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 "migration/blocker.h"
25 #include "migration/global_state.h"
26 #include "migration/misc.h"
27 #include "migration.h"
29 #include "qemu-file-channel.h"
30 #include "qemu-file.h"
31 #include "migration/vmstate.h"
32 #include "block/block.h"
33 #include "qapi/qmp/qerror.h"
34 #include "qapi/util.h"
37 #include "postcopy-ram.h"
38 #include "qemu/thread.h"
39 #include "qmp-commands.h"
41 #include "qapi-event.h"
42 #include "exec/target_page.h"
43 #include "io/channel-buffer.h"
44 #include "migration/colo.h"
45 #include "hw/boards.h"
46 #include "monitor/monitor.h"
48 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
50 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
52 #define BUFFER_DELAY 100
53 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
55 /* Time in milliseconds we are allowed to stop the source,
56 * for sending the last part */
57 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
59 /* Maximum migrate downtime set to 2000 seconds */
60 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
61 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
63 /* Default compression thread count */
64 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
65 /* Default decompression thread count, usually decompression is at
66 * least 4 times as fast as compression.*/
67 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
68 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
69 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
70 /* Define default autoconverge cpu throttle migration parameters */
71 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
72 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
74 /* Migration XBZRLE default cache size */
75 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
77 /* The delay time (in ms) between two COLO checkpoints
78 * Note: Please change this default value to 10000 when we support hybrid mode.
80 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
82 static NotifierList migration_state_notifiers
=
83 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
85 static bool deferred_incoming
;
87 /* Messages sent on the return path from destination to source */
88 enum mig_rp_message_type
{
89 MIG_RP_MSG_INVALID
= 0, /* Must be 0 */
90 MIG_RP_MSG_SHUT
, /* sibling will not send any more RP messages */
91 MIG_RP_MSG_PONG
, /* Response to a PING; data (seq: be32 ) */
93 MIG_RP_MSG_REQ_PAGES_ID
, /* data (start: be64, len: be32, id: string) */
94 MIG_RP_MSG_REQ_PAGES
, /* data (start: be64, len: be32) */
99 /* When we add fault tolerance, we could have several
100 migrations at once. For now we don't need to add
101 dynamic creation of migration */
103 static MigrationState
*current_migration
;
105 void migration_object_init(void)
107 MachineState
*ms
= MACHINE(qdev_get_machine());
109 /* This can only be called once. */
110 assert(!current_migration
);
111 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
114 * We cannot really do this in migration_instance_init() since at
115 * that time global properties are not yet applied, then this
116 * value will be definitely replaced by something else.
118 if (ms
->enforce_config_section
) {
119 current_migration
->send_configuration
= true;
124 MigrationState
*migrate_get_current(void)
126 /* This can only be called after the object created. */
127 assert(current_migration
);
128 return current_migration
;
131 MigrationIncomingState
*migration_incoming_get_current(void)
134 static MigrationIncomingState mis_current
;
137 mis_current
.state
= MIGRATION_STATUS_NONE
;
138 memset(&mis_current
, 0, sizeof(MigrationIncomingState
));
139 qemu_mutex_init(&mis_current
.rp_mutex
);
140 qemu_event_init(&mis_current
.main_thread_load_event
, false);
146 void migration_incoming_state_destroy(void)
148 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
150 if (mis
->to_src_file
) {
151 /* Tell source that we are done */
152 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
153 qemu_fclose(mis
->to_src_file
);
154 mis
->to_src_file
= NULL
;
157 if (mis
->from_src_file
) {
158 qemu_fclose(mis
->from_src_file
);
159 mis
->from_src_file
= NULL
;
162 qemu_event_destroy(&mis
->main_thread_load_event
);
165 static void migrate_generate_event(int new_state
)
167 if (migrate_use_events()) {
168 qapi_event_send_migration(new_state
, &error_abort
);
173 * Called on -incoming with a defer: uri.
174 * The migration can be started later after any parameters have been
177 static void deferred_incoming_migration(Error
**errp
)
179 if (deferred_incoming
) {
180 error_setg(errp
, "Incoming migration already deferred");
182 deferred_incoming
= true;
186 * Send a message on the return channel back to the source
189 static void migrate_send_rp_message(MigrationIncomingState
*mis
,
190 enum mig_rp_message_type message_type
,
191 uint16_t len
, void *data
)
193 trace_migrate_send_rp_message((int)message_type
, len
);
194 qemu_mutex_lock(&mis
->rp_mutex
);
195 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
196 qemu_put_be16(mis
->to_src_file
, len
);
197 qemu_put_buffer(mis
->to_src_file
, data
, len
);
198 qemu_fflush(mis
->to_src_file
);
199 qemu_mutex_unlock(&mis
->rp_mutex
);
202 /* Request a range of pages from the source VM at the given
204 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
205 * as the last request (a name must have been given previously)
206 * Start: Address offset within the RB
207 * Len: Length in bytes required - must be a multiple of pagesize
209 void migrate_send_rp_req_pages(MigrationIncomingState
*mis
, const char *rbname
,
210 ram_addr_t start
, size_t len
)
212 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
213 size_t msglen
= 12; /* start + len */
215 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
216 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
219 int rbname_len
= strlen(rbname
);
220 assert(rbname_len
< 256);
222 bufc
[msglen
++] = rbname_len
;
223 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
224 msglen
+= rbname_len
;
225 migrate_send_rp_message(mis
, MIG_RP_MSG_REQ_PAGES_ID
, msglen
, bufc
);
227 migrate_send_rp_message(mis
, MIG_RP_MSG_REQ_PAGES
, msglen
, bufc
);
231 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
235 qapi_event_send_migration(MIGRATION_STATUS_SETUP
, &error_abort
);
236 if (!strcmp(uri
, "defer")) {
237 deferred_incoming_migration(errp
);
238 } else if (strstart(uri
, "tcp:", &p
)) {
239 tcp_start_incoming_migration(p
, errp
);
241 } else if (strstart(uri
, "rdma:", &p
)) {
242 rdma_start_incoming_migration(p
, errp
);
244 } else if (strstart(uri
, "exec:", &p
)) {
245 exec_start_incoming_migration(p
, errp
);
246 } else if (strstart(uri
, "unix:", &p
)) {
247 unix_start_incoming_migration(p
, errp
);
248 } else if (strstart(uri
, "fd:", &p
)) {
249 fd_start_incoming_migration(p
, errp
);
251 error_setg(errp
, "unknown migration protocol: %s", uri
);
255 static void process_incoming_migration_bh(void *opaque
)
257 Error
*local_err
= NULL
;
258 MigrationIncomingState
*mis
= opaque
;
260 /* Make sure all file formats flush their mutable metadata.
