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);
163 init_dirty_bitmap_incoming_migration();
170 void migration_incoming_state_destroy(void)
172 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
174 if (mis
->to_src_file
) {
175 /* Tell source that we are done */
176 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
177 qemu_fclose(mis
->to_src_file
);
178 mis
->to_src_file
= NULL
;
181 if (mis
->from_src_file
) {
182 qemu_fclose(mis
->from_src_file
);
183 mis
->from_src_file
= NULL
;
185 if (mis
->postcopy_remote_fds
) {
186 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
187 mis
->postcopy_remote_fds
= NULL
;
190 qemu_event_reset(&mis
->main_thread_load_event
);
193 static void migrate_generate_event(int new_state
)
195 if (migrate_use_events()) {
196 qapi_event_send_migration(new_state
, &error_abort
);
201 * Called on -incoming with a defer: uri.
202 * The migration can be started later after any parameters have been
205 static void deferred_incoming_migration(Error
**errp
)
207 if (deferred_incoming
) {
208 error_setg(errp
, "Incoming migration already deferred");
210 deferred_incoming
= true;
214 * Send a message on the return channel back to the source
217 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
218 enum mig_rp_message_type message_type
,
219 uint16_t len
, void *data
)
223 trace_migrate_send_rp_message((int)message_type
, len
);
224 qemu_mutex_lock(&mis
->rp_mutex
);
227 * It's possible that the file handle got lost due to network
230 if (!mis
->to_src_file
) {
235 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
236 qemu_put_be16(mis
->to_src_file
, len
);
237 qemu_put_buffer(mis
->to_src_file
, data
, len
);
238 qemu_fflush(mis
->to_src_file
);
240 /* It's possible that qemu file got error during sending */
241 ret
= qemu_file_get_error(mis
->to_src_file
);
244 qemu_mutex_unlock(&mis
->rp_mutex
);
248 /* Request a range of pages from the source VM at the given
250 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
251 * as the last request (a name must have been given previously)
252 * Start: Address offset within the RB
253 * Len: Length in bytes required - must be a multiple of pagesize
255 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
, const char *rbname
,
256 ram_addr_t start
, size_t len
)
258 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
259 size_t msglen
= 12; /* start + len */
260 enum mig_rp_message_type msg_type
;
262 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
263 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
266 int rbname_len
= strlen(rbname
);
267 assert(rbname_len
< 256);
269 bufc
[msglen
++] = rbname_len
;
270 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
271 msglen
+= rbname_len
;
272 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
274 msg_type
= MIG_RP_MSG_REQ_PAGES
;
277 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
280 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
284 qapi_event_send_migration(MIGRATION_STATUS_SETUP
, &error_abort
);
285 if (!strcmp(uri
, "defer")) {
286 deferred_incoming_migration(errp
);
287 } else if (strstart(uri
, "tcp:", &p
)) {
288 tcp_start_incoming_migration(p
, errp
);
290 } else if (strstart(uri
, "rdma:", &p
)) {
291 rdma_start_incoming_migration(p
, errp
);
293 } else if (strstart(uri
, "exec:", &p
)) {
294 exec_start_incoming_migration(p
, errp
);
295 } else if (strstart(uri
, "unix:", &p
)) {
296 unix_start_incoming_migration(p
, errp
);
297 } else if (strstart(uri
, "fd:", &p
)) {
298 fd_start_incoming_migration(p
, errp
);
300 error_setg(errp
, "unknown migration protocol: %s", uri
);
304 static void process_incoming_migration_bh(void *opaque
)
306 Error
*local_err
= NULL
;
307 MigrationIncomingState
*mis
= opaque
;
309 /* Only fire up the block code now if we're going to restart the
310 * VM, else 'cont' will do it.
311 * This causes file locking to happen; so we don't want it to happen
312 * unless we really are starting the VM.
314 if (autostart
&& (!global_state_received() ||
315 global_state_get_runstate() == RUN_STATE_RUNNING
)) {
316 /* Make sure all file formats flush their mutable metadata.
317 * If we get an error here, just don't restart the VM yet. */
318 bdrv_invalidate_cache_all(&local_err
);
320 error_report_err(local_err
);
327 * This must happen after all error conditions are dealt with and
328 * we're sure the VM is going to be running on this host.
330 qemu_announce_self();
332 if (multifd_load_cleanup(&local_err
) != 0) {
333 error_report_err(local_err
);
336 /* If global state section was not received or we are in running
337 state, we need to obey autostart. Any other state is set with
340 dirty_bitmap_mig_before_vm_start();
342 if (!global_state_received() ||
343 global_state_get_runstate() == RUN_STATE_RUNNING
) {
347 runstate_set(RUN_STATE_PAUSED
);
350 runstate_set(global_state_get_runstate());
353 * This must happen after any state changes since as soon as an external
354 * observer sees this event they might start to prod at the VM assuming
357 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
358 MIGRATION_STATUS_COMPLETED
);
359 qemu_bh_delete(mis
->bh
);
360 migration_incoming_state_destroy();
363 static void process_incoming_migration_co(void *opaque
)
365 MigrationIncomingState
*mis
= migration_incoming_get_current();
369 assert(mis
->from_src_file
);
370 mis
->largest_page_size
= qemu_ram_pagesize_largest();
371 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
372 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
373 MIGRATION_STATUS_ACTIVE
);
374 ret
= qemu_loadvm_state(mis
->from_src_file
);
376 ps
= postcopy_state_get();
377 trace_process_incoming_migration_co_end(ret
, ps
);
378 if (ps
!= POSTCOPY_INCOMING_NONE
) {
379 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
381 * Where a migration had postcopy enabled (and thus went to advise)
382 * but managed to complete within the precopy period, we can use
385 postcopy_ram_incoming_cleanup(mis
);
386 } else if (ret
>= 0) {
388 * Postcopy was started, cleanup should happen at the end of the
391 trace_process_incoming_migration_co_postcopy_end_main();
394 /* Else if something went wrong then just fall out of the normal exit */
397 /* we get COLO info, and know if we are in COLO mode */
398 if (!ret
&& migration_incoming_enable_colo()) {
399 mis
->migration_incoming_co
= qemu_coroutine_self();
400 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
401 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
402 mis
->have_colo_incoming_thread
= true;
403 qemu_coroutine_yield();
405 /* Wait checkpoint incoming thread exit before free resource */
406 qemu_thread_join(&mis
->colo_incoming_thread
);
410 Error
*local_err
= NULL
;
412 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
413 MIGRATION_STATUS_FAILED
);
414 error_report("load of migration failed: %s", strerror(-ret
));
415 qemu_fclose(mis
->from_src_file
);
416 if (multifd_load_cleanup(&local_err
) != 0) {
417 error_report_err(local_err
);
421 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
422 qemu_bh_schedule(mis
->bh
);
425 static void migration_incoming_setup(QEMUFile
*f
)
427 MigrationIncomingState
*mis
= migration_incoming_get_current();
429 if (multifd_load_setup() != 0) {
430 /* We haven't been able to create multifd threads
431 nothing better to do */
435 if (!mis
->from_src_file
) {
436 mis
->from_src_file
= f
;
438 qemu_file_set_blocking(f
, false);
441 static void migration_incoming_process(void)
443 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
444 qemu_coroutine_enter(co
);
447 void migration_fd_process_incoming(QEMUFile
*f
)
449 migration_incoming_setup(f
);
450 migration_incoming_process();
453 void migration_ioc_process_incoming(QIOChannel
*ioc
)
455 MigrationIncomingState
*mis
= migration_incoming_get_current();
457 if (!mis
->from_src_file
) {
458 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
459 migration_fd_process_incoming(f
);
461 /* We still only have a single channel. Nothing to do here yet */
465 * @migration_has_all_channels: We have received all channels that we need
467 * Returns true when we have got connections to all the channels that
468 * we need for migration.
