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/error.h"
34 #include "qapi/qapi-commands-migration.h"
35 #include "qapi/qapi-events-migration.h"
36 #include "qapi/qmp/qerror.h"
37 #include "qapi/qmp/qnull.h"
40 #include "postcopy-ram.h"
41 #include "qemu/thread.h"
43 #include "exec/target_page.h"
44 #include "io/channel-buffer.h"
45 #include "migration/colo.h"
46 #include "hw/boards.h"
47 #include "monitor/monitor.h"
49 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
51 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
53 #define BUFFER_DELAY 100
54 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
56 /* Time in milliseconds we are allowed to stop the source,
57 * for sending the last part */
58 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
60 /* Maximum migrate downtime set to 2000 seconds */
61 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
62 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
64 /* Default compression thread count */
65 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
66 /* Default decompression thread count, usually decompression is at
67 * least 4 times as fast as compression.*/
68 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
69 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
70 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
71 /* Define default autoconverge cpu throttle migration parameters */
72 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
73 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
75 /* Migration XBZRLE default cache size */
76 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
78 /* The delay time (in ms) between two COLO checkpoints
79 * Note: Please change this default value to 10000 when we support hybrid mode.
81 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
82 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
83 #define DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT 16
85 static NotifierList migration_state_notifiers
=
86 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
88 static bool deferred_incoming
;
90 /* Messages sent on the return path from destination to source */
91 enum mig_rp_message_type
{
92 MIG_RP_MSG_INVALID
= 0, /* Must be 0 */
93 MIG_RP_MSG_SHUT
, /* sibling will not send any more RP messages */
94 MIG_RP_MSG_PONG
, /* Response to a PING; data (seq: be32 ) */
96 MIG_RP_MSG_REQ_PAGES_ID
, /* data (start: be64, len: be32, id: string) */
97 MIG_RP_MSG_REQ_PAGES
, /* data (start: be64, len: be32) */
102 /* When we add fault tolerance, we could have several
103 migrations at once. For now we don't need to add
104 dynamic creation of migration */
106 static MigrationState
*current_migration
;
108 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
109 static int migration_maybe_pause(MigrationState
*s
,
110 int *current_active_state
,
113 void migration_object_init(void)
115 MachineState
*ms
= MACHINE(qdev_get_machine());
118 /* This can only be called once. */
119 assert(!current_migration
);
120 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
122 if (!migration_object_check(current_migration
, &err
)) {
123 error_report_err(err
);
128 * We cannot really do this in migration_instance_init() since at
129 * that time global properties are not yet applied, then this
130 * value will be definitely replaced by something else.
132 if (ms
->enforce_config_section
) {
133 current_migration
->send_configuration
= true;
137 void migration_object_finalize(void)
139 object_unref(OBJECT(current_migration
));
143 MigrationState
*migrate_get_current(void)
145 /* This can only be called after the object created. */
146 assert(current_migration
);
147 return current_migration
;
150 MigrationIncomingState
*migration_incoming_get_current(void)
153 static MigrationIncomingState mis_current
;
156 mis_current
.state
= MIGRATION_STATUS_NONE
;
157 memset(&mis_current
, 0, sizeof(MigrationIncomingState
));
158 qemu_mutex_init(&mis_current
.rp_mutex
);
159 qemu_event_init(&mis_current
.main_thread_load_event
, false);
161 init_dirty_bitmap_incoming_migration();
168 void migration_incoming_state_destroy(void)
170 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
172 if (mis
->to_src_file
) {
173 /* Tell source that we are done */
174 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
175 qemu_fclose(mis
->to_src_file
);
176 mis
->to_src_file
= NULL
;
179 if (mis
->from_src_file
) {
180 qemu_fclose(mis
->from_src_file
);
181 mis
->from_src_file
= NULL
;
184 qemu_event_reset(&mis
->main_thread_load_event
);
187 static void migrate_generate_event(int new_state
)
189 if (migrate_use_events()) {
190 qapi_event_send_migration(new_state
, &error_abort
);
195 * Called on -incoming with a defer: uri.
196 * The migration can be started later after any parameters have been
199 static void deferred_incoming_migration(Error
**errp
)
201 if (deferred_incoming
) {
202 error_setg(errp
, "Incoming migration already deferred");
204 deferred_incoming
= true;
208 * Send a message on the return channel back to the source
211 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
212 enum mig_rp_message_type message_type
,
213 uint16_t len
, void *data
)
217 trace_migrate_send_rp_message((int)message_type
, len
);
218 qemu_mutex_lock(&mis
->rp_mutex
);
221 * It's possible that the file handle got lost due to network
224 if (!mis
->to_src_file
) {
229 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
230 qemu_put_be16(mis
->to_src_file
, len
);
231 qemu_put_buffer(mis
->to_src_file
, data
, len
);
232 qemu_fflush(mis
->to_src_file
);
234 /* It's possible that qemu file got error during sending */
235 ret
= qemu_file_get_error(mis
->to_src_file
);
238 qemu_mutex_unlock(&mis
->rp_mutex
);
242 /* Request a range of pages from the source VM at the given
244 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
245 * as the last request (a name must have been given previously)
246 * Start: Address offset within the RB
247 * Len: Length in bytes required - must be a multiple of pagesize
249 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
, const char *rbname
,
250 ram_addr_t start
, size_t len
)
252 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
253 size_t msglen
= 12; /* start + len */
254 enum mig_rp_message_type msg_type
;
256 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
257 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
260 int rbname_len
= strlen(rbname
);
261 assert(rbname_len
< 256);
263 bufc
[msglen
++] = rbname_len
;
264 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
265 msglen
+= rbname_len
;
266 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
268 msg_type
= MIG_RP_MSG_REQ_PAGES
;
271 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
274 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
278 qapi_event_send_migration(MIGRATION_STATUS_SETUP
, &error_abort
);
279 if (!strcmp(uri
, "defer")) {
280 deferred_incoming_migration(errp
);
281 } else if (strstart(uri
, "tcp:", &p
)) {
282 tcp_start_incoming_migration(p
, errp
);
284 } else if (strstart(uri
, "rdma:", &p
)) {
285 rdma_start_incoming_migration(p
, errp
);
287 } else if (strstart(uri
, "exec:", &p
)) {
288 exec_start_incoming_migration(p
, errp
);
289 } else if (strstart(uri
, "unix:", &p
)) {
290 unix_start_incoming_migration(p
, errp
);
291 } else if (strstart(uri
, "fd:", &p
)) {
292 fd_start_incoming_migration(p
, errp
);
294 error_setg(errp
, "unknown migration protocol: %s", uri
);
298 static void process_incoming_migration_bh(void *opaque
)
300 Error
*local_err
= NULL
;
301 MigrationIncomingState
*mis
= opaque
;
303 /* Make sure all file formats flush their mutable metadata.
304 * If we get an error here, just don't restart the VM yet. */
305 bdrv_invalidate_cache_all(&local_err
);
307 error_report_err(local_err
);
313 * This must happen after all error conditions are dealt with and
314 * we're sure the VM is going to be running on this host.
