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                              QObject **ret_data)
 137 {
 138     double d;
 139
 140     d = qdict_get_double(qdict, "value") * 1e9;
 141     d = MAX(0, MIN(UINT64_MAX, d));
 142     max_downtime = (uint64_t)d;
 143 }
 144
 145 static void migrate_print_status(Monitor *mon, const char *name,
 146                                  const QDict *status_dict)
 147 {
 148     QDict *qdict;
 149
 150     qdict = qobject_to_qdict(qdict_get(status_dict, name));
 151
 152     monitor_printf(mon, "transferred %s: %" PRIu64 " kbytes\n", name,
 153                         qdict_get_int(qdict, "transferred") >> 10);
 154     monitor_printf(mon, "remaining %s: %" PRIu64 " kbytes\n", name,
 155                         qdict_get_int(qdict, "remaining") >> 10);
 156     monitor_printf(mon, "total %s: %" PRIu64 " kbytes\n", name,
 157                         qdict_get_int(qdict, "total") >> 10);
 158 }
 159
 160 void do_info_migrate_print(Monitor *mon, const QObject *data)
 161 {
 162     QDict *qdict;
 163
 164     qdict = qobject_to_qdict(data);
 165
 166     monitor_printf(mon, "Migration status: %s\n",
 167                    qdict_get_str(qdict, "status"));
 168
 169     if (qdict_haskey(qdict, "ram")) {
 170         migrate_print_status(mon, "ram", qdict);
 171     }
 172
 173     if (qdict_haskey(qdict, "disk")) {
 174         migrate_print_status(mon, "disk", qdict);
 175     }
 176 }
 177
 178 static void migrate_put_status(QDict *qdict, const char *name,
 179                                uint64_t trans, uint64_t rem, uint64_t total)
 180 {
 181     QObject *obj;
 182
 183     obj = qobject_from_jsonf("{ 'transferred': %" PRId64 ", "
 184                                "'remaining': %" PRId64 ", "
 185                                "'total': %" PRId64 " }", trans, rem, total);
 186     qdict_put_obj(qdict, name, obj);
 187 }
 188
 189 /**
 190  * do_info_migrate(): Migration status
 191  *
 192  * Return a QDict. If migration is active there will be another
 193  * QDict with RAM migration status and if block migration is active
 194  * another one with block migration status.
 195  *
 196  * The main QDict contains the following:
 197  *
 198  * - "status": migration status
 199  * - "ram": only present if "status" is "active", it is a QDict with the
 200  *   following RAM information (in bytes):
 201  *          - "transferred": amount transferred
 202  *          - "remaining": amount remaining
 203  *          - "total": total
 204  * - "disk": only present if "status" is "active" and it is a block migration,
 205  *   it is a QDict with the following disk information (in bytes):
 206  *          - "transferred": amount transferred
 207  *          - "remaining": amount remaining
 208  *          - "total": total
 209  *
 210  * Examples:
 211  *
 212  * 1. Migration is "completed":
 213  *
 214  * { "status": "completed" }
 215  *
 216  * 2. Migration is "active" and it is not a block migration:
 217  *
 218  * { "status": "active",
 219  *            "ram": { "transferred": 123, "remaining": 123, "total": 246 } }
 220  *
 221  * 3. Migration is "active" and it is a block migration:
 222  *
 223  * { "status": "active",
 224  *   "ram": { "total": 1057024, "remaining": 1053304, "transferred": 3720 },
 225  *   "disk": { "total": 20971520, "remaining": 20880384, "transferred": 91136 }}
 226  */
 227 void do_info_migrate(Monitor *mon, QObject **ret_data)
 228 {
 229     QDict *qdict;
 230     MigrationState *s = current_migration;
 231
 232     if (s) {
 233         switch (s->get_status(s)) {
 234         case MIG_STATE_ACTIVE:
 235             qdict = qdict_new();
 236             qdict_put(qdict, "status", qstring_from_str("active"));
 237
 238             migrate_put_status(qdict, "ram", ram_bytes_transferred(),
 239                                ram_bytes_remaining(), ram_bytes_total());
 240
 241             if (blk_mig_active()) {
 242                 migrate_put_status(qdict, "disk", blk_mig_bytes_transferred(),
 243                                    blk_mig_bytes_remaining(),
 244                                    blk_mig_bytes_total());
 245             }
 246
 247             *ret_data = QOBJECT(qdict);
 248             break;
 249         case MIG_STATE_COMPLETED:
 250             *ret_data = qobject_from_jsonf("{ 'status': 'completed' }");
 251             break;
 252         case MIG_STATE_ERROR:
 253             *ret_data = qobject_from_jsonf("{ 'status': 'failed' }");
 254             break;
 255         case MIG_STATE_CANCELLED:
 256             *ret_data = qobject_from_jsonf("{ 'status': 'cancelled' }");
 257             break;
 258         }
 259     }
 260 }
 261
 262 /* shared migration helpers */
 263
 264 void migrate_fd_monitor_suspend(FdMigrationState *s, Monitor *mon)
 265 {
 266     s->mon = mon;
 267     if (monitor_suspend(mon) == 0) {
