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 "qemu-objects.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 /* Migration speed throttling */
  35 static uint32_t max_throttle = (32 << 20);
  36
  37 static MigrationState *current_migration;
  38
  39 int qemu_start_incoming_migration(const char *uri)
  40 {
  41     const char *p;
  42     int ret;
  43
  44     if (strstart(uri, "tcp:", &p))
  45         ret = tcp_start_incoming_migration(p);
  46 #if !defined(WIN32)
  47     else if (strstart(uri, "exec:", &p))
  48         ret =  exec_start_incoming_migration(p);
  49     else if (strstart(uri, "unix:", &p))
  50         ret = unix_start_incoming_migration(p);
  51     else if (strstart(uri, "fd:", &p))
  52         ret = fd_start_incoming_migration(p);
  53 #endif
  54     else {
  55         fprintf(stderr, "unknown migration protocol: %s\n", uri);
  56         ret = -EPROTONOSUPPORT;
  57     }
  58     return ret;
  59 }
  60
  61 void process_incoming_migration(QEMUFile *f)
  62 {
  63     if (qemu_loadvm_state(f) < 0) {
  64         fprintf(stderr, "load of migration failed\n");
  65         exit(0);
  66     }
  67     qemu_announce_self();
  68     DPRINTF("successfully loaded vm state\n");
  69
  70     incoming_expected = false;
  71
  72     if (autostart)
  73         vm_start();
  74 }
  75
  76 int do_migrate(Monitor *mon, const QDict *qdict, QObject **ret_data)
  77 {
  78     MigrationState *s = NULL;
  79     const char *p;
  80     int detach = qdict_get_try_bool(qdict, "detach", 0);
  81     int blk = qdict_get_try_bool(qdict, "blk", 0);
  82     int inc = qdict_get_try_bool(qdict, "inc", 0);
  83     const char *uri = qdict_get_str(qdict, "uri");
  84
  85     if (current_migration &&
  86         current_migration->get_status(current_migration) == MIG_STATE_ACTIVE) {
  87         monitor_printf(mon, "migration already in progress\n");
  88         return -1;
  89     }
  90
  91     if (strstart(uri, "tcp:", &p)) {
  92         s = tcp_start_outgoing_migration(mon, p, max_throttle, detach,
  93                                          blk, inc);
  94 #if !defined(WIN32)
  95     } else if (strstart(uri, "exec:", &p)) {
  96         s = exec_start_outgoing_migration(mon, p, max_throttle, detach,
  97                                           blk, inc);
  98     } else if (strstart(uri, "unix:", &p)) {
  99         s = unix_start_outgoing_migration(mon, p, max_throttle, detach,
 100                                           blk, inc);
 101     } else if (strstart(uri, "fd:", &p)) {
 102         s = fd_start_outgoing_migration(mon, p, max_throttle, detach,
 103                                         blk, inc);
 104 #endif
 105     } else {
 106         monitor_printf(mon, "unknown migration protocol: %s\n", uri);
 107         return -1;
 108     }
 109
 110     if (s == NULL) {
 111         monitor_printf(mon, "migration failed\n");
 112         return -1;
 113     }
 114
 115     if (current_migration) {
 116         current_migration->release(current_migration);
 117     }
 118
 119     current_migration = s;
 120     return 0;
 121 }
 122
 123 int do_migrate_cancel(Monitor *mon, const QDict *qdict, QObject **ret_data)
 124 {
 125     MigrationState *s = current_migration;
 126
 127     if (s)
 128         s->cancel(s);
 129
 130     return 0;
 131 }
 132
 133 int do_migrate_set_speed(Monitor *mon, const QDict *qdict, QObject **ret_data)
 134 {
 135     double d;
 136     FdMigrationState *s;
 137
 138     d = qdict_get_double(qdict, "value");
 139     d = MAX(0, MIN(UINT32_MAX, d));
 140     max_throttle = d;
 141
 142     s = migrate_to_fms(current_migration);
 143     if (s && s->file) {
 144         qemu_file_set_rate_limit(s->file, max_throttle);
 145     }
 146
 147     return 0;
 148 }
 149
 150 /* amount of nanoseconds we are willing to wait for migration to be down.
