migration/migration.c

   1 /*
   2  * QEMU live migration
   3  *
   4  * Copyright IBM, Corp. 2008
   5  *
   6  * Authors:
   7  *  Anthony Liguori   <aliguori@us.ibm.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2.  See
  10  * the COPYING file in the top-level directory.
  11  *
  12  * Contributions after 2012-01-13 are licensed under the terms of the
  13  * GNU GPL, version 2 or (at your option) any later version.
  14  */
  15
  16 #include "qemu/osdep.h"
  17 #include "qemu/cutils.h"
  18 #include "qemu/error-report.h"
  19 #include "qemu/main-loop.h"
  20 #include "migration/blocker.h"
  21 #include "exec.h"
  22 #include "fd.h"
  23 #include "socket.h"
  24 #include "sysemu/runstate.h"
  25 #include "sysemu/sysemu.h"
  26 #include "sysemu/cpu-throttle.h"
  27 #include "rdma.h"
  28 #include "ram.h"
  29 #include "migration/global_state.h"
  30 #include "migration/misc.h"
  31 #include "migration.h"
  32 #include "savevm.h"
  33 #include "qemu-file-channel.h"
  34 #include "qemu-file.h"
  35 #include "migration/vmstate.h"
  36 #include "block/block.h"
  37 #include "qapi/error.h"
  38 #include "qapi/clone-visitor.h"
  39 #include "qapi/qapi-visit-migration.h"
  40 #include "qapi/qapi-visit-sockets.h"
  41 #include "qapi/qapi-commands-migration.h"
  42 #include "qapi/qapi-events-migration.h"
  43 #include "qapi/qmp/qerror.h"
  44 #include "qapi/qmp/qnull.h"
  45 #include "qemu/rcu.h"
  46 #include "block.h"
  47 #include "postcopy-ram.h"
  48 #include "qemu/thread.h"
  49 #include "trace.h"
  50 #include "exec/target_page.h"
  51 #include "io/channel-buffer.h"
  52 #include "migration/colo.h"
  53 #include "hw/boards.h"
  54 #include "hw/qdev-properties.h"
  55 #include "hw/qdev-properties-system.h"
  56 #include "monitor/monitor.h"
  57 #include "net/announce.h"
  58 #include "qemu/queue.h"
  59 #include "multifd.h"
  60 #include "qemu/yank.h"
  61 #include "sysemu/cpus.h"
  62 #include "yank_functions.h"
  63 #include "sysemu/qtest.h"
  64
  65 #define MAX_THROTTLE  (128 << 20)      /* Migration transfer speed throttling */
  66
  67 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
  68  * data. */
  69 #define BUFFER_DELAY     100
  70 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
  71
  72 /* Time in milliseconds we are allowed to stop the source,
  73  * for sending the last part */
  74 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
  75
  76 /* Maximum migrate downtime set to 2000 seconds */
  77 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
  78 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
  79
  80 /* Default compression thread count */
  81 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
  82 /* Default decompression thread count, usually decompression is at
  83  * least 4 times as fast as compression.*/
  84 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
  85 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
  86 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
  87 /* Define default autoconverge cpu throttle migration parameters */
  88 #define DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD 50
  89 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
  90 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
  91 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
  92
  93 /* Migration XBZRLE default cache size */
  94 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
  95
  96 /* The delay time (in ms) between two COLO checkpoints */
  97 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
  98 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
  99 #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE
 100 /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */
 101 #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
 102 /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
 103 #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
 104
 105 /* Background transfer rate for postcopy, 0 means unlimited, note
 106  * that page requests can still exceed this limit.
 107  */
 108 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
 109
 110 /*
 111  * Parameters for self_announce_delay giving a stream of RARP/ARP
 112  * packets after migration.
 113  */
 114 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL  50
 115 #define DEFAULT_MIGRATE_ANNOUNCE_MAX     550
 116 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS    5
 117 #define DEFAULT_MIGRATE_ANNOUNCE_STEP    100
 118
 119 static NotifierList migration_state_notifiers =
 120     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
 121
 122 /* Messages sent on the return path from destination to source */
 123 enum mig_rp_message_type {
 124     MIG_RP_MSG_INVALID = 0,  /* Must be 0 */
 125     MIG_RP_MSG_SHUT,         /* sibling will not send any more RP messages */
 126     MIG_RP_MSG_PONG,         /* Response to a PING; data (seq: be32 ) */
 127
 128     MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
 129     MIG_RP_MSG_REQ_PAGES,    /* data (start: be64, len: be32) */
 130     MIG_RP_MSG_RECV_BITMAP,  /* send recved_bitmap back to source */
 131     MIG_RP_MSG_RESUME_ACK,   /* tell source that we are ready to resume */
 132
 133     MIG_RP_MSG_MAX
 134 };
 135
 136 /* Migration capabilities set */
 137 struct MigrateCapsSet {
 138     int size;                       /* Capability set size */
 139     MigrationCapability caps[];     /* Variadic array of capabilities */
 140 };
 141 typedef struct MigrateCapsSet MigrateCapsSet;
 142
 143 /* Define and initialize MigrateCapsSet */
 144 #define INITIALIZE_MIGRATE_CAPS_SET(_name, ...)   \
 145     MigrateCapsSet _name = {    \
 146         .size = sizeof((int []) { __VA_ARGS__ }) / sizeof(int), \
 147         .caps = { __VA_ARGS__ } \
 148     }
 149
 150 /* Background-snapshot compatibility check list */
 151 static const
 152 INITIALIZE_MIGRATE_CAPS_SET(check_caps_background_snapshot,
 153     MIGRATION_CAPABILITY_POSTCOPY_RAM,
 154     MIGRATION_CAPABILITY_DIRTY_BITMAPS,
 155     MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME,
 156     MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE,
 157     MIGRATION_CAPABILITY_RETURN_PATH,
 158     MIGRATION_CAPABILITY_MULTIFD,
 159     MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER,
 160     MIGRATION_CAPABILITY_AUTO_CONVERGE,
 161     MIGRATION_CAPABILITY_RELEASE_RAM,
 162     MIGRATION_CAPABILITY_RDMA_PIN_ALL,
 163     MIGRATION_CAPABILITY_COMPRESS,
 164     MIGRATION_CAPABILITY_XBZRLE,
 165     MIGRATION_CAPABILITY_X_COLO,
 166     MIGRATION_CAPABILITY_VALIDATE_UUID);
 167
 168 /* When we add fault tolerance, we could have several
 169    migrations at once.  For now we don't need to add
 170    dynamic creation of migration */
 171
 172 static MigrationState *current_migration;
 173 static MigrationIncomingState *current_incoming;
 174
 175 static GSList *migration_blockers;
 176
 177 static bool migration_object_check(MigrationState *ms, Error **errp);
 178 static int migration_maybe_pause(MigrationState *s,
 179                                  int *current_active_state,
 180                                  int new_state);
 181 static void migrate_fd_cancel(MigrationState *s);
 182
 183 static bool migrate_allow_multi_channels = true;
 184
 185 void migrate_protocol_allow_multi_channels(bool allow)
 186 {
 187     migrate_allow_multi_channels = allow;
 188 }
 189
 190 bool migrate_multi_channels_is_allowed(void)
 191 {
 192     return migrate_allow_multi_channels;
 193 }
 194
 195 static gint page_request_addr_cmp(gconstpointer ap, gconstpointer bp)
 196 {
 197     uintptr_t a = (uintptr_t) ap, b = (uintptr_t) bp;
 198
 199     return (a > b) - (a < b);
 200 }
 201
 202 void migration_object_init(void)
 203 {
 204     /* This can only be called once. */
 205     assert(!current_migration);
 206     current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
 207
 208     /*
 209      * Init the migrate incoming object as well no matter whether
 210      * we'll use it or not.
 211      */
 212     assert(!current_incoming);
 213     current_incoming = g_new0(MigrationIncomingState, 1);
 214     current_incoming->state = MIGRATION_STATUS_NONE;
 215     current_incoming->postcopy_remote_fds =
 216         g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD));
 217     qemu_mutex_init(&current_incoming->rp_mutex);
 218     qemu_event_init(&current_incoming->main_thread_load_event, false);
 219     qemu_sem_init(&current_incoming->postcopy_pause_sem_dst, 0);
 220     qemu_sem_init(&current_incoming->postcopy_pause_sem_fault, 0);
 221     qemu_mutex_init(&current_incoming->page_request_mutex);
 222     current_incoming->page_requested = g_tree_new(page_request_addr_cmp);
 223
 224     migration_object_check(current_migration, &error_fatal);
 225
 226     blk_mig_init();
 227     ram_mig_init();
 228     dirty_bitmap_mig_init();
 229 }
 230
 231 void migration_cancel(const Error *error)
 232 {
 233     if (error) {
 234         migrate_set_error(current_migration, error);
 235     }
 236     migrate_fd_cancel(current_migration);
 237 }
 238
 239 void migration_shutdown(void)
 240 {
 241     /*
 242      * When the QEMU main thread exit, the COLO thread
 243      * may wait a semaphore. So, we should wakeup the
 244      * COLO thread before migration shutdown.
 245      */
 246     colo_shutdown();
 247     /*
 248      * Cancel the current migration - that will (eventually)
 249      * stop the migration using this structure
 250      */
 251     migration_cancel(NULL);
 252     object_unref(OBJECT(current_migration));
 253
 254     /*
 255      * Cancel outgoing migration of dirty bitmaps. It should
 256      * at least unref used block nodes.
 257      */
 258     dirty_bitmap_mig_cancel_outgoing();
 259
 260     /*
 261      * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
 262      * are non-critical data, and their loss never considered as
 263      * something serious.
 264      */
 265     dirty_bitmap_mig_cancel_incoming();
 266 }
 267
 268 /* For outgoing */
 269 MigrationState *migrate_get_current(void)
 270 {
 271     /* This can only be called after the object created. */
 272     assert(current_migration);
 273     return current_migration;
 274 }
 275
 276 MigrationIncomingState *migration_incoming_get_current(void)
 277 {
 278     assert(current_incoming);
 279     return current_incoming;
 280 }
 281
 282 void migration_incoming_transport_cleanup(MigrationIncomingState *mis)
 283 {
 284     if (mis->socket_address_list) {
 285         qapi_free_SocketAddressList(mis->socket_address_list);
 286         mis->socket_address_list = NULL;
 287     }
 288
 289     if (mis->transport_cleanup) {
 290         mis->transport_cleanup(mis->transport_data);
 291         mis->transport_data = mis->transport_cleanup = NULL;
 292     }
 293 }
 294
 295 void migration_incoming_state_destroy(void)
 296 {
 297     struct MigrationIncomingState *mis = migration_incoming_get_current();
 298
 299     if (mis->to_src_file) {
 300         /* Tell source that we are done */
 301         migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
 302         qemu_fclose(mis->to_src_file);
 303         mis->to_src_file = NULL;
 304     }
 305
 306     if (mis->from_src_file) {
 307         migration_ioc_unregister_yank_from_file(mis->from_src_file);
 308         qemu_fclose(mis->from_src_file);
 309         mis->from_src_file = NULL;
 310     }
 311     if (mis->postcopy_remote_fds) {
 312         g_array_free(mis->postcopy_remote_fds, TRUE);
 313         mis->postcopy_remote_fds = NULL;
 314     }
 315
 316     migration_incoming_transport_cleanup(mis);
 317     qemu_event_reset(&mis->main_thread_load_event);
 318
 319     if (mis->page_requested) {
 320         g_tree_destroy(mis->page_requested);
 321         mis->page_requested = NULL;
 322     }
 323
 324     yank_unregister_instance(MIGRATION_YANK_INSTANCE);
 325 }
 326
 327 static void migrate_generate_event(int new_state)
 328 {
 329     if (migrate_use_events()) {
 330         qapi_event_send_migration(new_state);
 331     }
 332 }
 333
 334 static bool migrate_late_block_activate(void)
 335 {
 336     MigrationState *s;
 337
 338     s = migrate_get_current();
 339
 340     return s->enabled_capabilities[
 341         MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE];
 342 }
 343
 344 /*
 345  * Send a message on the return channel back to the source
 346  * of the migration.
 347  */
 348 static int migrate_send_rp_message(MigrationIncomingState *mis,
 349                                    enum mig_rp_message_type message_type,
 350                                    uint16_t len, void *data)
 351 {
 352     int ret = 0;
 353
 354     trace_migrate_send_rp_message((int)message_type, len);
 355     QEMU_LOCK_GUARD(&mis->rp_mutex);
 356
 357     /*
 358      * It's possible that the file handle got lost due to network
 359      * failures.
 360      */
 361     if (!mis->to_src_file) {
 362         ret = -EIO;
 363         return ret;
 364     }
 365
 366     qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
 367     qemu_put_be16(mis->to_src_file, len);
 368     qemu_put_buffer(mis->to_src_file, data, len);
 369     qemu_fflush(mis->to_src_file);
 370
 371     /* It's possible that qemu file got error during sending */
 372     ret = qemu_file_get_error(mis->to_src_file);
 373
 374     return ret;
 375 }
 376
 377 /* Request one page from the source VM at the given start address.
 378  *   rb: the RAMBlock to request the page in
 379  *   Start: Address offset within the RB
 380  *   Len: Length in bytes required - must be a multiple of pagesize
 381  */
 382 int migrate_send_rp_message_req_pages(MigrationIncomingState *mis,
 383                                       RAMBlock *rb, ram_addr_t start)
 384 {
 385     uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
 386     size_t msglen = 12; /* start + len */
 387     size_t len = qemu_ram_pagesize(rb);
 388     enum mig_rp_message_type msg_type;
 389     const char *rbname;
 390     int rbname_len;
 391
 392     *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
 393     *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
 394
 395     /*
 396      * We maintain the last ramblock that we requested for page.  Note that we
 397      * don't need locking because this function will only be called within the
 398      * postcopy ram fault thread.
 399      */
 400     if (rb != mis->last_rb) {
 401         mis->last_rb = rb;
 402
 403         rbname = qemu_ram_get_idstr(rb);
 404         rbname_len = strlen(rbname);
 405
 406         assert(rbname_len < 256);
 407
 408         bufc[msglen++] = rbname_len;
 409         memcpy(bufc + msglen, rbname, rbname_len);
 410         msglen += rbname_len;
 411         msg_type = MIG_RP_MSG_REQ_PAGES_ID;
 412     } else {
 413         msg_type = MIG_RP_MSG_REQ_PAGES;
 414     }
 415
 416     return migrate_send_rp_message(mis, msg_type, msglen, bufc);
 417 }
 418
 419 int migrate_send_rp_req_pages(MigrationIncomingState *mis,
 420                               RAMBlock *rb, ram_addr_t start, uint64_t haddr)
 421 {
 422     void *aligned = (void *)(uintptr_t)ROUND_DOWN(haddr, qemu_ram_pagesize(rb));
 423     bool received = false;
 424
 425     WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) {
 426         received = ramblock_recv_bitmap_test_byte_offset(rb, start);
 427         if (!received && !g_tree_lookup(mis->page_requested, aligned)) {
 428             /*
 429              * The page has not been received, and it's not yet in the page
 430              * request list.  Queue it.  Set the value of element to 1, so that
 431              * things like g_tree_lookup() will return TRUE (1) when found.
 432              */
 433             g_tree_insert(mis->page_requested, aligned, (gpointer)1);
 434             mis->page_requested_count++;
 435             trace_postcopy_page_req_add(aligned, mis->page_requested_count);
 436         }
 437     }
 438
 439     /*
 440      * If the page is there, skip sending the message.  We don't even need the
 441      * lock because as long as the page arrived, it'll be there forever.
 442      */
 443     if (received) {
 444         return 0;
 445     }
 446
 447     return migrate_send_rp_message_req_pages(mis, rb, start);
 448 }
 449
 450 static bool migration_colo_enabled;
 451 bool migration_incoming_colo_enabled(void)
 452 {
 453     return migration_colo_enabled;
 454 }
 455
 456 void migration_incoming_disable_colo(void)
 457 {
 458     ram_block_discard_disable(false);
 459     migration_colo_enabled = false;
 460 }
 461
 462 int migration_incoming_enable_colo(void)
 463 {
 464     if (ram_block_discard_disable(true)) {
 465         error_report("COLO: cannot disable RAM discard");
 466         return -EBUSY;
 467     }
 468     migration_colo_enabled = true;
 469     return 0;
 470 }
 471
 472 void migrate_add_address(SocketAddress *address)
 473 {
 474     MigrationIncomingState *mis = migration_incoming_get_current();
 475
 476     QAPI_LIST_PREPEND(mis->socket_address_list,
 477                       QAPI_CLONE(SocketAddress, address));
 478 }
 479
 480 static void qemu_start_incoming_migration(const char *uri, Error **errp)
 481 {
 482     const char *p = NULL;
 483
 484     migrate_protocol_allow_multi_channels(false); /* reset it anyway */
 485     qapi_event_send_migration(MIGRATION_STATUS_SETUP);
 486     if (strstart(uri, "tcp:", &p) ||
 487         strstart(uri, "unix:", NULL) ||
 488         strstart(uri, "vsock:", NULL)) {
 489         migrate_protocol_allow_multi_channels(true);
 490         socket_start_incoming_migration(p ? p : uri, errp);
 491 #ifdef CONFIG_RDMA
 492     } else if (strstart(uri, "rdma:", &p)) {
 493         rdma_start_incoming_migration(p, errp);
 494 #endif
 495     } else if (strstart(uri, "exec:", &p)) {
 496         exec_start_incoming_migration(p, errp);
 497     } else if (strstart(uri, "fd:", &p)) {
 498         fd_start_incoming_migration(p, errp);
 499     } else {
 500         error_setg(errp, "unknown migration protocol: %s", uri);
 501     }
 502 }
 503
 504 static void process_incoming_migration_bh(void *opaque)
 505 {
 506     Error *local_err = NULL;
 507     MigrationIncomingState *mis = opaque;
 508
 509     /* If capability late_block_activate is set:
 510      * Only fire up the block code now if we're going to restart the
 511      * VM, else 'cont' will do it.
 512      * This causes file locking to happen; so we don't want it to happen
 513      * unless we really are starting the VM.
 514      */
 515     if (!migrate_late_block_activate() ||
 516          (autostart && (!global_state_received() ||
 517             global_state_get_runstate() == RUN_STATE_RUNNING))) {
 518         /* Make sure all file formats throw away their mutable metadata.
