qemu

FORK: QEMU emulator
git clone https://git.neptards.moe/neptards/qemu.git
Log | Files | Refs | Submodules | LICENSE

migration.c (144203B)


      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.h"
     34 #include "migration/vmstate.h"
     35 #include "block/block.h"
     36 #include "qapi/error.h"
     37 #include "qapi/clone-visitor.h"
     38 #include "qapi/qapi-visit-migration.h"
     39 #include "qapi/qapi-visit-sockets.h"
     40 #include "qapi/qapi-commands-migration.h"
     41 #include "qapi/qapi-events-migration.h"
     42 #include "qapi/qmp/qerror.h"
     43 #include "qapi/qmp/qnull.h"
     44 #include "qemu/rcu.h"
     45 #include "block.h"
     46 #include "postcopy-ram.h"
     47 #include "qemu/thread.h"
     48 #include "trace.h"
     49 #include "exec/target_page.h"
     50 #include "io/channel-buffer.h"
     51 #include "io/channel-tls.h"
     52 #include "migration/colo.h"
     53 #include "hw/boards.h"
     54 #include "hw/qdev-properties.h"
     55 #include "hw/qdev-properties-system.h"
     56 #include "monitor/monitor.h"
     57 #include "net/announce.h"
     58 #include "qemu/queue.h"
     59 #include "multifd.h"
     60 #include "qemu/yank.h"
     61 #include "sysemu/cpus.h"
     62 #include "yank_functions.h"
     63 #include "sysemu/qtest.h"
     64 
     65 #define MAX_THROTTLE  (128 << 20)      /* Migration transfer speed throttling */
     66 
     67 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
     68  * data. */
     69 #define BUFFER_DELAY     100
     70 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
     71 
     72 /* Time in milliseconds we are allowed to stop the source,
     73  * for sending the last part */
     74 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
     75 
     76 /* Maximum migrate downtime set to 2000 seconds */
     77 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
     78 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
     79 
     80 /* Default compression thread count */
     81 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
     82 /* Default decompression thread count, usually decompression is at
     83  * least 4 times as fast as compression.*/
     84 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
     85 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
     86 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
     87 /* Define default autoconverge cpu throttle migration parameters */
     88 #define DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD 50
     89 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
     90 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
     91 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
     92 
     93 /* Migration XBZRLE default cache size */
     94 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
     95 
     96 /* The delay time (in ms) between two COLO checkpoints */
     97 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
     98 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
     99 #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE
    100 /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */
    101 #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
    102 /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
    103 #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
    104 
    105 /* Background transfer rate for postcopy, 0 means unlimited, note
    106  * that page requests can still exceed this limit.
    107  */
    108 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
    109 
    110 /*
    111  * Parameters for self_announce_delay giving a stream of RARP/ARP
    112  * packets after migration.
    113  */
    114 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL  50
    115 #define DEFAULT_MIGRATE_ANNOUNCE_MAX     550
    116 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS    5
    117 #define DEFAULT_MIGRATE_ANNOUNCE_STEP    100
    118 
    119 static NotifierList migration_state_notifiers =
    120     NOTIFIER_LIST_INITIALIZER(migration_state_notifiers);
    121 
    122 /* Messages sent on the return path from destination to source */
    123 enum mig_rp_message_type {
    124     MIG_RP_MSG_INVALID = 0,  /* Must be 0 */
    125     MIG_RP_MSG_SHUT,         /* sibling will not send any more RP messages */
    126     MIG_RP_MSG_PONG,         /* Response to a PING; data (seq: be32 ) */
    127 
    128     MIG_RP_MSG_REQ_PAGES_ID, /* data (start: be64, len: be32, id: string) */
    129     MIG_RP_MSG_REQ_PAGES,    /* data (start: be64, len: be32) */
    130     MIG_RP_MSG_RECV_BITMAP,  /* send recved_bitmap back to source */
    131     MIG_RP_MSG_RESUME_ACK,   /* tell source that we are ready to resume */
    132 
    133     MIG_RP_MSG_MAX
    134 };
    135 
    136 /* Migration capabilities set */
    137 struct MigrateCapsSet {
    138     int size;                       /* Capability set size */
    139     MigrationCapability caps[];     /* Variadic array of capabilities */
    140 };
    141 typedef struct MigrateCapsSet MigrateCapsSet;
    142 
    143 /* Define and initialize MigrateCapsSet */
    144 #define INITIALIZE_MIGRATE_CAPS_SET(_name, ...)   \
    145     MigrateCapsSet _name = {    \
    146         .size = sizeof((int []) { __VA_ARGS__ }) / sizeof(int), \
    147         .caps = { __VA_ARGS__ } \
    148     }
    149 
    150 /* Background-snapshot compatibility check list */
    151 static const
    152 INITIALIZE_MIGRATE_CAPS_SET(check_caps_background_snapshot,
    153     MIGRATION_CAPABILITY_POSTCOPY_RAM,
    154     MIGRATION_CAPABILITY_DIRTY_BITMAPS,
    155     MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME,
    156     MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE,
    157     MIGRATION_CAPABILITY_RETURN_PATH,
    158     MIGRATION_CAPABILITY_MULTIFD,
    159     MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER,
    160     MIGRATION_CAPABILITY_AUTO_CONVERGE,
    161     MIGRATION_CAPABILITY_RELEASE_RAM,
    162     MIGRATION_CAPABILITY_RDMA_PIN_ALL,
    163     MIGRATION_CAPABILITY_COMPRESS,
    164     MIGRATION_CAPABILITY_XBZRLE,
    165     MIGRATION_CAPABILITY_X_COLO,
    166     MIGRATION_CAPABILITY_VALIDATE_UUID,
    167     MIGRATION_CAPABILITY_ZERO_COPY_SEND);
    168 
    169 /* When we add fault tolerance, we could have several
    170    migrations at once.  For now we don't need to add
    171    dynamic creation of migration */
    172 
    173 static MigrationState *current_migration;
    174 static MigrationIncomingState *current_incoming;
    175 
    176 static GSList *migration_blockers;
    177 
    178 static bool migration_object_check(MigrationState *ms, Error **errp);
    179 static int migration_maybe_pause(MigrationState *s,
    180                                  int *current_active_state,
    181                                  int new_state);
    182 static void migrate_fd_cancel(MigrationState *s);
    183 
    184 static bool migrate_allow_multi_channels = true;
    185 
    186 void migrate_protocol_allow_multi_channels(bool allow)
    187 {
    188     migrate_allow_multi_channels = allow;
    189 }
    190 
    191 bool migrate_multi_channels_is_allowed(void)
    192 {
    193     return migrate_allow_multi_channels;
    194 }
    195 
    196 static gint page_request_addr_cmp(gconstpointer ap, gconstpointer bp)
    197 {
    198     uintptr_t a = (uintptr_t) ap, b = (uintptr_t) bp;
    199 
    200     return (a > b) - (a < b);
    201 }
    202 
    203 void migration_object_init(void)
    204 {
    205     /* This can only be called once. */
    206     assert(!current_migration);
    207     current_migration = MIGRATION_OBJ(object_new(TYPE_MIGRATION));
    208 
    209     /*
    210      * Init the migrate incoming object as well no matter whether
    211      * we'll use it or not.
    212      */
    213     assert(!current_incoming);
    214     current_incoming = g_new0(MigrationIncomingState, 1);
    215     current_incoming->state = MIGRATION_STATUS_NONE;
    216     current_incoming->postcopy_remote_fds =
    217         g_array_new(FALSE, TRUE, sizeof(struct PostCopyFD));
    218     qemu_mutex_init(&current_incoming->rp_mutex);
    219     qemu_mutex_init(&current_incoming->postcopy_prio_thread_mutex);
    220     qemu_event_init(&current_incoming->main_thread_load_event, false);
    221     qemu_sem_init(&current_incoming->postcopy_pause_sem_dst, 0);
    222     qemu_sem_init(&current_incoming->postcopy_pause_sem_fault, 0);
    223     qemu_sem_init(&current_incoming->postcopy_pause_sem_fast_load, 0);
    224     qemu_mutex_init(&current_incoming->page_request_mutex);
    225     current_incoming->page_requested = g_tree_new(page_request_addr_cmp);
    226 
    227     migration_object_check(current_migration, &error_fatal);
    228 
    229     blk_mig_init();
    230     ram_mig_init();
    231     dirty_bitmap_mig_init();
    232 }
    233 
    234 void migration_cancel(const Error *error)
    235 {
    236     if (error) {
    237         migrate_set_error(current_migration, error);
    238     }
    239     migrate_fd_cancel(current_migration);
    240 }
    241 
    242 void migration_shutdown(void)
    243 {
    244     /*
    245      * When the QEMU main thread exit, the COLO thread
    246      * may wait a semaphore. So, we should wakeup the
    247      * COLO thread before migration shutdown.
    248      */
    249     colo_shutdown();
    250     /*
    251      * Cancel the current migration - that will (eventually)
    252      * stop the migration using this structure
    253      */
    254     migration_cancel(NULL);
    255     object_unref(OBJECT(current_migration));
    256 
    257     /*
    258      * Cancel outgoing migration of dirty bitmaps. It should
    259      * at least unref used block nodes.
    260      */
    261     dirty_bitmap_mig_cancel_outgoing();
    262 
    263     /*
    264      * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
    265      * are non-critical data, and their loss never considered as
    266      * something serious.
    267      */
    268     dirty_bitmap_mig_cancel_incoming();
    269 }
    270 
    271 /* For outgoing */
    272 MigrationState *migrate_get_current(void)
    273 {
    274     /* This can only be called after the object created. */
    275     assert(current_migration);
    276     return current_migration;
    277 }
    278 
    279 MigrationIncomingState *migration_incoming_get_current(void)
    280 {
    281     assert(current_incoming);
    282     return current_incoming;
    283 }
    284 
    285 void migration_incoming_transport_cleanup(MigrationIncomingState *mis)
    286 {
    287     if (mis->socket_address_list) {
    288         qapi_free_SocketAddressList(mis->socket_address_list);
    289         mis->socket_address_list = NULL;
    290     }
    291 
    292     if (mis->transport_cleanup) {
    293         mis->transport_cleanup(mis->transport_data);
    294         mis->transport_data = mis->transport_cleanup = NULL;
    295     }
    296 }
    297 
    298 void migration_incoming_state_destroy(void)
    299 {
    300     struct MigrationIncomingState *mis = migration_incoming_get_current();
    301 
    302     if (mis->to_src_file) {
    303         /* Tell source that we are done */
    304         migrate_send_rp_shut(mis, qemu_file_get_error(mis->from_src_file) != 0);
    305         qemu_fclose(mis->to_src_file);
    306         mis->to_src_file = NULL;
    307     }
    308 
    309     if (mis->from_src_file) {
    310         migration_ioc_unregister_yank_from_file(mis->from_src_file);
    311         qemu_fclose(mis->from_src_file);
    312         mis->from_src_file = NULL;
    313     }
    314     if (mis->postcopy_remote_fds) {
    315         g_array_free(mis->postcopy_remote_fds, TRUE);
    316         mis->postcopy_remote_fds = NULL;
    317     }
    318 
    319     migration_incoming_transport_cleanup(mis);
    320     qemu_event_reset(&mis->main_thread_load_event);
    321 
    322     if (mis->page_requested) {
    323         g_tree_destroy(mis->page_requested);
    324         mis->page_requested = NULL;
    325     }
    326 
    327     if (mis->postcopy_qemufile_dst) {
    328         migration_ioc_unregister_yank_from_file(mis->postcopy_qemufile_dst);
    329         qemu_fclose(mis->postcopy_qemufile_dst);
    330         mis->postcopy_qemufile_dst = NULL;
    331     }
    332 
    333     yank_unregister_instance(MIGRATION_YANK_INSTANCE);
    334 }
    335 
    336 static void migrate_generate_event(int new_state)
    337 {
    338     if (migrate_use_events()) {
    339         qapi_event_send_migration(new_state);
    340     }
    341 }
    342 
    343 static bool migrate_late_block_activate(void)
    344 {
    345     MigrationState *s;
    346 
    347     s = migrate_get_current();
    348 
    349     return s->enabled_capabilities[
    350         MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE];
    351 }
    352 
    353 /*
    354  * Send a message on the return channel back to the source
    355  * of the migration.
    356  */
    357 static int migrate_send_rp_message(MigrationIncomingState *mis,
    358                                    enum mig_rp_message_type message_type,
    359                                    uint16_t len, void *data)
    360 {
    361     int ret = 0;
    362 
    363     trace_migrate_send_rp_message((int)message_type, len);
    364     QEMU_LOCK_GUARD(&mis->rp_mutex);
    365 
    366     /*
    367      * It's possible that the file handle got lost due to network
    368      * failures.
    369      */
    370     if (!mis->to_src_file) {
    371         ret = -EIO;
    372         return ret;
    373     }
    374 
    375     qemu_put_be16(mis->to_src_file, (unsigned int)message_type);
    376     qemu_put_be16(mis->to_src_file, len);
    377     qemu_put_buffer(mis->to_src_file, data, len);
    378     qemu_fflush(mis->to_src_file);
    379 
    380     /* It's possible that qemu file got error during sending */
    381     ret = qemu_file_get_error(mis->to_src_file);
    382 
    383     return ret;
    384 }
    385 
    386 /* Request one page from the source VM at the given start address.
    387  *   rb: the RAMBlock to request the page in
    388  *   Start: Address offset within the RB
    389  *   Len: Length in bytes required - must be a multiple of pagesize
    390  */
    391 int migrate_send_rp_message_req_pages(MigrationIncomingState *mis,
    392                                       RAMBlock *rb, ram_addr_t start)
    393 {
    394     uint8_t bufc[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
    395     size_t msglen = 12; /* start + len */
    396     size_t len = qemu_ram_pagesize(rb);
    397     enum mig_rp_message_type msg_type;
    398     const char *rbname;
    399     int rbname_len;
    400 
    401     *(uint64_t *)bufc = cpu_to_be64((uint64_t)start);
    402     *(uint32_t *)(bufc + 8) = cpu_to_be32((uint32_t)len);
    403 
    404     /*
    405      * We maintain the last ramblock that we requested for page.  Note that we
    406      * don't need locking because this function will only be called within the
    407      * postcopy ram fault thread.
    408      */
    409     if (rb != mis->last_rb) {
    410         mis->last_rb = rb;
    411 
    412         rbname = qemu_ram_get_idstr(rb);
    413         rbname_len = strlen(rbname);
    414 
    415         assert(rbname_len < 256);
    416 
    417         bufc[msglen++] = rbname_len;
    418         memcpy(bufc + msglen, rbname, rbname_len);
    419         msglen += rbname_len;
    420         msg_type = MIG_RP_MSG_REQ_PAGES_ID;
    421     } else {
    422         msg_type = MIG_RP_MSG_REQ_PAGES;
    423     }
    424 
    425     return migrate_send_rp_message(mis, msg_type, msglen, bufc);
    426 }
    427 
    428 int migrate_send_rp_req_pages(MigrationIncomingState *mis,
    429                               RAMBlock *rb, ram_addr_t start, uint64_t haddr)
    430 {
    431     void *aligned = (void *)(uintptr_t)ROUND_DOWN(haddr, qemu_ram_pagesize(rb));
    432     bool received = false;
    433 
    434     WITH_QEMU_LOCK_GUARD(&mis->page_request_mutex) {
    435         received = ramblock_recv_bitmap_test_byte_offset(rb, start);
    436         if (!received && !g_tree_lookup(mis->page_requested, aligned)) {
    437             /*
    438              * The page has not been received, and it's not yet in the page
    439              * request list.  Queue it.  Set the value of element to 1, so that
    440              * things like g_tree_lookup() will return TRUE (1) when found.
    441              */
    442             g_tree_insert(mis->page_requested, aligned, (gpointer)1);
    443             mis->page_requested_count++;
    444             trace_postcopy_page_req_add(aligned, mis->page_requested_count);
    445         }
    446     }
    447 
    448     /*
    449      * If the page is there, skip sending the message.  We don't even need the
    450      * lock because as long as the page arrived, it'll be there forever.
    451      */
    452     if (received) {
    453         return 0;
    454     }
    455 
    456     return migrate_send_rp_message_req_pages(mis, rb, start);
    457 }
    458 
    459 static bool migration_colo_enabled;
    460 bool migration_incoming_colo_enabled(void)
    461 {
    462     return migration_colo_enabled;
    463 }
    464 
    465 void migration_incoming_disable_colo(void)
    466 {
    467     ram_block_discard_disable(false);
    468     migration_colo_enabled = false;
    469 }
    470 
    471 int migration_incoming_enable_colo(void)
    472 {
    473     if (ram_block_discard_disable(true)) {
    474         error_report("COLO: cannot disable RAM discard");
    475         return -EBUSY;
    476     }
    477     migration_colo_enabled = true;
    478     return 0;
    479 }
    480 
    481 void migrate_add_address(SocketAddress *address)
    482 {
    483     MigrationIncomingState *mis = migration_incoming_get_current();
    484 
    485     QAPI_LIST_PREPEND(mis->socket_address_list,
    486                       QAPI_CLONE(SocketAddress, address));
    487 }
    488 
    489 static void qemu_start_incoming_migration(const char *uri, Error **errp)
    490 {
    491     const char *p = NULL;
    492 
    493     migrate_protocol_allow_multi_channels(false); /* reset it anyway */
    494     qapi_event_send_migration(MIGRATION_STATUS_SETUP);
    495     if (strstart(uri, "tcp:", &p) ||
    496         strstart(uri, "unix:", NULL) ||
    497         strstart(uri, "vsock:", NULL)) {
    498         migrate_protocol_allow_multi_channels(true);
    499         socket_start_incoming_migration(p ? p : uri, errp);
    500 #ifdef CONFIG_RDMA
    501     } else if (strstart(uri, "rdma:", &p)) {
    502         rdma_start_incoming_migration(p, errp);
    503 #endif
    504     } else if (strstart(uri, "exec:", &p)) {
    505         exec_start_incoming_migration(p, errp);
    506     } else if (strstart(uri, "fd:", &p)) {
    507         fd_start_incoming_migration(p, errp);
    508     } else {
    509         error_setg(errp, "unknown migration protocol: %s", uri);
    510     }
    511 }
    512 
    513 static void process_incoming_migration_bh(void *opaque)
    514 {
    515     Error *local_err = NULL;
    516     MigrationIncomingState *mis = opaque;
    517 
    518     /* If capability late_block_activate is set:
    519      * Only fire up the block code now if we're going to restart the
    520      * VM, else 'cont' will do it.
    521      * This causes file locking to happen; so we don't want it to happen
    522      * unless we really are starting the VM.
    523      */
    524     if (!migrate_late_block_activate() ||
    525          (autostart && (!global_state_received() ||
    526             global_state_get_runstate() == RUN_STATE_RUNNING))) {
    527         /* Make sure all file formats throw away their mutable metadata.
    528          * If we get an error here, just don't restart the VM yet. */
    529         bdrv_activate_all(&local_err);
    530         if (local_err) {
    531             error_report_err(local_err);
    532             local_err = NULL;
    533             autostart = false;
    534         }
    535     }
    536 
    537     /*
    538      * This must happen after all error conditions are dealt with and
    539      * we're sure the VM is going to be running on this host.
    540      */
    541     qemu_announce_self(&mis->announce_timer, migrate_announce_params());
    542 
    543     if (multifd_load_cleanup(&local_err) != 0) {
    544         error_report_err(local_err);
    545         autostart = false;
    546     }
    547     /* If global state section was not received or we are in running
    548        state, we need to obey autostart. Any other state is set with
    549        runstate_set. */
    550 
    551     dirty_bitmap_mig_before_vm_start();
    552 
    553     if (!global_state_received() ||
    554         global_state_get_runstate() == RUN_STATE_RUNNING) {
    555         if (autostart) {
    556             vm_start();
    557         } else {
    558             runstate_set(RUN_STATE_PAUSED);
    559         }
    560     } else if (migration_incoming_colo_enabled()) {
    561         migration_incoming_disable_colo();
    562         vm_start();
    563     } else {
    564         runstate_set(global_state_get_runstate());
    565     }
    566     /*
    567      * This must happen after any state changes since as soon as an external
    568      * observer sees this event they might start to prod at the VM assuming
    569      * it's ready to use.
    570      */
    571     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
    572                       MIGRATION_STATUS_COMPLETED);
    573     qemu_bh_delete(mis->bh);
    574     migration_incoming_state_destroy();
    575 }
    576 
    577 static void coroutine_fn
    578 process_incoming_migration_co(void *opaque)
    579 {
    580     MigrationIncomingState *mis = migration_incoming_get_current();
    581     PostcopyState ps;
    582     int ret;
    583     Error *local_err = NULL;
    584 
    585     assert(mis->from_src_file);
    586     mis->migration_incoming_co = qemu_coroutine_self();
    587     mis->largest_page_size = qemu_ram_pagesize_largest();
    588     postcopy_state_set(POSTCOPY_INCOMING_NONE);
    589     migrate_set_state(&mis->state, MIGRATION_STATUS_NONE,
    590                       MIGRATION_STATUS_ACTIVE);
    591     ret = qemu_loadvm_state(mis->from_src_file);
    592 
    593     ps = postcopy_state_get();
    594     trace_process_incoming_migration_co_end(ret, ps);
    595     if (ps != POSTCOPY_INCOMING_NONE) {
    596         if (ps == POSTCOPY_INCOMING_ADVISE) {
    597             /*
    598              * Where a migration had postcopy enabled (and thus went to advise)
    599              * but managed to complete within the precopy period, we can use
    600              * the normal exit.
    601              */
    602             postcopy_ram_incoming_cleanup(mis);
    603         } else if (ret >= 0) {
    604             /*
    605              * Postcopy was started, cleanup should happen at the end of the
    606              * postcopy thread.
