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 "qemu-file-channel.h"
23 #include "migration/qemu-file.h"
24 #include "sysemu/sysemu.h"
25 #include "block/block.h"
26 #include "qapi/qmp/qerror.h"
27 #include "qapi/util.h"
28 #include "qemu/sockets.h"
30 #include "migration/block.h"
31 #include "postcopy-ram.h"
32 #include "qemu/thread.h"
33 #include "qmp-commands.h"
35 #include "qapi-event.h"
37 #include "exec/memory.h"
38 #include "exec/address-spaces.h"
39 #include "io/channel-buffer.h"
40 #include "io/channel-tls.h"
41 #include "migration/colo.h"
43 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
45 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
47 #define BUFFER_DELAY 100
48 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
50 /* Time in milliseconds we are allowed to stop the source,
51 * for sending the last part */
52 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
54 /* Maximum migrate downtime set to 2000 seconds */
55 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
56 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
58 /* Default compression thread count */
59 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
60 /* Default decompression thread count, usually decompression is at
61 * least 4 times as fast as compression.*/
62 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
63 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
64 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
65 /* Define default autoconverge cpu throttle migration parameters */
66 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
67 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
69 /* Migration XBZRLE default cache size */
70 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
72 /* The delay time (in ms) between two COLO checkpoints
73 * Note: Please change this default value to 10000 when we support hybrid mode.
75 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
77 static NotifierList migration_state_notifiers
=
78 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
80 static bool deferred_incoming
;
82 /* When we add fault tolerance, we could have several
83 migrations at once. For now we don't need to add
84 dynamic creation of migration */
87 MigrationState
*migrate_get_current(void)
90 static MigrationState current_migration
= {
91 .state
= MIGRATION_STATUS_NONE
,
92 .xbzrle_cache_size
= DEFAULT_MIGRATE_CACHE_SIZE
,
95 .compress_level
= DEFAULT_MIGRATE_COMPRESS_LEVEL
,
96 .compress_threads
= DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
,
97 .decompress_threads
= DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
,
98 .cpu_throttle_initial
= DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
,
99 .cpu_throttle_increment
= DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
,
100 .max_bandwidth
= MAX_THROTTLE
,
101 .downtime_limit
= DEFAULT_MIGRATE_SET_DOWNTIME
,
102 .x_checkpoint_delay
= DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
,
107 current_migration
.parameters
.tls_creds
= g_strdup("");
108 current_migration
.parameters
.tls_hostname
= g_strdup("");
111 return ¤t_migration
;
114 MigrationIncomingState
*migration_incoming_get_current(void)
117 static MigrationIncomingState mis_current
;
120 mis_current
.state
= MIGRATION_STATUS_NONE
;
121 memset(&mis_current
, 0, sizeof(MigrationIncomingState
));
122 QLIST_INIT(&mis_current
.loadvm_handlers
);
123 qemu_mutex_init(&mis_current
.rp_mutex
);
124 qemu_event_init(&mis_current
.main_thread_load_event
, false);
130 void migration_incoming_state_destroy(void)
132 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
134 qemu_event_destroy(&mis
->main_thread_load_event
);
135 loadvm_free_handlers(mis
);
142 uint8_t runstate
[100];
147 static GlobalState global_state
;
149 int global_state_store(void)
151 if (!runstate_store((char *)global_state
.runstate
,
152 sizeof(global_state
.runstate
))) {
153 error_report("runstate name too big: %s", global_state
.runstate
);
154 trace_migrate_state_too_big();
160 void global_state_store_running(void)
162 const char *state
= RunState_lookup
[RUN_STATE_RUNNING
];
163 strncpy((char *)global_state
.runstate
,
164 state
, sizeof(global_state
.runstate
));
167 static bool global_state_received(void)
169 return global_state
.received
;
172 static RunState
global_state_get_runstate(void)
174 return global_state
.state
;
177 void global_state_set_optional(void)
179 global_state
.optional
= true;
182 static bool global_state_needed(void *opaque
)
184 GlobalState
*s
= opaque
;
185 char *runstate
= (char *)s
->runstate
;
187 /* If it is not optional, it is mandatory */
189 if (s
->optional
== false) {
193 /* If state is running or paused, it is not needed */
195 if (strcmp(runstate
, "running") == 0 ||
196 strcmp(runstate
, "paused") == 0) {
200 /* for any other state it is needed */
204 static int global_state_post_load(void *opaque
, int version_id
)
206 GlobalState
*s
= opaque
;
207 Error
*local_err
= NULL
;
209 char *runstate
= (char *)s
->runstate
;
212 trace_migrate_global_state_post_load(runstate
);
214 r
= qapi_enum_parse(RunState_lookup
, runstate
, RUN_STATE__MAX
,
219 error_report_err(local_err
);
228 static void global_state_pre_save(void *opaque
)
230 GlobalState
*s
= opaque
;
232 trace_migrate_global_state_pre_save((char *)s
->runstate
);
233 s
->size
= strlen((char *)s
->runstate
) + 1;
236 static const VMStateDescription vmstate_globalstate
= {
237 .name
= "globalstate",
239 .minimum_version_id
= 1,
240 .post_load
= global_state_post_load
,
241 .pre_save
= global_state_pre_save
,
242 .needed
= global_state_needed
,
243 .fields
= (VMStateField
[]) {
244 VMSTATE_UINT32(size
, GlobalState
),
245 VMSTATE_BUFFER(runstate
, GlobalState
),
246 VMSTATE_END_OF_LIST()
250 void register_global_state(void)
252 /* We would use it independently that we receive it */
253 strcpy((char *)&global_state
.runstate
, "");
254 global_state
.received
= false;
255 vmstate_register(NULL
, 0, &vmstate_globalstate
, &global_state
);
258 static void migrate_generate_event(int new_state
)
260 if (migrate_use_events()) {
261 qapi_event_send_migration(new_state
, &error_abort
);
266 * Called on -incoming with a defer: uri.
