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 mis_current
.postcopy_remote_fds
= g_array_new(FALSE
, TRUE
,
159 sizeof(struct PostCopyFD
));
160 qemu_mutex_init(&mis_current
.rp_mutex
);
161 qemu_event_init(&mis_current
.main_thread_load_event
, false);
162 qemu_sem_init(&mis_current
.postcopy_pause_sem_dst
, 0);
163 qemu_sem_init(&mis_current
.postcopy_pause_sem_fault
, 0);
165 init_dirty_bitmap_incoming_migration();
172 void migration_incoming_state_destroy(void)
174 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
176 if (mis
->to_src_file
) {
177 /* Tell source that we are done */
178 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
179 qemu_fclose(mis
->to_src_file
);
180 mis
->to_src_file
= NULL
;
183 if (mis
->from_src_file
) {
184 qemu_fclose(mis
->from_src_file
);
185 mis
->from_src_file
= NULL
;
187 if (mis
->postcopy_remote_fds
) {
188 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
189 mis
->postcopy_remote_fds
= NULL
;
192 qemu_event_reset(&mis
->main_thread_load_event
);
195 static void migrate_generate_event(int new_state
)
197 if (migrate_use_events()) {
198 qapi_event_send_migration(new_state
, &error_abort
);
203 * Called on -incoming with a defer: uri.
204 * The migration can be started later after any parameters have been
207 static void deferred_incoming_migration(Error
**errp
)
209 if (deferred_incoming
) {
210 error_setg(errp
, "Incoming migration already deferred");
212 deferred_incoming
= true;
216 * Send a message on the return channel back to the source
219 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
220 enum mig_rp_message_type message_type
,
221 uint16_t len
, void *data
)
225 trace_migrate_send_rp_message((int)message_type
, len
);
226 qemu_mutex_lock(&mis
->rp_mutex
);
229 * It's possible that the file handle got lost due to network
232 if (!mis
->to_src_file
) {
237 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
238 qemu_put_be16(mis
->to_src_file
, len
);
239 qemu_put_buffer(mis
->to_src_file
, data
, len
);
240 qemu_fflush(mis
->to_src_file
);
242 /* It's possible that qemu file got error during sending */
243 ret
= qemu_file_get_error(mis
->to_src_file
);
246 qemu_mutex_unlock(&mis
->rp_mutex
);
250 /* Request a range of pages from the source VM at the given
252 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
253 * as the last request (a name must have been given previously)
254 * Start: Address offset within the RB
255 * Len: Length in bytes required - must be a multiple of pagesize
257 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
, const char *rbname
,
258 ram_addr_t start
, size_t len
)
260 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
261 size_t msglen
= 12; /* start + len */
262 enum mig_rp_message_type msg_type
;
264 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
265 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
268 int rbname_len
= strlen(rbname
);
269 assert(rbname_len
< 256);
271 bufc
[msglen
++] = rbname_len
;
272 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
273 msglen
+= rbname_len
;
274 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
276 msg_type
= MIG_RP_MSG_REQ_PAGES
;
279 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
282 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
286 qapi_event_send_migration(MIGRATION_STATUS_SETUP
, &error_abort
);
287 if (!strcmp(uri
, "defer")) {
288 deferred_incoming_migration(errp
);
289 } else if (strstart(uri
, "tcp:", &p
)) {
290 tcp_start_incoming_migration(p
, errp
);
292 } else if (strstart(uri
, "rdma:", &p
)) {
293 rdma_start_incoming_migration(p
, errp
);
295 } else if (strstart(uri
, "exec:", &p
)) {
296 exec_start_incoming_migration(p
, errp
);
297 } else if (strstart(uri
, "unix:", &p
)) {
298 unix_start_incoming_migration(p
, errp
);
299 } else if (strstart(uri
, "fd:", &p
)) {
300 fd_start_incoming_migration(p
, errp
);
302 error_setg(errp
, "unknown migration protocol: %s", uri
);
306 static void process_incoming_migration_bh(void *opaque
)
308 Error
*local_err
= NULL
;
309 MigrationIncomingState
*mis
= opaque
;
311 /* Make sure all file formats flush their mutable metadata.
312 * If we get an error here, just don't restart the VM yet. */
313 bdrv_invalidate_cache_all(&local_err
);
315 error_report_err(local_err
);
321 * This must happen after all error conditions are dealt with and
322 * we're sure the VM is going to be running on this host.
324 qemu_announce_self();
326 if (multifd_load_cleanup(&local_err
) != 0) {
327 error_report_err(local_err
);
330 /* If global state section was not received or we are in running
331 state, we need to obey autostart. Any other state is set with
334 dirty_bitmap_mig_before_vm_start();
336 if (!global_state_received() ||
337 global_state_get_runstate() == RUN_STATE_RUNNING
) {
341 runstate_set(RUN_STATE_PAUSED
);
344 runstate_set(global_state_get_runstate());
347 * This must happen after any state changes since as soon as an external
348 * observer sees this event they might start to prod at the VM assuming
351 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
352 MIGRATION_STATUS_COMPLETED
);
353 qemu_bh_delete(mis
->bh
);
354 migration_incoming_state_destroy();
357 static void process_incoming_migration_co(void *opaque
)
359 MigrationIncomingState
*mis
= migration_incoming_get_current();
363 assert(mis
->from_src_file
);
364 mis
->largest_page_size
= qemu_ram_pagesize_largest();
365 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
366 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
367 MIGRATION_STATUS_ACTIVE
);
368 ret
= qemu_loadvm_state(mis
->from_src_file
);
370 ps
= postcopy_state_get();
371 trace_process_incoming_migration_co_end(ret
, ps
);
372 if (ps
!= POSTCOPY_INCOMING_NONE
) {
373 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
375 * Where a migration had postcopy enabled (and thus went to advise)
376 * but managed to complete within the precopy period, we can use
379 postcopy_ram_incoming_cleanup(mis
);
380 } else if (ret
>= 0) {
382 * Postcopy was started, cleanup should happen at the end of the
385 trace_process_incoming_migration_co_postcopy_end_main();
388 /* Else if something went wrong then just fall out of the normal exit */
391 /* we get COLO info, and know if we are in COLO mode */
392 if (!ret
&& migration_incoming_enable_colo()) {
393 mis
->migration_incoming_co
= qemu_coroutine_self();
394 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
395 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
396 mis
->have_colo_incoming_thread
= true;
397 qemu_coroutine_yield();
399 /* Wait checkpoint incoming thread exit before free resource */
400 qemu_thread_join(&mis
->colo_incoming_thread
);
404 Error
*local_err
= NULL
;
406 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
407 MIGRATION_STATUS_FAILED
);
408 error_report("load of migration failed: %s", strerror(-ret
));
409 qemu_fclose(mis
->from_src_file
);
410 if (multifd_load_cleanup(&local_err
) != 0) {
411 error_report_err(local_err
);
415 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
416 qemu_bh_schedule(mis
->bh
);
419 static void migration_incoming_setup(QEMUFile
*f
)
421 MigrationIncomingState
*mis
= migration_incoming_get_current();
423 if (multifd_load_setup() != 0) {
424 /* We haven't been able to create multifd threads
425 nothing better to do */
429 if (!mis
->from_src_file
) {
430 mis
->from_src_file
= f
;
432 qemu_file_set_blocking(f
, false);
435 void migration_incoming_process(void)
437 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
438 qemu_coroutine_enter(co
);
441 void migration_fd_process_incoming(QEMUFile
*f
)
443 migration_incoming_setup(f
);
444 migration_incoming_process();
447 void migration_ioc_process_incoming(QIOChannel
*ioc
)
449 MigrationIncomingState
*mis
= migration_incoming_get_current();
451 if (!mis
->from_src_file
) {
452 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
453 migration_incoming_setup(f
);
456 multifd_recv_new_channel(ioc
);
460 * @migration_has_all_channels: We have received all channels that we need
462 * Returns true when we have got connections to all the channels that
463 * we need for migration.