    607              */
    608             trace_process_incoming_migration_co_postcopy_end_main();
    609             return;
    610         }
    611         /* Else if something went wrong then just fall out of the normal exit */
    612     }
    613 
    614     /* we get COLO info, and know if we are in COLO mode */
    615     if (!ret && migration_incoming_colo_enabled()) {
    616         /* Make sure all file formats throw away their mutable metadata */
    617         bdrv_activate_all(&local_err);
    618         if (local_err) {
    619             error_report_err(local_err);
    620             goto fail;
    621         }
    622 
    623         qemu_thread_create(&mis->colo_incoming_thread, "COLO incoming",
    624              colo_process_incoming_thread, mis, QEMU_THREAD_JOINABLE);
    625         mis->have_colo_incoming_thread = true;
    626         qemu_coroutine_yield();
    627 
    628         qemu_mutex_unlock_iothread();
    629         /* Wait checkpoint incoming thread exit before free resource */
    630         qemu_thread_join(&mis->colo_incoming_thread);
    631         qemu_mutex_lock_iothread();
    632         /* We hold the global iothread lock, so it is safe here */
    633         colo_release_ram_cache();
    634     }
    635 
    636     if (ret < 0) {
    637         error_report("load of migration failed: %s", strerror(-ret));
    638         goto fail;
    639     }
    640     mis->bh = qemu_bh_new(process_incoming_migration_bh, mis);
    641     qemu_bh_schedule(mis->bh);
    642     mis->migration_incoming_co = NULL;
    643     return;
    644 fail:
    645     local_err = NULL;
    646     migrate_set_state(&mis->state, MIGRATION_STATUS_ACTIVE,
    647                       MIGRATION_STATUS_FAILED);
    648     qemu_fclose(mis->from_src_file);
    649     if (multifd_load_cleanup(&local_err) != 0) {
    650         error_report_err(local_err);
    651     }
    652     exit(EXIT_FAILURE);
    653 }
    654 
    655 /**
    656  * migration_incoming_setup: Setup incoming migration
    657  * @f: file for main migration channel
    658  * @errp: where to put errors
    659  *
    660  * Returns: %true on success, %false on error.
    661  */
    662 static bool migration_incoming_setup(QEMUFile *f, Error **errp)
    663 {
    664     MigrationIncomingState *mis = migration_incoming_get_current();
    665 
    666     if (multifd_load_setup(errp) != 0) {
    667         return false;
    668     }
    669 
    670     if (!mis->from_src_file) {
    671         mis->from_src_file = f;
    672     }
    673     qemu_file_set_blocking(f, false);
    674     return true;
    675 }
    676 
    677 void migration_incoming_process(void)
    678 {
    679     Coroutine *co = qemu_coroutine_create(process_incoming_migration_co, NULL);
    680     qemu_coroutine_enter(co);
    681 }
    682 
    683 /* Returns true if recovered from a paused migration, otherwise false */
    684 static bool postcopy_try_recover(void)
    685 {
    686     MigrationIncomingState *mis = migration_incoming_get_current();
    687 
    688     if (mis->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
    689         /* Resumed from a paused postcopy migration */
    690 
    691         /* This should be set already in migration_incoming_setup() */
    692         assert(mis->from_src_file);
    693         /* Postcopy has standalone thread to do vm load */
    694         qemu_file_set_blocking(mis->from_src_file, true);
    695 
    696         /* Re-configure the return path */
    697         mis->to_src_file = qemu_file_get_return_path(mis->from_src_file);
    698 
    699         migrate_set_state(&mis->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
    700                           MIGRATION_STATUS_POSTCOPY_RECOVER);
    701 
    702         /*
    703          * Here, we only wake up the main loading thread (while the
    704          * rest threads will still be waiting), so that we can receive
    705          * commands from source now, and answer it if needed. The
    706          * rest threads will be woken up afterwards until we are sure
    707          * that source is ready to reply to page requests.
    708          */
    709         qemu_sem_post(&mis->postcopy_pause_sem_dst);
    710         return true;
    711     }
    712 
    713     return false;
    714 }
    715 
    716 void migration_fd_process_incoming(QEMUFile *f, Error **errp)
    717 {
    718     if (!migration_incoming_setup(f, errp)) {
    719         return;
    720     }
    721     if (postcopy_try_recover()) {
    722         return;
    723     }
    724     migration_incoming_process();
    725 }
    726 
    727 static bool migration_needs_multiple_sockets(void)
    728 {
    729     return migrate_use_multifd() || migrate_postcopy_preempt();
    730 }
    731 
    732 void migration_ioc_process_incoming(QIOChannel *ioc, Error **errp)
    733 {
    734     MigrationIncomingState *mis = migration_incoming_get_current();
    735     Error *local_err = NULL;
    736     bool start_migration;
    737     QEMUFile *f;
    738 
    739     if (!mis->from_src_file) {
    740         /* The first connection (multifd may have multiple) */
    741         f = qemu_file_new_input(ioc);
    742 
    743         if (!migration_incoming_setup(f, errp)) {
    744             return;
    745         }
    746 
    747         /*
    748          * Common migration only needs one channel, so we can start
    749          * right now.  Some features need more than one channel, we wait.
    750          */
    751         start_migration = !migration_needs_multiple_sockets();
    752     } else {
    753         /* Multiple connections */
    754         assert(migration_needs_multiple_sockets());
    755         if (migrate_use_multifd()) {
    756             start_migration = multifd_recv_new_channel(ioc, &local_err);
    757         } else {
    758             assert(migrate_postcopy_preempt());
    759             f = qemu_file_new_input(ioc);
    760             start_migration = postcopy_preempt_new_channel(mis, f);
    761         }
    762         if (local_err) {
    763             error_propagate(errp, local_err);
    764             return;
    765         }
    766     }
    767 
    768     if (start_migration) {
    769         /* If it's a recovery, we're done */
    770         if (postcopy_try_recover()) {
    771             return;
    772         }
    773         migration_incoming_process();
    774     }
    775 }
    776 
    777 /**
    778  * @migration_has_all_channels: We have received all channels that we need
    779  *
    780  * Returns true when we have got connections to all the channels that
    781  * we need for migration.
    782  */
    783 bool migration_has_all_channels(void)
    784 {
    785     MigrationIncomingState *mis = migration_incoming_get_current();
    786 
    787     if (!mis->from_src_file) {
    788         return false;
    789     }
    790 
    791     if (migrate_use_multifd()) {
    792         return multifd_recv_all_channels_created();
    793     }
    794 
    795     if (migrate_postcopy_preempt()) {
    796         return mis->postcopy_qemufile_dst != NULL;
    797     }
    798 
    799     return true;
    800 }
    801 
    802 /*
    803  * Send a 'SHUT' message on the return channel with the given value
    804  * to indicate that we've finished with the RP.  Non-0 value indicates
    805  * error.
    806  */
    807 void migrate_send_rp_shut(MigrationIncomingState *mis,
    808                           uint32_t value)
    809 {
    810     uint32_t buf;
    811 
    812     buf = cpu_to_be32(value);
    813     migrate_send_rp_message(mis, MIG_RP_MSG_SHUT, sizeof(buf), &buf);
    814 }
    815 
    816 /*
    817  * Send a 'PONG' message on the return channel with the given value
    818  * (normally in response to a 'PING')
    819  */
    820 void migrate_send_rp_pong(MigrationIncomingState *mis,
    821                           uint32_t value)
    822 {
    823     uint32_t buf;
    824 
    825     buf = cpu_to_be32(value);
    826     migrate_send_rp_message(mis, MIG_RP_MSG_PONG, sizeof(buf), &buf);
    827 }
    828 
    829 void migrate_send_rp_recv_bitmap(MigrationIncomingState *mis,
    830                                  char *block_name)
    831 {
    832     char buf[512];
    833     int len;
    834     int64_t res;
    835 
    836     /*
    837      * First, we send the header part. It contains only the len of
    838      * idstr, and the idstr itself.
    839      */
    840     len = strlen(block_name);
    841     buf[0] = len;
    842     memcpy(buf + 1, block_name, len);
    843 
    844     if (mis->state != MIGRATION_STATUS_POSTCOPY_RECOVER) {
    845         error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
    846                      __func__);
    847         return;
    848     }
    849 
    850     migrate_send_rp_message(mis, MIG_RP_MSG_RECV_BITMAP, len + 1, buf);
    851 
    852     /*
    853      * Next, we dump the received bitmap to the stream.
    854      *
    855      * TODO: currently we are safe since we are the only one that is
    856      * using the to_src_file handle (fault thread is still paused),
    857      * and it's ok even not taking the mutex. However the best way is
    858      * to take the lock before sending the message header, and release
    859      * the lock after sending the bitmap.
    860      */
    861     qemu_mutex_lock(&mis->rp_mutex);
    862     res = ramblock_recv_bitmap_send(mis->to_src_file, block_name);
    863     qemu_mutex_unlock(&mis->rp_mutex);
    864 
    865     trace_migrate_send_rp_recv_bitmap(block_name, res);
    866 }
    867 
    868 void migrate_send_rp_resume_ack(MigrationIncomingState *mis, uint32_t value)
    869 {
    870     uint32_t buf;
    871 
    872     buf = cpu_to_be32(value);
    873     migrate_send_rp_message(mis, MIG_RP_MSG_RESUME_ACK, sizeof(buf), &buf);
    874 }
    875 
    876 MigrationCapabilityStatusList *qmp_query_migrate_capabilities(Error **errp)
    877 {
    878     MigrationCapabilityStatusList *head = NULL, **tail = &head;
    879     MigrationCapabilityStatus *caps;
    880     MigrationState *s = migrate_get_current();
    881     int i;
    882 
    883     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
    884 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
    885         if (i == MIGRATION_CAPABILITY_BLOCK) {
    886             continue;
    887         }
    888 #endif
    889         caps = g_malloc0(sizeof(*caps));
    890         caps->capability = i;
    891         caps->state = s->enabled_capabilities[i];
    892         QAPI_LIST_APPEND(tail, caps);
    893     }
    894 
    895     return head;
    896 }
    897 
    898 MigrationParameters *qmp_query_migrate_parameters(Error **errp)
    899 {
    900     MigrationParameters *params;
    901     MigrationState *s = migrate_get_current();
    902 
    903     /* TODO use QAPI_CLONE() instead of duplicating it inline */
    904     params = g_malloc0(sizeof(*params));
    905     params->has_compress_level = true;
    906     params->compress_level = s->parameters.compress_level;
    907     params->has_compress_threads = true;
    908     params->compress_threads = s->parameters.compress_threads;
    909     params->has_compress_wait_thread = true;
    910     params->compress_wait_thread = s->parameters.compress_wait_thread;
    911     params->has_decompress_threads = true;
    912     params->decompress_threads = s->parameters.decompress_threads;
    913     params->has_throttle_trigger_threshold = true;
    914     params->throttle_trigger_threshold = s->parameters.throttle_trigger_threshold;
    915     params->has_cpu_throttle_initial = true;
    916     params->cpu_throttle_initial = s->parameters.cpu_throttle_initial;
    917     params->has_cpu_throttle_increment = true;
    918     params->cpu_throttle_increment = s->parameters.cpu_throttle_increment;
    919     params->has_cpu_throttle_tailslow = true;
    920     params->cpu_throttle_tailslow = s->parameters.cpu_throttle_tailslow;
    921     params->has_tls_creds = true;
    922     params->tls_creds = g_strdup(s->parameters.tls_creds);
    923     params->has_tls_hostname = true;
    924     params->tls_hostname = g_strdup(s->parameters.tls_hostname);
    925     params->has_tls_authz = true;
    926     params->tls_authz = g_strdup(s->parameters.tls_authz ?
    927                                  s->parameters.tls_authz : "");
    928     params->has_max_bandwidth = true;
    929     params->max_bandwidth = s->parameters.max_bandwidth;
    930     params->has_downtime_limit = true;
    931     params->downtime_limit = s->parameters.downtime_limit;
    932     params->has_x_checkpoint_delay = true;
    933     params->x_checkpoint_delay = s->parameters.x_checkpoint_delay;
    934     params->has_block_incremental = true;
    935     params->block_incremental = s->parameters.block_incremental;
    936     params->has_multifd_channels = true;
    937     params->multifd_channels = s->parameters.multifd_channels;
    938     params->has_multifd_compression = true;
    939     params->multifd_compression = s->parameters.multifd_compression;
    940     params->has_multifd_zlib_level = true;
    941     params->multifd_zlib_level = s->parameters.multifd_zlib_level;
    942     params->has_multifd_zstd_level = true;
    943     params->multifd_zstd_level = s->parameters.multifd_zstd_level;
    944     params->has_xbzrle_cache_size = true;
    945     params->xbzrle_cache_size = s->parameters.xbzrle_cache_size;
    946     params->has_max_postcopy_bandwidth = true;
    947     params->max_postcopy_bandwidth = s->parameters.max_postcopy_bandwidth;
    948     params->has_max_cpu_throttle = true;
    949     params->max_cpu_throttle = s->parameters.max_cpu_throttle;
    950     params->has_announce_initial = true;
    951     params->announce_initial = s->parameters.announce_initial;
    952     params->has_announce_max = true;
    953     params->announce_max = s->parameters.announce_max;
    954     params->has_announce_rounds = true;
    955     params->announce_rounds = s->parameters.announce_rounds;
    956     params->has_announce_step = true;
    957     params->announce_step = s->parameters.announce_step;
    958 
    959     if (s->parameters.has_block_bitmap_mapping) {
    960         params->has_block_bitmap_mapping = true;
    961         params->block_bitmap_mapping =
    962             QAPI_CLONE(BitmapMigrationNodeAliasList,
    963                        s->parameters.block_bitmap_mapping);
    964     }
    965 
    966     return params;
    967 }
    968 
    969 AnnounceParameters *migrate_announce_params(void)
    970 {
    971     static AnnounceParameters ap;
    972 
    973     MigrationState *s = migrate_get_current();
    974 
    975     ap.initial = s->parameters.announce_initial;
    976     ap.max = s->parameters.announce_max;
    977     ap.rounds = s->parameters.announce_rounds;
    978     ap.step = s->parameters.announce_step;
    979 
    980     return &ap;
    981 }
    982 
    983 /*
    984  * Return true if we're already in the middle of a migration
    985  * (i.e. any of the active or setup states)
    986  */
    987 bool migration_is_setup_or_active(int state)
    988 {
    989     switch (state) {
    990     case MIGRATION_STATUS_ACTIVE:
    991     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
    992     case MIGRATION_STATUS_POSTCOPY_PAUSED:
    993     case MIGRATION_STATUS_POSTCOPY_RECOVER:
    994     case MIGRATION_STATUS_SETUP:
    995     case MIGRATION_STATUS_PRE_SWITCHOVER:
    996     case MIGRATION_STATUS_DEVICE:
    997     case MIGRATION_STATUS_WAIT_UNPLUG:
    998     case MIGRATION_STATUS_COLO:
    999         return true;
   1000 
   1001     default:
   1002         return false;
   1003 
   1004     }
   1005 }
   1006 
   1007 bool migration_is_running(int state)
   1008 {
   1009     switch (state) {
   1010     case MIGRATION_STATUS_ACTIVE:
   1011     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
   1012     case MIGRATION_STATUS_POSTCOPY_PAUSED:
   1013     case MIGRATION_STATUS_POSTCOPY_RECOVER:
   1014     case MIGRATION_STATUS_SETUP:
   1015     case MIGRATION_STATUS_PRE_SWITCHOVER:
   1016     case MIGRATION_STATUS_DEVICE:
   1017     case MIGRATION_STATUS_WAIT_UNPLUG:
   1018     case MIGRATION_STATUS_CANCELLING:
   1019         return true;
   1020 
   1021     default:
   1022         return false;
   1023 
   1024     }
   1025 }
   1026 
   1027 static void populate_time_info(MigrationInfo *info, MigrationState *s)
   1028 {
   1029     info->has_status = true;
   1030     info->has_setup_time = true;
   1031     info->setup_time = s->setup_time;
   1032     if (s->state == MIGRATION_STATUS_COMPLETED) {
   1033         info->has_total_time = true;
   1034         info->total_time = s->total_time;
   1035         info->has_downtime = true;
   1036         info->downtime = s->downtime;
   1037     } else {
   1038         info->has_total_time = true;
   1039         info->total_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) -
   1040                            s->start_time;
   1041         info->has_expected_downtime = true;
   1042         info->expected_downtime = s->expected_downtime;
   1043     }
   1044 }
   1045 
   1046 static void populate_ram_info(MigrationInfo *info, MigrationState *s)
   1047 {
   1048     size_t page_size = qemu_target_page_size();
   1049 
   1050     info->has_ram = true;
   1051     info->ram = g_malloc0(sizeof(*info->ram));
   1052     info->ram->transferred = ram_counters.transferred;
   1053     info->ram->total = ram_bytes_total();
   1054     info->ram->duplicate = ram_counters.duplicate;
   1055     /* legacy value.  It is not used anymore */
   1056     info->ram->skipped = 0;
   1057     info->ram->normal = ram_counters.normal;
   1058     info->ram->normal_bytes = ram_counters.normal * page_size;
   1059     info->ram->mbps = s->mbps;
   1060     info->ram->dirty_sync_count = ram_counters.dirty_sync_count;
   1061     info->ram->dirty_sync_missed_zero_copy =
   1062             ram_counters.dirty_sync_missed_zero_copy;
   1063     info->ram->postcopy_requests = ram_counters.postcopy_requests;
   1064     info->ram->page_size = page_size;
   1065     info->ram->multifd_bytes = ram_counters.multifd_bytes;
   1066     info->ram->pages_per_second = s->pages_per_second;
   1067     info->ram->precopy_bytes = ram_counters.precopy_bytes;
   1068     info->ram->downtime_bytes = ram_counters.downtime_bytes;
   1069     info->ram->postcopy_bytes = ram_counters.postcopy_bytes;
   1070 
   1071     if (migrate_use_xbzrle()) {
   1072         info->has_xbzrle_cache = true;
   1073         info->xbzrle_cache = g_malloc0(sizeof(*info->xbzrle_cache));
   1074         info->xbzrle_cache->cache_size = migrate_xbzrle_cache_size();
   1075         info->xbzrle_cache->bytes = xbzrle_counters.bytes;
   1076         info->xbzrle_cache->pages = xbzrle_counters.pages;
   1077         info->xbzrle_cache->cache_miss = xbzrle_counters.cache_miss;
   1078         info->xbzrle_cache->cache_miss_rate = xbzrle_counters.cache_miss_rate;
   1079         info->xbzrle_cache->encoding_rate = xbzrle_counters.encoding_rate;
   1080         info->xbzrle_cache->overflow = xbzrle_counters.overflow;
   1081     }
   1082 
   1083     if (migrate_use_compression()) {
   1084         info->has_compression = true;
   1085         info->compression = g_malloc0(sizeof(*info->compression));
   1086         info->compression->pages = compression_counters.pages;
   1087         info->compression->busy = compression_counters.busy;
   1088         info->compression->busy_rate = compression_counters.busy_rate;
   1089         info->compression->compressed_size =
   1090                                     compression_counters.compressed_size;
   1091         info->compression->compression_rate =
   1092                                     compression_counters.compression_rate;
   1093     }
   1094 
   1095     if (cpu_throttle_active()) {
   1096         info->has_cpu_throttle_percentage = true;
   1097         info->cpu_throttle_percentage = cpu_throttle_get_percentage();
   1098     }
   1099 
   1100     if (s->state != MIGRATION_STATUS_COMPLETED) {
   1101         info->ram->remaining = ram_bytes_remaining();
   1102         info->ram->dirty_pages_rate = ram_counters.dirty_pages_rate;
   1103     }
   1104 }
   1105 
   1106 static void populate_disk_info(MigrationInfo *info)
   1107 {
   1108     if (blk_mig_active()) {
   1109         info->has_disk = true;
   1110         info->disk = g_malloc0(sizeof(*info->disk));
   1111         info->disk->transferred = blk_mig_bytes_transferred();
   1112         info->disk->remaining = blk_mig_bytes_remaining();
   1113         info->disk->total = blk_mig_bytes_total();
   1114     }
   1115 }
   1116 
   1117 static void fill_source_migration_info(MigrationInfo *info)
   1118 {
   1119     MigrationState *s = migrate_get_current();
   1120     int state = qatomic_read(&s->state);
   1121     GSList *cur_blocker = migration_blockers;
   1122 
   1123     info->blocked_reasons = NULL;
   1124 
   1125     /*
   1126      * There are two types of reasons a migration might be blocked;
   1127      * a) devices marked in VMState as non-migratable, and
   1128      * b) Explicit migration blockers
   1129      * We need to add both of them here.
