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();
353 * This must happen after all error conditions are dealt with and
354 * we're sure the VM is going to be running on this host.
356 qemu_announce_self();
358 /* If global state section was not received or we are in running
359 state, we need to obey autostart. Any other state is set with
362 if (!global_state_received() ||
363 global_state_get_runstate() == RUN_STATE_RUNNING
) {
367 runstate_set(RUN_STATE_PAUSED
);
370 runstate_set(global_state_get_runstate());
372 migrate_decompress_threads_join();
374 * This must happen after any state changes since as soon as an external
375 * observer sees this event they might start to prod at the VM assuming
378 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
379 MIGRATION_STATUS_COMPLETED
);
380 qemu_bh_delete(mis
->bh
);
381 migration_incoming_state_destroy();
384 static void process_incoming_migration_co(void *opaque
)
386 QEMUFile
*f
= opaque
;
387 MigrationIncomingState
*mis
= migration_incoming_get_current();
391 mis
->from_src_file
= f
;
392 mis
->largest_page_size
= qemu_ram_pagesize_largest();
393 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
394 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
395 MIGRATION_STATUS_ACTIVE
);
396 ret
= qemu_loadvm_state(f
);
398 ps
= postcopy_state_get();
399 trace_process_incoming_migration_co_end(ret
, ps
);
400 if (ps
!= POSTCOPY_INCOMING_NONE
) {
401 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
403 * Where a migration had postcopy enabled (and thus went to advise)
404 * but managed to complete within the precopy period, we can use
407 postcopy_ram_incoming_cleanup(mis
);
408 } else if (ret
>= 0) {
410 * Postcopy was started, cleanup should happen at the end of the
413 trace_process_incoming_migration_co_postcopy_end_main();
416 /* Else if something went wrong then just fall out of the normal exit */
419 /* we get COLO info, and know if we are in COLO mode */
420 if (!ret
&& migration_incoming_enable_colo()) {
421 mis
->migration_incoming_co
= qemu_coroutine_self();
422 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
423 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
424 mis
->have_colo_incoming_thread
= true;
425 qemu_coroutine_yield();
427 /* Wait checkpoint incoming thread exit before free resource */
428 qemu_thread_join(&mis
->colo_incoming_thread
);
432 free_xbzrle_decoded_buf();
435 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
436 MIGRATION_STATUS_FAILED
);
437 error_report("load of migration failed: %s", strerror(-ret
));
438 migrate_decompress_threads_join();
442 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
443 qemu_bh_schedule(mis
->bh
);
446 void migration_fd_process_incoming(QEMUFile
*f
)
448 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, f
);
450 migrate_decompress_threads_create();
451 qemu_file_set_blocking(f
, false);
452 qemu_coroutine_enter(co
);
456 void migration_channel_process_incoming(MigrationState
*s
,
459 trace_migration_set_incoming_channel(
460 ioc
, object_get_typename(OBJECT(ioc
)));
462 if (s
->parameters
.tls_creds
&&
463 *s
->parameters
.tls_creds
&&
464 !object_dynamic_cast(OBJECT(ioc
),
465 TYPE_QIO_CHANNEL_TLS
)) {
466 Error
*local_err
= NULL
;
467 migration_tls_channel_process_incoming(s
, ioc
, &local_err
);
469 error_report_err(local_err
);
472 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
473 migration_fd_process_incoming(f
);
478 void migration_channel_connect(MigrationState
*s
,
480 const char *hostname
)
482 trace_migration_set_outgoing_channel(
483 ioc
, object_get_typename(OBJECT(ioc
)), hostname
);
485 if (s
->parameters
.tls_creds
&&
486 *s
->parameters
.tls_creds
&&
487 !object_dynamic_cast(OBJECT(ioc
),
488 TYPE_QIO_CHANNEL_TLS
)) {
489 Error
*local_err
= NULL
;
490 migration_tls_channel_connect(s
, ioc
, hostname
, &local_err
);
492 migrate_fd_error(s
, local_err
);
493 error_free(local_err
);
496 QEMUFile
*f
= qemu_fopen_channel_output(ioc
);
500 migrate_fd_connect(s
);
506 * Send a message on the return channel back to the source
509 void migrate_send_rp_message(MigrationIncomingState
*mis
,
510 enum mig_rp_message_type message_type
,
511 uint16_t len
, void *data
)
513 trace_migrate_send_rp_message((int)message_type
, len
);
514 qemu_mutex_lock(&mis
->rp_mutex
);
515 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
516 qemu_put_be16(mis
->to_src_file
, len
);
517 qemu_put_buffer(mis
->to_src_file
, data
, len
);
518 qemu_fflush(mis
->to_src_file
);
519 qemu_mutex_unlock(&mis
->rp_mutex
);
523 * Send a 'SHUT' message on the return channel with the given value
524 * to indicate that we've finished with the RP. Non-0 value indicates
527 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
532 buf
= cpu_to_be32(value
);
533 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
537 * Send a 'PONG' message on the return channel with the given value
538 * (normally in response to a 'PING')
540 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
545 buf
= cpu_to_be32(value
);
546 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
549 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
551 MigrationCapabilityStatusList
*head
= NULL
;
552 MigrationCapabilityStatusList
*caps
;
553 MigrationState
*s
= migrate_get_current();
556 caps
= NULL
; /* silence compiler warning */
557 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
558 if (i
== MIGRATION_CAPABILITY_X_COLO
&& !colo_supported()) {
562 head
= g_malloc0(sizeof(*caps
));
565 caps
->next
= g_malloc0(sizeof(*caps
));
569 g_malloc(sizeof(*caps
->value
));
570 caps
->value
->capability
= i
;
571 caps
->value
->state
= s
->enabled_capabilities
[i
];
577 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
579 MigrationParameters
*params
;
580 MigrationState
*s
= migrate_get_current();
582 params
= g_malloc0(sizeof(*params
));
583 params
->has_compress_level
= true;
584 params
->compress_level
= s
->parameters
.compress_level
;
585 params
->has_compress_threads
= true;
586 params
->compress_threads
= s
->parameters
.compress_threads
;
587 params
->has_decompress_threads
= true;
588 params
->decompress_threads
