migration/migration.c

   1 /*
   2  * QEMU live migration
   3  *
   4  * Copyright IBM, Corp. 2008
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <[email protected]>
   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/error.h"
  34 #include "qapi/qapi-commands-migration.h"
  35 #include "qapi/qapi-events-migration.h"
  36 #include "qapi/qmp/qerror.h"
  37 #include "qapi/qmp/qnull.h"
  38 #include "qemu/rcu.h"
  39 #include "block.h"
  40 #include "postcopy-ram.h"
  41 #include "qemu/thread.h"
  42 #include "trace.h"
  43 #include "exec/target_page.h"
  44 #include "io/channel-buffer.h"
  45 #include "migration/colo.h"
  46 #include "hw/boards.h"
  47 #include "monitor/monitor.h"
  48 #include "net/announce.h"
  49
  50 #define MAX_THROTTLE  (32 << 20)      /* Migration transfer speed throttling */
  51
  52 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
  53  * data. */
  54 #define BUFFER_DELAY     100
  55 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
  56
  57 /* Time in milliseconds we are allowed to stop the source,
  58  * for sending the last part */
  59 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
  60
  61 /* Maximum migrate downtime set to 2000 seconds */
  62 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
  63 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
  64
  65 /* Default compression thread count */
  66 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
  67 /* Default decompression thread count, usually decompression is at
  68  * least 4 times as fast as compression.*/
  69 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
  70 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
  71 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
  72 /* Define default autoconverge cpu throttle migration parameters */
  73 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
  74 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
  75 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
  76
  77 /* Migration XBZRLE default cache size */
  78 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
  79
  80 /* The delay time (in ms) between two COLO checkpoints */
  81 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
  82 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
  83 #define DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT 16
  84
  85 /* Background transfer rate for postcopy, 0 means unlimited, note
  86  * that page requests can still exceed this limit.
  87  */
  88 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
  89
  90 /*
  91  * Parameters for self_announce_delay giving a stream of RARP/ARP
  92  * packets after migration.
  93  */
  94 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL  50
  95 #define DEFAULT_MIGRATE_ANNOUNCE_MAX     550
  96 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS    5
  97 #define DEFAULT_MIGRATE_ANNOUNCE_STEP    100
  98
  99 static NotifierList migration_state_notifiers =
 100     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
 101
 102 static bool deferred_incoming;
 103
 104 /* Messages sent on the return path from destination to source */
 105 enum mig_rp_message_type {
 106     MIG_RP_MSG_INVALID = 0,  /* Must be 0 */
 107     MIG_RP_MSG_SHUT,         /* sibling will not send any more RP messages */
 108     MIG_RP_MSG_PONG,         /* Response to a PING; data (seq: be32 ) */
 109
 110     MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
 111     MIG_RP_MSG_REQ_PAGES,    /* data (start: be64, len: be32) */
 112     MIG_RP_MSG_RECV_BITMAP,  /* send recved_bitmap back to source */
 113     MIG_RP_MSG_RESUME_ACK,   /* tell source that we are ready to resume */
 114
 115     MIG_RP_MSG_MAX
 116 };
 117
 118 /* When we add fault tolerance, we could have several
 119    migrations at once.  For now we don't need to add
 120    dynamic creation of migration */
 121
 122 static MigrationState *current_migration;
 123 static MigrationIncomingState *current_incoming;
 124
 125 static bool migration_object_check(MigrationState *ms, Error **errp);
 126 static int migration_maybe_pause(MigrationState *s,
 127                                  int *current_active_state,
 128                                  int new_state);
 129 static void migrate_fd_cancel(MigrationState *s);
 130
 131 void migration_object_init(void)
 132 {
 133     MachineState *ms = MACHINE(qdev_get_machine());
 134     Error *err = NULL;
 135
 136     /* This can only be called once. */
 137     assert(!current_migration);
 138     current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
 139
 140     /*
 141      * Init the migrate incoming object as well no matter whether
 142      * we'll use it or not.
 143      */
 144     assert(!current_incoming);
 145     current_incoming = g_new0(MigrationIncomingState, 1);
 146     current_incoming->state = MIGRATION_STATUS_NONE;
 147     current_incoming->postcopy_remote_fds =
 148         g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD));
 149     qemu_mutex_init(&current_incoming->rp_mutex);
 150     qemu_event_init(&current_incoming->main_thread_load_event, false);
 151     qemu_sem_init(&current_incoming->postcopy_pause_sem_dst, 0);
 152     qemu_sem_init(&current_incoming->postcopy_pause_sem_fault, 0);
 153
 154     init_dirty_bitmap_incoming_migration();
 155
 156     if (!migration_object_check(current_migration, &err)) {
 157         error_report_err(err);
 158         exit(1);
 159     }
 160
 161     /*
 162      * We cannot really do this in migration_instance_init() since at
 163      * that time global properties are not yet applied, then this
 164      * value will be definitely replaced by something else.
 165      */
 166     if (ms->enforce_config_section) {
 167         current_migration->send_configuration = true;
 168     }
 169 }
 170
 171 void migration_shutdown(void)
 172 {
 173     /*
 174      * Cancel the current migration - that will (eventually)
 175      * stop the migration using this structure
 176      */
 177     migrate_fd_cancel(current_migration);
 178     object_unref(OBJECT(current_migration));
 179 }
 180
 181 /* For outgoing */
 182 MigrationState *migrate_get_current(void)
 183 {
 184     /* This can only be called after the object created. */
 185     assert(current_migration);
 186     return current_migration;
 187 }
 188
 189 MigrationIncomingState *migration_incoming_get_current(void)
 190 {
 191     assert(current_incoming);
 192     return current_incoming;
 193 }
 194
 195 void migration_incoming_state_destroy(void)
 196 {
 197     struct MigrationIncomingState *mis = migration_incoming_get_current();
 198
 199     if (mis->to_src_file) {
 200         /* Tell source that we are done */
 201         migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
 202         qemu_fclose(mis->to_src_file);
 203         mis->to_src_file = NULL;
 204     }
 205
 206     if (mis->from_src_file) {
 207         qemu_fclose(mis->from_src_file);
 208         mis->from_src_file = NULL;
 209     }
 210     if (mis->postcopy_remote_fds) {
 211         g_array_free(mis->postcopy_remote_fds, TRUE);
 212         mis->postcopy_remote_fds = NULL;
 213     }
 214
 215     qemu_event_reset(&mis->main_thread_load_event);
 216 }
 217
 218 static void migrate_generate_event(int new_state)
 219 {
 220     if (migrate_use_events()) {
 221         qapi_event_send_migration(new_state);
 222     }
 223 }
 224
 225 static bool migrate_late_block_activate(void)
 226 {
 227     MigrationState *s;
 228
 229     s = migrate_get_current();
 230
 231     return s->enabled_capabilities[
 232         MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE];
 233 }
 234
 235 /*
 236  * Called on -incoming with a defer: uri.
 237  * The migration can be started later after any parameters have been
 238  * changed.
 239  */
 240 static void deferred_incoming_migration(Error **errp)
 241 {
 242     if (deferred_incoming) {
 243         error_setg(errp, "Incoming migration already deferred");
 244     }
 245     deferred_incoming = true;
 246 }
 247
 248 /*
 249  * Send a message on the return channel back to the source
 250  * of the migration.
 251  */
 252 static int migrate_send_rp_message(MigrationIncomingState *mis,
 253                                    enum mig_rp_message_type message_type,
 254                                    uint16_t len, void *data)
 255 {
 256     int ret = 0;
 257
 258     trace_migrate_send_rp_message((int)message_type, len);
 259     qemu_mutex_lock(&mis->rp_mutex);
 260
 261     /*
 262      * It's possible that the file handle got lost due to network
 263      * failures.
 264      */
 265     if (!mis->to_src_file) {
 266         ret = -EIO;
 267         goto error;
 268     }
 269
 270     qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
 271     qemu_put_be16(mis->to_src_file, len);
 272     qemu_put_buffer(mis->to_src_file, data, len);
 273     qemu_fflush(mis->to_src_file);
 274
 275     /* It's possible that qemu file got error during sending */
 276     ret = qemu_file_get_error(mis->to_src_file);
 277
 278 error:
 279     qemu_mutex_unlock(&mis->rp_mutex);
 280     return ret;
 281 }
 282
 283 /* Request a range of pages from the source VM at the given
 284  * start address.
 285  *   rbname: Name of the RAMBlock to request the page in, if NULL it's the same
 286  *           as the last request (a name must have been given previously)
 287  *   Start: Address offset within the RB
 288  *   Len: Length in bytes required - must be a multiple of pagesize
 289  */
 290 int migrate_send_rp_req_pages(MigrationIncomingState *mis, const char *rbname,
 291                               ram_addr_t start, size_t len)
 292 {
 293     uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
 294     size_t msglen = 12; /* start + len */
 295     enum mig_rp_message_type msg_type;
 296
 297     *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
 298     *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
 299
 300     if (rbname) {
 301         int rbname_len = strlen(rbname);
 302         assert(rbname_len < 256);
 303
 304         bufc[msglen++] = rbname_len;
 305         memcpy(bufc + msglen, rbname, rbname_len);
 306         msglen += rbname_len;
 307         msg_type = MIG_RP_MSG_REQ_PAGES_ID;
 308     } else {
 309         msg_type = MIG_RP_MSG_REQ_PAGES;
 310     }
 311
 312     return migrate_send_rp_message(mis, msg_type, msglen, bufc);
 313 }
 314
 315 static bool migration_colo_enabled;
 316 bool migration_incoming_colo_enabled(void)
 317 {
 318     return migration_colo_enabled;
 319 }
 320
 321 void migration_incoming_disable_colo(void)
 322 {
 323     migration_colo_enabled = false;
 324 }
 325
 326 void migration_incoming_enable_colo(void)
 327 {
 328     migration_colo_enabled = true;
 329 }
 330
 331 void qemu_start_incoming_migration(const char *uri, Error **errp)
 332 {
 333     const char *p;
 334
 335     qapi_event_send_migration(MIGRATION_STATUS_SETUP);
 336     if (!strcmp(uri, "defer")) {
 337         deferred_incoming_migration(errp);
 338     } else if (strstart(uri, "tcp:", &p)) {
 339         tcp_start_incoming_migration(p, errp);
 340 #ifdef CONFIG_RDMA
 341     } else if (strstart(uri, "rdma:", &p)) {
 342         rdma_start_incoming_migration(p, errp);
 343 #endif
 344     } else if (strstart(uri, "exec:", &p)) {
 345         exec_start_incoming_migration(p, errp);
 346     } else if (strstart(uri, "unix:", &p)) {
 347         unix_start_incoming_migration(p, errp);
 348     } else if (strstart(uri, "fd:", &p)) {
 349         fd_start_incoming_migration(p, errp);
 350     } else {
 351         error_setg(errp, "unknown migration protocol: %s", uri);
 352     }
 353 }
 354
 355 static void process_incoming_migration_bh(void *opaque)
 356 {
 357     Error *local_err = NULL;
 358     MigrationIncomingState *mis = opaque;
 359
 360     /* If capability late_block_activate is set:
 361      * Only fire up the block code now if we're going to restart the
 362      * VM, else 'cont' will do it.
 363      * This causes file locking to happen; so we don't want it to happen
 364      * unless we really are starting the VM.
 365      */
 366     if (!migrate_late_block_activate() ||
 367          (autostart && (!global_state_received() ||
 368             global_state_get_runstate() == RUN_STATE_RUNNING))) {
 369         /* Make sure all file formats flush their mutable metadata.
 370          * If we get an error here, just don't restart the VM yet. */
 371         bdrv_invalidate_cache_all(&local_err);
 372         if (local_err) {
 373             error_report_err(local_err);
 374             local_err = NULL;
 375             autostart = false;
 376         }
 377     }
 378
 379     /*
 380      * This must happen after all error conditions are dealt with and
 381      * we're sure the VM is going to be running on this host.
 382      */
 383     qemu_announce_self(&mis->announce_timer, migrate_announce_params());
 384
 385     if (multifd_load_cleanup(&local_err) != 0) {
 386         error_report_err(local_err);
 387         autostart = false;
 388     }
 389     /* If global state section was not received or we are in running
 390        state, we need to obey autostart. Any other state is set with
 391        runstate_set. */
 392
 393     dirty_bitmap_mig_before_vm_start();
 394
 395     if (!global_state_received() ||
 396         global_state_get_runstate() == RUN_STATE_RUNNING) {
 397         if (autostart) {
 398             vm_start();
 399         } else {
 400             runstate_set(RUN_STATE_PAUSED);
 401         }
 402     } else {
 403         runstate_set(global_state_get_runstate());
 404     }
 405     /*
 406      * This must happen after any state changes since as soon as an external
 407      * observer sees this event they might start to prod at the VM assuming
 408      * it's ready to use.
 409      */
 410     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
 411                       MIGRATION_STATUS_COMPLETED);
 412     qemu_bh_delete(mis->bh);
 413     migration_incoming_state_destroy();
 414 }
 415
 416 static void process_incoming_migration_co(void *opaque)
 417 {
 418     MigrationIncomingState *mis = migration_incoming_get_current();
 419     PostcopyState ps;
 420     int ret;
 421     Error *local_err = NULL;
 422
 423     assert(mis->from_src_file);
 424     mis->migration_incoming_co = qemu_coroutine_self();
 425     mis->largest_page_size = qemu_ram_pagesize_largest();
 426     postcopy_state_set(POSTCOPY_INCOMING_NONE);
 427     migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
 428                       MIGRATION_STATUS_ACTIVE);
 429     ret = qemu_loadvm_state(mis->from_src_file);
 430
 431     ps = postcopy_state_get();
 432     trace_process_incoming_migration_co_end(ret, ps);
 433     if (ps != POSTCOPY_INCOMING_NONE) {
 434         if (ps == POSTCOPY_INCOMING_ADVISE) {
 435             /*
 436              * Where a migration had postcopy enabled (and thus went to advise)
 437              * but managed to complete within the precopy period, we can use
 438              * the normal exit.
 439              */
 440             postcopy_ram_incoming_cleanup(mis);
 441         } else if (ret >= 0) {
 442             /*
 443              * Postcopy was started, cleanup should happen at the end of the
 444              * postcopy thread.