   1130      */
   1131     qemu_savevm_non_migratable_list(&info->blocked_reasons);
   1132 
   1133     while (cur_blocker) {
   1134         QAPI_LIST_PREPEND(info->blocked_reasons,
   1135                           g_strdup(error_get_pretty(cur_blocker->data)));
   1136         cur_blocker = g_slist_next(cur_blocker);
   1137     }
   1138     info->has_blocked_reasons = info->blocked_reasons != NULL;
   1139 
   1140     switch (state) {
   1141     case MIGRATION_STATUS_NONE:
   1142         /* no migration has happened ever */
   1143         /* do not overwrite destination migration status */
   1144         return;
   1145     case MIGRATION_STATUS_SETUP:
   1146         info->has_status = true;
   1147         info->has_total_time = false;
   1148         break;
   1149     case MIGRATION_STATUS_ACTIVE:
   1150     case MIGRATION_STATUS_CANCELLING:
   1151     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
   1152     case MIGRATION_STATUS_PRE_SWITCHOVER:
   1153     case MIGRATION_STATUS_DEVICE:
   1154     case MIGRATION_STATUS_POSTCOPY_PAUSED:
   1155     case MIGRATION_STATUS_POSTCOPY_RECOVER:
   1156         /* TODO add some postcopy stats */
   1157         populate_time_info(info, s);
   1158         populate_ram_info(info, s);
   1159         populate_disk_info(info);
   1160         populate_vfio_info(info);
   1161         break;
   1162     case MIGRATION_STATUS_COLO:
   1163         info->has_status = true;
   1164         /* TODO: display COLO specific information (checkpoint info etc.) */
   1165         break;
   1166     case MIGRATION_STATUS_COMPLETED:
   1167         populate_time_info(info, s);
   1168         populate_ram_info(info, s);
   1169         populate_vfio_info(info);
   1170         break;
   1171     case MIGRATION_STATUS_FAILED:
   1172         info->has_status = true;
   1173         if (s->error) {
   1174             info->has_error_desc = true;
   1175             info->error_desc = g_strdup(error_get_pretty(s->error));
   1176         }
   1177         break;
   1178     case MIGRATION_STATUS_CANCELLED:
   1179         info->has_status = true;
   1180         break;
   1181     case MIGRATION_STATUS_WAIT_UNPLUG:
   1182         info->has_status = true;
   1183         break;
   1184     }
   1185     info->status = state;
   1186 }
   1187 
   1188 typedef enum WriteTrackingSupport {
   1189     WT_SUPPORT_UNKNOWN = 0,
   1190     WT_SUPPORT_ABSENT,
   1191     WT_SUPPORT_AVAILABLE,
   1192     WT_SUPPORT_COMPATIBLE
   1193 } WriteTrackingSupport;
   1194 
   1195 static
   1196 WriteTrackingSupport migrate_query_write_tracking(void)
   1197 {
   1198     /* Check if kernel supports required UFFD features */
   1199     if (!ram_write_tracking_available()) {
   1200         return WT_SUPPORT_ABSENT;
   1201     }
   1202     /*
   1203      * Check if current memory configuration is
   1204      * compatible with required UFFD features.
   1205      */
   1206     if (!ram_write_tracking_compatible()) {
   1207         return WT_SUPPORT_AVAILABLE;
   1208     }
   1209 
   1210     return WT_SUPPORT_COMPATIBLE;
   1211 }
   1212 
   1213 /**
   1214  * @migration_caps_check - check capability validity
   1215  *
   1216  * @cap_list: old capability list, array of bool
   1217  * @params: new capabilities to be applied soon
   1218  * @errp: set *errp if the check failed, with reason
   1219  *
   1220  * Returns true if check passed, otherwise false.
   1221  */
   1222 static bool migrate_caps_check(bool *cap_list,
   1223                                MigrationCapabilityStatusList *params,
   1224                                Error **errp)
   1225 {
   1226     MigrationCapabilityStatusList *cap;
   1227     bool old_postcopy_cap;
   1228     MigrationIncomingState *mis = migration_incoming_get_current();
   1229 
   1230     old_postcopy_cap = cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM];
   1231 
   1232     for (cap = params; cap; cap = cap->next) {
   1233         cap_list[cap->value->capability] = cap->value->state;
   1234     }
   1235 
   1236 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
   1237     if (cap_list[MIGRATION_CAPABILITY_BLOCK]) {
   1238         error_setg(errp, "QEMU compiled without old-style (blk/-b, inc/-i) "
   1239                    "block migration");
   1240         error_append_hint(errp, "Use drive_mirror+NBD instead.\n");
   1241         return false;
   1242     }
   1243 #endif
   1244 
   1245 #ifndef CONFIG_REPLICATION
   1246     if (cap_list[MIGRATION_CAPABILITY_X_COLO]) {
   1247         error_setg(errp, "QEMU compiled without replication module"
   1248                    " can't enable COLO");
   1249         error_append_hint(errp, "Please enable replication before COLO.\n");
   1250         return false;
   1251     }
   1252 #endif
   1253 
   1254     if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
   1255         /* This check is reasonably expensive, so only when it's being
   1256          * set the first time, also it's only the destination that needs
   1257          * special support.
   1258          */
   1259         if (!old_postcopy_cap && runstate_check(RUN_STATE_INMIGRATE) &&
   1260             !postcopy_ram_supported_by_host(mis)) {
   1261             /* postcopy_ram_supported_by_host will have emitted a more
   1262              * detailed message
   1263              */
   1264             error_setg(errp, "Postcopy is not supported");
   1265             return false;
   1266         }
   1267 
   1268         if (cap_list[MIGRATION_CAPABILITY_X_IGNORE_SHARED]) {
   1269             error_setg(errp, "Postcopy is not compatible with ignore-shared");
   1270             return false;
   1271         }
   1272     }
   1273 
   1274     if (cap_list[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT]) {
   1275         WriteTrackingSupport wt_support;
   1276         int idx;
   1277         /*
   1278          * Check if 'background-snapshot' capability is supported by
   1279          * host kernel and compatible with guest memory configuration.
   1280          */
   1281         wt_support = migrate_query_write_tracking();
   1282         if (wt_support < WT_SUPPORT_AVAILABLE) {
   1283             error_setg(errp, "Background-snapshot is not supported by host kernel");
   1284             return false;
   1285         }
   1286         if (wt_support < WT_SUPPORT_COMPATIBLE) {
   1287             error_setg(errp, "Background-snapshot is not compatible "
   1288                     "with guest memory configuration");
   1289             return false;
   1290         }
   1291 
   1292         /*
   1293          * Check if there are any migration capabilities
   1294          * incompatible with 'background-snapshot'.
   1295          */
   1296         for (idx = 0; idx < check_caps_background_snapshot.size; idx++) {
   1297             int incomp_cap = check_caps_background_snapshot.caps[idx];
   1298             if (cap_list[incomp_cap]) {
   1299                 error_setg(errp,
   1300                         "Background-snapshot is not compatible with %s",
   1301                         MigrationCapability_str(incomp_cap));
   1302                 return false;
   1303             }
   1304         }
   1305     }
   1306 
   1307 #ifdef CONFIG_LINUX
   1308     if (cap_list[MIGRATION_CAPABILITY_ZERO_COPY_SEND] &&
   1309         (!cap_list[MIGRATION_CAPABILITY_MULTIFD] ||
   1310          cap_list[MIGRATION_CAPABILITY_COMPRESS] ||
   1311          cap_list[MIGRATION_CAPABILITY_XBZRLE] ||
   1312          migrate_multifd_compression() ||
   1313          migrate_use_tls())) {
   1314         error_setg(errp,
   1315                    "Zero copy only available for non-compressed non-TLS multifd migration");
   1316         return false;
   1317     }
   1318 #else
   1319     if (cap_list[MIGRATION_CAPABILITY_ZERO_COPY_SEND]) {
   1320         error_setg(errp,
   1321                    "Zero copy currently only available on Linux");
   1322         return false;
   1323     }
   1324 #endif
   1325 
   1326 
   1327     /* incoming side only */
   1328     if (runstate_check(RUN_STATE_INMIGRATE) &&
   1329         !migrate_multi_channels_is_allowed() &&
   1330         cap_list[MIGRATION_CAPABILITY_MULTIFD]) {
   1331         error_setg(errp, "multifd is not supported by current protocol");
   1332         return false;
   1333     }
   1334 
   1335     if (cap_list[MIGRATION_CAPABILITY_POSTCOPY_PREEMPT]) {
   1336         if (!cap_list[MIGRATION_CAPABILITY_POSTCOPY_RAM]) {
   1337             error_setg(errp, "Postcopy preempt requires postcopy-ram");
   1338             return false;
   1339         }
   1340 
   1341         /*
   1342          * Preempt mode requires urgent pages to be sent in separate
   1343          * channel, OTOH compression logic will disorder all pages into
   1344          * different compression channels, which is not compatible with the
   1345          * preempt assumptions on channel assignments.
   1346          */
   1347         if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
   1348             error_setg(errp, "Postcopy preempt not compatible with compress");
   1349             return false;
   1350         }
   1351     }
   1352 
   1353     if (cap_list[MIGRATION_CAPABILITY_MULTIFD]) {
   1354         if (cap_list[MIGRATION_CAPABILITY_COMPRESS]) {
   1355             error_setg(errp, "Multifd is not compatible with compress");
   1356             return false;
   1357         }
   1358     }
   1359 
   1360     return true;
   1361 }
   1362 
   1363 static void fill_destination_migration_info(MigrationInfo *info)
   1364 {
   1365     MigrationIncomingState *mis = migration_incoming_get_current();
   1366 
   1367     if (mis->socket_address_list) {
   1368         info->has_socket_address = true;
   1369         info->socket_address =
   1370             QAPI_CLONE(SocketAddressList, mis->socket_address_list);
   1371     }
   1372 
   1373     switch (mis->state) {
   1374     case MIGRATION_STATUS_NONE:
   1375         return;
   1376     case MIGRATION_STATUS_SETUP:
   1377     case MIGRATION_STATUS_CANCELLING:
   1378     case MIGRATION_STATUS_CANCELLED:
   1379     case MIGRATION_STATUS_ACTIVE:
   1380     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
   1381     case MIGRATION_STATUS_POSTCOPY_PAUSED:
   1382     case MIGRATION_STATUS_POSTCOPY_RECOVER:
   1383     case MIGRATION_STATUS_FAILED:
   1384     case MIGRATION_STATUS_COLO:
   1385         info->has_status = true;
   1386         break;
   1387     case MIGRATION_STATUS_COMPLETED:
   1388         info->has_status = true;
   1389         fill_destination_postcopy_migration_info(info);
   1390         break;
   1391     }
   1392     info->status = mis->state;
   1393 }
   1394 
   1395 MigrationInfo *qmp_query_migrate(Error **errp)
   1396 {
   1397     MigrationInfo *info = g_malloc0(sizeof(*info));
   1398 
   1399     fill_destination_migration_info(info);
   1400     fill_source_migration_info(info);
   1401 
   1402     return info;
   1403 }
   1404 
   1405 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList *params,
   1406                                   Error **errp)
   1407 {
   1408     MigrationState *s = migrate_get_current();
   1409     MigrationCapabilityStatusList *cap;
   1410     bool cap_list[MIGRATION_CAPABILITY__MAX];
   1411 
   1412     if (migration_is_running(s->state)) {
   1413         error_setg(errp, QERR_MIGRATION_ACTIVE);
   1414         return;
   1415     }
   1416 
   1417     memcpy(cap_list, s->enabled_capabilities, sizeof(cap_list));
   1418     if (!migrate_caps_check(cap_list, params, errp)) {
   1419         return;
   1420     }
   1421 
   1422     for (cap = params; cap; cap = cap->next) {
   1423         s->enabled_capabilities[cap->value->capability] = cap->value->state;
   1424     }
   1425 }
   1426 
   1427 /*
   1428  * Check whether the parameters are valid. Error will be put into errp
   1429  * (if provided). Return true if valid, otherwise false.
   1430  */
   1431 static bool migrate_params_check(MigrationParameters *params, Error **errp)
   1432 {
   1433     if (params->has_compress_level &&
   1434         (params->compress_level > 9)) {
   1435         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "compress_level",
   1436                    "a value between 0 and 9");
   1437         return false;
   1438     }
   1439 
   1440     if (params->has_compress_threads && (params->compress_threads < 1)) {
   1441         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
   1442                    "compress_threads",
   1443                    "a value between 1 and 255");
   1444         return false;
   1445     }
   1446 
   1447     if (params->has_decompress_threads && (params->decompress_threads < 1)) {
   1448         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
   1449                    "decompress_threads",
   1450                    "a value between 1 and 255");
   1451         return false;
   1452     }
   1453 
   1454     if (params->has_throttle_trigger_threshold &&
   1455         (params->throttle_trigger_threshold < 1 ||
   1456          params->throttle_trigger_threshold > 100)) {
   1457         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
   1458                    "throttle_trigger_threshold",
   1459                    "an integer in the range of 1 to 100");
   1460         return false;
   1461     }
   1462 
   1463     if (params->has_cpu_throttle_initial &&
   1464         (params->cpu_throttle_initial < 1 ||
   1465          params->cpu_throttle_initial > 99)) {
   1466         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
   1467                    "cpu_throttle_initial",
   1468                    "an integer in the range of 1 to 99");
   1469         return false;
   1470     }
   1471 
   1472     if (params->has_cpu_throttle_increment &&
   1473         (params->cpu_throttle_increment < 1 ||
   1474          params->cpu_throttle_increment > 99)) {
   1475         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
   1476                    "cpu_throttle_increment",
   1477                    "an integer in the range of 1 to 99");
   1478         return false;
   1479     }
   1480 
   1481     if (params->has_max_bandwidth && (params->max_bandwidth > SIZE_MAX)) {
   1482         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
   1483                    "max_bandwidth",
   1484                    "an integer in the range of 0 to "stringify(SIZE_MAX)
   1485                    " bytes/second");
   1486         return false;
   1487     }
   1488 
   1489     if (params->has_downtime_limit &&
   1490         (params->downtime_limit > MAX_MIGRATE_DOWNTIME)) {
   1491         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
   1492                    "downtime_limit",
   1493                    "an integer in the range of 0 to "
   1494                     stringify(MAX_MIGRATE_DOWNTIME)" ms");
   1495         return false;
   1496     }
   1497 
   1498     /* x_checkpoint_delay is now always positive */
   1499 
   1500     if (params->has_multifd_channels && (params->multifd_channels < 1)) {
   1501         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
   1502                    "multifd_channels",
   1503                    "a value between 1 and 255");
   1504         return false;
   1505     }
   1506 
   1507     if (params->has_multifd_zlib_level &&
   1508         (params->multifd_zlib_level > 9)) {
   1509         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zlib_level",
   1510                    "a value between 0 and 9");
   1511         return false;
   1512     }
   1513 
   1514     if (params->has_multifd_zstd_level &&
   1515         (params->multifd_zstd_level > 20)) {
   1516         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "multifd_zstd_level",
   1517                    "a value between 0 and 20");
   1518         return false;
   1519     }
   1520 
   1521     if (params->has_xbzrle_cache_size &&
   1522         (params->xbzrle_cache_size < qemu_target_page_size() ||
   1523          !is_power_of_2(params->xbzrle_cache_size))) {
   1524         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
   1525                    "xbzrle_cache_size",
   1526                    "a power of two no less than the target page size");
   1527         return false;
   1528     }
   1529 
   1530     if (params->has_max_cpu_throttle &&
   1531         (params->max_cpu_throttle < params->cpu_throttle_initial ||
   1532          params->max_cpu_throttle > 99)) {
   1533         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
   1534                    "max_cpu_throttle",
   1535                    "an integer in the range of cpu_throttle_initial to 99");
   1536         return false;
   1537     }
   1538 
   1539     if (params->has_announce_initial &&
   1540         params->announce_initial > 100000) {
   1541         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
   1542                    "announce_initial",
   1543                    "a value between 0 and 100000");
   1544         return false;
   1545     }
   1546     if (params->has_announce_max &&
   1547         params->announce_max > 100000) {
   1548         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
   1549                    "announce_max",
   1550                    "a value between 0 and 100000");
   1551        return false;
   1552     }
   1553     if (params->has_announce_rounds &&
   1554         params->announce_rounds > 1000) {
   1555         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
   1556                    "announce_rounds",
   1557                    "a value between 0 and 1000");
   1558        return false;
   1559     }
   1560     if (params->has_announce_step &&
   1561         (params->announce_step < 1 ||
   1562         params->announce_step > 10000)) {
   1563         error_setg(errp, QERR_INVALID_PARAMETER_VALUE,
   1564                    "announce_step",
   1565                    "a value between 0 and 10000");
   1566        return false;
   1567     }
   1568 
   1569     if (params->has_block_bitmap_mapping &&
   1570         !check_dirty_bitmap_mig_alias_map(params->block_bitmap_mapping, errp)) {
   1571         error_prepend(errp, "Invalid mapping given for block-bitmap-mapping: ");
   1572         return false;
   1573     }
   1574 
   1575 #ifdef CONFIG_LINUX
   1576     if (migrate_use_zero_copy_send() &&
   1577         ((params->has_multifd_compression && params->multifd_compression) ||
   1578          (params->has_tls_creds && params->tls_creds && *params->tls_creds))) {
   1579         error_setg(errp,
   1580                    "Zero copy only available for non-compressed non-TLS multifd migration");
   1581         return false;
   1582     }
   1583 #endif
   1584 
   1585     return true;
   1586 }
   1587 
   1588 static void migrate_params_test_apply(MigrateSetParameters *params,
   1589                                       MigrationParameters *dest)
   1590 {
   1591     *dest = migrate_get_current()->parameters;
   1592 
   1593     /* TODO use QAPI_CLONE() instead of duplicating it inline */
   1594 
   1595     if (params->has_compress_level) {
   1596         dest->compress_level = params->compress_level;
   1597     }
   1598 
   1599     if (params->has_compress_threads) {
   1600         dest->compress_threads = params->compress_threads;
   1601     }
   1602 
   1603     if (params->has_compress_wait_thread) {
   1604         dest->compress_wait_thread = params->compress_wait_thread;
   1605     }
   1606 
   1607     if (params->has_decompress_threads) {
   1608         dest->decompress_threads = params->decompress_threads;
   1609     }
   1610 
   1611     if (params->has_throttle_trigger_threshold) {
   1612         dest->throttle_trigger_threshold = params->throttle_trigger_threshold;
   1613     }
   1614 
   1615     if (params->has_cpu_throttle_initial) {
   1616         dest->cpu_throttle_initial = params->cpu_throttle_initial;
   1617     }
   1618 
   1619     if (params->has_cpu_throttle_increment) {
   1620         dest->cpu_throttle_increment = params->cpu_throttle_increment;
   1621     }
   1622 
   1623     if (params->has_cpu_throttle_tailslow) {
   1624         dest->cpu_throttle_tailslow = params->cpu_throttle_tailslow;
   1625     }
   1626 
   1627     if (params->has_tls_creds) {
   1628         assert(params->tls_creds->type == QTYPE_QSTRING);
   1629         dest->tls_creds = params->tls_creds->u.s;
   1630     }
   1631 
   1632     if (params->has_tls_hostname) {
   1633         assert(params->tls_hostname->type == QTYPE_QSTRING);
   1634         dest->tls_hostname = params->tls_hostname->u.s;
   1635     }
   1636 
   1637     if (params->has_max_bandwidth) {
   1638         dest->max_bandwidth = params->max_bandwidth;
   1639     }
   1640 
   1641     if (params->has_downtime_limit) {
   1642         dest->downtime_limit = params->downtime_limit;
   1643     }
   1644 
   1645     if (params->has_x_checkpoint_delay) {
   1646         dest->x_checkpoint_delay = params->x_checkpoint_delay;
   1647     }
   1648 
   1649     if (params->has_block_incremental) {
   1650         dest->block_incremental = params->block_incremental;
   1651     }
   1652     if (params->has_multifd_channels) {
   1653         dest->multifd_channels = params->multifd_channels;
   1654     }
   1655     if (params->has_multifd_compression) {
   1656         dest->multifd_compression = params->multifd_compression;
   1657     }
   1658     if (params->has_xbzrle_cache_size) {
   1659         dest->xbzrle_cache_size = params->xbzrle_cache_size;
   1660     }
   1661     if (params->has_max_postcopy_bandwidth) {
   1662         dest->max_postcopy_bandwidth = params->max_postcopy_bandwidth;
   1663     }
   1664     if (params->has_max_cpu_throttle) {
   1665         dest->max_cpu_throttle = params->max_cpu_throttle;
   1666     }
   1667     if (params->has_announce_initial) {
   1668         dest->announce_initial = params->announce_initial;
   1669     }
   1670     if (params->has_announce_max) {
   1671         dest->announce_max = params->announce_max;
   1672     }
   1673     if (params->has_announce_rounds) {
   1674         dest->announce_rounds = params->announce_rounds;
   1675     }
   1676     if (params->has_announce_step) {
   1677         dest->announce_step = params->announce_step;
   1678     }
   1679 
   1680     if (params->has_block_bitmap_mapping) {
   1681         dest->has_block_bitmap_mapping = true;
   1682         dest->block_bitmap_mapping = params->block_bitmap_mapping;
   1683     }
   1684 }
   1685 
   1686 static void migrate_params_apply(MigrateSetParameters *params, Error **errp)
   1687 {
   1688     MigrationState *s = migrate_get_current();
   1689 
   1690     /* TODO use QAPI_CLONE() instead of duplicating it inline */
   1691 
   1692     if (params->has_compress_level) {
   1693         s->parameters.compress_level = params->compress_level;
   1694     }
   1695 
   1696     if (params->has_compress_threads) {
   1697         s->parameters.compress_threads = params->compress_threads;
   1698     }
   1699 
   1700     if (params->has_compress_wait_thread) {
   1701         s->parameters.compress_wait_thread = params->compress_wait_thread;
   1702     }
   1703 
   1704     if (params->has_decompress_threads) {
   1705         s->parameters.decompress_threads = params->decompress_threads;
   1706     }
   1707 
   1708     if (params->has_throttle_trigger_threshold) {
   1709         s->parameters.throttle_trigger_threshold = params->throttle_trigger_threshold;
   1710     }
   1711 
   1712     if (params->has_cpu_throttle_initial) {
   1713         s->parameters.cpu_throttle_initial = params->cpu_throttle_initial;
   1714     }
   1715 
   1716     if (params->has_cpu_throttle_increment) {
   1717         s->parameters.cpu_throttle_increment = params->cpu_throttle_increment;
   1718     }
   1719 
   1720     if (params->has_cpu_throttle_tailslow) {
   1721         s->parameters.cpu_throttle_tailslow = params->cpu_throttle_tailslow;
   1722     }
   1723 
   1724     if (params->has_tls_creds) {
   1725         g_free(s->parameters.tls_creds);
   1726         assert(params->tls_creds->type == QTYPE_QSTRING);
   1727         s->parameters.tls_creds = g_strdup(params->tls_creds->u.s);
   1728     }
   1729 
   1730     if (params->has_tls_hostname) {
   1731         g_free(s->parameters.tls_hostname);
   1732         assert(params->tls_hostname->type == QTYPE_QSTRING);
   1733         s->parameters.tls_hostname = g_strdup(params->tls_hostname->u.s);
   1734     }
   1735 
   1736     if (params->has_tls_authz) {
   1737         g_free(s->parameters.tls_authz);
   1738         assert(params->tls_authz->type == QTYPE_QSTRING);
   1739         s->parameters.tls_authz = g_strdup(params->tls_authz->u.s);
   1740     }
   1741 
   1742     if (params->has_max_bandwidth) {
   1743         s->parameters.max_bandwidth = params->max_bandwidth;
   1744         if (s->to_dst_file && !migration_in_postcopy()) {
   1745             qemu_file_set_rate_limit(s->to_dst_file,
   1746                                 s->parameters.max_bandwidth / XFER_LIMIT_RATIO);
   1747         }
   1748     }
   1749 
   1750     if (params->has_downtime_limit) {
   1751         s->parameters.downtime_limit = params->downtime_limit;
   1752     }
   1753 
   1754     if (params->has_x_checkpoint_delay) {
   1755         s->parameters.x_checkpoint_delay = params->x_checkpoint_delay;
   1756         if (migration_in_colo_state()) {
   1757             colo_checkpoint_notify(s);
   1758         }
   1759     }
   1760 
   1761     if (params->has_block_incremental) {
   1762         s->parameters.block_incremental = params->block_incremental;
   1763     }
   1764     if (params->has_multifd_channels) {
   1765         s->parameters.multifd_channels = params->multifd_channels;
   1766     }
   1767     if (params->has_multifd_compression) {
   1768         s->parameters.multifd_compression = params->multifd_compression;
   1769     }
   1770     if (params->has_xbzrle_cache_size) {
   1771         s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;
   1772         xbzrle_cache_resize(params->xbzrle_cache_size, errp);
   1773     }
   1774     if (params->has_max_postcopy_bandwidth) {
   1775         s->parameters.max_postcopy_bandwidth = params->max_postcopy_bandwidth;
   1776         if (s->to_dst_file && migration_in_postcopy()) {
   1777             qemu_file_set_rate_limit(s->to_dst_file,
   1778                     s->parameters.max_postcopy_bandwidth / XFER_LIMIT_RATIO);
   1779         }
   1780     }
   1781     if (params->has_max_cpu_throttle) {
   1782         s->parameters.max_cpu_throttle = params->max_cpu_throttle;
   1783     }
   1784     if (params->has_announce_initial) {
   1785         s->parameters.announce_initial = params->announce_initial;
   1786     }
   1787     if (params->has_announce_max) {
   1788         s->parameters.announce_max = params->announce_max;
   1789     }
   1790     if (params->has_announce_rounds) {
   1791         s->parameters.announce_rounds = params->announce_rounds;
   1792     }
   1793     if (params->has_announce_step) {
   1794         s->parameters.announce_step = params->announce_step;
   1795     }
   1796 
   1797     if (params->has_block_bitmap_mapping) {
   1798         qapi_free_BitmapMigrationNodeAliasList(
   1799             s->parameters.block_bitmap_mapping);
   1800 
   1801         s->parameters.has_block_bitmap_mapping = true;
   1802         s->parameters.block_bitmap_mapping =
   1803             QAPI_CLONE(BitmapMigrationNodeAliasList,
   1804                        params->block_bitmap_mapping);
   1805     }
   1806 }
   1807 
   1808 void qmp_migrate_set_parameters(MigrateSetParameters *params, Error **errp)
   1809 {
   1810     MigrationParameters tmp;
   1811 
   1812     /* TODO Rewrite "" to null instead */
   1813     if (params->has_tls_creds
   1814         && params->tls_creds->type == QTYPE_QNULL) {
   1815         qobject_unref(params->tls_creds->u.n);
   1816         params->tls_creds->type = QTYPE_QSTRING;
   1817         params->tls_creds->u.s = strdup("");
   1818     }
   1819     /* TODO Rewrite "" to null instead */
   1820     if (params->has_tls_hostname
   1821         && params->tls_hostname->type == QTYPE_QNULL) {
   1822         qobject_unref(params->tls_hostname->u.n);
   1823         params->tls_hostname->type = QTYPE_QSTRING;
   1824         params->tls_hostname->u.s = strdup("");
   1825     }
   1826 
   1827     migrate_params_test_apply(params, &tmp);
   1828 
   1829     if (!migrate_params_check(&tmp, errp)) {
   1830         /* Invalid parameter */
   1831         return;
   1832     }
   1833 
   1834     migrate_params_apply(params, errp);
   1835 }
   1836 
   1837 
   1838 void qmp_migrate_start_postcopy(Error **errp)
   1839 {
   1840     MigrationState *s = migrate_get_current();
   1841 
   1842     if (!migrate_postcopy()) {
   1843         error_setg(errp, "Enable postcopy with migrate_set_capability before"
   1844                          " the start of migration");
   1845         return;
   1846     }
   1847 
   1848     if (s->state == MIGRATION_STATUS_NONE) {
   1849         error_setg(errp, "Postcopy must be started after migration has been"
   1850                          " started");
   1851         return;
   1852     }
   1853     /*
   1854      * we don't error if migration has finished since that would be racy
   1855      * with issuing this command.
