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
;
595 params
->has_block_incremental
= true;
596 params
->block_incremental
= s
->parameters
.block_incremental
;
602 * Return true if we're already in the middle of a migration
603 * (i.e. any of the active or setup states)
605 static bool migration_is_setup_or_active(int state
)
608 case MIGRATION_STATUS_ACTIVE
:
609 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
610 case MIGRATION_STATUS_SETUP
:
619 static void get_xbzrle_cache_stats(MigrationInfo
*info
)
621 if (migrate_use_xbzrle()) {
622 info
->has_xbzrle_cache
= true;
623 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
624 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
625 info
->xbzrle_cache
->bytes
= xbzrle_mig_bytes_transferred();
626 info
->xbzrle_cache
->pages
= xbzrle_mig_pages_transferred();
627 info
->xbzrle_cache
->cache_miss
= xbzrle_mig_pages_cache_miss();
628 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_mig_cache_miss_rate();
629 info
->xbzrle_cache
->overflow
= xbzrle_mig_pages_overflow();
633 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
635 info
->has_ram
= true;
636 info
->ram
= g_malloc0(sizeof(*info
->ram
));
637 info
->ram
->transferred
= ram_bytes_transferred();
638 info
->ram
->total
= ram_bytes_total();
639 info
->ram
->duplicate
= dup_mig_pages_transferred();
640 /* legacy value. It is not used anymore */
641 info
->ram
->skipped
= 0;
642 info
->ram
->normal
= norm_mig_pages_transferred();
643 info
->ram
->normal_bytes
= norm_mig_pages_transferred() *
644 qemu_target_page_size();
645 info
->ram
->mbps
= s
->mbps
;
646 info
->ram
->dirty_sync_count
= ram_dirty_sync_count();
647 info
->ram
->postcopy_requests
= ram_postcopy_requests();
648 info
->ram
->page_size
= qemu_target_page_size();
650 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
651 info
->ram
->remaining
= ram_bytes_remaining();
652 info
->ram
->dirty_pages_rate
= ram_dirty_pages_rate();
656 MigrationInfo
*qmp_query_migrate(Error
**errp
)
658 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
659 MigrationState
*s
= migrate_get_current();
662 case MIGRATION_STATUS_NONE
:
663 /* no migration has happened ever */
665 case MIGRATION_STATUS_SETUP
:
666 info
->has_status
= true;
667 info
->has_total_time
= false;
669 case MIGRATION_STATUS_ACTIVE
:
670 case MIGRATION_STATUS_CANCELLING
:
671 info
->has_status
= true;
672 info
->has_total_time
= true;
673 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
675 info
->has_expected_downtime
= true;
676 info
->expected_downtime
= s
->expected_downtime
;
677 info
->has_setup_time
= true;
678 info
->setup_time
= s
->setup_time
;
680 populate_ram_info(info
, s
);
682 if (blk_mig_active()) {
683 info
->has_disk
= true;
684 info
->disk
= g_malloc0(sizeof(*info
->disk
));
685 info
->disk
->transferred
= blk_mig_bytes_transferred();
686 info
->disk
->remaining
= blk_mig_bytes_remaining();
687 info
->disk
->total
= blk_mig_bytes_total();
690 if (cpu_throttle_active()) {
691 info
->has_cpu_throttle_percentage
= true;
692 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
695 get_xbzrle_cache_stats(info
);
697 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
698 /* Mostly the same as active; TODO add some postcopy stats */
699 info
->has_status
= true;
700 info
->has_total_time
= true;
701 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
703 info
->has_expected_downtime
= true;
704 info
->expected_downtime
= s
->expected_downtime
;
705 info
->has_setup_time
= true;
706 info
->setup_time
= s
->setup_time
;
708 populate_ram_info(info
, s
);
710 if (blk_mig_active()) {
711 info
->has_disk
= true;
712 info
->disk
= g_malloc0(sizeof(*info
->disk
));
713 info
->disk
->transferred
= blk_mig_bytes_transferred();
714 info
->disk
->remaining
= blk_mig_bytes_remaining();
715 info
->disk
->total
= blk_mig_bytes_total();
718 get_xbzrle_cache_stats(info
);
720 case MIGRATION_STATUS_COLO
:
721 info
->has_status
= true;
722 /* TODO: display COLO specific information (checkpoint info etc.) */
724 case MIGRATION_STATUS_COMPLETED
:
725 get_xbzrle_cache_stats(info
);
727 info
->has_status
= true;
728 info
->has_total_time
= true;
729 info
->total_time
= s
->total_time
;
730 info
->has_downtime
= true;
731 info
->downtime
= s
->downtime
;
732 info
->has_setup_time
= true;
733 info
->setup_time
= s
->setup_time
;
735 populate_ram_info(info
, s
);
737 case MIGRATION_STATUS_FAILED
:
738 info
->has_status
= true;
740 info
->has_error_desc
= true;
741 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
744 case MIGRATION_STATUS_CANCELLED
:
745 info
->has_status
= true;
748 info
->status
= s
->state
;
753 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
756 MigrationState
*s
= migrate_get_current();
757 MigrationCapabilityStatusList
*cap
;
758 bool old_postcopy_cap
= migrate_postcopy_ram();
760 if (migration_is_setup_or_active(s
->state
)) {
761 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
765 for (cap
= params
; cap
; cap
= cap
->next
) {
766 if (cap
->value
->capability
== MIGRATION_CAPABILITY_X_COLO
) {
767 if (!colo_supported()) {
768 error_setg(errp
, "COLO is not currently supported, please"
769 " configure with --enable-colo option in order to"
770 " support COLO feature");
774 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
777 if (migrate_postcopy_ram()) {
778 if (migrate_use_compression()) {
779 /* The decompression threads asynchronously write into RAM
780 * rather than use the atomic copies needed to avoid
781 * userfaulting. It should be possible to fix the decompression
782 * threads for compatibility in future.
