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"
26 #include "migration/migration.h"
28 #include "qemu-file-channel.h"
29 #include "qemu-file.h"
30 #include "migration/vmstate.h"
31 #include "sysemu/sysemu.h"
32 #include "block/block.h"
33 #include "qapi/qmp/qerror.h"
34 #include "qapi/util.h"
35 #include "qemu/sockets.h"
38 #include "postcopy-ram.h"
39 #include "qemu/thread.h"
40 #include "qmp-commands.h"
42 #include "qapi-event.h"
44 #include "exec/memory.h"
45 #include "exec/address-spaces.h"
46 #include "exec/target_page.h"
47 #include "io/channel-buffer.h"
48 #include "migration/colo.h"
50 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
52 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
54 #define BUFFER_DELAY 100
55 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
57 /* Time in milliseconds we are allowed to stop the source,
58 * for sending the last part */
59 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
61 /* Maximum migrate downtime set to 2000 seconds */
62 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
63 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
65 /* Default compression thread count */
66 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
67 /* Default decompression thread count, usually decompression is at
68 * least 4 times as fast as compression.*/
69 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
70 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
71 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
72 /* Define default autoconverge cpu throttle migration parameters */
73 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
74 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
76 /* Migration XBZRLE default cache size */
77 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
79 /* The delay time (in ms) between two COLO checkpoints
80 * Note: Please change this default value to 10000 when we support hybrid mode.
82 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
84 static NotifierList migration_state_notifiers
=
85 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
87 static bool deferred_incoming
;
89 /* When we add fault tolerance, we could have several
90 migrations at once. For now we don't need to add
91 dynamic creation of migration */
94 MigrationState
*migrate_get_current(void)
97 static MigrationState current_migration
= {
98 .state
= MIGRATION_STATUS_NONE
,
99 .xbzrle_cache_size
= DEFAULT_MIGRATE_CACHE_SIZE
,
102 .compress_level
= DEFAULT_MIGRATE_COMPRESS_LEVEL
,
103 .compress_threads
= DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
,
104 .decompress_threads
= DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
,
105 .cpu_throttle_initial
= DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
,
106 .cpu_throttle_increment
= DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
,
107 .max_bandwidth
= MAX_THROTTLE
,
108 .downtime_limit
= DEFAULT_MIGRATE_SET_DOWNTIME
,
109 .x_checkpoint_delay
= DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
,
114 current_migration
.parameters
.tls_creds
= g_strdup("");
115 current_migration
.parameters
.tls_hostname
= g_strdup("");
118 return ¤t_migration
;
121 MigrationIncomingState
*migration_incoming_get_current(void)
124 static MigrationIncomingState mis_current
;
127 mis_current
.state
= MIGRATION_STATUS_NONE
;
128 memset(&mis_current
, 0, sizeof(MigrationIncomingState
));
129 qemu_mutex_init(&mis_current
.rp_mutex
);
130 qemu_event_init(&mis_current
.main_thread_load_event
, false);
136 void migration_incoming_state_destroy(void)
138 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
140 if (mis
->to_src_file
) {
141 /* Tell source that we are done */
142 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
143 qemu_fclose(mis
->to_src_file
);
144 mis
->to_src_file
= NULL
;
147 if (mis
->from_src_file
) {
148 qemu_fclose(mis
->from_src_file
);
149 mis
->from_src_file
= NULL
;
152 qemu_event_destroy(&mis
->main_thread_load_event
);
159 uint8_t runstate
[100];
164 static GlobalState global_state
;
166 int global_state_store(void)
168 if (!runstate_store((char *)global_state
.runstate
,
169 sizeof(global_state
.runstate
))) {
170 error_report("runstate name too big: %s", global_state
.runstate
);
171 trace_migrate_state_too_big();
177 void global_state_store_running(void)
179 const char *state
= RunState_lookup
[RUN_STATE_RUNNING
];
180 strncpy((char *)global_state
.runstate
,
181 state
, sizeof(global_state
.runstate
));
184 static bool global_state_received(void)
186 return global_state
.received
;
189 static RunState
global_state_get_runstate(void)
191 return global_state
.state
;
194 void global_state_set_optional(void)
196 global_state
.optional
= true;
199 static bool global_state_needed(void *opaque
)
201 GlobalState
*s
= opaque
;
202 char *runstate
= (char *)s
->runstate
;
204 /* If it is not optional, it is mandatory */
206 if (s
->optional
== false) {
210 /* If state is running or paused, it is not needed */
212 if (strcmp(runstate
, "running") == 0 ||
213 strcmp(runstate
, "paused") == 0) {
217 /* for any other state it is needed */
221 static int global_state_post_load(void *opaque
, int version_id
)
223 GlobalState
*s
= opaque
;
224 Error
*local_err
= NULL
;
226 char *runstate
= (char *)s
->runstate
;
229 trace_migrate_global_state_post_load(runstate
);
231 r
= qapi_enum_parse(RunState_lookup
, runstate
, RUN_STATE__MAX
,
236 error_report_err(local_err
);
245 static void global_state_pre_save(void *opaque
)
247 GlobalState
*s
= opaque
;
249 trace_migrate_global_state_pre_save((char *)s
->runstate
);
250 s
->size
= strlen((char *)s
->runstate
) + 1;
253 static const VMStateDescription vmstate_globalstate
= {
254 .name
= "globalstate",
256 .minimum_version_id
= 1,
257 .post_load
= global_state_post_load
,
258 .pre_save
= global_state_pre_save
,
259 .needed
= global_state_needed
,
260 .fields
= (VMStateField
[]) {
261 VMSTATE_UINT32(size
, GlobalState
),
262 VMSTATE_BUFFER(runstate
, GlobalState
),
263 VMSTATE_END_OF_LIST()
267 void register_global_state(void)
269 /* We would use it independently that we receive it */
270 strcpy((char *)&global_state
.runstate
, "");
271 global_state
.received
= false;
272 vmstate_register(NULL
, 0, &vmstate_globalstate
, &global_state
);
275 static void migrate_generate_event(int new_state
)
277 if (migrate_use_events()) {
278 qapi_event_send_migration(new_state
, &error_abort
);
283 * Called on -incoming with a defer: uri.
