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         info->ram->total_time = qemu_get_clock_ms(rt_clock)
 135             - s->total_time;
 136
 137         if (blk_mig_active()) {
 138             info->has_disk = true;
 139             info->disk = g_malloc0(sizeof(*info->disk));
 140             info->disk->transferred = blk_mig_bytes_transferred();
 141             info->disk->remaining = blk_mig_bytes_remaining();
 142             info->disk->total = blk_mig_bytes_total();
 143         }
 144         break;
 145     case MIG_STATE_COMPLETED:
 146         info->has_status = true;
 147         info->status = g_strdup("completed");
 148
 149         info->has_ram = true;
 150         info->ram = g_malloc0(sizeof(*info->ram));
 151         info->ram->transferred = ram_bytes_transferred();
 152         info->ram->remaining = 0;
 153         info->ram->total = ram_bytes_total();
 154         info->ram->total_time = s->total_time;
 155         break;
 156     case MIG_STATE_ERROR:
 157         info->has_status = true;
 158         info->status = g_strdup("failed");
 159         break;
 160     case MIG_STATE_CANCELLED:
 161         info->has_status = true;
 162         info->status = g_strdup("cancelled");
 163         break;
 164     }
 165
 166     return info;
 167 }
 168
 169 /* shared migration helpers */
 170
 171 static int migrate_fd_cleanup(MigrationState *s)
 172 {
 173     int ret = 0;
 174
 175     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 176
 177     if (s->file) {
 178         DPRINTF("closing file\n");
 179         ret = qemu_fclose(s->file);
 180         s->file = NULL;
 181     }
 182
 183     if (s->fd != -1) {
 184         close(s->fd);
 185         s->fd = -1;
 186     }
 187
 188     return ret;
 189 }
 190
 191 void migrate_fd_error(MigrationState *s)
 192 {
 193     DPRINTF("setting error state\n");
 194     s->state = MIG_STATE_ERROR;
 195     notifier_list_notify(&migration_state_notifiers, s);
 196     migrate_fd_cleanup(s);
 197 }
 198
 199 static void migrate_fd_completed(MigrationState *s)
 200 {
 201     DPRINTF("setting completed state\n");
 202     if (migrate_fd_cleanup(s) < 0) {
 203         s->state = MIG_STATE_ERROR;
 204     } else {
 205         s->state = MIG_STATE_COMPLETED;
 206         runstate_set(RUN_STATE_POSTMIGRATE);
 207     }
 208     notifier_list_notify(&migration_state_notifiers, s);
 209 }
 210
 211 static void migrate_fd_put_notify(void *opaque)
 212 {
 213     MigrationState *s = opaque;
 214
 215     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 216     qemu_file_put_notify(s->file);
 217     if (s->file && qemu_file_get_error(s->file)) {
 218         migrate_fd_error(s);
 219     }
 220 }
 221
 222 static ssize_t migrate_fd_put_buffer(void *opaque, const void *data,
 223                                      size_t size)
 224 {
 225     MigrationState *s = opaque;
 226     ssize_t ret;
 227
 228     if (s->state != MIG_STATE_ACTIVE) {
 229         return -EIO;
 230     }
 231
 232     do {
 233         ret = s->write(s, data, size);
 234     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 235
 236     if (ret == -1)
 237         ret = -(s->get_error(s));
 238
 239     if (ret == -EAGAIN) {
 240         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 241     }
 242
 243     return ret;
 244 }
 245
 246 static void migrate_fd_put_ready(void *opaque)
 247 {
 248     MigrationState *s = opaque;
 249     int ret;
 250
 251     if (s->state != MIG_STATE_ACTIVE) {
 252         DPRINTF("put_ready returning because of non-active state\n");
 253         return;
 254     }
 255
 256     DPRINTF("iterate\n");
 257     ret = qemu_savevm_state_iterate(s->file);
 258     if (ret < 0) {
 259         migrate_fd_error(s);
 260     } else if (ret == 1) {
 261         int old_vm_running = runstate_is_running();
 262
 263         DPRINTF("done iterating\n");
 264         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
 265         vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
 266
 267         if (qemu_savevm_state_complete(s->file) < 0) {
 268             migrate_fd_error(s);
 269         } else {
 270             migrate_fd_completed(s);
 271         }
 272         s->total_time = qemu_get_clock_ms(rt_clock) - s->total_time;
 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->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     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 323     return s->close(s);
 324 }
 325
 326 void add_migration_state_change_notifier(Notifier *notify)
