4 * Copyright IBM, Corp. 2008
7 * Anthony Liguori <aliguori@us.ibm.com>
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
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.
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/migration.h"
21 #include "migration/qemu-file.h"
22 #include "sysemu/sysemu.h"
23 #include "block/block.h"
24 #include "qapi/qmp/qerror.h"
25 #include "qapi/util.h"
26 #include "qemu/sockets.h"
28 #include "migration/block.h"
29 #include "migration/postcopy-ram.h"
30 #include "qemu/thread.h"
31 #include "qmp-commands.h"
33 #include "qapi-event.h"
35 #include "exec/memory.h"
36 #include "exec/address-spaces.h"
37 #include "io/channel-buffer.h"
38 #include "io/channel-tls.h"
39 #include "migration/colo.h"
41 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
43 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
45 #define BUFFER_DELAY 100
46 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
48 /* Time in milliseconds we are allowed to stop the source,
49 * for sending the last part */
50 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
52 /* Maximum migrate downtime set to 2000 seconds */
53 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
54 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
56 /* Default compression thread count */
57 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
58 /* Default decompression thread count, usually decompression is at
59 * least 4 times as fast as compression.*/
60 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
61 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
62 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
63 /* Define default autoconverge cpu throttle migration parameters */
64 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
65 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
67 /* Migration XBZRLE default cache size */
68 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
70 /* The delay time (in ms) between two COLO checkpoints
71 * Note: Please change this default value to 10000 when we support hybrid mode.
73 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
75 static NotifierList migration_state_notifiers
=
76 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
78 static bool deferred_incoming
;
81 * Current state of incoming postcopy; note this is not part of
82 * MigrationIncomingState since it's state is used during cleanup
83 * at the end as MIS is being freed.
85 static PostcopyState incoming_postcopy_state
;
87 /* When we add fault tolerance, we could have several
88 migrations at once. For now we don't need to add
89 dynamic creation of migration */
92 MigrationState
*migrate_get_current(void)
95 static MigrationState current_migration
= {
96 .state
= MIGRATION_STATUS_NONE
,
97 .xbzrle_cache_size
= DEFAULT_MIGRATE_CACHE_SIZE
,
100 .compress_level
= DEFAULT_MIGRATE_COMPRESS_LEVEL
,
101 .compress_threads
= DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
,
102 .decompress_threads
= DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
,
103 .cpu_throttle_initial
= DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
,
104 .cpu_throttle_increment
= DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
,
105 .max_bandwidth
= MAX_THROTTLE
,
106 .downtime_limit
= DEFAULT_MIGRATE_SET_DOWNTIME
,
107 .x_checkpoint_delay
= DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
,
112 qemu_mutex_init(¤t_migration
.src_page_req_mutex
);
115 return ¤t_migration
;
118 MigrationIncomingState
*migration_incoming_get_current(void)
121 static MigrationIncomingState mis_current
;
124 mis_current
.state
= MIGRATION_STATUS_NONE
;
125 memset(&mis_current
, 0, sizeof(MigrationIncomingState
));
126 QLIST_INIT(&mis_current
.loadvm_handlers
);
127 qemu_mutex_init(&mis_current
.rp_mutex
);
128 qemu_event_init(&mis_current
.main_thread_load_event
, false);
134 void migration_incoming_state_destroy(void)
136 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
138 qemu_event_destroy(&mis
->main_thread_load_event
);
139 loadvm_free_handlers(mis
);
146 uint8_t runstate
[100];
151 static GlobalState global_state
;
153 int global_state_store(void)
155 if (!runstate_store((char *)global_state
.runstate
,
156 sizeof(global_state
.runstate
))) {
157 error_report("runstate name too big: %s", global_state
.runstate
);
158 trace_migrate_state_too_big();
164 void global_state_store_running(void)
166 const char *state
= RunState_lookup
[RUN_STATE_RUNNING
];
167 strncpy((char *)global_state
.runstate
,
168 state
, sizeof(global_state
.runstate
));
171 static bool global_state_received(void)
173 return global_state
.received
;
176 static RunState
global_state_get_runstate(void)
178 return global_state
.state
;
181 void global_state_set_optional(void)
183 global_state
.optional
= true;
186 static bool global_state_needed(void *opaque
)
188 GlobalState
*s
= opaque
;
189 char *runstate
= (char *)s
->runstate
;
191 /* If it is not optional, it is mandatory */
193 if (s
->optional
== false) {
197 /* If state is running or paused, it is not needed */
199 if (strcmp(runstate
, "running") == 0 ||
200 strcmp(runstate
, "paused") == 0) {
204 /* for any other state it is needed */
208 static int global_state_post_load(void *opaque
, int version_id
)
210 GlobalState
*s
= opaque
;
211 Error
*local_err
= NULL
;
213 char *runstate
= (char *)s
->runstate
;
216 trace_migrate_global_state_post_load(runstate
);
218 r
= qapi_enum_parse(RunState_lookup
, runstate
, RUN_STATE__MAX
,
223 error_report_err(local_err
);
232 static void global_state_pre_save(void *opaque
)
234 GlobalState
*s
= opaque
;
236 trace_migrate_global_state_pre_save((char *)s
->runstate
);
237 s
->size
= strlen((char *)s
->runstate
) + 1;
240 static const VMStateDescription vmstate_globalstate
= {
241 .name
= "globalstate",
243 .minimum_version_id
= 1,
244 .post_load
= global_state_post_load
,
245 .pre_save
= global_state_pre_save
,
246 .needed
= global_state_needed
,
247 .fields
= (VMStateField
[]) {
248 VMSTATE_UINT32(size
, GlobalState
),
249 VMSTATE_BUFFER(runstate
, GlobalState
),
250 VMSTATE_END_OF_LIST()
254 void register_global_state(void)
256 /* We would use it independently that we receive it */
257 strcpy((char *)&global_state
.runstate
, "");
258 global_state
.received
= false;
259 vmstate_register(NULL
, 0, &vmstate_globalstate
, &global_state
);
262 static void migrate_generate_event(int new_state
)
264 if (migrate_use_events()) {
265 qapi_event_send_migration(new_state
, &error_abort
);
270 * Called on -incoming with a defer: uri.