= s
->parameters
.decompress_threads
;
589 params
->has_cpu_throttle_initial
= true;
590 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
591 params
->has_cpu_throttle_increment
= true;
592 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
593 params
->has_tls_creds
= !!s
->parameters
.tls_creds
;
594 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
595 params
->has_tls_hostname
= !!s
->parameters
.tls_hostname
;
596 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
597 params
->has_max_bandwidth
= true;
598 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
599 params
->has_downtime_limit
= true;
600 params
->downtime_limit
= s
->parameters
.downtime_limit
;
601 params
->has_x_checkpoint_delay
= true;
602 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
608 * Return true if we're already in the middle of a migration
609 * (i.e. any of the active or setup states)
611 static bool migration_is_setup_or_active(int state
)
614 case MIGRATION_STATUS_ACTIVE
:
615 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
616 case MIGRATION_STATUS_SETUP
:
625 static void get_xbzrle_cache_stats(MigrationInfo
*info
)
627 if (migrate_use_xbzrle()) {
628 info
->has_xbzrle_cache
= true;
629 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
630 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
631 info
->xbzrle_cache
->bytes
= xbzrle_mig_bytes_transferred();
632 info
->xbzrle_cache
->pages
= xbzrle_mig_pages_transferred();
633 info
->xbzrle_cache
->cache_miss
= xbzrle_mig_pages_cache_miss();
634 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_mig_cache_miss_rate();
635 info
->xbzrle_cache
->overflow
= xbzrle_mig_pages_overflow();
639 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
641 info
->has_ram
= true;
642 info
->ram
= g_malloc0(sizeof(*info
->ram
));
643 info
->ram
->transferred
= ram_bytes_transferred();
644 info
->ram
->total
= ram_bytes_total();
645 info
->ram
->duplicate
= dup_mig_pages_transferred();
646 info
->ram
->skipped
= skipped_mig_pages_transferred();
647 info
->ram
->normal
= norm_mig_pages_transferred();
648 info
->ram
->normal_bytes
= norm_mig_bytes_transferred();
649 info
->ram
->mbps
= s
->mbps
;
650 info
->ram
->dirty_sync_count
= s
->dirty_sync_count
;
651 info
->ram
->postcopy_requests
= s
->postcopy_requests
;
653 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
654 info
->ram
->remaining
= ram_bytes_remaining();
655 info
->ram
->dirty_pages_rate
= s
->dirty_pages_rate
;
659 MigrationInfo
*qmp_query_migrate(Error
**errp
)
661 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
662 MigrationState
*s
= migrate_get_current();
665 case MIGRATION_STATUS_NONE
:
666 /* no migration has happened ever */
668 case MIGRATION_STATUS_SETUP
:
669 info
->has_status
= true;
670 info
->has_total_time
= false;
672 case MIGRATION_STATUS_ACTIVE
:
673 case MIGRATION_STATUS_CANCELLING
:
674 info
->has_status
= true;
675 info
->has_total_time
= true;
676 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
678 info
->has_expected_downtime
= true;
679 info
->expected_downtime
= s
->expected_downtime
;
680 info
->has_setup_time
= true;
681 info
->setup_time
= s
->setup_time
;
683 populate_ram_info(info
, s
);
685 if (blk_mig_active()) {
686 info
->has_disk
= true;
687 info
->disk
= g_malloc0(sizeof(*info
->disk
));
688 info
->disk
->transferred
= blk_mig_bytes_transferred();
689 info
->disk
->remaining
= blk_mig_bytes_remaining();
690 info
->disk
->total
= blk_mig_bytes_total();
693 if (cpu_throttle_active()) {
694 info
->has_cpu_throttle_percentage
= true;
695 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
698 get_xbzrle_cache_stats(info
);
700 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
701 /* Mostly the same as active; TODO add some postcopy stats */
702 info
->has_status
= true;
703 info
->has_total_time
= true;
704 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
706 info
->has_expected_downtime
= true;
707 info
->expected_downtime
= s
->expected_downtime
;
708 info
->has_setup_time
= true;
709 info
->setup_time
= s
->setup_time
;
711 populate_ram_info(info
, s
);
713 if (blk_mig_active()) {
714 info
->has_disk
= true;
715 info
->disk
= g_malloc0(sizeof(*info
->disk
));
716 info
->disk
->transferred
= blk_mig_bytes_transferred();
717 info
->disk
->remaining
= blk_mig_bytes_remaining();
718 info
->disk
->total
= blk_mig_bytes_total();
721 get_xbzrle_cache_stats(info
);
723 case MIGRATION_STATUS_COLO
:
724 info
->has_status
= true;
725 /* TODO: display COLO specific information (checkpoint info etc.) */
727 case MIGRATION_STATUS_COMPLETED
:
728 get_xbzrle_cache_stats(info
);
730 info
->has_status
= true;
731 info
->has_total_time
= true;
732 info
->total_time
= s
->total_time
;
733 info
->has_downtime
= true;
734 info
->downtime
= s
->downtime
;
735 info
->has_setup_time
= true;
736 info
->setup_time
= s
->setup_time
;
738 populate_ram_info(info
, s
);
740 case MIGRATION_STATUS_FAILED
:
741 info
->has_status
= true;
743 info
->has_error_desc
= true;
744 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
747 case MIGRATION_STATUS_CANCELLED
:
748 info
->has_status
= true;
751 info
->status
= s
->state
;
756 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
759 MigrationState
*s
= migrate_get_current();
760 MigrationCapabilityStatusList
*cap
;
761 bool old_postcopy_cap
= migrate_postcopy_ram();
763 if (migration_is_setup_or_active(s
->state
)) {
764 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
768 for (cap
= params
; cap
; cap
= cap
->next
) {
769 if (cap
->value
->capability
== MIGRATION_CAPABILITY_X_COLO
) {
770 if (!colo_supported()) {
771 error_setg(errp
, "COLO is not currently supported, please"
772 " configure with --enable-colo option in order to"
773 " support COLO feature");
777 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
780 if (migrate_postcopy_ram()) {
781 if (migrate_use_compression()) {
782 /* The decompression threads asynchronously write into RAM
783 * rather than use the atomic copies needed to avoid
784 * userfaulting. It should be possible to fix the decompression
785 * threads for compatibility in future.
