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