 519          * If we get an error here, just don't restart the VM yet. */
 520         bdrv_activate_all(&local_err);
 521         if (local_err) {
 522             error_report_err(local_err);
 523             local_err = NULL;
 524             autostart = false;
 525         }
 526     }
 527
 528     /*
 529      * This must happen after all error conditions are dealt with and
 530      * we're sure the VM is going to be running on this host.
 531      */
 532     qemu_announce_self(&mis->announce_timer, migrate_announce_params());
 533
 534     if (multifd_load_cleanup(&local_err) != 0) {
 535         error_report_err(local_err);
 536         autostart = false;
 537     }
 538     /* If global state section was not received or we are in running
 539        state, we need to obey autostart. Any other state is set with
 540        runstate_set. */
 541
 542     dirty_bitmap_mig_before_vm_start();
 543
 544     if (!global_state_received() ||
 545         global_state_get_runstate() == RUN_STATE_RUNNING) {
 546         if (autostart) {
 547             vm_start();
 548         } else {
 549             runstate_set(RUN_STATE_PAUSED);
 550         }
 551     } else if (migration_incoming_colo_enabled()) {
 552         migration_incoming_disable_colo();
 553         vm_start();
 554     } else {
 555         runstate_set(global_state_get_runstate());
 556     }
 557     /*
 558      * This must happen after any state changes since as soon as an external
 559      * observer sees this event they might start to prod at the VM assuming
 560      * it's ready to use.
 561      */
 562     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
 563                       MIGRATION_STATUS_COMPLETED);
 564     qemu_bh_delete(mis->bh);
 565     migration_incoming_state_destroy();
 566 }
 567
 568 static void process_incoming_migration_co(void *opaque)
 569 {
 570     MigrationIncomingState *mis = migration_incoming_get_current();
 571     PostcopyState ps;
 572     int ret;
 573     Error *local_err = NULL;
 574
 575     assert(mis->from_src_file);
 576     mis->migration_incoming_co = qemu_coroutine_self();
 577     mis->largest_page_size = qemu_ram_pagesize_largest();
 578     postcopy_state_set(POSTCOPY_INCOMING_NONE);
 579     migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
 580                       MIGRATION_STATUS_ACTIVE);
 581     ret = qemu_loadvm_state(mis->from_src_file);
 582
 583     ps = postcopy_state_get();
 584     trace_process_incoming_migration_co_end(ret, ps);
 585     if (ps != POSTCOPY_INCOMING_NONE) {
 586         if (ps == POSTCOPY_INCOMING_ADVISE) {
 587             /*
 588              * Where a migration had postcopy enabled (and thus went to advise)
 589              * but managed to complete within the precopy period, we can use
 590              * the normal exit.
 591              */
 592             postcopy_ram_incoming_cleanup(mis);
 593         } else if (ret >= 0) {
 594             /*
 595              * Postcopy was started, cleanup should happen at the end of the
 596              * postcopy thread.
 597              */
 598             trace_process_incoming_migration_co_postcopy_end_main();
 599             return;
 600         }
 601         /* Else if something went wrong then just fall out of the normal exit */
 602     }
 603
 604     /* we get COLO info, and know if we are in COLO mode */
 605     if (!ret && migration_incoming_colo_enabled()) {
 606         /* Make sure all file formats throw away their mutable metadata */
 607         bdrv_activate_all(&local_err);
 608         if (local_err) {
 609             error_report_err(local_err);
 610             goto fail;
 611         }
 612
 613         qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
 614              colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
 615         mis->have_colo_incoming_thread = true;
 616         qemu_coroutine_yield();
 617
 618         qemu_mutex_unlock_iothread();
 619         /* Wait checkpoint incoming thread exit before free resource */
 620         qemu_thread_join(&mis->colo_incoming_thread);
 621         qemu_mutex_lock_iothread();
 622         /* We hold the global iothread lock, so it is safe here */
 623         colo_release_ram_cache();
 624     }
 625
 626     if (ret < 0) {
 627         error_report("load of migration failed: %s", strerror(-ret));
 628         goto fail;
 629     }
 630     mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
 631     qemu_bh_schedule(mis->bh);
 632     mis->migration_incoming_co = NULL;
 633     return;
 634 fail:
 635     local_err = NULL;
 636     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
 637                       MIGRATION_STATUS_FAILED);
 638     qemu_fclose(mis->from_src_file);
 639     if (multifd_load_cleanup(&local_err) != 0) {
 640         error_report_err(local_err);
 641     }
 642     exit(EXIT_FAILURE);
 643 }
 644
 645 /**
 646  * migration_incoming_setup: Setup incoming migration
 647  * @f: file for main migration channel
 648  * @errp: where to put errors
 649  *
 650  * Returns: %true on success, %false on error.
 651  */
 652 static bool migration_incoming_setup(QEMUFile *f, Error **errp)
 653 {
 654     MigrationIncomingState *mis = migration_incoming_get_current();
 655
 656     if (multifd_load_setup(errp) != 0) {
 657         return false;
 658     }
 659
 660     if (!mis->from_src_file) {
 661         mis->from_src_file = f;
 662     }
 663     qemu_file_set_blocking(f, false);
 664     return true;
 665 }
 666
 667 void migration_incoming_process(void)
 668 {
 669     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
 670     qemu_coroutine_enter(co);
 671 }
 672
 673 /* Returns true if recovered from a paused migration, otherwise false */
 674 static bool postcopy_try_recover(void)
 675 {
 676     MigrationIncomingState *mis = migration_incoming_get_current();
 677
 678     if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
 679         /* Resumed from a paused postcopy migration */
 680
 681         /* This should be set already in migration_incoming_setup() */
 682         assert(mis->from_src_file);
 683         /* Postcopy has standalone thread to do vm load */
 684         qemu_file_set_blocking(mis->from_src_file, true);
 685
 686         /* Re-configure the return path */
 687         mis->to_src_file = qemu_file_get_return_path(mis->from_src_file);
 688
 689         migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
 690                           MIGRATION_STATUS_POSTCOPY_RECOVER);
 691
 692         /*
 693          * Here, we only wake up the main loading thread (while the
 694          * fault thread will still be waiting), so that we can receive
 695          * commands from source now, and answer it if needed. The
 696          * fault thread will be woken up afterwards until we are sure
 697          * that source is ready to reply to page requests.
 698          */
 699         qemu_sem_post(&mis->postcopy_pause_sem_dst);
 700         return true;
 701     }
 702
 703     return false;
 704 }
 705
 706 void migration_fd_process_incoming(QEMUFile *f, Error **errp)
 707 {
 708     if (!migration_incoming_setup(f, errp)) {
 709         return;
 710     }
 711     if (postcopy_try_recover()) {
 712         return;
 713     }
 714     migration_incoming_process();
 715 }
 716
 717 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp)
 718 {
 719     MigrationIncomingState *mis = migration_incoming_get_current();
 720     Error *local_err = NULL;
 721     bool start_migration;
 722
 723     if (!mis->from_src_file) {
 724         /* The first connection (multifd may have multiple) */
 725         QEMUFile *f = qemu_fopen_channel_input(ioc);
 726
 727         if (!migration_incoming_setup(f, errp)) {
 728             return;
 729         }
 730
 731         /*
 732          * Common migration only needs one channel, so we can start
 733          * right now.  Multifd needs more than one channel, we wait.
 734          */
 735         start_migration = !migrate_use_multifd();
 736     } else {
 737         /* Multiple connections */
 738         assert(migrate_use_multifd());
 739         start_migration = multifd_recv_new_channel(ioc, &local_err);
 740         if (local_err) {
 741             error_propagate(errp, local_err);
 742             return;
 743         }
 744     }
 745
 746     if (start_migration) {
 747         /* If it's a recovery, we're done */
 748         if (postcopy_try_recover()) {
 749             return;
 750         }
 751         migration_incoming_process();
 752     }
 753 }
 754
 755 /**
 756  * @migration_has_all_channels: We have received all channels that we need
 757  *
 758  * Returns true when we have got connections to all the channels that
 759  * we need for migration.
 760  */
 761 bool migration_has_all_channels(void)
 762 {
 763     MigrationIncomingState *mis = migration_incoming_get_current();
 764     bool all_channels;
 765
 766     all_channels = multifd_recv_all_channels_created();
 767
 768     return all_channels && mis->from_src_file != NULL;
 769 }
 770
 771 /*
 772  * Send a 'SHUT' message on the return channel with the given value
 773  * to indicate that we've finished with the RP.  Non-0 value indicates
 774  * error.
 775  */
 776 void migrate_send_rp_shut(MigrationIncomingState *mis,
 777                           uint32_t value)
 778 {
 779     uint32_t buf;
 780
 781     buf = cpu_to_be32(value);
 782     migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
 783 }
 784
 785 /*
 786  * Send a 'PONG' message on the return channel with the given value
 787  * (normally in response to a 'PING')
 788  */
 789 void migrate_send_rp_pong(MigrationIncomingState *mis,
 790                           uint32_t value)
 791 {
 792     uint32_t buf;
 793
 794     buf = cpu_to_be32(value);
 795     migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
 796 }
 797
 798 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
 799                                  char *block_name)
 800 {
 801     char buf[512];
 802     int len;
 803     int64_t res;
 804
 805     /*
 806      * First, we send the header part. It contains only the len of
 807      * idstr, and the idstr itself.
 808      */
 809     len = strlen(block_name);
 810     buf[0] = len;
 811     memcpy(buf + 1, block_name, len);
 812
 813     if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
 814         error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
 815                      __func__);
 816         return;
 817     }
 818
 819     migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf);
 820
 821     /*
 822      * Next, we dump the received bitmap to the stream.
 823      *
 824      * TODO: currently we are safe since we are the only one that is
 825      * using the to_src_file handle (fault thread is still paused),
 826      * and it's ok even not taking the mutex. However the best way is
 827      * to take the lock before sending the message header, and release
 828      * the lock after sending the bitmap.
 829      */
 830     qemu_mutex_lock(&mis->rp_mutex);
 831     res = ramblock_recv_bitmap_send(mis->to_src_file, block_name);
 832     qemu_mutex_unlock(&mis->rp_mutex);
 833
 834     trace_migrate_send_rp_recv_bitmap(block_name, res);
 835 }
 836
 837 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value)
 838 {
 839     uint32_t buf;
 840
 841     buf = cpu_to_be32(value);
 842     migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf);
 843 }
 844
 845 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
 846 {
 847     MigrationCapabilityStatusList *head = NULL, **tail = &head;
 848     MigrationCapabilityStatus *caps;
 849     MigrationState *s = migrate_get_current();
 850     int i;
 851
 852     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
 853 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
 854         if (i == MIGRATION_CAPABILITY_BLOCK) {
 855             continue;
 856         }
 857 #endif
 858         caps = g_malloc0(sizeof(*caps));
 859         caps->capability = i;
 860         caps->state = s->enabled_capabilities[i];
 861         QAPI_LIST_APPEND(tail, caps);
 862     }
 863
 864     return head;
 865 }
 866
 867 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
 868 {
 869     MigrationParameters *params;
 870     MigrationState *s = migrate_get_current();
 871
 872     /* TODO use QAPI_CLONE() instead of duplicating it inline */
 873     params = g_malloc0(sizeof(*params));
 874     params->has_compress_level = true;
 875     params->compress_level = s->parameters.compress_level;
 876     params->has_compress_threads = true;
 877     params->compress_threads = s->parameters.compress_threads;
 878     params->has_compress_wait_thread = true;
 879     params->compress_wait_thread = s->parameters.compress_wait_thread;
 880     params->has_decompress_threads = true;
 881     params->decompress_threads = s->parameters.decompress_threads;
 882     params->has_throttle_trigger_threshold = true;
 883     params->throttle_trigger_threshold = s->parameters.throttle_trigger_threshold;
 884     params->has_cpu_throttle_initial = true;
 885     params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
 886     params->has_cpu_throttle_increment = true;
 887     params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
 888     params->has_cpu_throttle_tailslow = true;
 889     params->cpu_throttle_tailslow = s->parameters.cpu_throttle_tailslow;
 890     params->has_tls_creds = true;
 891     params->tls_creds = g_strdup(s->parameters.tls_creds);
 892     params->has_tls_hostname = true;
 893     params->tls_hostname = g_strdup(s->parameters.tls_hostname);
 894     params->has_tls_authz = true;
 895     params->tls_authz = g_strdup(s->parameters.tls_authz ?
 896                                  s->parameters.tls_authz : "");
 897     params->has_max_bandwidth = true;
 898     params->max_bandwidth = s->parameters.max_bandwidth;
 899     params->has_downtime_limit = true;
 900     params->downtime_limit = s->parameters.downtime_limit;
 901     params->has_x_checkpoint_delay = true;
 902     params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
 903     params->has_block_incremental = true;
 904     params->block_incremental = s->parameters.block_incremental;
 905     params->has_multifd_channels = true;
 906     params->multifd_channels = s->parameters.multifd_channels;
 907     params->has_multifd_compression = true;
 908     params->multifd_compression = s->parameters.multifd_compression;
 909     params->has_multifd_zlib_level = true;
 910     params->multifd_zlib_level = s->parameters.multifd_zlib_level;
 911     params->has_multifd_zstd_level = true;
 912     params->multifd_zstd_level = s->parameters.multifd_zstd_level;
 913     params->has_xbzrle_cache_size = true;
 914     params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
 915     params->has_max_postcopy_bandwidth = true;
 916     params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth;
 917     params->has_max_cpu_throttle = true;
 918     params->max_cpu_throttle = s->parameters.max_cpu_throttle;
 919     params->has_announce_initial = true;
 920     params->announce_initial = s->parameters.announce_initial;
 921     params->has_announce_max = true;
 922     params->announce_max = s->parameters.announce_max;
 923     params->has_announce_rounds = true;
 924     params->announce_rounds = s->parameters.announce_rounds;
 925     params->has_announce_step = true;
 926     params->announce_step = s->parameters.announce_step;
 927
 928     if (s->parameters.has_block_bitmap_mapping) {
 929         params->has_block_bitmap_mapping = true;
 930         params->block_bitmap_mapping =
 931             QAPI_CLONE(BitmapMigrationNodeAliasList,
 932                        s->parameters.block_bitmap_mapping);
 933     }
 934
 935     return params;
 936 }
 937
 938 AnnounceParameters *migrate_announce_params(void)
 939 {
 940     static AnnounceParameters ap;
 941
 942     MigrationState *s = migrate_get_current();
 943
 944     ap.initial = s->parameters.announce_initial;
 945     ap.max = s->parameters.announce_max;
 946     ap.rounds = s->parameters.announce_rounds;
 947     ap.step = s->parameters.announce_step;
 948
 949     return &ap;
 950 }
 951
 952 /*
 953  * Return true if we're already in the middle of a migration
 954  * (i.e. any of the active or setup states)
 955  */
 956 bool migration_is_setup_or_active(int state)
 957 {
 958     switch (state) {
 959     case MIGRATION_STATUS_ACTIVE:
 960     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
 961     case MIGRATION_STATUS_POSTCOPY_PAUSED:
 962     case MIGRATION_STATUS_POSTCOPY_RECOVER:
 963     case MIGRATION_STATUS_SETUP:
 964     case MIGRATION_STATUS_PRE_SWITCHOVER:
 965     case MIGRATION_STATUS_DEVICE:
 966     case MIGRATION_STATUS_WAIT_UNPLUG:
 967     case MIGRATION_STATUS_COLO:
 968         return true;
 969
 970     default:
 971         return false;
 972
 973     }
 974 }
 975
 976 bool migration_is_running(int state)
 977 {
 978     switch (state) {
 979     case MIGRATION_STATUS_ACTIVE:
 980     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
 981     case MIGRATION_STATUS_POSTCOPY_PAUSED:
 982     case MIGRATION_STATUS_POSTCOPY_RECOVER:
 983     case MIGRATION_STATUS_SETUP:
 984     case MIGRATION_STATUS_PRE_SWITCHOVER:
 985     case MIGRATION_STATUS_DEVICE:
 986     case MIGRATION_STATUS_WAIT_UNPLUG:
 987     case MIGRATION_STATUS_CANCELLING:
 988         return true;
 989
 990     default:
 991         return false;
 992
 993     }
 994 }
 995
 996 static void populate_time_info(MigrationInfo *info, MigrationState *s)
 997 {
 998     info->has_status = true;
 999     info->has_setup_time = true;
1000     info->setup_time = s->setup_time;
1001     if (s->state == MIGRATION_STATUS_COMPLETED) {
1002         info->has_total_time = true;
1003         info->total_time = s->total_time;
1004         info->has_downtime = true;
1005         info->downtime = s->downtime;
1006     } else {
1007         info->has_total_time = true;
1008         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) -
1009                            s->start_time;
1010         info->has_expected_downtime = true;
1011         info->expected_downtime = s->expected_downtime;
1012     }
1013 }
1014
1015 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
1016 {
1017     size_t page_size = qemu_target_page_size();
1018
1019     info->has_ram = true;
1020     info->ram = g_malloc0(sizeof(*info->ram));
1021     info->ram->transferred = ram_counters.transferred;
1022     info->ram->total = ram_bytes_total();
1023     info->ram->duplicate = ram_counters.duplicate;
1024     /* legacy value.  It is not used anymore */
1025     info->ram->skipped = 0;
1026     info->ram->normal = ram_counters.normal;
1027     info->ram->normal_bytes = ram_counters.normal * page_size;
1028     info->ram->mbps = s->mbps;
1029     info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
1030     info->ram->postcopy_requests = ram_counters.postcopy_requests;
1031     info->ram->page_size = page_size;
1032     info->ram->multifd_bytes = ram_counters.multifd_bytes;
1033     info->ram->pages_per_second = s->pages_per_second;
1034     info->ram->precopy_bytes = ram_counters.precopy_bytes;
1035     info->ram->downtime_bytes = ram_counters.downtime_bytes;
1036     info->ram->postcopy_bytes = ram_counters.postcopy_bytes;
1037
1038     if (migrate_use_xbzrle()) {
1039         info->has_xbzrle_cache = true;
1040         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
1041         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
1042         info->xbzrle_cache->bytes = xbzrle_counters.bytes;
1043         info->xbzrle_cache->pages = xbzrle_counters.pages;
1044         info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
1045         info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
1046         info->xbzrle_cache->encoding_rate = xbzrle_counters.encoding_rate;
1047         info->xbzrle_cache->overflow = xbzrle_counters.overflow;
1048     }
1049
1050     if (migrate_use_compression()) {
1051         info->has_compression = true;
1052         info->compression = g_malloc0(sizeof(*info->compression));
1053         info->compression->pages = compression_counters.pages;
1054         info->compression->busy = compression_counters.busy;
1055         info->compression->busy_rate = compression_counters.busy_rate;
1056         info->compression->compressed_size =
1057                                     compression_counters.compressed_size;
1058         info->compression->compression_rate =
1059                                     compression_counters.compression_rate;
1060     }
1061
1062     if (cpu_throttle_active()) {
1063         info->has_cpu_throttle_percentage = true;
1064         info->cpu_throttle_percentage = cpu_throttle_get_percentage();
1065     }
1066
1067     if (s->state != MIGRATION_STATUS_COMPLETED) {
1068         info->ram->remaining = ram_bytes_remaining();
1069         info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
1070     }
1071 }
1072
1073 static void populate_disk_info(MigrationInfo *info)
1074 {
1075     if (blk_mig_active()) {
1076         info->has_disk = true;
1077         info->disk = g_malloc0(sizeof(*info->disk));
1078         info->disk->transferred = blk_mig_bytes_transferred();
1079         info->disk->remaining = blk_mig_bytes_remaining();
1080         info->disk->total = blk_mig_bytes_total();
1081     }
1082 }
1083
1084 static void fill_source_migration_info(MigrationInfo *info)
1085 {
1086     MigrationState *s = migrate_get_current();
1087     int state = qatomic_read(&s->state);
1088     GSList *cur_blocker = migration_blockers;
1089
1090     info->blocked_reasons = NULL;
1091
1092     /*
1093      * There are two types of reasons a migration might be blocked;
1094      * a) devices marked in VMState as non-migratable, and
1095      * b) Explicit migration blockers
1096      * We need to add both of them here.
