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 void qemu_start_incoming_migration(const char *uri)
  40 {
  41     const char *p;
  42
  43     if (strstart(uri, "tcp:", &p))
  44         tcp_start_incoming_migration(p);
  45 #if !defined(WIN32)
  46     else if (strstart(uri, "exec:", &p))
  47         exec_start_incoming_migration(p);
  48     else if (strstart(uri, "unix:", &p))
  49         unix_start_incoming_migration(p);
  50     else if (strstart(uri, "fd:", &p))
  51         fd_start_incoming_migration(p);
  52 #endif
  53     else
  54         fprintf(stderr, "unknown migration protocol: %s\n", uri);
  55 }
  56
  57 void do_migrate(Monitor *mon, const QDict *qdict, QObject **ret_data)
  58 {
  59     MigrationState *s = NULL;
  60     const char *p;
  61     int detach = qdict_get_int(qdict, "detach");
  62     const char *uri = qdict_get_str(qdict, "uri");
  63
  64     if (current_migration &&
  65         current_migration->get_status(current_migration) == MIG_STATE_ACTIVE) {
  66         monitor_printf(mon, "migration already in progress\n");
  67         return;
  68     }
  69
  70     if (strstart(uri, "tcp:", &p))
  71         s = tcp_start_outgoing_migration(mon, p, max_throttle, detach,
  72                                          (int)qdict_get_int(qdict, "blk"),
  73                                          (int)qdict_get_int(qdict, "inc"));
  74 #if !defined(WIN32)
  75     else if (strstart(uri, "exec:", &p))
  76         s = exec_start_outgoing_migration(mon, p, max_throttle, detach,
  77                                           (int)qdict_get_int(qdict, "blk"),
  78                                           (int)qdict_get_int(qdict, "inc"));
  79     else if (strstart(uri, "unix:", &p))
  80         s = unix_start_outgoing_migration(mon, p, max_throttle, detach,
  81                                           (int)qdict_get_int(qdict, "blk"),
  82                                           (int)qdict_get_int(qdict, "inc"));
  83     else if (strstart(uri, "fd:", &p))
  84         s = fd_start_outgoing_migration(mon, p, max_throttle, detach,
  85                                         (int)qdict_get_int(qdict, "blk"),
  86                                         (int)qdict_get_int(qdict, "inc"));
  87 #endif
  88     else
  89         monitor_printf(mon, "unknown migration protocol: %s\n", uri);
  90
  91     if (s == NULL)
  92         monitor_printf(mon, "migration failed\n");
  93     else {
  94         if (current_migration)
  95             current_migration->release(current_migration);
  96
  97         current_migration = s;
  98     }
  99 }
 100
 101 void do_migrate_cancel(Monitor *mon, const QDict *qdict, QObject **ret_data)
 102 {
 103     MigrationState *s = current_migration;
 104
 105     if (s)
 106         s->cancel(s);
 107 }
 108
 109 void do_migrate_set_speed(Monitor *mon, const QDict *qdict, QObject **ret_data)
 110 {
 111     double d;
 112     FdMigrationState *s;
 113
 114     d = qdict_get_double(qdict, "value");
 115     d = MAX(0, MIN(UINT32_MAX, d));
 116     max_throttle = d;
 117
 118     s = migrate_to_fms(current_migration);
 119     if (s && s->file) {
 120         qemu_file_set_rate_limit(s->file, max_throttle);
 121     }
 122 }
 123
 124 /* amount of nanoseconds we are willing to wait for migration to be down.
