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  * Contributions after 2012-01-13 are licensed under the terms of the
  13  * GNU GPL, version 2 or (at your option) any later version.
  14  */
  15
  16 #include "qemu-common.h"
  17 #include "migration.h"
  18 #include "monitor.h"
  19 #include "buffered_file.h"
  20 #include "sysemu.h"
  21 #include "block.h"
  22 #include "qemu_socket.h"
  23 #include "block-migration.h"
  24 #include "qmp-commands.h"
  25
  26 //#define DEBUG_MIGRATION
  27
  28 #ifdef DEBUG_MIGRATION
  29 #define DPRINTF(fmt, ...) \
  30     do { printf("migration: " fmt, ## __VA_ARGS__); } while (0)
  31 #else
  32 #define DPRINTF(fmt, ...) \
  33     do { } while (0)
  34 #endif
  35
  36 enum {
  37     MIG_STATE_ERROR,
  38     MIG_STATE_SETUP,
  39     MIG_STATE_CANCELLED,
  40     MIG_STATE_ACTIVE,
  41     MIG_STATE_COMPLETED,
  42 };
  43
  44 #define MAX_THROTTLE  (32 << 20)      /* Migration speed throttling */
  45
  46 static NotifierList migration_state_notifiers =
  47     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
  48
  49 /* When we add fault tolerance, we could have several
  50    migrations at once.  For now we don't need to add
  51    dynamic creation of migration */
  52
  53 static MigrationState *migrate_get_current(void)
  54 {
  55     static MigrationState current_migration = {
  56         .state = MIG_STATE_SETUP,
  57         .bandwidth_limit = MAX_THROTTLE,
  58     };
  59
  60     return &current_migration;
  61 }
  62
  63 int qemu_start_incoming_migration(const char *uri, Error **errp)
  64 {
  65     const char *p;
  66     int ret;
  67
  68     if (strstart(uri, "tcp:", &p))
  69         ret = tcp_start_incoming_migration(p, errp);
  70 #if !defined(WIN32)
  71     else if (strstart(uri, "exec:", &p))
  72         ret =  exec_start_incoming_migration(p);
  73     else if (strstart(uri, "unix:", &p))
  74         ret = unix_start_incoming_migration(p);
  75     else if (strstart(uri, "fd:", &p))
  76         ret = fd_start_incoming_migration(p);
  77 #endif
  78     else {
  79         fprintf(stderr, "unknown migration protocol: %s\n", uri);
  80         ret = -EPROTONOSUPPORT;
  81     }
  82     return ret;
  83 }
  84
  85 void process_incoming_migration(QEMUFile *f)
  86 {
  87     if (qemu_loadvm_state(f) < 0) {
  88         fprintf(stderr, "load of migration failed\n");
  89         exit(0);
  90     }
  91     qemu_announce_self();
  92     DPRINTF("successfully loaded vm state\n");
  93
  94     bdrv_clear_incoming_migration_all();
  95     /* Make sure all file formats flush their mutable metadata */
  96     bdrv_invalidate_cache_all();
  97
  98     if (autostart) {
  99         vm_start();
 100     } else {
 101         runstate_set(RUN_STATE_PRELAUNCH);
 102     }
 103 }
 104
 105 /* amount of nanoseconds we are willing to wait for migration to be down.
 106  * the choice of nanoseconds is because it is the maximum resolution that
 107  * get_clock() can achieve. It is an internal measure. All user-visible
 108  * units must be in seconds */
 109 static uint64_t max_downtime = 30000000;
 110
 111 uint64_t migrate_max_downtime(void)
 112 {
 113     return max_downtime;
 114 }
 115
 116 MigrationInfo *qmp_query_migrate(Error **errp)
 117 {
 118     MigrationInfo *info = g_malloc0(sizeof(*info));
 119     MigrationState *s = migrate_get_current();
 120
 121     switch (s->state) {
 122     case MIG_STATE_SETUP:
 123         /* no migration has happened ever */
 124         break;
 125     case MIG_STATE_ACTIVE:
 126         info->has_status = true;
 127         info->status = g_strdup("active");
 128
 129         info->has_ram = true;
 130         info->ram = g_malloc0(sizeof(*info->ram));
 131         info->ram->transferred = ram_bytes_transferred();
 132         info->ram->remaining = ram_bytes_remaining();
 133         info->ram->total = ram_bytes_total();
 134
 135         if (blk_mig_active()) {
 136             info->has_disk = true;
 137             info->disk = g_malloc0(sizeof(*info->disk));
 138             info->disk->transferred = blk_mig_bytes_transferred();
 139             info->disk->remaining = blk_mig_bytes_remaining();
 140             info->disk->total = blk_mig_bytes_total();
 141         }
 142         break;
 143     case MIG_STATE_COMPLETED:
 144         info->has_status = true;
 145         info->status = g_strdup("completed");
 146         break;
 147     case MIG_STATE_ERROR:
 148         info->has_status = true;