   1856      */
   1857     qatomic_set(&s->start_postcopy, true);
   1858 }
   1859 
   1860 /* shared migration helpers */
   1861 
   1862 void migrate_set_state(int *state, int old_state, int new_state)
   1863 {
   1864     assert(new_state < MIGRATION_STATUS__MAX);
   1865     if (qatomic_cmpxchg(state, old_state, new_state) == old_state) {
   1866         trace_migrate_set_state(MigrationStatus_str(new_state));
   1867         migrate_generate_event(new_state);
   1868     }
   1869 }
   1870 
   1871 static MigrationCapabilityStatus *migrate_cap_add(MigrationCapability index,
   1872                                                   bool state)
   1873 {
   1874     MigrationCapabilityStatus *cap;
   1875 
   1876     cap = g_new0(MigrationCapabilityStatus, 1);
   1877     cap->capability = index;
   1878     cap->state = state;
   1879 
   1880     return cap;
   1881 }
   1882 
   1883 void migrate_set_block_enabled(bool value, Error **errp)
   1884 {
   1885     MigrationCapabilityStatusList *cap = NULL;
   1886 
   1887     QAPI_LIST_PREPEND(cap, migrate_cap_add(MIGRATION_CAPABILITY_BLOCK, value));
   1888     qmp_migrate_set_capabilities(cap, errp);
   1889     qapi_free_MigrationCapabilityStatusList(cap);
   1890 }
   1891 
   1892 static void migrate_set_block_incremental(MigrationState *s, bool value)
   1893 {
   1894     s->parameters.block_incremental = value;
   1895 }
   1896 
   1897 static void block_cleanup_parameters(MigrationState *s)
   1898 {
   1899     if (s->must_remove_block_options) {
   1900         /* setting to false can never fail */
   1901         migrate_set_block_enabled(false, &error_abort);
   1902         migrate_set_block_incremental(s, false);
   1903         s->must_remove_block_options = false;
   1904     }
   1905 }
   1906 
   1907 static void migrate_fd_cleanup(MigrationState *s)
   1908 {
   1909     qemu_bh_delete(s->cleanup_bh);
   1910     s->cleanup_bh = NULL;
   1911 
   1912     g_free(s->hostname);
   1913     s->hostname = NULL;
   1914 
   1915     qemu_savevm_state_cleanup();
   1916 
   1917     if (s->to_dst_file) {
   1918         QEMUFile *tmp;
   1919 
   1920         trace_migrate_fd_cleanup();
   1921         qemu_mutex_unlock_iothread();
   1922         if (s->migration_thread_running) {
   1923             qemu_thread_join(&s->thread);
   1924             s->migration_thread_running = false;
   1925         }
   1926         qemu_mutex_lock_iothread();
   1927 
   1928         multifd_save_cleanup();
   1929         qemu_mutex_lock(&s->qemu_file_lock);
   1930         tmp = s->to_dst_file;
   1931         s->to_dst_file = NULL;
   1932         qemu_mutex_unlock(&s->qemu_file_lock);
   1933         /*
   1934          * Close the file handle without the lock to make sure the
   1935          * critical section won't block for long.
   1936          */
   1937         migration_ioc_unregister_yank_from_file(tmp);
   1938         qemu_fclose(tmp);
   1939     }
   1940 
   1941     if (s->postcopy_qemufile_src) {
   1942         migration_ioc_unregister_yank_from_file(s->postcopy_qemufile_src);
   1943         qemu_fclose(s->postcopy_qemufile_src);
   1944         s->postcopy_qemufile_src = NULL;
   1945     }
   1946 
   1947     assert(!migration_is_active(s));
   1948 
   1949     if (s->state == MIGRATION_STATUS_CANCELLING) {
   1950         migrate_set_state(&s->state, MIGRATION_STATUS_CANCELLING,
   1951                           MIGRATION_STATUS_CANCELLED);
   1952     }
   1953 
   1954     if (s->error) {
   1955         /* It is used on info migrate.  We can't free it */
   1956         error_report_err(error_copy(s->error));
   1957     }
   1958     notifier_list_notify(&migration_state_notifiers, s);
   1959     block_cleanup_parameters(s);
   1960     yank_unregister_instance(MIGRATION_YANK_INSTANCE);
   1961 }
   1962 
   1963 static void migrate_fd_cleanup_schedule(MigrationState *s)
   1964 {
   1965     /*
   1966      * Ref the state for bh, because it may be called when
   1967      * there're already no other refs
   1968      */
   1969     object_ref(OBJECT(s));
   1970     qemu_bh_schedule(s->cleanup_bh);
   1971 }
   1972 
   1973 static void migrate_fd_cleanup_bh(void *opaque)
   1974 {
   1975     MigrationState *s = opaque;
   1976     migrate_fd_cleanup(s);
   1977     object_unref(OBJECT(s));
   1978 }
   1979 
   1980 void migrate_set_error(MigrationState *s, const Error *error)
   1981 {
   1982     QEMU_LOCK_GUARD(&s->error_mutex);
   1983     if (!s->error) {
   1984         s->error = error_copy(error);
   1985     }
   1986 }
   1987 
   1988 static void migrate_error_free(MigrationState *s)
   1989 {
   1990     QEMU_LOCK_GUARD(&s->error_mutex);
   1991     if (s->error) {
   1992         error_free(s->error);
   1993         s->error = NULL;
   1994     }
   1995 }
   1996 
   1997 void migrate_fd_error(MigrationState *s, const Error *error)
   1998 {
   1999     trace_migrate_fd_error(error_get_pretty(error));
   2000     assert(s->to_dst_file == NULL);
   2001     migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
   2002                       MIGRATION_STATUS_FAILED);
   2003     migrate_set_error(s, error);
   2004 }
   2005 
   2006 static void migrate_fd_cancel(MigrationState *s)
   2007 {
   2008     int old_state ;
   2009     QEMUFile *f = migrate_get_current()->to_dst_file;
   2010     trace_migrate_fd_cancel();
   2011 
   2012     WITH_QEMU_LOCK_GUARD(&s->qemu_file_lock) {
   2013         if (s->rp_state.from_dst_file) {
   2014             /* shutdown the rp socket, so causing the rp thread to shutdown */
   2015             qemu_file_shutdown(s->rp_state.from_dst_file);
   2016         }
   2017     }
   2018 
   2019     do {
   2020         old_state = s->state;
   2021         if (!migration_is_running(old_state)) {
   2022             break;
   2023         }
   2024         /* If the migration is paused, kick it out of the pause */
   2025         if (old_state == MIGRATION_STATUS_PRE_SWITCHOVER) {
   2026             qemu_sem_post(&s->pause_sem);
   2027         }
   2028         migrate_set_state(&s->state, old_state, MIGRATION_STATUS_CANCELLING);
   2029     } while (s->state != MIGRATION_STATUS_CANCELLING);
   2030 
   2031     /*
   2032      * If we're unlucky the migration code might be stuck somewhere in a
   2033      * send/write while the network has failed and is waiting to timeout;
   2034      * if we've got shutdown(2) available then we can force it to quit.
   2035      * The outgoing qemu file gets closed in migrate_fd_cleanup that is
   2036      * called in a bh, so there is no race against this cancel.
   2037      */
   2038     if (s->state == MIGRATION_STATUS_CANCELLING && f) {
   2039         qemu_file_shutdown(f);
   2040     }
   2041     if (s->state == MIGRATION_STATUS_CANCELLING && s->block_inactive) {
   2042         Error *local_err = NULL;
   2043 
   2044         bdrv_activate_all(&local_err);
   2045         if (local_err) {
   2046             error_report_err(local_err);
   2047         } else {
   2048             s->block_inactive = false;
   2049         }
   2050     }
   2051 }
   2052 
   2053 void add_migration_state_change_notifier(Notifier *notify)
   2054 {
   2055     notifier_list_add(&migration_state_notifiers, notify);
   2056 }
   2057 
   2058 void remove_migration_state_change_notifier(Notifier *notify)
   2059 {
   2060     notifier_remove(notify);
   2061 }
   2062 
   2063 bool migration_in_setup(MigrationState *s)
   2064 {
   2065     return s->state == MIGRATION_STATUS_SETUP;
   2066 }
   2067 
   2068 bool migration_has_finished(MigrationState *s)
   2069 {
   2070     return s->state == MIGRATION_STATUS_COMPLETED;
   2071 }
   2072 
   2073 bool migration_has_failed(MigrationState *s)
   2074 {
   2075     return (s->state == MIGRATION_STATUS_CANCELLED ||
   2076             s->state == MIGRATION_STATUS_FAILED);
   2077 }
   2078 
   2079 bool migration_in_postcopy(void)
   2080 {
   2081     MigrationState *s = migrate_get_current();
   2082 
   2083     switch (s->state) {
   2084     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
   2085     case MIGRATION_STATUS_POSTCOPY_PAUSED:
   2086     case MIGRATION_STATUS_POSTCOPY_RECOVER:
   2087         return true;
   2088     default:
   2089         return false;
   2090     }
   2091 }
   2092 
   2093 bool migration_in_postcopy_after_devices(MigrationState *s)
   2094 {
   2095     return migration_in_postcopy() && s->postcopy_after_devices;
   2096 }
   2097 
   2098 bool migration_in_incoming_postcopy(void)
   2099 {
   2100     PostcopyState ps = postcopy_state_get();
   2101 
   2102     return ps >= POSTCOPY_INCOMING_DISCARD && ps < POSTCOPY_INCOMING_END;
   2103 }
   2104 
   2105 bool migration_in_bg_snapshot(void)
   2106 {
   2107     MigrationState *s = migrate_get_current();
   2108 
   2109     return migrate_background_snapshot() &&
   2110             migration_is_setup_or_active(s->state);
   2111 }
   2112 
   2113 bool migration_is_idle(void)
   2114 {
   2115     MigrationState *s = current_migration;
   2116 
   2117     if (!s) {
   2118         return true;
   2119     }
   2120 
   2121     switch (s->state) {
   2122     case MIGRATION_STATUS_NONE:
   2123     case MIGRATION_STATUS_CANCELLED:
   2124     case MIGRATION_STATUS_COMPLETED:
   2125     case MIGRATION_STATUS_FAILED:
   2126         return true;
   2127     case MIGRATION_STATUS_SETUP:
   2128     case MIGRATION_STATUS_CANCELLING:
   2129     case MIGRATION_STATUS_ACTIVE:
   2130     case MIGRATION_STATUS_POSTCOPY_ACTIVE:
   2131     case MIGRATION_STATUS_COLO:
   2132     case MIGRATION_STATUS_PRE_SWITCHOVER:
   2133     case MIGRATION_STATUS_DEVICE:
   2134     case MIGRATION_STATUS_WAIT_UNPLUG:
   2135         return false;
   2136     case MIGRATION_STATUS__MAX:
   2137         g_assert_not_reached();
   2138     }
   2139 
   2140     return false;
   2141 }
   2142 
   2143 bool migration_is_active(MigrationState *s)
   2144 {
   2145     return (s->state == MIGRATION_STATUS_ACTIVE ||
   2146             s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
   2147 }
   2148 
   2149 void migrate_init(MigrationState *s)
   2150 {
   2151     /*
   2152      * Reinitialise all migration state, except
   2153      * parameters/capabilities that the user set, and
   2154      * locks.
   2155      */
   2156     s->cleanup_bh = 0;
   2157     s->vm_start_bh = 0;
   2158     s->to_dst_file = NULL;
   2159     s->state = MIGRATION_STATUS_NONE;
   2160     s->rp_state.from_dst_file = NULL;
   2161     s->rp_state.error = false;
   2162     s->mbps = 0.0;
   2163     s->pages_per_second = 0.0;
   2164     s->downtime = 0;
   2165     s->expected_downtime = 0;
   2166     s->setup_time = 0;
   2167     s->start_postcopy = false;
   2168     s->postcopy_after_devices = false;
   2169     s->migration_thread_running = false;
   2170     error_free(s->error);
   2171     s->error = NULL;
   2172     s->hostname = NULL;
   2173 
   2174     migrate_set_state(&s->state, MIGRATION_STATUS_NONE, MIGRATION_STATUS_SETUP);
   2175 
   2176     s->start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
   2177     s->total_time = 0;
   2178     s->vm_was_running = false;
   2179     s->iteration_initial_bytes = 0;
   2180     s->threshold_size = 0;
   2181 }
   2182 
   2183 int migrate_add_blocker_internal(Error *reason, Error **errp)
   2184 {
   2185     /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */
   2186     if (runstate_check(RUN_STATE_SAVE_VM) || !migration_is_idle()) {
   2187         error_propagate_prepend(errp, error_copy(reason),
   2188                                 "disallowing migration blocker "
   2189                                 "(migration/snapshot in progress) for: ");
   2190         return -EBUSY;
   2191     }
   2192 
   2193     migration_blockers = g_slist_prepend(migration_blockers, reason);
   2194     return 0;
   2195 }
   2196 
   2197 int migrate_add_blocker(Error *reason, Error **errp)
   2198 {
   2199     if (only_migratable) {
   2200         error_propagate_prepend(errp, error_copy(reason),
   2201                                 "disallowing migration blocker "
   2202                                 "(--only-migratable) for: ");
   2203         return -EACCES;
   2204     }
   2205 
   2206     return migrate_add_blocker_internal(reason, errp);
   2207 }
   2208 
   2209 void migrate_del_blocker(Error *reason)
   2210 {
   2211     migration_blockers = g_slist_remove(migration_blockers, reason);
   2212 }
   2213 
   2214 void qmp_migrate_incoming(const char *uri, Error **errp)
   2215 {
   2216     Error *local_err = NULL;
   2217     static bool once = true;
   2218 
   2219     if (!once) {
   2220         error_setg(errp, "The incoming migration has already been started");
   2221         return;
   2222     }
   2223     if (!runstate_check(RUN_STATE_INMIGRATE)) {
   2224         error_setg(errp, "'-incoming' was not specified on the command line");
   2225         return;
   2226     }
   2227 
   2228     if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) {
   2229         return;
   2230     }
   2231 
   2232     qemu_start_incoming_migration(uri, &local_err);
   2233 
   2234     if (local_err) {
   2235         yank_unregister_instance(MIGRATION_YANK_INSTANCE);
   2236         error_propagate(errp, local_err);
   2237         return;
   2238     }
   2239 
   2240     once = false;
   2241 }
   2242 
   2243 void qmp_migrate_recover(const char *uri, Error **errp)
   2244 {
   2245     MigrationIncomingState *mis = migration_incoming_get_current();
   2246 
   2247     /*
   2248      * Don't even bother to use ERRP_GUARD() as it _must_ always be set by
   2249      * callers (no one should ignore a recover failure); if there is, it's a
   2250      * programming error.
   2251      */
   2252     assert(errp);
   2253 
   2254     if (mis->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
   2255         error_setg(errp, "Migrate recover can only be run "
   2256                    "when postcopy is paused.");
   2257         return;
   2258     }
   2259 
   2260     /* If there's an existing transport, release it */
   2261     migration_incoming_transport_cleanup(mis);
   2262 
   2263     /*
   2264      * Note that this call will never start a real migration; it will
   2265      * only re-setup the migration stream and poke existing migration
   2266      * to continue using that newly established channel.
