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 get_xbzrle_cache_stats(MigrationInfo
*info
)
593 if (migrate_use_xbzrle()) {
594 info
->has_xbzrle_cache
= true;
595 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
596 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
597 info
->xbzrle_cache
->bytes
= xbzrle_mig_bytes_transferred();
598 info
->xbzrle_cache
->pages
= xbzrle_mig_pages_transferred();
599 info
->xbzrle_cache
->cache_miss
= xbzrle_mig_pages_cache_miss();
600 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_mig_cache_miss_rate();
601 info
->xbzrle_cache
->overflow
= xbzrle_mig_pages_overflow();
605 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
607 info
->has_ram
= true;
608 info
->ram
= g_malloc0(sizeof(*info
->ram
));
609 info
->ram
->transferred
= ram_bytes_transferred();
610 info
->ram
->total
= ram_bytes_total();
611 info
->ram
->duplicate
= dup_mig_pages_transferred();
612 /* legacy value. It is not used anymore */
613 info
->ram
->skipped
= 0;
614 info
->ram
->normal
= norm_mig_pages_transferred();
615 info
->ram
->normal_bytes
= norm_mig_pages_transferred() *
616 qemu_target_page_size();
617 info
->ram
->mbps
= s
->mbps
;
618 info
->ram
->dirty_sync_count
= ram_dirty_sync_count();
619 info
->ram
->postcopy_requests
= ram_postcopy_requests();
620 info
->ram
->page_size
= qemu_target_page_size();
622 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
623 info
->ram
->remaining
= ram_bytes_remaining();
624 info
->ram
->dirty_pages_rate
= ram_dirty_pages_rate();
628 MigrationInfo
*qmp_query_migrate(Error
**errp
)
630 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
631 MigrationState
*s
= migrate_get_current();
634 case MIGRATION_STATUS_NONE
:
635 /* no migration has happened ever */
637 case MIGRATION_STATUS_SETUP
:
638 info
->has_status
= true;
639 info
->has_total_time
= false;
641 case MIGRATION_STATUS_ACTIVE
:
642 case MIGRATION_STATUS_CANCELLING
:
643 info
->has_status
= true;
644 info
->has_total_time
= true;
645 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
647 info
->has_expected_downtime
= true;
648 info
->expected_downtime
= s
->expected_downtime
;
649 info
->has_setup_time
= true;
650 info
->setup_time
= s
->setup_time
;
652 populate_ram_info(info
, s
);
654 if (blk_mig_active()) {
655 info
->has_disk
= true;
656 info
->disk
= g_malloc0(sizeof(*info
->disk
));
657 info
->disk
->transferred
= blk_mig_bytes_transferred();
658 info
->disk
->remaining
= blk_mig_bytes_remaining();
659 info
->disk
->total
= blk_mig_bytes_total();
662 if (cpu_throttle_active()) {
663 info
->has_cpu_throttle_percentage
= true;
664 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
667 get_xbzrle_cache_stats(info
);
669 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
670 /* Mostly the same as active; TODO add some postcopy stats */
671 info
->has_status
= true;
672 info
->has_total_time
= true;
673 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
675 info
->has_expected_downtime
= true;
676 info
->expected_downtime
= s
->expected_downtime
;
677 info
->has_setup_time
= true;
678 info
->setup_time
= s
->setup_time
;
680 populate_ram_info(info
, s
);
682 if (blk_mig_active()) {
683 info
->has_disk
= true;
684 info
->disk
= g_malloc0(sizeof(*info
->disk
));
685 info
->disk
->transferred
= blk_mig_bytes_transferred();
686 info
->disk
->remaining
= blk_mig_bytes_remaining();
687 info
->disk
->total
= blk_mig_bytes_total();
690 get_xbzrle_cache_stats(info
);
692 case MIGRATION_STATUS_COLO
:
693 info
->has_status
= true;
694 /* TODO: display COLO specific information (checkpoint info etc.) */
696 case MIGRATION_STATUS_COMPLETED
:
697 get_xbzrle_cache_stats(info
);
699 info
->has_status
= true;
700 info
->has_total_time
= true;
701 info
->total_time
= s
->total_time
;
702 info
->has_downtime
= true;
703 info
->downtime
= s
->downtime
;
704 info
->has_setup_time
= true;
705 info
->setup_time
= s
->setup_time
;
707 populate_ram_info(info
, s
);
709 case MIGRATION_STATUS_FAILED
:
710 info
->has_status
= true;
712 info
->has_error_desc
= true;
713 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
716 case MIGRATION_STATUS_CANCELLED
:
717 info
->has_status
= true;
720 info
->status
= s
->state
;
725 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
728 MigrationState
*s
= migrate_get_current();
729 MigrationCapabilityStatusList
*cap
;
730 bool old_postcopy_cap
= migrate_postcopy_ram();
732 if (migration_is_setup_or_active(s
->state
)) {
733 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
737 for (cap
= params
; cap
; cap
= cap
->next
) {
738 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
739 if (cap
->value
->capability
== MIGRATION_CAPABILITY_BLOCK
740 && cap
->value
->state
) {
741 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
743 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
747 if (cap
->value
->capability
== MIGRATION_CAPABILITY_X_COLO
) {
748 if (!colo_supported()) {
749 error_setg(errp
, "COLO is not currently supported, please"
750 " configure with --enable-colo option in order to"
751 " support COLO feature");
755 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
758 if (migrate_postcopy_ram()) {
759 if (migrate_use_compression()) {
760 /* The decompression threads asynchronously write into RAM
761 * rather than use the atomic copies needed to avoid
762 * userfaulting. It should be possible to fix the decompression
763 * threads for compatibility in future.