 268         DPRINTF("suspending monitor\n");
 269     } else {
 270         monitor_printf(mon, "terminal does not allow synchronous "
 271                        "migration, continuing detached\n");
 272     }
 273 }
 274
 275 void migrate_fd_error(FdMigrationState *s)
 276 {
 277     DPRINTF("setting error state\n");
 278     s->state = MIG_STATE_ERROR;
 279     migrate_fd_cleanup(s);
 280 }
 281
 282 void migrate_fd_cleanup(FdMigrationState *s)
 283 {
 284     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 285
 286     if (s->file) {
 287         DPRINTF("closing file\n");
 288         qemu_fclose(s->file);
 289         s->file = NULL;
 290     }
 291
 292     if (s->fd != -1)
 293         close(s->fd);
 294
 295     /* Don't resume monitor until we've flushed all of the buffers */
 296     if (s->mon) {
 297         monitor_resume(s->mon);
 298     }
 299
 300     s->fd = -1;
 301 }
 302
 303 void migrate_fd_put_notify(void *opaque)
 304 {
 305     FdMigrationState *s = opaque;
 306
 307     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 308     qemu_file_put_notify(s->file);
 309 }
 310
 311 ssize_t migrate_fd_put_buffer(void *opaque, const void *data, size_t size)
 312 {
 313     FdMigrationState *s = opaque;
 314     ssize_t ret;
 315
 316     do {
 317         ret = s->write(s, data, size);
 318     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 319
 320     if (ret == -1)
 321         ret = -(s->get_error(s));
 322
 323     if (ret == -EAGAIN)
 324         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 325
 326     return ret;
 327 }
 328
 329 void migrate_fd_connect(FdMigrationState *s)
 330 {
 331     int ret;
 332
 333     s->file = qemu_fopen_ops_buffered(s,
 334                                       s->bandwidth_limit,
 335                                       migrate_fd_put_buffer,
 336                                       migrate_fd_put_ready,
 337                                       migrate_fd_wait_for_unfreeze,
 338                                       migrate_fd_close);
 339
 340     DPRINTF("beginning savevm\n");
 341     ret = qemu_savevm_state_begin(s->mon, s->file, s->mig_state.blk,
 342                                   s->mig_state.shared);
 343     if (ret < 0) {
 344         DPRINTF("failed, %d\n", ret);
 345         migrate_fd_error(s);
 346         return;
 347     }
 348
 349     migrate_fd_put_ready(s);
 350 }
 351
 352 void migrate_fd_put_ready(void *opaque)
 353 {
 354     FdMigrationState *s = opaque;
 355
 356     if (s->state != MIG_STATE_ACTIVE) {
 357         DPRINTF("put_ready returning because of non-active state\n");
 358         return;
 359     }
 360
 361     DPRINTF("iterate\n");
 362     if (qemu_savevm_state_iterate(s->mon, s->file) == 1) {
 363         int state;
 364         int old_vm_running = vm_running;
 365
 366         DPRINTF("done iterating\n");
 367         vm_stop(0);
 368
 369         qemu_aio_flush();
 370         bdrv_flush_all();
 371         if ((qemu_savevm_state_complete(s->mon, s->file)) < 0) {
 372             if (old_vm_running) {
 373                 vm_start();
 374             }
 375             state = MIG_STATE_ERROR;
 376         } else {
 377             state = MIG_STATE_COMPLETED;
 378         }
 379         migrate_fd_cleanup(s);
 380         s->state = state;
 381     }
 382 }
 383
 384 int migrate_fd_get_status(MigrationState *mig_state)
 385 {
 386     FdMigrationState *s = migrate_to_fms(mig_state);
 387     return s->state;
 388 }
 389
 390 void migrate_fd_cancel(MigrationState *mig_state)
 391 {
 392     FdMigrationState *s = migrate_to_fms(mig_state);
 393
 394     if (s->state != MIG_STATE_ACTIVE)
 395         return;
 396
 397     DPRINTF("cancelling migration\n");
 398
 399     s->state = MIG_STATE_CANCELLED;
 400     qemu_savevm_state_cancel(s->mon, s->file);
 401
 402     migrate_fd_cleanup(s);
 403 }
 404
 405 void migrate_fd_release(MigrationState *mig_state)
 406 {
 407     FdMigrationState *s = migrate_to_fms(mig_state);
 408
 409     DPRINTF("releasing state\n");
 410
 411     if (s->state == MIG_STATE_ACTIVE) {
 412         s->state = MIG_STATE_CANCELLED;
 413         migrate_fd_cleanup(s);
 414     }
 415     free(s);
 416 }
 417
 418 void migrate_fd_wait_for_unfreeze(void *opaque)
 419 {
 420     FdMigrationState *s = opaque;
 421     int ret;
 422
 423     DPRINTF("wait for unfreeze\n");
 424     if (s->state != MIG_STATE_ACTIVE)
 425         return;
 426
 427     do {
 428         fd_set wfds;
 429
 430         FD_ZERO(&wfds);
 431         FD_SET(s->fd, &wfds);
 432
 433         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 434     } while (ret == -1 && (s->get_error(s)) == EINTR);
 435 }
 436
 437 int migrate_fd_close(void *opaque)
 438 {
 439     FdMigrationState *s = opaque;
 440
 441     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 442     return s->close(s);
 443 }