 149         info->status = g_strdup("failed");
 150         break;
 151     case MIG_STATE_CANCELLED:
 152         info->has_status = true;
 153         info->status = g_strdup("cancelled");
 154         break;
 155     }
 156
 157     return info;
 158 }
 159
 160 /* shared migration helpers */
 161
 162 static int migrate_fd_cleanup(MigrationState *s)
 163 {
 164     int ret = 0;
 165
 166     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 167
 168     if (s->file) {
 169         DPRINTF("closing file\n");
 170         ret = qemu_fclose(s->file);
 171         s->file = NULL;
 172     }
 173
 174     if (s->fd != -1) {
 175         close(s->fd);
 176         s->fd = -1;
 177     }
 178
 179     return ret;
 180 }
 181
 182 void migrate_fd_error(MigrationState *s)
 183 {
 184     DPRINTF("setting error state\n");
 185     s->state = MIG_STATE_ERROR;
 186     notifier_list_notify(&migration_state_notifiers, s);
 187     migrate_fd_cleanup(s);
 188 }
 189
 190 static void migrate_fd_completed(MigrationState *s)
 191 {
 192     DPRINTF("setting completed state\n");
 193     if (migrate_fd_cleanup(s) < 0) {
 194         s->state = MIG_STATE_ERROR;
 195     } else {
 196         s->state = MIG_STATE_COMPLETED;
 197         runstate_set(RUN_STATE_POSTMIGRATE);
 198     }
 199     notifier_list_notify(&migration_state_notifiers, s);
 200 }
 201
 202 static void migrate_fd_put_notify(void *opaque)
 203 {
 204     MigrationState *s = opaque;
 205
 206     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 207     qemu_file_put_notify(s->file);
 208     if (s->file && qemu_file_get_error(s->file)) {
 209         migrate_fd_error(s);
 210     }
 211 }
 212
 213 static ssize_t migrate_fd_put_buffer(void *opaque, const void *data,
 214                                      size_t size)
 215 {
 216     MigrationState *s = opaque;
 217     ssize_t ret;
 218
 219     if (s->state != MIG_STATE_ACTIVE) {
 220         return -EIO;
 221     }
 222
 223     do {
 224         ret = s->write(s, data, size);
 225     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 226
 227     if (ret == -1)
 228         ret = -(s->get_error(s));
 229
 230     if (ret == -EAGAIN) {
 231         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 232     }
 233
 234     return ret;
 235 }
 236
 237 static void migrate_fd_put_ready(void *opaque)
 238 {
 239     MigrationState *s = opaque;
 240     int ret;
 241
 242     if (s->state != MIG_STATE_ACTIVE) {
 243         DPRINTF("put_ready returning because of non-active state\n");
 244         return;
 245     }
 246
 247     DPRINTF("iterate\n");
 248     ret = qemu_savevm_state_iterate(s->file);
 249     if (ret < 0) {
 250         migrate_fd_error(s);
 251     } else if (ret == 1) {
 252         int old_vm_running = runstate_is_running();
 253
 254         DPRINTF("done iterating\n");
 255         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 256         vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 257
 258         if (qemu_savevm_state_complete(s->file) < 0) {
 259             migrate_fd_error(s);
 260         } else {
 261             migrate_fd_completed(s);
 262         }
 263         if (s->state != MIG_STATE_COMPLETED) {
 264             if (old_vm_running) {
 265                 vm_start();
 266             }
 267         }
 268     }
 269 }
 270
 271 static void migrate_fd_cancel(MigrationState *s)
 272 {
 273     if (s->state != MIG_STATE_ACTIVE)
 274         return;
 275
 276     DPRINTF("cancelling migration\n");
 277
 278     s->state = MIG_STATE_CANCELLED;
 279     notifier_list_notify(&migration_state_notifiers, s);
 280     qemu_savevm_state_cancel(s->file);
 281
 282     migrate_fd_cleanup(s);
 283 }
 284
 285 static void migrate_fd_wait_for_unfreeze(void *opaque)
 286 {
 287     MigrationState *s = opaque;
 288     int ret;
 289
 290     DPRINTF("wait for unfreeze\n");
 291     if (s->state != MIG_STATE_ACTIVE)
 292         return;
 293
 294     do {
 295         fd_set wfds;
 296
 297         FD_ZERO(&wfds);
 298         FD_SET(s->fd, &wfds);
 299
 300         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 301     } while (ret == -1 && (s->get_error(s)) == EINTR);
 302
 303     if (ret == -1) {
 304         qemu_file_set_error(s->file, -s->get_error(s));
 305     }
 306 }
 307
 308 static int migrate_fd_close(void *opaque)
 309 {
 310     MigrationState *s = opaque;
 311
 312     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 313     return s->close(s);
