migration/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  * Contributions after 2012-01-13 are licensed under the terms of the
  13  * GNU GPL, version 2 or (at your option) any later version.
  14  */
  15
  16 #include "qemu/osdep.h"
  17 #include "qemu/cutils.h"
  18 #include "qemu/error-report.h"
  19 #include "migration/blocker.h"
  20 #include "exec.h"
  21 #include "fd.h"
  22 #include "socket.h"
  23 #include "rdma.h"
  24 #include "ram.h"
  25 #include "migration/global_state.h"
  26 #include "migration/misc.h"
  27 #include "migration.h"
  28 #include "savevm.h"
  29 #include "qemu-file-channel.h"
  30 #include "qemu-file.h"
  31 #include "migration/vmstate.h"
  32 #include "block/block.h"
  33 #include "qapi/qmp/qerror.h"
  34 #include "qemu/rcu.h"
  35 #include "block.h"
  36 #include "postcopy-ram.h"
  37 #include "qemu/thread.h"
  38 #include "qmp-commands.h"
  39 #include "trace.h"
  40 #include "qapi-event.h"
  41 #include "exec/target_page.h"
  42 #include "io/channel-buffer.h"
  43 #include "migration/colo.h"
  44 #include "hw/boards.h"
  45 #include "monitor/monitor.h"
  46
  47 #define MAX_THROTTLE  (32 << 20)      /* Migration transfer speed throttling */
  48
  49 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
  50  * data. */
  51 #define BUFFER_DELAY     100
  52 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
  53
  54 /* Time in milliseconds we are allowed to stop the source,
  55  * for sending the last part */
  56 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
  57
  58 /* Maximum migrate downtime set to 2000 seconds */
  59 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
  60 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
  61
  62 /* Default compression thread count */
  63 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
  64 /* Default decompression thread count, usually decompression is at
  65  * least 4 times as fast as compression.*/
  66 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
  67 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
  68 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
  69 /* Define default autoconverge cpu throttle migration parameters */
  70 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
  71 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
  72
  73 /* Migration XBZRLE default cache size */
  74 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
  75
  76 /* The delay time (in ms) between two COLO checkpoints
  77  * Note: Please change this default value to 10000 when we support hybrid mode.
  78  */
  79 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
  80 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
  81 #define DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT 16
  82
  83 static NotifierList migration_state_notifiers =
  84     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
  85
  86 static bool deferred_incoming;
  87
  88 /* Messages sent on the return path from destination to source */
  89 enum mig_rp_message_type {
  90     MIG_RP_MSG_INVALID = 0,  /* Must be 0 */
  91     MIG_RP_MSG_SHUT,         /* sibling will not send any more RP messages */
  92     MIG_RP_MSG_PONG,         /* Response to a PING; data (seq: be32 ) */
  93
  94     MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
  95     MIG_RP_MSG_REQ_PAGES,    /* data (start: be64, len: be32) */
  96
  97     MIG_RP_MSG_MAX
  98 };
  99
 100 /* When we add fault tolerance, we could have several
 101    migrations at once.  For now we don't need to add
 102    dynamic creation of migration */
 103
 104 static MigrationState *current_migration;
 105
 106 static bool migration_object_check(MigrationState *ms, Error **errp);
 107 static int migration_maybe_pause(MigrationState *s,
 108                                  int *current_active_state,
 109                                  int new_state);
 110
 111 void migration_object_init(void)
 112 {
 113     MachineState *ms = MACHINE(qdev_get_machine());
 114     Error *err = NULL;
 115
 116     /* This can only be called once. */
 117     assert(!current_migration);
 118     current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
 119
 120     if (!migration_object_check(current_migration, &err)) {
 121         error_report_err(err);
 122         exit(1);
 123     }
 124
 125     /*
 126      * We cannot really do this in migration_instance_init() since at
 127      * that time global properties are not yet applied, then this
 128      * value will be definitely replaced by something else.
 129      */
 130     if (ms->enforce_config_section) {
 131         current_migration->send_configuration = true;
 132     }
 133 }
 134
 135 /* For outgoing */
 136 MigrationState *migrate_get_current(void)
 137 {
 138     /* This can only be called after the object created. */
 139     assert(current_migration);
 140     return current_migration;
 141 }
 142
 143 MigrationIncomingState *migration_incoming_get_current(void)
 144 {
 145     static bool once;
 146     static MigrationIncomingState mis_current;
 147
 148     if (!once) {
 149         mis_current.state = MIGRATION_STATUS_NONE;
 150         memset(&mis_current, 0, sizeof(MigrationIncomingState));
 151         qemu_mutex_init(&mis_current.rp_mutex);
 152         qemu_event_init(&mis_current.main_thread_load_event, false);
 153         once = true;
 154     }
 155     return &mis_current;
 156 }
 157
 158 void migration_incoming_state_destroy(void)
 159 {
 160     struct MigrationIncomingState *mis = migration_incoming_get_current();
 161
 162     if (mis->to_src_file) {
 163         /* Tell source that we are done */
 164         migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
 165         qemu_fclose(mis->to_src_file);
 166         mis->to_src_file = NULL;
 167     }
 168
 169     if (mis->from_src_file) {
 170         qemu_fclose(mis->from_src_file);
 171         mis->from_src_file = NULL;
 172     }
 173
 174     qemu_event_reset(&mis->main_thread_load_event);
 175 }
 176
 177 static void migrate_generate_event(int new_state)
 178 {
 179     if (migrate_use_events()) {
 180         qapi_event_send_migration(new_state, &error_abort);
 181     }
 182 }
 183
 184 /*
 185  * Called on -incoming with a defer: uri.
 186  * The migration can be started later after any parameters have been
 187  * changed.
 188  */
 189 static void deferred_incoming_migration(Error **errp)
 190 {
 191     if (deferred_incoming) {
 192         error_setg(errp, "Incoming migration already deferred");
 193     }
 194     deferred_incoming = true;
 195 }
 196
 197 /*
 198  * Send a message on the return channel back to the source
 199  * of the migration.
 200  */
 201 static void migrate_send_rp_message(MigrationIncomingState *mis,
 202                                     enum mig_rp_message_type message_type,
 203                                     uint16_t len, void *data)
 204 {
 205     trace_migrate_send_rp_message((int)message_type, len);
 206     qemu_mutex_lock(&mis->rp_mutex);
 207     qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
 208     qemu_put_be16(mis->to_src_file, len);
 209     qemu_put_buffer(mis->to_src_file, data, len);
 210     qemu_fflush(mis->to_src_file);
 211     qemu_mutex_unlock(&mis->rp_mutex);
 212 }
 213
 214 /* Request a range of pages from the source VM at the given
 215  * start address.
 216  *   rbname: Name of the RAMBlock to request the page in, if NULL it's the same
 217  *           as the last request (a name must have been given previously)
 218  *   Start: Address offset within the RB
 219  *   Len: Length in bytes required - must be a multiple of pagesize
 220  */
 221 void migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
 222                                ram_addr_t start, size_t len)
 223 {
 224     uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
 225     size_t msglen = 12; /* start + len */
 226
 227     *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
 228     *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
 229
 230     if (rbname) {
 231         int rbname_len = strlen(rbname);
 232         assert(rbname_len < 256);
 233
 234         bufc[msglen++] = rbname_len;
 235         memcpy(bufc + msglen, rbname, rbname_len);
 236         msglen += rbname_len;
 237         migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES_ID, msglen, bufc);
 238     } else {
 239         migrate_send_rp_message(mis, MIG_RP_MSG_REQ_PAGES, msglen, bufc);
 240     }
 241 }
 242
 243 void qemu_start_incoming_migration(const char *uri, Error **errp)
 244 {
 245     const char *p;
 246
 247     qapi_event_send_migration(MIGRATION_STATUS_SETUP, &error_abort);
 248     if (!strcmp(uri, "defer")) {
 249         deferred_incoming_migration(errp);
 250     } else if (strstart(uri, "tcp:", &p)) {
 251         tcp_start_incoming_migration(p, errp);
 252 #ifdef CONFIG_RDMA
 253     } else if (strstart(uri, "rdma:", &p)) {
 254         rdma_start_incoming_migration(p, errp);
 255 #endif
 256     } else if (strstart(uri, "exec:", &p)) {
 257         exec_start_incoming_migration(p, errp);
 258     } else if (strstart(uri, "unix:", &p)) {
 259         unix_start_incoming_migration(p, errp);
 260     } else if (strstart(uri, "fd:", &p)) {
 261         fd_start_incoming_migration(p, errp);
 262     } else {
 263         error_setg(errp, "unknown migration protocol: %s", uri);
 264     }
 265 }
 266
 267 static void process_incoming_migration_bh(void *opaque)
 268 {
 269     Error *local_err = NULL;
 270     MigrationIncomingState *mis = opaque;
 271
 272     /* Make sure all file formats flush their mutable metadata.
 273      * If we get an error here, just don't restart the VM yet. */
 274     bdrv_invalidate_cache_all(&local_err);
 275     if (local_err) {
 276         error_report_err(local_err);
 277         local_err = NULL;
 278         autostart = false;
 279     }
 280
 281     /*
 282      * This must happen after all error conditions are dealt with and
 283      * we're sure the VM is going to be running on this host.
 284      */
 285     qemu_announce_self();
 286
 287     if (multifd_load_cleanup(&local_err) != 0) {
 288         error_report_err(local_err);
 289         autostart = false;
 290     }
 291     /* If global state section was not received or we are in running
 292        state, we need to obey autostart. Any other state is set with
 293        runstate_set. */
 294
 295     if (!global_state_received() ||
 296         global_state_get_runstate() == RUN_STATE_RUNNING) {
 297         if (autostart) {
 298             vm_start();
 299         } else {
 300             runstate_set(RUN_STATE_PAUSED);
 301         }
 302     } else {
 303         runstate_set(global_state_get_runstate());
 304     }
 305     /*
 306      * This must happen after any state changes since as soon as an external
 307      * observer sees this event they might start to prod at the VM assuming
 308      * it's ready to use.
