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
);
113 current_migration
.parameters
.tls_creds
= g_strdup("");
114 current_migration
.parameters
.tls_hostname
= g_strdup("");
117 return ¤t_migration
;
120 MigrationIncomingState
*migration_incoming_get_current(void)
123 static MigrationIncomingState mis_current
;
126 mis_current
.state
= MIGRATION_STATUS_NONE
;
127 memset(&mis_current
, 0, sizeof(MigrationIncomingState
));
128 QLIST_INIT(&mis_current
.loadvm_handlers
);
129 qemu_mutex_init(&mis_current
.rp_mutex
);
130 qemu_event_init(&mis_current
.main_thread_load_event
, false);
136 void migration_incoming_state_destroy(void)
138 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
140 qemu_event_destroy(&mis
->main_thread_load_event
);
141 loadvm_free_handlers(mis
);
148 uint8_t runstate
[100];
153 static GlobalState global_state
;
155 int global_state_store(void)
157 if (!runstate_store((char *)global_state
.runstate
,
158 sizeof(global_state
.runstate
))) {
159 error_report("runstate name too big: %s", global_state
.runstate
);
160 trace_migrate_state_too_big();
166 void global_state_store_running(void)
168 const char *state
= RunState_lookup
[RUN_STATE_RUNNING
];
169 strncpy((char *)global_state
.runstate
,
170 state
, sizeof(global_state
.runstate
));
173 static bool global_state_received(void)
175 return global_state
.received
;
178 static RunState
global_state_get_runstate(void)
180 return global_state
.state
;
183 void global_state_set_optional(void)
185 global_state
.optional
= true;
188 static bool global_state_needed(void *opaque
)
190 GlobalState
*s
= opaque
;
191 char *runstate
= (char *)s
->runstate
;
193 /* If it is not optional, it is mandatory */
195 if (s
->optional
== false) {
199 /* If state is running or paused, it is not needed */
201 if (strcmp(runstate
, "running") == 0 ||
202 strcmp(runstate
, "paused") == 0) {
206 /* for any other state it is needed */
210 static int global_state_post_load(void *opaque
, int version_id
)
212 GlobalState
*s
= opaque
;
213 Error
*local_err
= NULL
;
215 char *runstate
= (char *)s
->runstate
;
218 trace_migrate_global_state_post_load(runstate
);
220 r
= qapi_enum_parse(RunState_lookup
, runstate
, RUN_STATE__MAX
,
225 error_report_err(local_err
);
234 static void global_state_pre_save(void *opaque
)
236 GlobalState
*s
= opaque
;
238 trace_migrate_global_state_pre_save((char *)s
->runstate
);
239 s
->size
= strlen((char *)s
->runstate
) + 1;
242 static const VMStateDescription vmstate_globalstate
= {
243 .name
= "globalstate",
245 .minimum_version_id
= 1,
246 .post_load
= global_state_post_load
,
247 .pre_save
= global_state_pre_save
,
248 .needed
= global_state_needed
,
249 .fields
= (VMStateField
[]) {
250 VMSTATE_UINT32(size
, GlobalState
),
251 VMSTATE_BUFFER(runstate
, GlobalState
),
252 VMSTATE_END_OF_LIST()
256 void register_global_state(void)
258 /* We would use it independently that we receive it */
259 strcpy((char *)&global_state
.runstate
, "");
260 global_state
.received
= false;
261 vmstate_register(NULL
, 0, &vmstate_globalstate
, &global_state
);
264 static void migrate_generate_event(int new_state
)
266 if (migrate_use_events()) {
267 qapi_event_send_migration(new_state
, &error_abort
);
272 * Called on -incoming with a defer: uri.
273 * The migration can be started later after any parameters have been
276 static void deferred_incoming_migration(Error
**errp
)
278 if (deferred_incoming
) {
279 error_setg(errp
, "Incoming migration already deferred");
281 deferred_incoming
= true;
284 /* Request a range of pages from the source VM at the given
286 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
287 * as the last request (a name must have been given previously)
288 * Start: Address offset within the RB
289 * Len: Length in bytes required - must be a multiple of pagesize
291 void migrate_send_rp_req_pages(MigrationIncomingState
*mis
, const char *rbname
,
292 ram_addr_t start
, size_t len
)
294 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
295 size_t msglen
= 12; /* start + len */
297 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
298 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
301 int rbname_len
= strlen(rbname
);
302 assert(rbname_len
< 256);
304 bufc
[msglen
++] = rbname_len
;
305 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
306 msglen
+= rbname_len
;
307 migrate_send_rp_message(mis
, MIG_RP_MSG_REQ_PAGES_ID
, msglen
, bufc
);
309 migrate_send_rp_message(mis
, MIG_RP_MSG_REQ_PAGES
, msglen
, bufc
);
313 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
317 qapi_event_send_migration(MIGRATION_STATUS_SETUP
, &error_abort
);
318 if (!strcmp(uri
, "defer")) {
319 deferred_incoming_migration(errp
);
320 } else if (strstart(uri
, "tcp:", &p
)) {
321 tcp_start_incoming_migration(p
, errp
);
323 } else if (strstart(uri
, "rdma:", &p
)) {
324 rdma_start_incoming_migration(p
, errp
);
326 } else if (strstart(uri
, "exec:", &p
)) {
327 exec_start_incoming_migration(p
, errp
);
328 } else if (strstart(uri
, "unix:", &p
)) {
329 unix_start_incoming_migration(p
, errp
);
330 } else if (strstart(uri
, "fd:", &p
)) {
331 fd_start_incoming_migration(p
, errp
);
333 error_setg(errp
, "unknown migration protocol: %s", uri
);
337 static void process_incoming_migration_bh(void *opaque
)
339 Error
*local_err
= NULL
;
340 MigrationIncomingState
*mis
= opaque
;
342 /* Make sure all file formats flush their mutable metadata */
343 bdrv_invalidate_cache_all(&local_err
);
345 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
346 MIGRATION_STATUS_FAILED
);
347 error_report_err(local_err
);
348 migrate_decompress_threads_join();
352 /* If we get an error here, just don't restart the VM yet. */
353 blk_resume_after_migration(&local_err
);
355 error_free(local_err
);
361 * This must happen after all error conditions are dealt with and
362 * we're sure the VM is going to be running on this host.