316 qemu_announce_self();
318 if (multifd_load_cleanup(&local_err
) != 0) {
319 error_report_err(local_err
);
322 /* If global state section was not received or we are in running
323 state, we need to obey autostart. Any other state is set with
326 dirty_bitmap_mig_before_vm_start();
328 if (!global_state_received() ||
329 global_state_get_runstate() == RUN_STATE_RUNNING
) {
333 runstate_set(RUN_STATE_PAUSED
);
336 runstate_set(global_state_get_runstate());
339 * This must happen after any state changes since as soon as an external
340 * observer sees this event they might start to prod at the VM assuming
343 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
344 MIGRATION_STATUS_COMPLETED
);
345 qemu_bh_delete(mis
->bh
);
346 migration_incoming_state_destroy();
349 static void process_incoming_migration_co(void *opaque
)
351 MigrationIncomingState
*mis
= migration_incoming_get_current();
355 assert(mis
->from_src_file
);
356 mis
->largest_page_size
= qemu_ram_pagesize_largest();
357 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
358 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
359 MIGRATION_STATUS_ACTIVE
);
360 ret
= qemu_loadvm_state(mis
->from_src_file
);
362 ps
= postcopy_state_get();
363 trace_process_incoming_migration_co_end(ret
, ps
);
364 if (ps
!= POSTCOPY_INCOMING_NONE
) {
365 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
367 * Where a migration had postcopy enabled (and thus went to advise)
368 * but managed to complete within the precopy period, we can use
371 postcopy_ram_incoming_cleanup(mis
);
372 } else if (ret
>= 0) {
374 * Postcopy was started, cleanup should happen at the end of the
377 trace_process_incoming_migration_co_postcopy_end_main();
380 /* Else if something went wrong then just fall out of the normal exit */
383 /* we get COLO info, and know if we are in COLO mode */
384 if (!ret
&& migration_incoming_enable_colo()) {
385 mis
->migration_incoming_co
= qemu_coroutine_self();
386 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
387 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
388 mis
->have_colo_incoming_thread
= true;
389 qemu_coroutine_yield();
391 /* Wait checkpoint incoming thread exit before free resource */
392 qemu_thread_join(&mis
->colo_incoming_thread
);
396 Error
*local_err
= NULL
;
398 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
399 MIGRATION_STATUS_FAILED
);
400 error_report("load of migration failed: %s", strerror(-ret
));
401 qemu_fclose(mis
->from_src_file
);
402 if (multifd_load_cleanup(&local_err
) != 0) {
403 error_report_err(local_err
);
407 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
408 qemu_bh_schedule(mis
->bh
);
411 static void migration_incoming_setup(QEMUFile
*f
)
413 MigrationIncomingState
*mis
= migration_incoming_get_current();
415 if (multifd_load_setup() != 0) {
416 /* We haven't been able to create multifd threads
417 nothing better to do */
421 if (!mis
->from_src_file
) {
422 mis
->from_src_file
= f
;
424 qemu_file_set_blocking(f
, false);
427 static void migration_incoming_process(void)
429 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
430 qemu_coroutine_enter(co
);
433 void migration_fd_process_incoming(QEMUFile
*f
)
435 migration_incoming_setup(f
);
436 migration_incoming_process();
439 void migration_ioc_process_incoming(QIOChannel
*ioc
)
441 MigrationIncomingState
*mis
= migration_incoming_get_current();
443 if (!mis
->from_src_file
) {
444 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
445 migration_fd_process_incoming(f
);
447 /* We still only have a single channel. Nothing to do here yet */
451 * @migration_has_all_channels: We have received all channels that we need
453 * Returns true when we have got connections to all the channels that
454 * we need for migration.
456 bool migration_has_all_channels(void)
462 * Send a 'SHUT' message on the return channel with the given value
463 * to indicate that we've finished with the RP. Non-0 value indicates
466 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
471 buf
= cpu_to_be32(value
);
472 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
476 * Send a 'PONG' message on the return channel with the given value
477 * (normally in response to a 'PING')
479 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
484 buf
= cpu_to_be32(value
);
485 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
488 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
490 MigrationCapabilityStatusList
*head
= NULL
;
491 MigrationCapabilityStatusList
*caps
;
492 MigrationState
*s
= migrate_get_current();
495 caps
= NULL
; /* silence compiler warning */
496 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
497 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
498 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
503 head
= g_malloc0(sizeof(*caps
));
506 caps
->next
= g_malloc0(sizeof(*caps
));
510 g_malloc(sizeof(*caps
->value
));
511 caps
->value
->capability
= i
;
512 caps
->value
->state
= s
->enabled_capabilities
[i
];
518 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
520 MigrationParameters
*params
;
521 MigrationState
*s
= migrate_get_current();
523 /* TODO use QAPI_CLONE() instead of duplicating it inline */
524 params
= g_malloc0(sizeof(*params
));
525 params
->has_compress_level
= true;
526 params
->compress_level
= s
->parameters
.compress_level
;
527 params
->has_compress_threads
= true;
528 params
->compress_threads
= s
->parameters
.compress_threads
;
529 params
->has_decompress_threads
= true;
530 params
->decompress_threads
= s
->parameters
.decompress_threads
;
531 params
->has_cpu_throttle_initial
= true;
532 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
533 params
->has_cpu_throttle_increment
= true;
534 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
535 params
->has_tls_creds
= true;
536 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
537 params
->has_tls_hostname
= true;
538 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
539 params
->has_max_bandwidth
= true;
540 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
541 params
->has_downtime_limit
= true;
542 params
->downtime_limit
= s
->parameters
.downtime_limit
;
543 params
->has_x_checkpoint_delay
= true;
544 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
545 params
->has_block_incremental
= true;
546 params
->block_incremental
= s
->parameters
.block_incremental
;
547 params
->has_x_multifd_channels
= true;
548 params
->x_multifd_channels
= s
->parameters
.x_multifd_channels
;
549 params
->has_x_multifd_page_count
= true;
550 params
->x_multifd_page_count
= s
->parameters
.x_multifd_page_count
;
551 params
->has_xbzrle_cache_size
= true;
552 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
558 * Return true if we're already in the middle of a migration
559 * (i.e. any of the active or setup states)
561 static bool migration_is_setup_or_active(int state
)
564 case MIGRATION_STATUS_ACTIVE
:
565 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
566 case MIGRATION_STATUS_SETUP
:
567 case MIGRATION_STATUS_PRE_SWITCHOVER
:
568 case MIGRATION_STATUS_DEVICE
:
577 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
579 info
->has_ram
= true;
580 info
->ram
= g_malloc0(sizeof(*info
->ram
));
581 info
->ram
->transferred
= ram_counters
.transferred
;
582 info
->ram
->total
= ram_bytes_total();
583 info
->ram
->duplicate
= ram_counters
.duplicate
;
584 /* legacy value. It is not used anymore */
585 info
->ram
->skipped
= 0;
586 info
->ram
->normal
= ram_counters
.normal
;
587 info
->ram
->normal_bytes
= ram_counters
.normal
*
588 qemu_target_page_size();
589 info
->ram
->mbps
= s
->mbps
;
590 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
591 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
592 info
->ram
->page_size
= qemu_target_page_size();
594 if (migrate_use_xbzrle()) {
595 info
->has_xbzrle_cache
= true;
596 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
597 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
598 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
599 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
600 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
601 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
602 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
605 if (cpu_throttle_active()) {
606 info
->has_cpu_throttle_percentage
= true;
607 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
610 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
611 info
->ram
->remaining
= ram_bytes_remaining();
612 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
616 static void populate_disk_info(MigrationInfo
*info
)
618 if (blk_mig_active()) {
619 info
->has_disk
= true;
620 info
->disk
= g_malloc0(sizeof(*info
->disk
));
621 info
->disk
->transferred
= blk_mig_bytes_transferred();
622 info
->disk
->remaining
= blk_mig_bytes_remaining();
623 info
->disk
->total
= blk_mig_bytes_total();
627 MigrationInfo
*qmp_query_migrate(Error
**errp
)
629 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
630 MigrationState
*s
= migrate_get_current();
633 case MIGRATION_STATUS_NONE
:
634 /* no migration has happened ever */
636 case MIGRATION_STATUS_SETUP
:
637 info
->has_status
= true;
638 info
->has_total_time
= false;
640 case MIGRATION_STATUS_ACTIVE
:
641 case MIGRATION_STATUS_CANCELLING
:
642 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
643 case MIGRATION_STATUS_PRE_SWITCHOVER
:
644 case MIGRATION_STATUS_DEVICE
:
645 /* TODO add some postcopy stats */
646 info
->has_status
= true;
647 info
->has_total_time
= true;
648 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
650 info
->has_expected_downtime
= true;
651 info
->expected_downtime
= s
->expected_downtime
;
652 info
->has_setup_time
= true;
653 info
->setup_time
= s
->setup_time
;
655 populate_ram_info(info
, s
);
656 populate_disk_info(info
);
658 case MIGRATION_STATUS_COLO
:
659 info
->has_status
= true;
660 /* TODO: display COLO specific information (checkpoint info etc.) */
662 case MIGRATION_STATUS_COMPLETED
:
663 info
->has_status
= true;
664 info
->has_total_time
= true;
665 info
->total_time
= s
->total_time
;
666 info
->has_downtime
= true;
667 info
->downtime
= s
->downtime
;
668 info
->has_setup_time
= true;
669 info
->setup_time
= s
->setup_time
;
671 populate_ram_info(info
, s
);
673 case MIGRATION_STATUS_FAILED
:
674 info
->has_status
= true;
676 info
->has_error_desc
= true;
677 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
680 case MIGRATION_STATUS_CANCELLED
:
681 info
->has_status
= true;
684 info
->status
= s
->state
;
690 * @migration_caps_check - check capability validity
692 * @cap_list: old capability list, array of bool
693 * @params: new capabilities to be applied soon
694 * @errp: set *errp if the check failed, with reason
696 * Returns true if check passed, otherwise false.