470 bool migration_has_all_channels(void)
476 * Send a 'SHUT' message on the return channel with the given value
477 * to indicate that we've finished with the RP. Non-0 value indicates
480 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
485 buf
= cpu_to_be32(value
);
486 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
490 * Send a 'PONG' message on the return channel with the given value
491 * (normally in response to a 'PING')
493 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
498 buf
= cpu_to_be32(value
);
499 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
502 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
504 MigrationCapabilityStatusList
*head
= NULL
;
505 MigrationCapabilityStatusList
*caps
;
506 MigrationState
*s
= migrate_get_current();
509 caps
= NULL
; /* silence compiler warning */
510 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
511 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
512 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
517 head
= g_malloc0(sizeof(*caps
));
520 caps
->next
= g_malloc0(sizeof(*caps
));
524 g_malloc(sizeof(*caps
->value
));
525 caps
->value
->capability
= i
;
526 caps
->value
->state
= s
->enabled_capabilities
[i
];
532 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
534 MigrationParameters
*params
;
535 MigrationState
*s
= migrate_get_current();
537 /* TODO use QAPI_CLONE() instead of duplicating it inline */
538 params
= g_malloc0(sizeof(*params
));
539 params
->has_compress_level
= true;
540 params
->compress_level
= s
->parameters
.compress_level
;
541 params
->has_compress_threads
= true;
542 params
->compress_threads
= s
->parameters
.compress_threads
;
543 params
->has_decompress_threads
= true;
544 params
->decompress_threads
= s
->parameters
.decompress_threads
;
545 params
->has_cpu_throttle_initial
= true;
546 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
547 params
->has_cpu_throttle_increment
= true;
548 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
549 params
->has_tls_creds
= true;
550 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
551 params
->has_tls_hostname
= true;
552 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
553 params
->has_max_bandwidth
= true;
554 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
555 params
->has_downtime_limit
= true;
556 params
->downtime_limit
= s
->parameters
.downtime_limit
;
557 params
->has_x_checkpoint_delay
= true;
558 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
559 params
->has_block_incremental
= true;
560 params
->block_incremental
= s
->parameters
.block_incremental
;
561 params
->has_x_multifd_channels
= true;
562 params
->x_multifd_channels
= s
->parameters
.x_multifd_channels
;
563 params
->has_x_multifd_page_count
= true;
564 params
->x_multifd_page_count
= s
->parameters
.x_multifd_page_count
;
565 params
->has_xbzrle_cache_size
= true;
566 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
572 * Return true if we're already in the middle of a migration
573 * (i.e. any of the active or setup states)
575 static bool migration_is_setup_or_active(int state
)
578 case MIGRATION_STATUS_ACTIVE
:
579 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
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 MigrationInfo
*qmp_query_migrate(Error
**errp
)
643 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
644 MigrationState
*s
= migrate_get_current();
647 case MIGRATION_STATUS_NONE
:
648 /* no migration has happened ever */
650 case MIGRATION_STATUS_SETUP
:
651 info
->has_status
= true;
652 info
->has_total_time
= false;
654 case MIGRATION_STATUS_ACTIVE
:
655 case MIGRATION_STATUS_CANCELLING
:
656 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
657 case MIGRATION_STATUS_PRE_SWITCHOVER
:
658 case MIGRATION_STATUS_DEVICE
:
659 /* TODO add some postcopy stats */
660 info
->has_status
= true;
661 info
->has_total_time
= true;
662 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
664 info
->has_expected_downtime
= true;
665 info
->expected_downtime
= s
->expected_downtime
;
666 info
->has_setup_time
= true;
667 info
->setup_time
= s
->setup_time
;
669 populate_ram_info(info
, s
);
670 populate_disk_info(info
);
672 case MIGRATION_STATUS_COLO
:
673 info
->has_status
= true;
674 /* TODO: display COLO specific information (checkpoint info etc.) */
676 case MIGRATION_STATUS_COMPLETED
:
677 info
->has_status
= true;
678 info
->has_total_time
= true;
679 info
->total_time
= s
->total_time
;
680 info
->has_downtime
= true;
681 info
->downtime
= s
->downtime
;
682 info
->has_setup_time
= true;
683 info
->setup_time
= s
->setup_time
;
685 populate_ram_info(info
, s
);
687 case MIGRATION_STATUS_FAILED
:
688 info
->has_status
= true;
690 info
->has_error_desc
= true;
691 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
694 case MIGRATION_STATUS_CANCELLED
:
695 info
->has_status
= true;
698 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 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
767 MigrationState
*s
= migrate_get_current();
768 MigrationCapabilityStatusList
*cap
;
769 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
771 if (migration_is_setup_or_active(s
->state
)) {
772 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
776 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
777 if (!migrate_caps_check(cap_list
, params
, errp
)) {
781 for (cap
= params
; cap
; cap
= cap
->next
) {
782 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
787 * Check whether the parameters are valid. Error will be put into errp
788 * (if provided). Return true if valid, otherwise false.