 445              */
 446             trace_process_incoming_migration_co_postcopy_end_main();
 447             return;
 448         }
 449         /* Else if something went wrong then just fall out of the normal exit */
 450     }
 451
 452     /* we get COLO info, and know if we are in COLO mode */
 453     if (!ret && migration_incoming_colo_enabled()) {
 454         /* Make sure all file formats flush their mutable metadata */
 455         bdrv_invalidate_cache_all(&local_err);
 456         if (local_err) {
 457             error_report_err(local_err);
 458             goto fail;
 459         }
 460
 461         if (colo_init_ram_cache() < 0) {
 462             error_report("Init ram cache failed");
 463             goto fail;
 464         }
 465
 466         qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
 467              colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
 468         mis->have_colo_incoming_thread = true;
 469         qemu_coroutine_yield();
 470
 471         /* Wait checkpoint incoming thread exit before free resource */
 472         qemu_thread_join(&mis->colo_incoming_thread);
 473         /* We hold the global iothread lock, so it is safe here */
 474         colo_release_ram_cache();
 475     }
 476
 477     if (ret < 0) {
 478         error_report("load of migration failed: %s", strerror(-ret));
 479         goto fail;
 480     }
 481     mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
 482     qemu_bh_schedule(mis->bh);
 483     mis->migration_incoming_co = NULL;
 484     return;
 485 fail:
 486     local_err = NULL;
 487     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
 488                       MIGRATION_STATUS_FAILED);
 489     qemu_fclose(mis->from_src_file);
 490     if (multifd_load_cleanup(&local_err) != 0) {
 491         error_report_err(local_err);
 492     }
 493     exit(EXIT_FAILURE);
 494 }
 495
 496 static void migration_incoming_setup(QEMUFile *f)
 497 {
 498     MigrationIncomingState *mis = migration_incoming_get_current();
 499
 500     if (multifd_load_setup() != 0) {
 501         /* We haven't been able to create multifd threads
 502            nothing better to do */
 503         exit(EXIT_FAILURE);
 504     }
 505
 506     if (!mis->from_src_file) {
 507         mis->from_src_file = f;
 508     }
 509     qemu_file_set_blocking(f, false);
 510 }
 511
 512 void migration_incoming_process(void)
 513 {
 514     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
 515     qemu_coroutine_enter(co);
 516 }
 517
 518 /* Returns true if recovered from a paused migration, otherwise false */
 519 static bool postcopy_try_recover(QEMUFile *f)
 520 {
 521     MigrationIncomingState *mis = migration_incoming_get_current();
 522
 523     if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
 524         /* Resumed from a paused postcopy migration */
 525
 526         mis->from_src_file = f;
 527         /* Postcopy has standalone thread to do vm load */
 528         qemu_file_set_blocking(f, true);
 529
 530         /* Re-configure the return path */
 531         mis->to_src_file = qemu_file_get_return_path(f);
 532
 533         migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
 534                           MIGRATION_STATUS_POSTCOPY_RECOVER);
 535
 536         /*
 537          * Here, we only wake up the main loading thread (while the
 538          * fault thread will still be waiting), so that we can receive
 539          * commands from source now, and answer it if needed. The
 540          * fault thread will be woken up afterwards until we are sure
 541          * that source is ready to reply to page requests.
 542          */
 543         qemu_sem_post(&mis->postcopy_pause_sem_dst);
 544         return true;
 545     }
 546
 547     return false;
 548 }
 549
 550 void migration_fd_process_incoming(QEMUFile *f)
 551 {
 552     if (postcopy_try_recover(f)) {
 553         return;
 554     }
 555
 556     migration_incoming_setup(f);
 557     migration_incoming_process();
 558 }
 559
 560 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp)
 561 {
 562     MigrationIncomingState *mis = migration_incoming_get_current();
 563     bool start_migration;
 564
 565     if (!mis->from_src_file) {
 566         /* The first connection (multifd may have multiple) */
 567         QEMUFile *f = qemu_fopen_channel_input(ioc);
 568
 569         /* If it's a recovery, we're done */
 570         if (postcopy_try_recover(f)) {
 571             return;
 572         }
 573
 574         migration_incoming_setup(f);
 575
 576         /*
 577          * Common migration only needs one channel, so we can start
 578          * right now.  Multifd needs more than one channel, we wait.
 579          */
 580         start_migration = !migrate_use_multifd();
 581     } else {
 582         Error *local_err = NULL;
 583         /* Multiple connections */
 584         assert(migrate_use_multifd());
 585         start_migration = multifd_recv_new_channel(ioc, &local_err);
 586         if (local_err) {
 587             error_propagate(errp, local_err);
 588             return;
 589         }
 590     }
 591
 592     if (start_migration) {
 593         migration_incoming_process();
 594     }
 595 }
 596
 597 /**
 598  * @migration_has_all_channels: We have received all channels that we need
 599  *
 600  * Returns true when we have got connections to all the channels that
 601  * we need for migration.
 602  */
 603 bool migration_has_all_channels(void)
 604 {
 605     MigrationIncomingState *mis = migration_incoming_get_current();
 606     bool all_channels;
 607
 608     all_channels = multifd_recv_all_channels_created();
 609
 610     return all_channels && mis->from_src_file != NULL;
 611 }
 612
 613 /*
 614  * Send a 'SHUT' message on the return channel with the given value
 615  * to indicate that we've finished with the RP.  Non-0 value indicates
 616  * error.
 617  */
 618 void migrate_send_rp_shut(MigrationIncomingState *mis,
 619                           uint32_t value)
 620 {
 621     uint32_t buf;
 622
 623     buf = cpu_to_be32(value);
 624     migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
 625 }
 626
 627 /*
 628  * Send a 'PONG' message on the return channel with the given value
 629  * (normally in response to a 'PING')
 630  */
 631 void migrate_send_rp_pong(MigrationIncomingState *mis,
 632                           uint32_t value)
 633 {
 634     uint32_t buf;
 635
 636     buf = cpu_to_be32(value);
 637     migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
 638 }
 639
 640 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
 641                                  char *block_name)
 642 {
 643     char buf[512];
 644     int len;
 645     int64_t res;
 646
 647     /*
 648      * First, we send the header part. It contains only the len of
 649      * idstr, and the idstr itself.
 650      */
 651     len = strlen(block_name);
 652     buf[0] = len;
 653     memcpy(buf + 1, block_name, len);
 654
 655     if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
 656         error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
 657                      __func__);
 658         return;
 659     }
 660
 661     migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf);
 662
 663     /*
 664      * Next, we dump the received bitmap to the stream.
 665      *
 666      * TODO: currently we are safe since we are the only one that is
 667      * using the to_src_file handle (fault thread is still paused),
 668      * and it's ok even not taking the mutex. However the best way is
 669      * to take the lock before sending the message header, and release
 670      * the lock after sending the bitmap.
 671      */
 672     qemu_mutex_lock(&mis->rp_mutex);
 673     res = ramblock_recv_bitmap_send(mis->to_src_file, block_name);
 674     qemu_mutex_unlock(&mis->rp_mutex);
 675
 676     trace_migrate_send_rp_recv_bitmap(block_name, res);
 677 }
 678
 679 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value)
 680 {
 681     uint32_t buf;
 682
 683     buf = cpu_to_be32(value);
 684     migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf);
 685 }
 686
 687 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
 688 {
 689     MigrationCapabilityStatusList *head = NULL;
 690     MigrationCapabilityStatusList *caps;
 691     MigrationState *s = migrate_get_current();
 692     int i;
 693
 694     caps = NULL; /* silence compiler warning */
 695     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
 696 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
 697         if (i == MIGRATION_CAPABILITY_BLOCK) {
 698             continue;
 699         }
 700 #endif
 701         if (head == NULL) {
 702             head = g_malloc0(sizeof(*caps));
 703             caps = head;
 704         } else {
 705             caps->next = g_malloc0(sizeof(*caps));
 706             caps = caps->next;
 707         }
 708         caps->value =
 709             g_malloc(sizeof(*caps->value));
 710         caps->value->capability = i;
 711         caps->value->state = s->enabled_capabilities[i];
 712     }
 713
 714     return head;
 715 }
 716
 717 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
 718 {
 719     MigrationParameters *params;
 720     MigrationState *s = migrate_get_current();
 721
 722     /* TODO use QAPI_CLONE() instead of duplicating it inline */
 723     params = g_malloc0(sizeof(*params));
 724     params->has_compress_level = true;
 725     params->compress_level = s->parameters.compress_level;
 726     params->has_compress_threads = true;
 727     params->compress_threads = s->parameters.compress_threads;
 728     params->has_compress_wait_thread = true;
 729     params->compress_wait_thread = s->parameters.compress_wait_thread;
 730     params->has_decompress_threads = true;
 731     params->decompress_threads = s->parameters.decompress_threads;
 732     params->has_cpu_throttle_initial = true;
 733     params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
 734     params->has_cpu_throttle_increment = true;
 735     params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
 736     params->has_tls_creds = true;
 737     params->tls_creds = g_strdup(s->parameters.tls_creds);
 738     params->has_tls_hostname = true;
 739     params->tls_hostname = g_strdup(s->parameters.tls_hostname);
 740     params->has_max_bandwidth = true;
 741     params->max_bandwidth = s->parameters.max_bandwidth;
 742     params->has_downtime_limit = true;
 743     params->downtime_limit = s->parameters.downtime_limit;
 744     params->has_x_checkpoint_delay = true;
 745     params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
 746     params->has_block_incremental = true;
 747     params->block_incremental = s->parameters.block_incremental;
 748     params->has_x_multifd_channels = true;
 749     params->x_multifd_channels = s->parameters.x_multifd_channels;
 750     params->has_x_multifd_page_count = true;
 751     params->x_multifd_page_count = s->parameters.x_multifd_page_count;
 752     params->has_xbzrle_cache_size = true;
 753     params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
 754     params->has_max_postcopy_bandwidth = true;
 755     params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth;
 756     params->has_max_cpu_throttle = true;
 757     params->max_cpu_throttle = s->parameters.max_cpu_throttle;
 758     params->has_announce_initial = true;
 759     params->announce_initial = s->parameters.announce_initial;
 760     params->has_announce_max = true;
 761     params->announce_max = s->parameters.announce_max;
 762     params->has_announce_rounds = true;
 763     params->announce_rounds = s->parameters.announce_rounds;
 764     params->has_announce_step = true;
 765     params->announce_step = s->parameters.announce_step;
 766
 767     return params;
 768 }
 769
 770 AnnounceParameters *migrate_announce_params(void)
 771 {
 772     static AnnounceParameters ap;
 773
 774     MigrationState *s = migrate_get_current();
 775
 776     ap.initial = s->parameters.announce_initial;
 777     ap.max = s->parameters.announce_max;
 778     ap.rounds = s->parameters.announce_rounds;
 779     ap.step = s->parameters.announce_step;
 780
 781     return &ap;
 782 }
 783
 784 /*
 785  * Return true if we're already in the middle of a migration
 786  * (i.e. any of the active or setup states)
 787  */
 788 bool migration_is_setup_or_active(int state)
 789 {
 790     switch (state) {
 791     case MIGRATION_STATUS_ACTIVE:
 792     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
 793     case MIGRATION_STATUS_POSTCOPY_PAUSED:
 794     case MIGRATION_STATUS_POSTCOPY_RECOVER:
 795     case MIGRATION_STATUS_SETUP:
 796     case MIGRATION_STATUS_PRE_SWITCHOVER:
 797     case MIGRATION_STATUS_DEVICE:
 798         return true;
 799
 800     default:
 801         return false;
 802
 803     }
 804 }
 805
 806 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
 807 {
 808     info->has_ram = true;
 809     info->ram = g_malloc0(sizeof(*info->ram));
 810     info->ram->transferred = ram_counters.transferred;
 811     info->ram->total = ram_bytes_total();
 812     info->ram->duplicate = ram_counters.duplicate;
 813     /* legacy value.  It is not used anymore */
 814     info->ram->skipped = 0;
 815     info->ram->normal = ram_counters.normal;
 816     info->ram->normal_bytes = ram_counters.normal *
 817         qemu_target_page_size();
 818     info->ram->mbps = s->mbps;
 819     info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
 820     info->ram->postcopy_requests = ram_counters.postcopy_requests;
 821     info->ram->page_size = qemu_target_page_size();
 822     info->ram->multifd_bytes = ram_counters.multifd_bytes;
 823     info->ram->pages_per_second = s->pages_per_second;
 824
 825     if (migrate_use_xbzrle()) {
 826         info->has_xbzrle_cache = true;
 827         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
 828         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
 829         info->xbzrle_cache->bytes = xbzrle_counters.bytes;
 830         info->xbzrle_cache->pages = xbzrle_counters.pages;
 831         info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
 832         info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
 833         info->xbzrle_cache->overflow = xbzrle_counters.overflow;
 834     }
 835
 836     if (migrate_use_compression()) {
 837         info->has_compression = true;
 838         info->compression = g_malloc0(sizeof(*info->compression));
 839         info->compression->pages = compression_counters.pages;
 840         info->compression->busy = compression_counters.busy;
 841         info->compression->busy_rate = compression_counters.busy_rate;
 842         info->compression->compressed_size =
 843                                     compression_counters.compressed_size;
 844         info->compression->compression_rate =
 845                                     compression_counters.compression_rate;
 846     }
 847
 848     if (cpu_throttle_active()) {
 849         info->has_cpu_throttle_percentage = true;
 850         info->cpu_throttle_percentage = cpu_throttle_get_percentage();
 851     }
 852
 853     if (s->state != MIGRATION_STATUS_COMPLETED) {
 854         info->ram->remaining = ram_bytes_remaining();
 855         info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
 856     }
 857 }
 858
 859 static void populate_disk_info(MigrationInfo *info)
 860 {
 861     if (blk_mig_active()) {
 862         info->has_disk = true;
 863         info->disk = g_malloc0(sizeof(*info->disk));
 864         info->disk->transferred = blk_mig_bytes_transferred();
 865         info->disk->remaining = blk_mig_bytes_remaining();
 866         info->disk->total = blk_mig_bytes_total();
 867     }
 868 }
 869
 870 static void fill_source_migration_info(MigrationInfo *info)
 871 {
 872     MigrationState *s = migrate_get_current();
 873
 874     switch (s->state) {
 875     case MIGRATION_STATUS_NONE:
 876         /* no migration has happened ever */
 877         /* do not overwrite destination migration status */
 878         return;
 879         break;
 880     case MIGRATION_STATUS_SETUP:
 881         info->has_status = true;
 882         info->has_total_time = false;
 883         break;
 884     case MIGRATION_STATUS_ACTIVE:
 885     case MIGRATION_STATUS_CANCELLING:
 886     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
 887     case MIGRATION_STATUS_PRE_SWITCHOVER:
 888     case MIGRATION_STATUS_DEVICE:
 889     case MIGRATION_STATUS_POSTCOPY_PAUSED:
 890     case MIGRATION_STATUS_POSTCOPY_RECOVER:
 891          /* TODO add some postcopy stats */
 892         info->has_status = true;
 893         info->has_total_time = true;
 894         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME)
 895             - s->start_time;
 896         info->has_expected_downtime = true;
 897         info->expected_downtime = s->expected_downtime;
 898         info->has_setup_time = true;
 899         info->setup_time = s->setup_time;
 900
 901         populate_ram_info(info, s);
 902         populate_disk_info(info);
 903         break;
 904     case MIGRATION_STATUS_COLO:
 905         info->has_status = true;
 906         /* TODO: display COLO specific information (checkpoint info etc.) */
 907         break;
 908     case MIGRATION_STATUS_COMPLETED:
 909         info->has_status = true;
 910         info->has_total_time = true;
 911         info->total_time = s->total_time;
 912         info->has_downtime = true;
 913         info->downtime = s->downtime;
 914         info->has_setup_time = true;
 915         info->setup_time = s->setup_time;
 916
 917         populate_ram_info(info, s);
 918         break;
 919     case MIGRATION_STATUS_FAILED:
 920         info->has_status = true;
 921         if (s->error) {
 922             info->has_error_desc = true;
 923             info->error_desc = g_strdup(error_get_pretty(s->error));
 924         }
 925         break;
 926     case MIGRATION_STATUS_CANCELLED:
 927         info->has_status = true;
 928         break;
 929     }
 930     info->status = s->state;
 931 }
 932
 933 /**
 934  * @migration_caps_check - check capability validity
 935  *
 936  * @cap_list: old capability list, array of bool
 937  * @params: new capabilities to be applied soon
 938  * @errp: set *errp if the check failed, with reason
 939  *
 940  * Returns true if check passed, otherwise false.