1097      */
1098     qemu_savevm_non_migratable_list(&info->blocked_reasons);
1099
1100     while (cur_blocker) {
1101         QAPI_LIST_PREPEND(info->blocked_reasons,
1102                           g_strdup(error_get_pretty(cur_blocker->data)));
1103         cur_blocker = g_slist_next(cur_blocker);
1104     }
1105     info->has_blocked_reasons = info->blocked_reasons != NULL;
1106
1107     switch (state) {
1108     case MIGRATION_STATUS_NONE:
1109         /* no migration has happened ever */
1110         /* do not overwrite destination migration status */
1111         return;
1112     case MIGRATION_STATUS_SETUP:
1113         info->has_status = true;
1114         info->has_total_time = false;
1115         break;
1116     case MIGRATION_STATUS_ACTIVE:
1117     case MIGRATION_STATUS_CANCELLING:
1118     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1119     case MIGRATION_STATUS_PRE_SWITCHOVER:
1120     case MIGRATION_STATUS_DEVICE:
1121     case MIGRATION_STATUS_POSTCOPY_PAUSED:
1122     case MIGRATION_STATUS_POSTCOPY_RECOVER:
1123         /* TODO add some postcopy stats */
1124         populate_time_info(info, s);
1125         populate_ram_info(info, s);
1126         populate_disk_info(info);
1127         populate_vfio_info(info);
1128         break;
1129     case MIGRATION_STATUS_COLO:
1130         info->has_status = true;
1131         /* TODO: display COLO specific information (checkpoint info etc.) */
1132         break;
1133     case MIGRATION_STATUS_COMPLETED:
1134         populate_time_info(info, s);
1135         populate_ram_info(info, s);
1136         populate_vfio_info(info);
1137         break;
1138     case MIGRATION_STATUS_FAILED:
1139         info->has_status = true;
1140         if (s->error) {
1141             info->has_error_desc = true;
1142             info->error_desc = g_strdup(error_get_pretty(s->error));
1143         }
1144         break;
1145     case MIGRATION_STATUS_CANCELLED:
1146         info->has_status = true;
1147         break;
1148     case MIGRATION_STATUS_WAIT_UNPLUG:
1149         info->has_status = true;
1150         break;
1151     }
1152     info->status = state;
1153 }
1154
1155 typedef enum WriteTrackingSupport {
1156     WT_SUPPORT_UNKNOWN = 0,
1157     WT_SUPPORT_ABSENT,
1158     WT_SUPPORT_AVAILABLE,
1159     WT_SUPPORT_COMPATIBLE
1160 } WriteTrackingSupport;
1161
1162 static
1163 WriteTrackingSupport migrate_query_write_tracking(void)
1164 {
1165     /* Check if kernel supports required UFFD features */
1166     if (!ram_write_tracking_available()) {
1167         return WT_SUPPORT_ABSENT;
1168     }
1169     /*
1170      * Check if current memory configuration is
1171      * compatible with required UFFD features.
1172      */
1173     if (!ram_write_tracking_compatible()) {
1174         return WT_SUPPORT_AVAILABLE;
1175     }
1176
1177     return WT_SUPPORT_COMPATIBLE;
1178 }
1179
1180 /**
1181  * @migration_caps_check - check capability validity
1182  *
1183  * @cap_list: old capability list, array of bool
1184  * @params: new capabilities to be applied soon
1185  * @errp: set *errp if the check failed, with reason
1186  *
1187  * Returns true if check passed, otherwise false.
1188  */
1189 static bool migrate_caps_check(bool *cap_list,
1190                                MigrationCapabilityStatusList *params,
1191                                Error **errp)
1192 {
1193     MigrationCapabilityStatusList *cap;
1194     bool old_postcopy_cap;
1195     MigrationIncomingState *mis = migration_incoming_get_current();
1196
1197     old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
1198
1199     for (cap = params; cap; cap = cap->next) {
1200         cap_list[cap->value->capability] = cap->value->state;
1201     }
1202
1203 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1204     if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
1205         error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
1206                    "block migration");
1207         error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
1208         return false;
1209     }
1210 #endif
1211
1212 #ifndef CONFIG_REPLICATION
1213     if (cap_list[MIGRATION_CAPABILITY_X_COLO]) {
1214         error_setg(errp, "QEMU compiled without replication module"
1215                    " can't enable COLO");
1216         error_append_hint(errp, "Please enable replication before COLO.\n");
1217         return false;
1218     }
1219 #endif
1220
1221     if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
1222         /* This check is reasonably expensive, so only when it's being
1223          * set the first time, also it's only the destination that needs
1224          * special support.
1225          */
1226         if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
1227             !postcopy_ram_supported_by_host(mis)) {
1228             /* postcopy_ram_supported_by_host will have emitted a more
1229              * detailed message
1230              */
1231             error_setg(errp, "Postcopy is not supported");
1232             return false;
1233         }
1234
1235         if (cap_list[MIGRATION_CAPABILITY_X_IGNORE_SHARED]) {
1236             error_setg(errp, "Postcopy is not compatible with ignore-shared");
1237             return false;
1238         }
1239     }
1240
1241     if (cap_list[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT]) {
1242         WriteTrackingSupport wt_support;
1243         int idx;
1244         /*
1245          * Check if 'background-snapshot' capability is supported by
1246          * host kernel and compatible with guest memory configuration.
1247          */
1248         wt_support = migrate_query_write_tracking();
1249         if (wt_support < WT_SUPPORT_AVAILABLE) {
1250             error_setg(errp, "Background-snapshot is not supported by host kernel");
1251             return false;
1252         }
1253         if (wt_support < WT_SUPPORT_COMPATIBLE) {
1254             error_setg(errp, "Background-snapshot is not compatible "
1255                     "with guest memory configuration");
1256             return false;
1257         }
1258
1259         /*
1260          * Check if there are any migration capabilities
1261          * incompatible with 'background-snapshot'.
1262          */
1263         for (idx = 0; idx < check_caps_background_snapshot.size; idx++) {
1264             int incomp_cap = check_caps_background_snapshot.caps[idx];
1265             if (cap_list[incomp_cap]) {
1266                 error_setg(errp,
1267                         "Background-snapshot is not compatible with %s",
1268                         MigrationCapability_str(incomp_cap));
1269                 return false;
1270             }
1271         }
1272     }
1273
1274     /* incoming side only */
1275     if (runstate_check(RUN_STATE_INMIGRATE) &&
1276         !migrate_multi_channels_is_allowed() &&
1277         cap_list[MIGRATION_CAPABILITY_MULTIFD]) {
1278         error_setg(errp, "multifd is not supported by current protocol");
1279         return false;
1280     }
1281
1282     return true;
1283 }
1284
1285 static void fill_destination_migration_info(MigrationInfo *info)
1286 {
1287     MigrationIncomingState *mis = migration_incoming_get_current();
1288
1289     if (mis->socket_address_list) {
1290         info->has_socket_address = true;
1291         info->socket_address =
1292             QAPI_CLONE(SocketAddressList, mis->socket_address_list);
1293     }
1294
1295     switch (mis->state) {
1296     case MIGRATION_STATUS_NONE:
1297         return;
1298     case MIGRATION_STATUS_SETUP:
1299     case MIGRATION_STATUS_CANCELLING:
1300     case MIGRATION_STATUS_CANCELLED:
1301     case MIGRATION_STATUS_ACTIVE:
1302     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1303     case MIGRATION_STATUS_POSTCOPY_PAUSED:
1304     case MIGRATION_STATUS_POSTCOPY_RECOVER:
1305     case MIGRATION_STATUS_FAILED:
1306     case MIGRATION_STATUS_COLO:
1307         info->has_status = true;
1308         break;
1309     case MIGRATION_STATUS_COMPLETED:
1310         info->has_status = true;
1311         fill_destination_postcopy_migration_info(info);
1312         break;
1313     }
1314     info->status = mis->state;
1315 }
1316
1317 MigrationInfo *qmp_query_migrate(Error **errp)
1318 {
1319     MigrationInfo *info = g_malloc0(sizeof(*info));
1320
1321     fill_destination_migration_info(info);
1322     fill_source_migration_info(info);
1323
1324     return info;
1325 }
1326
1327 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
1328                                   Error **errp)
1329 {
1330     MigrationState *s = migrate_get_current();
1331     MigrationCapabilityStatusList *cap;
1332     bool cap_list[MIGRATION_CAPABILITY__MAX];
1333
1334     if (migration_is_running(s->state)) {
1335         error_setg(errp, QERR_MIGRATION_ACTIVE);
1336         return;
1337     }
1338
1339     memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list));
1340     if (!migrate_caps_check(cap_list, params, errp)) {
1341         return;
1342     }
1343
1344     for (cap = params; cap; cap = cap->next) {
1345         s->enabled_capabilities[cap->value->capability] = cap->value->state;
1346     }
1347 }
1348
1349 /*
1350  * Check whether the parameters are valid. Error will be put into errp
1351  * (if provided). Return true if valid, otherwise false.
1352  */
1353 static bool migrate_params_check(MigrationParameters *params, Error **errp)
1354 {
1355     if (params->has_compress_level &&
1356         (params->compress_level > 9)) {
1357         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
1358                    "a value between 0 and 9");
1359         return false;
1360     }
1361
1362     if (params->has_compress_threads && (params->compress_threads < 1)) {
1363         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1364                    "compress_threads",
1365                    "a value between 1 and 255");
1366         return false;
1367     }
1368
1369     if (params->has_decompress_threads && (params->decompress_threads < 1)) {
1370         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1371                    "decompress_threads",
1372                    "a value between 1 and 255");
1373         return false;
1374     }
1375
1376     if (params->has_throttle_trigger_threshold &&
1377         (params->throttle_trigger_threshold < 1 ||
1378          params->throttle_trigger_threshold > 100)) {
1379         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1380                    "throttle_trigger_threshold",
1381                    "an integer in the range of 1 to 100");
1382         return false;
1383     }
1384
1385     if (params->has_cpu_throttle_initial &&
1386         (params->cpu_throttle_initial < 1 ||
1387          params->cpu_throttle_initial > 99)) {
1388         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1389                    "cpu_throttle_initial",
1390                    "an integer in the range of 1 to 99");
1391         return false;
1392     }
1393
1394     if (params->has_cpu_throttle_increment &&
1395         (params->cpu_throttle_increment < 1 ||
1396          params->cpu_throttle_increment > 99)) {
1397         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1398                    "cpu_throttle_increment",
1399                    "an integer in the range of 1 to 99");
1400         return false;
1401     }
1402
1403     if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
1404         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1405                    "max_bandwidth",
1406                    "an integer in the range of 0 to "stringify(SIZE_MAX)
1407                    " bytes/second");
1408         return false;
1409     }
1410
1411     if (params->has_downtime_limit &&
1412         (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
1413         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1414                    "downtime_limit",
1415                    "an integer in the range of 0 to "
1416                     stringify(MAX_MIGRATE_DOWNTIME)" ms");
1417         return false;
1418     }
1419
1420     /* x_checkpoint_delay is now always positive */
1421
1422     if (params->has_multifd_channels && (params->multifd_channels < 1)) {
1423         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1424                    "multifd_channels",
1425                    "a value between 1 and 255");
1426         return false;
1427     }
1428
1429     if (params->has_multifd_zlib_level &&
1430         (params->multifd_zlib_level > 9)) {
1431         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zlib_level",
1432                    "a value between 0 and 9");
1433         return false;
1434     }
1435
1436     if (params->has_multifd_zstd_level &&
1437         (params->multifd_zstd_level > 20)) {
1438         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zstd_level",
1439                    "a value between 0 and 20");
1440         return false;
1441     }
1442
1443     if (params->has_xbzrle_cache_size &&
1444         (params->xbzrle_cache_size < qemu_target_page_size() ||
1445          !is_power_of_2(params->xbzrle_cache_size))) {
1446         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1447                    "xbzrle_cache_size",
1448                    "a power of two no less than the target page size");
1449         return false;
1450     }
1451
1452     if (params->has_max_cpu_throttle &&
1453         (params->max_cpu_throttle < params->cpu_throttle_initial ||
1454          params->max_cpu_throttle > 99)) {
1455         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1456                    "max_cpu_throttle",
1457                    "an integer in the range of cpu_throttle_initial to 99");
1458         return false;
1459     }
1460
1461     if (params->has_announce_initial &&
1462         params->announce_initial > 100000) {
1463         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1464                    "announce_initial",
1465                    "a value between 0 and 100000");
1466         return false;
1467     }
1468     if (params->has_announce_max &&
1469         params->announce_max > 100000) {
1470         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1471                    "announce_max",
1472                    "a value between 0 and 100000");
1473        return false;
1474     }
1475     if (params->has_announce_rounds &&
1476         params->announce_rounds > 1000) {
1477         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1478                    "announce_rounds",
1479                    "a value between 0 and 1000");
1480        return false;
1481     }
1482     if (params->has_announce_step &&
1483         (params->announce_step < 1 ||
1484         params->announce_step > 10000)) {
1485         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
1486                    "announce_step",
1487                    "a value between 0 and 10000");
1488        return false;
1489     }
1490
1491     if (params->has_block_bitmap_mapping &&
1492         !check_dirty_bitmap_mig_alias_map(params->block_bitmap_mapping, errp)) {
1493         error_prepend(errp, "Invalid mapping given for block-bitmap-mapping: ");
1494         return false;
1495     }
1496
1497     return true;
1498 }
1499
1500 static void migrate_params_test_apply(MigrateSetParameters *params,
1501                                       MigrationParameters *dest)
1502 {
1503     *dest = migrate_get_current()->parameters;
1504
1505     /* TODO use QAPI_CLONE() instead of duplicating it inline */
1506
1507     if (params->has_compress_level) {
1508         dest->compress_level = params->compress_level;
1509     }
1510
1511     if (params->has_compress_threads) {
1512         dest->compress_threads = params->compress_threads;
1513     }
1514
1515     if (params->has_compress_wait_thread) {
1516         dest->compress_wait_thread = params->compress_wait_thread;
1517     }
1518
1519     if (params->has_decompress_threads) {
1520         dest->decompress_threads = params->decompress_threads;
1521     }
1522
1523     if (params->has_throttle_trigger_threshold) {
1524         dest->throttle_trigger_threshold = params->throttle_trigger_threshold;
1525     }
1526
1527     if (params->has_cpu_throttle_initial) {
1528         dest->cpu_throttle_initial = params->cpu_throttle_initial;
1529     }
1530
1531     if (params->has_cpu_throttle_increment) {
1532         dest->cpu_throttle_increment = params->cpu_throttle_increment;
1533     }
1534
1535     if (params->has_cpu_throttle_tailslow) {
1536         dest->cpu_throttle_tailslow = params->cpu_throttle_tailslow;
1537     }
1538
1539     if (params->has_tls_creds) {
1540         assert(params->tls_creds->type == QTYPE_QSTRING);
1541         dest->tls_creds = params->tls_creds->u.s;
1542     }
1543
1544     if (params->has_tls_hostname) {
1545         assert(params->tls_hostname->type == QTYPE_QSTRING);
1546         dest->tls_hostname = params->tls_hostname->u.s;
1547     }
1548
1549     if (params->has_max_bandwidth) {
1550         dest->max_bandwidth = params->max_bandwidth;
1551     }
1552
1553     if (params->has_downtime_limit) {
1554         dest->downtime_limit = params->downtime_limit;
1555     }
1556
1557     if (params->has_x_checkpoint_delay) {
1558         dest->x_checkpoint_delay = params->x_checkpoint_delay;
1559     }
1560
1561     if (params->has_block_incremental) {
1562         dest->block_incremental = params->block_incremental;
1563     }
1564     if (params->has_multifd_channels) {
1565         dest->multifd_channels = params->multifd_channels;
1566     }
1567     if (params->has_multifd_compression) {
1568         dest->multifd_compression = params->multifd_compression;
1569     }
1570     if (params->has_xbzrle_cache_size) {
1571         dest->xbzrle_cache_size = params->xbzrle_cache_size;
1572     }
1573     if (params->has_max_postcopy_bandwidth) {
1574         dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1575     }
1576     if (params->has_max_cpu_throttle) {
1577         dest->max_cpu_throttle = params->max_cpu_throttle;
1578     }
1579     if (params->has_announce_initial) {
1580         dest->announce_initial = params->announce_initial;
1581     }
1582     if (params->has_announce_max) {
1583         dest->announce_max = params->announce_max;
1584     }
1585     if (params->has_announce_rounds) {
1586         dest->announce_rounds = params->announce_rounds;
1587     }
1588     if (params->has_announce_step) {
1589         dest->announce_step = params->announce_step;
1590     }
1591
1592     if (params->has_block_bitmap_mapping) {
1593         dest->has_block_bitmap_mapping = true;
1594         dest->block_bitmap_mapping = params->block_bitmap_mapping;
