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-common.h"
17 #include "qemu/error-report.h"
18 #include "qemu/main-loop.h"
19 #include "migration/migration.h"
20 #include "migration/qemu-file.h"
21 #include "sysemu/sysemu.h"
22 #include "block/block.h"
23 #include "qapi/qmp/qerror.h"
24 #include "qapi/util.h"
25 #include "qemu/sockets.h"
27 #include "migration/block.h"
28 #include "migration/postcopy-ram.h"
29 #include "qemu/thread.h"
30 #include "qmp-commands.h"
32 #include "qapi-event.h"
34 #include "exec/memory.h"
35 #include "exec/address-spaces.h"
37 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
39 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
41 #define BUFFER_DELAY 100
42 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
44 /* Default compression thread count */
45 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
46 /* Default decompression thread count, usually decompression is at
47 * least 4 times as fast as compression.*/
48 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
49 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
50 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
51 /* Define default autoconverge cpu throttle migration parameters */
52 #define DEFAULT_MIGRATE_X_CPU_THROTTLE_INITIAL 20
53 #define DEFAULT_MIGRATE_X_CPU_THROTTLE_INCREMENT 10
55 /* Migration XBZRLE default cache size */
56 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
58 static NotifierList migration_state_notifiers
=
59 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
61 static bool deferred_incoming
;
64 * Current state of incoming postcopy; note this is not part of
65 * MigrationIncomingState since it's state is used during cleanup
66 * at the end as MIS is being freed.
68 static PostcopyState incoming_postcopy_state
;
70 /* When we add fault tolerance, we could have several
71 migrations at once. For now we don't need to add
72 dynamic creation of migration */
75 MigrationState
*migrate_get_current(void)
78 static MigrationState current_migration
= {
79 .state
= MIGRATION_STATUS_NONE
,
80 .bandwidth_limit
= MAX_THROTTLE
,
81 .xbzrle_cache_size
= DEFAULT_MIGRATE_CACHE_SIZE
,
83 .parameters
[MIGRATION_PARAMETER_COMPRESS_LEVEL
] =
84 DEFAULT_MIGRATE_COMPRESS_LEVEL
,
85 .parameters
[MIGRATION_PARAMETER_COMPRESS_THREADS
] =
86 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
,
87 .parameters
[MIGRATION_PARAMETER_DECOMPRESS_THREADS
] =
88 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
,
89 .parameters
[MIGRATION_PARAMETER_X_CPU_THROTTLE_INITIAL
] =
90 DEFAULT_MIGRATE_X_CPU_THROTTLE_INITIAL
,
91 .parameters
[MIGRATION_PARAMETER_X_CPU_THROTTLE_INCREMENT
] =
92 DEFAULT_MIGRATE_X_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 QLIST_INIT(&mis_current
->loadvm_handlers
);
115 qemu_mutex_init(&mis_current
->rp_mutex
);
116 qemu_event_init(&mis_current
->main_thread_load_event
, false);
121 void migration_incoming_state_destroy(void)
123 qemu_event_destroy(&mis_current
->main_thread_load_event
);
124 loadvm_free_handlers(mis_current
);
133 uint8_t runstate
[100];
138 static GlobalState global_state
;
140 int global_state_store(void)
142 if (!runstate_store((char *)global_state
.runstate
,
143 sizeof(global_state
.runstate
))) {
144 error_report("runstate name too big: %s", global_state
.runstate
);
145 trace_migrate_state_too_big();
151 void global_state_store_running(void)
153 const char *state
= RunState_lookup
[RUN_STATE_RUNNING
];
154 strncpy((char *)global_state
.runstate
,
155 state
, sizeof(global_state
.runstate
));
158 static bool global_state_received(void)
160 return global_state
.received
;
163 static RunState
global_state_get_runstate(void)
165 return global_state
.state
;
168 void global_state_set_optional(void)
170 global_state
.optional
= true;
173 static bool global_state_needed(void *opaque
)
175 GlobalState
*s
= opaque
;
176 char *runstate
= (char *)s
->runstate
;
178 /* If it is not optional, it is mandatory */
180 if (s
->optional
== false) {
184 /* If state is running or paused, it is not needed */
186 if (strcmp(runstate
, "running") == 0 ||
187 strcmp(runstate
, "paused") == 0) {
191 /* for any other state it is needed */
195 static int global_state_post_load(void *opaque
, int version_id
)
197 GlobalState
*s
= opaque
;
198 Error
*local_err
= NULL
;
200 char *runstate
= (char *)s
->runstate
;
203 trace_migrate_global_state_post_load(runstate
);
205 r
= qapi_enum_parse(RunState_lookup
, runstate
, RUN_STATE_MAX
,
210 error_report_err(local_err
);
219 static void global_state_pre_save(void *opaque
)
221 GlobalState
*s
= opaque
;
223 trace_migrate_global_state_pre_save((char *)s
->runstate
);
224 s
->size
= strlen((char *)s
->runstate
) + 1;
227 static const VMStateDescription vmstate_globalstate
= {
228 .name
= "globalstate",
230 .minimum_version_id
= 1,
231 .post_load
= global_state_post_load
,
232 .pre_save
= global_state_pre_save
,
233 .needed
= global_state_needed
,
234 .fields
= (VMStateField
[]) {
235 VMSTATE_UINT32(size
, GlobalState
),
236 VMSTATE_BUFFER(runstate
, GlobalState
),
237 VMSTATE_END_OF_LIST()
241 void register_global_state(void)
243 /* We would use it independently that we receive it */
244 strcpy((char *)&global_state
.runstate
, "");
245 global_state
.received
= false;
246 vmstate_register(NULL
, 0, &vmstate_globalstate
, &global_state
);
249 static void migrate_generate_event(int new_state
)
251 if (migrate_use_events()) {
252 qapi_event_send_migration(new_state
, &error_abort
);
257 * Called on -incoming with a defer: uri.