284 * The migration can be started later after any parameters have been
287 static void deferred_incoming_migration(Error
**errp
)
289 if (deferred_incoming
) {
290 error_setg(errp
, "Incoming migration already deferred");
292 deferred_incoming
= true;
295 /* Request a range of pages from the source VM at the given
297 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
298 * as the last request (a name must have been given previously)
299 * Start: Address offset within the RB
300 * Len: Length in bytes required - must be a multiple of pagesize
302 void migrate_send_rp_req_pages(MigrationIncomingState
*mis
, const char *rbname
,
303 ram_addr_t start
, size_t len
)
305 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
306 size_t msglen
= 12; /* start + len */
308 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
309 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
312 int rbname_len
= strlen(rbname
);
313 assert(rbname_len
< 256);
315 bufc
[msglen
++] = rbname_len
;
316 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
317 msglen
+= rbname_len
;
318 migrate_send_rp_message(mis
, MIG_RP_MSG_REQ_PAGES_ID
, msglen
, bufc
);
320 migrate_send_rp_message(mis
, MIG_RP_MSG_REQ_PAGES
, msglen
, bufc
);
324 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
328 qapi_event_send_migration(MIGRATION_STATUS_SETUP
, &error_abort
);
329 if (!strcmp(uri
, "defer")) {
330 deferred_incoming_migration(errp
);
331 } else if (strstart(uri
, "tcp:", &p
)) {
332 tcp_start_incoming_migration(p
, errp
);
334 } else if (strstart(uri
, "rdma:", &p
)) {
335 rdma_start_incoming_migration(p
, errp
);
337 } else if (strstart(uri
, "exec:", &p
)) {
338 exec_start_incoming_migration(p
, errp
);
339 } else if (strstart(uri
, "unix:", &p
)) {
340 unix_start_incoming_migration(p
, errp
);
341 } else if (strstart(uri
, "fd:", &p
)) {
342 fd_start_incoming_migration(p
, errp
);
344 error_setg(errp
, "unknown migration protocol: %s", uri
);
348 static void process_incoming_migration_bh(void *opaque
)
350 Error
*local_err
= NULL
;
351 MigrationIncomingState
*mis
= opaque
;
353 /* Make sure all file formats flush their mutable metadata.
354 * If we get an error here, just don't restart the VM yet. */
355 bdrv_invalidate_cache_all(&local_err
);
357 error_report_err(local_err
);
363 * This must happen after all error conditions are dealt with and
364 * we're sure the VM is going to be running on this host.
366 qemu_announce_self();
368 /* If global state section was not received or we are in running
369 state, we need to obey autostart. Any other state is set with
372 if (!global_state_received() ||
373 global_state_get_runstate() == RUN_STATE_RUNNING
) {
377 runstate_set(RUN_STATE_PAUSED
);
380 runstate_set(global_state_get_runstate());
382 migrate_decompress_threads_join();
384 * This must happen after any state changes since as soon as an external
385 * observer sees this event they might start to prod at the VM assuming
388 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
389 MIGRATION_STATUS_COMPLETED
);
390 qemu_bh_delete(mis
->bh
);
391 migration_incoming_state_destroy();
394 static void process_incoming_migration_co(void *opaque
)
396 QEMUFile
*f
= opaque
;
397 MigrationIncomingState
*mis
= migration_incoming_get_current();
401 mis
->from_src_file
= f
;
402 mis
->largest_page_size
= qemu_ram_pagesize_largest();
403 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
404 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
405 MIGRATION_STATUS_ACTIVE
);
406 ret
= qemu_loadvm_state(f
);
408 ps
= postcopy_state_get();
409 trace_process_incoming_migration_co_end(ret
, ps
);
410 if (ps
!= POSTCOPY_INCOMING_NONE
) {
411 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
413 * Where a migration had postcopy enabled (and thus went to advise)
414 * but managed to complete within the precopy period, we can use
417 postcopy_ram_incoming_cleanup(mis
);
418 } else if (ret
>= 0) {
420 * Postcopy was started, cleanup should happen at the end of the
423 trace_process_incoming_migration_co_postcopy_end_main();
426 /* Else if something went wrong then just fall out of the normal exit */
429 /* we get COLO info, and know if we are in COLO mode */
430 if (!ret
&& migration_incoming_enable_colo()) {
431 mis
->migration_incoming_co
= qemu_coroutine_self();
432 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
433 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
434 mis
->have_colo_incoming_thread
= true;
435 qemu_coroutine_yield();
437 /* Wait checkpoint incoming thread exit before free resource */
438 qemu_thread_join(&mis
->colo_incoming_thread
);
442 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
443 MIGRATION_STATUS_FAILED
);
444 error_report("load of migration failed: %s", strerror(-ret
));
445 migrate_decompress_threads_join();
449 free_xbzrle_decoded_buf();
451 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
452 qemu_bh_schedule(mis
->bh
);
455 void migration_fd_process_incoming(QEMUFile
*f
)
457 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, f
);
459 migrate_decompress_threads_create();
460 qemu_file_set_blocking(f
, false);
461 qemu_coroutine_enter(co
);
465 * Send a message on the return channel back to the source
468 void migrate_send_rp_message(MigrationIncomingState
*mis
,
469 enum mig_rp_message_type message_type
,
470 uint16_t len
, void *data
)
472 trace_migrate_send_rp_message((int)message_type
, len
);
473 qemu_mutex_lock(&mis
->rp_mutex
);
474 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
475 qemu_put_be16(mis
->to_src_file
, len
);
476 qemu_put_buffer(mis
->to_src_file
, data
, len
);
477 qemu_fflush(mis
->to_src_file
);
478 qemu_mutex_unlock(&mis
->rp_mutex
);
482 * Send a 'SHUT' message on the return channel with the given value
483 * to indicate that we've finished with the RP. Non-0 value indicates
486 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
491 buf
= cpu_to_be32(value
);
492 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
496 * Send a 'PONG' message on the return channel with the given value
497 * (normally in response to a 'PING')
499 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
504 buf
= cpu_to_be32(value
);
505 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
508 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
510 MigrationCapabilityStatusList
*head
= NULL
;
511 MigrationCapabilityStatusList
*caps
;
512 MigrationState
*s
= migrate_get_current();
515 caps
= NULL
; /* silence compiler warning */
516 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
517 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
518 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
522 if (i
== MIGRATION_CAPABILITY_X_COLO
&& !colo_supported()) {
526 head
= g_malloc0(sizeof(*caps
));
529 caps
->next
= g_malloc0(sizeof(*caps
));
533 g_malloc(sizeof(*caps
->value
));
534 caps
->value
->capability
= i
;