765 error_report("Postcopy is not currently compatible with "
767 s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
] =
770 /* This check is reasonably expensive, so only when it's being
771 * set the first time, also it's only the destination that needs
774 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
775 !postcopy_ram_supported_by_host()) {
776 /* postcopy_ram_supported_by_host will have emitted a more
779 error_report("Postcopy is not supported");
780 s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
] =
786 void qmp_migrate_set_parameters(MigrationParameters
*params
, Error
**errp
)
788 MigrationState
*s
= migrate_get_current();
790 if (params
->has_compress_level
&&
791 (params
->compress_level
< 0 || params
->compress_level
> 9)) {
792 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
793 "is invalid, it should be in the range of 0 to 9");
796 if (params
->has_compress_threads
&&
797 (params
->compress_threads
< 1 || params
->compress_threads
> 255)) {
798 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
800 "is invalid, it should be in the range of 1 to 255");
803 if (params
->has_decompress_threads
&&
804 (params
->decompress_threads
< 1 || params
->decompress_threads
> 255)) {
805 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
806 "decompress_threads",
807 "is invalid, it should be in the range of 1 to 255");
810 if (params
->has_cpu_throttle_initial
&&
811 (params
->cpu_throttle_initial
< 1 ||
812 params
->cpu_throttle_initial
> 99)) {
813 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
814 "cpu_throttle_initial",
815 "an integer in the range of 1 to 99");
818 if (params
->has_cpu_throttle_increment
&&
819 (params
->cpu_throttle_increment
< 1 ||
820 params
->cpu_throttle_increment
> 99)) {
821 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
822 "cpu_throttle_increment",
823 "an integer in the range of 1 to 99");
826 if (params
->has_max_bandwidth
&&
827 (params
->max_bandwidth
< 0 || params
->max_bandwidth
> SIZE_MAX
)) {
828 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
829 " range of 0 to %zu bytes/second", SIZE_MAX
);
832 if (params
->has_downtime_limit
&&
833 (params
->downtime_limit
< 0 ||
834 params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
835 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
836 "the range of 0 to %d milliseconds",
837 MAX_MIGRATE_DOWNTIME
);
840 if (params
->has_x_checkpoint_delay
&& (params
->x_checkpoint_delay
< 0)) {
841 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
842 "x_checkpoint_delay",
843 "is invalid, it should be positive");
846 if (params
->has_compress_level
) {
847 s
->parameters
.compress_level
= params
->compress_level
;
849 if (params
->has_compress_threads
) {
850 s
->parameters
.compress_threads
= params
->compress_threads
;
852 if (params
->has_decompress_threads
) {
853 s
->parameters
.decompress_threads
= params
->decompress_threads
;
855 if (params
->has_cpu_throttle_initial
) {
856 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
858 if (params
->has_cpu_throttle_increment
) {
859 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
861 if (params
->has_tls_creds
) {
862 g_free(s
->parameters
.tls_creds
);
863 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
);
865 if (params
->has_tls_hostname
) {
866 g_free(s
->parameters
.tls_hostname
);
867 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
);
869 if (params
->has_max_bandwidth
) {
870 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
871 if (s
->to_dst_file
) {
872 qemu_file_set_rate_limit(s
->to_dst_file
,
873 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
876 if (params
->has_downtime_limit
) {
877 s
->parameters
.downtime_limit
= params
->downtime_limit
;
880 if (params
->has_x_checkpoint_delay
) {
881 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
882 if (migration_in_colo_state()) {
883 colo_checkpoint_notify(s
);
886 if (params
->has_block_incremental
) {
887 s
->parameters
.block_incremental
= params
->block_incremental
;
892 void qmp_migrate_start_postcopy(Error
**errp
)
894 MigrationState
*s
= migrate_get_current();
896 if (!migrate_postcopy_ram()) {
897 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
898 " the start of migration");
902 if (s
->state
== MIGRATION_STATUS_NONE
) {
903 error_setg(errp
, "Postcopy must be started after migration has been"
908 * we don't error if migration has finished since that would be racy
909 * with issuing this command.