258 * The migration can be started later after any parameters have been
261 static void deferred_incoming_migration(Error
**errp
)
263 if (deferred_incoming
) {
264 error_setg(errp
, "Incoming migration already deferred");
266 deferred_incoming
= true;
269 /* Request a range of pages from the source VM at the given
271 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
272 * as the last request (a name must have been given previously)
273 * Start: Address offset within the RB
274 * Len: Length in bytes required - must be a multiple of pagesize
276 void migrate_send_rp_req_pages(MigrationIncomingState
*mis
, const char *rbname
,
277 ram_addr_t start
, size_t len
)
279 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname upto 256 */
280 size_t msglen
= 12; /* start + len */
282 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
283 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
286 int rbname_len
= strlen(rbname
);
287 assert(rbname_len
< 256);
289 bufc
[msglen
++] = rbname_len
;
290 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
291 msglen
+= rbname_len
;
292 migrate_send_rp_message(mis
, MIG_RP_MSG_REQ_PAGES_ID
, msglen
, bufc
);
294 migrate_send_rp_message(mis
, MIG_RP_MSG_REQ_PAGES
, msglen
, bufc
);
298 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
302 qapi_event_send_migration(MIGRATION_STATUS_SETUP
, &error_abort
);
303 if (!strcmp(uri
, "defer")) {
304 deferred_incoming_migration(errp
);
305 } else if (strstart(uri
, "tcp:", &p
)) {
306 tcp_start_incoming_migration(p
, errp
);
308 } else if (strstart(uri
, "rdma:", &p
)) {
309 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
);
320 error_setg(errp
, "unknown migration protocol: %s", uri
);
324 static void process_incoming_migration_co(void *opaque
)
326 QEMUFile
*f
= opaque
;
327 Error
*local_err
= NULL
;
330 migration_incoming_state_new(f
);
331 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
332 migrate_generate_event(MIGRATION_STATUS_ACTIVE
);
333 ret
= qemu_loadvm_state(f
);
336 free_xbzrle_decoded_buf();
337 migration_incoming_state_destroy();
340 migrate_generate_event(MIGRATION_STATUS_FAILED
);
341 error_report("load of migration failed: %s", strerror(-ret
));
342 migrate_decompress_threads_join();
346 /* Make sure all file formats flush their mutable metadata */
347 bdrv_invalidate_cache_all(&local_err
);
349 migrate_generate_event(MIGRATION_STATUS_FAILED
);
350 error_report_err(local_err
);
351 migrate_decompress_threads_join();
356 * This must happen after all error conditions are dealt with and
357 * we're sure the VM is going to be running on this host.
359 qemu_announce_self();
361 /* If global state section was not received or we are in running
362 state, we need to obey autostart. Any other state is set with
365 if (!global_state_received() ||
366 global_state_get_runstate() == RUN_STATE_RUNNING
) {
370 runstate_set(RUN_STATE_PAUSED
);
373 runstate_set(global_state_get_runstate());
375 migrate_decompress_threads_join();
377 * This must happen after any state changes since as soon as an external
378 * observer sees this event they might start to prod at the VM assuming
381 migrate_generate_event(MIGRATION_STATUS_COMPLETED
);
384 void process_incoming_migration(QEMUFile
*f
)
386 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
);
387 int fd
= qemu_get_fd(f
);
390 migrate_decompress_threads_create();
391 qemu_set_nonblock(fd
);
392 qemu_coroutine_enter(co
, f
);
396 * Send a message on the return channel back to the source
399 void migrate_send_rp_message(MigrationIncomingState
*mis
,
400 enum mig_rp_message_type message_type
,
401 uint16_t len
, void *data
)
403 trace_migrate_send_rp_message((int)message_type
, len
);
404 qemu_mutex_lock(&mis
->rp_mutex
);
405 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
406 qemu_put_be16(mis
->to_src_file
, len
);
407 qemu_put_buffer(mis
->to_src_file
, data
, len
);
408 qemu_fflush(mis
->to_src_file
);
409 qemu_mutex_unlock(&mis
->rp_mutex
);
413 * Send a 'SHUT' message on the return channel with the given value
414 * to indicate that we've finished with the RP. Non-0 value indicates
417 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
422 buf
= cpu_to_be32(value
);
423 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
427 * Send a 'PONG' message on the return channel with the given value
428 * (normally in response to a 'PING')
430 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
435 buf
= cpu_to_be32(value
);
436 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
439 /* amount of nanoseconds we are willing to wait for migration to be down.
440 * the choice of nanoseconds is because it is the maximum resolution that
441 * get_clock() can achieve. It is an internal measure. All user-visible
442 * units must be in seconds */
443 static uint64_t max_downtime
= 300000000;
445 uint64_t migrate_max_downtime(void)
450 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
452 MigrationCapabilityStatusList
*head
= NULL
;
453 MigrationCapabilityStatusList
*caps
;
454 MigrationState
*s
= migrate_get_current();
457 caps
= NULL
; /* silence compiler warning */
458 for (i
= 0; i
< MIGRATION_CAPABILITY_MAX
; i
++) {
460 head
= g_malloc0(sizeof(*caps
));
463 caps
->next
= g_malloc0(sizeof(*caps
));
467 g_malloc(sizeof(*caps
->value
));
468 caps
->value
->capability
= i
;
469 caps
->value
->state
= s
->enabled_capabilities
[i
];
475 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
477 MigrationParameters
*params
;
478 MigrationState
*s
= migrate_get_current();
480 params
= g_malloc0(sizeof(*params
));
481 params
->compress_level
= s
->parameters
[MIGRATION_PARAMETER_COMPRESS_LEVEL
];
482 params
->compress_threads
=
483 s
->parameters
[MIGRATION_PARAMETER_COMPRESS_THREADS
];
484 params
->decompress_threads
=
485 s
->parameters
[MIGRATION_PARAMETER_DECOMPRESS_THREADS
];