364 qemu_announce_self();
366 /* If global state section was not received or we are in running
367 state, we need to obey autostart. Any other state is set with
370 if (!global_state_received() ||
371 global_state_get_runstate() == RUN_STATE_RUNNING
) {
375 runstate_set(RUN_STATE_PAUSED
);
378 runstate_set(global_state_get_runstate());
380 migrate_decompress_threads_join();
382 * This must happen after any state changes since as soon as an external
383 * observer sees this event they might start to prod at the VM assuming
386 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
387 MIGRATION_STATUS_COMPLETED
);
388 qemu_bh_delete(mis
->bh
);
389 migration_incoming_state_destroy();
392 static void process_incoming_migration_co(void *opaque
)
394 QEMUFile
*f
= opaque
;
395 MigrationIncomingState
*mis
= migration_incoming_get_current();
399 mis
->from_src_file
= f
;
400 mis
->largest_page_size
= qemu_ram_pagesize_largest();
401 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
402 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
403 MIGRATION_STATUS_ACTIVE
);
404 ret
= qemu_loadvm_state(f
);
406 ps
= postcopy_state_get();
407 trace_process_incoming_migration_co_end(ret
, ps
);
408 if (ps
!= POSTCOPY_INCOMING_NONE
) {
409 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
411 * Where a migration had postcopy enabled (and thus went to advise)
412 * but managed to complete within the precopy period, we can use
415 postcopy_ram_incoming_cleanup(mis
);
416 } else if (ret
>= 0) {
418 * Postcopy was started, cleanup should happen at the end of the
421 trace_process_incoming_migration_co_postcopy_end_main();
424 /* Else if something went wrong then just fall out of the normal exit */
427 /* we get COLO info, and know if we are in COLO mode */
428 if (!ret
&& migration_incoming_enable_colo()) {
429 mis
->migration_incoming_co
= qemu_coroutine_self();
430 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
431 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
432 mis
->have_colo_incoming_thread
= true;
433 qemu_coroutine_yield();
435 /* Wait checkpoint incoming thread exit before free resource */
436 qemu_thread_join(&mis
->colo_incoming_thread
);
440 free_xbzrle_decoded_buf();
443 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
444 MIGRATION_STATUS_FAILED
);
445 error_report("load of migration failed: %s", strerror(-ret
));
446 migrate_decompress_threads_join();
450 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
451 qemu_bh_schedule(mis
->bh
);
454 void migration_fd_process_incoming(QEMUFile
*f
)
456 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, f
);
458 migrate_decompress_threads_create();
459 qemu_file_set_blocking(f
, false);
460 qemu_coroutine_enter(co
);
464 void migration_channel_process_incoming(MigrationState
*s
,
467 trace_migration_set_incoming_channel(
468 ioc
, object_get_typename(OBJECT(ioc
)));
470 if (s
->parameters
.tls_creds
&&
471 *s
->parameters
.tls_creds
&&
472 !object_dynamic_cast(OBJECT(ioc
),
473 TYPE_QIO_CHANNEL_TLS
)) {
474 Error
*local_err
= NULL
;
475 migration_tls_channel_process_incoming(s
, ioc
, &local_err
);
477 error_report_err(local_err
);
480 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
481 migration_fd_process_incoming(f
);
486 void migration_channel_connect(MigrationState
*s
,
488 const char *hostname
)
490 trace_migration_set_outgoing_channel(
491 ioc
, object_get_typename(OBJECT(ioc
)), hostname
);
493 if (s
->parameters
.tls_creds
&&
494 *s
->parameters
.tls_creds
&&
495 !object_dynamic_cast(OBJECT(ioc
),
496 TYPE_QIO_CHANNEL_TLS
)) {
497 Error
*local_err
= NULL
;
498 migration_tls_channel_connect(s
, ioc
, hostname
, &local_err
);
500 migrate_fd_error(s
, local_err
);
501 error_free(local_err
);
504 QEMUFile
*f
= qemu_fopen_channel_output(ioc
);
508 migrate_fd_connect(s
);
514 * Send a message on the return channel back to the source
517 void migrate_send_rp_message(MigrationIncomingState
*mis
,
518 enum mig_rp_message_type message_type
,
519 uint16_t len
, void *data
)
521 trace_migrate_send_rp_message((int)message_type
, len
);
522 qemu_mutex_lock(&mis
->rp_mutex
);
523 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
524 qemu_put_be16(mis
->to_src_file
, len
);
525 qemu_put_buffer(mis
->to_src_file
, data
, len
);
526 qemu_fflush(mis
->to_src_file
);
527 qemu_mutex_unlock(&mis
->rp_mutex
);
531 * Send a 'SHUT' message on the return channel with the given value
532 * to indicate that we've finished with the RP. Non-0 value indicates
535 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
540 buf
= cpu_to_be32(value
);
541 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
545 * Send a 'PONG' message on the return channel with the given value
546 * (normally in response to a 'PING')
548 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
553 buf
= cpu_to_be32(value
);
554 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
557 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
559 MigrationCapabilityStatusList
*head
= NULL
;
560 MigrationCapabilityStatusList
*caps
;
561 MigrationState
*s
= migrate_get_current();
564 caps
= NULL
; /* silence compiler warning */
565 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
566 if (i
== MIGRATION_CAPABILITY_X_COLO
&& !colo_supported()) {
570 head
= g_malloc0(sizeof(*caps
));
573 caps
->next
= g_malloc0(sizeof(*caps
));
577 g_malloc(sizeof(*caps
->value
));
578 caps
->value
->capability
= i
;
579 caps
->value
->state
= s
->enabled_capabilities
[i
];
585 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
587 MigrationParameters
*params
;
588 MigrationState
*s
= migrate_get_current();
590 params
= g_malloc0(sizeof(*params
));
591 params
->has_compress_level
= true;
592 params
->compress_level
= s
->parameters
.compress_level
;
593 params
->has_compress_threads
= true;
594 params
->compress_threads
= s
->parameters
.compress_threads
;
595 params
->has_decompress_threads
= true;
596 params
->decompress_threads
= s
->parameters
.decompress_threads
;
597 params
->has_cpu_throttle_initial
= true;