465 bool migration_has_all_channels(void)
469 all_channels
= multifd_recv_all_channels_created();
475 * Send a 'SHUT' message on the return channel with the given value
476 * to indicate that we've finished with the RP. Non-0 value indicates
479 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
484 buf
= cpu_to_be32(value
);
485 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
489 * Send a 'PONG' message on the return channel with the given value
490 * (normally in response to a 'PING')
492 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
497 buf
= cpu_to_be32(value
);
498 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
501 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
503 MigrationCapabilityStatusList
*head
= NULL
;
504 MigrationCapabilityStatusList
*caps
;
505 MigrationState
*s
= migrate_get_current();
508 caps
= NULL
; /* silence compiler warning */
509 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
510 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
511 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
516 head
= g_malloc0(sizeof(*caps
));
519 caps
->next
= g_malloc0(sizeof(*caps
));
523 g_malloc(sizeof(*caps
->value
));
524 caps
->value
->capability
= i
;
525 caps
->value
->state
= s
->enabled_capabilities
[i
];
531 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
533 MigrationParameters
*params
;
534 MigrationState
*s
= migrate_get_current();
536 /* TODO use QAPI_CLONE() instead of duplicating it inline */
537 params
= g_malloc0(sizeof(*params
));
538 params
->has_compress_level
= true;
539 params
->compress_level
= s
->parameters
.compress_level
;
540 params
->has_compress_threads
= true;
541 params
->compress_threads
= s
->parameters
.compress_threads
;
542 params
->has_decompress_threads
= true;
543 params
->decompress_threads
= s
->parameters
.decompress_threads
;
544 params
->has_cpu_throttle_initial
= true;
545 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
546 params
->has_cpu_throttle_increment
= true;
547 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
548 params
->has_tls_creds
= true;
549 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
550 params
->has_tls_hostname
= true;
551 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
552 params
->has_max_bandwidth
= true;
553 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
554 params
->has_downtime_limit
= true;
555 params
->downtime_limit
= s
->parameters
.downtime_limit
;
556 params
->has_x_checkpoint_delay
= true;
557 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
558 params
->has_block_incremental
= true;
559 params
->block_incremental
= s
->parameters
.block_incremental
;
560 params
->has_x_multifd_channels
= true;
561 params
->x_multifd_channels
= s
->parameters
.x_multifd_channels
;
562 params
->has_x_multifd_page_count
= true;
563 params
->x_multifd_page_count
= s
->parameters
.x_multifd_page_count
;
564 params
->has_xbzrle_cache_size
= true;
565 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
571 * Return true if we're already in the middle of a migration
572 * (i.e. any of the active or setup states)
574 static bool migration_is_setup_or_active(int state
)
577 case MIGRATION_STATUS_ACTIVE
:
578 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
579 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
580 case MIGRATION_STATUS_SETUP
:
581 case MIGRATION_STATUS_PRE_SWITCHOVER
:
582 case MIGRATION_STATUS_DEVICE
:
591 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
593 info
->has_ram
= true;
594 info
->ram
= g_malloc0(sizeof(*info
->ram
));
595 info
->ram
->transferred
= ram_counters
.transferred
;
596 info
->ram
->total
= ram_bytes_total();
597 info
->ram
->duplicate
= ram_counters
.duplicate
;
598 /* legacy value. It is not used anymore */
599 info
->ram
->skipped
= 0;
600 info
->ram
->normal
= ram_counters
.normal
;
601 info
->ram
->normal_bytes
= ram_counters
.normal
*
602 qemu_target_page_size();
603 info
->ram
->mbps
= s
->mbps
;
604 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
605 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
606 info
->ram
->page_size
= qemu_target_page_size();
608 if (migrate_use_xbzrle()) {
609 info
->has_xbzrle_cache
= true;
610 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
611 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
612 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
613 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
614 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
615 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
616 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
619 if (cpu_throttle_active()) {
620 info
->has_cpu_throttle_percentage
= true;
621 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
624 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
625 info
->ram
->remaining
= ram_bytes_remaining();
626 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
630 static void populate_disk_info(MigrationInfo
*info
)
632 if (blk_mig_active()) {
633 info
->has_disk
= true;
634 info
->disk
= g_malloc0(sizeof(*info
->disk
));
635 info
->disk
->transferred
= blk_mig_bytes_transferred();
636 info
->disk
->remaining
= blk_mig_bytes_remaining();
637 info
->disk
->total
= blk_mig_bytes_total();
641 static void fill_source_migration_info(MigrationInfo
*info
)
643 MigrationState
*s
= migrate_get_current();
646 case MIGRATION_STATUS_NONE
:
647 /* no migration has happened ever */
648 /* do not overwrite destination migration status */
651 case MIGRATION_STATUS_SETUP
:
652 info
->has_status
= true;
653 info
->has_total_time
= false;
655 case MIGRATION_STATUS_ACTIVE
:
656 case MIGRATION_STATUS_CANCELLING
:
657 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
658 case MIGRATION_STATUS_PRE_SWITCHOVER
:
659 case MIGRATION_STATUS_DEVICE
:
660 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
661 /* TODO add some postcopy stats */
662 info
->has_status
= true;
663 info
->has_total_time
= true;
664 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
666 info
->has_expected_downtime
= true;
667 info
->expected_downtime
= s
->expected_downtime
;
668 info
->has_setup_time
= true;
669 info
->setup_time
= s
->setup_time
;
671 populate_ram_info(info
, s
);
672 populate_disk_info(info
);
674 case MIGRATION_STATUS_COLO
:
675 info
->has_status
= true;
676 /* TODO: display COLO specific information (checkpoint info etc.) */
678 case MIGRATION_STATUS_COMPLETED
:
679 info
->has_status
= true;
680 info
->has_total_time
= true;
681 info
->total_time
= s
->total_time
;
682 info
->has_downtime
= true;
683 info
->downtime
= s
->downtime
;
684 info
->has_setup_time
= true;
685 info
->setup_time
= s
->setup_time
;
687 populate_ram_info(info
, s
);
689 case MIGRATION_STATUS_FAILED
:
690 info
->has_status
= true;
692 info
->has_error_desc
= true;
693 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
696 case MIGRATION_STATUS_CANCELLED
:
697 info
->has_status
= true;
700 info
->status
= s
->state
;
704 * @migration_caps_check - check capability validity
706 * @cap_list: old capability list, array of bool
707 * @params: new capabilities to be applied soon
708 * @errp: set *errp if the check failed, with reason
710 * Returns true if check passed, otherwise false.
712 static bool migrate_caps_check(bool *cap_list
,
713 MigrationCapabilityStatusList
*params
,
716 MigrationCapabilityStatusList
*cap
;
717 bool old_postcopy_cap
;
718 MigrationIncomingState
*mis
= migration_incoming_get_current();
720 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
722 for (cap
= params
; cap
; cap
= cap
->next
) {
723 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
726 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
727 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
728 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
730 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
735 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
736 if (cap_list
[MIGRATION_CAPABILITY_COMPRESS
]) {
737 /* The decompression threads asynchronously write into RAM
738 * rather than use the atomic copies needed to avoid
739 * userfaulting. It should be possible to fix the decompression
740 * threads for compatibility in future.
742 error_setg(errp
, "Postcopy is not currently compatible "
747 /* This check is reasonably expensive, so only when it's being
748 * set the first time, also it's only the destination that needs
751 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
752 !postcopy_ram_supported_by_host(mis
)) {
753 /* postcopy_ram_supported_by_host will have emitted a more
756 error_setg(errp
, "Postcopy is not supported");
764 static void fill_destination_migration_info(MigrationInfo
*info
)
766 MigrationIncomingState
*mis
= migration_incoming_get_current();
768 switch (mis
->state
) {
769 case MIGRATION_STATUS_NONE
:
772 case MIGRATION_STATUS_SETUP
:
773 case MIGRATION_STATUS_CANCELLING
:
774 case MIGRATION_STATUS_CANCELLED
:
775 case MIGRATION_STATUS_ACTIVE
:
776 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
777 case MIGRATION_STATUS_FAILED
:
778 case MIGRATION_STATUS_COLO
:
779 info
->has_status
= true;
781 case MIGRATION_STATUS_COMPLETED
:
782 info
->has_status
= true;
783 fill_destination_postcopy_migration_info(info
);
786 info
->status
= mis
->state
;
789 MigrationInfo
*qmp_query_migrate(Error
**errp
)
791 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
793 fill_destination_migration_info(info
);
794 fill_source_migration_info(info
);
799 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
802 MigrationState
*s
= migrate_get_current();
803 MigrationCapabilityStatusList
*cap
;
804 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
806 if (migration_is_setup_or_active(s
->state
)) {
807 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
811 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
812 if (!migrate_caps_check(cap_list
, params
, errp
)) {
816 for (cap
= params
; cap
; cap
= cap
->next
) {
817 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
822 * Check whether the parameters are valid. Error will be put into errp
823 * (if provided). Return true if valid, otherwise false.