911 atomic_set(&s
->start_postcopy
, true);
914 /* shared migration helpers */
916 void migrate_set_state(int *state
, int old_state
, int new_state
)
918 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
919 trace_migrate_set_state(new_state
);
920 migrate_generate_event(new_state
);
924 void migrate_set_block_enabled(bool value
, Error
**errp
)
926 MigrationCapabilityStatusList
*cap
;
928 cap
= g_new0(MigrationCapabilityStatusList
, 1);
929 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
930 cap
->value
->capability
= MIGRATION_CAPABILITY_BLOCK
;
931 cap
->value
->state
= value
;
932 qmp_migrate_set_capabilities(cap
, errp
);
933 qapi_free_MigrationCapabilityStatusList(cap
);
936 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
938 s
->parameters
.block_incremental
= value
;
941 static void block_cleanup_parameters(MigrationState
*s
)
943 if (s
->must_remove_block_options
) {
944 /* setting to false can never fail */
945 migrate_set_block_enabled(false, &error_abort
);
946 migrate_set_block_incremental(s
, false);
947 s
->must_remove_block_options
= false;
951 static void migrate_fd_cleanup(void *opaque
)
953 MigrationState
*s
= opaque
;
955 qemu_bh_delete(s
->cleanup_bh
);
956 s
->cleanup_bh
= NULL
;
958 migration_page_queue_free();
960 if (s
->to_dst_file
) {
961 trace_migrate_fd_cleanup();
962 qemu_mutex_unlock_iothread();
963 if (s
->migration_thread_running
) {
964 qemu_thread_join(&s
->thread
);
965 s
->migration_thread_running
= false;
967 qemu_mutex_lock_iothread();
969 migrate_compress_threads_join();
970 qemu_fclose(s
->to_dst_file
);
971 s
->to_dst_file
= NULL
;
974 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
975 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
977 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
978 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
979 MIGRATION_STATUS_CANCELLED
);
982 notifier_list_notify(&migration_state_notifiers
, s
);
983 block_cleanup_parameters(s
);
986 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
988 trace_migrate_fd_error(error_get_pretty(error
));
989 assert(s
->to_dst_file
== NULL
);
990 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
991 MIGRATION_STATUS_FAILED
);
993 s
->error
= error_copy(error
);
995 notifier_list_notify(&migration_state_notifiers
, s
);
996 block_cleanup_parameters(s
);
999 static void migrate_fd_cancel(MigrationState
*s
)
1002 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1003 trace_migrate_fd_cancel();
1005 if (s
->rp_state
.from_dst_file
) {
1006 /* shutdown the rp socket, so causing the rp thread to shutdown */
1007 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1011 old_state
= s
->state
;
1012 if (!migration_is_setup_or_active(old_state
)) {
1015 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1016 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1019 * If we're unlucky the migration code might be stuck somewhere in a
1020 * send/write while the network has failed and is waiting to timeout;
1021 * if we've got shutdown(2) available then we can force it to quit.
1022 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1023 * called in a bh, so there is no race against this cancel.
1025 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1026 qemu_file_shutdown(f
);
1028 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1029 Error
*local_err
= NULL
;
1031 bdrv_invalidate_cache_all(&local_err
);
1033 error_report_err(local_err
);
1035 s
->block_inactive
= false;
1038 block_cleanup_parameters(s
);
1041 void add_migration_state_change_notifier(Notifier
*notify
)
1043 notifier_list_add(&migration_state_notifiers
, notify
);
1046 void remove_migration_state_change_notifier(Notifier
*notify
)
1048 notifier_remove(notify
);
1051 bool migration_in_setup(MigrationState
*s
)
1053 return s
->state
== MIGRATION_STATUS_SETUP
;
1056 bool migration_has_finished(MigrationState
*s
)
1058 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1061 bool migration_has_failed(MigrationState
*s
)
1063 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1064 s
->state
== MIGRATION_STATUS_FAILED
);
1067 bool migration_in_postcopy(void)
1069 MigrationState
*s
= migrate_get_current();
1071 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1074 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1076 return migration_in_postcopy() && s
->postcopy_after_devices
;
1079 bool migration_is_idle(void)
1081 MigrationState
*s
= migrate_get_current();
1084 case MIGRATION_STATUS_NONE
:
1085 case MIGRATION_STATUS_CANCELLED
:
1086 case MIGRATION_STATUS_COMPLETED
:
1087 case MIGRATION_STATUS_FAILED
:
1089 case MIGRATION_STATUS_SETUP
:
1090 case MIGRATION_STATUS_CANCELLING
:
1091 case MIGRATION_STATUS_ACTIVE
:
1092 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1093 case MIGRATION_STATUS_COLO
:
1095 case MIGRATION_STATUS__MAX
:
1096 g_assert_not_reached();
1102 MigrationState
*migrate_init(void)
1104 MigrationState
*s
= migrate_get_current();
1107 * Reinitialise all migration state, except