787 error_report("Postcopy is not currently compatible with "
789 s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
] =
792 /* This check is reasonably expensive, so only when it's being
793 * set the first time, also it's only the destination that needs
796 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
797 !postcopy_ram_supported_by_host()) {
798 /* postcopy_ram_supported_by_host will have emitted a more
801 error_report("Postcopy is not supported");
802 s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
] =
808 void qmp_migrate_set_parameters(MigrationParameters
*params
, Error
**errp
)
810 MigrationState
*s
= migrate_get_current();
812 if (params
->has_compress_level
&&
813 (params
->compress_level
< 0 || params
->compress_level
> 9)) {
814 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
815 "is invalid, it should be in the range of 0 to 9");
818 if (params
->has_compress_threads
&&
819 (params
->compress_threads
< 1 || params
->compress_threads
> 255)) {
820 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
822 "is invalid, it should be in the range of 1 to 255");
825 if (params
->has_decompress_threads
&&
826 (params
->decompress_threads
< 1 || params
->decompress_threads
> 255)) {
827 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
828 "decompress_threads",
829 "is invalid, it should be in the range of 1 to 255");
832 if (params
->has_cpu_throttle_initial
&&
833 (params
->cpu_throttle_initial
< 1 ||
834 params
->cpu_throttle_initial
> 99)) {
835 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
836 "cpu_throttle_initial",
837 "an integer in the range of 1 to 99");
840 if (params
->has_cpu_throttle_increment
&&
841 (params
->cpu_throttle_increment
< 1 ||
842 params
->cpu_throttle_increment
> 99)) {
843 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
844 "cpu_throttle_increment",
845 "an integer in the range of 1 to 99");
848 if (params
->has_max_bandwidth
&&
849 (params
->max_bandwidth
< 0 || params
->max_bandwidth
> SIZE_MAX
)) {
850 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
851 " range of 0 to %zu bytes/second", SIZE_MAX
);
854 if (params
->has_downtime_limit
&&
855 (params
->downtime_limit
< 0 ||
856 params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
857 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
858 "the range of 0 to %d milliseconds",
859 MAX_MIGRATE_DOWNTIME
);
862 if (params
->has_x_checkpoint_delay
&& (params
->x_checkpoint_delay
< 0)) {
863 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
864 "x_checkpoint_delay",
865 "is invalid, it should be positive");
868 if (params
->has_compress_level
) {
869 s
->parameters
.compress_level
= params
->compress_level
;
871 if (params
->has_compress_threads
) {
872 s
->parameters
.compress_threads
= params
->compress_threads
;
874 if (params
->has_decompress_threads
) {
875 s
->parameters
.decompress_threads
= params
->decompress_threads
;
877 if (params
->has_cpu_throttle_initial
) {
878 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
880 if (params
->has_cpu_throttle_increment
) {
881 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
883 if (params
->has_tls_creds
) {
884 g_free(s
->parameters
.tls_creds
);
885 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
);
887 if (params
->has_tls_hostname
) {
888 g_free(s
->parameters
.tls_hostname
);
889 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
);
891 if (params
->has_max_bandwidth
) {
892 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
893 if (s
->to_dst_file
) {
894 qemu_file_set_rate_limit(s
->to_dst_file
,
895 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
898 if (params
->has_downtime_limit
) {
899 s
->parameters
.downtime_limit
= params
->downtime_limit
;
902 if (params
->has_x_checkpoint_delay
) {
903 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
904 if (migration_in_colo_state()) {
905 colo_checkpoint_notify(s
);
911 void qmp_migrate_start_postcopy(Error
**errp
)
913 MigrationState
*s
= migrate_get_current();
915 if (!migrate_postcopy_ram()) {
916 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
917 " the start of migration");
921 if (s
->state
== MIGRATION_STATUS_NONE
) {
922 error_setg(errp
, "Postcopy must be started after migration has been"
927 * we don't error if migration has finished since that would be racy
928 * with issuing this command.