598 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
599 params
->has_cpu_throttle_increment
= true;
600 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
601 params
->has_tls_creds
= !!s
->parameters
.tls_creds
;
602 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
603 params
->has_tls_hostname
= !!s
->parameters
.tls_hostname
;
604 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
605 params
->has_max_bandwidth
= true;
606 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
607 params
->has_downtime_limit
= true;
608 params
->downtime_limit
= s
->parameters
.downtime_limit
;
609 params
->has_x_checkpoint_delay
= true;
610 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
616 * Return true if we're already in the middle of a migration
617 * (i.e. any of the active or setup states)
619 static bool migration_is_setup_or_active(int state
)
622 case MIGRATION_STATUS_ACTIVE
:
623 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
624 case MIGRATION_STATUS_SETUP
:
633 static void get_xbzrle_cache_stats(MigrationInfo
*info
)
635 if (migrate_use_xbzrle()) {
636 info
->has_xbzrle_cache
= true;
637 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
638 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
639 info
->xbzrle_cache
->bytes
= xbzrle_mig_bytes_transferred();
640 info
->xbzrle_cache
->pages
= xbzrle_mig_pages_transferred();
641 info
->xbzrle_cache
->cache_miss
= xbzrle_mig_pages_cache_miss();
642 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_mig_cache_miss_rate();
643 info
->xbzrle_cache
->overflow
= xbzrle_mig_pages_overflow();
647 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
649 info
->has_ram
= true;
650 info
->ram
= g_malloc0(sizeof(*info
->ram
));
651 info
->ram
->transferred
= ram_bytes_transferred();
652 info
->ram
->total
= ram_bytes_total();
653 info
->ram
->duplicate
= dup_mig_pages_transferred();
654 info
->ram
->skipped
= skipped_mig_pages_transferred();
655 info
->ram
->normal
= norm_mig_pages_transferred();
656 info
->ram
->normal_bytes
= norm_mig_bytes_transferred();
657 info
->ram
->mbps
= s
->mbps
;
658 info
->ram
->dirty_sync_count
= s
->dirty_sync_count
;
659 info
->ram
->postcopy_requests
= s
->postcopy_requests
;
661 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
662 info
->ram
->remaining
= ram_bytes_remaining();
663 info
->ram
->dirty_pages_rate
= s
->dirty_pages_rate
;
667 MigrationInfo
*qmp_query_migrate(Error
**errp
)
669 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
670 MigrationState
*s
= migrate_get_current();
673 case MIGRATION_STATUS_NONE
:
674 /* no migration has happened ever */
676 case MIGRATION_STATUS_SETUP
:
677 info
->has_status
= true;
678 info
->has_total_time
= false;
680 case MIGRATION_STATUS_ACTIVE
:
681 case MIGRATION_STATUS_CANCELLING
:
682 info
->has_status
= true;
683 info
->has_total_time
= true;
684 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
686 info
->has_expected_downtime
= true;
687 info
->expected_downtime
= s
->expected_downtime
;
688 info
->has_setup_time
= true;
689 info
->setup_time
= s
->setup_time
;
691 populate_ram_info(info
, s
);
693 if (blk_mig_active()) {
694 info
->has_disk
= true;
695 info
->disk
= g_malloc0(sizeof(*info
->disk
));
696 info
->disk
->transferred
= blk_mig_bytes_transferred();
697 info
->disk
->remaining
= blk_mig_bytes_remaining();
698 info
->disk
->total
= blk_mig_bytes_total();
701 if (cpu_throttle_active()) {
702 info
->has_cpu_throttle_percentage
= true;
703 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
706 get_xbzrle_cache_stats(info
);
708 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
709 /* Mostly the same as active; TODO add some postcopy stats */
710 info
->has_status
= true;
711 info
->has_total_time
= true;
712 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
714 info
->has_expected_downtime
= true;
715 info
->expected_downtime
= s
->expected_downtime
;
716 info
->has_setup_time
= true;
717 info
->setup_time
= s
->setup_time
;
719 populate_ram_info(info
, s
);
721 if (blk_mig_active()) {
722 info
->has_disk
= true;
723 info
->disk
= g_malloc0(sizeof(*info
->disk
));
724 info
->disk
->transferred
= blk_mig_bytes_transferred();
725 info
->disk
->remaining
= blk_mig_bytes_remaining();
726 info
->disk
->total
= blk_mig_bytes_total();
729 get_xbzrle_cache_stats(info
);
731 case MIGRATION_STATUS_COLO
:
732 info
->has_status
= true;
733 /* TODO: display COLO specific information (checkpoint info etc.) */
735 case MIGRATION_STATUS_COMPLETED
:
736 get_xbzrle_cache_stats(info
);
738 info
->has_status
= true;
739 info
->has_total_time
= true;
740 info
->total_time
= s
->total_time
;
741 info
->has_downtime
= true;
742 info
->downtime
= s
->downtime
;
743 info
->has_setup_time
= true;
744 info
->setup_time
= s
->setup_time
;
746 populate_ram_info(info
, s
);
748 case MIGRATION_STATUS_FAILED
:
749 info
->has_status
= true;
751 info
->has_error_desc
= true;
752 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
755 case MIGRATION_STATUS_CANCELLED
:
756 info
->has_status
= true;
759 info
->status
= s
->state
;
764 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
767 MigrationState
*s
= migrate_get_current();
768 MigrationCapabilityStatusList
*cap
;
769 bool old_postcopy_cap
= migrate_postcopy_ram();
771 if (migration_is_setup_or_active(s
->state
)) {
772 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
776 for (cap
= params
; cap
; cap
= cap
->next
) {
777 if (cap
->value
->capability
== MIGRATION_CAPABILITY_X_COLO
) {
778 if (!colo_supported()) {
779 error_setg(errp
, "COLO is not currently supported, please"
780 " configure with --enable-colo option in order to"
781 " support COLO feature");
785 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
788 if (migrate_postcopy_ram()) {
789 if (migrate_use_compression()) {
790 /* The decompression threads asynchronously write into RAM
791 * rather than use the atomic copies needed to avoid
792 * userfaulting. It should be possible to fix the decompression
793 * threads for compatibility in future.