 125  * the choice of nanoseconds is because it is the maximum resolution that
 126  * get_clock() can achieve. It is an internal measure. All user-visible
 127  * units must be in seconds */
 128 static uint64_t max_downtime = 30000000;
 129
 130 uint64_t migrate_max_downtime(void)
 131 {
 132     return max_downtime;
 133 }
 134
 135 void do_migrate_set_downtime(Monitor *mon, const QDict *qdict)
 136 {
 137     char *ptr;
 138     double d;
 139     const char *value = qdict_get_str(qdict, "value");
 140
 141     d = strtod(value, &ptr);
 142     if (!strcmp(ptr,"ms")) {
 143         d *= 1000000;
 144     } else if (!strcmp(ptr,"us")) {
 145         d *= 1000;
 146     } else if (!strcmp(ptr,"ns")) {
 147     } else {
 148         /* all else considered to be seconds */
 149         d *= 1000000000;
 150     }
 151
 152     max_downtime = (uint64_t)d;
 153 }
 154
 155 static void migrate_print_status(Monitor *mon, const char *name,
 156                                  const QDict *status_dict)
 157 {
 158     QDict *qdict;
 159
 160     qdict = qobject_to_qdict(qdict_get(status_dict, name));
 161
 162     monitor_printf(mon, "transferred %s: %" PRIu64 " kbytes\n", name,
 163                         qdict_get_int(qdict, "transferred") >> 10);
 164     monitor_printf(mon, "remaining %s: %" PRIu64 " kbytes\n", name,
 165                         qdict_get_int(qdict, "remaining") >> 10);
 166     monitor_printf(mon, "total %s: %" PRIu64 " kbytes\n", name,
 167                         qdict_get_int(qdict, "total") >> 10);
 168 }
 169
 170 void do_info_migrate_print(Monitor *mon, const QObject *data)
 171 {
 172     QDict *qdict;
 173
 174     qdict = qobject_to_qdict(data);
 175
 176     monitor_printf(mon, "Migration status: %s\n",
 177                    qdict_get_str(qdict, "status"));
 178
 179     if (qdict_haskey(qdict, "ram")) {
 180         migrate_print_status(mon, "ram", qdict);
 181     }
 182
 183     if (qdict_haskey(qdict, "disk")) {
 184         migrate_print_status(mon, "disk", qdict);
 185     }
 186 }
 187
 188 static void migrate_put_status(QDict *qdict, const char *name,
 189                                uint64_t trans, uint64_t rem, uint64_t total)
 190 {
 191     QObject *obj;
 192
 193     obj = qobject_from_jsonf("{ 'transferred': %" PRId64 ", "
 194                                "'remaining': %" PRId64 ", "
 195                                "'total': %" PRId64 " }", trans, rem, total);
 196     assert(obj != NULL);
 197
 198     qdict_put_obj(qdict, name, obj);
 199 }
 200
 201 /**
 202  * do_info_migrate(): Migration status
 203  *
 204  * Return a QDict. If migration is active there will be another
 205  * QDict with RAM migration status and if block migration is active
 206  * another one with block migration status.
 207  *
 208  * The main QDict contains the following:
 209  *
 210  * - "status": migration status
 211  * - "ram": only present if "status" is "active", it is a QDict with the
 212  *   following RAM information (in bytes):
 213  *          - "transferred": amount transferred
 214  *          - "remaining": amount remaining
 215  *          - "total": total
 216  * - "disk": only present if "status" is "active" and it is a block migration,
 217  *   it is a QDict with the following disk information (in bytes):
 218  *          - "transferred": amount transferred
 219  *          - "remaining": amount remaining
 220  *          - "total": total
 221  *
 222  * Examples:
 223  *
 224  * 1. Migration is "completed":
 225  *
 226  * { "status": "completed" }
 227  *
 228  * 2. Migration is "active" and it is not a block migration:
 229  *
 230  * { "status": "active",
 231  *            "ram": { "transferred": 123, "remaining": 123, "total": 246 } }
 232  *
 233  * 3. Migration is "active" and it is a block migration:
 234  *
 235  * { "status": "active",
 236  *   "ram": { "total": 1057024, "remaining": 1053304, "transferred": 3720 },
 237  *   "disk": { "total": 20971520, "remaining": 20880384, "transferred": 91136 }}
 238  */
 239 void do_info_migrate(Monitor *mon, QObject **ret_data)
 240 {
 241     QDict *qdict;
 242     MigrationState *s = current_migration;
 243
 244     if (s) {
 245         switch (s->get_status(s)) {
 246         case MIG_STATE_ACTIVE:
 247             qdict = qdict_new();
 248             qdict_put(qdict, "status", qstring_from_str("active"));
 249
 250             migrate_put_status(qdict, "ram", ram_bytes_transferred(),
 251                                ram_bytes_remaining(), ram_bytes_total());
 252
 253             if (blk_mig_active()) {
 254                 migrate_put_status(qdict, "disk", blk_mig_bytes_transferred(),
 255                                    blk_mig_bytes_remaining(),
 256                                    blk_mig_bytes_total());
 257             }
 258
 259             *ret_data = QOBJECT(qdict);
 260             break;
 261         case MIG_STATE_COMPLETED:
 262             *ret_data = qobject_from_jsonf("{ 'status': 'completed' }");
 263             break;
 264         case MIG_STATE_ERROR:
 265             *ret_data = qobject_from_jsonf("{ 'status': 'failed' }");
 266             break;
 267         case MIG_STATE_CANCELLED:
 268             *ret_data = qobject_from_jsonf("{ 'status': 'cancelled' }");
 269             break;
 270         }
 271         assert(*ret_data != NULL);
 272     }
 273 }
 274
 275 /* shared migration helpers */
 276
 277 void migrate_fd_monitor_suspend(FdMigrationState *s, Monitor *mon)
 278 {
 279     s->mon = mon;
 280     if (monitor_suspend(mon) == 0) {
 281         dprintf("suspending monitor\n");
 282     } else {
 283         monitor_printf(mon, "terminal does not allow synchronous "
 284                        "migration, continuing detached\n");
 285     }
 286 }
 287
 288 void migrate_fd_error(FdMigrationState *s)
 289 {
 290     dprintf("setting error state\n");
 291     s->state = MIG_STATE_ERROR;
 292     migrate_fd_cleanup(s);
 293 }
 294
 295 void migrate_fd_cleanup(FdMigrationState *s)
 296 {
 297     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 298
 299     if (s->file) {
 300         dprintf("closing file\n");
 301         qemu_fclose(s->file);
 302         s->file = NULL;
 303     }
 304
 305     if (s->fd != -1)
 306         close(s->fd);
 307
 308     /* Don't resume monitor until we've flushed all of the buffers */
 309     if (s->mon) {
 310         monitor_resume(s->mon);
 311     }
 312
 313     s->fd = -1;
 314 }
 315
 316 void migrate_fd_put_notify(void *opaque)
 317 {
 318     FdMigrationState *s = opaque;
 319
 320     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 321     qemu_file_put_notify(s->file);
 322 }
 323
 324 ssize_t migrate_fd_put_buffer(void *opaque, const void *data, size_t size)
 325 {
 326     FdMigrationState *s = opaque;
 327     ssize_t ret;
 328
 329     do {
 330         ret = s->write(s, data, size);
 331     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 332
 333     if (ret == -1)
 334         ret = -(s->get_error(s));
 335
 336     if (ret == -EAGAIN)
 337         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 338
 339     return ret;
 340 }
 341
 342 void migrate_fd_connect(FdMigrationState *s)
 343 {
 344     int ret;
 345
 346     s->file = qemu_fopen_ops_buffered(s,
 347                                       s->bandwidth_limit,
 348                                       migrate_fd_put_buffer,
 349                                       migrate_fd_put_ready,
 350                                       migrate_fd_wait_for_unfreeze,
 351                                       migrate_fd_close);
 352
 353     dprintf("beginning savevm\n");
 354     ret = qemu_savevm_state_begin(s->mon, s->file, s->mig_state.blk,
 355                                   s->mig_state.shared);
 356     if (ret < 0) {
 357         dprintf("failed, %d\n", ret);
 358         migrate_fd_error(s);
 359         return;
 360     }
 361
 362     migrate_fd_put_ready(s);
 363 }
 364
 365 void migrate_fd_put_ready(void *opaque)
 366 {
 367     FdMigrationState *s = opaque;
 368
 369     if (s->state != MIG_STATE_ACTIVE) {
 370         dprintf("put_ready returning because of non-active state\n");
 371         return;
 372     }
 373
 374     dprintf("iterate\n");
 375     if (qemu_savevm_state_iterate(s->mon, s->file) == 1) {
 376         int state;
 377         int old_vm_running = vm_running;
 378
 379         dprintf("done iterating\n");
 380         vm_stop(0);
 381
 382         qemu_aio_flush();
 383         bdrv_flush_all();
 384         if ((qemu_savevm_state_complete(s->mon, s->file)) < 0) {
 385             if (old_vm_running) {
 386                 vm_start();
 387             }
 388             state = MIG_STATE_ERROR;
 389         } else {
 390             state = MIG_STATE_COMPLETED;
 391         }
 392         migrate_fd_cleanup(s);
 393         s->state = state;
 394     }
 395 }
 396
 397 int migrate_fd_get_status(MigrationState *mig_state)
 398 {
 399     FdMigrationState *s = migrate_to_fms(mig_state);
 400     return s->state;
 401 }
 402
 403 void migrate_fd_cancel(MigrationState *mig_state)
 404 {
 405     FdMigrationState *s = migrate_to_fms(mig_state);
 406
 407     if (s->state != MIG_STATE_ACTIVE)
 408         return;
 409
 410     dprintf("cancelling migration\n");
 411
 412     s->state = MIG_STATE_CANCELLED;
 413     qemu_savevm_state_cancel(s->mon, s->file);
 414
 415     migrate_fd_cleanup(s);
 416 }
 417
 418 void migrate_fd_release(MigrationState *mig_state)
 419 {
 420     FdMigrationState *s = migrate_to_fms(mig_state);
 421
 422     dprintf("releasing state\n");
 423
 424     if (s->state == MIG_STATE_ACTIVE) {
 425         s->state = MIG_STATE_CANCELLED;
 426         migrate_fd_cleanup(s);
 427     }
 428     free(s);
 429 }
 430
 431 void migrate_fd_wait_for_unfreeze(void *opaque)
 432 {
 433     FdMigrationState *s = opaque;
 434     int ret;
 435
 436     dprintf("wait for unfreeze\n");
 437     if (s->state != MIG_STATE_ACTIVE)
 438         return;
 439
 440     do {
 441         fd_set wfds;
 442
 443         FD_ZERO(&wfds);
 444         FD_SET(s->fd, &wfds);
 445
 446         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 447     } while (ret == -1 && (s->get_error(s)) == EINTR);
 448 }
 449
 450 int migrate_fd_close(void *opaque)
 451 {
 452     FdMigrationState *s = opaque;
 453
 454     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 455     return s->close(s);
 456 }