 941  */
 942 static bool migrate_caps_check(bool *cap_list,
 943                                MigrationCapabilityStatusList *params,
 944                                Error **errp)
 945 {
 946     MigrationCapabilityStatusList *cap;
 947     bool old_postcopy_cap;
 948     MigrationIncomingState *mis = migration_incoming_get_current();
 949
 950     old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
 951
 952     for (cap = params; cap; cap = cap->next) {
 953         cap_list[cap->value->capability] = cap->value->state;
 954     }
 955
 956 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
 957     if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
 958         error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
 959                    "block migration");
 960         error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
 961         return false;
 962     }
 963 #endif
 964
 965 #ifndef CONFIG_REPLICATION
 966     if (cap_list[MIGRATION_CAPABILITY_X_COLO]) {
 967         error_setg(errp, "QEMU compiled without replication module"
 968                    " can't enable COLO");
 969         error_append_hint(errp, "Please enable replication before COLO.\n");
 970         return false;
 971     }
 972 #endif
 973
 974     if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
 975         if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
 976             /* The decompression threads asynchronously write into RAM
 977              * rather than use the atomic copies needed to avoid
 978              * userfaulting.  It should be possible to fix the decompression
 979              * threads for compatibility in future.
 980              */
 981             error_setg(errp, "Postcopy is not currently compatible "
 982                        "with compression");
 983             return false;
 984         }
 985
 986         /* This check is reasonably expensive, so only when it's being
 987          * set the first time, also it's only the destination that needs
 988          * special support.
 989          */
 990         if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
 991             !postcopy_ram_supported_by_host(mis)) {
 992             /* postcopy_ram_supported_by_host will have emitted a more
 993              * detailed message
 994              */
 995             error_setg(errp, "Postcopy is not supported");
 996             return false;
 997         }
 998
 999         if (cap_list[MIGRATION_CAPABILITY_X_IGNORE_SHARED]) {
1000             error_setg(errp, "Postcopy is not compatible with ignore-shared");
1001             return false;
1002         }
1003     }
1004
1005     return true;
1006 }
1007
1008 static void fill_destination_migration_info(MigrationInfo *info)
1009 {
1010     MigrationIncomingState *mis = migration_incoming_get_current();
1011
1012     switch (mis->state) {
1013     case MIGRATION_STATUS_NONE:
1014         return;
1015         break;
1016     case MIGRATION_STATUS_SETUP:
1017     case MIGRATION_STATUS_CANCELLING:
1018     case MIGRATION_STATUS_CANCELLED:
1019     case MIGRATION_STATUS_ACTIVE:
1020     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1021     case MIGRATION_STATUS_POSTCOPY_PAUSED:
1022     case MIGRATION_STATUS_POSTCOPY_RECOVER:
1023     case MIGRATION_STATUS_FAILED:
1024     case MIGRATION_STATUS_COLO:
1025         info->has_status = true;
1026         break;
1027     case MIGRATION_STATUS_COMPLETED:
1028         info->has_status = true;
1029         fill_destination_postcopy_migration_info(info);
1030         break;
1031     }
1032     info->status = mis->state;
1033 }
1034
1035 MigrationInfo *qmp_query_migrate(Error **errp)
1036 {
1037     MigrationInfo *info = g_malloc0(sizeof(*info));
1038
1039     fill_destination_migration_info(info);
1040     fill_source_migration_info(info);
1041
1042     return info;
1043 }
1044
1045 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
1046                                   Error **errp)
1047 {
1048     MigrationState *s = migrate_get_current();
1049     MigrationCapabilityStatusList *cap;
1050     bool cap_list[MIGRATION_CAPABILITY__MAX];
1051
1052     if (migration_is_setup_or_active(s->state)) {
1053         error_setg(errp, QERR_MIGRATION_ACTIVE);
1054         return;
1055     }
1056
1057     memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list));
1058     if (!migrate_caps_check(cap_list, params, errp)) {
1059         return;
1060     }
1061
1062     for (cap = params; cap; cap = cap->next) {
1063         s->enabled_capabilities[cap->value->capability] = cap->value->state;
1064     }
1065 }
1066
1067 /*
1068  * Check whether the parameters are valid. Error will be put into errp
1069  * (if provided). Return true if valid, otherwise false.
1070  */
1071 static bool migrate_params_check(MigrationParameters *params, Error **errp)
1072 {
1073     if (params->has_compress_level &&
1074         (params->compress_level > 9)) {
1075         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
1076                    "is invalid, it should be in the range of 0 to 9");
1077         return false;
1078     }
1079
1080     if (params->has_compress_threads && (params->compress_threads < 1)) {
1081         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1082                    "compress_threads",
1083                    "is invalid, it should be in the range of 1 to 255");
1084         return false;
1085     }
1086
1087     if (params->has_decompress_threads && (params->decompress_threads < 1)) {
1088         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1089                    "decompress_threads",
1090                    "is invalid, it should be in the range of 1 to 255");
1091         return false;
1092     }
1093
1094     if (params->has_cpu_throttle_initial &&
1095         (params->cpu_throttle_initial < 1 ||
1096          params->cpu_throttle_initial > 99)) {
1097         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1098                    "cpu_throttle_initial",
1099                    "an integer in the range of 1 to 99");
1100         return false;
1101     }
1102
1103     if (params->has_cpu_throttle_increment &&
1104         (params->cpu_throttle_increment < 1 ||
1105          params->cpu_throttle_increment > 99)) {
1106         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1107                    "cpu_throttle_increment",
1108                    "an integer in the range of 1 to 99");
1109         return false;
1110     }
1111
1112     if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
1113         error_setg(errp, "Parameter 'max_bandwidth' expects an integer in the"
1114                          " range of 0 to %zu bytes/second", SIZE_MAX);
1115         return false;
1116     }
1117
1118     if (params->has_downtime_limit &&
1119         (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
1120         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1121                          "the range of 0 to %d milliseconds",
1122                          MAX_MIGRATE_DOWNTIME);
1123         return false;
1124     }
1125
1126     /* x_checkpoint_delay is now always positive */
1127
1128     if (params->has_x_multifd_channels && (params->x_multifd_channels < 1)) {
1129         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1130                    "multifd_channels",
1131                    "is invalid, it should be in the range of 1 to 255");
1132         return false;
1133     }
1134     if (params->has_x_multifd_page_count &&
1135         (params->x_multifd_page_count < 1 ||
1136          params->x_multifd_page_count > 10000)) {
1137         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1138                    "multifd_page_count",
1139                    "is invalid, it should be in the range of 1 to 10000");
1140         return false;
1141     }
1142
1143     if (params->has_xbzrle_cache_size &&
1144         (params->xbzrle_cache_size < qemu_target_page_size() ||
1145          !is_power_of_2(params->xbzrle_cache_size))) {
1146         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1147                    "xbzrle_cache_size",
1148                    "is invalid, it should be bigger than target page size"
1149                    " and a power of two");
1150         return false;
1151     }
1152
1153     if (params->has_max_cpu_throttle &&
1154         (params->max_cpu_throttle < params->cpu_throttle_initial ||
1155          params->max_cpu_throttle > 99)) {
1156         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1157                    "max_cpu_throttle",
1158                    "an integer in the range of cpu_throttle_initial to 99");
1159         return false;
1160     }
1161
1162     if (params->has_announce_initial &&
1163         params->announce_initial > 100000) {
1164         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1165                    "announce_initial",
1166                    "is invalid, it must be less than 100000 ms");
1167         return false;
1168     }
1169     if (params->has_announce_max &&
1170         params->announce_max > 100000) {
1171         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1172                    "announce_max",
1173                    "is invalid, it must be less than 100000 ms");
1174        return false;
1175     }
1176     if (params->has_announce_rounds &&
1177         params->announce_rounds > 1000) {
1178         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1179                    "announce_rounds",
1180                    "is invalid, it must be in the range of 0 to 1000");
1181        return false;
1182     }
1183     if (params->has_announce_step &&
1184         (params->announce_step < 1 ||
1185         params->announce_step > 10000)) {
1186         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1187                    "announce_step",
1188                    "is invalid, it must be in the range of 1 to 10000 ms");
1189        return false;
1190     }
1191     return true;
1192 }
1193
1194 static void migrate_params_test_apply(MigrateSetParameters *params,
1195                                       MigrationParameters *dest)
1196 {
1197     *dest = migrate_get_current()->parameters;
1198
1199     /* TODO use QAPI_CLONE() instead of duplicating it inline */
1200
1201     if (params->has_compress_level) {
1202         dest->compress_level = params->compress_level;
1203     }
1204
1205     if (params->has_compress_threads) {
1206         dest->compress_threads = params->compress_threads;
1207     }
1208
1209     if (params->has_compress_wait_thread) {
1210         dest->compress_wait_thread = params->compress_wait_thread;
1211     }
1212
1213     if (params->has_decompress_threads) {
1214         dest->decompress_threads = params->decompress_threads;
1215     }
1216
1217     if (params->has_cpu_throttle_initial) {
1218         dest->cpu_throttle_initial = params->cpu_throttle_initial;
1219     }
1220
1221     if (params->has_cpu_throttle_increment) {
1222         dest->cpu_throttle_increment = params->cpu_throttle_increment;
1223     }
1224
1225     if (params->has_tls_creds) {
1226         assert(params->tls_creds->type == QTYPE_QSTRING);
1227         dest->tls_creds = g_strdup(params->tls_creds->u.s);
1228     }
1229
1230     if (params->has_tls_hostname) {
1231         assert(params->tls_hostname->type == QTYPE_QSTRING);
1232         dest->tls_hostname = g_strdup(params->tls_hostname->u.s);
1233     }
1234
1235     if (params->has_max_bandwidth) {
1236         dest->max_bandwidth = params->max_bandwidth;
1237     }
1238
1239     if (params->has_downtime_limit) {
1240         dest->downtime_limit = params->downtime_limit;
1241     }
1242
1243     if (params->has_x_checkpoint_delay) {
1244         dest->x_checkpoint_delay = params->x_checkpoint_delay;
1245     }
1246
1247     if (params->has_block_incremental) {
1248         dest->block_incremental = params->block_incremental;
1249     }
1250     if (params->has_x_multifd_channels) {
1251         dest->x_multifd_channels = params->x_multifd_channels;
1252     }
1253     if (params->has_x_multifd_page_count) {
1254         dest->x_multifd_page_count = params->x_multifd_page_count;
1255     }
1256     if (params->has_xbzrle_cache_size) {
1257         dest->xbzrle_cache_size = params->xbzrle_cache_size;
1258     }
1259     if (params->has_max_postcopy_bandwidth) {
1260         dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1261     }
1262     if (params->has_max_cpu_throttle) {
1263         dest->max_cpu_throttle = params->max_cpu_throttle;
1264     }
1265     if (params->has_announce_initial) {
1266         dest->announce_initial = params->announce_initial;
1267     }
1268     if (params->has_announce_max) {
1269         dest->announce_max = params->announce_max;
1270     }
1271     if (params->has_announce_rounds) {
1272         dest->announce_rounds = params->announce_rounds;
1273     }
1274     if (params->has_announce_step) {
1275         dest->announce_step = params->announce_step;
1276     }
1277 }
1278
1279 static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
1280 {
1281     MigrationState *s = migrate_get_current();
1282
1283     /* TODO use QAPI_CLONE() instead of duplicating it inline */
1284
1285     if (params->has_compress_level) {
1286         s->parameters.compress_level = params->compress_level;
1287     }
1288
1289     if (params->has_compress_threads) {
1290         s->parameters.compress_threads = params->compress_threads;
1291     }
1292
1293     if (params->has_compress_wait_thread) {
1294         s->parameters.compress_wait_thread = params->compress_wait_thread;
1295     }
1296
1297     if (params->has_decompress_threads) {
1298         s->parameters.decompress_threads = params->decompress_threads;
1299     }
1300
1301     if (params->has_cpu_throttle_initial) {
1302         s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1303     }
1304
1305     if (params->has_cpu_throttle_increment) {
1306         s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1307     }
1308
1309     if (params->has_tls_creds) {
1310         g_free(s->parameters.tls_creds);
1311         assert(params->tls_creds->type == QTYPE_QSTRING);
1312         s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
1313     }
1314
1315     if (params->has_tls_hostname) {
1316         g_free(s->parameters.tls_hostname);
1317         assert(params->tls_hostname->type == QTYPE_QSTRING);
1318         s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
1319     }
1320
1321     if (params->has_max_bandwidth) {
1322         s->parameters.max_bandwidth = params->max_bandwidth;
1323         if (s->to_dst_file) {
1324             qemu_file_set_rate_limit(s->to_dst_file,
1325                                 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
1326         }
1327     }
1328
1329     if (params->has_downtime_limit) {
1330         s->parameters.downtime_limit = params->downtime_limit;
1331     }
1332
1333     if (params->has_x_checkpoint_delay) {
1334         s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
1335         if (migration_in_colo_state()) {
1336             colo_checkpoint_notify(s);
1337         }
1338     }
1339
1340     if (params->has_block_incremental) {
1341         s->parameters.block_incremental = params->block_incremental;
1342     }
1343     if (params->has_x_multifd_channels) {
1344         s->parameters.x_multifd_channels = params->x_multifd_channels;
1345     }
1346     if (params->has_x_multifd_page_count) {
1347         s->parameters.x_multifd_page_count = params->x_multifd_page_count;
1348     }
1349     if (params->has_xbzrle_cache_size) {
1350         s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
1351         xbzrle_cache_resize(params->xbzrle_cache_size, errp);
1352     }
1353     if (params->has_max_postcopy_bandwidth) {
1354         s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1355     }
1356     if (params->has_max_cpu_throttle) {
1357         s->parameters.max_cpu_throttle = params->max_cpu_throttle;
1358     }
1359     if (params->has_announce_initial) {
1360         s->parameters.announce_initial = params->announce_initial;
1361     }
1362     if (params->has_announce_max) {
1363         s->parameters.announce_max = params->announce_max;
1364     }
1365     if (params->has_announce_rounds) {
1366         s->parameters.announce_rounds = params->announce_rounds;
1367     }
1368     if (params->has_announce_step) {
1369         s->parameters.announce_step = params->announce_step;
1370     }
1371 }
1372
1373 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
1374 {
1375     MigrationParameters tmp;
1376
1377     /* TODO Rewrite "" to null instead */
1378     if (params->has_tls_creds
1379         && params->tls_creds->type == QTYPE_QNULL) {
1380         qobject_unref(params->tls_creds->u.n);
1381         params->tls_creds->type = QTYPE_QSTRING;
1382         params->tls_creds->u.s = strdup("");
1383     }
1384     /* TODO Rewrite "" to null instead */
1385     if (params->has_tls_hostname
1386         && params->tls_hostname->type == QTYPE_QNULL) {
1387         qobject_unref(params->tls_hostname->u.n);
1388         params->tls_hostname->type = QTYPE_QSTRING;
1389         params->tls_hostname->u.s = strdup("");
1390     }
1391
1392     migrate_params_test_apply(params, &tmp);
1393
1394     if (!migrate_params_check(&tmp, errp)) {
1395         /* Invalid parameter */
1396         return;
1397     }
1398
1399     migrate_params_apply(params, errp);
1400 }
1401
1402
1403 void qmp_migrate_start_postcopy(Error **errp)
1404 {
1405     MigrationState *s = migrate_get_current();
1406
1407     if (!migrate_postcopy()) {
1408         error_setg(errp, "Enable postcopy with migrate_set_capability before"
1409                          " the start of migration");
1410         return;
1411     }
1412
1413     if (s->state == MIGRATION_STATUS_NONE) {
1414         error_setg(errp, "Postcopy must be started after migration has been"
1415                          " started");
1416         return;
1417     }
1418     /*
1419      * we don't error if migration has finished since that would be racy
1420      * with issuing this command.
