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 int64_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     int64_t d;
 136     FdMigrationState *s;
 137
 138     d = qdict_get_int(qdict, "value");
 139     if (d < 0) {
 140         d = 0;
 141     }
 142     max_throttle = d;
 143
 144     s = migrate_to_fms(current_migration);
 145     if (s && s->file) {
 146         qemu_file_set_rate_limit(s->file, max_throttle);
 147     }
 148
 149     return 0;
 150 }
 151
 152 /* amount of nanoseconds we are willing to wait for migration to be down.
 153  * the choice of nanoseconds is because it is the maximum resolution that
 154  * get_clock() can achieve. It is an internal measure. All user-visible
 155  * units must be in seconds */
 156 static uint64_t max_downtime = 30000000;
 157
 158 uint64_t migrate_max_downtime(void)
 159 {
 160     return max_downtime;
 161 }
 162
 163 int do_migrate_set_downtime(Monitor *mon, const QDict *qdict,
 164                             QObject **ret_data)
 165 {
 166     double d;
 167
 168     d = qdict_get_double(qdict, "value") * 1e9;
 169     d = MAX(0, MIN(UINT64_MAX, d));
 170     max_downtime = (uint64_t)d;
 171
 172     return 0;
 173 }
 174
 175 static void migrate_print_status(Monitor *mon, const char *name,
 176                                  const QDict *status_dict)
 177 {
 178     QDict *qdict;
 179
 180     qdict = qobject_to_qdict(qdict_get(status_dict, name));
 181
 182     monitor_printf(mon, "transferred %s: %" PRIu64 " kbytes\n", name,
 183                         qdict_get_int(qdict, "transferred") >> 10);
 184     monitor_printf(mon, "remaining %s: %" PRIu64 " kbytes\n", name,
 185                         qdict_get_int(qdict, "remaining") >> 10);
 186     monitor_printf(mon, "total %s: %" PRIu64 " kbytes\n", name,
 187                         qdict_get_int(qdict, "total") >> 10);
 188 }
 189
 190 void do_info_migrate_print(Monitor *mon, const QObject *data)
 191 {
 192     QDict *qdict;
 193
 194     qdict = qobject_to_qdict(data);
 195
 196     monitor_printf(mon, "Migration status: %s\n",
 197                    qdict_get_str(qdict, "status"));
 198
 199     if (qdict_haskey(qdict, "ram")) {
 200         migrate_print_status(mon, "ram", qdict);
 201     }
 202
 203     if (qdict_haskey(qdict, "disk")) {
 204         migrate_print_status(mon, "disk", qdict);
 205     }
 206 }
 207
 208 static void migrate_put_status(QDict *qdict, const char *name,
 209                                uint64_t trans, uint64_t rem, uint64_t total)
 210 {
 211     QObject *obj;
 212
 213     obj = qobject_from_jsonf("{ 'transferred': %" PRId64 ", "
 214                                "'remaining': %" PRId64 ", "
 215                                "'total': %" PRId64 " }", trans, rem, total);
 216     qdict_put_obj(qdict, name, obj);
 217 }
 218
 219 void do_info_migrate(Monitor *mon, QObject **ret_data)
 220 {
 221     QDict *qdict;
 222     MigrationState *s = current_migration;
 223
 224     if (s) {
 225         switch (s->get_status(s)) {
 226         case MIG_STATE_ACTIVE:
 227             qdict = qdict_new();
 228             qdict_put(qdict, "status", qstring_from_str("active"));
 229
 230             migrate_put_status(qdict, "ram", ram_bytes_transferred(),
 231                                ram_bytes_remaining(), ram_bytes_total());
 232
 233             if (blk_mig_active()) {
 234                 migrate_put_status(qdict, "disk", blk_mig_bytes_transferred(),
 235                                    blk_mig_bytes_remaining(),
 236                                    blk_mig_bytes_total());
 237             }
 238
 239             *ret_data = QOBJECT(qdict);
 240             break;
 241         case MIG_STATE_COMPLETED:
 242             *ret_data = qobject_from_jsonf("{ 'status': 'completed' }");
 243             break;
 244         case MIG_STATE_ERROR:
 245             *ret_data = qobject_from_jsonf("{ 'status': 'failed' }");
 246             break;
 247         case MIG_STATE_CANCELLED:
 248             *ret_data = qobject_from_jsonf("{ 'status': 'cancelled' }");
 249             break;
 250         }
 251     }
 252 }
 253
 254 /* shared migration helpers */
 255
 256 void migrate_fd_monitor_suspend(FdMigrationState *s, Monitor *mon)
 257 {
 258     s->mon = mon;
 259     if (monitor_suspend(mon) == 0) {
 260         DPRINTF("suspending monitor\n");
 261     } else {
 262         monitor_printf(mon, "terminal does not allow synchronous "
 263                        "migration, continuing detached\n");
 264     }
 265 }
 266
 267 void migrate_fd_error(FdMigrationState *s)
 268 {
 269     DPRINTF("setting error state\n");
 270     s->state = MIG_STATE_ERROR;
 271     migrate_fd_cleanup(s);
 272 }
 273
 274 int migrate_fd_cleanup(FdMigrationState *s)
 275 {
 276     int ret = 0;
 277