535 caps
->value
->state
= s
->enabled_capabilities
[i
];
541 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
543 MigrationParameters
*params
;
544 MigrationState
*s
= migrate_get_current();
546 params
= g_malloc0(sizeof(*params
));
547 params
->has_compress_level
= true;
548 params
->compress_level
= s
->parameters
.compress_level
;
549 params
->has_compress_threads
= true;
550 params
->compress_threads
= s
->parameters
.compress_threads
;
551 params
->has_decompress_threads
= true;
552 params
->decompress_threads
= s
->parameters
.decompress_threads
;
553 params
->has_cpu_throttle_initial
= true;
554 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
555 params
->has_cpu_throttle_increment
= true;
556 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
557 params
->has_tls_creds
= !!s
->parameters
.tls_creds
;
558 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
559 params
->has_tls_hostname
= !!s
->parameters
.tls_hostname
;
560 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
561 params
->has_max_bandwidth
= true;
562 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
563 params
->has_downtime_limit
= true;
564 params
->downtime_limit
= s
->parameters
.downtime_limit
;
565 params
->has_x_checkpoint_delay
= true;
566 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
567 params
->has_block_incremental
= true;
568 params
->block_incremental
= s
->parameters
.block_incremental
;
574 * Return true if we're already in the middle of a migration
575 * (i.e. any of the active or setup states)
577 static bool migration_is_setup_or_active(int state
)
580 case MIGRATION_STATUS_ACTIVE
:
581 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
582 case MIGRATION_STATUS_SETUP
:
591 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
593 info
->has_ram
= true;
594 info
->ram
= g_malloc0(sizeof(*info
->ram
));
595 info
->ram
->transferred
= ram_counters
.transferred
;
596 info
->ram
->total
= ram_bytes_total();
597 info
->ram
->duplicate
= ram_counters
.duplicate
;
598 /* legacy value. It is not used anymore */
599 info
->ram
->skipped
= 0;
600 info
->ram
->normal
= ram_counters
.normal
;
601 info
->ram
->normal_bytes
= ram_counters
.normal
*
602 qemu_target_page_size();
603 info
->ram
->mbps
= s
->mbps
;
604 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
605 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
606 info
->ram
->page_size
= qemu_target_page_size();
608 if (migrate_use_xbzrle()) {
609 info
->has_xbzrle_cache
= true;
610 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
611 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
612 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
613 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
614 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
615 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
616 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
619 if (cpu_throttle_active()) {
620 info
->has_cpu_throttle_percentage
= true;
621 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
624 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
625 info
->ram
->remaining
= ram_bytes_remaining();
626 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
630 MigrationInfo
*qmp_query_migrate(Error
**errp
)
632 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
633 MigrationState
*s
= migrate_get_current();
636 case MIGRATION_STATUS_NONE
:
637 /* no migration has happened ever */
639 case MIGRATION_STATUS_SETUP
:
640 info
->has_status
= true;
641 info
->has_total_time
= false;
643 case MIGRATION_STATUS_ACTIVE
:
644 case MIGRATION_STATUS_CANCELLING
:
645 info
->has_status
= true;
646 info
->has_total_time
= true;
647 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
649 info
->has_expected_downtime
= true;
650 info
->expected_downtime
= s
->expected_downtime
;
651 info
->has_setup_time
= true;
652 info
->setup_time
= s
->setup_time
;
654 populate_ram_info(info
, s
);
656 if (blk_mig_active()) {
657 info
->has_disk
= true;
658 info
->disk
= g_malloc0(sizeof(*info
->disk
));
659 info
->disk
->transferred
= blk_mig_bytes_transferred();
660 info
->disk
->remaining
= blk_mig_bytes_remaining();
661 info
->disk
->total
= blk_mig_bytes_total();
665 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
666 /* Mostly the same as active; TODO add some postcopy stats */
667 info
->has_status
= true;
668 info
->has_total_time
= true;
669 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
671 info
->has_expected_downtime
= true;
672 info
->expected_downtime
= s
->expected_downtime
;
673 info
->has_setup_time
= true;
674 info
->setup_time
= s
->setup_time
;
676 populate_ram_info(info
, s
);
678 if (blk_mig_active()) {
679 info
->has_disk
= true;
680 info
->disk
= g_malloc0(sizeof(*info
->disk
));
681 info
->disk
->transferred
= blk_mig_bytes_transferred();
682 info
->disk
->remaining
= blk_mig_bytes_remaining();
683 info
->disk
->total
= blk_mig_bytes_total();
687 case MIGRATION_STATUS_COLO
:
688 info
->has_status
= true;
689 /* TODO: display COLO specific information (checkpoint info etc.) */
691 case MIGRATION_STATUS_COMPLETED
:
692 info
->has_status
= true;
693 info
->has_total_time
= true;
694 info
->total_time
= s
->total_time
;
695 info
->has_downtime
= true;
696 info
->downtime
= s
->downtime
;
697 info
->has_setup_time
= true;
698 info
->setup_time
= s
->setup_time
;
700 populate_ram_info(info
, s
);
702 case MIGRATION_STATUS_FAILED
:
703 info
->has_status
= true;
705 info
->has_error_desc
= true;
706 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
709 case MIGRATION_STATUS_CANCELLED
:
710 info
->has_status
= true;
713 info
->status
= s
->state
;
718 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
721 MigrationState
*s
= migrate_get_current();
722 MigrationCapabilityStatusList
*cap
;
723 bool old_postcopy_cap
= migrate_postcopy_ram();
725 if (migration_is_setup_or_active(s
->state
)) {
726 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
730 for (cap
= params
; cap
; cap
= cap
->next
) {
731 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
732 if (cap
->value
->capability
== MIGRATION_CAPABILITY_BLOCK
733 && cap
->value
->state
) {
734 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
736 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
740 if (cap
->value
->capability
== MIGRATION_CAPABILITY_X_COLO
) {
741 if (!colo_supported()) {
742 error_setg(errp
, "COLO is not currently supported, please"
743 " configure with --enable-colo option in order to"
744 " support COLO feature");
748 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
751 if (migrate_postcopy_ram()) {
752 if (migrate_use_compression()) {
753 /* The decompression threads asynchronously write into RAM
754 * rather than use the atomic copies needed to avoid
755 * userfaulting. It should be possible to fix the decompression
756 * threads for compatibility in future.
