migration/migration.c

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