   2267      */
   2268     qemu_start_incoming_migration(uri, errp);
   2269 }
   2270 
   2271 void qmp_migrate_pause(Error **errp)
   2272 {
   2273     MigrationState *ms = migrate_get_current();
   2274     MigrationIncomingState *mis = migration_incoming_get_current();
   2275     int ret;
   2276 
   2277     if (ms->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
   2278         /* Source side, during postcopy */
   2279         qemu_mutex_lock(&ms->qemu_file_lock);
   2280         ret = qemu_file_shutdown(ms->to_dst_file);
   2281         qemu_mutex_unlock(&ms->qemu_file_lock);
   2282         if (ret) {
   2283             error_setg(errp, "Failed to pause source migration");
   2284         }
   2285         return;
   2286     }
   2287 
   2288     if (mis->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
   2289         ret = qemu_file_shutdown(mis->from_src_file);
   2290         if (ret) {
   2291             error_setg(errp, "Failed to pause destination migration");
   2292         }
   2293         return;
   2294     }
   2295 
   2296     error_setg(errp, "migrate-pause is currently only supported "
   2297                "during postcopy-active state");
   2298 }
   2299 
   2300 bool migration_is_blocked(Error **errp)
   2301 {
   2302     if (qemu_savevm_state_blocked(errp)) {
   2303         return true;
   2304     }
   2305 
   2306     if (migration_blockers) {
   2307         error_propagate(errp, error_copy(migration_blockers->data));
   2308         return true;
   2309     }
   2310 
   2311     return false;
   2312 }
   2313 
   2314 /* Returns true if continue to migrate, or false if error detected */
   2315 static bool migrate_prepare(MigrationState *s, bool blk, bool blk_inc,
   2316                             bool resume, Error **errp)
   2317 {
   2318     Error *local_err = NULL;
   2319 
   2320     if (resume) {
   2321         if (s->state != MIGRATION_STATUS_POSTCOPY_PAUSED) {
   2322             error_setg(errp, "Cannot resume if there is no "
   2323                        "paused migration");
   2324             return false;
   2325         }
   2326 
   2327         /*
   2328          * Postcopy recovery won't work well with release-ram
   2329          * capability since release-ram will drop the page buffer as
   2330          * long as the page is put into the send buffer.  So if there
   2331          * is a network failure happened, any page buffers that have
   2332          * not yet reached the destination VM but have already been
   2333          * sent from the source VM will be lost forever.  Let's refuse
   2334          * the client from resuming such a postcopy migration.
   2335          * Luckily release-ram was designed to only be used when src
   2336          * and destination VMs are on the same host, so it should be
   2337          * fine.
   2338          */
   2339         if (migrate_release_ram()) {
   2340             error_setg(errp, "Postcopy recovery cannot work "
   2341                        "when release-ram capability is set");
   2342             return false;
   2343         }
   2344 
   2345         /* This is a resume, skip init status */
   2346         return true;
   2347     }
   2348 
   2349     if (migration_is_running(s->state)) {
   2350         error_setg(errp, QERR_MIGRATION_ACTIVE);
   2351         return false;
   2352     }
   2353 
   2354     if (runstate_check(RUN_STATE_INMIGRATE)) {
   2355         error_setg(errp, "Guest is waiting for an incoming migration");
   2356         return false;
   2357     }
   2358 
   2359     if (runstate_check(RUN_STATE_POSTMIGRATE)) {
   2360         error_setg(errp, "Can't migrate the vm that was paused due to "
   2361                    "previous migration");
   2362         return false;
   2363     }
   2364 
   2365     if (migration_is_blocked(errp)) {
   2366         return false;
   2367     }
   2368 
   2369     if (blk || blk_inc) {
   2370         if (migrate_colo_enabled()) {
   2371             error_setg(errp, "No disk migration is required in COLO mode");
   2372             return false;
   2373         }
   2374         if (migrate_use_block() || migrate_use_block_incremental()) {
   2375             error_setg(errp, "Command options are incompatible with "
   2376                        "current migration capabilities");
   2377             return false;
   2378         }
   2379         migrate_set_block_enabled(true, &local_err);
   2380         if (local_err) {
   2381             error_propagate(errp, local_err);
   2382             return false;
   2383         }
   2384         s->must_remove_block_options = true;
   2385     }
   2386 
   2387     if (blk_inc) {
   2388         migrate_set_block_incremental(s, true);
   2389     }
   2390 
   2391     migrate_init(s);
   2392     /*
   2393      * set ram_counters compression_counters memory to zero for a
   2394      * new migration
   2395      */
   2396     memset(&ram_counters, 0, sizeof(ram_counters));
   2397     memset(&compression_counters, 0, sizeof(compression_counters));
   2398 
   2399     return true;
   2400 }
   2401 
   2402 void qmp_migrate(const char *uri, bool has_blk, bool blk,
   2403                  bool has_inc, bool inc, bool has_detach, bool detach,
   2404                  bool has_resume, bool resume, Error **errp)
   2405 {
   2406     Error *local_err = NULL;
   2407     MigrationState *s = migrate_get_current();
   2408     const char *p = NULL;
   2409 
   2410     if (!migrate_prepare(s, has_blk && blk, has_inc && inc,
   2411                          has_resume && resume, errp)) {
   2412         /* Error detected, put into errp */
   2413         return;
   2414     }
   2415 
   2416     if (!(has_resume && resume)) {
   2417         if (!yank_register_instance(MIGRATION_YANK_INSTANCE, errp)) {
   2418             return;
   2419         }
   2420     }
   2421 
   2422     migrate_protocol_allow_multi_channels(false);
   2423     if (strstart(uri, "tcp:", &p) ||
   2424         strstart(uri, "unix:", NULL) ||
   2425         strstart(uri, "vsock:", NULL)) {
   2426         migrate_protocol_allow_multi_channels(true);
   2427         socket_start_outgoing_migration(s, p ? p : uri, &local_err);
   2428 #ifdef CONFIG_RDMA
   2429     } else if (strstart(uri, "rdma:", &p)) {
   2430         rdma_start_outgoing_migration(s, p, &local_err);
   2431 #endif
   2432     } else if (strstart(uri, "exec:", &p)) {
   2433         exec_start_outgoing_migration(s, p, &local_err);
   2434     } else if (strstart(uri, "fd:", &p)) {
   2435         fd_start_outgoing_migration(s, p, &local_err);
   2436     } else {
   2437         if (!(has_resume && resume)) {
   2438             yank_unregister_instance(MIGRATION_YANK_INSTANCE);
   2439         }
   2440         error_setg(errp, QERR_INVALID_PARAMETER_VALUE, "uri",
   2441                    "a valid migration protocol");
   2442         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
   2443                           MIGRATION_STATUS_FAILED);
   2444         block_cleanup_parameters(s);
   2445         return;
   2446     }
   2447 
   2448     if (local_err) {
   2449         if (!(has_resume && resume)) {
   2450             yank_unregister_instance(MIGRATION_YANK_INSTANCE);
   2451         }
   2452         migrate_fd_error(s, local_err);
   2453         error_propagate(errp, local_err);
   2454         return;
   2455     }
   2456 }
   2457 
   2458 void qmp_migrate_cancel(Error **errp)
   2459 {
   2460     migration_cancel(NULL);
   2461 }
   2462 
   2463 void qmp_migrate_continue(MigrationStatus state, Error **errp)
   2464 {
   2465     MigrationState *s = migrate_get_current();
   2466     if (s->state != state) {
   2467         error_setg(errp,  "Migration not in expected state: %s",
   2468                    MigrationStatus_str(s->state));
   2469         return;
   2470     }
   2471     qemu_sem_post(&s->pause_sem);
   2472 }
   2473 
   2474 bool migrate_release_ram(void)
   2475 {
   2476     MigrationState *s;
   2477 
   2478     s = migrate_get_current();
   2479 
   2480     return s->enabled_capabilities[MIGRATION_CAPABILITY_RELEASE_RAM];
   2481 }
   2482 
   2483 bool migrate_postcopy_ram(void)
   2484 {
   2485     MigrationState *s;
   2486 
   2487     s = migrate_get_current();
   2488 
   2489     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_RAM];
   2490 }
   2491 
   2492 bool migrate_postcopy(void)
   2493 {
   2494     return migrate_postcopy_ram() || migrate_dirty_bitmaps();
   2495 }
   2496 
   2497 bool migrate_auto_converge(void)
   2498 {
   2499     MigrationState *s;
   2500 
   2501     s = migrate_get_current();
   2502 
   2503     return s->enabled_capabilities[MIGRATION_CAPABILITY_AUTO_CONVERGE];
   2504 }
   2505 
   2506 bool migrate_zero_blocks(void)
   2507 {
   2508     MigrationState *s;
   2509 
   2510     s = migrate_get_current();
   2511 
   2512     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_BLOCKS];
   2513 }
   2514 
   2515 bool migrate_postcopy_blocktime(void)
   2516 {
   2517     MigrationState *s;
   2518 
   2519     s = migrate_get_current();
   2520 
   2521     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME];
   2522 }
   2523 
   2524 bool migrate_use_compression(void)
   2525 {
   2526     MigrationState *s;
   2527 
   2528     s = migrate_get_current();
   2529 
   2530     return s->enabled_capabilities[MIGRATION_CAPABILITY_COMPRESS];
   2531 }
   2532 
   2533 int migrate_compress_level(void)
   2534 {
   2535     MigrationState *s;
   2536 
   2537     s = migrate_get_current();
   2538 
   2539     return s->parameters.compress_level;
   2540 }
   2541 
   2542 int migrate_compress_threads(void)
   2543 {
   2544     MigrationState *s;
   2545 
   2546     s = migrate_get_current();
   2547 
   2548     return s->parameters.compress_threads;
   2549 }
   2550 
   2551 int migrate_compress_wait_thread(void)
   2552 {
   2553     MigrationState *s;
   2554 
   2555     s = migrate_get_current();
   2556 
   2557     return s->parameters.compress_wait_thread;
   2558 }
   2559 
   2560 int migrate_decompress_threads(void)
   2561 {
   2562     MigrationState *s;
   2563 
   2564     s = migrate_get_current();
   2565 
   2566     return s->parameters.decompress_threads;
   2567 }
   2568 
   2569 bool migrate_dirty_bitmaps(void)
   2570 {
   2571     MigrationState *s;
   2572 
   2573     s = migrate_get_current();
   2574 
   2575     return s->enabled_capabilities[MIGRATION_CAPABILITY_DIRTY_BITMAPS];
   2576 }
   2577 
   2578 bool migrate_ignore_shared(void)
   2579 {
   2580     MigrationState *s;
   2581 
   2582     s = migrate_get_current();
   2583 
   2584     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_IGNORE_SHARED];
   2585 }
   2586 
   2587 bool migrate_validate_uuid(void)
   2588 {
   2589     MigrationState *s;
   2590 
   2591     s = migrate_get_current();
   2592 
   2593     return s->enabled_capabilities[MIGRATION_CAPABILITY_VALIDATE_UUID];
   2594 }
   2595 
   2596 bool migrate_use_events(void)
   2597 {
   2598     MigrationState *s;
   2599 
   2600     s = migrate_get_current();
   2601 
   2602     return s->enabled_capabilities[MIGRATION_CAPABILITY_EVENTS];
   2603 }
   2604 
   2605 bool migrate_use_multifd(void)
   2606 {
   2607     MigrationState *s;
   2608 
   2609     s = migrate_get_current();
   2610 
   2611     return s->enabled_capabilities[MIGRATION_CAPABILITY_MULTIFD];
   2612 }
   2613 
   2614 bool migrate_pause_before_switchover(void)
   2615 {
   2616     MigrationState *s;
   2617 
   2618     s = migrate_get_current();
   2619 
   2620     return s->enabled_capabilities[
   2621         MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER];
   2622 }
   2623 
   2624 int migrate_multifd_channels(void)
   2625 {
   2626     MigrationState *s;
   2627 
   2628     s = migrate_get_current();
   2629 
   2630     return s->parameters.multifd_channels;
   2631 }
   2632 
   2633 MultiFDCompression migrate_multifd_compression(void)
   2634 {
   2635     MigrationState *s;
   2636 
   2637     s = migrate_get_current();
   2638 
   2639     assert(s->parameters.multifd_compression < MULTIFD_COMPRESSION__MAX);
   2640     return s->parameters.multifd_compression;
   2641 }
   2642 
   2643 int migrate_multifd_zlib_level(void)
   2644 {
   2645     MigrationState *s;
   2646 
   2647     s = migrate_get_current();
   2648 
   2649     return s->parameters.multifd_zlib_level;
   2650 }
   2651 
   2652 int migrate_multifd_zstd_level(void)
   2653 {
   2654     MigrationState *s;
   2655 
   2656     s = migrate_get_current();
   2657 
   2658     return s->parameters.multifd_zstd_level;
   2659 }
   2660 
   2661 #ifdef CONFIG_LINUX
   2662 bool migrate_use_zero_copy_send(void)
   2663 {
   2664     MigrationState *s;
   2665 
   2666     s = migrate_get_current();
   2667 
   2668     return s->enabled_capabilities[MIGRATION_CAPABILITY_ZERO_COPY_SEND];
   2669 }
   2670 #endif
   2671 
   2672 int migrate_use_tls(void)
   2673 {
   2674     MigrationState *s;
   2675 
   2676     s = migrate_get_current();
   2677 
   2678     return s->parameters.tls_creds && *s->parameters.tls_creds;
   2679 }
   2680 
   2681 int migrate_use_xbzrle(void)
   2682 {
   2683     MigrationState *s;
   2684 
   2685     s = migrate_get_current();
   2686 
   2687     return s->enabled_capabilities[MIGRATION_CAPABILITY_XBZRLE];
   2688 }
   2689 
   2690 uint64_t migrate_xbzrle_cache_size(void)
   2691 {
   2692     MigrationState *s;
   2693 
   2694     s = migrate_get_current();
   2695 
   2696     return s->parameters.xbzrle_cache_size;
   2697 }
   2698 
   2699 static int64_t migrate_max_postcopy_bandwidth(void)
   2700 {
   2701     MigrationState *s;
   2702 
   2703     s = migrate_get_current();
   2704 
   2705     return s->parameters.max_postcopy_bandwidth;
   2706 }
   2707 
   2708 bool migrate_use_block(void)
   2709 {
   2710     MigrationState *s;
   2711 
   2712     s = migrate_get_current();
   2713 
   2714     return s->enabled_capabilities[MIGRATION_CAPABILITY_BLOCK];
   2715 }
   2716 
   2717 bool migrate_use_return_path(void)
   2718 {
   2719     MigrationState *s;
   2720 
   2721     s = migrate_get_current();
   2722 
   2723     return s->enabled_capabilities[MIGRATION_CAPABILITY_RETURN_PATH];
   2724 }
   2725 
   2726 bool migrate_use_block_incremental(void)
   2727 {
   2728     MigrationState *s;
   2729 
   2730     s = migrate_get_current();
   2731 
   2732     return s->parameters.block_incremental;
   2733 }
   2734 
   2735 bool migrate_background_snapshot(void)
   2736 {
   2737     MigrationState *s;
   2738 
   2739     s = migrate_get_current();
   2740 
   2741     return s->enabled_capabilities[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT];
   2742 }
   2743 
   2744 bool migrate_postcopy_preempt(void)
   2745 {
   2746     MigrationState *s;
   2747 
   2748     s = migrate_get_current();
   2749 
   2750     return s->enabled_capabilities[MIGRATION_CAPABILITY_POSTCOPY_PREEMPT];
   2751 }
   2752 
   2753 /* migration thread support */
   2754 /*
   2755  * Something bad happened to the RP stream, mark an error
   2756  * The caller shall print or trace something to indicate why
   2757  */
   2758 static void mark_source_rp_bad(MigrationState *s)
   2759 {
   2760     s->rp_state.error = true;
   2761 }
   2762 
   2763 static struct rp_cmd_args {
   2764     ssize_t     len; /* -1 = variable */
   2765     const char *name;
   2766 } rp_cmd_args[] = {
   2767     [MIG_RP_MSG_INVALID]        = { .len = -1, .name = "INVALID" },
   2768     [MIG_RP_MSG_SHUT]           = { .len =  4, .name = "SHUT" },
   2769     [MIG_RP_MSG_PONG]           = { .len =  4, .name = "PONG" },
   2770     [MIG_RP_MSG_REQ_PAGES]      = { .len = 12, .name = "REQ_PAGES" },
   2771     [MIG_RP_MSG_REQ_PAGES_ID]   = { .len = -1, .name = "REQ_PAGES_ID" },
   2772     [MIG_RP_MSG_RECV_BITMAP]    = { .len = -1, .name = "RECV_BITMAP" },
   2773     [MIG_RP_MSG_RESUME_ACK]     = { .len =  4, .name = "RESUME_ACK" },
   2774     [MIG_RP_MSG_MAX]            = { .len = -1, .name = "MAX" },
   2775 };
   2776 
   2777 /*
   2778  * Process a request for pages received on the return path,
   2779  * We're allowed to send more than requested (e.g. to round to our page size)
   2780  * and we don't need to send pages that have already been sent.
   2781  */
   2782 static void migrate_handle_rp_req_pages(MigrationState *ms, const char* rbname,
   2783                                        ram_addr_t start, size_t len)
   2784 {
   2785     long our_host_ps = qemu_real_host_page_size();
   2786 
   2787     trace_migrate_handle_rp_req_pages(rbname, start, len);
   2788 
   2789     /*
   2790      * Since we currently insist on matching page sizes, just sanity check
   2791      * we're being asked for whole host pages.
   2792      */
   2793     if (!QEMU_IS_ALIGNED(start, our_host_ps) ||
   2794         !QEMU_IS_ALIGNED(len, our_host_ps)) {
   2795         error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
   2796                      " len: %zd", __func__, start, len);
   2797         mark_source_rp_bad(ms);
   2798         return;
   2799     }
   2800 
   2801     if (ram_save_queue_pages(rbname, start, len)) {
   2802         mark_source_rp_bad(ms);
   2803     }
   2804 }
   2805 
   2806 /* Return true to retry, false to quit */
   2807 static bool postcopy_pause_return_path_thread(MigrationState *s)
   2808 {
   2809     trace_postcopy_pause_return_path();
   2810 
   2811     qemu_sem_wait(&s->postcopy_pause_rp_sem);
   2812 
   2813     trace_postcopy_pause_return_path_continued();
   2814 
   2815     return true;
   2816 }
   2817 
   2818 static int migrate_handle_rp_recv_bitmap(MigrationState *s, char *block_name)
   2819 {
   2820     RAMBlock *block = qemu_ram_block_by_name(block_name);
   2821 
   2822     if (!block) {
   2823         error_report("%s: invalid block name '%s'", __func__, block_name);
   2824         return -EINVAL;
   2825     }
   2826 
   2827     /* Fetch the received bitmap and refresh the dirty bitmap */
   2828     return ram_dirty_bitmap_reload(s, block);
   2829 }
   2830 
   2831 static int migrate_handle_rp_resume_ack(MigrationState *s, uint32_t value)
   2832 {
   2833     trace_source_return_path_thread_resume_ack(value);
   2834 
   2835     if (value != MIGRATION_RESUME_ACK_VALUE) {
   2836         error_report("%s: illegal resume_ack value %"PRIu32,
   2837                      __func__, value);
   2838         return -1;
   2839     }
   2840 
   2841     /* Now both sides are active. */
   2842     migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_RECOVER,
   2843                       MIGRATION_STATUS_POSTCOPY_ACTIVE);
   2844 
   2845     /* Notify send thread that time to continue send pages */
   2846     qemu_sem_post(&s->rp_state.rp_sem);
   2847 
   2848     return 0;
   2849 }
   2850 
   2851 /* Release ms->rp_state.from_dst_file in a safe way */
   2852 static void migration_release_from_dst_file(MigrationState *ms)
   2853 {
   2854     QEMUFile *file;
   2855 
   2856     WITH_QEMU_LOCK_GUARD(&ms->qemu_file_lock) {
   2857         /*
   2858          * Reset the from_dst_file pointer first before releasing it, as we
   2859          * can't block within lock section
   2860          */
   2861         file = ms->rp_state.from_dst_file;
   2862         ms->rp_state.from_dst_file = NULL;
   2863     }
   2864 
   2865     qemu_fclose(file);
   2866 }
   2867 
   2868 /*
   2869  * Handles messages sent on the return path towards the source VM
   2870  *
   2871  */
   2872 static void *source_return_path_thread(void *opaque)
   2873 {
   2874     MigrationState *ms = opaque;
   2875     QEMUFile *rp = ms->rp_state.from_dst_file;
   2876     uint16_t header_len, header_type;
   2877     uint8_t buf[512];
   2878     uint32_t tmp32, sibling_error;
   2879     ram_addr_t start = 0; /* =0 to silence warning */
   2880     size_t  len = 0, expected_len;
   2881     int res;
   2882 
   2883     trace_source_return_path_thread_entry();
   2884     rcu_register_thread();
   2885 
   2886 retry:
   2887     while (!ms->rp_state.error && !qemu_file_get_error(rp) &&
   2888            migration_is_setup_or_active(ms->state)) {
   2889         trace_source_return_path_thread_loop_top();
   2890         header_type = qemu_get_be16(rp);
   2891         header_len = qemu_get_be16(rp);
   2892 
   2893         if (qemu_file_get_error(rp)) {
   2894             mark_source_rp_bad(ms);
   2895             goto out;
   2896         }
   2897 
   2898         if (header_type >= MIG_RP_MSG_MAX ||
   2899             header_type == MIG_RP_MSG_INVALID) {
   2900             error_report("RP: Received invalid message 0x%04x length 0x%04x",
   2901                          header_type, header_len);
   2902             mark_source_rp_bad(ms);
   2903             goto out;
   2904         }
   2905 
   2906         if ((rp_cmd_args[header_type].len != -1 &&
   2907             header_len != rp_cmd_args[header_type].len) ||
   2908             header_len > sizeof(buf)) {
   2909             error_report("RP: Received '%s' message (0x%04x) with"
   2910                          "incorrect length %d expecting %zu",
   2911                          rp_cmd_args[header_type].name, header_type, header_len,
   2912                          (size_t)rp_cmd_args[header_type].len);
   2913             mark_source_rp_bad(ms);
   2914             goto out;
   2915         }
   2916 
   2917         /* We know we've got a valid header by this point */
   2918         res = qemu_get_buffer(rp, buf, header_len);
   2919         if (res != header_len) {
   2920             error_report("RP: Failed reading data for message 0x%04x"
   2921                          " read %d expected %d",
   2922                          header_type, res, header_len);
   2923             mark_source_rp_bad(ms);
   2924             goto out;
   2925         }
   2926 
   2927         /* OK, we have the message and the data */
   2928         switch (header_type) {
   2929         case MIG_RP_MSG_SHUT:
   2930             sibling_error = ldl_be_p(buf);
   2931             trace_source_return_path_thread_shut(sibling_error);
   2932             if (sibling_error) {
   2933                 error_report("RP: Sibling indicated error %d", sibling_error);
   2934                 mark_source_rp_bad(ms);
   2935             }
   2936             /*
   2937              * We'll let the main thread deal with closing the RP
   2938              * we could do a shutdown(2) on it, but we're the only user
   2939              * anyway, so there's nothing gained.