758 error_report("Postcopy is not currently compatible with "
760 s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
] =
763 /* This check is reasonably expensive, so only when it's being
764 * set the first time, also it's only the destination that needs
767 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
768 !postcopy_ram_supported_by_host()) {
769 /* postcopy_ram_supported_by_host will have emitted a more
772 error_report("Postcopy is not supported");
773 s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
] =
779 void qmp_migrate_set_parameters(MigrationParameters
*params
, Error
**errp
)
781 MigrationState
*s
= migrate_get_current();
783 if (params
->has_compress_level
&&
784 (params
->compress_level
< 0 || params
->compress_level
> 9)) {
785 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
786 "is invalid, it should be in the range of 0 to 9");
789 if (params
->has_compress_threads
&&
790 (params
->compress_threads
< 1 || params
->compress_threads
> 255)) {
791 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
793 "is invalid, it should be in the range of 1 to 255");
796 if (params
->has_decompress_threads
&&
797 (params
->decompress_threads
< 1 || params
->decompress_threads
> 255)) {
798 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
799 "decompress_threads",
800 "is invalid, it should be in the range of 1 to 255");
803 if (params
->has_cpu_throttle_initial
&&
804 (params
->cpu_throttle_initial
< 1 ||
805 params
->cpu_throttle_initial
> 99)) {
806 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
807 "cpu_throttle_initial",
808 "an integer in the range of 1 to 99");
811 if (params
->has_cpu_throttle_increment
&&
812 (params
->cpu_throttle_increment
< 1 ||
813 params
->cpu_throttle_increment
> 99)) {
814 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
815 "cpu_throttle_increment",
816 "an integer in the range of 1 to 99");
819 if (params
->has_max_bandwidth
&&
820 (params
->max_bandwidth
< 0 || params
->max_bandwidth
> SIZE_MAX
)) {
821 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
822 " range of 0 to %zu bytes/second", SIZE_MAX
);
825 if (params
->has_downtime_limit
&&
826 (params
->downtime_limit
< 0 ||
827 params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
828 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
829 "the range of 0 to %d milliseconds",
830 MAX_MIGRATE_DOWNTIME
);
833 if (params
->has_x_checkpoint_delay
&& (params
->x_checkpoint_delay
< 0)) {
834 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
835 "x_checkpoint_delay",
836 "is invalid, it should be positive");
839 if (params
->has_compress_level
) {
840 s
->parameters
.compress_level
= params
->compress_level
;
842 if (params
->has_compress_threads
) {
843 s
->parameters
.compress_threads
= params
->compress_threads
;
845 if (params
->has_decompress_threads
) {
846 s
->parameters
.decompress_threads
= params
->decompress_threads
;
848 if (params
->has_cpu_throttle_initial
) {
849 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
851 if (params
->has_cpu_throttle_increment
) {
852 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
854 if (params
->has_tls_creds
) {
855 g_free(s
->parameters
.tls_creds
);
856 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
);
858 if (params
->has_tls_hostname
) {
859 g_free(s
->parameters
.tls_hostname
);
860 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
);
862 if (params
->has_max_bandwidth
) {
863 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
864 if (s
->to_dst_file
) {
865 qemu_file_set_rate_limit(s
->to_dst_file
,
866 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
869 if (params
->has_downtime_limit
) {
870 s
->parameters
.downtime_limit
= params
->downtime_limit
;
873 if (params
->has_x_checkpoint_delay
) {
874 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
875 if (migration_in_colo_state()) {
876 colo_checkpoint_notify(s
);
879 if (params
->has_block_incremental
) {
880 s
->parameters
.block_incremental
= params
->block_incremental
;
885 void qmp_migrate_start_postcopy(Error
**errp
)
887 MigrationState
*s
= migrate_get_current();
889 if (!migrate_postcopy_ram()) {
890 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
891 " the start of migration");
895 if (s
->state
== MIGRATION_STATUS_NONE
) {
896 error_setg(errp
, "Postcopy must be started after migration has been"
901 * we don't error if migration has finished since that would be racy
902 * with issuing this command.
904 atomic_set(&s
->start_postcopy
, true);
907 /* shared migration helpers */
909 void migrate_set_state(int *state
, int old_state
, int new_state
)
911 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
912 trace_migrate_set_state(new_state
);
913 migrate_generate_event(new_state
);
917 void migrate_set_block_enabled(bool value
, Error
**errp
)
919 MigrationCapabilityStatusList
*cap
;
921 cap
= g_new0(MigrationCapabilityStatusList
, 1);
922 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
923 cap
->value
->capability
= MIGRATION_CAPABILITY_BLOCK
;
924 cap
->value
->state
= value
;
925 qmp_migrate_set_capabilities(cap
, errp
);
926 qapi_free_MigrationCapabilityStatusList(cap
);
929 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
931 s
->parameters
.block_incremental
= value
;
934 static void block_cleanup_parameters(MigrationState
*s
)
936 if (s
->must_remove_block_options
) {
937 /* setting to false can never fail */
938 migrate_set_block_enabled(false, &error_abort
);
939 migrate_set_block_incremental(s
, false);
940 s
->must_remove_block_options
= false;
944 static void migrate_fd_cleanup(void *opaque
)
946 MigrationState
*s
= opaque
;
948 qemu_bh_delete(s
->cleanup_bh
);
949 s
->cleanup_bh
= NULL
;
951 if (s
->to_dst_file
) {
952 trace_migrate_fd_cleanup();
953 qemu_mutex_unlock_iothread();
954 if (s
->migration_thread_running
) {
955 qemu_thread_join(&s
->thread
);
956 s
->migration_thread_running
= false;
958 qemu_mutex_lock_iothread();
960 migrate_compress_threads_join();
961 qemu_fclose(s
->to_dst_file
);
962 s
->to_dst_file
= NULL
;
965 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
966 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
968 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
969 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
970 MIGRATION_STATUS_CANCELLED
);
973 notifier_list_notify(&migration_state_notifiers
, s
);
974 block_cleanup_parameters(s
);
977 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
979 trace_migrate_fd_error(error_get_pretty(error
));
980 assert(s
->to_dst_file
== NULL
);
981 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
982 MIGRATION_STATUS_FAILED
);
984 s
->error
= error_copy(error
);
986 notifier_list_notify(&migration_state_notifiers
, s
);
987 block_cleanup_parameters(s
);
990 static void migrate_fd_cancel(MigrationState
*s
)
993 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
994 trace_migrate_fd_cancel();
996 if (s
->rp_state
.from_dst_file
) {
997 /* shutdown the rp socket, so causing the rp thread to shutdown */
998 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1002 old_state
= s
->state
;
1003 if (!migration_is_setup_or_active(old_state
)) {
1006 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1007 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1010 * If we're unlucky the migration code might be stuck somewhere in a
1011 * send/write while the network has failed and is waiting to timeout;
1012 * if we've got shutdown(2) available then we can force it to quit.
1013 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1014 * called in a bh, so there is no race against this cancel.