267 * The migration can be started later after any parameters have been
270 static void deferred_incoming_migration(Error
**errp
)
272 if (deferred_incoming
) {
273 error_setg(errp
, "Incoming migration already deferred");
275 deferred_incoming
= true;
278 /* Request a range of pages from the source VM at the given
280 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
281 * as the last request (a name must have been given previously)
282 * Start: Address offset within the RB
283 * Len: Length in bytes required - must be a multiple of pagesize
285 void migrate_send_rp_req_pages(MigrationIncomingState
*mis
, const char *rbname
,
286 ram_addr_t start
, size_t len
)
288 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
289 size_t msglen
= 12; /* start + len */
291 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
292 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
295 int rbname_len
= strlen(rbname
);
296 assert(rbname_len
< 256);
298 bufc
[msglen
++] = rbname_len
;
299 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
300 msglen
+= rbname_len
;
301 migrate_send_rp_message(mis
, MIG_RP_MSG_REQ_PAGES_ID
, msglen
, bufc
);
303 migrate_send_rp_message(mis
, MIG_RP_MSG_REQ_PAGES
, msglen
, bufc
);
307 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
311 qapi_event_send_migration(MIGRATION_STATUS_SETUP
, &error_abort
);
312 if (!strcmp(uri
, "defer")) {
313 deferred_incoming_migration(errp
);
314 } else if (strstart(uri
, "tcp:", &p
)) {
315 tcp_start_incoming_migration(p
, errp
);
317 } else if (strstart(uri
, "rdma:", &p
)) {
318 rdma_start_incoming_migration(p
, errp
);
320 } else if (strstart(uri
, "exec:", &p
)) {
321 exec_start_incoming_migration(p
, errp
);
322 } else if (strstart(uri
, "unix:", &p
)) {
323 unix_start_incoming_migration(p
, errp
);
324 } else if (strstart(uri
, "fd:", &p
)) {
325 fd_start_incoming_migration(p
, errp
);
327 error_setg(errp
, "unknown migration protocol: %s", uri
);
331 static void process_incoming_migration_bh(void *opaque
)
333 Error
*local_err
= NULL
;
334 MigrationIncomingState
*mis
= opaque
;
336 /* Make sure all file formats flush their mutable metadata.
337 * If we get an error here, just don't restart the VM yet. */
338 bdrv_invalidate_cache_all(&local_err
);
340 error_report_err(local_err
);
346 * This must happen after all error conditions are dealt with and
347 * we're sure the VM is going to be running on this host.
349 qemu_announce_self();
351 /* If global state section was not received or we are in running
352 state, we need to obey autostart. Any other state is set with
355 if (!global_state_received() ||
356 global_state_get_runstate() == RUN_STATE_RUNNING
) {
360 runstate_set(RUN_STATE_PAUSED
);
363 runstate_set(global_state_get_runstate());
365 migrate_decompress_threads_join();
367 * This must happen after any state changes since as soon as an external
368 * observer sees this event they might start to prod at the VM assuming
371 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
372 MIGRATION_STATUS_COMPLETED
);
373 qemu_bh_delete(mis
->bh
);
374 migration_incoming_state_destroy();
377 static void process_incoming_migration_co(void *opaque
)
379 QEMUFile
*f
= opaque
;
380 MigrationIncomingState
*mis
= migration_incoming_get_current();
384 mis
->from_src_file
= f
;
385 mis
->largest_page_size
= qemu_ram_pagesize_largest();
386 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
387 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
388 MIGRATION_STATUS_ACTIVE
);
389 ret
= qemu_loadvm_state(f
);
391 ps
= postcopy_state_get();
392 trace_process_incoming_migration_co_end(ret
, ps
);
393 if (ps
!= POSTCOPY_INCOMING_NONE
) {
394 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
396 * Where a migration had postcopy enabled (and thus went to advise)
397 * but managed to complete within the precopy period, we can use
400 postcopy_ram_incoming_cleanup(mis
);
401 } else if (ret
>= 0) {
403 * Postcopy was started, cleanup should happen at the end of the
406 trace_process_incoming_migration_co_postcopy_end_main();
409 /* Else if something went wrong then just fall out of the normal exit */
412 /* we get COLO info, and know if we are in COLO mode */
413 if (!ret
&& migration_incoming_enable_colo()) {
414 mis
->migration_incoming_co
= qemu_coroutine_self();
415 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
416 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
417 mis
->have_colo_incoming_thread
= true;
418 qemu_coroutine_yield();
420 /* Wait checkpoint incoming thread exit before free resource */
421 qemu_thread_join(&mis
->colo_incoming_thread
);
425 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
426 MIGRATION_STATUS_FAILED
);
427 error_report("load of migration failed: %s", strerror(-ret
));
428 migrate_decompress_threads_join();
433 free_xbzrle_decoded_buf();
435 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
436 qemu_bh_schedule(mis
->bh
);
439 void migration_fd_process_incoming(QEMUFile
*f
)
441 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, f
);
443 migrate_decompress_threads_create();
444 qemu_file_set_blocking(f
, false);
445 qemu_coroutine_enter(co
);
449 * Send a message on the return channel back to the source
452 void migrate_send_rp_message(MigrationIncomingState
*mis
,
453 enum mig_rp_message_type message_type
,
454 uint16_t len
, void *data
)
456 trace_migrate_send_rp_message((int)message_type
, len
);
457 qemu_mutex_lock(&mis
->rp_mutex
);
458 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
459 qemu_put_be16(mis
->to_src_file
, len
);
460 qemu_put_buffer(mis
->to_src_file
, data
, len
);
461 qemu_fflush(mis
->to_src_file
);
462 qemu_mutex_unlock(&mis
->rp_mutex
);
466 * Send a 'SHUT' message on the return channel with the given value
467 * to indicate that we've finished with the RP. Non-0 value indicates
470 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
475 buf
= cpu_to_be32(value
);
476 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
480 * Send a 'PONG' message on the return channel with the given value
481 * (normally in response to a 'PING')
483 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
488 buf
= cpu_to_be32(value
);
489 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
492 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
494 MigrationCapabilityStatusList
*head
= NULL
;
495 MigrationCapabilityStatusList
*caps
;
496 MigrationState
*s
= migrate_get_current();
499 caps
= NULL
; /* silence compiler warning */
500 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
501 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
502 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
506 if (i
== MIGRATION_CAPABILITY_X_COLO
&& !colo_supported()) {
510 head
= g_malloc0(sizeof(*caps
));
513 caps
->next
= g_malloc0(sizeof(*caps
));
517 g_malloc(sizeof(*caps
->value
));
518 caps
->value
->capability
= i
;
519 caps
->value
->state
= s
->enabled_capabilities
[i
];
525 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