   2940              */
   2941             goto out;
   2942 
   2943         case MIG_RP_MSG_PONG:
   2944             tmp32 = ldl_be_p(buf);
   2945             trace_source_return_path_thread_pong(tmp32);
   2946             break;
   2947 
   2948         case MIG_RP_MSG_REQ_PAGES:
   2949             start = ldq_be_p(buf);
   2950             len = ldl_be_p(buf + 8);
   2951             migrate_handle_rp_req_pages(ms, NULL, start, len);
   2952             break;
   2953 
   2954         case MIG_RP_MSG_REQ_PAGES_ID:
   2955             expected_len = 12 + 1; /* header + termination */
   2956 
   2957             if (header_len >= expected_len) {
   2958                 start = ldq_be_p(buf);
   2959                 len = ldl_be_p(buf + 8);
   2960                 /* Now we expect an idstr */
   2961                 tmp32 = buf[12]; /* Length of the following idstr */
   2962                 buf[13 + tmp32] = '\0';
   2963                 expected_len += tmp32;
   2964             }
   2965             if (header_len != expected_len) {
   2966                 error_report("RP: Req_Page_id with length %d expecting %zd",
   2967                              header_len, expected_len);
   2968                 mark_source_rp_bad(ms);
   2969                 goto out;
   2970             }
   2971             migrate_handle_rp_req_pages(ms, (char *)&buf[13], start, len);
   2972             break;
   2973 
   2974         case MIG_RP_MSG_RECV_BITMAP:
   2975             if (header_len < 1) {
   2976                 error_report("%s: missing block name", __func__);
   2977                 mark_source_rp_bad(ms);
   2978                 goto out;
   2979             }
   2980             /* Format: len (1B) + idstr (<255B). This ends the idstr. */
   2981             buf[buf[0] + 1] = '\0';
   2982             if (migrate_handle_rp_recv_bitmap(ms, (char *)(buf + 1))) {
   2983                 mark_source_rp_bad(ms);
   2984                 goto out;
   2985             }
   2986             break;
   2987 
   2988         case MIG_RP_MSG_RESUME_ACK:
   2989             tmp32 = ldl_be_p(buf);
   2990             if (migrate_handle_rp_resume_ack(ms, tmp32)) {
   2991                 mark_source_rp_bad(ms);
   2992                 goto out;
   2993             }
   2994             break;
   2995 
   2996         default:
   2997             break;
   2998         }
   2999     }
   3000 
   3001 out:
   3002     res = qemu_file_get_error(rp);
   3003     if (res) {
   3004         if (res && migration_in_postcopy()) {
   3005             /*
   3006              * Maybe there is something we can do: it looks like a
   3007              * network down issue, and we pause for a recovery.
   3008              */
   3009             migration_release_from_dst_file(ms);
   3010             rp = NULL;
   3011             if (postcopy_pause_return_path_thread(ms)) {
   3012                 /*
   3013                  * Reload rp, reset the rest.  Referencing it is safe since
   3014                  * it's reset only by us above, or when migration completes
   3015                  */
   3016                 rp = ms->rp_state.from_dst_file;
   3017                 ms->rp_state.error = false;
   3018                 goto retry;
   3019             }
   3020         }
   3021 
   3022         trace_source_return_path_thread_bad_end();
   3023         mark_source_rp_bad(ms);
   3024     }
   3025 
   3026     trace_source_return_path_thread_end();
   3027     migration_release_from_dst_file(ms);
   3028     rcu_unregister_thread();
   3029     return NULL;
   3030 }
   3031 
   3032 static int open_return_path_on_source(MigrationState *ms,
   3033                                       bool create_thread)
   3034 {
   3035     ms->rp_state.from_dst_file = qemu_file_get_return_path(ms->to_dst_file);
   3036     if (!ms->rp_state.from_dst_file) {
   3037         return -1;
   3038     }
   3039 
   3040     trace_open_return_path_on_source();
   3041 
   3042     if (!create_thread) {
   3043         /* We're done */
   3044         return 0;
   3045     }
   3046 
   3047     qemu_thread_create(&ms->rp_state.rp_thread, "return path",
   3048                        source_return_path_thread, ms, QEMU_THREAD_JOINABLE);
   3049     ms->rp_state.rp_thread_created = true;
   3050 
   3051     trace_open_return_path_on_source_continue();
   3052 
   3053     return 0;
   3054 }
   3055 
   3056 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
   3057 static int await_return_path_close_on_source(MigrationState *ms)
   3058 {
   3059     /*
   3060      * If this is a normal exit then the destination will send a SHUT and the
   3061      * rp_thread will exit, however if there's an error we need to cause
   3062      * it to exit.
   3063      */
   3064     if (qemu_file_get_error(ms->to_dst_file) && ms->rp_state.from_dst_file) {
   3065         /*
   3066          * shutdown(2), if we have it, will cause it to unblock if it's stuck
   3067          * waiting for the destination.
   3068          */
   3069         qemu_file_shutdown(ms->rp_state.from_dst_file);
   3070         mark_source_rp_bad(ms);
   3071     }
   3072     trace_await_return_path_close_on_source_joining();
   3073     qemu_thread_join(&ms->rp_state.rp_thread);
   3074     ms->rp_state.rp_thread_created = false;
   3075     trace_await_return_path_close_on_source_close();
   3076     return ms->rp_state.error;
   3077 }
   3078 
   3079 /*
   3080  * Switch from normal iteration to postcopy
   3081  * Returns non-0 on error
   3082  */
   3083 static int postcopy_start(MigrationState *ms)
   3084 {
   3085     int ret;
   3086     QIOChannelBuffer *bioc;
   3087     QEMUFile *fb;
   3088     int64_t time_at_stop = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
   3089     int64_t bandwidth = migrate_max_postcopy_bandwidth();
   3090     bool restart_block = false;
   3091     int cur_state = MIGRATION_STATUS_ACTIVE;
   3092 
   3093     if (postcopy_preempt_wait_channel(ms)) {
   3094         migrate_set_state(&ms->state, ms->state, MIGRATION_STATUS_FAILED);
   3095         return -1;
   3096     }
   3097 
   3098     if (!migrate_pause_before_switchover()) {
   3099         migrate_set_state(&ms->state, MIGRATION_STATUS_ACTIVE,
   3100                           MIGRATION_STATUS_POSTCOPY_ACTIVE);
   3101     }
   3102 
   3103     trace_postcopy_start();
   3104     qemu_mutex_lock_iothread();
   3105     trace_postcopy_start_set_run();
   3106 
   3107     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
   3108     global_state_store();
   3109     ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
   3110     if (ret < 0) {
   3111         goto fail;
   3112     }
   3113 
   3114     ret = migration_maybe_pause(ms, &cur_state,
   3115                                 MIGRATION_STATUS_POSTCOPY_ACTIVE);
   3116     if (ret < 0) {
   3117         goto fail;
   3118     }
   3119 
   3120     ret = bdrv_inactivate_all();
   3121     if (ret < 0) {
   3122         goto fail;
   3123     }
   3124     restart_block = true;
   3125 
   3126     /*
   3127      * Cause any non-postcopiable, but iterative devices to
   3128      * send out their final data.
   3129      */
   3130     qemu_savevm_state_complete_precopy(ms->to_dst_file, true, false);
   3131 
   3132     /*
   3133      * in Finish migrate and with the io-lock held everything should
   3134      * be quiet, but we've potentially still got dirty pages and we
   3135      * need to tell the destination to throw any pages it's already received
   3136      * that are dirty
   3137      */
   3138     if (migrate_postcopy_ram()) {
   3139         ram_postcopy_send_discard_bitmap(ms);
   3140     }
   3141 
   3142     /*
   3143      * send rest of state - note things that are doing postcopy
   3144      * will notice we're in POSTCOPY_ACTIVE and not actually
   3145      * wrap their state up here
   3146      */
   3147     /* 0 max-postcopy-bandwidth means unlimited */
   3148     if (!bandwidth) {
   3149         qemu_file_set_rate_limit(ms->to_dst_file, INT64_MAX);
   3150     } else {
   3151         qemu_file_set_rate_limit(ms->to_dst_file, bandwidth / XFER_LIMIT_RATIO);
   3152     }
   3153     if (migrate_postcopy_ram()) {
   3154         /* Ping just for debugging, helps line traces up */
   3155         qemu_savevm_send_ping(ms->to_dst_file, 2);
   3156     }
   3157 
   3158     /*
   3159      * While loading the device state we may trigger page transfer
   3160      * requests and the fd must be free to process those, and thus
   3161      * the destination must read the whole device state off the fd before
   3162      * it starts processing it.  Unfortunately the ad-hoc migration format
   3163      * doesn't allow the destination to know the size to read without fully
   3164      * parsing it through each devices load-state code (especially the open
   3165      * coded devices that use get/put).
   3166      * So we wrap the device state up in a package with a length at the start;
   3167      * to do this we use a qemu_buf to hold the whole of the device state.
   3168      */
   3169     bioc = qio_channel_buffer_new(4096);
   3170     qio_channel_set_name(QIO_CHANNEL(bioc), "migration-postcopy-buffer");
   3171     fb = qemu_file_new_output(QIO_CHANNEL(bioc));
   3172     object_unref(OBJECT(bioc));
   3173 
   3174     /*
   3175      * Make sure the receiver can get incoming pages before we send the rest
   3176      * of the state
   3177      */
   3178     qemu_savevm_send_postcopy_listen(fb);
   3179 
   3180     qemu_savevm_state_complete_precopy(fb, false, false);
   3181     if (migrate_postcopy_ram()) {
   3182         qemu_savevm_send_ping(fb, 3);
   3183     }
   3184 
   3185     qemu_savevm_send_postcopy_run(fb);
   3186 
   3187     /* <><> end of stuff going into the package */
   3188 
   3189     /* Last point of recovery; as soon as we send the package the destination
   3190      * can open devices and potentially start running.
   3191      * Lets just check again we've not got any errors.
   3192      */
   3193     ret = qemu_file_get_error(ms->to_dst_file);
   3194     if (ret) {
   3195         error_report("postcopy_start: Migration stream errored (pre package)");
   3196         goto fail_closefb;
   3197     }
   3198 
   3199     restart_block = false;
   3200 
   3201     /* Now send that blob */
   3202     if (qemu_savevm_send_packaged(ms->to_dst_file, bioc->data, bioc->usage)) {
   3203         goto fail_closefb;
   3204     }
   3205     qemu_fclose(fb);
   3206 
   3207     /* Send a notify to give a chance for anything that needs to happen
   3208      * at the transition to postcopy and after the device state; in particular
   3209      * spice needs to trigger a transition now
   3210      */
   3211     ms->postcopy_after_devices = true;
   3212     notifier_list_notify(&migration_state_notifiers, ms);
   3213 
   3214     ms->downtime =  qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - time_at_stop;
   3215 
   3216     qemu_mutex_unlock_iothread();
   3217 
   3218     if (migrate_postcopy_ram()) {
   3219         /*
   3220          * Although this ping is just for debug, it could potentially be
   3221          * used for getting a better measurement of downtime at the source.
   3222          */
   3223         qemu_savevm_send_ping(ms->to_dst_file, 4);
   3224     }
   3225 
   3226     if (migrate_release_ram()) {
   3227         ram_postcopy_migrated_memory_release(ms);
   3228     }
   3229 
   3230     ret = qemu_file_get_error(ms->to_dst_file);
   3231     if (ret) {
   3232         error_report("postcopy_start: Migration stream errored");
   3233         migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
   3234                               MIGRATION_STATUS_FAILED);
   3235     }
   3236 
   3237     trace_postcopy_preempt_enabled(migrate_postcopy_preempt());
   3238 
   3239     return ret;
   3240 
   3241 fail_closefb:
   3242     qemu_fclose(fb);
   3243 fail:
   3244     migrate_set_state(&ms->state, MIGRATION_STATUS_POSTCOPY_ACTIVE,
   3245                           MIGRATION_STATUS_FAILED);
   3246     if (restart_block) {
   3247         /* A failure happened early enough that we know the destination hasn't
   3248          * accessed block devices, so we're safe to recover.
   3249          */
   3250         Error *local_err = NULL;
   3251 
   3252         bdrv_activate_all(&local_err);
   3253         if (local_err) {
   3254             error_report_err(local_err);
   3255         }
   3256     }
   3257     qemu_mutex_unlock_iothread();
   3258     return -1;
   3259 }
   3260 
   3261 /**
   3262  * migration_maybe_pause: Pause if required to by
   3263  * migrate_pause_before_switchover called with the iothread locked
   3264  * Returns: 0 on success
   3265  */
   3266 static int migration_maybe_pause(MigrationState *s,
   3267                                  int *current_active_state,
   3268                                  int new_state)
   3269 {
   3270     if (!migrate_pause_before_switchover()) {
   3271         return 0;
   3272     }
   3273 
   3274     /* Since leaving this state is not atomic with posting the semaphore
   3275      * it's possible that someone could have issued multiple migrate_continue
   3276      * and the semaphore is incorrectly positive at this point;
   3277      * the docs say it's undefined to reinit a semaphore that's already
   3278      * init'd, so use timedwait to eat up any existing posts.
   3279      */
   3280     while (qemu_sem_timedwait(&s->pause_sem, 1) == 0) {
   3281         /* This block intentionally left blank */
   3282     }
   3283 
   3284     /*
   3285      * If the migration is cancelled when it is in the completion phase,
   3286      * the migration state is set to MIGRATION_STATUS_CANCELLING.
   3287      * So we don't need to wait a semaphore, otherwise we would always
   3288      * wait for the 'pause_sem' semaphore.
   3289      */
   3290     if (s->state != MIGRATION_STATUS_CANCELLING) {
   3291         qemu_mutex_unlock_iothread();
   3292         migrate_set_state(&s->state, *current_active_state,
   3293                           MIGRATION_STATUS_PRE_SWITCHOVER);
   3294         qemu_sem_wait(&s->pause_sem);
   3295         migrate_set_state(&s->state, MIGRATION_STATUS_PRE_SWITCHOVER,
   3296                           new_state);
   3297         *current_active_state = new_state;
   3298         qemu_mutex_lock_iothread();
   3299     }
   3300 
   3301     return s->state == new_state ? 0 : -EINVAL;
   3302 }
   3303 
   3304 /**
   3305  * migration_completion: Used by migration_thread when there's not much left.
   3306  *   The caller 'breaks' the loop when this returns.
   3307  *
   3308  * @s: Current migration state
   3309  */
   3310 static void migration_completion(MigrationState *s)
   3311 {
   3312     int ret;
   3313     int current_active_state = s->state;
   3314 
   3315     if (s->state == MIGRATION_STATUS_ACTIVE) {
   3316         qemu_mutex_lock_iothread();
   3317         s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
   3318         qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
   3319         s->vm_was_running = runstate_is_running();
   3320         ret = global_state_store();
   3321 
   3322         if (!ret) {
   3323             bool inactivate = !migrate_colo_enabled();
   3324             ret = vm_stop_force_state(RUN_STATE_FINISH_MIGRATE);
   3325             trace_migration_completion_vm_stop(ret);
   3326             if (ret >= 0) {
   3327                 ret = migration_maybe_pause(s, &current_active_state,
   3328                                             MIGRATION_STATUS_DEVICE);
   3329             }
   3330             if (ret >= 0) {
   3331                 qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
   3332                 ret = qemu_savevm_state_complete_precopy(s->to_dst_file, false,
   3333                                                          inactivate);
   3334             }
   3335             if (inactivate && ret >= 0) {
   3336                 s->block_inactive = true;
   3337             }
   3338         }
   3339         qemu_mutex_unlock_iothread();
   3340 
   3341         if (ret < 0) {
   3342             goto fail;
   3343         }
   3344     } else if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
   3345         trace_migration_completion_postcopy_end();
   3346 
   3347         qemu_mutex_lock_iothread();
   3348         qemu_savevm_state_complete_postcopy(s->to_dst_file);
   3349         qemu_mutex_unlock_iothread();
   3350 
   3351         /* Shutdown the postcopy fast path thread */
   3352         if (migrate_postcopy_preempt()) {
   3353             postcopy_preempt_shutdown_file(s);
   3354         }
   3355 
   3356         trace_migration_completion_postcopy_end_after_complete();
   3357     } else {
   3358         goto fail;
   3359     }
   3360 
   3361     /*
   3362      * If rp was opened we must clean up the thread before
   3363      * cleaning everything else up (since if there are no failures
   3364      * it will wait for the destination to send it's status in
   3365      * a SHUT command).
   3366      */
   3367     if (s->rp_state.rp_thread_created) {
   3368         int rp_error;
   3369         trace_migration_return_path_end_before();
   3370         rp_error = await_return_path_close_on_source(s);
   3371         trace_migration_return_path_end_after(rp_error);
   3372         if (rp_error) {
   3373             goto fail_invalidate;
   3374         }
   3375     }
   3376 
   3377     if (qemu_file_get_error(s->to_dst_file)) {
   3378         trace_migration_completion_file_err();
   3379         goto fail_invalidate;
   3380     }
   3381 
   3382     if (migrate_colo_enabled() && s->state == MIGRATION_STATUS_ACTIVE) {
   3383         /* COLO does not support postcopy */
   3384         migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE,
   3385                           MIGRATION_STATUS_COLO);
   3386     } else {
   3387         migrate_set_state(&s->state, current_active_state,
   3388                           MIGRATION_STATUS_COMPLETED);
   3389     }
   3390 
   3391     return;
   3392 
   3393 fail_invalidate:
   3394     /* If not doing postcopy, vm_start() will be called: let's regain
   3395      * control on images.
   3396      */
   3397     if (s->state == MIGRATION_STATUS_ACTIVE ||
   3398         s->state == MIGRATION_STATUS_DEVICE) {
   3399         Error *local_err = NULL;
   3400 
   3401         qemu_mutex_lock_iothread();
   3402         bdrv_activate_all(&local_err);
   3403         if (local_err) {
   3404             error_report_err(local_err);
   3405         } else {
   3406             s->block_inactive = false;
   3407         }
   3408         qemu_mutex_unlock_iothread();
   3409     }
   3410 
   3411 fail:
   3412     migrate_set_state(&s->state, current_active_state,
   3413                       MIGRATION_STATUS_FAILED);
   3414 }
   3415 
   3416 /**
   3417  * bg_migration_completion: Used by bg_migration_thread when after all the
   3418  *   RAM has been saved. The caller 'breaks' the loop when this returns.
   3419  *
   3420  * @s: Current migration state
   3421  */
   3422 static void bg_migration_completion(MigrationState *s)
   3423 {
   3424     int current_active_state = s->state;
   3425 
   3426     /*
   3427      * Stop tracking RAM writes - un-protect memory, un-register UFFD
   3428      * memory ranges, flush kernel wait queues and wake up threads
   3429      * waiting for write fault to be resolved.
   3430      */
   3431     ram_write_tracking_stop();
   3432 
   3433     if (s->state == MIGRATION_STATUS_ACTIVE) {
   3434         /*
   3435          * By this moment we have RAM content saved into the migration stream.
   3436          * The next step is to flush the non-RAM content (device state)
   3437          * right after the ram content. The device state has been stored into
   3438          * the temporary buffer before RAM saving started.
   3439          */
   3440         qemu_put_buffer(s->to_dst_file, s->bioc->data, s->bioc->usage);
   3441         qemu_fflush(s->to_dst_file);
   3442     } else if (s->state == MIGRATION_STATUS_CANCELLING) {
   3443         goto fail;
   3444     }
   3445 
   3446     if (qemu_file_get_error(s->to_dst_file)) {
   3447         trace_migration_completion_file_err();
   3448         goto fail;
   3449     }
   3450 
   3451     migrate_set_state(&s->state, current_active_state,
   3452                       MIGRATION_STATUS_COMPLETED);
   3453     return;
   3454 
   3455 fail:
   3456     migrate_set_state(&s->state, current_active_state,
   3457                       MIGRATION_STATUS_FAILED);
   3458 }
   3459 
   3460 bool migrate_colo_enabled(void)
   3461 {
   3462     MigrationState *s = migrate_get_current();
   3463     return s->enabled_capabilities[MIGRATION_CAPABILITY_X_COLO];
   3464 }
   3465 
   3466 typedef enum MigThrError {
   3467     /* No error detected */
   3468     MIG_THR_ERR_NONE = 0,
   3469     /* Detected error, but resumed successfully */
   3470     MIG_THR_ERR_RECOVERED = 1,
   3471     /* Detected fatal error, need to exit */
   3472     MIG_THR_ERR_FATAL = 2,
   3473 } MigThrError;
   3474 
   3475 static int postcopy_resume_handshake(MigrationState *s)
   3476 {
   3477     qemu_savevm_send_postcopy_resume(s->to_dst_file);
   3478 
   3479     while (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
   3480         qemu_sem_wait(&s->rp_state.rp_sem);
   3481     }
   3482 
   3483     if (s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE) {
   3484         return 0;
   3485     }
   3486 
   3487     return -1;
   3488 }
   3489 
   3490 /* Return zero if success, or <0 for error */
   3491 static int postcopy_do_resume(MigrationState *s)
   3492 {
   3493     int ret;
   3494 
   3495     /*
   3496      * Call all the resume_prepare() hooks, so that modules can be
   3497      * ready for the migration resume.