1016 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1017 qemu_file_shutdown(f
);
1019 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1020 Error
*local_err
= NULL
;
1022 bdrv_invalidate_cache_all(&local_err
);
1024 error_report_err(local_err
);
1026 s
->block_inactive
= false;
1029 block_cleanup_parameters(s
);
1032 void add_migration_state_change_notifier(Notifier
*notify
)
1034 notifier_list_add(&migration_state_notifiers
, notify
);
1037 void remove_migration_state_change_notifier(Notifier
*notify
)
1039 notifier_remove(notify
);
1042 bool migration_in_setup(MigrationState
*s
)
1044 return s
->state
== MIGRATION_STATUS_SETUP
;
1047 bool migration_has_finished(MigrationState
*s
)
1049 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1052 bool migration_has_failed(MigrationState
*s
)
1054 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1055 s
->state
== MIGRATION_STATUS_FAILED
);
1058 bool migration_in_postcopy(void)
1060 MigrationState
*s
= migrate_get_current();
1062 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1065 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1067 return migration_in_postcopy() && s
->postcopy_after_devices
;
1070 bool migration_is_idle(void)
1072 MigrationState
*s
= migrate_get_current();
1075 case MIGRATION_STATUS_NONE
:
1076 case MIGRATION_STATUS_CANCELLED
:
1077 case MIGRATION_STATUS_COMPLETED
:
1078 case MIGRATION_STATUS_FAILED
:
1080 case MIGRATION_STATUS_SETUP
:
1081 case MIGRATION_STATUS_CANCELLING
:
1082 case MIGRATION_STATUS_ACTIVE
:
1083 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1084 case MIGRATION_STATUS_COLO
:
1086 case MIGRATION_STATUS__MAX
:
1087 g_assert_not_reached();
1093 MigrationState
*migrate_init(void)
1095 MigrationState
*s
= migrate_get_current();
1098 * Reinitialise all migration state, except
1099 * parameters/capabilities that the user set, and
1105 s
->to_dst_file
= NULL
;
1106 s
->state
= MIGRATION_STATUS_NONE
;
1107 s
->rp_state
.from_dst_file
= NULL
;
1108 s
->rp_state
.error
= false;
1111 s
->expected_downtime
= 0;
1113 s
->start_postcopy
= false;
1114 s
->postcopy_after_devices
= false;
1115 s
->migration_thread_running
= false;
1116 error_free(s
->error
);
1119 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1121 s
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1125 static GSList
*migration_blockers
;
1127 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1129 if (only_migratable
) {
1130 error_propagate(errp
, error_copy(reason
));
1131 error_prepend(errp
, "disallowing migration blocker "
1132 "(--only_migratable) for: ");
1136 if (migration_is_idle()) {
1137 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1141 error_propagate(errp
, error_copy(reason
));
1142 error_prepend(errp
, "disallowing migration blocker (migration in "
1147 void migrate_del_blocker(Error
*reason
)
1149 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1152 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1154 Error
*local_err
= NULL
;
1155 static bool once
= true;
1157 if (!deferred_incoming
) {
1158 error_setg(errp
, "For use with '-incoming defer'");
1162 error_setg(errp
, "The incoming migration has already been started");
1165 qemu_start_incoming_migration(uri
, &local_err
);
1168 error_propagate(errp
, local_err
);
1175 bool migration_is_blocked(Error
**errp
)
1177 if (qemu_savevm_state_blocked(errp
)) {
1181 if (migration_blockers
) {
1182 *errp
= error_copy(migration_blockers
->data
);
1189 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1190 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1193 Error
*local_err
= NULL
;
1194 MigrationState
*s
= migrate_get_current();
1197 if (migration_is_setup_or_active(s
->state
) ||
1198 s
->state
== MIGRATION_STATUS_CANCELLING
||
1199 s
->state
== MIGRATION_STATUS_COLO
) {
1200 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1203 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1204 error_setg(errp
, "Guest is waiting for an incoming migration");
1208 if (migration_is_blocked(errp
)) {
1212 if ((has_blk
&& blk
) || (has_inc
&& inc
)) {
1213 if (migrate_use_block() || migrate_use_block_incremental()) {
1214 error_setg(errp
, "Command options are incompatible with "
1215 "current migration capabilities");
1218 migrate_set_block_enabled(true, &local_err
);
1220 error_propagate(errp
, local_err
);
1223 s
->must_remove_block_options
= true;
1226 if (has_inc
&& inc
) {
1227 migrate_set_block_incremental(s
, true);
1232 if (strstart(uri
, "tcp:", &p
)) {
1233 tcp_start_outgoing_migration(s
, p
, &local_err
);
1235 } else if (strstart(uri
, "rdma:", &p
)) {
1236 rdma_start_outgoing_migration(s
, p
, &local_err
);
1238 } else if (strstart(uri
, "exec:", &p
)) {
1239 exec_start_outgoing_migration(s
, p
, &local_err
);
1240 } else if (strstart(uri
, "unix:", &p
)) {
1241 unix_start_outgoing_migration(s
, p
, &local_err
);
1242 } else if (strstart(uri
, "fd:", &p
)) {
1243 fd_start_outgoing_migration(s
, p
, &local_err
);
1245 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1246 "a valid migration protocol");
1247 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1248 MIGRATION_STATUS_FAILED
);
1253 migrate_fd_error(s
, local_err
);
1254 error_propagate(errp
, local_err
);
1259 void qmp_migrate_cancel(Error
**errp
)
1261 migrate_fd_cancel(migrate_get_current());
1264 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1266 MigrationState
*s
= migrate_get_current();
1269 /* Check for truncation */
1270 if (value
!= (size_t)value
) {
1271 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1272 "exceeding address space");
1276 /* Cache should not be larger than guest ram size */
1277 if (value
> ram_bytes_total()) {
1278 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1279 "exceeds guest ram size ");
1283 new_size
= xbzrle_cache_resize(value
);
1285 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1286 "is smaller than page size");
1290 s
->xbzrle_cache_size
= new_size
;
1293 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1295 return migrate_xbzrle_cache_size();
1298 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1300 MigrationParameters p
= {
1301 .has_max_bandwidth
= true,
1302 .max_bandwidth
= value
,
1305 qmp_migrate_set_parameters(&p
, errp
);
1308 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1310 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
1311 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1312 "the range of 0 to %d seconds",
1313 MAX_MIGRATE_DOWNTIME_SECONDS
);
1317 value
*= 1000; /* Convert to milliseconds */
1318 value
= MAX(0, MIN(INT64_MAX
, value
));
1320 MigrationParameters p
= {
1321 .has_downtime_limit
= true,
1322 .downtime_limit
= value
,
1325 qmp_migrate_set_parameters(&p
, errp
);
1328 bool migrate_release_ram(void)
1332 s
= migrate_get_current();
1334 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
1337 bool migrate_postcopy_ram(void)
1341 s
= migrate_get_current();
1343 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1346 bool migrate_auto_converge(void)
1350 s
= migrate_get_current();
1352 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1355 bool migrate_zero_blocks(void)
1359 s
= migrate_get_current();
1361 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
1364 bool migrate_use_compression(void)
1368 s
= migrate_get_current();
1370 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
1373 int migrate_compress_level(void)
1377 s
= migrate_get_current();
1379 return s
->parameters
.compress_level
;
1382 int migrate_compress_threads(void)
1386 s
= migrate_get_current();
1388 return s
->parameters
.compress_threads
;
1391 int migrate_decompress_threads(void)
1395 s
= migrate_get_current();
1397 return s
->parameters
.decompress_threads
;
1400 bool migrate_use_events(void)
1404 s
= migrate_get_current();
1406 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
1409 int migrate_use_xbzrle(void)
1413 s
= migrate_get_current();
1415 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
1418 int64_t migrate_xbzrle_cache_size(void)
1422 s
= migrate_get_current();
1424 return s
->xbzrle_cache_size
;
1427 bool migrate_use_block(void)
1431 s
= migrate_get_current();
1433 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
1436 bool migrate_use_block_incremental(void)
1440 s
= migrate_get_current();
1442 return s
->parameters
.block_incremental
;
1445 /* migration thread support */
1447 * Something bad happened to the RP stream, mark an error
1448 * The caller shall print or trace something to indicate why
1450 static void mark_source_rp_bad(MigrationState
*s
)
1452 s
->rp_state
.error
= true;
1455 static struct rp_cmd_args
{
1456 ssize_t len
; /* -1 = variable */
1459 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1460 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1461 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1462 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1463 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1464 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1468 * Process a request for pages received on the return path,
1469 * We're allowed to send more than requested (e.g. to round to our page size)
1470 * and we don't need to send pages that have already been sent.