930 atomic_set(&s
->start_postcopy
, true);
933 /* shared migration helpers */
935 void migrate_set_state(int *state
, int old_state
, int new_state
)
937 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
938 trace_migrate_set_state(new_state
);
939 migrate_generate_event(new_state
);
943 static void migrate_fd_cleanup(void *opaque
)
945 MigrationState
*s
= opaque
;
947 qemu_bh_delete(s
->cleanup_bh
);
948 s
->cleanup_bh
= NULL
;
952 if (s
->to_dst_file
) {
953 trace_migrate_fd_cleanup();
954 qemu_mutex_unlock_iothread();
955 if (s
->migration_thread_running
) {
956 qemu_thread_join(&s
->thread
);
957 s
->migration_thread_running
= false;
959 qemu_mutex_lock_iothread();
961 migrate_compress_threads_join();
962 qemu_fclose(s
->to_dst_file
);
963 s
->to_dst_file
= NULL
;
966 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
967 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
969 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
970 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
971 MIGRATION_STATUS_CANCELLED
);
974 notifier_list_notify(&migration_state_notifiers
, s
);
977 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
979 trace_migrate_fd_error(error_get_pretty(error
));
980 assert(s
->to_dst_file
== NULL
);
981 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
982 MIGRATION_STATUS_FAILED
);
984 s
->error
= error_copy(error
);
986 notifier_list_notify(&migration_state_notifiers
, s
);
989 static void migrate_fd_cancel(MigrationState
*s
)
992 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
993 trace_migrate_fd_cancel();
995 if (s
->rp_state
.from_dst_file
) {
996 /* shutdown the rp socket, so causing the rp thread to shutdown */
997 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1001 old_state
= s
->state
;
1002 if (!migration_is_setup_or_active(old_state
)) {
1005 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1006 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1009 * If we're unlucky the migration code might be stuck somewhere in a
1010 * send/write while the network has failed and is waiting to timeout;
1011 * if we've got shutdown(2) available then we can force it to quit.
1012 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1013 * called in a bh, so there is no race against this cancel.
1015 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1016 qemu_file_shutdown(f
);
1018 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1019 Error
*local_err
= NULL
;
1021 bdrv_invalidate_cache_all(&local_err
);
1023 error_report_err(local_err
);
1025 s
->block_inactive
= false;
1030 void add_migration_state_change_notifier(Notifier
*notify
)
1032 notifier_list_add(&migration_state_notifiers
, notify
);
1035 void remove_migration_state_change_notifier(Notifier
*notify
)
1037 notifier_remove(notify
);
1040 bool migration_in_setup(MigrationState
*s
)
1042 return s
->state
== MIGRATION_STATUS_SETUP
;
1045 bool migration_has_finished(MigrationState
*s
)
1047 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1050 bool migration_has_failed(MigrationState
*s
)
1052 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1053 s
->state
== MIGRATION_STATUS_FAILED
);
1056 bool migration_in_postcopy(MigrationState
*s
)
1058 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1061 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1063 return migration_in_postcopy(s
) && s
->postcopy_after_devices
;
1066 bool migration_is_idle(MigrationState
*s
)
1069 s
= migrate_get_current();
1073 case MIGRATION_STATUS_NONE
:
1074 case MIGRATION_STATUS_CANCELLED
:
1075 case MIGRATION_STATUS_COMPLETED
:
1076 case MIGRATION_STATUS_FAILED
:
1078 case MIGRATION_STATUS_SETUP
:
1079 case MIGRATION_STATUS_CANCELLING
:
1080 case MIGRATION_STATUS_ACTIVE
:
1081 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1082 case MIGRATION_STATUS_COLO
:
1084 case MIGRATION_STATUS__MAX
:
1085 g_assert_not_reached();
1091 MigrationState
*migrate_init(const MigrationParams
*params
)
1093 MigrationState
*s
= migrate_get_current();
1096 * Reinitialise all migration state, except
1097 * parameters/capabilities that the user set, and
1103 s
->to_dst_file
= NULL
;
1104 s
->state
= MIGRATION_STATUS_NONE
;
1105 s
->params
= *params
;
1106 s
->rp_state
.from_dst_file
= NULL
;
1107 s
->rp_state
.error
= false;
1110 s
->expected_downtime
= 0;
1111 s
->dirty_pages_rate
= 0;
1112 s
->dirty_bytes_rate
= 0;
1114 s
->dirty_sync_count
= 0;
1115 s
->start_postcopy
= false;
1116 s
->postcopy_after_devices
= false;
1117 s
->postcopy_requests
= 0;
1118 s
->migration_thread_running
= false;
1119 s
->last_req_rb
= NULL
;
1120 error_free(s
->error
);
1123 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1125 QSIMPLEQ_INIT(&s
->src_page_requests
);
1127 s
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1131 static GSList
*migration_blockers
;
1133 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1135 if (only_migratable
) {
1136 error_propagate(errp
, error_copy(reason
));
1137 error_prepend(errp
, "disallowing migration blocker "
1138 "(--only_migratable) for: ");
1142 if (migration_is_idle(NULL
)) {
1143 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1147 error_propagate(errp
, error_copy(reason
));
1148 error_prepend(errp
, "disallowing migration blocker (migration in "
1153 void migrate_del_blocker(Error
*reason
)
1155 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1158 int check_migratable(Object
*obj
, Error
**err
)
1160 DeviceClass
*dc
= DEVICE_GET_CLASS(obj
);
1161 if (only_migratable
&& dc
->vmsd
) {
1162 if (dc
->vmsd
->unmigratable
) {
1163 error_setg(err
, "Device %s is not migratable, but "
1164 "--only-migratable was specified",
1165 object_get_typename(obj
));
1173 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1175 Error
*local_err
= NULL
;
1176 static bool once
= true;
1178 if (!deferred_incoming
) {
1179 error_setg(errp
, "For use with '-incoming defer'");
