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 int 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     return 0;
 109 }
 110
 111 int do_migrate_set_speed(Monitor *mon, const QDict *qdict, QObject **ret_data)
 112 {
 113     double d;
 114     FdMigrationState *s;
 115
 116     d = qdict_get_double(qdict, "value");
 117     d = MAX(0, MIN(UINT32_MAX, d));
 118     max_throttle = d;
 119
 120     s = migrate_to_fms(current_migration);
 121     if (s && s->file) {
 122         qemu_file_set_rate_limit(s->file, max_throttle);
 123     }
 124
 125     return 0;
 126 }
 127
 128 /* amount of nanoseconds we are willing to wait for migration to be down.
 129  * the choice of nanoseconds is because it is the maximum resolution that
 130  * get_clock() can achieve. It is an internal measure. All user-visible
 131  * units must be in seconds */
 132 static uint64_t max_downtime = 30000000;
 133
 134 uint64_t migrate_max_downtime(void)
 135 {
 136     return max_downtime;
 137 }
 138
 139 int do_migrate_set_downtime(Monitor *mon, const QDict *qdict,
 140                             QObject **ret_data)
 141 {
 142     double d;
 143
 144     d = qdict_get_double(qdict, "value") * 1e9;
 145     d = MAX(0, MIN(UINT64_MAX, d));
 146     max_downtime = (uint64_t)d;
 147
 148     return 0;
 149 }
 150
 151 static void migrate_print_status(Monitor *mon, const char *name,
 152                                  const QDict *status_dict)
 153 {
 154     QDict *qdict;
 155
 156     qdict = qobject_to_qdict(qdict_get(status_dict, name));
 157
 158     monitor_printf(mon, "transferred %s: %" PRIu64 " kbytes\n", name,
 159                         qdict_get_int(qdict, "transferred") >> 10);
 160     monitor_printf(mon, "remaining %s: %" PRIu64 " kbytes\n", name,
 161                         qdict_get_int(qdict, "remaining") >> 10);
 162     monitor_printf(mon, "total %s: %" PRIu64 " kbytes\n", name,
 163                         qdict_get_int(qdict, "total") >> 10);
 164 }
 165
 166 void do_info_migrate_print(Monitor *mon, const QObject *data)
 167 {
 168     QDict *qdict;
 169
 170     qdict = qobject_to_qdict(data);
 171
 172     monitor_printf(mon, "Migration status: %s\n",
 173                    qdict_get_str(qdict, "status"));
 174
 175     if (qdict_haskey(qdict, "ram")) {
 176         migrate_print_status(mon, "ram", qdict);
 177     }
 178
 179     if (qdict_haskey(qdict, "disk")) {
 180         migrate_print_status(mon, "disk", qdict);
 181     }
 182 }
 183
 184 static void migrate_put_status(QDict *qdict, const char *name,
 185                                uint64_t trans, uint64_t rem, uint64_t total)
 186 {
 187     QObject *obj;
 188
 189     obj = qobject_from_jsonf("{ 'transferred': %" PRId64 ", "
 190                                "'remaining': %" PRId64 ", "
 191                                "'total': %" PRId64 " }", trans, rem, total);
 192     qdict_put_obj(qdict, name, obj);
 193 }
 194
 195 /**
 196  * do_info_migrate(): Migration status
 197  *
 198  * Return a QDict. If migration is active there will be another
 199  * QDict with RAM migration status and if block migration is active
 200  * another one with block migration status.
