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"
40 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
42 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
44 #define BUFFER_DELAY 100
45 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
47 /* Default compression thread count */
48 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
49 /* Default decompression thread count, usually decompression is at
50 * least 4 times as fast as compression.*/
51 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
52 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
53 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
54 /* Define default autoconverge cpu throttle migration parameters */
55 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
56 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
58 /* Migration XBZRLE default cache size */
59 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
61 static NotifierList migration_state_notifiers
=
62 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
64 static bool deferred_incoming
;
67 * Current state of incoming postcopy; note this is not part of
68 * MigrationIncomingState since it's state is used during cleanup
69 * at the end as MIS is being freed.
71 static PostcopyState incoming_postcopy_state
;
73 /* When we add fault tolerance, we could have several
74 migrations at once. For now we don't need to add
75 dynamic creation of migration */
78 MigrationState
*migrate_get_current(void)
81 static MigrationState current_migration
= {
82 .state
= MIGRATION_STATUS_NONE
,
83 .bandwidth_limit
= MAX_THROTTLE
,
84 .xbzrle_cache_size
= DEFAULT_MIGRATE_CACHE_SIZE
,
87 .compress_level
= DEFAULT_MIGRATE_COMPRESS_LEVEL
,
88 .compress_threads
= DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
,
89 .decompress_threads
= DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
,
90 .cpu_throttle_initial
= DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
,
91 .cpu_throttle_increment
= DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
,
96 qemu_mutex_init(¤t_migration
.src_page_req_mutex
);
99 return ¤t_migration
;
103 static MigrationIncomingState
*mis_current
;
105 MigrationIncomingState
*migration_incoming_get_current(void)
110 MigrationIncomingState
*migration_incoming_state_new(QEMUFile
* f
)
112 mis_current
= g_new0(MigrationIncomingState
, 1);
113 mis_current
->from_src_file
= f
;
114 mis_current
->state
= MIGRATION_STATUS_NONE
;
115 QLIST_INIT(&mis_current
->loadvm_handlers
);
116 qemu_mutex_init(&mis_current
->rp_mutex
);
117 qemu_event_init(&mis_current
->main_thread_load_event
, false);
122 void migration_incoming_state_destroy(void)
124 qemu_event_destroy(&mis_current
->main_thread_load_event
);
125 loadvm_free_handlers(mis_current
);
134 uint8_t runstate
[100];
139 static GlobalState global_state
;
141 int global_state_store(void)
143 if (!runstate_store((char *)global_state
.runstate
,
144 sizeof(global_state
.runstate
))) {
145 error_report("runstate name too big: %s", global_state
.runstate
);
146 trace_migrate_state_too_big();
152 void global_state_store_running(void)
154 const char *state
= RunState_lookup
[RUN_STATE_RUNNING
];
155 strncpy((char *)global_state
.runstate
,
156 state
, sizeof(global_state
.runstate
));
159 static bool global_state_received(void)
161 return global_state
.received
;
164 static RunState
global_state_get_runstate(void)
166 return global_state
.state
;
169 void global_state_set_optional(void)
171 global_state
.optional
= true;
174 static bool global_state_needed(void *opaque
)
176 GlobalState
*s
= opaque
;
177 char *runstate
= (char *)s
->runstate
;
179 /* If it is not optional, it is mandatory */
181 if (s
->optional
== false) {
185 /* If state is running or paused, it is not needed */
187 if (strcmp(runstate
, "running") == 0 ||
188 strcmp(runstate
, "paused") == 0) {
192 /* for any other state it is needed */
196 static int global_state_post_load(void *opaque
, int version_id
)
198 GlobalState
*s
= opaque
;
199 Error
*local_err
= NULL
;
201 char *runstate
= (char *)s
->runstate
;
204 trace_migrate_global_state_post_load(runstate
);
206 r
= qapi_enum_parse(RunState_lookup
, runstate
, RUN_STATE__MAX
,
211 error_report_err(local_err
);
220 static void global_state_pre_save(void *opaque
)
222 GlobalState
*s
= opaque
;
224 trace_migrate_global_state_pre_save((char *)s
->runstate
);
225 s
->size
= strlen((char *)s
->runstate
) + 1;
228 static const VMStateDescription vmstate_globalstate
= {
229 .name
= "globalstate",
231 .minimum_version_id
= 1,
232 .post_load
= global_state_post_load
,
233 .pre_save
= global_state_pre_save
,
234 .needed
= global_state_needed
,
235 .fields
= (VMStateField
[]) {
236 VMSTATE_UINT32(size
, GlobalState
),
237 VMSTATE_BUFFER(runstate
, GlobalState
),
238 VMSTATE_END_OF_LIST()
242 void register_global_state(void)
244 /* We would use it independently that we receive it */
245 strcpy((char *)&global_state
.runstate
, "");
246 global_state
.received
= false;
247 vmstate_register(NULL
, 0, &vmstate_globalstate
, &global_state
);
250 static void migrate_generate_event(int new_state
)
252 if (migrate_use_events()) {
253 qapi_event_send_migration(new_state
, &error_abort
);
258 * Called on -incoming with a defer: uri.