790 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
792 if (params
->has_compress_level
&&
793 (params
->compress_level
> 9)) {
794 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
795 "is invalid, it should be in the range of 0 to 9");
799 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
800 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
802 "is invalid, it should be in the range of 1 to 255");
806 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
807 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
808 "decompress_threads",
809 "is invalid, it should be in the range of 1 to 255");
813 if (params
->has_cpu_throttle_initial
&&
814 (params
->cpu_throttle_initial
< 1 ||
815 params
->cpu_throttle_initial
> 99)) {
816 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
817 "cpu_throttle_initial",
818 "an integer in the range of 1 to 99");
822 if (params
->has_cpu_throttle_increment
&&
823 (params
->cpu_throttle_increment
< 1 ||
824 params
->cpu_throttle_increment
> 99)) {
825 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
826 "cpu_throttle_increment",
827 "an integer in the range of 1 to 99");
831 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
832 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
833 " range of 0 to %zu bytes/second", SIZE_MAX
);
837 if (params
->has_downtime_limit
&&
838 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
839 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
840 "the range of 0 to %d milliseconds",
841 MAX_MIGRATE_DOWNTIME
);
845 /* x_checkpoint_delay is now always positive */
847 if (params
->has_x_multifd_channels
&& (params
->x_multifd_channels
< 1)) {
848 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
850 "is invalid, it should be in the range of 1 to 255");
853 if (params
->has_x_multifd_page_count
&&
854 (params
->x_multifd_page_count
< 1 ||
855 params
->x_multifd_page_count
> 10000)) {
856 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
857 "multifd_page_count",
858 "is invalid, it should be in the range of 1 to 10000");
862 if (params
->has_xbzrle_cache_size
&&
863 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
864 !is_power_of_2(params
->xbzrle_cache_size
))) {
865 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
867 "is invalid, it should be bigger than target page size"
868 " and a power of two");
875 static void migrate_params_test_apply(MigrateSetParameters
*params
,
876 MigrationParameters
*dest
)
878 *dest
= migrate_get_current()->parameters
;
880 /* TODO use QAPI_CLONE() instead of duplicating it inline */
882 if (params
->has_compress_level
) {
883 dest
->compress_level
= params
->compress_level
;
886 if (params
->has_compress_threads
) {
887 dest
->compress_threads
= params
->compress_threads
;
890 if (params
->has_decompress_threads
) {
891 dest
->decompress_threads
= params
->decompress_threads
;
894 if (params
->has_cpu_throttle_initial
) {
895 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
898 if (params
->has_cpu_throttle_increment
) {
899 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
902 if (params
->has_tls_creds
) {
903 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
904 dest
->tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
907 if (params
->has_tls_hostname
) {
908 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
909 dest
->tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
912 if (params
->has_max_bandwidth
) {
913 dest
->max_bandwidth
= params
->max_bandwidth
;
916 if (params
->has_downtime_limit
) {
917 dest
->downtime_limit
= params
->downtime_limit
;
920 if (params
->has_x_checkpoint_delay
) {
921 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
924 if (params
->has_block_incremental
) {
925 dest
->block_incremental
= params
->block_incremental
;
927 if (params
->has_x_multifd_channels
) {
928 dest
->x_multifd_channels
= params
->x_multifd_channels
;
930 if (params
->has_x_multifd_page_count
) {
931 dest
->x_multifd_page_count
= params
->x_multifd_page_count
;
933 if (params
->has_xbzrle_cache_size
) {
934 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
938 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
940 MigrationState
*s
= migrate_get_current();
942 /* TODO use QAPI_CLONE() instead of duplicating it inline */
944 if (params
->has_compress_level
) {
945 s
->parameters
.compress_level
= params
->compress_level
;
948 if (params
->has_compress_threads
) {
949 s
->parameters
.compress_threads
= params
->compress_threads
;
952 if (params
->has_decompress_threads
) {
953 s
->parameters
.decompress_threads
= params
->decompress_threads
;
956 if (params
->has_cpu_throttle_initial
) {
957 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
960 if (params
->has_cpu_throttle_increment
) {
961 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
964 if (params
->has_tls_creds
) {
965 g_free(s
->parameters
.tls_creds
);
966 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
967 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
970 if (params
->has_tls_hostname
) {
971 g_free(s
->parameters
.tls_hostname
);
972 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
973 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
976 if (params
->has_max_bandwidth
) {
977 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
978 if (s
->to_dst_file
) {
979 qemu_file_set_rate_limit(s
->to_dst_file
,
980 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
984 if (params
->has_downtime_limit
) {
985 s
->parameters
.downtime_limit
= params
->downtime_limit
;
988 if (params
->has_x_checkpoint_delay
) {
989 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
990 if (migration_in_colo_state()) {
991 colo_checkpoint_notify(s
);
995 if (params
->has_block_incremental
) {
996 s
->parameters
.block_incremental
= params
->block_incremental
;
998 if (params
->has_x_multifd_channels
) {
999 s
->parameters
.x_multifd_channels
= params
->x_multifd_channels
;
1001 if (params
->has_x_multifd_page_count
) {
1002 s
->parameters
.x_multifd_page_count
= params
->x_multifd_page_count
;
1004 if (params
->has_xbzrle_cache_size
) {
1005 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1006 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1010 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1012 MigrationParameters tmp
;
1014 /* TODO Rewrite "" to null instead */
1015 if (params
->has_tls_creds
1016 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1017 QDECREF(params
->tls_creds
->u
.n
);
1018 params
->tls_creds
->type
= QTYPE_QSTRING
;
1019 params
->tls_creds
->u
.s
= strdup("");
1021 /* TODO Rewrite "" to null instead */
1022 if (params
->has_tls_hostname
1023 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1024 QDECREF(params
->tls_hostname
->u
.n
);
1025 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1026 params
->tls_hostname
->u
.s
= strdup("");
1029 migrate_params_test_apply(params
, &tmp
);
1031 if (!migrate_params_check(&tmp
, errp
)) {
1032 /* Invalid parameter */
1036 migrate_params_apply(params
, errp
);
1040 void qmp_migrate_start_postcopy(Error
**errp
)
1042 MigrationState
*s
= migrate_get_current();
1044 if (!migrate_postcopy()) {
1045 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1046 " the start of migration");
1050 if (s
->state
== MIGRATION_STATUS_NONE
) {
1051 error_setg(errp
, "Postcopy must be started after migration has been"
1056 * we don't error if migration has finished since that would be racy
1057 * with issuing this command.
1059 atomic_set(&s
->start_postcopy
, true);
1062 /* shared migration helpers */
1064 void migrate_set_state(int *state
, int old_state
, int new_state
)
1066 assert(new_state
< MIGRATION_STATUS__MAX
);
1067 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1068 trace_migrate_set_state(MigrationStatus_str(new_state
));
1069 migrate_generate_event(new_state
);
1073 static MigrationCapabilityStatusList
*migrate_cap_add(
1074 MigrationCapabilityStatusList
*list
,
1075 MigrationCapability index
,
1078 MigrationCapabilityStatusList
*cap
;
1080 cap
= g_new0(MigrationCapabilityStatusList
, 1);
1081 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
1082 cap
->value
->capability
= index
;
1083 cap
->value
->state
= state
;
1089 void migrate_set_block_enabled(bool value
, Error
**errp
)
1091 MigrationCapabilityStatusList
*cap
;
1093 cap
= migrate_cap_add(NULL
, MIGRATION_CAPABILITY_BLOCK
, value
);
1094 qmp_migrate_set_capabilities(cap
, errp
);
1095 qapi_free_MigrationCapabilityStatusList(cap
);
1098 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1100 s
->parameters
.block_incremental
= value
;
1103 static void block_cleanup_parameters(MigrationState
*s
)
1105 if (s
->must_remove_block_options
) {
1106 /* setting to false can never fail */
1107 migrate_set_block_enabled(false, &error_abort
);
1108 migrate_set_block_incremental(s
, false);
1109 s
->must_remove_block_options
= false;
1113 static void migrate_fd_cleanup(void *opaque
)
1115 MigrationState
*s
= opaque
;
1117 qemu_bh_delete(s
->cleanup_bh
);
1118 s
->cleanup_bh
= NULL
;
1120 qemu_savevm_state_cleanup();
1122 if (s
->to_dst_file
) {
1123 Error
*local_err
= NULL
;
1125 trace_migrate_fd_cleanup();
1126 qemu_mutex_unlock_iothread();
1127 if (s
->migration_thread_running
) {
1128 qemu_thread_join(&s
->thread
);
1129 s
->migration_thread_running
= false;
1131 qemu_mutex_lock_iothread();
1133 if (multifd_save_cleanup(&local_err
) != 0) {
1134 error_report_err(local_err
);
1136 qemu_fclose(s
->to_dst_file
);
1137 s
->to_dst_file
= NULL
;
1140 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
1141 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
1143 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1144 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1145 MIGRATION_STATUS_CANCELLED
);
1149 /* It is used on info migrate. We can't free it */
1150 error_report_err(error_copy(s
->error
));
1152 notifier_list_notify(&migration_state_notifiers
, s
);
1153 block_cleanup_parameters(s
);
1156 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1158 qemu_mutex_lock(&s
->error_mutex
);
1160 s
->error
= error_copy(error
);
1162 qemu_mutex_unlock(&s
->error_mutex
);
1165 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1167 trace_migrate_fd_error(error_get_pretty(error
));
1168 assert(s
->to_dst_file
== NULL
);
1169 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1170 MIGRATION_STATUS_FAILED
);
1171 migrate_set_error(s
, error
);
1174 static void migrate_fd_cancel(MigrationState
*s
)
1177 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1178 trace_migrate_fd_cancel();
1180 if (s
->rp_state
.from_dst_file
) {
1181 /* shutdown the rp socket, so causing the rp thread to shutdown */
1182 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1186 old_state
= s
->state
;
1187 if (!migration_is_setup_or_active(old_state
)) {
1190 /* If the migration is paused, kick it out of the pause */
1191 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1192 qemu_sem_post(&s
->pause_sem
);
1194 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1195 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1198 * If we're unlucky the migration code might be stuck somewhere in a
1199 * send/write while the network has failed and is waiting to timeout;
1200 * if we've got shutdown(2) available then we can force it to quit.