1108 * parameters/capabilities that the user set, and
1114 s
->to_dst_file
= NULL
;
1115 s
->state
= MIGRATION_STATUS_NONE
;
1116 s
->rp_state
.from_dst_file
= NULL
;
1117 s
->rp_state
.error
= false;
1120 s
->expected_downtime
= 0;
1122 s
->start_postcopy
= false;
1123 s
->postcopy_after_devices
= false;
1124 s
->migration_thread_running
= false;
1125 error_free(s
->error
);
1128 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1130 s
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1134 static GSList
*migration_blockers
;
1136 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1138 if (only_migratable
) {
1139 error_propagate(errp
, error_copy(reason
));
1140 error_prepend(errp
, "disallowing migration blocker "
1141 "(--only_migratable) for: ");
1145 if (migration_is_idle()) {
1146 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1150 error_propagate(errp
, error_copy(reason
));
1151 error_prepend(errp
, "disallowing migration blocker (migration in "
1156 void migrate_del_blocker(Error
*reason
)
1158 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1161 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1163 Error
*local_err
= NULL
;
1164 static bool once
= true;
1166 if (!deferred_incoming
) {
1167 error_setg(errp
, "For use with '-incoming defer'");
1171 error_setg(errp
, "The incoming migration has already been started");
1174 qemu_start_incoming_migration(uri
, &local_err
);
1177 error_propagate(errp
, local_err
);
1184 bool migration_is_blocked(Error
**errp
)
1186 if (qemu_savevm_state_blocked(errp
)) {
1190 if (migration_blockers
) {
1191 *errp
= error_copy(migration_blockers
->data
);
1198 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1199 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1202 Error
*local_err
= NULL
;
1203 MigrationState
*s
= migrate_get_current();
1206 if (migration_is_setup_or_active(s
->state
) ||
1207 s
->state
== MIGRATION_STATUS_CANCELLING
||
1208 s
->state
== MIGRATION_STATUS_COLO
) {
1209 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1212 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1213 error_setg(errp
, "Guest is waiting for an incoming migration");
1217 if (migration_is_blocked(errp
)) {
1221 if ((has_blk
&& blk
) || (has_inc
&& inc
)) {
1222 if (migrate_use_block() || migrate_use_block_incremental()) {
1223 error_setg(errp
, "Command options are incompatible with "
1224 "current migration capabilities");
1227 migrate_set_block_enabled(true, &local_err
);
1229 error_propagate(errp
, local_err
);
1232 s
->must_remove_block_options
= true;
1235 if (has_inc
&& inc
) {
1236 migrate_set_block_incremental(s
, true);
1241 if (strstart(uri
, "tcp:", &p
)) {
1242 tcp_start_outgoing_migration(s
, p
, &local_err
);
1244 } else if (strstart(uri
, "rdma:", &p
)) {
1245 rdma_start_outgoing_migration(s
, p
, &local_err
);
1247 } else if (strstart(uri
, "exec:", &p
)) {
1248 exec_start_outgoing_migration(s
, p
, &local_err
);
1249 } else if (strstart(uri
, "unix:", &p
)) {
1250 unix_start_outgoing_migration(s
, p
, &local_err
);
1251 } else if (strstart(uri
, "fd:", &p
)) {
1252 fd_start_outgoing_migration(s
, p
, &local_err
);
1254 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1255 "a valid migration protocol");
1256 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1257 MIGRATION_STATUS_FAILED
);
1262 migrate_fd_error(s
, local_err
);
1263 error_propagate(errp
, local_err
);
1268 void qmp_migrate_cancel(Error
**errp
)
1270 migrate_fd_cancel(migrate_get_current());
1273 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1275 MigrationState
*s
= migrate_get_current();
1278 /* Check for truncation */
1279 if (value
!= (size_t)value
) {
1280 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1281 "exceeding address space");
1285 /* Cache should not be larger than guest ram size */
1286 if (value
> ram_bytes_total()) {
1287 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1288 "exceeds guest ram size ");
1292 new_size
= xbzrle_cache_resize(value
);
1294 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1295 "is smaller than page size");
1299 s
->xbzrle_cache_size
= new_size
;
1302 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1304 return migrate_xbzrle_cache_size();
1307 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1309 MigrationParameters p
= {
1310 .has_max_bandwidth
= true,
1311 .max_bandwidth
= value
,
1314 qmp_migrate_set_parameters(&p
, errp
);
1317 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1319 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
1320 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1321 "the range of 0 to %d seconds",
1322 MAX_MIGRATE_DOWNTIME_SECONDS
);
1326 value
*= 1000; /* Convert to milliseconds */
1327 value
= MAX(0, MIN(INT64_MAX
, value
));
1329 MigrationParameters p
= {
1330 .has_downtime_limit
= true,
1331 .downtime_limit
= value
,
1334 qmp_migrate_set_parameters(&p
, errp
);
1337 bool migrate_release_ram(void)
1341 s
= migrate_get_current();
1343 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