527 MigrationParameters
*params
;
528 MigrationState
*s
= migrate_get_current();
530 params
= g_malloc0(sizeof(*params
));
531 params
->has_compress_level
= true;
532 params
->compress_level
= s
->parameters
.compress_level
;
533 params
->has_compress_threads
= true;
534 params
->compress_threads
= s
->parameters
.compress_threads
;
535 params
->has_decompress_threads
= true;
536 params
->decompress_threads
= s
->parameters
.decompress_threads
;
537 params
->has_cpu_throttle_initial
= true;
538 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
539 params
->has_cpu_throttle_increment
= true;
540 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
541 params
->has_tls_creds
= !!s
->parameters
.tls_creds
;
542 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
543 params
->has_tls_hostname
= !!s
->parameters
.tls_hostname
;
544 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
545 params
->has_max_bandwidth
= true;
546 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
547 params
->has_downtime_limit
= true;
548 params
->downtime_limit
= s
->parameters
.downtime_limit
;
549 params
->has_x_checkpoint_delay
= true;
550 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
551 params
->has_block_incremental
= true;
552 params
->block_incremental
= s
->parameters
.block_incremental
;
558 * Return true if we're already in the middle of a migration
559 * (i.e. any of the active or setup states)
561 static bool migration_is_setup_or_active(int state
)
564 case MIGRATION_STATUS_ACTIVE
:
565 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
566 case MIGRATION_STATUS_SETUP
:
575 static void get_xbzrle_cache_stats(MigrationInfo
*info
)
577 if (migrate_use_xbzrle()) {
578 info
->has_xbzrle_cache
= true;
579 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
580 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
581 info
->xbzrle_cache
->bytes
= xbzrle_mig_bytes_transferred();
582 info
->xbzrle_cache
->pages
= xbzrle_mig_pages_transferred();
583 info
->xbzrle_cache
->cache_miss
= xbzrle_mig_pages_cache_miss();
584 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_mig_cache_miss_rate();
585 info
->xbzrle_cache
->overflow
= xbzrle_mig_pages_overflow();
589 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
591 info
->has_ram
= true;
592 info
->ram
= g_malloc0(sizeof(*info
->ram
));
593 info
->ram
->transferred
= ram_bytes_transferred();
594 info
->ram
->total
= ram_bytes_total();
595 info
->ram
->duplicate
= dup_mig_pages_transferred();
596 /* legacy value. It is not used anymore */
597 info
->ram
->skipped
= 0;
598 info
->ram
->normal
= norm_mig_pages_transferred();
599 info
->ram
->normal_bytes
= norm_mig_pages_transferred() *
600 qemu_target_page_size();
601 info
->ram
->mbps
= s
->mbps
;
602 info
->ram
->dirty_sync_count
= ram_dirty_sync_count();
603 info
->ram
->postcopy_requests
= ram_postcopy_requests();
604 info
->ram
->page_size
= qemu_target_page_size();
606 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
607 info
->ram
->remaining
= ram_bytes_remaining();
608 info
->ram
->dirty_pages_rate
= ram_dirty_pages_rate();
612 MigrationInfo
*qmp_query_migrate(Error
**errp
)
614 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
615 MigrationState
*s
= migrate_get_current();
618 case MIGRATION_STATUS_NONE
:
619 /* no migration has happened ever */
621 case MIGRATION_STATUS_SETUP
:
622 info
->has_status
= true;
623 info
->has_total_time
= false;
625 case MIGRATION_STATUS_ACTIVE
:
626 case MIGRATION_STATUS_CANCELLING
:
627 info
->has_status
= true;
628 info
->has_total_time
= true;
629 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
631 info
->has_expected_downtime
= true;
632 info
->expected_downtime
= s
->expected_downtime
;
633 info
->has_setup_time
= true;
634 info
->setup_time
= s
->setup_time
;
636 populate_ram_info(info
, s
);
638 if (blk_mig_active()) {
639 info
->has_disk
= true;
640 info
->disk
= g_malloc0(sizeof(*info
->disk
));
641 info
->disk
->transferred
= blk_mig_bytes_transferred();
642 info
->disk
->remaining
= blk_mig_bytes_remaining();
643 info
->disk
->total
= blk_mig_bytes_total();
646 if (cpu_throttle_active()) {
647 info
->has_cpu_throttle_percentage
= true;
648 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
651 get_xbzrle_cache_stats(info
);
653 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
654 /* Mostly the same as active; TODO add some postcopy stats */
655 info
->has_status
= true;
656 info
->has_total_time
= true;
657 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
659 info
->has_expected_downtime
= true;
660 info
->expected_downtime
= s
->expected_downtime
;
661 info
->has_setup_time
= true;
662 info
->setup_time
= s
->setup_time
;
664 populate_ram_info(info
, s
);
666 if (blk_mig_active()) {
667 info
->has_disk
= true;
668 info
->disk
= g_malloc0(sizeof(*info
->disk
));
669 info
->disk
->transferred
= blk_mig_bytes_transferred();
670 info
->disk
->remaining
= blk_mig_bytes_remaining();
671 info
->disk
->total
= blk_mig_bytes_total();
674 get_xbzrle_cache_stats(info
);
676 case MIGRATION_STATUS_COLO
:
677 info
->has_status
= true;
678 /* TODO: display COLO specific information (checkpoint info etc.) */
680 case MIGRATION_STATUS_COMPLETED
:
681 get_xbzrle_cache_stats(info
);
683 info
->has_status
= true;
684 info
->has_total_time
= true;
685 info
->total_time
= s
->total_time
;
686 info
->has_downtime
= true;
687 info
->downtime
= s
->downtime
;
688 info
->has_setup_time
= true;
689 info
->setup_time
= s
->setup_time
;
691 populate_ram_info(info
, s
);
693 case MIGRATION_STATUS_FAILED
:
694 info
->has_status
= true;
696 info
->has_error_desc
= true;
697 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
700 case MIGRATION_STATUS_CANCELLED
:
701 info
->has_status
= true;
704 info
->status
= s
->state
;
709 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
712 MigrationState
*s
= migrate_get_current();
713 MigrationCapabilityStatusList
*cap
;
714 bool old_postcopy_cap
= migrate_postcopy_ram();
716 if (migration_is_setup_or_active(s
->state
)) {
717 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
721 for (cap
= params
; cap
; cap
= cap
->next
) {
722 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
723 if (cap
->value
->capability
== MIGRATION_CAPABILITY_BLOCK
724 && cap
->value
->state
) {
725 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
727 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
731 if (cap
->value
->capability
== MIGRATION_CAPABILITY_X_COLO
) {
732 if (!colo_supported()) {
733 error_setg(errp
, "COLO is not currently supported, please"
734 " configure with --enable-colo option in order to"
735 " support COLO feature");
739 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
742 if (migrate_postcopy_ram()) {
743 if (migrate_use_compression()) {
744 /* The decompression threads asynchronously write into RAM
745 * rather than use the atomic copies needed to avoid
746 * userfaulting. It should be possible to fix the decompression
747 * threads for compatibility in future.