486 params
->x_cpu_throttle_initial
=
487 s
->parameters
[MIGRATION_PARAMETER_X_CPU_THROTTLE_INITIAL
];
488 params
->x_cpu_throttle_increment
=
489 s
->parameters
[MIGRATION_PARAMETER_X_CPU_THROTTLE_INCREMENT
];
495 * Return true if we're already in the middle of a migration
496 * (i.e. any of the active or setup states)
498 static bool migration_is_setup_or_active(int state
)
501 case MIGRATION_STATUS_ACTIVE
:
502 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
503 case MIGRATION_STATUS_SETUP
:
512 static void get_xbzrle_cache_stats(MigrationInfo
*info
)
514 if (migrate_use_xbzrle()) {
515 info
->has_xbzrle_cache
= true;
516 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
517 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
518 info
->xbzrle_cache
->bytes
= xbzrle_mig_bytes_transferred();
519 info
->xbzrle_cache
->pages
= xbzrle_mig_pages_transferred();
520 info
->xbzrle_cache
->cache_miss
= xbzrle_mig_pages_cache_miss();
521 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_mig_cache_miss_rate();
522 info
->xbzrle_cache
->overflow
= xbzrle_mig_pages_overflow();
526 MigrationInfo
*qmp_query_migrate(Error
**errp
)
528 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
529 MigrationState
*s
= migrate_get_current();
532 case MIGRATION_STATUS_NONE
:
533 /* no migration has happened ever */
535 case MIGRATION_STATUS_SETUP
:
536 info
->has_status
= true;
537 info
->has_total_time
= false;
539 case MIGRATION_STATUS_ACTIVE
:
540 case MIGRATION_STATUS_CANCELLING
:
541 info
->has_status
= true;
542 info
->has_total_time
= true;
543 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
545 info
->has_expected_downtime
= true;
546 info
->expected_downtime
= s
->expected_downtime
;
547 info
->has_setup_time
= true;
548 info
->setup_time
= s
->setup_time
;
550 info
->has_ram
= true;
551 info
->ram
= g_malloc0(sizeof(*info
->ram
));
552 info
->ram
->transferred
= ram_bytes_transferred();
553 info
->ram
->remaining
= ram_bytes_remaining();
554 info
->ram
->total
= ram_bytes_total();
555 info
->ram
->duplicate
= dup_mig_pages_transferred();
556 info
->ram
->skipped
= skipped_mig_pages_transferred();
557 info
->ram
->normal
= norm_mig_pages_transferred();
558 info
->ram
->normal_bytes
= norm_mig_bytes_transferred();
559 info
->ram
->dirty_pages_rate
= s
->dirty_pages_rate
;
560 info
->ram
->mbps
= s
->mbps
;
561 info
->ram
->dirty_sync_count
= s
->dirty_sync_count
;
563 if (blk_mig_active()) {
564 info
->has_disk
= true;
565 info
->disk
= g_malloc0(sizeof(*info
->disk
));
566 info
->disk
->transferred
= blk_mig_bytes_transferred();
567 info
->disk
->remaining
= blk_mig_bytes_remaining();
568 info
->disk
->total
= blk_mig_bytes_total();
571 if (cpu_throttle_active()) {
572 info
->has_x_cpu_throttle_percentage
= true;
573 info
->x_cpu_throttle_percentage
= cpu_throttle_get_percentage();
576 get_xbzrle_cache_stats(info
);
578 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
579 /* Mostly the same as active; TODO add some postcopy stats */
580 info
->has_status
= true;
581 info
->has_total_time
= true;
582 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
584 info
->has_expected_downtime
= true;
585 info
->expected_downtime
= s
->expected_downtime
;
586 info
->has_setup_time
= true;
587 info
->setup_time
= s
->setup_time
;
589 info
->has_ram
= true;
590 info
->ram
= g_malloc0(sizeof(*info
->ram
));
591 info
->ram
->transferred
= ram_bytes_transferred();
592 info
->ram
->remaining
= ram_bytes_remaining();
593 info
->ram
->total
= ram_bytes_total();
594 info
->ram
->duplicate
= dup_mig_pages_transferred();
595 info
->ram
->skipped
= skipped_mig_pages_transferred();
596 info
->ram
->normal
= norm_mig_pages_transferred();
597 info
->ram
->normal_bytes
= norm_mig_bytes_transferred();
598 info
->ram
->dirty_pages_rate
= s
->dirty_pages_rate
;
599 info
->ram
->mbps
= s
->mbps
;
601 if (blk_mig_active()) {
602 info
->has_disk
= true;
603 info
->disk
= g_malloc0(sizeof(*info
->disk
));
604 info
->disk
->transferred
= blk_mig_bytes_transferred();
605 info
->disk
->remaining
= blk_mig_bytes_remaining();
606 info
->disk
->total
= blk_mig_bytes_total();
609 get_xbzrle_cache_stats(info
);
611 case MIGRATION_STATUS_COMPLETED
:
612 get_xbzrle_cache_stats(info
);
614 info
->has_status
= true;
615 info
->has_total_time
= true;
616 info
->total_time
= s
->total_time
;
617 info
->has_downtime
= true;
618 info
->downtime
= s
->downtime
;
619 info
->has_setup_time
= true;
620 info
->setup_time
= s
->setup_time
;
622 info
->has_ram
= true;
623 info
->ram
= g_malloc0(sizeof(*info
->ram
));
624 info
->ram
->transferred
= ram_bytes_transferred();
625 info
->ram
->remaining
= 0;
626 info
->ram
->total
= ram_bytes_total();
627 info
->ram
->duplicate
= dup_mig_pages_transferred();
628 info
->ram
->skipped
= skipped_mig_pages_transferred();
629 info
->ram
->normal
= norm_mig_pages_transferred();
630 info
->ram
->normal_bytes
= norm_mig_bytes_transferred();
631 info
->ram
->mbps
= s
->mbps
;
632 info
->ram
->dirty_sync_count
= s
->dirty_sync_count
;
634 case MIGRATION_STATUS_FAILED
:
635 info
->has_status
= true;
637 case MIGRATION_STATUS_CANCELLED
:
638 info
->has_status
= true;
641 info
->status
= s
->state
;
646 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
649 MigrationState
*s
= migrate_get_current();
650 MigrationCapabilityStatusList
*cap
;
652 if (migration_is_setup_or_active(s
->state
)) {
653 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
657 for (cap
= params
; cap
; cap
= cap
->next
) {
658 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
661 if (migrate_postcopy_ram()) {
662 if (migrate_use_compression()) {
663 /* The decompression threads asynchronously write into RAM
664 * rather than use the atomic copies needed to avoid
665 * userfaulting. It should be possible to fix the decompression
666 * threads for compatibility in future.