1421      */
1422     atomic_set(&s->start_postcopy, true);
1423 }
1424
1425 /* shared migration helpers */
1426
1427 void migrate_set_state(int *state, int old_state, int new_state)
1428 {
1429     assert(new_state < MIGRATION_STATUS__MAX);
1430     if (atomic_cmpxchg(state, old_state, new_state) == old_state) {
1431         trace_migrate_set_state(MigrationStatus_str(new_state));
1432         migrate_generate_event(new_state);
1433     }
1434 }
1435
1436 static MigrationCapabilityStatusList *migrate_cap_add(
1437     MigrationCapabilityStatusList *list,
1438     MigrationCapability index,
1439     bool state)
1440 {
1441     MigrationCapabilityStatusList *cap;
1442
1443     cap = g_new0(MigrationCapabilityStatusList, 1);
1444     cap->value = g_new0(MigrationCapabilityStatus, 1);
1445     cap->value->capability = index;
1446     cap->value->state = state;
1447     cap->next = list;
1448
1449     return cap;
1450 }
1451
1452 void migrate_set_block_enabled(bool value, Error **errp)
1453 {
1454     MigrationCapabilityStatusList *cap;
1455
1456     cap = migrate_cap_add(NULL, MIGRATION_CAPABILITY_BLOCK, value);
1457     qmp_migrate_set_capabilities(cap, errp);
1458     qapi_free_MigrationCapabilityStatusList(cap);
1459 }
1460
1461 static void migrate_set_block_incremental(MigrationState *s, bool value)
1462 {
1463     s->parameters.block_incremental = value;
1464 }
1465
1466 static void block_cleanup_parameters(MigrationState *s)
1467 {
1468     if (s->must_remove_block_options) {
1469         /* setting to false can never fail */
1470         migrate_set_block_enabled(false, &error_abort);
1471         migrate_set_block_incremental(s, false);
1472         s->must_remove_block_options = false;
1473     }
1474 }
1475
1476 static void migrate_fd_cleanup(void *opaque)
1477 {
1478     MigrationState *s = opaque;
1479
1480     qemu_bh_delete(s->cleanup_bh);
1481     s->cleanup_bh = NULL;
1482
1483     qemu_savevm_state_cleanup();
1484
1485     if (s->to_dst_file) {
1486         QEMUFile *tmp;
1487
1488         trace_migrate_fd_cleanup();
1489         qemu_mutex_unlock_iothread();
1490         if (s->migration_thread_running) {
1491             qemu_thread_join(&s->thread);
1492             s->migration_thread_running = false;
1493         }
1494         qemu_mutex_lock_iothread();
1495
1496         multifd_save_cleanup();
1497         qemu_mutex_lock(&s->qemu_file_lock);
1498         tmp = s->to_dst_file;
1499         s->to_dst_file = NULL;
1500         qemu_mutex_unlock(&s->qemu_file_lock);
1501         /*
1502          * Close the file handle without the lock to make sure the
1503          * critical section won't block for long.
1504          */
1505         qemu_fclose(tmp);
1506     }
1507
1508     assert((s->state != MIGRATION_STATUS_ACTIVE) &&
1509            (s->state != MIGRATION_STATUS_POSTCOPY_ACTIVE));
1510
1511     if (s->state == MIGRATION_STATUS_CANCELLING) {
1512         migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1513                           MIGRATION_STATUS_CANCELLED);
1514     }
1515
1516     if (s->error) {
1517         /* It is used on info migrate.  We can't free it */
1518         error_report_err(error_copy(s->error));
1519     }
1520     notifier_list_notify(&migration_state_notifiers, s);
1521     block_cleanup_parameters(s);
1522 }
1523
1524 void migrate_set_error(MigrationState *s, const Error *error)
1525 {
1526     qemu_mutex_lock(&s->error_mutex);
1527     if (!s->error) {
1528         s->error = error_copy(error);
1529     }
1530     qemu_mutex_unlock(&s->error_mutex);
1531 }
1532
1533 void migrate_fd_error(MigrationState *s, const Error *error)
1534 {
1535     trace_migrate_fd_error(error_get_pretty(error));
1536     assert(s->to_dst_file == NULL);
1537     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1538                       MIGRATION_STATUS_FAILED);
1539     migrate_set_error(s, error);
1540 }
1541
1542 static void migrate_fd_cancel(MigrationState *s)
1543 {
1544     int old_state ;
1545     QEMUFile *f = migrate_get_current()->to_dst_file;
1546     trace_migrate_fd_cancel();
1547
1548     if (s->rp_state.from_dst_file) {
1549         /* shutdown the rp socket, so causing the rp thread to shutdown */
1550         qemu_file_shutdown(s->rp_state.from_dst_file);
1551     }
1552
1553     do {
1554         old_state = s->state;
1555         if (!migration_is_setup_or_active(old_state)) {
1556             break;
1557         }
1558         /* If the migration is paused, kick it out of the pause */
1559         if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1560             qemu_sem_post(&s->pause_sem);
1561         }
1562         migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1563     } while (s->state != MIGRATION_STATUS_CANCELLING);
1564
1565     /*
1566      * If we're unlucky the migration code might be stuck somewhere in a
1567      * send/write while the network has failed and is waiting to timeout;
1568      * if we've got shutdown(2) available then we can force it to quit.
1569      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1570      * called in a bh, so there is no race against this cancel.
1571      */
1572     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1573         qemu_file_shutdown(f);
1574     }
1575     if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1576         Error *local_err = NULL;
1577
1578         bdrv_invalidate_cache_all(&local_err);
1579         if (local_err) {
1580             error_report_err(local_err);
1581         } else {
1582             s->block_inactive = false;
1583         }
1584     }
1585 }
1586
1587 void add_migration_state_change_notifier(Notifier *notify)
1588 {
1589     notifier_list_add(&migration_state_notifiers, notify);
1590 }
1591
1592 void remove_migration_state_change_notifier(Notifier *notify)
1593 {
1594     notifier_remove(notify);
1595 }
1596
1597 bool migration_in_setup(MigrationState *s)
1598 {
1599     return s->state == MIGRATION_STATUS_SETUP;
1600 }
1601
1602 bool migration_has_finished(MigrationState *s)
1603 {
1604     return s->state == MIGRATION_STATUS_COMPLETED;
1605 }
1606
1607 bool migration_has_failed(MigrationState *s)
1608 {
1609     return (s->state == MIGRATION_STATUS_CANCELLED ||
1610             s->state == MIGRATION_STATUS_FAILED);
1611 }
1612
1613 bool migration_in_postcopy(void)
1614 {
1615     MigrationState *s = migrate_get_current();
1616
1617     return (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
1618 }
1619
1620 bool migration_in_postcopy_after_devices(MigrationState *s)
1621 {
1622     return migration_in_postcopy() && s->postcopy_after_devices;
1623 }
1624
1625 bool migration_is_idle(void)
1626 {
1627     MigrationState *s = migrate_get_current();
1628
1629     switch (s->state) {
1630     case MIGRATION_STATUS_NONE:
1631     case MIGRATION_STATUS_CANCELLED:
1632     case MIGRATION_STATUS_COMPLETED:
1633     case MIGRATION_STATUS_FAILED:
1634         return true;
1635     case MIGRATION_STATUS_SETUP:
1636     case MIGRATION_STATUS_CANCELLING:
1637     case MIGRATION_STATUS_ACTIVE:
1638     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1639     case MIGRATION_STATUS_COLO:
1640     case MIGRATION_STATUS_PRE_SWITCHOVER:
1641     case MIGRATION_STATUS_DEVICE:
1642         return false;
1643     case MIGRATION_STATUS__MAX:
1644         g_assert_not_reached();
1645     }
1646
1647     return false;
1648 }
1649
1650 void migrate_init(MigrationState *s)
1651 {
1652     /*
1653      * Reinitialise all migration state, except
1654      * parameters/capabilities that the user set, and
1655      * locks.
1656      */
1657     s->bytes_xfer = 0;
1658     s->xfer_limit = 0;
1659     s->cleanup_bh = 0;
1660     s->to_dst_file = NULL;
1661     s->state = MIGRATION_STATUS_NONE;
1662     s->rp_state.from_dst_file = NULL;
1663     s->rp_state.error = false;
1664     s->mbps = 0.0;
1665     s->pages_per_second = 0.0;
1666     s->downtime = 0;
1667     s->expected_downtime = 0;
1668     s->setup_time = 0;
1669     s->start_postcopy = false;
1670     s->postcopy_after_devices = false;
1671     s->migration_thread_running = false;
1672     error_free(s->error);
1673     s->error = NULL;
1674
1675     migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
1676
1677     s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
1678     s->total_time = 0;
1679     s->vm_was_running = false;
1680     s->iteration_initial_bytes = 0;
1681     s->threshold_size = 0;
1682 }
1683
1684 static GSList *migration_blockers;
1685
1686 int migrate_add_blocker(Error *reason, Error **errp)
1687 {
1688     if (migrate_get_current()->only_migratable) {
1689         error_propagate_prepend(errp, error_copy(reason),
1690                                 "disallowing migration blocker "
1691                                 "(--only_migratable) for: ");
1692         return -EACCES;
1693     }
1694
1695     if (migration_is_idle()) {
1696         migration_blockers = g_slist_prepend(migration_blockers, reason);
1697         return 0;
1698     }
1699
1700     error_propagate_prepend(errp, error_copy(reason),
1701                             "disallowing migration blocker "
1702                             "(migration in progress) for: ");
1703     return -EBUSY;
1704 }
1705
1706 void migrate_del_blocker(Error *reason)
1707 {
1708     migration_blockers = g_slist_remove(migration_blockers, reason);
1709 }
1710
1711 void qmp_migrate_incoming(const char *uri, Error **errp)
1712 {
1713     Error *local_err = NULL;
1714     static bool once = true;
1715
1716     if (!deferred_incoming) {
1717         error_setg(errp, "For use with '-incoming defer'");
1718         return;
1719     }
1720     if (!once) {
1721         error_setg(errp, "The incoming migration has already been started");
1722     }
1723
1724     qemu_start_incoming_migration(uri, &local_err);
1725
1726     if (local_err) {
1727         error_propagate(errp, local_err);
1728         return;
1729     }
1730
1731     once = false;
1732 }
1733
1734 void qmp_migrate_recover(const char *uri, Error **errp)
1735 {
1736     MigrationIncomingState *mis = migration_incoming_get_current();
1737
1738     if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1739         error_setg(errp, "Migrate recover can only be run "
1740                    "when postcopy is paused.");
1741         return;
1742     }
1743
1744     if (atomic_cmpxchg(&mis->postcopy_recover_triggered,
1745                        false, true) == true) {
1746         error_setg(errp, "Migrate recovery is triggered already");
1747         return;
1748     }
1749
1750     /*
1751      * Note that this call will never start a real migration; it will
1752      * only re-setup the migration stream and poke existing migration
1753      * to continue using that newly established channel.