749 error_report("Postcopy is not currently compatible with "
751 s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
] =
754 /* This check is reasonably expensive, so only when it's being
755 * set the first time, also it's only the destination that needs
758 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
759 !postcopy_ram_supported_by_host()) {
760 /* postcopy_ram_supported_by_host will have emitted a more
763 error_report("Postcopy is not supported");
764 s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
] =
770 void qmp_migrate_set_parameters(MigrationParameters
*params
, Error
**errp
)
772 MigrationState
*s
= migrate_get_current();
774 if (params
->has_compress_level
&&
775 (params
->compress_level
< 0 || params
->compress_level
> 9)) {
776 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
777 "is invalid, it should be in the range of 0 to 9");
780 if (params
->has_compress_threads
&&
781 (params
->compress_threads
< 1 || params
->compress_threads
> 255)) {
782 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
784 "is invalid, it should be in the range of 1 to 255");
787 if (params
->has_decompress_threads
&&
788 (params
->decompress_threads
< 1 || params
->decompress_threads
> 255)) {
789 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
790 "decompress_threads",
791 "is invalid, it should be in the range of 1 to 255");
794 if (params
->has_cpu_throttle_initial
&&
795 (params
->cpu_throttle_initial
< 1 ||
796 params
->cpu_throttle_initial
> 99)) {
797 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
798 "cpu_throttle_initial",
799 "an integer in the range of 1 to 99");
802 if (params
->has_cpu_throttle_increment
&&
803 (params
->cpu_throttle_increment
< 1 ||
804 params
->cpu_throttle_increment
> 99)) {
805 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
806 "cpu_throttle_increment",
807 "an integer in the range of 1 to 99");
810 if (params
->has_max_bandwidth
&&
811 (params
->max_bandwidth
< 0 || params
->max_bandwidth
> SIZE_MAX
)) {
812 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
813 " range of 0 to %zu bytes/second", SIZE_MAX
);
816 if (params
->has_downtime_limit
&&
817 (params
->downtime_limit
< 0 ||
818 params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
819 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
820 "the range of 0 to %d milliseconds",
821 MAX_MIGRATE_DOWNTIME
);
824 if (params
->has_x_checkpoint_delay
&& (params
->x_checkpoint_delay
< 0)) {
825 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
826 "x_checkpoint_delay",
827 "is invalid, it should be positive");
830 if (params
->has_compress_level
) {
831 s
->parameters
.compress_level
= params
->compress_level
;
833 if (params
->has_compress_threads
) {
834 s
->parameters
.compress_threads
= params
->compress_threads
;
836 if (params
->has_decompress_threads
) {
837 s
->parameters
.decompress_threads
= params
->decompress_threads
;
839 if (params
->has_cpu_throttle_initial
) {
840 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
842 if (params
->has_cpu_throttle_increment
) {
843 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
845 if (params
->has_tls_creds
) {
846 g_free(s
->parameters
.tls_creds
);
847 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
);
849 if (params
->has_tls_hostname
) {
850 g_free(s
->parameters
.tls_hostname
);
851 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
);
853 if (params
->has_max_bandwidth
) {
854 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
855 if (s
->to_dst_file
) {
856 qemu_file_set_rate_limit(s
->to_dst_file
,
857 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
860 if (params
->has_downtime_limit
) {
861 s
->parameters
.downtime_limit
= params
->downtime_limit
;
864 if (params
->has_x_checkpoint_delay
) {
865 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
866 if (migration_in_colo_state()) {
867 colo_checkpoint_notify(s
);
870 if (params
->has_block_incremental
) {
871 s
->parameters
.block_incremental
= params
->block_incremental
;
876 void qmp_migrate_start_postcopy(Error
**errp
)
878 MigrationState
*s
= migrate_get_current();
880 if (!migrate_postcopy_ram()) {
881 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
882 " the start of migration");
886 if (s
->state
== MIGRATION_STATUS_NONE
) {
887 error_setg(errp
, "Postcopy must be started after migration has been"
892 * we don't error if migration has finished since that would be racy
893 * with issuing this command.
895 atomic_set(&s
->start_postcopy
, true);
898 /* shared migration helpers */
900 void migrate_set_state(int *state
, int old_state
, int new_state
)
902 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
903 trace_migrate_set_state(new_state
);
904 migrate_generate_event(new_state
);
908 void migrate_set_block_enabled(bool value
, Error
**errp
)
910 MigrationCapabilityStatusList
*cap
;
912 cap
= g_new0(MigrationCapabilityStatusList
, 1);
913 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
914 cap
->value
->capability
= MIGRATION_CAPABILITY_BLOCK
;
915 cap
->value
->state
= value
;
916 qmp_migrate_set_capabilities(cap
, errp
);
917 qapi_free_MigrationCapabilityStatusList(cap
);
920 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
922 s
->parameters
.block_incremental
= value
;
925 static void block_cleanup_parameters(MigrationState
*s
)
927 if (s
->must_remove_block_options
) {
928 /* setting to false can never fail */
929 migrate_set_block_enabled(false, &error_abort
);
930 migrate_set_block_incremental(s
, false);
931 s
->must_remove_block_options
= false;
935 static void migrate_fd_cleanup(void *opaque
)
937 MigrationState
*s
= opaque
;
939 qemu_bh_delete(s
->cleanup_bh
);
940 s
->cleanup_bh
= NULL
;
942 migration_page_queue_free();
944 if (s
->to_dst_file
) {
945 trace_migrate_fd_cleanup();
946 qemu_mutex_unlock_iothread();
947 if (s
->migration_thread_running
) {
948 qemu_thread_join(&s
->thread
);
949 s
->migration_thread_running
= false;
951 qemu_mutex_lock_iothread();
953 migrate_compress_threads_join();
954 qemu_fclose(s
->to_dst_file
);
955 s
->to_dst_file
= NULL
;
958 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
959 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
961 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
962 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
963 MIGRATION_STATUS_CANCELLED
);
966 notifier_list_notify(&migration_state_notifiers
, s
);
967 block_cleanup_parameters(s
);
970 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
972 trace_migrate_fd_error(error_get_pretty(error
));
973 assert(s
->to_dst_file
== NULL
);
974 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
975 MIGRATION_STATUS_FAILED
);
977 s
->error
= error_copy(error
);
979 notifier_list_notify(&migration_state_notifiers
, s
);
980 block_cleanup_parameters(s
);
983 static void migrate_fd_cancel(MigrationState
*s
)
986 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
987 trace_migrate_fd_cancel();
989 if (s
->rp_state
.from_dst_file
) {
990 /* shutdown the rp socket, so causing the rp thread to shutdown */
991 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
995 old_state
= s
->state
;
996 if (!migration_is_setup_or_active(old_state
)) {
999 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1000 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1003 * If we're unlucky the migration code might be stuck somewhere in a
1004 * send/write while the network has failed and is waiting to timeout;
1005 * if we've got shutdown(2) available then we can force it to quit.
1006 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1007 * called in a bh, so there is no race against this cancel.