668 error_report("Postcopy is not currently compatible with "
670 s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_POSTCOPY_RAM
] =
676 void qmp_migrate_set_parameters(bool has_compress_level
,
677 int64_t compress_level
,
678 bool has_compress_threads
,
679 int64_t compress_threads
,
680 bool has_decompress_threads
,
681 int64_t decompress_threads
,
682 bool has_x_cpu_throttle_initial
,
683 int64_t x_cpu_throttle_initial
,
684 bool has_x_cpu_throttle_increment
,
685 int64_t x_cpu_throttle_increment
, Error
**errp
)
687 MigrationState
*s
= migrate_get_current();
689 if (has_compress_level
&& (compress_level
< 0 || compress_level
> 9)) {
690 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
691 "is invalid, it should be in the range of 0 to 9");
694 if (has_compress_threads
&&
695 (compress_threads
< 1 || compress_threads
> 255)) {
696 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
698 "is invalid, it should be in the range of 1 to 255");
701 if (has_decompress_threads
&&
702 (decompress_threads
< 1 || decompress_threads
> 255)) {
703 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
704 "decompress_threads",
705 "is invalid, it should be in the range of 1 to 255");
708 if (has_x_cpu_throttle_initial
&&
709 (x_cpu_throttle_initial
< 1 || x_cpu_throttle_initial
> 99)) {
710 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
711 "x_cpu_throttle_initial",
712 "an integer in the range of 1 to 99");
714 if (has_x_cpu_throttle_increment
&&
715 (x_cpu_throttle_increment
< 1 || x_cpu_throttle_increment
> 99)) {
716 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
717 "x_cpu_throttle_increment",
718 "an integer in the range of 1 to 99");
721 if (has_compress_level
) {
722 s
->parameters
[MIGRATION_PARAMETER_COMPRESS_LEVEL
] = compress_level
;
724 if (has_compress_threads
) {
725 s
->parameters
[MIGRATION_PARAMETER_COMPRESS_THREADS
] = compress_threads
;
727 if (has_decompress_threads
) {
728 s
->parameters
[MIGRATION_PARAMETER_DECOMPRESS_THREADS
] =
731 if (has_x_cpu_throttle_initial
) {
732 s
->parameters
[MIGRATION_PARAMETER_X_CPU_THROTTLE_INITIAL
] =
733 x_cpu_throttle_initial
;
736 if (has_x_cpu_throttle_increment
) {
737 s
->parameters
[MIGRATION_PARAMETER_X_CPU_THROTTLE_INCREMENT
] =
738 x_cpu_throttle_increment
;
742 void qmp_migrate_start_postcopy(Error
**errp
)
744 MigrationState
*s
= migrate_get_current();
746 if (!migrate_postcopy_ram()) {
747 error_setg(errp
, "Enable postcopy with migration_set_capability before"
748 " the start of migration");
752 if (s
->state
== MIGRATION_STATUS_NONE
) {
753 error_setg(errp
, "Postcopy must be started after migration has been"
758 * we don't error if migration has finished since that would be racy
759 * with issuing this command.
761 atomic_set(&s
->start_postcopy
, true);
764 /* shared migration helpers */
766 static void migrate_set_state(MigrationState
*s
, int old_state
, int new_state
)
768 if (atomic_cmpxchg(&s
->state
, old_state
, new_state
) == old_state
) {
769 trace_migrate_set_state(new_state
);
770 migrate_generate_event(new_state
);
774 static void migrate_fd_cleanup(void *opaque
)
776 MigrationState
*s
= opaque
;
778 qemu_bh_delete(s
->cleanup_bh
);
779 s
->cleanup_bh
= NULL
;
784 trace_migrate_fd_cleanup();
785 qemu_mutex_unlock_iothread();
786 if (s
->migration_thread_running
) {
787 qemu_thread_join(&s
->thread
);
788 s
->migration_thread_running
= false;
790 qemu_mutex_lock_iothread();
792 migrate_compress_threads_join();
793 qemu_fclose(s
->file
);
797 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
798 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
800 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
801 migrate_set_state(s
, MIGRATION_STATUS_CANCELLING
,
802 MIGRATION_STATUS_CANCELLED
);
805 notifier_list_notify(&migration_state_notifiers
, s
);
808 void migrate_fd_error(MigrationState
*s
)
810 trace_migrate_fd_error();
811 assert(s
->file
== NULL
);
812 migrate_set_state(s
, MIGRATION_STATUS_SETUP
, MIGRATION_STATUS_FAILED
);
813 notifier_list_notify(&migration_state_notifiers
, s
);
816 static void migrate_fd_cancel(MigrationState
*s
)
819 QEMUFile
*f
= migrate_get_current()->file
;
820 trace_migrate_fd_cancel();
822 if (s
->rp_state
.from_dst_file
) {
823 /* shutdown the rp socket, so causing the rp thread to shutdown */
824 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
828 old_state
= s
->state
;
829 if (!migration_is_setup_or_active(old_state
)) {
832 migrate_set_state(s
, old_state
, MIGRATION_STATUS_CANCELLING
);
833 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
836 * If we're unlucky the migration code might be stuck somewhere in a
837 * send/write while the network has failed and is waiting to timeout;
838 * if we've got shutdown(2) available then we can force it to quit.
839 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
840 * called in a bh, so there is no race against this cancel.