825 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
827 if (params
->has_compress_level
&&
828 (params
->compress_level
> 9)) {
829 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
830 "is invalid, it should be in the range of 0 to 9");
834 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
835 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
837 "is invalid, it should be in the range of 1 to 255");
841 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
842 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
843 "decompress_threads",
844 "is invalid, it should be in the range of 1 to 255");
848 if (params
->has_cpu_throttle_initial
&&
849 (params
->cpu_throttle_initial
< 1 ||
850 params
->cpu_throttle_initial
> 99)) {
851 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
852 "cpu_throttle_initial",
853 "an integer in the range of 1 to 99");
857 if (params
->has_cpu_throttle_increment
&&
858 (params
->cpu_throttle_increment
< 1 ||
859 params
->cpu_throttle_increment
> 99)) {
860 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
861 "cpu_throttle_increment",
862 "an integer in the range of 1 to 99");
866 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
867 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
868 " range of 0 to %zu bytes/second", SIZE_MAX
);
872 if (params
->has_downtime_limit
&&
873 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
874 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
875 "the range of 0 to %d milliseconds",
876 MAX_MIGRATE_DOWNTIME
);
880 /* x_checkpoint_delay is now always positive */
882 if (params
->has_x_multifd_channels
&& (params
->x_multifd_channels
< 1)) {
883 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
885 "is invalid, it should be in the range of 1 to 255");
888 if (params
->has_x_multifd_page_count
&&
889 (params
->x_multifd_page_count
< 1 ||
890 params
->x_multifd_page_count
> 10000)) {
891 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
892 "multifd_page_count",
893 "is invalid, it should be in the range of 1 to 10000");
897 if (params
->has_xbzrle_cache_size
&&
898 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
899 !is_power_of_2(params
->xbzrle_cache_size
))) {
900 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
902 "is invalid, it should be bigger than target page size"
903 " and a power of two");
910 static void migrate_params_test_apply(MigrateSetParameters
*params
,
911 MigrationParameters
*dest
)
913 *dest
= migrate_get_current()->parameters
;
915 /* TODO use QAPI_CLONE() instead of duplicating it inline */
917 if (params
->has_compress_level
) {
918 dest
->compress_level
= params
->compress_level
;
921 if (params
->has_compress_threads
) {
922 dest
->compress_threads
= params
->compress_threads
;
925 if (params
->has_decompress_threads
) {
926 dest
->decompress_threads
= params
->decompress_threads
;
929 if (params
->has_cpu_throttle_initial
) {
930 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
933 if (params
->has_cpu_throttle_increment
) {
934 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
937 if (params
->has_tls_creds
) {
938 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
939 dest
->tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
942 if (params
->has_tls_hostname
) {
943 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
944 dest
->tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
947 if (params
->has_max_bandwidth
) {
948 dest
->max_bandwidth
= params
->max_bandwidth
;
951 if (params
->has_downtime_limit
) {
952 dest
->downtime_limit
= params
->downtime_limit
;
955 if (params
->has_x_checkpoint_delay
) {
956 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
959 if (params
->has_block_incremental
) {
960 dest
->block_incremental
= params
->block_incremental
;
962 if (params
->has_x_multifd_channels
) {
963 dest
->x_multifd_channels
= params
->x_multifd_channels
;
965 if (params
->has_x_multifd_page_count
) {
966 dest
->x_multifd_page_count
= params
->x_multifd_page_count
;
968 if (params
->has_xbzrle_cache_size
) {
969 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
973 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
975 MigrationState
*s
= migrate_get_current();
977 /* TODO use QAPI_CLONE() instead of duplicating it inline */
979 if (params
->has_compress_level
) {
980 s
->parameters
.compress_level
= params
->compress_level
;
983 if (params
->has_compress_threads
) {
984 s
->parameters
.compress_threads
= params
->compress_threads
;
987 if (params
->has_decompress_threads
) {
988 s
->parameters
.decompress_threads
= params
->decompress_threads
;
991 if (params
->has_cpu_throttle_initial
) {
992 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
995 if (params
->has_cpu_throttle_increment
) {
996 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
999 if (params
->has_tls_creds
) {
1000 g_free(s
->parameters
.tls_creds
);
1001 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1002 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1005 if (params
->has_tls_hostname
) {
1006 g_free(s
->parameters
.tls_hostname
);
1007 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1008 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1011 if (params
->has_max_bandwidth
) {
1012 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1013 if (s
->to_dst_file
) {
1014 qemu_file_set_rate_limit(s
->to_dst_file
,
1015 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1019 if (params
->has_downtime_limit
) {
1020 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1023 if (params
->has_x_checkpoint_delay
) {
1024 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1025 if (migration_in_colo_state()) {
1026 colo_checkpoint_notify(s
);
1030 if (params
->has_block_incremental
) {
1031 s
->parameters
.block_incremental
= params
->block_incremental
;
1033 if (params
->has_x_multifd_channels
) {
1034 s
->parameters
.x_multifd_channels
= params
->x_multifd_channels
;
1036 if (params
->has_x_multifd_page_count
) {
1037 s
->parameters
.x_multifd_page_count
= params
->x_multifd_page_count
;
1039 if (params
->has_xbzrle_cache_size
) {
1040 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1041 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1045 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1047 MigrationParameters tmp
;
1049 /* TODO Rewrite "" to null instead */
1050 if (params
->has_tls_creds
1051 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1052 qobject_unref(params
->tls_creds
->u
.n
);
1053 params
->tls_creds
->type
= QTYPE_QSTRING
;
1054 params
->tls_creds
->u
.s
= strdup("");
1056 /* TODO Rewrite "" to null instead */
1057 if (params
->has_tls_hostname
1058 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1059 qobject_unref(params
->tls_hostname
->u
.n
);
1060 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1061 params
->tls_hostname
->u
.s
= strdup("");
1064 migrate_params_test_apply(params
, &tmp
);
1066 if (!migrate_params_check(&tmp
, errp
)) {
1067 /* Invalid parameter */
1071 migrate_params_apply(params
, errp
);
1075 void qmp_migrate_start_postcopy(Error
**errp
)
1077 MigrationState
*s
= migrate_get_current();
1079 if (!migrate_postcopy()) {
1080 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1081 " the start of migration");
1085 if (s
->state
== MIGRATION_STATUS_NONE
) {
1086 error_setg(errp
, "Postcopy must be started after migration has been"
1091 * we don't error if migration has finished since that would be racy
1092 * with issuing this command.