   3498      */
   3499     ret = qemu_savevm_state_resume_prepare(s);
   3500     if (ret) {
   3501         error_report("%s: resume_prepare() failure detected: %d",
   3502                      __func__, ret);
   3503         return ret;
   3504     }
   3505 
   3506     /*
   3507      * Last handshake with destination on the resume (destination will
   3508      * switch to postcopy-active afterwards)
   3509      */
   3510     ret = postcopy_resume_handshake(s);
   3511     if (ret) {
   3512         error_report("%s: handshake failed: %d", __func__, ret);
   3513         return ret;
   3514     }
   3515 
   3516     return 0;
   3517 }
   3518 
   3519 /*
   3520  * We don't return until we are in a safe state to continue current
   3521  * postcopy migration.  Returns MIG_THR_ERR_RECOVERED if recovered, or
   3522  * MIG_THR_ERR_FATAL if unrecovery failure happened.
   3523  */
   3524 static MigThrError postcopy_pause(MigrationState *s)
   3525 {
   3526     assert(s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE);
   3527 
   3528     while (true) {
   3529         QEMUFile *file;
   3530 
   3531         /*
   3532          * Current channel is possibly broken. Release it.  Note that this is
   3533          * guaranteed even without lock because to_dst_file should only be
   3534          * modified by the migration thread.  That also guarantees that the
   3535          * unregister of yank is safe too without the lock.  It should be safe
   3536          * even to be within the qemu_file_lock, but we didn't do that to avoid
   3537          * taking more mutex (yank_lock) within qemu_file_lock.  TL;DR: we make
   3538          * the qemu_file_lock critical section as small as possible.
   3539          */
   3540         assert(s->to_dst_file);
   3541         migration_ioc_unregister_yank_from_file(s->to_dst_file);
   3542         qemu_mutex_lock(&s->qemu_file_lock);
   3543         file = s->to_dst_file;
   3544         s->to_dst_file = NULL;
   3545         qemu_mutex_unlock(&s->qemu_file_lock);
   3546 
   3547         qemu_file_shutdown(file);
   3548         qemu_fclose(file);
   3549 
   3550         /*
   3551          * Do the same to postcopy fast path socket too if there is.  No
   3552          * locking needed because no racer as long as we do this before setting
   3553          * status to paused.
   3554          */
   3555         if (s->postcopy_qemufile_src) {
   3556             migration_ioc_unregister_yank_from_file(s->postcopy_qemufile_src);
   3557             qemu_file_shutdown(s->postcopy_qemufile_src);
   3558             qemu_fclose(s->postcopy_qemufile_src);
   3559             s->postcopy_qemufile_src = NULL;
   3560         }
   3561 
   3562         migrate_set_state(&s->state, s->state,
   3563                           MIGRATION_STATUS_POSTCOPY_PAUSED);
   3564 
   3565         error_report("Detected IO failure for postcopy. "
   3566                      "Migration paused.");
   3567 
   3568         /*
   3569          * We wait until things fixed up. Then someone will setup the
   3570          * status back for us.
   3571          */
   3572         while (s->state == MIGRATION_STATUS_POSTCOPY_PAUSED) {
   3573             qemu_sem_wait(&s->postcopy_pause_sem);
   3574         }
   3575 
   3576         if (s->state == MIGRATION_STATUS_POSTCOPY_RECOVER) {
   3577             /* Woken up by a recover procedure. Give it a shot */
   3578 
   3579             if (postcopy_preempt_wait_channel(s)) {
   3580                 /*
   3581                  * Preempt enabled, and new channel create failed; loop
   3582                  * back to wait for another recovery.
   3583                  */
   3584                 continue;
   3585             }
   3586 
   3587             /*
   3588              * Firstly, let's wake up the return path now, with a new
   3589              * return path channel.
   3590              */
   3591             qemu_sem_post(&s->postcopy_pause_rp_sem);
   3592 
   3593             /* Do the resume logic */
   3594             if (postcopy_do_resume(s) == 0) {
   3595                 /* Let's continue! */
   3596                 trace_postcopy_pause_continued();
   3597                 return MIG_THR_ERR_RECOVERED;
   3598             } else {
   3599                 /*
   3600                  * Something wrong happened during the recovery, let's
   3601                  * pause again. Pause is always better than throwing
   3602                  * data away.
   3603                  */
   3604                 continue;
   3605             }
   3606         } else {
   3607             /* This is not right... Time to quit. */
   3608             return MIG_THR_ERR_FATAL;
   3609         }
   3610     }
   3611 }
   3612 
   3613 static MigThrError migration_detect_error(MigrationState *s)
   3614 {
   3615     int ret;
   3616     int state = s->state;
   3617     Error *local_error = NULL;
   3618 
   3619     if (state == MIGRATION_STATUS_CANCELLING ||
   3620         state == MIGRATION_STATUS_CANCELLED) {
   3621         /* End the migration, but don't set the state to failed */
   3622         return MIG_THR_ERR_FATAL;
   3623     }
   3624 
   3625     /*
   3626      * Try to detect any file errors.  Note that postcopy_qemufile_src will
   3627      * be NULL when postcopy preempt is not enabled.
   3628      */
   3629     ret = qemu_file_get_error_obj_any(s->to_dst_file,
   3630                                       s->postcopy_qemufile_src,
   3631                                       &local_error);
   3632     if (!ret) {
   3633         /* Everything is fine */
   3634         assert(!local_error);
   3635         return MIG_THR_ERR_NONE;
   3636     }
   3637 
   3638     if (local_error) {
   3639         migrate_set_error(s, local_error);
   3640         error_free(local_error);
   3641     }
   3642 
   3643     if (state == MIGRATION_STATUS_POSTCOPY_ACTIVE && ret) {
   3644         /*
   3645          * For postcopy, we allow the network to be down for a
   3646          * while. After that, it can be continued by a
   3647          * recovery phase.
   3648          */
   3649         return postcopy_pause(s);
   3650     } else {
   3651         /*
   3652          * For precopy (or postcopy with error outside IO), we fail
   3653          * with no time.
   3654          */
   3655         migrate_set_state(&s->state, state, MIGRATION_STATUS_FAILED);
   3656         trace_migration_thread_file_err();
   3657 
   3658         /* Time to stop the migration, now. */
   3659         return MIG_THR_ERR_FATAL;
   3660     }
   3661 }
   3662 
   3663 /* How many bytes have we transferred since the beginning of the migration */
   3664 static uint64_t migration_total_bytes(MigrationState *s)
   3665 {
   3666     return qemu_file_total_transferred(s->to_dst_file) +
   3667         ram_counters.multifd_bytes;
   3668 }
   3669 
   3670 static void migration_calculate_complete(MigrationState *s)
   3671 {
   3672     uint64_t bytes = migration_total_bytes(s);
   3673     int64_t end_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
   3674     int64_t transfer_time;
   3675 
   3676     s->total_time = end_time - s->start_time;
   3677     if (!s->downtime) {
   3678         /*
   3679          * It's still not set, so we are precopy migration.  For
   3680          * postcopy, downtime is calculated during postcopy_start().
   3681          */
   3682         s->downtime = end_time - s->downtime_start;
   3683     }
   3684 
   3685     transfer_time = s->total_time - s->setup_time;
   3686     if (transfer_time) {
   3687         s->mbps = ((double) bytes * 8.0) / transfer_time / 1000;
   3688     }
   3689 }
   3690 
   3691 static void update_iteration_initial_status(MigrationState *s)
   3692 {
   3693     /*
   3694      * Update these three fields at the same time to avoid mismatch info lead
   3695      * wrong speed calculation.
   3696      */
   3697     s->iteration_start_time = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
   3698     s->iteration_initial_bytes = migration_total_bytes(s);
   3699     s->iteration_initial_pages = ram_get_total_transferred_pages();
   3700 }
   3701 
   3702 static void migration_update_counters(MigrationState *s,
   3703                                       int64_t current_time)
   3704 {
   3705     uint64_t transferred, transferred_pages, time_spent;
   3706     uint64_t current_bytes; /* bytes transferred since the beginning */
   3707     double bandwidth;
   3708 
   3709     if (current_time < s->iteration_start_time + BUFFER_DELAY) {
   3710         return;
   3711     }
   3712 
   3713     current_bytes = migration_total_bytes(s);
   3714     transferred = current_bytes - s->iteration_initial_bytes;
   3715     time_spent = current_time - s->iteration_start_time;
   3716     bandwidth = (double)transferred / time_spent;
   3717     s->threshold_size = bandwidth * s->parameters.downtime_limit;
   3718 
   3719     s->mbps = (((double) transferred * 8.0) /
   3720                ((double) time_spent / 1000.0)) / 1000.0 / 1000.0;
   3721 
   3722     transferred_pages = ram_get_total_transferred_pages() -
   3723                             s->iteration_initial_pages;
   3724     s->pages_per_second = (double) transferred_pages /
   3725                              (((double) time_spent / 1000.0));
   3726 
   3727     /*
   3728      * if we haven't sent anything, we don't want to
   3729      * recalculate. 10000 is a small enough number for our purposes
   3730      */
   3731     if (ram_counters.dirty_pages_rate && transferred > 10000) {
   3732         s->expected_downtime = ram_counters.remaining / bandwidth;
   3733     }
   3734 
   3735     qemu_file_reset_rate_limit(s->to_dst_file);
   3736 
   3737     update_iteration_initial_status(s);
   3738 
   3739     trace_migrate_transferred(transferred, time_spent,
   3740                               bandwidth, s->threshold_size);
   3741 }
   3742 
   3743 /* Migration thread iteration status */
   3744 typedef enum {
   3745     MIG_ITERATE_RESUME,         /* Resume current iteration */
   3746     MIG_ITERATE_SKIP,           /* Skip current iteration */
   3747     MIG_ITERATE_BREAK,          /* Break the loop */
   3748 } MigIterateState;
   3749 
   3750 /*
   3751  * Return true if continue to the next iteration directly, false
   3752  * otherwise.
   3753  */
   3754 static MigIterateState migration_iteration_run(MigrationState *s)
   3755 {
   3756     uint64_t pending_size, pend_pre, pend_compat, pend_post;
   3757     bool in_postcopy = s->state == MIGRATION_STATUS_POSTCOPY_ACTIVE;
   3758 
   3759     qemu_savevm_state_pending(s->to_dst_file, s->threshold_size, &pend_pre,
   3760                               &pend_compat, &pend_post);
   3761     pending_size = pend_pre + pend_compat + pend_post;
   3762 
   3763     trace_migrate_pending(pending_size, s->threshold_size,
   3764                           pend_pre, pend_compat, pend_post);
   3765 
   3766     if (pending_size && pending_size >= s->threshold_size) {
   3767         /* Still a significant amount to transfer */
   3768         if (!in_postcopy && pend_pre <= s->threshold_size &&
   3769             qatomic_read(&s->start_postcopy)) {
   3770             if (postcopy_start(s)) {
   3771                 error_report("%s: postcopy failed to start", __func__);
   3772             }
   3773             return MIG_ITERATE_SKIP;
   3774         }
   3775         /* Just another iteration step */
   3776         qemu_savevm_state_iterate(s->to_dst_file, in_postcopy);
   3777     } else {
   3778         trace_migration_thread_low_pending(pending_size);
   3779         migration_completion(s);
   3780         return MIG_ITERATE_BREAK;
   3781     }
   3782 
   3783     return MIG_ITERATE_RESUME;
   3784 }
   3785 
   3786 static void migration_iteration_finish(MigrationState *s)
   3787 {
   3788     /* If we enabled cpu throttling for auto-converge, turn it off. */
   3789     cpu_throttle_stop();
   3790 
   3791     qemu_mutex_lock_iothread();
   3792     switch (s->state) {
   3793     case MIGRATION_STATUS_COMPLETED:
   3794         migration_calculate_complete(s);
   3795         runstate_set(RUN_STATE_POSTMIGRATE);
   3796         break;
   3797     case MIGRATION_STATUS_COLO:
   3798         if (!migrate_colo_enabled()) {
   3799             error_report("%s: critical error: calling COLO code without "
   3800                          "COLO enabled", __func__);
   3801         }
   3802         migrate_start_colo_process(s);
   3803         s->vm_was_running = true;
   3804         /* Fallthrough */
   3805     case MIGRATION_STATUS_FAILED:
   3806     case MIGRATION_STATUS_CANCELLED:
   3807     case MIGRATION_STATUS_CANCELLING:
   3808         if (s->vm_was_running) {
   3809             if (!runstate_check(RUN_STATE_SHUTDOWN)) {
   3810                 vm_start();
   3811             }
   3812         } else {
   3813             if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
   3814                 runstate_set(RUN_STATE_POSTMIGRATE);
   3815             }
   3816         }
   3817         break;
   3818 
   3819     default:
   3820         /* Should not reach here, but if so, forgive the VM. */
   3821         error_report("%s: Unknown ending state %d", __func__, s->state);
   3822         break;
   3823     }
   3824     migrate_fd_cleanup_schedule(s);
   3825     qemu_mutex_unlock_iothread();
   3826 }
   3827 
   3828 static void bg_migration_iteration_finish(MigrationState *s)
   3829 {
   3830     qemu_mutex_lock_iothread();
   3831     switch (s->state) {
   3832     case MIGRATION_STATUS_COMPLETED:
   3833         migration_calculate_complete(s);
   3834         break;
   3835 
   3836     case MIGRATION_STATUS_ACTIVE:
   3837     case MIGRATION_STATUS_FAILED:
   3838     case MIGRATION_STATUS_CANCELLED:
   3839     case MIGRATION_STATUS_CANCELLING:
   3840         break;
   3841 
   3842     default:
   3843         /* Should not reach here, but if so, forgive the VM. */
   3844         error_report("%s: Unknown ending state %d", __func__, s->state);
   3845         break;
   3846     }
   3847 
   3848     migrate_fd_cleanup_schedule(s);
   3849     qemu_mutex_unlock_iothread();
   3850 }
   3851 
   3852 /*
   3853  * Return true if continue to the next iteration directly, false
   3854  * otherwise.
   3855  */
   3856 static MigIterateState bg_migration_iteration_run(MigrationState *s)
   3857 {
   3858     int res;
   3859 
   3860     res = qemu_savevm_state_iterate(s->to_dst_file, false);
   3861     if (res > 0) {
   3862         bg_migration_completion(s);
   3863         return MIG_ITERATE_BREAK;
   3864     }
   3865 
   3866     return MIG_ITERATE_RESUME;
   3867 }
   3868 
   3869 void migration_make_urgent_request(void)
   3870 {
   3871     qemu_sem_post(&migrate_get_current()->rate_limit_sem);
   3872 }
   3873 
   3874 void migration_consume_urgent_request(void)
   3875 {
   3876     qemu_sem_wait(&migrate_get_current()->rate_limit_sem);
   3877 }
   3878 
   3879 /* Returns true if the rate limiting was broken by an urgent request */
   3880 bool migration_rate_limit(void)
   3881 {
   3882     int64_t now = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
   3883     MigrationState *s = migrate_get_current();
   3884 
   3885     bool urgent = false;
   3886     migration_update_counters(s, now);
   3887     if (qemu_file_rate_limit(s->to_dst_file)) {
   3888 
   3889         if (qemu_file_get_error(s->to_dst_file)) {
   3890             return false;
   3891         }
   3892         /*
   3893          * Wait for a delay to do rate limiting OR
   3894          * something urgent to post the semaphore.
   3895          */
   3896         int ms = s->iteration_start_time + BUFFER_DELAY - now;
   3897         trace_migration_rate_limit_pre(ms);
   3898         if (qemu_sem_timedwait(&s->rate_limit_sem, ms) == 0) {
   3899             /*
   3900              * We were woken by one or more urgent things but
   3901              * the timedwait will have consumed one of them.
   3902              * The service routine for the urgent wake will dec
   3903              * the semaphore itself for each item it consumes,
   3904              * so add this one we just eat back.
   3905              */
   3906             qemu_sem_post(&s->rate_limit_sem);
   3907             urgent = true;
   3908         }
   3909         trace_migration_rate_limit_post(urgent);
   3910     }
   3911     return urgent;
   3912 }
   3913 
   3914 /*
   3915  * if failover devices are present, wait they are completely
   3916  * unplugged
   3917  */
   3918 
   3919 static void qemu_savevm_wait_unplug(MigrationState *s, int old_state,
   3920                                     int new_state)
   3921 {
   3922     if (qemu_savevm_state_guest_unplug_pending()) {
   3923         migrate_set_state(&s->state, old_state, MIGRATION_STATUS_WAIT_UNPLUG);
   3924 
   3925         while (s->state == MIGRATION_STATUS_WAIT_UNPLUG &&
   3926                qemu_savevm_state_guest_unplug_pending()) {
   3927             qemu_sem_timedwait(&s->wait_unplug_sem, 250);
   3928         }
   3929         if (s->state != MIGRATION_STATUS_WAIT_UNPLUG) {
   3930             int timeout = 120; /* 30 seconds */
   3931             /*
   3932              * migration has been canceled
   3933              * but as we have started an unplug we must wait the end
   3934              * to be able to plug back the card
   3935              */
   3936             while (timeout-- && qemu_savevm_state_guest_unplug_pending()) {
   3937                 qemu_sem_timedwait(&s->wait_unplug_sem, 250);
   3938             }
   3939             if (qemu_savevm_state_guest_unplug_pending() &&
   3940                 !qtest_enabled()) {
   3941                 warn_report("migration: partially unplugged device on "
   3942                             "failure");
   3943             }
   3944         }
   3945 
   3946         migrate_set_state(&s->state, MIGRATION_STATUS_WAIT_UNPLUG, new_state);
   3947     } else {
   3948         migrate_set_state(&s->state, old_state, new_state);
   3949     }
   3950 }
   3951 
   3952 /*
   3953  * Master migration thread on the source VM.
   3954  * It drives the migration and pumps the data down the outgoing channel.
   3955  */
   3956 static void *migration_thread(void *opaque)
   3957 {
   3958     MigrationState *s = opaque;
   3959     int64_t setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
   3960     MigThrError thr_error;
   3961     bool urgent = false;
   3962 
   3963     rcu_register_thread();
   3964 
   3965     object_ref(OBJECT(s));
   3966     update_iteration_initial_status(s);
   3967 
   3968     qemu_savevm_state_header(s->to_dst_file);
   3969 
   3970     /*
   3971      * If we opened the return path, we need to make sure dst has it
   3972      * opened as well.
   3973      */
   3974     if (s->rp_state.rp_thread_created) {
   3975         /* Now tell the dest that it should open its end so it can reply */
   3976         qemu_savevm_send_open_return_path(s->to_dst_file);
   3977 
   3978         /* And do a ping that will make stuff easier to debug */
   3979         qemu_savevm_send_ping(s->to_dst_file, 1);
   3980     }
   3981 
   3982     if (migrate_postcopy()) {
   3983         /*
   3984          * Tell the destination that we *might* want to do postcopy later;
   3985          * if the other end can't do postcopy it should fail now, nice and
   3986          * early.
   3987          */
   3988         qemu_savevm_send_postcopy_advise(s->to_dst_file);
   3989     }
   3990 
   3991     if (migrate_colo_enabled()) {
   3992         /* Notify migration destination that we enable COLO */
   3993         qemu_savevm_send_colo_enable(s->to_dst_file);
   3994     }
   3995 
   3996     qemu_savevm_state_setup(s->to_dst_file);
   3997 
   3998     qemu_savevm_wait_unplug(s, MIGRATION_STATUS_SETUP,
   3999                                MIGRATION_STATUS_ACTIVE);
   4000 
   4001     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
   4002 
   4003     trace_migration_thread_setup_complete();
   4004 
   4005     while (migration_is_active(s)) {
   4006         if (urgent || !qemu_file_rate_limit(s->to_dst_file)) {
   4007             MigIterateState iter_state = migration_iteration_run(s);
   4008             if (iter_state == MIG_ITERATE_SKIP) {
   4009                 continue;
   4010             } else if (iter_state == MIG_ITERATE_BREAK) {
   4011                 break;
   4012             }
   4013         }
   4014 
   4015         /*
   4016          * Try to detect any kind of failures, and see whether we
   4017          * should stop the migration now.
   4018          */
   4019         thr_error = migration_detect_error(s);
   4020         if (thr_error == MIG_THR_ERR_FATAL) {
   4021             /* Stop migration */
   4022             break;
   4023         } else if (thr_error == MIG_THR_ERR_RECOVERED) {
   4024             /*
   4025              * Just recovered from a e.g. network failure, reset all
   4026              * the local variables. This is important to avoid
   4027              * breaking transferred_bytes and bandwidth calculation
   4028              */
   4029             update_iteration_initial_status(s);
   4030         }
   4031 
   4032         urgent = migration_rate_limit();
   4033     }
   4034 
   4035     trace_migration_thread_after_loop();
   4036     migration_iteration_finish(s);
   4037     object_unref(OBJECT(s));
   4038     rcu_unregister_thread();
   4039     return NULL;
   4040 }
   4041 
   4042 static void bg_migration_vm_start_bh(void *opaque)
   4043 {
   4044     MigrationState *s = opaque;
   4045 
   4046     qemu_bh_delete(s->vm_start_bh);
   4047     s->vm_start_bh = NULL;
   4048 
   4049     vm_start();
   4050     s->downtime = qemu_clock_get_ms(QEMU_CLOCK_REALTIME) - s->downtime_start;
   4051 }
   4052 
   4053 /**
   4054  * Background snapshot thread, based on live migration code.
   4055  * This is an alternative implementation of live migration mechanism
   4056  * introduced specifically to support background snapshots.
   4057  *
   4058  * It takes advantage of userfault_fd write protection mechanism introduced
   4059  * in v5.7 kernel. Compared to existing dirty page logging migration much
   4060  * lesser stream traffic is produced resulting in smaller snapshot images,
   4061  * simply cause of no page duplicates can get into the stream.