842 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
843 qemu_file_shutdown(f
);
847 void add_migration_state_change_notifier(Notifier
*notify
)
849 notifier_list_add(&migration_state_notifiers
, notify
);
852 void remove_migration_state_change_notifier(Notifier
*notify
)
854 notifier_remove(notify
);
857 bool migration_in_setup(MigrationState
*s
)
859 return s
->state
== MIGRATION_STATUS_SETUP
;
862 bool migration_has_finished(MigrationState
*s
)
864 return s
->state
== MIGRATION_STATUS_COMPLETED
;
867 bool migration_has_failed(MigrationState
*s
)
869 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
870 s
->state
== MIGRATION_STATUS_FAILED
);
873 bool migration_in_postcopy(MigrationState
*s
)
875 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
878 MigrationState
*migrate_init(const MigrationParams
*params
)
880 MigrationState
*s
= migrate_get_current();
881 int64_t bandwidth_limit
= s
->bandwidth_limit
;
882 bool enabled_capabilities
[MIGRATION_CAPABILITY_MAX
];
883 int64_t xbzrle_cache_size
= s
->xbzrle_cache_size
;
884 int compress_level
= s
->parameters
[MIGRATION_PARAMETER_COMPRESS_LEVEL
];
885 int compress_thread_count
=
886 s
->parameters
[MIGRATION_PARAMETER_COMPRESS_THREADS
];
887 int decompress_thread_count
=
888 s
->parameters
[MIGRATION_PARAMETER_DECOMPRESS_THREADS
];
889 int x_cpu_throttle_initial
=
890 s
->parameters
[MIGRATION_PARAMETER_X_CPU_THROTTLE_INITIAL
];
891 int x_cpu_throttle_increment
=
892 s
->parameters
[MIGRATION_PARAMETER_X_CPU_THROTTLE_INCREMENT
];
894 memcpy(enabled_capabilities
, s
->enabled_capabilities
,
895 sizeof(enabled_capabilities
));
897 memset(s
, 0, sizeof(*s
));
899 memcpy(s
->enabled_capabilities
, enabled_capabilities
,
900 sizeof(enabled_capabilities
));
901 s
->xbzrle_cache_size
= xbzrle_cache_size
;
903 s
->parameters
[MIGRATION_PARAMETER_COMPRESS_LEVEL
] = compress_level
;
904 s
->parameters
[MIGRATION_PARAMETER_COMPRESS_THREADS
] =
905 compress_thread_count
;
906 s
->parameters
[MIGRATION_PARAMETER_DECOMPRESS_THREADS
] =
907 decompress_thread_count
;
908 s
->parameters
[MIGRATION_PARAMETER_X_CPU_THROTTLE_INITIAL
] =
909 x_cpu_throttle_initial
;
910 s
->parameters
[MIGRATION_PARAMETER_X_CPU_THROTTLE_INCREMENT
] =
911 x_cpu_throttle_increment
;
912 s
->bandwidth_limit
= bandwidth_limit
;
913 migrate_set_state(s
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
915 QSIMPLEQ_INIT(&s
->src_page_requests
);
917 s
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
921 static GSList
*migration_blockers
;
923 void migrate_add_blocker(Error
*reason
)
925 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
928 void migrate_del_blocker(Error
*reason
)
930 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
933 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
935 Error
*local_err
= NULL
;
936 static bool once
= true;
938 if (!deferred_incoming
) {
939 error_setg(errp
, "For use with '-incoming defer'");
943 error_setg(errp
, "The incoming migration has already been started");
946 qemu_start_incoming_migration(uri
, &local_err
);
949 error_propagate(errp
, local_err
);
956 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
957 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
960 Error
*local_err
= NULL
;
961 MigrationState
*s
= migrate_get_current();
962 MigrationParams params
;
965 params
.blk
= has_blk
&& blk
;
966 params
.shared
= has_inc
&& inc
;
968 if (migration_is_setup_or_active(s
->state
) ||
969 s
->state
== MIGRATION_STATUS_CANCELLING
) {
970 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
973 if (runstate_check(RUN_STATE_INMIGRATE
)) {
974 error_setg(errp
, "Guest is waiting for an incoming migration");
978 if (qemu_savevm_state_blocked(errp
)) {
982 if (migration_blockers
) {
983 *errp
= error_copy(migration_blockers
->data
);
987 /* We are starting a new migration, so we want to start in a clean
988 state. This change is only needed if previous migration
989 failed/was cancelled. We don't use migrate_set_state() because
990 we are setting the initial state, not changing it. */
991 s
->state
= MIGRATION_STATUS_NONE
;
993 s
= migrate_init(¶ms
);
995 if (strstart(uri
, "tcp:", &p
)) {
996 tcp_start_outgoing_migration(s
, p
, &local_err
);
998 } else if (strstart(uri
, "rdma:", &p
)) {
999 rdma_start_outgoing_migration(s
, p
, &local_err
);
1002 } else if (strstart(uri
, "exec:", &p
)) {
1003 exec_start_outgoing_migration(s
, p
, &local_err
);
1004 } else if (strstart(uri
, "unix:", &p
)) {
1005 unix_start_outgoing_migration(s
, p
, &local_err
);
1006 } else if (strstart(uri
, "fd:", &p
)) {
1007 fd_start_outgoing_migration(s
, p
, &local_err
);
1010 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1011 "a valid migration protocol");
1012 migrate_set_state(s
, MIGRATION_STATUS_SETUP
, MIGRATION_STATUS_FAILED
);
1017 migrate_fd_error(s
);
1018 error_propagate(errp
, local_err
);
1023 void qmp_migrate_cancel(Error
**errp
)
1025 migrate_fd_cancel(migrate_get_current());
1028 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1030 MigrationState
*s
= migrate_get_current();
1033 /* Check for truncation */
1034 if (value
!= (size_t)value
) {
1035 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1036 "exceeding address space");
1040 /* Cache should not be larger than guest ram size */
1041 if (value
> ram_bytes_total()) {
1042 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1043 "exceeds guest ram size ");
1047 new_size
= xbzrle_cache_resize(value
);
1049 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1050 "is smaller than page size");
1054 s
->xbzrle_cache_size
= new_size
;
1057 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1059 return migrate_xbzrle_cache_size();
1062 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1069 if (value
> SIZE_MAX
) {
1073 s
= migrate_get_current();
1074 s
->bandwidth_limit
= value
;
1076 qemu_file_set_rate_limit(s
->file
, s
->bandwidth_limit
/ XFER_LIMIT_RATIO
);
1080 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1083 value
= MAX(0, MIN(UINT64_MAX
, value
));
1084 max_downtime
= (uint64_t)value
;
1087 bool migrate_postcopy_ram(void)
1091 s
= migrate_get_current();
1093 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_POSTCOPY_RAM
];
1096 bool migrate_auto_converge(void)
1100 s
= migrate_get_current();
1102 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1105 bool migrate_zero_blocks(void)
1109 s
= migrate_get_current();
1111 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
1114 bool migrate_use_compression(void)
1118 s
= migrate_get_current();
1120 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
1123 int migrate_compress_level(void)
1127 s
= migrate_get_current();
1129 return s
->parameters
[MIGRATION_PARAMETER_COMPRESS_LEVEL
];
1132 int migrate_compress_threads(void)
1136 s
= migrate_get_current();
1138 return s
->parameters
[MIGRATION_PARAMETER_COMPRESS_THREADS
];
1141 int migrate_decompress_threads(void)
1145 s
= migrate_get_current();
1147 return s
->parameters
[MIGRATION_PARAMETER_DECOMPRESS_THREADS
];
1150 bool migrate_use_events(void)
1154 s
= migrate_get_current();
1156 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
1159 int migrate_use_xbzrle(void)
1163 s
= migrate_get_current();
1165 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
1168 int64_t migrate_xbzrle_cache_size(void)
1172 s
= migrate_get_current();
1174 return s
->xbzrle_cache_size
;
1177 /* migration thread support */
1179 * Something bad happened to the RP stream, mark an error
1180 * The caller shall print or trace something to indicate why
1182 static void mark_source_rp_bad(MigrationState
*s
)
1184 s
->rp_state
.error
= true;
1187 static struct rp_cmd_args
{
1188 ssize_t len
; /* -1 = variable */
1191 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1192 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1193 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1194 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1195 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1196 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1200 * Process a request for pages received on the return path,
1201 * We're allowed to send more than requested (e.g. to round to our page size)
1202 * and we don't need to send pages that have already been sent.