1094 atomic_set(&s
->start_postcopy
, true);
1097 /* shared migration helpers */
1099 void migrate_set_state(int *state
, int old_state
, int new_state
)
1101 assert(new_state
< MIGRATION_STATUS__MAX
);
1102 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1103 trace_migrate_set_state(MigrationStatus_str(new_state
));
1104 migrate_generate_event(new_state
);
1108 static MigrationCapabilityStatusList
*migrate_cap_add(
1109 MigrationCapabilityStatusList
*list
,
1110 MigrationCapability index
,
1113 MigrationCapabilityStatusList
*cap
;
1115 cap
= g_new0(MigrationCapabilityStatusList
, 1);
1116 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
1117 cap
->value
->capability
= index
;
1118 cap
->value
->state
= state
;
1124 void migrate_set_block_enabled(bool value
, Error
**errp
)
1126 MigrationCapabilityStatusList
*cap
;
1128 cap
= migrate_cap_add(NULL
, MIGRATION_CAPABILITY_BLOCK
, value
);
1129 qmp_migrate_set_capabilities(cap
, errp
);
1130 qapi_free_MigrationCapabilityStatusList(cap
);
1133 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1135 s
->parameters
.block_incremental
= value
;
1138 static void block_cleanup_parameters(MigrationState
*s
)
1140 if (s
->must_remove_block_options
) {
1141 /* setting to false can never fail */
1142 migrate_set_block_enabled(false, &error_abort
);
1143 migrate_set_block_incremental(s
, false);
1144 s
->must_remove_block_options
= false;
1148 static void migrate_fd_cleanup(void *opaque
)
1150 MigrationState
*s
= opaque
;
1152 qemu_bh_delete(s
->cleanup_bh
);
1153 s
->cleanup_bh
= NULL
;
1155 qemu_savevm_state_cleanup();
1157 if (s
->to_dst_file
) {
1158 Error
*local_err
= NULL
;
1160 trace_migrate_fd_cleanup();
1161 qemu_mutex_unlock_iothread();
1162 if (s
->migration_thread_running
) {
1163 qemu_thread_join(&s
->thread
);
1164 s
->migration_thread_running
= false;
1166 qemu_mutex_lock_iothread();
1168 if (multifd_save_cleanup(&local_err
) != 0) {
1169 error_report_err(local_err
);
1171 qemu_fclose(s
->to_dst_file
);
1172 s
->to_dst_file
= NULL
;
1175 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
1176 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
1178 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1179 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1180 MIGRATION_STATUS_CANCELLED
);
1184 /* It is used on info migrate. We can't free it */
1185 error_report_err(error_copy(s
->error
));
1187 notifier_list_notify(&migration_state_notifiers
, s
);
1188 block_cleanup_parameters(s
);
1191 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1193 qemu_mutex_lock(&s
->error_mutex
);
1195 s
->error
= error_copy(error
);
1197 qemu_mutex_unlock(&s
->error_mutex
);
1200 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1202 trace_migrate_fd_error(error_get_pretty(error
));
1203 assert(s
->to_dst_file
== NULL
);
1204 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1205 MIGRATION_STATUS_FAILED
);
1206 migrate_set_error(s
, error
);
1209 static void migrate_fd_cancel(MigrationState
*s
)
1212 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1213 trace_migrate_fd_cancel();
1215 if (s
->rp_state
.from_dst_file
) {
1216 /* shutdown the rp socket, so causing the rp thread to shutdown */
1217 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1221 old_state
= s
->state
;
1222 if (!migration_is_setup_or_active(old_state
)) {
1225 /* If the migration is paused, kick it out of the pause */
1226 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1227 qemu_sem_post(&s
->pause_sem
);
1229 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1230 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1233 * If we're unlucky the migration code might be stuck somewhere in a
1234 * send/write while the network has failed and is waiting to timeout;
1235 * if we've got shutdown(2) available then we can force it to quit.
1236 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1237 * called in a bh, so there is no race against this cancel.
1239 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1240 qemu_file_shutdown(f
);
1242 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1243 Error
*local_err
= NULL
;
1245 bdrv_invalidate_cache_all(&local_err
);
1247 error_report_err(local_err
);
1249 s
->block_inactive
= false;
1254 void add_migration_state_change_notifier(Notifier
*notify
)
1256 notifier_list_add(&migration_state_notifiers
, notify
);
1259 void remove_migration_state_change_notifier(Notifier
*notify
)
1261 notifier_remove(notify
);
1264 bool migration_in_setup(MigrationState
*s
)
1266 return s
->state
== MIGRATION_STATUS_SETUP
;
1269 bool migration_has_finished(MigrationState
*s
)
1271 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1274 bool migration_has_failed(MigrationState
*s
)
1276 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1277 s
->state
== MIGRATION_STATUS_FAILED
);
1280 bool migration_in_postcopy(void)
1282 MigrationState
*s
= migrate_get_current();
1284 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1287 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1289 return migration_in_postcopy() && s
->postcopy_after_devices
;
1292 bool migration_is_idle(void)
1294 MigrationState
*s
= migrate_get_current();
1297 case MIGRATION_STATUS_NONE
:
1298 case MIGRATION_STATUS_CANCELLED
:
1299 case MIGRATION_STATUS_COMPLETED
:
1300 case MIGRATION_STATUS_FAILED
:
1302 case MIGRATION_STATUS_SETUP
:
1303 case MIGRATION_STATUS_CANCELLING
:
1304 case MIGRATION_STATUS_ACTIVE
:
1305 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1306 case MIGRATION_STATUS_COLO
:
1307 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1308 case MIGRATION_STATUS_DEVICE
:
1310 case MIGRATION_STATUS__MAX
:
1311 g_assert_not_reached();
1317 void migrate_init(MigrationState
*s
)
1320 * Reinitialise all migration state, except
1321 * parameters/capabilities that the user set, and
1327 s
->to_dst_file
= NULL
;
1328 s
->state
= MIGRATION_STATUS_NONE
;
1329 s
->rp_state
.from_dst_file
= NULL
;
1330 s
->rp_state
.error
= false;
1333 s
->expected_downtime
= 0;
1335 s
->start_postcopy
= false;
1336 s
->postcopy_after_devices
= false;
1337 s
->migration_thread_running
= false;
1338 error_free(s
->error
);
1341 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1343 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1345 s
->vm_was_running
= false;
1346 s
->iteration_initial_bytes
= 0;
1347 s
->threshold_size
= 0;
1350 static GSList
*migration_blockers
;
1352 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1354 if (migrate_get_current()->only_migratable
) {
1355 error_propagate(errp
, error_copy(reason
));
1356 error_prepend(errp
, "disallowing migration blocker "
1357 "(--only_migratable) for: ");
1361 if (migration_is_idle()) {
1362 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1366 error_propagate(errp
, error_copy(reason
));
1367 error_prepend(errp
, "disallowing migration blocker (migration in "
1372 void migrate_del_blocker(Error
*reason
)
1374 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1377 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1379 Error
*local_err
= NULL
;
1380 static bool once
= true;
1382 if (!deferred_incoming
) {
1383 error_setg(errp
, "For use with '-incoming defer'");
1387 error_setg(errp
, "The incoming migration has already been started");
1390 qemu_start_incoming_migration(uri
, &local_err
);
1393 error_propagate(errp
, local_err
);
1400 bool migration_is_blocked(Error
**errp
)
1402 if (qemu_savevm_state_blocked(errp
)) {
1406 if (migration_blockers
) {
1407 error_propagate(errp
, error_copy(migration_blockers
->data
));
1414 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1415 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1416 bool has_resume
, bool resume
, Error
**errp
)
1418 Error
*local_err
= NULL
;
1419 MigrationState
*s
= migrate_get_current();
1422 if (migration_is_setup_or_active(s
->state
) ||
1423 s
->state
== MIGRATION_STATUS_CANCELLING
||
1424 s
->state
== MIGRATION_STATUS_COLO
) {
1425 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1428 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1429 error_setg(errp
, "Guest is waiting for an incoming migration");
1433 if (migration_is_blocked(errp
)) {
1437 if ((has_blk
&& blk
) || (has_inc
&& inc
)) {
1438 if (migrate_use_block() || migrate_use_block_incremental()) {
1439 error_setg(errp
, "Command options are incompatible with "
1440 "current migration capabilities");
1443 migrate_set_block_enabled(true, &local_err
);
1445 error_propagate(errp
, local_err
);
1448 s
->must_remove_block_options
= true;
1451 if (has_inc
&& inc
) {
1452 migrate_set_block_incremental(s
, true);
1457 if (strstart(uri
, "tcp:", &p
)) {
1458 tcp_start_outgoing_migration(s
, p
, &local_err
);
1460 } else if (strstart(uri
, "rdma:", &p
)) {
1461 rdma_start_outgoing_migration(s
, p
, &local_err
);
1463 } else if (strstart(uri
, "exec:", &p
)) {
1464 exec_start_outgoing_migration(s
, p
, &local_err
);
1465 } else if (strstart(uri
, "unix:", &p
)) {
1466 unix_start_outgoing_migration(s
, p
, &local_err
);
1467 } else if (strstart(uri
, "fd:", &p
)) {
1468 fd_start_outgoing_migration(s
, p
, &local_err
);