271 * The migration can be started later after any parameters have been
274 static void deferred_incoming_migration(Error
**errp
)
276 if (deferred_incoming
) {
277 error_setg(errp
, "Incoming migration already deferred");
279 deferred_incoming
= true;
282 /* Request a range of pages from the source VM at the given
284 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
285 * as the last request (a name must have been given previously)
286 * Start: Address offset within the RB
287 * Len: Length in bytes required - must be a multiple of pagesize
289 void migrate_send_rp_req_pages(MigrationIncomingState
*mis
, const char *rbname
,
290 ram_addr_t start
, size_t len
)
292 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
293 size_t msglen
= 12; /* start + len */
295 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
296 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
299 int rbname_len
= strlen(rbname
);
300 assert(rbname_len
< 256);
302 bufc
[msglen
++] = rbname_len
;
303 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
304 msglen
+= rbname_len
;
305 migrate_send_rp_message(mis
, MIG_RP_MSG_REQ_PAGES_ID
, msglen
, bufc
);
307 migrate_send_rp_message(mis
, MIG_RP_MSG_REQ_PAGES
, msglen
, bufc
);
311 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
315 qapi_event_send_migration(MIGRATION_STATUS_SETUP
, &error_abort
);
316 if (!strcmp(uri
, "defer")) {
317 deferred_incoming_migration(errp
);
318 } else if (strstart(uri
, "tcp:", &p
)) {
319 tcp_start_incoming_migration(p
, errp
);
321 } else if (strstart(uri
, "rdma:", &p
)) {
322 rdma_start_incoming_migration(p
, errp
);
324 } else if (strstart(uri
, "exec:", &p
)) {
325 exec_start_incoming_migration(p
, errp
);
326 } else if (strstart(uri
, "unix:", &p
)) {
327 unix_start_incoming_migration(p
, errp
);
328 } else if (strstart(uri
, "fd:", &p
)) {
329 fd_start_incoming_migration(p
, errp
);
331 error_setg(errp
, "unknown migration protocol: %s", uri
);
335 static void process_incoming_migration_bh(void *opaque
)
337 Error
*local_err
= NULL
;
338 MigrationIncomingState
*mis
= opaque
;
340 /* Make sure all file formats flush their mutable metadata */
341 bdrv_invalidate_cache_all(&local_err
);
343 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
344 MIGRATION_STATUS_FAILED
);
345 error_report_err(local_err
);
346 migrate_decompress_threads_join();
351 * This must happen after all error conditions are dealt with and
352 * we're sure the VM is going to be running on this host.
354 qemu_announce_self();
356 /* If global state section was not received or we are in running
357 state, we need to obey autostart. Any other state is set with
360 if (!global_state_received() ||
361 global_state_get_runstate() == RUN_STATE_RUNNING
) {
365 runstate_set(RUN_STATE_PAUSED
);
368 runstate_set(global_state_get_runstate());
370 migrate_decompress_threads_join();
372 * This must happen after any state changes since as soon as an external
373 * observer sees this event they might start to prod at the VM assuming
376 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
377 MIGRATION_STATUS_COMPLETED
);
378 qemu_bh_delete(mis
->bh
);
379 migration_incoming_state_destroy();
382 static void process_incoming_migration_co(void *opaque
)
384 QEMUFile
*f
= opaque
;
385 MigrationIncomingState
*mis
= migration_incoming_get_current();
389 mis
->from_src_file
= f
;
390 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
391 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
392 MIGRATION_STATUS_ACTIVE
);
393 ret
= qemu_loadvm_state(f
);
395 ps
= postcopy_state_get();
396 trace_process_incoming_migration_co_end(ret
, ps
);
397 if (ps
!= POSTCOPY_INCOMING_NONE
) {
398 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
400 * Where a migration had postcopy enabled (and thus went to advise)
401 * but managed to complete within the precopy period, we can use
404 postcopy_ram_incoming_cleanup(mis
);
405 } else if (ret
>= 0) {
407 * Postcopy was started, cleanup should happen at the end of the
410 trace_process_incoming_migration_co_postcopy_end_main();
413 /* Else if something went wrong then just fall out of the normal exit */
416 /* we get COLO info, and know if we are in COLO mode */
417 if (!ret
&& migration_incoming_enable_colo()) {
418 mis
->migration_incoming_co
= qemu_coroutine_self();
419 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
420 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
421 mis
->have_colo_incoming_thread
= true;
422 qemu_coroutine_yield();
424 /* Wait checkpoint incoming thread exit before free resource */
425 qemu_thread_join(&mis
->colo_incoming_thread
);
429 free_xbzrle_decoded_buf();
432 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
433 MIGRATION_STATUS_FAILED
);
434 error_report("load of migration failed: %s", strerror(-ret
));
435 migrate_decompress_threads_join();
439 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
440 qemu_bh_schedule(mis
->bh
);
443 void migration_fd_process_incoming(QEMUFile
*f
)
445 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, f
);
447 migrate_decompress_threads_create();
448 qemu_file_set_blocking(f
, false);
449 qemu_coroutine_enter(co
);
453 void migration_channel_process_incoming(MigrationState
*s
,
456 trace_migration_set_incoming_channel(
457 ioc
, object_get_typename(OBJECT(ioc
)));
459 if (s
->parameters
.tls_creds
&&
460 !object_dynamic_cast(OBJECT(ioc
),
461 TYPE_QIO_CHANNEL_TLS
)) {
462 Error
*local_err
= NULL
;
463 migration_tls_channel_process_incoming(s
, ioc
, &local_err
);
465 error_report_err(local_err
);
468 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
469 migration_fd_process_incoming(f
);
474 void migration_channel_connect(MigrationState
*s
,
476 const char *hostname
)
478 trace_migration_set_outgoing_channel(
479 ioc
, object_get_typename(OBJECT(ioc
)), hostname
);
481 if (s
->parameters
.tls_creds
&&
482 !object_dynamic_cast(OBJECT(ioc
),
483 TYPE_QIO_CHANNEL_TLS
)) {
484 Error
*local_err
= NULL
;
485 migration_tls_channel_connect(s
, ioc
, hostname
, &local_err
);
487 migrate_fd_error(s
, local_err
);
488 error_free(local_err
);
491 QEMUFile
*f
= qemu_fopen_channel_output(ioc
);
495 migrate_fd_connect(s
);
501 * Send a message on the return channel back to the source
504 void migrate_send_rp_message(MigrationIncomingState
*mis
,
505 enum mig_rp_message_type message_type
,
506 uint16_t len
, void *data
)
508 trace_migrate_send_rp_message((int)message_type
, len
);
509 qemu_mutex_lock(&mis
->rp_mutex
);
510 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
511 qemu_put_be16(mis
->to_src_file
, len
);
512 qemu_put_buffer(mis
->to_src_file
, data
, len
);
513 qemu_fflush(mis
->to_src_file
);
514 qemu_mutex_unlock(&mis
->rp_mutex
);
518 * Send a 'SHUT' message on the return channel with the given value
519 * to indicate that we've finished with the RP. Non-0 value indicates
522 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
527 buf
= cpu_to_be32(value
);
528 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
532 * Send a 'PONG' message on the return channel with the given value
533 * (normally in response to a 'PING')
535 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
540 buf
= cpu_to_be32(value
);