1754      */
1755     qemu_start_incoming_migration(uri, errp);
1756 }
1757
1758 void qmp_migrate_pause(Error **errp)
1759 {
1760     MigrationState *ms = migrate_get_current();
1761     MigrationIncomingState *mis = migration_incoming_get_current();
1762     int ret;
1763
1764     if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1765         /* Source side, during postcopy */
1766         qemu_mutex_lock(&ms->qemu_file_lock);
1767         ret = qemu_file_shutdown(ms->to_dst_file);
1768         qemu_mutex_unlock(&ms->qemu_file_lock);
1769         if (ret) {
1770             error_setg(errp, "Failed to pause source migration");
1771         }
1772         return;
1773     }
1774
1775     if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
1776         ret = qemu_file_shutdown(mis->from_src_file);
1777         if (ret) {
1778             error_setg(errp, "Failed to pause destination migration");
1779         }
1780         return;
1781     }
1782
1783     error_setg(errp, "migrate-pause is currently only supported "
1784                "during postcopy-active state");
1785 }
1786
1787 bool migration_is_blocked(Error **errp)
1788 {
1789     if (qemu_savevm_state_blocked(errp)) {
1790         return true;
1791     }
1792
1793     if (migration_blockers) {
1794         error_propagate(errp, error_copy(migration_blockers->data));
1795         return true;
1796     }
1797
1798     return false;
1799 }
1800
1801 /* Returns true if continue to migrate, or false if error detected */
1802 static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
1803                             bool resume, Error **errp)
1804 {
1805     Error *local_err = NULL;
1806
1807     if (resume) {
1808         if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
1809             error_setg(errp, "Cannot resume if there is no "
1810                        "paused migration");
1811             return false;
1812         }
1813
1814         /*
1815          * Postcopy recovery won't work well with release-ram
1816          * capability since release-ram will drop the page buffer as
1817          * long as the page is put into the send buffer.  So if there
1818          * is a network failure happened, any page buffers that have
1819          * not yet reached the destination VM but have already been
1820          * sent from the source VM will be lost forever.  Let's refuse
1821          * the client from resuming such a postcopy migration.
1822          * Luckily release-ram was designed to only be used when src
1823          * and destination VMs are on the same host, so it should be
1824          * fine.
1825          */
1826         if (migrate_release_ram()) {
1827             error_setg(errp, "Postcopy recovery cannot work "
1828                        "when release-ram capability is set");
1829             return false;
1830         }
1831
1832         /* This is a resume, skip init status */
1833         return true;
1834     }
1835
1836     if (migration_is_setup_or_active(s->state) ||
1837         s->state == MIGRATION_STATUS_CANCELLING ||
1838         s->state == MIGRATION_STATUS_COLO) {
1839         error_setg(errp, QERR_MIGRATION_ACTIVE);
1840         return false;
1841     }
1842
1843     if (runstate_check(RUN_STATE_INMIGRATE)) {
1844         error_setg(errp, "Guest is waiting for an incoming migration");
1845         return false;
1846     }
1847
1848     if (migration_is_blocked(errp)) {
1849         return false;
1850     }
1851
1852     if (blk || blk_inc) {
1853         if (migrate_use_block() || migrate_use_block_incremental()) {
1854             error_setg(errp, "Command options are incompatible with "
1855                        "current migration capabilities");
1856             return false;
1857         }
1858         migrate_set_block_enabled(true, &local_err);
1859         if (local_err) {
1860             error_propagate(errp, local_err);
1861             return false;
1862         }
1863         s->must_remove_block_options = true;
1864     }
1865
1866     if (blk_inc) {
1867         migrate_set_block_incremental(s, true);
1868     }
1869
1870     migrate_init(s);
1871
1872     return true;
1873 }
1874
1875 void qmp_migrate(const char *uri, bool has_blk, bool blk,
1876                  bool has_inc, bool inc, bool has_detach, bool detach,
1877                  bool has_resume, bool resume, Error **errp)
1878 {
1879     Error *local_err = NULL;
1880     MigrationState *s = migrate_get_current();
1881     const char *p;
1882
1883     if (!migrate_prepare(s, has_blk && blk, has_inc && inc,
1884                          has_resume && resume, errp)) {
1885         /* Error detected, put into errp */
1886         return;
1887     }
1888
1889     if (strstart(uri, "tcp:", &p)) {
1890         tcp_start_outgoing_migration(s, p, &local_err);
1891 #ifdef CONFIG_RDMA
1892     } else if (strstart(uri, "rdma:", &p)) {
1893         rdma_start_outgoing_migration(s, p, &local_err);
1894 #endif
1895     } else if (strstart(uri, "exec:", &p)) {
1896         exec_start_outgoing_migration(s, p, &local_err);
1897     } else if (strstart(uri, "unix:", &p)) {
1898         unix_start_outgoing_migration(s, p, &local_err);
1899     } else if (strstart(uri, "fd:", &p)) {
1900         fd_start_outgoing_migration(s, p, &local_err);
1901     } else {
1902         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
1903                    "a valid migration protocol");
1904         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1905                           MIGRATION_STATUS_FAILED);
1906         block_cleanup_parameters(s);
1907         return;
1908     }
1909
1910     if (local_err) {
1911         migrate_fd_error(s, local_err);
1912         error_propagate(errp, local_err);
1913         return;
1914     }
1915 }
1916
1917 void qmp_migrate_cancel(Error **errp)
1918 {
1919     migrate_fd_cancel(migrate_get_current());
1920 }
1921
1922 void qmp_migrate_continue(MigrationStatus state, Error **errp)
1923 {
1924     MigrationState *s = migrate_get_current();
1925     if (s->state != state) {
1926         error_setg(errp,  "Migration not in expected state: %s",
1927                    MigrationStatus_str(s->state));
1928         return;
1929     }
1930     qemu_sem_post(&s->pause_sem);
1931 }
1932
1933 void qmp_migrate_set_cache_size(int64_t value, Error **errp)
1934 {
1935     MigrateSetParameters p = {
1936         .has_xbzrle_cache_size = true,
1937         .xbzrle_cache_size = value,
1938     };
1939
1940     qmp_migrate_set_parameters(&p, errp);
1941 }
1942
1943 int64_t qmp_query_migrate_cache_size(Error **errp)
1944 {
1945     return migrate_xbzrle_cache_size();
1946 }
1947
1948 void qmp_migrate_set_speed(int64_t value, Error **errp)
1949 {
1950     MigrateSetParameters p = {
1951         .has_max_bandwidth = true,
1952         .max_bandwidth = value,
1953     };
1954
1955     qmp_migrate_set_parameters(&p, errp);
1956 }
1957
1958 void qmp_migrate_set_downtime(double value, Error **errp)
1959 {
1960     if (value < 0 || value > MAX_MIGRATE_DOWNTIME_SECONDS) {
1961         error_setg(errp, "Parameter 'downtime_limit' expects an integer in "
1962                          "the range of 0 to %d seconds",
1963                          MAX_MIGRATE_DOWNTIME_SECONDS);
1964         return;
1965     }
1966
1967     value *= 1000; /* Convert to milliseconds */
1968     value = MAX(0, MIN(INT64_MAX, value));
1969
1970     MigrateSetParameters p = {
1971         .has_downtime_limit = true,
1972         .downtime_limit = value,
1973     };
1974
1975     qmp_migrate_set_parameters(&p, errp);
1976 }
1977
1978 bool migrate_release_ram(void)
1979 {
1980     MigrationState *s;
1981
1982     s = migrate_get_current();
1983
1984     return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
1985 }
1986
1987 bool migrate_postcopy_ram(void)
1988 {
1989     MigrationState *s;
1990
1991     s = migrate_get_current();
1992
1993     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1994 }
1995
1996 bool migrate_postcopy(void)
1997 {
1998     return migrate_postcopy_ram() || migrate_dirty_bitmaps();
1999 }
2000
2001 bool migrate_auto_converge(void)
2002 {
2003     MigrationState *s;
2004
2005     s = migrate_get_current();
2006
2007     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
2008 }
2009
2010 bool migrate_zero_blocks(void)
2011 {
2012     MigrationState *s;
2013
2014     s = migrate_get_current();
2015
2016     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
2017 }
2018
2019 bool migrate_postcopy_blocktime(void)
2020 {
2021     MigrationState *s;
2022
2023     s = migrate_get_current();
2024
2025     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
2026 }
2027
2028 bool migrate_use_compression(void)
2029 {
2030     MigrationState *s;
2031
2032     s = migrate_get_current();
2033
2034     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
2035 }
2036
2037 int migrate_compress_level(void)
2038 {
2039     MigrationState *s;
2040
2041     s = migrate_get_current();
2042
2043     return s->parameters.compress_level;
2044 }
2045
2046 int migrate_compress_threads(void)
2047 {
2048     MigrationState *s;
2049
2050     s = migrate_get_current();
2051
2052     return s->parameters.compress_threads;
2053 }
2054
2055 int migrate_compress_wait_thread(void)
2056 {
2057     MigrationState *s;
2058
2059     s = migrate_get_current();
2060
2061     return s->parameters.compress_wait_thread;
2062 }
2063
2064 int migrate_decompress_threads(void)
2065 {
2066     MigrationState *s;
2067
2068     s = migrate_get_current();
2069
2070     return s->parameters.decompress_threads;
2071 }
2072
2073 bool migrate_dirty_bitmaps(void)
2074 {
2075     MigrationState *s;
2076
2077     s = migrate_get_current();
2078
2079     return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS];
2080 }
2081
2082 bool migrate_ignore_shared(void)
2083 {
2084     MigrationState *s;
2085
2086     s = migrate_get_current();
2087
2088     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_IGNORE_SHARED];
2089 }
2090
2091 bool migrate_use_events(void)
2092 {
2093     MigrationState *s;
2094
2095     s = migrate_get_current();
2096
2097     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
2098 }
2099
2100 bool migrate_use_multifd(void)
2101 {
2102     MigrationState *s;
2103
2104     s = migrate_get_current();
2105
2106     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_MULTIFD];
2107 }
2108
2109 bool migrate_pause_before_switchover(void)
2110 {
2111     MigrationState *s;
2112
2113     s = migrate_get_current();
2114
2115     return s->enabled_capabilities[
2116         MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
2117 }
2118
2119 int migrate_multifd_channels(void)
2120 {
2121     MigrationState *s;
2122
2123     s = migrate_get_current();
2124
2125     return s->parameters.x_multifd_channels;
2126 }
2127
2128 int migrate_multifd_page_count(void)
2129 {
2130     MigrationState *s;
2131
2132     s = migrate_get_current();
2133
2134     return s->parameters.x_multifd_page_count;
2135 }
2136
2137 int migrate_use_xbzrle(void)
2138 {
2139     MigrationState *s;
2140
2141     s = migrate_get_current();
2142
2143     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
2144 }
2145
2146 int64_t migrate_xbzrle_cache_size(void)
2147 {
2148     MigrationState *s;
2149
2150     s = migrate_get_current();
2151
2152     return s->parameters.xbzrle_cache_size;
2153 }
2154
2155 static int64_t migrate_max_postcopy_bandwidth(void)
2156 {
2157     MigrationState *s;
2158
2159     s = migrate_get_current();
2160
2161     return s->parameters.max_postcopy_bandwidth;
2162 }
2163
2164 bool migrate_use_block(void)
2165 {
2166     MigrationState *s;
2167
2168     s = migrate_get_current();
2169
2170     return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
2171 }
2172
2173 bool migrate_use_return_path(void)
2174 {
2175     MigrationState *s;
2176
2177     s = migrate_get_current();
2178
2179     return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
2180 }
2181
2182 bool migrate_use_block_incremental(void)
2183 {
2184     MigrationState *s;
2185
2186     s = migrate_get_current();
2187
2188     return s->parameters.block_incremental;
2189 }
2190
2191 /* migration thread support */
2192 /*
2193  * Something bad happened to the RP stream, mark an error
2194  * The caller shall print or trace something to indicate why
2195  */
2196 static void mark_source_rp_bad(MigrationState *s)
2197 {
2198     s->rp_state.error = true;
2199 }
2200
2201 static struct rp_cmd_args {
2202     ssize_t     len; /* -1 = variable */
2203     const char *name;
2204 } rp_cmd_args[] = {
2205     [MIG_RP_MSG_INVALID]        = { .len = -1, .name = "INVALID" },
2206     [MIG_RP_MSG_SHUT]           = { .len =  4, .name = "SHUT" },
2207     [MIG_RP_MSG_PONG]           = { .len =  4, .name = "PONG" },
2208     [MIG_RP_MSG_REQ_PAGES]      = { .len = 12, .name = "REQ_PAGES" },
2209     [MIG_RP_MSG_REQ_PAGES_ID]   = { .len = -1, .name = "REQ_PAGES_ID" },
2210     [MIG_RP_MSG_RECV_BITMAP]    = { .len = -1, .name = "RECV_BITMAP" },
2211     [MIG_RP_MSG_RESUME_ACK]     = { .len =  4, .name = "RESUME_ACK" },
2212     [MIG_RP_MSG_MAX]            = { .len = -1, .name = "MAX" },
2213 };
2214
2215 /*
2216  * Process a request for pages received on the return path,
2217  * We're allowed to send more than requested (e.g. to round to our page size)
2218  * and we don't need to send pages that have already been sent.
2219  */
2220 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
2221                                        ram_addr_t start, size_t len)
2222 {
2223     long our_host_ps = getpagesize();
2224
2225     trace_migrate_handle_rp_req_pages(rbname, start, len);
2226
2227     /*
2228      * Since we currently insist on matching page sizes, just sanity check
2229      * we're being asked for whole host pages.