261 * If we get an error here, just don't restart the VM yet. */
262 bdrv_invalidate_cache_all(&local_err
);
264 error_report_err(local_err
);
270 * This must happen after all error conditions are dealt with and
271 * we're sure the VM is going to be running on this host.
273 qemu_announce_self();
275 /* If global state section was not received or we are in running
276 state, we need to obey autostart. Any other state is set with
279 if (!global_state_received() ||
280 global_state_get_runstate() == RUN_STATE_RUNNING
) {
284 runstate_set(RUN_STATE_PAUSED
);
287 runstate_set(global_state_get_runstate());
289 migrate_decompress_threads_join();
291 * This must happen after any state changes since as soon as an external
292 * observer sees this event they might start to prod at the VM assuming
295 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
296 MIGRATION_STATUS_COMPLETED
);
297 qemu_bh_delete(mis
->bh
);
298 migration_incoming_state_destroy();
301 static void process_incoming_migration_co(void *opaque
)
303 QEMUFile
*f
= opaque
;
304 MigrationIncomingState
*mis
= migration_incoming_get_current();
308 mis
->from_src_file
= f
;
309 mis
->largest_page_size
= qemu_ram_pagesize_largest();
310 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
311 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
312 MIGRATION_STATUS_ACTIVE
);
313 ret
= qemu_loadvm_state(f
);
315 ps
= postcopy_state_get();
316 trace_process_incoming_migration_co_end(ret
, ps
);
317 if (ps
!= POSTCOPY_INCOMING_NONE
) {
318 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
320 * Where a migration had postcopy enabled (and thus went to advise)
321 * but managed to complete within the precopy period, we can use
324 postcopy_ram_incoming_cleanup(mis
);
325 } else if (ret
>= 0) {
327 * Postcopy was started, cleanup should happen at the end of the
330 trace_process_incoming_migration_co_postcopy_end_main();
333 /* Else if something went wrong then just fall out of the normal exit */
336 /* we get COLO info, and know if we are in COLO mode */
337 if (!ret
&& migration_incoming_enable_colo()) {
338 mis
->migration_incoming_co
= qemu_coroutine_self();
339 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
340 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
341 mis
->have_colo_incoming_thread
= true;
342 qemu_coroutine_yield();
344 /* Wait checkpoint incoming thread exit before free resource */
345 qemu_thread_join(&mis
->colo_incoming_thread
);
349 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
350 MIGRATION_STATUS_FAILED
);
351 error_report("load of migration failed: %s", strerror(-ret
));
352 migrate_decompress_threads_join();
356 free_xbzrle_decoded_buf();
358 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
359 qemu_bh_schedule(mis
->bh
);
362 void migration_fd_process_incoming(QEMUFile
*f
)
364 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, f
);
366 migrate_decompress_threads_create();
367 qemu_file_set_blocking(f
, false);
368 qemu_coroutine_enter(co
);
372 * Send a 'SHUT' message on the return channel with the given value
373 * to indicate that we've finished with the RP. Non-0 value indicates
376 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
381 buf
= cpu_to_be32(value
);
382 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
386 * Send a 'PONG' message on the return channel with the given value
387 * (normally in response to a 'PING')
389 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
394 buf
= cpu_to_be32(value
);
395 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
398 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
400 MigrationCapabilityStatusList
*head
= NULL
;
401 MigrationCapabilityStatusList
*caps
;
402 MigrationState
*s
= migrate_get_current();
405 caps
= NULL
; /* silence compiler warning */
406 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
407 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
408 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
412 if (i
== MIGRATION_CAPABILITY_X_COLO
&& !colo_supported()) {
416 head
= g_malloc0(sizeof(*caps
));
419 caps
->next
= g_malloc0(sizeof(*caps
));
423 g_malloc(sizeof(*caps
->value
));
424 caps
->value
->capability
= i
;
425 caps
->value
->state
= s
->enabled_capabilities
[i
];
431 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
433 MigrationParameters
*params
;
434 MigrationState
*s
= migrate_get_current();
436 params
= g_malloc0(sizeof(*params
));
437 params
->has_compress_level
= true;
438 params
->compress_level
= s
->parameters
.compress_level
;
439 params
->has_compress_threads
= true;
440 params
->compress_threads
= s
->parameters
.compress_threads
;
441 params
->has_decompress_threads
= true;
442 params
->decompress_threads
= s
->parameters
.decompress_threads
;
443 params
->has_cpu_throttle_initial
= true;
444 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
445 params
->has_cpu_throttle_increment
= true;
446 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
447 params
->has_tls_creds
= !!s
->parameters
.tls_creds
;
448 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
449 params
->has_tls_hostname
= !!s
->parameters
.tls_hostname
;
450 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
451 params
->has_max_bandwidth
= true;
452 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
453 params
->has_downtime_limit
= true;
454 params
->downtime_limit
= s
->parameters
.downtime_limit
;
455 params
->has_x_checkpoint_delay
= true;
456 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
457 params
->has_block_incremental
= true;
458 params
->block_incremental
= s
->parameters
.block_incremental
;
464 * Return true if we're already in the middle of a migration
465 * (i.e. any of the active or setup states)
467 static bool migration_is_setup_or_active(int state
)
470 case MIGRATION_STATUS_ACTIVE
:
471 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
472 case MIGRATION_STATUS_SETUP
:
481 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
483 info
->has_ram
= true;
484 info
->ram
= g_malloc0(sizeof(*info
->ram
));
485 info
->ram
->transferred
= ram_counters
.transferred
;
486 info
->ram
->total
= ram_bytes_total();
487 info
->ram
->duplicate
= ram_counters
.duplicate
;
488 /* legacy value. It is not used anymore */
489 info
->ram
->skipped
= 0;
490 info
->ram
->normal
= ram_counters
.normal
;
491 info
->ram
->normal_bytes
= ram_counters
.normal
*
492 qemu_target_page_size();
493 info
->ram
->mbps
= s
->mbps
;
494 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
495 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
496 info
->ram
->page_size
= qemu_target_page_size();
498 if (migrate_use_xbzrle()) {
499 info
->has_xbzrle_cache
= true;
500 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
501 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
502 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
503 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
504 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
505 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
506 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
509 if (cpu_throttle_active()) {
510 info
->has_cpu_throttle_percentage
= true;
511 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
514 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
515 info
->ram
->remaining
= ram_bytes_remaining();
516 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
520 static void populate_disk_info(MigrationInfo
*info
)
522 if (blk_mig_active()) {
523 info