1470 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1471 "a valid migration protocol");
1472 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1473 MIGRATION_STATUS_FAILED
);
1474 block_cleanup_parameters(s
);
1479 migrate_fd_error(s
, local_err
);
1480 error_propagate(errp
, local_err
);
1485 void qmp_migrate_cancel(Error
**errp
)
1487 migrate_fd_cancel(migrate_get_current());
1490 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
1492 MigrationState
*s
= migrate_get_current();
1493 if (s
->state
!= state
) {
1494 error_setg(errp
, "Migration not in expected state: %s",
1495 MigrationStatus_str(s
->state
));
1498 qemu_sem_post(&s
->pause_sem
);
1501 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1503 MigrateSetParameters p
= {
1504 .has_xbzrle_cache_size
= true,
1505 .xbzrle_cache_size
= value
,
1508 qmp_migrate_set_parameters(&p
, errp
);
1511 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1513 return migrate_xbzrle_cache_size();
1516 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1518 MigrateSetParameters p
= {
1519 .has_max_bandwidth
= true,
1520 .max_bandwidth
= value
,
1523 qmp_migrate_set_parameters(&p
, errp
);
1526 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1528 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
1529 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1530 "the range of 0 to %d seconds",
1531 MAX_MIGRATE_DOWNTIME_SECONDS
);
1535 value
*= 1000; /* Convert to milliseconds */
1536 value
= MAX(0, MIN(INT64_MAX
, value
));
1538 MigrateSetParameters p
= {
1539 .has_downtime_limit
= true,
1540 .downtime_limit
= value
,
1543 qmp_migrate_set_parameters(&p
, errp
);
1546 bool migrate_release_ram(void)
1550 s
= migrate_get_current();
1552 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
1555 bool migrate_postcopy_ram(void)
1559 s
= migrate_get_current();
1561 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1564 bool migrate_postcopy(void)
1566 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
1569 bool migrate_auto_converge(void)
1573 s
= migrate_get_current();
1575 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1578 bool migrate_zero_blocks(void)
1582 s
= migrate_get_current();
1584 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
1587 bool migrate_postcopy_blocktime(void)
1591 s
= migrate_get_current();
1593 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
1596 bool migrate_use_compression(void)
1600 s
= migrate_get_current();
1602 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
1605 int migrate_compress_level(void)
1609 s
= migrate_get_current();
1611 return s
->parameters
.compress_level
;
1614 int migrate_compress_threads(void)
1618 s
= migrate_get_current();
1620 return s
->parameters
.compress_threads
;
1623 int migrate_decompress_threads(void)
1627 s
= migrate_get_current();
1629 return s
->parameters
.decompress_threads
;
1632 bool migrate_dirty_bitmaps(void)
1636 s
= migrate_get_current();
1638 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
1641 bool migrate_use_events(void)
1645 s
= migrate_get_current();
1647 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
1650 bool migrate_use_multifd(void)
1654 s
= migrate_get_current();
1656 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_MULTIFD
];
1659 bool migrate_pause_before_switchover(void)
1663 s
= migrate_get_current();
1665 return s
->enabled_capabilities
[
1666 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
1669 int migrate_multifd_channels(void)
1673 s
= migrate_get_current();
1675 return s
->parameters
.x_multifd_channels
;
1678 int migrate_multifd_page_count(void)
1682 s
= migrate_get_current();
1684 return s
->parameters
.x_multifd_page_count
;
1687 int migrate_use_xbzrle(void)
1691 s
= migrate_get_current();
1693 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
1696 int64_t migrate_xbzrle_cache_size(void)
1700 s
= migrate_get_current();
1702 return s
->parameters
.xbzrle_cache_size
;
1705 bool migrate_use_block(void)
1709 s
= migrate_get_current();
1711 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
1714 bool migrate_use_return_path(void)
1718 s
= migrate_get_current();
1720 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
1723 bool migrate_use_block_incremental(void)
1727 s
= migrate_get_current();
1729 return s
->parameters
.block_incremental
;
1732 /* migration thread support */
1734 * Something bad happened to the RP stream, mark an error
1735 * The caller shall print or trace something to indicate why
1737 static void mark_source_rp_bad(MigrationState
*s
)
1739 s
->rp_state
.error
= true;
1742 static struct rp_cmd_args
{
1743 ssize_t len
; /* -1 = variable */
1746 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1747 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1748 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1749 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1750 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1751 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1755 * Process a request for pages received on the return path,
1756 * We're allowed to send more than requested (e.g. to round to our page size)
1757 * and we don't need to send pages that have already been sent.
1759 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1760 ram_addr_t start
, size_t len
)
1762 long our_host_ps
= getpagesize();
1764 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1767 * Since we currently insist on matching page sizes, just sanity check
1768 * we're being asked for whole host pages.
1770 if (start
& (our_host_ps
-1) ||
1771 (len
& (our_host_ps
-1))) {
1772 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1773 " len: %zd", __func__
, start
, len
);
1774 mark_source_rp_bad(ms
);
1778 if (ram_save_queue_pages(rbname
, start
, len
)) {
1779 mark_source_rp_bad(ms
);
1783 /* Return true to retry, false to quit */
1784 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
1786 trace_postcopy_pause_return_path();
1788 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
1790 trace_postcopy_pause_return_path_continued();
1796 * Handles messages sent on the return path towards the source VM
1799 static void *source_return_path_thread(void *opaque
)
1801 MigrationState
*ms
= opaque
;
1802 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1803 uint16_t header_len
, header_type
;
1805 uint32_t tmp32
, sibling_error
;
1806 ram_addr_t start
= 0; /* =0 to silence warning */
1807 size_t len
= 0, expected_len
;
1810 trace_source_return_path_thread_entry();
1813 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1814 migration_is_setup_or_active(ms
->state
)) {
1815 trace_source_return_path_thread_loop_top();
1816 header_type
= qemu_get_be16(rp
);
1817 header_len
= qemu_get_be16(rp
);
1819 if (qemu_file_get_error(rp
)) {
1820 mark_source_rp_bad(ms
);
1824 if (header_type
>= MIG_RP_MSG_MAX
||
1825 header_type
== MIG_RP_MSG_INVALID
) {
1826 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1827 header_type
, header_len
);
1828 mark_source_rp_bad(ms
);
1832 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1833 header_len
!= rp_cmd_args
[header_type
].len
) ||
1834 header_len
> sizeof(buf
)) {
1835 error_report("RP: Received '%s' message (0x%04x) with"
1836 "incorrect length %d expecting %zu",
1837 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1838 (size_t)rp_cmd_args
[header_type
].len
);
1839 mark_source_rp_bad(ms
);
1843 /* We know we've got a valid header by this point */
1844 res
= qemu_get_buffer(rp
, buf
, header_len
);
1845 if (res
!= header_len
) {
1846 error_report("RP: Failed reading data for message 0x%04x"
1847 " read %d expected %d",
1848 header_type
, res
, header_len
);
1849 mark_source_rp_bad(ms
);
1853 /* OK, we have the message and the data */
1854 switch (header_type
) {
1855 case MIG_RP_MSG_SHUT
:
1856 sibling_error
= ldl_be_p(buf
);
1857 trace_source_return_path_thread_shut(sibling_error
);
1858 if (sibling_error
) {
1859 error_report("RP: Sibling indicated error %d", sibling_error
);
1860 mark_source_rp_bad(ms
);
1863 * We'll let the main thread deal with closing the RP
1864 * we could do a shutdown(2) on it, but we're the only user
1865 * anyway, so there's nothing gained.
1869 case MIG_RP_MSG_PONG
:
1870 tmp32
= ldl_be_p(buf
);
1871 trace_source_return_path_thread_pong(tmp32
);
1874 case MIG_RP_MSG_REQ_PAGES
:
1875 start
= ldq_be_p(buf
);
1876 len
= ldl_be_p(buf
+ 8);
1877 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1880 case MIG_RP_MSG_REQ_PAGES_ID
:
1881 expected_len
= 12 + 1; /* header + termination */
1883 if (header_len
>= expected_len
) {
1884 start
= ldq_be_p(buf
);
1885 len
= ldl_be_p(buf
+ 8);
1886 /* Now we expect an idstr */
1887 tmp32
= buf
[12]; /* Length of the following idstr */
1888 buf
[13 + tmp32
] = '\0';
1889 expected_len
+= tmp32
;
1891 if (header_len
!= expected_len
) {
1892 error_report("RP: Req_Page_id with length %d expecting %zd",
1893 header_len
, expected_len
);
1894 mark_source_rp_bad(ms
);
1897 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
1906 res
= qemu_file_get_error(rp
);
1910 * Maybe there is something we can do: it looks like a
1911 * network down issue, and we pause for a recovery.
