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 /* Make sure all file formats flush their mutable metadata.
310 * If we get an error here, just don't restart the VM yet. */
311 bdrv_invalidate_cache_all(&local_err
);
313 error_report_err(local_err
);
319 * This must happen after all error conditions are dealt with and
320 * we're sure the VM is going to be running on this host.
322 qemu_announce_self();
324 if (multifd_load_cleanup(&local_err
) != 0) {
325 error_report_err(local_err
);
328 /* If global state section was not received or we are in running
329 state, we need to obey autostart. Any other state is set with
332 dirty_bitmap_mig_before_vm_start();
334 if (!global_state_received() ||
335 global_state_get_runstate() == RUN_STATE_RUNNING
) {
339 runstate_set(RUN_STATE_PAUSED
);
342 runstate_set(global_state_get_runstate());
345 * This must happen after any state changes since as soon as an external
346 * observer sees this event they might start to prod at the VM assuming
349 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
350 MIGRATION_STATUS_COMPLETED
);
351 qemu_bh_delete(mis
->bh
);
352 migration_incoming_state_destroy();
355 static void process_incoming_migration_co(void *opaque
)
357 MigrationIncomingState
*mis
= migration_incoming_get_current();
361 assert(mis
->from_src_file
);
362 mis
->largest_page_size
= qemu_ram_pagesize_largest();
363 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
364 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
365 MIGRATION_STATUS_ACTIVE
);
366 ret
= qemu_loadvm_state(mis
->from_src_file
);
368 ps
= postcopy_state_get();
369 trace_process_incoming_migration_co_end(ret
, ps
);
370 if (ps
!= POSTCOPY_INCOMING_NONE
) {
371 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
373 * Where a migration had postcopy enabled (and thus went to advise)
374 * but managed to complete within the precopy period, we can use
377 postcopy_ram_incoming_cleanup(mis
);
378 } else if (ret
>= 0) {
380 * Postcopy was started, cleanup should happen at the end of the
383 trace_process_incoming_migration_co_postcopy_end_main();
386 /* Else if something went wrong then just fall out of the normal exit */
389 /* we get COLO info, and know if we are in COLO mode */
390 if (!ret
&& migration_incoming_enable_colo()) {
391 mis
->migration_incoming_co
= qemu_coroutine_self();
392 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
393 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
394 mis
->have_colo_incoming_thread
= true;
395 qemu_coroutine_yield();
397 /* Wait checkpoint incoming thread exit before free resource */
398 qemu_thread_join(&mis
->colo_incoming_thread
);
402 Error
*local_err
= NULL
;
404 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
405 MIGRATION_STATUS_FAILED
);
406 error_report("load of migration failed: %s", strerror(-ret
));
407 qemu_fclose(mis
->from_src_file
);
408 if (multifd_load_cleanup(&local_err
) != 0) {
409 error_report_err(local_err
);
413 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
414 qemu_bh_schedule(mis
->bh
);
417 static void migration_incoming_setup(QEMUFile
*f
)
419 MigrationIncomingState
*mis
= migration_incoming_get_current();
421 if (multifd_load_setup() != 0) {
422 /* We haven't been able to create multifd threads
423 nothing better to do */
427 if (!mis
->from_src_file
) {
428 mis
->from_src_file
= f
;
430 qemu_file_set_blocking(f
, false);
433 static void migration_incoming_process(void)
435 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
436 qemu_coroutine_enter(co
);
439 void migration_fd_process_incoming(QEMUFile
*f
)
441 migration_incoming_setup(f
);
442 migration_incoming_process();
445 void migration_ioc_process_incoming(QIOChannel
*ioc
)
447 MigrationIncomingState
*mis
= migration_incoming_get_current();
449 if (!mis
->from_src_file
) {
450 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
451 migration_fd_process_incoming(f
);
453 /* We still only have a single channel. Nothing to do here yet */
457 * @migration_has_all_channels: We have received all channels that we need
459 * Returns true when we have got connections to all the channels that
460 * we need for migration.