1346 bool migrate_postcopy_ram(void)
1350 s
= migrate_get_current();
1352 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1355 bool migrate_auto_converge(void)
1359 s
= migrate_get_current();
1361 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1364 bool migrate_zero_blocks(void)
1368 s
= migrate_get_current();
1370 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
1373 bool migrate_use_compression(void)
1377 s
= migrate_get_current();
1379 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
1382 int migrate_compress_level(void)
1386 s
= migrate_get_current();
1388 return s
->parameters
.compress_level
;
1391 int migrate_compress_threads(void)
1395 s
= migrate_get_current();
1397 return s
->parameters
.compress_threads
;
1400 int migrate_decompress_threads(void)
1404 s
= migrate_get_current();
1406 return s
->parameters
.decompress_threads
;
1409 bool migrate_use_events(void)
1413 s
= migrate_get_current();
1415 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
1418 int migrate_use_xbzrle(void)
1422 s
= migrate_get_current();
1424 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
1427 int64_t migrate_xbzrle_cache_size(void)
1431 s
= migrate_get_current();
1433 return s
->xbzrle_cache_size
;
1436 bool migrate_use_block(void)
1440 s
= migrate_get_current();
1442 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
1445 bool migrate_use_block_incremental(void)
1449 s
= migrate_get_current();
1451 return s
->parameters
.block_incremental
;
1454 /* migration thread support */
1456 * Something bad happened to the RP stream, mark an error
1457 * The caller shall print or trace something to indicate why
1459 static void mark_source_rp_bad(MigrationState
*s
)
1461 s
->rp_state
.error
= true;
1464 static struct rp_cmd_args
{
1465 ssize_t len
; /* -1 = variable */
1468 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1469 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1470 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1471 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1472 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1473 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1477 * Process a request for pages received on the return path,
1478 * We're allowed to send more than requested (e.g. to round to our page size)
1479 * and we don't need to send pages that have already been sent.
1481 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1482 ram_addr_t start
, size_t len
)
1484 long our_host_ps
= getpagesize();
1486 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1489 * Since we currently insist on matching page sizes, just sanity check
1490 * we're being asked for whole host pages.
1492 if (start
& (our_host_ps
-1) ||
1493 (len
& (our_host_ps
-1))) {
1494 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1495 " len: %zd", __func__
, start
, len
);
1496 mark_source_rp_bad(ms
);
1500 if (ram_save_queue_pages(rbname
, start
, len
)) {
1501 mark_source_rp_bad(ms
);
1506 * Handles messages sent on the return path towards the source VM
1509 static void *source_return_path_thread(void *opaque
)
1511 MigrationState
*ms
= opaque
;
1512 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1513 uint16_t header_len
, header_type
;
1515 uint32_t tmp32
, sibling_error
;
1516 ram_addr_t start
= 0; /* =0 to silence warning */
1517 size_t len
= 0, expected_len
;
1520 trace_source_return_path_thread_entry();
1521 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1522 migration_is_setup_or_active(ms
->state
)) {
1523 trace_source_return_path_thread_loop_top();
1524 header_type
= qemu_get_be16(rp
);
1525 header_len
= qemu_get_be16(rp
);
1527 if (header_type
>= MIG_RP_MSG_MAX
||
1528 header_type
== MIG_RP_MSG_INVALID
) {
1529 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1530 header_type
, header_len
);
1531 mark_source_rp_bad(ms
);
1535 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1536 header_len
!= rp_cmd_args
[header_type
].len
) ||
1537 header_len
> sizeof(buf
)) {
1538 error_report("RP: Received '%s' message (0x%04x) with"
1539 "incorrect length %d expecting %zu",
1540 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1541 (size_t)rp_cmd_args
[header_type
].len
);
1542 mark_source_rp_bad(ms
);
1546 /* We know we've got a valid header by this point */
1547 res
= qemu_get_buffer(rp
, buf
, header_len
);
1548 if (res
!= header_len
) {
1549 error_report("RP: Failed reading data for message 0x%04x"
1550 " read %d expected %d",
1551 header_type
, res
, header_len
);
1552 mark_source_rp_bad(ms
);
1556 /* OK, we have the message and the data */
1557 switch (header_type
) {
1558 case MIG_RP_MSG_SHUT
:
1559 sibling_error
= ldl_be_p(buf
);
1560 trace_source_return_path_thread_shut(sibling_error
);
1561 if (sibling_error
) {
1562 error_report("RP: Sibling indicated error %d", sibling_error
);
1563 mark_source_rp_bad(ms
);
1566 * We'll let the main thread deal with closing the RP
1567 * we could do a shutdown(2) on it, but we're the only user
1568 * anyway, so there's nothing gained.
