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/blocker.h"
21 #include "migration/migration.h"
22 #include "migration/qemu-file.h"
23 #include "sysemu/sysemu.h"
24 #include "block/block.h"
25 #include "qapi/qmp/qerror.h"
26 #include "qapi/util.h"
27 #include "qemu/sockets.h"
29 #include "migration/block.h"
30 #include "postcopy-ram.h"
31 #include "qemu/thread.h"
32 #include "qmp-commands.h"
34 #include "qapi-event.h"
36 #include "exec/memory.h"
37 #include "exec/address-spaces.h"
38 #include "io/channel-buffer.h"
39 #include "io/channel-tls.h"
40 #include "migration/colo.h"
42 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
44 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
46 #define BUFFER_DELAY 100
47 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
49 /* Time in milliseconds we are allowed to stop the source,
50 * for sending the last part */
51 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
53 /* Maximum migrate downtime set to 2000 seconds */
54 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
55 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
57 /* Default compression thread count */
58 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
59 /* Default decompression thread count, usually decompression is at
60 * least 4 times as fast as compression.*/
61 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
62 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
63 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
64 /* Define default autoconverge cpu throttle migration parameters */
65 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
66 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
68 /* Migration XBZRLE default cache size */
69 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
71 /* The delay time (in ms) between two COLO checkpoints
72 * Note: Please change this default value to 10000 when we support hybrid mode.
74 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
76 static NotifierList migration_state_notifiers
=
77 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
79 static bool deferred_incoming
;
81 /* When we add fault tolerance, we could have several
82 migrations at once. For now we don't need to add
83 dynamic creation of migration */
86 MigrationState
*migrate_get_current(void)
89 static MigrationState current_migration
= {
90 .state
= MIGRATION_STATUS_NONE
,
91 .xbzrle_cache_size
= DEFAULT_MIGRATE_CACHE_SIZE
,
94 .compress_level
= DEFAULT_MIGRATE_COMPRESS_LEVEL
,
95 .compress_threads
= DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
,
96 .decompress_threads
= DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
,
97 .cpu_throttle_initial
= DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
,
98 .cpu_throttle_increment
= DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
,
99 .max_bandwidth
= MAX_THROTTLE
,
100 .downtime_limit
= DEFAULT_MIGRATE_SET_DOWNTIME
,
101 .x_checkpoint_delay
= DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
,
106 current_migration
.parameters
.tls_creds
= g_strdup("");
107 current_migration
.parameters
.tls_hostname
= g_strdup("");
110 return ¤t_migration
;
113 MigrationIncomingState
*migration_incoming_get_current(void)
116 static MigrationIncomingState mis_current
;
119 mis_current
.state
= MIGRATION_STATUS_NONE
;
120 memset(&mis_current
, 0, sizeof(MigrationIncomingState
));
121 QLIST_INIT(&mis_current
.loadvm_handlers
);
122 qemu_mutex_init(&mis_current
.rp_mutex
);
123 qemu_event_init(&mis_current
.main_thread_load_event
, false);
129 void migration_incoming_state_destroy(void)
131 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
133 qemu_event_destroy(&mis
->main_thread_load_event
);
134 loadvm_free_handlers(mis
);
141 uint8_t runstate
[100];
146 static GlobalState global_state
;
148 int global_state_store(void)
150 if (!runstate_store((char *)global_state
.runstate
,
151 sizeof(global_state
.runstate
))) {
152 error_report("runstate name too big: %s", global_state
.runstate
);
153 trace_migrate_state_too_big();
159 void global_state_store_running(void)
161 const char *state
= RunState_lookup
[RUN_STATE_RUNNING
];
162 strncpy((char *)global_state
.runstate
,
163 state
, sizeof(global_state
.runstate
));
166 static bool global_state_received(void)
168 return global_state
.received
;
171 static RunState
global_state_get_runstate(void)
173 return global_state
.state
;
176 void global_state_set_optional(void)
178 global_state
.optional
= true;
181 static bool global_state_needed(void *opaque
)
183 GlobalState
*s
= opaque
;
184 char *runstate
= (char *)s
->runstate
;
186 /* If it is not optional, it is mandatory */
188 if (s
->optional
== false) {
192 /* If state is running or paused, it is not needed */
194 if (strcmp(runstate
, "running") == 0 ||
195 strcmp(runstate
, "paused") == 0) {
199 /* for any other state it is needed */
203 static int global_state_post_load(void *opaque
, int version_id
)
205 GlobalState
*s
= opaque
;
206 Error
*local_err
= NULL
;
208 char *runstate
= (char *)s
->runstate
;
211 trace_migrate_global_state_post_load(runstate
);
213 r
= qapi_enum_parse(RunState_lookup
, runstate
, RUN_STATE__MAX
,
218 error_report_err(local_err
);
227 static void global_state_pre_save(void *opaque
)
229 GlobalState
*s
= opaque
;
231 trace_migrate_global_state_pre_save((char *)s
->runstate
);
232 s
->size
= strlen((char *)s
->runstate
) + 1;
235 static const VMStateDescription vmstate_globalstate
= {
236 .name
= "globalstate",
238 .minimum_version_id
= 1,
239 .post_load
= global_state_post_load
,
240 .pre_save
= global_state_pre_save
,
241 .needed
= global_state_needed
,
242 .fields
= (VMStateField
[]) {
243 VMSTATE_UINT32(size
, GlobalState
),
244 VMSTATE_BUFFER(runstate
, GlobalState
),
245 VMSTATE_END_OF_LIST()
249 void register_global_state(void)
251 /* We would use it independently that we receive it */
252 strcpy((char *)&global_state
.runstate
, "");
253 global_state
.received
= false;
254 vmstate_register(NULL
, 0, &vmstate_globalstate
, &global_state
);
257 static void migrate_generate_event(int new_state
)
259 if (migrate_use_events()) {
260 qapi_event_send_migration(new_state
, &error_abort
);
265 * Called on -incoming with a defer: uri.