1204 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1205 ram_addr_t start
, size_t len
)
1207 long our_host_ps
= getpagesize();
1209 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1212 * Since we currently insist on matching page sizes, just sanity check
1213 * we're being asked for whole host pages.
1215 if (start
& (our_host_ps
-1) ||
1216 (len
& (our_host_ps
-1))) {
1217 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1218 " len: %zd", __func__
, start
, len
);
1219 mark_source_rp_bad(ms
);
1223 if (ram_save_queue_pages(ms
, rbname
, start
, len
)) {
1224 mark_source_rp_bad(ms
);
1229 * Handles messages sent on the return path towards the source VM
1232 static void *source_return_path_thread(void *opaque
)
1234 MigrationState
*ms
= opaque
;
1235 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1236 uint16_t header_len
, header_type
;
1237 const int max_len
= 512;
1238 uint8_t buf
[max_len
];
1239 uint32_t tmp32
, sibling_error
;
1240 ram_addr_t start
= 0; /* =0 to silence warning */
1241 size_t len
= 0, expected_len
;
1244 trace_source_return_path_thread_entry();
1245 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1246 migration_is_setup_or_active(ms
->state
)) {
1247 trace_source_return_path_thread_loop_top();
1248 header_type
= qemu_get_be16(rp
);
1249 header_len
= qemu_get_be16(rp
);
1251 if (header_type
>= MIG_RP_MSG_MAX
||
1252 header_type
== MIG_RP_MSG_INVALID
) {
1253 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1254 header_type
, header_len
);
1255 mark_source_rp_bad(ms
);
1259 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1260 header_len
!= rp_cmd_args
[header_type
].len
) ||
1261 header_len
> max_len
) {
1262 error_report("RP: Received '%s' message (0x%04x) with"
1263 "incorrect length %d expecting %zu",
1264 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1265 (size_t)rp_cmd_args
[header_type
].len
);
1266 mark_source_rp_bad(ms
);
1270 /* We know we've got a valid header by this point */
1271 res
= qemu_get_buffer(rp
, buf
, header_len
);
1272 if (res
!= header_len
) {
1273 error_report("RP: Failed reading data for message 0x%04x"
1274 " read %d expected %d",
1275 header_type
, res
, header_len
);
1276 mark_source_rp_bad(ms
);
1280 /* OK, we have the message and the data */
1281 switch (header_type
) {
1282 case MIG_RP_MSG_SHUT
:
1283 sibling_error
= be32_to_cpup((uint32_t *)buf
);
1284 trace_source_return_path_thread_shut(sibling_error
);
1285 if (sibling_error
) {
1286 error_report("RP: Sibling indicated error %d", sibling_error
);
1287 mark_source_rp_bad(ms
);
1290 * We'll let the main thread deal with closing the RP
1291 * we could do a shutdown(2) on it, but we're the only user
1292 * anyway, so there's nothing gained.
1296 case MIG_RP_MSG_PONG
:
1297 tmp32
= be32_to_cpup((uint32_t *)buf
);
1298 trace_source_return_path_thread_pong(tmp32
);
1301 case MIG_RP_MSG_REQ_PAGES
:
1302 start
= be64_to_cpup((uint64_t *)buf
);
1303 len
= be32_to_cpup((uint32_t *)(buf
+ 8));
1304 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1307 case MIG_RP_MSG_REQ_PAGES_ID
:
1308 expected_len
= 12 + 1; /* header + termination */
1310 if (header_len
>= expected_len
) {
1311 start
= be64_to_cpup((uint64_t *)buf
);
1312 len
= be32_to_cpup((uint32_t *)(buf
+ 8));
1313 /* Now we expect an idstr */
1314 tmp32
= buf
[12]; /* Length of the following idstr */
1315 buf
[13 + tmp32
] = '\0';
1316 expected_len
+= tmp32
;
1318 if (header_len
!= expected_len
) {
1319 error_report("RP: Req_Page_id with length %d expecting %zd",
1320 header_len
, expected_len
);
1321 mark_source_rp_bad(ms
);
1324 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
1331 if (rp
&& qemu_file_get_error(rp
)) {
1332 trace_source_return_path_thread_bad_end();
1333 mark_source_rp_bad(ms
);
1336 trace_source_return_path_thread_end();
1338 ms
->rp_state
.from_dst_file
= NULL
;
1343 static int open_return_path_on_source(MigrationState
*ms
)
1346 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->file
);
1347 if (!ms
->rp_state
.from_dst_file
) {
1351 trace_open_return_path_on_source();
1352 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
1353 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
1355 trace_open_return_path_on_source_continue();
1360 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1361 static int await_return_path_close_on_source(MigrationState
*ms
)
1364 * If this is a normal exit then the destination will send a SHUT and the
1365 * rp_thread will exit, however if there's an error we need to cause
1368 if (qemu_file_get_error(ms
->file
) && ms
->rp_state
.from_dst_file
) {
1370 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1371 * waiting for the destination.