 151  * the choice of nanoseconds is because it is the maximum resolution that
 152  * get_clock() can achieve. It is an internal measure. All user-visible
 153  * units must be in seconds */
 154 static uint64_t max_downtime = 30000000;
 155
 156 uint64_t migrate_max_downtime(void)
 157 {
 158     return max_downtime;
 159 }
 160
 161 int do_migrate_set_downtime(Monitor *mon, const QDict *qdict,
 162                             QObject **ret_data)
 163 {
 164     double d;
 165
 166     d = qdict_get_double(qdict, "value") * 1e9;
 167     d = MAX(0, MIN(UINT64_MAX, d));
 168     max_downtime = (uint64_t)d;
 169
 170     return 0;
 171 }
 172
 173 static void migrate_print_status(Monitor *mon, const char *name,
 174                                  const QDict *status_dict)
 175 {
 176     QDict *qdict;
 177
 178     qdict = qobject_to_qdict(qdict_get(status_dict, name));
 179
 180     monitor_printf(mon, "transferred %s: %" PRIu64 " kbytes\n", name,
 181                         qdict_get_int(qdict, "transferred") >> 10);
 182     monitor_printf(mon, "remaining %s: %" PRIu64 " kbytes\n", name,
 183                         qdict_get_int(qdict, "remaining") >> 10);
 184     monitor_printf(mon, "total %s: %" PRIu64 " kbytes\n", name,
 185                         qdict_get_int(qdict, "total") >> 10);
 186 }
 187
 188 void do_info_migrate_print(Monitor *mon, const QObject *data)
 189 {
 190     QDict *qdict;
 191
 192     qdict = qobject_to_qdict(data);
 193
 194     monitor_printf(mon, "Migration status: %s\n",
 195                    qdict_get_str(qdict, "status"));
 196
 197     if (qdict_haskey(qdict, "ram")) {
 198         migrate_print_status(mon, "ram", qdict);
 199     }
 200
 201     if (qdict_haskey(qdict, "disk")) {
 202         migrate_print_status(mon, "disk", qdict);
 203     }
 204 }
 205
 206 static void migrate_put_status(QDict *qdict, const char *name,
 207                                uint64_t trans, uint64_t rem, uint64_t total)
 208 {
 209     QObject *obj;
 210
 211     obj = qobject_from_jsonf("{ 'transferred': %" PRId64 ", "
 212                                "'remaining': %" PRId64 ", "
 213                                "'total': %" PRId64 " }", trans, rem, total);
 214     qdict_put_obj(qdict, name, obj);
 215 }
 216
 217 void do_info_migrate(Monitor *mon, QObject **ret_data)
 218 {
 219     QDict *qdict;
 220     MigrationState *s = current_migration;
 221
 222     if (s) {
 223         switch (s->get_status(s)) {
 224         case MIG_STATE_ACTIVE:
 225             qdict = qdict_new();
 226             qdict_put(qdict, "status", qstring_from_str("active"));
 227
 228             migrate_put_status(qdict, "ram", ram_bytes_transferred(),
 229                                ram_bytes_remaining(), ram_bytes_total());
 230
 231             if (blk_mig_active()) {
 232                 migrate_put_status(qdict, "disk", blk_mig_bytes_transferred(),
 233                                    blk_mig_bytes_remaining(),
 234                                    blk_mig_bytes_total());
 235             }
 236
 237             *ret_data = QOBJECT(qdict);
 238             break;
 239         case MIG_STATE_COMPLETED:
 240             *ret_data = qobject_from_jsonf("{ 'status': 'completed' }");
 241             break;
 242         case MIG_STATE_ERROR:
 243             *ret_data = qobject_from_jsonf("{ 'status': 'failed' }");
 244             break;
 245         case MIG_STATE_CANCELLED:
 246             *ret_data = qobject_from_jsonf("{ 'status': 'cancelled' }");
 247             break;
 248         }
 249     }
 250 }
 251
 252 /* shared migration helpers */
 253
 254 void migrate_fd_monitor_suspend(FdMigrationState *s, Monitor *mon)
 255 {
 256     s->mon = mon;
 257     if (monitor_suspend(mon) == 0) {
 258         DPRINTF("suspending monitor\n");
 259     } else {
 260         monitor_printf(mon, "terminal does not allow synchronous "
 261                        "migration, continuing detached\n");
 262     }
 263 }
 264
 265 void migrate_fd_error(FdMigrationState *s)
 266 {
 267     DPRINTF("setting error state\n");
 268     s->state = MIG_STATE_ERROR;
 269     migrate_fd_cleanup(s);
 270 }
 271
 272 int migrate_fd_cleanup(FdMigrationState *s)
 273 {
 274     int ret = 0;
 275
 276     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 277
 278     if (s->file) {
 279         DPRINTF("closing file\n");
 280         if (qemu_fclose(s->file) != 0) {
 281             ret = -1;
 282         }
 283         s->file = NULL;
 284     }
 285
 286     if (s->fd != -1)
