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