 327 {
 328     notifier_list_add(&migration_state_notifiers, notify);
 329 }
 330
 331 void remove_migration_state_change_notifier(Notifier *notify)
 332 {
 333     notifier_remove(notify);
 334 }
 335
 336 bool migration_is_active(MigrationState *s)
 337 {
 338     return s->state == MIG_STATE_ACTIVE;
 339 }
 340
 341 bool migration_has_finished(MigrationState *s)
 342 {
 343     return s->state == MIG_STATE_COMPLETED;
 344 }
 345
 346 bool migration_has_failed(MigrationState *s)
 347 {
 348     return (s->state == MIG_STATE_CANCELLED ||
 349             s->state == MIG_STATE_ERROR);
 350 }
 351
 352 void migrate_fd_connect(MigrationState *s)
 353 {
 354     int ret;
 355
 356     s->state = MIG_STATE_ACTIVE;
 357     s->file = qemu_fopen_ops_buffered(s,
 358                                       s->bandwidth_limit,
 359                                       migrate_fd_put_buffer,
 360                                       migrate_fd_put_ready,
 361                                       migrate_fd_wait_for_unfreeze,
 362                                       migrate_fd_close);
 363
 364     DPRINTF("beginning savevm\n");
 365     ret = qemu_savevm_state_begin(s->file, &s->params);
 366     if (ret < 0) {
 367         DPRINTF("failed, %d\n", ret);
 368         migrate_fd_error(s);
 369         return;
 370     }
 371     migrate_fd_put_ready(s);
 372 }
 373
 374 static MigrationState *migrate_init(const MigrationParams *params)
 375 {
 376     MigrationState *s = migrate_get_current();
 377     int64_t bandwidth_limit = s->bandwidth_limit;
 378
 379     memset(s, 0, sizeof(*s));
 380     s->bandwidth_limit = bandwidth_limit;
 381     s->params = *params;
 382
 383     s->bandwidth_limit = bandwidth_limit;
 384     s->state = MIG_STATE_SETUP;
 385     s->total_time = qemu_get_clock_ms(rt_clock);
 386
 387     return s;
 388 }
 389
 390 static GSList *migration_blockers;
 391
 392 void migrate_add_blocker(Error *reason)
 393 {
 394     migration_blockers = g_slist_prepend(migration_blockers, reason);
 395 }
 396
 397 void migrate_del_blocker(Error *reason)
 398 {
 399     migration_blockers = g_slist_remove(migration_blockers, reason);
 400 }
 401
 402 void qmp_migrate(const char *uri, bool has_blk, bool blk,
 403                  bool has_inc, bool inc, bool has_detach, bool detach,
 404                  Error **errp)
 405 {
 406     MigrationState *s = migrate_get_current();
 407     MigrationParams params;
 408     const char *p;
 409     int ret;
 410
 411     params.blk = blk;
 412     params.shared = inc;
 413
 414     if (s->state == MIG_STATE_ACTIVE) {
 415         error_set(errp, QERR_MIGRATION_ACTIVE);
 416         return;
 417     }
 418
 419     if (qemu_savevm_state_blocked(errp)) {
 420         return;
 421     }
 422
 423     if (migration_blockers) {
 424         *errp = error_copy(migration_blockers->data);
 425         return;
 426     }
 427
 428     s = migrate_init(&params);
 429
 430     if (strstart(uri, "tcp:", &p)) {
 431         ret = tcp_start_outgoing_migration(s, p, errp);
 432 #if !defined(WIN32)
 433     } else if (strstart(uri, "exec:", &p)) {
 434         ret = exec_start_outgoing_migration(s, p);
 435     } else if (strstart(uri, "unix:", &p)) {
 436         ret = unix_start_outgoing_migration(s, p);
 437     } else if (strstart(uri, "fd:", &p)) {
 438         ret = fd_start_outgoing_migration(s, p);
 439 #endif
 440     } else {
 441         error_set(errp, QERR_INVALID_PARAMETER_VALUE, "uri", "a valid migration protocol");
 442         return;
 443     }
 444
 445     if (ret < 0) {
 446         if (!error_is_set(errp)) {
 447             DPRINTF("migration failed: %s\n", strerror(-ret));
 448             /* FIXME: we should return meaningful errors */
 449             error_set(errp, QERR_UNDEFINED_ERROR);
 450         }
 451         return;
 452     }
 453
 454     notifier_list_notify(&migration_state_notifiers, s);
 455 }
 456
 457 void qmp_migrate_cancel(Error **errp)
 458 {
 459     migrate_fd_cancel(migrate_get_current());
 460 }
 461
 462 void qmp_migrate_set_speed(int64_t value, Error **errp)
 463 {
 464     MigrationState *s;
 465
 466     if (value < 0) {
 467         value = 0;
 468     }
 469
 470     s = migrate_get_current();
 471     s->bandwidth_limit = value;
 472     qemu_file_set_rate_limit(s->file, s->bandwidth_limit);
 473 }
 474
 475 void qmp_migrate_set_downtime(double value, Error **errp)
 476 {
 477     value *= 1e9;
 478     value = MAX(0, MIN(UINT64_MAX, value));
 479     max_downtime = (uint64_t)value;
 480 }