259 * The migration can be started later after any parameters have been
262 static void deferred_incoming_migration(Error
**errp
)
264 if (deferred_incoming
) {
265 error_setg(errp
, "Incoming migration already deferred");
267 deferred_incoming
= true;
270 /* Request a range of pages from the source VM at the given
272 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
273 * as the last request (a name must have been given previously)
274 * Start: Address offset within the RB
275 * Len: Length in bytes required - must be a multiple of pagesize
277 void migrate_send_rp_req_pages(MigrationIncomingState
*mis
, const char *rbname
,
278 ram_addr_t start
, size_t len
)
280 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
281 size_t msglen
= 12; /* start + len */
283 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
284 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
287 int rbname_len
= strlen(rbname
);
288 assert(rbname_len
< 256);
290 bufc
[msglen
++] = rbname_len
;
291 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
292 msglen
+= rbname_len
;
293 migrate_send_rp_message(mis
, MIG_RP_MSG_REQ_PAGES_ID
, msglen
, bufc
);
295 migrate_send_rp_message(mis
, MIG_RP_MSG_REQ_PAGES
, msglen
, bufc
);
299 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
303 qapi_event_send_migration(MIGRATION_STATUS_SETUP
, &error_abort
);
304 if (!strcmp(uri
, "defer")) {
305 deferred_incoming_migration(errp
);
306 } else if (strstart(uri
, "tcp:", &p
)) {
307 tcp_start_incoming_migration(p
, errp
);
309 } else if (strstart(uri
, "rdma:", &p
)) {
310 rdma_start_incoming_migration(p
, errp
);
312 } else if (strstart(uri
, "exec:", &p
)) {
313 exec_start_incoming_migration(p
, errp
);
314 } else if (strstart(uri
, "unix:", &p
)) {
315 unix_start_incoming_migration(p
, errp
);
316 } else if (strstart(uri
, "fd:", &p
)) {
317 fd_start_incoming_migration(p
, errp
);
319 error_setg(errp
, "unknown migration protocol: %s", uri
);
323 static void process_incoming_migration_bh(void *opaque
)
325 Error
*local_err
= NULL
;
326 MigrationIncomingState
*mis
= opaque
;
328 /* Make sure all file formats flush their mutable metadata */
329 bdrv_invalidate_cache_all(&local_err
);
331 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
332 MIGRATION_STATUS_FAILED
);
333 error_report_err(local_err
);
334 migrate_decompress_threads_join();
339 * This must happen after all error conditions are dealt with and
340 * we're sure the VM is going to be running on this host.
342 qemu_announce_self();
344 /* If global state section was not received or we are in running
345 state, we need to obey autostart. Any other state is set with
348 if (!global_state_received() ||
349 global_state_get_runstate() == RUN_STATE_RUNNING
) {
353 runstate_set(RUN_STATE_PAUSED
);
356 runstate_set(global_state_get_runstate());
358 migrate_decompress_threads_join();
360 * This must happen after any state changes since as soon as an external
361 * observer sees this event they might start to prod at the VM assuming
364 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
365 MIGRATION_STATUS_COMPLETED
);
366 qemu_bh_delete(mis
->bh
);
367 migration_incoming_state_destroy();
370 static void process_incoming_migration_co(void *opaque
)
372 QEMUFile
*f
= opaque
;
373 MigrationIncomingState
*mis
;
377 mis
= migration_incoming_state_new(f
);
378 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
379 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
380 MIGRATION_STATUS_ACTIVE
);
381 ret
= qemu_loadvm_state(f
);
383 ps
= postcopy_state_get();
384 trace_process_incoming_migration_co_end(ret
, ps
);
385 if (ps
!= POSTCOPY_INCOMING_NONE
) {
386 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
388 * Where a migration had postcopy enabled (and thus went to advise)
389 * but managed to complete within the precopy period, we can use
392 postcopy_ram_incoming_cleanup(mis
);
393 } else if (ret
>= 0) {
395 * Postcopy was started, cleanup should happen at the end of the
398 trace_process_incoming_migration_co_postcopy_end_main();
401 /* Else if something went wrong then just fall out of the normal exit */
405 free_xbzrle_decoded_buf();
408 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
409 MIGRATION_STATUS_FAILED
);
410 error_report("load of migration failed: %s", strerror(-ret
));
411 migrate_decompress_threads_join();
415 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
416 qemu_bh_schedule(mis
->bh
);
419 void migration_fd_process_incoming(QEMUFile
*f
)
421 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, f
);
423 migrate_decompress_threads_create();
424 qemu_file_set_blocking(f
, false);
425 qemu_coroutine_enter(co
);
429 void migration_channel_process_incoming(MigrationState
*s
,
432 trace_migration_set_incoming_channel(
433 ioc
, object_get_typename(OBJECT(ioc
)));
435 if (s
->parameters
.tls_creds
&&
436 !object_dynamic_cast(OBJECT(ioc
),
437 TYPE_QIO_CHANNEL_TLS
)) {
438 Error
*local_err
= NULL
;
439 migration_tls_channel_process_incoming(s
, ioc
, &local_err
);
441 error_report_err(local_err
);
444 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
445 migration_fd_process_incoming(f
);
450 void migration_channel_connect(MigrationState
*s
,
452 const char *hostname
)
454 trace_migration_set_outgoing_channel(
455 ioc
, object_get_typename(OBJECT(ioc
)), hostname
);
457 if (s
->parameters
.tls_creds
&&
458 !object_dynamic_cast(OBJECT(ioc
),
459 TYPE_QIO_CHANNEL_TLS
)) {
460 Error
*local_err
= NULL
;
461 migration_tls_channel_connect(s
, ioc
, hostname
, &local_err
);
463 migrate_fd_error(s
, local_err
);
464 error_free(local_err
);
467 QEMUFile
*f
= qemu_fopen_channel_output(ioc
);
471 migrate_fd_connect(s
);
477 * Send a message on the return channel back to the source
480 void migrate_send_rp_message(MigrationIncomingState
*mis
,
481 enum mig_rp_message_type message_type
,
482 uint16_t len
, void *data
)
484 trace_migrate_send_rp_message((int)message_type
, len
);
485 qemu_mutex_lock(&mis
->rp_mutex
);
486 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
487 qemu_put_be16(mis
->to_src_file
, len
);
488 qemu_put_buffer(mis
->to_src_file
, data
, len
);
489 qemu_fflush(mis
->to_src_file
);
490 qemu_mutex_unlock(&mis
->rp_mutex
);
494 * Send a 'SHUT' message on the return channel with the given value
495 * to indicate that we've finished with the RP. Non-0 value indicates
498 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
503 buf
= cpu_to_be32(value
);
504 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
508 * Send a 'PONG' message on the return channel with the given value
509 * (normally in response to a 'PING')
511 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
516 buf
= cpu_to_be32(value
);
517 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
520 /* amount of nanoseconds we are willing to wait for migration to be down.