 314 }
 315
 316 void add_migration_state_change_notifier(Notifier *notify)
 317 {
 318     notifier_list_add(&migration_state_notifiers, notify);
 319 }
 320
 321 void remove_migration_state_change_notifier(Notifier *notify)
 322 {
 323     notifier_remove(notify);
 324 }
 325
 326 bool migration_is_active(MigrationState *s)
 327 {
 328     return s->state == MIG_STATE_ACTIVE;
 329 }
 330
 331 bool migration_has_finished(MigrationState *s)
 332 {
 333     return s->state == MIG_STATE_COMPLETED;
 334 }
 335
 336 bool migration_has_failed(MigrationState *s)
 337 {
 338     return (s->state == MIG_STATE_CANCELLED ||
 339             s->state == MIG_STATE_ERROR);
 340 }
 341
 342 void migrate_fd_connect(MigrationState *s)
 343 {
 344     int ret;
 345
 346     s->state = MIG_STATE_ACTIVE;
 347     s->file = qemu_fopen_ops_buffered(s,
 348                                       s->bandwidth_limit,
 349                                       migrate_fd_put_buffer,
 350                                       migrate_fd_put_ready,
 351                                       migrate_fd_wait_for_unfreeze,
 352                                       migrate_fd_close);
 353
 354     DPRINTF("beginning savevm\n");
 355     ret = qemu_savevm_state_begin(s->file, s->blk, s->shared);
 356     if (ret < 0) {
 357         DPRINTF("failed, %d\n", ret);
 358         migrate_fd_error(s);
 359         return;
 360     }
 361     migrate_fd_put_ready(s);
 362 }
 363
 364 static MigrationState *migrate_init(int blk, int inc)
 365 {
 366     MigrationState *s = migrate_get_current();
 367     int64_t bandwidth_limit = s->bandwidth_limit;
 368
 369     memset(s, 0, sizeof(*s));
 370     s->bandwidth_limit = bandwidth_limit;
 371     s->blk = blk;
 372     s->shared = inc;
 373
 374     s->bandwidth_limit = bandwidth_limit;
 375     s->state = MIG_STATE_SETUP;
 376
 377     return s;
 378 }
 379
 380 static GSList *migration_blockers;
 381
 382 void migrate_add_blocker(Error *reason)
 383 {
 384     migration_blockers = g_slist_prepend(migration_blockers, reason);
 385 }
 386
 387 void migrate_del_blocker(Error *reason)
 388 {
 389     migration_blockers = g_slist_remove(migration_blockers, reason);
 390 }
 391
 392 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 393                  bool has_inc, bool inc, bool has_detach, bool detach,
 394                  Error **errp)
 395 {
 396     MigrationState *s = migrate_get_current();
 397     const char *p;
 398     int ret;
 399
 400     if (s->state == MIG_STATE_ACTIVE) {
 401         error_set(errp, QERR_MIGRATION_ACTIVE);
 402         return;
 403     }
 404
 405     if (qemu_savevm_state_blocked(errp)) {
 406         return;
 407     }
 408
 409     if (migration_blockers) {
 410         *errp = error_copy(migration_blockers->data);
 411         return;
 412     }
 413
 414     s = migrate_init(blk, inc);
 415
 416     if (strstart(uri, "tcp:", &p)) {
 417         ret = tcp_start_outgoing_migration(s, p, errp);
 418 #if !defined(WIN32)
 419     } else if (strstart(uri, "exec:", &p)) {
 420         ret = exec_start_outgoing_migration(s, p);
 421     } else if (strstart(uri, "unix:", &p)) {
 422         ret = unix_start_outgoing_migration(s, p);
 423     } else if (strstart(uri, "fd:", &p)) {
 424         ret = fd_start_outgoing_migration(s, p);
 425 #endif
 426     } else {
 427         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "uri", "a valid migration protocol");
 428         return;
 429     }
 430
 431     if (ret < 0) {
 432         if (!error_is_set(errp)) {
 433             DPRINTF("migration failed: %s\n", strerror(-ret));
 434             /* FIXME: we should return meaningful errors */
 435             error_set(errp, QERR_UNDEFINED_ERROR);
 436         }
 437         return;
 438     }
 439
 440     notifier_list_notify(&migration_state_notifiers, s);
 441 }
 442
 443 void qmp_migrate_cancel(Error **errp)
 444 {
 445     migrate_fd_cancel(migrate_get_current());
 446 }
 447
 448 void qmp_migrate_set_speed(int64_t value, Error **errp)
 449 {
 450     MigrationState *s;
 451
 452     if (value < 0) {
 453         value = 0;
 454     }
 455
 456     s = migrate_get_current();
 457     s->bandwidth_limit = value;
 458     qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
 459 }
 460
 461 void qmp_migrate_set_downtime(double value, Error **errp)
 462 {
 463     value *= 1e9;
 464     value = MAX(0, MIN(UINT64_MAX, value));
 465     max_downtime = (uint64_t)value;
 466 }