1595     }
1596 }
1597
1598 static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
1599 {
1600     MigrationState *s = migrate_get_current();
1601
1602     /* TODO use QAPI_CLONE() instead of duplicating it inline */
1603
1604     if (params->has_compress_level) {
1605         s->parameters.compress_level = params->compress_level;
1606     }
1607
1608     if (params->has_compress_threads) {
1609         s->parameters.compress_threads = params->compress_threads;
1610     }
1611
1612     if (params->has_compress_wait_thread) {
1613         s->parameters.compress_wait_thread = params->compress_wait_thread;
1614     }
1615
1616     if (params->has_decompress_threads) {
1617         s->parameters.decompress_threads = params->decompress_threads;
1618     }
1619
1620     if (params->has_throttle_trigger_threshold) {
1621         s->parameters.throttle_trigger_threshold = params->throttle_trigger_threshold;
1622     }
1623
1624     if (params->has_cpu_throttle_initial) {
1625         s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
1626     }
1627
1628     if (params->has_cpu_throttle_increment) {
1629         s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
1630     }
1631
1632     if (params->has_cpu_throttle_tailslow) {
1633         s->parameters.cpu_throttle_tailslow = params->cpu_throttle_tailslow;
1634     }
1635
1636     if (params->has_tls_creds) {
1637         g_free(s->parameters.tls_creds);
1638         assert(params->tls_creds->type == QTYPE_QSTRING);
1639         s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
1640     }
1641
1642     if (params->has_tls_hostname) {
1643         g_free(s->parameters.tls_hostname);
1644         assert(params->tls_hostname->type == QTYPE_QSTRING);
1645         s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
1646     }
1647
1648     if (params->has_tls_authz) {
1649         g_free(s->parameters.tls_authz);
1650         assert(params->tls_authz->type == QTYPE_QSTRING);
1651         s->parameters.tls_authz = g_strdup(params->tls_authz->u.s);
1652     }
1653
1654     if (params->has_max_bandwidth) {
1655         s->parameters.max_bandwidth = params->max_bandwidth;
1656         if (s->to_dst_file && !migration_in_postcopy()) {
1657             qemu_file_set_rate_limit(s->to_dst_file,
1658                                 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
1659         }
1660     }
1661
1662     if (params->has_downtime_limit) {
1663         s->parameters.downtime_limit = params->downtime_limit;
1664     }
1665
1666     if (params->has_x_checkpoint_delay) {
1667         s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
1668         if (migration_in_colo_state()) {
1669             colo_checkpoint_notify(s);
1670         }
1671     }
1672
1673     if (params->has_block_incremental) {
1674         s->parameters.block_incremental = params->block_incremental;
1675     }
1676     if (params->has_multifd_channels) {
1677         s->parameters.multifd_channels = params->multifd_channels;
1678     }
1679     if (params->has_multifd_compression) {
1680         s->parameters.multifd_compression = params->multifd_compression;
1681     }
1682     if (params->has_xbzrle_cache_size) {
1683         s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
1684         xbzrle_cache_resize(params->xbzrle_cache_size, errp);
1685     }
1686     if (params->has_max_postcopy_bandwidth) {
1687         s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth;
1688         if (s->to_dst_file && migration_in_postcopy()) {
1689             qemu_file_set_rate_limit(s->to_dst_file,
1690                     s->parameters.max_postcopy_bandwidth / XFER_LIMIT_RATIO);
1691         }
1692     }
1693     if (params->has_max_cpu_throttle) {
1694         s->parameters.max_cpu_throttle = params->max_cpu_throttle;
1695     }
1696     if (params->has_announce_initial) {
1697         s->parameters.announce_initial = params->announce_initial;
1698     }
1699     if (params->has_announce_max) {
1700         s->parameters.announce_max = params->announce_max;
1701     }
1702     if (params->has_announce_rounds) {
1703         s->parameters.announce_rounds = params->announce_rounds;
1704     }
1705     if (params->has_announce_step) {
1706         s->parameters.announce_step = params->announce_step;
1707     }
1708
1709     if (params->has_block_bitmap_mapping) {
1710         qapi_free_BitmapMigrationNodeAliasList(
1711             s->parameters.block_bitmap_mapping);
1712
1713         s->parameters.has_block_bitmap_mapping = true;
1714         s->parameters.block_bitmap_mapping =
1715             QAPI_CLONE(BitmapMigrationNodeAliasList,
1716                        params->block_bitmap_mapping);
1717     }
1718 }
1719
1720 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
1721 {
1722     MigrationParameters tmp;
1723
1724     /* TODO Rewrite "" to null instead */
1725     if (params->has_tls_creds
1726         && params->tls_creds->type == QTYPE_QNULL) {
1727         qobject_unref(params->tls_creds->u.n);
1728         params->tls_creds->type = QTYPE_QSTRING;
1729         params->tls_creds->u.s = strdup("");
1730     }
1731     /* TODO Rewrite "" to null instead */
1732     if (params->has_tls_hostname
1733         && params->tls_hostname->type == QTYPE_QNULL) {
1734         qobject_unref(params->tls_hostname->u.n);
1735         params->tls_hostname->type = QTYPE_QSTRING;
1736         params->tls_hostname->u.s = strdup("");
1737     }
1738
1739     migrate_params_test_apply(params, &tmp);
1740
1741     if (!migrate_params_check(&tmp, errp)) {
1742         /* Invalid parameter */
1743         return;
1744     }
1745
1746     migrate_params_apply(params, errp);
1747 }
1748
1749
1750 void qmp_migrate_start_postcopy(Error **errp)
1751 {
1752     MigrationState *s = migrate_get_current();
1753
1754     if (!migrate_postcopy()) {
1755         error_setg(errp, "Enable postcopy with migrate_set_capability before"
1756                          " the start of migration");
1757         return;
1758     }
1759
1760     if (s->state == MIGRATION_STATUS_NONE) {
1761         error_setg(errp, "Postcopy must be started after migration has been"
1762                          " started");
1763         return;
1764     }
1765     /*
1766      * we don't error if migration has finished since that would be racy
1767      * with issuing this command.
1768      */
1769     qatomic_set(&s->start_postcopy, true);
1770 }
1771
1772 /* shared migration helpers */
1773
1774 void migrate_set_state(int *state, int old_state, int new_state)
1775 {
1776     assert(new_state < MIGRATION_STATUS__MAX);
1777     if (qatomic_cmpxchg(state, old_state, new_state) == old_state) {
1778         trace_migrate_set_state(MigrationStatus_str(new_state));
1779         migrate_generate_event(new_state);
1780     }
1781 }
1782
1783 static MigrationCapabilityStatus *migrate_cap_add(MigrationCapability index,
1784                                                   bool state)
1785 {
1786     MigrationCapabilityStatus *cap;
1787
1788     cap = g_new0(MigrationCapabilityStatus, 1);
1789     cap->capability = index;
1790     cap->state = state;
1791
1792     return cap;
1793 }
1794
1795 void migrate_set_block_enabled(bool value, Error **errp)
1796 {
1797     MigrationCapabilityStatusList *cap = NULL;
1798
1799     QAPI_LIST_PREPEND(cap, migrate_cap_add(MIGRATION_CAPABILITY_BLOCK, value));
1800     qmp_migrate_set_capabilities(cap, errp);
1801     qapi_free_MigrationCapabilityStatusList(cap);
1802 }
1803
1804 static void migrate_set_block_incremental(MigrationState *s, bool value)
1805 {
1806     s->parameters.block_incremental = value;
1807 }
1808
1809 static void block_cleanup_parameters(MigrationState *s)
1810 {
1811     if (s->must_remove_block_options) {
1812         /* setting to false can never fail */
1813         migrate_set_block_enabled(false, &error_abort);
1814         migrate_set_block_incremental(s, false);
1815         s->must_remove_block_options = false;
1816     }
1817 }
1818
1819 static void migrate_fd_cleanup(MigrationState *s)
1820 {
1821     qemu_bh_delete(s->cleanup_bh);
1822     s->cleanup_bh = NULL;
1823
1824     g_free(s->hostname);
1825     s->hostname = NULL;
1826
1827     qemu_savevm_state_cleanup();
1828
1829     if (s->to_dst_file) {
1830         QEMUFile *tmp;
1831
1832         trace_migrate_fd_cleanup();
1833         qemu_mutex_unlock_iothread();
1834         if (s->migration_thread_running) {
1835             qemu_thread_join(&s->thread);
1836             s->migration_thread_running = false;
1837         }
1838         qemu_mutex_lock_iothread();
1839
1840         multifd_save_cleanup();
1841         qemu_mutex_lock(&s->qemu_file_lock);
1842         tmp = s->to_dst_file;
1843         s->to_dst_file = NULL;
1844         qemu_mutex_unlock(&s->qemu_file_lock);
1845         /*
1846          * Close the file handle without the lock to make sure the
1847          * critical section won't block for long.
1848          */
1849         migration_ioc_unregister_yank_from_file(tmp);
1850         qemu_fclose(tmp);
1851     }
1852
1853     assert(!migration_is_active(s));
1854
1855     if (s->state == MIGRATION_STATUS_CANCELLING) {
1856         migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
1857                           MIGRATION_STATUS_CANCELLED);
1858     }
1859
1860     if (s->error) {
1861         /* It is used on info migrate.  We can't free it */
1862         error_report_err(error_copy(s->error));
1863     }
1864     notifier_list_notify(&migration_state_notifiers, s);
1865     block_cleanup_parameters(s);
1866     yank_unregister_instance(MIGRATION_YANK_INSTANCE);
1867 }
1868
1869 static void migrate_fd_cleanup_schedule(MigrationState *s)
1870 {
1871     /*
1872      * Ref the state for bh, because it may be called when
1873      * there're already no other refs
1874      */
1875     object_ref(OBJECT(s));
1876     qemu_bh_schedule(s->cleanup_bh);
1877 }
1878
1879 static void migrate_fd_cleanup_bh(void *opaque)
1880 {
1881     MigrationState *s = opaque;
1882     migrate_fd_cleanup(s);
1883     object_unref(OBJECT(s));
1884 }
1885
1886 void migrate_set_error(MigrationState *s, const Error *error)
1887 {
1888     QEMU_LOCK_GUARD(&s->error_mutex);
1889     if (!s->error) {
1890         s->error = error_copy(error);
1891     }
1892 }
1893
1894 static void migrate_error_free(MigrationState *s)
1895 {
1896     QEMU_LOCK_GUARD(&s->error_mutex);
1897     if (s->error) {
1898         error_free(s->error);
1899         s->error = NULL;
1900     }
1901 }
1902
1903 void migrate_fd_error(MigrationState *s, const Error *error)
1904 {
1905     trace_migrate_fd_error(error_get_pretty(error));
1906     assert(s->to_dst_file == NULL);
1907     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
1908                       MIGRATION_STATUS_FAILED);
1909     migrate_set_error(s, error);
1910 }
1911
1912 static void migrate_fd_cancel(MigrationState *s)
1913 {
1914     int old_state ;
1915     QEMUFile *f = migrate_get_current()->to_dst_file;
1916     trace_migrate_fd_cancel();
1917
1918     WITH_QEMU_LOCK_GUARD(&s->qemu_file_lock) {
1919         if (s->rp_state.from_dst_file) {
1920             /* shutdown the rp socket, so causing the rp thread to shutdown */
1921             qemu_file_shutdown(s->rp_state.from_dst_file);
1922         }
1923     }
1924
1925     do {
1926         old_state = s->state;
1927         if (!migration_is_running(old_state)) {
1928             break;
1929         }
1930         /* If the migration is paused, kick it out of the pause */
1931         if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
1932             qemu_sem_post(&s->pause_sem);
1933         }
1934         migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
1935     } while (s->state != MIGRATION_STATUS_CANCELLING);
1936
1937     /*
1938      * If we're unlucky the migration code might be stuck somewhere in a
1939      * send/write while the network has failed and is waiting to timeout;
1940      * if we've got shutdown(2) available then we can force it to quit.
1941      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1942      * called in a bh, so there is no race against this cancel.
1943      */
1944     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
1945         qemu_file_shutdown(f);
1946     }
1947     if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
1948         Error *local_err = NULL;
1949
1950         bdrv_activate_all(&local_err);
1951         if (local_err) {
1952             error_report_err(local_err);
1953         } else {
1954             s->block_inactive = false;
1955         }
1956     }
1957 }
1958
1959 void add_migration_state_change_notifier(Notifier *notify)
1960 {
1961     notifier_list_add(&migration_state_notifiers, notify);
1962 }
1963
1964 void remove_migration_state_change_notifier(Notifier *notify)
1965 {
1966     notifier_remove(notify);
1967 }
1968
1969 bool migration_in_setup(MigrationState *s)
1970 {
1971     return s->state == MIGRATION_STATUS_SETUP;
1972 }
1973
1974 bool migration_has_finished(MigrationState *s)
1975 {
1976     return s->state == MIGRATION_STATUS_COMPLETED;
1977 }
1978
1979 bool migration_has_failed(MigrationState *s)
1980 {
1981     return (s->state == MIGRATION_STATUS_CANCELLED ||
1982             s->state == MIGRATION_STATUS_FAILED);
1983 }
1984
1985 bool migration_in_postcopy(void)
1986 {
1987     MigrationState *s = migrate_get_current();
1988
1989     switch (s->state) {
1990     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
1991     case MIGRATION_STATUS_POSTCOPY_PAUSED:
1992     case MIGRATION_STATUS_POSTCOPY_RECOVER:
1993         return true;
1994     default:
1995         return false;
1996     }
1997 }
1998
1999 bool migration_in_postcopy_after_devices(MigrationState *s)
2000 {
2001     return migration_in_postcopy() && s->postcopy_after_devices;
2002 }
2003
2004 bool migration_in_incoming_postcopy(void)
2005 {
2006     PostcopyState ps = postcopy_state_get();
2007
2008     return ps >= POSTCOPY_INCOMING_DISCARD && ps < POSTCOPY_INCOMING_END;
2009 }
2010
2011 bool migration_in_bg_snapshot(void)
2012 {
2013     MigrationState *s = migrate_get_current();
2014
2015     return migrate_background_snapshot() &&
2016             migration_is_setup_or_active(s->state);
2017 }
2018
2019 bool migration_is_idle(void)
2020 {
2021     MigrationState *s = current_migration;
2022
2023     if (!s) {
2024         return true;
2025     }
2026
2027     switch (s->state) {
2028     case MIGRATION_STATUS_NONE:
2029     case MIGRATION_STATUS_CANCELLED:
2030     case MIGRATION_STATUS_COMPLETED:
2031     case MIGRATION_STATUS_FAILED:
2032         return true;
2033     case MIGRATION_STATUS_SETUP:
2034     case MIGRATION_STATUS_CANCELLING:
2035     case MIGRATION_STATUS_ACTIVE:
2036     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
2037     case MIGRATION_STATUS_COLO:
2038     case MIGRATION_STATUS_PRE_SWITCHOVER:
2039     case MIGRATION_STATUS_DEVICE:
2040     case MIGRATION_STATUS_WAIT_UNPLUG:
2041         return false;
2042     case MIGRATION_STATUS__MAX:
2043         g_assert_not_reached();
2044     }
2045
2046     return false;
2047 }
2048
2049 bool migration_is_active(MigrationState *s)
2050 {
2051     return (s->state == MIGRATION_STATUS_ACTIVE ||
2052             s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
2053 }
2054
2055 void migrate_init(MigrationState *s)
2056 {
2057     /*
2058      * Reinitialise all migration state, except
2059      * parameters/capabilities that the user set, and
2060      * locks.
2061      */
2062     s->cleanup_bh = 0;
2063     s->vm_start_bh = 0;
2064     s->to_dst_file = NULL;
2065     s->state = MIGRATION_STATUS_NONE;
2066     s->rp_state.from_dst_file = NULL;
2067     s->rp_state.error = false;
2068     s->mbps = 0.0;
2069     s->pages_per_second = 0.0;
2070     s->downtime = 0;
2071     s->expected_downtime = 0;
2072     s->setup_time = 0;
2073     s->start_postcopy = false;
2074     s->postcopy_after_devices = false;
2075     s->migration_thread_running = false;
2076     error_free(s->error);
2077     s->error = NULL;
2078     s->hostname = NULL;
2079
2080     migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
2081
2082     s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2083     s->total_time = 0;
2084     s->vm_was_running = false;
2085     s->iteration_initial_bytes = 0;
2086     s->threshold_size = 0;
2087 }
2088
2089 int migrate_add_blocker_internal(Error *reason, Error **errp)
2090 {
2091     /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */
2092     if (runstate_check(RUN_STATE_SAVE_VM) || !migration_is_idle()) {
2093         error_propagate_prepend(errp, error_copy(reason),
2094                                 "disallowing migration blocker "
2095                                 "(migration/snapshot in progress) for: ");
2096         return -EBUSY;
2097     }
2098
2099     migration_blockers = g_slist_prepend(migration_blockers, reason);
2100     return 0;
2101 }
2102
2103 int migrate_add_blocker(Error *reason, Error **errp)
2104 {
2105     if (only_migratable) {
2106         error_propagate_prepend(errp, error_copy(reason),
2107                                 "disallowing migration blocker "
2108                                 "(--only-migratable) for: ");
2109         return -EACCES;
2110     }
2111
2112     return migrate_add_blocker_internal(reason, errp);
2113 }
2114
2115 void migrate_del_blocker(Error *reason)
2116 {
2117     migration_blockers = g_slist_remove(migration_blockers, reason);
2118 }
2119
2120 void qmp_migrate_incoming(const char *uri, Error **errp)
2121 {
2122     Error *local_err = NULL;
2123     static bool once = true;
2124
2125     if (!once) {
2126         error_setg(errp, "The incoming migration has already been started");
2127         return;
2128     }
2129     if (!runstate_check(RUN_STATE_INMIGRATE)) {
2130         error_setg(errp, "'-incoming' was not specified on the command line");
2131         return;
2132     }
2133
2134     if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) {
2135         return;
2136     }
2137
2138     qemu_start_incoming_migration(uri, &local_err);
2139
2140     if (local_err) {
2141         yank_unregister_instance(MIGRATION_YANK_INSTANCE);
2142         error_propagate(errp, local_err);
2143         return;
2144     }
2145
2146     once = false;
2147 }
2148
2149 void qmp_migrate_recover(const char *uri, Error **errp)
2150 {
2151     MigrationIncomingState *mis = migration_incoming_get_current();
2152
2153     /*
2154      * Don't even bother to use ERRP_GUARD() as it _must_ always be set by
2155      * callers (no one should ignore a recover failure); if there is, it's a
2156      * programming error.