698 static bool migrate_caps_check(bool *cap_list
,
699 MigrationCapabilityStatusList
*params
,
702 MigrationCapabilityStatusList
*cap
;
703 bool old_postcopy_cap
;
704 MigrationIncomingState
*mis
= migration_incoming_get_current();
706 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
708 for (cap
= params
; cap
; cap
= cap
->next
) {
709 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
712 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
713 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
714 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
716 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
721 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
722 if (cap_list
[MIGRATION_CAPABILITY_COMPRESS
]) {
723 /* The decompression threads asynchronously write into RAM
724 * rather than use the atomic copies needed to avoid
725 * userfaulting. It should be possible to fix the decompression
726 * threads for compatibility in future.
728 error_setg(errp
, "Postcopy is not currently compatible "
733 /* This check is reasonably expensive, so only when it's being
734 * set the first time, also it's only the destination that needs
737 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
738 !postcopy_ram_supported_by_host(mis
)) {
739 /* postcopy_ram_supported_by_host will have emitted a more
742 error_setg(errp
, "Postcopy is not supported");
750 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
753 MigrationState
*s
= migrate_get_current();
754 MigrationCapabilityStatusList
*cap
;
755 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
757 if (migration_is_setup_or_active(s
->state
)) {
758 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
762 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
763 if (!migrate_caps_check(cap_list
, params
, errp
)) {
767 for (cap
= params
; cap
; cap
= cap
->next
) {
768 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
773 * Check whether the parameters are valid. Error will be put into errp
774 * (if provided). Return true if valid, otherwise false.
776 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
778 if (params
->has_compress_level
&&
779 (params
->compress_level
> 9)) {
780 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
781 "is invalid, it should be in the range of 0 to 9");
785 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
786 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
788 "is invalid, it should be in the range of 1 to 255");
792 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
793 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
794 "decompress_threads",
795 "is invalid, it should be in the range of 1 to 255");
799 if (params
->has_cpu_throttle_initial
&&
800 (params
->cpu_throttle_initial
< 1 ||
801 params
->cpu_throttle_initial
> 99)) {
802 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
803 "cpu_throttle_initial",
804 "an integer in the range of 1 to 99");
808 if (params
->has_cpu_throttle_increment
&&
809 (params
->cpu_throttle_increment
< 1 ||
810 params
->cpu_throttle_increment
> 99)) {
811 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
812 "cpu_throttle_increment",
813 "an integer in the range of 1 to 99");
817 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
818 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
819 " range of 0 to %zu bytes/second", SIZE_MAX
);
823 if (params
->has_downtime_limit
&&
824 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
825 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
826 "the range of 0 to %d milliseconds",
827 MAX_MIGRATE_DOWNTIME
);
831 /* x_checkpoint_delay is now always positive */
833 if (params
->has_x_multifd_channels
&& (params
->x_multifd_channels
< 1)) {
834 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
836 "is invalid, it should be in the range of 1 to 255");
839 if (params
->has_x_multifd_page_count
&&
840 (params
->x_multifd_page_count
< 1 ||
841 params
->x_multifd_page_count
> 10000)) {
842 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
843 "multifd_page_count",
844 "is invalid, it should be in the range of 1 to 10000");
848 if (params
->has_xbzrle_cache_size
&&
849 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
850 !is_power_of_2(params
->xbzrle_cache_size
))) {
851 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
853 "is invalid, it should be bigger than target page size"
854 " and a power of two");
861 static void migrate_params_test_apply(MigrateSetParameters
*params
,
862 MigrationParameters
*dest
)
864 *dest
= migrate_get_current()->parameters
;
866 /* TODO use QAPI_CLONE() instead of duplicating it inline */
868 if (params
->has_compress_level
) {
869 dest
->compress_level
= params
->compress_level
;
872 if (params
->has_compress_threads
) {
873 dest
->compress_threads
= params
->compress_threads
;
876 if (params
->has_decompress_threads
) {
877 dest
->decompress_threads
= params
->decompress_threads
;
880 if (params
->has_cpu_throttle_initial
) {
881 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
884 if (params
->has_cpu_throttle_increment
) {
885 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
888 if (params
->has_tls_creds
) {
889 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
890 dest
->tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
893 if (params
->has_tls_hostname
) {
894 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
895 dest
->tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
898 if (params
->has_max_bandwidth
) {
899 dest
->max_bandwidth
= params
->max_bandwidth
;
902 if (params
->has_downtime_limit
) {
903 dest
->downtime_limit
= params
->downtime_limit
;
906 if (params
->has_x_checkpoint_delay
) {
907 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
910 if (params
->has_block_incremental
) {
911 dest
->block_incremental
= params
->block_incremental
;
913 if (params
->has_x_multifd_channels
) {
914 dest
->x_multifd_channels
= params
->x_multifd_channels
;
916 if (params
->has_x_multifd_page_count
) {
917 dest
->x_multifd_page_count
= params
->x_multifd_page_count
;
919 if (params
->has_xbzrle_cache_size
) {
920 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
924 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
926 MigrationState
*s
= migrate_get_current();
928 /* TODO use QAPI_CLONE() instead of duplicating it inline */
930 if (params
->has_compress_level
) {
931 s
->parameters
.compress_level
= params
->compress_level
;
934 if (params
->has_compress_threads
) {
935 s
->parameters
.compress_threads
= params
->compress_threads
;
938 if (params
->has_decompress_threads
) {
939 s
->parameters
.decompress_threads
= params
->decompress_threads
;
942 if (params
->has_cpu_throttle_initial
) {
943 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
946 if (params
->has_cpu_throttle_increment
) {
947 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
950 if (params
->has_tls_creds
) {
951 g_free(s
->parameters
.tls_creds
);
952 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
953 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
956 if (params
->has_tls_hostname
) {
957 g_free(s
->parameters
.tls_hostname
);
958 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
959 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
962 if (params
->has_max_bandwidth
) {
963 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
964 if (s
->to_dst_file
) {
965 qemu_file_set_rate_limit(s
->to_dst_file
,
966 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
970 if (params
->has_downtime_limit
) {
971 s
->parameters
.downtime_limit
= params
->downtime_limit
;
974 if (params
->has_x_checkpoint_delay
) {
975 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
976 if (migration_in_colo_state()) {
977 colo_checkpoint_notify(s
);
981 if (params
->has_block_incremental
) {
982 s
->parameters
.block_incremental
= params
->block_incremental
;
984 if (params
->has_x_multifd_channels
) {
985 s
->parameters
.x_multifd_channels
= params
->x_multifd_channels
;
987 if (params
->has_x_multifd_page_count
) {
988 s
->parameters
.x_multifd_page_count
= params
->x_multifd_page_count
;
990 if (params
->has_xbzrle_cache_size
) {
991 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
992 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
996 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
998 MigrationParameters tmp
;
1000 /* TODO Rewrite "" to null instead */
1001 if (params
->has_tls_creds
1002 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1003 QDECREF(params
->tls_creds
->u
.n
);
1004 params
->tls_creds
->type
= QTYPE_QSTRING
;
1005 params
->tls_creds
->u
.s
= strdup("");
1007 /* TODO Rewrite "" to null instead */
1008 if (params
->has_tls_hostname
1009 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1010 QDECREF(params
->tls_hostname
->u
.n
);
1011 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1012 params
->tls_hostname
->u
.s
= strdup("");
1015 migrate_params_test_apply(params
, &tmp
);
1017 if (!migrate_params_check(&tmp
, errp
)) {
1018 /* Invalid parameter */
1022 migrate_params_apply(params
, errp
);
1026 void qmp_migrate_start_postcopy(Error
**errp
)
1028 MigrationState
*s
= migrate_get_current();
1030 if (!migrate_postcopy()) {
1031 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1032 " the start of migration");
1036 if (s
->state
== MIGRATION_STATUS_NONE
) {
1037 error_setg(errp
, "Postcopy must be started after migration has been"
1042 * we don't error if migration has finished since that would be racy
1043 * with issuing this command.