795 error_report("Postcopy is not currently compatible with "
797 s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
] =
800 /* This check is reasonably expensive, so only when it's being
801 * set the first time, also it's only the destination that needs
804 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
805 !postcopy_ram_supported_by_host()) {
806 /* postcopy_ram_supported_by_host will have emitted a more
809 error_report("Postcopy is not supported");
810 s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
] =
816 void qmp_migrate_set_parameters(MigrationParameters
*params
, Error
**errp
)
818 MigrationState
*s
= migrate_get_current();
820 if (params
->has_compress_level
&&
821 (params
->compress_level
< 0 || params
->compress_level
> 9)) {
822 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
823 "is invalid, it should be in the range of 0 to 9");
826 if (params
->has_compress_threads
&&
827 (params
->compress_threads
< 1 || params
->compress_threads
> 255)) {
828 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
830 "is invalid, it should be in the range of 1 to 255");
833 if (params
->has_decompress_threads
&&
834 (params
->decompress_threads
< 1 || params
->decompress_threads
> 255)) {
835 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
836 "decompress_threads",
837 "is invalid, it should be in the range of 1 to 255");
840 if (params
->has_cpu_throttle_initial
&&
841 (params
->cpu_throttle_initial
< 1 ||
842 params
->cpu_throttle_initial
> 99)) {
843 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
844 "cpu_throttle_initial",
845 "an integer in the range of 1 to 99");
848 if (params
->has_cpu_throttle_increment
&&
849 (params
->cpu_throttle_increment
< 1 ||
850 params
->cpu_throttle_increment
> 99)) {
851 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
852 "cpu_throttle_increment",
853 "an integer in the range of 1 to 99");
856 if (params
->has_max_bandwidth
&&
857 (params
->max_bandwidth
< 0 || params
->max_bandwidth
> SIZE_MAX
)) {
858 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
859 " range of 0 to %zu bytes/second", SIZE_MAX
);
862 if (params
->has_downtime_limit
&&
863 (params
->downtime_limit
< 0 ||
864 params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
865 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
866 "the range of 0 to %d milliseconds",
867 MAX_MIGRATE_DOWNTIME
);
870 if (params
->has_x_checkpoint_delay
&& (params
->x_checkpoint_delay
< 0)) {
871 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
872 "x_checkpoint_delay",
873 "is invalid, it should be positive");
876 if (params
->has_compress_level
) {
877 s
->parameters
.compress_level
= params
->compress_level
;
879 if (params
->has_compress_threads
) {
880 s
->parameters
.compress_threads
= params
->compress_threads
;
882 if (params
->has_decompress_threads
) {
883 s
->parameters
.decompress_threads
= params
->decompress_threads
;
885 if (params
->has_cpu_throttle_initial
) {
886 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
888 if (params
->has_cpu_throttle_increment
) {
889 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
891 if (params
->has_tls_creds
) {
892 g_free(s
->parameters
.tls_creds
);
893 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
);
895 if (params
->has_tls_hostname
) {
896 g_free(s
->parameters
.tls_hostname
);
897 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
);
899 if (params
->has_max_bandwidth
) {
900 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
901 if (s
->to_dst_file
) {
902 qemu_file_set_rate_limit(s
->to_dst_file
,
903 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
906 if (params
->has_downtime_limit
) {
907 s
->parameters
.downtime_limit
= params
->downtime_limit
;
910 if (params
->has_x_checkpoint_delay
) {
911 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
912 if (migration_in_colo_state()) {
913 colo_checkpoint_notify(s
);
919 void qmp_migrate_start_postcopy(Error
**errp
)
921 MigrationState
*s
= migrate_get_current();
923 if (!migrate_postcopy_ram()) {
924 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
925 " the start of migration");
929 if (s
->state
== MIGRATION_STATUS_NONE
) {
930 error_setg(errp
, "Postcopy must be started after migration has been"
935 * we don't error if migration has finished since that would be racy
936 * with issuing this command.
938 atomic_set(&s
->start_postcopy
, true);
941 /* shared migration helpers */
943 void migrate_set_state(int *state
, int old_state
, int new_state
)
945 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
946 trace_migrate_set_state(new_state
);
947 migrate_generate_event(new_state
);
951 static void migrate_fd_cleanup(void *opaque
)
953 MigrationState
*s
= opaque
;
955 qemu_bh_delete(s
->cleanup_bh
);
956 s
->cleanup_bh
= NULL
;
960 if (s
->to_dst_file
) {
961 trace_migrate_fd_cleanup();
962 qemu_mutex_unlock_iothread();
963 if (s
->migration_thread_running
) {
964 qemu_thread_join(&s
->thread
);
965 s
->migration_thread_running
= false;
967 qemu_mutex_lock_iothread();
969 migrate_compress_threads_join();
970 qemu_fclose(s
->to_dst_file
);
971 s
->to_dst_file
= NULL
;
974 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
975 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
977 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
978 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
979 MIGRATION_STATUS_CANCELLED
);
982 notifier_list_notify(&migration_state_notifiers
, s
);
985 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
987 trace_migrate_fd_error(error_get_pretty(error
));
988 assert(s
->to_dst_file
== NULL
);
989 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
990 MIGRATION_STATUS_FAILED
);
992 s
->error
= error_copy(error
);
994 notifier_list_notify(&migration_state_notifiers
, s
);
997 static void migrate_fd_cancel(MigrationState
*s
)
1000 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1001 trace_migrate_fd_cancel();
1003 if (s
->rp_state
.from_dst_file
) {
1004 /* shutdown the rp socket, so causing the rp thread to shutdown */
1005 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1009 old_state
= s
->state
;
1010 if (!migration_is_setup_or_active(old_state
)) {
1013 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1014 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1017 * If we're unlucky the migration code might be stuck somewhere in a
1018 * send/write while the network has failed and is waiting to timeout;
1019 * if we've got shutdown(2) available then we can force it to quit.
1020 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1021 * called in a bh, so there is no race against this cancel.
