migration/migration.c

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