1045 atomic_set(&s
->start_postcopy
, true);
1048 /* shared migration helpers */
1050 void migrate_set_state(int *state
, int old_state
, int new_state
)
1052 assert(new_state
< MIGRATION_STATUS__MAX
);
1053 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1054 trace_migrate_set_state(MigrationStatus_str(new_state
));
1055 migrate_generate_event(new_state
);
1059 static MigrationCapabilityStatusList
*migrate_cap_add(
1060 MigrationCapabilityStatusList
*list
,
1061 MigrationCapability index
,
1064 MigrationCapabilityStatusList
*cap
;
1066 cap
= g_new0(MigrationCapabilityStatusList
, 1);
1067 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
1068 cap
->value
->capability
= index
;
1069 cap
->value
->state
= state
;
1075 void migrate_set_block_enabled(bool value
, Error
**errp
)
1077 MigrationCapabilityStatusList
*cap
;
1079 cap
= migrate_cap_add(NULL
, MIGRATION_CAPABILITY_BLOCK
, value
);
1080 qmp_migrate_set_capabilities(cap
, errp
);
1081 qapi_free_MigrationCapabilityStatusList(cap
);
1084 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1086 s
->parameters
.block_incremental
= value
;
1089 static void block_cleanup_parameters(MigrationState
*s
)
1091 if (s
->must_remove_block_options
) {
1092 /* setting to false can never fail */
1093 migrate_set_block_enabled(false, &error_abort
);
1094 migrate_set_block_incremental(s
, false);
1095 s
->must_remove_block_options
= false;
1099 static void migrate_fd_cleanup(void *opaque
)
1101 MigrationState
*s
= opaque
;
1103 qemu_bh_delete(s
->cleanup_bh
);
1104 s
->cleanup_bh
= NULL
;
1106 qemu_savevm_state_cleanup();
1108 if (s
->to_dst_file
) {
1109 Error
*local_err
= NULL
;
1111 trace_migrate_fd_cleanup();
1112 qemu_mutex_unlock_iothread();
1113 if (s
->migration_thread_running
) {
1114 qemu_thread_join(&s
->thread
);
1115 s
->migration_thread_running
= false;
1117 qemu_mutex_lock_iothread();
1119 if (multifd_save_cleanup(&local_err
) != 0) {
1120 error_report_err(local_err
);
1122 qemu_fclose(s
->to_dst_file
);
1123 s
->to_dst_file
= NULL
;
1126 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
1127 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
1129 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1130 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1131 MIGRATION_STATUS_CANCELLED
);
1135 /* It is used on info migrate. We can't free it */
1136 error_report_err(error_copy(s
->error
));
1138 notifier_list_notify(&migration_state_notifiers
, s
);
1139 block_cleanup_parameters(s
);
1142 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1144 qemu_mutex_lock(&s
->error_mutex
);
1146 s
->error
= error_copy(error
);
1148 qemu_mutex_unlock(&s
->error_mutex
);
1151 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1153 trace_migrate_fd_error(error_get_pretty(error
));
1154 assert(s
->to_dst_file
== NULL
);
1155 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1156 MIGRATION_STATUS_FAILED
);
1157 migrate_set_error(s
, error
);
1160 static void migrate_fd_cancel(MigrationState
*s
)
1163 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1164 trace_migrate_fd_cancel();
1166 if (s
->rp_state
.from_dst_file
) {
1167 /* shutdown the rp socket, so causing the rp thread to shutdown */
1168 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1172 old_state
= s
->state
;
1173 if (!migration_is_setup_or_active(old_state
)) {
1176 /* If the migration is paused, kick it out of the pause */
1177 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1178 qemu_sem_post(&s
->pause_sem
);
1180 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1181 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1184 * If we're unlucky the migration code might be stuck somewhere in a
1185 * send/write while the network has failed and is waiting to timeout;
1186 * if we've got shutdown(2) available then we can force it to quit.
1187 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1188 * called in a bh, so there is no race against this cancel.
1190 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1191 qemu_file_shutdown(f
);
1193 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1194 Error
*local_err
= NULL
;
1196 bdrv_invalidate_cache_all(&local_err
);
1198 error_report_err(local_err
);
1200 s
->block_inactive
= false;
1205 void add_migration_state_change_notifier(Notifier
*notify
)
1207 notifier_list_add(&migration_state_notifiers
, notify
);
1210 void remove_migration_state_change_notifier(Notifier
*notify
)
1212 notifier_remove(notify
);
1215 bool migration_in_setup(MigrationState
*s
)
1217 return s
->state
== MIGRATION_STATUS_SETUP
;
1220 bool migration_has_finished(MigrationState
*s
)
1222 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1225 bool migration_has_failed(MigrationState
*s
)
1227 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1228 s
->state
== MIGRATION_STATUS_FAILED
);
1231 bool migration_in_postcopy(void)
1233 MigrationState
*s
= migrate_get_current();
1235 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1238 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1240 return migration_in_postcopy() && s
->postcopy_after_devices
;
1243 bool migration_is_idle(void)
1245 MigrationState
*s
= migrate_get_current();
1248 case MIGRATION_STATUS_NONE
:
1249 case MIGRATION_STATUS_CANCELLED
:
1250 case MIGRATION_STATUS_COMPLETED
:
1251 case MIGRATION_STATUS_FAILED
:
1253 case MIGRATION_STATUS_SETUP
:
1254 case MIGRATION_STATUS_CANCELLING
:
1255 case MIGRATION_STATUS_ACTIVE
:
1256 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1257 case MIGRATION_STATUS_COLO
:
1258 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1259 case MIGRATION_STATUS_DEVICE
:
1261 case MIGRATION_STATUS__MAX
:
1262 g_assert_not_reached();
1268 void migrate_init(MigrationState
*s
)
1271 * Reinitialise all migration state, except
1272 * parameters/capabilities that the user set, and
1278 s
->to_dst_file
= NULL
;
1279 s
->state
= MIGRATION_STATUS_NONE
;
1280 s
->rp_state
.from_dst_file
= NULL
;
1281 s
->rp_state
.error
= false;
1284 s
->expected_downtime
= 0;
1286 s
->start_postcopy
= false;
1287 s
->postcopy_after_devices
= false;
1288 s
->migration_thread_running
= false;
1289 error_free(s
->error
);
1292 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1294 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1296 s
->vm_was_running
= false;
1297 s
->iteration_initial_bytes
= 0;
1298 s
->threshold_size
= 0;
1301 static GSList
*migration_blockers
;
1303 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1305 if (migrate_get_current()->only_migratable
) {
1306 error_propagate(errp
, error_copy(reason
));
1307 error_prepend(errp
, "disallowing migration blocker "
1308 "(--only_migratable) for: ");
1312 if (migration_is_idle()) {
1313 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1317 error_propagate(errp
, error_copy(reason
));
1318 error_prepend(errp
, "disallowing migration blocker (migration in "
1323 void migrate_del_blocker(Error
*reason
)
1325 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1328 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1330 Error
*local_err
= NULL
;
1331 static bool once
= true;
1333 if (!deferred_incoming
) {
1334 error_setg(errp
, "For use with '-incoming defer'");
1338 error_setg(errp
, "The incoming migration has already been started");
1341 qemu_start_incoming_migration(uri
, &local_err
);
1344 error_propagate(errp
, local_err
);
1351 bool migration_is_blocked(Error
**errp
)
1353 if (qemu_savevm_state_blocked(errp
)) {
1357 if (migration_blockers
) {
1358 error_propagate(errp
, error_copy(migration_blockers
->data
));
1365 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1366 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1369 Error
*local_err
= NULL
;
1370 MigrationState
*s
= migrate_get_current();
1373 if (migration_is_setup_or_active(s
->state
) ||
1374 s
->state
== MIGRATION_STATUS_CANCELLING
||
1375 s
->state
== MIGRATION_STATUS_COLO
) {
1376 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1379 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1380 error_setg(errp
, "Guest is waiting for an incoming migration");
1384 if (migration_is_blocked(errp
)) {
1388 if ((has_blk
&& blk
) || (has_inc
&& inc
)) {
1389 if (migrate_use_block() || migrate_use_block_incremental()) {
1390 error_setg(errp
, "Command options are incompatible with "
1391 "current migration capabilities");
1394 migrate_set_block_enabled(true, &local_err
);