1183 error_setg(errp
, "The incoming migration has already been started");
1186 qemu_start_incoming_migration(uri
, &local_err
);
1189 error_propagate(errp
, local_err
);
1196 bool migration_is_blocked(Error
**errp
)
1198 if (qemu_savevm_state_blocked(errp
)) {
1202 if (migration_blockers
) {
1203 *errp
= error_copy(migration_blockers
->data
);
1210 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1211 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1214 Error
*local_err
= NULL
;
1215 MigrationState
*s
= migrate_get_current();
1216 MigrationParams params
;
1219 params
.blk
= has_blk
&& blk
;
1220 params
.shared
= has_inc
&& inc
;
1222 if (migration_is_setup_or_active(s
->state
) ||
1223 s
->state
== MIGRATION_STATUS_CANCELLING
||
1224 s
->state
== MIGRATION_STATUS_COLO
) {
1225 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1228 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1229 error_setg(errp
, "Guest is waiting for an incoming migration");
1233 if (migration_is_blocked(errp
)) {
1237 s
= migrate_init(¶ms
);
1239 if (strstart(uri
, "tcp:", &p
)) {
1240 tcp_start_outgoing_migration(s
, p
, &local_err
);
1242 } else if (strstart(uri
, "rdma:", &p
)) {
1243 rdma_start_outgoing_migration(s
, p
, &local_err
);
1245 } else if (strstart(uri
, "exec:", &p
)) {
1246 exec_start_outgoing_migration(s
, p
, &local_err
);
1247 } else if (strstart(uri
, "unix:", &p
)) {
1248 unix_start_outgoing_migration(s
, p
, &local_err
);
1249 } else if (strstart(uri
, "fd:", &p
)) {
1250 fd_start_outgoing_migration(s
, p
, &local_err
);
1252 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1253 "a valid migration protocol");
1254 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1255 MIGRATION_STATUS_FAILED
);
1260 migrate_fd_error(s
, local_err
);
1261 error_propagate(errp
, local_err
);
1266 void qmp_migrate_cancel(Error
**errp
)
1268 migrate_fd_cancel(migrate_get_current());
1271 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1273 MigrationState
*s
= migrate_get_current();
1276 /* Check for truncation */
1277 if (value
!= (size_t)value
) {
1278 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1279 "exceeding address space");
1283 /* Cache should not be larger than guest ram size */
1284 if (value
> ram_bytes_total()) {
1285 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1286 "exceeds guest ram size ");
1290 new_size
= xbzrle_cache_resize(value
);
1292 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1293 "is smaller than page size");
1297 s
->xbzrle_cache_size
= new_size
;
1300 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1302 return migrate_xbzrle_cache_size();
1305 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1307 MigrationParameters p
= {
1308 .has_max_bandwidth
= true,
1309 .max_bandwidth
= value
,
1312 qmp_migrate_set_parameters(&p
, errp
);
1315 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1317 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
1318 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1319 "the range of 0 to %d seconds",
1320 MAX_MIGRATE_DOWNTIME_SECONDS
);
1324 value
*= 1000; /* Convert to milliseconds */
1325 value
= MAX(0, MIN(INT64_MAX
, value
));
1327 MigrationParameters p
= {
1328 .has_downtime_limit
= true,
1329 .downtime_limit
= value
,
1332 qmp_migrate_set_parameters(&p
, errp
);
1335 bool migrate_release_ram(void)
1339 s
= migrate_get_current();
1341 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
1344 bool migrate_postcopy_ram(void)
1348 s
= migrate_get_current();
1350 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1353 bool migrate_auto_converge(void)
1357 s
= migrate_get_current();
1359 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1362 bool migrate_zero_blocks(void)
1366 s
= migrate_get_current();
1368 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
1371 bool migrate_use_compression(void)
1375 s
= migrate_get_current();
1377 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
1380 int migrate_compress_level(void)
1384 s
= migrate_get_current();
1386 return s
->parameters
.compress_level
;
1389 int migrate_compress_threads(void)
1393 s
= migrate_get_current();
1395 return s
->parameters
.compress_threads
;
1398 int migrate_decompress_threads(void)
1402 s
= migrate_get_current();
1404 return s
->parameters
.decompress_threads
;
1407 bool migrate_use_events(void)
1411 s
= migrate_get_current();
1413 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
1416 int migrate_use_xbzrle(void)
1420 s
= migrate_get_current();
1422 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
1425 int64_t migrate_xbzrle_cache_size(void)
1429 s
= migrate_get_current();
1431 return s
->xbzrle_cache_size
;
1434 /* migration thread support */
1436 * Something bad happened to the RP stream, mark an error
1437 * The caller shall print or trace something to indicate why
1439 static void mark_source_rp_bad(MigrationState
*s
)
1441 s
->rp_state
.error
= true;
1444 static struct rp_cmd_args
{
1445 ssize_t len
; /* -1 = variable */
1448 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1449 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1450 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1451 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1452 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1453 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1457 * Process a request for pages received on the return path,
1458 * We're allowed to send more than requested (e.g. to round to our page size)
1459 * and we don't need to send pages that have already been sent.
1461 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1462 ram_addr_t start
, size_t len
)
1464 long our_host_ps
= getpagesize();
1466 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1469 * Since we currently insist on matching page sizes, just sanity check
1470 * we're being asked for whole host pages.