521 * the choice of nanoseconds is because it is the maximum resolution that
522 * get_clock() can achieve. It is an internal measure. All user-visible
523 * units must be in seconds */
524 static uint64_t max_downtime
= 300000000;
526 uint64_t migrate_max_downtime(void)
531 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
533 MigrationCapabilityStatusList
*head
= NULL
;
534 MigrationCapabilityStatusList
*caps
;
535 MigrationState
*s
= migrate_get_current();
538 caps
= NULL
; /* silence compiler warning */
539 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
541 head
= g_malloc0(sizeof(*caps
));
544 caps
->next
= g_malloc0(sizeof(*caps
));
548 g_malloc(sizeof(*caps
->value
));
549 caps
->value
->capability
= i
;
550 caps
->value
->state
= s
->enabled_capabilities
[i
];
556 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
558 MigrationParameters
*params
;
559 MigrationState
*s
= migrate_get_current();
561 params
= g_malloc0(sizeof(*params
));
562 params
->has_compress_level
= true;
563 params
->compress_level
= s
->parameters
.compress_level
;
564 params
->has_compress_threads
= true;
565 params
->compress_threads
= s
->parameters
.compress_threads
;
566 params
->has_decompress_threads
= true;
567 params
->decompress_threads
= s
->parameters
.decompress_threads
;
568 params
->has_cpu_throttle_initial
= true;
569 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
570 params
->has_cpu_throttle_increment
= true;
571 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
572 params
->has_tls_creds
= !!s
->parameters
.tls_creds
;
573 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
574 params
->has_tls_hostname
= !!s
->parameters
.tls_hostname
;
575 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
581 * Return true if we're already in the middle of a migration
582 * (i.e. any of the active or setup states)
584 static bool migration_is_setup_or_active(int state
)
587 case MIGRATION_STATUS_ACTIVE
:
588 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
589 case MIGRATION_STATUS_SETUP
:
598 static void get_xbzrle_cache_stats(MigrationInfo
*info
)
600 if (migrate_use_xbzrle()) {
601 info
->has_xbzrle_cache
= true;
602 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
603 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
604 info
->xbzrle_cache
->bytes
= xbzrle_mig_bytes_transferred();
605 info
->xbzrle_cache
->pages
= xbzrle_mig_pages_transferred();
606 info
->xbzrle_cache
->cache_miss
= xbzrle_mig_pages_cache_miss();
607 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_mig_cache_miss_rate();
608 info
->xbzrle_cache
->overflow
= xbzrle_mig_pages_overflow();
612 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
614 info
->has_ram
= true;
615 info
->ram
= g_malloc0(sizeof(*info
->ram
));
616 info
->ram
->transferred
= ram_bytes_transferred();
617 info
->ram
->total
= ram_bytes_total();
618 info
->ram
->duplicate
= dup_mig_pages_transferred();
619 info
->ram
->skipped
= skipped_mig_pages_transferred();
620 info
->ram
->normal
= norm_mig_pages_transferred();
621 info
->ram
->normal_bytes
= norm_mig_bytes_transferred();
622 info
->ram
->mbps
= s
->mbps
;
623 info
->ram
->dirty_sync_count
= s
->dirty_sync_count
;
624 info
->ram
->postcopy_requests
= s
->postcopy_requests
;
626 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
627 info
->ram
->remaining
= ram_bytes_remaining();
628 info
->ram
->dirty_pages_rate
= s
->dirty_pages_rate
;
632 MigrationInfo
*qmp_query_migrate(Error
**errp
)
634 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
635 MigrationState
*s
= migrate_get_current();
638 case MIGRATION_STATUS_NONE
:
639 /* no migration has happened ever */
641 case MIGRATION_STATUS_SETUP
:
642 info
->has_status
= true;
643 info
->has_total_time
= false;
645 case MIGRATION_STATUS_ACTIVE
:
646 case MIGRATION_STATUS_CANCELLING
:
647 info
->has_status
= true;
648 info
->has_total_time
= true;
649 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
651 info
->has_expected_downtime
= true;
652 info
->expected_downtime
= s
->expected_downtime
;
653 info
->has_setup_time
= true;
654 info
->setup_time
= s
->setup_time
;
656 populate_ram_info(info
, s
);
658 if (blk_mig_active()) {
659 info
->has_disk
= true;
660 info
->disk
= g_malloc0(sizeof(*info
->disk
));
661 info
->disk
->transferred
= blk_mig_bytes_transferred();
662 info
->disk
->remaining
= blk_mig_bytes_remaining();
663 info
->disk
->total
= blk_mig_bytes_total();
666 if (cpu_throttle_active()) {
667 info
->has_cpu_throttle_percentage
= true;
668 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
671 get_xbzrle_cache_stats(info
);
673 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
674 /* Mostly the same as active; TODO add some postcopy stats */
675 info
->has_status
= true;
676 info
->has_total_time
= true;
677 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
679 info
->has_expected_downtime
= true;
680 info
->expected_downtime
= s
->expected_downtime
;
681 info
->has_setup_time
= true;
682 info
->setup_time
= s
->setup_time
;
684 populate_ram_info(info
, s
);
686 if (blk_mig_active()) {
687 info
->has_disk
= true;
688 info
->disk
= g_malloc0(sizeof(*info
->disk
));
689 info
->disk
->transferred
= blk_mig_bytes_transferred();
690 info
->disk
->remaining
= blk_mig_bytes_remaining();
691 info
->disk
->total
= blk_mig_bytes_total();
694 get_xbzrle_cache_stats(info
);
696 case MIGRATION_STATUS_COMPLETED
:
697 get_xbzrle_cache_stats(info
);
699 info
->has_status
= true;
700 info
->has_total_time
= true;
701 info
->total_time
= s
->total_time
;
702 info
->has_downtime
= true;
703 info
->downtime
= s
->downtime
;
704 info
->has_setup_time
= true;
705 info
->setup_time
= s
->setup_time
;
707 populate_ram_info(info
, s
);
709 case MIGRATION_STATUS_FAILED
:
710 info
->has_status
= true;
712 info
->has_error_desc
= true;
713 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
716 case MIGRATION_STATUS_CANCELLED
:
717 info
->has_status
= true;
720 info
->status
= s
->state
;
725 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
728 MigrationState
*s
= migrate_get_current();
729 MigrationCapabilityStatusList
*cap
;
730 bool old_postcopy_cap
= migrate_postcopy_ram();
732 if (migration_is_setup_or_active(s
->state
)) {
733 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
737 for (cap
= params
; cap
; cap
= cap
->next
) {
738 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
741 if (migrate_postcopy_ram()) {
742 if (migrate_use_compression()) {
743 /* The decompression threads asynchronously write into RAM
744 * rather than use the atomic copies needed to avoid
745 * userfaulting. It should be possible to fix the decompression
746 * threads for compatibility in future.