2230      */
2231     if (start & (our_host_ps-1) ||
2232        (len & (our_host_ps-1))) {
2233         error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2234                      " len: %zd", __func__, start, len);
2235         mark_source_rp_bad(ms);
2236         return;
2237     }
2238
2239     if (ram_save_queue_pages(rbname, start, len)) {
2240         mark_source_rp_bad(ms);
2241     }
2242 }
2243
2244 /* Return true to retry, false to quit */
2245 static bool postcopy_pause_return_path_thread(MigrationState *s)
2246 {
2247     trace_postcopy_pause_return_path();
2248
2249     qemu_sem_wait(&s->postcopy_pause_rp_sem);
2250
2251     trace_postcopy_pause_return_path_continued();
2252
2253     return true;
2254 }
2255
2256 static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name)
2257 {
2258     RAMBlock *block = qemu_ram_block_by_name(block_name);
2259
2260     if (!block) {
2261         error_report("%s: invalid block name '%s'", __func__, block_name);
2262         return -EINVAL;
2263     }
2264
2265     /* Fetch the received bitmap and refresh the dirty bitmap */
2266     return ram_dirty_bitmap_reload(s, block);
2267 }
2268
2269 static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
2270 {
2271     trace_source_return_path_thread_resume_ack(value);
2272
2273     if (value != MIGRATION_RESUME_ACK_VALUE) {
2274         error_report("%s: illegal resume_ack value %"PRIu32,
2275                      __func__, value);
2276         return -1;
2277     }
2278
2279     /* Now both sides are active. */
2280     migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2281                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
2282
2283     /* Notify send thread that time to continue send pages */
2284     qemu_sem_post(&s->rp_state.rp_sem);
2285
2286     return 0;
2287 }
2288
2289 /*
2290  * Handles messages sent on the return path towards the source VM
2291  *
2292  */
2293 static void *source_return_path_thread(void *opaque)
2294 {
2295     MigrationState *ms = opaque;
2296     QEMUFile *rp = ms->rp_state.from_dst_file;
2297     uint16_t header_len, header_type;
2298     uint8_t buf[512];
2299     uint32_t tmp32, sibling_error;
2300     ram_addr_t start = 0; /* =0 to silence warning */
2301     size_t  len = 0, expected_len;
2302     int res;
2303
2304     trace_source_return_path_thread_entry();
2305     rcu_register_thread();
2306
2307 retry:
2308     while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
2309            migration_is_setup_or_active(ms->state)) {
2310         trace_source_return_path_thread_loop_top();
2311         header_type = qemu_get_be16(rp);
2312         header_len = qemu_get_be16(rp);
2313
2314         if (qemu_file_get_error(rp)) {
2315             mark_source_rp_bad(ms);
2316             goto out;
2317         }
2318
2319         if (header_type >= MIG_RP_MSG_MAX ||
2320             header_type == MIG_RP_MSG_INVALID) {
2321             error_report("RP: Received invalid message 0x%04x length 0x%04x",
2322                     header_type, header_len);
2323             mark_source_rp_bad(ms);
2324             goto out;
2325         }
2326
2327         if ((rp_cmd_args[header_type].len != -1 &&
2328             header_len != rp_cmd_args[header_type].len) ||
2329             header_len > sizeof(buf)) {
2330             error_report("RP: Received '%s' message (0x%04x) with"
2331                     "incorrect length %d expecting %zu",
2332                     rp_cmd_args[header_type].name, header_type, header_len,
2333                     (size_t)rp_cmd_args[header_type].len);
2334             mark_source_rp_bad(ms);
2335             goto out;
2336         }
2337
2338         /* We know we've got a valid header by this point */
2339         res = qemu_get_buffer(rp, buf, header_len);
2340         if (res != header_len) {
2341             error_report("RP: Failed reading data for message 0x%04x"
2342                          " read %d expected %d",
2343                          header_type, res, header_len);
2344             mark_source_rp_bad(ms);
2345             goto out;
2346         }
2347
2348         /* OK, we have the message and the data */
2349         switch (header_type) {
2350         case MIG_RP_MSG_SHUT:
2351             sibling_error = ldl_be_p(buf);
2352             trace_source_return_path_thread_shut(sibling_error);
2353             if (sibling_error) {
2354                 error_report("RP: Sibling indicated error %d", sibling_error);
2355                 mark_source_rp_bad(ms);
2356             }
2357             /*
2358              * We'll let the main thread deal with closing the RP
2359              * we could do a shutdown(2) on it, but we're the only user
2360              * anyway, so there's nothing gained.
2361              */
2362             goto out;
2363
2364         case MIG_RP_MSG_PONG:
2365             tmp32 = ldl_be_p(buf);
2366             trace_source_return_path_thread_pong(tmp32);
2367             break;
2368
2369         case MIG_RP_MSG_REQ_PAGES:
2370             start = ldq_be_p(buf);
2371             len = ldl_be_p(buf + 8);
2372             migrate_handle_rp_req_pages(ms, NULL, start, len);
2373             break;
2374
2375         case MIG_RP_MSG_REQ_PAGES_ID:
2376             expected_len = 12 + 1; /* header + termination */
2377
2378             if (header_len >= expected_len) {
2379                 start = ldq_be_p(buf);
2380                 len = ldl_be_p(buf + 8);
2381                 /* Now we expect an idstr */
2382                 tmp32 = buf[12]; /* Length of the following idstr */
2383                 buf[13 + tmp32] = '\0';
2384                 expected_len += tmp32;
2385             }
2386             if (header_len != expected_len) {
2387                 error_report("RP: Req_Page_id with length %d expecting %zd",
2388                         header_len, expected_len);
2389                 mark_source_rp_bad(ms);
2390                 goto out;
2391             }
2392             migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
2393             break;
2394
2395         case MIG_RP_MSG_RECV_BITMAP:
2396             if (header_len < 1) {
2397                 error_report("%s: missing block name", __func__);
2398                 mark_source_rp_bad(ms);
2399                 goto out;
2400             }
2401             /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2402             buf[buf[0] + 1] = '\0';
2403             if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) {
2404                 mark_source_rp_bad(ms);
2405                 goto out;
2406             }
2407             break;
2408
2409         case MIG_RP_MSG_RESUME_ACK:
2410             tmp32 = ldl_be_p(buf);
2411             if (migrate_handle_rp_resume_ack(ms, tmp32)) {
2412                 mark_source_rp_bad(ms);
2413                 goto out;
2414             }
2415             break;
2416
2417         default:
2418             break;
2419         }
2420     }
2421
2422 out:
2423     res = qemu_file_get_error(rp);
2424     if (res) {
2425         if (res == -EIO) {
2426             /*
2427              * Maybe there is something we can do: it looks like a
2428              * network down issue, and we pause for a recovery.
2429              */
2430             if (postcopy_pause_return_path_thread(ms)) {
2431                 /* Reload rp, reset the rest */
2432                 if (rp != ms->rp_state.from_dst_file) {
2433                     qemu_fclose(rp);
2434                     rp = ms->rp_state.from_dst_file;
2435                 }
2436                 ms->rp_state.error = false;
2437                 goto retry;
2438             }
2439         }
2440
2441         trace_source_return_path_thread_bad_end();
2442         mark_source_rp_bad(ms);
2443     }
2444
2445     trace_source_return_path_thread_end();
2446     ms->rp_state.from_dst_file = NULL;
2447     qemu_fclose(rp);
2448     rcu_unregister_thread();
2449     return NULL;
2450 }
2451
2452 static int open_return_path_on_source(MigrationState *ms,
2453                                       bool create_thread)
2454 {
2455
2456     ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
2457     if (!ms->rp_state.from_dst_file) {
2458         return -1;
2459     }
2460
2461     trace_open_return_path_on_source();
2462
2463     if (!create_thread) {
2464         /* We're done */
2465         return 0;
2466     }
2467
2468     qemu_thread_create(&ms->rp_state.rp_thread, "return path",
2469                        source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
2470
2471     trace_open_return_path_on_source_continue();
2472
2473     return 0;
2474 }
2475
2476 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2477 static int await_return_path_close_on_source(MigrationState *ms)
2478 {
2479     /*
2480      * If this is a normal exit then the destination will send a SHUT and the
2481      * rp_thread will exit, however if there's an error we need to cause
2482      * it to exit.
2483      */
2484     if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
2485         /*
2486          * shutdown(2), if we have it, will cause it to unblock if it's stuck
2487          * waiting for the destination.
2488          */
2489         qemu_file_shutdown(ms->rp_state.from_dst_file);
2490         mark_source_rp_bad(ms);
2491     }
2492     trace_await_return_path_close_on_source_joining();
2493     qemu_thread_join(&ms->rp_state.rp_thread);
2494     trace_await_return_path_close_on_source_close();
2495     return ms->rp_state.error;
2496 }
2497
2498 /*
2499  * Switch from normal iteration to postcopy
2500  * Returns non-0 on error
2501  */
2502 static int postcopy_start(MigrationState *ms)
2503 {
2504     int ret;
2505     QIOChannelBuffer *bioc;
2506     QEMUFile *fb;
2507     int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2508     int64_t bandwidth = migrate_max_postcopy_bandwidth();
2509     bool restart_block = false;
2510     int cur_state = MIGRATION_STATUS_ACTIVE;
2511     if (!migrate_pause_before_switchover()) {
2512         migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
2513                           MIGRATION_STATUS_POSTCOPY_ACTIVE);
2514     }
2515
2516     trace_postcopy_start();
2517     qemu_mutex_lock_iothread();
2518     trace_postcopy_start_set_run();
2519
2520     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2521     global_state_store();
2522     ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2523     if (ret < 0) {
2524         goto fail;
2525     }
2526
2527     ret = migration_maybe_pause(ms, &cur_state,
2528                                 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2529     if (ret < 0) {
2530         goto fail;
2531     }
2532
2533     ret = bdrv_inactivate_all();
2534     if (ret < 0) {
2535         goto fail;
2536     }
2537     restart_block = true;
2538
2539     /*
2540      * Cause any non-postcopiable, but iterative devices to
2541      * send out their final data.
2542      */
2543     qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
2544
2545     /*
2546      * in Finish migrate and with the io-lock held everything should
2547      * be quiet, but we've potentially still got dirty pages and we
2548      * need to tell the destination to throw any pages it's already received
2549      * that are dirty
2550      */
2551     if (migrate_postcopy_ram()) {
2552         if (ram_postcopy_send_discard_bitmap(ms)) {
2553             error_report("postcopy send discard bitmap failed");
2554             goto fail;
2555         }
2556     }
2557
2558     /*
2559      * send rest of state - note things that are doing postcopy
2560      * will notice we're in POSTCOPY_ACTIVE and not actually
2561      * wrap their state up here
2562      */
2563     /* 0 max-postcopy-bandwidth means unlimited */
2564     if (!bandwidth) {
2565         qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
2566     } else {
2567         qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO);
2568     }
2569     if (migrate_postcopy_ram()) {
2570         /* Ping just for debugging, helps line traces up */
2571         qemu_savevm_send_ping(ms->to_dst_file, 2);
2572     }
2573
2574     /*
2575      * While loading the device state we may trigger page transfer
2576      * requests and the fd must be free to process those, and thus
2577      * the destination must read the whole device state off the fd before
2578      * it starts processing it.  Unfortunately the ad-hoc migration format
2579      * doesn't allow the destination to know the size to read without fully
2580      * parsing it through each devices load-state code (especially the open
2581      * coded devices that use get/put).
2582      * So we wrap the device state up in a package with a length at the start;
2583      * to do this we use a qemu_buf to hold the whole of the device state.
2584      */
2585     bioc = qio_channel_buffer_new(4096);
2586     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
2587     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
2588     object_unref(OBJECT(bioc));
2589
2590     /*
2591      * Make sure the receiver can get incoming pages before we send the rest
2592      * of the state
2593      */
2594     qemu_savevm_send_postcopy_listen(fb);
2595
2596     qemu_savevm_state_complete_precopy(fb, false, false);
2597     if (migrate_postcopy_ram()) {
2598         qemu_savevm_send_ping(fb, 3);
2599     }
2600
2601     qemu_savevm_send_postcopy_run(fb);
2602
2603     /* <><> end of stuff going into the package */
2604
2605     /* Last point of recovery; as soon as we send the package the destination
2606      * can open devices and potentially start running.
2607      * Lets just check again we've not got any errors.
2608      */
2609     ret = qemu_file_get_error(ms->to_dst_file);
2610     if (ret) {
2611         error_report("postcopy_start: Migration stream errored (pre package)");
2612         goto fail_closefb;
2613     }
2614
2615     restart_block = false;
2616
2617     /* Now send that blob */
2618     if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
2619         goto fail_closefb;
2620     }
2621     qemu_fclose(fb);
2622
2623     /* Send a notify to give a chance for anything that needs to happen
2624      * at the transition to postcopy and after the device state; in particular
2625      * spice needs to trigger a transition now
2626      */
2627     ms->postcopy_after_devices = true;
2628     notifier_list_notify(&migration_state_notifiers, ms);
2629
2630     ms->downtime =  qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
2631
2632     qemu_mutex_unlock_iothread();
2633
2634     if (migrate_postcopy_ram()) {
2635         /*
2636          * Although this ping is just for debug, it could potentially be
2637          * used for getting a better measurement of downtime at the source.
2638          */
2639         qemu_savevm_send_ping(ms->to_dst_file, 4);
2640     }
2641
2642     if (migrate_release_ram()) {
2643         ram_postcopy_migrated_memory_release(ms);
2644     }
2645
2646     ret = qemu_file_get_error(ms->to_dst_file);
2647     if (ret) {
2648         error_report("postcopy_start: Migration stream errored");
2649         migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2650                               MIGRATION_STATUS_FAILED);
2651     }
2652
2653     return ret;
2654
2655 fail_closefb:
2656     qemu_fclose(fb);
2657 fail:
2658     migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
2659                           MIGRATION_STATUS_FAILED);
2660     if (restart_block) {
2661         /* A failure happened early enough that we know the destination hasn't
2662          * accessed block devices, so we're safe to recover.
2663          */
2664         Error *local_err = NULL;
2665
2666         bdrv_invalidate_cache_all(&local_err);
2667         if (local_err) {
2668             error_report_err(local_err);
2669         }
2670     }
2671     qemu_mutex_unlock_iothread();
2672     return -1;
2673 }
2674
2675 /**
2676  * migration_maybe_pause: Pause if required to by
2677  * migrate_pause_before_switchover called with the iothread locked
2678  * Returns: 0 on success
2679  */
2680 static int migration_maybe_pause(MigrationState *s,
2681                                  int *current_active_state,
2682                                  int new_state)
2683 {
2684     if (!migrate_pause_before_switchover()) {
2685         return 0;
2686     }
2687
2688     /* Since leaving this state is not atomic with posting the semaphore
2689      * it's possible that someone could have issued multiple migrate_continue
2690      * and the semaphore is incorrectly positive at this point;
2691      * the docs say it's undefined to reinit a semaphore that's already
2692      * init'd, so use timedwait to eat up any existing posts.
2693      */
2694     while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
2695         /* This block intentionally left blank */
2696     }
2697
2698     qemu_mutex_unlock_iothread();
2699     migrate_set_state(&s->state, *current_active_state,
2700                       MIGRATION_STATUS_PRE_SWITCHOVER);
2701     qemu_sem_wait(&s->pause_sem);
2702     migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
2703                       new_state);
2704     *current_active_state = new_state;
2705     qemu_mutex_lock_iothread();
2706
2707     return s->state == new_state ? 0 : -EINVAL;
2708 }
2709
2710 /**
2711  * migration_completion: Used by migration_thread when there's not much left.