1373 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
1374 mark_source_rp_bad(ms
);
1376 trace_await_return_path_close_on_source_joining();
1377 qemu_thread_join(&ms
->rp_state
.rp_thread
);
1378 trace_await_return_path_close_on_source_close();
1379 return ms
->rp_state
.error
;
1383 * Switch from normal iteration to postcopy
1384 * Returns non-0 on error
1386 static int postcopy_start(MigrationState
*ms
, bool *old_vm_running
)
1389 const QEMUSizedBuffer
*qsb
;
1390 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1391 migrate_set_state(ms
, MIGRATION_STATUS_ACTIVE
,
1392 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1394 trace_postcopy_start();
1395 qemu_mutex_lock_iothread();
1396 trace_postcopy_start_set_run();
1398 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1399 *old_vm_running
= runstate_is_running();
1400 global_state_store();
1401 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1408 * in Finish migrate and with the io-lock held everything should
1409 * be quiet, but we've potentially still got dirty pages and we
1410 * need to tell the destination to throw any pages it's already received
1413 if (ram_postcopy_send_discard_bitmap(ms
)) {
1414 error_report("postcopy send discard bitmap failed");
1419 * send rest of state - note things that are doing postcopy
1420 * will notice we're in POSTCOPY_ACTIVE and not actually
1421 * wrap their state up here
1423 qemu_file_set_rate_limit(ms
->file
, INT64_MAX
);
1424 /* Ping just for debugging, helps line traces up */
1425 qemu_savevm_send_ping(ms
->file
, 2);
1428 * While loading the device state we may trigger page transfer
1429 * requests and the fd must be free to process those, and thus
1430 * the destination must read the whole device state off the fd before
1431 * it starts processing it. Unfortunately the ad-hoc migration format
1432 * doesn't allow the destination to know the size to read without fully
1433 * parsing it through each devices load-state code (especially the open
1434 * coded devices that use get/put).
1435 * So we wrap the device state up in a package with a length at the start;
1436 * to do this we use a qemu_buf to hold the whole of the device state.
1438 QEMUFile
*fb
= qemu_bufopen("w", NULL
);
1440 error_report("Failed to create buffered file");
1444 qemu_savevm_state_complete_precopy(fb
);
1445 qemu_savevm_send_ping(fb
, 3);
1447 qemu_savevm_send_postcopy_run(fb
);
1449 /* <><> end of stuff going into the package */
1450 qsb
= qemu_buf_get(fb
);
1452 /* Now send that blob */
1453 if (qemu_savevm_send_packaged(ms
->file
, qsb
)) {
1457 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
1459 qemu_mutex_unlock_iothread();
1462 * Although this ping is just for debug, it could potentially be
1463 * used for getting a better measurement of downtime at the source.
1465 qemu_savevm_send_ping(ms
->file
, 4);
1467 ret
= qemu_file_get_error(ms
->file
);
1469 error_report("postcopy_start: Migration stream errored");
1470 migrate_set_state(ms
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1471 MIGRATION_STATUS_FAILED
);
1479 migrate_set_state(ms
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1480 MIGRATION_STATUS_FAILED
);
1481 qemu_mutex_unlock_iothread();
1486 * migration_completion: Used by migration_thread when there's not much left.
1487 * The caller 'breaks' the loop when this returns.
1489 * @s: Current migration state
1490 * @current_active_state: The migration state we expect to be in
1491 * @*old_vm_running: Pointer to old_vm_running flag
1492 * @*start_time: Pointer to time to update
1494 static void migration_completion(MigrationState
*s
, int current_active_state
,
1495 bool *old_vm_running
,
1496 int64_t *start_time
)
1500 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1501 qemu_mutex_lock_iothread();
1502 *start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1503 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1504 *old_vm_running
= runstate_is_running();
1505 ret
= global_state_store();
1508 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1510 qemu_file_set_rate_limit(s
->file
, INT64_MAX
);
1511 qemu_savevm_state_complete_precopy(s
->file
);
1514 qemu_mutex_unlock_iothread();
1519 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1520 trace_migration_completion_postcopy_end();
1522 qemu_savevm_state_complete_postcopy(s
->file
);
1523 trace_migration_completion_postcopy_end_after_complete();
1527 * If rp was opened we must clean up the thread before
1528 * cleaning everything else up (since if there are no failures
1529 * it will wait for the destination to send it's status in
1531 * Postcopy opens rp if enabled (even if it's not avtivated)
1533 if (migrate_postcopy_ram()) {
1535 trace_migration_completion_postcopy_end_before_rp();
1536 rp_error
= await_return_path_close_on_source(s
);
1537 trace_migration_completion_postcopy_end_after_rp(rp_error
);
1543 if (qemu_file_get_error(s
->file
)) {
1544 trace_migration_completion_file_err();
1548 migrate_set_state(s
, current_active_state
, MIGRATION_STATUS_COMPLETED
);
1552 migrate_set_state(s
, current_active_state
, MIGRATION_STATUS_FAILED
);
1556 * Master migration thread on the source VM.
1557 * It drives the migration and pumps the data down the outgoing channel.