 201  *
 202  * The main QDict contains the following:
 203  *
 204  * - "status": migration status
 205  * - "ram": only present if "status" is "active", it is a QDict with the
 206  *   following RAM information (in bytes):
 207  *          - "transferred": amount transferred
 208  *          - "remaining": amount remaining
 209  *          - "total": total
 210  * - "disk": only present if "status" is "active" and it is a block migration,
 211  *   it is a QDict with the following disk information (in bytes):
 212  *          - "transferred": amount transferred
 213  *          - "remaining": amount remaining
 214  *          - "total": total
 215  *
 216  * Examples:
 217  *
 218  * 1. Migration is "completed":
 219  *
 220  * { "status": "completed" }
 221  *
 222  * 2. Migration is "active" and it is not a block migration:
 223  *
 224  * { "status": "active",
 225  *            "ram": { "transferred": 123, "remaining": 123, "total": 246 } }
 226  *
 227  * 3. Migration is "active" and it is a block migration:
 228  *
 229  * { "status": "active",
 230  *   "ram": { "total": 1057024, "remaining": 1053304, "transferred": 3720 },
 231  *   "disk": { "total": 20971520, "remaining": 20880384, "transferred": 91136 }}
 232  */
 233 void do_info_migrate(Monitor *mon, QObject **ret_data)
 234 {
 235     QDict *qdict;
 236     MigrationState *s = current_migration;
 237
 238     if (s) {
 239         switch (s->get_status(s)) {
 240         case MIG_STATE_ACTIVE:
 241             qdict = qdict_new();
 242             qdict_put(qdict, "status", qstring_from_str("active"));
 243
 244             migrate_put_status(qdict, "ram", ram_bytes_transferred(),
 245                                ram_bytes_remaining(), ram_bytes_total());
 246
 247             if (blk_mig_active()) {
 248                 migrate_put_status(qdict, "disk", blk_mig_bytes_transferred(),
 249                                    blk_mig_bytes_remaining(),
 250                                    blk_mig_bytes_total());
 251             }
 252
 253             *ret_data = QOBJECT(qdict);
 254             break;
 255         case MIG_STATE_COMPLETED:
 256             *ret_data = qobject_from_jsonf("{ 'status': 'completed' }");
 257             break;
 258         case MIG_STATE_ERROR:
 259             *ret_data = qobject_from_jsonf("{ 'status': 'failed' }");
 260             break;
 261         case MIG_STATE_CANCELLED:
 262             *ret_data = qobject_from_jsonf("{ 'status': 'cancelled' }");
 263             break;
 264         }
 265     }
 266 }
 267
 268 /* shared migration helpers */
 269
 270 void migrate_fd_monitor_suspend(FdMigrationState *s, Monitor *mon)
 271 {
 272     s->mon = mon;
 273     if (monitor_suspend(mon) == 0) {
 274         DPRINTF("suspending monitor\n");
 275     } else {
 276         monitor_printf(mon, "terminal does not allow synchronous "
 277                        "migration, continuing detached\n");
 278     }
 279 }
 280
 281 void migrate_fd_error(FdMigrationState *s)
 282 {
 283     DPRINTF("setting error state\n");
 284     s->state = MIG_STATE_ERROR;
 285     migrate_fd_cleanup(s);
 286 }
 287
 288 void migrate_fd_cleanup(FdMigrationState *s)
 289 {
 290     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 291
 292     if (s->file) {
 293         DPRINTF("closing file\n");
 294         qemu_fclose(s->file);
 295         s->file = NULL;
 296     }
 297
 298     if (s->fd != -1)
 299         close(s->fd);
 300
 301     /* Don't resume monitor until we've flushed all of the buffers */
 302     if (s->mon) {
 303         monitor_resume(s->mon);
 304     }
 305
 306     s->fd = -1;
 307 }
 308
 309 void migrate_fd_put_notify(void *opaque)
 310 {
 311     FdMigrationState *s = opaque;
 312
 313     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 314     qemu_file_put_notify(s->file);