 278     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 279
 280     if (s->file) {
 281         DPRINTF("closing file\n");
 282         if (qemu_fclose(s->file) != 0) {
 283             ret = -1;
 284         }
 285         s->file = NULL;
 286     }
 287
 288     if (s->fd != -1)
 289         close(s->fd);
 290
 291     /* Don't resume monitor until we've flushed all of the buffers */
 292     if (s->mon) {
 293         monitor_resume(s->mon);
 294     }
 295
 296     s->fd = -1;
 297
 298     return ret;
 299 }
 300
 301 void migrate_fd_put_notify(void *opaque)
 302 {
 303     FdMigrationState *s = opaque;
 304
 305     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 306     qemu_file_put_notify(s->file);
 307 }
 308
 309 ssize_t migrate_fd_put_buffer(void *opaque, const void *data, size_t size)
 310 {
 311     FdMigrationState *s = opaque;
 312     ssize_t ret;
 313
 314     do {
 315         ret = s->write(s, data, size);
 316     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 317
 318     if (ret == -1)
 319         ret = -(s->get_error(s));
 320
 321     if (ret == -EAGAIN) {
 322         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 323     } else if (ret < 0) {
 324         if (s->mon) {
 325             monitor_resume(s->mon);
 326         }
 327         s->state = MIG_STATE_ERROR;
 328     }
 329
 330     return ret;
 331 }
 332
 333 void migrate_fd_connect(FdMigrationState *s)
 334 {
 335     int ret;
 336
 337     s->file = qemu_fopen_ops_buffered(s,
 338                                       s->bandwidth_limit,
 339                                       migrate_fd_put_buffer,
 340                                       migrate_fd_put_ready,
 341                                       migrate_fd_wait_for_unfreeze,
 342                                       migrate_fd_close);
 343
 344     DPRINTF("beginning savevm\n");
 345     ret = qemu_savevm_state_begin(s->mon, s->file, s->mig_state.blk,
 346                                   s->mig_state.shared);
 347     if (ret < 0) {
 348         DPRINTF("failed, %d\n", ret);
 349         migrate_fd_error(s);
 350         return;
 351     }
 352
 353     migrate_fd_put_ready(s);
 354 }
 355
 356 void migrate_fd_put_ready(void *opaque)
 357 {
 358     FdMigrationState *s = opaque;
 359
 360     if (s->state != MIG_STATE_ACTIVE) {
 361         DPRINTF("put_ready returning because of non-active state\n");
 362         return;
 363     }
 364
 365     DPRINTF("iterate\n");
 366     if (qemu_savevm_state_iterate(s->mon, s->file) == 1) {
 367         int state;
 368         int old_vm_running = vm_running;
 369
 370         DPRINTF("done iterating\n");
 371         vm_stop(0);
 372
 373         qemu_aio_flush();
 374         bdrv_flush_all();
 375         if ((qemu_savevm_state_complete(s->mon, s->file)) < 0) {
 376             if (old_vm_running) {
 377                 vm_start();
 378             }
 379             state = MIG_STATE_ERROR;
 380         } else {
 381             state = MIG_STATE_COMPLETED;
 382         }
 383         if (migrate_fd_cleanup(s) < 0) {
 384             if (old_vm_running) {
 385                 vm_start();
 386             }
 387             state = MIG_STATE_ERROR;
 388         }
 389         s->state = state;
 390     }
 391 }
 392
 393 int migrate_fd_get_status(MigrationState *mig_state)
 394 {
 395     FdMigrationState *s = migrate_to_fms(mig_state);
 396     return s->state;
 397 }
 398
 399 void migrate_fd_cancel(MigrationState *mig_state)
 400 {
 401     FdMigrationState *s = migrate_to_fms(mig_state);
 402
 403     if (s->state != MIG_STATE_ACTIVE)
 404         return;
 405
 406     DPRINTF("cancelling migration\n");
 407
 408     s->state = MIG_STATE_CANCELLED;
 409     qemu_savevm_state_cancel(s->mon, s->file);
 410
 411     migrate_fd_cleanup(s);
 412 }
 413
 414 void migrate_fd_release(MigrationState *mig_state)
 415 {
 416     FdMigrationState *s = migrate_to_fms(mig_state);
 417
 418     DPRINTF("releasing state\n");
 419
 420     if (s->state == MIG_STATE_ACTIVE) {
 421         s->state = MIG_STATE_CANCELLED;
 422         migrate_fd_cleanup(s);
 423     }
 424     qemu_free(s);
 425 }
 426
 427 void migrate_fd_wait_for_unfreeze(void *opaque)
 428 {
 429     FdMigrationState *s = opaque;
 430     int ret;
 431
 432     DPRINTF("wait for unfreeze\n");
 433     if (s->state != MIG_STATE_ACTIVE)
 434         return;
 435
 436     do {
 437         fd_set wfds;
 438
 439         FD_ZERO(&wfds);
 440         FD_SET(s->fd, &wfds);
 441
 442         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 443     } while (ret == -1 && (s->get_error(s)) == EINTR);
 444 }
 445
 446 int migrate_fd_close(void *opaque)
 447 {
 448     FdMigrationState *s = opaque;
 449
 450     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 451     return s->close(s);
 452 }