->has_disk
= true;
524 info
->disk
= g_malloc0(sizeof(*info
->disk
));
525 info
->disk
->transferred
= blk_mig_bytes_transferred();
526 info
->disk
->remaining
= blk_mig_bytes_remaining();
527 info
->disk
->total
= blk_mig_bytes_total();
531 MigrationInfo
*qmp_query_migrate(Error
**errp
)
533 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
534 MigrationState
*s
= migrate_get_current();
537 case MIGRATION_STATUS_NONE
:
538 /* no migration has happened ever */
540 case MIGRATION_STATUS_SETUP
:
541 info
->has_status
= true;
542 info
->has_total_time
= false;
544 case MIGRATION_STATUS_ACTIVE
:
545 case MIGRATION_STATUS_CANCELLING
:
546 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
547 /* TODO add some postcopy stats */
548 info
->has_status
= true;
549 info
->has_total_time
= true;
550 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
552 info
->has_expected_downtime
= true;
553 info
->expected_downtime
= s
->expected_downtime
;
554 info
->has_setup_time
= true;
555 info
->setup_time
= s
->setup_time
;
557 populate_ram_info(info
, s
);
558 populate_disk_info(info
);
560 case MIGRATION_STATUS_COLO
:
561 info
->has_status
= true;
562 /* TODO: display COLO specific information (checkpoint info etc.) */
564 case MIGRATION_STATUS_COMPLETED
:
565 info
->has_status
= true;
566 info
->has_total_time
= true;
567 info
->total_time
= s
->total_time
;
568 info
->has_downtime
= true;
569 info
->downtime
= s
->downtime
;
570 info
->has_setup_time
= true;
571 info
->setup_time
= s
->setup_time
;
573 populate_ram_info(info
, s
);
575 case MIGRATION_STATUS_FAILED
:
576 info
->has_status
= true;
578 info
->has_error_desc
= true;
579 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
582 case MIGRATION_STATUS_CANCELLED
:
583 info
->has_status
= true;
586 info
->status
= s
->state
;
591 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
594 MigrationState
*s
= migrate_get_current();
595 MigrationCapabilityStatusList
*cap
;
596 bool old_postcopy_cap
= migrate_postcopy_ram();
598 if (migration_is_setup_or_active(s
->state
)) {
599 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
603 for (cap
= params
; cap
; cap
= cap
->next
) {
604 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
605 if (cap
->value
->capability
== MIGRATION_CAPABILITY_BLOCK
606 && cap
->value
->state
) {
607 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
609 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
613 if (cap
->value
->capability
== MIGRATION_CAPABILITY_X_COLO
) {
614 if (!colo_supported()) {
615 error_setg(errp
, "COLO is not currently supported, please"
616 " configure with --enable-colo option in order to"
617 " support COLO feature");
621 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
624 if (migrate_postcopy_ram()) {
625 if (migrate_use_compression()) {
626 /* The decompression threads asynchronously write into RAM
627 * rather than use the atomic copies needed to avoid
628 * userfaulting. It should be possible to fix the decompression
629 * threads for compatibility in future.
631 error_report("Postcopy is not currently compatible with "
633 s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
] =
636 /* This check is reasonably expensive, so only when it's being
637 * set the first time, also it's only the destination that needs
640 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
641 !postcopy_ram_supported_by_host()) {
642 /* postcopy_ram_supported_by_host will have emitted a more
645 error_report("Postcopy is not supported");
646 s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
] =
652 void qmp_migrate_set_parameters(MigrationParameters
*params
, Error
**errp
)
654 MigrationState
*s
= migrate_get_current();
656 if (params
->has_compress_level
&&
657 (params
->compress_level
< 0 || params
->compress_level
> 9)) {
658 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
659 "is invalid, it should be in the range of 0 to 9");
662 if (params
->has_compress_threads
&&
663 (params
->compress_threads
< 1 || params
->compress_threads
> 255)) {
664 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
666 "is invalid, it should be in the range of 1 to 255");
669 if (params
->has_decompress_threads
&&
670 (params
->decompress_threads
< 1 || params
->decompress_threads
> 255)) {
671 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
672 "decompress_threads",
673 "is invalid, it should be in the range of 1 to 255");
676 if (params
->has_cpu_throttle_initial
&&
677 (params
->cpu_throttle_initial
< 1 ||
678 params
->cpu_throttle_initial
> 99)) {
679 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
680 "cpu_throttle_initial",
681 "an integer in the range of 1 to 99");
684 if (params
->has_cpu_throttle_increment
&&
685 (params
->cpu_throttle_increment
< 1 ||
686 params
->cpu_throttle_increment
> 99)) {
687 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
688 "cpu_throttle_increment",
689 "an integer in the range of 1 to 99");
692 if (params
->has_max_bandwidth
&&
693 (params
->max_bandwidth
< 0 || params
->max_bandwidth
> SIZE_MAX
)) {
694 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
695 " range of 0 to %zu bytes/second", SIZE_MAX
);
698 if (params
->has_downtime_limit
&&
699 (params
->downtime_limit
< 0 ||
700 params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
701 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
702 "the range of 0 to %d milliseconds",
703 MAX_MIGRATE_DOWNTIME
);
706 if (params
->has_x_checkpoint_delay
&& (params
->x_checkpoint_delay
< 0)) {
707 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
708 "x_checkpoint_delay",
709 "is invalid, it should be positive");
712 if (params
->has_compress_level
) {
713 s
->parameters
.compress_level
= params
->compress_level
;
715 if (params
->has_compress_threads
) {
716 s
->parameters
.compress_threads
= params
->compress_threads
;
718 if (params
->has_decompress_threads
) {
719 s
->parameters
.decompress_threads
= params
->decompress_threads
;
721 if (params
->has_cpu_throttle_initial
) {
722 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
724 if (params
->has_cpu_throttle_increment
) {
725 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
727 if (params
->has_tls_creds
) {
728 g_free(s
->parameters
.tls_creds
);
729 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
);
731 if (params
->has_tls_hostname
) {
732 g_free(s
->parameters
.tls_hostname
);
733 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
);
735 if (params
->has_max_bandwidth
) {
736 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
737 if (s
->to_dst_file
) {
738 qemu_file_set_rate_limit(s
->to_dst_file
,
739 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
742 if (params
->has_downtime_limit
) {
743 s
->parameters
.downtime_limit
= params
->downtime_limit
;
746 if (params
->has_x_checkpoint_delay
) {
747 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
748 if (migration_in_colo_state()) {
749 colo_checkpoint_notify(s
);
752 if (params
->has_block_incremental
) {
753 s
->parameters
.block_incremental
= params
->block_incremental
;
758 void qmp_migrate_start_postcopy(Error
**errp
)
760 MigrationState
*s
= migrate_get_current();
762 if (!migrate_postcopy_ram()) {
763 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
764 " the start of migration");
768 if (s
->state
== MIGRATION_STATUS_NONE
) {
769 error_setg(errp
, "Postcopy must be started after migration has been"
774 * we don't error if migration has finished since that would be racy
775 * with issuing this command.