 315 }
 316
 317 ssize_t migrate_fd_put_buffer(void *opaque, const void *data, size_t size)
 318 {
 319     FdMigrationState *s = opaque;
 320     ssize_t ret;
 321
 322     do {
 323         ret = s->write(s, data, size);
 324     } while (ret == -1 && ((s->get_error(s)) == EINTR));
 325
 326     if (ret == -1)
 327         ret = -(s->get_error(s));
 328
 329     if (ret == -EAGAIN)
 330         qemu_set_fd_handler2(s->fd, NULL, NULL, migrate_fd_put_notify, s);
 331
 332     return ret;
 333 }
 334
 335 void migrate_fd_connect(FdMigrationState *s)
 336 {
 337     int ret;
 338
 339     s->file = qemu_fopen_ops_buffered(s,
 340                                       s->bandwidth_limit,
 341                                       migrate_fd_put_buffer,
 342                                       migrate_fd_put_ready,
 343                                       migrate_fd_wait_for_unfreeze,
 344                                       migrate_fd_close);
 345
 346     DPRINTF("beginning savevm\n");
 347     ret = qemu_savevm_state_begin(s->mon, s->file, s->mig_state.blk,
 348                                   s->mig_state.shared);
 349     if (ret < 0) {
 350         DPRINTF("failed, %d\n", ret);
 351         migrate_fd_error(s);
 352         return;
 353     }
 354
 355     migrate_fd_put_ready(s);
 356 }
 357
 358 void migrate_fd_put_ready(void *opaque)
 359 {
 360     FdMigrationState *s = opaque;
 361
 362     if (s->state != MIG_STATE_ACTIVE) {
 363         DPRINTF("put_ready returning because of non-active state\n");
 364         return;
 365     }
 366
 367     DPRINTF("iterate\n");
 368     if (qemu_savevm_state_iterate(s->mon, s->file) == 1) {
 369         int state;
 370         int old_vm_running = vm_running;
 371
 372         DPRINTF("done iterating\n");
 373         vm_stop(0);
 374
 375         qemu_aio_flush();
 376         bdrv_flush_all();
 377         if ((qemu_savevm_state_complete(s->mon, s->file)) < 0) {
 378             if (old_vm_running) {
 379                 vm_start();
 380             }
 381             state = MIG_STATE_ERROR;
 382         } else {
 383             state = MIG_STATE_COMPLETED;
 384         }
 385         migrate_fd_cleanup(s);
 386         s->state = state;
 387     }
 388 }
 389
 390 int migrate_fd_get_status(MigrationState *mig_state)
 391 {
 392     FdMigrationState *s = migrate_to_fms(mig_state);
 393     return s->state;
 394 }
 395
 396 void migrate_fd_cancel(MigrationState *mig_state)
 397 {
 398     FdMigrationState *s = migrate_to_fms(mig_state);
 399
 400     if (s->state != MIG_STATE_ACTIVE)
 401         return;
 402
 403     DPRINTF("cancelling migration\n");
 404
 405     s->state = MIG_STATE_CANCELLED;
 406     qemu_savevm_state_cancel(s->mon, s->file);
 407
 408     migrate_fd_cleanup(s);
 409 }
 410
 411 void migrate_fd_release(MigrationState *mig_state)
 412 {
 413     FdMigrationState *s = migrate_to_fms(mig_state);
 414
 415     DPRINTF("releasing state\n");
 416
 417     if (s->state == MIG_STATE_ACTIVE) {
 418         s->state = MIG_STATE_CANCELLED;
 419         migrate_fd_cleanup(s);
 420     }
 421     free(s);
 422 }
 423
 424 void migrate_fd_wait_for_unfreeze(void *opaque)
 425 {
 426     FdMigrationState *s = opaque;
 427     int ret;
 428
 429     DPRINTF("wait for unfreeze\n");
 430     if (s->state != MIG_STATE_ACTIVE)
 431         return;
 432
 433     do {
 434         fd_set wfds;
 435
 436         FD_ZERO(&wfds);
 437         FD_SET(s->fd, &wfds);
 438
 439         ret = select(s->fd + 1, NULL, &wfds, NULL, NULL);
 440     } while (ret == -1 && (s->get_error(s)) == EINTR);
 441 }
 442
 443 int migrate_fd_close(void *opaque)
 444 {
 445     FdMigrationState *s = opaque;
 446
 447     qemu_set_fd_handler2(s->fd, NULL, NULL, NULL, NULL);
 448     return s->close(s);
 449 }