748 error_report("Postcopy is not currently compatible with "
750 s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
] =
753 /* This check is reasonably expensive, so only when it's being
754 * set the first time, also it's only the destination that needs
757 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
758 !postcopy_ram_supported_by_host()) {
759 /* postcopy_ram_supported_by_host will have emitted a more
762 error_report("Postcopy is not supported");
763 s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
] =
769 void qmp_migrate_set_parameters(bool has_compress_level
,
770 int64_t compress_level
,
771 bool has_compress_threads
,
772 int64_t compress_threads
,
773 bool has_decompress_threads
,
774 int64_t decompress_threads
,
775 bool has_cpu_throttle_initial
,
776 int64_t cpu_throttle_initial
,
777 bool has_cpu_throttle_increment
,
778 int64_t cpu_throttle_increment
,
780 const char *tls_creds
,
781 bool has_tls_hostname
,
782 const char *tls_hostname
,
785 MigrationState
*s
= migrate_get_current();
787 if (has_compress_level
&& (compress_level
< 0 || compress_level
> 9)) {
788 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
789 "is invalid, it should be in the range of 0 to 9");
792 if (has_compress_threads
&&
793 (compress_threads
< 1 || compress_threads
> 255)) {
794 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
796 "is invalid, it should be in the range of 1 to 255");
799 if (has_decompress_threads
&&
800 (decompress_threads
< 1 || decompress_threads
> 255)) {
801 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
802 "decompress_threads",
803 "is invalid, it should be in the range of 1 to 255");
806 if (has_cpu_throttle_initial
&&
807 (cpu_throttle_initial
< 1 || cpu_throttle_initial
> 99)) {
808 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
809 "cpu_throttle_initial",
810 "an integer in the range of 1 to 99");
812 if (has_cpu_throttle_increment
&&
813 (cpu_throttle_increment
< 1 || cpu_throttle_increment
> 99)) {
814 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
815 "cpu_throttle_increment",
816 "an integer in the range of 1 to 99");
819 if (has_compress_level
) {
820 s
->parameters
.compress_level
= compress_level
;
822 if (has_compress_threads
) {
823 s
->parameters
.compress_threads
= compress_threads
;
825 if (has_decompress_threads
) {
826 s
->parameters
.decompress_threads
= decompress_threads
;
828 if (has_cpu_throttle_initial
) {
829 s
->parameters
.cpu_throttle_initial
= cpu_throttle_initial
;
831 if (has_cpu_throttle_increment
) {
832 s
->parameters
.cpu_throttle_increment
= cpu_throttle_increment
;
835 g_free(s
->parameters
.tls_creds
);
836 s
->parameters
.tls_creds
= g_strdup(tls_creds
);
838 if (has_tls_hostname
) {
839 g_free(s
->parameters
.tls_hostname
);
840 s
->parameters
.tls_hostname
= g_strdup(tls_hostname
);
845 void qmp_migrate_start_postcopy(Error
**errp
)
847 MigrationState
*s
= migrate_get_current();
849 if (!migrate_postcopy_ram()) {
850 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
851 " the start of migration");
855 if (s
->state
== MIGRATION_STATUS_NONE
) {
856 error_setg(errp
, "Postcopy must be started after migration has been"
861 * we don't error if migration has finished since that would be racy
862 * with issuing this command.
864 atomic_set(&s
->start_postcopy
, true);
867 /* shared migration helpers */
869 void migrate_set_state(int *state
, int old_state
, int new_state
)
871 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
872 trace_migrate_set_state(new_state
);
873 migrate_generate_event(new_state
);
877 static void migrate_fd_cleanup(void *opaque
)
879 MigrationState
*s
= opaque
;
881 qemu_bh_delete(s
->cleanup_bh
);
882 s
->cleanup_bh
= NULL
;
886 if (s
->to_dst_file
) {
887 trace_migrate_fd_cleanup();
888 qemu_mutex_unlock_iothread();
889 if (s
->migration_thread_running
) {
890 qemu_thread_join(&s
->thread
);
891 s
->migration_thread_running
= false;
893 qemu_mutex_lock_iothread();
895 migrate_compress_threads_join();
896 qemu_fclose(s
->to_dst_file
);
897 s
->to_dst_file
= NULL
;
900 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
901 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
903 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
904 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
905 MIGRATION_STATUS_CANCELLED
);
908 notifier_list_notify(&migration_state_notifiers
, s
);
911 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
913 trace_migrate_fd_error(error
? error_get_pretty(error
) : "");
914 assert(s
->to_dst_file
== NULL
);
915 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
916 MIGRATION_STATUS_FAILED
);
918 s
->error
= error_copy(error
);
920 notifier_list_notify(&migration_state_notifiers
, s
);
923 static void migrate_fd_cancel(MigrationState
*s
)
926 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
927 trace_migrate_fd_cancel();
929 if (s
->rp_state
.from_dst_file
) {
930 /* shutdown the rp socket, so causing the rp thread to shutdown */
931 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
935 old_state
= s
->state
;
936 if (!migration_is_setup_or_active(old_state
)) {
939 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
940 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
943 * If we're unlucky the migration code might be stuck somewhere in a
944 * send/write while the network has failed and is waiting to timeout;
945 * if we've got shutdown(2) available then we can force it to quit.