 309      */
 310     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
 311                       MIGRATION_STATUS_COMPLETED);
 312     qemu_bh_delete(mis->bh);
 313     migration_incoming_state_destroy();
 314 }
 315
 316 static void process_incoming_migration_co(void *opaque)
 317 {
 318     MigrationIncomingState *mis = migration_incoming_get_current();
 319     PostcopyState ps;
 320     int ret;
 321
 322     assert(mis->from_src_file);
 323     mis->largest_page_size = qemu_ram_pagesize_largest();
 324     postcopy_state_set(POSTCOPY_INCOMING_NONE);
 325     migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
 326                       MIGRATION_STATUS_ACTIVE);
 327     ret = qemu_loadvm_state(mis->from_src_file);
 328
 329     ps = postcopy_state_get();
 330     trace_process_incoming_migration_co_end(ret, ps);
 331     if (ps != POSTCOPY_INCOMING_NONE) {
 332         if (ps == POSTCOPY_INCOMING_ADVISE) {
 333             /*
 334              * Where a migration had postcopy enabled (and thus went to advise)
 335              * but managed to complete within the precopy period, we can use
 336              * the normal exit.
 337              */
 338             postcopy_ram_incoming_cleanup(mis);
 339         } else if (ret >= 0) {
 340             /*
 341              * Postcopy was started, cleanup should happen at the end of the
 342              * postcopy thread.
 343              */
 344             trace_process_incoming_migration_co_postcopy_end_main();
 345             return;
 346         }
 347         /* Else if something went wrong then just fall out of the normal exit */
 348     }
 349
 350     /* we get COLO info, and know if we are in COLO mode */
 351     if (!ret && migration_incoming_enable_colo()) {
 352         mis->migration_incoming_co = qemu_coroutine_self();
 353         qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
 354              colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
 355         mis->have_colo_incoming_thread = true;
 356         qemu_coroutine_yield();
 357
 358         /* Wait checkpoint incoming thread exit before free resource */
 359         qemu_thread_join(&mis->colo_incoming_thread);
 360     }
 361
 362     if (ret < 0) {
 363         Error *local_err = NULL;
 364
 365         migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
 366                           MIGRATION_STATUS_FAILED);
 367         error_report("load of migration failed: %s", strerror(-ret));
 368         qemu_fclose(mis->from_src_file);
 369         if (multifd_load_cleanup(&local_err) != 0) {
 370             error_report_err(local_err);
 371         }
 372         exit(EXIT_FAILURE);
 373     }
 374     mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
 375     qemu_bh_schedule(mis->bh);
 376 }
 377
 378 static void migration_incoming_setup(QEMUFile *f)
 379 {
 380     MigrationIncomingState *mis = migration_incoming_get_current();
 381
 382     if (multifd_load_setup() != 0) {
 383         /* We haven't been able to create multifd threads
 384            nothing better to do */
 385         exit(EXIT_FAILURE);
 386     }
 387
 388     if (!mis->from_src_file) {
 389         mis->from_src_file = f;
 390     }
 391     qemu_file_set_blocking(f, false);
 392 }
 393
 394 static void migration_incoming_process(void)
 395 {
 396     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
 397     qemu_coroutine_enter(co);
 398 }
 399
 400 void migration_fd_process_incoming(QEMUFile *f)
 401 {
 402     migration_incoming_setup(f);
 403     migration_incoming_process();
 404 }
 405
 406 void migration_ioc_process_incoming(QIOChannel *ioc)
 407 {
 408     MigrationIncomingState *mis = migration_incoming_get_current();
 409
 410     if (!mis->from_src_file) {
 411         QEMUFile *f = qemu_fopen_channel_input(ioc);
 412         migration_fd_process_incoming(f);
 413     }
 414     /* We still only have a single channel.  Nothing to do here yet */
 415 }
 416
 417 /**
 418  * @migration_has_all_channels: We have received all channels that we need
 419  *
 420  * Returns true when we have got connections to all the channels that
 421  * we need for migration.
 422  */
 423 bool migration_has_all_channels(void)
 424 {
 425     return true;
 426 }
 427
 428 /*
 429  * Send a 'SHUT' message on the return channel with the given value
 430  * to indicate that we've finished with the RP.  Non-0 value indicates
 431  * error.
 432  */
 433 void migrate_send_rp_shut(MigrationIncomingState *mis,
 434                           uint32_t value)
 435 {
 436     uint32_t buf;
 437
 438     buf = cpu_to_be32(value);
 439     migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
 440 }
 441
 442 /*
 443  * Send a 'PONG' message on the return channel with the given value
 444  * (normally in response to a 'PING')
 445  */
 446 void migrate_send_rp_pong(MigrationIncomingState *mis,
 447                           uint32_t value)
 448 {
 449     uint32_t buf;
 450
 451     buf = cpu_to_be32(value);
 452     migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
 453 }
 454
 455 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
 456 {
 457     MigrationCapabilityStatusList *head = NULL;
 458     MigrationCapabilityStatusList *caps;
 459     MigrationState *s = migrate_get_current();
 460     int i;
 461
 462     caps = NULL; /* silence compiler warning */
 463     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
 464 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
 465         if (i == MIGRATION_CAPABILITY_BLOCK) {
 466             continue;
 467         }
 468 #endif
 469         if (head == NULL) {
 470             head = g_malloc0(sizeof(*caps));
 471             caps = head;
 472         } else {
 473             caps->next = g_malloc0(sizeof(*caps));
 474             caps = caps->next;
 475         }
 476         caps->value =
 477             g_malloc(sizeof(*caps->value));
 478         caps->value->capability = i;
 479         caps->value->state = s->enabled_capabilities[i];
 480     }
 481
 482     return head;
 483 }
 484
 485 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
 486 {
 487     MigrationParameters *params;
 488     MigrationState *s = migrate_get_current();
 489
 490     /* TODO use QAPI_CLONE() instead of duplicating it inline */
 491     params = g_malloc0(sizeof(*params));
 492     params->has_compress_level = true;
 493     params->compress_level = s->parameters.compress_level;
 494     params->has_compress_threads = true;
 495     params->compress_threads = s->parameters.compress_threads;
 496     params->has_decompress_threads = true;
 497     params->decompress_threads = s->parameters.decompress_threads;
 498     params->has_cpu_throttle_initial = true;
 499     params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
 500     params->has_cpu_throttle_increment = true;
 501     params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
 502     params->has_tls_creds = true;
 503     params->tls_creds = g_strdup(s->parameters.tls_creds);
 504     params->has_tls_hostname = true;
 505     params->tls_hostname = g_strdup(s->parameters.tls_hostname);
 506     params->has_max_bandwidth = true;
 507     params->max_bandwidth = s->parameters.max_bandwidth;
 508     params->has_downtime_limit = true;
 509     params->downtime_limit = s->parameters.downtime_limit;
 510     params->has_x_checkpoint_delay = true;
 511     params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
 512     params->has_block_incremental = true;
 513     params->block_incremental = s->parameters.block_incremental;
 514     params->has_x_multifd_channels = true;
 515     params->x_multifd_channels = s->parameters.x_multifd_channels;
 516     params->has_x_multifd_page_count = true;
 517     params->x_multifd_page_count = s->parameters.x_multifd_page_count;
 518     params->has_xbzrle_cache_size = true;
 519     params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
 520
 521     return params;
 522 }
 523
 524 /*
 525  * Return true if we're already in the middle of a migration
 526  * (i.e. any of the active or setup states)
 527  */
 528 static bool migration_is_setup_or_active(int state)
 529 {
 530     switch (state) {
 531     case MIGRATION_STATUS_ACTIVE:
 532     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
 533     case MIGRATION_STATUS_SETUP:
 534     case MIGRATION_STATUS_PRE_SWITCHOVER:
 535     case MIGRATION_STATUS_DEVICE:
 536         return true;
 537
 538     default:
 539         return false;
 540
 541     }
 542 }
 543
 544 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
 545 {
 546     info->has_ram = true;
 547     info->ram = g_malloc0(sizeof(*info->ram));
 548     info->ram->transferred = ram_counters.transferred;
 549     info->ram->total = ram_bytes_total();
 550     info->ram->duplicate = ram_counters.duplicate;
 551     /* legacy value.  It is not used anymore */
 552     info->ram->skipped = 0;
 553     info->ram->normal = ram_counters.normal;
 554     info->ram->normal_bytes = ram_counters.normal *
 555         qemu_target_page_size();
 556     info->ram->mbps = s->mbps;
 557     info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
 558     info->ram->postcopy_requests = ram_counters.postcopy_requests;
 559     info->ram->page_size = qemu_target_page_size();
 560
 561     if (migrate_use_xbzrle()) {
 562         info->has_xbzrle_cache = true;
 563         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
 564         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
 565         info->xbzrle_cache->bytes = xbzrle_counters.bytes;
 566         info->xbzrle_cache->pages = xbzrle_counters.pages;
 567         info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
 568         info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
 569         info->xbzrle_cache->overflow = xbzrle_counters.overflow;
 570     }
 571
 572     if (cpu_throttle_active()) {
 573         info->has_cpu_throttle_percentage = true;
 574         info->cpu_throttle_percentage = cpu_throttle_get_percentage();
 575     }
 576
 577     if (s->state != MIGRATION_STATUS_COMPLETED) {
 578         info->ram->remaining = ram_bytes_remaining();
 579         info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
 580     }
 581 }
 582
 583 static void populate_disk_info(MigrationInfo *info)
 584 {
 585     if (blk_mig_active()) {
 586         info->has_disk = true;
 587         info->disk = g_malloc0(sizeof(*info->disk));
 588         info->disk->transferred = blk_mig_bytes_transferred();
 589         info->disk->remaining = blk_mig_bytes_remaining();
 590         info->disk->total = blk_mig_bytes_total();
 591     }
 592 }
 593
 594 MigrationInfo *qmp_query_migrate(Error **errp)
 595 {
 596     MigrationInfo *info = g_malloc0(sizeof(*info));
 597     MigrationState *s = migrate_get_current();
 598
 599     switch (s->state) {
 600     case MIGRATION_STATUS_NONE:
 601         /* no migration has happened ever */
 602         break;
 603     case MIGRATION_STATUS_SETUP:
 604         info->has_status = true;
 605         info->has_total_time = false;
 606         break;
 607     case MIGRATION_STATUS_ACTIVE:
 608     case MIGRATION_STATUS_CANCELLING:
 609     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
 610     case MIGRATION_STATUS_PRE_SWITCHOVER:
 611     case MIGRATION_STATUS_DEVICE:
 612          /* TODO add some postcopy stats */
 613         info->has_status = true;
 614         info->has_total_time = true;
 615         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
 616             - s->total_time;
 617         info->has_expected_downtime = true;
 618         info->expected_downtime = s->expected_downtime;
 619         info->has_setup_time = true;
 620         info->setup_time = s->setup_time;
 621
 622         populate_ram_info(info, s);
 623         populate_disk_info(info);
 624         break;
 625     case MIGRATION_STATUS_COLO:
 626         info->has_status = true;
 627         /* TODO: display COLO specific information (checkpoint info etc.) */
 628         break;
 629     case MIGRATION_STATUS_COMPLETED:
 630         info->has_status = true;
 631         info->has_total_time = true;
 632         info->total_time = s->total_time;
 633         info->has_downtime = true;
 634         info->downtime = s->downtime;
 635         info->has_setup_time = true;
 636         info->setup_time = s->setup_time;
 637
 638         populate_ram_info(info, s);
 639         break;
 640     case MIGRATION_STATUS_FAILED:
 641         info->has_status = true;
 642         if (s->error) {
 643             info->has_error_desc = true;
 644             info->error_desc = g_strdup(error_get_pretty(s->error));
 645         }
 646         break;
 647     case MIGRATION_STATUS_CANCELLED:
 648         info->has_status = true;
 649         break;
 650     }
 651     info->status = s->state;
 652
 653     return info;
 654 }
 655
 656 /**
 657  * @migration_caps_check - check capability validity
 658  *
 659  * @cap_list: old capability list, array of bool
 660  * @params: new capabilities to be applied soon
 661  * @errp: set *errp if the check failed, with reason
 662  *
 663  * Returns true if check passed, otherwise false.