2157      */
2158     assert(errp);
2159
2160     if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
2161         error_setg(errp, "Migrate recover can only be run "
2162                    "when postcopy is paused.");
2163         return;
2164     }
2165
2166     /* If there's an existing transport, release it */
2167     migration_incoming_transport_cleanup(mis);
2168
2169     /*
2170      * Note that this call will never start a real migration; it will
2171      * only re-setup the migration stream and poke existing migration
2172      * to continue using that newly established channel.
2173      */
2174     qemu_start_incoming_migration(uri, errp);
2175 }
2176
2177 void qmp_migrate_pause(Error **errp)
2178 {
2179     MigrationState *ms = migrate_get_current();
2180     MigrationIncomingState *mis = migration_incoming_get_current();
2181     int ret;
2182
2183     if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2184         /* Source side, during postcopy */
2185         qemu_mutex_lock(&ms->qemu_file_lock);
2186         ret = qemu_file_shutdown(ms->to_dst_file);
2187         qemu_mutex_unlock(&ms->qemu_file_lock);
2188         if (ret) {
2189             error_setg(errp, "Failed to pause source migration");
2190         }
2191         return;
2192     }
2193
2194     if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
2195         ret = qemu_file_shutdown(mis->from_src_file);
2196         if (ret) {
2197             error_setg(errp, "Failed to pause destination migration");
2198         }
2199         return;
2200     }
2201
2202     error_setg(errp, "migrate-pause is currently only supported "
2203                "during postcopy-active state");
2204 }
2205
2206 bool migration_is_blocked(Error **errp)
2207 {
2208     if (qemu_savevm_state_blocked(errp)) {
2209         return true;
2210     }
2211
2212     if (migration_blockers) {
2213         error_propagate(errp, error_copy(migration_blockers->data));
2214         return true;
2215     }
2216
2217     return false;
2218 }
2219
2220 /* Returns true if continue to migrate, or false if error detected */
2221 static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
2222                             bool resume, Error **errp)
2223 {
2224     Error *local_err = NULL;
2225
2226     if (resume) {
2227         if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
2228             error_setg(errp, "Cannot resume if there is no "
2229                        "paused migration");
2230             return false;
2231         }
2232
2233         /*
2234          * Postcopy recovery won't work well with release-ram
2235          * capability since release-ram will drop the page buffer as
2236          * long as the page is put into the send buffer.  So if there
2237          * is a network failure happened, any page buffers that have
2238          * not yet reached the destination VM but have already been
2239          * sent from the source VM will be lost forever.  Let's refuse
2240          * the client from resuming such a postcopy migration.
2241          * Luckily release-ram was designed to only be used when src
2242          * and destination VMs are on the same host, so it should be
2243          * fine.
2244          */
2245         if (migrate_release_ram()) {
2246             error_setg(errp, "Postcopy recovery cannot work "
2247                        "when release-ram capability is set");
2248             return false;
2249         }
2250
2251         /* This is a resume, skip init status */
2252         return true;
2253     }
2254
2255     if (migration_is_running(s->state)) {
2256         error_setg(errp, QERR_MIGRATION_ACTIVE);
2257         return false;
2258     }
2259
2260     if (runstate_check(RUN_STATE_INMIGRATE)) {
2261         error_setg(errp, "Guest is waiting for an incoming migration");
2262         return false;
2263     }
2264
2265     if (runstate_check(RUN_STATE_POSTMIGRATE)) {
2266         error_setg(errp, "Can't migrate the vm that was paused due to "
2267                    "previous migration");
2268         return false;
2269     }
2270
2271     if (migration_is_blocked(errp)) {
2272         return false;
2273     }
2274
2275     if (blk || blk_inc) {
2276         if (migrate_colo_enabled()) {
2277             error_setg(errp, "No disk migration is required in COLO mode");
2278             return false;
2279         }
2280         if (migrate_use_block() || migrate_use_block_incremental()) {
2281             error_setg(errp, "Command options are incompatible with "
2282                        "current migration capabilities");
2283             return false;
2284         }
2285         migrate_set_block_enabled(true, &local_err);
2286         if (local_err) {
2287             error_propagate(errp, local_err);
2288             return false;
2289         }
2290         s->must_remove_block_options = true;
2291     }
2292
2293     if (blk_inc) {
2294         migrate_set_block_incremental(s, true);
2295     }
2296
2297     migrate_init(s);
2298     /*
2299      * set ram_counters compression_counters memory to zero for a
2300      * new migration
2301      */
2302     memset(&ram_counters, 0, sizeof(ram_counters));
2303     memset(&compression_counters, 0, sizeof(compression_counters));
2304
2305     return true;
2306 }
2307
2308 void qmp_migrate(const char *uri, bool has_blk, bool blk,
2309                  bool has_inc, bool inc, bool has_detach, bool detach,
2310                  bool has_resume, bool resume, Error **errp)
2311 {
2312     Error *local_err = NULL;
2313     MigrationState *s = migrate_get_current();
2314     const char *p = NULL;
2315
2316     if (!migrate_prepare(s, has_blk && blk, has_inc && inc,
2317                          has_resume && resume, errp)) {
2318         /* Error detected, put into errp */
2319         return;
2320     }
2321
2322     if (!(has_resume && resume)) {
2323         if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) {
2324             return;
2325         }
2326     }
2327
2328     migrate_protocol_allow_multi_channels(false);
2329     if (strstart(uri, "tcp:", &p) ||
2330         strstart(uri, "unix:", NULL) ||
2331         strstart(uri, "vsock:", NULL)) {
2332         migrate_protocol_allow_multi_channels(true);
2333         socket_start_outgoing_migration(s, p ? p : uri, &local_err);
2334 #ifdef CONFIG_RDMA
2335     } else if (strstart(uri, "rdma:", &p)) {
2336         rdma_start_outgoing_migration(s, p, &local_err);
2337 #endif
2338     } else if (strstart(uri, "exec:", &p)) {
2339         exec_start_outgoing_migration(s, p, &local_err);
2340     } else if (strstart(uri, "fd:", &p)) {
2341         fd_start_outgoing_migration(s, p, &local_err);
2342     } else {
2343         if (!(has_resume && resume)) {
2344             yank_unregister_instance(MIGRATION_YANK_INSTANCE);
2345         }
2346         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
2347                    "a valid migration protocol");
2348         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
2349                           MIGRATION_STATUS_FAILED);
2350         block_cleanup_parameters(s);
2351         return;
2352     }
2353
2354     if (local_err) {
2355         if (!(has_resume && resume)) {
2356             yank_unregister_instance(MIGRATION_YANK_INSTANCE);
2357         }
2358         migrate_fd_error(s, local_err);
2359         error_propagate(errp, local_err);
2360         return;
2361     }
2362 }
2363
2364 void qmp_migrate_cancel(Error **errp)
2365 {
2366     migration_cancel(NULL);
2367 }
2368
2369 void qmp_migrate_continue(MigrationStatus state, Error **errp)
2370 {
2371     MigrationState *s = migrate_get_current();
2372     if (s->state != state) {
2373         error_setg(errp,  "Migration not in expected state: %s",
2374                    MigrationStatus_str(s->state));
2375         return;
2376     }
2377     qemu_sem_post(&s->pause_sem);
2378 }
2379
2380 bool migrate_release_ram(void)
2381 {
2382     MigrationState *s;
2383
2384     s = migrate_get_current();
2385
2386     return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
2387 }
2388
2389 bool migrate_postcopy_ram(void)
2390 {
2391     MigrationState *s;
2392
2393     s = migrate_get_current();
2394
2395     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
2396 }
2397
2398 bool migrate_postcopy(void)
2399 {
2400     return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2401 }
2402
2403 bool migrate_auto_converge(void)
2404 {
2405     MigrationState *s;
2406
2407     s = migrate_get_current();
2408
2409     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
2410 }
2411
2412 bool migrate_zero_blocks(void)
2413 {
2414     MigrationState *s;
2415
2416     s = migrate_get_current();
2417
2418     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
2419 }
2420
2421 bool migrate_postcopy_blocktime(void)
2422 {
2423     MigrationState *s;
2424
2425     s = migrate_get_current();
2426
2427     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
2428 }
2429
2430 bool migrate_use_compression(void)
2431 {
2432     MigrationState *s;
2433
2434     s = migrate_get_current();
2435
2436     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
2437 }
2438
2439 int migrate_compress_level(void)
2440 {
2441     MigrationState *s;
2442
2443     s = migrate_get_current();
2444
2445     return s->parameters.compress_level;
2446 }
2447
2448 int migrate_compress_threads(void)
2449 {
2450     MigrationState *s;
2451
2452     s = migrate_get_current();
2453
2454     return s->parameters.compress_threads;
2455 }
2456
2457 int migrate_compress_wait_thread(void)
2458 {
2459     MigrationState *s;
2460
2461     s = migrate_get_current();
2462
2463     return s->parameters.compress_wait_thread;
2464 }
2465
2466 int migrate_decompress_threads(void)
2467 {
2468     MigrationState *s;
2469
2470     s = migrate_get_current();
2471
2472     return s->parameters.decompress_threads;
2473 }
2474
2475 bool migrate_dirty_bitmaps(void)
2476 {
2477     MigrationState *s;
2478
2479     s = migrate_get_current();
2480
2481     return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS];
2482 }
2483
2484 bool migrate_ignore_shared(void)
2485 {
2486     MigrationState *s;
2487
2488     s = migrate_get_current();
2489
2490     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_IGNORE_SHARED];
2491 }
2492
2493 bool migrate_validate_uuid(void)
2494 {
2495     MigrationState *s;
2496
2497     s = migrate_get_current();
2498
2499     return s->enabled_capabilities[MIGRATION_CAPABILITY_VALIDATE_UUID];
2500 }
2501
2502 bool migrate_use_events(void)
2503 {
2504     MigrationState *s;
2505
2506     s = migrate_get_current();
2507
2508     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
2509 }
2510
2511 bool migrate_use_multifd(void)
2512 {
2513     MigrationState *s;
2514
2515     s = migrate_get_current();
2516
2517     return s->enabled_capabilities[MIGRATION_CAPABILITY_MULTIFD];
2518 }
2519
2520 bool migrate_pause_before_switchover(void)
2521 {
2522     MigrationState *s;
2523
2524     s = migrate_get_current();
2525
2526     return s->enabled_capabilities[
2527         MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
2528 }
2529
2530 int migrate_multifd_channels(void)
2531 {
2532     MigrationState *s;
2533
2534     s = migrate_get_current();
2535
2536     return s->parameters.multifd_channels;
2537 }
2538
2539 MultiFDCompression migrate_multifd_compression(void)
2540 {
2541     MigrationState *s;
2542
2543     s = migrate_get_current();
2544
2545     return s->parameters.multifd_compression;
2546 }
2547
2548 int migrate_multifd_zlib_level(void)
2549 {
2550     MigrationState *s;
2551
2552     s = migrate_get_current();
2553
2554     return s->parameters.multifd_zlib_level;
2555 }
2556
2557 int migrate_multifd_zstd_level(void)
2558 {
2559     MigrationState *s;
2560
2561     s = migrate_get_current();
2562
2563     return s->parameters.multifd_zstd_level;
2564 }
2565
2566 int migrate_use_xbzrle(void)
2567 {
2568     MigrationState *s;
2569
2570     s = migrate_get_current();
2571
2572     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
2573 }
2574
2575 uint64_t migrate_xbzrle_cache_size(void)
2576 {
2577     MigrationState *s;
2578
2579     s = migrate_get_current();
2580
2581     return s->parameters.xbzrle_cache_size;
2582 }
2583
2584 static int64_t migrate_max_postcopy_bandwidth(void)
2585 {
2586     MigrationState *s;
2587
2588     s = migrate_get_current();
2589
2590     return s->parameters.max_postcopy_bandwidth;
2591 }
2592
2593 bool migrate_use_block(void)
2594 {
2595     MigrationState *s;
2596
2597     s = migrate_get_current();
2598
2599     return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
2600 }
2601
2602 bool migrate_use_return_path(void)
2603 {
2604     MigrationState *s;
2605
2606     s = migrate_get_current();
2607
2608     return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
2609 }
2610
2611 bool migrate_use_block_incremental(void)
2612 {
2613     MigrationState *s;
2614
2615     s = migrate_get_current();
2616
2617     return s->parameters.block_incremental;
2618 }
2619
2620 bool migrate_background_snapshot(void)
2621 {
2622     MigrationState *s;
2623
2624     s = migrate_get_current();
2625
2626     return s->enabled_capabilities[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT];
2627 }
2628
2629 /* migration thread support */
2630 /*
2631  * Something bad happened to the RP stream, mark an error
2632  * The caller shall print or trace something to indicate why
2633  */
2634 static void mark_source_rp_bad(MigrationState *s)
2635 {
2636     s->rp_state.error = true;
2637 }
2638
2639 static struct rp_cmd_args {
2640     ssize_t     len; /* -1 = variable */
2641     const char *name;
2642 } rp_cmd_args[] = {
2643     [MIG_RP_MSG_INVALID]        = { .len = -1, .name = "INVALID" },
2644     [MIG_RP_MSG_SHUT]           = { .len =  4, .name = "SHUT" },
2645     [MIG_RP_MSG_PONG]           = { .len =  4, .name = "PONG" },
2646     [MIG_RP_MSG_REQ_PAGES]      = { .len = 12, .name = "REQ_PAGES" },
2647     [MIG_RP_MSG_REQ_PAGES_ID]   = { .len = -1, .name = "REQ_PAGES_ID" },
2648     [MIG_RP_MSG_RECV_BITMAP]    = { .len = -1, .name = "RECV_BITMAP" },
2649     [MIG_RP_MSG_RESUME_ACK]     = { .len =  4, .name = "RESUME_ACK" },
2650     [MIG_RP_MSG_MAX]            = { .len = -1, .name = "MAX" },
2651 };
2652
2653 /*
2654  * Process a request for pages received on the return path,
2655  * We're allowed to send more than requested (e.g. to round to our page size)
2656  * and we don't need to send pages that have already been sent.
2657  */
2658 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
2659                                        ram_addr_t start, size_t len)
2660 {
2661     long our_host_ps = qemu_real_host_page_size();
2662
2663     trace_migrate_handle_rp_req_pages(rbname, start, len);
2664
2665     /*
2666      * Since we currently insist on matching page sizes, just sanity check
2667      * we're being asked for whole host pages.
2668      */
2669     if (!QEMU_IS_ALIGNED(start, our_host_ps) ||
2670         !QEMU_IS_ALIGNED(len, our_host_ps)) {
2671         error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2672                      " len: %zd", __func__, start, len);
2673         mark_source_rp_bad(ms);
2674         return;
2675     }
2676
2677     if (ram_save_queue_pages(rbname, start, len)) {
2678         mark_source_rp_bad(ms);
2679     }
2680 }
2681
2682 /* Return true to retry, false to quit */
2683 static bool postcopy_pause_return_path_thread(MigrationState *s)
2684 {
2685     trace_postcopy_pause_return_path();
2686
2687     qemu_sem_wait(&s->postcopy_pause_rp_sem);
2688
2689     trace_postcopy_pause_return_path_continued();
2690
2691     return true;
2692 }
2693
2694 static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name)
2695 {
2696     RAMBlock *block = qemu_ram_block_by_name(block_name);
2697
2698     if (!block) {
2699         error_report("%s: invalid block name '%s'", __func__, block_name);
2700         return -EINVAL;
2701     }
2702
2703     /* Fetch the received bitmap and refresh the dirty bitmap */
2704     return ram_dirty_bitmap_reload(s, block);
2705 }
2706
2707 static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
2708 {
2709     trace_source_return_path_thread_resume_ack(value);
2710
2711     if (value != MIGRATION_RESUME_ACK_VALUE) {
2712         error_report("%s: illegal resume_ack value %"PRIu32,
2713                      __func__, value);
2714         return -1;
2715     }
2716
2717     /* Now both sides are active. */
2718     migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
2719                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
2720
2721     /* Notify send thread that time to continue send pages */
2722     qemu_sem_post(&s->rp_state.rp_sem);
2723
2724     return 0;
2725 }
2726
2727 /* Release ms->rp_state.from_dst_file in a safe way */
2728 static void migration_release_from_dst_file(MigrationState *ms)
2729 {
2730     QEMUFile *file;
2731
2732     WITH_QEMU_LOCK_GUARD(&ms->qemu_file_lock) {
2733         /*
2734          * Reset the from_dst_file pointer first before releasing it, as we
2735          * can't block within lock section
2736          */
2737         file = ms->rp_state.from_dst_file;
2738         ms->rp_state.from_dst_file = NULL;
2739     }
2740
2741     qemu_fclose(file);
2742 }
2743
2744 /*
2745  * Handles messages sent on the return path towards the source VM
2746  *
2747  */
2748 static void *source_return_path_thread(void *opaque)
2749 {
2750     MigrationState *ms = opaque;
2751     QEMUFile *rp = ms->rp_state.from_dst_file;
2752     uint16_t header_len, header_type;
2753     uint8_t buf[512];
2754     uint32_t tmp32, sibling_error;
2755     ram_addr_t start = 0; /* =0 to silence warning */
2756     size_t  len = 0, expected_len;
2757     int res;
2758
2759     trace_source_return_path_thread_entry();
2760     rcu_register_thread();
2761
2762 retry:
2763     while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
2764            migration_is_setup_or_active(ms->state)) {
2765         trace_source_return_path_thread_loop_top();
2766         header_type = qemu_get_be16(rp);
2767         header_len = qemu_get_be16(rp);
2768
2769         if (qemu_file_get_error(rp)) {
2770             mark_source_rp_bad(ms);
2771             goto out;
2772         }
2773
2774         if (header_type >= MIG_RP_MSG_MAX ||
2775             header_type == MIG_RP_MSG_INVALID) {
2776             error_report("RP: Received invalid message 0x%04x length 0x%04x",
2777                          header_type, header_len);
2778             mark_source_rp_bad(ms);
2779             goto out;
2780         }
2781
2782         if ((rp_cmd_args[header_type].len != -1 &&
2783             header_len != rp_cmd_args[header_type].len) ||
2784             header_len > sizeof(buf)) {
2785             error_report("RP: Received '%s' message (0x%04x) with"
2786                          "incorrect length %d expecting %zu",
2787                          rp_cmd_args[header_type].name, header_type, header_len,
2788                          (size_t)rp_cmd_args[header_type].len);
2789             mark_source_rp_bad(ms);
2790             goto out;
2791         }
2792
2793         /* We know we've got a valid header by this point */
2794         res = qemu_get_buffer(rp, buf, header_len);
2795         if (res != header_len) {
2796             error_report("RP: Failed reading data for message 0x%04x"
2797                          " read %d expected %d",
2798                          header_type, res, header_len);
2799             mark_source_rp_bad(ms);
2800             goto out;
2801         }
2802
2803         /* OK, we have the message and the data */
2804         switch (header_type) {
2805         case MIG_RP_MSG_SHUT:
2806             sibling_error = ldl_be_p(buf);
2807             trace_source_return_path_thread_shut(sibling_error);
2808             if (sibling_error) {
2809                 error_report("RP: Sibling indicated error %d", sibling_error);
2810                 mark_source_rp_bad(ms);
2811             }
2812             /*
2813              * We'll let the main thread deal with closing the RP
2814              * we could do a shutdown(2) on it, but we're the only user
2815              * anyway, so there's nothing gained.