1572 case MIG_RP_MSG_PONG
:
1573 tmp32
= ldl_be_p(buf
);
1574 trace_source_return_path_thread_pong(tmp32
);
1577 case MIG_RP_MSG_REQ_PAGES
:
1578 start
= ldq_be_p(buf
);
1579 len
= ldl_be_p(buf
+ 8);
1580 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1583 case MIG_RP_MSG_REQ_PAGES_ID
:
1584 expected_len
= 12 + 1; /* header + termination */
1586 if (header_len
>= expected_len
) {
1587 start
= ldq_be_p(buf
);
1588 len
= ldl_be_p(buf
+ 8);
1589 /* Now we expect an idstr */
1590 tmp32
= buf
[12]; /* Length of the following idstr */
1591 buf
[13 + tmp32
] = '\0';
1592 expected_len
+= tmp32
;
1594 if (header_len
!= expected_len
) {
1595 error_report("RP: Req_Page_id with length %d expecting %zd",
1596 header_len
, expected_len
);
1597 mark_source_rp_bad(ms
);
1600 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
1607 if (qemu_file_get_error(rp
)) {
1608 trace_source_return_path_thread_bad_end();
1609 mark_source_rp_bad(ms
);
1612 trace_source_return_path_thread_end();
1614 ms
->rp_state
.from_dst_file
= NULL
;
1619 static int open_return_path_on_source(MigrationState
*ms
)
1622 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
1623 if (!ms
->rp_state
.from_dst_file
) {
1627 trace_open_return_path_on_source();
1628 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
1629 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
1631 trace_open_return_path_on_source_continue();
1636 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1637 static int await_return_path_close_on_source(MigrationState
*ms
)
1640 * If this is a normal exit then the destination will send a SHUT and the
1641 * rp_thread will exit, however if there's an error we need to cause
1644 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
1646 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1647 * waiting for the destination.
1649 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
1650 mark_source_rp_bad(ms
);
1652 trace_await_return_path_close_on_source_joining();
1653 qemu_thread_join(&ms
->rp_state
.rp_thread
);
1654 trace_await_return_path_close_on_source_close();
1655 return ms
->rp_state
.error
;
1659 * Switch from normal iteration to postcopy
1660 * Returns non-0 on error
1662 static int postcopy_start(MigrationState
*ms
, bool *old_vm_running
)
1665 QIOChannelBuffer
*bioc
;
1667 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1668 bool restart_block
= false;
1669 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
1670 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1672 trace_postcopy_start();
1673 qemu_mutex_lock_iothread();
1674 trace_postcopy_start_set_run();
1676 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1677 *old_vm_running
= runstate_is_running();
1678 global_state_store();
1679 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1684 ret
= bdrv_inactivate_all();
1688 restart_block
= true;
1691 * Cause any non-postcopiable, but iterative devices to
1692 * send out their final data.
1694 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true);
1697 * in Finish migrate and with the io-lock held everything should
1698 * be quiet, but we've potentially still got dirty pages and we
1699 * need to tell the destination to throw any pages it's already received
1702 if (ram_postcopy_send_discard_bitmap(ms
)) {
1703 error_report("postcopy send discard bitmap failed");
1708 * send rest of state - note things that are doing postcopy
1709 * will notice we're in POSTCOPY_ACTIVE and not actually
1710 * wrap their state up here
1712 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
1713 /* Ping just for debugging, helps line traces up */
1714 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
1717 * While loading the device state we may trigger page transfer
1718 * requests and the fd must be free to process those, and thus
1719 * the destination must read the whole device state off the fd before
1720 * it starts processing it. Unfortunately the ad-hoc migration format
1721 * doesn't allow the destination to know the size to read without fully
1722 * parsing it through each devices load-state code (especially the open
1723 * coded devices that use get/put).
1724 * So we wrap the device state up in a package with a length at the start;
1725 * to do this we use a qemu_buf to hold the whole of the device state.
1727 bioc
= qio_channel_buffer_new(4096);
1728 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
1729 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
1730 object_unref(OBJECT(bioc
));
1733 * Make sure the receiver can get incoming pages before we send the rest
1736 qemu_savevm_send_postcopy_listen(fb
);
1738 qemu_savevm_state_complete_precopy(fb
, false);
1739 qemu_savevm_send_ping(fb
, 3);
1741 qemu_savevm_send_postcopy_run(fb
);
1743 /* <><> end of stuff going into the package */
1745 /* Last point of recovery; as soon as we send the package the destination
1746 * can open devices and potentially start running.
1747 * Lets just check again we've not got any errors.
1749 ret
= qemu_file_get_error(ms
->to_dst_file
);
1751 error_report("postcopy_start: Migration stream errored (pre package)");
1755 restart_block
= false;
1757 /* Now send that blob */
1758 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
1763 /* Send a notify to give a chance for anything that needs to happen
1764 * at the transition to postcopy and after the device state; in particular
1765 * spice needs to trigger a transition now
1767 ms
->postcopy_after_devices
= true;
1768 notifier_list_notify(&migration_state_notifiers
, ms
);
1770 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
1772 qemu_mutex_unlock_iothread();
1775 * Although this ping is just for debug, it could potentially be
1776 * used for getting a better measurement of downtime at the source.
1778 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
1780 if (migrate_release_ram()) {
1781 ram_postcopy_migrated_memory_release(ms
);
1784 ret
= qemu_file_get_error(ms
->to_dst_file
);
1786 error_report("postcopy_start: Migration stream errored");
1787 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1788 MIGRATION_STATUS_FAILED
);
1796 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1797 MIGRATION_STATUS_FAILED
);
1798 if (restart_block
) {
1799 /* A failure happened early enough that we know the destination hasn't
1800 * accessed block devices, so we're safe to recover.