946 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
947 * called in a bh, so there is no race against this cancel.
949 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
950 qemu_file_shutdown(f
);
954 void add_migration_state_change_notifier(Notifier
*notify
)
956 notifier_list_add(&migration_state_notifiers
, notify
);
959 void remove_migration_state_change_notifier(Notifier
*notify
)
961 notifier_remove(notify
);
964 bool migration_in_setup(MigrationState
*s
)
966 return s
->state
== MIGRATION_STATUS_SETUP
;
969 bool migration_has_finished(MigrationState
*s
)
971 return s
->state
== MIGRATION_STATUS_COMPLETED
;
974 bool migration_has_failed(MigrationState
*s
)
976 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
977 s
->state
== MIGRATION_STATUS_FAILED
);
980 bool migration_in_postcopy(MigrationState
*s
)
982 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
985 bool migration_in_postcopy_after_devices(MigrationState
*s
)
987 return migration_in_postcopy(s
) && s
->postcopy_after_devices
;
990 MigrationState
*migrate_init(const MigrationParams
*params
)
992 MigrationState
*s
= migrate_get_current();
995 * Reinitialise all migration state, except
996 * parameters/capabilities that the user set, and
1002 s
->to_dst_file
= NULL
;
1003 s
->state
= MIGRATION_STATUS_NONE
;
1004 s
->params
= *params
;
1005 s
->rp_state
.from_dst_file
= NULL
;
1006 s
->rp_state
.error
= false;
1009 s
->expected_downtime
= 0;
1010 s
->dirty_pages_rate
= 0;
1011 s
->dirty_bytes_rate
= 0;
1013 s
->dirty_sync_count
= 0;
1014 s
->start_postcopy
= false;
1015 s
->postcopy_after_devices
= false;
1016 s
->postcopy_requests
= 0;
1017 s
->migration_thread_running
= false;
1018 s
->last_req_rb
= NULL
;
1019 error_free(s
->error
);
1022 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1024 QSIMPLEQ_INIT(&s
->src_page_requests
);
1026 s
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1030 static GSList
*migration_blockers
;
1032 void migrate_add_blocker(Error
*reason
)
1034 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1037 void migrate_del_blocker(Error
*reason
)
1039 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1042 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1044 Error
*local_err
= NULL
;
1045 static bool once
= true;
1047 if (!deferred_incoming
) {
1048 error_setg(errp
, "For use with '-incoming defer'");
1052 error_setg(errp
, "The incoming migration has already been started");
1055 qemu_start_incoming_migration(uri
, &local_err
);
1058 error_propagate(errp
, local_err
);
1065 bool migration_is_blocked(Error
**errp
)
1067 if (qemu_savevm_state_blocked(errp
)) {
1071 if (migration_blockers
) {
1072 *errp
= error_copy(migration_blockers
->data
);
1079 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1080 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1083 Error
*local_err
= NULL
;
1084 MigrationState
*s
= migrate_get_current();
1085 MigrationParams params
;
1088 params
.blk
= has_blk
&& blk
;
1089 params
.shared
= has_inc
&& inc
;
1091 if (migration_is_setup_or_active(s
->state
) ||
1092 s
->state
== MIGRATION_STATUS_CANCELLING
) {
1093 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1096 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1097 error_setg(errp
, "Guest is waiting for an incoming migration");
1101 if (migration_is_blocked(errp
)) {
1105 s
= migrate_init(¶ms
);
1107 if (strstart(uri
, "tcp:", &p
)) {
1108 tcp_start_outgoing_migration(s
, p
, &local_err
);
1110 } else if (strstart(uri
, "rdma:", &p
)) {
1111 rdma_start_outgoing_migration(s
, p
, &local_err
);
1113 } else if (strstart(uri
, "exec:", &p
)) {
1114 exec_start_outgoing_migration(s
, p
, &local_err
);
1115 } else if (strstart(uri
, "unix:", &p
)) {
1116 unix_start_outgoing_migration(s
, p
, &local_err
);
1117 } else if (strstart(uri
, "fd:", &p
)) {
1118 fd_start_outgoing_migration(s
, p
, &local_err
);
1120 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1121 "a valid migration protocol");
1122 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1123 MIGRATION_STATUS_FAILED
);
1128 migrate_fd_error(s
, local_err
);
1129 error_propagate(errp
, local_err
);
1134 void qmp_migrate_cancel(Error
**errp
)
1136 migrate_fd_cancel(migrate_get_current());
1139 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1141 MigrationState
*s
= migrate_get_current();
1144 /* Check for truncation */
1145 if (value
!= (size_t)value
) {
1146 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1147 "exceeding address space");
1151 /* Cache should not be larger than guest ram size */
1152 if (value
> ram_bytes_total()) {
1153 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1154 "exceeds guest ram size ");
1158 new_size
= xbzrle_cache_resize(value
);
1160 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1161 "is smaller than page size");
1165 s
->xbzrle_cache_size
= new_size
;
1168 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1170 return migrate_xbzrle_cache_size();
1173 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1180 if (value
> SIZE_MAX
) {
1184 s
= migrate_get_current();
1185 s
->bandwidth_limit
= value
;
1186 if (s
->to_dst_file
) {
1187 qemu_file_set_rate_limit(s
->to_dst_file
,
1188 s
->bandwidth_limit
/ XFER_LIMIT_RATIO
);
1192 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1195 value
= MAX(0, MIN(UINT64_MAX
, value
));
1196 max_downtime
= (uint64_t)value
;
1199 bool migrate_postcopy_ram(void)
1203 s
= migrate_get_current();
1205 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1208 bool migrate_auto_converge(void)
1212 s
= migrate_get_current();
1214 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1217 bool migrate_zero_blocks(void)
1221 s
= migrate_get_current();
1223 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
1226 bool migrate_use_compression(void)
1230 s
= migrate_get_current();
1232 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
1235 int migrate_compress_level(void)
1239 s
= migrate_get_current();
1241 return s
->parameters
.compress_level
;
1244 int migrate_compress_threads(void)
1248 s
= migrate_get_current();
1250 return s
->parameters
.compress_threads
;
1253 int migrate_decompress_threads(void)
1257 s
= migrate_get_current();
1259 return s
->parameters
.decompress_threads
;
1262 bool migrate_use_events(void)
1266 s
= migrate_get_current();
1268 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
1271 int migrate_use_xbzrle(void)
1275 s
= migrate_get_current();
1277 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
1280 int64_t migrate_xbzrle_cache_size(void)
1284 s
= migrate_get_current();
1286 return s
->xbzrle_cache_size
;
1289 /* migration thread support */
1291 * Something bad happened to the RP stream, mark an error
1292 * The caller shall print or trace something to indicate why
1294 static void mark_source_rp_bad(MigrationState
*s
)
1296 s
->rp_state
.error
= true;
1299 static struct rp_cmd_args
{
1300 ssize_t len
; /* -1 = variable */
1303 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1304 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1305 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1306 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1307 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1308 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1312 * Process a request for pages received on the return path,
1313 * We're allowed to send more than requested (e.g. to round to our page size)
1314 * and we don't need to send pages that have already been sent.