777 atomic_set(&s
->start_postcopy
, true);
780 /* shared migration helpers */
782 void migrate_set_state(int *state
, int old_state
, int new_state
)
784 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
785 trace_migrate_set_state(new_state
);
786 migrate_generate_event(new_state
);
790 void migrate_set_block_enabled(bool value
, Error
**errp
)
792 MigrationCapabilityStatusList
*cap
;
794 cap
= g_new0(MigrationCapabilityStatusList
, 1);
795 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
796 cap
->value
->capability
= MIGRATION_CAPABILITY_BLOCK
;
797 cap
->value
->state
= value
;
798 qmp_migrate_set_capabilities(cap
, errp
);
799 qapi_free_MigrationCapabilityStatusList(cap
);
802 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
804 s
->parameters
.block_incremental
= value
;
807 static void block_cleanup_parameters(MigrationState
*s
)
809 if (s
->must_remove_block_options
) {
810 /* setting to false can never fail */
811 migrate_set_block_enabled(false, &error_abort
);
812 migrate_set_block_incremental(s
, false);
813 s
->must_remove_block_options
= false;
817 static void migrate_fd_cleanup(void *opaque
)
819 MigrationState
*s
= opaque
;
821 qemu_bh_delete(s
->cleanup_bh
);
822 s
->cleanup_bh
= NULL
;
824 if (s
->to_dst_file
) {
825 trace_migrate_fd_cleanup();
826 qemu_mutex_unlock_iothread();
827 if (s
->migration_thread_running
) {
828 qemu_thread_join(&s
->thread
);
829 s
->migration_thread_running
= false;
831 qemu_mutex_lock_iothread();
833 migrate_compress_threads_join();
834 qemu_fclose(s
->to_dst_file
);
835 s
->to_dst_file
= NULL
;
838 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
839 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
841 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
842 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
843 MIGRATION_STATUS_CANCELLED
);
846 notifier_list_notify(&migration_state_notifiers
, s
);
847 block_cleanup_parameters(s
);
850 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
852 trace_migrate_fd_error(error_get_pretty(error
));
853 assert(s
->to_dst_file
== NULL
);
854 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
855 MIGRATION_STATUS_FAILED
);
857 s
->error
= error_copy(error
);
859 notifier_list_notify(&migration_state_notifiers
, s
);
860 block_cleanup_parameters(s
);
863 static void migrate_fd_cancel(MigrationState
*s
)
866 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
867 trace_migrate_fd_cancel();
869 if (s
->rp_state
.from_dst_file
) {
870 /* shutdown the rp socket, so causing the rp thread to shutdown */
871 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
875 old_state
= s
->state
;
876 if (!migration_is_setup_or_active(old_state
)) {
879 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
880 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
883 * If we're unlucky the migration code might be stuck somewhere in a
884 * send/write while the network has failed and is waiting to timeout;
885 * if we've got shutdown(2) available then we can force it to quit.
886 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
887 * called in a bh, so there is no race against this cancel.
889 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
890 qemu_file_shutdown(f
);
892 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
893 Error
*local_err
= NULL
;
895 bdrv_invalidate_cache_all(&local_err
);
897 error_report_err(local_err
);
899 s
->block_inactive
= false;
902 block_cleanup_parameters(s
);
905 void add_migration_state_change_notifier(Notifier
*notify
)
907 notifier_list_add(&migration_state_notifiers
, notify
);
910 void remove_migration_state_change_notifier(Notifier
*notify
)
912 notifier_remove(notify
);
915 bool migration_in_setup(MigrationState
*s
)
917 return s
->state
== MIGRATION_STATUS_SETUP
;
920 bool migration_has_finished(MigrationState
*s
)
922 return s
->state
== MIGRATION_STATUS_COMPLETED
;
925 bool migration_has_failed(MigrationState
*s
)
927 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
928 s
->state
== MIGRATION_STATUS_FAILED
);
931 bool migration_in_postcopy(void)
933 MigrationState
*s
= migrate_get_current();
935 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
938 bool migration_in_postcopy_after_devices(MigrationState
*s
)
940 return migration_in_postcopy() && s
->postcopy_after_devices
;
943 bool migration_is_idle(void)
945 MigrationState
*s
= migrate_get_current();
948 case MIGRATION_STATUS_NONE
:
949 case MIGRATION_STATUS_CANCELLED
:
950 case MIGRATION_STATUS_COMPLETED
:
951 case MIGRATION_STATUS_FAILED
:
953 case MIGRATION_STATUS_SETUP
:
954 case MIGRATION_STATUS_CANCELLING
:
955 case MIGRATION_STATUS_ACTIVE
:
956 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
957 case MIGRATION_STATUS_COLO
:
959 case MIGRATION_STATUS__MAX
:
960 g_assert_not_reached();
966 MigrationState
*migrate_init(void)
968 MigrationState
*s
= migrate_get_current();
971 * Reinitialise all migration state, except
972 * parameters/capabilities that the user set, and
978 s
->to_dst_file
= NULL
;
979 s
->state
= MIGRATION_STATUS_NONE
;
980 s
->rp_state
.from_dst_file
= NULL
;
981 s
->rp_state
.error
= false;
984 s
->expected_downtime
= 0;
986 s
->start_postcopy
= false;
987 s
->postcopy_after_devices
= false;
988 s
->migration_thread_running
= false;
989 error_free(s
->error
);
992 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
994 s
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
998 static GSList
*migration_blockers
;
1000 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1002 if (migrate_get_current()->only_migratable
) {
1003 error_propagate(errp
, error_copy(reason
));
1004 error_prepend(errp
, "disallowing migration blocker "
1005 "(--only_migratable) for: ");
1009 if (migration_is_idle()) {
1010 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1014 error_propagate(errp
, error_copy(reason
));
1015 error_prepend(errp
, "disallowing migration blocker (migration in "
1020 void migrate_del_blocker(Error
*reason
)
1022 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1025 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1027 Error
*local_err
= NULL
;
1028 static bool once
= true;
1030 if (!deferred_incoming
) {
1031 error_setg(errp
, "For use with '-incoming defer'");
1035 error_setg(errp
, "The incoming migration has already been started");
1038 qemu_start_incoming_migration(uri
, &local_err
);
1041 error_propagate(errp
, local_err