1802 Error
*local_err
= NULL
;
1804 bdrv_invalidate_cache_all(&local_err
);
1806 error_report_err(local_err
);
1809 qemu_mutex_unlock_iothread();
1814 * migration_completion: Used by migration_thread when there's not much left.
1815 * The caller 'breaks' the loop when this returns.
1817 * @s: Current migration state
1818 * @current_active_state: The migration state we expect to be in
1819 * @*old_vm_running: Pointer to old_vm_running flag
1820 * @*start_time: Pointer to time to update
1822 static void migration_completion(MigrationState
*s
, int current_active_state
,
1823 bool *old_vm_running
,
1824 int64_t *start_time
)
1828 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1829 qemu_mutex_lock_iothread();
1830 *start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1831 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1832 *old_vm_running
= runstate_is_running();
1833 ret
= global_state_store();
1836 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1838 * Don't mark the image with BDRV_O_INACTIVE flag if
1839 * we will go into COLO stage later.
1841 if (ret
>= 0 && !migrate_colo_enabled()) {
1842 ret
= bdrv_inactivate_all();
1845 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
1846 qemu_savevm_state_complete_precopy(s
->to_dst_file
, false);
1847 s
->block_inactive
= true;
1850 qemu_mutex_unlock_iothread();
1855 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1856 trace_migration_completion_postcopy_end();
1858 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
1859 trace_migration_completion_postcopy_end_after_complete();
1863 * If rp was opened we must clean up the thread before
1864 * cleaning everything else up (since if there are no failures
1865 * it will wait for the destination to send it's status in
1867 * Postcopy opens rp if enabled (even if it's not avtivated)
1869 if (migrate_postcopy_ram()) {
1871 trace_migration_completion_postcopy_end_before_rp();
1872 rp_error
= await_return_path_close_on_source(s
);
1873 trace_migration_completion_postcopy_end_after_rp(rp_error
);
1875 goto fail_invalidate
;
1879 if (qemu_file_get_error(s
->to_dst_file
)) {
1880 trace_migration_completion_file_err();
1881 goto fail_invalidate
;
1884 if (!migrate_colo_enabled()) {
1885 migrate_set_state(&s
->state
, current_active_state
,
1886 MIGRATION_STATUS_COMPLETED
);
1892 /* If not doing postcopy, vm_start() will be called: let's regain
1893 * control on images.
1895 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1896 Error
*local_err
= NULL
;
1898 qemu_mutex_lock_iothread();
1899 bdrv_invalidate_cache_all(&local_err
);
1901 error_report_err(local_err
);
1903 s
->block_inactive
= false;
1905 qemu_mutex_unlock_iothread();
1909 migrate_set_state(&s
->state
, current_active_state
,
1910 MIGRATION_STATUS_FAILED
);
1913 bool migrate_colo_enabled(void)
1915 MigrationState
*s
= migrate_get_current();
1916 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
1920 * Master migration thread on the source VM.
1921 * It drives the migration and pumps the data down the outgoing channel.
1923 static void *migration_thread(void *opaque
)
1925 MigrationState
*s
= opaque
;
1926 /* Used by the bandwidth calcs, updated later */
1927 int64_t initial_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1928 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
1929 int64_t initial_bytes
= 0;
1931 * The final stage happens when the remaining data is smaller than
1932 * this threshold; it's calculated from the requested downtime and
1933 * measured bandwidth
1935 int64_t threshold_size
= 0;
1936 int64_t start_time
= initial_time
;
1938 bool old_vm_running
= false;
1939 bool entered_postcopy
= false;
1940 /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
1941 enum MigrationStatus current_active_state
= MIGRATION_STATUS_ACTIVE
;
1942 bool enable_colo
= migrate_colo_enabled();
1944 rcu_register_thread();
1946 qemu_savevm_state_header(s
->to_dst_file
);
1948 if (migrate_postcopy_ram()) {
1949 /* Now tell the dest that it should open its end so it can reply */
1950 qemu_savevm_send_open_return_path(s
->to_dst_file
);
1952 /* And do a ping that will make stuff easier to debug */
1953 qemu_savevm_send_ping(s
->to_dst_file
, 1);
1956 * Tell the destination that we *might* want to do postcopy later;
1957 * if the other end can't do postcopy it should fail now, nice and
1960 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
1963 qemu_savevm_state_begin(s
->to_dst_file
);
1965 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
1966 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1967 MIGRATION_STATUS_ACTIVE
);
1969 trace_migration_thread_setup_complete();
1971 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
