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     double d;
 138
 139     d = qdict_get_double(qdict, "value") * 1e9;
 140     d = MAX(0, MIN(UINT64_MAX, d));
 141     max_downtime = (uint64_t)d;
 142 }
 143
 144 static void migrate_print_status(Monitor *mon, const char *name,
 145                                  const QDict *status_dict)
 146 {
 147     QDict *qdict;
 148
 149     qdict = qobject_to_qdict(qdict_get(status_dict, name));
 150
 151     monitor_printf(mon, "transferred %s: %" PRIu64 " kbytes\n", name,
 152                         qdict_get_int(qdict, "transferred") >> 10);
 153     monitor_printf(mon, "remaining %s: %" PRIu64 " kbytes\n", name,
 154                         qdict_get_int(qdict, "remaining") >> 10);
 155     monitor_printf(mon, "total %s: %" PRIu64 " kbytes\n", name,
 156                         qdict_get_int(qdict, "total") >> 10);
 157 }
 158
 159 void do_info_migrate_print(Monitor *mon, const QObject *data)
 160 {
 161     QDict *qdict;
 162
 163     qdict = qobject_to_qdict(data);
 164
 165     monitor_printf(mon, "Migration status: %s\n",
 166                    qdict_get_str(qdict, "status"));
 167
 168     if (qdict_haskey(qdict, "ram")) {
 169         migrate_print_status(mon, "ram", qdict);
 170     }
 171
 172     if (qdict_haskey(qdict, "disk")) {
 173         migrate_print_status(mon, "disk", qdict);
 174     }
 175 }
 176
 177 static void migrate_put_status(QDict *qdict, const char *name,
 178                                uint64_t trans, uint64_t rem, uint64_t total)
 179 {
 180     QObject *obj;
 181
 182     obj = qobject_from_jsonf("{ 'transferred': %" PRId64 ", "
 183                                "'remaining': %" PRId64 ", "
 184                                "'total': %" PRId64 " }", trans, rem, total);
 185     assert(obj != NULL);
 186
 187     qdict_put_obj(qdict, name, obj);
 188 }
 189
 190 /**
 191  * do_info_migrate(): Migration status
 192  *
 193  * Return a QDict. If migration is active there will be another
 194  * QDict with RAM migration status and if block migration is active
 195  * another one with block migration status.
 196  *
 197  * The main QDict contains the following:
 198  *
 199  * - "status": migration status
 200  * - "ram": only present if "status" is "active", it is a QDict with the
 201  *   following RAM information (in bytes):
 202  *          - "transferred": amount transferred
 203  *          - "remaining": amount remaining
 204  *          - "total": total
 205  * - "disk": only present if "status" is "active" and it is a block migration,
 206  *   it is a QDict with the following disk information (in bytes):
 207  *          - "transferred": amount transferred
 208  *          - "remaining": amount remaining
 209  *          - "total": total
 210  *
 211  * Examples:
 212  *
 213  * 1. Migration is "completed":
 214  *
 215  * { "status": "completed" }
 216  *
 217  * 2. Migration is "active" and it is not a block migration:
 218  *
 219  * { "status": "active",
 220  *            "ram": { "transferred": 123, "remaining": 123, "total": 246 } }
 221  *
 222  * 3. Migration is "active" and it is a block migration:
 223  *
 224  * { "status": "active",
 225  *   "ram": { "total": 1057024, "remaining": 1053304, "transferred": 3720 },
 226  *   "disk": { "total": 20971520, "remaining": 20880384, "transferred": 91136 }}
 227  */
 228 void do_info_migrate(Monitor *mon, QObject **ret_data)
 229 {
 230     QDict *qdict;
 231     MigrationState *s = current_migration;
 232
 233     if (s) {
 234         switch (s->get_status(s)) {
 235         case MIG_STATE_ACTIVE:
 236             qdict = qdict_new();
 237             qdict_put(qdict, "status", qstring_from_str("active"));
 238
 239             migrate_put_status(qdict, "ram", ram_bytes_transferred(),
 240                                ram_bytes_remaining(), ram_bytes_total());
 241
 242             if (blk_mig_active()) {
 243                 migrate_put_status(qdict, "disk", blk_mig_bytes_transferred(),
 244                                    blk_mig_bytes_remaining(),