1009 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1010 qemu_file_shutdown(f
);
1012 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1013 Error
*local_err
= NULL
;
1015 bdrv_invalidate_cache_all(&local_err
);
1017 error_report_err(local_err
);
1019 s
->block_inactive
= false;
1022 block_cleanup_parameters(s
);
1025 void add_migration_state_change_notifier(Notifier
*notify
)
1027 notifier_list_add(&migration_state_notifiers
, notify
);
1030 void remove_migration_state_change_notifier(Notifier
*notify
)
1032 notifier_remove(notify
);
1035 bool migration_in_setup(MigrationState
*s
)
1037 return s
->state
== MIGRATION_STATUS_SETUP
;
1040 bool migration_has_finished(MigrationState
*s
)
1042 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1045 bool migration_has_failed(MigrationState
*s
)
1047 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1048 s
->state
== MIGRATION_STATUS_FAILED
);
1051 bool migration_in_postcopy(void)
1053 MigrationState
*s
= migrate_get_current();
1055 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1058 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1060 return migration_in_postcopy() && s
->postcopy_after_devices
;
1063 bool migration_is_idle(void)
1065 MigrationState
*s
= migrate_get_current();
1068 case MIGRATION_STATUS_NONE
:
1069 case MIGRATION_STATUS_CANCELLED
:
1070 case MIGRATION_STATUS_COMPLETED
:
1071 case MIGRATION_STATUS_FAILED
:
1073 case MIGRATION_STATUS_SETUP
:
1074 case MIGRATION_STATUS_CANCELLING
:
1075 case MIGRATION_STATUS_ACTIVE
:
1076 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1077 case MIGRATION_STATUS_COLO
:
1079 case MIGRATION_STATUS__MAX
:
1080 g_assert_not_reached();
1086 MigrationState
*migrate_init(void)
1088 MigrationState
*s
= migrate_get_current();
1091 * Reinitialise all migration state, except
1092 * parameters/capabilities that the user set, and
1098 s
->to_dst_file
= NULL
;
1099 s
->state
= MIGRATION_STATUS_NONE
;
1100 s
->rp_state
.from_dst_file
= NULL
;
1101 s
->rp_state
.error
= false;
1104 s
->expected_downtime
= 0;
1106 s
->start_postcopy
= false;
1107 s
->postcopy_after_devices
= false;
1108 s
->migration_thread_running
= false;
1109 error_free(s
->error
);
1112 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1114 s
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1118 static GSList
*migration_blockers
;
1120 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1122 if (only_migratable
) {
1123 error_propagate(errp
, error_copy(reason
));
1124 error_prepend(errp
, "disallowing migration blocker "
1125 "(--only_migratable) for: ");
1129 if (migration_is_idle()) {
1130 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1134 error_propagate(errp
, error_copy(reason
));
1135 error_prepend(errp
, "disallowing migration blocker (migration in "
1140 void migrate_del_blocker(Error
*reason
)
1142 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1145 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1147 Error
*local_err
= NULL
;
1148 static bool once
= true;
1150 if (!deferred_incoming
) {
1151 error_setg(errp
, "For use with '-incoming defer'");
1155 error_setg(errp
, "The incoming migration has already been started");
1158 qemu_start_incoming_migration(uri
, &local_err
);
1161 error_propagate(errp
, local_err
);
1168 bool migration_is_blocked(Error
**errp
)
1170 if (qemu_savevm_state_blocked(errp
)) {
1174 if (migration_blockers
) {
1175 *errp
= error_copy(migration_blockers
->data
);
1182 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1183 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1186 Error
*local_err
= NULL
;
1187 MigrationState
*s
= migrate_get_current();
1190 if (migration_is_setup_or_active(s
->state
) ||
1191 s
->state
== MIGRATION_STATUS_CANCELLING
||
1192 s
->state
== MIGRATION_STATUS_COLO
) {
1193 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1196 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1197 error_setg(errp
, "Guest is waiting for an incoming migration");
1201 if (migration_is_blocked(errp
)) {
1205 if ((has_blk
&& blk
) || (has_inc
&& inc
)) {
1206 if (migrate_use_block() || migrate_use_block_incremental()) {
1207 error_setg(errp
, "Command options are incompatible with "
1208 "current migration capabilities");
1211 migrate_set_block_enabled(true, &local_err
);
1213 error_propagate(errp
, local_err
);
1216 s
->must_remove_block_options
= true;
1219 if (has_inc
&& inc
) {
1220 migrate_set_block_incremental(s
, true);
1225 if (strstart(uri
, "tcp:", &p
)) {
1226 tcp_start_outgoing_migration(s
, p
, &local_err
);
1228 } else if (strstart(uri
, "rdma:", &p
)) {
1229 rdma_start_outgoing_migration(s
, p
, &local_err
);
1231 } else if (strstart(uri
, "exec:", &p
)) {
1232 exec_start_outgoing_migration(s
, p
, &local_err
);
1233 } else if (strstart(uri
, "unix:", &p
)) {
1234 unix_start_outgoing_migration(s
, p
, &local_err
);
1235 } else if (strstart(uri
, "fd:", &p
)) {
1236 fd_start_outgoing_migration(s
, p
, &local_err
);
1238 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1239 "a valid migration protocol");
1240 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1241 MIGRATION_STATUS_FAILED
);
1246 migrate_fd_error(s
, local_err
);
1247 error_propagate(errp
, local_err
);
1252 void qmp_migrate_cancel(Error
**errp
)
1254 migrate_fd_cancel(migrate_get_current());
1257 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1259 MigrationState
*s
= migrate_get_current();
1262 /* Check for truncation */
1263 if (value
!= (size_t)value
) {
1264 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1265 "exceeding address space");
1269 /* Cache should not be larger than guest ram size */
1270 if (value
> ram_bytes_total()) {
1271 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1272 "exceeds guest ram size ");
1276 new_size
= xbzrle_cache_resize(value
);
1278 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1279 "is smaller than page size");
1283 s
->xbzrle_cache_size
= new_size
;
1286 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1288 return migrate_xbzrle_cache_size();
1291 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1293 MigrationParameters p
= {
1294 .has_max_bandwidth
= true,
1295 .max_bandwidth
= value
,
1298 qmp_migrate_set_parameters(&p
, errp
);
1301 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1303 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
1304 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1305 "the range of 0 to %d seconds",
1306 MAX_MIGRATE_DOWNTIME_SECONDS
);
1310 value
*= 1000; /* Convert to milliseconds */
1311 value
= MAX(0, MIN(INT64_MAX
, value
));
1313 MigrationParameters p
= {
1314 .has_downtime_limit
= true,
1315 .downtime_limit
= value
,
1318 qmp_migrate_set_parameters(&p
, errp
);
1321 bool migrate_release_ram(void)
1325 s
= migrate_get_current();
1327 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
1330 bool migrate_postcopy_ram(void)
1334 s
= migrate_get_current();
1336 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1339 bool migrate_auto_converge(void)
1343 s
= migrate_get_current();
1345 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1348 bool migrate_zero_blocks(void)
1352 s
= migrate_get_current();
1354 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
1357 bool migrate_use_compression(void)
1361 s
= migrate_get_current();
1363 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
1366 int migrate_compress_level(void)
1370 s
= migrate_get_current();
1372 return s
->parameters
.compress_level
;
1375 int migrate_compress_threads(void)
1379 s
= migrate_get_current();
1381 return s
->parameters
.compress_threads
;
1384 int migrate_decompress_threads(void)
1388 s
= migrate_get_current();
1390 return s
->parameters
.decompress_threads
;
1393 bool migrate_use_events(void)
1397 s
= migrate_get_current();
1399 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
1402 int migrate_use_xbzrle(void)
1406 s
= migrate_get_current();
1408 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
1411 int64_t migrate_xbzrle_cache_size(void)
1415 s
= migrate_get_current();
1417 return s
->xbzrle_cache_size
;
1420 bool migrate_use_block(void)
1424 s
= migrate_get_current();
1426 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
1429 bool migrate_use_block_incremental(void)
1433 s
= migrate_get_current();
1435 return s
->parameters
.block_incremental
;
1438 /* migration thread support */
1440 * Something bad happened to the RP stream, mark an error
1441 * The caller shall print or trace something to indicate why
1443 static void mark_source_rp_bad(MigrationState
*s
)
1445 s
->rp_state
.error
= true;
1448 static struct rp_cmd_args
{
1449 ssize_t len
; /* -1 = variable */
1452 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1453 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1454 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1455 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1456 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1457 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1461 * Process a request for pages received on the return path,
1462 * We're allowed to send more than requested (e.g. to round to our page size)
1463 * and we don't need to send pages that have already been sent.