 664  */
 665 static bool migrate_caps_check(bool *cap_list,
 666                                MigrationCapabilityStatusList *params,
 667                                Error **errp)
 668 {
 669     MigrationCapabilityStatusList *cap;
 670     bool old_postcopy_cap;
 671     MigrationIncomingState *mis = migration_incoming_get_current();
 672
 673     old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
 674
 675     for (cap = params; cap; cap = cap->next) {
 676         cap_list[cap->value->capability] = cap->value->state;
 677     }
 678
 679 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
 680     if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
 681         error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
 682                    "block migration");
 683         error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
 684         return false;
 685     }
 686 #endif
 687
 688     if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
 689         if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
 690             /* The decompression threads asynchronously write into RAM
 691              * rather than use the atomic copies needed to avoid
 692              * userfaulting.  It should be possible to fix the decompression
 693              * threads for compatibility in future.
 694              */
 695             error_setg(errp, "Postcopy is not currently compatible "
 696                        "with compression");
 697             return false;
 698         }
 699
 700         /* This check is reasonably expensive, so only when it's being
 701          * set the first time, also it's only the destination that needs
 702          * special support.
 703          */
 704         if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
 705             !postcopy_ram_supported_by_host(mis)) {
 706             /* postcopy_ram_supported_by_host will have emitted a more
 707              * detailed message
 708              */
 709             error_setg(errp, "Postcopy is not supported");
 710             return false;
 711         }
 712     }
 713
 714     return true;
 715 }
 716
 717 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
 718                                   Error **errp)
 719 {
 720     MigrationState *s = migrate_get_current();
 721     MigrationCapabilityStatusList *cap;
 722
 723     if (migration_is_setup_or_active(s->state)) {
 724         error_setg(errp, QERR_MIGRATION_ACTIVE);
 725         return;
 726     }
 727
 728     if (!migrate_caps_check(s->enabled_capabilities, params, errp)) {
 729         return;
 730     }
 731
 732     for (cap = params; cap; cap = cap->next) {
 733         s->enabled_capabilities[cap->value->capability] = cap->value->state;
 734     }
 735 }
 736
 737 /*
 738  * Check whether the parameters are valid. Error will be put into errp
 739  * (if provided). Return true if valid, otherwise false.
 740  */
 741 static bool migrate_params_check(MigrationParameters *params, Error **errp)
 742 {
 743     if (params->has_compress_level &&
 744         (params->compress_level < 0 || params->compress_level > 9)) {
 745         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
 746                    "is invalid, it should be in the range of 0 to 9");
 747         return false;
 748     }
 749
 750     if (params->has_compress_threads &&
 751         (params->compress_threads < 1 || params->compress_threads > 255)) {
 752         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 753                    "compress_threads",
 754                    "is invalid, it should be in the range of 1 to 255");
 755         return false;
 756     }
 757
 758     if (params->has_decompress_threads &&
 759         (params->decompress_threads < 1 || params->decompress_threads > 255)) {
 760         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 761                    "decompress_threads",
 762                    "is invalid, it should be in the range of 1 to 255");
 763         return false;
 764     }
 765
 766     if (params->has_cpu_throttle_initial &&
 767         (params->cpu_throttle_initial < 1 ||
 768          params->cpu_throttle_initial > 99)) {
 769         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 770                    "cpu_throttle_initial",
 771                    "an integer in the range of 1 to 99");
 772         return false;
 773     }
 774
 775     if (params->has_cpu_throttle_increment &&
 776         (params->cpu_throttle_increment < 1 ||
 777          params->cpu_throttle_increment > 99)) {
 778         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 779                    "cpu_throttle_increment",
 780                    "an integer in the range of 1 to 99");
 781         return false;
 782     }
 783
 784     if (params->has_max_bandwidth &&
 785         (params->max_bandwidth < 0 || params->max_bandwidth > SIZE_MAX)) {
 786         error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
 787                          " range of 0 to %zu bytes/second", SIZE_MAX);
 788         return false;
 789     }
 790
 791     if (params->has_downtime_limit &&
 792         (params->downtime_limit < 0 ||
 793          params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
 794         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
 795                          "the range of 0 to %d milliseconds",
 796                          MAX_MIGRATE_DOWNTIME);
 797         return false;
 798     }
 799
 800     if (params->has_x_checkpoint_delay && (params->x_checkpoint_delay < 0)) {
 801         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 802                     "x_checkpoint_delay",
 803                     "is invalid, it should be positive");
 804         return false;
 805     }
 806     if (params->has_x_multifd_channels &&
 807         (params->x_multifd_channels < 1 || params->x_multifd_channels > 255)) {
 808         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 809                    "multifd_channels",
 810                    "is invalid, it should be in the range of 1 to 255");
 811         return false;
 812     }
 813     if (params->has_x_multifd_page_count &&
 814             (params->x_multifd_page_count < 1 ||
 815              params->x_multifd_page_count > 10000)) {
 816         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 817                    "multifd_page_count",
 818                    "is invalid, it should be in the range of 1 to 10000");
 819         return false;
 820     }
 821
 822     if (params->has_xbzrle_cache_size &&
 823         (params->xbzrle_cache_size < qemu_target_page_size() ||
 824          !is_power_of_2(params->xbzrle_cache_size))) {
 825         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
 826                    "xbzrle_cache_size",
 827                    "is invalid, it should be bigger than target page size"
 828                    " and a power of two");
 829         return false;
 830     }
 831
 832     return true;
 833 }
 834
 835 static void migrate_params_test_apply(MigrateSetParameters *params,
 836                                       MigrationParameters *dest)
 837 {
 838     *dest = migrate_get_current()->parameters;
 839
 840     /* TODO use QAPI_CLONE() instead of duplicating it inline */
 841
 842     if (params->has_compress_level) {
 843         dest->compress_level = params->compress_level;
 844     }
 845
 846     if (params->has_compress_threads) {
 847         dest->compress_threads = params->compress_threads;
 848     }
 849
 850     if (params->has_decompress_threads) {
 851         dest->decompress_threads = params->decompress_threads;
 852     }
 853
 854     if (params->has_cpu_throttle_initial) {
 855         dest->cpu_throttle_initial = params->cpu_throttle_initial;
 856     }
 857
 858     if (params->has_cpu_throttle_increment) {
 859         dest->cpu_throttle_increment = params->cpu_throttle_increment;
 860     }
 861
 862     if (params->has_tls_creds) {
 863         assert(params->tls_creds->type == QTYPE_QSTRING);
 864         dest->tls_creds = g_strdup(params->tls_creds->u.s);
 865     }
 866
 867     if (params->has_tls_hostname) {
 868         assert(params->tls_hostname->type == QTYPE_QSTRING);
 869         dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
 870     }
 871
 872     if (params->has_max_bandwidth) {
 873         dest->max_bandwidth = params->max_bandwidth;
 874     }
 875
 876     if (params->has_downtime_limit) {
 877         dest->downtime_limit = params->downtime_limit;
 878     }
 879
 880     if (params->has_x_checkpoint_delay) {
 881         dest->x_checkpoint_delay = params->x_checkpoint_delay;
 882     }
 883
 884     if (params->has_block_incremental) {
 885         dest->block_incremental = params->block_incremental;
 886     }
 887     if (params->has_x_multifd_channels) {
 888         dest->x_multifd_channels = params->x_multifd_channels;
 889     }
 890     if (params->has_x_multifd_page_count) {
 891         dest->x_multifd_page_count = params->x_multifd_page_count;
 892     }
 893     if (params->has_xbzrle_cache_size) {
 894         dest->xbzrle_cache_size = params->xbzrle_cache_size;
 895     }
 896 }
 897
 898 static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
 899 {
 900     MigrationState *s = migrate_get_current();
 901
 902     /* TODO use QAPI_CLONE() instead of duplicating it inline */
 903
 904     if (params->has_compress_level) {
 905         s->parameters.compress_level = params->compress_level;
 906     }
 907
 908     if (params->has_compress_threads) {
 909         s->parameters.compress_threads = params->compress_threads;
 910     }
 911
 912     if (params->has_decompress_threads) {
 913         s->parameters.decompress_threads = params->decompress_threads;
 914     }
 915
 916     if (params->has_cpu_throttle_initial) {
 917         s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
 918     }
 919
 920     if (params->has_cpu_throttle_increment) {
 921         s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
 922     }
 923
 924     if (params->has_tls_creds) {
 925         g_free(s->parameters.tls_creds);
 926         assert(params->tls_creds->type == QTYPE_QSTRING);
 927         s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
 928     }
 929
 930     if (params->has_tls_hostname) {
 931         g_free(s->parameters.tls_hostname);
 932         assert(params->tls_hostname->type == QTYPE_QSTRING);
 933         s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
 934     }
 935
 936     if (params->has_max_bandwidth) {
 937         s->parameters.max_bandwidth = params->max_bandwidth;
 938         if (s->to_dst_file) {
 939             qemu_file_set_rate_limit(s->to_dst_file,
 940                                 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
 941         }
 942     }
 943
 944     if (params->has_downtime_limit) {
 945         s->parameters.downtime_limit = params->downtime_limit;
 946     }
 947
 948     if (params->has_x_checkpoint_delay) {
 949         s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
 950         if (migration_in_colo_state()) {
 951             colo_checkpoint_notify(s);
 952         }
 953     }
 954
 955     if (params->has_block_incremental) {
 956         s->parameters.block_incremental = params->block_incremental;
 957     }
 958     if (params->has_x_multifd_channels) {
 959         s->parameters.x_multifd_channels = params->x_multifd_channels;
 960     }
 961     if (params->has_x_multifd_page_count) {
 962         s->parameters.x_multifd_page_count = params->x_multifd_page_count;
 963     }
 964     if (params->has_xbzrle_cache_size) {
 965         s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
 966         xbzrle_cache_resize(params->xbzrle_cache_size, errp);
 967     }
 968 }
 969
 970 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
 971 {
 972     MigrationParameters tmp;
 973
 974     /* TODO Rewrite "" to null instead */
 975     if (params->has_tls_creds
 976         && params->tls_creds->type == QTYPE_QNULL) {
 977         QDECREF(params->tls_creds->u.n);
 978         params->tls_creds->type = QTYPE_QSTRING;
 979         params->tls_creds->u.s = strdup("");
 980     }
 981     /* TODO Rewrite "" to null instead */
 982     if (params->has_tls_hostname
 983         && params->tls_hostname->type == QTYPE_QNULL) {
 984         QDECREF(params->tls_hostname->u.n);
 985         params->tls_hostname->type = QTYPE_QSTRING;
 986         params->tls_hostname->u.s = strdup("");
 987     }
 988
 989     migrate_params_test_apply(params, &tmp);
 990
 991     if (!migrate_params_check(&tmp, errp)) {
 992         /* Invalid parameter */
 993         return;
 994     }
 995
 996     migrate_params_apply(params, errp);
 997 }
 998
 999
1000 void qmp_migrate_start_postcopy(Error **errp)
1001 {
1002     MigrationState *s = migrate_get_current();
1003
1004     if (!migrate_postcopy_ram()) {
1005         error_setg(errp, "Enable postcopy with migrate_set_capability before"
1006                          " the start of migration");
1007         return;
1008     }
1009
1010     if (s->state == MIGRATION_STATUS_NONE) {
1011         error_setg(errp, "Postcopy must be started after migration has been"
1012                          " started");
1013         return;
1014     }
1015     /*
1016      * we don't error if migration has finished since that would be racy
1017      * with issuing this command.