1023 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1024 qemu_file_shutdown(f
);
1026 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1027 Error
*local_err
= NULL
;
1029 bdrv_invalidate_cache_all(&local_err
);
1031 error_report_err(local_err
);
1033 s
->block_inactive
= false;
1038 void add_migration_state_change_notifier(Notifier
*notify
)
1040 notifier_list_add(&migration_state_notifiers
, notify
);
1043 void remove_migration_state_change_notifier(Notifier
*notify
)
1045 notifier_remove(notify
);
1048 bool migration_in_setup(MigrationState
*s
)
1050 return s
->state
== MIGRATION_STATUS_SETUP
;
1053 bool migration_has_finished(MigrationState
*s
)
1055 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1058 bool migration_has_failed(MigrationState
*s
)
1060 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1061 s
->state
== MIGRATION_STATUS_FAILED
);
1064 bool migration_in_postcopy(MigrationState
*s
)
1066 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1069 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1071 return migration_in_postcopy(s
) && s
->postcopy_after_devices
;
1074 bool migration_is_idle(MigrationState
*s
)
1077 s
= migrate_get_current();
1081 case MIGRATION_STATUS_NONE
:
1082 case MIGRATION_STATUS_CANCELLED
:
1083 case MIGRATION_STATUS_COMPLETED
:
1084 case MIGRATION_STATUS_FAILED
:
1086 case MIGRATION_STATUS_SETUP
:
1087 case MIGRATION_STATUS_CANCELLING
:
1088 case MIGRATION_STATUS_ACTIVE
:
1089 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1090 case MIGRATION_STATUS_COLO
:
1092 case MIGRATION_STATUS__MAX
:
1093 g_assert_not_reached();
1099 MigrationState
*migrate_init(const MigrationParams
*params
)
1101 MigrationState
*s
= migrate_get_current();
1104 * Reinitialise all migration state, except
1105 * parameters/capabilities that the user set, and
1111 s
->to_dst_file
= NULL
;
1112 s
->state
= MIGRATION_STATUS_NONE
;
1113 s
->params
= *params
;
1114 s
->rp_state
.from_dst_file
= NULL
;
1115 s
->rp_state
.error
= false;
1118 s
->expected_downtime
= 0;
1119 s
->dirty_pages_rate
= 0;
1120 s
->dirty_bytes_rate
= 0;
1122 s
->dirty_sync_count
= 0;
1123 s
->start_postcopy
= false;
1124 s
->postcopy_after_devices
= false;
1125 s
->postcopy_requests
= 0;
1126 s
->migration_thread_running
= false;
1127 s
->last_req_rb
= NULL
;
1128 error_free(s
->error
);
1131 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1133 QSIMPLEQ_INIT(&s
->src_page_requests
);
1135 s
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1139 static GSList
*migration_blockers
;
1141 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1143 if (only_migratable
) {
1144 error_propagate(errp
, error_copy(reason
));
1145 error_prepend(errp
, "disallowing migration blocker "
1146 "(--only_migratable) for: ");
1150 if (migration_is_idle(NULL
)) {
1151 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1155 error_propagate(errp
, error_copy(reason
));
1156 error_prepend(errp
, "disallowing migration blocker (migration in "
1161 void migrate_del_blocker(Error
*reason
)
1163 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1166 int check_migratable(Object
*obj
, Error
**err
)
1168 DeviceClass
*dc
= DEVICE_GET_CLASS(obj
);
1169 if (only_migratable
&& dc
->vmsd
) {
1170 if (dc
->vmsd
->unmigratable
) {
1171 error_setg(err
, "Device %s is not migratable, but "
1172 "--only-migratable was specified",
1173 object_get_typename(obj
));
1181 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1183 Error
*local_err
= NULL
;
1184 static bool once
= true;
1186 if (!deferred_incoming
) {
1187 error_setg(errp
, "For use with '-incoming defer'");
1191 error_setg(errp
, "The incoming migration has already been started");
1194 qemu_start_incoming_migration(uri
, &local_err
);
1197 error_propagate(errp
, local_err
);
1204 bool migration_is_blocked(Error
**errp
)
1206 if (qemu_savevm_state_blocked(errp
)) {
1210 if (migration_blockers
) {
1211 *errp
= error_copy(migration_blockers
->data
);
1218 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1219 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1222 Error
*local_err
= NULL
;
1223 MigrationState
*s
= migrate_get_current();
1224 MigrationParams params
;
1227 params
.blk
= has_blk
&& blk
;
1228 params
.shared
= has_inc
&& inc
;
1230 if (migration_is_setup_or_active(s
->state
) ||
1231 s
->state
== MIGRATION_STATUS_CANCELLING
||
1232 s
->state
== MIGRATION_STATUS_COLO
) {
1233 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1236 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1237 error_setg(errp
, "Guest is waiting for an incoming migration");
1241 if (migration_is_blocked(errp
)) {
1245 s
= migrate_init(¶ms
);
1247 if (strstart(uri
, "tcp:", &p
)) {
1248 tcp_start_outgoing_migration(s
, p
, &local_err
);
1250 } else if (strstart(uri
, "rdma:", &p
)) {
1251 rdma_start_outgoing_migration(s
, p
, &local_err
);
1253 } else if (strstart(uri
, "exec:", &p
)) {
1254 exec_start_outgoing_migration(s
, p
, &local_err
);
1255 } else if (strstart(uri
, "unix:", &p
)) {
1256 unix_start_outgoing_migration(s
, p
, &local_err
);
1257 } else if (strstart(uri
, "fd:", &p
)) {
1258 fd_start_outgoing_migration(s
, p
, &local_err
);
1260 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1261 "a valid migration protocol");
1262 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1263 MIGRATION_STATUS_FAILED
);
1268 migrate_fd_error(s
, local_err
);
1269 error_propagate(errp
, local_err
);
1274 void qmp_migrate_cancel(Error
**errp
)
1276 migrate_fd_cancel(migrate_get_current());
1279 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1281 MigrationState
*s
= migrate_get_current();
1284 /* Check for truncation */
1285 if (value
!= (size_t)value
) {
1286 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1287 "exceeding address space");
1291 /* Cache should not be larger than guest ram size */
1292 if (value
> ram_bytes_total()) {
1293 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1294 "exceeds guest ram size ");
1298 new_size
= xbzrle_cache_resize(value
);
1300 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1301 "is smaller than page size");
1305 s
->xbzrle_cache_size
= new_size
;
1308 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1310 return migrate_xbzrle_cache_size();
1313 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1315 MigrationParameters p
= {
1316 .has_max_bandwidth
= true,
1317 .max_bandwidth
= value
,
1320 qmp_migrate_set_parameters(&p
, errp
);
1323 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1325 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
1326 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1327 "the range of 0 to %d seconds",
1328 MAX_MIGRATE_DOWNTIME_SECONDS
);
1332 value
*= 1000; /* Convert to milliseconds */
1333 value
= MAX(0, MIN(INT64_MAX
, value
));
1335 MigrationParameters p
= {
1336 .has_downtime_limit
= true,
1337 .downtime_limit
= value
,
1340 qmp_migrate_set_parameters(&p
, errp
);
1343 bool migrate_release_ram(void)
1347 s
= migrate_get_current();
1349 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
1352 bool migrate_postcopy_ram(void)
1356 s
= migrate_get_current();
1358 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1361 bool migrate_auto_converge(void)
1365 s
= migrate_get_current();
1367 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1370 bool migrate_zero_blocks(void)
1374 s
= migrate_get_current();
1376 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
1379 bool migrate_use_compression(void)
1383 s
= migrate_get_current();
1385 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
1388 int migrate_compress_level(void)
1392 s
= migrate_get_current();
1394 return s
->parameters
.compress_level
;
1397 int migrate_compress_threads(void)
1401 s
= migrate_get_current();
1403 return s
->parameters
.compress_threads
;
1406 int migrate_decompress_threads(void)
1410 s
= migrate_get_current();
1412 return s
->parameters
.decompress_threads
;
1415 bool migrate_use_events(void)
1419 s
= migrate_get_current();
1421 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
1424 int migrate_use_xbzrle(void)
1428 s
= migrate_get_current();
1430 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
1433 int64_t migrate_xbzrle_cache_size(void)
1437 s
= migrate_get_current();
1439 return s
->xbzrle_cache_size
;
1442 /* migration thread support */
1444 * Something bad happened to the RP stream, mark an error
1445 * The caller shall print or trace something to indicate why
1447 static void mark_source_rp_bad(MigrationState
*s
)
1449 s
->rp_state
.error
= true;
1452 static struct rp_cmd_args
{
1453 ssize_t len
; /* -1 = variable */
1456 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1457 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1458 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1459 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1460 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1461 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1465 * Process a request for pages received on the return path,
1466 * We're allowed to send more than requested (e.g. to round to our page size)
1467 * and we don't need to send pages that have already been sent.
