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     if (monitor_suspend(mon) == 0) {
 159         DPRINTF("suspending monitor\n");
 160     } else {
 161         monitor_printf(mon, "terminal does not allow synchronous "
 162                        "migration, continuing detached\n");
 163     }
 164 }
 165
 166 static int migrate_fd_cleanup(MigrationState *s)
 167 {
 168     int ret = 0;
 169
 170     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 171
 172     if (s->file) {
 173         DPRINTF("closing file\n");
 174         if (qemu_fclose(s->file) != 0) {
 175             ret = -1;
 176         }
 177         s->file = NULL;
 178     } else {
 179         if (s->mon) {
 180             monitor_resume(s->mon);
 181         }
 182     }
 183
 184     if (s->fd != -1) {
 185         close(s->fd);
 186         s->fd = -1;
 187     }
 188
 189     return ret;
 190 }
 191
 192 void migrate_fd_error(MigrationState *s)
 193 {
 194     DPRINTF("setting error state\n");
 195     s->state = MIG_STATE_ERROR;
 196     notifier_list_notify(&migration_state_notifiers, s);
 197     migrate_fd_cleanup(s);
 198 }
 199
 200 static void migrate_fd_completed(MigrationState *s)
 201 {
 202     DPRINTF("setting completed state\n");
 203     if (migrate_fd_cleanup(s) < 0) {
 204         s->state = MIG_STATE_ERROR;
 205     } else {
 206         s->state = MIG_STATE_COMPLETED;
 207         runstate_set(RUN_STATE_POSTMIGRATE);
 208     }
 209     notifier_list_notify(&migration_state_notifiers, s);
 210 }
 211
 212 static void migrate_fd_put_notify(void *opaque)
 213 {
 214     MigrationState *s = opaque;
 215
 216     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 217     qemu_file_put_notify(s->file);
 218     if (s->file && qemu_file_get_error(s->file)) {
 219         migrate_fd_error(s);
 220     }
 221 }
 222
 223 static ssize_t migrate_fd_put_buffer(void *opaque, const void *data,
 224                                      size_t size)
 225 {
 226     MigrationState *s = opaque;
 227     ssize_t ret;
 228
 229     if (s->state != MIG_STATE_ACTIVE) {
 230         return -EIO;
 231     }
 232
 233     do {
 234         ret = s->write(s, data, size);
 235     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 236
 237     if (ret == -1)
 238         ret = -(s->get_error(s));
 239
 240     if (ret == -EAGAIN) {
 241         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 242     }
 243
 244     return ret;
 245 }
 246
 247 static void migrate_fd_put_ready(void *opaque)
 248 {
 249     MigrationState *s = opaque;
 250     int ret;
 251
 252     if (s->state != MIG_STATE_ACTIVE) {
 253         DPRINTF("put_ready returning because of non-active state\n");
 254         return;
 255     }
 256
 257     DPRINTF("iterate\n");
 258     ret = qemu_savevm_state_iterate(s->mon, s->file);
 259     if (ret < 0) {
 260         migrate_fd_error(s);
 261     } else if (ret == 1) {
 262         int old_vm_running = runstate_is_running();
 263
 264         DPRINTF("done iterating\n");
 265         vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 266
 267         if (qemu_savevm_state_complete(s->mon, s->file) < 0) {
 268             migrate_fd_error(s);
 269         } else {
 270             migrate_fd_completed(s);
 271         }
 272         if (s->state != MIG_STATE_COMPLETED) {
 273             if (old_vm_running) {
 274                 vm_start();
 275             }
 276         }
 277     }
 278 }
 279
 280 static void migrate_fd_cancel(MigrationState *s)
 281 {
 282     if (s->state != MIG_STATE_ACTIVE)
 283         return;
 284
 285     DPRINTF("cancelling migration\n");
 286
 287     s->state = MIG_STATE_CANCELLED;
 288     notifier_list_notify(&migration_state_notifiers, s);
 289     qemu_savevm_state_cancel(s->mon, s->file);
 290
 291     migrate_fd_cleanup(s);
 292 }
 293
 294 static void migrate_fd_wait_for_unfreeze(void *opaque)
 295 {
 296     MigrationState *s = opaque;
 297     int ret;
 298
 299     DPRINTF("wait for unfreeze\n");
 300     if (s->state != MIG_STATE_ACTIVE)
 301         return;
 302
 303     do {
 304         fd_set wfds;
 305
 306         FD_ZERO(&wfds);
 307         FD_SET(s->fd, &wfds);
 308
 309         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 310     } while (ret == -1 && (s->get_error(s)) == EINTR);
 311
 312     if (ret == -1) {
 313         qemu_file_set_error(s->file, -s->get_error(s));
 314     }
 315 }
 316
 317 static int migrate_fd_close(void *opaque)
 318 {
 319     MigrationState *s = opaque;
 320
 321     if (s->mon) {
 322         monitor_resume(s->mon);
 323     }
 324     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 325     return s->close(s);
 326 }
 327