1316 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1317 ram_addr_t start
, size_t len
)
1319 long our_host_ps
= getpagesize();
1321 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1324 * Since we currently insist on matching page sizes, just sanity check
1325 * we're being asked for whole host pages.
1327 if (start
& (our_host_ps
-1) ||
1328 (len
& (our_host_ps
-1))) {
1329 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1330 " len: %zd", __func__
, start
, len
);
1331 mark_source_rp_bad(ms
);
1335 if (ram_save_queue_pages(ms
, rbname
, start
, len
)) {
1336 mark_source_rp_bad(ms
);
1341 * Handles messages sent on the return path towards the source VM
1344 static void *source_return_path_thread(void *opaque
)
1346 MigrationState
*ms
= opaque
;
1347 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1348 uint16_t header_len
, header_type
;
1350 uint32_t tmp32
, sibling_error
;
1351 ram_addr_t start
= 0; /* =0 to silence warning */
1352 size_t len
= 0, expected_len
;
1355 trace_source_return_path_thread_entry();
1356 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1357 migration_is_setup_or_active(ms
->state
)) {
1358 trace_source_return_path_thread_loop_top();
1359 header_type
= qemu_get_be16(rp
);
1360 header_len
= qemu_get_be16(rp
);
1362 if (header_type
>= MIG_RP_MSG_MAX
||
1363 header_type
== MIG_RP_MSG_INVALID
) {
1364 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1365 header_type
, header_len
);
1366 mark_source_rp_bad(ms
);
1370 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1371 header_len
!= rp_cmd_args
[header_type
].len
) ||
1372 header_len
> sizeof(buf
)) {
1373 error_report("RP: Received '%s' message (0x%04x) with"
1374 "incorrect length %d expecting %zu",
1375 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1376 (size_t)rp_cmd_args
[header_type
].len
);
1377 mark_source_rp_bad(ms
);
1381 /* We know we've got a valid header by this point */
1382 res
= qemu_get_buffer(rp
, buf
, header_len
);
1383 if (res
!= header_len
) {
1384 error_report("RP: Failed reading data for message 0x%04x"
1385 " read %d expected %d",
1386 header_type
, res
, header_len
);
1387 mark_source_rp_bad(ms
);
1391 /* OK, we have the message and the data */
1392 switch (header_type
) {
1393 case MIG_RP_MSG_SHUT
:
1394 sibling_error
= ldl_be_p(buf
);
1395 trace_source_return_path_thread_shut(sibling_error
);
1396 if (sibling_error
) {
1397 error_report("RP: Sibling indicated error %d", sibling_error
);
1398 mark_source_rp_bad(ms
);
1401 * We'll let the main thread deal with closing the RP
1402 * we could do a shutdown(2) on it, but we're the only user
1403 * anyway, so there's nothing gained.
1407 case MIG_RP_MSG_PONG
:
1408 tmp32
= ldl_be_p(buf
);
1409 trace_source_return_path_thread_pong(tmp32
);
1412 case MIG_RP_MSG_REQ_PAGES
:
1413 start
= ldq_be_p(buf
);
1414 len
= ldl_be_p(buf
+ 8);
1415 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1418 case MIG_RP_MSG_REQ_PAGES_ID
:
1419 expected_len
= 12 + 1; /* header + termination */
1421 if (header_len
>= expected_len
) {
1422 start
= ldq_be_p(buf
);
1423 len
= ldl_be_p(buf
+ 8);
1424 /* Now we expect an idstr */
1425 tmp32
= buf
[12]; /* Length of the following idstr */
1426 buf
[13 + tmp32
] = '\0';
1427 expected_len
+= tmp32
;
1429 if (header_len
!= expected_len
) {
1430 error_report("RP: Req_Page_id with length %d expecting %zd",
1431 header_len
, expected_len
);
1432 mark_source_rp_bad(ms
);
1435 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
1442 if (qemu_file_get_error(rp
)) {
1443 trace_source_return_path_thread_bad_end();
1444 mark_source_rp_bad(ms
);
1447 trace_source_return_path_thread_end();
1449 ms
->rp_state
.from_dst_file
= NULL
;
1454 static int open_return_path_on_source(MigrationState
*ms
)
1457 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
1458 if (!ms
->rp_state
.from_dst_file
) {
1462 trace_open_return_path_on_source();
1463 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
1464 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
1466 trace_open_return_path_on_source_continue();
1471 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1472 static int await_return_path_close_on_source(MigrationState
*ms
)
1475 * If this is a normal exit then the destination will send a SHUT and the
1476 * rp_thread will exit, however if there's an error we need to cause
1479 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
1481 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1482 * waiting for the destination.