1559 static void *migration_thread(void *opaque
)
1561 MigrationState
*s
= opaque
;
1562 /* Used by the bandwidth calcs, updated later */
1563 int64_t initial_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1564 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
1565 int64_t initial_bytes
= 0;
1566 int64_t max_size
= 0;
1567 int64_t start_time
= initial_time
;
1569 bool old_vm_running
= false;
1570 bool entered_postcopy
= false;
1571 /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
1572 enum MigrationStatus current_active_state
= MIGRATION_STATUS_ACTIVE
;
1574 rcu_register_thread();
1576 qemu_savevm_state_header(s
->file
);
1578 if (migrate_postcopy_ram()) {
1579 /* Now tell the dest that it should open its end so it can reply */
1580 qemu_savevm_send_open_return_path(s
->file
);
1582 /* And do a ping that will make stuff easier to debug */
1583 qemu_savevm_send_ping(s
->file
, 1);
1586 * Tell the destination that we *might* want to do postcopy later;
1587 * if the other end can't do postcopy it should fail now, nice and
1590 qemu_savevm_send_postcopy_advise(s
->file
);
1593 qemu_savevm_state_begin(s
->file
, &s
->params
);
1595 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
1596 current_active_state
= MIGRATION_STATUS_ACTIVE
;
1597 migrate_set_state(s
, MIGRATION_STATUS_SETUP
, MIGRATION_STATUS_ACTIVE
);
1599 trace_migration_thread_setup_complete();
1601 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
1602 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1603 int64_t current_time
;
1604 uint64_t pending_size
;
1606 if (!qemu_file_rate_limit(s
->file
)) {
1607 uint64_t pend_post
, pend_nonpost
;
1609 qemu_savevm_state_pending(s
->file
, max_size
, &pend_nonpost
,
1611 pending_size
= pend_nonpost
+ pend_post
;
1612 trace_migrate_pending(pending_size
, max_size
,
1613 pend_post
, pend_nonpost
);
1614 if (pending_size
&& pending_size
>= max_size
) {
1615 /* Still a significant amount to transfer */
1617 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1618 if (migrate_postcopy_ram() &&
1619 s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
&&
1620 pend_nonpost
<= max_size
&&
1621 atomic_read(&s
->start_postcopy
)) {
1623 if (!postcopy_start(s
, &old_vm_running
)) {
1624 current_active_state
= MIGRATION_STATUS_POSTCOPY_ACTIVE
;
1625 entered_postcopy
= true;
1630 /* Just another iteration step */
1631 qemu_savevm_state_iterate(s
->file
, entered_postcopy
);
1633 trace_migration_thread_low_pending(pending_size
);
1634 migration_completion(s
, current_active_state
,
1635 &old_vm_running
, &start_time
);
1640 if (qemu_file_get_error(s
->file
)) {
1641 migrate_set_state(s
, current_active_state
, MIGRATION_STATUS_FAILED
);
1642 trace_migration_thread_file_err();
1645 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1646 if (current_time
>= initial_time
+ BUFFER_DELAY
) {
1647 uint64_t transferred_bytes
= qemu_ftell(s
->file
) - initial_bytes
;
1648 uint64_t time_spent
= current_time
- initial_time
;
1649 double bandwidth
= transferred_bytes
/ time_spent
;
1650 max_size
= bandwidth
* migrate_max_downtime() / 1000000;
1652 s
->mbps
= time_spent
? (((double) transferred_bytes
* 8.0) /
1653 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0 : -1;
1655 trace_migrate_transferred(transferred_bytes
, time_spent
,
1656 bandwidth
, max_size
);
1657 /* if we haven't sent anything, we don't want to recalculate
1658 10000 is a small enough number for our purposes */
1659 if (s
->dirty_bytes_rate
&& transferred_bytes
> 10000) {
1660 s
->expected_downtime
= s
->dirty_bytes_rate
/ bandwidth
;
1663 qemu_file_reset_rate_limit(s
->file
);
1664 initial_time
= current_time
;
1665 initial_bytes
= qemu_ftell(s
->file
);
1667 if (qemu_file_rate_limit(s
->file
)) {
1668 /* usleep expects microseconds */
1669 g_usleep((initial_time
+ BUFFER_DELAY
- current_time
)*1000);
1673 trace_migration_thread_after_loop();
1674 /* If we enabled cpu throttling for auto-converge, turn it off. */
1675 cpu_throttle_stop();
1676 end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1678 qemu_mutex_lock_iothread();
1679 qemu_savevm_state_cleanup();
1680 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
1681 uint64_t transferred_bytes
= qemu_ftell(s
->file
);
1682 s
->total_time
= end_time
- s
->total_time
;
1683 if (!entered_postcopy
) {
1684 s
->downtime
= end_time
- start_time
;
1686 if (s
->total_time
) {
1687 s
->mbps
= (((double) transferred_bytes
* 8.0) /
1688 ((double) s
->total_time
)) / 1000;
1690 runstate_set(RUN_STATE_POSTMIGRATE
);
1692 if (old_vm_running
&& !entered_postcopy
) {
1696 qemu_bh_schedule(s
->cleanup_bh
);
1697 qemu_mutex_unlock_iothread();
1699 rcu_unregister_thread();
1703 void migrate_fd_connect(MigrationState
*s
)
1705 /* This is a best 1st approximation. ns to ms */
1706 s
->expected_downtime
= max_downtime
/1000000;
1707 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
1709 qemu_file_set_rate_limit(s
->file
,
1710 s
->bandwidth_limit
/ XFER_LIMIT_RATIO
);
1712 /* Notify before starting migration thread */
1713 notifier_list_notify(&migration_state_notifiers
, s
);
1716 * Open the return path; currently for postcopy but other things might
1719 if (migrate_postcopy_ram()) {
1720 if (open_return_path_on_source(s
)) {
1721 error_report("Unable to open return-path for postcopy");
1722 migrate_set_state(s
, MIGRATION_STATUS_SETUP
,
1723 MIGRATION_STATUS_FAILED
);
1724 migrate_fd_cleanup(s
);
1729 migrate_compress_threads_create();
1730 qemu_thread_create(&s
->thread
, "migration", migration_thread
, s
,
1731 QEMU_THREAD_JOINABLE
);
1732 s
->migration_thread_running
= true;
1735 PostcopyState
postcopy_state_get(void)
1737 return atomic_mb_read(&incoming_postcopy_state
);
1740 /* Set the state and return the old state */
1741 PostcopyState
postcopy_state_set(PostcopyState new_state
)
1743 return atomic_xchg(&incoming_postcopy_state
, new_state
);