1396 error_propagate(errp
, local_err
);
1399 s
->must_remove_block_options
= true;
1402 if (has_inc
&& inc
) {
1403 migrate_set_block_incremental(s
, true);
1408 if (strstart(uri
, "tcp:", &p
)) {
1409 tcp_start_outgoing_migration(s
, p
, &local_err
);
1411 } else if (strstart(uri
, "rdma:", &p
)) {
1412 rdma_start_outgoing_migration(s
, p
, &local_err
);
1414 } else if (strstart(uri
, "exec:", &p
)) {
1415 exec_start_outgoing_migration(s
, p
, &local_err
);
1416 } else if (strstart(uri
, "unix:", &p
)) {
1417 unix_start_outgoing_migration(s
, p
, &local_err
);
1418 } else if (strstart(uri
, "fd:", &p
)) {
1419 fd_start_outgoing_migration(s
, p
, &local_err
);
1421 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1422 "a valid migration protocol");
1423 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1424 MIGRATION_STATUS_FAILED
);
1429 migrate_fd_error(s
, local_err
);
1430 error_propagate(errp
, local_err
);
1435 void qmp_migrate_cancel(Error
**errp
)
1437 migrate_fd_cancel(migrate_get_current());
1440 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
1442 MigrationState
*s
= migrate_get_current();
1443 if (s
->state
!= state
) {
1444 error_setg(errp
, "Migration not in expected state: %s",
1445 MigrationStatus_str(s
->state
));
1448 qemu_sem_post(&s
->pause_sem
);
1451 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1453 MigrateSetParameters p
= {
1454 .has_xbzrle_cache_size
= true,
1455 .xbzrle_cache_size
= value
,
1458 qmp_migrate_set_parameters(&p
, errp
);
1461 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1463 return migrate_xbzrle_cache_size();
1466 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1468 MigrateSetParameters p
= {
1469 .has_max_bandwidth
= true,
1470 .max_bandwidth
= value
,
1473 qmp_migrate_set_parameters(&p
, errp
);
1476 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1478 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
1479 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1480 "the range of 0 to %d seconds",
1481 MAX_MIGRATE_DOWNTIME_SECONDS
);
1485 value
*= 1000; /* Convert to milliseconds */
1486 value
= MAX(0, MIN(INT64_MAX
, value
));
1488 MigrateSetParameters p
= {
1489 .has_downtime_limit
= true,
1490 .downtime_limit
= value
,
1493 qmp_migrate_set_parameters(&p
, errp
);
1496 bool migrate_release_ram(void)
1500 s
= migrate_get_current();
1502 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
1505 bool migrate_postcopy_ram(void)
1509 s
= migrate_get_current();
1511 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1514 bool migrate_postcopy(void)
1516 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
1519 bool migrate_auto_converge(void)
1523 s
= migrate_get_current();
1525 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1528 bool migrate_zero_blocks(void)
1532 s
= migrate_get_current();
1534 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
1537 bool migrate_use_compression(void)
1541 s
= migrate_get_current();
1543 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
1546 int migrate_compress_level(void)
1550 s
= migrate_get_current();
1552 return s
->parameters
.compress_level
;
1555 int migrate_compress_threads(void)
1559 s
= migrate_get_current();
1561 return s
->parameters
.compress_threads
;
1564 int migrate_decompress_threads(void)
1568 s
= migrate_get_current();
1570 return s
->parameters
.decompress_threads
;
1573 bool migrate_dirty_bitmaps(void)
1577 s
= migrate_get_current();
1579 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
1582 bool migrate_use_events(void)
1586 s
= migrate_get_current();
1588 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
1591 bool migrate_use_multifd(void)
1595 s
= migrate_get_current();
1597 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_MULTIFD
];
1600 bool migrate_pause_before_switchover(void)
1604 s
= migrate_get_current();
1606 return s
->enabled_capabilities
[
1607 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
1610 int migrate_multifd_channels(void)
1614 s
= migrate_get_current();
1616 return s
->parameters
.x_multifd_channels
;
1619 int migrate_multifd_page_count(void)
1623 s
= migrate_get_current();
1625 return s
->parameters
.x_multifd_page_count
;
1628 int migrate_use_xbzrle(void)
1632 s
= migrate_get_current();
1634 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
1637 int64_t migrate_xbzrle_cache_size(void)
1641 s
= migrate_get_current();
1643 return s
->parameters
.xbzrle_cache_size
;
1646 bool migrate_use_block(void)
1650 s
= migrate_get_current();
1652 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
1655 bool migrate_use_return_path(void)
1659 s
= migrate_get_current();
1661 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
1664 bool migrate_use_block_incremental(void)
1668 s
= migrate_get_current();
1670 return s
->parameters
.block_incremental
;
1673 /* migration thread support */
1675 * Something bad happened to the RP stream, mark an error
1676 * The caller shall print or trace something to indicate why
1678 static void mark_source_rp_bad(MigrationState
*s
)
1680 s
->rp_state
.error
= true;
1683 static struct rp_cmd_args
{
1684 ssize_t len
; /* -1 = variable */
1687 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1688 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1689 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1690 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1691 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1692 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1696 * Process a request for pages received on the return path,
1697 * We're allowed to send more than requested (e.g. to round to our page size)
1698 * and we don't need to send pages that have already been sent.
1700 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1701 ram_addr_t start
, size_t len
)
1703 long our_host_ps
= getpagesize();
1705 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1708 * Since we currently insist on matching page sizes, just sanity check
1709 * we're being asked for whole host pages.
1711 if (start
& (our_host_ps
-1) ||
1712 (len
& (our_host_ps
-1))) {
1713 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1714 " len: %zd", __func__
, start
, len
);
1715 mark_source_rp_bad(ms
);
1719 if (ram_save_queue_pages(rbname
, start
, len
)) {
1720 mark_source_rp_bad(ms
);
1725 * Handles messages sent on the return path towards the source VM
1728 static void *source_return_path_thread(void *opaque
)
1730 MigrationState
*ms
= opaque
;
1731 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1732 uint16_t header_len
, header_type
;
1734 uint32_t tmp32
, sibling_error
;
1735 ram_addr_t start
= 0; /* =0 to silence warning */
1736 size_t len
= 0, expected_len
;
1739 trace_source_return_path_thread_entry();
1740 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1741 migration_is_setup_or_active(ms
->state
)) {
1742 trace_source_return_path_thread_loop_top();
1743 header_type
= qemu_get_be16(rp
);
1744 header_len
= qemu_get_be16(rp
);
1746 if (qemu_file_get_error(rp
)) {
1747 mark_source_rp_bad(ms
);
1751 if (header_type
>= MIG_RP_MSG_MAX
||
1752 header_type
== MIG_RP_MSG_INVALID
) {
1753 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1754 header_type
, header_len
);
1755 mark_source_rp_bad(ms
);
1759 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1760 header_len
!= rp_cmd_args
[header_type
].len
) ||
1761 header_len
> sizeof(buf
)) {
1762 error_report("RP: Received '%s' message (0x%04x) with"
1763 "incorrect length %d expecting %zu",
1764 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1765 (size_t)rp_cmd_args
[header_type
].len
);
1766 mark_source_rp_bad(ms
);
1770 /* We know we've got a valid header by this point */
1771 res
= qemu_get_buffer(rp
, buf
, header_len
);
1772 if (res
!= header_len
) {
1773 error_report("RP: Failed reading data for message 0x%04x"
1774 " read %d expected %d",
1775 header_type
, res
, header_len
);
1776 mark_source_rp_bad(ms
);
1780 /* OK, we have the message and the data */
1781 switch (header_type
) {
1782 case MIG_RP_MSG_SHUT
:
1783 sibling_error
= ldl_be_p(buf
);
1784 trace_source_return_path_thread_shut(sibling_error
);
1785 if (sibling_error
) {
1786 error_report("RP: Sibling indicated error %d", sibling_error
);
1787 mark_source_rp_bad(ms
);
1790 * We'll let the main thread deal with closing the RP
1791 * we could do a shutdown(2) on it, but we're the only user
1792 * anyway, so there's nothing gained.