541 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
544 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
546 MigrationCapabilityStatusList
*head
= NULL
;
547 MigrationCapabilityStatusList
*caps
;
548 MigrationState
*s
= migrate_get_current();
551 caps
= NULL
; /* silence compiler warning */
552 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
553 if (i
== MIGRATION_CAPABILITY_X_COLO
&& !colo_supported()) {
557 head
= g_malloc0(sizeof(*caps
));
560 caps
->next
= g_malloc0(sizeof(*caps
));
564 g_malloc(sizeof(*caps
->value
));
565 caps
->value
->capability
= i
;
566 caps
->value
->state
= s
->enabled_capabilities
[i
];
572 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
574 MigrationParameters
*params
;
575 MigrationState
*s
= migrate_get_current();
577 params
= g_malloc0(sizeof(*params
));
578 params
->has_compress_level
= true;
579 params
->compress_level
= s
->parameters
.compress_level
;
580 params
->has_compress_threads
= true;
581 params
->compress_threads
= s
->parameters
.compress_threads
;
582 params
->has_decompress_threads
= true;
583 params
->decompress_threads
= s
->parameters
.decompress_threads
;
584 params
->has_cpu_throttle_initial
= true;
585 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
586 params
->has_cpu_throttle_increment
= true;
587 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
588 params
->has_tls_creds
= !!s
->parameters
.tls_creds
;
589 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
590 params
->has_tls_hostname
= !!s
->parameters
.tls_hostname
;
591 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
592 params
->has_max_bandwidth
= true;
593 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
594 params
->has_downtime_limit
= true;
595 params
->downtime_limit
= s
->parameters
.downtime_limit
;
596 params
->has_x_checkpoint_delay
= true;
597 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
603 * Return true if we're already in the middle of a migration
604 * (i.e. any of the active or setup states)
606 static bool migration_is_setup_or_active(int state
)
609 case MIGRATION_STATUS_ACTIVE
:
610 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
611 case MIGRATION_STATUS_SETUP
:
620 static void get_xbzrle_cache_stats(MigrationInfo
*info
)
622 if (migrate_use_xbzrle()) {
623 info
->has_xbzrle_cache
= true;
624 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
625 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
626 info
->xbzrle_cache
->bytes
= xbzrle_mig_bytes_transferred();
627 info
->xbzrle_cache
->pages
= xbzrle_mig_pages_transferred();
628 info
->xbzrle_cache
->cache_miss
= xbzrle_mig_pages_cache_miss();
629 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_mig_cache_miss_rate();
630 info
->xbzrle_cache
->overflow
= xbzrle_mig_pages_overflow();
634 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
636 info
->has_ram
= true;
637 info
->ram
= g_malloc0(sizeof(*info
->ram
));
638 info
->ram
->transferred
= ram_bytes_transferred();
639 info
->ram
->total
= ram_bytes_total();
640 info
->ram
->duplicate
= dup_mig_pages_transferred();
641 info
->ram
->skipped
= skipped_mig_pages_transferred();
642 info
->ram
->normal
= norm_mig_pages_transferred();
643 info
->ram
->normal_bytes
= norm_mig_bytes_transferred();
644 info
->ram
->mbps
= s
->mbps
;
645 info
->ram
->dirty_sync_count
= s
->dirty_sync_count
;
646 info
->ram
->postcopy_requests
= s
->postcopy_requests
;
648 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
649 info
->ram
->remaining
= ram_bytes_remaining();
650 info
->ram
->dirty_pages_rate
= s
->dirty_pages_rate
;
654 MigrationInfo
*qmp_query_migrate(Error
**errp
)
656 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
657 MigrationState
*s
= migrate_get_current();
660 case MIGRATION_STATUS_NONE
:
661 /* no migration has happened ever */
663 case MIGRATION_STATUS_SETUP
:
664 info
->has_status
= true;
665 info
->has_total_time
= false;
667 case MIGRATION_STATUS_ACTIVE
:
668 case MIGRATION_STATUS_CANCELLING
:
669 info
->has_status
= true;
670 info
->has_total_time
= true;
671 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
673 info
->has_expected_downtime
= true;
674 info
->expected_downtime
= s
->expected_downtime
;
675 info
->has_setup_time
= true;
676 info
->setup_time
= s
->setup_time
;
678 populate_ram_info(info
, s
);
680 if (blk_mig_active()) {
681 info
->has_disk
= true;
682 info
->disk
= g_malloc0(sizeof(*info
->disk
));
683 info
->disk
->transferred
= blk_mig_bytes_transferred();
684 info
->disk
->remaining
= blk_mig_bytes_remaining();
685 info
->disk
->total
= blk_mig_bytes_total();
688 if (cpu_throttle_active()) {
689 info
->has_cpu_throttle_percentage
= true;
690 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
693 get_xbzrle_cache_stats(info
);
695 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
696 /* Mostly the same as active; TODO add some postcopy stats */
697 info
->has_status
= true;
698 info
->has_total_time
= true;
699 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
701 info
->has_expected_downtime
= true;
702 info
->expected_downtime
= s
->expected_downtime
;
703 info
->has_setup_time
= true;
704 info
->setup_time
= s
->setup_time
;
706 populate_ram_info(info
, s
);
708 if (blk_mig_active()) {
709 info
->has_disk
= true;
710 info
->disk
= g_malloc0(sizeof(*info
->disk
));
711 info
->disk
->transferred
= blk_mig_bytes_transferred();
712 info
->disk
->remaining
= blk_mig_bytes_remaining();
713 info
->disk
->total
= blk_mig_bytes_total();
716 get_xbzrle_cache_stats(info
);
718 case MIGRATION_STATUS_COLO
:
719 info
->has_status
= true;
720 /* TODO: display COLO specific information (checkpoint info etc.) */
722 case MIGRATION_STATUS_COMPLETED
:
723 get_xbzrle_cache_stats(info
);
725 info
->has_status
= true;
726 info
->has_total_time
= true;
727 info
->total_time
= s
->total_time
;
728 info
->has_downtime
= true;
729 info
->downtime
= s
->downtime
;
730 info
->has_setup_time
= true;
731 info
->setup_time
= s
->setup_time
;
733 populate_ram_info(info
, s
);
735 case MIGRATION_STATUS_FAILED
:
736 info
->has_status
= true;
738 info
->has_error_desc
= true;
739 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
742 case MIGRATION_STATUS_CANCELLED
:
743 info
->has_status
= true;
746 info
->status
= s
->state
;
751 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
754 MigrationState
*s
= migrate_get_current();
755 MigrationCapabilityStatusList
*cap
;
756 bool old_postcopy_cap
= migrate_postcopy_ram();
758 if (migration_is_setup_or_active(s
->state
)) {
759 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
763 for (cap
= params
; cap
; cap
= cap
->next
) {
764 if (cap
->value
->capability
== MIGRATION_CAPABILITY_X_COLO
) {
765 if (!colo_supported()) {
766 error_setg(errp
, "COLO is not currently supported, please"
767 " configure with --enable-colo option in order to"
768 " support COLO feature");
772 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
775 if (migrate_postcopy_ram()) {
776 if (migrate_use_compression()) {
777 /* The decompression threads asynchronously write into RAM
778 * rather than use the atomic copies needed to avoid
779 * userfaulting. It should be possible to fix the decompression
780 * threads for compatibility in future.
