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