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