 328 void add_migration_state_change_notifier(Notifier *notify)
 329 {
 330     notifier_list_add(&migration_state_notifiers, notify);
 331 }
 332
 333 void remove_migration_state_change_notifier(Notifier *notify)
 334 {
 335     notifier_list_remove(&migration_state_notifiers, notify);
 336 }
 337
 338 bool migration_is_active(MigrationState *s)
 339 {
 340     return s->state == MIG_STATE_ACTIVE;
 341 }
 342
 343 bool migration_has_finished(MigrationState *s)
 344 {
 345     return s->state == MIG_STATE_COMPLETED;
 346 }
 347
 348 bool migration_has_failed(MigrationState *s)
 349 {
 350     return (s->state == MIG_STATE_CANCELLED ||
 351             s->state == MIG_STATE_ERROR);
 352 }
 353
 354 void migrate_fd_connect(MigrationState *s)
 355 {
 356     int ret;
 357
 358     s->state = MIG_STATE_ACTIVE;
 359     s->file = qemu_fopen_ops_buffered(s,
 360                                       s->bandwidth_limit,
 361                                       migrate_fd_put_buffer,
 362                                       migrate_fd_put_ready,
 363                                       migrate_fd_wait_for_unfreeze,
 364                                       migrate_fd_close);
 365
 366     DPRINTF("beginning savevm\n");
 367     ret = qemu_savevm_state_begin(s->mon, s->file, s->blk, s->shared);
 368     if (ret < 0) {
 369         DPRINTF("failed, %d\n", ret);
 370         migrate_fd_error(s);
 371         return;
 372     }
 373     migrate_fd_put_ready(s);
 374 }
 375
 376 static MigrationState *migrate_init(Monitor *mon, int detach, int blk, int inc)
 377 {
 378     MigrationState *s = migrate_get_current();
 379     int64_t bandwidth_limit = s->bandwidth_limit;
 380
 381     memset(s, 0, sizeof(*s));
 382     s->bandwidth_limit = bandwidth_limit;
 383     s->blk = blk;
 384     s->shared = inc;
 385
 386     /* s->mon is used for two things:
 387        - pass fd in fd migration
 388        - suspend/resume monitor for not detached migration
 389     */
 390     s->mon = mon;
 391     s->bandwidth_limit = bandwidth_limit;
 392     s->state = MIG_STATE_SETUP;
 393
 394     if (!detach) {
 395         migrate_fd_monitor_suspend(s, mon);
 396     }
 397
 398     return s;
 399 }
 400
 401 int do_migrate(Monitor *mon, const QDict *qdict, QObject **ret_data)
 402 {
 403     MigrationState *s = migrate_get_current();
 404     const char *p;
 405     int detach = qdict_get_try_bool(qdict, "detach", 0);
 406     int blk = qdict_get_try_bool(qdict, "blk", 0);
 407     int inc = qdict_get_try_bool(qdict, "inc", 0);
 408     const char *uri = qdict_get_str(qdict, "uri");
 409     int ret;
 410
 411     if (s->state == MIG_STATE_ACTIVE) {
 412         monitor_printf(mon, "migration already in progress\n");
 413         return -1;
 414     }
 415
 416     if (qemu_savevm_state_blocked(mon)) {
 417         return -1;
 418     }
 419
 420     s = migrate_init(mon, detach, blk, inc);
 421
 422     if (strstart(uri, "tcp:", &p)) {
 423         ret = tcp_start_outgoing_migration(s, p);
 424 #if !defined(WIN32)
 425     } else if (strstart(uri, "exec:", &p)) {
 426         ret = exec_start_outgoing_migration(s, p);
 427     } else if (strstart(uri, "unix:", &p)) {
 428         ret = unix_start_outgoing_migration(s, p);
 429     } else if (strstart(uri, "fd:", &p)) {
 430         ret = fd_start_outgoing_migration(s, p);
 431 #endif
 432     } else {
 433         monitor_printf(mon, "unknown migration protocol: %s\n", uri);
 434         ret  = -EINVAL;
 435     }
 436
 437     if (ret < 0) {
 438         monitor_printf(mon, "migration failed: %s\n", strerror(-ret));
 439         return ret;
 440     }
 441
 442     if (detach) {
 443         s->mon = NULL;
 444     }
 445
 446     notifier_list_notify(&migration_state_notifiers, s);
 447     return 0;
 448 }
 449
 450 int do_migrate_cancel(Monitor *mon, const QDict *qdict, QObject **ret_data)
 451 {
 452     migrate_fd_cancel(migrate_get_current());
 453     return 0;
 454 }
 455
 456 int do_migrate_set_speed(Monitor *mon, const QDict *qdict, QObject **ret_data)
 457 {
 458     int64_t d;
 459     MigrationState *s;
 460
 461     d = qdict_get_int(qdict, "value");
 462     if (d < 0) {
 463         d = 0;
 464     }
 465
 466     s = migrate_get_current();
 467     s->bandwidth_limit = d;
 468     qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
 469
 470     return 0;
 471 }
 472
 473 int do_migrate_set_downtime(Monitor *mon, const QDict *qdict,
 474                             QObject **ret_data)
 475 {
 476     double d;
 477
 478     d = qdict_get_double(qdict, "value") * 1e9;
 479     d = MAX(0, MIN(UINT64_MAX, d));
 480     max_downtime = (uint64_t)d;
 481
 482     return 0;
 483 }