266 * The migration can be started later after any parameters have been
269 static void deferred_incoming_migration(Error
**errp
)
271 if (deferred_incoming
) {
272 error_setg(errp
, "Incoming migration already deferred");
274 deferred_incoming
= true;
277 /* Request a range of pages from the source VM at the given
279 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
280 * as the last request (a name must have been given previously)
281 * Start: Address offset within the RB
282 * Len: Length in bytes required - must be a multiple of pagesize
284 void migrate_send_rp_req_pages(MigrationIncomingState
*mis
, const char *rbname
,
285 ram_addr_t start
, size_t len
)
287 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
288 size_t msglen
= 12; /* start + len */
290 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
291 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
294 int rbname_len
= strlen(rbname
);
295 assert(rbname_len
< 256);
297 bufc
[msglen
++] = rbname_len
;
298 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
299 msglen
+= rbname_len
;
300 migrate_send_rp_message(mis
, MIG_RP_MSG_REQ_PAGES_ID
, msglen
, bufc
);
302 migrate_send_rp_message(mis
, MIG_RP_MSG_REQ_PAGES
, msglen
, bufc
);
306 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
310 qapi_event_send_migration(MIGRATION_STATUS_SETUP
, &error_abort
);
311 if (!strcmp(uri
, "defer")) {
312 deferred_incoming_migration(errp
);
313 } else if (strstart(uri
, "tcp:", &p
)) {
314 tcp_start_incoming_migration(p
, errp
);
316 } else if (strstart(uri
, "rdma:", &p
)) {
317 rdma_start_incoming_migration(p
, errp
);
319 } else if (strstart(uri
, "exec:", &p
)) {
320 exec_start_incoming_migration(p
, errp
);
321 } else if (strstart(uri
, "unix:", &p
)) {
322 unix_start_incoming_migration(p
, errp
);
323 } else if (strstart(uri
, "fd:", &p
)) {
324 fd_start_incoming_migration(p
, errp
);
326 error_setg(errp
, "unknown migration protocol: %s", uri
);
330 static void process_incoming_migration_bh(void *opaque
)
332 Error
*local_err
= NULL
;
333 MigrationIncomingState
*mis
= opaque
;
335 /* Make sure all file formats flush their mutable metadata.
336 * If we get an error here, just don't restart the VM yet. */
337 bdrv_invalidate_cache_all(&local_err
);
339 error_report_err(local_err
);
345 * This must happen after all error conditions are dealt with and
346 * we're sure the VM is going to be running on this host.
348 qemu_announce_self();
350 /* If global state section was not received or we are in running
351 state, we need to obey autostart. Any other state is set with
354 if (!global_state_received() ||
355 global_state_get_runstate() == RUN_STATE_RUNNING
) {
359 runstate_set(RUN_STATE_PAUSED
);
362 runstate_set(global_state_get_runstate());
364 migrate_decompress_threads_join();
366 * This must happen after any state changes since as soon as an external
367 * observer sees this event they might start to prod at the VM assuming
370 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
371 MIGRATION_STATUS_COMPLETED
);
372 qemu_bh_delete(mis
->bh
);
373 migration_incoming_state_destroy();
376 static void process_incoming_migration_co(void *opaque
)
378 QEMUFile
*f
= opaque
;
379 MigrationIncomingState
*mis
= migration_incoming_get_current();
383 mis
->from_src_file
= f
;
384 mis
->largest_page_size
= qemu_ram_pagesize_largest();
385 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
386 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
387 MIGRATION_STATUS_ACTIVE
);
388 ret
= qemu_loadvm_state(f
);
390 ps
= postcopy_state_get();
391 trace_process_incoming_migration_co_end(ret
, ps
);
392 if (ps
!= POSTCOPY_INCOMING_NONE
) {
393 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
395 * Where a migration had postcopy enabled (and thus went to advise)
396 * but managed to complete within the precopy period, we can use
399 postcopy_ram_incoming_cleanup(mis
);
400 } else if (ret
>= 0) {
402 * Postcopy was started, cleanup should happen at the end of the
405 trace_process_incoming_migration_co_postcopy_end_main();
408 /* Else if something went wrong then just fall out of the normal exit */
411 /* we get COLO info, and know if we are in COLO mode */
412 if (!ret
&& migration_incoming_enable_colo()) {
413 mis
->migration_incoming_co
= qemu_coroutine_self();
414 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
415 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
416 mis
->have_colo_incoming_thread
= true;
417 qemu_coroutine_yield();
419 /* Wait checkpoint incoming thread exit before free resource */
420 qemu_thread_join(&mis
->colo_incoming_thread
);
424 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
425 MIGRATION_STATUS_FAILED
);
426 error_report("load of migration failed: %s", strerror(-ret
));
427 migrate_decompress_threads_join();
432 free_xbzrle_decoded_buf();
434 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
435 qemu_bh_schedule(mis
->bh
);
438 void migration_fd_process_incoming(QEMUFile
*f
)
440 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, f
);
442 migrate_decompress_threads_create();
443 qemu_file_set_blocking(f
, false);
444 qemu_coroutine_enter(co
);
448 void migration_channel_process_incoming(MigrationState
*s
,
451 trace_migration_set_incoming_channel(
452 ioc
, object_get_typename(OBJECT(ioc
)));
454 if (s
->parameters
.tls_creds
&&
455 *s
->parameters
.tls_creds
&&
456 !object_dynamic_cast(OBJECT(ioc
),
457 TYPE_QIO_CHANNEL_TLS
)) {
458 Error
*local_err
= NULL
;
459 migration_tls_channel_process_incoming(s
, ioc
, &local_err
);
461 error_report_err(local_err
);
464 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
465 migration_fd_process_incoming(f
);
470 void migration_channel_connect(MigrationState
*s
,
472 const char *hostname
)
474 trace_migration_set_outgoing_channel(
475 ioc
, object_get_typename(OBJECT(ioc
)), hostname
);
477 if (s
->parameters
.tls_creds
&&
478 *s
->parameters
.tls_creds
&&
479 !object_dynamic_cast(OBJECT(ioc
),
480 TYPE_QIO_CHANNEL_TLS
)) {
481 Error
*local_err
= NULL
;
482 migration_tls_channel_connect(s
, ioc
, hostname
, &local_err
);
484 migrate_fd_error(s
, local_err
);
485 error_free(local_err
);
488 QEMUFile
*f
= qemu_fopen_channel_output(ioc
);
492 migrate_fd_connect(s
);
498 * Send a message on the return channel back to the source
501 void migrate_send_rp_message(MigrationIncomingState
*mis
,
502 enum mig_rp_message_type message_type
,
503 uint16_t len
, void *data
)
505 trace_migrate_send_rp_message((int)message_type
, len
);
506 qemu_mutex_lock(&mis
->rp_mutex
);
507 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
508 qemu_put_be16(mis
->to_src_file
, len
);
509 qemu_put_buffer(mis
->to_src_file
, data
, len
);
510 qemu_fflush(mis
->to_src_file
);
511 qemu_mutex_unlock(&mis
->rp_mutex
);
515 * Send a 'SHUT' message on the return channel with the given value
516 * to indicate that we've finished with the RP. Non-0 value indicates
519 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
524 buf
= cpu_to_be32(value
);
525 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
529 * Send a 'PONG' message on the return channel with the given value
530 * (normally in response to a 'PING')
532 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
537 buf
= cpu_to_be32(value
);
538 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