784 error_report("Postcopy is not currently compatible with "
786 s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
] =
789 /* This check is reasonably expensive, so only when it's being
790 * set the first time, also it's only the destination that needs
793 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
794 !postcopy_ram_supported_by_host()) {
795 /* postcopy_ram_supported_by_host will have emitted a more
798 error_report("Postcopy is not supported");
799 s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
] =
805 void qmp_migrate_set_parameters(MigrationParameters
*params
, Error
**errp
)
807 MigrationState
*s
= migrate_get_current();
809 if (params
->has_compress_level
&&
810 (params
->compress_level
< 0 || params
->compress_level
> 9)) {
811 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
812 "is invalid, it should be in the range of 0 to 9");
815 if (params
->has_compress_threads
&&
816 (params
->compress_threads
< 1 || params
->compress_threads
> 255)) {
817 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
819 "is invalid, it should be in the range of 1 to 255");
822 if (params
->has_decompress_threads
&&
823 (params
->decompress_threads
< 1 || params
->decompress_threads
> 255)) {
824 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
825 "decompress_threads",
826 "is invalid, it should be in the range of 1 to 255");
829 if (params
->has_cpu_throttle_initial
&&
830 (params
->cpu_throttle_initial
< 1 ||
831 params
->cpu_throttle_initial
> 99)) {
832 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
833 "cpu_throttle_initial",
834 "an integer in the range of 1 to 99");
837 if (params
->has_cpu_throttle_increment
&&
838 (params
->cpu_throttle_increment
< 1 ||
839 params
->cpu_throttle_increment
> 99)) {
840 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
841 "cpu_throttle_increment",
842 "an integer in the range of 1 to 99");
845 if (params
->has_max_bandwidth
&&
846 (params
->max_bandwidth
< 0 || params
->max_bandwidth
> SIZE_MAX
)) {
847 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
848 " range of 0 to %zu bytes/second", SIZE_MAX
);
851 if (params
->has_downtime_limit
&&
852 (params
->downtime_limit
< 0 ||
853 params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
854 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
855 "the range of 0 to %d milliseconds",
856 MAX_MIGRATE_DOWNTIME
);
859 if (params
->has_x_checkpoint_delay
&& (params
->x_checkpoint_delay
< 0)) {
860 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
861 "x_checkpoint_delay",
862 "is invalid, it should be positive");
865 if (params
->has_compress_level
) {
866 s
->parameters
.compress_level
= params
->compress_level
;
868 if (params
->has_compress_threads
) {
869 s
->parameters
.compress_threads
= params
->compress_threads
;
871 if (params
->has_decompress_threads
) {
872 s
->parameters
.decompress_threads
= params
->decompress_threads
;
874 if (params
->has_cpu_throttle_initial
) {
875 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
877 if (params
->has_cpu_throttle_increment
) {
878 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
880 if (params
->has_tls_creds
) {
881 g_free(s
->parameters
.tls_creds
);
882 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
);
884 if (params
->has_tls_hostname
) {
885 g_free(s
->parameters
.tls_hostname
);
886 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
);
888 if (params
->has_max_bandwidth
) {
889 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
890 if (s
->to_dst_file
) {
891 qemu_file_set_rate_limit(s
->to_dst_file
,
892 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
895 if (params
->has_downtime_limit
) {
896 s
->parameters
.downtime_limit
= params
->downtime_limit
;
899 if (params
->has_x_checkpoint_delay
) {
900 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
901 if (migration_in_colo_state()) {
902 colo_checkpoint_notify(s
);
905 if (params
->has_block_incremental
) {
906 s
->parameters
.block_incremental
= params
->block_incremental
;
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 void migrate_set_block_enabled(bool value
, Error
**errp
)
945 MigrationCapabilityStatusList
*cap
;
947 cap
= g_new0(MigrationCapabilityStatusList
, 1);
948 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
949 cap
->value
->capability
= MIGRATION_CAPABILITY_BLOCK
;
950 cap
->value
->state
= value
;
951 qmp_migrate_set_capabilities(cap
, errp
);
952 qapi_free_MigrationCapabilityStatusList(cap
);
955 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
957 s
->parameters
.block_incremental
= value
;
960 static void block_cleanup_parameters(MigrationState
*s
)
962 if (s
->must_remove_block_options
) {
963 /* setting to false can never fail */
964 migrate_set_block_enabled(false, &error_abort