);
1048 bool migration_is_blocked(Error
**errp
)
1050 if (qemu_savevm_state_blocked(errp
)) {
1054 if (migration_blockers
) {
1055 error_propagate(errp
, error_copy(migration_blockers
->data
));
1062 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1063 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1066 Error
*local_err
= NULL
;
1067 MigrationState
*s
= migrate_get_current();
1070 if (migration_is_setup_or_active(s
->state
) ||
1071 s
->state
== MIGRATION_STATUS_CANCELLING
||
1072 s
->state
== MIGRATION_STATUS_COLO
) {
1073 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1076 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1077 error_setg(errp
, "Guest is waiting for an incoming migration");
1081 if (migration_is_blocked(errp
)) {
1085 if ((has_blk
&& blk
) || (has_inc
&& inc
)) {
1086 if (migrate_use_block() || migrate_use_block_incremental()) {
1087 error_setg(errp
, "Command options are incompatible with "
1088 "current migration capabilities");
1091 migrate_set_block_enabled(true, &local_err
);
1093 error_propagate(errp
, local_err
);
1096 s
->must_remove_block_options
= true;
1099 if (has_inc
&& inc
) {
1100 migrate_set_block_incremental(s
, true);
1105 if (strstart(uri
, "tcp:", &p
)) {
1106 tcp_start_outgoing_migration(s
, p
, &local_err
);
1108 } else if (strstart(uri
, "rdma:", &p
)) {
1109 rdma_start_outgoing_migration(s
, p
, &local_err
);
1111 } else if (strstart(uri
, "exec:", &p
)) {
1112 exec_start_outgoing_migration(s
, p
, &local_err
);
1113 } else if (strstart(uri
, "unix:", &p
)) {
1114 unix_start_outgoing_migration(s
, p
, &local_err
);
1115 } else if (strstart(uri
, "fd:", &p
)) {
1116 fd_start_outgoing_migration(s
, p
, &local_err
);
1118 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1119 "a valid migration protocol");
1120 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1121 MIGRATION_STATUS_FAILED
);
1126 migrate_fd_error(s
, local_err
);
1127 error_propagate(errp
, local_err
);
1132 void qmp_migrate_cancel(Error
**errp
)
1134 migrate_fd_cancel(migrate_get_current());
1137 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1139 MigrationState
*s
= migrate_get_current();
1142 /* Check for truncation */
1143 if (value
!= (size_t)value
) {
1144 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1145 "exceeding address space");
1149 /* Cache should not be larger than guest ram size */
1150 if (value
> ram_bytes_total()) {
1151 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1152 "exceeds guest ram size ");
1156 new_size
= xbzrle_cache_resize(value
);
1158 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
1159 "is smaller than page size");
1163 s
->xbzrle_cache_size
= new_size
;
1166 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1168 return migrate_xbzrle_cache_size();
1171 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1173 MigrationParameters p
= {
1174 .has_max_bandwidth
= true,
1175 .max_bandwidth
= value
,
1178 qmp_migrate_set_parameters(&p
, errp
);
1181 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1183 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
1184 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1185 "the range of 0 to %d seconds",
1186 MAX_MIGRATE_DOWNTIME_SECONDS
);
1190 value
*= 1000; /* Convert to milliseconds */
1191 value
= MAX(0, MIN(INT64_MAX
, value
));
1193 MigrationParameters p
= {
1194 .has_downtime_limit
= true,
1195 .downtime_limit
= value
,
1198 qmp_migrate_set_parameters(&p
, errp
);
1201 bool migrate_release_ram(void)
1205 s
= migrate_get_current();
1207 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
1210 bool migrate_postcopy_ram(void)
1214 s
= migrate_get_current();
1216 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1219 bool migrate_auto_converge(void)
1223 s
= migrate_get_current();
1225 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1228 bool migrate_zero_blocks(void)
1232 s
= migrate_get_current();
1234 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
1237 bool migrate_use_compression(void)
1241 s
= migrate_get_current();
1243 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
1246 int migrate_compress_level(void)
1250 s
= migrate_get_current();
1252 return s
->parameters
.compress_level
;
1255 int migrate_compress_threads(void)
1259 s
= migrate_get_current();
1261 return s
->parameters
.compress_threads
;
1264 int migrate_decompress_threads(void)
1268 s
= migrate_get_current();
1270 return s
->parameters
.decompress_threads
;
1273 bool migrate_use_events(void)
1277 s
= migrate_get_current();
1279 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
1282 int migrate_use_xbzrle(void)
1286 s
= migrate_get_current();
1288 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
1291 int64_t migrate_xbzrle_cache_size(void)
1295 s
= migrate_get_current();
1297 return s
->xbzrle_cache_size
;
1300 bool migrate_use_block(void)
1304 s
= migrate_get_current();
1306 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
1309 bool migrate_use_return_path(void)
1313 s
= migrate_get_current();
1315 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
1318 bool migrate_use_block_incremental(void)
1322 s
= migrate_get_current();
1324 return s
->parameters
.block_incremental
;
1327 /* migration thread support */
1329 * Something bad happened to the RP stream, mark an error
1330 * The caller shall print or trace something to indicate why
1332 static void mark_source_rp_bad(MigrationState
*s
)
1334 s
->rp_state
.error
= true;
1337 static struct rp_cmd_args
{
1338 ssize_t len
; /* -1 = variable */
1341 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1342 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1343 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1344 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1345 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1346 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1350 * Process a request for pages received on the return path,
1351 * We're allowed to send more than requested (e.g. to round to our page size)
1352 * and we don't need to send pages that have already been sent.
1354 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1355 ram_addr_t start
, size_t len
)
1357 long our_host_ps
= getpagesize();
1359 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1362 * Since we currently insist on matching page sizes, just sanity check
1363 * we're being asked for whole host pages.