541 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
543 MigrationCapabilityStatusList
*head
= NULL
;
544 MigrationCapabilityStatusList
*caps
;
545 MigrationState
*s
= migrate_get_current();
548 caps
= NULL
; /* silence compiler warning */
549 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
550 if (i
== MIGRATION_CAPABILITY_X_COLO
&& !colo_supported()) {
554 head
= g_malloc0(sizeof(*caps
));
557 caps
->next
= g_malloc0(sizeof(*caps
));
561 g_malloc(sizeof(*caps
->value
));
562 caps
->value
->capability
= i
;
563 caps
->value
->state
= s
->enabled_capabilities
[i
];
569 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
571 MigrationParameters
*params
;
572 MigrationState
*s
= migrate_get_current();
574 params
= g_malloc0(sizeof(*params
));
575 params
->has_compress_level
= true;
576 params
->compress_level
= s
->parameters
.compress_level
;
577 params
->has_compress_threads
= true;
578 params
->compress_threads
= s
->parameters
.compress_threads
;
579 params
->has_decompress_threads
= true;
580 params
->decompress_threads
= s
->parameters
.decompress_threads
;
581 params
->has_cpu_throttle_initial
= true;
582 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
583 params
->has_cpu_throttle_increment
= true;
584 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
585 params
->has_tls_creds
= !!s
->parameters
.tls_creds
;
586 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
587 params
->has_tls_hostname
= !!s
->parameters
.tls_hostname
;
588 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
589 params
->has_max_bandwidth
= true;
590 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
591 params
->has_downtime_limit
= true;
592 params
->downtime_limit
= s
->parameters
.downtime_limit
;
593 params
->has_x_checkpoint_delay
= true;
594 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
600 * Return true if we're already in the middle of a migration
601 * (i.e. any of the active or setup states)
603 static bool migration_is_setup_or_active(int state
)
606 case MIGRATION_STATUS_ACTIVE
:
607 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
608 case MIGRATION_STATUS_SETUP
:
617 static void get_xbzrle_cache_stats(MigrationInfo
*info
)
619 if (migrate_use_xbzrle()) {
620 info
->has_xbzrle_cache
= true;
621 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
622 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
623 info
->xbzrle_cache
->bytes
= xbzrle_mig_bytes_transferred();
624 info
->xbzrle_cache
->pages
= xbzrle_mig_pages_transferred();
625 info
->xbzrle_cache
->cache_miss
= xbzrle_mig_pages_cache_miss();
626 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_mig_cache_miss_rate();
627 info
->xbzrle_cache
->overflow
= xbzrle_mig_pages_overflow();
631 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
633 info
->has_ram
= true;
634 info
->ram
= g_malloc0(sizeof(*info
->ram
));
635 info
->ram
->transferred
= ram_bytes_transferred();
636 info
->ram
->total
= ram_bytes_total();
637 info
->ram
->duplicate
= dup_mig_pages_transferred();
638 /* legacy value. It is not used anymore */
639 info
->ram
->skipped
= 0;
640 info
->ram
->normal
= norm_mig_pages_transferred();
641 info
->ram
->normal_bytes
= norm_mig_pages_transferred() *
642 qemu_target_page_size();
643 info
->ram
->mbps
= s
->mbps
;
644 info
->ram
->dirty_sync_count
= ram_dirty_sync_count();
645 info
->ram
->postcopy_requests
= ram_postcopy_requests();
646 info
->ram
->page_size
= qemu_target_page_size();
648 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
649 info
->ram
->remaining
= ram_bytes_remaining();
650 info
->ram
->dirty_pages_rate
= ram_dirty_pages_rate();
654 MigrationInfo
*qmp_query_migrate(Error
**errp
)
656 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
657 MigrationState
*s
= migrate_get_current();
660 case MIGRATION_STATUS_NONE
:
661 /* no migration has happened ever */
663 case MIGRATION_STATUS_SETUP
:
664 info
->has_status
= true;
665 info
->has_total_time
= false;
667 case MIGRATION_STATUS_ACTIVE
:
668 case MIGRATION_STATUS_CANCELLING
:
669 info
->has_status
= true;
670 info
->has_total_time
= true;
671 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
673 info
->has_expected_downtime
= true;
674 info
->expected_downtime
= s
->expected_downtime
;
675 info
->has_setup_time
= true;
676 info
->setup_time
= s
->setup_time
;
678 populate_ram_info(info
, s
);
680 if (blk_mig_active()) {
681 info
->has_disk
= true;
682 info
->disk
= g_malloc0(sizeof(*info
->disk
));
683 info
->disk
->transferred
= blk_mig_bytes_transferred();
684 info
->disk
->remaining
= blk_mig_bytes_remaining();
685 info
->disk
->total
= blk_mig_bytes_total();
688 if (cpu_throttle_active()) {
689 info
->has_cpu_throttle_percentage
= true;
690 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
693 get_xbzrle_cache_stats(info
);
695 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
696 /* Mostly the same as active; TODO add some postcopy stats */
697 info
->has_status
= true;
698 info
->has_total_time
= true;
699 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
701 info
->has_expected_downtime
= true;
702 info
->expected_downtime
= s
->expected_downtime
;
703 info
->has_setup_time
= true;
704 info
->setup_time
= s
->setup_time
;
706 populate_ram_info(info
, s
);
708 if (blk_mig_active()) {
709 info
->has_disk
= true;
710 info
->disk
= g_malloc0(sizeof(*info
->disk
));
711 info
->disk
->transferred
= blk_mig_bytes_transferred();
712 info
->disk
->remaining
= blk_mig_bytes_remaining();
713 info
->disk
->total
= blk_mig_bytes_total();
716 get_xbzrle_cache_stats(info
);
718 case MIGRATION_STATUS_COLO
:
719 info
->has_status
= true;
720 /* TODO: display COLO specific information (checkpoint info etc.) */
722 case MIGRATION_STATUS_COMPLETED
:
723 get_xbzrle_cache_stats(info
);
725 info
->has_status
= true;
726 info
->has_total_time
= true;
727 info
->total_time
= s
->total_time
;
728 info
->has_downtime
= true;
729 info
->downtime
= s
->downtime
;
730 info
->has_setup_time
= true;
731 info
->setup_time
= s
->setup_time
;
733 populate_ram_info(info
, s
);
735 case MIGRATION_STATUS_FAILED
:
736 info
->has_status
= true;
738 info
->has_error_desc
= true;
739 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
742 case MIGRATION_STATUS_CANCELLED
:
743 info
->has_status
= true;
746 info
->status
= s
->state
;
751 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
754 MigrationState
*s
= migrate_get_current();
755 MigrationCapabilityStatusList
*cap
;
756 bool old_postcopy_cap
= migrate_postcopy_ram();
758 if (migration_is_setup_or_active(s
->state
)) {
759 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
763 for (cap
= params
; cap
; cap
= cap
->next
) {
764 if (cap
->value
->capability
== MIGRATION_CAPABILITY_X_COLO
) {
765 if (!colo_supported()) {
766 error_setg(errp
, "COLO is not currently supported, please"
767 " configure with --enable-colo option in order to"
768 " support COLO feature");
772 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
775 if (migrate_postcopy_ram()) {
776 if (migrate_use_compression()) {
777 /* The decompression threads asynchronously write into RAM
778 * rather than use the atomic copies needed to avoid
779 * userfaulting. It should be possible to fix the decompression
780 * threads for compatibility in future.