 245                                    blk_mig_bytes_total());
 246             }
 247
 248             *ret_data = QOBJECT(qdict);
 249             break;
 250         case MIG_STATE_COMPLETED:
 251             *ret_data = qobject_from_jsonf("{ 'status': 'completed' }");
 252             break;
 253         case MIG_STATE_ERROR:
 254             *ret_data = qobject_from_jsonf("{ 'status': 'failed' }");
 255             break;
 256         case MIG_STATE_CANCELLED:
 257             *ret_data = qobject_from_jsonf("{ 'status': 'cancelled' }");
 258             break;
 259         }
 260         assert(*ret_data != NULL);
 261     }
 262 }
 263
 264 /* shared migration helpers */
 265
 266 void migrate_fd_monitor_suspend(FdMigrationState *s, Monitor *mon)
 267 {
 268     s->mon = mon;
 269     if (monitor_suspend(mon) == 0) {
 270         dprintf("suspending monitor\n");
 271     } else {
 272         monitor_printf(mon, "terminal does not allow synchronous "
 273                        "migration, continuing detached\n");
 274     }
 275 }
 276
 277 void migrate_fd_error(FdMigrationState *s)
 278 {
 279     dprintf("setting error state\n");
 280     s->state = MIG_STATE_ERROR;
 281     migrate_fd_cleanup(s);
 282 }
 283
 284 void migrate_fd_cleanup(FdMigrationState *s)
 285 {
 286     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 287
 288     if (s->file) {
 289         dprintf("closing file\n");
 290         qemu_fclose(s->file);
 291         s->file = NULL;
 292     }
 293
 294     if (s->fd != -1)
 295         close(s->fd);
 296
 297     /* Don't resume monitor until we've flushed all of the buffers */
 298     if (s->mon) {
 299         monitor_resume(s->mon);
 300     }
 301
 302     s->fd = -1;
 303 }
 304
 305 void migrate_fd_put_notify(void *opaque)
 306 {
 307     FdMigrationState *s = opaque;
 308
 309     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 310     qemu_file_put_notify(s->file);
 311 }
 312
 313 ssize_t migrate_fd_put_buffer(void *opaque, const void *data, size_t size)
 314 {
 315     FdMigrationState *s = opaque;
 316     ssize_t ret;
 317
 318     do {
 319         ret = s->write(s, data, size);
 320     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 321
 322     if (ret == -1)
 323         ret = -(s->get_error(s));
 324
 325     if (ret == -EAGAIN)
 326         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 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         migrate_fd_cleanup(s);
 382         s->state = state;
 383     }
 384 }
 385
 386 int migrate_fd_get_status(MigrationState *mig_state)
 387 {
 388     FdMigrationState *s = migrate_to_fms(mig_state);
 389     return s->state;
 390 }
 391
 392 void migrate_fd_cancel(MigrationState *mig_state)
 393 {
 394     FdMigrationState *s = migrate_to_fms(mig_state);
 395
 396     if (s->state != MIG_STATE_ACTIVE)
 397         return;
 398
 399     dprintf("cancelling migration\n");
 400
 401     s->state = MIG_STATE_CANCELLED;
 402     qemu_savevm_state_cancel(s->mon, s->file);
 403
 404     migrate_fd_cleanup(s);
 405 }
 406
 407 void migrate_fd_release(MigrationState *mig_state)
 408 {
 409     FdMigrationState *s = migrate_to_fms(mig_state);
 410
 411     dprintf("releasing state\n");
 412
 413     if (s->state == MIG_STATE_ACTIVE) {
 414         s->state = MIG_STATE_CANCELLED;
 415         migrate_fd_cleanup(s);
 416     }
 417     free(s);
 418 }
 419
 420 void migrate_fd_wait_for_unfreeze(void *opaque)
 421 {
 422     FdMigrationState *s = opaque;
 423     int ret;
 424
 425     dprintf("wait for unfreeze\n");
 426     if (s->state != MIG_STATE_ACTIVE)
 427         return;
 428
 429     do {
 430         fd_set wfds;
 431
 432         FD_ZERO(&wfds);
 433         FD_SET(s->fd, &wfds);
 434
 435         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 436     } while (ret == -1 && (s->get_error(s)) == EINTR);
 437 }
 438
 439 int migrate_fd_close(void *opaque)
 440 {
 441     FdMigrationState *s = opaque;
 442
 443     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 444     return s->close(s);
 445 }