2712  *   The caller 'breaks' the loop when this returns.
2713  *
2714  * @s: Current migration state
2715  */
2716 static void migration_completion(MigrationState *s)
2717 {
2718     int ret;
2719     int current_active_state = s->state;
2720
2721     if (s->state == MIGRATION_STATUS_ACTIVE) {
2722         qemu_mutex_lock_iothread();
2723         s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2724         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2725         s->vm_was_running = runstate_is_running();
2726         ret = global_state_store();
2727
2728         if (!ret) {
2729             bool inactivate = !migrate_colo_enabled();
2730             ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2731             if (ret >= 0) {
2732                 ret = migration_maybe_pause(s, &current_active_state,
2733                                             MIGRATION_STATUS_DEVICE);
2734             }
2735             if (ret >= 0) {
2736                 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
2737                 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
2738                                                          inactivate);
2739             }
2740             if (inactivate && ret >= 0) {
2741                 s->block_inactive = true;
2742             }
2743         }
2744         qemu_mutex_unlock_iothread();
2745
2746         if (ret < 0) {
2747             goto fail;
2748         }
2749     } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2750         trace_migration_completion_postcopy_end();
2751
2752         qemu_savevm_state_complete_postcopy(s->to_dst_file);
2753         trace_migration_completion_postcopy_end_after_complete();
2754     }
2755
2756     /*
2757      * If rp was opened we must clean up the thread before
2758      * cleaning everything else up (since if there are no failures
2759      * it will wait for the destination to send it's status in
2760      * a SHUT command).
2761      */
2762     if (s->rp_state.from_dst_file) {
2763         int rp_error;
2764         trace_migration_return_path_end_before();
2765         rp_error = await_return_path_close_on_source(s);
2766         trace_migration_return_path_end_after(rp_error);
2767         if (rp_error) {
2768             goto fail_invalidate;
2769         }
2770     }
2771
2772     if (qemu_file_get_error(s->to_dst_file)) {
2773         trace_migration_completion_file_err();
2774         goto fail_invalidate;
2775     }
2776
2777     if (!migrate_colo_enabled()) {
2778         migrate_set_state(&s->state, current_active_state,
2779                           MIGRATION_STATUS_COMPLETED);
2780     }
2781
2782     return;
2783
2784 fail_invalidate:
2785     /* If not doing postcopy, vm_start() will be called: let's regain
2786      * control on images.
2787      */
2788     if (s->state == MIGRATION_STATUS_ACTIVE ||
2789         s->state == MIGRATION_STATUS_DEVICE) {
2790         Error *local_err = NULL;
2791
2792         qemu_mutex_lock_iothread();
2793         bdrv_invalidate_cache_all(&local_err);
2794         if (local_err) {
2795             error_report_err(local_err);
2796         } else {
2797             s->block_inactive = false;
2798         }
2799         qemu_mutex_unlock_iothread();
2800     }
2801
2802 fail:
2803     migrate_set_state(&s->state, current_active_state,
2804                       MIGRATION_STATUS_FAILED);
2805 }
2806
2807 bool migrate_colo_enabled(void)
2808 {
2809     MigrationState *s = migrate_get_current();
2810     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
2811 }
2812
2813 typedef enum MigThrError {
2814     /* No error detected */
2815     MIG_THR_ERR_NONE = 0,
2816     /* Detected error, but resumed successfully */
2817     MIG_THR_ERR_RECOVERED = 1,
2818     /* Detected fatal error, need to exit */
2819     MIG_THR_ERR_FATAL = 2,
2820 } MigThrError;
2821
2822 static int postcopy_resume_handshake(MigrationState *s)
2823 {
2824     qemu_savevm_send_postcopy_resume(s->to_dst_file);
2825
2826     while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
2827         qemu_sem_wait(&s->rp_state.rp_sem);
2828     }
2829
2830     if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2831         return 0;
2832     }
2833
2834     return -1;
2835 }
2836
2837 /* Return zero if success, or <0 for error */
2838 static int postcopy_do_resume(MigrationState *s)
2839 {
2840     int ret;
2841
2842     /*
2843      * Call all the resume_prepare() hooks, so that modules can be
2844      * ready for the migration resume.
2845      */
2846     ret = qemu_savevm_state_resume_prepare(s);
2847     if (ret) {
2848         error_report("%s: resume_prepare() failure detected: %d",
2849                      __func__, ret);
2850         return ret;
2851     }
2852
2853     /*
2854      * Last handshake with destination on the resume (destination will
2855      * switch to postcopy-active afterwards)
2856      */
2857     ret = postcopy_resume_handshake(s);
2858     if (ret) {
2859         error_report("%s: handshake failed: %d", __func__, ret);
2860         return ret;
2861     }
2862
2863     return 0;
2864 }
2865
2866 /*
2867  * We don't return until we are in a safe state to continue current
2868  * postcopy migration.  Returns MIG_THR_ERR_RECOVERED if recovered, or
2869  * MIG_THR_ERR_FATAL if unrecovery failure happened.
2870  */
2871 static MigThrError postcopy_pause(MigrationState *s)
2872 {
2873     assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
2874
2875     while (true) {
2876         QEMUFile *file;
2877
2878         migrate_set_state(&s->state, s->state,
2879                           MIGRATION_STATUS_POSTCOPY_PAUSED);
2880
2881         /* Current channel is possibly broken. Release it. */
2882         assert(s->to_dst_file);
2883         qemu_mutex_lock(&s->qemu_file_lock);
2884         file = s->to_dst_file;
2885         s->to_dst_file = NULL;
2886         qemu_mutex_unlock(&s->qemu_file_lock);
2887
2888         qemu_file_shutdown(file);
2889         qemu_fclose(file);
2890
2891         error_report("Detected IO failure for postcopy. "
2892                      "Migration paused.");
2893
2894         /*
2895          * We wait until things fixed up. Then someone will setup the
2896          * status back for us.
2897          */
2898         while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
2899             qemu_sem_wait(&s->postcopy_pause_sem);
2900         }
2901
2902         if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
2903             /* Woken up by a recover procedure. Give it a shot */
2904
2905             /*
2906              * Firstly, let's wake up the return path now, with a new
2907              * return path channel.
2908              */
2909             qemu_sem_post(&s->postcopy_pause_rp_sem);
2910
2911             /* Do the resume logic */
2912             if (postcopy_do_resume(s) == 0) {
2913                 /* Let's continue! */
2914                 trace_postcopy_pause_continued();
2915                 return MIG_THR_ERR_RECOVERED;
2916             } else {
2917                 /*
2918                  * Something wrong happened during the recovery, let's
2919                  * pause again. Pause is always better than throwing
2920                  * data away.
2921                  */
2922                 continue;
2923             }
2924         } else {
2925             /* This is not right... Time to quit. */
2926             return MIG_THR_ERR_FATAL;
2927         }
2928     }
2929 }
2930
2931 static MigThrError migration_detect_error(MigrationState *s)
2932 {
2933     int ret;
2934     int state = s->state;
2935
2936     if (state == MIGRATION_STATUS_CANCELLING ||
2937         state == MIGRATION_STATUS_CANCELLED) {
2938         /* End the migration, but don't set the state to failed */
2939         return MIG_THR_ERR_FATAL;
2940     }
2941
2942     /* Try to detect any file errors */
2943     ret = qemu_file_get_error(s->to_dst_file);
2944
2945     if (!ret) {
2946         /* Everything is fine */
2947         return MIG_THR_ERR_NONE;
2948     }
2949
2950     if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret == -EIO) {
2951         /*
2952          * For postcopy, we allow the network to be down for a
2953          * while. After that, it can be continued by a
2954          * recovery phase.
2955          */
2956         return postcopy_pause(s);
2957     } else {
2958         /*
2959          * For precopy (or postcopy with error outside IO), we fail
2960          * with no time.
2961          */
2962         migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED);
2963         trace_migration_thread_file_err();
2964
2965         /* Time to stop the migration, now. */
2966         return MIG_THR_ERR_FATAL;
2967     }
2968 }
2969
2970 /* How many bytes have we transferred since the beggining of the migration */
2971 static uint64_t migration_total_bytes(MigrationState *s)
2972 {
2973     return qemu_ftell(s->to_dst_file) + ram_counters.multifd_bytes;
2974 }
2975
2976 static void migration_calculate_complete(MigrationState *s)
2977 {
2978     uint64_t bytes = migration_total_bytes(s);
2979     int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2980     int64_t transfer_time;
2981
2982     s->total_time = end_time - s->start_time;
2983     if (!s->downtime) {
2984         /*
2985          * It's still not set, so we are precopy migration.  For
2986          * postcopy, downtime is calculated during postcopy_start().
2987          */
2988         s->downtime = end_time - s->downtime_start;
2989     }
2990
2991     transfer_time = s->total_time - s->setup_time;
2992     if (transfer_time) {
2993         s->mbps = ((double) bytes * 8.0) / transfer_time / 1000;
2994     }
2995 }
2996
2997 static void migration_update_counters(MigrationState *s,
2998                                       int64_t current_time)
2999 {
3000     uint64_t transferred, transferred_pages, time_spent;
3001     uint64_t current_bytes; /* bytes transferred since the beginning */
3002     double bandwidth;
3003
3004     if (current_time < s->iteration_start_time + BUFFER_DELAY) {
3005         return;
3006     }
3007
3008     current_bytes = migration_total_bytes(s);
3009     transferred = current_bytes - s->iteration_initial_bytes;
3010     time_spent = current_time - s->iteration_start_time;
3011     bandwidth = (double)transferred / time_spent;
3012     s->threshold_size = bandwidth * s->parameters.downtime_limit;
3013
3014     s->mbps = (((double) transferred * 8.0) /
3015                ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
3016
3017     transferred_pages = ram_get_total_transferred_pages() -
3018                             s->iteration_initial_pages;
3019     s->pages_per_second = (double) transferred_pages /
3020                              (((double) time_spent / 1000.0));
3021
3022     /*
3023      * if we haven't sent anything, we don't want to
3024      * recalculate. 10000 is a small enough number for our purposes
3025      */
3026     if (ram_counters.dirty_pages_rate && transferred > 10000) {
3027         s->expected_downtime = ram_counters.remaining / bandwidth;
3028     }
3029
3030     qemu_file_reset_rate_limit(s->to_dst_file);
3031
3032     s->iteration_start_time = current_time;
3033     s->iteration_initial_bytes = current_bytes;
3034     s->iteration_initial_pages = ram_get_total_transferred_pages();
3035
3036     trace_migrate_transferred(transferred, time_spent,
3037                               bandwidth, s->threshold_size);
3038 }
3039
3040 /* Migration thread iteration status */
3041 typedef enum {
3042     MIG_ITERATE_RESUME,         /* Resume current iteration */
3043     MIG_ITERATE_SKIP,           /* Skip current iteration */
3044     MIG_ITERATE_BREAK,          /* Break the loop */
3045 } MigIterateState;
3046
3047 /*
3048  * Return true if continue to the next iteration directly, false
3049  * otherwise.
3050  */
3051 static MigIterateState migration_iteration_run(MigrationState *s)
3052 {
3053     uint64_t pending_size, pend_pre, pend_compat, pend_post;
3054     bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
3055
3056     qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre,
3057                               &pend_compat, &pend_post);
3058     pending_size = pend_pre + pend_compat + pend_post;
3059
3060     trace_migrate_pending(pending_size, s->threshold_size,
3061                           pend_pre, pend_compat, pend_post);
3062
3063     if (pending_size && pending_size >= s->threshold_size) {
3064         /* Still a significant amount to transfer */
3065         if (migrate_postcopy() && !in_postcopy &&
3066             pend_pre <= s->threshold_size &&
3067             atomic_read(&s->start_postcopy)) {
3068             if (postcopy_start(s)) {
3069                 error_report("%s: postcopy failed to start", __func__);
3070             }
3071             return MIG_ITERATE_SKIP;
3072         }
3073         /* Just another iteration step */
3074         qemu_savevm_state_iterate(s->to_dst_file,
3075             s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
3076     } else {
3077         trace_migration_thread_low_pending(pending_size);
3078         migration_completion(s);
3079         return MIG_ITERATE_BREAK;
3080     }
3081
3082     return MIG_ITERATE_RESUME;
3083 }
3084
3085 static void migration_iteration_finish(MigrationState *s)
3086 {
3087     /* If we enabled cpu throttling for auto-converge, turn it off. */
3088     cpu_throttle_stop();
3089
3090     qemu_mutex_lock_iothread();
3091     switch (s->state) {
3092     case MIGRATION_STATUS_COMPLETED:
3093         migration_calculate_complete(s);
3094         runstate_set(RUN_STATE_POSTMIGRATE);
3095         break;
3096
3097     case MIGRATION_STATUS_ACTIVE:
3098         /*
3099          * We should really assert here, but since it's during
3100          * migration, let's try to reduce the usage of assertions.
3101          */
3102         if (!migrate_colo_enabled()) {
3103             error_report("%s: critical error: calling COLO code without "
3104                          "COLO enabled", __func__);
3105         }
3106         migrate_start_colo_process(s);
3107         /*
3108          * Fixme: we will run VM in COLO no matter its old running state.
3109          * After exited COLO, we will keep running.
3110          */
3111         s->vm_was_running = true;
3112         /* Fallthrough */
3113     case MIGRATION_STATUS_FAILED:
3114     case MIGRATION_STATUS_CANCELLED:
3115     case MIGRATION_STATUS_CANCELLING:
3116         if (s->vm_was_running) {
3117             vm_start();
3118         } else {
3119             if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
3120                 runstate_set(RUN_STATE_POSTMIGRATE);
3121             }
3122         }
3123         break;
3124
3125     default:
3126         /* Should not reach here, but if so, forgive the VM. */
3127         error_report("%s: Unknown ending state %d", __func__, s->state);
3128         break;
3129     }
3130     qemu_bh_schedule(s->cleanup_bh);
3131     qemu_mutex_unlock_iothread();
3132 }
3133
3134 void migration_make_urgent_request(void)
3135 {
3136     qemu_sem_post(&migrate_get_current()->rate_limit_sem);
3137 }
3138
3139 void migration_consume_urgent_request(void)
3140 {
3141     qemu_sem_wait(&migrate_get_current()->rate_limit_sem);
3142 }
3143
3144 /*
3145  * Master migration thread on the source VM.