);
965 migrate_set_block_incremental(s
, false);
966 s
->must_remove_block_options
= false;
970 static void migrate_fd_cleanup(void *opaque
)
972 MigrationState
*s
= opaque
;
974 qemu_bh_delete(s
->cleanup_bh
);
975 s
->cleanup_bh
= NULL
;
977 migration_page_queue_free();
979 if (s
->to_dst_file
) {
980 trace_migrate_fd_cleanup();
981 qemu_mutex_unlock_iothread();
982 if (s
->migration_thread_running
) {
983 qemu_thread_join(&s
->thread
);
984 s
->migration_thread_running
= false;
986 qemu_mutex_lock_iothread();
988 migrate_compress_threads_join();
989 qemu_fclose(s
->to_dst_file
);
990 s
->to_dst_file
= NULL
;
993 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
994 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
996 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
997 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
998 MIGRATION_STATUS_CANCELLED
);
1001 notifier_list_notify(&migration_state_notifiers
, s
);
1002 block_cleanup_parameters(s
);
1005 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1007 trace_migrate_fd_error(error_get_pretty(error
));
1008 assert(s
->to_dst_file
== NULL
);
1009 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1010 MIGRATION_STATUS_FAILED
);
1012 s
->error
= error_copy(error
);
1014 notifier_list_notify(&migration_state_notifiers
, s
);
1015 block_cleanup_parameters(s
);
1018 static void migrate_fd_cancel(MigrationState
*s
)
1021 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1022 trace_migrate_fd_cancel();
1024 if (s
->rp_state
.from_dst_file
) {
1025 /* shutdown the rp socket, so causing the rp thread to shutdown */
1026 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1030 old_state
= s
->state
;
1031 if (!migration_is_setup_or_active(old_state
)) {
1034 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1035 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1038 * If we're unlucky the migration code might be stuck somewhere in a
1039 * send/write while the network has failed and is waiting to timeout;
1040 * if we've got shutdown(2) available then we can force it to quit.
1041 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1042 * called in a bh, so there is no race against this cancel.
1044 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1045 qemu_file_shutdown(f
);
1047 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1048 Error
*local_err
= NULL
;
1050 bdrv_invalidate_cache_all(&local_err
);
1052 error_report_err(local_err
);
1054 s
->block_inactive
= false;
1057 block_cleanup_parameters(s
);
1060 void add_migration_state_change_notifier(Notifier
*notify
)
1062 notifier_list_add(&migration_state_notifiers
, notify
);
1065 void remove_migration_state_change_notifier(Notifier
*notify
)
1067 notifier_remove(notify
);
1070 bool migration_in_setup(MigrationState
*s
)
1072 return s
->state
== MIGRATION_STATUS_SETUP
;
1075 bool migration_has_finished(MigrationState
*s
)
1077 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1080 bool migration_has_failed(MigrationState
*s
)
1082 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1083 s
->state
== MIGRATION_STATUS_FAILED
);
1086 bool migration_in_postcopy(void)
1088 MigrationState
*s
= migrate_get_current();
1090 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1093 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1095 return migration_in_postcopy() && s
->postcopy_after_devices
;
1098 bool migration_is_idle(void)
1100 MigrationState
*s
= migrate_get_current();
1103 case MIGRATION_STATUS_NONE
:
1104 case MIGRATION_STATUS_CANCELLED
:
1105 case MIGRATION_STATUS_COMPLETED
:
1106 case MIGRATION_STATUS_FAILED
:
1108 case MIGRATION_STATUS_SETUP
:
1109 case MIGRATION_STATUS_CANCELLING
:
1110 case MIGRATION_STATUS_ACTIVE
:
1111 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1112 case MIGRATION_STATUS_COLO
:
1114 case MIGRATION_STATUS__MAX
:
1115 g_assert_not_reached();
1121 MigrationState
*migrate_init(const MigrationParams
*params
)
1123 MigrationState
*s
= migrate_get_current();
1126 * Reinitialise all migration state, except
1127 * parameters/capabilities that the user set, and
1133 s
->to_dst_file
= NULL
;
1134 s
->state
= MIGRATION_STATUS_NONE
;
1135 s
->params
= *params
;
1136 s
->rp_state
.from_dst_file
= NULL
;
1137 s
->rp_state
.error
= false;
1140 s
->expected_downtime
= 0;
1142 s
->start_postcopy
= false;
1143 s
->postcopy_after_devices
= false;
1144 s
->migration_thread_running
= false;
1145 error_free(s
->error
);
1148 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1150 s
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1154 static GSList
*migration_blockers
;
1156 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1158 if (only_migratable
) {
1159 error_propagate(errp
, error_copy(reason