1913 if (postcopy_pause_return_path_thread(ms
)) {
1914 /* Reload rp, reset the rest */
1915 rp
= ms
->rp_state
.from_dst_file
;
1916 ms
->rp_state
.error
= false;
1921 trace_source_return_path_thread_bad_end();
1922 mark_source_rp_bad(ms
);
1925 trace_source_return_path_thread_end();
1926 ms
->rp_state
.from_dst_file
= NULL
;
1931 static int open_return_path_on_source(MigrationState
*ms
)
1934 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
1935 if (!ms
->rp_state
.from_dst_file
) {
1939 trace_open_return_path_on_source();
1940 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
1941 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
1943 trace_open_return_path_on_source_continue();
1948 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1949 static int await_return_path_close_on_source(MigrationState
*ms
)
1952 * If this is a normal exit then the destination will send a SHUT and the
1953 * rp_thread will exit, however if there's an error we need to cause
1956 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
1958 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1959 * waiting for the destination.
1961 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
1962 mark_source_rp_bad(ms
);
1964 trace_await_return_path_close_on_source_joining();
1965 qemu_thread_join(&ms
->rp_state
.rp_thread
);
1966 trace_await_return_path_close_on_source_close();
1967 return ms
->rp_state
.error
;
1971 * Switch from normal iteration to postcopy
1972 * Returns non-0 on error
1974 static int postcopy_start(MigrationState
*ms
)
1977 QIOChannelBuffer
*bioc
;
1979 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1980 bool restart_block
= false;
1981 int cur_state
= MIGRATION_STATUS_ACTIVE
;
1982 if (!migrate_pause_before_switchover()) {
1983 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
1984 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1987 trace_postcopy_start();
1988 qemu_mutex_lock_iothread();
1989 trace_postcopy_start_set_run();
1991 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1992 global_state_store();
1993 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1998 ret
= migration_maybe_pause(ms
, &cur_state
,
1999 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2004 ret
= bdrv_inactivate_all();
2008 restart_block
= true;
2011 * Cause any non-postcopiable, but iterative devices to
2012 * send out their final data.
2014 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2017 * in Finish migrate and with the io-lock held everything should
2018 * be quiet, but we've potentially still got dirty pages and we
2019 * need to tell the destination to throw any pages it's already received
2022 if (migrate_postcopy_ram()) {
2023 if (ram_postcopy_send_discard_bitmap(ms
)) {
2024 error_report("postcopy send discard bitmap failed");
2030 * send rest of state - note things that are doing postcopy
2031 * will notice we're in POSTCOPY_ACTIVE and not actually
2032 * wrap their state up here
2034 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
2035 if (migrate_postcopy_ram()) {
2036 /* Ping just for debugging, helps line traces up */
2037 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
2041 * While loading the device state we may trigger page transfer
2042 * requests and the fd must be free to process those, and thus
2043 * the destination must read the whole device state off the fd before
2044 * it starts processing it. Unfortunately the ad-hoc migration format
2045 * doesn't allow the destination to know the size to read without fully
2046 * parsing it through each devices load-state code (especially the open
2047 * coded devices that use get/put).
2048 * So we wrap the device state up in a package with a length at the start;
2049 * to do this we use a qemu_buf to hold the whole of the device state.
2051 bioc
= qio_channel_buffer_new(4096);
2052 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
2053 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
2054 object_unref(OBJECT(bioc
));
2057 * Make sure the receiver can get incoming pages before we send the rest
2060 qemu_savevm_send_postcopy_listen(fb
);
2062 qemu_savevm_state_complete_precopy(fb
, false, false);
2063 if (migrate_postcopy_ram()) {
2064 qemu_savevm_send_ping(fb
, 3);
2067 qemu_savevm_send_postcopy_run(fb
);
2069 /* <><> end of stuff going into the package */
2071 /* Last point of recovery; as soon as we send the package the destination
2072 * can open devices and potentially start running.
2073 * Lets just check again we've not got any errors.
2075 ret
= qemu_file_get_error(ms
->to_dst_file
);
2077 error_report("postcopy_start: Migration stream errored (pre package)");
2081 restart_block
= false;
2083 /* Now send that blob */
2084 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2089 /* Send a notify to give a chance for anything that needs to happen
2090 * at the transition to postcopy and after the device state; in particular
2091 * spice needs to trigger a transition now
2093 ms
->postcopy_after_devices
= true;
2094 notifier_list_notify(&migration_state_notifiers
, ms
);
2096 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2098 qemu_mutex_unlock_iothread();
2100 if (migrate_postcopy_ram()) {
2102 * Although this ping is just for debug, it could potentially be
2103 * used for getting a better measurement of downtime at the source.
2105 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2108 if (migrate_release_ram()) {
2109 ram_postcopy_migrated_memory_release(ms
);
2112 ret
= qemu_file_get_error(ms
->to_dst_file
);
2114 error_report("postcopy_start: Migration stream errored");
2115 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2116 MIGRATION_STATUS_FAILED
);
2124 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2125 MIGRATION_STATUS_FAILED
);
2126 if (restart_block
) {
2127 /* A failure happened early enough that we know the destination hasn't
2128 * accessed block devices, so we're safe to recover.
2130 Error
*local_err
= NULL
;
2132 bdrv_invalidate_cache_all(&local_err
);
2134 error_report_err(local_err
);
2137 qemu_mutex_unlock_iothread();
2142 * migration_maybe_pause: Pause if required to by
2143 * migrate_pause_before_switchover called with the iothread locked
2144 * Returns: 0 on success
2146 static int migration_maybe_pause(MigrationState
*s
,
2147 int *current_active_state
,
2150 if (!migrate_pause_before_switchover()) {
2154 /* Since leaving this state is not atomic with posting the semaphore
2155 * it's possible that someone could have issued multiple migrate_continue
2156 * and the semaphore is incorrectly positive at this point;
2157 * the docs say it's undefined to reinit a semaphore that's already
2158 * init'd, so use timedwait to eat up any existing posts.
2160 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2161 /* This block intentionally left blank */
2164 qemu_mutex_unlock_iothread();
2165 migrate_set_state(&s
->state
, *current_active_state
,
2166 MIGRATION_STATUS_PRE_SWITCHOVER
);
2167 qemu_sem_wait(&s
->pause_sem
);
2168 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2170 *current_active_state
= new_state
;
2171 qemu_mutex_lock_iothread();
2173 return s
->state
== new_state
? 0 : -EINVAL
;
2177 * migration_completion: Used by migration_thread when there's not much left.
2178 * The caller 'breaks' the loop when this returns.
2180 * @s: Current migration state
2182 static void migration_completion(MigrationState
*s
)
2185 int current_active_state
= s
->state
;
2187 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2188 qemu_mutex_lock_iothread();
2189 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2190 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
2191 s
->vm_was_running
= runstate_is_running();
2192 ret
= global_state_store();
2195 bool inactivate
= !migrate_colo_enabled();
2196 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2198 ret
= migration_maybe_pause(s
, ¤t_active_state
,
2199 MIGRATION_STATUS_DEVICE
);
2202 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
2203 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
2206 if (inactivate
&& ret
>= 0) {
2207 s
->block_inactive
= true;
2210 qemu_mutex_unlock_iothread();
2215 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2216 trace_migration_completion_postcopy_end();
2218 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
2219 trace_migration_completion_postcopy_end_after_complete();
2223 * If rp was opened we must clean up the thread before
2224 * cleaning everything else up (since if there are no failures
2225 * it will wait for the destination to send it's status in
2228 if (s
->rp_state
.from_dst_file
) {
2230 trace_migration_return_path_end_before();
2231 rp_error
= await_return_path_close_on_source(s
);
2232 trace_migration_return_path_end_after(rp_error
);
2234 goto fail_invalidate
;
2238 if (qemu_file_get_error(s
->to_dst_file
)) {
2239 trace_migration_completion_file_err();
2240 goto fail_invalidate
;
2243 if (!migrate_colo_enabled()) {
2244 migrate_set_state(&s
->state
, current_active_state
,
2245 MIGRATION_STATUS_COMPLETED
);
2251 /* If not doing postcopy, vm_start() will be called: let's regain
2252 * control on images.