1018      */
1019     atomic_set(&s->start_postcopy, true);
1020 }
1021
1022 /* shared migration helpers */
1023
1024 void migrate_set_state(int *state, int old_state, int new_state)
1025 {
1026     assert(new_state < MIGRATION_STATUS__MAX);
1027     if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
1028         trace_migrate_set_state(MigrationStatus_str(new_state));
1029         migrate_generate_event(new_state);
1030     }
1031 }
1032
1033 static MigrationCapabilityStatusList *migrate_cap_add(
1034     MigrationCapabilityStatusList *list,
1035     MigrationCapability index,
1036     bool state)
1037 {
1038     MigrationCapabilityStatusList *cap;
1039
1040     cap = g_new0(MigrationCapabilityStatusList, 1);
1041     cap->value = g_new0(MigrationCapabilityStatus, 1);
1042     cap->value->capability = index;
1043     cap->value->state = state;
1044     cap->next = list;
1045
1046     return cap;
1047 }
1048
1049 void migrate_set_block_enabled(bool value, Error **errp)
1050 {
1051     MigrationCapabilityStatusList *cap;
1052
1053     cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
1054     qmp_migrate_set_capabilities(cap, errp);
1055     qapi_free_MigrationCapabilityStatusList(cap);
1056 }
1057
1058 static void migrate_set_block_incremental(MigrationState *s, bool value)
1059 {
1060     s->parameters.block_incremental = value;
1061 }
1062
1063 static void block_cleanup_parameters(MigrationState *s)
1064 {
1065     if (s->must_remove_block_options) {
1066         /* setting to false can never fail */
1067         migrate_set_block_enabled(false, &error_abort);
1068         migrate_set_block_incremental(s, false);
1069         s->must_remove_block_options = false;
1070     }
1071 }
1072
1073 static void migrate_fd_cleanup(void *opaque)
1074 {
1075     MigrationState *s = opaque;
1076
1077     qemu_bh_delete(s->cleanup_bh);
1078     s->cleanup_bh = NULL;
1079
1080     if (s->to_dst_file) {
1081         Error *local_err = NULL;
1082
1083         trace_migrate_fd_cleanup();
1084         qemu_mutex_unlock_iothread();
1085         if (s->migration_thread_running) {
1086             qemu_thread_join(&s->thread);
1087             s->migration_thread_running = false;
1088         }
1089         qemu_mutex_lock_iothread();
1090
1091         if (multifd_save_cleanup(&local_err) != 0) {
1092             error_report_err(local_err);
1093         }
1094         qemu_fclose(s->to_dst_file);
1095         s->to_dst_file = NULL;
1096     }
1097
1098     assert((s->state != MIGRATION_STATUS_ACTIVE) &&
1099            (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
1100
1101     if (s->state == MIGRATION_STATUS_CANCELLING) {
1102         migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1103                           MIGRATION_STATUS_CANCELLED);
1104     }
1105
1106     if (s->error) {
1107         /* It is used on info migrate.  We can't free it */
1108         error_report_err(error_copy(s->error));
1109     }
1110     notifier_list_notify(&migration_state_notifiers, s);
1111     block_cleanup_parameters(s);
1112 }
1113
1114 void migrate_set_error(MigrationState *s, const Error *error)
1115 {
1116     qemu_mutex_lock(&s->error_mutex);
1117     if (!s->error) {
1118         s->error = error_copy(error);
1119     }
1120     qemu_mutex_unlock(&s->error_mutex);
1121 }
1122
1123 void migrate_fd_error(MigrationState *s, const Error *error)
1124 {
1125     trace_migrate_fd_error(error_get_pretty(error));
1126     assert(s->to_dst_file == NULL);
1127     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1128                       MIGRATION_STATUS_FAILED);
1129     migrate_set_error(s, error);
1130     notifier_list_notify(&migration_state_notifiers, s);
1131     block_cleanup_parameters(s);
1132 }
1133
1134 static void migrate_fd_cancel(MigrationState *s)
1135 {
1136     int old_state ;
1137     QEMUFile *f = migrate_get_current()->to_dst_file;
1138     trace_migrate_fd_cancel();
1139
1140     if (s->rp_state.from_dst_file) {
1141         /* shutdown the rp socket, so causing the rp thread to shutdown */
1142         qemu_file_shutdown(s->rp_state.from_dst_file);
1143     }
1144
1145     do {
1146         old_state = s->state;
1147         if (!migration_is_setup_or_active(old_state)) {
1148             break;
1149         }
1150         /* If the migration is paused, kick it out of the pause */
1151         if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1152             qemu_sem_post(&s->pause_sem);
1153         }
1154         migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1155     } while (s->state != MIGRATION_STATUS_CANCELLING);
1156
1157     /*
1158      * If we're unlucky the migration code might be stuck somewhere in a
1159      * send/write while the network has failed and is waiting to timeout;
1160      * if we've got shutdown(2) available then we can force it to quit.
1161      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1162      * called in a bh, so there is no race against this cancel.
1163      */
1164     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1165         qemu_file_shutdown(f);
1166     }
1167     if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1168         Error *local_err = NULL;
1169
1170         bdrv_invalidate_cache_all(&local_err);
1171         if (local_err) {
1172             error_report_err(local_err);
1173         } else {
1174             s->block_inactive = false;
1175         }
1176     }
1177     block_cleanup_parameters(s);
1178 }
1179
1180 void add_migration_state_change_notifier(Notifier *notify)
1181 {
1182     notifier_list_add(&migration_state_notifiers, notify);
1183 }
1184
1185 void remove_migration_state_change_notifier(Notifier *notify)
1186 {
1187     notifier_remove(notify);
1188 }
1189
1190 bool migration_in_setup(MigrationState *s)
1191 {
1192     return s->state == MIGRATION_STATUS_SETUP;
1193 }
1194
1195 bool migration_has_finished(MigrationState *s)
1196 {
1197     return s->state == MIGRATION_STATUS_COMPLETED;
1198 }
1199
1200 bool migration_has_failed(MigrationState *s)
1201 {
1202     return (s->state == MIGRATION_STATUS_CANCELLED ||
1203             s->state == MIGRATION_STATUS_FAILED);
1204 }
1205
1206 bool migration_in_postcopy(void)
1207 {
1208     MigrationState *s = migrate_get_current();
1209
1210     return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1211 }
1212
1213 bool migration_in_postcopy_after_devices(MigrationState *s)
1214 {
1215     return migration_in_postcopy() && s->postcopy_after_devices;
1216 }
1217
1218 bool migration_is_idle(void)
1219 {
1220     MigrationState *s = migrate_get_current();
1221
1222     switch (s->state) {
1223     case MIGRATION_STATUS_NONE:
1224     case MIGRATION_STATUS_CANCELLED:
1225     case MIGRATION_STATUS_COMPLETED:
1226     case MIGRATION_STATUS_FAILED:
1227         return true;
1228     case MIGRATION_STATUS_SETUP:
1229     case MIGRATION_STATUS_CANCELLING:
1230     case MIGRATION_STATUS_ACTIVE:
1231     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1232     case MIGRATION_STATUS_COLO:
1233     case MIGRATION_STATUS_PRE_SWITCHOVER:
1234     case MIGRATION_STATUS_DEVICE:
1235         return false;
1236     case MIGRATION_STATUS__MAX:
1237         g_assert_not_reached();
1238     }
1239
1240     return false;
1241 }
1242
1243 MigrationState *migrate_init(void)
1244 {
1245     MigrationState *s = migrate_get_current();
1246
1247     /*
1248      * Reinitialise all migration state, except
1249      * parameters/capabilities that the user set, and
1250      * locks.