));
1160 error_prepend(errp
, "disallowing migration blocker "
1161 "(--only_migratable) for: ");
1165 if (migration_is_idle()) {
1166 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1170 error_propagate(errp
, error_copy(reason
));
1171 error_prepend(errp
, "disallowing migration blocker (migration in "
1176 void migrate_del_blocker(Error
*reason
)
1178 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1181 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1183 Error
*local_err
= NULL
;
1184 static bool once
= true;
1186 if (!deferred_incoming
) {
1187 error_setg(errp
, "For use with '-incoming defer'");
1191 error_setg(errp
, "The incoming migration has already been started");
1194 qemu_start_incoming_migration(uri
, &local_err
);
1197 error_propagate(errp
, local_err
);
1204 bool migration_is_blocked(Error
**errp
)
1206 if (qemu_savevm_state_blocked(errp
)) {
1210 if (migration_blockers
) {
1211 *errp
= error_copy(migration_blockers
->data
);
1218 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1219 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1222 Error
*local_err
= NULL
;
1223 MigrationState
*s
= migrate_get_current();
1224 MigrationParams params
;
1227 params
.blk
= has_blk
&& blk
;
1228 params
.shared
= has_inc
&& inc
;
1230 if (migration_is_setup_or_active(s
->state
) ||
1231 s
->state
== MIGRATION_STATUS_CANCELLING
||
1232 s
->state
== MIGRATION_STATUS_COLO
) {
1233 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1236 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1237 error_setg(errp
, "Guest is waiting for an incoming migration");
1241 if (migration_is_blocked(errp
)) {
1245 if ((has_blk
&& blk
) || (has_inc
&& inc
)) {
1246 if (migrate_use_block() || migrate_use_block_incremental()) {
1247 error_setg(errp
, "Command options are incompatible with "
1248 "current migration capabilities");
1251 migrate_set_block_enabled(true, &local_err
);
1253 error_propagate(errp
, local_err
);
1256 s
->must_remove_block_options
= true;
1259 if (has_inc
&& inc
) {
1260 migrate_set_block_incremental(s
, true);
1263 s
= migrate_init(¶ms
);
1265 if (strstart(uri
, "tcp:", &p
)) {
1266 tcp_start_outgoing_migration(s
, p
, &local_err
);
1268 } else if (strstart(uri
, "rdma:", &p
)) {
1269 rdma_start_outgoing_migration(s
, p
, &local_err
);
1271 } else if (strstart(uri
, "exec:", &p
)) {
1272 exec_start_outgoing_migration(s
, p
, &local_err
);
1273 } else if (strstart(uri
, "unix:", &p
)) {
1274 unix_start_outgoing_migration(s
, p
, &local_err
);
1275 } else if (strstart(uri
, "fd:", &p
)) {
1276 fd_start_outgoing_migration(s
, p
, &local_err
);
1278 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1279 "a valid migration protocol");
1280 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1281 MIGRATION_STATUS_FAILED
);
1286 migrate_fd_error(s
, local_err
);
1287 error_propagate(errp
, local_err
);
1292 void qmp_migrate_cancel(Error
**errp
)
1294 migrate_fd_cancel(migrate_get_current());
1297 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1299 MigrationState
*s
= migrate_get_current();
1302 /* Check for truncation */
1303 if (value
!= (size_t)value
) {
1304 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1305 "exceeding address space");
1309 /* Cache should not be larger than guest ram size */
1310 if (value
> ram_bytes_total()) {
1311 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1312 "exceeds guest ram size ");
1316 new_size
= xbzrle_cache_resize(value
);
1318 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1319 "is smaller than page size");
1323 s
->xbzrle_cache_size
= new_size
;
1326 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1328 return migrate_xbzrle_cache_size();
1331 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1333 MigrationParameters p
= {
1334 .has_max_bandwidth
= true,
1335 .max_bandwidth
= value
,
1338 qmp_migrate_set_parameters(&p
, errp
);
1341 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1343 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
1344 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1345 "the range of 0 to %d seconds",
1346 MAX_MIGRATE_DOWNTIME_SECONDS
);
1350 value
*= 1000; /* Convert to milliseconds */
1351 value
= MAX(0, MIN(INT64_MAX
, value
));
1353 MigrationParameters p
= {
1354 .has_downtime_limit
= true,
1355 .downtime_limit
= value
,
1358 qmp_migrate_set_parameters(&p
, errp
);
1361 bool migrate_release_ram(void)
1365 s
= migrate_get_current();
1367 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
1370 bool migrate_postcopy_ram(void)
1374 s
= migrate_get_current();
1376 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1379 bool migrate_auto_converge(void)
1383 s
= migrate_get_current();
1385 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1388 bool migrate_zero_blocks(void)