782 error_report("Postcopy is not currently compatible with "
784 s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
] =
787 /* This check is reasonably expensive, so only when it's being
788 * set the first time, also it's only the destination that needs
791 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
792 !postcopy_ram_supported_by_host()) {
793 /* postcopy_ram_supported_by_host will have emitted a more
796 error_report("Postcopy is not supported");
797 s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
] =
803 void qmp_migrate_set_parameters(MigrationParameters
*params
, Error
**errp
)
805 MigrationState
*s
= migrate_get_current();
807 if (params
->has_compress_level
&&
808 (params
->compress_level
< 0 || params
->compress_level
> 9)) {
809 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
810 "is invalid, it should be in the range of 0 to 9");
813 if (params
->has_compress_threads
&&
814 (params
->compress_threads
< 1 || params
->compress_threads
> 255)) {
815 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
817 "is invalid, it should be in the range of 1 to 255");
820 if (params
->has_decompress_threads
&&
821 (params
->decompress_threads
< 1 || params
->decompress_threads
> 255)) {
822 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
823 "decompress_threads",
824 "is invalid, it should be in the range of 1 to 255");
827 if (params
->has_cpu_throttle_initial
&&
828 (params
->cpu_throttle_initial
< 1 ||
829 params
->cpu_throttle_initial
> 99)) {
830 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
831 "cpu_throttle_initial",
832 "an integer in the range of 1 to 99");
835 if (params
->has_cpu_throttle_increment
&&
836 (params
->cpu_throttle_increment
< 1 ||
837 params
->cpu_throttle_increment
> 99)) {
838 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
839 "cpu_throttle_increment",
840 "an integer in the range of 1 to 99");
843 if (params
->has_max_bandwidth
&&
844 (params
->max_bandwidth
< 0 || params
->max_bandwidth
> SIZE_MAX
)) {
845 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
846 " range of 0 to %zu bytes/second", SIZE_MAX
);
849 if (params
->has_downtime_limit
&&
850 (params
->downtime_limit
< 0 ||
851 params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
852 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
853 "the range of 0 to %d milliseconds",
854 MAX_MIGRATE_DOWNTIME
);
857 if (params
->has_x_checkpoint_delay
&& (params
->x_checkpoint_delay
< 0)) {
858 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
859 "x_checkpoint_delay",
860 "is invalid, it should be positive");
863 if (params
->has_compress_level
) {
864 s
->parameters
.compress_level
= params
->compress_level
;
866 if (params
->has_compress_threads
) {
867 s
->parameters
.compress_threads
= params
->compress_threads
;
869 if (params
->has_decompress_threads
) {
870 s
->parameters
.decompress_threads
= params
->decompress_threads
;
872 if (params
->has_cpu_throttle_initial
) {
873 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
875 if (params
->has_cpu_throttle_increment
) {
876 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
878 if (params
->has_tls_creds
) {
879 g_free(s
->parameters
.tls_creds
);
880 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
);
882 if (params
->has_tls_hostname
) {
883 g_free(s
->parameters
.tls_hostname
);
884 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
);
886 if (params
->has_max_bandwidth
) {
887 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
888 if (s
->to_dst_file
) {
889 qemu_file_set_rate_limit(s
->to_dst_file
,
890 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
893 if (params
->has_downtime_limit
) {
894 s
->parameters
.downtime_limit
= params
->downtime_limit
;
897 if (params
->has_x_checkpoint_delay
) {
898 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
899 if (migration_in_colo_state()) {
900 colo_checkpoint_notify(s
);
906 void qmp_migrate_start_postcopy(Error
**errp
)
908 MigrationState
*s
= migrate_get_current();
910 if (!migrate_postcopy_ram()) {
911 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
912 " the start of migration");
916 if (s
->state
== MIGRATION_STATUS_NONE
) {
917 error_setg(errp
, "Postcopy must be started after migration has been"
922 * we don't error if migration has finished since that would be racy
923 * with issuing this command.
925 atomic_set(&s
->start_postcopy
, true);
928 /* shared migration helpers */
930 void migrate_set_state(int *state
, int old_state
, int new_state
)
932 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
933 trace_migrate_set_state(new_state
);
934 migrate_generate_event(new_state
);
938 static void migrate_fd_cleanup(void *opaque
)
940 MigrationState
*s
= opaque
;
942 qemu_bh_delete(s
->cleanup_bh
);
943 s
->cleanup_bh
= NULL
;
945 migration_page_queue_free();
947 if (s
->to_dst_file
) {
948 trace_migrate_fd_cleanup();
949 qemu_mutex_unlock_iothread();
950 if (s
->migration_thread_running
) {
951 qemu_thread_join(&s
->thread
);
952 s
->migration_thread_running
= false;
954 qemu_mutex_lock_iothread();
956 migrate_compress_threads_join();
957 qemu_fclose(s
->to_dst_file
);
958 s
->to_dst_file
= NULL
;
961 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
962 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
964 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
965 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
966 MIGRATION_STATUS_CANCELLED
);
969 notifier_list_notify(&migration_state_notifiers
, s
);
972 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
974 trace_migrate_fd_error(error_get_pretty(error
));
975 assert(s
->to_dst_file
== NULL
);
976 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
977 MIGRATION_STATUS_FAILED
);
979 s
->error
= error_copy(error
);
981 notifier_list_notify(&migration_state_notifiers
, s
);
984 static void migrate_fd_cancel(MigrationState
*s
)
987 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
988 trace_migrate_fd_cancel();
990 if (s
->rp_state
.from_dst_file
) {
991 /* shutdown the rp socket, so causing the rp thread to shutdown */
992 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
996 old_state
= s
->state
;
997 if (!migration_is_setup_or_active(old_state
)) {
1000 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1001 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1004 * If we're unlucky the migration code might be stuck somewhere in a
1005 * send/write while the network has failed and is waiting to timeout;
1006 * if we've got shutdown(2) available then we can force it to quit.