782 error_report("Postcopy is not currently compatible with "
784 s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
] =
787 /* This check is reasonably expensive, so only when it's being
788 * set the first time, also it's only the destination that needs
791 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
792 !postcopy_ram_supported_by_host()) {
793 /* postcopy_ram_supported_by_host will have emitted a more
796 error_report("Postcopy is not supported");
797 s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
] =
803 void qmp_migrate_set_parameters(MigrationParameters
*params
, Error
**errp
)
805 MigrationState
*s
= migrate_get_current();
807 if (params
->has_compress_level
&&
808 (params
->compress_level
< 0 || params
->compress_level
> 9)) {
809 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
810 "is invalid, it should be in the range of 0 to 9");
813 if (params
->has_compress_threads
&&
814 (params
->compress_threads
< 1 || params
->compress_threads
> 255)) {
815 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
817 "is invalid, it should be in the range of 1 to 255");
820 if (params
->has_decompress_threads
&&
821 (params
->decompress_threads
< 1 || params
->decompress_threads
> 255)) {
822 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
823 "decompress_threads",
824 "is invalid, it should be in the range of 1 to 255");
827 if (params
->has_cpu_throttle_initial
&&
828 (params
->cpu_throttle_initial
< 1 ||
829 params
->cpu_throttle_initial
> 99)) {
830 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
831 "cpu_throttle_initial",
832 "an integer in the range of 1 to 99");
835 if (params
->has_cpu_throttle_increment
&&
836 (params
->cpu_throttle_increment
< 1 ||
837 params
->cpu_throttle_increment
> 99)) {
838 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
839 "cpu_throttle_increment",
840 "an integer in the range of 1 to 99");
843 if (params
->has_max_bandwidth
&&
844 (params
->max_bandwidth
< 0 || params
->max_bandwidth
> SIZE_MAX
)) {
845 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
846 " range of 0 to %zu bytes/second", SIZE_MAX
);
849 if (params
->has_downtime_limit
&&
850 (params
->downtime_limit
< 0 ||
851 params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
852 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
853 "the range of 0 to %d milliseconds",
854 MAX_MIGRATE_DOWNTIME
);
857 if (params
->has_x_checkpoint_delay
&& (params
->x_checkpoint_delay
< 0)) {
858 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
859 "x_checkpoint_delay",
860 "is invalid, it should be positive");
863 if (params
->has_compress_level
) {
864 s
->parameters
.compress_level
= params
->compress_level
;
866 if (params
->has_compress_threads
) {
867 s
->parameters
.compress_threads
= params
->compress_threads
;
869 if (params
->has_decompress_threads
) {
870 s
->parameters
.decompress_threads
= params
->decompress_threads
;
872 if (params
->has_cpu_throttle_initial
) {
873 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
875 if (params
->has_cpu_throttle_increment
) {
876 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
878 if (params
->has_tls_creds
) {
879 g_free(s
->parameters
.tls_creds
);
880 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
);
882 if (params
->has_tls_hostname
) {
883 g_free(s
->parameters
.tls_hostname
);
884 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
);
886 if (params
->has_max_bandwidth
) {
887 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
888 if (s
->to_dst_file
) {
889 qemu_file_set_rate_limit(s
->to_dst_file
,
890 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
893 if (params
->has_downtime_limit
) {
894 s
->parameters
.downtime_limit
= params
->downtime_limit
;
897 if (params
->has_x_checkpoint_delay
) {
898 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
899 if (migration_in_colo_state()) {
900 colo_checkpoint_notify(s
);
906 void qmp_migrate_start_postcopy(Error
**errp
)
908 MigrationState
*s
= migrate_get_current();
910 if (!migrate_postcopy_ram()) {
911 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
912 " the start of migration");
916 if (s
->state
== MIGRATION_STATUS_NONE
) {
917 error_setg(errp
, "Postcopy must be started after migration has been"
922 * we don't error if migration has finished since that would be racy
923 * with issuing this command.
925 atomic_set(&s
->start_postcopy
, true);
928 /* shared migration helpers */
930 void migrate_set_state(int *state
, int old_state
, int new_state
)
932 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
933 trace_migrate_set_state(new_state
);
934 migrate_generate_event(new_state
);
938 static void migrate_fd_cleanup(void *opaque
)
940 MigrationState
*s
= opaque
;
942 qemu_bh_delete(s
->cleanup_bh
);
943 s
->cleanup_bh
= NULL
;
947 if (s
->to_dst_file
) {
948 trace_migrate_fd_cleanup();
949 qemu_mutex_unlock_iothread();
950 if (s
->migration_thread_running
) {
951 qemu_thread_join(&s
->thread
);
952 s
->migration_thread_running
= false;
954 qemu_mutex_lock_iothread();
956 migrate_compress_threads_join();
957 qemu_fclose(s
->to_dst_file
);
958 s
->to_dst_file
= NULL
;
961 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
962 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
964 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
965 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
966 MIGRATION_STATUS_CANCELLED
);
969 notifier_list_notify(&migration_state_notifiers
, s
);
972 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
974 trace_migrate_fd_error(error_get_pretty(error
));
975 assert(s
->to_dst_file
== NULL
);
976 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
977 MIGRATION_STATUS_FAILED
);
979 s
->error
= error_copy(error
);
981 notifier_list_notify(&migration_state_notifiers
, s
);
984 static void migrate_fd_cancel(MigrationState
*s
)
987 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
988 trace_migrate_fd_cancel();
990 if (s
->rp_state
.from_dst_file
) {
991 /* shutdown the rp socket, so causing the rp thread to shutdown */
992 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
996 old_state
= s
->state
;
997 if (!migration_is_setup_or_active(old_state
)) {
1000 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1001 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1004 * If we're unlucky the migration code might be stuck somewhere in a
1005 * send/write while the network has failed and is waiting to timeout;
1006 * if we've got shutdown(2) available then we can force it to quit.
1007 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1008 * called in a bh, so there is no race against this cancel.