1972 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1973 int64_t current_time
;
1974 uint64_t pending_size
;
1976 if (!qemu_file_rate_limit(s
->to_dst_file
)) {
1977 uint64_t pend_post
, pend_nonpost
;
1979 qemu_savevm_state_pending(s
->to_dst_file
, threshold_size
,
1980 &pend_nonpost
, &pend_post
);
1981 pending_size
= pend_nonpost
+ pend_post
;
1982 trace_migrate_pending(pending_size
, threshold_size
,
1983 pend_post
, pend_nonpost
);
1984 if (pending_size
&& pending_size
>= threshold_size
) {
1985 /* Still a significant amount to transfer */
1987 if (migrate_postcopy_ram() &&
1988 s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
&&
1989 pend_nonpost
<= threshold_size
&&
1990 atomic_read(&s
->start_postcopy
)) {
1992 if (!postcopy_start(s
, &old_vm_running
)) {
1993 current_active_state
= MIGRATION_STATUS_POSTCOPY_ACTIVE
;
1994 entered_postcopy
= true;
1999 /* Just another iteration step */
2000 qemu_savevm_state_iterate(s
->to_dst_file
, entered_postcopy
);
2002 trace_migration_thread_low_pending(pending_size
);
2003 migration_completion(s
, current_active_state
,
2004 &old_vm_running
, &start_time
);
2009 if (qemu_file_get_error(s
->to_dst_file
)) {
2010 migrate_set_state(&s
->state
, current_active_state
,
2011 MIGRATION_STATUS_FAILED
);
2012 trace_migration_thread_file_err();
2015 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2016 if (current_time
>= initial_time
+ BUFFER_DELAY
) {
2017 uint64_t transferred_bytes
= qemu_ftell(s
->to_dst_file
) -
2019 uint64_t time_spent
= current_time
- initial_time
;
2020 double bandwidth
= (double)transferred_bytes
/ time_spent
;
2021 threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
2023 s
->mbps
= (((double) transferred_bytes
* 8.0) /
2024 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2026 trace_migrate_transferred(transferred_bytes
, time_spent
,
2027 bandwidth
, threshold_size
);
2028 /* if we haven't sent anything, we don't want to recalculate
2029 10000 is a small enough number for our purposes */
2030 if (ram_dirty_pages_rate() && transferred_bytes
> 10000) {
2031 s
->expected_downtime
= ram_dirty_pages_rate() *
2032 qemu_target_page_size() / bandwidth
;
2035 qemu_file_reset_rate_limit(s
->to_dst_file
);
2036 initial_time
= current_time
;
2037 initial_bytes
= qemu_ftell(s
->to_dst_file
);
2039 if (qemu_file_rate_limit(s
->to_dst_file
)) {
2040 /* usleep expects microseconds */
2041 g_usleep((initial_time
+ BUFFER_DELAY
- current_time
)*1000);
2045 trace_migration_thread_after_loop();
2046 /* If we enabled cpu throttling for auto-converge, turn it off. */
2047 cpu_throttle_stop();
2048 end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2050 qemu_mutex_lock_iothread();
2052 * The resource has been allocated by migration will be reused in COLO
2053 * process, so don't release them.
2056 qemu_savevm_state_cleanup();
2058 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
2059 uint64_t transferred_bytes
= qemu_ftell(s
->to_dst_file
);
2060 s
->total_time
= end_time
- s
->total_time
;
2061 if (!entered_postcopy
) {
2062 s
->downtime
= end_time
- start_time
;
2064 if (s
->total_time
) {
2065 s
->mbps
= (((double) transferred_bytes
* 8.0) /
2066 ((double) s
->total_time
)) / 1000;
2068 runstate_set(RUN_STATE_POSTMIGRATE
);
2070 if (s
->state
== MIGRATION_STATUS_ACTIVE
&& enable_colo
) {
2071 migrate_start_colo_process(s
);
2072 qemu_savevm_state_cleanup();
2074 * Fixme: we will run VM in COLO no matter its old running state.
2075 * After exited COLO, we will keep running.
2077 old_vm_running
= true;
2079 if (old_vm_running
&& !entered_postcopy
) {
2082 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2083 runstate_set(RUN_STATE_POSTMIGRATE
);
2087 qemu_bh_schedule(s
->cleanup_bh
);
2088 qemu_mutex_unlock_iothread();
2090 rcu_unregister_thread();
2094 void migrate_fd_connect(MigrationState
*s
)
2096 s
->expected_downtime
= s
->parameters
.downtime_limit
;
2097 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
2099 qemu_file_set_blocking(s
->to_dst_file
, true);
2100 qemu_file_set_rate_limit(s
->to_dst_file
,
2101 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
2103 /* Notify before starting migration thread */
2104 notifier_list_notify(&migration_state_notifiers
, s
);
2107 * Open the return path; currently for postcopy but other things might
2110 if (migrate_postcopy_ram()) {
2111 if (open_return_path_on_source(s
)) {
2112 error_report("Unable to open return-path for postcopy");
2113 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2114 MIGRATION_STATUS_FAILED
);
2115 migrate_fd_cleanup(s
);
2120 migrate_compress_threads_create();
2121 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
2122 QEMU_THREAD_JOINABLE
);
2123 s
->migration_thread_running
= true;