1472 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1473 ram_addr_t start
, size_t len
)
1475 long our_host_ps
= getpagesize();
1477 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1480 * Since we currently insist on matching page sizes, just sanity check
1481 * we're being asked for whole host pages.
1483 if (start
& (our_host_ps
-1) ||
1484 (len
& (our_host_ps
-1))) {
1485 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1486 " len: %zd", __func__
, start
, len
);
1487 mark_source_rp_bad(ms
);
1491 if (ram_save_queue_pages(rbname
, start
, len
)) {
1492 mark_source_rp_bad(ms
);
1497 * Handles messages sent on the return path towards the source VM
1500 static void *source_return_path_thread(void *opaque
)
1502 MigrationState
*ms
= opaque
;
1503 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1504 uint16_t header_len
, header_type
;
1506 uint32_t tmp32
, sibling_error
;
1507 ram_addr_t start
= 0; /* =0 to silence warning */
1508 size_t len
= 0, expected_len
;
1511 trace_source_return_path_thread_entry();
1512 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1513 migration_is_setup_or_active(ms
->state
)) {
1514 trace_source_return_path_thread_loop_top();
1515 header_type
= qemu_get_be16(rp
);
1516 header_len
= qemu_get_be16(rp
);
1518 if (header_type
>= MIG_RP_MSG_MAX
||
1519 header_type
== MIG_RP_MSG_INVALID
) {
1520 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1521 header_type
, header_len
);
1522 mark_source_rp_bad(ms
);
1526 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1527 header_len
!= rp_cmd_args
[header_type
].len
) ||
1528 header_len
> sizeof(buf
)) {
1529 error_report("RP: Received '%s' message (0x%04x) with"
1530 "incorrect length %d expecting %zu",
1531 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1532 (size_t)rp_cmd_args
[header_type
].len
);
1533 mark_source_rp_bad(ms
);
1537 /* We know we've got a valid header by this point */
1538 res
= qemu_get_buffer(rp
, buf
, header_len
);
1539 if (res
!= header_len
) {
1540 error_report("RP: Failed reading data for message 0x%04x"
1541 " read %d expected %d",
1542 header_type
, res
, header_len
);
1543 mark_source_rp_bad(ms
);
1547 /* OK, we have the message and the data */
1548 switch (header_type
) {
1549 case MIG_RP_MSG_SHUT
:
1550 sibling_error
= ldl_be_p(buf
);
1551 trace_source_return_path_thread_shut(sibling_error
);
1552 if (sibling_error
) {
1553 error_report("RP: Sibling indicated error %d", sibling_error
);
1554 mark_source_rp_bad(ms
);
1557 * We'll let the main thread deal with closing the RP
1558 * we could do a shutdown(2) on it, but we're the only user
1559 * anyway, so there's nothing gained.
1563 case MIG_RP_MSG_PONG
:
1564 tmp32
= ldl_be_p(buf
);
1565 trace_source_return_path_thread_pong(tmp32
);
1568 case MIG_RP_MSG_REQ_PAGES
:
1569 start
= ldq_be_p(buf
);
1570 len
= ldl_be_p(buf
+ 8);
1571 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1574 case MIG_RP_MSG_REQ_PAGES_ID
:
1575 expected_len
= 12 + 1; /* header + termination */
1577 if (header_len
>= expected_len
) {
1578 start
= ldq_be_p(buf
);
1579 len
= ldl_be_p(buf
+ 8);
1580 /* Now we expect an idstr */
1581 tmp32
= buf
[12]; /* Length of the following idstr */
1582 buf
[13 + tmp32
] = '\0';
1583 expected_len
+= tmp32
;
1585 if (header_len
!= expected_len
) {
1586 error_report("RP: Req_Page_id with length %d expecting %zd",
1587 header_len
, expected_len
);
1588 mark_source_rp_bad(ms
);
1591 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
1598 if (qemu_file_get_error(rp
)) {
1599 trace_source_return_path_thread_bad_end();
1600 mark_source_rp_bad(ms
);
1603 trace_source_return_path_thread_end();
1605 ms
->rp_state
.from_dst_file
= NULL
;
1610 static int open_return_path_on_source(MigrationState
*ms
)
1613 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
1614 if (!ms
->rp_state
.from_dst_file
) {
1618 trace_open_return_path_on_source();
1619 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
1620 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
1622 trace_open_return_path_on_source_continue();
1627 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1628 static int await_return_path_close_on_source(MigrationState
*ms
)
1631 * If this is a normal exit then the destination will send a SHUT and the
1632 * rp_thread will exit, however if there's an error we need to cause
1635 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
1637 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1638 * waiting for the destination.