1201 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1202 * called in a bh, so there is no race against this cancel.
1204 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1205 qemu_file_shutdown(f
);
1207 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1208 Error
*local_err
= NULL
;
1210 bdrv_invalidate_cache_all(&local_err
);
1212 error_report_err(local_err
);
1214 s
->block_inactive
= false;
1219 void add_migration_state_change_notifier(Notifier
*notify
)
1221 notifier_list_add(&migration_state_notifiers
, notify
);
1224 void remove_migration_state_change_notifier(Notifier
*notify
)
1226 notifier_remove(notify
);
1229 bool migration_in_setup(MigrationState
*s
)
1231 return s
->state
== MIGRATION_STATUS_SETUP
;
1234 bool migration_has_finished(MigrationState
*s
)
1236 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1239 bool migration_has_failed(MigrationState
*s
)
1241 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1242 s
->state
== MIGRATION_STATUS_FAILED
);
1245 bool migration_in_postcopy(void)
1247 MigrationState
*s
= migrate_get_current();
1249 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1252 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1254 return migration_in_postcopy() && s
->postcopy_after_devices
;
1257 bool migration_is_idle(void)
1259 MigrationState
*s
= migrate_get_current();
1262 case MIGRATION_STATUS_NONE
:
1263 case MIGRATION_STATUS_CANCELLED
:
1264 case MIGRATION_STATUS_COMPLETED
:
1265 case MIGRATION_STATUS_FAILED
:
1267 case MIGRATION_STATUS_SETUP
:
1268 case MIGRATION_STATUS_CANCELLING
:
1269 case MIGRATION_STATUS_ACTIVE
:
1270 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1271 case MIGRATION_STATUS_COLO
:
1272 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1273 case MIGRATION_STATUS_DEVICE
:
1275 case MIGRATION_STATUS__MAX
:
1276 g_assert_not_reached();
1282 void migrate_init(MigrationState
*s
)
1285 * Reinitialise all migration state, except
1286 * parameters/capabilities that the user set, and
1292 s
->to_dst_file
= NULL
;
1293 s
->state
= MIGRATION_STATUS_NONE
;
1294 s
->rp_state
.from_dst_file
= NULL
;
1295 s
->rp_state
.error
= false;
1298 s
->expected_downtime
= 0;
1300 s
->start_postcopy
= false;
1301 s
->postcopy_after_devices
= false;
1302 s
->migration_thread_running
= false;
1303 error_free(s
->error
);
1306 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1308 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1310 s
->vm_was_running
= false;
1311 s
->iteration_initial_bytes
= 0;
1312 s
->threshold_size
= 0;
1315 static GSList
*migration_blockers
;
1317 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1319 if (migrate_get_current()->only_migratable
) {
1320 error_propagate(errp
, error_copy(reason
));
1321 error_prepend(errp
, "disallowing migration blocker "
1322 "(--only_migratable) for: ");
1326 if (migration_is_idle()) {
1327 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1331 error_propagate(errp
, error_copy(reason
));
1332 error_prepend(errp
, "disallowing migration blocker (migration in "
1337 void migrate_del_blocker(Error
*reason
)
1339 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1342 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1344 Error
*local_err
= NULL
;
1345 static bool once
= true;
1347 if (!deferred_incoming
) {
1348 error_setg(errp
, "For use with '-incoming defer'");
1352 error_setg(errp
, "The incoming migration has already been started");
1355 qemu_start_incoming_migration(uri
, &local_err
);
1358 error_propagate(errp
, local_err
);
1365 bool migration_is_blocked(Error
**errp
)
1367 if (qemu_savevm_state_blocked(errp
)) {
1371 if (migration_blockers
) {
1372 error_propagate(errp
, error_copy(migration_blockers
->data
));
1379 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1380 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1383 Error
*local_err
= NULL
;
1384 MigrationState
*s
= migrate_get_current();
1387 if (migration_is_setup_or_active(s
->state
) ||
1388 s
->state
== MIGRATION_STATUS_CANCELLING
||
1389 s
->state
== MIGRATION_STATUS_COLO
) {
1390 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1393 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1394 error_setg(errp
, "Guest is waiting for an incoming migration");
1398 if (migration_is_blocked(errp
)) {
1402 if ((has_blk
&& blk
) || (has_inc
&& inc
)) {
1403 if (migrate_use_block() || migrate_use_block_incremental()) {
1404 error_setg(errp
, "Command options are incompatible with "
1405 "current migration capabilities");
1408 migrate_set_block_enabled(true, &local_err
);
1410 error_propagate(errp
, local_err
);
1413 s
->must_remove_block_options
= true;
1416 if (has_inc
&& inc
) {
1417 migrate_set_block_incremental(s
, true);
1422 if (strstart(uri
, "tcp:", &p
)) {
1423 tcp_start_outgoing_migration(s
, p
, &local_err
);
1425 } else if (strstart(uri
, "rdma:", &p
)) {
1426 rdma_start_outgoing_migration(s
, p
, &local_err
);
1428 } else if (strstart(uri
, "exec:", &p
)) {
1429 exec_start_outgoing_migration(s
, p
, &local_err
);
1430 } else if (strstart(uri
, "unix:", &p
)) {
1431 unix_start_outgoing_migration(s
, p
, &local_err
);
1432 } else if (strstart(uri
, "fd:", &p
)) {
1433 fd_start_outgoing_migration(s
, p
, &local_err
);
1435 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1436 "a valid migration protocol");
1437 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1438 MIGRATION_STATUS_FAILED
);
1439 block_cleanup_parameters(s
);
1444 migrate_fd_error(s
, local_err
);
1445 error_propagate(errp
, local_err
);
1450 void qmp_migrate_cancel(Error
**errp
)
1452 migrate_fd_cancel(migrate_get_current());
1455 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
1457 MigrationState
*s
= migrate_get_current();
1458 if (s
->state
!= state
) {
1459 error_setg(errp
, "Migration not in expected state: %s",
1460 MigrationStatus_str(s
->state
));
1463 qemu_sem_post(&s
->pause_sem
);
1466 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1468 MigrateSetParameters p
= {
1469 .has_xbzrle_cache_size
= true,
1470 .xbzrle_cache_size
= value
,
1473 qmp_migrate_set_parameters(&p
, errp
);
1476 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1478 return migrate_xbzrle_cache_size();
1481 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1483 MigrateSetParameters p
= {
1484 .has_max_bandwidth
= true,
1485 .max_bandwidth
= value
,
1488 qmp_migrate_set_parameters(&p
, errp
);
1491 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1493 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
1494 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1495 "the range of 0 to %d seconds",
1496 MAX_MIGRATE_DOWNTIME_SECONDS
);
1500 value
*= 1000; /* Convert to milliseconds */
1501 value
= MAX(0, MIN(INT64_MAX
, value
));
1503 MigrateSetParameters p
= {
1504 .has_downtime_limit
= true,
1505 .downtime_limit
= value
,
1508 qmp_migrate_set_parameters(&p
, errp
);
1511 bool migrate_release_ram(void)
1515 s
= migrate_get_current();
1517 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
1520 bool migrate_postcopy_ram(void)
1524 s
= migrate_get_current();
1526 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1529 bool migrate_postcopy(void)
1531 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
1534 bool migrate_auto_converge(void)
1538 s
= migrate_get_current();
1540 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1543 bool migrate_zero_blocks(void)
1547 s
= migrate_get_current();