1007 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1008 * called in a bh, so there is no race against this cancel.
1010 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1011 qemu_file_shutdown(f
);
1013 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1014 Error
*local_err
= NULL
;
1016 bdrv_invalidate_cache_all(&local_err
);
1018 error_report_err(local_err
);
1020 s
->block_inactive
= false;
1025 void add_migration_state_change_notifier(Notifier
*notify
)
1027 notifier_list_add(&migration_state_notifiers
, notify
);
1030 void remove_migration_state_change_notifier(Notifier
*notify
)
1032 notifier_remove(notify
);
1035 bool migration_in_setup(MigrationState
*s
)
1037 return s
->state
== MIGRATION_STATUS_SETUP
;
1040 bool migration_has_finished(MigrationState
*s
)
1042 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1045 bool migration_has_failed(MigrationState
*s
)
1047 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1048 s
->state
== MIGRATION_STATUS_FAILED
);
1051 bool migration_in_postcopy(void)
1053 MigrationState
*s
= migrate_get_current();
1055 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1058 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1060 return migration_in_postcopy() && s
->postcopy_after_devices
;
1063 bool migration_is_idle(void)
1065 MigrationState
*s
= migrate_get_current();
1068 case MIGRATION_STATUS_NONE
:
1069 case MIGRATION_STATUS_CANCELLED
:
1070 case MIGRATION_STATUS_COMPLETED
:
1071 case MIGRATION_STATUS_FAILED
:
1073 case MIGRATION_STATUS_SETUP
:
1074 case MIGRATION_STATUS_CANCELLING
:
1075 case MIGRATION_STATUS_ACTIVE
:
1076 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1077 case MIGRATION_STATUS_COLO
:
1079 case MIGRATION_STATUS__MAX
:
1080 g_assert_not_reached();
1086 MigrationState
*migrate_init(const MigrationParams
*params
)
1088 MigrationState
*s
= migrate_get_current();
1091 * Reinitialise all migration state, except
1092 * parameters/capabilities that the user set, and
1098 s
->to_dst_file
= NULL
;
1099 s
->state
= MIGRATION_STATUS_NONE
;
1100 s
->params
= *params
;
1101 s
->rp_state
.from_dst_file
= NULL
;
1102 s
->rp_state
.error
= false;
1105 s
->expected_downtime
= 0;
1107 s
->start_postcopy
= false;
1108 s
->postcopy_after_devices
= false;
1109 s
->migration_thread_running
= false;
1110 error_free(s
->error
);
1113 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1115 s
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1119 static GSList
*migration_blockers
;
1121 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1123 if (only_migratable
) {
1124 error_propagate(errp
, error_copy(reason
));
1125 error_prepend(errp
, "disallowing migration blocker "
1126 "(--only_migratable) for: ");
1130 if (migration_is_idle()) {
1131 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1135 error_propagate(errp
, error_copy(reason
));
1136 error_prepend(errp
, "disallowing migration blocker (migration in "
1141 void migrate_del_blocker(Error
*reason
)
1143 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1146 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1148 Error
*local_err
= NULL
;
1149 static bool once
= true;
1151 if (!deferred_incoming
) {
1152 error_setg(errp
, "For use with '-incoming defer'");
1156 error_setg(errp
, "The incoming migration has already been started");
1159 qemu_start_incoming_migration(uri
, &local_err
);
1162 error_propagate(errp
, local_err
);
1169 bool migration_is_blocked(Error
**errp
)
1171 if (qemu_savevm_state_blocked(errp
)) {
1175 if (migration_blockers
) {
1176 *errp
= error_copy(migration_blockers
->data
);
1183 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1184 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1187 Error
*local_err
= NULL
;
1188 MigrationState
*s
= migrate_get_current();
1189 MigrationParams params
;
1192 params
.blk
= has_blk
&& blk
;
1193 params
.shared
= has_inc
&& inc
;
1195 if (migration_is_setup_or_active(s
->state
) ||
1196 s
->state
== MIGRATION_STATUS_CANCELLING
||
1197 s
->state
== MIGRATION_STATUS_COLO
) {
1198 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1201 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1202 error_setg(errp
, "Guest is waiting for an incoming migration");
1206 if (migration_is_blocked(errp
)) {
1210 s
= migrate_init(¶ms
);
1212 if (strstart(uri
, "tcp:", &p
)) {
1213 tcp_start_outgoing_migration(s
, p
, &local_err
);
1215 } else if (strstart(uri
, "rdma:", &p
)) {
1216 rdma_start_outgoing_migration(s
, p
, &local_err
);
1218 } else if (strstart(uri
, "exec:", &p
)) {
1219 exec_start_outgoing_migration(s
, p
, &local_err
);
1220 } else if (strstart(uri
, "unix:", &p
)) {
1221 unix_start_outgoing_migration(s
, p
, &local_err
);
1222 } else if (strstart(uri
, "fd:", &p
)) {
1223 fd_start_outgoing_migration(s
, p
, &local_err
);
1225 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1226 "a valid migration protocol");
1227 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1228 MIGRATION_STATUS_FAILED
);
1233 migrate_fd_error(s
, local_err
);
1234 error_propagate(errp
, local_err
);
1239 void qmp_migrate_cancel(Error
**errp
)
1241 migrate_fd_cancel(migrate_get_current());
1244 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1246 MigrationState
*s
= migrate_get_current();
1249 /* Check for truncation */
1250 if (value
!= (size_t)value
) {
1251 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1252 "exceeding address space");
1256 /* Cache should not be larger than guest ram size */
1257 if (value
> ram_bytes_total()) {
1258 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1259 "exceeds guest ram size ");
1263 new_size
= xbzrle_cache_resize(value
);
1265 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1266 "is smaller than page size");
1270 s
->xbzrle_cache_size
= new_size
;
1273 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1275 return migrate_xbzrle_cache_size();
1278 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1280 MigrationParameters p
= {
1281 .has_max_bandwidth
= true,
1282 .max_bandwidth
= value
,
1285 qmp_migrate_set_parameters(&p
, errp
);
1288 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1290 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
1291 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1292 "the range of 0 to %d seconds",
1293 MAX_MIGRATE_DOWNTIME_SECONDS
);
1297 value
*= 1000; /* Convert to milliseconds */
1298 value
= MAX(0, MIN(INT64_MAX
, value
));
1300 MigrationParameters p
= {
1301 .has_downtime_limit
= true,
1302 .downtime_limit
= value
,
1305 qmp_migrate_set_parameters(&p
, errp
);
1308 bool migrate_release_ram(void)
1312 s
= migrate_get_current();
1314 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
1317 bool migrate_postcopy_ram(void)
1321 s
= migrate_get_current();
1323 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1326 bool migrate_auto_converge(void)
1330 s
= migrate_get_current();
1332 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1335 bool migrate_zero_blocks(void)
1339 s
= migrate_get_current();
1341 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
1344 bool migrate_use_compression(void)
1348 s
= migrate_get_current();
1350 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
1353 int migrate_compress_level(void)
1357 s
= migrate_get_current();
1359 return s
->parameters
.compress_level
;
1362 int migrate_compress_threads(void)
1366 s
= migrate_get_current();
1368 return s
->parameters
.compress_threads
;
1371 int migrate_decompress_threads(void)
1375 s
= migrate_get_current();
1377 return s
->parameters
.decompress_threads
;
1380 bool migrate_use_events(void)
1384 s
= migrate_get_current();
1386 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
1389 int migrate_use_xbzrle(void)
1393 s
= migrate_get_current();
1395 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
1398 int64_t migrate_xbzrle_cache_size(void)
1402 s
= migrate_get_current();
1404 return s
->xbzrle_cache_size
;
1407 /* migration thread support */
1409 * Something bad happened to the RP stream, mark an error
1410 * The caller shall print or trace something to indicate why
1412 static void mark_source_rp_bad(MigrationState
*s
)
1414 s
->rp_state
.error
= true;
1417 static struct rp_cmd_args
{
1418 ssize_t len
; /* -1 = variable */
1421 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1422 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1423 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1424 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1425 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1426 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1430 * Process a request for pages received on the return path,
1431 * We're allowed to send more than requested (e.g. to round to our page size)
1432 * and we don't need to send pages that have already been sent.