 287         close(s->fd);
 288
 289     /* Don't resume monitor until we've flushed all of the buffers */
 290     if (s->mon) {
 291         monitor_resume(s->mon);
 292     }
 293
 294     s->fd = -1;
 295
 296     return ret;
 297 }
 298
 299 void migrate_fd_put_notify(void *opaque)
 300 {
 301     FdMigrationState *s = opaque;
 302
 303     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 304     qemu_file_put_notify(s->file);
 305 }
 306
 307 ssize_t migrate_fd_put_buffer(void *opaque, const void *data, size_t size)
 308 {
 309     FdMigrationState *s = opaque;
 310     ssize_t ret;
 311
 312     do {
 313         ret = s->write(s, data, size);
 314     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 315
 316     if (ret == -1)
 317         ret = -(s->get_error(s));
 318
 319     if (ret == -EAGAIN) {
 320         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 321     } else if (ret < 0) {
 322         if (s->mon) {
 323             monitor_resume(s->mon);
 324         }
 325         s->state = MIG_STATE_ERROR;
 326     }
 327
 328     return ret;
 329 }
 330
 331 void migrate_fd_connect(FdMigrationState *s)
 332 {
 333     int ret;
 334
 335     s->file = qemu_fopen_ops_buffered(s,
 336                                       s->bandwidth_limit,
 337                                       migrate_fd_put_buffer,
 338                                       migrate_fd_put_ready,
 339                                       migrate_fd_wait_for_unfreeze,
 340                                       migrate_fd_close);
 341
 342     DPRINTF("beginning savevm\n");
 343     ret = qemu_savevm_state_begin(s->mon, s->file, s->mig_state.blk,
 344                                   s->mig_state.shared);
 345     if (ret < 0) {
 346         DPRINTF("failed, %d\n", ret);
 347         migrate_fd_error(s);
 348         return;
 349     }
 350
 351     migrate_fd_put_ready(s);
 352 }
 353
 354 void migrate_fd_put_ready(void *opaque)
 355 {
 356     FdMigrationState *s = opaque;
 357
 358     if (s->state != MIG_STATE_ACTIVE) {
 359         DPRINTF("put_ready returning because of non-active state\n");
 360         return;
 361     }
 362
 363     DPRINTF("iterate\n");
 364     if (qemu_savevm_state_iterate(s->mon, s->file) == 1) {
 365         int state;
 366         int old_vm_running = vm_running;
 367
 368         DPRINTF("done iterating\n");
 369         vm_stop(0);
 370
 371         qemu_aio_flush();
 372         bdrv_flush_all();
 373         if ((qemu_savevm_state_complete(s->mon, s->file)) < 0) {
 374             if (old_vm_running) {
 375                 vm_start();
 376             }
 377             state = MIG_STATE_ERROR;
 378         } else {
 379             state = MIG_STATE_COMPLETED;
 380         }
 381         if (migrate_fd_cleanup(s) < 0) {
 382             if (old_vm_running) {
 383                 vm_start();
 384             }
 385             state = MIG_STATE_ERROR;
 386         }
 387         s->state = state;
 388     }
 389 }
 390
 391 int migrate_fd_get_status(MigrationState *mig_state)
 392 {
 393     FdMigrationState *s = migrate_to_fms(mig_state);
 394     return s->state;
 395 }
 396
 397 void migrate_fd_cancel(MigrationState *mig_state)
 398 {
 399     FdMigrationState *s = migrate_to_fms(mig_state);
 400
 401     if (s->state != MIG_STATE_ACTIVE)
 402         return;
 403
 404     DPRINTF("cancelling migration\n");
 405
 406     s->state = MIG_STATE_CANCELLED;
 407     qemu_savevm_state_cancel(s->mon, s->file);
 408
 409     migrate_fd_cleanup(s);
 410 }
 411
 412 void migrate_fd_release(MigrationState *mig_state)
 413 {
 414     FdMigrationState *s = migrate_to_fms(mig_state);
 415
 416     DPRINTF("releasing state\n");
 417
 418     if (s->state == MIG_STATE_ACTIVE) {
 419         s->state = MIG_STATE_CANCELLED;
 420         migrate_fd_cleanup(s);
 421     }
 422     qemu_free(s);
 423 }
 424
 425 void migrate_fd_wait_for_unfreeze(void *opaque)
 426 {
 427     FdMigrationState *s = opaque;
 428     int ret;
 429
 430     DPRINTF("wait for unfreeze\n");
 431     if (s->state != MIG_STATE_ACTIVE)
 432         return;
 433
 434     do {
 435         fd_set wfds;
 436
 437         FD_ZERO(&wfds);
 438         FD_SET(s->fd, &wfds);
 439
 440         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 441     } while (ret == -1 && (s->get_error(s)) == EINTR);
 442 }
 443
 444 int migrate_fd_close(void *opaque)
 445 {
 446     FdMigrationState *s = opaque;
 447
 448     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 449     return s->close(s);
 450 }