1251      */
1252     s->bytes_xfer = 0;
1253     s->xfer_limit = 0;
1254     s->cleanup_bh = 0;
1255     s->to_dst_file = NULL;
1256     s->state = MIGRATION_STATUS_NONE;
1257     s->rp_state.from_dst_file = NULL;
1258     s->rp_state.error = false;
1259     s->mbps = 0.0;
1260     s->downtime = 0;
1261     s->expected_downtime = 0;
1262     s->setup_time = 0;
1263     s->start_postcopy = false;
1264     s->postcopy_after_devices = false;
1265     s->migration_thread_running = false;
1266     error_free(s->error);
1267     s->error = NULL;
1268
1269     migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1270
1271     s->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1272     return s;
1273 }
1274
1275 static GSList *migration_blockers;
1276
1277 int migrate_add_blocker(Error *reason, Error **errp)
1278 {
1279     if (migrate_get_current()->only_migratable) {
1280         error_propagate(errp, error_copy(reason));
1281         error_prepend(errp, "disallowing migration blocker "
1282                           "(--only_migratable) for: ");
1283         return -EACCES;
1284     }
1285
1286     if (migration_is_idle()) {
1287         migration_blockers = g_slist_prepend(migration_blockers, reason);
1288         return 0;
1289     }
1290
1291     error_propagate(errp, error_copy(reason));
1292     error_prepend(errp, "disallowing migration blocker (migration in "
1293                       "progress) for: ");
1294     return -EBUSY;
1295 }
1296
1297 void migrate_del_blocker(Error *reason)
1298 {
1299     migration_blockers = g_slist_remove(migration_blockers, reason);
1300 }
1301
1302 void qmp_migrate_incoming(const char *uri, Error **errp)
1303 {
1304     Error *local_err = NULL;
1305     static bool once = true;
1306
1307     if (!deferred_incoming) {
1308         error_setg(errp, "For use with '-incoming defer'");
1309         return;
1310     }
1311     if (!once) {
1312         error_setg(errp, "The incoming migration has already been started");
1313     }
1314
1315     qemu_start_incoming_migration(uri, &local_err);
1316
1317     if (local_err) {
1318         error_propagate(errp, local_err);
1319         return;
1320     }
1321
1322     once = false;
1323 }
1324
1325 bool migration_is_blocked(Error **errp)
1326 {
1327     if (qemu_savevm_state_blocked(errp)) {
1328         return true;
1329     }
1330
1331     if (migration_blockers) {
1332         error_propagate(errp, error_copy(migration_blockers->data));
1333         return true;
1334     }
1335
1336     return false;
1337 }
1338
1339 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1340                  bool has_inc, bool inc, bool has_detach, bool detach,
1341                  Error **errp)
1342 {
1343     Error *local_err = NULL;
1344     MigrationState *s = migrate_get_current();
1345     const char *p;
1346
1347     if (migration_is_setup_or_active(s->state) ||
1348         s->state == MIGRATION_STATUS_CANCELLING ||
1349         s->state == MIGRATION_STATUS_COLO) {
1350         error_setg(errp, QERR_MIGRATION_ACTIVE);
1351         return;
1352     }
1353     if (runstate_check(RUN_STATE_INMIGRATE)) {
1354         error_setg(errp, "Guest is waiting for an incoming migration");
1355         return;
1356     }
1357
1358     if (migration_is_blocked(errp)) {
1359         return;
1360     }
1361
1362     if ((has_blk && blk) || (has_inc && inc)) {
1363         if (migrate_use_block() || migrate_use_block_incremental()) {
1364             error_setg(errp, "Command options are incompatible with "
1365                        "current migration capabilities");
1366             return;
1367         }
1368         migrate_set_block_enabled(true, &local_err);
1369         if (local_err) {
1370             error_propagate(errp, local_err);
1371             return;
1372         }
1373         s->must_remove_block_options = true;
1374     }
1375
1376     if (has_inc && inc) {
1377         migrate_set_block_incremental(s, true);
1378     }
1379
1380     s = migrate_init();
1381
1382     if (strstart(uri, "tcp:", &p)) {
1383         tcp_start_outgoing_migration(s, p, &local_err);
1384 #ifdef CONFIG_RDMA
1385     } else if (strstart(uri, "rdma:", &p)) {
1386         rdma_start_outgoing_migration(s, p, &local_err);
1387 #endif
1388     } else if (strstart(uri, "exec:", &p)) {
1389         exec_start_outgoing_migration(s, p, &local_err);
1390     } else if (strstart(uri, "unix:", &p)) {
1391         unix_start_outgoing_migration(s, p, &local_err);
1392     } else if (strstart(uri, "fd:", &p)) {
1393         fd_start_outgoing_migration(s, p, &local_err);
1394     } else {
1395         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1396                    "a valid migration protocol");
1397         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1398                           MIGRATION_STATUS_FAILED);
1399         return;
1400     }
1401
1402     if (local_err) {
1403         migrate_fd_error(s, local_err);
1404         error_propagate(errp, local_err);
1405         return;
1406     }
1407 }
1408
1409 void qmp_migrate_cancel(Error **errp)
1410 {
1411     migrate_fd_cancel(migrate_get_current());
1412 }
1413
1414 void qmp_migrate_continue(MigrationStatus state, Error **errp)
1415 {
1416     MigrationState *s = migrate_get_current();
1417     if (s->state != state) {
1418         error_setg(errp,  "Migration not in expected state: %s",
1419                    MigrationStatus_str(s->state));
1420         return;
1421     }
1422     qemu_sem_post(&s->pause_sem);
1423 }
1424
1425 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1426 {
1427     MigrateSetParameters p = {
1428         .has_xbzrle_cache_size = true,
1429         .xbzrle_cache_size = value,
1430     };
1431
1432     qmp_migrate_set_parameters(&p, errp);
1433 }
1434
1435 int64_t qmp_query_migrate_cache_size(Error **errp)
1436 {
1437     return migrate_xbzrle_cache_size();
1438 }
1439
1440 void qmp_migrate_set_speed(int64_t value, Error **errp)
1441 {
1442     MigrateSetParameters p = {
1443         .has_max_bandwidth = true,
1444         .max_bandwidth = value,
1445     };
1446
1447     qmp_migrate_set_parameters(&p, errp);
1448 }
1449
1450 void qmp_migrate_set_downtime(double value, Error **errp)
1451 {
1452     if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1453         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1454                          "the range of 0 to %d seconds",
1455                          MAX_MIGRATE_DOWNTIME_SECONDS);
1456         return;
1457     }
1458
1459     value *= 1000; /* Convert to milliseconds */
1460     value = MAX(0, MIN(INT64_MAX, value));
1461
1462     MigrateSetParameters p = {
1463         .has_downtime_limit = true,
1464         .downtime_limit = value,
1465     };
1466
1467     qmp_migrate_set_parameters(&p, errp);
1468 }
1469
1470 bool migrate_release_ram(void)
1471 {
1472     MigrationState *s;
1473
1474     s = migrate_get_current();
1475
1476     return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1477 }
1478
1479 bool migrate_postcopy_ram(void)
1480 {
1481     MigrationState *s;
1482
1483     s = migrate_get_current();
1484
1485     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1486 }
1487
1488 bool migrate_postcopy(void)
1489 {
1490     return migrate_postcopy_ram();
1491 }
1492
1493 bool migrate_auto_converge(void)
1494 {
1495     MigrationState *s;
1496
1497     s = migrate_get_current();
1498
1499     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
1500 }
1501
1502 bool migrate_zero_blocks(void)
1503 {
1504     MigrationState *s;
1505
1506     s = migrate_get_current();
1507
1508     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
1509 }
1510
1511 bool migrate_use_compression(void)
1512 {
1513     MigrationState *s;
1514
1515     s = migrate_get_current();
1516
1517     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
1518 }
1519
1520 int migrate_compress_level(void)
1521 {
1522     MigrationState *s;
1523
1524     s = migrate_get_current();
1525
1526     return s->parameters.compress_level;
1527 }
1528
1529 int migrate_compress_threads(void)
1530 {
1531     MigrationState *s;
1532
1533     s = migrate_get_current();
1534
1535     return s->parameters.compress_threads;
1536 }
1537
1538 int migrate_decompress_threads(void)
1539 {
1540     MigrationState *s;
1541
1542     s = migrate_get_current();
1543
1544     return s->parameters.decompress_threads;
1545 }
1546
1547 bool migrate_use_events(void)
1548 {
1549     MigrationState *s;
1550
1551     s = migrate_get_current();
1552
1553     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
1554 }
1555
1556 bool migrate_use_multifd(void)
1557 {
1558     MigrationState *s;
1559
1560     s = migrate_get_current();
1561
1562     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_MULTIFD];
1563 }
1564
1565 bool migrate_pause_before_switchover(void)
1566 {
1567     MigrationState *s;
1568
1569     s = migrate_get_current();
1570
1571     return s->enabled_capabilities[
1572         MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
1573 }
1574
1575 int migrate_multifd_channels(void)
1576 {
1577     MigrationState *s;
1578
1579     s = migrate_get_current();
1580
1581     return s->parameters.x_multifd_channels;
1582 }
1583
1584 int migrate_multifd_page_count(void)
1585 {
1586     MigrationState *s;
1587
1588     s = migrate_get_current();
1589
1590     return s->parameters.x_multifd_page_count;
1591 }
1592
1593 int migrate_use_xbzrle(void)
1594 {
1595     MigrationState *s;
1596
1597     s = migrate_get_current();
1598
1599     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
1600 }
1601
1602 int64_t migrate_xbzrle_cache_size(void)
1603 {
1604     MigrationState *s;
1605
1606     s = migrate_get_current();
1607
1608     return s->parameters.xbzrle_cache_size;
1609 }
1610
1611 bool migrate_use_block(void)
1612 {
1613     MigrationState *s;
1614
1615     s = migrate_get_current();
1616
1617     return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
1618 }
1619
1620 bool migrate_use_return_path(void)
1621 {
1622     MigrationState *s;
1623
1624     s = migrate_get_current();
1625
1626     return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
1627 }
1628
1629 bool migrate_use_block_incremental(void)
1630 {
1631     MigrationState *s;
1632
1633     s = migrate_get_current();
1634
1635     return s->parameters.block_incremental;
1636 }
1637
1638 /* migration thread support */
1639 /*
1640  * Something bad happened to the RP stream, mark an error
1641  * The caller shall print or trace something to indicate why
1642  */
1643 static void mark_source_rp_bad(MigrationState *s)
1644 {
1645     s->rp_state.error = true;
1646 }
1647
1648 static struct rp_cmd_args {
1649     ssize_t     len; /* -1 = variable */
1650     const char *name;
1651 } rp_cmd_args[] = {
1652     [MIG_RP_MSG_INVALID]        = { .len = -1, .name = "INVALID" },
1653     [MIG_RP_MSG_SHUT]           = { .len =  4, .name = "SHUT" },
1654     [MIG_RP_MSG_PONG]           = { .len =  4, .name = "PONG" },
1655     [MIG_RP_MSG_REQ_PAGES]      = { .len = 12, .name = "REQ_PAGES" },
1656     [MIG_RP_MSG_REQ_PAGES_ID]   = { .len = -1, .name = "REQ_PAGES_ID" },
1657     [MIG_RP_MSG_MAX]            = { .len = -1, .name = "MAX" },
1658 };
1659
1660 /*
1661  * Process a request for pages received on the return path,
1662  * We're allowed to send more than requested (e.g. to round to our page size)
1663  * and we don't need to send pages that have already been sent.