2254 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
2255 s
->state
== MIGRATION_STATUS_DEVICE
) {
2256 Error
*local_err
= NULL
;
2258 qemu_mutex_lock_iothread();
2259 bdrv_invalidate_cache_all(&local_err
);
2261 error_report_err(local_err
);
2263 s
->block_inactive
= false;
2265 qemu_mutex_unlock_iothread();
2269 migrate_set_state(&s
->state
, current_active_state
,
2270 MIGRATION_STATUS_FAILED
);
2273 bool migrate_colo_enabled(void)
2275 MigrationState
*s
= migrate_get_current();
2276 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
2279 typedef enum MigThrError
{
2280 /* No error detected */
2281 MIG_THR_ERR_NONE
= 0,
2282 /* Detected error, but resumed successfully */
2283 MIG_THR_ERR_RECOVERED
= 1,
2284 /* Detected fatal error, need to exit */
2285 MIG_THR_ERR_FATAL
= 2,
2289 * We don't return until we are in a safe state to continue current
2290 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2291 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2293 static MigThrError
postcopy_pause(MigrationState
*s
)
2295 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2296 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2297 MIGRATION_STATUS_POSTCOPY_PAUSED
);
2299 /* Current channel is possibly broken. Release it. */
2300 assert(s
->to_dst_file
);
2301 qemu_file_shutdown(s
->to_dst_file
);
2302 qemu_fclose(s
->to_dst_file
);
2303 s
->to_dst_file
= NULL
;
2305 error_report("Detected IO failure for postcopy. "
2306 "Migration paused.");
2309 * We wait until things fixed up. Then someone will setup the
2310 * status back for us.
2312 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2313 qemu_sem_wait(&s
->postcopy_pause_sem
);
2316 trace_postcopy_pause_continued();
2318 return MIG_THR_ERR_RECOVERED
;
2321 static MigThrError
migration_detect_error(MigrationState
*s
)
2325 /* Try to detect any file errors */
2326 ret
= qemu_file_get_error(s
->to_dst_file
);
2329 /* Everything is fine */
2330 return MIG_THR_ERR_NONE
;
2333 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
== -EIO
) {
2335 * For postcopy, we allow the network to be down for a
2336 * while. After that, it can be continued by a
2339 return postcopy_pause(s
);
2342 * For precopy (or postcopy with error outside IO), we fail
2345 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
2346 trace_migration_thread_file_err();
2348 /* Time to stop the migration, now. */
2349 return MIG_THR_ERR_FATAL
;
2353 static void migration_calculate_complete(MigrationState
*s
)
2355 uint64_t bytes
= qemu_ftell(s
->to_dst_file
);
2356 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2358 s
->total_time
= end_time
- s
->start_time
;
2361 * It's still not set, so we are precopy migration. For
2362 * postcopy, downtime is calculated during postcopy_start().
2364 s
->downtime
= end_time
- s
->downtime_start
;
2367 if (s
->total_time
) {
2368 s
->mbps
= ((double) bytes
* 8.0) / s
->total_time
/ 1000;
2372 static void migration_update_counters(MigrationState
*s
,
2373 int64_t current_time
)
2375 uint64_t transferred
, time_spent
;
2378 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
2382 transferred
= qemu_ftell(s
->to_dst_file
) - s
->iteration_initial_bytes
;
2383 time_spent
= current_time
- s
->iteration_start_time
;
2384 bandwidth
= (double)transferred
/ time_spent
;
2385 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
2387 s
->mbps
= (((double) transferred
* 8.0) /
2388 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2391 * if we haven't sent anything, we don't want to
2392 * recalculate. 10000 is a small enough number for our purposes
2394 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
2395 s
->expected_downtime
= ram_counters
.dirty_pages_rate
*
2396 qemu_target_page_size() / bandwidth
;
2399 qemu_file_reset_rate_limit(s
->to_dst_file
);
2401 s
->iteration_start_time
= current_time
;
2402 s
->iteration_initial_bytes
= qemu_ftell(s
->to_dst_file
);
2404 trace_migrate_transferred(transferred
, time_spent
,
2405 bandwidth
, s
->threshold_size
);
2408 /* Migration thread iteration status */
2410 MIG_ITERATE_RESUME
, /* Resume current iteration */
2411 MIG_ITERATE_SKIP
, /* Skip current iteration */
2412 MIG_ITERATE_BREAK
, /* Break the loop */
2416 * Return true if continue to the next iteration directly, false
2419 static MigIterateState
migration_iteration_run(MigrationState
*s
)
2421 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
2422 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
2424 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
2425 &pend_compat
, &pend_post
);
2426 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
2428 trace_migrate_pending(pending_size
, s
->threshold_size
,
2429 pend_pre
, pend_compat
, pend_post
);
2431 if (pending_size
&& pending_size
>= s
->threshold_size
) {
2432 /* Still a significant amount to transfer */
2433 if (migrate_postcopy() && !in_postcopy
&&
2434 pend_pre
<= s
->threshold_size
&&
2435 atomic_read(&s
->start_postcopy
)) {
2436 if (postcopy_start(s
)) {
2437 error_report("%s: postcopy failed to start", __func__
);
2439 return MIG_ITERATE_SKIP
;
2441 /* Just another iteration step */
2442 qemu_savevm_state_iterate(s
->to_dst_file
,
2443 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2445 trace_migration_thread_low_pending(pending_size
);
2446 migration_completion(s
);
2447 return MIG_ITERATE_BREAK
;
2450 return MIG_ITERATE_RESUME
;
2453 static void migration_iteration_finish(MigrationState
*s
)
2455 /* If we enabled cpu throttling for auto-converge, turn it off. */
2456 cpu_throttle_stop();
2458 qemu_mutex_lock_iothread();
2460 case MIGRATION_STATUS_COMPLETED
:
2461 migration_calculate_complete(s
);
2462 runstate_set(RUN_STATE_POSTMIGRATE
);
2465 case MIGRATION_STATUS_ACTIVE
:
2467 * We should really assert here, but since it's during
2468 * migration, let's try to reduce the usage of assertions.
2470 if (!migrate_colo_enabled()) {
2471 error_report("%s: critical error: calling COLO code without "
2472 "COLO enabled", __func__
);
2474 migrate_start_colo_process(s
);
2476 * Fixme: we will run VM in COLO no matter its old running state.
2477 * After exited COLO, we will keep running.
2479 s
->vm_was_running
= true;
2481 case MIGRATION_STATUS_FAILED
:
2482 case MIGRATION_STATUS_CANCELLED
:
2483 if (s
->vm_was_running
) {
2486 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2487 runstate_set(RUN_STATE_POSTMIGRATE
);
2493 /* Should not reach here, but if so, forgive the VM. */
2494 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
2497 qemu_bh_schedule(s
->cleanup_bh
);
2498 qemu_mutex_unlock_iothread();
2502 * Master migration thread on the source VM.
2503 * It drives the migration and pumps the data down the outgoing channel.
2505 static void *migration_thread(void *opaque
)
2507 MigrationState
*s
= opaque
;
2508 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
2509 MigThrError thr_error
;
2511 rcu_register_thread();
2513 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2515 qemu_savevm_state_header(s
->to_dst_file
);
2518 * If we opened the return path, we need to make sure dst has it
2521 if (s
->rp_state
.from_dst_file
) {
2522 /* Now tell the dest that it should open its end so it can reply */
2523 qemu_savevm_send_open_return_path(s
->to_dst_file
);
2525 /* And do a ping that will make stuff easier to debug */
2526 qemu_savevm_send_ping(s
->to_dst_file
, 1);
2529 if (migrate_postcopy()) {
2531 * Tell the destination that we *might* want to do postcopy later;
2532 * if the other end can't do postcopy it should fail now, nice and
2535 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
2538 qemu_savevm_state_setup(s
->to_dst_file
);
2540 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
2541 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2542 MIGRATION_STATUS_ACTIVE
);
2544 trace_migration_thread_setup_complete();
2546 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
2547 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2548 int64_t current_time
;
2550 if (!qemu_file_rate_limit(s
->to_dst_file
)) {
2551 MigIterateState iter_state
= migration_iteration_run(s
);
2552 if (iter_state
== MIG_ITERATE_SKIP
) {
2554 } else if (iter_state
== MIG_ITERATE_BREAK
) {
2560 * Try to detect any kind of failures, and see whether we
2561 * should stop the migration now.