1472 if (start
& (our_host_ps
-1) ||
1473 (len
& (our_host_ps
-1))) {
1474 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1475 " len: %zd", __func__
, start
, len
);
1476 mark_source_rp_bad(ms
);
1480 if (ram_save_queue_pages(ms
, rbname
, start
, len
)) {
1481 mark_source_rp_bad(ms
);
1486 * Handles messages sent on the return path towards the source VM
1489 static void *source_return_path_thread(void *opaque
)
1491 MigrationState
*ms
= opaque
;
1492 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1493 uint16_t header_len
, header_type
;
1495 uint32_t tmp32
, sibling_error
;
1496 ram_addr_t start
= 0; /* =0 to silence warning */
1497 size_t len
= 0, expected_len
;
1500 trace_source_return_path_thread_entry();
1501 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1502 migration_is_setup_or_active(ms
->state
)) {
1503 trace_source_return_path_thread_loop_top();
1504 header_type
= qemu_get_be16(rp
);
1505 header_len
= qemu_get_be16(rp
);
1507 if (header_type
>= MIG_RP_MSG_MAX
||
1508 header_type
== MIG_RP_MSG_INVALID
) {
1509 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1510 header_type
, header_len
);
1511 mark_source_rp_bad(ms
);
1515 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1516 header_len
!= rp_cmd_args
[header_type
].len
) ||
1517 header_len
> sizeof(buf
)) {
1518 error_report("RP: Received '%s' message (0x%04x) with"
1519 "incorrect length %d expecting %zu",
1520 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1521 (size_t)rp_cmd_args
[header_type
].len
);
1522 mark_source_rp_bad(ms
);
1526 /* We know we've got a valid header by this point */
1527 res
= qemu_get_buffer(rp
, buf
, header_len
);
1528 if (res
!= header_len
) {
1529 error_report("RP: Failed reading data for message 0x%04x"
1530 " read %d expected %d",
1531 header_type
, res
, header_len
);
1532 mark_source_rp_bad(ms
);
1536 /* OK, we have the message and the data */
1537 switch (header_type
) {
1538 case MIG_RP_MSG_SHUT
:
1539 sibling_error
= ldl_be_p(buf
);
1540 trace_source_return_path_thread_shut(sibling_error
);
1541 if (sibling_error
) {
1542 error_report("RP: Sibling indicated error %d", sibling_error
);
1543 mark_source_rp_bad(ms
);
1546 * We'll let the main thread deal with closing the RP
1547 * we could do a shutdown(2) on it, but we're the only user
1548 * anyway, so there's nothing gained.
1552 case MIG_RP_MSG_PONG
:
1553 tmp32
= ldl_be_p(buf
);
1554 trace_source_return_path_thread_pong(tmp32
);
1557 case MIG_RP_MSG_REQ_PAGES
:
1558 start
= ldq_be_p(buf
);
1559 len
= ldl_be_p(buf
+ 8);
1560 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1563 case MIG_RP_MSG_REQ_PAGES_ID
:
1564 expected_len
= 12 + 1; /* header + termination */
1566 if (header_len
>= expected_len
) {
1567 start
= ldq_be_p(buf
);
1568 len
= ldl_be_p(buf
+ 8);
1569 /* Now we expect an idstr */
1570 tmp32
= buf
[12]; /* Length of the following idstr */
1571 buf
[13 + tmp32
] = '\0';
1572 expected_len
+= tmp32
;
1574 if (header_len
!= expected_len
) {
1575 error_report("RP: Req_Page_id with length %d expecting %zd",
1576 header_len
, expected_len
);
1577 mark_source_rp_bad(ms
);
1580 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
1587 if (qemu_file_get_error(rp
)) {
1588 trace_source_return_path_thread_bad_end();
1589 mark_source_rp_bad(ms
);
1592 trace_source_return_path_thread_end();
1594 ms
->rp_state
.from_dst_file
= NULL
;
1599 static int open_return_path_on_source(MigrationState
*ms
)
1602 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
1603 if (!ms
->rp_state
.from_dst_file
) {
1607 trace_open_return_path_on_source();
1608 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
1609 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
1611 trace_open_return_path_on_source_continue();
1616 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1617 static int await_return_path_close_on_source(MigrationState
*ms
)
1620 * If this is a normal exit then the destination will send a SHUT and the
1621 * rp_thread will exit, however if there's an error we need to cause
1624 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
1626 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1627 * waiting for the destination.
1629 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
1630 mark_source_rp_bad(ms
);
1632 trace_await_return_path_close_on_source_joining();
1633 qemu_thread_join(&ms
->rp_state
.rp_thread
);
1634 trace_await_return_path_close_on_source_close();
1635 return ms
->rp_state
.error
;
1639 * Switch from normal iteration to postcopy
1640 * Returns non-0 on error
1642 static int postcopy_start(MigrationState
*ms
, bool *old_vm_running
)
1645 QIOChannelBuffer
*bioc
;
1647 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1648 bool restart_block
= false;
1649 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
1650 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1652 trace_postcopy_start();
1653 qemu_mutex_lock_iothread();
1654 trace_postcopy_start_set_run();
1656 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1657 *old_vm_running
= runstate_is_running();
1658 global_state_store();
1659 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1664 ret
= bdrv_inactivate_all();
1668 restart_block
= true;
1671 * Cause any non-postcopiable, but iterative devices to
1672 * send out their final data.
1674 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true);
1677 * in Finish migrate and with the io-lock held everything should
1678 * be quiet, but we've potentially still got dirty pages and we
1679 * need to tell the destination to throw any pages it's already received
1682 if (ram_postcopy_send_discard_bitmap(ms
)) {
1683 error_report("postcopy send discard bitmap failed");
1688 * send rest of state - note things that are doing postcopy
1689 * will notice we're in POSTCOPY_ACTIVE and not actually
1690 * wrap their state up here
1692 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
1693 /* Ping just for debugging, helps line traces up */
1694 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
1697 * While loading the device state we may trigger page transfer
1698 * requests and the fd must be free to process those, and thus
1699 * the destination must read the whole device state off the fd before
1700 * it starts processing it. Unfortunately the ad-hoc migration format
1701 * doesn't allow the destination to know the size to read without fully
1702 * parsing it through each devices load-state code (especially the open
1703 * coded devices that use get/put).
1704 * So we wrap the device state up in a package with a length at the start;
1705 * to do this we use a qemu_buf to hold the whole of the device state.
1707 bioc
= qio_channel_buffer_new(4096);
1708 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
1709 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
1710 object_unref(OBJECT(bioc
));
1713 * Make sure the receiver can get incoming pages before we send the rest
1716 qemu_savevm_send_postcopy_listen(fb
);
1718 qemu_savevm_state_complete_precopy(fb
, false);
1719 qemu_savevm_send_ping(fb
, 3);
1721 qemu_savevm_send_postcopy_run(fb
);
1723 /* <><> end of stuff going into the package */
1725 /* Last point of recovery; as soon as we send the package the destination
1726 * can open devices and potentially start running.