1010 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1011 qemu_file_shutdown(f
);
1013 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1014 Error
*local_err
= NULL
;
1016 bdrv_invalidate_cache_all(&local_err
);
1018 error_report_err(local_err
);
1020 s
->block_inactive
= false;
1025 void add_migration_state_change_notifier(Notifier
*notify
)
1027 notifier_list_add(&migration_state_notifiers
, notify
);
1030 void remove_migration_state_change_notifier(Notifier
*notify
)
1032 notifier_remove(notify
);
1035 bool migration_in_setup(MigrationState
*s
)
1037 return s
->state
== MIGRATION_STATUS_SETUP
;
1040 bool migration_has_finished(MigrationState
*s
)
1042 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1045 bool migration_has_failed(MigrationState
*s
)
1047 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1048 s
->state
== MIGRATION_STATUS_FAILED
);
1051 bool migration_in_postcopy(MigrationState
*s
)
1053 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1056 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1058 return migration_in_postcopy(s
) && s
->postcopy_after_devices
;
1061 bool migration_is_idle(MigrationState
*s
)
1064 s
= migrate_get_current();
1068 case MIGRATION_STATUS_NONE
:
1069 case MIGRATION_STATUS_CANCELLED
:
1070 case MIGRATION_STATUS_COMPLETED
:
1071 case MIGRATION_STATUS_FAILED
:
1073 case MIGRATION_STATUS_SETUP
:
1074 case MIGRATION_STATUS_CANCELLING
:
1075 case MIGRATION_STATUS_ACTIVE
:
1076 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1077 case MIGRATION_STATUS_COLO
:
1079 case MIGRATION_STATUS__MAX
:
1080 g_assert_not_reached();
1086 MigrationState
*migrate_init(const MigrationParams
*params
)
1088 MigrationState
*s
= migrate_get_current();
1091 * Reinitialise all migration state, except
1092 * parameters/capabilities that the user set, and
1098 s
->to_dst_file
= NULL
;
1099 s
->state
= MIGRATION_STATUS_NONE
;
1100 s
->params
= *params
;
1101 s
->rp_state
.from_dst_file
= NULL
;
1102 s
->rp_state
.error
= false;
1105 s
->expected_downtime
= 0;
1106 s
->dirty_pages_rate
= 0;
1107 s
->dirty_bytes_rate
= 0;
1109 s
->dirty_sync_count
= 0;
1110 s
->start_postcopy
= false;
1111 s
->postcopy_after_devices
= false;
1112 s
->postcopy_requests
= 0;
1113 s
->migration_thread_running
= false;
1114 s
->last_req_rb
= NULL
;
1115 error_free(s
->error
);
1118 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1120 QSIMPLEQ_INIT(&s
->src_page_requests
);
1122 s
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1126 static GSList
*migration_blockers
;
1128 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1130 if (only_migratable
) {
1131 error_propagate(errp
, error_copy(reason
));
1132 error_prepend(errp
, "disallowing migration blocker "
1133 "(--only_migratable) for: ");
1137 if (migration_is_idle(NULL
)) {
1138 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1142 error_propagate(errp
, error_copy(reason
));
1143 error_prepend(errp
, "disallowing migration blocker (migration in "
1148 void migrate_del_blocker(Error
*reason
)
1150 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1153 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1155 Error
*local_err
= NULL
;
1156 static bool once
= true;
1158 if (!deferred_incoming
) {
1159 error_setg(errp
, "For use with '-incoming defer'");
1163 error_setg(errp
, "The incoming migration has already been started");
1166 qemu_start_incoming_migration(uri
, &local_err
);
1169 error_propagate(errp
, local_err
);
1176 bool migration_is_blocked(Error
**errp
)
1178 if (qemu_savevm_state_blocked(errp
)) {
1182 if (migration_blockers
) {
1183 *errp
= error_copy(migration_blockers
->data
);
1190 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1191 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1194 Error
*local_err
= NULL
;
1195 MigrationState
*s
= migrate_get_current();
1196 MigrationParams params
;
1199 params
.blk
= has_blk
&& blk
;
1200 params
.shared
= has_inc
&& inc
;
1202 if (migration_is_setup_or_active(s
->state
) ||
1203 s
->state
== MIGRATION_STATUS_CANCELLING
||
1204 s
->state
== MIGRATION_STATUS_COLO
) {
1205 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1208 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1209 error_setg(errp
, "Guest is waiting for an incoming migration");
1213 if (migration_is_blocked(errp
)) {
1217 s
= migrate_init(¶ms
);
1219 if (strstart(uri
, "tcp:", &p
)) {
1220 tcp_start_outgoing_migration(s
, p
, &local_err
);
1222 } else if (strstart(uri
, "rdma:", &p
)) {
1223 rdma_start_outgoing_migration(s
, p
, &local_err
);
1225 } else if (strstart(uri
, "exec:", &p
)) {
1226 exec_start_outgoing_migration(s
, p
, &local_err
);
1227 } else if (strstart(uri
, "unix:", &p
)) {
1228 unix_start_outgoing_migration(s
, p
, &local_err
);
1229 } else if (strstart(uri
, "fd:", &p
)) {
1230 fd_start_outgoing_migration(s
, p
, &local_err
);
1232 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1233 "a valid migration protocol");
1234 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1235 MIGRATION_STATUS_FAILED
);
1240 migrate_fd_error(s
, local_err
);
1241 error_propagate(errp
, local_err
);
1246 void qmp_migrate_cancel(Error
**errp
)
1248 migrate_fd_cancel(migrate_get_current());
1251 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1253 MigrationState
*s
= migrate_get_current();
1256 /* Check for truncation */
1257 if (value
!= (size_t)value
) {
1258 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1259 "exceeding address space");
1263 /* Cache should not be larger than guest ram size */
1264 if (value
> ram_bytes_total()) {
1265 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1266 "exceeds guest ram size ");
1270 new_size
= xbzrle_cache_resize(value
);
1272 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1273 "is smaller than page size");
1277 s
->xbzrle_cache_size
= new_size
;
1280 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1282 return migrate_xbzrle_cache_size();
1285 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1287 MigrationParameters p
= {
1288 .has_max_bandwidth
= true,
1289 .max_bandwidth
= value
,
1292 qmp_migrate_set_parameters(&p
, errp
);
1295 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1297 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
1298 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1299 "the range of 0 to %d seconds",
1300 MAX_MIGRATE_DOWNTIME_SECONDS
);
1304 value
*= 1000; /* Convert to milliseconds */
1305 value
= MAX(0, MIN(INT64_MAX
, value
));
1307 MigrationParameters p
= {
1308 .has_downtime_limit
= true,
1309 .downtime_limit
= value
,
1312 qmp_migrate_set_parameters(&p
, errp
);
1315 bool migrate_release_ram(void)
1319 s
= migrate_get_current();
1321 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
1324 bool migrate_postcopy_ram(void)
1328 s
= migrate_get_current();
1330 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1333 bool migrate_auto_converge(void)
1337 s
= migrate_get_current();
1339 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1342 bool migrate_zero_blocks(void)
1346 s
= migrate_get_current();
1348 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
1351 bool migrate_use_compression(void)
1355 s
= migrate_get_current();
1357 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
1360 int migrate_compress_level(void)
1364 s
= migrate_get_current();
1366 return s
->parameters
.compress_level
;
1369 int migrate_compress_threads(void)
1373 s
= migrate_get_current();
1375 return s
->parameters
.compress_threads
;
1378 int migrate_decompress_threads(void)
1382 s
= migrate_get_current();
1384 return s
->parameters
.decompress_threads
;
1387 bool migrate_use_events(void)
1391 s
= migrate_get_current();
1393 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
1396 int migrate_use_xbzrle(void)
1400 s
= migrate_get_current();
1402 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
1405 int64_t migrate_xbzrle_cache_size(void)
1409 s
= migrate_get_current();
1411 return s
->xbzrle_cache_size
;
1414 /* migration thread support */
1416 * Something bad happened to the RP stream, mark an error
1417 * The caller shall print or trace something to indicate why
1419 static void mark_source_rp_bad(MigrationState
*s
)
1421 s
->rp_state
.error
= true;
1424 static struct rp_cmd_args
{
1425 ssize_t len
; /* -1 = variable */
1428 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1429 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1430 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1431 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1432 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1433 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1437 * Process a request for pages received on the return path,
1438 * We're allowed to send more than requested (e.g. to round to our page size)
1439 * and we don't need to send pages that have already been sent.