1365 if (start
& (our_host_ps
-1) ||
1366 (len
& (our_host_ps
-1))) {
1367 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1368 " len: %zd", __func__
, start
, len
);
1369 mark_source_rp_bad(ms
);
1373 if (ram_save_queue_pages(rbname
, start
, len
)) {
1374 mark_source_rp_bad(ms
);
1379 * Handles messages sent on the return path towards the source VM
1382 static void *source_return_path_thread(void *opaque
)
1384 MigrationState
*ms
= opaque
;
1385 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1386 uint16_t header_len
, header_type
;
1388 uint32_t tmp32
, sibling_error
;
1389 ram_addr_t start
= 0; /* =0 to silence warning */
1390 size_t len
= 0, expected_len
;
1393 trace_source_return_path_thread_entry();
1394 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1395 migration_is_setup_or_active(ms
->state
)) {
1396 trace_source_return_path_thread_loop_top();
1397 header_type
= qemu_get_be16(rp
);
1398 header_len
= qemu_get_be16(rp
);
1400 if (header_type
>= MIG_RP_MSG_MAX
||
1401 header_type
== MIG_RP_MSG_INVALID
) {
1402 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1403 header_type
, header_len
);
1404 mark_source_rp_bad(ms
);
1408 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1409 header_len
!= rp_cmd_args
[header_type
].len
) ||
1410 header_len
> sizeof(buf
)) {
1411 error_report("RP: Received '%s' message (0x%04x) with"
1412 "incorrect length %d expecting %zu",
1413 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1414 (size_t)rp_cmd_args
[header_type
].len
);
1415 mark_source_rp_bad(ms
);
1419 /* We know we've got a valid header by this point */
1420 res
= qemu_get_buffer(rp
, buf
, header_len
);
1421 if (res
!= header_len
) {
1422 error_report("RP: Failed reading data for message 0x%04x"
1423 " read %d expected %d",
1424 header_type
, res
, header_len
);
1425 mark_source_rp_bad(ms
);
1429 /* OK, we have the message and the data */
1430 switch (header_type
) {
1431 case MIG_RP_MSG_SHUT
:
1432 sibling_error
= ldl_be_p(buf
);
1433 trace_source_return_path_thread_shut(sibling_error
);
1434 if (sibling_error
) {
1435 error_report("RP: Sibling indicated error %d", sibling_error
);
1436 mark_source_rp_bad(ms
);
1439 * We'll let the main thread deal with closing the RP
1440 * we could do a shutdown(2) on it, but we're the only user
1441 * anyway, so there's nothing gained.
1445 case MIG_RP_MSG_PONG
:
1446 tmp32
= ldl_be_p(buf
);
1447 trace_source_return_path_thread_pong(tmp32
);
1450 case MIG_RP_MSG_REQ_PAGES
:
1451 start
= ldq_be_p(buf
);
1452 len
= ldl_be_p(buf
+ 8);
1453 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1456 case MIG_RP_MSG_REQ_PAGES_ID
:
1457 expected_len
= 12 + 1; /* header + termination */
1459 if (header_len
>= expected_len
) {
1460 start
= ldq_be_p(buf
);
1461 len
= ldl_be_p(buf
+ 8);
1462 /* Now we expect an idstr */
1463 tmp32
= buf
[12]; /* Length of the following idstr */
1464 buf
[13 + tmp32
] = '\0';
1465 expected_len
+= tmp32
;
1467 if (header_len
!= expected_len
) {
1468 error_report("RP: Req_Page_id with length %d expecting %zd",
1469 header_len
, expected_len
);
1470 mark_source_rp_bad(ms
);
1473 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
1480 if (qemu_file_get_error(rp
)) {
1481 trace_source_return_path_thread_bad_end();
1482 mark_source_rp_bad(ms
);
1485 trace_source_return_path_thread_end();
1487 ms
->rp_state
.from_dst_file
= NULL
;
1492 static int open_return_path_on_source(MigrationState
*ms
)
1495 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
1496 if (!ms
->rp_state
.from_dst_file
) {
1500 trace_open_return_path_on_source();
1501 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
1502 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
1504 trace_open_return_path_on_source_continue();
1509 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1510 static int await_return_path_close_on_source(MigrationState
*ms
)
1513 * If this is a normal exit then the destination will send a SHUT and the
1514 * rp_thread will exit, however if there's an error we need to cause
1517 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
1519 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1520 * waiting for the destination.
1522 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
1523 mark_source_rp_bad(ms
);
1525 trace_await_return_path_close_on_source_joining();
1526 qemu_thread_join(&ms
->rp_state
.rp_thread
);
1527 trace_await_return_path_close_on_source_close();
1528 return ms
->rp_state
.error
;
1532 * Switch from normal iteration to postcopy
1533 * Returns non-0 on error
1535 static int postcopy_start(MigrationState
*ms
, bool *old_vm_running
)
1538 QIOChannelBuffer
*bioc
;
1540 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1541 bool restart_block
= false;
1542 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
1543 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1545 trace_postcopy_start();
1546 qemu_mutex_lock_iothread();
1547 trace_postcopy_start_set_run();
1549 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1550 *old_vm_running
= runstate_is_running();
1551 global_state_store();
1552 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1557 ret
= bdrv_inactivate_all();
1561 restart_block
= true;
1564 * Cause any non-postcopiable, but iterative devices to
1565 * send out their final data.
1567 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
1570 * in Finish migrate and with the io-lock held everything should
1571 * be quiet, but we've potentially still got dirty pages and we
1572 * need to tell the destination to throw any pages it's already received
1575 if (ram_postcopy_send_discard_bitmap(ms
)) {
1576 error_report("postcopy send discard bitmap failed");
1581 * send rest of state - note things that are doing postcopy
1582 * will notice we're in POSTCOPY_ACTIVE and not actually
1583 * wrap their state up here
1585 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
1586 /* Ping just for debugging, helps line traces up */
1587 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
1590 * While loading the device state we may trigger page transfer
1591 * requests and the fd must be free to process those, and thus
1592 * the destination must read the whole device state off the fd before
1593 * it starts processing it. Unfortunately the ad-hoc migration format
1594 * doesn't allow the destination to know the size to read without fully
1595 * parsing it through each devices load-state code (especially the open
1596 * coded devices that use get/put).
1597 * So we wrap the device state up in a package with a length at the start;
1598 * to do this we use a qemu_buf to hold the whole of the device state.
1600 bioc
= qio_channel_buffer_new(4096);
1601 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
1602 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
1603 object_unref(OBJECT(bioc
));
1606 * Make sure the receiver can get incoming pages before we send the rest
1609 qemu_savevm_send_postcopy_listen(fb
);
1611 qemu_savevm_state_complete_precopy(fb
, false, false);
1612 qemu_savevm_send_ping(fb
, 3);
1614 qemu_savevm_send_postcopy_run(fb
);
1616 /* <><> end of stuff going into the package */
1618 /* Last point of recovery; as soon as we send the package the destination
1619 * can open devices and potentially start running.
1620 * Lets just check again we've not got any errors.
1622 ret
= qemu_file_get_error(ms
->to_dst_file
);
1624 error_report("postcopy_start: Migration stream errored (pre package)");
1628 restart_block
= false;
1630 /* Now send that blob */
1631 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
1636 /* Send a notify to give a chance for anything that needs to happen
1637 * at the transition to postcopy and after the device state; in particular
1638 * spice needs to trigger a transition now
1640 ms
->postcopy_after_devices
= true;
1641 notifier_list_notify(&migration_state_notifiers
, ms
);
1643 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
1645 qemu_mutex_unlock_iothread();
1648 * Although this ping is just for debug, it could potentially be
1649 * used for getting a better measurement of downtime at the source.