1640 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
1641 mark_source_rp_bad(ms
);
1643 trace_await_return_path_close_on_source_joining();
1644 qemu_thread_join(&ms
->rp_state
.rp_thread
);
1645 trace_await_return_path_close_on_source_close();
1646 return ms
->rp_state
.error
;
1650 * Switch from normal iteration to postcopy
1651 * Returns non-0 on error
1653 static int postcopy_start(MigrationState
*ms
, bool *old_vm_running
)
1656 QIOChannelBuffer
*bioc
;
1658 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1659 bool restart_block
= false;
1660 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
1661 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1663 trace_postcopy_start();
1664 qemu_mutex_lock_iothread();
1665 trace_postcopy_start_set_run();
1667 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1668 *old_vm_running
= runstate_is_running();
1669 global_state_store();
1670 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1675 ret
= bdrv_inactivate_all();
1679 restart_block
= true;
1682 * Cause any non-postcopiable, but iterative devices to
1683 * send out their final data.
1685 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true);
1688 * in Finish migrate and with the io-lock held everything should
1689 * be quiet, but we've potentially still got dirty pages and we
1690 * need to tell the destination to throw any pages it's already received
1693 if (ram_postcopy_send_discard_bitmap(ms
)) {
1694 error_report("postcopy send discard bitmap failed");
1699 * send rest of state - note things that are doing postcopy
1700 * will notice we're in POSTCOPY_ACTIVE and not actually
1701 * wrap their state up here
1703 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
1704 /* Ping just for debugging, helps line traces up */
1705 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
1708 * While loading the device state we may trigger page transfer
1709 * requests and the fd must be free to process those, and thus
1710 * the destination must read the whole device state off the fd before
1711 * it starts processing it. Unfortunately the ad-hoc migration format
1712 * doesn't allow the destination to know the size to read without fully
1713 * parsing it through each devices load-state code (especially the open
1714 * coded devices that use get/put).
1715 * So we wrap the device state up in a package with a length at the start;
1716 * to do this we use a qemu_buf to hold the whole of the device state.
1718 bioc
= qio_channel_buffer_new(4096);
1719 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
1720 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
1721 object_unref(OBJECT(bioc
));
1724 * Make sure the receiver can get incoming pages before we send the rest
1727 qemu_savevm_send_postcopy_listen(fb
);
1729 qemu_savevm_state_complete_precopy(fb
, false);
1730 qemu_savevm_send_ping(fb
, 3);
1732 qemu_savevm_send_postcopy_run(fb
);
1734 /* <><> end of stuff going into the package */
1736 /* Last point of recovery; as soon as we send the package the destination
1737 * can open devices and potentially start running.
1738 * Lets just check again we've not got any errors.
1740 ret
= qemu_file_get_error(ms
->to_dst_file
);
1742 error_report("postcopy_start: Migration stream errored (pre package)");
1746 restart_block
= false;
1748 /* Now send that blob */
1749 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
1754 /* Send a notify to give a chance for anything that needs to happen
1755 * at the transition to postcopy and after the device state; in particular
1756 * spice needs to trigger a transition now
1758 ms
->postcopy_after_devices
= true;
1759 notifier_list_notify(&migration_state_notifiers
, ms
);
1761 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
1763 qemu_mutex_unlock_iothread();
1766 * Although this ping is just for debug, it could potentially be
1767 * used for getting a better measurement of downtime at the source.
1769 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
1771 if (migrate_release_ram()) {
1772 ram_postcopy_migrated_memory_release(ms
);
1775 ret
= qemu_file_get_error(ms
->to_dst_file
);
1777 error_report("postcopy_start: Migration stream errored");
1778 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1779 MIGRATION_STATUS_FAILED
);
1787 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1788 MIGRATION_STATUS_FAILED
);
1789 if (restart_block
) {
1790 /* A failure happened early enough that we know the destination hasn't
1791 * accessed block devices, so we're safe to recover.
1793 Error
*local_err
= NULL
;
1795 bdrv_invalidate_cache_all(&local_err
);
1797 error_report_err(local_err
);
1800 qemu_mutex_unlock_iothread();
1805 * migration_completion: Used by migration_thread when there's not much left.
1806 * The caller 'breaks' the loop when this returns.
1808 * @s: Current migration state
1809 * @current_active_state: The migration state we expect to be in
1810 * @*old_vm_running: Pointer to old_vm_running flag
1811 * @*start_time: Pointer to time to update
1813 static void migration_completion(MigrationState
*s
, int current_active_state
,
1814 bool *old_vm_running
,
1815 int64_t *start_time
)
1819 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1820 qemu_mutex_lock_iothread();
1821 *start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1822 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1823 *old_vm_running
= runstate_is_running();
1824 ret
= global_state_store();
1827 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1829 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
1830 qemu_savevm_state_complete_precopy(s
->to_dst_file
, false);
1833 * Don't mark the image with BDRV_O_INACTIVE flag if
1834 * we will go into COLO stage later.
1836 if (ret
>= 0 && !migrate_colo_enabled()) {
1837 ret
= bdrv_inactivate_all();
1839 s
->block_inactive
= true;
1843 qemu_mutex_unlock_iothread();
1848 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1849 trace_migration_completion_postcopy_end();
1851 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
1852 trace_migration_completion_postcopy_end_after_complete();
1856 * If rp was opened we must clean up the thread before
1857 * cleaning everything else up (since if there are no failures
1858 * it will wait for the destination to send it's status in
1860 * Postcopy opens rp if enabled (even if it's not avtivated)
1862 if (migrate_postcopy_ram()) {
1864 trace_migration_completion_postcopy_end_before_rp();
1865 rp_error
= await_return_path_close_on_source(s
);
1866 trace_migration_completion_postcopy_end_after_rp(rp_error
);
1868 goto fail_invalidate
;
1872 if (qemu_file_get_error(s
->to_dst_file
)) {
1873 trace_migration_completion_file_err();
1874 goto fail_invalidate
;
1877 if (!migrate_colo_enabled()) {
1878 migrate_set_state(&s
->state
, current_active_state
,
1879 MIGRATION_STATUS_COMPLETED
);
1885 /* If not doing postcopy, vm_start() will be called: let's regain
1886 * control on images.
1888 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1889 Error
*local_err
= NULL
;
1891 qemu_mutex_lock_iothread();
1892 bdrv_invalidate_cache_all(&local_err
);
1894 error_report_err(local_err
);
1896 s
->block_inactive
= false;
1898 qemu_mutex_unlock_iothread();
1902 migrate_set_state(&s
->state
, current_active_state
,
1903 MIGRATION_STATUS_FAILED
);
1906 bool migrate_colo_enabled(void)
1908 MigrationState
*s
= migrate_get_current();
1909 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
1913 * Master migration thread on the source VM.
1914 * It drives the migration and pumps the data down the outgoing channel.