1796 case MIG_RP_MSG_PONG
:
1797 tmp32
= ldl_be_p(buf
);
1798 trace_source_return_path_thread_pong(tmp32
);
1801 case MIG_RP_MSG_REQ_PAGES
:
1802 start
= ldq_be_p(buf
);
1803 len
= ldl_be_p(buf
+ 8);
1804 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1807 case MIG_RP_MSG_REQ_PAGES_ID
:
1808 expected_len
= 12 + 1; /* header + termination */
1810 if (header_len
>= expected_len
) {
1811 start
= ldq_be_p(buf
);
1812 len
= ldl_be_p(buf
+ 8);
1813 /* Now we expect an idstr */
1814 tmp32
= buf
[12]; /* Length of the following idstr */
1815 buf
[13 + tmp32
] = '\0';
1816 expected_len
+= tmp32
;
1818 if (header_len
!= expected_len
) {
1819 error_report("RP: Req_Page_id with length %d expecting %zd",
1820 header_len
, expected_len
);
1821 mark_source_rp_bad(ms
);
1824 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
1831 if (qemu_file_get_error(rp
)) {
1832 trace_source_return_path_thread_bad_end();
1833 mark_source_rp_bad(ms
);
1836 trace_source_return_path_thread_end();
1838 ms
->rp_state
.from_dst_file
= NULL
;
1843 static int open_return_path_on_source(MigrationState
*ms
)
1846 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
1847 if (!ms
->rp_state
.from_dst_file
) {
1851 trace_open_return_path_on_source();
1852 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
1853 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
1855 trace_open_return_path_on_source_continue();
1860 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1861 static int await_return_path_close_on_source(MigrationState
*ms
)
1864 * If this is a normal exit then the destination will send a SHUT and the
1865 * rp_thread will exit, however if there's an error we need to cause
1868 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
1870 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1871 * waiting for the destination.
1873 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
1874 mark_source_rp_bad(ms
);
1876 trace_await_return_path_close_on_source_joining();
1877 qemu_thread_join(&ms
->rp_state
.rp_thread
);
1878 trace_await_return_path_close_on_source_close();
1879 return ms
->rp_state
.error
;
1883 * Switch from normal iteration to postcopy
1884 * Returns non-0 on error
1886 static int postcopy_start(MigrationState
*ms
)
1889 QIOChannelBuffer
*bioc
;
1891 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1892 bool restart_block
= false;
1893 int cur_state
= MIGRATION_STATUS_ACTIVE
;
1894 if (!migrate_pause_before_switchover()) {
1895 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
1896 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1899 trace_postcopy_start();
1900 qemu_mutex_lock_iothread();
1901 trace_postcopy_start_set_run();
1903 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1904 global_state_store();
1905 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1910 ret
= migration_maybe_pause(ms
, &cur_state
,
1911 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1916 ret
= bdrv_inactivate_all();
1920 restart_block
= true;
1923 * Cause any non-postcopiable, but iterative devices to
1924 * send out their final data.
1926 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
1929 * in Finish migrate and with the io-lock held everything should
1930 * be quiet, but we've potentially still got dirty pages and we
1931 * need to tell the destination to throw any pages it's already received
1934 if (migrate_postcopy_ram()) {
1935 if (ram_postcopy_send_discard_bitmap(ms
)) {
1936 error_report("postcopy send discard bitmap failed");
1942 * send rest of state - note things that are doing postcopy
1943 * will notice we're in POSTCOPY_ACTIVE and not actually
1944 * wrap their state up here
1946 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
1947 if (migrate_postcopy_ram()) {
1948 /* Ping just for debugging, helps line traces up */
1949 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
1953 * While loading the device state we may trigger page transfer
1954 * requests and the fd must be free to process those, and thus
1955 * the destination must read the whole device state off the fd before
1956 * it starts processing it. Unfortunately the ad-hoc migration format
1957 * doesn't allow the destination to know the size to read without fully
1958 * parsing it through each devices load-state code (especially the open
1959 * coded devices that use get/put).
1960 * So we wrap the device state up in a package with a length at the start;
1961 * to do this we use a qemu_buf to hold the whole of the device state.
1963 bioc
= qio_channel_buffer_new(4096);
1964 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
1965 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
1966 object_unref(OBJECT(bioc
));
1969 * Make sure the receiver can get incoming pages before we send the rest
1972 qemu_savevm_send_postcopy_listen(fb
);
1974 qemu_savevm_state_complete_precopy(fb
, false, false);
1975 if (migrate_postcopy_ram()) {
1976 qemu_savevm_send_ping(fb
, 3);
1979 qemu_savevm_send_postcopy_run(fb
);
1981 /* <><> end of stuff going into the package */
1983 /* Last point of recovery; as soon as we send the package the destination
1984 * can open devices and potentially start running.
1985 * Lets just check again we've not got any errors.
1987 ret
= qemu_file_get_error(ms
->to_dst_file
);
1989 error_report("postcopy_start: Migration stream errored (pre package)");
1993 restart_block
= false;
1995 /* Now send that blob */
1996 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2001 /* Send a notify to give a chance for anything that needs to happen
2002 * at the transition to postcopy and after the device state; in particular
2003 * spice needs to trigger a transition now
2005 ms
->postcopy_after_devices
= true;
2006 notifier_list_notify(&migration_state_notifiers
, ms
);
2008 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2010 qemu_mutex_unlock_iothread();
2012 if (migrate_postcopy_ram()) {
2014 * Although this ping is just for debug, it could potentially be
2015 * used for getting a better measurement of downtime at the source.
2017 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2020 if (migrate_release_ram()) {
2021 ram_postcopy_migrated_memory_release(ms
);
2024 ret
= qemu_file_get_error(ms
->to_dst_file
);
2026 error_report("postcopy_start: Migration stream errored");
2027 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2028 MIGRATION_STATUS_FAILED
);
2036 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2037 MIGRATION_STATUS_FAILED
);
2038 if (restart_block
) {
2039 /* A failure happened early enough that we know the destination hasn't
2040 * accessed block devices, so we're safe to recover.
2042 Error
*local_err
= NULL
;
2044 bdrv_invalidate_cache_all(&local_err
);
2046 error_report_err(local_err
);
2049 qemu_mutex_unlock_iothread();
2054 * migration_maybe_pause: Pause if required to by
2055 * migrate_pause_before_switchover called with the iothread locked
2056 * Returns: 0 on success
2058 static int migration_maybe_pause(MigrationState
*s
,
2059 int *current_active_state
,
2062 if (!migrate_pause_before_switchover()) {
2066 /* Since leaving this state is not atomic with posting the semaphore
2067 * it's possible that someone could have issued multiple migrate_continue
2068 * and the semaphore is incorrectly positive at this point;
2069 * the docs say it's undefined to reinit a semaphore that's already
2070 * init'd, so use timedwait to eat up any existing posts.