1549 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
1552 bool migrate_use_compression(void)
1556 s
= migrate_get_current();
1558 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
1561 int migrate_compress_level(void)
1565 s
= migrate_get_current();
1567 return s
->parameters
.compress_level
;
1570 int migrate_compress_threads(void)
1574 s
= migrate_get_current();
1576 return s
->parameters
.compress_threads
;
1579 int migrate_decompress_threads(void)
1583 s
= migrate_get_current();
1585 return s
->parameters
.decompress_threads
;
1588 bool migrate_dirty_bitmaps(void)
1592 s
= migrate_get_current();
1594 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
1597 bool migrate_use_events(void)
1601 s
= migrate_get_current();
1603 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
1606 bool migrate_use_multifd(void)
1610 s
= migrate_get_current();
1612 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_MULTIFD
];
1615 bool migrate_pause_before_switchover(void)
1619 s
= migrate_get_current();
1621 return s
->enabled_capabilities
[
1622 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
1625 int migrate_multifd_channels(void)
1629 s
= migrate_get_current();
1631 return s
->parameters
.x_multifd_channels
;
1634 int migrate_multifd_page_count(void)
1638 s
= migrate_get_current();
1640 return s
->parameters
.x_multifd_page_count
;
1643 int migrate_use_xbzrle(void)
1647 s
= migrate_get_current();
1649 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
1652 int64_t migrate_xbzrle_cache_size(void)
1656 s
= migrate_get_current();
1658 return s
->parameters
.xbzrle_cache_size
;
1661 bool migrate_use_block(void)
1665 s
= migrate_get_current();
1667 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
1670 bool migrate_use_return_path(void)
1674 s
= migrate_get_current();
1676 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
1679 bool migrate_use_block_incremental(void)
1683 s
= migrate_get_current();
1685 return s
->parameters
.block_incremental
;
1688 /* migration thread support */
1690 * Something bad happened to the RP stream, mark an error
1691 * The caller shall print or trace something to indicate why
1693 static void mark_source_rp_bad(MigrationState
*s
)
1695 s
->rp_state
.error
= true;
1698 static struct rp_cmd_args
{
1699 ssize_t len
; /* -1 = variable */
1702 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1703 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1704 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1705 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1706 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1707 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1711 * Process a request for pages received on the return path,
1712 * We're allowed to send more than requested (e.g. to round to our page size)
1713 * and we don't need to send pages that have already been sent.
1715 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1716 ram_addr_t start
, size_t len
)
1718 long our_host_ps
= getpagesize();
1720 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1723 * Since we currently insist on matching page sizes, just sanity check
1724 * we're being asked for whole host pages.
1726 if (start
& (our_host_ps
-1) ||
1727 (len
& (our_host_ps
-1))) {
1728 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1729 " len: %zd", __func__
, start
, len
);
1730 mark_source_rp_bad(ms
);
1734 if (ram_save_queue_pages(rbname
, start
, len
)) {
1735 mark_source_rp_bad(ms
);
1740 * Handles messages sent on the return path towards the source VM
1743 static void *source_return_path_thread(void *opaque
)
1745 MigrationState
*ms
= opaque
;
1746 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1747 uint16_t header_len
, header_type
;
1749 uint32_t tmp32
, sibling_error
;
1750 ram_addr_t start
= 0; /* =0 to silence warning */
1751 size_t len
= 0, expected_len
;
1754 trace_source_return_path_thread_entry();
1755 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1756 migration_is_setup_or_active(ms
->state
)) {
1757 trace_source_return_path_thread_loop_top();
1758 header_type
= qemu_get_be16(rp
);
1759 header_len
= qemu_get_be16(rp
);
1761 if (qemu_file_get_error(rp
)) {
1762 mark_source_rp_bad(ms
);
1766 if (header_type
>= MIG_RP_MSG_MAX
||
1767 header_type
== MIG_RP_MSG_INVALID
) {
1768 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1769 header_type
, header_len
);
1770 mark_source_rp_bad(ms
);
1774 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1775 header_len
!= rp_cmd_args
[header_type
].len
) ||
1776 header_len
> sizeof(buf
)) {
1777 error_report("RP: Received '%s' message (0x%04x) with"
1778 "incorrect length %d expecting %zu",
1779 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1780 (size_t)rp_cmd_args
[header_type
].len
);
1781 mark_source_rp_bad(ms
);
1785 /* We know we've got a valid header by this point */
1786 res
= qemu_get_buffer(rp
, buf
, header_len
);
1787 if (res
!= header_len
) {
1788 error_report("RP: Failed reading data for message 0x%04x"
1789 " read %d expected %d",
1790 header_type
, res
, header_len
);
1791 mark_source_rp_bad(ms
);
1795 /* OK, we have the message and the data */
1796 switch (header_type
) {
1797 case MIG_RP_MSG_SHUT
:
1798 sibling_error
= ldl_be_p(buf
);
1799 trace_source_return_path_thread_shut(sibling_error
);
1800 if (sibling_error
) {
1801 error_report("RP: Sibling indicated error %d", sibling_error
);
1802 mark_source_rp_bad(ms
);
1805 * We'll let the main thread deal with closing the RP
1806 * we could do a shutdown(2) on it, but we're the only user
1807 * anyway, so there's nothing gained.
1811 case MIG_RP_MSG_PONG
:
1812 tmp32
= ldl_be_p(buf
);
1813 trace_source_return_path_thread_pong(tmp32
);
1816 case MIG_RP_MSG_REQ_PAGES
:
1817 start
= ldq_be_p(buf
);
1818 len
= ldl_be_p(buf
+ 8);
1819 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1822 case MIG_RP_MSG_REQ_PAGES_ID
:
1823 expected_len
= 12 + 1; /* header + termination */
1825 if (header_len
>= expected_len
) {
1826 start
= ldq_be_p(buf
);
1827 len
= ldl_be_p(buf
+ 8);
1828 /* Now we expect an idstr */
1829 tmp32
= buf
[12]; /* Length of the following idstr */
1830 buf
[13 + tmp32
] = '\0';
1831 expected_len
+= tmp32
;
1833 if (header_len
!= expected_len
) {
1834 error_report("RP: Req_Page_id with length %d expecting %zd",
1835 header_len
, expected_len
);
1836 mark_source_rp_bad(ms
);
1839 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
1846 if (qemu_file_get_error(rp
)) {
1847 trace_source_return_path_thread_bad_end();
1848 mark_source_rp_bad(ms
);
1851 trace_source_return_path_thread_end();
1853 ms
->rp_state
.from_dst_file
= NULL
;
1858 static int open_return_path_on_source(MigrationState
*ms
)
1861 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
1862 if (!ms
->rp_state
.from_dst_file
) {
1866 trace_open_return_path_on_source();
1867 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
1868 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
1870 trace_open_return_path_on_source_continue();
1875 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1876 static int await_return_path_close_on_source(MigrationState
*ms
)
1879 * If this is a normal exit then the destination will send a SHUT and the
1880 * rp_thread will exit, however if there's an error we need to cause
1883 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
1885 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1886 * waiting for the destination.