1465 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1466 ram_addr_t start
, size_t len
)
1468 long our_host_ps
= getpagesize();
1470 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1473 * Since we currently insist on matching page sizes, just sanity check
1474 * we're being asked for whole host pages.
1476 if (start
& (our_host_ps
-1) ||
1477 (len
& (our_host_ps
-1))) {
1478 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1479 " len: %zd", __func__
, start
, len
);
1480 mark_source_rp_bad(ms
);
1484 if (ram_save_queue_pages(rbname
, start
, len
)) {
1485 mark_source_rp_bad(ms
);
1490 * Handles messages sent on the return path towards the source VM
1493 static void *source_return_path_thread(void *opaque
)
1495 MigrationState
*ms
= opaque
;
1496 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1497 uint16_t header_len
, header_type
;
1499 uint32_t tmp32
, sibling_error
;
1500 ram_addr_t start
= 0; /* =0 to silence warning */
1501 size_t len
= 0, expected_len
;
1504 trace_source_return_path_thread_entry();
1505 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1506 migration_is_setup_or_active(ms
->state
)) {
1507 trace_source_return_path_thread_loop_top();
1508 header_type
= qemu_get_be16(rp
);
1509 header_len
= qemu_get_be16(rp
);
1511 if (header_type
>= MIG_RP_MSG_MAX
||
1512 header_type
== MIG_RP_MSG_INVALID
) {
1513 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1514 header_type
, header_len
);
1515 mark_source_rp_bad(ms
);
1519 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1520 header_len
!= rp_cmd_args
[header_type
].len
) ||
1521 header_len
> sizeof(buf
)) {
1522 error_report("RP: Received '%s' message (0x%04x) with"
1523 "incorrect length %d expecting %zu",
1524 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1525 (size_t)rp_cmd_args
[header_type
].len
);
1526 mark_source_rp_bad(ms
);
1530 /* We know we've got a valid header by this point */
1531 res
= qemu_get_buffer(rp
, buf
, header_len
);
1532 if (res
!= header_len
) {
1533 error_report("RP: Failed reading data for message 0x%04x"
1534 " read %d expected %d",
1535 header_type
, res
, header_len
);
1536 mark_source_rp_bad(ms
);
1540 /* OK, we have the message and the data */
1541 switch (header_type
) {
1542 case MIG_RP_MSG_SHUT
:
1543 sibling_error
= ldl_be_p(buf
);
1544 trace_source_return_path_thread_shut(sibling_error
);
1545 if (sibling_error
) {
1546 error_report("RP: Sibling indicated error %d", sibling_error
);
1547 mark_source_rp_bad(ms
);
1550 * We'll let the main thread deal with closing the RP
1551 * we could do a shutdown(2) on it, but we're the only user
1552 * anyway, so there's nothing gained.
1556 case MIG_RP_MSG_PONG
:
1557 tmp32
= ldl_be_p(buf
);
1558 trace_source_return_path_thread_pong(tmp32
);
1561 case MIG_RP_MSG_REQ_PAGES
:
1562 start
= ldq_be_p(buf
);
1563 len
= ldl_be_p(buf
+ 8);
1564 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1567 case MIG_RP_MSG_REQ_PAGES_ID
:
1568 expected_len
= 12 + 1; /* header + termination */
1570 if (header_len
>= expected_len
) {
1571 start
= ldq_be_p(buf
);
1572 len
= ldl_be_p(buf
+ 8);
1573 /* Now we expect an idstr */
1574 tmp32
= buf
[12]; /* Length of the following idstr */
1575 buf
[13 + tmp32
] = '\0';
1576 expected_len
+= tmp32
;
1578 if (header_len
!= expected_len
) {
1579 error_report("RP: Req_Page_id with length %d expecting %zd",
1580 header_len
, expected_len
);
1581 mark_source_rp_bad(ms
);
1584 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
1591 if (qemu_file_get_error(rp
)) {
1592 trace_source_return_path_thread_bad_end();
1593 mark_source_rp_bad(ms
);
1596 trace_source_return_path_thread_end();
1598 ms
->rp_state
.from_dst_file
= NULL
;
1603 static int open_return_path_on_source(MigrationState
*ms
)
1606 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
1607 if (!ms
->rp_state
.from_dst_file
) {
1611 trace_open_return_path_on_source();
1612 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
1613 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
1615 trace_open_return_path_on_source_continue();
1620 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1621 static int await_return_path_close_on_source(MigrationState
*ms
)
1624 * If this is a normal exit then the destination will send a SHUT and the
1625 * rp_thread will exit, however if there's an error we need to cause
1628 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
1630 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1631 * waiting for the destination.