3146  * It drives the migration and pumps the data down the outgoing channel.
3147  */
3148 static void *migration_thread(void *opaque)
3149 {
3150     MigrationState *s = opaque;
3151     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
3152     MigThrError thr_error;
3153     bool urgent = false;
3154
3155     rcu_register_thread();
3156
3157     object_ref(OBJECT(s));
3158     s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3159
3160     qemu_savevm_state_header(s->to_dst_file);
3161
3162     /*
3163      * If we opened the return path, we need to make sure dst has it
3164      * opened as well.
3165      */
3166     if (s->rp_state.from_dst_file) {
3167         /* Now tell the dest that it should open its end so it can reply */
3168         qemu_savevm_send_open_return_path(s->to_dst_file);
3169
3170         /* And do a ping that will make stuff easier to debug */
3171         qemu_savevm_send_ping(s->to_dst_file, 1);
3172     }
3173
3174     if (migrate_postcopy()) {
3175         /*
3176          * Tell the destination that we *might* want to do postcopy later;
3177          * if the other end can't do postcopy it should fail now, nice and
3178          * early.
3179          */
3180         qemu_savevm_send_postcopy_advise(s->to_dst_file);
3181     }
3182
3183     if (migrate_colo_enabled()) {
3184         /* Notify migration destination that we enable COLO */
3185         qemu_savevm_send_colo_enable(s->to_dst_file);
3186     }
3187
3188     qemu_savevm_state_setup(s->to_dst_file);
3189
3190     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
3191     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3192                       MIGRATION_STATUS_ACTIVE);
3193
3194     trace_migration_thread_setup_complete();
3195
3196     while (s->state == MIGRATION_STATUS_ACTIVE ||
3197            s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
3198         int64_t current_time;
3199
3200         if (urgent || !qemu_file_rate_limit(s->to_dst_file)) {
3201             MigIterateState iter_state = migration_iteration_run(s);
3202             if (iter_state == MIG_ITERATE_SKIP) {
3203                 continue;
3204             } else if (iter_state == MIG_ITERATE_BREAK) {
3205                 break;
3206             }
3207         }
3208
3209         /*
3210          * Try to detect any kind of failures, and see whether we
3211          * should stop the migration now.
3212          */
3213         thr_error = migration_detect_error(s);
3214         if (thr_error == MIG_THR_ERR_FATAL) {
3215             /* Stop migration */
3216             break;
3217         } else if (thr_error == MIG_THR_ERR_RECOVERED) {
3218             /*
3219              * Just recovered from a e.g. network failure, reset all
3220              * the local variables. This is important to avoid
3221              * breaking transferred_bytes and bandwidth calculation
3222              */
3223             s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3224             s->iteration_initial_bytes = 0;
3225         }
3226
3227         current_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3228
3229         migration_update_counters(s, current_time);
3230
3231         urgent = false;
3232         if (qemu_file_rate_limit(s->to_dst_file)) {
3233             /* Wait for a delay to do rate limiting OR
3234              * something urgent to post the semaphore.
3235              */
3236             int ms = s->iteration_start_time + BUFFER_DELAY - current_time;
3237             trace_migration_thread_ratelimit_pre(ms);
3238             if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) {
3239                 /* We were worken by one or more urgent things but
3240                  * the timedwait will have consumed one of them.
3241                  * The service routine for the urgent wake will dec
3242                  * the semaphore itself for each item it consumes,
3243                  * so add this one we just eat back.
3244                  */
3245                 qemu_sem_post(&s->rate_limit_sem);
3246                 urgent = true;
3247             }
3248             trace_migration_thread_ratelimit_post(urgent);
3249         }
3250     }
3251
3252     trace_migration_thread_after_loop();
3253     migration_iteration_finish(s);
3254     object_unref(OBJECT(s));
3255     rcu_unregister_thread();
3256     return NULL;
3257 }
3258
3259 void migrate_fd_connect(MigrationState *s, Error *error_in)
3260 {
3261     int64_t rate_limit;
3262     bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED;
3263
3264     s->expected_downtime = s->parameters.downtime_limit;
3265     s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup, s);
3266     if (error_in) {
3267         migrate_fd_error(s, error_in);
3268         migrate_fd_cleanup(s);
3269         return;
3270     }
3271
3272     if (resume) {
3273         /* This is a resumed migration */
3274         rate_limit = INT64_MAX;
3275     } else {
3276         /* This is a fresh new migration */
3277         rate_limit = s->parameters.max_bandwidth / XFER_LIMIT_RATIO;
3278
3279         /* Notify before starting migration thread */
3280         notifier_list_notify(&migration_state_notifiers, s);
3281     }
3282
3283     qemu_file_set_rate_limit(s->to_dst_file, rate_limit);
3284     qemu_file_set_blocking(s->to_dst_file, true);
3285
3286     /*
3287      * Open the return path. For postcopy, it is used exclusively. For
3288      * precopy, only if user specified "return-path" capability would
3289      * QEMU uses the return path.
3290      */
3291     if (migrate_postcopy_ram() || migrate_use_return_path()) {
3292         if (open_return_path_on_source(s, !resume)) {
3293             error_report("Unable to open return-path for postcopy");
3294             migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED);
3295             migrate_fd_cleanup(s);
3296             return;
3297         }
3298     }
3299
3300     if (resume) {
3301         /* Wakeup the main migration thread to do the recovery */
3302         migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
3303                           MIGRATION_STATUS_POSTCOPY_RECOVER);
3304         qemu_sem_post(&s->postcopy_pause_sem);
3305         return;
3306     }
3307
3308     if (multifd_save_setup() != 0) {
3309         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
3310                           MIGRATION_STATUS_FAILED);
3311         migrate_fd_cleanup(s);
3312         return;
3313     }
3314     qemu_thread_create(&s->thread, "live_migration", migration_thread, s,
3315                        QEMU_THREAD_JOINABLE);
3316     s->migration_thread_running = true;
3317 }
3318
3319 void migration_global_dump(Monitor *mon)
3320 {
3321     MigrationState *ms = migrate_get_current();
3322
3323     monitor_printf(mon, "globals:\n");
3324     monitor_printf(mon, "store-global-state: %s\n",
3325                    ms->store_global_state ? "on" : "off");
3326     monitor_printf(mon, "only-migratable: %s\n",
3327                    ms->only_migratable ? "on" : "off");
3328     monitor_printf(mon, "send-configuration: %s\n",
3329                    ms->send_configuration ? "on" : "off");
3330     monitor_printf(mon, "send-section-footer: %s\n",
3331                    ms->send_section_footer ? "on" : "off");
3332     monitor_printf(mon, "decompress-error-check: %s\n",
3333                    ms->decompress_error_check ? "on" : "off");
3334 }
3335
3336 #define DEFINE_PROP_MIG_CAP(name, x)             \
3337     DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3338
3339 static Property migration_properties[] = {
3340     DEFINE_PROP_BOOL("store-global-state", MigrationState,
3341                      store_global_state, true),
3342     DEFINE_PROP_BOOL("only-migratable", MigrationState, only_migratable, false),
3343     DEFINE_PROP_BOOL("send-configuration", MigrationState,
3344                      send_configuration, true),
3345     DEFINE_PROP_BOOL("send-section-footer", MigrationState,
3346                      send_section_footer, true),
3347     DEFINE_PROP_BOOL("decompress-error-check", MigrationState,
3348                       decompress_error_check, true),
3349
3350     /* Migration parameters */
3351     DEFINE_PROP_UINT8("x-compress-level", MigrationState,
3352                       parameters.compress_level,
3353                       DEFAULT_MIGRATE_COMPRESS_LEVEL),
3354     DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
3355                       parameters.compress_threads,
3356                       DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
3357     DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState,
3358                       parameters.compress_wait_thread, true),
3359     DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
3360                       parameters.decompress_threads,
3361                       DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
3362     DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
3363                       parameters.cpu_throttle_initial,
3364                       DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
3365     DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
3366                       parameters.cpu_throttle_increment,
3367                       DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
3368     DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
3369                       parameters.max_bandwidth, MAX_THROTTLE),
3370     DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
3371                       parameters.downtime_limit,
3372                       DEFAULT_MIGRATE_SET_DOWNTIME),
3373     DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
3374                       parameters.x_checkpoint_delay,
3375                       DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
3376     DEFINE_PROP_UINT8("x-multifd-channels", MigrationState,
3377                       parameters.x_multifd_channels,
3378                       DEFAULT_MIGRATE_MULTIFD_CHANNELS),
3379     DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState,
3380                       parameters.x_multifd_page_count,
3381                       DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT),
3382     DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
3383                       parameters.xbzrle_cache_size,
3384                       DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
3385     DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState,
3386                       parameters.max_postcopy_bandwidth,
3387                       DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH),
3388     DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState,
3389                       parameters.max_cpu_throttle,
3390                       DEFAULT_MIGRATE_MAX_CPU_THROTTLE),
3391     DEFINE_PROP_SIZE("announce-initial", MigrationState,
3392                       parameters.announce_initial,
3393                       DEFAULT_MIGRATE_ANNOUNCE_INITIAL),
3394     DEFINE_PROP_SIZE("announce-max", MigrationState,
3395                       parameters.announce_max,
3396                       DEFAULT_MIGRATE_ANNOUNCE_MAX),
3397     DEFINE_PROP_SIZE("announce-rounds", MigrationState,
3398                       parameters.announce_rounds,
3399                       DEFAULT_MIGRATE_ANNOUNCE_ROUNDS),
3400     DEFINE_PROP_SIZE("announce-step", MigrationState,
3401                       parameters.announce_step,
3402                       DEFAULT_MIGRATE_ANNOUNCE_STEP),
3403
3404     /* Migration capabilities */
3405     DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
3406     DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
3407     DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
3408     DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
3409     DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
3410     DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
3411     DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
3412     DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
3413     DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
3414     DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
3415     DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
3416     DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD),
3417
3418     DEFINE_PROP_END_OF_LIST(),
3419 };
3420
3421 static void migration_class_init(ObjectClass *klass, void *data)
3422 {
3423     DeviceClass *dc = DEVICE_CLASS(klass);
3424
3425     dc->user_creatable = false;
3426     dc->props = migration_properties;
3427 }
3428
3429 static void migration_instance_finalize(Object *obj)
3430 {
3431     MigrationState *ms = MIGRATION_OBJ(obj);
3432     MigrationParameters *params = &ms->parameters;
3433
3434     qemu_mutex_destroy(&ms->error_mutex);
3435     qemu_mutex_destroy(&ms->qemu_file_lock);
3436     g_free(params->tls_hostname);
3437     g_free(params->tls_creds);
3438     qemu_sem_destroy(&ms->rate_limit_sem);
3439     qemu_sem_destroy(&ms->pause_sem);
3440     qemu_sem_destroy(&ms->postcopy_pause_sem);
3441     qemu_sem_destroy(&ms->postcopy_pause_rp_sem);
3442     qemu_sem_destroy(&ms->rp_state.rp_sem);
3443     error_free(ms->error);
3444 }
3445
3446 static void migration_instance_init(Object *obj)
3447 {
3448     MigrationState *ms = MIGRATION_OBJ(obj);
3449     MigrationParameters *params = &ms->parameters;
3450
3451     ms->state = MIGRATION_STATUS_NONE;
3452     ms->mbps = -1;
3453     ms->pages_per_second = -1;
3454     qemu_sem_init(&ms->pause_sem, 0);
3455     qemu_mutex_init(&ms->error_mutex);
3456
3457     params->tls_hostname = g_strdup("");
3458     params->tls_creds = g_strdup("");
3459
3460     /* Set has_* up only for parameter checks */
3461     params->has_compress_level = true;
3462     params->has_compress_threads = true;
3463     params->has_decompress_threads = true;
3464     params->has_cpu_throttle_initial = true;
3465     params->has_cpu_throttle_increment = true;
3466     params->has_max_bandwidth = true;
3467     params->has_downtime_limit = true;
3468     params->has_x_checkpoint_delay = true;
3469     params->has_block_incremental = true;
3470     params->has_x_multifd_channels = true;
3471     params->has_x_multifd_page_count = true;
3472     params->has_xbzrle_cache_size = true;
3473     params->has_max_postcopy_bandwidth = true;
3474     params->has_max_cpu_throttle = true;
3475     params->has_announce_initial = true;
3476     params->has_announce_max = true;
3477     params->has_announce_rounds = true;
3478     params->has_announce_step = true;
3479
3480     qemu_sem_init(&ms->postcopy_pause_sem, 0);
3481     qemu_sem_init(&ms->postcopy_pause_rp_sem, 0);
3482     qemu_sem_init(&ms->rp_state.rp_sem, 0);
3483     qemu_sem_init(&ms->rate_limit_sem, 0);
3484     qemu_mutex_init(&ms->qemu_file_lock);
3485 }
3486
3487 /*
3488  * Return true if check pass, false otherwise. Error will be put
3489  * inside errp if provided.
3490  */
3491 static bool migration_object_check(MigrationState *ms, Error **errp)
3492 {
3493     MigrationCapabilityStatusList *head = NULL;
3494     /* Assuming all off */
3495     bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
3496     int i;
3497
3498     if (!migrate_params_check(&ms->parameters, errp)) {
3499         return false;
3500     }
3501
3502     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
3503         if (ms->enabled_capabilities[i]) {
3504             head = migrate_cap_add(head, i, true);
3505         }
3506     }
3507
3508     ret = migrate_caps_check(cap_list, head, errp);
3509
3510     /* It works with head == NULL */
3511     qapi_free_MigrationCapabilityStatusList(head);
3512
3513     return ret;
3514 }
3515
3516 static const TypeInfo migration_type = {
3517     .name = TYPE_MIGRATION,
3518     /*
3519      * NOTE: TYPE_MIGRATION is not really a device, as the object is
3520      * not created using qdev_create(), it is not attached to the qdev
3521      * device tree, and it is never realized.
3522      *
3523      * TODO: Make this TYPE_OBJECT once QOM provides something like
3524      * TYPE_DEVICE's "-global" properties.
3525      */
3526     .parent = TYPE_DEVICE,
3527     .class_init = migration_class_init,
3528     .class_size = sizeof(MigrationClass),
3529     .instance_size = sizeof(MigrationState),
3530     .instance_init = migration_instance_init,
3531     .instance_finalize = migration_instance_finalize,
3532 };
3533
3534 static void register_migration_types(void)
3535 {
3536     type_register_static(&migration_type);
3537 }
3538
3539 type_init(register_migration_types);