1392 s
= migrate_get_current();
1394 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
1397 bool migrate_use_compression(void)
1401 s
= migrate_get_current();
1403 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
1406 int migrate_compress_level(void)
1410 s
= migrate_get_current();
1412 return s
->parameters
.compress_level
;
1415 int migrate_compress_threads(void)
1419 s
= migrate_get_current();
1421 return s
->parameters
.compress_threads
;
1424 int migrate_decompress_threads(void)
1428 s
= migrate_get_current();
1430 return s
->parameters
.decompress_threads
;
1433 bool migrate_use_events(void)
1437 s
= migrate_get_current();
1439 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
1442 int migrate_use_xbzrle(void)
1446 s
= migrate_get_current();
1448 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
1451 int64_t migrate_xbzrle_cache_size(void)
1455 s
= migrate_get_current();
1457 return s
->xbzrle_cache_size
;
1460 bool migrate_use_block(void)
1464 s
= migrate_get_current();
1466 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
1469 bool migrate_use_block_incremental(void)
1473 s
= migrate_get_current();
1475 return s
->parameters
.block_incremental
;
1478 /* migration thread support */
1480 * Something bad happened to the RP stream, mark an error
1481 * The caller shall print or trace something to indicate why
1483 static void mark_source_rp_bad(MigrationState
*s
)
1485 s
->rp_state
.error
= true;
1488 static struct rp_cmd_args
{
1489 ssize_t len
; /* -1 = variable */
1492 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1493 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1494 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1495 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1496 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1497 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1501 * Process a request for pages received on the return path,
1502 * We're allowed to send more than requested (e.g. to round to our page size)
1503 * and we don't need to send pages that have already been sent.
1505 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1506 ram_addr_t start
, size_t len
)
1508 long our_host_ps
= getpagesize();
1510 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1513 * Since we currently insist on matching page sizes, just sanity check
1514 * we're being asked for whole host pages.
1516 if (start
& (our_host_ps
-1) ||
1517 (len
& (our_host_ps
-1))) {
1518 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1519 " len: %zd", __func__
, start
, len
);
1520 mark_source_rp_bad(ms
);
1524 if (ram_save_queue_pages(rbname
, start
, len
)) {
1525 mark_source_rp_bad(ms
);
1530 * Handles messages sent on the return path towards the source VM
1533 static void *source_return_path_thread(void *opaque
)
1535 MigrationState
*ms
= opaque
;
1536 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1537 uint16_t header_len
, header_type
;
1539 uint32_t tmp32
, sibling_error
;
1540 ram_addr_t start
= 0; /* =0 to silence warning */
1541 size_t len
= 0, expected_len
;
1544 trace_source_return_path_thread_entry();
1545 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1546 migration_is_setup_or_active(ms
->state
)) {
1547 trace_source_return_path_thread_loop_top();
1548 header_type
= qemu_get_be16(rp
);
1549 header_len
= qemu_get_be16(rp
);
1551 if (header_type
>= MIG_RP_MSG_MAX
||
1552 header_type
== MIG_RP_MSG_INVALID
) {
1553 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1554 header_type
, header_len
);
1555 mark_source_rp_bad(ms
);
1559 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1560 header_len
!= rp_cmd_args
[header_type
].len
) ||
1561 header_len
> sizeof(buf
)) {
1562 error_report("RP: Received '%s' message (0x%04x) with"
1563 "incorrect length %d expecting %zu",
1564 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1565 (size_t)rp_cmd_args
[header_type
].len
);
1566 mark_source_rp_bad(ms
);
1570 /* We know we've got a valid header by this point */
1571 res
= qemu_get_buffer(rp
, buf
, header_len
);
1572 if (res
!= header_len
) {
1573 error_report("RP: Failed reading data for message 0x%04x"
1574 " read %d expected %d",
1575 header_type
, res
, header_len
);
1576 mark_source_rp_bad(ms
);
1580 /* OK, we have the message and the data */
1581 switch (header_type
) {
1582 case MIG_RP_MSG_SHUT
:
1583 sibling_error
= ldl_be_p(buf
);
1584 trace_source_return_path_thread_shut(sibling_error
);
1585 if (sibling_error
) {
1586 error_report("RP: Sibling indicated error %d", sibling_error
);
1587 mark_source_rp_bad(ms
);
1590 * We'll let the main thread deal with closing the RP
1591 * we could do a shutdown(2) on it, but we're the only user
1592 * anyway, so there's nothing gained.