2816              */
2817             goto out;
2818
2819         case MIG_RP_MSG_PONG:
2820             tmp32 = ldl_be_p(buf);
2821             trace_source_return_path_thread_pong(tmp32);
2822             break;
2823
2824         case MIG_RP_MSG_REQ_PAGES:
2825             start = ldq_be_p(buf);
2826             len = ldl_be_p(buf + 8);
2827             migrate_handle_rp_req_pages(ms, NULL, start, len);
2828             break;
2829
2830         case MIG_RP_MSG_REQ_PAGES_ID:
2831             expected_len = 12 + 1; /* header + termination */
2832
2833             if (header_len >= expected_len) {
2834                 start = ldq_be_p(buf);
2835                 len = ldl_be_p(buf + 8);
2836                 /* Now we expect an idstr */
2837                 tmp32 = buf[12]; /* Length of the following idstr */
2838                 buf[13 + tmp32] = '\0';
2839                 expected_len += tmp32;
2840             }
2841             if (header_len != expected_len) {
2842                 error_report("RP: Req_Page_id with length %d expecting %zd",
2843                              header_len, expected_len);
2844                 mark_source_rp_bad(ms);
2845                 goto out;
2846             }
2847             migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
2848             break;
2849
2850         case MIG_RP_MSG_RECV_BITMAP:
2851             if (header_len < 1) {
2852                 error_report("%s: missing block name", __func__);
2853                 mark_source_rp_bad(ms);
2854                 goto out;
2855             }
2856             /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2857             buf[buf[0] + 1] = '\0';
2858             if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) {
2859                 mark_source_rp_bad(ms);
2860                 goto out;
2861             }
2862             break;
2863
2864         case MIG_RP_MSG_RESUME_ACK:
2865             tmp32 = ldl_be_p(buf);
2866             if (migrate_handle_rp_resume_ack(ms, tmp32)) {
2867                 mark_source_rp_bad(ms);
2868                 goto out;
2869             }
2870             break;
2871
2872         default:
2873             break;
2874         }
2875     }
2876
2877 out:
2878     res = qemu_file_get_error(rp);
2879     if (res) {
2880         if (res && migration_in_postcopy()) {
2881             /*
2882              * Maybe there is something we can do: it looks like a
2883              * network down issue, and we pause for a recovery.
2884              */
2885             migration_release_from_dst_file(ms);
2886             rp = NULL;
2887             if (postcopy_pause_return_path_thread(ms)) {
2888                 /*
2889                  * Reload rp, reset the rest.  Referencing it is safe since
2890                  * it's reset only by us above, or when migration completes
2891                  */
2892                 rp = ms->rp_state.from_dst_file;
2893                 ms->rp_state.error = false;
2894                 goto retry;
2895             }
2896         }
2897
2898         trace_source_return_path_thread_bad_end();
2899         mark_source_rp_bad(ms);
2900     }
2901
2902     trace_source_return_path_thread_end();
2903     migration_release_from_dst_file(ms);
2904     rcu_unregister_thread();
2905     return NULL;
2906 }
2907
2908 static int open_return_path_on_source(MigrationState *ms,
2909                                       bool create_thread)
2910 {
2911     ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
2912     if (!ms->rp_state.from_dst_file) {
2913         return -1;
2914     }
2915
2916     trace_open_return_path_on_source();
2917
2918     if (!create_thread) {
2919         /* We're done */
2920         return 0;
2921     }
2922
2923     qemu_thread_create(&ms->rp_state.rp_thread, "return path",
2924                        source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
2925     ms->rp_state.rp_thread_created = true;
2926
2927     trace_open_return_path_on_source_continue();
2928
2929     return 0;
2930 }
2931
2932 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2933 static int await_return_path_close_on_source(MigrationState *ms)
2934 {
2935     /*
2936      * If this is a normal exit then the destination will send a SHUT and the
2937      * rp_thread will exit, however if there's an error we need to cause
2938      * it to exit.
2939      */
2940     if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
2941         /*
2942          * shutdown(2), if we have it, will cause it to unblock if it's stuck
2943          * waiting for the destination.
2944          */
2945         qemu_file_shutdown(ms->rp_state.from_dst_file);
2946         mark_source_rp_bad(ms);
2947     }
2948     trace_await_return_path_close_on_source_joining();
2949     qemu_thread_join(&ms->rp_state.rp_thread);
2950     ms->rp_state.rp_thread_created = false;
2951     trace_await_return_path_close_on_source_close();
2952     return ms->rp_state.error;
2953 }
2954
2955 /*
2956  * Switch from normal iteration to postcopy
2957  * Returns non-0 on error
2958  */
2959 static int postcopy_start(MigrationState *ms)
2960 {
2961     int ret;
2962     QIOChannelBuffer *bioc;
2963     QEMUFile *fb;
2964     int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
2965     int64_t bandwidth = migrate_max_postcopy_bandwidth();
2966     bool restart_block = false;
2967     int cur_state = MIGRATION_STATUS_ACTIVE;
2968     if (!migrate_pause_before_switchover()) {
2969         migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
2970                           MIGRATION_STATUS_POSTCOPY_ACTIVE);
2971     }
2972
2973     trace_postcopy_start();
2974     qemu_mutex_lock_iothread();
2975     trace_postcopy_start_set_run();
2976
2977     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
2978     global_state_store();
2979     ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
2980     if (ret < 0) {
2981         goto fail;
2982     }
2983
2984     ret = migration_maybe_pause(ms, &cur_state,
2985                                 MIGRATION_STATUS_POSTCOPY_ACTIVE);
2986     if (ret < 0) {
2987         goto fail;
2988     }
2989
2990     ret = bdrv_inactivate_all();
2991     if (ret < 0) {
2992         goto fail;
2993     }
2994     restart_block = true;
2995
2996     /*
2997      * Cause any non-postcopiable, but iterative devices to
2998      * send out their final data.
2999      */
3000     qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
3001
3002     /*
3003      * in Finish migrate and with the io-lock held everything should
3004      * be quiet, but we've potentially still got dirty pages and we
3005      * need to tell the destination to throw any pages it's already received
3006      * that are dirty
3007      */
3008     if (migrate_postcopy_ram()) {
3009         ram_postcopy_send_discard_bitmap(ms);
3010     }
3011
3012     /*
3013      * send rest of state - note things that are doing postcopy
3014      * will notice we're in POSTCOPY_ACTIVE and not actually
3015      * wrap their state up here
3016      */
3017     /* 0 max-postcopy-bandwidth means unlimited */
3018     if (!bandwidth) {
3019         qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
3020     } else {
3021         qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO);
3022     }
3023     if (migrate_postcopy_ram()) {
3024         /* Ping just for debugging, helps line traces up */
3025         qemu_savevm_send_ping(ms->to_dst_file, 2);
3026     }
3027
3028     /*
3029      * While loading the device state we may trigger page transfer
3030      * requests and the fd must be free to process those, and thus
3031      * the destination must read the whole device state off the fd before
3032      * it starts processing it.  Unfortunately the ad-hoc migration format
3033      * doesn't allow the destination to know the size to read without fully
3034      * parsing it through each devices load-state code (especially the open
3035      * coded devices that use get/put).
3036      * So we wrap the device state up in a package with a length at the start;
3037      * to do this we use a qemu_buf to hold the whole of the device state.
3038      */
3039     bioc = qio_channel_buffer_new(4096);
3040     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
3041     fb = qemu_fopen_channel_output(QIO_CHANNEL(bioc));
3042     object_unref(OBJECT(bioc));
3043
3044     /*
3045      * Make sure the receiver can get incoming pages before we send the rest
3046      * of the state
3047      */
3048     qemu_savevm_send_postcopy_listen(fb);
3049
3050     qemu_savevm_state_complete_precopy(fb, false, false);
3051     if (migrate_postcopy_ram()) {
3052         qemu_savevm_send_ping(fb, 3);
3053     }
3054
3055     qemu_savevm_send_postcopy_run(fb);
3056
3057     /* <><> end of stuff going into the package */
3058
3059     /* Last point of recovery; as soon as we send the package the destination
3060      * can open devices and potentially start running.
3061      * Lets just check again we've not got any errors.
3062      */
3063     ret = qemu_file_get_error(ms->to_dst_file);
3064     if (ret) {
3065         error_report("postcopy_start: Migration stream errored (pre package)");
3066         goto fail_closefb;
3067     }
3068
3069     restart_block = false;
3070
3071     /* Now send that blob */
3072     if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
3073         goto fail_closefb;
3074     }
3075     qemu_fclose(fb);
3076
3077     /* Send a notify to give a chance for anything that needs to happen
3078      * at the transition to postcopy and after the device state; in particular
3079      * spice needs to trigger a transition now
3080      */
3081     ms->postcopy_after_devices = true;
3082     notifier_list_notify(&migration_state_notifiers, ms);
3083
3084     ms->downtime =  qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
3085
3086     qemu_mutex_unlock_iothread();
3087
3088     if (migrate_postcopy_ram()) {
3089         /*
3090          * Although this ping is just for debug, it could potentially be
3091          * used for getting a better measurement of downtime at the source.
3092          */
3093         qemu_savevm_send_ping(ms->to_dst_file, 4);
3094     }
3095
3096     if (migrate_release_ram()) {
3097         ram_postcopy_migrated_memory_release(ms);
3098     }
3099
3100     ret = qemu_file_get_error(ms->to_dst_file);
3101     if (ret) {
3102         error_report("postcopy_start: Migration stream errored");
3103         migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
3104                               MIGRATION_STATUS_FAILED);
3105     }
3106
3107     return ret;
3108
3109 fail_closefb:
3110     qemu_fclose(fb);
3111 fail:
3112     migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
3113                           MIGRATION_STATUS_FAILED);
3114     if (restart_block) {
3115         /* A failure happened early enough that we know the destination hasn't
3116          * accessed block devices, so we're safe to recover.
3117          */
3118         Error *local_err = NULL;
3119
3120         bdrv_activate_all(&local_err);
3121         if (local_err) {
3122             error_report_err(local_err);
3123         }
3124     }
3125     qemu_mutex_unlock_iothread();
3126     return -1;
3127 }
3128
3129 /**
3130  * migration_maybe_pause: Pause if required to by
3131  * migrate_pause_before_switchover called with the iothread locked
3132  * Returns: 0 on success
3133  */
3134 static int migration_maybe_pause(MigrationState *s,
3135                                  int *current_active_state,
3136                                  int new_state)
3137 {
3138     if (!migrate_pause_before_switchover()) {
3139         return 0;
3140     }
3141
3142     /* Since leaving this state is not atomic with posting the semaphore
3143      * it's possible that someone could have issued multiple migrate_continue
3144      * and the semaphore is incorrectly positive at this point;
3145      * the docs say it's undefined to reinit a semaphore that's already
3146      * init'd, so use timedwait to eat up any existing posts.
3147      */
3148     while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
3149         /* This block intentionally left blank */
3150     }
3151
3152     /*
3153      * If the migration is cancelled when it is in the completion phase,
3154      * the migration state is set to MIGRATION_STATUS_CANCELLING.
3155      * So we don't need to wait a semaphore, otherwise we would always
3156      * wait for the 'pause_sem' semaphore.
3157      */
3158     if (s->state != MIGRATION_STATUS_CANCELLING) {
3159         qemu_mutex_unlock_iothread();
3160         migrate_set_state(&s->state, *current_active_state,
3161                           MIGRATION_STATUS_PRE_SWITCHOVER);
3162         qemu_sem_wait(&s->pause_sem);
3163         migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
3164                           new_state);
3165         *current_active_state = new_state;
3166         qemu_mutex_lock_iothread();
3167     }
3168
3169     return s->state == new_state ? 0 : -EINVAL;
3170 }
3171
3172 /**
3173  * migration_completion: Used by migration_thread when there's not much left.
3174  *   The caller 'breaks' the loop when this returns.
3175  *
3176  * @s: Current migration state
3177  */
3178 static void migration_completion(MigrationState *s)
3179 {
3180     int ret;
3181     int current_active_state = s->state;
3182
3183     if (s->state == MIGRATION_STATUS_ACTIVE) {
3184         qemu_mutex_lock_iothread();
3185         s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3186         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
3187         s->vm_was_running = runstate_is_running();
3188         ret = global_state_store();
3189
3190         if (!ret) {
3191             bool inactivate = !migrate_colo_enabled();
3192             ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
3193             trace_migration_completion_vm_stop(ret);
3194             if (ret >= 0) {
3195                 ret = migration_maybe_pause(s, &current_active_state,
3196                                             MIGRATION_STATUS_DEVICE);
3197             }
3198             if (ret >= 0) {
3199                 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
3200                 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
3201                                                          inactivate);
3202             }
3203             if (inactivate && ret >= 0) {
3204                 s->block_inactive = true;
3205             }
3206         }
3207         qemu_mutex_unlock_iothread();
3208
3209         if (ret < 0) {
3210             goto fail;
3211         }
3212     } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
3213         trace_migration_completion_postcopy_end();
3214
3215         qemu_mutex_lock_iothread();
3216         qemu_savevm_state_complete_postcopy(s->to_dst_file);
3217         qemu_mutex_unlock_iothread();
3218
3219         trace_migration_completion_postcopy_end_after_complete();
3220     } else {
3221         goto fail;
3222     }
3223
3224     /*
3225      * If rp was opened we must clean up the thread before
3226      * cleaning everything else up (since if there are no failures
3227      * it will wait for the destination to send it's status in
3228      * a SHUT command).
3229      */
3230     if (s->rp_state.rp_thread_created) {
3231         int rp_error;
3232         trace_migration_return_path_end_before();
3233         rp_error = await_return_path_close_on_source(s);
3234         trace_migration_return_path_end_after(rp_error);
3235         if (rp_error) {
3236             goto fail_invalidate;
3237         }
3238     }
3239
3240     if (qemu_file_get_error(s->to_dst_file)) {
3241         trace_migration_completion_file_err();
3242         goto fail_invalidate;
3243     }
3244
3245     if (migrate_colo_enabled() && s->state == MIGRATION_STATUS_ACTIVE) {
3246         /* COLO does not support postcopy */
3247         migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE,
3248                           MIGRATION_STATUS_COLO);
3249     } else {
3250         migrate_set_state(&s->state, current_active_state,
3251                           MIGRATION_STATUS_COMPLETED);
3252     }
3253
3254     return;
3255
3256 fail_invalidate:
3257     /* If not doing postcopy, vm_start() will be called: let's regain
3258      * control on images.
3259      */
3260     if (s->state == MIGRATION_STATUS_ACTIVE ||
3261         s->state == MIGRATION_STATUS_DEVICE) {
3262         Error *local_err = NULL;
3263
3264         qemu_mutex_lock_iothread();
3265         bdrv_activate_all(&local_err);
3266         if (local_err) {
3267             error_report_err(local_err);
3268         } else {
3269             s->block_inactive = false;
3270         }
3271         qemu_mutex_unlock_iothread();
3272     }
3273
3274 fail:
3275     migrate_set_state(&s->state, current_active_state,
3276                       MIGRATION_STATUS_FAILED);
3277 }
3278
3279 /**
3280  * bg_migration_completion: Used by bg_migration_thread when after all the
3281  *   RAM has been saved. The caller 'breaks' the loop when this returns.
3282  *
3283  * @s: Current migration state
3284  */
3285 static void bg_migration_completion(MigrationState *s)
3286 {
3287     int current_active_state = s->state;
3288
3289     /*
3290      * Stop tracking RAM writes - un-protect memory, un-register UFFD
3291      * memory ranges, flush kernel wait queues and wake up threads
3292      * waiting for write fault to be resolved.
3293      */
3294     ram_write_tracking_stop();
3295
3296     if (s->state == MIGRATION_STATUS_ACTIVE) {
3297         /*
3298          * By this moment we have RAM content saved into the migration stream.
3299          * The next step is to flush the non-RAM content (device state)
3300          * right after the ram content. The device state has been stored into
3301          * the temporary buffer before RAM saving started.
3302          */
3303         qemu_put_buffer(s->to_dst_file, s->bioc->data, s->bioc->usage);
3304         qemu_fflush(s->to_dst_file);
3305     } else if (s->state == MIGRATION_STATUS_CANCELLING) {
3306         goto fail;
3307     }
3308
3309     if (qemu_file_get_error(s->to_dst_file)) {
3310         trace_migration_completion_file_err();
3311         goto fail;
3312     }
3313
3314     migrate_set_state(&s->state, current_active_state,
3315                       MIGRATION_STATUS_COMPLETED);
3316     return;
3317
3318 fail:
3319     migrate_set_state(&s->state, current_active_state,
3320                       MIGRATION_STATUS_FAILED);
3321 }
3322
3323 bool migrate_colo_enabled(void)
3324 {
3325     MigrationState *s = migrate_get_current();
3326     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
3327 }
3328
3329 typedef enum MigThrError {
3330     /* No error detected */
3331     MIG_THR_ERR_NONE = 0,
3332     /* Detected error, but resumed successfully */
3333     MIG_THR_ERR_RECOVERED = 1,
3334     /* Detected fatal error, need to exit */
3335     MIG_THR_ERR_FATAL = 2,
3336 } MigThrError;
3337
3338 static int postcopy_resume_handshake(MigrationState *s)
3339 {
3340     qemu_savevm_send_postcopy_resume(s->to_dst_file);
3341
3342     while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
3343         qemu_sem_wait(&s->rp_state.rp_sem);
3344     }
3345
3346     if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
3347         return 0;
3348     }
3349
3350     return -1;
3351 }
3352
3353 /* Return zero if success, or <0 for error */
3354 static int postcopy_do_resume(MigrationState *s)
3355 {
3356     int ret;
3357
3358     /*
3359      * Call all the resume_prepare() hooks, so that modules can be
3360      * ready for the migration resume.