2072 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2073 /* This block intentionally left blank */
2076 qemu_mutex_unlock_iothread();
2077 migrate_set_state(&s
->state
, *current_active_state
,
2078 MIGRATION_STATUS_PRE_SWITCHOVER
);
2079 qemu_sem_wait(&s
->pause_sem
);
2080 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2082 *current_active_state
= new_state
;
2083 qemu_mutex_lock_iothread();
2085 return s
->state
== new_state
? 0 : -EINVAL
;
2089 * migration_completion: Used by migration_thread when there's not much left.
2090 * The caller 'breaks' the loop when this returns.
2092 * @s: Current migration state
2094 static void migration_completion(MigrationState
*s
)
2097 int current_active_state
= s
->state
;
2099 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2100 qemu_mutex_lock_iothread();
2101 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2102 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
2103 s
->vm_was_running
= runstate_is_running();
2104 ret
= global_state_store();
2107 bool inactivate
= !migrate_colo_enabled();
2108 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2110 ret
= migration_maybe_pause(s
, ¤t_active_state
,
2111 MIGRATION_STATUS_DEVICE
);
2114 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
2115 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
2118 if (inactivate
&& ret
>= 0) {
2119 s
->block_inactive
= true;
2122 qemu_mutex_unlock_iothread();
2127 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2128 trace_migration_completion_postcopy_end();
2130 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
2131 trace_migration_completion_postcopy_end_after_complete();
2135 * If rp was opened we must clean up the thread before
2136 * cleaning everything else up (since if there are no failures
2137 * it will wait for the destination to send it's status in
2140 if (s
->rp_state
.from_dst_file
) {
2142 trace_migration_return_path_end_before();
2143 rp_error
= await_return_path_close_on_source(s
);
2144 trace_migration_return_path_end_after(rp_error
);
2146 goto fail_invalidate
;
2150 if (qemu_file_get_error(s
->to_dst_file
)) {
2151 trace_migration_completion_file_err();
2152 goto fail_invalidate
;
2155 if (!migrate_colo_enabled()) {
2156 migrate_set_state(&s
->state
, current_active_state
,
2157 MIGRATION_STATUS_COMPLETED
);
2163 /* If not doing postcopy, vm_start() will be called: let's regain
2164 * control on images.
2166 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
2167 s
->state
== MIGRATION_STATUS_DEVICE
) {
2168 Error
*local_err
= NULL
;
2170 qemu_mutex_lock_iothread();
2171 bdrv_invalidate_cache_all(&local_err
);
2173 error_report_err(local_err
);
2175 s
->block_inactive
= false;
2177 qemu_mutex_unlock_iothread();
2181 migrate_set_state(&s
->state
, current_active_state
,
2182 MIGRATION_STATUS_FAILED
);
2185 bool migrate_colo_enabled(void)
2187 MigrationState
*s
= migrate_get_current();
2188 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
2191 static void migration_calculate_complete(MigrationState
*s
)
2193 uint64_t bytes
= qemu_ftell(s
->to_dst_file
);
2194 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2196 s
->total_time
= end_time
- s
->start_time
;
2199 * It's still not set, so we are precopy migration. For
2200 * postcopy, downtime is calculated during postcopy_start().
2202 s
->downtime
= end_time
- s
->downtime_start
;
2205 if (s
->total_time
) {
2206 s
->mbps
= ((double) bytes
* 8.0) / s
->total_time
/ 1000;
2210 static void migration_update_counters(MigrationState
*s
,
2211 int64_t current_time
)
2213 uint64_t transferred
, time_spent
;
2216 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
2220 transferred
= qemu_ftell(s
->to_dst_file
) - s
->iteration_initial_bytes
;
2221 time_spent
= current_time
- s
->iteration_start_time
;
2222 bandwidth
= (double)transferred
/ time_spent
;
2223 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
2225 s
->mbps
= (((double) transferred
* 8.0) /
2226 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2229 * if we haven't sent anything, we don't want to
2230 * recalculate. 10000 is a small enough number for our purposes
2232 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
2233 s
->expected_downtime
= ram_counters
.dirty_pages_rate
*
2234 qemu_target_page_size() / bandwidth
;
2237 qemu_file_reset_rate_limit(s
->to_dst_file
);
2239 s
->iteration_start_time
= current_time
;
2240 s
->iteration_initial_bytes
= qemu_ftell(s
->to_dst_file
);
2242 trace_migrate_transferred(transferred
, time_spent
,
2243 bandwidth
, s
->threshold_size
);
2246 /* Migration thread iteration status */
2248 MIG_ITERATE_RESUME
, /* Resume current iteration */
2249 MIG_ITERATE_SKIP
, /* Skip current iteration */
2250 MIG_ITERATE_BREAK
, /* Break the loop */
2254 * Return true if continue to the next iteration directly, false
2257 static MigIterateState
migration_iteration_run(MigrationState
*s
)
2259 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
2260 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
2262 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
2263 &pend_compat
, &pend_post
);
2264 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
2266 trace_migrate_pending(pending_size
, s
->threshold_size
,
2267 pend_pre
, pend_compat
, pend_post
);
2269 if (pending_size
&& pending_size
>= s
->threshold_size
) {
2270 /* Still a significant amount to transfer */
2271 if (migrate_postcopy() && !in_postcopy
&&
2272 pend_pre
<= s
->threshold_size
&&
2273 atomic_read(&s
->start_postcopy
)) {
2274 if (postcopy_start(s
)) {
2275 error_report("%s: postcopy failed to start", __func__
);
2277 return MIG_ITERATE_SKIP
;
2279 /* Just another iteration step */
2280 qemu_savevm_state_iterate(s
->to_dst_file
,
2281 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2283 trace_migration_thread_low_pending(pending_size
);
2284 migration_completion(s
);
2285 return MIG_ITERATE_BREAK
;
2288 return MIG_ITERATE_RESUME
;
2291 static void migration_iteration_finish(MigrationState
*s
)
2293 /* If we enabled cpu throttling for auto-converge, turn it off. */
2294 cpu_throttle_stop();
2296 qemu_mutex_lock_iothread();
2298 case MIGRATION_STATUS_COMPLETED
:
2299 migration_calculate_complete(s
);
2300 runstate_set(RUN_STATE_POSTMIGRATE
);
2303 case MIGRATION_STATUS_ACTIVE
:
2305 * We should really assert here, but since it's during
2306 * migration, let's try to reduce the usage of assertions.
2308 if (!migrate_colo_enabled()) {
2309 error_report("%s: critical error: calling COLO code without "
2310 "COLO enabled", __func__
);
2312 migrate_start_colo_process(s
);
2314 * Fixme: we will run VM in COLO no matter its old running state.
2315 * After exited COLO, we will keep running.
2317 s
->vm_was_running
= true;
2319 case MIGRATION_STATUS_FAILED
:
2320 case MIGRATION_STATUS_CANCELLED
:
2321 if (s
->vm_was_running
) {
2324 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2325 runstate_set(RUN_STATE_POSTMIGRATE
);
2331 /* Should not reach here, but if so, forgive the VM. */
2332 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
2335 qemu_bh_schedule(s
->cleanup_bh
);
2336 qemu_mutex_unlock_iothread();
2340 * Master migration thread on the source VM.
2341 * It drives the migration and pumps the data down the outgoing channel.