1633 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
1634 mark_source_rp_bad(ms
);
1636 trace_await_return_path_close_on_source_joining();
1637 qemu_thread_join(&ms
->rp_state
.rp_thread
);
1638 trace_await_return_path_close_on_source_close();
1639 return ms
->rp_state
.error
;
1643 * Switch from normal iteration to postcopy
1644 * Returns non-0 on error
1646 static int postcopy_start(MigrationState
*ms
, bool *old_vm_running
)
1649 QIOChannelBuffer
*bioc
;
1651 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1652 bool restart_block
= false;
1653 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
1654 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1656 trace_postcopy_start();
1657 qemu_mutex_lock_iothread();
1658 trace_postcopy_start_set_run();
1660 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1661 *old_vm_running
= runstate_is_running();
1662 global_state_store();
1663 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1668 ret
= bdrv_inactivate_all();
1672 restart_block
= true;
1675 * Cause any non-postcopiable, but iterative devices to
1676 * send out their final data.
1678 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true);
1681 * in Finish migrate and with the io-lock held everything should
1682 * be quiet, but we've potentially still got dirty pages and we
1683 * need to tell the destination to throw any pages it's already received
1686 if (ram_postcopy_send_discard_bitmap(ms
)) {
1687 error_report("postcopy send discard bitmap failed");
1692 * send rest of state - note things that are doing postcopy
1693 * will notice we're in POSTCOPY_ACTIVE and not actually
1694 * wrap their state up here
1696 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
1697 /* Ping just for debugging, helps line traces up */
1698 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
1701 * While loading the device state we may trigger page transfer
1702 * requests and the fd must be free to process those, and thus
1703 * the destination must read the whole device state off the fd before
1704 * it starts processing it. Unfortunately the ad-hoc migration format
1705 * doesn't allow the destination to know the size to read without fully
1706 * parsing it through each devices load-state code (especially the open
1707 * coded devices that use get/put).
1708 * So we wrap the device state up in a package with a length at the start;
1709 * to do this we use a qemu_buf to hold the whole of the device state.
1711 bioc
= qio_channel_buffer_new(4096);
1712 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
1713 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
1714 object_unref(OBJECT(bioc
));
1717 * Make sure the receiver can get incoming pages before we send the rest
1720 qemu_savevm_send_postcopy_listen(fb
);
1722 qemu_savevm_state_complete_precopy(fb
, false);
1723 qemu_savevm_send_ping(fb
, 3);
1725 qemu_savevm_send_postcopy_run(fb
);
1727 /* <><> end of stuff going into the package */
1729 /* Last point of recovery; as soon as we send the package the destination
1730 * can open devices and potentially start running.
1731 * Lets just check again we've not got any errors.
1733 ret
= qemu_file_get_error(ms
->to_dst_file
);
1735 error_report("postcopy_start: Migration stream errored (pre package)");
1739 restart_block
= false;
1741 /* Now send that blob */
1742 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
1747 /* Send a notify to give a chance for anything that needs to happen
1748 * at the transition to postcopy and after the device state; in particular
1749 * spice needs to trigger a transition now
1751 ms
->postcopy_after_devices
= true;
1752 notifier_list_notify(&migration_state_notifiers
, ms
);
1754 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
1756 qemu_mutex_unlock_iothread();
1759 * Although this ping is just for debug, it could potentially be
1760 * used for getting a better measurement of downtime at the source.
1762 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
1764 if (migrate_release_ram()) {
1765 ram_postcopy_migrated_memory_release(ms
);
1768 ret
= qemu_file_get_error(ms
->to_dst_file
);
1770 error_report("postcopy_start: Migration stream errored");
1771 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1772 MIGRATION_STATUS_FAILED
);
1780 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1781 MIGRATION_STATUS_FAILED
);
1782 if (restart_block
) {
1783 /* A failure happened early enough that we know the destination hasn't
1784 * accessed block devices, so we're safe to recover.
1786 Error
*local_err
= NULL
;
1788 bdrv_invalidate_cache_all(&local_err
);
1790 error_report_err(local_err
);
1793 qemu_mutex_unlock_iothread();
1798 * migration_completion: Used by migration_thread when there's not much left.
1799 * The caller 'breaks' the loop when this returns.
1801 * @s: Current migration state
1802 * @current_active_state: The migration state we expect to be in
1803 * @*old_vm_running: Pointer to old_vm_running flag
1804 * @*start_time: Pointer to time to update
1806 static void migration_completion(MigrationState
*s
, int current_active_state
,
1807 bool *old_vm_running
,
1808 int64_t *start_time
)
1812 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1813 qemu_mutex_lock_iothread();
1814 *start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1815 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1816 *old_vm_running
= runstate_is_running();
1817 ret
= global_state_store();
1820 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1822 * Don't mark the image with BDRV_O_INACTIVE flag if
1823 * we will go into COLO stage later.
1825 if (ret
>= 0 && !migrate_colo_enabled()) {
1826 ret
= bdrv_inactivate_all();
1829 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
1830 qemu_savevm_state_complete_precopy(s
->to_dst_file
, false);
1831 s
->block_inactive
= true;
1834 qemu_mutex_unlock_iothread();
1839 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1840 trace_migration_completion_postcopy_end();
1842 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
1843 trace_migration_completion_postcopy_end_after_complete();
1847 * If rp was opened we must clean up the thread before
1848 * cleaning everything else up (since if there are no failures
1849 * it will wait for the destination to send it's status in
1851 * Postcopy opens rp if enabled (even if it's not avtivated)
1853 if (migrate_postcopy_ram()) {
1855 trace_migration_completion_postcopy_end_before_rp();
1856 rp_error
= await_return_path_close_on_source(s
);
1857 trace_migration_completion_postcopy_end_after_rp(rp_error
);
1859 goto fail_invalidate
;
1863 if (qemu_file_get_error(s
->to_dst_file
)) {
1864 trace_migration_completion_file_err();
1865 goto fail_invalidate
;
1868 if (!migrate_colo_enabled()) {
1869 migrate_set_state(&s
->state
, current_active_state
,
1870 MIGRATION_STATUS_COMPLETED
);
1876 /* If not doing postcopy, vm_start() will be called: let's regain
1877 * control on images.
1879 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1880 Error
*local_err
= NULL
;
1882 qemu_mutex_lock_iothread();
1883 bdrv_invalidate_cache_all(&local_err
);
1885 error_report_err(local_err
);
1887 s
->block_inactive
= false;
1889 qemu_mutex_unlock_iothread();
1893 migrate_set_state(&s
->state
, current_active_state
,
1894 MIGRATION_STATUS_FAILED
);
1897 bool migrate_colo_enabled(void)
1899 MigrationState
*s
= migrate_get_current();
1900 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
1904 * Master migration thread on the source VM.
1905 * It drives the migration and pumps the data down the outgoing channel.