1727 * Lets just check again we've not got any errors.
1729 ret
= qemu_file_get_error(ms
->to_dst_file
);
1731 error_report("postcopy_start: Migration stream errored (pre package)");
1735 restart_block
= false;
1737 /* Now send that blob */
1738 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
1743 /* Send a notify to give a chance for anything that needs to happen
1744 * at the transition to postcopy and after the device state; in particular
1745 * spice needs to trigger a transition now
1747 ms
->postcopy_after_devices
= true;
1748 notifier_list_notify(&migration_state_notifiers
, ms
);
1750 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
1752 qemu_mutex_unlock_iothread();
1755 * Although this ping is just for debug, it could potentially be
1756 * used for getting a better measurement of downtime at the source.
1758 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
1760 if (migrate_release_ram()) {
1761 ram_postcopy_migrated_memory_release(ms
);
1764 ret
= qemu_file_get_error(ms
->to_dst_file
);
1766 error_report("postcopy_start: Migration stream errored");
1767 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1768 MIGRATION_STATUS_FAILED
);
1776 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1777 MIGRATION_STATUS_FAILED
);
1778 if (restart_block
) {
1779 /* A failure happened early enough that we know the destination hasn't
1780 * accessed block devices, so we're safe to recover.
1782 Error
*local_err
= NULL
;
1784 bdrv_invalidate_cache_all(&local_err
);
1786 error_report_err(local_err
);
1789 qemu_mutex_unlock_iothread();
1794 * migration_completion: Used by migration_thread when there's not much left.
1795 * The caller 'breaks' the loop when this returns.
1797 * @s: Current migration state
1798 * @current_active_state: The migration state we expect to be in
1799 * @*old_vm_running: Pointer to old_vm_running flag
1800 * @*start_time: Pointer to time to update
1802 static void migration_completion(MigrationState
*s
, int current_active_state
,
1803 bool *old_vm_running
,
1804 int64_t *start_time
)
1808 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1809 qemu_mutex_lock_iothread();
1810 *start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1811 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1812 *old_vm_running
= runstate_is_running();
1813 ret
= global_state_store();
1816 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1818 * Don't mark the image with BDRV_O_INACTIVE flag if
1819 * we will go into COLO stage later.
1821 if (ret
>= 0 && !migrate_colo_enabled()) {
1822 ret
= bdrv_inactivate_all();
1825 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
1826 qemu_savevm_state_complete_precopy(s
->to_dst_file
, false);
1827 s
->block_inactive
= true;
1830 qemu_mutex_unlock_iothread();
1835 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1836 trace_migration_completion_postcopy_end();
1838 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
1839 trace_migration_completion_postcopy_end_after_complete();
1843 * If rp was opened we must clean up the thread before
1844 * cleaning everything else up (since if there are no failures
1845 * it will wait for the destination to send it's status in
1847 * Postcopy opens rp if enabled (even if it's not avtivated)
1849 if (migrate_postcopy_ram()) {
1851 trace_migration_completion_postcopy_end_before_rp();
1852 rp_error
= await_return_path_close_on_source(s
);
1853 trace_migration_completion_postcopy_end_after_rp(rp_error
);
1855 goto fail_invalidate
;
1859 if (qemu_file_get_error(s
->to_dst_file
)) {
1860 trace_migration_completion_file_err();
1861 goto fail_invalidate
;
1864 if (!migrate_colo_enabled()) {
1865 migrate_set_state(&s
->state
, current_active_state
,
1866 MIGRATION_STATUS_COMPLETED
);
1872 /* If not doing postcopy, vm_start() will be called: let's regain
1873 * control on images.
1875 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1876 Error
*local_err
= NULL
;
1878 qemu_mutex_lock_iothread();
1879 bdrv_invalidate_cache_all(&local_err
);
1881 error_report_err(local_err
);
1883 s
->block_inactive
= false;
1885 qemu_mutex_unlock_iothread();
1889 migrate_set_state(&s
->state
, current_active_state
,
1890 MIGRATION_STATUS_FAILED
);
1893 bool migrate_colo_enabled(void)
1895 MigrationState
*s
= migrate_get_current();
1896 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
1900 * Master migration thread on the source VM.
1901 * It drives the migration and pumps the data down the outgoing channel.
