1;
 431     }
 432
 433     if (migration_blockers) {
 434         Error *err = migration_blockers->data;
 435         qerror_report_err(err);
 436         return -1;
 437     }
 438
 439     s = migrate_init(mon, detach, blk, inc);
 440
 441     if (strstart(uri, "tcp:", &p)) {
 442         ret = tcp_start_outgoing_migration(s, p);
 443 #if !defined(WIN32)
 444     } else if (strstart(uri, "exec:", &p)) {
 445         ret = exec_start_outgoing_migration(s, p);
 446     } else if (strstart(uri, "unix:", &p)) {
 447         ret = unix_start_outgoing_migration(s, p);
 448     } else if (strstart(uri, "fd:", &p)) {
 449         ret = fd_start_outgoing_migration(s, p);
 450 #endif
 451     } else {
 452         monitor_printf(mon, "unknown migration protocol: %s\n", uri);
 453         ret  = -EINVAL;
 454     }
 455
 456     if (ret < 0) {
 457         monitor_printf(mon, "migration failed: %s\n", strerror(-ret));
 458         return ret;
 459     }
 460
 461     if (detach) {
 462         s->mon = NULL;
 463     }
 464
 465     notifier_list_notify(&migration_state_notifiers, s);
 466     return 0;
 467 }
 468
 469 void qmp_migrate_cancel(Error **errp)
 470 {
 471     migrate_fd_cancel(migrate_get_current());
 472 }
 473
 474 void qmp_migrate_set_speed(int64_t value, Error **errp)
 475 {
 476     MigrationState *s;
 477
 478     if (value < 0) {
 479         value = 0;
 480     }
 481
 482     s = migrate_get_current();
 483     s->bandwidth_limit = value;
 484     qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
 485 }
 486
 487 void qmp_migrate_set_downtime(double value, Error **errp)
 488 {
 489     value *= 1e9;
 490     value = MAX(0, MIN(UINT64_MAX, value));
 491     max_downtime = (uint64_t)value;
 492 }