2343 static void *migration_thread(void *opaque
)
2345 MigrationState
*s
= opaque
;
2346 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
2348 rcu_register_thread();
2350 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2352 qemu_savevm_state_header(s
->to_dst_file
);
2355 * If we opened the return path, we need to make sure dst has it
2358 if (s
->rp_state
.from_dst_file
) {
2359 /* Now tell the dest that it should open its end so it can reply */
2360 qemu_savevm_send_open_return_path(s
->to_dst_file
);
2362 /* And do a ping that will make stuff easier to debug */
2363 qemu_savevm_send_ping(s
->to_dst_file
, 1);
2366 if (migrate_postcopy()) {
2368 * Tell the destination that we *might* want to do postcopy later;
2369 * if the other end can't do postcopy it should fail now, nice and
2372 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
2375 qemu_savevm_state_setup(s
->to_dst_file
);
2377 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
2378 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2379 MIGRATION_STATUS_ACTIVE
);
2381 trace_migration_thread_setup_complete();
2383 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
2384 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2385 int64_t current_time
;
2387 if (!qemu_file_rate_limit(s
->to_dst_file
)) {
2388 MigIterateState iter_state
= migration_iteration_run(s
);
2389 if (iter_state
== MIG_ITERATE_SKIP
) {
2391 } else if (iter_state
== MIG_ITERATE_BREAK
) {
2396 if (qemu_file_get_error(s
->to_dst_file
)) {
2397 if (migration_is_setup_or_active(s
->state
)) {
2398 migrate_set_state(&s
->state
, s
->state
,
2399 MIGRATION_STATUS_FAILED
);
2401 trace_migration_thread_file_err();
2405 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2407 migration_update_counters(s
, current_time
);
2409 if (qemu_file_rate_limit(s
->to_dst_file
)) {
2410 /* usleep expects microseconds */
2411 g_usleep((s
->iteration_start_time
+ BUFFER_DELAY
-
2412 current_time
) * 1000);
2416 trace_migration_thread_after_loop();
2417 migration_iteration_finish(s
);
2418 rcu_unregister_thread();
2422 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
2424 s
->expected_downtime
= s
->parameters
.downtime_limit
;
2425 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
2427 migrate_fd_error(s
, error_in
);
2428 migrate_fd_cleanup(s
);
2432 qemu_file_set_blocking(s
->to_dst_file
, true);
2433 qemu_file_set_rate_limit(s
->to_dst_file
,
2434 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
2436 /* Notify before starting migration thread */
2437 notifier_list_notify(&migration_state_notifiers
, s
);
2440 * Open the return path. For postcopy, it is used exclusively. For
2441 * precopy, only if user specified "return-path" capability would
2442 * QEMU uses the return path.
2444 if (migrate_postcopy_ram() || migrate_use_return_path()) {
2445 if (open_return_path_on_source(s
)) {
2446 error_report("Unable to open return-path for postcopy");
2447 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2448 MIGRATION_STATUS_FAILED
);
2449 migrate_fd_cleanup(s
);
2454 if (multifd_save_setup() != 0) {
2455 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2456 MIGRATION_STATUS_FAILED
);
2457 migrate_fd_cleanup(s
);
2460 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
2461 QEMU_THREAD_JOINABLE
);
2462 s
->migration_thread_running
= true;
2465 void migration_global_dump(Monitor
*mon
)
2467 MigrationState
*ms
= migrate_get_current();
2469 monitor_printf(mon
, "globals:\n");
2470 monitor_printf(mon
, "store-global-state: %s\n",
2471 ms
->store_global_state
? "on" : "off");
2472 monitor_printf(mon
, "only-migratable: %s\n",
2473 ms
->only_migratable
? "on" : "off");
2474 monitor_printf(mon
, "send-configuration: %s\n",
2475 ms
->send_configuration
? "on" : "off");
2476 monitor_printf(mon
, "send-section-footer: %s\n",
2477 ms
->send_section_footer
? "on" : "off");
2480 #define DEFINE_PROP_MIG_CAP(name, x) \
2481 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
2483 static Property migration_properties
[] = {
2484 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
2485 store_global_state
, true),
2486 DEFINE_PROP_BOOL("only-migratable", MigrationState
, only_migratable
, false),
2487 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
2488 send_configuration
, true),
2489 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
2490 send_section_footer
, true),
2492 /* Migration parameters */
2493 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
2494 parameters
.compress_level
,
2495 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
2496 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
2497 parameters
.compress_threads
,
2498 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
2499 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
2500 parameters
.decompress_threads
,
2501 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
2502 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
2503 parameters
.cpu_throttle_initial
,
2504 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
2505 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
2506 parameters
.cpu_throttle_increment
,
2507 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
2508 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
2509 parameters
.max_bandwidth
, MAX_THROTTLE
),
2510 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
2511 parameters
.downtime_limit
,
2512 DEFAULT_MIGRATE_SET_DOWNTIME
),
2513 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
2514 parameters
.x_checkpoint_delay
,
2515 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
2516 DEFINE_PROP_UINT8("x-multifd-channels", MigrationState
,
2517 parameters
.x_multifd_channels
,
2518 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
2519 DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState
,
2520 parameters
.x_multifd_page_count
,
2521 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT
),
2522 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
2523 parameters
.xbzrle_cache_size
,
2524 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
2526 /* Migration capabilities */
2527 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
2528 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
2529 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
2530 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
2531 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
2532 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
2533 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
2534 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
2535 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
2536 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
2537 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
2538 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD
),
2540 DEFINE_PROP_END_OF_LIST(),
2543 static void migration_class_init(ObjectClass
*klass
, void *data
)
2545 DeviceClass
*dc
= DEVICE_CLASS(klass
);
2547 dc
->user_creatable
= false;
2548 dc
->props
= migration_properties
;
2551 static void migration_instance_finalize(Object
*obj
)
2553 MigrationState
*ms
= MIGRATION_OBJ(obj
);
2554 MigrationParameters
*params
= &ms
->parameters
;
2556 qemu_mutex_destroy(&ms
->error_mutex
);
2557 g_free(params
->tls_hostname
);
2558 g_free(params
->tls_creds
);
2559 qemu_sem_destroy(&ms
->pause_sem
);
2560 error_free(ms
->error
);
2563 static void migration_instance_init(Object
*obj
)
2565 MigrationState
*ms
= MIGRATION_OBJ(obj
);
2566 MigrationParameters
*params
= &ms
->parameters
;
2568 ms
->state
= MIGRATION_STATUS_NONE
;
2570 qemu_sem_init(&ms
->pause_sem
, 0);
2571 qemu_mutex_init(&ms
->error_mutex
);
2573 params
->tls_hostname
= g_strdup("");
2574 params
->tls_creds
= g_strdup("");
2576 /* Set has_* up only for parameter checks */
2577 params
->has_compress_level
= true;
2578 params
->has_compress_threads
= true;
2579 params
->has_decompress_threads
= true;
2580 params
->has_cpu_throttle_initial
= true;
2581 params
->has_cpu_throttle_increment
= true;
2582 params
->has_max_bandwidth
= true;
2583 params
->has_downtime_limit
= true;
2584 params
->has_x_checkpoint_delay
= true;
2585 params
->has_block_incremental
= true;
2586 params
->has_x_multifd_channels
= true;
2587 params
->has_x_multifd_page_count
= true;
2588 params
->has_xbzrle_cache_size
= true;
2592 * Return true if check pass, false otherwise. Error will be put
2593 * inside errp if provided.
2595 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
2597 MigrationCapabilityStatusList
*head
= NULL
;
2598 /* Assuming all off */
2599 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
2602 if (!migrate_params_check(&ms
->parameters
, errp
)) {
2606 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
2607 if (ms
->enabled_capabilities
[i
]) {
2608 head
= migrate_cap_add(head
, i
, true);
2612 ret
= migrate_caps_check(cap_list
, head
, errp
);
2614 /* It works with head == NULL */
2615 qapi_free_MigrationCapabilityStatusList(head
);
2620 static const TypeInfo migration_type
= {
2621 .name
= TYPE_MIGRATION
,
2623 * NOTE: TYPE_MIGRATION is not really a device, as the object is
2624 * not created using qdev_create(), it is not attached to the qdev
2625 * device tree, and it is never realized.
2627 * TODO: Make this TYPE_OBJECT once QOM provides something like
2628 * TYPE_DEVICE's "-global" properties.
2630 .parent
= TYPE_DEVICE
,
2631 .class_init
= migration_class_init
,
2632 .class_size
= sizeof(MigrationClass
),
2633 .instance_size
= sizeof(MigrationState
),
2634 .instance_init
= migration_instance_init
,
2635 .instance_finalize
= migration_instance_finalize
,
2638 static void register_migration_types(void)
2640 type_register_static(&migration_type
);
2643 type_init(register_migration_types
);