1903 static void *migration_thread(void *opaque
)
1905 MigrationState
*s
= opaque
;
1906 /* Used by the bandwidth calcs, updated later */
1907 int64_t initial_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1908 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
1909 int64_t initial_bytes
= 0;
1910 int64_t max_size
= 0;
1911 int64_t start_time
= initial_time
;
1913 bool old_vm_running
= false;
1914 bool entered_postcopy
= false;
1915 /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
1916 enum MigrationStatus current_active_state
= MIGRATION_STATUS_ACTIVE
;
1917 bool enable_colo
= migrate_colo_enabled();
1919 rcu_register_thread();
1921 qemu_savevm_state_header(s
->to_dst_file
);
1923 if (migrate_postcopy_ram()) {
1924 /* Now tell the dest that it should open its end so it can reply */
1925 qemu_savevm_send_open_return_path(s
->to_dst_file
);
1927 /* And do a ping that will make stuff easier to debug */
1928 qemu_savevm_send_ping(s
->to_dst_file
, 1);
1931 * Tell the destination that we *might* want to do postcopy later;
1932 * if the other end can't do postcopy it should fail now, nice and
1935 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
1938 qemu_savevm_state_begin(s
->to_dst_file
, &s
->params
);
1940 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
1941 current_active_state
= MIGRATION_STATUS_ACTIVE
;
1942 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1943 MIGRATION_STATUS_ACTIVE
);
1945 trace_migration_thread_setup_complete();
1947 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
1948 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1949 int64_t current_time
;
1950 uint64_t pending_size
;
1952 if (!qemu_file_rate_limit(s
->to_dst_file
)) {
1953 uint64_t pend_post
, pend_nonpost
;
1955 qemu_savevm_state_pending(s
->to_dst_file
, max_size
, &pend_nonpost
,
1957 pending_size
= pend_nonpost
+ pend_post
;
1958 trace_migrate_pending(pending_size
, max_size
,
1959 pend_post
, pend_nonpost
);
1960 if (pending_size
&& pending_size
>= max_size
) {
1961 /* Still a significant amount to transfer */
1963 if (migrate_postcopy_ram() &&
1964 s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
&&
1965 pend_nonpost
<= max_size
&&
1966 atomic_read(&s
->start_postcopy
)) {
1968 if (!postcopy_start(s
, &old_vm_running
)) {
1969 current_active_state
= MIGRATION_STATUS_POSTCOPY_ACTIVE
;
1970 entered_postcopy
= true;
1975 /* Just another iteration step */
1976 qemu_savevm_state_iterate(s
->to_dst_file
, entered_postcopy
);
1978 trace_migration_thread_low_pending(pending_size
);
1979 migration_completion(s
, current_active_state
,
1980 &old_vm_running
, &start_time
);
1985 if (qemu_file_get_error(s
->to_dst_file
)) {
1986 migrate_set_state(&s
->state
, current_active_state
,
1987 MIGRATION_STATUS_FAILED
);
1988 trace_migration_thread_file_err();
1991 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1992 if (current_time
>= initial_time
+ BUFFER_DELAY
) {
1993 uint64_t transferred_bytes
= qemu_ftell(s
->to_dst_file
) -
1995 uint64_t time_spent
= current_time
- initial_time
;
1996 double bandwidth
= (double)transferred_bytes
/ time_spent
;
1997 max_size
= bandwidth
* s
->parameters
.downtime_limit
;
1999 s
->mbps
= (((double) transferred_bytes
* 8.0) /
2000 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2002 trace_migrate_transferred(transferred_bytes
, time_spent
,
2003 bandwidth
, max_size
);
2004 /* if we haven't sent anything, we don't want to recalculate
2005 10000 is a small enough number for our purposes */
2006 if (s
->dirty_bytes_rate
&& transferred_bytes
> 10000) {
2007 s
->expected_downtime
= s
->dirty_bytes_rate
/ bandwidth
;
2010 qemu_file_reset_rate_limit(s
->to_dst_file
);
2011 initial_time
= current_time
;
2012 initial_bytes
= qemu_ftell(s
->to_dst_file
);
2014 if (qemu_file_rate_limit(s
->to_dst_file
)) {
2015 /* usleep expects microseconds */
2016 g_usleep((initial_time
+ BUFFER_DELAY
- current_time
)*1000);
2020 trace_migration_thread_after_loop();
2021 /* If we enabled cpu throttling for auto-converge, turn it off. */
2022 cpu_throttle_stop();
2023 end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2025 qemu_mutex_lock_iothread();
2027 * The resource has been allocated by migration will be reused in COLO
2028 * process, so don't release them.
2031 qemu_savevm_state_cleanup();
2033 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
2034 uint64_t transferred_bytes
= qemu_ftell(s
->to_dst_file
);
2035 s
->total_time
= end_time
- s
->total_time
;
2036 if (!entered_postcopy
) {
2037 s
->downtime
= end_time
- start_time
;
2039 if (s
->total_time
) {
2040 s
->mbps
= (((double) transferred_bytes
* 8.0) /
2041 ((double) s
->total_time
)) / 1000;
2043 runstate_set(RUN_STATE_POSTMIGRATE
);
2045 if (s
->state
== MIGRATION_STATUS_ACTIVE
&& enable_colo
) {
2046 migrate_start_colo_process(s
);
2047 qemu_savevm_state_cleanup();
2049 * Fixme: we will run VM in COLO no matter its old running state.
2050 * After exited COLO, we will keep running.
2052 old_vm_running
= true;
2054 if (old_vm_running
&& !entered_postcopy
) {
2057 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2058 runstate_set(RUN_STATE_POSTMIGRATE
);
2062 qemu_bh_schedule(s
->cleanup_bh
);
2063 qemu_mutex_unlock_iothread();
2065 rcu_unregister_thread();
2069 void migrate_fd_connect(MigrationState
*s
)
2071 s
->expected_downtime
= s
->parameters
.downtime_limit
;
2072 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
2074 qemu_file_set_blocking(s
->to_dst_file
, true);
2075 qemu_file_set_rate_limit(s
->to_dst_file
,
2076 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
2078 /* Notify before starting migration thread */
2079 notifier_list_notify(&migration_state_notifiers
, s
);
2082 * Open the return path; currently for postcopy but other things might
2085 if (migrate_postcopy_ram()) {
2086 if (open_return_path_on_source(s
)) {
2087 error_report("Unable to open return-path for postcopy");
2088 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2089 MIGRATION_STATUS_FAILED
);
2090 migrate_fd_cleanup(s
);
2095 migrate_compress_threads_create();
2096 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
2097 QEMU_THREAD_JOINABLE
);
2098 s
->migration_thread_running
= true;
2101 PostcopyState
postcopy_state_get(void)
2103 return atomic_mb_read(&incoming_postcopy_state
);
2106 /* Set the state and return the old state */
2107 PostcopyState
postcopy_state_set(PostcopyState new_state
)
2109 return atomic_xchg(&incoming_postcopy_state
, new_state
);