3361      */
3362     ret = qemu_savevm_state_resume_prepare(s);
3363     if (ret) {
3364         error_report("%s: resume_prepare() failure detected: %d",
3365                      __func__, ret);
3366         return ret;
3367     }
3368
3369     /*
3370      * Last handshake with destination on the resume (destination will
3371      * switch to postcopy-active afterwards)
3372      */
3373     ret = postcopy_resume_handshake(s);
3374     if (ret) {
3375         error_report("%s: handshake failed: %d", __func__, ret);
3376         return ret;
3377     }
3378
3379     return 0;
3380 }
3381
3382 /*
3383  * We don't return until we are in a safe state to continue current
3384  * postcopy migration.  Returns MIG_THR_ERR_RECOVERED if recovered, or
3385  * MIG_THR_ERR_FATAL if unrecovery failure happened.
3386  */
3387 static MigThrError postcopy_pause(MigrationState *s)
3388 {
3389     assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
3390
3391     while (true) {
3392         QEMUFile *file;
3393
3394         /*
3395          * Current channel is possibly broken. Release it.  Note that this is
3396          * guaranteed even without lock because to_dst_file should only be
3397          * modified by the migration thread.  That also guarantees that the
3398          * unregister of yank is safe too without the lock.  It should be safe
3399          * even to be within the qemu_file_lock, but we didn't do that to avoid
3400          * taking more mutex (yank_lock) within qemu_file_lock.  TL;DR: we make
3401          * the qemu_file_lock critical section as small as possible.
3402          */
3403         assert(s->to_dst_file);
3404         migration_ioc_unregister_yank_from_file(s->to_dst_file);
3405         qemu_mutex_lock(&s->qemu_file_lock);
3406         file = s->to_dst_file;
3407         s->to_dst_file = NULL;
3408         qemu_mutex_unlock(&s->qemu_file_lock);
3409
3410         qemu_file_shutdown(file);
3411         qemu_fclose(file);
3412
3413         migrate_set_state(&s->state, s->state,
3414                           MIGRATION_STATUS_POSTCOPY_PAUSED);
3415
3416         error_report("Detected IO failure for postcopy. "
3417                      "Migration paused.");
3418
3419         /*
3420          * We wait until things fixed up. Then someone will setup the
3421          * status back for us.
3422          */
3423         while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
3424             qemu_sem_wait(&s->postcopy_pause_sem);
3425         }
3426
3427         if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
3428             /* Woken up by a recover procedure. Give it a shot */
3429
3430             /*
3431              * Firstly, let's wake up the return path now, with a new
3432              * return path channel.
3433              */
3434             qemu_sem_post(&s->postcopy_pause_rp_sem);
3435
3436             /* Do the resume logic */
3437             if (postcopy_do_resume(s) == 0) {
3438                 /* Let's continue! */
3439                 trace_postcopy_pause_continued();
3440                 return MIG_THR_ERR_RECOVERED;
3441             } else {
3442                 /*
3443                  * Something wrong happened during the recovery, let's
3444                  * pause again. Pause is always better than throwing
3445                  * data away.
3446                  */
3447                 continue;
3448             }
3449         } else {
3450             /* This is not right... Time to quit. */
3451             return MIG_THR_ERR_FATAL;
3452         }
3453     }
3454 }
3455
3456 static MigThrError migration_detect_error(MigrationState *s)
3457 {
3458     int ret;
3459     int state = s->state;
3460     Error *local_error = NULL;
3461
3462     if (state == MIGRATION_STATUS_CANCELLING ||
3463         state == MIGRATION_STATUS_CANCELLED) {
3464         /* End the migration, but don't set the state to failed */
3465         return MIG_THR_ERR_FATAL;
3466     }
3467
3468     /* Try to detect any file errors */
3469     ret = qemu_file_get_error_obj(s->to_dst_file, &local_error);
3470     if (!ret) {
3471         /* Everything is fine */
3472         assert(!local_error);
3473         return MIG_THR_ERR_NONE;
3474     }
3475
3476     if (local_error) {
3477         migrate_set_error(s, local_error);
3478         error_free(local_error);
3479     }
3480
3481     if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret) {
3482         /*
3483          * For postcopy, we allow the network to be down for a
3484          * while. After that, it can be continued by a
3485          * recovery phase.
3486          */
3487         return postcopy_pause(s);
3488     } else {
3489         /*
3490          * For precopy (or postcopy with error outside IO), we fail
3491          * with no time.
3492          */
3493         migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED);
3494         trace_migration_thread_file_err();
3495
3496         /* Time to stop the migration, now. */
3497         return MIG_THR_ERR_FATAL;
3498     }
3499 }
3500
3501 /* How many bytes have we transferred since the beginning of the migration */
3502 static uint64_t migration_total_bytes(MigrationState *s)
3503 {
3504     return qemu_ftell(s->to_dst_file) + ram_counters.multifd_bytes;
3505 }
3506
3507 static void migration_calculate_complete(MigrationState *s)
3508 {
3509     uint64_t bytes = migration_total_bytes(s);
3510     int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3511     int64_t transfer_time;
3512
3513     s->total_time = end_time - s->start_time;
3514     if (!s->downtime) {
3515         /*
3516          * It's still not set, so we are precopy migration.  For
3517          * postcopy, downtime is calculated during postcopy_start().
3518          */
3519         s->downtime = end_time - s->downtime_start;
3520     }
3521
3522     transfer_time = s->total_time - s->setup_time;
3523     if (transfer_time) {
3524         s->mbps = ((double) bytes * 8.0) / transfer_time / 1000;
3525     }
3526 }
3527
3528 static void update_iteration_initial_status(MigrationState *s)
3529 {
3530     /*
3531      * Update these three fields at the same time to avoid mismatch info lead
3532      * wrong speed calculation.
3533      */
3534     s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3535     s->iteration_initial_bytes = migration_total_bytes(s);
3536     s->iteration_initial_pages = ram_get_total_transferred_pages();
3537 }
3538
3539 static void migration_update_counters(MigrationState *s,
3540                                       int64_t current_time)
3541 {
3542     uint64_t transferred, transferred_pages, time_spent;
3543     uint64_t current_bytes; /* bytes transferred since the beginning */
3544     double bandwidth;
3545
3546     if (current_time < s->iteration_start_time + BUFFER_DELAY) {
3547         return;
3548     }
3549
3550     current_bytes = migration_total_bytes(s);
3551     transferred = current_bytes - s->iteration_initial_bytes;
3552     time_spent = current_time - s->iteration_start_time;
3553     bandwidth = (double)transferred / time_spent;
3554     s->threshold_size = bandwidth * s->parameters.downtime_limit;
3555
3556     s->mbps = (((double) transferred * 8.0) /
3557                ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
3558
3559     transferred_pages = ram_get_total_transferred_pages() -
3560                             s->iteration_initial_pages;
3561     s->pages_per_second = (double) transferred_pages /
3562                              (((double) time_spent / 1000.0));
3563
3564     /*
3565      * if we haven't sent anything, we don't want to
3566      * recalculate. 10000 is a small enough number for our purposes
3567      */
3568     if (ram_counters.dirty_pages_rate && transferred > 10000) {
3569         s->expected_downtime = ram_counters.remaining / bandwidth;
3570     }
3571
3572     qemu_file_reset_rate_limit(s->to_dst_file);
3573
3574     update_iteration_initial_status(s);
3575
3576     trace_migrate_transferred(transferred, time_spent,
3577                               bandwidth, s->threshold_size);
3578 }
3579
3580 /* Migration thread iteration status */
3581 typedef enum {
3582     MIG_ITERATE_RESUME,         /* Resume current iteration */
3583     MIG_ITERATE_SKIP,           /* Skip current iteration */
3584     MIG_ITERATE_BREAK,          /* Break the loop */
3585 } MigIterateState;
3586
3587 /*
3588  * Return true if continue to the next iteration directly, false
3589  * otherwise.
3590  */
3591 static MigIterateState migration_iteration_run(MigrationState *s)
3592 {
3593     uint64_t pending_size, pend_pre, pend_compat, pend_post;
3594     bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
3595
3596     qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre,
3597                               &pend_compat, &pend_post);
3598     pending_size = pend_pre + pend_compat + pend_post;
3599
3600     trace_migrate_pending(pending_size, s->threshold_size,
3601                           pend_pre, pend_compat, pend_post);
3602
3603     if (pending_size && pending_size >= s->threshold_size) {
3604         /* Still a significant amount to transfer */
3605         if (!in_postcopy && pend_pre <= s->threshold_size &&
3606             qatomic_read(&s->start_postcopy)) {
3607             if (postcopy_start(s)) {
3608                 error_report("%s: postcopy failed to start", __func__);
3609             }
3610             return MIG_ITERATE_SKIP;
3611         }
3612         /* Just another iteration step */
3613         qemu_savevm_state_iterate(s->to_dst_file, in_postcopy);
3614     } else {
3615         trace_migration_thread_low_pending(pending_size);
3616         migration_completion(s);
3617         return MIG_ITERATE_BREAK;
3618     }
3619
3620     return MIG_ITERATE_RESUME;
3621 }
3622
3623 static void migration_iteration_finish(MigrationState *s)
3624 {
3625     /* If we enabled cpu throttling for auto-converge, turn it off. */
3626     cpu_throttle_stop();
3627
3628     qemu_mutex_lock_iothread();
3629     switch (s->state) {
3630     case MIGRATION_STATUS_COMPLETED:
3631         migration_calculate_complete(s);
3632         runstate_set(RUN_STATE_POSTMIGRATE);
3633         break;
3634     case MIGRATION_STATUS_COLO:
3635         if (!migrate_colo_enabled()) {
3636             error_report("%s: critical error: calling COLO code without "
3637                          "COLO enabled", __func__);
3638         }
3639         migrate_start_colo_process(s);
3640         s->vm_was_running = true;
3641         /* Fallthrough */
3642     case MIGRATION_STATUS_FAILED:
3643     case MIGRATION_STATUS_CANCELLED:
3644     case MIGRATION_STATUS_CANCELLING:
3645         if (s->vm_was_running) {
3646             if (!runstate_check(RUN_STATE_SHUTDOWN)) {
3647                 vm_start();
3648             }
3649         } else {
3650             if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
3651                 runstate_set(RUN_STATE_POSTMIGRATE);
3652             }
3653         }
3654         break;
3655
3656     default:
3657         /* Should not reach here, but if so, forgive the VM. */
3658         error_report("%s: Unknown ending state %d", __func__, s->state);
3659         break;
3660     }
3661     migrate_fd_cleanup_schedule(s);
3662     qemu_mutex_unlock_iothread();
3663 }
3664
3665 static void bg_migration_iteration_finish(MigrationState *s)
3666 {
3667     qemu_mutex_lock_iothread();
3668     switch (s->state) {
3669     case MIGRATION_STATUS_COMPLETED:
3670         migration_calculate_complete(s);
3671         break;
3672
3673     case MIGRATION_STATUS_ACTIVE:
3674     case MIGRATION_STATUS_FAILED:
3675     case MIGRATION_STATUS_CANCELLED:
3676     case MIGRATION_STATUS_CANCELLING:
3677         break;
3678
3679     default:
3680         /* Should not reach here, but if so, forgive the VM. */
3681         error_report("%s: Unknown ending state %d", __func__, s->state);
3682         break;
3683     }
3684
3685     migrate_fd_cleanup_schedule(s);
3686     qemu_mutex_unlock_iothread();
3687 }
3688
3689 /*
3690  * Return true if continue to the next iteration directly, false
3691  * otherwise.
3692  */
3693 static MigIterateState bg_migration_iteration_run(MigrationState *s)
3694 {
3695     int res;
3696
3697     res = qemu_savevm_state_iterate(s->to_dst_file, false);
3698     if (res > 0) {
3699         bg_migration_completion(s);
3700         return MIG_ITERATE_BREAK;
3701     }
3702
3703     return MIG_ITERATE_RESUME;
3704 }
3705
3706 void migration_make_urgent_request(void)
3707 {
3708     qemu_sem_post(&migrate_get_current()->rate_limit_sem);
3709 }
3710
3711 void migration_consume_urgent_request(void)
3712 {
3713     qemu_sem_wait(&migrate_get_current()->rate_limit_sem);
3714 }
3715
3716 /* Returns true if the rate limiting was broken by an urgent request */
3717 bool migration_rate_limit(void)
3718 {
3719     int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3720     MigrationState *s = migrate_get_current();
3721
3722     bool urgent = false;
3723     migration_update_counters(s, now);
3724     if (qemu_file_rate_limit(s->to_dst_file)) {
3725
3726         if (qemu_file_get_error(s->to_dst_file)) {
3727             return false;
3728         }
3729         /*
3730          * Wait for a delay to do rate limiting OR
3731          * something urgent to post the semaphore.
3732          */
3733         int ms = s->iteration_start_time + BUFFER_DELAY - now;
3734         trace_migration_rate_limit_pre(ms);
3735         if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) {
3736             /*
3737              * We were woken by one or more urgent things but
3738              * the timedwait will have consumed one of them.
3739              * The service routine for the urgent wake will dec
3740              * the semaphore itself for each item it consumes,
3741              * so add this one we just eat back.
3742              */
3743             qemu_sem_post(&s->rate_limit_sem);
3744             urgent = true;
3745         }
3746         trace_migration_rate_limit_post(urgent);
3747     }
3748     return urgent;
3749 }
3750
3751 /*
3752  * if failover devices are present, wait they are completely
3753  * unplugged
3754  */
3755
3756 static void qemu_savevm_wait_unplug(MigrationState *s, int old_state,
3757                                     int new_state)
3758 {
3759     if (qemu_savevm_state_guest_unplug_pending()) {
3760         migrate_set_state(&s->state, old_state, MIGRATION_STATUS_WAIT_UNPLUG);
3761
3762         while (s->state == MIGRATION_STATUS_WAIT_UNPLUG &&
3763                qemu_savevm_state_guest_unplug_pending()) {
3764             qemu_sem_timedwait(&s->wait_unplug_sem, 250);
3765         }
3766         if (s->state != MIGRATION_STATUS_WAIT_UNPLUG) {
3767             int timeout = 120; /* 30 seconds */
3768             /*
3769              * migration has been canceled
3770              * but as we have started an unplug we must wait the end
3771              * to be able to plug back the card
3772              */
3773             while (timeout-- && qemu_savevm_state_guest_unplug_pending()) {
3774                 qemu_sem_timedwait(&s->wait_unplug_sem, 250);
3775             }
3776             if (qemu_savevm_state_guest_unplug_pending() &&
3777                 !qtest_enabled()) {
3778                 warn_report("migration: partially unplugged device on "
3779                             "failure");
3780             }
3781         }
3782
3783         migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG, new_state);
3784     } else {
3785         migrate_set_state(&s->state, old_state, new_state);
3786     }
3787 }
3788
3789 /*
3790  * Master migration thread on the source VM.
3791  * It drives the migration and pumps the data down the outgoing channel.
3792  */
3793 static void *migration_thread(void *opaque)
3794 {
3795     MigrationState *s = opaque;
3796     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
3797     MigThrError thr_error;
3798     bool urgent = false;
3799
3800     rcu_register_thread();
3801
3802     object_ref(OBJECT(s));
3803     update_iteration_initial_status(s);
3804
3805     qemu_savevm_state_header(s->to_dst_file);
3806
3807     /*
3808      * If we opened the return path, we need to make sure dst has it
3809      * opened as well.
3810      */
3811     if (s->rp_state.rp_thread_created) {
3812         /* Now tell the dest that it should open its end so it can reply */
3813         qemu_savevm_send_open_return_path(s->to_dst_file);
3814
3815         /* And do a ping that will make stuff easier to debug */
3816         qemu_savevm_send_ping(s->to_dst_file, 1);
3817     }
3818
3819     if (migrate_postcopy()) {
3820         /*
3821          * Tell the destination that we *might* want to do postcopy later;
3822          * if the other end can't do postcopy it should fail now, nice and
3823          * early.
3824          */
3825         qemu_savevm_send_postcopy_advise(s->to_dst_file);
3826     }
3827
3828     if (migrate_colo_enabled()) {
3829         /* Notify migration destination that we enable COLO */
3830         qemu_savevm_send_colo_enable(s->to_dst_file);
3831     }
3832
3833     qemu_savevm_state_setup(s->to_dst_file);
3834
3835     qemu_savevm_wait_unplug(s, MIGRATION_STATUS_SETUP,
3836                                MIGRATION_STATUS_ACTIVE);
3837
3838     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
3839
3840     trace_migration_thread_setup_complete();
3841
3842     while (migration_is_active(s)) {
3843         if (urgent || !qemu_file_rate_limit(s->to_dst_file)) {
3844             MigIterateState iter_state = migration_iteration_run(s);
3845             if (iter_state == MIG_ITERATE_SKIP) {
3846                 continue;
3847             } else if (iter_state == MIG_ITERATE_BREAK) {
3848                 break;
3849             }
3850         }
3851
3852         /*
3853          * Try to detect any kind of failures, and see whether we
3854          * should stop the migration now.
3855          */
3856         thr_error = migration_detect_error(s);
3857         if (thr_error == MIG_THR_ERR_FATAL) {
3858             /* Stop migration */
3859             break;
3860         } else if (thr_error == MIG_THR_ERR_RECOVERED) {
3861             /*
3862              * Just recovered from a e.g. network failure, reset all
3863              * the local variables. This is important to avoid
3864              * breaking transferred_bytes and bandwidth calculation
3865              */
3866             update_iteration_initial_status(s);
3867         }
3868
3869         urgent = migration_rate_limit();
3870     }
3871
3872     trace_migration_thread_after_loop();
3873     migration_iteration_finish(s);
3874     object_unref(OBJECT(s));
3875     rcu_unregister_thread();
3876     return NULL;
3877 }
3878
3879 static void bg_migration_vm_start_bh(void *opaque)
3880 {
3881     MigrationState *s = opaque;
3882
3883     qemu_bh_delete(s->vm_start_bh);
3884     s->vm_start_bh = NULL;
3885
3886     vm_start();
3887     s->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - s->downtime_start;
3888 }
3889
3890 /**
3891  * Background snapshot thread, based on live migration code.
