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         if (qemu_fclose(s->file) != 0) {
 178             ret = -1;
 179         }
 180         s->file = NULL;
 181     } else {
 182         if (s->mon) {
 183             monitor_resume(s->mon);
 184         }
 185     }
 186
 187     if (s->fd != -1) {
 188         close(s->fd);
 189         s->fd = -1;
 190     }
 191
 192     return ret;
 193 }
 194
 195 void migrate_fd_error(MigrationState *s)
 196 {
 197     DPRINTF("setting error state\n");
 198     s->state = MIG_STATE_ERROR;
 199     notifier_list_notify(&migration_state_notifiers, s);
 200     migrate_fd_cleanup(s);
 201 }
 202
 203 static void migrate_fd_completed(MigrationState *s)
 204 {
 205     DPRINTF("setting completed state\n");
 206     if (migrate_fd_cleanup(s) < 0) {
 207         s->state = MIG_STATE_ERROR;
 208     } else {
 209         s->state = MIG_STATE_COMPLETED;
 210         runstate_set(RUN_STATE_POSTMIGRATE);
 211     }
 212     notifier_list_notify(&migration_state_notifiers, s);
 213 }
 214
 215 static void migrate_fd_put_notify(void *opaque)
 216 {
 217     MigrationState *s = opaque;
 218
 219     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 220     qemu_file_put_notify(s->file);
 221     if (s->file && qemu_file_get_error(s->file)) {
 222         migrate_fd_error(s);
 223     }
 224 }
 225
 226 static ssize_t migrate_fd_put_buffer(void *opaque, const void *data,
 227                                      size_t size)
 228 {
 229     MigrationState *s = opaque;
 230     ssize_t ret;
 231
 232     if (s->state != MIG_STATE_ACTIVE) {
 233         return -EIO;
 234     }
 235
 236     do {
 237         ret = s->write(s, data, size);
 238     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 239
 240     if (ret == -1)
 241         ret = -(s->get_error(s));
 242
 243     if (ret == -EAGAIN) {
 244         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 245     }
 246
 247     return ret;
 248 }
 249
 250 static void migrate_fd_put_ready(void *opaque)
 251 {
 252     MigrationState *s = opaque;
 253     int ret;
 254
 255     if (s->state != MIG_STATE_ACTIVE) {
 256         DPRINTF("put_ready returning because of non-active state\n");
 257         return;
 258     }
 259
 260     DPRINTF("iterate\n");
 261     ret = qemu_savevm_state_iterate(s->mon, s->file);
 262     if (ret < 0) {
 263         migrate_fd_error(s);
 264     } else if (ret == 1) {
 265         int old_vm_running = runstate_is_running();
 266
 267         DPRINTF("done iterating\n");
 268         vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 269
 270         if (qemu_savevm_state_complete(s->mon, s->file) < 0) {
 271             migrate_fd_error(s);
 272         } else {
 273             migrate_fd_completed(s);
 274         }
 275         if (s->state != MIG_STATE_COMPLETED) {
 276             if (old_vm_running) {
 277                 vm_start();
 278             }
 279         }
 280     }
 281 }
 282
 283 static void migrate_fd_cancel(MigrationState *s)
 284 {
 285     if (s->state != MIG_STATE_ACTIVE)
 286         return;
 287
 288     DPRINTF("cancelling migration\n");
 289
 290     s->state = MIG_STATE_CANCELLED;
 291     notifier_list_notify(&migration_state_notifiers, s);
 292     qemu_savevm_state_cancel(s->mon, s->file);
 293
 294     migrate_fd_cleanup(s);
 295 }
 296
 297 static void migrate_fd_wait_for_unfreeze(void *opaque)
 298 {
 299     MigrationState *s = opaque;
 300     int ret;
 301
 302     DPRINTF("wait for unfreeze\n");
 303     if (s->state != MIG_STATE_ACTIVE)
 304         return;
 305
 306     do {
 307         fd_set wfds;
 308
 309         FD_ZERO(&wfds);
 310         FD_SET(s->fd, &wfds);
 311
 312         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 313     } while (ret == -1 && (s->get_error(s)) == EINTR);
 314
 315     if (ret == -1) {
 316         qemu_file_set_error(s->file, -s->get_error(s));
 317     }
 318 }
 319
 320 static int migrate_fd_close(void *opaque)
 321 {
 322     MigrationState *s = opaque;
 323
 324     if (s->mon) {
 325         monitor_resume(s->mon);
 326     }
 327     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 328     return s->close(s);
 329 }
 330
 331 void add_migration_state_change_notifier(Notifier *notify)
 332 {