1434 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1435 ram_addr_t start
, size_t len
)
1437 long our_host_ps
= getpagesize();
1439 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1442 * Since we currently insist on matching page sizes, just sanity check
1443 * we're being asked for whole host pages.
1445 if (start
& (our_host_ps
-1) ||
1446 (len
& (our_host_ps
-1))) {
1447 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1448 " len: %zd", __func__
, start
, len
);
1449 mark_source_rp_bad(ms
);
1453 if (ram_save_queue_pages(rbname
, start
, len
)) {
1454 mark_source_rp_bad(ms
);
1459 * Handles messages sent on the return path towards the source VM
1462 static void *source_return_path_thread(void *opaque
)
1464 MigrationState
*ms
= opaque
;
1465 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1466 uint16_t header_len
, header_type
;
1468 uint32_t tmp32
, sibling_error
;
1469 ram_addr_t start
= 0; /* =0 to silence warning */
1470 size_t len
= 0, expected_len
;
1473 trace_source_return_path_thread_entry();
1474 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1475 migration_is_setup_or_active(ms
->state
)) {
1476 trace_source_return_path_thread_loop_top();
1477 header_type
= qemu_get_be16(rp
);
1478 header_len
= qemu_get_be16(rp
);
1480 if (header_type
>= MIG_RP_MSG_MAX
||
1481 header_type
== MIG_RP_MSG_INVALID
) {
1482 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1483 header_type
, header_len
);
1484 mark_source_rp_bad(ms
);
1488 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1489 header_len
!= rp_cmd_args
[header_type
].len
) ||
1490 header_len
> sizeof(buf
)) {
1491 error_report("RP: Received '%s' message (0x%04x) with"
1492 "incorrect length %d expecting %zu",
1493 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1494 (size_t)rp_cmd_args
[header_type
].len
);
1495 mark_source_rp_bad(ms
);
1499 /* We know we've got a valid header by this point */
1500 res
= qemu_get_buffer(rp
, buf
, header_len
);
1501 if (res
!= header_len
) {
1502 error_report("RP: Failed reading data for message 0x%04x"
1503 " read %d expected %d",
1504 header_type
, res
, header_len
);
1505 mark_source_rp_bad(ms
);
1509 /* OK, we have the message and the data */
1510 switch (header_type
) {
1511 case MIG_RP_MSG_SHUT
:
1512 sibling_error
= ldl_be_p(buf
);
1513 trace_source_return_path_thread_shut(sibling_error
);
1514 if (sibling_error
) {
1515 error_report("RP: Sibling indicated error %d", sibling_error
);
1516 mark_source_rp_bad(ms
);
1519 * We'll let the main thread deal with closing the RP
1520 * we could do a shutdown(2) on it, but we're the only user
1521 * anyway, so there's nothing gained.
1525 case MIG_RP_MSG_PONG
:
1526 tmp32
= ldl_be_p(buf
);
1527 trace_source_return_path_thread_pong(tmp32
);
1530 case MIG_RP_MSG_REQ_PAGES
:
1531 start
= ldq_be_p(buf
);
1532 len
= ldl_be_p(buf
+ 8);
1533 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1536 case MIG_RP_MSG_REQ_PAGES_ID
:
1537 expected_len
= 12 + 1; /* header + termination */
1539 if (header_len
>= expected_len
) {
1540 start
= ldq_be_p(buf
);
1541 len
= ldl_be_p(buf
+ 8);
1542 /* Now we expect an idstr */
1543 tmp32
= buf
[12]; /* Length of the following idstr */
1544 buf
[13 + tmp32
] = '\0';
1545 expected_len
+= tmp32
;
1547 if (header_len
!= expected_len
) {
1548 error_report("RP: Req_Page_id with length %d expecting %zd",
1549 header_len
, expected_len
);
1550 mark_source_rp_bad(ms
);
1553 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
1560 if (qemu_file_get_error(rp
)) {
1561 trace_source_return_path_thread_bad_end();
1562 mark_source_rp_bad(ms
);
1565 trace_source_return_path_thread_end();
1567 ms
->rp_state
.from_dst_file
= NULL
;
1572 static int open_return_path_on_source(MigrationState
*ms
)
1575 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
1576 if (!ms
->rp_state
.from_dst_file
) {
1580 trace_open_return_path_on_source();
1581 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
1582 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
1584 trace_open_return_path_on_source_continue();
1589 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1590 static int await_return_path_close_on_source(MigrationState
*ms
)
1593 * If this is a normal exit then the destination will send a SHUT and the
1594 * rp_thread will exit, however if there's an error we need to cause
1597 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
1599 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1600 * waiting for the destination.