1441 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1442 ram_addr_t start
, size_t len
)
1444 long our_host_ps
= getpagesize();
1446 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1449 * Since we currently insist on matching page sizes, just sanity check
1450 * we're being asked for whole host pages.
1452 if (start
& (our_host_ps
-1) ||
1453 (len
& (our_host_ps
-1))) {
1454 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1455 " len: %zd", __func__
, start
, len
);
1456 mark_source_rp_bad(ms
);
1460 if (ram_save_queue_pages(ms
, rbname
, start
, len
)) {
1461 mark_source_rp_bad(ms
);
1466 * Handles messages sent on the return path towards the source VM
1469 static void *source_return_path_thread(void *opaque
)
1471 MigrationState
*ms
= opaque
;
1472 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1473 uint16_t header_len
, header_type
;
1475 uint32_t tmp32
, sibling_error
;
1476 ram_addr_t start
= 0; /* =0 to silence warning */
1477 size_t len
= 0, expected_len
;
1480 trace_source_return_path_thread_entry();
1481 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1482 migration_is_setup_or_active(ms
->state
)) {
1483 trace_source_return_path_thread_loop_top();
1484 header_type
= qemu_get_be16(rp
);
1485 header_len
= qemu_get_be16(rp
);
1487 if (header_type
>= MIG_RP_MSG_MAX
||
1488 header_type
== MIG_RP_MSG_INVALID
) {
1489 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1490 header_type
, header_len
);
1491 mark_source_rp_bad(ms
);
1495 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1496 header_len
!= rp_cmd_args
[header_type
].len
) ||
1497 header_len
> sizeof(buf
)) {
1498 error_report("RP: Received '%s' message (0x%04x) with"
1499 "incorrect length %d expecting %zu",
1500 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1501 (size_t)rp_cmd_args
[header_type
].len
);
1502 mark_source_rp_bad(ms
);
1506 /* We know we've got a valid header by this point */
1507 res
= qemu_get_buffer(rp
, buf
, header_len
);
1508 if (res
!= header_len
) {
1509 error_report("RP: Failed reading data for message 0x%04x"
1510 " read %d expected %d",
1511 header_type
, res
, header_len
);
1512 mark_source_rp_bad(ms
);
1516 /* OK, we have the message and the data */
1517 switch (header_type
) {
1518 case MIG_RP_MSG_SHUT
:
1519 sibling_error
= ldl_be_p(buf
);
1520 trace_source_return_path_thread_shut(sibling_error
);
1521 if (sibling_error
) {
1522 error_report("RP: Sibling indicated error %d", sibling_error
);
1523 mark_source_rp_bad(ms
);
1526 * We'll let the main thread deal with closing the RP
1527 * we could do a shutdown(2) on it, but we're the only user
1528 * anyway, so there's nothing gained.
1532 case MIG_RP_MSG_PONG
:
1533 tmp32
= ldl_be_p(buf
);
1534 trace_source_return_path_thread_pong(tmp32
);
1537 case MIG_RP_MSG_REQ_PAGES
:
1538 start
= ldq_be_p(buf
);
1539 len
= ldl_be_p(buf
+ 8);
1540 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1543 case MIG_RP_MSG_REQ_PAGES_ID
:
1544 expected_len
= 12 + 1; /* header + termination */
1546 if (header_len
>= expected_len
) {
1547 start
= ldq_be_p(buf
);
1548 len
= ldl_be_p(buf
+ 8);
1549 /* Now we expect an idstr */
1550 tmp32
= buf
[12]; /* Length of the following idstr */
1551 buf
[13 + tmp32
] = '\0';
1552 expected_len
+= tmp32
;
1554 if (header_len
!= expected_len
) {
1555 error_report("RP: Req_Page_id with length %d expecting %zd",
1556 header_len
, expected_len
);
1557 mark_source_rp_bad(ms
);
1560 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
1567 if (qemu_file_get_error(rp
)) {
1568 trace_source_return_path_thread_bad_end();
1569 mark_source_rp_bad(ms
);
1572 trace_source_return_path_thread_end();
1574 ms
->rp_state
.from_dst_file
= NULL
;
1579 static int open_return_path_on_source(MigrationState
*ms
)
1582 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
1583 if (!ms
->rp_state
.from_dst_file
) {
1587 trace_open_return_path_on_source();
1588 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
1589 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
1591 trace_open_return_path_on_source_continue();
1596 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1597 static int await_return_path_close_on_source(MigrationState
*ms
)
1600 * If this is a normal exit then the destination will send a SHUT and the
1601 * rp_thread will exit, however if there's an error we need to cause
1604 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
1606 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1607 * waiting for the destination.