1907 static void *migration_thread(void *opaque
)
1909 MigrationState
*s
= opaque
;
1910 /* Used by the bandwidth calcs, updated later */
1911 int64_t initial_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1912 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
1913 int64_t initial_bytes
= 0;
1915 * The final stage happens when the remaining data is smaller than
1916 * this threshold; it's calculated from the requested downtime and
1917 * measured bandwidth
1919 int64_t threshold_size
= 0;
1920 int64_t start_time
= initial_time
;
1922 bool old_vm_running
= false;
1923 bool entered_postcopy
= false;
1924 /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
1925 enum MigrationStatus current_active_state
= MIGRATION_STATUS_ACTIVE
;
1926 bool enable_colo
= migrate_colo_enabled();
1928 rcu_register_thread();
1930 qemu_savevm_state_header(s
->to_dst_file
);
1932 if (migrate_postcopy_ram()) {
1933 /* Now tell the dest that it should open its end so it can reply */
1934 qemu_savevm_send_open_return_path(s
->to_dst_file
);
1936 /* And do a ping that will make stuff easier to debug */
1937 qemu_savevm_send_ping(s
->to_dst_file
, 1);
1940 * Tell the destination that we *might* want to do postcopy later;
1941 * if the other end can't do postcopy it should fail now, nice and
1944 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
1947 qemu_savevm_state_begin(s
->to_dst_file
);
1949 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
1950 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1951 MIGRATION_STATUS_ACTIVE
);
1953 trace_migration_thread_setup_complete();
1955 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
1956 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1957 int64_t current_time
;
1958 uint64_t pending_size
;
1960 if (!qemu_file_rate_limit(s
->to_dst_file
)) {
1961 uint64_t pend_post
, pend_nonpost
;
1963 qemu_savevm_state_pending(s
->to_dst_file
, threshold_size
,
1964 &pend_nonpost
, &pend_post
);
1965 pending_size
= pend_nonpost
+ pend_post
;
1966 trace_migrate_pending(pending_size
, threshold_size
,
1967 pend_post
, pend_nonpost
);
1968 if (pending_size
&& pending_size
>= threshold_size
) {
1969 /* Still a significant amount to transfer */
1971 if (migrate_postcopy_ram() &&
1972 s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
&&
1973 pend_nonpost
<= threshold_size
&&
1974 atomic_read(&s
->start_postcopy
)) {
1976 if (!postcopy_start(s
, &old_vm_running
)) {
1977 current_active_state
= MIGRATION_STATUS_POSTCOPY_ACTIVE
;
1978 entered_postcopy
= true;
1983 /* Just another iteration step */
1984 qemu_savevm_state_iterate(s
->to_dst_file
, entered_postcopy
);
1986 trace_migration_thread_low_pending(pending_size
);
1987 migration_completion(s
, current_active_state
,
1988 &old_vm_running
, &start_time
);
1993 if (qemu_file_get_error(s
->to_dst_file
)) {
1994 migrate_set_state(&s
->state
, current_active_state
,
1995 MIGRATION_STATUS_FAILED
);
1996 trace_migration_thread_file_err();
1999 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2000 if (current_time
>= initial_time
+ BUFFER_DELAY
) {
2001 uint64_t transferred_bytes
= qemu_ftell(s
->to_dst_file
) -
2003 uint64_t time_spent
= current_time
- initial_time
;
2004 double bandwidth
= (double)transferred_bytes
/ time_spent
;
2005 threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
2007 s
->mbps
= (((double) transferred_bytes
* 8.0) /
2008 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2010 trace_migrate_transferred(transferred_bytes
, time_spent
,
2011 bandwidth
, threshold_size
);
2012 /* if we haven't sent anything, we don't want to recalculate
2013 10000 is a small enough number for our purposes */
2014 if (ram_dirty_pages_rate() && transferred_bytes
> 10000) {
2015 s
->expected_downtime
= ram_dirty_pages_rate() *
2016 qemu_target_page_size() / bandwidth
;
2019 qemu_file_reset_rate_limit(s
->to_dst_file
);
2020 initial_time
= current_time
;
2021 initial_bytes
= qemu_ftell(s
->to_dst_file
);
2023 if (qemu_file_rate_limit(s
->to_dst_file
)) {
2024 /* usleep expects microseconds */
2025 g_usleep((initial_time
+ BUFFER_DELAY
- current_time
)*1000);
2029 trace_migration_thread_after_loop();
2030 /* If we enabled cpu throttling for auto-converge, turn it off. */
2031 cpu_throttle_stop();
2032 end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2034 qemu_mutex_lock_iothread();
2036 * The resource has been allocated by migration will be reused in COLO
2037 * process, so don't release them.
2040 qemu_savevm_state_cleanup();
2042 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
2043 uint64_t transferred_bytes
= qemu_ftell(s
->to_dst_file
);
2044 s
->total_time
= end_time
- s
->total_time
;
2045 if (!entered_postcopy
) {
2046 s
->downtime
= end_time
- start_time
;
2048 if (s
->total_time
) {
2049 s
->mbps
= (((double) transferred_bytes
* 8.0) /
2050 ((double) s
->total_time
)) / 1000;
2052 runstate_set(RUN_STATE_POSTMIGRATE
);
2054 if (s
->state
== MIGRATION_STATUS_ACTIVE
&& enable_colo
) {
2055 migrate_start_colo_process(s
);
2056 qemu_savevm_state_cleanup();
2058 * Fixme: we will run VM in COLO no matter its old running state.
2059 * After exited COLO, we will keep running.
2061 old_vm_running
= true;
2063 if (old_vm_running
&& !entered_postcopy
) {
2066 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2067 runstate_set(RUN_STATE_POSTMIGRATE
);
2071 qemu_bh_schedule(s
->cleanup_bh
);
2072 qemu_mutex_unlock_iothread();
2074 rcu_unregister_thread();
2078 void migrate_fd_connect(MigrationState
*s
)
2080 s
->expected_downtime
= s
->parameters
.downtime_limit
;
2081 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
2083 qemu_file_set_blocking(s
->to_dst_file
, true);
2084 qemu_file_set_rate_limit(s
->to_dst_file
,
2085 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
2087 /* Notify before starting migration thread */
2088 notifier_list_notify(&migration_state_notifiers
, s
);
2091 * Open the return path; currently for postcopy but other things might
2094 if (migrate_postcopy_ram()) {
2095 if (open_return_path_on_source(s
)) {
2096 error_report("Unable to open return-path for postcopy");
2097 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2098 MIGRATION_STATUS_FAILED
);
2099 migrate_fd_cleanup(s
);
2104 migrate_compress_threads_create();
2105 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
2106 QEMU_THREAD_JOINABLE
);
2107 s
->migration_thread_running
= true;