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     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 MigrationInfo *qmp_query_migrate(Error **errp)
 111 {
 112     MigrationInfo *info = g_malloc0(sizeof(*info));
 113     MigrationState *s = migrate_get_current();
 114
 115     switch (s->state) {
 116     case MIG_STATE_SETUP:
 117         /* no migration has happened ever */
 118         break;
 119     case MIG_STATE_ACTIVE:
 120         info->has_status = true;
 121         info->status = g_strdup("active");
 122
 123         info->has_ram = true;
 124         info->ram = g_malloc0(sizeof(*info->ram));
 125         info->ram->transferred = ram_bytes_transferred();
 126         info->ram->remaining = ram_bytes_remaining();
 127         info->ram->total = ram_bytes_total();
 128
 129         if (blk_mig_active()) {
 130             info->has_disk = true;
 131             info->disk = g_malloc0(sizeof(*info->disk));
 132             info->disk->transferred = blk_mig_bytes_transferred();
 133             info->disk->remaining = blk_mig_bytes_remaining();
 134             info->disk->total = blk_mig_bytes_total();
 135         }
 136         break;
 137     case MIG_STATE_COMPLETED:
 138         info->has_status = true;
 139         info->status = g_strdup("completed");
 140         break;
 141     case MIG_STATE_ERROR:
 142         info->has_status = true;
 143         info->status = g_strdup("failed");
 144         break;
 145     case MIG_STATE_CANCELLED:
 146         info->has_status = true;
 147         info->status = g_strdup("cancelled");
 148         break;
 149     }
 150
 151     return info;
 152 }
 153
 154 /* shared migration helpers */
 155
 156 static void migrate_fd_monitor_suspend(MigrationState *s, Monitor *mon)
 157 {
 158     s->mon = mon;
 159     if (monitor_suspend(mon) == 0) {
 160         DPRINTF("suspending monitor\n");
 161     } else {
 162         monitor_printf(mon, "terminal does not allow synchronous "
 163                        "migration, continuing detached\n");
 164     }
 165 }
 166
 167 static int migrate_fd_cleanup(MigrationState *s)
 168 {
 169     int ret = 0;
 170
 171     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 172
 173     if (s->file) {
 174         DPRINTF("closing file\n");
 175         if (qemu_fclose(s->file) != 0) {
 176             ret = -1;
 177         }
 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     s->mon = NULL;
 387     s->bandwidth_limit = bandwidth_limit;
 388     s->state = MIG_STATE_SETUP;
 389
 390     if (!detach) {
 391         migrate_fd_monitor_suspend(s, mon);
 392     }
 393
 394     return s;
 395 }
 396
 397 int do_migrate(Monitor *mon, const QDict *qdict, QObject **ret_data)
 398 {
 399     MigrationState *s = migrate_get_current();
 400     const char *p;
 401     int detach = qdict_get_try_bool(qdict, "detach", 0);
 402     int blk = qdict_get_try_bool(qdict, "blk", 0);
 403     int inc = qdict_get_try_bool(qdict, "inc", 0);
 404     const char *uri = qdict_get_str(qdict, "uri");
 405     int ret;
 406
 407     if (s->state == MIG_STATE_ACTIVE) {
 408         monitor_printf(mon, "migration already in progress\n");
 409         return -1;
 410     }
 411
 412     if (qemu_savevm_state_blocked(mon)) {
 413         return -1;
 414     }
 415
 416     s = migrate_init(mon, detach, blk, inc);
 417
 418     if (strstart(uri, "tcp:", &p)) {
 419         ret = tcp_start_outgoing_migration(s, p);
 420 #if !defined(WIN32)
 421     } else if (strstart(uri, "exec:", &p)) {
 422         ret = exec_start_outgoing_migration(s, p);
 423     } else if (strstart(uri, "unix:", &p)) {
 424         ret = unix_start_outgoing_migration(s, p);
 425     } else if (strstart(uri, "fd:", &p)) {
 426         ret = fd_start_outgoing_migration(s, p);
 427 #endif
 428     } else {
 429         monitor_printf(mon, "unknown migration protocol: %s\n", uri);
 430         ret  = -EINVAL;
 431     }
 432
 433     if (ret < 0) {
 434         monitor_printf(mon, "migration failed: %s\n", strerror(-ret));
 435         return ret;
 436     }
 437
 438     notifier_list_notify(&migration_state_notifiers, s);
 439     return 0;
 440 }
 441
 442 int do_migrate_cancel(Monitor *mon, const QDict *qdict, QObject **ret_data)
 443 {
 444     migrate_fd_cancel(migrate_get_current());
 445     return 0;
 446 }
 447
 448 int do_migrate_set_speed(Monitor *mon, const QDict *qdict, QObject **ret_data)
 449 {
 450     int64_t d;
 451     MigrationState *s;
 452
 453     d = qdict_get_int(qdict, "value");
 454     if (d < 0) {
 455         d = 0;
 456     }
 457
 458     s = migrate_get_current();
 459     s->bandwidth_limit = d;
 460     qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
 461
 462     return 0;
 463 }
 464
 465 int do_migrate_set_downtime(Monitor *mon, const QDict *qdict,
 466                             QObject **ret_data)
 467 {
 468     double d;
 469
 470     d = qdict_get_double(qdict, "value") * 1e9;
 471     d = MAX(0, MIN(UINT64_MAX, d));
 472     max_downtime = (uint64_t)d;
 473
 474     return 0;
 475 }