1609 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
1610 mark_source_rp_bad(ms
);
1612 trace_await_return_path_close_on_source_joining();
1613 qemu_thread_join(&ms
->rp_state
.rp_thread
);
1614 trace_await_return_path_close_on_source_close();
1615 return ms
->rp_state
.error
;
1619 * Switch from normal iteration to postcopy
1620 * Returns non-0 on error
1622 static int postcopy_start(MigrationState
*ms
, bool *old_vm_running
)
1625 QIOChannelBuffer
*bioc
;
1627 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1628 bool restart_block
= false;
1629 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
1630 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1632 trace_postcopy_start();
1633 qemu_mutex_lock_iothread();
1634 trace_postcopy_start_set_run();
1636 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1637 *old_vm_running
= runstate_is_running();
1638 global_state_store();
1639 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1644 ret
= bdrv_inactivate_all();
1648 restart_block
= true;
1651 * Cause any non-postcopiable, but iterative devices to
1652 * send out their final data.
1654 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true);
1657 * in Finish migrate and with the io-lock held everything should
1658 * be quiet, but we've potentially still got dirty pages and we
1659 * need to tell the destination to throw any pages it's already received
1662 if (ram_postcopy_send_discard_bitmap(ms
)) {
1663 error_report("postcopy send discard bitmap failed");
1668 * send rest of state - note things that are doing postcopy
1669 * will notice we're in POSTCOPY_ACTIVE and not actually
1670 * wrap their state up here
1672 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
1673 /* Ping just for debugging, helps line traces up */
1674 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
1677 * While loading the device state we may trigger page transfer
1678 * requests and the fd must be free to process those, and thus
1679 * the destination must read the whole device state off the fd before
1680 * it starts processing it. Unfortunately the ad-hoc migration format
1681 * doesn't allow the destination to know the size to read without fully
1682 * parsing it through each devices load-state code (especially the open
1683 * coded devices that use get/put).
1684 * So we wrap the device state up in a package with a length at the start;
1685 * to do this we use a qemu_buf to hold the whole of the device state.
1687 bioc
= qio_channel_buffer_new(4096);
1688 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
1689 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
1690 object_unref(OBJECT(bioc
));
1693 * Make sure the receiver can get incoming pages before we send the rest
1696 qemu_savevm_send_postcopy_listen(fb
);
1698 qemu_savevm_state_complete_precopy(fb
, false);
1699 qemu_savevm_send_ping(fb
, 3);
1701 qemu_savevm_send_postcopy_run(fb
);
1703 /* <><> end of stuff going into the package */
1705 /* Last point of recovery; as soon as we send the package the destination
1706 * can open devices and potentially start running.
1707 * Lets just check again we've not got any errors.
1709 ret
= qemu_file_get_error(ms
->to_dst_file
);
1711 error_report("postcopy_start: Migration stream errored (pre package)");
1715 restart_block
= false;
1717 /* Now send that blob */
1718 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
1723 /* Send a notify to give a chance for anything that needs to happen
1724 * at the transition to postcopy and after the device state; in particular
1725 * spice needs to trigger a transition now
1727 ms
->postcopy_after_devices
= true;
1728 notifier_list_notify(&migration_state_notifiers
, ms
);
1730 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
1732 qemu_mutex_unlock_iothread();
1735 * Although this ping is just for debug, it could potentially be
1736 * used for getting a better measurement of downtime at the source.
1738 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
1740 if (migrate_release_ram()) {
1741 ram_postcopy_migrated_memory_release(ms
);
1744 ret
= qemu_file_get_error(ms
->to_dst_file
);
1746 error_report("postcopy_start: Migration stream errored");
1747 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1748 MIGRATION_STATUS_FAILED
);
1756 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1757 MIGRATION_STATUS_FAILED
);
1758 if (restart_block
) {
1759 /* A failure happened early enough that we know the destination hasn't
1760 * accessed block devices, so we're safe to recover.
1762 Error
*local_err
= NULL
;
1764 bdrv_invalidate_cache_all(&local_err
);
1766 error_report_err(local_err
);
1769 qemu_mutex_unlock_iothread();
1774 * migration_completion: Used by migration_thread when there's not much left.
1775 * The caller 'breaks' the loop when this returns.
1777 * @s: Current migration state
1778 * @current_active_state: The migration state we expect to be in
1779 * @*old_vm_running: Pointer to old_vm_running flag
1780 * @*start_time: Pointer to time to update
1782 static void migration_completion(MigrationState
*s
, int current_active_state
,
1783 bool *old_vm_running
,
1784 int64_t *start_time
)
1788 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1789 qemu_mutex_lock_iothread();
1790 *start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1791 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1792 *old_vm_running
= runstate_is_running();
1793 ret
= global_state_store();
1796 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1798 * Don't mark the image with BDRV_O_INACTIVE flag if
1799 * we will go into COLO stage later.
1801 if (ret
>= 0 && !migrate_colo_enabled()) {
1802 ret
= bdrv_inactivate_all();
1805 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
1806 qemu_savevm_state_complete_precopy(s
->to_dst_file
, false);
1807 s
->block_inactive
= true;
1810 qemu_mutex_unlock_iothread();
1815 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1816 trace_migration_completion_postcopy_end();
1818 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
1819 trace_migration_completion_postcopy_end_after_complete();
1823 * If rp was opened we must clean up the thread before
1824 * cleaning everything else up (since if there are no failures
1825 * it will wait for the destination to send it's status in
1827 * Postcopy opens rp if enabled (even if it's not avtivated)
1829 if (migrate_postcopy_ram()) {
1831 trace_migration_completion_postcopy_end_before_rp();
1832 rp_error
= await_return_path_close_on_source(s
);
1833 trace_migration_completion_postcopy_end_after_rp(rp_error
);
1835 goto fail_invalidate
;
1839 if (qemu_file_get_error(s
->to_dst_file
)) {
1840 trace_migration_completion_file_err();
1841 goto fail_invalidate
;
1844 if (!migrate_colo_enabled()) {
1845 migrate_set_state(&s
->state
, current_active_state
,
1846 MIGRATION_STATUS_COMPLETED
);
1852 /* If not doing postcopy, vm_start() will be called: let's regain
1853 * control on images.
1855 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1856 Error
*local_err
= NULL
;
1858 qemu_mutex_lock_iothread();
1859 bdrv_invalidate_cache_all(&local_err
);
1861 error_report_err(local_err
);
1863 s
->block_inactive
= false;
1865 qemu_mutex_unlock_iothread();
1869 migrate_set_state(&s
->state
, current_active_state
,
1870 MIGRATION_STATUS_FAILED
);
1873 bool migrate_colo_enabled(void)
1875 MigrationState
*s
= migrate_get_current();
1876 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
1880 * Master migration thread on the source VM.
1881 * It drives the migration and pumps the data down the outgoing channel.