 333     notifier_list_add(&migration_state_notifiers, notify);
 334 }
 335
 336 void remove_migration_state_change_notifier(Notifier *notify)
 337 {
 338     notifier_list_remove(&migration_state_notifiers, notify);
 339 }
 340
 341 bool migration_is_active(MigrationState *s)
 342 {
 343     return s->state == MIG_STATE_ACTIVE;
 344 }
 345
 346 bool migration_has_finished(MigrationState *s)
 347 {
 348     return s->state == MIG_STATE_COMPLETED;
 349 }
 350
 351 bool migration_has_failed(MigrationState *s)
 352 {
 353     return (s->state == MIG_STATE_CANCELLED ||
 354             s->state == MIG_STATE_ERROR);
 355 }
 356
 357 void migrate_fd_connect(MigrationState *s)
 358 {
 359     int ret;
 360
 361     s->state = MIG_STATE_ACTIVE;
 362     s->file = qemu_fopen_ops_buffered(s,
 363                                       s->bandwidth_limit,
 364                                       migrate_fd_put_buffer,
 365                                       migrate_fd_put_ready,
 366                                       migrate_fd_wait_for_unfreeze,
 367                                       migrate_fd_close);
 368
 369     DPRINTF("beginning savevm\n");
 370     ret = qemu_savevm_state_begin(s->mon, s->file, s->blk, s->shared);
 371     if (ret < 0) {
 372         DPRINTF("failed, %d\n", ret);
 373         migrate_fd_error(s);
 374         return;
 375     }
 376     migrate_fd_put_ready(s);
 377 }
 378
 379 static MigrationState *migrate_init(Monitor *mon, int detach, int blk, int inc)
 380 {
 381     MigrationState *s = migrate_get_current();
 382     int64_t bandwidth_limit = s->bandwidth_limit;
 383
 384     memset(s, 0, sizeof(*s));
 385     s->bandwidth_limit = bandwidth_limit;
 386     s->blk = blk;
 387     s->shared = inc;
 388
 389     /* s->mon is used for two things:
 390        - pass fd in fd migration
 391        - suspend/resume monitor for not detached migration
 392     */
 393     s->mon = mon;
 394     s->bandwidth_limit = bandwidth_limit;
 395     s->state = MIG_STATE_SETUP;
 396
 397     if (!detach) {
 398         migrate_fd_monitor_suspend(s, mon);
 399     }
 400
 401     return s;
 402 }
 403
 404 static GSList *migration_blockers;
 405
 406 void migrate_add_blocker(Error *reason)
 407 {
 408     migration_blockers = g_slist_prepend(migration_blockers, reason);
 409 }
 410
 411 void migrate_del_blocker(Error *reason)
 412 {
 413     migration_blockers = g_slist_remove(migration_blockers, reason);
 414 }
 415
 416 int do_migrate(Monitor *mon, const QDict *qdict, QObject **ret_data)
 417 {
 418     MigrationState *s = migrate_get_current();
 419     const char *p;
 420     int detach = qdict_get_try_bool(qdict, "detach", 0);
 421     int blk = qdict_get_try_bool(qdict, "blk", 0);
 422     int inc = qdict_get_try_bool(qdict, "inc", 0);
 423     const char *uri = qdict_get_str(qdict, "uri");
 424     int ret;
 425
 426     if (s->state == MIG_STATE_ACTIVE) {
 427         monitor_printf(mon, "migration already in progress\n");
 428         return -1;
 429     }
 430
 431     if (qemu_savevm_state_blocked(mon)) {
 432         return -1;
 433     }
 434
 435     if (migration_blockers) {
 436         Error *err = migration_blockers->data;
 437         qerror_report_err(err);
 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     if (detach) {
 464         s->mon = NULL;
 465     }
 466
 467     notifier_list_notify(&migration_state_notifiers, s);
 468     return 0;
 469 }
 470
 471 int do_migrate_cancel(Monitor *mon, const QDict *qdict, QObject **ret_data)
 472 {
 473     migrate_fd_cancel(migrate_get_current());
 474     return 0;
 475 }
 476
 477 int do_migrate_set_speed(Monitor *mon, const QDict *qdict, QObject **ret_data)
 478 {
 479     int64_t d;
 480     MigrationState *s;
 481
 482     d = qdict_get_int(qdict, "value");
 483     if (d < 0) {
 484         d = 0;
 485     }
 486
 487     s = migrate_get_current();
 488     s->bandwidth_limit = d;
 489     qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
 490
 491     return 0;
 492 }
 493
 494 int do_migrate_set_downtime(Monitor *mon, const QDict *qdict,
 495                             QObject **ret_data)
 496 {
 497     double d;
 498
 499     d = qdict_get_double(qdict, "value") * 1e9;
 500     d = MAX(0, MIN(UINT64_MAX, d));
 501     max_downtime = (uint64_t)d;
 502
 503     return 0;
 504 }