1664  */
1665 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
1666                                        ram_addr_t start, size_t len)
1667 {
1668     long our_host_ps = getpagesize();
1669
1670     trace_migrate_handle_rp_req_pages(rbname, start, len);
1671
1672     /*
1673      * Since we currently insist on matching page sizes, just sanity check
1674      * we're being asked for whole host pages.
1675      */
1676     if (start & (our_host_ps-1) ||
1677        (len & (our_host_ps-1))) {
1678         error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1679                      " len: %zd", __func__, start, len);
1680         mark_source_rp_bad(ms);
1681         return;
1682     }
1683
1684     if (ram_save_queue_pages(rbname, start, len)) {
1685         mark_source_rp_bad(ms);
1686     }
1687 }
1688
1689 /*
1690  * Handles messages sent on the return path towards the source VM
1691  *
1692  */
1693 static void *source_return_path_thread(void *opaque)
1694 {
1695     MigrationState *ms = opaque;
1696     QEMUFile *rp = ms->rp_state.from_dst_file;
1697     uint16_t header_len, header_type;
1698     uint8_t buf[512];
1699     uint32_t tmp32, sibling_error;
1700     ram_addr_t start = 0; /* =0 to silence warning */
1701     size_t  len = 0, expected_len;
1702     int res;
1703
1704     trace_source_return_path_thread_entry();
1705     while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
1706            migration_is_setup_or_active(ms->state)) {
1707         trace_source_return_path_thread_loop_top();
1708         header_type = qemu_get_be16(rp);
1709         header_len = qemu_get_be16(rp);
1710
1711         if (header_type >= MIG_RP_MSG_MAX ||
1712             header_type == MIG_RP_MSG_INVALID) {
1713             error_report("RP: Received invalid message 0x%04x length 0x%04x",
1714                     header_type, header_len);
1715             mark_source_rp_bad(ms);
1716             goto out;
1717         }
1718
1719         if ((rp_cmd_args[header_type].len != -1 &&
1720             header_len != rp_cmd_args[header_type].len) ||
1721             header_len > sizeof(buf)) {
1722             error_report("RP: Received '%s' message (0x%04x) with"
1723                     "incorrect length %d expecting %zu",
1724                     rp_cmd_args[header_type].name, header_type, header_len,
1725                     (size_t)rp_cmd_args[header_type].len);
1726             mark_source_rp_bad(ms);
1727             goto out;
1728         }
1729
1730         /* We know we've got a valid header by this point */
1731         res = qemu_get_buffer(rp, buf, header_len);
1732         if (res != header_len) {
1733             error_report("RP: Failed reading data for message 0x%04x"
1734                          " read %d expected %d",
1735                          header_type, res, header_len);
1736             mark_source_rp_bad(ms);
1737             goto out;
1738         }
1739
1740         /* OK, we have the message and the data */
1741         switch (header_type) {
1742         case MIG_RP_MSG_SHUT:
1743             sibling_error = ldl_be_p(buf);
1744             trace_source_return_path_thread_shut(sibling_error);
1745             if (sibling_error) {
1746                 error_report("RP: Sibling indicated error %d", sibling_error);
1747                 mark_source_rp_bad(ms);
1748             }
1749             /*
1750              * We'll let the main thread deal with closing the RP
1751              * we could do a shutdown(2) on it, but we're the only user
1752              * anyway, so there's nothing gained.
1753              */
1754             goto out;
1755
1756         case MIG_RP_MSG_PONG:
1757             tmp32 = ldl_be_p(buf);
1758             trace_source_return_path_thread_pong(tmp32);
1759             break;
1760
1761         case MIG_RP_MSG_REQ_PAGES:
1762             start = ldq_be_p(buf);
1763             len = ldl_be_p(buf + 8);
1764             migrate_handle_rp_req_pages(ms, NULL, start, len);
1765             break;
1766
1767         case MIG_RP_MSG_REQ_PAGES_ID:
1768             expected_len = 12 + 1; /* header + termination */
1769
1770             if (header_len >= expected_len) {
1771                 start = ldq_be_p(buf);
1772                 len = ldl_be_p(buf + 8);
1773                 /* Now we expect an idstr */
1774                 tmp32 = buf[12]; /* Length of the following idstr */
1775                 buf[13 + tmp32] = '\0';
1776                 expected_len += tmp32;
1777             }
1778             if (header_len != expected_len) {
1779                 error_report("RP: Req_Page_id with length %d expecting %zd",
1780                         header_len, expected_len);
1781                 mark_source_rp_bad(ms);
1782                 goto out;
1783             }
1784             migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
1785             break;
1786
1787         default:
1788             break;
1789         }
1790     }
1791     if (qemu_file_get_error(rp)) {
1792         trace_source_return_path_thread_bad_end();
1793         mark_source_rp_bad(ms);
1794     }
1795
1796     trace_source_return_path_thread_end();
1797 out:
1798     ms->rp_state.from_dst_file = NULL;
1799     qemu_fclose(rp);
1800     return NULL;
1801 }
1802
1803 static int open_return_path_on_source(MigrationState *ms)
1804 {
1805
1806     ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
1807     if (!ms->rp_state.from_dst_file) {
1808         return -1;
1809     }
1810
1811     trace_open_return_path_on_source();
1812     qemu_thread_create(&ms->rp_state.rp_thread, "return path",
1813                        source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
1814
1815     trace_open_return_path_on_source_continue();
1816
1817     return 0;
1818 }
1819
1820 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1821 static int await_return_path_close_on_source(MigrationState *ms)
1822 {
1823     /*
1824      * If this is a normal exit then the destination will send a SHUT and the
1825      * rp_thread will exit, however if there's an error we need to cause
1826      * it to exit.
1827      */
1828     if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
1829         /*
1830          * shutdown(2), if we have it, will cause it to unblock if it's stuck
1831          * waiting for the destination.
1832          */
1833         qemu_file_shutdown(ms->rp_state.from_dst_file);
1834         mark_source_rp_bad(ms);
1835     }
1836     trace_await_return_path_close_on_source_joining();
1837     qemu_thread_join(&ms->rp_state.rp_thread);
1838     trace_await_return_path_close_on_source_close();
1839     return ms->rp_state.error;
1840 }
1841
1842 /*
1843  * Switch from normal iteration to postcopy
1844  * Returns non-0 on error
1845  */
1846 static int postcopy_start(MigrationState *ms, bool *old_vm_running)
1847 {
1848     int ret;
1849     QIOChannelBuffer *bioc;
1850     QEMUFile *fb;
1851     int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1852     bool restart_block = false;
1853     int cur_state = MIGRATION_STATUS_ACTIVE;
1854     if (!migrate_pause_before_switchover()) {
1855         migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
1856                           MIGRATION_STATUS_POSTCOPY_ACTIVE);
1857     }
1858
1859     trace_postcopy_start();
1860     qemu_mutex_lock_iothread();
1861     trace_postcopy_start_set_run();
1862
1863     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
1864     *old_vm_running = runstate_is_running();
1865     global_state_store();
1866     ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
1867     if (ret < 0) {
1868         goto fail;
1869     }
1870
1871     ret = migration_maybe_pause(ms, &cur_state,
1872                                 MIGRATION_STATUS_POSTCOPY_ACTIVE);
1873     if (ret < 0) {
1874         goto fail;
1875     }
1876
1877     ret = bdrv_inactivate_all();
1878     if (ret < 0) {
1879         goto fail;
1880     }
1881     restart_block = true;
1882
1883     /*
1884      * Cause any non-postcopiable, but iterative devices to
1885      * send out their final data.
1886      */
1887     qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
1888
1889     /*
1890      * in Finish migrate and with the io-lock held everything should
1891      * be quiet, but we've potentially still got dirty pages and we
1892      * need to tell the destination to throw any pages it's already received
1893      * that are dirty
1894      */
1895     if (migrate_postcopy_ram()) {
1896         if (ram_postcopy_send_discard_bitmap(ms)) {
1897             error_report("postcopy send discard bitmap failed");
1898             goto fail;
1899         }
1900     }
1901
1902     /*
1903      * send rest of state - note things that are doing postcopy
1904      * will notice we're in POSTCOPY_ACTIVE and not actually
1905      * wrap their state up here
1906      */
1907     qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
1908     if (migrate_postcopy_ram()) {
1909         /* Ping just for debugging, helps line traces up */
1910         qemu_savevm_send_ping(ms->to_dst_file, 2);
1911     }
1912
1913     /*
1914      * While loading the device state we may trigger page transfer
1915      * requests and the fd must be free to process those, and thus
1916      * the destination must read the whole device state off the fd before
1917      * it starts processing it.  Unfortunately the ad-hoc migration format
1918      * doesn't allow the destination to know the size to read without fully
1919      * parsing it through each devices load-state code (especially the open
1920      * coded devices that use get/put).
1921      * So we wrap the device state up in a package with a length at the start;
1922      * to do this we use a qemu_buf to hold the whole of the device state.
1923      */
1924     bioc = qio_channel_buffer_new(4096);
1925     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
1926     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
1927     object_unref(OBJECT(bioc));
1928
1929     /*
1930      * Make sure the receiver can get incoming pages before we send the rest
1931      * of the state
1932      */
1933     qemu_savevm_send_postcopy_listen(fb);
1934
1935     qemu_savevm_state_complete_precopy(fb, false, false);
1936     if (migrate_postcopy_ram()) {
1937         qemu_savevm_send_ping(fb, 3);
1938     }
1939
1940     qemu_savevm_send_postcopy_run(fb);
1941
1942     /* <><> end of stuff going into the package */
1943
1944     /* Last point of recovery; as soon as we send the package the destination
1945      * can open devices and potentially start running.
1946      * Lets just check again we've not got any errors.