1883 static void *migration_thread(void *opaque
)
1885 MigrationState
*s
= opaque
;
1886 /* Used by the bandwidth calcs, updated later */
1887 int64_t initial_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1888 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
1889 int64_t initial_bytes
= 0;
1890 int64_t max_size
= 0;
1891 int64_t start_time
= initial_time
;
1893 bool old_vm_running
= false;
1894 bool entered_postcopy
= false;
1895 /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
1896 enum MigrationStatus current_active_state
= MIGRATION_STATUS_ACTIVE
;
1897 bool enable_colo
= migrate_colo_enabled();
1899 rcu_register_thread();
1901 qemu_savevm_state_header(s
->to_dst_file
);
1903 if (migrate_postcopy_ram()) {
1904 /* Now tell the dest that it should open its end so it can reply */
1905 qemu_savevm_send_open_return_path(s
->to_dst_file
);
1907 /* And do a ping that will make stuff easier to debug */
1908 qemu_savevm_send_ping(s
->to_dst_file
, 1);
1911 * Tell the destination that we *might* want to do postcopy later;
1912 * if the other end can't do postcopy it should fail now, nice and
1915 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
1918 qemu_savevm_state_begin(s
->to_dst_file
, &s
->params
);
1920 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
1921 current_active_state
= MIGRATION_STATUS_ACTIVE
;
1922 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1923 MIGRATION_STATUS_ACTIVE
);
1925 trace_migration_thread_setup_complete();
1927 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
1928 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1929 int64_t current_time
;
1930 uint64_t pending_size
;
1932 if (!qemu_file_rate_limit(s
->to_dst_file
)) {
1933 uint64_t pend_post
, pend_nonpost
;
1935 qemu_savevm_state_pending(s
->to_dst_file
, max_size
, &pend_nonpost
,
1937 pending_size
= pend_nonpost
+ pend_post
;
1938 trace_migrate_pending(pending_size
, max_size
,
1939 pend_post
, pend_nonpost
);
1940 if (pending_size
&& pending_size
>= max_size
) {
1941 /* Still a significant amount to transfer */
1943 if (migrate_postcopy_ram() &&
1944 s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
&&
1945 pend_nonpost
<= max_size
&&
1946 atomic_read(&s
->start_postcopy
)) {
1948 if (!postcopy_start(s
, &old_vm_running
)) {
1949 current_active_state
= MIGRATION_STATUS_POSTCOPY_ACTIVE
;
1950 entered_postcopy
= true;
1955 /* Just another iteration step */
1956 qemu_savevm_state_iterate(s
->to_dst_file
, entered_postcopy
);
1958 trace_migration_thread_low_pending(pending_size
);
1959 migration_completion(s
, current_active_state
,
1960 &old_vm_running
, &start_time
);
1965 if (qemu_file_get_error(s
->to_dst_file
)) {
1966 migrate_set_state(&s
->state
, current_active_state
,
1967 MIGRATION_STATUS_FAILED
);
1968 trace_migration_thread_file_err();
1971 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1972 if (current_time
>= initial_time
+ BUFFER_DELAY
) {
1973 uint64_t transferred_bytes
= qemu_ftell(s
->to_dst_file
) -
1975 uint64_t time_spent
= current_time
- initial_time
;
1976 double bandwidth
= (double)transferred_bytes
/ time_spent
;
1977 max_size
= bandwidth
* s
->parameters
.downtime_limit
;
1979 s
->mbps
= (((double) transferred_bytes
* 8.0) /
1980 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
1982 trace_migrate_transferred(transferred_bytes
, time_spent
,
1983 bandwidth
, max_size
);
1984 /* if we haven't sent anything, we don't want to recalculate
1985 10000 is a small enough number for our purposes */
1986 if (s
->dirty_bytes_rate
&& transferred_bytes
> 10000) {
1987 s
->expected_downtime
= s
->dirty_bytes_rate
/ bandwidth
;
1990 qemu_file_reset_rate_limit(s
->to_dst_file
);
1991 initial_time
= current_time
;
1992 initial_bytes
= qemu_ftell(s
->to_dst_file
);
1994 if (qemu_file_rate_limit(s
->to_dst_file
)) {
1995 /* usleep expects microseconds */
1996 g_usleep((initial_time
+ BUFFER_DELAY
- current_time
)*1000);
2000 trace_migration_thread_after_loop();
2001 /* If we enabled cpu throttling for auto-converge, turn it off. */
2002 cpu_throttle_stop();
2003 end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2005 qemu_mutex_lock_iothread();
2007 * The resource has been allocated by migration will be reused in COLO
2008 * process, so don't release them.
2011 qemu_savevm_state_cleanup();
2013 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
2014 uint64_t transferred_bytes
= qemu_ftell(s
->to_dst_file
);
2015 s
->total_time
= end_time
- s
->total_time
;
2016 if (!entered_postcopy
) {
2017 s
->downtime
= end_time
- start_time
;
2019 if (s
->total_time
) {
2020 s
->mbps
= (((double) transferred_bytes
* 8.0) /
2021 ((double) s
->total_time
)) / 1000;
2023 runstate_set(RUN_STATE_POSTMIGRATE
);
2025 if (s
->state
== MIGRATION_STATUS_ACTIVE
&& enable_colo
) {
2026 migrate_start_colo_process(s
);
2027 qemu_savevm_state_cleanup();
2029 * Fixme: we will run VM in COLO no matter its old running state.
2030 * After exited COLO, we will keep running.
2032 old_vm_running
= true;
2034 if (old_vm_running
&& !entered_postcopy
) {
2037 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2038 runstate_set(RUN_STATE_POSTMIGRATE
);
2042 qemu_bh_schedule(s
->cleanup_bh
);
2043 qemu_mutex_unlock_iothread();
2045 rcu_unregister_thread();
2049 void migrate_fd_connect(MigrationState
*s
)
2051 s
->expected_downtime
= s
->parameters
.downtime_limit
;
2052 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
2054 qemu_file_set_blocking(s
->to_dst_file
, true);
2055 qemu_file_set_rate_limit(s
->to_dst_file
,
2056 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
2058 /* Notify before starting migration thread */
2059 notifier_list_notify(&migration_state_notifiers
, s
);
2062 * Open the return path; currently for postcopy but other things might
2065 if (migrate_postcopy_ram()) {
2066 if (open_return_path_on_source(s
)) {
2067 error_report("Unable to open return-path for postcopy");
2068 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2069 MIGRATION_STATUS_FAILED
);
2070 migrate_fd_cleanup(s
);
2075 migrate_compress_threads_create();
2076 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
2077 QEMU_THREAD_JOINABLE
);
2078 s
->migration_thread_running
= true;
2081 PostcopyState
postcopy_state_get(void)
2083 return atomic_mb_read(&incoming_postcopy_state
);
2086 /* Set the state and return the old state */
2087 PostcopyState
postcopy_state_set(PostcopyState new_state
)
2089 return atomic_xchg(&incoming_postcopy_state
, new_state
);