1602 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
1603 mark_source_rp_bad(ms
);
1605 trace_await_return_path_close_on_source_joining();
1606 qemu_thread_join(&ms
->rp_state
.rp_thread
);
1607 trace_await_return_path_close_on_source_close();
1608 return ms
->rp_state
.error
;
1612 * Switch from normal iteration to postcopy
1613 * Returns non-0 on error
1615 static int postcopy_start(MigrationState
*ms
, bool *old_vm_running
)
1618 QIOChannelBuffer
*bioc
;
1620 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1621 bool restart_block
= false;
1622 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
1623 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1625 trace_postcopy_start();
1626 qemu_mutex_lock_iothread();
1627 trace_postcopy_start_set_run();
1629 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1630 *old_vm_running
= runstate_is_running();
1631 global_state_store();
1632 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1637 ret
= bdrv_inactivate_all();
1641 restart_block
= true;
1644 * Cause any non-postcopiable, but iterative devices to
1645 * send out their final data.
1647 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true);
1650 * in Finish migrate and with the io-lock held everything should
1651 * be quiet, but we've potentially still got dirty pages and we
1652 * need to tell the destination to throw any pages it's already received
1655 if (ram_postcopy_send_discard_bitmap(ms
)) {
1656 error_report("postcopy send discard bitmap failed");
1661 * send rest of state - note things that are doing postcopy
1662 * will notice we're in POSTCOPY_ACTIVE and not actually
1663 * wrap their state up here
1665 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
1666 /* Ping just for debugging, helps line traces up */
1667 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
1670 * While loading the device state we may trigger page transfer
1671 * requests and the fd must be free to process those, and thus
1672 * the destination must read the whole device state off the fd before
1673 * it starts processing it. Unfortunately the ad-hoc migration format
1674 * doesn't allow the destination to know the size to read without fully
1675 * parsing it through each devices load-state code (especially the open
1676 * coded devices that use get/put).
1677 * So we wrap the device state up in a package with a length at the start;
1678 * to do this we use a qemu_buf to hold the whole of the device state.
1680 bioc
= qio_channel_buffer_new(4096);
1681 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
1682 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
1683 object_unref(OBJECT(bioc
));
1686 * Make sure the receiver can get incoming pages before we send the rest
1689 qemu_savevm_send_postcopy_listen(fb
);
1691 qemu_savevm_state_complete_precopy(fb
, false);
1692 qemu_savevm_send_ping(fb
, 3);
1694 qemu_savevm_send_postcopy_run(fb
);
1696 /* <><> end of stuff going into the package */
1698 /* Last point of recovery; as soon as we send the package the destination
1699 * can open devices and potentially start running.
1700 * Lets just check again we've not got any errors.
1702 ret
= qemu_file_get_error(ms
->to_dst_file
);
1704 error_report("postcopy_start: Migration stream errored (pre package)");
1708 restart_block
= false;
1710 /* Now send that blob */
1711 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
1716 /* Send a notify to give a chance for anything that needs to happen
1717 * at the transition to postcopy and after the device state; in particular
1718 * spice needs to trigger a transition now
1720 ms
->postcopy_after_devices
= true;
1721 notifier_list_notify(&migration_state_notifiers
, ms
);
1723 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
1725 qemu_mutex_unlock_iothread();
1728 * Although this ping is just for debug, it could potentially be
1729 * used for getting a better measurement of downtime at the source.
1731 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
1733 if (migrate_release_ram()) {
1734 ram_postcopy_migrated_memory_release(ms
);
1737 ret
= qemu_file_get_error(ms
->to_dst_file
);
1739 error_report("postcopy_start: Migration stream errored");
1740 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1741 MIGRATION_STATUS_FAILED
);
1749 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1750 MIGRATION_STATUS_FAILED
);
1751 if (restart_block
) {
1752 /* A failure happened early enough that we know the destination hasn't
1753 * accessed block devices, so we're safe to recover.
1755 Error
*local_err
= NULL
;
1757 bdrv_invalidate_cache_all(&local_err
);
1759 error_report_err(local_err
);
1762 qemu_mutex_unlock_iothread();
1767 * migration_completion: Used by migration_thread when there's not much left.
1768 * The caller 'breaks' the loop when this returns.
1770 * @s: Current migration state
1771 * @current_active_state: The migration state we expect to be in
1772 * @*old_vm_running: Pointer to old_vm_running flag
1773 * @*start_time: Pointer to time to update
1775 static void migration_completion(MigrationState
*s
, int current_active_state
,
1776 bool *old_vm_running
,
1777 int64_t *start_time
)
1781 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1782 qemu_mutex_lock_iothread();
1783 *start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1784 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1785 *old_vm_running
= runstate_is_running();
1786 ret
= global_state_store();
1789 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1791 * Don't mark the image with BDRV_O_INACTIVE flag if
1792 * we will go into COLO stage later.
1794 if (ret
>= 0 && !migrate_colo_enabled()) {
1795 ret
= bdrv_inactivate_all();
1798 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
1799 qemu_savevm_state_complete_precopy(s
->to_dst_file
, false);
1800 s
->block_inactive
= true;
1803 qemu_mutex_unlock_iothread();
1808 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1809 trace_migration_completion_postcopy_end();
1811 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
1812 trace_migration_completion_postcopy_end_after_complete();
1816 * If rp was opened we must clean up the thread before
1817 * cleaning everything else up (since if there are no failures
1818 * it will wait for the destination to send it's status in
1820 * Postcopy opens rp if enabled (even if it's not avtivated)
1822 if (migrate_postcopy_ram()) {
1824 trace_migration_completion_postcopy_end_before_rp();
1825 rp_error
= await_return_path_close_on_source(s
);
1826 trace_migration_completion_postcopy_end_after_rp(rp_error
);
1828 goto fail_invalidate
;
1832 if (qemu_file_get_error(s
->to_dst_file
)) {
1833 trace_migration_completion_file_err();
1834 goto fail_invalidate
;
1837 if (!migrate_colo_enabled()) {
1838 migrate_set_state(&s
->state
, current_active_state
,
1839 MIGRATION_STATUS_COMPLETED
);
1845 /* If not doing postcopy, vm_start() will be called: let's regain
1846 * control on images.
1848 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1849 Error
*local_err
= NULL
;
1851 qemu_mutex_lock_iothread();
1852 bdrv_invalidate_cache_all(&local_err
);
1854 error_report_err(local_err
);
1856 s
->block_inactive
= false;
1858 qemu_mutex_unlock_iothread();
1862 migrate_set_state(&s
->state
, current_active_state
,
1863 MIGRATION_STATUS_FAILED
);
1866 bool migrate_colo_enabled(void)
1868 MigrationState
*s
= migrate_get_current();
1869 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
1873 * Master migration thread on the source VM.
1874 * It drives the migration and pumps the data down the outgoing channel.