2563 thr_error
= migration_detect_error(s
);
2564 if (thr_error
== MIG_THR_ERR_FATAL
) {
2565 /* Stop migration */
2567 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
2569 * Just recovered from a e.g. network failure, reset all
2570 * the local variables. This is important to avoid
2571 * breaking transferred_bytes and bandwidth calculation
2573 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2574 s
->iteration_initial_bytes
= 0;
2577 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2579 migration_update_counters(s
, current_time
);
2581 if (qemu_file_rate_limit(s
->to_dst_file
)) {
2582 /* usleep expects microseconds */
2583 g_usleep((s
->iteration_start_time
+ BUFFER_DELAY
-
2584 current_time
) * 1000);
2588 trace_migration_thread_after_loop();
2589 migration_iteration_finish(s
);
2590 rcu_unregister_thread();
2594 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
2596 s
->expected_downtime
= s
->parameters
.downtime_limit
;
2597 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
2599 migrate_fd_error(s
, error_in
);
2600 migrate_fd_cleanup(s
);
2604 qemu_file_set_blocking(s
->to_dst_file
, true);
2605 qemu_file_set_rate_limit(s
->to_dst_file
,
2606 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
2608 /* Notify before starting migration thread */
2609 notifier_list_notify(&migration_state_notifiers
, s
);
2612 * Open the return path. For postcopy, it is used exclusively. For
2613 * precopy, only if user specified "return-path" capability would
2614 * QEMU uses the return path.
2616 if (migrate_postcopy_ram() || migrate_use_return_path()) {
2617 if (open_return_path_on_source(s
)) {
2618 error_report("Unable to open return-path for postcopy");
2619 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2620 MIGRATION_STATUS_FAILED
);
2621 migrate_fd_cleanup(s
);
2626 if (multifd_save_setup() != 0) {
2627 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2628 MIGRATION_STATUS_FAILED
);
2629 migrate_fd_cleanup(s
);
2632 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
2633 QEMU_THREAD_JOINABLE
);
2634 s
->migration_thread_running
= true;
2637 void migration_global_dump(Monitor
*mon
)
2639 MigrationState
*ms
= migrate_get_current();
2641 monitor_printf(mon
, "globals:\n");
2642 monitor_printf(mon
, "store-global-state: %s\n",
2643 ms
->store_global_state
? "on" : "off");
2644 monitor_printf(mon
, "only-migratable: %s\n",
2645 ms
->only_migratable
? "on" : "off");
2646 monitor_printf(mon
, "send-configuration: %s\n",
2647 ms
->send_configuration
? "on" : "off");
2648 monitor_printf(mon
, "send-section-footer: %s\n",
2649 ms
->send_section_footer
? "on" : "off");
2652 #define DEFINE_PROP_MIG_CAP(name, x) \
2653 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
2655 static Property migration_properties
[] = {
2656 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
2657 store_global_state
, true),
2658 DEFINE_PROP_BOOL("only-migratable", MigrationState
, only_migratable
, false),
2659 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
2660 send_configuration
, true),
2661 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
2662 send_section_footer
, true),
2664 /* Migration parameters */
2665 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
2666 parameters
.compress_level
,
2667 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
2668 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
2669 parameters
.compress_threads
,
2670 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
2671 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
2672 parameters
.decompress_threads
,
2673 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
2674 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
2675 parameters
.cpu_throttle_initial
,
2676 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
2677 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
2678 parameters
.cpu_throttle_increment
,
2679 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
2680 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
2681 parameters
.max_bandwidth
, MAX_THROTTLE
),
2682 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
2683 parameters
.downtime_limit
,
2684 DEFAULT_MIGRATE_SET_DOWNTIME
),
2685 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
2686 parameters
.x_checkpoint_delay
,
2687 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
2688 DEFINE_PROP_UINT8("x-multifd-channels", MigrationState
,
2689 parameters
.x_multifd_channels
,
2690 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
2691 DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState
,
2692 parameters
.x_multifd_page_count
,
2693 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT
),
2694 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
2695 parameters
.xbzrle_cache_size
,
2696 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
2698 /* Migration capabilities */
2699 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
2700 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
2701 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
2702 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
2703 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
2704 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
2705 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
2706 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
2707 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
2708 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
2709 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
2710 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD
),
2712 DEFINE_PROP_END_OF_LIST(),
2715 static void migration_class_init(ObjectClass
*klass
, void *data
)
2717 DeviceClass
*dc
= DEVICE_CLASS(klass
);
2719 dc
->user_creatable
= false;
2720 dc
->props
= migration_properties
;
2723 static void migration_instance_finalize(Object
*obj
)
2725 MigrationState
*ms
= MIGRATION_OBJ(obj
);
2726 MigrationParameters
*params
= &ms
->parameters
;
2728 qemu_mutex_destroy(&ms
->error_mutex
);
2729 g_free(params
->tls_hostname
);
2730 g_free(params
->tls_creds
);
2731 qemu_sem_destroy(&ms
->pause_sem
);
2732 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
2733 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
2734 error_free(ms
->error
);
2737 static void migration_instance_init(Object
*obj
)
2739 MigrationState
*ms
= MIGRATION_OBJ(obj
);
2740 MigrationParameters
*params
= &ms
->parameters
;
2742 ms
->state
= MIGRATION_STATUS_NONE
;
2744 qemu_sem_init(&ms
->pause_sem
, 0);
2745 qemu_mutex_init(&ms
->error_mutex
);
2747 params
->tls_hostname
= g_strdup("");
2748 params
->tls_creds
= g_strdup("");
2750 /* Set has_* up only for parameter checks */
2751 params
->has_compress_level
= true;
2752 params
->has_compress_threads
= true;
2753 params
->has_decompress_threads
= true;
2754 params
->has_cpu_throttle_initial
= true;
2755 params
->has_cpu_throttle_increment
= true;
2756 params
->has_max_bandwidth
= true;
2757 params
->has_downtime_limit
= true;
2758 params
->has_x_checkpoint_delay
= true;
2759 params
->has_block_incremental
= true;
2760 params
->has_x_multifd_channels
= true;
2761 params
->has_x_multifd_page_count
= true;
2762 params
->has_xbzrle_cache_size
= true;
2764 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
2765 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
2769 * Return true if check pass, false otherwise. Error will be put
2770 * inside errp if provided.
2772 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
2774 MigrationCapabilityStatusList
*head
= NULL
;
2775 /* Assuming all off */
2776 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
2779 if (!migrate_params_check(&ms
->parameters
, errp
)) {
2783 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
2784 if (ms
->enabled_capabilities
[i
]) {
2785 head
= migrate_cap_add(head
, i
, true);
2789 ret
= migrate_caps_check(cap_list
, head
, errp
);
2791 /* It works with head == NULL */
2792 qapi_free_MigrationCapabilityStatusList(head
);
2797 static const TypeInfo migration_type
= {
2798 .name
= TYPE_MIGRATION
,
2800 * NOTE: TYPE_MIGRATION is not really a device, as the object is
2801 * not created using qdev_create(), it is not attached to the qdev
2802 * device tree, and it is never realized.
2804 * TODO: Make this TYPE_OBJECT once QOM provides something like
2805 * TYPE_DEVICE's "-global" properties.
2807 .parent
= TYPE_DEVICE
,
2808 .class_init
= migration_class_init
,
2809 .class_size
= sizeof(MigrationClass
),
2810 .instance_size
= sizeof(MigrationState
),
2811 .instance_init
= migration_instance_init
,
2812 .instance_finalize
= migration_instance_finalize
,
2815 static void register_migration_types(void)
2817 type_register_static(&migration_type
);
2820 type_init(register_migration_types
);