1916 static void *migration_thread(void *opaque
)
1918 MigrationState
*s
= opaque
;
1919 /* Used by the bandwidth calcs, updated later */
1920 int64_t initial_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1921 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
1922 int64_t initial_bytes
= 0;
1924 * The final stage happens when the remaining data is smaller than
1925 * this threshold; it's calculated from the requested downtime and
1926 * measured bandwidth
1928 int64_t threshold_size
= 0;
1929 int64_t start_time
= initial_time
;
1931 bool old_vm_running
= false;
1932 bool entered_postcopy
= false;
1933 /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
1934 enum MigrationStatus current_active_state
= MIGRATION_STATUS_ACTIVE
;
1935 bool enable_colo
= migrate_colo_enabled();
1937 rcu_register_thread();
1939 qemu_savevm_state_header(s
->to_dst_file
);
1941 if (migrate_postcopy_ram()) {
1942 /* Now tell the dest that it should open its end so it can reply */
1943 qemu_savevm_send_open_return_path(s
->to_dst_file
);
1945 /* And do a ping that will make stuff easier to debug */
1946 qemu_savevm_send_ping(s
->to_dst_file
, 1);
1949 * Tell the destination that we *might* want to do postcopy later;
1950 * if the other end can't do postcopy it should fail now, nice and
1953 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
1956 qemu_savevm_state_begin(s
->to_dst_file
);
1958 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
1959 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1960 MIGRATION_STATUS_ACTIVE
);
1962 trace_migration_thread_setup_complete();
1964 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
1965 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1966 int64_t current_time
;
1967 uint64_t pending_size
;
1969 if (!qemu_file_rate_limit(s
->to_dst_file
)) {
1970 uint64_t pend_post
, pend_nonpost
;
1972 qemu_savevm_state_pending(s
->to_dst_file
, threshold_size
,
1973 &pend_nonpost
, &pend_post
);
1974 pending_size
= pend_nonpost
+ pend_post
;
1975 trace_migrate_pending(pending_size
, threshold_size
,
1976 pend_post
, pend_nonpost
);
1977 if (pending_size
&& pending_size
>= threshold_size
) {
1978 /* Still a significant amount to transfer */
1980 if (migrate_postcopy_ram() &&
1981 s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
&&
1982 pend_nonpost
<= threshold_size
&&
1983 atomic_read(&s
->start_postcopy
)) {
1985 if (!postcopy_start(s
, &old_vm_running
)) {
1986 current_active_state
= MIGRATION_STATUS_POSTCOPY_ACTIVE
;
1987 entered_postcopy
= true;
1992 /* Just another iteration step */
1993 qemu_savevm_state_iterate(s
->to_dst_file
, entered_postcopy
);
1995 trace_migration_thread_low_pending(pending_size
);
1996 migration_completion(s
, current_active_state
,
1997 &old_vm_running
, &start_time
);
2002 if (qemu_file_get_error(s
->to_dst_file
)) {
2003 migrate_set_state(&s
->state
, current_active_state
,
2004 MIGRATION_STATUS_FAILED
);
2005 trace_migration_thread_file_err();
2008 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2009 if (current_time
>= initial_time
+ BUFFER_DELAY
) {
2010 uint64_t transferred_bytes
= qemu_ftell(s
->to_dst_file
) -
2012 uint64_t time_spent
= current_time
- initial_time
;
2013 double bandwidth
= (double)transferred_bytes
/ time_spent
;
2014 threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
2016 s
->mbps
= (((double) transferred_bytes
* 8.0) /
2017 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2019 trace_migrate_transferred(transferred_bytes
, time_spent
,
2020 bandwidth
, threshold_size
);
2021 /* if we haven't sent anything, we don't want to recalculate
2022 10000 is a small enough number for our purposes */
2023 if (ram_counters
.dirty_pages_rate
&& transferred_bytes
> 10000) {
2024 s
->expected_downtime
= ram_counters
.dirty_pages_rate
*
2025 qemu_target_page_size() / bandwidth
;
2028 qemu_file_reset_rate_limit(s
->to_dst_file
);
2029 initial_time
= current_time
;
2030 initial_bytes
= qemu_ftell(s
->to_dst_file
);
2032 if (qemu_file_rate_limit(s
->to_dst_file
)) {
2033 /* usleep expects microseconds */
2034 g_usleep((initial_time
+ BUFFER_DELAY
- current_time
)*1000);
2038 trace_migration_thread_after_loop();
2039 /* If we enabled cpu throttling for auto-converge, turn it off. */
2040 cpu_throttle_stop();
2041 end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2043 qemu_mutex_lock_iothread();
2045 * The resource has been allocated by migration will be reused in COLO
2046 * process, so don't release them.
2049 qemu_savevm_state_cleanup();
2051 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
2052 uint64_t transferred_bytes
= qemu_ftell(s
->to_dst_file
);
2053 s
->total_time
= end_time
- s
->total_time
;
2054 if (!entered_postcopy
) {
2055 s
->downtime
= end_time
- start_time
;
2057 if (s
->total_time
) {
2058 s
->mbps
= (((double) transferred_bytes
* 8.0) /
2059 ((double) s
->total_time
)) / 1000;
2061 runstate_set(RUN_STATE_POSTMIGRATE
);
2063 if (s
->state
== MIGRATION_STATUS_ACTIVE
&& enable_colo
) {
2064 migrate_start_colo_process(s
);
2065 qemu_savevm_state_cleanup();
2067 * Fixme: we will run VM in COLO no matter its old running state.
2068 * After exited COLO, we will keep running.
2070 old_vm_running
= true;
2072 if (old_vm_running
&& !entered_postcopy
) {
2075 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2076 runstate_set(RUN_STATE_POSTMIGRATE
);
2080 qemu_bh_schedule(s
->cleanup_bh
);
2081 qemu_mutex_unlock_iothread();
2083 rcu_unregister_thread();
2087 void migrate_fd_connect(MigrationState
*s
)
2089 s
->expected_downtime
= s
->parameters
.downtime_limit
;
2090 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
2092 qemu_file_set_blocking(s
->to_dst_file
, true);
2093 qemu_file_set_rate_limit(s
->to_dst_file
,
2094 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
2096 /* Notify before starting migration thread */
2097 notifier_list_notify(&migration_state_notifiers
, s
);
2100 * Open the return path; currently for postcopy but other things might
2103 if (migrate_postcopy_ram()) {
2104 if (open_return_path_on_source(s
)) {
2105 error_report("Unable to open return-path for postcopy");
2106 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2107 MIGRATION_STATUS_FAILED
);
2108 migrate_fd_cleanup(s
);
2113 migrate_compress_threads_create();
2114 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
2115 QEMU_THREAD_JOINABLE
);
2116 s
->migration_thread_running
= true;