1469 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1470 ram_addr_t start
, size_t len
)
1472 long our_host_ps
= getpagesize();
1474 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1477 * Since we currently insist on matching page sizes, just sanity check
1478 * we're being asked for whole host pages.
1480 if (start
& (our_host_ps
-1) ||
1481 (len
& (our_host_ps
-1))) {
1482 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1483 " len: %zd", __func__
, start
, len
);
1484 mark_source_rp_bad(ms
);
1488 if (ram_save_queue_pages(ms
, rbname
, start
, len
)) {
1489 mark_source_rp_bad(ms
);
1494 * Handles messages sent on the return path towards the source VM
1497 static void *source_return_path_thread(void *opaque
)
1499 MigrationState
*ms
= opaque
;
1500 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1501 uint16_t header_len
, header_type
;
1503 uint32_t tmp32
, sibling_error
;
1504 ram_addr_t start
= 0; /* =0 to silence warning */
1505 size_t len
= 0, expected_len
;
1508 trace_source_return_path_thread_entry();
1509 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1510 migration_is_setup_or_active(ms
->state
)) {
1511 trace_source_return_path_thread_loop_top();
1512 header_type
= qemu_get_be16(rp
);
1513 header_len
= qemu_get_be16(rp
);
1515 if (header_type
>= MIG_RP_MSG_MAX
||
1516 header_type
== MIG_RP_MSG_INVALID
) {
1517 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1518 header_type
, header_len
);
1519 mark_source_rp_bad(ms
);
1523 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1524 header_len
!= rp_cmd_args
[header_type
].len
) ||
1525 header_len
> sizeof(buf
)) {
1526 error_report("RP: Received '%s' message (0x%04x) with"
1527 "incorrect length %d expecting %zu",
1528 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1529 (size_t)rp_cmd_args
[header_type
].len
);
1530 mark_source_rp_bad(ms
);
1534 /* We know we've got a valid header by this point */
1535 res
= qemu_get_buffer(rp
, buf
, header_len
);
1536 if (res
!= header_len
) {
1537 error_report("RP: Failed reading data for message 0x%04x"
1538 " read %d expected %d",
1539 header_type
, res
, header_len
);
1540 mark_source_rp_bad(ms
);
1544 /* OK, we have the message and the data */
1545 switch (header_type
) {
1546 case MIG_RP_MSG_SHUT
:
1547 sibling_error
= ldl_be_p(buf
);
1548 trace_source_return_path_thread_shut(sibling_error
);
1549 if (sibling_error
) {
1550 error_report("RP: Sibling indicated error %d", sibling_error
);
1551 mark_source_rp_bad(ms
);
1554 * We'll let the main thread deal with closing the RP
1555 * we could do a shutdown(2) on it, but we're the only user
1556 * anyway, so there's nothing gained.
1560 case MIG_RP_MSG_PONG
:
1561 tmp32
= ldl_be_p(buf
);
1562 trace_source_return_path_thread_pong(tmp32
);
1565 case MIG_RP_MSG_REQ_PAGES
:
1566 start
= ldq_be_p(buf
);
1567 len
= ldl_be_p(buf
+ 8);
1568 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1571 case MIG_RP_MSG_REQ_PAGES_ID
:
1572 expected_len
= 12 + 1; /* header + termination */
1574 if (header_len
>= expected_len
) {
1575 start
= ldq_be_p(buf
);
1576 len
= ldl_be_p(buf
+ 8);
1577 /* Now we expect an idstr */
1578 tmp32
= buf
[12]; /* Length of the following idstr */
1579 buf
[13 + tmp32
] = '\0';
1580 expected_len
+= tmp32
;
1582 if (header_len
!= expected_len
) {
1583 error_report("RP: Req_Page_id with length %d expecting %zd",
1584 header_len
, expected_len
);
1585 mark_source_rp_bad(ms
);
1588 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
1595 if (qemu_file_get_error(rp
)) {
1596 trace_source_return_path_thread_bad_end();
1597 mark_source_rp_bad(ms
);
1600 trace_source_return_path_thread_end();
1602 ms
->rp_state
.from_dst_file
= NULL
;
1607 static int open_return_path_on_source(MigrationState
*ms
)
1610 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
1611 if (!ms
->rp_state
.from_dst_file
) {
1615 trace_open_return_path_on_source();
1616 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
1617 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
1619 trace_open_return_path_on_source_continue();
1624 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1625 static int await_return_path_close_on_source(MigrationState
*ms
)
1628 * If this is a normal exit then the destination will send a SHUT and the
1629 * rp_thread will exit, however if there's an error we need to cause
1632 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
1634 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1635 * waiting for the destination.
1637 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
1638 mark_source_rp_bad(ms
);
1640 trace_await_return_path_close_on_source_joining();
1641 qemu_thread_join(&ms
->rp_state
.rp_thread
);
1642 trace_await_return_path_close_on_source_close();
1643 return ms
->rp_state
.error
;
1647 * Switch from normal iteration to postcopy
1648 * Returns non-0 on error
1650 static int postcopy_start(MigrationState
*ms
, bool *old_vm_running
)
1653 QIOChannelBuffer
*bioc
;
1655 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1656 bool restart_block
= false;
1657 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
1658 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1660 trace_postcopy_start();
1661 qemu_mutex_lock_iothread();
1662 trace_postcopy_start_set_run();
1664 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1665 *old_vm_running
= runstate_is_running();
1666 global_state_store();
1667 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1672 ret
= bdrv_inactivate_all();
1676 restart_block
= true;
1679 * Cause any non-postcopiable, but iterative devices to
1680 * send out their final data.