   4062  *
   4063  * Another key point is that generated vmstate stream reflects machine state
   4064  * 'frozen' at the beginning of snapshot creation compared to dirty page logging
   4065  * mechanism, which effectively results in that saved snapshot is the state of VM
   4066  * at the end of the process.
   4067  */
   4068 static void *bg_migration_thread(void *opaque)
   4069 {
   4070     MigrationState *s = opaque;
   4071     int64_t setup_start;
   4072     MigThrError thr_error;
   4073     QEMUFile *fb;
   4074     bool early_fail = true;
   4075 
   4076     rcu_register_thread();
   4077     object_ref(OBJECT(s));
   4078 
   4079     qemu_file_set_rate_limit(s->to_dst_file, INT64_MAX);
   4080 
   4081     setup_start = qemu_clock_get_ms(QEMU_CLOCK_HOST);
   4082     /*
   4083      * We want to save vmstate for the moment when migration has been
   4084      * initiated but also we want to save RAM content while VM is running.
   4085      * The RAM content should appear first in the vmstate. So, we first
   4086      * stash the non-RAM part of the vmstate to the temporary buffer,
   4087      * then write RAM part of the vmstate to the migration stream
   4088      * with vCPUs running and, finally, write stashed non-RAM part of
   4089      * the vmstate from the buffer to the migration stream.
   4090      */
   4091     s->bioc = qio_channel_buffer_new(512 * 1024);
   4092     qio_channel_set_name(QIO_CHANNEL(s->bioc), "vmstate-buffer");
   4093     fb = qemu_file_new_output(QIO_CHANNEL(s->bioc));
   4094     object_unref(OBJECT(s->bioc));
   4095 
   4096     update_iteration_initial_status(s);
   4097 
   4098     /*
   4099      * Prepare for tracking memory writes with UFFD-WP - populate
   4100      * RAM pages before protecting.
   4101      */
   4102 #ifdef __linux__
   4103     ram_write_tracking_prepare();
   4104 #endif
   4105 
   4106     qemu_savevm_state_header(s->to_dst_file);
   4107     qemu_savevm_state_setup(s->to_dst_file);
   4108 
   4109     qemu_savevm_wait_unplug(s, MIGRATION_STATUS_SETUP,
   4110                                MIGRATION_STATUS_ACTIVE);
   4111 
   4112     s->setup_time = qemu_clock_get_ms(QEMU_CLOCK_HOST) - setup_start;
   4113 
   4114     trace_migration_thread_setup_complete();
   4115     s->downtime_start = qemu_clock_get_ms(QEMU_CLOCK_REALTIME);
   4116 
   4117     qemu_mutex_lock_iothread();
   4118 
   4119     /*
   4120      * If VM is currently in suspended state, then, to make a valid runstate
   4121      * transition in vm_stop_force_state() we need to wakeup it up.
   4122      */
   4123     qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER, NULL);
   4124     s->vm_was_running = runstate_is_running();
   4125 
   4126     if (global_state_store()) {
   4127         goto fail;
   4128     }
   4129     /* Forcibly stop VM before saving state of vCPUs and devices */
   4130     if (vm_stop_force_state(RUN_STATE_PAUSED)) {
   4131         goto fail;
   4132     }
   4133     /*
   4134      * Put vCPUs in sync with shadow context structures, then
   4135      * save their state to channel-buffer along with devices.
   4136      */
   4137     cpu_synchronize_all_states();
   4138     if (qemu_savevm_state_complete_precopy_non_iterable(fb, false, false)) {
   4139         goto fail;
   4140     }
   4141     /*
   4142      * Since we are going to get non-iterable state data directly
   4143      * from s->bioc->data, explicit flush is needed here.
   4144      */
   4145     qemu_fflush(fb);
   4146 
   4147     /* Now initialize UFFD context and start tracking RAM writes */
   4148     if (ram_write_tracking_start()) {
   4149         goto fail;
   4150     }
   4151     early_fail = false;
   4152 
   4153     /*
   4154      * Start VM from BH handler to avoid write-fault lock here.
   4155      * UFFD-WP protection for the whole RAM is already enabled so
   4156      * calling VM state change notifiers from vm_start() would initiate
   4157      * writes to virtio VQs memory which is in write-protected region.
   4158      */
   4159     s->vm_start_bh = qemu_bh_new(bg_migration_vm_start_bh, s);
   4160     qemu_bh_schedule(s->vm_start_bh);
   4161 
   4162     qemu_mutex_unlock_iothread();
   4163 
   4164     while (migration_is_active(s)) {
   4165         MigIterateState iter_state = bg_migration_iteration_run(s);
   4166         if (iter_state == MIG_ITERATE_SKIP) {
   4167             continue;
   4168         } else if (iter_state == MIG_ITERATE_BREAK) {
   4169             break;
   4170         }
   4171 
   4172         /*
   4173          * Try to detect any kind of failures, and see whether we
   4174          * should stop the migration now.
   4175          */
   4176         thr_error = migration_detect_error(s);
   4177         if (thr_error == MIG_THR_ERR_FATAL) {
   4178             /* Stop migration */
   4179             break;
   4180         }
   4181 
   4182         migration_update_counters(s, qemu_clock_get_ms(QEMU_CLOCK_REALTIME));
   4183     }
   4184 
   4185     trace_migration_thread_after_loop();
   4186 
   4187 fail:
   4188     if (early_fail) {
   4189         migrate_set_state(&s->state, MIGRATION_STATUS_ACTIVE,
   4190                 MIGRATION_STATUS_FAILED);
   4191         qemu_mutex_unlock_iothread();
   4192     }
   4193 
   4194     bg_migration_iteration_finish(s);
   4195 
   4196     qemu_fclose(fb);
   4197     object_unref(OBJECT(s));
   4198     rcu_unregister_thread();
   4199 
   4200     return NULL;
   4201 }
   4202 
   4203 void migrate_fd_connect(MigrationState *s, Error *error_in)
   4204 {
   4205     Error *local_err = NULL;
   4206     int64_t rate_limit;
   4207     bool resume = s->state == MIGRATION_STATUS_POSTCOPY_PAUSED;
   4208 
   4209     /*
   4210      * If there's a previous error, free it and prepare for another one.
   4211      * Meanwhile if migration completes successfully, there won't have an error
   4212      * dumped when calling migrate_fd_cleanup().
   4213      */
   4214     migrate_error_free(s);
   4215 
   4216     s->expected_downtime = s->parameters.downtime_limit;
   4217     if (resume) {
   4218         assert(s->cleanup_bh);
   4219     } else {
   4220         assert(!s->cleanup_bh);
   4221         s->cleanup_bh = qemu_bh_new(migrate_fd_cleanup_bh, s);
   4222     }
   4223     if (error_in) {
   4224         migrate_fd_error(s, error_in);
   4225         if (resume) {
   4226             /*
   4227              * Don't do cleanup for resume if channel is invalid, but only dump
   4228              * the error.  We wait for another channel connect from the user.
   4229              * The error_report still gives HMP user a hint on what failed.
   4230              * It's normally done in migrate_fd_cleanup(), but call it here
   4231              * explicitly.
   4232              */
   4233             error_report_err(error_copy(s->error));
   4234         } else {
   4235             migrate_fd_cleanup(s);
   4236         }
   4237         return;
   4238     }
   4239 
   4240     if (resume) {
   4241         /* This is a resumed migration */
   4242         rate_limit = s->parameters.max_postcopy_bandwidth /
   4243             XFER_LIMIT_RATIO;
   4244     } else {
   4245         /* This is a fresh new migration */
   4246         rate_limit = s->parameters.max_bandwidth / XFER_LIMIT_RATIO;
   4247 
   4248         /* Notify before starting migration thread */
   4249         notifier_list_notify(&migration_state_notifiers, s);
   4250     }
   4251 
   4252     qemu_file_set_rate_limit(s->to_dst_file, rate_limit);
   4253     qemu_file_set_blocking(s->to_dst_file, true);
   4254 
   4255     /*
   4256      * Open the return path. For postcopy, it is used exclusively. For
   4257      * precopy, only if user specified "return-path" capability would
   4258      * QEMU uses the return path.
   4259      */
   4260     if (migrate_postcopy_ram() || migrate_use_return_path()) {
   4261         if (open_return_path_on_source(s, !resume)) {
   4262             error_report("Unable to open return-path for postcopy");
   4263             migrate_set_state(&s->state, s->state, MIGRATION_STATUS_FAILED);
   4264             migrate_fd_cleanup(s);
   4265             return;
   4266         }
   4267     }
   4268 
   4269     /* This needs to be done before resuming a postcopy */
   4270     if (postcopy_preempt_setup(s, &local_err)) {
   4271         error_report_err(local_err);
   4272         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
   4273                           MIGRATION_STATUS_FAILED);
   4274         migrate_fd_cleanup(s);
   4275         return;
   4276     }
   4277 
   4278     if (resume) {
   4279         /* Wakeup the main migration thread to do the recovery */
   4280         migrate_set_state(&s->state, MIGRATION_STATUS_POSTCOPY_PAUSED,
   4281                           MIGRATION_STATUS_POSTCOPY_RECOVER);
   4282         qemu_sem_post(&s->postcopy_pause_sem);
   4283         return;
   4284     }
   4285 
   4286     if (multifd_save_setup(&local_err) != 0) {
   4287         error_report_err(local_err);
   4288         migrate_set_state(&s->state, MIGRATION_STATUS_SETUP,
   4289                           MIGRATION_STATUS_FAILED);
   4290         migrate_fd_cleanup(s);
   4291         return;
   4292     }
   4293 
   4294     if (migrate_background_snapshot()) {
   4295         qemu_thread_create(&s->thread, "bg_snapshot",
   4296                 bg_migration_thread, s, QEMU_THREAD_JOINABLE);
   4297     } else {
   4298         qemu_thread_create(&s->thread, "live_migration",
   4299                 migration_thread, s, QEMU_THREAD_JOINABLE);
   4300     }
   4301     s->migration_thread_running = true;
   4302 }
   4303 
   4304 void migration_global_dump(Monitor *mon)
   4305 {
   4306     MigrationState *ms = migrate_get_current();
   4307 
   4308     monitor_printf(mon, "globals:\n");
   4309     monitor_printf(mon, "store-global-state: %s\n",
   4310                    ms->store_global_state ? "on" : "off");
   4311     monitor_printf(mon, "only-migratable: %s\n",
   4312                    only_migratable ? "on" : "off");
   4313     monitor_printf(mon, "send-configuration: %s\n",
   4314                    ms->send_configuration ? "on" : "off");
   4315     monitor_printf(mon, "send-section-footer: %s\n",
   4316                    ms->send_section_footer ? "on" : "off");
   4317     monitor_printf(mon, "decompress-error-check: %s\n",
   4318                    ms->decompress_error_check ? "on" : "off");
   4319     monitor_printf(mon, "clear-bitmap-shift: %u\n",
   4320                    ms->clear_bitmap_shift);
   4321 }
   4322 
   4323 #define DEFINE_PROP_MIG_CAP(name, x)             \
   4324     DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
   4325 
   4326 static Property migration_properties[] = {
   4327     DEFINE_PROP_BOOL("store-global-state", MigrationState,
   4328                      store_global_state, true),
   4329     DEFINE_PROP_BOOL("send-configuration", MigrationState,
   4330                      send_configuration, true),
   4331     DEFINE_PROP_BOOL("send-section-footer", MigrationState,
   4332                      send_section_footer, true),
   4333     DEFINE_PROP_BOOL("decompress-error-check", MigrationState,
   4334                       decompress_error_check, true),
   4335     DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState,
   4336                       clear_bitmap_shift, CLEAR_BITMAP_SHIFT_DEFAULT),
   4337 
   4338     /* Migration parameters */
   4339     DEFINE_PROP_UINT8("x-compress-level", MigrationState,
   4340                       parameters.compress_level,
   4341                       DEFAULT_MIGRATE_COMPRESS_LEVEL),
   4342     DEFINE_PROP_UINT8("x-compress-threads", MigrationState,
   4343                       parameters.compress_threads,
   4344                       DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT),
   4345     DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState,
   4346                       parameters.compress_wait_thread, true),
   4347     DEFINE_PROP_UINT8("x-decompress-threads", MigrationState,
   4348                       parameters.decompress_threads,
   4349                       DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT),
   4350     DEFINE_PROP_UINT8("x-throttle-trigger-threshold", MigrationState,
   4351                       parameters.throttle_trigger_threshold,
   4352                       DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD),
   4353     DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState,
   4354                       parameters.cpu_throttle_initial,
   4355                       DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL),
   4356     DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState,
   4357                       parameters.cpu_throttle_increment,
   4358                       DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT),
   4359     DEFINE_PROP_BOOL("x-cpu-throttle-tailslow", MigrationState,
   4360                       parameters.cpu_throttle_tailslow, false),
   4361     DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState,
   4362                       parameters.max_bandwidth, MAX_THROTTLE),
   4363     DEFINE_PROP_UINT64("x-downtime-limit", MigrationState,
   4364                       parameters.downtime_limit,
   4365                       DEFAULT_MIGRATE_SET_DOWNTIME),
   4366     DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState,
   4367                       parameters.x_checkpoint_delay,
   4368                       DEFAULT_MIGRATE_X_CHECKPOINT_DELAY),
   4369     DEFINE_PROP_UINT8("multifd-channels", MigrationState,
   4370                       parameters.multifd_channels,
   4371                       DEFAULT_MIGRATE_MULTIFD_CHANNELS),
   4372     DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState,
   4373                       parameters.multifd_compression,
   4374                       DEFAULT_MIGRATE_MULTIFD_COMPRESSION),
   4375     DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState,
   4376                       parameters.multifd_zlib_level,
   4377                       DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL),
   4378     DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState,
   4379                       parameters.multifd_zstd_level,
   4380                       DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL),
   4381     DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState,
   4382                       parameters.xbzrle_cache_size,
   4383                       DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
   4384     DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState,
   4385                       parameters.max_postcopy_bandwidth,
   4386                       DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH),
   4387     DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState,
   4388                       parameters.max_cpu_throttle,
   4389                       DEFAULT_MIGRATE_MAX_CPU_THROTTLE),
   4390     DEFINE_PROP_SIZE("announce-initial", MigrationState,
   4391                       parameters.announce_initial,
   4392                       DEFAULT_MIGRATE_ANNOUNCE_INITIAL),
   4393     DEFINE_PROP_SIZE("announce-max", MigrationState,
   4394                       parameters.announce_max,
   4395                       DEFAULT_MIGRATE_ANNOUNCE_MAX),
   4396     DEFINE_PROP_SIZE("announce-rounds", MigrationState,
   4397                       parameters.announce_rounds,
   4398                       DEFAULT_MIGRATE_ANNOUNCE_ROUNDS),
   4399     DEFINE_PROP_SIZE("announce-step", MigrationState,
   4400                       parameters.announce_step,
   4401                       DEFAULT_MIGRATE_ANNOUNCE_STEP),
   4402     DEFINE_PROP_BOOL("x-postcopy-preempt-break-huge", MigrationState,
   4403                       postcopy_preempt_break_huge, true),
   4404     DEFINE_PROP_STRING("tls-creds", MigrationState, parameters.tls_creds),
   4405     DEFINE_PROP_STRING("tls-hostname", MigrationState, parameters.tls_hostname),
   4406     DEFINE_PROP_STRING("tls-authz", MigrationState, parameters.tls_authz),
   4407 
   4408     /* Migration capabilities */
   4409     DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE),
   4410     DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL),
   4411     DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE),
   4412     DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS),
   4413     DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS),
   4414     DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS),
   4415     DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM),
   4416     DEFINE_PROP_MIG_CAP("x-postcopy-preempt",
   4417                         MIGRATION_CAPABILITY_POSTCOPY_PREEMPT),
   4418     DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO),
   4419     DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM),
   4420     DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK),
   4421     DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH),
   4422     DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD),
   4423     DEFINE_PROP_MIG_CAP("x-background-snapshot",
   4424             MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT),
   4425 #ifdef CONFIG_LINUX
   4426     DEFINE_PROP_MIG_CAP("x-zero-copy-send",
   4427             MIGRATION_CAPABILITY_ZERO_COPY_SEND),
   4428 #endif
   4429 
   4430     DEFINE_PROP_END_OF_LIST(),
   4431 };
   4432 
   4433 static void migration_class_init(ObjectClass *klass, void *data)
   4434 {
   4435     DeviceClass *dc = DEVICE_CLASS(klass);
   4436 
   4437     dc->user_creatable = false;
   4438     device_class_set_props(dc, migration_properties);
   4439 }
   4440 
   4441 static void migration_instance_finalize(Object *obj)
   4442 {
   4443     MigrationState *ms = MIGRATION_OBJ(obj);
   4444 
   4445     qemu_mutex_destroy(&ms->error_mutex);
   4446     qemu_mutex_destroy(&ms->qemu_file_lock);
   4447     qemu_sem_destroy(&ms->wait_unplug_sem);
   4448     qemu_sem_destroy(&ms->rate_limit_sem);
   4449     qemu_sem_destroy(&ms->pause_sem);
   4450     qemu_sem_destroy(&ms->postcopy_pause_sem);
   4451     qemu_sem_destroy(&ms->postcopy_pause_rp_sem);
   4452     qemu_sem_destroy(&ms->rp_state.rp_sem);
   4453     qemu_sem_destroy(&ms->postcopy_qemufile_src_sem);
   4454     error_free(ms->error);
   4455 }
   4456 
   4457 static void migration_instance_init(Object *obj)
   4458 {
   4459     MigrationState *ms = MIGRATION_OBJ(obj);
   4460     MigrationParameters *params = &ms->parameters;
   4461 
   4462     ms->state = MIGRATION_STATUS_NONE;
   4463     ms->mbps = -1;
   4464     ms->pages_per_second = -1;
   4465     qemu_sem_init(&ms->pause_sem, 0);
   4466     qemu_mutex_init(&ms->error_mutex);
   4467 
   4468     params->tls_hostname = g_strdup("");
   4469     params->tls_creds = g_strdup("");
   4470 
   4471     /* Set has_* up only for parameter checks */
   4472     params->has_compress_level = true;
   4473     params->has_compress_threads = true;
   4474     params->has_compress_wait_thread = true;
   4475     params->has_decompress_threads = true;
   4476     params->has_throttle_trigger_threshold = true;
   4477     params->has_cpu_throttle_initial = true;
   4478     params->has_cpu_throttle_increment = true;
   4479     params->has_cpu_throttle_tailslow = true;
   4480     params->has_max_bandwidth = true;
   4481     params->has_downtime_limit = true;
   4482     params->has_x_checkpoint_delay = true;
   4483     params->has_block_incremental = true;
   4484     params->has_multifd_channels = true;
   4485     params->has_multifd_compression = true;
   4486     params->has_multifd_zlib_level = true;
   4487     params->has_multifd_zstd_level = true;
   4488     params->has_xbzrle_cache_size = true;
   4489     params->has_max_postcopy_bandwidth = true;
   4490     params->has_max_cpu_throttle = true;
   4491     params->has_announce_initial = true;
   4492     params->has_announce_max = true;
   4493     params->has_announce_rounds = true;
   4494     params->has_announce_step = true;
   4495     params->has_tls_creds = true;
   4496     params->has_tls_hostname = true;
   4497     params->has_tls_authz = true;
   4498 
   4499     qemu_sem_init(&ms->postcopy_pause_sem, 0);
   4500     qemu_sem_init(&ms->postcopy_pause_rp_sem, 0);
   4501     qemu_sem_init(&ms->rp_state.rp_sem, 0);
   4502     qemu_sem_init(&ms->rate_limit_sem, 0);
   4503     qemu_sem_init(&ms->wait_unplug_sem, 0);
   4504     qemu_sem_init(&ms->postcopy_qemufile_src_sem, 0);
   4505     qemu_mutex_init(&ms->qemu_file_lock);
   4506 }
   4507 
   4508 /*
   4509  * Return true if check pass, false otherwise. Error will be put
   4510  * inside errp if provided.
   4511  */
   4512 static bool migration_object_check(MigrationState *ms, Error **errp)
   4513 {
   4514     MigrationCapabilityStatusList *head = NULL;
   4515     /* Assuming all off */
   4516     bool cap_list[MIGRATION_CAPABILITY__MAX] = { 0 }, ret;
   4517     int i;
   4518 
   4519     if (!migrate_params_check(&ms->parameters, errp)) {
   4520         return false;
   4521     }
   4522 
   4523     for (i = 0; i < MIGRATION_CAPABILITY__MAX; i++) {
   4524         if (ms->enabled_capabilities[i]) {
   4525             QAPI_LIST_PREPEND(head, migrate_cap_add(i, true));
   4526         }
   4527     }
   4528 
   4529     ret = migrate_caps_check(cap_list, head, errp);
   4530 
   4531     /* It works with head == NULL */
   4532     qapi_free_MigrationCapabilityStatusList(head);
   4533 
   4534     return ret;
   4535 }
   4536 
   4537 static const TypeInfo migration_type = {
   4538     .name = TYPE_MIGRATION,
   4539     /*
   4540      * NOTE: TYPE_MIGRATION is not really a device, as the object is
   4541      * not created using qdev_new(), it is not attached to the qdev
   4542      * device tree, and it is never realized.
   4543      *
   4544      * TODO: Make this TYPE_OBJECT once QOM provides something like
   4545      * TYPE_DEVICE's "-global" properties.
   4546      */
   4547     .parent = TYPE_DEVICE,
   4548     .class_init = migration_class_init,
   4549     .class_size = sizeof(MigrationClass),
   4550     .instance_size = sizeof(MigrationState),
   4551     .instance_init = migration_instance_init,
   4552     .instance_finalize = migration_instance_finalize,
   4553 };
   4554 
   4555 static void register_migration_types(void)
   4556 {
   4557     type_register_static(&migration_type);
   4558 }
   4559 
   4560 type_init(register_migration_types);