1947      */
1948     ret = qemu_file_get_error(ms->to_dst_file);
1949     if (ret) {
1950         error_report("postcopy_start: Migration stream errored (pre package)");
1951         goto fail_closefb;
1952     }
1953
1954     restart_block = false;
1955
1956     /* Now send that blob */
1957     if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
1958         goto fail_closefb;
1959     }
1960     qemu_fclose(fb);
1961
1962     /* Send a notify to give a chance for anything that needs to happen
1963      * at the transition to postcopy and after the device state; in particular
1964      * spice needs to trigger a transition now
1965      */
1966     ms->postcopy_after_devices = true;
1967     notifier_list_notify(&migration_state_notifiers, ms);
1968
1969     ms->downtime =  qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
1970
1971     qemu_mutex_unlock_iothread();
1972
1973     if (migrate_postcopy_ram()) {
1974         /*
1975          * Although this ping is just for debug, it could potentially be
1976          * used for getting a better measurement of downtime at the source.
1977          */
1978         qemu_savevm_send_ping(ms->to_dst_file, 4);
1979     }
1980
1981     if (migrate_release_ram()) {
1982         ram_postcopy_migrated_memory_release(ms);
1983     }
1984
1985     ret = qemu_file_get_error(ms->to_dst_file);
1986     if (ret) {
1987         error_report("postcopy_start: Migration stream errored");
1988         migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1989                               MIGRATION_STATUS_FAILED);
1990     }
1991
1992     return ret;
1993
1994 fail_closefb:
1995     qemu_fclose(fb);
1996 fail:
1997     migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
1998                           MIGRATION_STATUS_FAILED);
1999     if (restart_block) {
2000         /* A failure happened early enough that we know the destination hasn't
2001          * accessed block devices, so we're safe to recover.
2002          */
2003         Error *local_err = NULL;
2004
2005         bdrv_invalidate_cache_all(&local_err);
2006         if (local_err) {
2007             error_report_err(local_err);
2008         }
2009     }
2010     qemu_mutex_unlock_iothread();
2011     return -1;
2012 }
2013
2014 /**
2015  * migration_maybe_pause: Pause if required to by
2016  * migrate_pause_before_switchover called with the iothread locked
2017  * Returns: 0 on success
2018  */
2019 static int migration_maybe_pause(MigrationState *s,
2020                                  int *current_active_state,
2021                                  int new_state)
2022 {
2023     if (!migrate_pause_before_switchover()) {
2024         return 0;
2025     }
2026
2027     /* Since leaving this state is not atomic with posting the semaphore
2028      * it's possible that someone could have issued multiple migrate_continue
2029      * and the semaphore is incorrectly positive at this point;
2030      * the docs say it's undefined to reinit a semaphore that's already
2031      * init'd, so use timedwait to eat up any existing posts.
2032      */
2033     while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2034         /* This block intentionally left blank */
2035     }
2036
2037     qemu_mutex_unlock_iothread();
2038     migrate_set_state(&s->state, *current_active_state,
2039                       MIGRATION_STATUS_PRE_SWITCHOVER);
2040     qemu_sem_wait(&s->pause_sem);
2041     migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
2042                       new_state);
2043     *current_active_state = new_state;
2044     qemu_mutex_lock_iothread();
2045
2046     return s->state == new_state ? 0 : -EINVAL;
2047 }
2048
2049 /**
2050  * migration_completion: Used by migration_thread when there's not much left.
2051  *   The caller 'breaks' the loop when this returns.
2052  *
2053  * @s: Current migration state
2054  * @current_active_state: The migration state we expect to be in
2055  * @*old_vm_running: Pointer to old_vm_running flag
2056  * @*start_time: Pointer to time to update
2057  */
2058 static void migration_completion(MigrationState *s, int current_active_state,
2059                                  bool *old_vm_running,
2060                                  int64_t *start_time)
2061 {
2062     int ret;
2063
2064     if (s->state == MIGRATION_STATUS_ACTIVE) {
2065         qemu_mutex_lock_iothread();
2066         *start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2067         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER);
2068         *old_vm_running = runstate_is_running();
2069         ret = global_state_store();
2070
2071         if (!ret) {
2072             bool inactivate = !migrate_colo_enabled();
2073             ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2074             if (ret >= 0) {
2075                 ret = migration_maybe_pause(s, &current_active_state,
2076                                             MIGRATION_STATUS_DEVICE);
2077             }
2078             if (ret >= 0) {
2079                 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
2080                 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2081                                                          inactivate);
2082             }
2083             if (inactivate && ret >= 0) {
2084                 s->block_inactive = true;
2085             }
2086         }
2087         qemu_mutex_unlock_iothread();
2088
2089         if (ret < 0) {
2090             goto fail;
2091         }
2092     } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2093         trace_migration_completion_postcopy_end();
2094
2095         qemu_savevm_state_complete_postcopy(s->to_dst_file);
2096         trace_migration_completion_postcopy_end_after_complete();
2097     }
2098
2099     /*
2100      * If rp was opened we must clean up the thread before
2101      * cleaning everything else up (since if there are no failures
2102      * it will wait for the destination to send it's status in
2103      * a SHUT command).
2104      */
2105     if (s->rp_state.from_dst_file) {
2106         int rp_error;
2107         trace_migration_return_path_end_before();
2108         rp_error = await_return_path_close_on_source(s);
2109         trace_migration_return_path_end_after(rp_error);
2110         if (rp_error) {
2111             goto fail_invalidate;
2112         }
2113     }
2114
2115     if (qemu_file_get_error(s->to_dst_file)) {
2116         trace_migration_completion_file_err();
2117         goto fail_invalidate;
2118     }
2119
2120     if (!migrate_colo_enabled()) {
2121         migrate_set_state(&s->state, current_active_state,
2122                           MIGRATION_STATUS_COMPLETED);
2123     }
2124
2125     return;
2126
2127 fail_invalidate:
2128     /* If not doing postcopy, vm_start() will be called: let's regain
2129      * control on images.
2130      */
2131     if (s->state == MIGRATION_STATUS_ACTIVE) {
2132         Error *local_err = NULL;
2133
2134         qemu_mutex_lock_iothread();
2135         bdrv_invalidate_cache_all(&local_err);
2136         if (local_err) {
2137             error_report_err(local_err);
2138         } else {
2139             s->block_inactive = false;
2140         }
2141         qemu_mutex_unlock_iothread();
2142     }
2143
2144 fail:
2145     migrate_set_state(&s->state, current_active_state,
2146                       MIGRATION_STATUS_FAILED);
2147 }
2148
2149 bool migrate_colo_enabled(void)
2150 {
2151     MigrationState *s = migrate_get_current();
2152     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2153 }
2154
2155 /*
2156  * Master migration thread on the source VM.
2157  * It drives the migration and pumps the data down the outgoing channel.
2158  */
2159 static void *migration_thread(void *opaque)
2160 {
2161     MigrationState *s = opaque;
2162     /* Used by the bandwidth calcs, updated later */
2163     int64_t initial_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2164     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
2165     int64_t initial_bytes = 0;
2166     /*
2167      * The final stage happens when the remaining data is smaller than
2168      * this threshold; it's calculated from the requested downtime and
2169      * measured bandwidth
2170      */
2171     int64_t threshold_size = 0;
2172     int64_t start_time = initial_time;
2173     int64_t end_time;
2174     bool old_vm_running = false;
2175     bool entered_postcopy = false;
2176     /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
2177     enum MigrationStatus current_active_state = MIGRATION_STATUS_ACTIVE;
2178     bool enable_colo = migrate_colo_enabled();
2179
2180     rcu_register_thread();
2181
2182     qemu_savevm_state_header(s->to_dst_file);
2183
2184     /*
2185      * If we opened the return path, we need to make sure dst has it
2186      * opened as well.
2187      */
2188     if (s->rp_state.from_dst_file) {
2189         /* Now tell the dest that it should open its end so it can reply */
2190         qemu_savevm_send_open_return_path(s->to_dst_file);
2191
2192         /* And do a ping that will make stuff easier to debug */
2193         qemu_savevm_send_ping(s->to_dst_file, 1);
2194     }
2195
2196     if (migrate_postcopy()) {
2197         /*
2198          * Tell the destination that we *might* want to do postcopy later;
2199          * if the other end can't do postcopy it should fail now, nice and
2200          * early.
2201          */
2202         qemu_savevm_send_postcopy_advise(s->to_dst_file);
2203     }
2204
2205     qemu_savevm_state_setup(s->to_dst_file);
2206
2207     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
2208     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2209                       MIGRATION_STATUS_ACTIVE);
2210
2211     trace_migration_thread_setup_complete();
2212
2213     while (s->state == MIGRATION_STATUS_ACTIVE ||
2214            s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2215         int64_t current_time;
2216         uint64_t pending_size;
2217
2218         if (!qemu_file_rate_limit(s->to_dst_file)) {
2219             uint64_t pend_post, pend_nonpost;
2220
2221             qemu_savevm_state_pending(s->to_dst_file, threshold_size,
2222                                       &pend_nonpost, &pend_post);
2223             pending_size = pend_nonpost + pend_post;
2224             trace_migrate_pending(pending_size, threshold_size,
2225                                   pend_post, pend_nonpost);
2226             if (pending_size && pending_size >= threshold_size) {
2227                 /* Still a significant amount to transfer */
2228
2229                 if (migrate_postcopy() &&
2230                     s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE &&
2231                     pend_nonpost <= threshold_size &&
2232                     atomic_read(&s->start_postcopy)) {
2233
2234                     if (!postcopy_start(s, &old_vm_running)) {
2235                         current_active_state = MIGRATION_STATUS_POSTCOPY_ACTIVE;
2236                         entered_postcopy = true;
2237                     }
2238
2239                     continue;
2240                 }
2241                 /* Just another iteration step */
2242                 qemu_savevm_state_iterate(s->to_dst_file, entered_postcopy);
2243             } else {
2244                 trace_migration_thread_low_pending(pending_size);
2245                 migration_completion(s, current_active_state,
2246                                      &old_vm_running, &start_time);
2247                 break;
2248             }
2249         }
2250
2251         if (qemu_file_get_error(s->to_dst_file)) {
2252             migrate_set_state(&s->state, current_active_state,
2253                               MIGRATION_STATUS_FAILED);
2254             trace_migration_thread_file_err();
2255             break;
2256         }
2257         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2258         if (current_time >= initial_time + BUFFER_DELAY) {
2259             uint64_t transferred_bytes = qemu_ftell(s->to_dst_file) -
2260                                          initial_bytes;
2261             uint64_t time_spent = current_time - initial_time;
2262             double bandwidth = (double)transferred_bytes / time_spent;
2263             threshold_size = bandwidth * s->parameters.downtime_limit;
2264
2265             s->mbps = (((double) transferred_bytes * 8.0) /
2266                     ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
2267
2268             trace_migrate_transferred(transferred_bytes, time_spent,
2269                                       bandwidth, threshold_size);
2270             /* if we haven't sent anything, we don't want to recalculate
2271                10000 is a small enough number for our purposes */
2272             if (ram_counters.dirty_pages_rate && transferred_bytes > 10000) {
2273                 s->expected_downtime = ram_counters.dirty_pages_rate *
2274                     qemu_target_page_size() / bandwidth;
2275             }
2276
2277             qemu_file_reset_rate_limit(s->to_dst_file);
2278             initial_time = current_time;
2279             initial_bytes = qemu_ftell(s->to_dst_file);
2280         }
2281         if (qemu_file_rate_limit(s->to_dst_file)) {
2282             /* usleep expects microseconds */
2283             g_usleep((initial_time + BUFFER_DELAY - current_time)*1000);
2284         }
2285     }
2286
2287     trace_migration_thread_after_loop();
2288     /* If we enabled cpu throttling for auto-converge, turn it off. */
2289     cpu_throttle_stop();
2290     end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2291
2292     qemu_mutex_lock_iothread();
2293     /*
2294      * The resource has been allocated by migration will be reused in COLO
2295      * process, so don't release them.