1876 static void *migration_thread(void *opaque
)
1878 MigrationState
*s
= opaque
;
1879 /* Used by the bandwidth calcs, updated later */
1880 int64_t initial_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1881 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
1882 int64_t initial_bytes
= 0;
1884 * The final stage happens when the remaining data is smaller than
1885 * this threshold; it's calculated from the requested downtime and
1886 * measured bandwidth
1888 int64_t threshold_size
= 0;
1889 int64_t start_time
= initial_time
;
1891 bool old_vm_running
= false;
1892 bool entered_postcopy
= false;
1893 /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
1894 enum MigrationStatus current_active_state
= MIGRATION_STATUS_ACTIVE
;
1895 bool enable_colo
= migrate_colo_enabled();
1897 rcu_register_thread();
1899 qemu_savevm_state_header(s
->to_dst_file
);
1901 if (migrate_postcopy_ram()) {
1902 /* Now tell the dest that it should open its end so it can reply */
1903 qemu_savevm_send_open_return_path(s
->to_dst_file
);
1905 /* And do a ping that will make stuff easier to debug */
1906 qemu_savevm_send_ping(s
->to_dst_file
, 1);
1909 * Tell the destination that we *might* want to do postcopy later;
1910 * if the other end can't do postcopy it should fail now, nice and
1913 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
1916 qemu_savevm_state_begin(s
->to_dst_file
, &s
->params
);
1918 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
1919 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1920 MIGRATION_STATUS_ACTIVE
);
1922 trace_migration_thread_setup_complete();
1924 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
1925 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1926 int64_t current_time
;
1927 uint64_t pending_size
;
1929 if (!qemu_file_rate_limit(s
->to_dst_file
)) {
1930 uint64_t pend_post
, pend_nonpost
;
1932 qemu_savevm_state_pending(s
->to_dst_file
, threshold_size
,
1933 &pend_nonpost
, &pend_post
);
1934 pending_size
= pend_nonpost
+ pend_post
;
1935 trace_migrate_pending(pending_size
, threshold_size
,
1936 pend_post
, pend_nonpost
);
1937 if (pending_size
&& pending_size
>= threshold_size
) {
1938 /* Still a significant amount to transfer */
1940 if (migrate_postcopy_ram() &&
1941 s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
&&
1942 pend_nonpost
<= threshold_size
&&
1943 atomic_read(&s
->start_postcopy
)) {
1945 if (!postcopy_start(s
, &old_vm_running
)) {
1946 current_active_state
= MIGRATION_STATUS_POSTCOPY_ACTIVE
;
1947 entered_postcopy
= true;
1952 /* Just another iteration step */
1953 qemu_savevm_state_iterate(s
->to_dst_file
, entered_postcopy
);
1955 trace_migration_thread_low_pending(pending_size
);
1956 migration_completion(s
, current_active_state
,
1957 &old_vm_running
, &start_time
);
1962 if (qemu_file_get_error(s
->to_dst_file
)) {
1963 migrate_set_state(&s
->state
, current_active_state
,
1964 MIGRATION_STATUS_FAILED
);
1965 trace_migration_thread_file_err();
1968 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1969 if (current_time
>= initial_time
+ BUFFER_DELAY
) {
1970 uint64_t transferred_bytes
= qemu_ftell(s
->to_dst_file
) -
1972 uint64_t time_spent
= current_time
- initial_time
;
1973 double bandwidth
= (double)transferred_bytes
/ time_spent
;
1974 threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
1976 s
->mbps
= (((double) transferred_bytes
* 8.0) /
1977 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
1979 trace_migrate_transferred(transferred_bytes
, time_spent
,
1980 bandwidth
, threshold_size
);
1981 /* if we haven't sent anything, we don't want to recalculate
1982 10000 is a small enough number for our purposes */
1983 if (ram_dirty_pages_rate() && transferred_bytes
> 10000) {
1984 s
->expected_downtime
= ram_dirty_pages_rate() *
1985 qemu_target_page_size() / bandwidth
;
1988 qemu_file_reset_rate_limit(s
->to_dst_file
);
1989 initial_time
= current_time
;
1990 initial_bytes
= qemu_ftell(s
->to_dst_file
);
1992 if (qemu_file_rate_limit(s
->to_dst_file
)) {
1993 /* usleep expects microseconds */
1994 g_usleep((initial_time
+ BUFFER_DELAY
- current_time
)*1000);
1998 trace_migration_thread_after_loop();
1999 /* If we enabled cpu throttling for auto-converge, turn it off. */
2000 cpu_throttle_stop();
2001 end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2003 qemu_mutex_lock_iothread();
2005 * The resource has been allocated by migration will be reused in COLO
2006 * process, so don't release them.
2009 qemu_savevm_state_cleanup();
2011 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
2012 uint64_t transferred_bytes
= qemu_ftell(s
->to_dst_file
);
2013 s
->total_time
= end_time
- s
->total_time
;
2014 if (!entered_postcopy
) {
2015 s
->downtime
= end_time
- start_time
;
2017 if (s
->total_time
) {
2018 s
->mbps
= (((double) transferred_bytes
* 8.0) /
2019 ((double) s
->total_time
)) / 1000;
2021 runstate_set(RUN_STATE_POSTMIGRATE
);
2023 if (s
->state
== MIGRATION_STATUS_ACTIVE
&& enable_colo
) {
2024 migrate_start_colo_process(s
);
2025 qemu_savevm_state_cleanup();
2027 * Fixme: we will run VM in COLO no matter its old running state.
2028 * After exited COLO, we will keep running.
2030 old_vm_running
= true;
2032 if (old_vm_running
&& !entered_postcopy
) {
2035 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2036 runstate_set(RUN_STATE_POSTMIGRATE
);
2040 qemu_bh_schedule(s
->cleanup_bh
);
2041 qemu_mutex_unlock_iothread();
2043 rcu_unregister_thread();
2047 void migrate_fd_connect(MigrationState
*s
)
2049 s
->expected_downtime
= s
->parameters
.downtime_limit
;
2050 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
2052 qemu_file_set_blocking(s
->to_dst_file
, true);
2053 qemu_file_set_rate_limit(s
->to_dst_file
,
2054 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
2056 /* Notify before starting migration thread */
2057 notifier_list_notify(&migration_state_notifiers
, s
);
2060 * Open the return path; currently for postcopy but other things might
2063 if (migrate_postcopy_ram()) {
2064 if (open_return_path_on_source(s
)) {
2065 error_report("Unable to open return-path for postcopy");
2066 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2067 MIGRATION_STATUS_FAILED
);
2068 migrate_fd_cleanup(s
);
2073 migrate_compress_threads_create();
2074 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
2075 QEMU_THREAD_JOINABLE
);
2076 s
->migration_thread_running
= true;