1682 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true);
1685 * in Finish migrate and with the io-lock held everything should
1686 * be quiet, but we've potentially still got dirty pages and we
1687 * need to tell the destination to throw any pages it's already received
1690 if (ram_postcopy_send_discard_bitmap(ms
)) {
1691 error_report("postcopy send discard bitmap failed");
1696 * send rest of state - note things that are doing postcopy
1697 * will notice we're in POSTCOPY_ACTIVE and not actually
1698 * wrap their state up here
1700 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
1701 /* Ping just for debugging, helps line traces up */
1702 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
1705 * While loading the device state we may trigger page transfer
1706 * requests and the fd must be free to process those, and thus
1707 * the destination must read the whole device state off the fd before
1708 * it starts processing it. Unfortunately the ad-hoc migration format
1709 * doesn't allow the destination to know the size to read without fully
1710 * parsing it through each devices load-state code (especially the open
1711 * coded devices that use get/put).
1712 * So we wrap the device state up in a package with a length at the start;
1713 * to do this we use a qemu_buf to hold the whole of the device state.
1715 bioc
= qio_channel_buffer_new(4096);
1716 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
1717 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
1718 object_unref(OBJECT(bioc
));
1721 * Make sure the receiver can get incoming pages before we send the rest
1724 qemu_savevm_send_postcopy_listen(fb
);
1726 qemu_savevm_state_complete_precopy(fb
, false);
1727 qemu_savevm_send_ping(fb
, 3);
1729 qemu_savevm_send_postcopy_run(fb
);
1731 /* <><> end of stuff going into the package */
1733 /* Last point of recovery; as soon as we send the package the destination
1734 * can open devices and potentially start running.
1735 * Lets just check again we've not got any errors.
1737 ret
= qemu_file_get_error(ms
->to_dst_file
);
1739 error_report("postcopy_start: Migration stream errored (pre package)");
1743 restart_block
= false;
1745 /* Now send that blob */
1746 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
1751 /* Send a notify to give a chance for anything that needs to happen
1752 * at the transition to postcopy and after the device state; in particular
1753 * spice needs to trigger a transition now
1755 ms
->postcopy_after_devices
= true;
1756 notifier_list_notify(&migration_state_notifiers
, ms
);
1758 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
1760 qemu_mutex_unlock_iothread();
1763 * Although this ping is just for debug, it could potentially be
1764 * used for getting a better measurement of downtime at the source.
1766 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
1768 if (migrate_release_ram()) {
1769 ram_postcopy_migrated_memory_release(ms
);
1772 ret
= qemu_file_get_error(ms
->to_dst_file
);
1774 error_report("postcopy_start: Migration stream errored");
1775 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1776 MIGRATION_STATUS_FAILED
);
1784 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1785 MIGRATION_STATUS_FAILED
);
1786 if (restart_block
) {
1787 /* A failure happened early enough that we know the destination hasn't
1788 * accessed block devices, so we're safe to recover.
1790 Error
*local_err
= NULL
;
1792 bdrv_invalidate_cache_all(&local_err
);
1794 error_report_err(local_err
);
1797 qemu_mutex_unlock_iothread();
1802 * migration_completion: Used by migration_thread when there's not much left.
1803 * The caller 'breaks' the loop when this returns.
1805 * @s: Current migration state
1806 * @current_active_state: The migration state we expect to be in
1807 * @*old_vm_running: Pointer to old_vm_running flag
1808 * @*start_time: Pointer to time to update
1810 static void migration_completion(MigrationState
*s
, int current_active_state
,
1811 bool *old_vm_running
,
1812 int64_t *start_time
)
1816 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1817 qemu_mutex_lock_iothread();
1818 *start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1819 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1820 *old_vm_running
= runstate_is_running();
1821 ret
= global_state_store();
1824 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1826 * Don't mark the image with BDRV_O_INACTIVE flag if
1827 * we will go into COLO stage later.
1829 if (ret
>= 0 && !migrate_colo_enabled()) {
1830 ret
= bdrv_inactivate_all();
1833 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
1834 qemu_savevm_state_complete_precopy(s
->to_dst_file
, false);
1835 s
->block_inactive
= true;
1838 qemu_mutex_unlock_iothread();
1843 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1844 trace_migration_completion_postcopy_end();
1846 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
1847 trace_migration_completion_postcopy_end_after_complete();
1851 * If rp was opened we must clean up the thread before
1852 * cleaning everything else up (since if there are no failures
1853 * it will wait for the destination to send it's status in
1855 * Postcopy opens rp if enabled (even if it's not avtivated)
1857 if (migrate_postcopy_ram()) {
1859 trace_migration_completion_postcopy_end_before_rp();
1860 rp_error
= await_return_path_close_on_source(s
);
1861 trace_migration_completion_postcopy_end_after_rp(rp_error
);
1863 goto fail_invalidate
;
1867 if (qemu_file_get_error(s
->to_dst_file
)) {
1868 trace_migration_completion_file_err();
1869 goto fail_invalidate
;
1872 if (!migrate_colo_enabled()) {
1873 migrate_set_state(&s
->state
, current_active_state
,
1874 MIGRATION_STATUS_COMPLETED
);
1880 /* If not doing postcopy, vm_start() will be called: let's regain
1881 * control on images.
1883 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1884 Error
*local_err
= NULL
;
1886 qemu_mutex_lock_iothread();
1887 bdrv_invalidate_cache_all(&local_err
);
1889 error_report_err(local_err
);
1891 s
->block_inactive
= false;
1893 qemu_mutex_unlock_iothread();
1897 migrate_set_state(&s
->state
, current_active_state
,
1898 MIGRATION_STATUS_FAILED
);
1901 bool migrate_colo_enabled(void)
1903 MigrationState
*s
= migrate_get_current();
1904 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
1908 * Master migration thread on the source VM.
1909 * It drives the migration and pumps the data down the outgoing channel.