1484 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
1485 mark_source_rp_bad(ms
);
1487 trace_await_return_path_close_on_source_joining();
1488 qemu_thread_join(&ms
->rp_state
.rp_thread
);
1489 trace_await_return_path_close_on_source_close();
1490 return ms
->rp_state
.error
;
1494 * Switch from normal iteration to postcopy
1495 * Returns non-0 on error
1497 static int postcopy_start(MigrationState
*ms
, bool *old_vm_running
)
1500 QIOChannelBuffer
*bioc
;
1502 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1503 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
1504 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1506 trace_postcopy_start();
1507 qemu_mutex_lock_iothread();
1508 trace_postcopy_start_set_run();
1510 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1511 *old_vm_running
= runstate_is_running();
1512 global_state_store();
1513 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1518 ret
= bdrv_inactivate_all();
1524 * Cause any non-postcopiable, but iterative devices to
1525 * send out their final data.
1527 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true);
1530 * in Finish migrate and with the io-lock held everything should
1531 * be quiet, but we've potentially still got dirty pages and we
1532 * need to tell the destination to throw any pages it's already received
1535 if (ram_postcopy_send_discard_bitmap(ms
)) {
1536 error_report("postcopy send discard bitmap failed");
1541 * send rest of state - note things that are doing postcopy
1542 * will notice we're in POSTCOPY_ACTIVE and not actually
1543 * wrap their state up here
1545 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
1546 /* Ping just for debugging, helps line traces up */
1547 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
1550 * While loading the device state we may trigger page transfer
1551 * requests and the fd must be free to process those, and thus
1552 * the destination must read the whole device state off the fd before
1553 * it starts processing it. Unfortunately the ad-hoc migration format
1554 * doesn't allow the destination to know the size to read without fully
1555 * parsing it through each devices load-state code (especially the open
1556 * coded devices that use get/put).
1557 * So we wrap the device state up in a package with a length at the start;
1558 * to do this we use a qemu_buf to hold the whole of the device state.
1560 bioc
= qio_channel_buffer_new(4096);
1561 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
1562 object_unref(OBJECT(bioc
));
1565 * Make sure the receiver can get incoming pages before we send the rest
1568 qemu_savevm_send_postcopy_listen(fb
);
1570 qemu_savevm_state_complete_precopy(fb
, false);
1571 qemu_savevm_send_ping(fb
, 3);
1573 qemu_savevm_send_postcopy_run(fb
);
1575 /* <><> end of stuff going into the package */
1577 /* Now send that blob */
1578 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
1583 /* Send a notify to give a chance for anything that needs to happen
1584 * at the transition to postcopy and after the device state; in particular
1585 * spice needs to trigger a transition now
1587 ms
->postcopy_after_devices
= true;
1588 notifier_list_notify(&migration_state_notifiers
, ms
);
1590 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
1592 qemu_mutex_unlock_iothread();
1595 * Although this ping is just for debug, it could potentially be
1596 * used for getting a better measurement of downtime at the source.
1598 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
1600 ret
= qemu_file_get_error(ms
->to_dst_file
);
1602 error_report("postcopy_start: Migration stream errored");
1603 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1604 MIGRATION_STATUS_FAILED
);
1612 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1613 MIGRATION_STATUS_FAILED
);
1614 qemu_mutex_unlock_iothread();
1619 * migration_completion: Used by migration_thread when there's not much left.
1620 * The caller 'breaks' the loop when this returns.
1622 * @s: Current migration state
1623 * @current_active_state: The migration state we expect to be in
1624 * @*old_vm_running: Pointer to old_vm_running flag
1625 * @*start_time: Pointer to time to update
1627 static void migration_completion(MigrationState
*s
, int current_active_state
,
1628 bool *old_vm_running
,
1629 int64_t *start_time
)
1633 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1634 qemu_mutex_lock_iothread();
1635 *start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1636 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1637 *old_vm_running
= runstate_is_running();
1638 ret
= global_state_store();
1641 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1643 ret
= bdrv_inactivate_all();
1646 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
1647 qemu_savevm_state_complete_precopy(s
->to_dst_file
, false);
1650 qemu_mutex_unlock_iothread();
1655 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1656 trace_migration_completion_postcopy_end();
1658 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
1659 trace_migration_completion_postcopy_end_after_complete();
1663 * If rp was opened we must clean up the thread before
1664 * cleaning everything else up (since if there are no failures
1665 * it will wait for the destination to send it's status in
1667 * Postcopy opens rp if enabled (even if it's not avtivated)
1669 if (migrate_postcopy_ram()) {
1671 trace_migration_completion_postcopy_end_before_rp();
1672 rp_error
= await_return_path_close_on_source(s
);
1673 trace_migration_completion_postcopy_end_after_rp(rp_error
);
1675 goto fail_invalidate
;
1679 if (qemu_file_get_error(s
->to_dst_file
)) {
1680 trace_migration_completion_file_err();
1681 goto fail_invalidate
;
1684 migrate_set_state(&s
->state
, current_active_state
,
1685 MIGRATION_STATUS_COMPLETED
);
1689 /* If not doing postcopy, vm_start() will be called: let's regain
1690 * control on images.
1692 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1693 Error
*local_err
= NULL
;
1695 bdrv_invalidate_cache_all(&local_err
);
1697 error_report_err(local_err
);
1702 migrate_set_state(&s
->state
, current_active_state
,
1703 MIGRATION_STATUS_FAILED
);
1707 * Master migration thread on the source VM.
1708 * It drives the migration and pumps the data down the outgoing channel.