1596 case MIG_RP_MSG_PONG
:
1597 tmp32
= ldl_be_p(buf
);
1598 trace_source_return_path_thread_pong(tmp32
);
1601 case MIG_RP_MSG_REQ_PAGES
:
1602 start
= ldq_be_p(buf
);
1603 len
= ldl_be_p(buf
+ 8);
1604 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1607 case MIG_RP_MSG_REQ_PAGES_ID
:
1608 expected_len
= 12 + 1; /* header + termination */
1610 if (header_len
>= expected_len
) {
1611 start
= ldq_be_p(buf
);
1612 len
= ldl_be_p(buf
+ 8);
1613 /* Now we expect an idstr */
1614 tmp32
= buf
[12]; /* Length of the following idstr */
1615 buf
[13 + tmp32
] = '\0';
1616 expected_len
+= tmp32
;
1618 if (header_len
!= expected_len
) {
1619 error_report("RP: Req_Page_id with length %d expecting %zd",
1620 header_len
, expected_len
);
1621 mark_source_rp_bad(ms
);
1624 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
1631 if (qemu_file_get_error(rp
)) {
1632 trace_source_return_path_thread_bad_end();
1633 mark_source_rp_bad(ms
);
1636 trace_source_return_path_thread_end();
1638 ms
->rp_state
.from_dst_file
= NULL
;
1643 static int open_return_path_on_source(MigrationState
*ms
)
1646 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
1647 if (!ms
->rp_state
.from_dst_file
) {
1651 trace_open_return_path_on_source();
1652 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
1653 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
1655 trace_open_return_path_on_source_continue();
1660 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1661 static int await_return_path_close_on_source(MigrationState
*ms
)
1664 * If this is a normal exit then the destination will send a SHUT and the
1665 * rp_thread will exit, however if there's an error we need to cause
1668 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
1670 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1671 * waiting for the destination.
1673 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
1674 mark_source_rp_bad(ms
);
1676 trace_await_return_path_close_on_source_joining();
1677 qemu_thread_join(&ms
->rp_state
.rp_thread
);
1678 trace_await_return_path_close_on_source_close();
1679 return ms
->rp_state
.error
;
1683 * Switch from normal iteration to postcopy
1684 * Returns non-0 on error
1686 static int postcopy_start(MigrationState
*ms
, bool *old_vm_running
)
1689 QIOChannelBuffer
*bioc
;
1691 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1692 bool restart_block
= false;
1693 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
1694 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1696 trace_postcopy_start();
1697 qemu_mutex_lock_iothread();
1698 trace_postcopy_start_set_run();
1700 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1701 *old_vm_running
= runstate_is_running();
1702 global_state_store();
1703 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1708 ret
= bdrv_inactivate_all();
1712 restart_block
= true;
1715 * Cause any non-postcopiable, but iterative devices to
1716 * send out their final data.
1718 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true);
1721 * in Finish migrate and with the io-lock held everything should
1722 * be quiet, but we've potentially still got dirty pages and we
1723 * need to tell the destination to throw any pages it's already received
1726 if (ram_postcopy_send_discard_bitmap(ms
)) {
1727 error_report("postcopy send discard bitmap failed");
1732 * send rest of state - note things that are doing postcopy
1733 * will notice we're in POSTCOPY_ACTIVE and not actually
1734 * wrap their state up here
1736 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
1737 /* Ping just for debugging, helps line traces up */
1738 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
1741 * While loading the device state we may trigger page transfer
1742 * requests and the fd must be free to process those, and thus
1743 * the destination must read the whole device state off the fd before
1744 * it starts processing it. Unfortunately the ad-hoc migration format
1745 * doesn't allow the destination to know the size to read without fully
1746 * parsing it through each devices load-state code (especially the open
1747 * coded devices that use get/put).
1748 * So we wrap the device state up in a package with a length at the start;
1749 * to do this we use a qemu_buf to hold the whole of the device state.
1751 bioc
= qio_channel_buffer_new(4096);
1752 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
1753 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
1754 object_unref(OBJECT(bioc
));
1757 * Make sure the receiver can get incoming pages before we send the rest
1760 qemu_savevm_send_postcopy_listen(fb
);
1762 qemu_savevm_state_complete_precopy(fb
, false);
1763 qemu_savevm_send_ping(fb
, 3);
1765 qemu_savevm_send_postcopy_run(fb
);
1767 /* <><> end of stuff going into the package */
1769 /* Last point of recovery; as soon as we send the package the destination
1770 * can open devices and potentially start running.
1771 * Lets just check again we've not got any errors.
1773 ret
= qemu_file_get_error(ms
->to_dst_file
);
1775 error_report("postcopy_start: Migration stream errored (pre package)");
1779 restart_block
= false;
1781 /* Now send that blob */
1782 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
1787 /* Send a notify to give a chance for anything that needs to happen
1788 * at the transition to postcopy and after the device state; in particular
1789 * spice needs to trigger a transition now
1791 ms
->postcopy_after_devices
= true;
1792 notifier_list_notify(&migration_state_notifiers
, ms
);
1794 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
1796 qemu_mutex_unlock_iothread();
1799 * Although this ping is just for debug, it could potentially be
1800 * used for getting a better measurement of downtime at the source.
1802 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
1804 if (migrate_release_ram()) {
1805 ram_postcopy_migrated_memory_release(ms
);
1808 ret
= qemu_file_get_error(ms
->to_dst_file
);
1810 error_report("postcopy_start: Migration stream errored");
1811 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1812 MIGRATION_STATUS_FAILED
);
1820 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1821 MIGRATION_STATUS_FAILED
);
1822 if (restart_block
) {
1823 /* A failure happened early enough that we know the destination hasn't
1824 * accessed block devices, so we're safe to recover.