1651 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
1653 if (migrate_release_ram()) {
1654 ram_postcopy_migrated_memory_release(ms
);
1657 ret
= qemu_file_get_error(ms
->to_dst_file
);
1659 error_report("postcopy_start: Migration stream errored");
1660 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1661 MIGRATION_STATUS_FAILED
);
1669 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1670 MIGRATION_STATUS_FAILED
);
1671 if (restart_block
) {
1672 /* A failure happened early enough that we know the destination hasn't
1673 * accessed block devices, so we're safe to recover.
1675 Error
*local_err
= NULL
;
1677 bdrv_invalidate_cache_all(&local_err
);
1679 error_report_err(local_err
);
1682 qemu_mutex_unlock_iothread();
1687 * migration_completion: Used by migration_thread when there's not much left.
1688 * The caller 'breaks' the loop when this returns.
1690 * @s: Current migration state
1691 * @current_active_state: The migration state we expect to be in
1692 * @*old_vm_running: Pointer to old_vm_running flag
1693 * @*start_time: Pointer to time to update
1695 static void migration_completion(MigrationState
*s
, int current_active_state
,
1696 bool *old_vm_running
,
1697 int64_t *start_time
)
1701 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1702 qemu_mutex_lock_iothread();
1703 *start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1704 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1705 *old_vm_running
= runstate_is_running();
1706 ret
= global_state_store();
1709 bool inactivate
= !migrate_colo_enabled();
1710 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1712 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
1713 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
1716 if (inactivate
&& ret
>= 0) {
1717 s
->block_inactive
= true;
1720 qemu_mutex_unlock_iothread();
1725 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1726 trace_migration_completion_postcopy_end();
1728 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
1729 trace_migration_completion_postcopy_end_after_complete();
1733 * If rp was opened we must clean up the thread before
1734 * cleaning everything else up (since if there are no failures
1735 * it will wait for the destination to send it's status in
1738 if (s
->rp_state
.from_dst_file
) {
1740 trace_migration_return_path_end_before();
1741 rp_error
= await_return_path_close_on_source(s
);
1742 trace_migration_return_path_end_after(rp_error
);
1744 goto fail_invalidate
;
1748 if (qemu_file_get_error(s
->to_dst_file
)) {
1749 trace_migration_completion_file_err();
1750 goto fail_invalidate
;
1753 if (!migrate_colo_enabled()) {
1754 migrate_set_state(&s
->state
, current_active_state
,
1755 MIGRATION_STATUS_COMPLETED
);
1761 /* If not doing postcopy, vm_start() will be called: let's regain
1762 * control on images.
1764 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
1765 Error
*local_err
= NULL
;
1767 qemu_mutex_lock_iothread();
1768 bdrv_invalidate_cache_all(&local_err
);
1770 error_report_err(local_err
);
1772 s
->block_inactive
= false;
1774 qemu_mutex_unlock_iothread();
1778 migrate_set_state(&s
->state
, current_active_state
,
1779 MIGRATION_STATUS_FAILED
);
1782 bool migrate_colo_enabled(void)
1784 MigrationState
*s
= migrate_get_current();
1785 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
1789 * Master migration thread on the source VM.
1790 * It drives the migration and pumps the data down the outgoing channel.
1792 static void *migration_thread(void *opaque
)
1794 MigrationState
*s
= opaque
;
1795 /* Used by the bandwidth calcs, updated later */
1796 int64_t initial_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1797 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
1798 int64_t initial_bytes
= 0;
1800 * The final stage happens when the remaining data is smaller than
1801 * this threshold; it's calculated from the requested downtime and
1802 * measured bandwidth
1804 int64_t threshold_size
= 0;
1805 int64_t start_time
= initial_time
;
1807 bool old_vm_running
= false;
1808 bool entered_postcopy
= false;
1809 /* The active state we expect to be in; ACTIVE or POSTCOPY_ACTIVE */
1810 enum MigrationStatus current_active_state
= MIGRATION_STATUS_ACTIVE
;
1811 bool enable_colo
= migrate_colo_enabled();
1813 rcu_register_thread();
1815 qemu_savevm_state_header(s
->to_dst_file
);
1818 * If we opened the return path, we need to make sure dst has it
1821 if (s
->rp_state
.from_dst_file
) {
1822 /* Now tell the dest that it should open its end so it can reply */
1823 qemu_savevm_send_open_return_path(s
->to_dst_file
);
1825 /* And do a ping that will make stuff easier to debug */
1826 qemu_savevm_send_ping(s
->to_dst_file
, 1);
1829 if (migrate_postcopy_ram()) {
1831 * Tell the destination that we *might* want to do postcopy later;
1832 * if the other end can't do postcopy it should fail now, nice and
1835 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
1838 qemu_savevm_state_begin(s
->to_dst_file
);
1840 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
1841 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1842 MIGRATION_STATUS_ACTIVE
);
1844 trace_migration_thread_setup_complete();
1846 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
1847 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1848 int64_t current_time
;
1849 uint64_t pending_size
;
1851 if (!qemu_file_rate_limit(s
->to_dst_file
)) {
1852 uint64_t pend_post
, pend_nonpost
;
1854 qemu_savevm_state_pending(s
->to_dst_file
, threshold_size
,
1855 &pend_nonpost
, &pend_post
);
1856 pending_size
= pend_nonpost
+ pend_post
;
1857 trace_migrate_pending(pending_size
, threshold_size
,
1858 pend_post
, pend_nonpost
);
1859 if (pending_size
&& pending_size
>= threshold_size
) {
1860 /* Still a significant amount to transfer */
1862 if (migrate_postcopy_ram() &&
1863 s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
&&
1864 pend_nonpost
<= threshold_size
&&
1865 atomic_read(&s
->start_postcopy
)) {
1867 if (!postcopy_start(s
, &old_vm_running
)) {
1868 current_active_state
= MIGRATION_STATUS_POSTCOPY_ACTIVE
;
1869 entered_postcopy
= true;
1874 /* Just another iteration step */
1875 qemu_savevm_state_iterate(s
->to_dst_file
, entered_postcopy
);
1877 trace_migration_thread_low_pending(pending_size
);
1878 migration_completion(s
, current_active_state
,
1879 &old_vm_running
, &start_time
);
1884 if (qemu_file_get_error(s
->to_dst_file
)) {
1885 migrate_set_state(&s
->state
, current_active_state
,
1886 MIGRATION_STATUS_FAILED
);
1887 trace_migration_thread_file_err();
1890 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1891 if (current_time
>= initial_time
+ BUFFER_DELAY
) {
1892 uint64_t transferred_bytes
= qemu_ftell(s
->to_dst_file
) -
1894 uint64_t time_spent
= current_time
- initial_time
;
1895 double bandwidth
= (double)transferred_bytes
/ time_spent
;
1896 threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
1898 s
->mbps
= (((double) transferred_bytes
* 8.0) /
1899 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
1901 trace_migrate_transferred(transferred_bytes
, time_spent
,
1902 bandwidth
, threshold_size
);
1903 /* if we haven't sent anything, we don't want to recalculate
1904 10000 is a small enough number for our purposes */
1905 if (ram_counters
.dirty_pages_rate
&& transferred_bytes
> 10000) {
1906 s
->expected_downtime
= ram_counters
.dirty_pages_rate
*
1907 qemu_target_page_size() / bandwidth
;
1910 qemu_file_reset_rate_limit(s
->to_dst_file
);
1911 initial_time
= current_time
;
1912 initial_bytes
= qemu_ftell(s
->to_dst_file
);
1914 if (qemu_file_rate_limit(s
->to_dst_file
)) {
1915 /* usleep expects microseconds */
1916 g_usleep((initial_time
+ BUFFER_DELAY
- current_time
)*1000);
1920 trace_migration_thread_after_loop();
1921 /* If we enabled cpu throttling for auto-converge, turn it off. */
1922 cpu_throttle_stop();
1923 end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1925 qemu_mutex_lock_iothread();
1927 * The resource has been allocated by migration will be reused in COLO
1928 * process, so don't release them.