1888 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
1889 mark_source_rp_bad(ms
);
1891 trace_await_return_path_close_on_source_joining();
1892 qemu_thread_join(&ms
->rp_state
.rp_thread
);
1893 trace_await_return_path_close_on_source_close();
1894 return ms
->rp_state
.error
;
1898 * Switch from normal iteration to postcopy
1899 * Returns non-0 on error
1901 static int postcopy_start(MigrationState
*ms
)
1904 QIOChannelBuffer
*bioc
;
1906 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1907 bool restart_block
= false;
1908 int cur_state
= MIGRATION_STATUS_ACTIVE
;
1909 if (!migrate_pause_before_switchover()) {
1910 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
1911 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1914 trace_postcopy_start();
1915 qemu_mutex_lock_iothread();
1916 trace_postcopy_start_set_run();
1918 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1919 global_state_store();
1920 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1925 ret
= migration_maybe_pause(ms
, &cur_state
,
1926 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1931 ret
= bdrv_inactivate_all();
1935 restart_block
= true;
1938 * Cause any non-postcopiable, but iterative devices to
1939 * send out their final data.
1941 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
1944 * in Finish migrate and with the io-lock held everything should
1945 * be quiet, but we've potentially still got dirty pages and we
1946 * need to tell the destination to throw any pages it's already received
1949 if (migrate_postcopy_ram()) {
1950 if (ram_postcopy_send_discard_bitmap(ms
)) {
1951 error_report("postcopy send discard bitmap failed");
1957 * send rest of state - note things that are doing postcopy
1958 * will notice we're in POSTCOPY_ACTIVE and not actually
1959 * wrap their state up here
1961 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
1962 if (migrate_postcopy_ram()) {
1963 /* Ping just for debugging, helps line traces up */
1964 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
1968 * While loading the device state we may trigger page transfer
1969 * requests and the fd must be free to process those, and thus
1970 * the destination must read the whole device state off the fd before
1971 * it starts processing it. Unfortunately the ad-hoc migration format
1972 * doesn't allow the destination to know the size to read without fully
1973 * parsing it through each devices load-state code (especially the open
1974 * coded devices that use get/put).
1975 * So we wrap the device state up in a package with a length at the start;
1976 * to do this we use a qemu_buf to hold the whole of the device state.
1978 bioc
= qio_channel_buffer_new(4096);
1979 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
1980 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
1981 object_unref(OBJECT(bioc
));
1984 * Make sure the receiver can get incoming pages before we send the rest
1987 qemu_savevm_send_postcopy_listen(fb
);
1989 qemu_savevm_state_complete_precopy(fb
, false, false);
1990 if (migrate_postcopy_ram()) {
1991 qemu_savevm_send_ping(fb
, 3);
1994 qemu_savevm_send_postcopy_run(fb
);
1996 /* <><> end of stuff going into the package */
1998 /* Last point of recovery; as soon as we send the package the destination
1999 * can open devices and potentially start running.
2000 * Lets just check again we've not got any errors.
2002 ret
= qemu_file_get_error(ms
->to_dst_file
);
2004 error_report("postcopy_start: Migration stream errored (pre package)");
2008 restart_block
= false;
2010 /* Now send that blob */
2011 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2016 /* Send a notify to give a chance for anything that needs to happen
2017 * at the transition to postcopy and after the device state; in particular
2018 * spice needs to trigger a transition now
2020 ms
->postcopy_after_devices
= true;
2021 notifier_list_notify(&migration_state_notifiers
, ms
);
2023 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2025 qemu_mutex_unlock_iothread();
2027 if (migrate_postcopy_ram()) {
2029 * Although this ping is just for debug, it could potentially be
2030 * used for getting a better measurement of downtime at the source.
2032 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2035 if (migrate_release_ram()) {
2036 ram_postcopy_migrated_memory_release(ms
);
2039 ret
= qemu_file_get_error(ms
->to_dst_file
);
2041 error_report("postcopy_start: Migration stream errored");
2042 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2043 MIGRATION_STATUS_FAILED
);
2051 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2052 MIGRATION_STATUS_FAILED
);
2053 if (restart_block
) {
2054 /* A failure happened early enough that we know the destination hasn't
2055 * accessed block devices, so we're safe to recover.
2057 Error
*local_err
= NULL
;
2059 bdrv_invalidate_cache_all(&local_err
);
2061 error_report_err(local_err
);
2064 qemu_mutex_unlock_iothread();
2069 * migration_maybe_pause: Pause if required to by
2070 * migrate_pause_before_switchover called with the iothread locked
2071 * Returns: 0 on success
2073 static int migration_maybe_pause(MigrationState
*s
,
2074 int *current_active_state
,
2077 if (!migrate_pause_before_switchover()) {
2081 /* Since leaving this state is not atomic with posting the semaphore
2082 * it's possible that someone could have issued multiple migrate_continue
2083 * and the semaphore is incorrectly positive at this point;
2084 * the docs say it's undefined to reinit a semaphore that's already
2085 * init'd, so use timedwait to eat up any existing posts.
2087 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2088 /* This block intentionally left blank */
2091 qemu_mutex_unlock_iothread();
2092 migrate_set_state(&s
->state
, *current_active_state
,
2093 MIGRATION_STATUS_PRE_SWITCHOVER
);
2094 qemu_sem_wait(&s
->pause_sem
);
2095 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2097 *current_active_state
= new_state
;
2098 qemu_mutex_lock_iothread();
2100 return s
->state
== new_state
? 0 : -EINVAL
;
2104 * migration_completion: Used by migration_thread when there's not much left.
2105 * The caller 'breaks' the loop when this returns.
2107 * @s: Current migration state
2109 static void migration_completion(MigrationState
*s
)
2112 int current_active_state
= s
->state
;
2114 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2115 qemu_mutex_lock_iothread();
2116 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2117 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
2118 s
->vm_was_running
= runstate_is_running();
2119 ret
= global_state_store();
2122 bool inactivate
= !migrate_colo_enabled();
2123 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2125 ret
= migration_maybe_pause(s
, ¤t_active_state
,
2126 MIGRATION_STATUS_DEVICE
);
2129 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
2130 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
2133 if (inactivate
&& ret
>= 0) {
2134 s
->block_inactive
= true;
2137 qemu_mutex_unlock_iothread();
2142 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2143 trace_migration_completion_postcopy_end();
2145 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
2146 trace_migration_completion_postcopy_end_after_complete();
2150 * If rp was opened we must clean up the thread before
2151 * cleaning everything else up (since if there are no failures
2152 * it will wait for the destination to send it's status in
2155 if (s
->rp_state
.from_dst_file
) {
2157 trace_migration_return_path_end_before();
2158 rp_error
= await_return_path_close_on_source(s
);
2159 trace_migration_return_path_end_after(rp_error
);
2161 goto fail_invalidate
;
2165 if (qemu_file_get_error(s
->to_dst_file
)) {
2166 trace_migration_completion_file_err();
2167 goto fail_invalidate
;
2170 if (!migrate_colo_enabled()) {
2171 migrate_set_state(&s
->state
, current_active_state
,
2172 MIGRATION_STATUS_COMPLETED
);
2178 /* If not doing postcopy, vm_start() will be called: let's regain
2179 * control on images.