1826 Error
*local_err
= NULL
;
1828 bdrv_invalidate_cache_all(&local_err
);
1830 error_report_err(local_err
);
1833 qemu_mutex_unlock_iothread();
1838 * migration_completion: Used by migration_thread when there's not much left.
1839 * The caller 'breaks' the loop when this returns.
1841 * @s: Current migration state
1842 * @current_active_state: The migration state we expect to be in
1843 * @*old_vm_running: Pointer to old_vm_running flag
1844 * @*start_time: Pointer to time to update
1846 static void migration_completion(MigrationState
*s
, int current_active_state
,
1847 bool *old_vm_running
,
1848 int64_t *start_time
)
1852 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1853 qemu_mutex_lock_iothread();
1854 *start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1855 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1856 *old_vm_running
= runstate_is_running();
1857 ret
= global_state_store();
1860 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1862 * Don't mark the image with BDRV_O_INACTIVE flag if
1863 * we will go into COLO stage later.
1865 if (ret
>= 0 && !migrate_colo_enabled()) {
1866 ret
= bdrv_inactivate_all();
1869 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
1870 qemu_savevm_state_complete_precopy(s
->to_dst_file
, false);
1871 s
->block_inactive
= true;
1874 qemu_mutex_unlock_iothread();
1879 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1880 trace_migration_completion_postcopy_end();
1882 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
1883 trace_migration_completion_postcopy_end_after_complete();
1887 * If rp was opened we must clean up the thread before
1888 * cleaning everything else up (since if there are no failures
1889 * it will wait for the destination to send it's status in
1891 * Postcopy opens rp if enabled (even if it's not avtivated)
1893 if (migrate_postcopy_ram()) {
1895 trace_migration_completion_postcopy_end_before_rp();
1896 rp_error
= await_return_path_close_on_source(s
);
1897 trace_migration_completion_postcopy_end_after_rp(rp_error
);
1899 goto fail_invalidate
;
1903 if (qemu_file_get_error(s
->to_dst_file
)) {
1904 trace_migration_completion_file_err();
1905 goto fail_invalidate
;
1908 if (!migrate_colo_enabled()) {
1909 migrate_set_state(&s
->state
, current_active_state
,
1910 MIGRATION_STATUS_COMPLETED
);
1916 /* If not doing postcopy, vm_start() will be called: let's regain
1917 * control on images.
1919 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1920 Error
*local_err
= NULL
;
1922 qemu_mutex_lock_iothread();
1923 bdrv_invalidate_cache_all(&local_err
);
1925 error_report_err(local_err
);
1927 s
->block_inactive
= false;
1929 qemu_mutex_unlock_iothread();
1933 migrate_set_state(&s
->state
, current_active_state
,
1934 MIGRATION_STATUS_FAILED
);
1937 bool migrate_colo_enabled(void)
1939 MigrationState
*s
= migrate_get_current();
1940 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
1944 * Master migration thread on the source VM.
1945 * It drives the migration and pumps the data down the outgoing channel.
1947 static void *migration_thread(void *opaque
)
1949 MigrationState
*s
= opaque
;
1950 /* Used by the bandwidth calcs, updated later */
1951 int64_t initial_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1952 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
1953 int64_t initial_bytes
= 0;
1955 * The final stage happens when the remaining data is smaller than
1956 * this threshold; it's calculated from the requested downtime and
1957 * measured bandwidth
1959 int64_t threshold_size
= 0;
1960 int64_t start_time
= initial_time
;
1962 bool old_vm_running
= false;
1963 bool entered_postcopy
= false;
1964 /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
1965 enum MigrationStatus current_active_state
= MIGRATION_STATUS_ACTIVE
;
1966 bool enable_colo
= migrate_colo_enabled();
1968 rcu_register_thread();
1970 qemu_savevm_state_header(s
->to_dst_file
);
1972 if (migrate_postcopy_ram()) {
1973 /* Now tell the dest that it should open its end so it can reply */
1974 qemu_savevm_send_open_return_path(s
->to_dst_file
);
1976 /* And do a ping that will make stuff easier to debug */
1977 qemu_savevm_send_ping(s
->to_dst_file
, 1);
1980 * Tell the destination that we *might* want to do postcopy later;
1981 * if the other end can't do postcopy it should fail now, nice and
1984 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
1987 qemu_savevm_state_begin(s
->to_dst_file
, &s
->params
);
1989 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
1990 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1991 MIGRATION_STATUS_ACTIVE
);
1993 trace_migration_thread_setup_complete();
1995 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