2296      */
2297     if (!enable_colo) {
2298         qemu_savevm_state_cleanup();
2299     }
2300     if (s->state == MIGRATION_STATUS_COMPLETED) {
2301         uint64_t transferred_bytes = qemu_ftell(s->to_dst_file);
2302         s->total_time = end_time - s->total_time;
2303         if (!entered_postcopy) {
2304             s->downtime = end_time - start_time;
2305         }
2306         if (s->total_time) {
2307             s->mbps = (((double) transferred_bytes * 8.0) /
2308                        ((double) s->total_time)) / 1000;
2309         }
2310         runstate_set(RUN_STATE_POSTMIGRATE);
2311     } else {
2312         if (s->state == MIGRATION_STATUS_ACTIVE && enable_colo) {
2313             migrate_start_colo_process(s);
2314             qemu_savevm_state_cleanup();
2315             /*
2316             * Fixme: we will run VM in COLO no matter its old running state.
2317             * After exited COLO, we will keep running.
2318             */
2319             old_vm_running = true;
2320         }
2321         if (old_vm_running && !entered_postcopy) {
2322             vm_start();
2323         } else {
2324             if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
2325                 runstate_set(RUN_STATE_POSTMIGRATE);
2326             }
2327         }
2328     }
2329     qemu_bh_schedule(s->cleanup_bh);
2330     qemu_mutex_unlock_iothread();
2331
2332     rcu_unregister_thread();
2333     return NULL;
2334 }
2335
2336 void migrate_fd_connect(MigrationState *s)
2337 {
2338     s->expected_downtime = s->parameters.downtime_limit;
2339     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
2340
2341     qemu_file_set_blocking(s->to_dst_file, true);
2342     qemu_file_set_rate_limit(s->to_dst_file,
2343                              s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
2344
2345     /* Notify before starting migration thread */
2346     notifier_list_notify(&migration_state_notifiers, s);
2347
2348     /*
2349      * Open the return path. For postcopy, it is used exclusively. For
2350      * precopy, only if user specified "return-path" capability would
2351      * QEMU uses the return path.
2352      */
2353     if (migrate_postcopy_ram() || migrate_use_return_path()) {
2354         if (open_return_path_on_source(s)) {
2355             error_report("Unable to open return-path for postcopy");
2356             migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2357                               MIGRATION_STATUS_FAILED);
2358             migrate_fd_cleanup(s);
2359             return;
2360         }
2361     }
2362
2363     if (multifd_save_setup() != 0) {
2364         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2365                           MIGRATION_STATUS_FAILED);
2366         migrate_fd_cleanup(s);
2367         return;
2368     }
2369     qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
2370                        QEMU_THREAD_JOINABLE);
2371     s->migration_thread_running = true;
2372 }
2373
2374 void migration_global_dump(Monitor *mon)
2375 {
2376     MigrationState *ms = migrate_get_current();
2377
2378     monitor_printf(mon, "globals: store-global-state=%d, only_migratable=%d, "
2379                    "send-configuration=%d, send-section-footer=%d\n",
2380                    ms->store_global_state, ms->only_migratable,
2381                    ms->send_configuration, ms->send_section_footer);
2382 }
2383
2384 #define DEFINE_PROP_MIG_CAP(name, x)             \
2385     DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
2386
2387 static Property migration_properties[] = {
2388     DEFINE_PROP_BOOL("store-global-state", MigrationState,
2389                      store_global_state, true),
2390     DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false),
2391     DEFINE_PROP_BOOL("send-configuration", MigrationState,
2392                      send_configuration, true),
2393     DEFINE_PROP_BOOL("send-section-footer", MigrationState,
2394                      send_section_footer, true),
2395
2396     /* Migration parameters */
2397     DEFINE_PROP_INT64("x-compress-level", MigrationState,
2398                       parameters.compress_level,
2399                       DEFAULT_MIGRATE_COMPRESS_LEVEL),
2400     DEFINE_PROP_INT64("x-compress-threads", MigrationState,
2401                       parameters.compress_threads,
2402                       DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
2403     DEFINE_PROP_INT64("x-decompress-threads", MigrationState,
2404                       parameters.decompress_threads,
2405                       DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
2406     DEFINE_PROP_INT64("x-cpu-throttle-initial", MigrationState,
2407                       parameters.cpu_throttle_initial,
2408                       DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
2409     DEFINE_PROP_INT64("x-cpu-throttle-increment", MigrationState,
2410                       parameters.cpu_throttle_increment,
2411                       DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
2412     DEFINE_PROP_INT64("x-max-bandwidth", MigrationState,
2413                       parameters.max_bandwidth, MAX_THROTTLE),
2414     DEFINE_PROP_INT64("x-downtime-limit", MigrationState,
2415                       parameters.downtime_limit,
2416                       DEFAULT_MIGRATE_SET_DOWNTIME),
2417     DEFINE_PROP_INT64("x-checkpoint-delay", MigrationState,
2418                       parameters.x_checkpoint_delay,
2419                       DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
2420     DEFINE_PROP_INT64("x-multifd-channels", MigrationState,
2421                       parameters.x_multifd_channels,
2422                       DEFAULT_MIGRATE_MULTIFD_CHANNELS),
2423     DEFINE_PROP_INT64("x-multifd-page-count", MigrationState,
2424                       parameters.x_multifd_page_count,
2425                       DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT),
2426     DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
2427                       parameters.xbzrle_cache_size,
2428                       DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
2429
2430     /* Migration capabilities */
2431     DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
2432     DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
2433     DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
2434     DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
2435     DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
2436     DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
2437     DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
2438     DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
2439     DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
2440     DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
2441     DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
2442     DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD),
2443
2444     DEFINE_PROP_END_OF_LIST(),
2445 };
2446
2447 static void migration_class_init(ObjectClass *klass, void *data)
2448 {
2449     DeviceClass *dc = DEVICE_CLASS(klass);
2450
2451     dc->user_creatable = false;
2452     dc->props = migration_properties;
2453 }
2454
2455 static void migration_instance_finalize(Object *obj)
2456 {
2457     MigrationState *ms = MIGRATION_OBJ(obj);
2458     MigrationParameters *params = &ms->parameters;
2459
2460     qemu_mutex_destroy(&ms->error_mutex);
2461     g_free(params->tls_hostname);
2462     g_free(params->tls_creds);
2463     qemu_sem_destroy(&ms->pause_sem);
2464 }
2465
2466 static void migration_instance_init(Object *obj)
2467 {
2468     MigrationState *ms = MIGRATION_OBJ(obj);
2469     MigrationParameters *params = &ms->parameters;
2470
2471     ms->state = MIGRATION_STATUS_NONE;
2472     ms->mbps = -1;
2473     qemu_sem_init(&ms->pause_sem, 0);
2474     qemu_mutex_init(&ms->error_mutex);
2475
2476     params->tls_hostname = g_strdup("");
2477     params->tls_creds = g_strdup("");
2478
2479     /* Set has_* up only for parameter checks */
2480     params->has_compress_level = true;
2481     params->has_compress_threads = true;
2482     params->has_decompress_threads = true;
2483     params->has_cpu_throttle_initial = true;
2484     params->has_cpu_throttle_increment = true;
2485     params->has_max_bandwidth = true;
2486     params->has_downtime_limit = true;
2487     params->has_x_checkpoint_delay = true;
2488     params->has_block_incremental = true;
2489     params->has_x_multifd_channels = true;
2490     params->has_x_multifd_page_count = true;
2491     params->has_xbzrle_cache_size = true;
2492 }
2493
2494 /*
2495  * Return true if check pass, false otherwise. Error will be put
2496  * inside errp if provided.
2497  */
2498 static bool migration_object_check(MigrationState *ms, Error **errp)
2499 {
2500     MigrationCapabilityStatusList *head = NULL;
2501     /* Assuming all off */
2502     bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
2503     int i;
2504
2505     if (!migrate_params_check(&ms->parameters, errp)) {
2506         return false;
2507     }
2508
2509     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
2510         if (ms->enabled_capabilities[i]) {
2511             head = migrate_cap_add(head, i, true);
2512         }
2513     }
2514
2515     ret = migrate_caps_check(cap_list, head, errp);
2516
2517     /* It works with head == NULL */
2518     qapi_free_MigrationCapabilityStatusList(head);
2519
2520     return ret;
2521 }
2522
2523 static const TypeInfo migration_type = {
2524     .name = TYPE_MIGRATION,
2525     /*
2526      * NOTE: TYPE_MIGRATION is not really a device, as the object is
2527      * not created using qdev_create(), it is not attached to the qdev
2528      * device tree, and it is never realized.
2529      *
2530      * TODO: Make this TYPE_OBJECT once QOM provides something like
2531      * TYPE_DEVICE's "-global" properties.
2532      */
2533     .parent = TYPE_DEVICE,
2534     .class_init = migration_class_init,
2535     .class_size = sizeof(MigrationClass),
2536     .instance_size = sizeof(MigrationState),
2537     .instance_init = migration_instance_init,
2538     .instance_finalize = migration_instance_finalize,
2539 };
2540
2541 static void register_migration_types(void)
2542 {
2543     type_register_static(&migration_type);
2544 }
2545
2546 type_init(register_migration_types);