2181 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
2182 s
->state
== MIGRATION_STATUS_DEVICE
) {
2183 Error
*local_err
= NULL
;
2185 qemu_mutex_lock_iothread();
2186 bdrv_invalidate_cache_all(&local_err
);
2188 error_report_err(local_err
);
2190 s
->block_inactive
= false;
2192 qemu_mutex_unlock_iothread();
2196 migrate_set_state(&s
->state
, current_active_state
,
2197 MIGRATION_STATUS_FAILED
);
2200 bool migrate_colo_enabled(void)
2202 MigrationState
*s
= migrate_get_current();
2203 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
2206 static void migration_calculate_complete(MigrationState
*s
)
2208 uint64_t bytes
= qemu_ftell(s
->to_dst_file
);
2209 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2211 s
->total_time
= end_time
- s
->start_time
;
2214 * It's still not set, so we are precopy migration. For
2215 * postcopy, downtime is calculated during postcopy_start().
2217 s
->downtime
= end_time
- s
->downtime_start
;
2220 if (s
->total_time
) {
2221 s
->mbps
= ((double) bytes
* 8.0) / s
->total_time
/ 1000;
2225 static void migration_update_counters(MigrationState
*s
,
2226 int64_t current_time
)
2228 uint64_t transferred
, time_spent
;
2231 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
2235 transferred
= qemu_ftell(s
->to_dst_file
) - s
->iteration_initial_bytes
;
2236 time_spent
= current_time
- s
->iteration_start_time
;
2237 bandwidth
= (double)transferred
/ time_spent
;
2238 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
2240 s
->mbps
= (((double) transferred
* 8.0) /
2241 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2244 * if we haven't sent anything, we don't want to
2245 * recalculate. 10000 is a small enough number for our purposes
2247 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
2248 s
->expected_downtime
= ram_counters
.dirty_pages_rate
*
2249 qemu_target_page_size() / bandwidth
;
2252 qemu_file_reset_rate_limit(s
->to_dst_file
);
2254 s
->iteration_start_time
= current_time
;
2255 s
->iteration_initial_bytes
= qemu_ftell(s
->to_dst_file
);
2257 trace_migrate_transferred(transferred
, time_spent
,
2258 bandwidth
, s
->threshold_size
);
2261 /* Migration thread iteration status */
2263 MIG_ITERATE_RESUME
, /* Resume current iteration */
2264 MIG_ITERATE_SKIP
, /* Skip current iteration */
2265 MIG_ITERATE_BREAK
, /* Break the loop */
2269 * Return true if continue to the next iteration directly, false
2272 static MigIterateState
migration_iteration_run(MigrationState
*s
)
2274 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
2275 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
2277 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
2278 &pend_compat
, &pend_post
);
2279 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
2281 trace_migrate_pending(pending_size
, s
->threshold_size
,
2282 pend_pre
, pend_compat
, pend_post
);
2284 if (pending_size
&& pending_size
>= s
->threshold_size
) {
2285 /* Still a significant amount to transfer */
2286 if (migrate_postcopy() && !in_postcopy
&&
2287 pend_pre
<= s
->threshold_size
&&
2288 atomic_read(&s
->start_postcopy
)) {
2289 if (postcopy_start(s
)) {
2290 error_report("%s: postcopy failed to start", __func__
);
2292 return MIG_ITERATE_SKIP
;
2294 /* Just another iteration step */
2295 qemu_savevm_state_iterate(s
->to_dst_file
,
2296 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2298 trace_migration_thread_low_pending(pending_size
);
2299 migration_completion(s
);
2300 return MIG_ITERATE_BREAK
;
2303 return MIG_ITERATE_RESUME
;
2306 static void migration_iteration_finish(MigrationState
*s
)
2308 /* If we enabled cpu throttling for auto-converge, turn it off. */
2309 cpu_throttle_stop();
2311 qemu_mutex_lock_iothread();
2313 case MIGRATION_STATUS_COMPLETED
:
2314 migration_calculate_complete(s
);
2315 runstate_set(RUN_STATE_POSTMIGRATE
);
2318 case MIGRATION_STATUS_ACTIVE
:
2320 * We should really assert here, but since it's during
2321 * migration, let's try to reduce the usage of assertions.
2323 if (!migrate_colo_enabled()) {
2324 error_report("%s: critical error: calling COLO code without "
2325 "COLO enabled", __func__
);
2327 migrate_start_colo_process(s
);
2329 * Fixme: we will run VM in COLO no matter its old running state.
2330 * After exited COLO, we will keep running.
2332 s
->vm_was_running
= true;
2334 case MIGRATION_STATUS_FAILED
:
2335 case MIGRATION_STATUS_CANCELLED
:
2336 if (s
->vm_was_running
) {
2339 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2340 runstate_set(RUN_STATE_POSTMIGRATE
);
2346 /* Should not reach here, but if so, forgive the VM. */
2347 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
2350 qemu_bh_schedule(s
->cleanup_bh
);
2351 qemu_mutex_unlock_iothread();
2355 * Master migration thread on the source VM.
2356 * It drives the migration and pumps the data down the outgoing channel.
2358 static void *migration_thread(void *opaque
)
2360 MigrationState
*s
= opaque
;
2361 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
2363 rcu_register_thread();
2365 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2367 qemu_savevm_state_header(s
->to_dst_file
);
2370 * If we opened the return path, we need to make sure dst has it
2373 if (s
->rp_state
.from_dst_file
) {
2374 /* Now tell the dest that it should open its end so it can reply */
2375 qemu_savevm_send_open_return_path(s
->to_dst_file
);
2377 /* And do a ping that will make stuff easier to debug */
2378 qemu_savevm_send_ping(s
->to_dst_file
, 1);
2381 if (migrate_postcopy()) {
2383 * Tell the destination that we *might* want to do postcopy later;
2384 * if the other end can't do postcopy it should fail now, nice and
2387 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
2390 qemu_savevm_state_setup(s
->to_dst_file
);
2392 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
2393 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2394 MIGRATION_STATUS_ACTIVE
);
2396 trace_migration_thread_setup_complete();
2398 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
2399 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2400 int64_t current_time
;
2402 if (!qemu_file_rate_limit(s
->to_dst_file
)) {
2403 MigIterateState iter_state
= migration_iteration_run(s
);
2404 if (iter_state
== MIG_ITERATE_SKIP
) {
2406 } else if (iter_state
== MIG_ITERATE_BREAK
) {
2411 if (qemu_file_get_error(s
->to_dst_file
)) {
2412 if (migration_is_setup_or_active(s
->state
)) {
2413 migrate_set_state(&s
->state
, s
->state
,
2414 MIGRATION_STATUS_FAILED
);
2416 trace_migration_thread_file_err();
2420 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2422 migration_update_counters(s
, current_time
);
2424 if (qemu_file_rate_limit(s
->to_dst_file
)) {
2425 /* usleep expects microseconds */
2426 g_usleep((s
->iteration_start_time
+ BUFFER_DELAY
-
2427 current_time
) * 1000);
2431 trace_migration_thread_after_loop();
2432 migration_iteration_finish(s
);
2433 rcu_unregister_thread();
2437 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
2439 s
->expected_downtime
= s
->parameters
.downtime_limit
;
2440 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
2442 migrate_fd_error(s
, error_in
);
2443 migrate_fd_cleanup(s
);
2447 qemu_file_set_blocking(s
->to_dst_file
, true);
2448 qemu_file_set_rate_limit(s
->to_dst_file
,
2449 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
2451 /* Notify before starting migration thread */
2452 notifier_list_notify(&migration_state_notifiers
, s
);
2455 * Open the return path. For postcopy, it is used exclusively. For
2456 * precopy, only if user specified "return-path" capability would
2457 * QEMU uses the return path.