1996 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1997 int64_t current_time
;
1998 uint64_t pending_size
;
2000 if (!qemu_file_rate_limit(s
->to_dst_file
)) {
2001 uint64_t pend_post
, pend_nonpost
;
2003 qemu_savevm_state_pending(s
->to_dst_file
, threshold_size
,
2004 &pend_nonpost
, &pend_post
);
2005 pending_size
= pend_nonpost
+ pend_post
;
2006 trace_migrate_pending(pending_size
, threshold_size
,
2007 pend_post
, pend_nonpost
);
2008 if (pending_size
&& pending_size
>= threshold_size
) {
2009 /* Still a significant amount to transfer */
2011 if (migrate_postcopy_ram() &&
2012 s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
&&
2013 pend_nonpost
<= threshold_size
&&
2014 atomic_read(&s
->start_postcopy
)) {
2016 if (!postcopy_start(s
, &old_vm_running
)) {
2017 current_active_state
= MIGRATION_STATUS_POSTCOPY_ACTIVE
;
2018 entered_postcopy
= true;
2023 /* Just another iteration step */
2024 qemu_savevm_state_iterate(s
->to_dst_file
, entered_postcopy
);
2026 trace_migration_thread_low_pending(pending_size
);
2027 migration_completion(s
, current_active_state
,
2028 &old_vm_running
, &start_time
);
2033 if (qemu_file_get_error(s
->to_dst_file
)) {
2034 migrate_set_state(&s
->state
, current_active_state
,
2035 MIGRATION_STATUS_FAILED
);
2036 trace_migration_thread_file_err();
2039 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2040 if (current_time
>= initial_time
+ BUFFER_DELAY
) {
2041 uint64_t transferred_bytes
= qemu_ftell(s
->to_dst_file
) -
2043 uint64_t time_spent
= current_time
- initial_time
;
2044 double bandwidth
= (double)transferred_bytes
/ time_spent
;
2045 threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
2047 s
->mbps
= (((double) transferred_bytes
* 8.0) /
2048 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2050 trace_migrate_transferred(transferred_bytes
, time_spent
,
2051 bandwidth
, threshold_size
);
2052 /* if we haven't sent anything, we don't want to recalculate
2053 10000 is a small enough number for our purposes */
2054 if (ram_dirty_pages_rate() && transferred_bytes
> 10000) {
2055 s
->expected_downtime
= ram_dirty_pages_rate() *
2056 qemu_target_page_size() / bandwidth
;
2059 qemu_file_reset_rate_limit(s
->to_dst_file
);
2060 initial_time
= current_time
;
2061 initial_bytes
= qemu_ftell(s
->to_dst_file
);
2063 if (qemu_file_rate_limit(s
->to_dst_file
)) {
2064 /* usleep expects microseconds */
2065 g_usleep((initial_time
+ BUFFER_DELAY
- current_time
)*1000);
2069 trace_migration_thread_after_loop();
2070 /* If we enabled cpu throttling for auto-converge, turn it off. */
2071 cpu_throttle_stop();
2072 end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2074 qemu_mutex_lock_iothread();
2076 * The resource has been allocated by migration will be reused in COLO
2077 * process, so don't release them.
2080 qemu_savevm_state_cleanup();
2082 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
2083 uint64_t transferred_bytes
= qemu_ftell(s
->to_dst_file
);
2084 s
->total_time
= end_time
- s
->total_time
;
2085 if (!entered_postcopy
) {
2086 s
->downtime
= end_time
- start_time
;
2088 if (s
->total_time
) {
2089 s
->mbps
= (((double) transferred_bytes
* 8.0) /
2090 ((double) s
->total_time
)) / 1000;
2092 runstate_set(RUN_STATE_POSTMIGRATE
);
2094 if (s
->state
== MIGRATION_STATUS_ACTIVE
&& enable_colo
) {
2095 migrate_start_colo_process(s
);
2096 qemu_savevm_state_cleanup();
2098 * Fixme: we will run VM in COLO no matter its old running state.
2099 * After exited COLO, we will keep running.
2101 old_vm_running
= true;
2103 if (old_vm_running
&& !entered_postcopy
) {
2106 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2107 runstate_set(RUN_STATE_POSTMIGRATE
);
2111 qemu_bh_schedule(s
->cleanup_bh
);
2112 qemu_mutex_unlock_iothread();
2114 rcu_unregister_thread();
2118 void migrate_fd_connect(MigrationState
*s
)
2120 s
->expected_downtime
= s
->parameters
.downtime_limit
;
2121 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
2123 qemu_file_set_blocking(s
->to_dst_file
, true);
2124 qemu_file_set_rate_limit(s
->to_dst_file
,
2125 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
2127 /* Notify before starting migration thread */
2128 notifier_list_notify(&migration_state_notifiers
, s
);
2131 * Open the return path; currently for postcopy but other things might
2134 if (migrate_postcopy_ram()) {
2135 if (open_return_path_on_source(s
)) {
2136 error_report("Unable to open return-path for postcopy");
2137 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2138 MIGRATION_STATUS_FAILED
);
2139 migrate_fd_cleanup(s
);
2144 migrate_compress_threads_create();
2145 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
2146 QEMU_THREAD_JOINABLE
);
2147 s
->migration_thread_running
= true;