3892  * This is an alternative implementation of live migration mechanism
3893  * introduced specifically to support background snapshots.
3894  *
3895  * It takes advantage of userfault_fd write protection mechanism introduced
3896  * in v5.7 kernel. Compared to existing dirty page logging migration much
3897  * lesser stream traffic is produced resulting in smaller snapshot images,
3898  * simply cause of no page duplicates can get into the stream.
3899  *
3900  * Another key point is that generated vmstate stream reflects machine state
3901  * 'frozen' at the beginning of snapshot creation compared to dirty page logging
3902  * mechanism, which effectively results in that saved snapshot is the state of VM
3903  * at the end of the process.
3904  */
3905 static void *bg_migration_thread(void *opaque)
3906 {
3907     MigrationState *s = opaque;
3908     int64_t setup_start;
3909     MigThrError thr_error;
3910     QEMUFile *fb;
3911     bool early_fail = true;
3912
3913     rcu_register_thread();
3914     object_ref(OBJECT(s));
3915
3916     qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
3917
3918     setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
3919     /*
3920      * We want to save vmstate for the moment when migration has been
3921      * initiated but also we want to save RAM content while VM is running.
3922      * The RAM content should appear first in the vmstate. So, we first
3923      * stash the non-RAM part of the vmstate to the temporary buffer,
3924      * then write RAM part of the vmstate to the migration stream
3925      * with vCPUs running and, finally, write stashed non-RAM part of
3926      * the vmstate from the buffer to the migration stream.
3927      */
3928     s->bioc = qio_channel_buffer_new(512 * 1024);
3929     qio_channel_set_name(QIO_CHANNEL(s->bioc), "vmstate-buffer");
3930     fb = qemu_fopen_channel_output(QIO_CHANNEL(s->bioc));
3931     object_unref(OBJECT(s->bioc));
3932
3933     update_iteration_initial_status(s);
3934
3935     /*
3936      * Prepare for tracking memory writes with UFFD-WP - populate
3937      * RAM pages before protecting.
3938      */
3939 #ifdef __linux__
3940     ram_write_tracking_prepare();
3941 #endif
3942
3943     qemu_savevm_state_header(s->to_dst_file);
3944     qemu_savevm_state_setup(s->to_dst_file);
3945
3946     qemu_savevm_wait_unplug(s, MIGRATION_STATUS_SETUP,
3947                                MIGRATION_STATUS_ACTIVE);
3948
3949     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
3950
3951     trace_migration_thread_setup_complete();
3952     s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
3953
3954     qemu_mutex_lock_iothread();
3955
3956     /*
3957      * If VM is currently in suspended state, then, to make a valid runstate
3958      * transition in vm_stop_force_state() we need to wakeup it up.
3959      */
3960     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
3961     s->vm_was_running = runstate_is_running();
3962
3963     if (global_state_store()) {
3964         goto fail;
3965     }
3966     /* Forcibly stop VM before saving state of vCPUs and devices */
3967     if (vm_stop_force_state(RUN_STATE_PAUSED)) {
3968         goto fail;
3969     }
3970     /*
3971      * Put vCPUs in sync with shadow context structures, then
3972      * save their state to channel-buffer along with devices.
3973      */
3974     cpu_synchronize_all_states();
3975     if (qemu_savevm_state_complete_precopy_non_iterable(fb, false, false)) {
3976         goto fail;
3977     }
3978     /*
3979      * Since we are going to get non-iterable state data directly
3980      * from s->bioc->data, explicit flush is needed here.
3981      */
3982     qemu_fflush(fb);
3983
3984     /* Now initialize UFFD context and start tracking RAM writes */
3985     if (ram_write_tracking_start()) {
3986         goto fail;
3987     }
3988     early_fail = false;
3989
3990     /*
3991      * Start VM from BH handler to avoid write-fault lock here.
3992      * UFFD-WP protection for the whole RAM is already enabled so
3993      * calling VM state change notifiers from vm_start() would initiate
3994      * writes to virtio VQs memory which is in write-protected region.
3995      */
3996     s->vm_start_bh = qemu_bh_new(bg_migration_vm_start_bh, s);
3997     qemu_bh_schedule(s->vm_start_bh);
3998
3999     qemu_mutex_unlock_iothread();
4000
4001     while (migration_is_active(s)) {
4002         MigIterateState iter_state = bg_migration_iteration_run(s);
4003         if (iter_state == MIG_ITERATE_SKIP) {
4004             continue;
4005         } else if (iter_state == MIG_ITERATE_BREAK) {
4006             break;
4007         }
4008
4009         /*
4010          * Try to detect any kind of failures, and see whether we
4011          * should stop the migration now.
4012          */
4013         thr_error = migration_detect_error(s);
4014         if (thr_error == MIG_THR_ERR_FATAL) {
4015             /* Stop migration */
4016             break;
4017         }
4018
4019         migration_update_counters(s, qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
4020     }
4021
4022     trace_migration_thread_after_loop();
4023
4024 fail:
4025     if (early_fail) {
4026         migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE,
4027                 MIGRATION_STATUS_FAILED);
4028         qemu_mutex_unlock_iothread();
4029     }
4030
4031     bg_migration_iteration_finish(s);
4032
4033     qemu_fclose(fb);
4034     object_unref(OBJECT(s));
4035     rcu_unregister_thread();
4036
4037     return NULL;
4038 }
4039
4040 void migrate_fd_connect(MigrationState *s, Error *error_in)
4041 {
4042     Error *local_err = NULL;
4043     int64_t rate_limit;
4044     bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED;
4045
4046     /*
4047      * If there's a previous error, free it and prepare for another one.
4048      * Meanwhile if migration completes successfully, there won't have an error
4049      * dumped when calling migrate_fd_cleanup().
4050      */
4051     migrate_error_free(s);
4052
4053     s->expected_downtime = s->parameters.downtime_limit;
4054     if (resume) {
4055         assert(s->cleanup_bh);
4056     } else {
4057         assert(!s->cleanup_bh);
4058         s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup_bh, s);
4059     }
4060     if (error_in) {
4061         migrate_fd_error(s, error_in);
4062         if (resume) {
4063             /*
4064              * Don't do cleanup for resume if channel is invalid, but only dump
4065              * the error.  We wait for another channel connect from the user.
4066              * The error_report still gives HMP user a hint on what failed.
4067              * It's normally done in migrate_fd_cleanup(), but call it here
4068              * explicitly.
4069              */
4070             error_report_err(error_copy(s->error));
4071         } else {
4072             migrate_fd_cleanup(s);
4073         }
4074         return;
4075     }
4076
4077     if (resume) {
4078         /* This is a resumed migration */
4079         rate_limit = s->parameters.max_postcopy_bandwidth /
4080             XFER_LIMIT_RATIO;
4081     } else {
4082         /* This is a fresh new migration */
4083         rate_limit = s->parameters.max_bandwidth / XFER_LIMIT_RATIO;
4084
4085         /* Notify before starting migration thread */
4086         notifier_list_notify(&migration_state_notifiers, s);
4087     }
4088
4089     qemu_file_set_rate_limit(s->to_dst_file, rate_limit);
4090     qemu_file_set_blocking(s->to_dst_file, true);
4091
4092     /*
4093      * Open the return path. For postcopy, it is used exclusively. For
4094      * precopy, only if user specified "return-path" capability would
4095      * QEMU uses the return path.
4096      */
4097     if (migrate_postcopy_ram() || migrate_use_return_path()) {
4098         if (open_return_path_on_source(s, !resume)) {
4099             error_report("Unable to open return-path for postcopy");
4100             migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED);
4101             migrate_fd_cleanup(s);
4102             return;
4103         }
4104     }
4105
4106     if (resume) {
4107         /* Wakeup the main migration thread to do the recovery */
4108         migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
4109                           MIGRATION_STATUS_POSTCOPY_RECOVER);
4110         qemu_sem_post(&s->postcopy_pause_sem);
4111         return;
4112     }
4113
4114     if (multifd_save_setup(&local_err) != 0) {
4115         error_report_err(local_err);
4116         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
4117                           MIGRATION_STATUS_FAILED);
4118         migrate_fd_cleanup(s);
4119         return;
4120     }
4121
4122     if (migrate_background_snapshot()) {
4123         qemu_thread_create(&s->thread, "bg_snapshot",
4124                 bg_migration_thread, s, QEMU_THREAD_JOINABLE);
4125     } else {
4126         qemu_thread_create(&s->thread, "live_migration",
4127                 migration_thread, s, QEMU_THREAD_JOINABLE);
4128     }
4129     s->migration_thread_running = true;
4130 }
4131
4132 void migration_global_dump(Monitor *mon)
4133 {
4134     MigrationState *ms = migrate_get_current();
4135
4136     monitor_printf(mon, "globals:\n");
4137     monitor_printf(mon, "store-global-state: %s\n",
4138                    ms->store_global_state ? "on" : "off");
4139     monitor_printf(mon, "only-migratable: %s\n",
4140                    only_migratable ? "on" : "off");
4141     monitor_printf(mon, "send-configuration: %s\n",
4142                    ms->send_configuration ? "on" : "off");
4143     monitor_printf(mon, "send-section-footer: %s\n",
4144                    ms->send_section_footer ? "on" : "off");
4145     monitor_printf(mon, "decompress-error-check: %s\n",
4146                    ms->decompress_error_check ? "on" : "off");
4147     monitor_printf(mon, "clear-bitmap-shift: %u\n",
4148                    ms->clear_bitmap_shift);
4149 }
4150
4151 #define DEFINE_PROP_MIG_CAP(name, x)             \
4152     DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
4153
4154 static Property migration_properties[] = {
4155     DEFINE_PROP_BOOL("store-global-state", MigrationState,
4156                      store_global_state, true),
4157     DEFINE_PROP_BOOL("send-configuration", MigrationState,
4158                      send_configuration, true),
4159     DEFINE_PROP_BOOL("send-section-footer", MigrationState,
4160                      send_section_footer, true),
4161     DEFINE_PROP_BOOL("decompress-error-check", MigrationState,
4162                       decompress_error_check, true),
4163     DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState,
4164                       clear_bitmap_shift, CLEAR_BITMAP_SHIFT_DEFAULT),
4165
4166     /* Migration parameters */
4167     DEFINE_PROP_UINT8("x-compress-level", MigrationState,
4168                       parameters.compress_level,
4169                       DEFAULT_MIGRATE_COMPRESS_LEVEL),
4170     DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
4171                       parameters.compress_threads,
4172                       DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
4173     DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState,
4174                       parameters.compress_wait_thread, true),
4175     DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
4176                       parameters.decompress_threads,
4177                       DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
4178     DEFINE_PROP_UINT8("x-throttle-trigger-threshold", MigrationState,
4179                       parameters.throttle_trigger_threshold,
4180                       DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD),
4181     DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
4182                       parameters.cpu_throttle_initial,
4183                       DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
4184     DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
4185                       parameters.cpu_throttle_increment,
4186                       DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
4187     DEFINE_PROP_BOOL("x-cpu-throttle-tailslow", MigrationState,
4188                       parameters.cpu_throttle_tailslow, false),
4189     DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
4190                       parameters.max_bandwidth, MAX_THROTTLE),
4191     DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
4192                       parameters.downtime_limit,
4193                       DEFAULT_MIGRATE_SET_DOWNTIME),
4194     DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
4195                       parameters.x_checkpoint_delay,
4196                       DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
4197     DEFINE_PROP_UINT8("multifd-channels", MigrationState,
4198                       parameters.multifd_channels,
4199                       DEFAULT_MIGRATE_MULTIFD_CHANNELS),
4200     DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState,
4201                       parameters.multifd_compression,
4202                       DEFAULT_MIGRATE_MULTIFD_COMPRESSION),
4203     DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState,
4204                       parameters.multifd_zlib_level,
4205                       DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL),
4206     DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState,
4207                       parameters.multifd_zstd_level,
4208                       DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL),
4209     DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
4210                       parameters.xbzrle_cache_size,
4211                       DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
4212     DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState,
4213                       parameters.max_postcopy_bandwidth,
4214                       DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH),
4215     DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState,
4216                       parameters.max_cpu_throttle,
4217                       DEFAULT_MIGRATE_MAX_CPU_THROTTLE),
4218     DEFINE_PROP_SIZE("announce-initial", MigrationState,
4219                       parameters.announce_initial,
4220                       DEFAULT_MIGRATE_ANNOUNCE_INITIAL),
4221     DEFINE_PROP_SIZE("announce-max", MigrationState,
4222                       parameters.announce_max,
4223                       DEFAULT_MIGRATE_ANNOUNCE_MAX),
4224     DEFINE_PROP_SIZE("announce-rounds", MigrationState,
4225                       parameters.announce_rounds,
4226                       DEFAULT_MIGRATE_ANNOUNCE_ROUNDS),
4227     DEFINE_PROP_SIZE("announce-step", MigrationState,
4228                       parameters.announce_step,
4229                       DEFAULT_MIGRATE_ANNOUNCE_STEP),
4230
4231     /* Migration capabilities */
4232     DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
4233     DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
4234     DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
4235     DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
4236     DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
4237     DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
4238     DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
4239     DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
4240     DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
4241     DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
4242     DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
4243     DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD),
4244     DEFINE_PROP_MIG_CAP("x-background-snapshot",
4245             MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT),
4246
4247     DEFINE_PROP_END_OF_LIST(),
4248 };
4249
4250 static void migration_class_init(ObjectClass *klass, void *data)
4251 {
4252     DeviceClass *dc = DEVICE_CLASS(klass);
4253
4254     dc->user_creatable = false;
4255     device_class_set_props(dc, migration_properties);
4256 }
4257
4258 static void migration_instance_finalize(Object *obj)
4259 {
4260     MigrationState *ms = MIGRATION_OBJ(obj);
4261     MigrationParameters *params = &ms->parameters;
4262
4263     qemu_mutex_destroy(&ms->error_mutex);
4264     qemu_mutex_destroy(&ms->qemu_file_lock);
4265     g_free(params->tls_hostname);
4266     g_free(params->tls_creds);
4267     qemu_sem_destroy(&ms->wait_unplug_sem);
4268     qemu_sem_destroy(&ms->rate_limit_sem);
4269     qemu_sem_destroy(&ms->pause_sem);
4270     qemu_sem_destroy(&ms->postcopy_pause_sem);
4271     qemu_sem_destroy(&ms->postcopy_pause_rp_sem);
4272     qemu_sem_destroy(&ms->rp_state.rp_sem);
4273     error_free(ms->error);
4274 }
4275
4276 static void migration_instance_init(Object *obj)
4277 {
4278     MigrationState *ms = MIGRATION_OBJ(obj);
4279     MigrationParameters *params = &ms->parameters;
4280
4281     ms->state = MIGRATION_STATUS_NONE;
4282     ms->mbps = -1;
4283     ms->pages_per_second = -1;
4284     qemu_sem_init(&ms->pause_sem, 0);
4285     qemu_mutex_init(&ms->error_mutex);
4286
4287     params->tls_hostname = g_strdup("");
4288     params->tls_creds = g_strdup("");
4289
4290     /* Set has_* up only for parameter checks */
4291     params->has_compress_level = true;
4292     params->has_compress_threads = true;
4293     params->has_decompress_threads = true;
4294     params->has_throttle_trigger_threshold = true;
4295     params->has_cpu_throttle_initial = true;
4296     params->has_cpu_throttle_increment = true;
4297     params->has_cpu_throttle_tailslow = true;
4298     params->has_max_bandwidth = true;
4299     params->has_downtime_limit = true;
4300     params->has_x_checkpoint_delay = true;
4301     params->has_block_incremental = true;
4302     params->has_multifd_channels = true;
4303     params->has_multifd_compression = true;
4304     params->has_multifd_zlib_level = true;
4305     params->has_multifd_zstd_level = true;
4306     params->has_xbzrle_cache_size = true;
4307     params->has_max_postcopy_bandwidth = true;
4308     params->has_max_cpu_throttle = true;
4309     params->has_announce_initial = true;
4310     params->has_announce_max = true;
4311     params->has_announce_rounds = true;
4312     params->has_announce_step = true;
4313
4314     qemu_sem_init(&ms->postcopy_pause_sem, 0);
4315     qemu_sem_init(&ms->postcopy_pause_rp_sem, 0);
4316     qemu_sem_init(&ms->rp_state.rp_sem, 0);
4317     qemu_sem_init(&ms->rate_limit_sem, 0);
4318     qemu_sem_init(&ms->wait_unplug_sem, 0);
4319     qemu_mutex_init(&ms->qemu_file_lock);
4320 }
4321
4322 /*
4323  * Return true if check pass, false otherwise. Error will be put
4324  * inside errp if provided.
4325  */
4326 static bool migration_object_check(MigrationState *ms, Error **errp)
4327 {
4328     MigrationCapabilityStatusList *head = NULL;
4329     /* Assuming all off */
4330     bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
4331     int i;
4332
4333     if (!migrate_params_check(&ms->parameters, errp)) {
4334         return false;
4335     }
4336
4337     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
4338         if (ms->enabled_capabilities[i]) {
4339             QAPI_LIST_PREPEND(head, migrate_cap_add(i, true));
4340         }
4341     }
4342
4343     ret = migrate_caps_check(cap_list, head, errp);
4344
4345     /* It works with head == NULL */
4346     qapi_free_MigrationCapabilityStatusList(head);
4347
4348     return ret;
4349 }
4350
4351 static const TypeInfo migration_type = {
4352     .name = TYPE_MIGRATION,
4353     /*
4354      * NOTE: TYPE_MIGRATION is not really a device, as the object is
4355      * not created using qdev_new(), it is not attached to the qdev
4356      * device tree, and it is never realized.
4357      *
4358      * TODO: Make this TYPE_OBJECT once QOM provides something like
4359      * TYPE_DEVICE's "-global" properties.
4360      */
4361     .parent = TYPE_DEVICE,
4362     .class_init = migration_class_init,
4363     .class_size = sizeof(MigrationClass),
4364     .instance_size = sizeof(MigrationState),
4365     .instance_init = migration_instance_init,
4366     .instance_finalize = migration_instance_finalize,
4367 };
4368
4369 static void register_migration_types(void)
4370 {
4371     type_register_static(&migration_type);
4372 }
4373
4374 type_init(register_migration_types);