1710 static void *migration_thread(void *opaque
)
1712 MigrationState
*s
= opaque
;
1713 /* Used by the bandwidth calcs, updated later */
1714 int64_t initial_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1715 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
1716 int64_t initial_bytes
= 0;
1717 int64_t max_size
= 0;
1718 int64_t start_time
= initial_time
;
1720 bool old_vm_running
= false;
1721 bool entered_postcopy
= false;
1722 /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
1723 enum MigrationStatus current_active_state
= MIGRATION_STATUS_ACTIVE
;
1725 rcu_register_thread();
1727 qemu_savevm_state_header(s
->to_dst_file
);
1729 if (migrate_postcopy_ram()) {
1730 /* Now tell the dest that it should open its end so it can reply */
1731 qemu_savevm_send_open_return_path(s
->to_dst_file
);
1733 /* And do a ping that will make stuff easier to debug */
1734 qemu_savevm_send_ping(s
->to_dst_file
, 1);
1737 * Tell the destination that we *might* want to do postcopy later;
1738 * if the other end can't do postcopy it should fail now, nice and
1741 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
1744 qemu_savevm_state_begin(s
->to_dst_file
, &s
->params
);
1746 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
1747 current_active_state
= MIGRATION_STATUS_ACTIVE
;
1748 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1749 MIGRATION_STATUS_ACTIVE
);
1751 trace_migration_thread_setup_complete();
1753 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
1754 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1755 int64_t current_time
;
1756 uint64_t pending_size
;
1758 if (!qemu_file_rate_limit(s
->to_dst_file
)) {
1759 uint64_t pend_post
, pend_nonpost
;
1761 qemu_savevm_state_pending(s
->to_dst_file
, max_size
, &pend_nonpost
,
1763 pending_size
= pend_nonpost
+ pend_post
;
1764 trace_migrate_pending(pending_size
, max_size
,
1765 pend_post
, pend_nonpost
);
1766 if (pending_size
&& pending_size
>= max_size
) {
1767 /* Still a significant amount to transfer */
1769 if (migrate_postcopy_ram() &&
1770 s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
&&
1771 pend_nonpost
<= max_size
&&
1772 atomic_read(&s
->start_postcopy
)) {
1774 if (!postcopy_start(s
, &old_vm_running
)) {
1775 current_active_state
= MIGRATION_STATUS_POSTCOPY_ACTIVE
;
1776 entered_postcopy
= true;
1781 /* Just another iteration step */
1782 qemu_savevm_state_iterate(s
->to_dst_file
, entered_postcopy
);
1784 trace_migration_thread_low_pending(pending_size
);
1785 migration_completion(s
, current_active_state
,
1786 &old_vm_running
, &start_time
);
1791 if (qemu_file_get_error(s
->to_dst_file
)) {
1792 migrate_set_state(&s
->state
, current_active_state
,
1793 MIGRATION_STATUS_FAILED
);
1794 trace_migration_thread_file_err();
1797 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1798 if (current_time
>= initial_time
+ BUFFER_DELAY
) {
1799 uint64_t transferred_bytes
= qemu_ftell(s
->to_dst_file
) -
1801 uint64_t time_spent
= current_time
- initial_time
;
1802 double bandwidth
= (double)transferred_bytes
/ time_spent
;
1803 max_size
= bandwidth
* migrate_max_downtime() / 1000000;
1805 s
->mbps
= (((double) transferred_bytes
* 8.0) /
1806 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
1808 trace_migrate_transferred(transferred_bytes
, time_spent
,
1809 bandwidth
, max_size
);
1810 /* if we haven't sent anything, we don't want to recalculate
1811 10000 is a small enough number for our purposes */
1812 if (s
->dirty_bytes_rate
&& transferred_bytes
> 10000) {
1813 s
->expected_downtime
= s
->dirty_bytes_rate
/ bandwidth
;
1816 qemu_file_reset_rate_limit(s
->to_dst_file
);
1817 initial_time
= current_time
;
1818 initial_bytes
= qemu_ftell(s
->to_dst_file
);
1820 if (qemu_file_rate_limit(s
->to_dst_file
)) {
1821 /* usleep expects microseconds */
1822 g_usleep((initial_time
+ BUFFER_DELAY
- current_time
)*1000);
1826 trace_migration_thread_after_loop();
1827 /* If we enabled cpu throttling for auto-converge, turn it off. */
1828 cpu_throttle_stop();
1829 end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1831 qemu_mutex_lock_iothread();
1832 qemu_savevm_state_cleanup();
1833 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
1834 uint64_t transferred_bytes
= qemu_ftell(s
->to_dst_file
);
1835 s
->total_time
= end_time
- s
->total_time
;
1836 if (!entered_postcopy
) {
1837 s
->downtime
= end_time
- start_time
;
1839 if (s
->total_time
) {
1840 s
->mbps
= (((double) transferred_bytes
* 8.0) /
1841 ((double) s
->total_time
)) / 1000;
1843 runstate_set(RUN_STATE_POSTMIGRATE
);
1845 if (old_vm_running
&& !entered_postcopy
) {
1848 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
1849 runstate_set(RUN_STATE_POSTMIGRATE
);
1853 qemu_bh_schedule(s
->cleanup_bh
);
1854 qemu_mutex_unlock_iothread();
1856 rcu_unregister_thread();
1860 void migrate_fd_connect(MigrationState
*s
)
1862 /* This is a best 1st approximation. ns to ms */
1863 s
->expected_downtime
= max_downtime
/1000000;
1864 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
1866 qemu_file_set_blocking(s
->to_dst_file
, true);
1867 qemu_file_set_rate_limit(s
->to_dst_file
,
1868 s
->bandwidth_limit
/ XFER_LIMIT_RATIO
);
1870 /* Notify before starting migration thread */
1871 notifier_list_notify(&migration_state_notifiers
, s
);
1874 * Open the return path; currently for postcopy but other things might
1877 if (migrate_postcopy_ram()) {
1878 if (open_return_path_on_source(s
)) {
1879 error_report("Unable to open return-path for postcopy");
1880 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1881 MIGRATION_STATUS_FAILED
);
1882 migrate_fd_cleanup(s
);
1887 migrate_compress_threads_create();
1888 qemu_thread_create(&s
->thread
, "migration", migration_thread
, s
,
1889 QEMU_THREAD_JOINABLE
);
1890 s
->migration_thread_running
= true;
1893 PostcopyState
postcopy_state_get(void)
1895 return atomic_mb_read(&incoming_postcopy_state
);
1898 /* Set the state and return the old state */
1899 PostcopyState
postcopy_state_set(PostcopyState new_state
)
1901 return atomic_xchg(&incoming_postcopy_state
, new_state
);