1931 qemu_savevm_state_cleanup();
1933 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
1934 uint64_t transferred_bytes
= qemu_ftell(s
->to_dst_file
);
1935 s
->total_time
= end_time
- s
->total_time
;
1936 if (!entered_postcopy
) {
1937 s
->downtime
= end_time
- start_time
;
1939 if (s
->total_time
) {
1940 s
->mbps
= (((double) transferred_bytes
* 8.0) /
1941 ((double) s
->total_time
)) / 1000;
1943 runstate_set(RUN_STATE_POSTMIGRATE
);
1945 if (s
->state
== MIGRATION_STATUS_ACTIVE
&& enable_colo
) {
1946 migrate_start_colo_process(s
);
1947 qemu_savevm_state_cleanup();
1949 * Fixme: we will run VM in COLO no matter its old running state.
1950 * After exited COLO, we will keep running.
1952 old_vm_running
= true;
1954 if (old_vm_running
&& !entered_postcopy
) {
1957 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
1958 runstate_set(RUN_STATE_POSTMIGRATE
);
1962 qemu_bh_schedule(s
->cleanup_bh
);
1963 qemu_mutex_unlock_iothread();
1965 rcu_unregister_thread();
1969 void migrate_fd_connect(MigrationState
*s
)
1971 s
->expected_downtime
= s
->parameters
.downtime_limit
;
1972 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
1974 qemu_file_set_blocking(s
->to_dst_file
, true);
1975 qemu_file_set_rate_limit(s
->to_dst_file
,
1976 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1978 /* Notify before starting migration thread */
1979 notifier_list_notify(&migration_state_notifiers
, s
);
1982 * Open the return path. For postcopy, it is used exclusively. For
1983 * precopy, only if user specified "return-path" capability would
1984 * QEMU uses the return path.
1986 if (migrate_postcopy_ram() || migrate_use_return_path()) {
1987 if (open_return_path_on_source(s
)) {
1988 error_report("Unable to open return-path for postcopy");
1989 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1990 MIGRATION_STATUS_FAILED
);
1991 migrate_fd_cleanup(s
);
1996 migrate_compress_threads_create();
1997 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
1998 QEMU_THREAD_JOINABLE
);
1999 s
->migration_thread_running
= true;
2002 void migration_global_dump(Monitor
*mon
)
2004 MigrationState
*ms
= migrate_get_current();
2006 monitor_printf(mon
, "globals: store-global-state=%d, only_migratable=%d, "
2007 "send-configuration=%d, send-section-footer=%d\n",
2008 ms
->store_global_state
, ms
->only_migratable
,
2009 ms
->send_configuration
, ms
->send_section_footer
);
2012 static Property migration_properties
[] = {
2013 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
2014 store_global_state
, true),
2015 DEFINE_PROP_BOOL("only-migratable", MigrationState
, only_migratable
, false),
2016 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
2017 send_configuration
, true),
2018 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
2019 send_section_footer
, true),
2020 DEFINE_PROP_END_OF_LIST(),
2023 static void migration_class_init(ObjectClass
*klass
, void *data
)
2025 DeviceClass
*dc
= DEVICE_CLASS(klass
);
2027 dc
->user_creatable
= false;
2028 dc
->props
= migration_properties
;
2031 static void migration_instance_init(Object
*obj
)
2033 MigrationState
*ms
= MIGRATION_OBJ(obj
);
2035 ms
->state
= MIGRATION_STATUS_NONE
;
2036 ms
->xbzrle_cache_size
= DEFAULT_MIGRATE_CACHE_SIZE
;
2038 ms
->parameters
= (MigrationParameters
) {
2039 .compress_level
= DEFAULT_MIGRATE_COMPRESS_LEVEL
,
2040 .compress_threads
= DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
,
2041 .decompress_threads
= DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
,
2042 .cpu_throttle_initial
= DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
,
2043 .cpu_throttle_increment
= DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
,
2044 .max_bandwidth
= MAX_THROTTLE
,
2045 .downtime_limit
= DEFAULT_MIGRATE_SET_DOWNTIME
,
2046 .x_checkpoint_delay
= DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
,
2048 ms
->parameters
.tls_creds
= g_strdup("");
2049 ms
->parameters
.tls_hostname
= g_strdup("");
2052 static const TypeInfo migration_type
= {
2053 .name
= TYPE_MIGRATION
,
2055 * NOTE: "migration" itself is not really a device. We used
2056 * TYPE_DEVICE here only to leverage some existing QDev features
2057 * like "-global" properties, and HW_COMPAT_* fields (which are
2058 * finally applied as global properties as well). If one day the
2059 * global property feature can be migrated from QDev to QObject in
2060 * general, then we can switch to QObject as well.
2062 .parent
= TYPE_DEVICE
,
2063 .class_init
= migration_class_init
,
2064 .class_size
= sizeof(MigrationClass
),
2065 .instance_size
= sizeof(MigrationState
),
2066 .instance_init
= migration_instance_init
,
2069 static void register_migration_types(void)
2071 type_register_static(&migration_type
);
2074 type_init(register_migration_types
);