1911 static void *migration_thread(void *opaque
)
1913 MigrationState
*s
= opaque
;
1914 /* Used by the bandwidth calcs, updated later */
1915 int64_t initial_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1916 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
1917 int64_t initial_bytes
= 0;
1918 int64_t max_size
= 0;
1919 int64_t start_time
= initial_time
;
1921 bool old_vm_running
= false;
1922 bool entered_postcopy
= false;
1923 /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
1924 enum MigrationStatus current_active_state
= MIGRATION_STATUS_ACTIVE
;
1925 bool enable_colo
= migrate_colo_enabled();
1927 rcu_register_thread();
1929 qemu_savevm_state_header(s
->to_dst_file
);
1931 if (migrate_postcopy_ram()) {
1932 /* Now tell the dest that it should open its end so it can reply */
1933 qemu_savevm_send_open_return_path(s
->to_dst_file
);
1935 /* And do a ping that will make stuff easier to debug */
1936 qemu_savevm_send_ping(s
->to_dst_file
, 1);
1939 * Tell the destination that we *might* want to do postcopy later;
1940 * if the other end can't do postcopy it should fail now, nice and
1943 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
1946 qemu_savevm_state_begin(s
->to_dst_file
, &s
->params
);
1948 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
1949 current_active_state
= MIGRATION_STATUS_ACTIVE
;
1950 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1951 MIGRATION_STATUS_ACTIVE
);
1953 trace_migration_thread_setup_complete();
1955 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
1956 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1957 int64_t current_time
;
1958 uint64_t pending_size
;
1960 if (!qemu_file_rate_limit(s
->to_dst_file
)) {
1961 uint64_t pend_post
, pend_nonpost
;
1963 qemu_savevm_state_pending(s
->to_dst_file
, max_size
, &pend_nonpost
,
1965 pending_size
= pend_nonpost
+ pend_post
;
1966 trace_migrate_pending(pending_size
, max_size
,
1967 pend_post
, pend_nonpost
);
1968 if (pending_size
&& pending_size
>= max_size
) {
1969 /* Still a significant amount to transfer */
1971 if (migrate_postcopy_ram() &&
1972 s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
&&
1973 pend_nonpost
<= max_size
&&
1974 atomic_read(&s
->start_postcopy
)) {
1976 if (!postcopy_start(s
, &old_vm_running
)) {
1977 current_active_state
= MIGRATION_STATUS_POSTCOPY_ACTIVE
;
1978 entered_postcopy
= true;
1983 /* Just another iteration step */
1984 qemu_savevm_state_iterate(s
->to_dst_file
, entered_postcopy
);
1986 trace_migration_thread_low_pending(pending_size
);
1987 migration_completion(s
, current_active_state
,
1988 &old_vm_running
, &start_time
);
1993 if (qemu_file_get_error(s
->to_dst_file
)) {
1994 migrate_set_state(&s
->state
, current_active_state
,
1995 MIGRATION_STATUS_FAILED
);
1996 trace_migration_thread_file_err();
1999 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2000 if (current_time
>= initial_time
+ BUFFER_DELAY
) {
2001 uint64_t transferred_bytes
= qemu_ftell(s
->to_dst_file
) -
2003 uint64_t time_spent
= current_time
- initial_time
;
2004 double bandwidth
= (double)transferred_bytes
/ time_spent
;
2005 max_size
= bandwidth
* s
->parameters
.downtime_limit
;
2007 s
->mbps
= (((double) transferred_bytes
* 8.0) /
2008 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2010 trace_migrate_transferred(transferred_bytes
, time_spent
,
2011 bandwidth
, max_size
);
2012 /* if we haven't sent anything, we don't want to recalculate
2013 10000 is a small enough number for our purposes */
2014 if (s
->dirty_bytes_rate
&& transferred_bytes
> 10000) {
2015 s
->expected_downtime
= s
->dirty_bytes_rate
/ bandwidth
;
2018 qemu_file_reset_rate_limit(s
->to_dst_file
);
2019 initial_time
= current_time
;
2020 initial_bytes
= qemu_ftell(s
->to_dst_file
);
2022 if (qemu_file_rate_limit(s
->to_dst_file
)) {
2023 /* usleep expects microseconds */
2024 g_usleep((initial_time
+ BUFFER_DELAY
- current_time
)*1000);
2028 trace_migration_thread_after_loop();
2029 /* If we enabled cpu throttling for auto-converge, turn it off. */
2030 cpu_throttle_stop();
2031 end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2033 qemu_mutex_lock_iothread();
2035 * The resource has been allocated by migration will be reused in COLO
2036 * process, so don't release them.
2039 qemu_savevm_state_cleanup();
2041 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
2042 uint64_t transferred_bytes
= qemu_ftell(s
->to_dst_file
);
2043 s
->total_time
= end_time
- s
->total_time
;
2044 if (!entered_postcopy
) {
2045 s
->downtime
= end_time
- start_time
;
2047 if (s
->total_time
) {
2048 s
->mbps
= (((double) transferred_bytes
* 8.0) /
2049 ((double) s
->total_time
)) / 1000;
2051 runstate_set(RUN_STATE_POSTMIGRATE
);
2053 if (s
->state
== MIGRATION_STATUS_ACTIVE
&& enable_colo
) {
2054 migrate_start_colo_process(s
);
2055 qemu_savevm_state_cleanup();
2057 * Fixme: we will run VM in COLO no matter its old running state.
2058 * After exited COLO, we will keep running.
2060 old_vm_running
= true;
2062 if (old_vm_running
&& !entered_postcopy
) {
2065 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2066 runstate_set(RUN_STATE_POSTMIGRATE
);
2070 qemu_bh_schedule(s
->cleanup_bh
);
2071 qemu_mutex_unlock_iothread();
2073 rcu_unregister_thread();
2077 void migrate_fd_connect(MigrationState
*s
)
2079 s
->expected_downtime
= s
->parameters
.downtime_limit
;
2080 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
2082 qemu_file_set_blocking(s
->to_dst_file
, true);
2083 qemu_file_set_rate_limit(s
->to_dst_file
,
2084 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
2086 /* Notify before starting migration thread */
2087 notifier_list_notify(&migration_state_notifiers
, s
);
2090 * Open the return path; currently for postcopy but other things might
2093 if (migrate_postcopy_ram()) {
2094 if (open_return_path_on_source(s
)) {
2095 error_report("Unable to open return-path for postcopy");
2096 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2097 MIGRATION_STATUS_FAILED
);
2098 migrate_fd_cleanup(s
);
2103 migrate_compress_threads_create();
2104 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
2105 QEMU_THREAD_JOINABLE
);
2106 s
->migration_thread_running
= true;
2109 PostcopyState
postcopy_state_get(void)
2111 return atomic_mb_read(&incoming_postcopy_state
);
2114 /* Set the state and return the old state */
2115 PostcopyState
postcopy_state_set(PostcopyState new_state
)
2117 return atomic_xchg(&incoming_postcopy_state
, new_state
);