2459 if (migrate_postcopy_ram() || migrate_use_return_path()) {
2460 if (open_return_path_on_source(s
)) {
2461 error_report("Unable to open return-path for postcopy");
2462 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2463 MIGRATION_STATUS_FAILED
);
2464 migrate_fd_cleanup(s
);
2469 if (multifd_save_setup() != 0) {
2470 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2471 MIGRATION_STATUS_FAILED
);
2472 migrate_fd_cleanup(s
);
2475 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
2476 QEMU_THREAD_JOINABLE
);
2477 s
->migration_thread_running
= true;
2480 void migration_global_dump(Monitor
*mon
)
2482 MigrationState
*ms
= migrate_get_current();
2484 monitor_printf(mon
, "globals:\n");
2485 monitor_printf(mon
, "store-global-state: %s\n",
2486 ms
->store_global_state
? "on" : "off");
2487 monitor_printf(mon
, "only-migratable: %s\n",
2488 ms
->only_migratable
? "on" : "off");
2489 monitor_printf(mon
, "send-configuration: %s\n",
2490 ms
->send_configuration
? "on" : "off");
2491 monitor_printf(mon
, "send-section-footer: %s\n",
2492 ms
->send_section_footer
? "on" : "off");
2495 #define DEFINE_PROP_MIG_CAP(name, x) \
2496 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
2498 static Property migration_properties
[] = {
2499 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
2500 store_global_state
, true),
2501 DEFINE_PROP_BOOL("only-migratable", MigrationState
, only_migratable
, false),
2502 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
2503 send_configuration
, true),
2504 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
2505 send_section_footer
, true),
2507 /* Migration parameters */
2508 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
2509 parameters
.compress_level
,
2510 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
2511 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
2512 parameters
.compress_threads
,
2513 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
2514 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
2515 parameters
.decompress_threads
,
2516 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
2517 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
2518 parameters
.cpu_throttle_initial
,
2519 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
2520 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
2521 parameters
.cpu_throttle_increment
,
2522 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
2523 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
2524 parameters
.max_bandwidth
, MAX_THROTTLE
),
2525 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
2526 parameters
.downtime_limit
,
2527 DEFAULT_MIGRATE_SET_DOWNTIME
),
2528 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
2529 parameters
.x_checkpoint_delay
,
2530 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
2531 DEFINE_PROP_UINT8("x-multifd-channels", MigrationState
,
2532 parameters
.x_multifd_channels
,
2533 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
2534 DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState
,
2535 parameters
.x_multifd_page_count
,
2536 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT
),
2537 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
2538 parameters
.xbzrle_cache_size
,
2539 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
2541 /* Migration capabilities */
2542 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
2543 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
2544 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
2545 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
2546 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
2547 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
2548 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
2549 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
2550 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
2551 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
2552 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
2553 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD
),
2555 DEFINE_PROP_END_OF_LIST(),
2558 static void migration_class_init(ObjectClass
*klass
, void *data
)
2560 DeviceClass
*dc
= DEVICE_CLASS(klass
);
2562 dc
->user_creatable
= false;
2563 dc
->props
= migration_properties
;
2566 static void migration_instance_finalize(Object
*obj
)
2568 MigrationState
*ms
= MIGRATION_OBJ(obj
);
2569 MigrationParameters
*params
= &ms
->parameters
;
2571 qemu_mutex_destroy(&ms
->error_mutex
);
2572 g_free(params
->tls_hostname
);
2573 g_free(params
->tls_creds
);
2574 qemu_sem_destroy(&ms
->pause_sem
);
2575 error_free(ms
->error
);
2578 static void migration_instance_init(Object
*obj
)
2580 MigrationState
*ms
= MIGRATION_OBJ(obj
);
2581 MigrationParameters
*params
= &ms
->parameters
;
2583 ms
->state
= MIGRATION_STATUS_NONE
;
2585 qemu_sem_init(&ms
->pause_sem
, 0);
2586 qemu_mutex_init(&ms
->error_mutex
);
2588 params
->tls_hostname
= g_strdup("");
2589 params
->tls_creds
= g_strdup("");
2591 /* Set has_* up only for parameter checks */
2592 params
->has_compress_level
= true;
2593 params
->has_compress_threads
= true;
2594 params
->has_decompress_threads
= true;
2595 params
->has_cpu_throttle_initial
= true;
2596 params
->has_cpu_throttle_increment
= true;
2597 params
->has_max_bandwidth
= true;
2598 params
->has_downtime_limit
= true;
2599 params
->has_x_checkpoint_delay
= true;
2600 params
->has_block_incremental
= true;
2601 params
->has_x_multifd_channels
= true;
2602 params
->has_x_multifd_page_count
= true;
2603 params
->has_xbzrle_cache_size
= true;
2607 * Return true if check pass, false otherwise. Error will be put
2608 * inside errp if provided.
2610 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
2612 MigrationCapabilityStatusList
*head
= NULL
;
2613 /* Assuming all off */
2614 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
2617 if (!migrate_params_check(&ms
->parameters
, errp
)) {
2621 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
2622 if (ms
->enabled_capabilities
[i
]) {
2623 head
= migrate_cap_add(head
, i
, true);
2627 ret
= migrate_caps_check(cap_list
, head
, errp
);
2629 /* It works with head == NULL */
2630 qapi_free_MigrationCapabilityStatusList(head
);
2635 static const TypeInfo migration_type
= {
2636 .name
= TYPE_MIGRATION
,
2638 * NOTE: TYPE_MIGRATION is not really a device, as the object is
2639 * not created using qdev_create(), it is not attached to the qdev
2640 * device tree, and it is never realized.
2642 * TODO: Make this TYPE_OBJECT once QOM provides something like
2643 * TYPE_DEVICE's "-global" properties.
2645 .parent
= TYPE_DEVICE
,
2646 .class_init
= migration_class_init
,
2647 .class_size
= sizeof(MigrationClass
),
2648 .instance_size
= sizeof(MigrationState
),
2649 .instance_init
= migration_instance_init
,
2650 .instance_finalize
= migration_instance_finalize
,
2653 static void register_migration_types(void)
2655 type_register_static(&migration_type
);
2658 type_init(register_migration_types
);