462 bool migration_has_all_channels(void)
468 * Send a 'SHUT' message on the return channel with the given value
469 * to indicate that we've finished with the RP. Non-0 value indicates
472 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
477 buf
= cpu_to_be32(value
);
478 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
482 * Send a 'PONG' message on the return channel with the given value
483 * (normally in response to a 'PING')
485 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
490 buf
= cpu_to_be32(value
);
491 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
494 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
496 MigrationCapabilityStatusList
*head
= NULL
;
497 MigrationCapabilityStatusList
*caps
;
498 MigrationState
*s
= migrate_get_current();
501 caps
= NULL
; /* silence compiler warning */
502 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
503 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
504 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
509 head
= g_malloc0(sizeof(*caps
));
512 caps
->next
= g_malloc0(sizeof(*caps
));
516 g_malloc(sizeof(*caps
->value
));
517 caps
->value
->capability
= i
;
518 caps
->value
->state
= s
->enabled_capabilities
[i
];
524 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
526 MigrationParameters
*params
;
527 MigrationState
*s
= migrate_get_current();
529 /* TODO use QAPI_CLONE() instead of duplicating it inline */
530 params
= g_malloc0(sizeof(*params
));
531 params
->has_compress_level
= true;
532 params
->compress_level
= s
->parameters
.compress_level
;
533 params
->has_compress_threads
= true;
534 params
->compress_threads
= s
->parameters
.compress_threads
;
535 params
->has_decompress_threads
= true;
536 params
->decompress_threads
= s
->parameters
.decompress_threads
;
537 params
->has_cpu_throttle_initial
= true;
538 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
539 params
->has_cpu_throttle_increment
= true;
540 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
541 params
->has_tls_creds
= true;
542 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
543 params
->has_tls_hostname
= true;
544 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
545 params
->has_max_bandwidth
= true;
546 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
547 params
->has_downtime_limit
= true;
548 params
->downtime_limit
= s
->parameters
.downtime_limit
;
549 params
->has_x_checkpoint_delay
= true;
550 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
551 params
->has_block_incremental
= true;
552 params
->block_incremental
= s
->parameters
.block_incremental
;
553 params
->has_x_multifd_channels
= true;
554 params
->x_multifd_channels
= s
->parameters
.x_multifd_channels
;
555 params
->has_x_multifd_page_count
= true;
556 params
->x_multifd_page_count
= s
->parameters
.x_multifd_page_count
;
557 params
->has_xbzrle_cache_size
= true;
558 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
564 * Return true if we're already in the middle of a migration
565 * (i.e. any of the active or setup states)
567 static bool migration_is_setup_or_active(int state
)
570 case MIGRATION_STATUS_ACTIVE
:
571 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
572 case MIGRATION_STATUS_SETUP
:
573 case MIGRATION_STATUS_PRE_SWITCHOVER
:
574 case MIGRATION_STATUS_DEVICE
:
583 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
585 info
->has_ram
= true;
586 info
->ram
= g_malloc0(sizeof(*info
->ram
));
587 info
->ram
->transferred
= ram_counters
.transferred
;
588 info
->ram
->total
= ram_bytes_total();
589 info
->ram
->duplicate
= ram_counters
.duplicate
;
590 /* legacy value. It is not used anymore */
591 info
->ram
->skipped
= 0;
592 info
->ram
->normal
= ram_counters
.normal
;
593 info
->ram
->normal_bytes
= ram_counters
.normal
*
594 qemu_target_page_size();
595 info
->ram
->mbps
= s
->mbps
;
596 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
597 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
598 info
->ram
->page_size
= qemu_target_page_size();
600 if (migrate_use_xbzrle()) {
601 info
->has_xbzrle_cache
= true;
602 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
603 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
604 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
605 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
606 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
607 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
608 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
611 if (cpu_throttle_active()) {
612 info
->has_cpu_throttle_percentage
= true;
613 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
616 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
617 info
->ram
->remaining
= ram_bytes_remaining();
618 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
622 static void populate_disk_info(MigrationInfo
*info
)
624 if (blk_mig_active()) {
625 info
->has_disk
= true;
626 info
->disk
= g_malloc0(sizeof(*info
->disk
));
627 info
->disk
->transferred
= blk_mig_bytes_transferred();
628 info
->disk
->remaining
= blk_mig_bytes_remaining();
629 info
->disk
->total
= blk_mig_bytes_total();
633 MigrationInfo
*qmp_query_migrate(Error
**errp
)
635 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
636 MigrationState
*s
= migrate_get_current();
639 case MIGRATION_STATUS_NONE
:
640 /* no migration has happened ever */
642 case MIGRATION_STATUS_SETUP
:
643 info
->has_status
= true;
644 info
->has_total_time
= false;
646 case MIGRATION_STATUS_ACTIVE
:
647 case MIGRATION_STATUS_CANCELLING
:
648 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
649 case MIGRATION_STATUS_PRE_SWITCHOVER
:
650 case MIGRATION_STATUS_DEVICE
:
651 /* TODO add some postcopy stats */
652 info
->has_status
= true;
653 info
->has_total_time
= true;
654 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
656 info
->has_expected_downtime
= true;
657 info
->expected_downtime
= s
->expected_downtime
;
658 info
->has_setup_time
= true;
659 info
->setup_time
= s
->setup_time
;
661 populate_ram_info(info
, s
);
662 populate_disk_info(info
);
664 case MIGRATION_STATUS_COLO
:
665 info
->has_status
= true;
666 /* TODO: display COLO specific information (checkpoint info etc.) */
668 case MIGRATION_STATUS_COMPLETED
:
669 info
->has_status
= true;
670 info
->has_total_time
= true;
671 info
->total_time
= s
->total_time
;
672 info
->has_downtime
= true;
673 info
->downtime
= s
->downtime
;
674 info
->has_setup_time
= true;
675 info
->setup_time
= s
->setup_time
;
677 populate_ram_info(info
, s
);
679 case MIGRATION_STATUS_FAILED
:
680 info
->has_status
= true;
682 info
->has_error_desc
= true;
683 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
686 case MIGRATION_STATUS_CANCELLED
:
687 info
->has_status
= true;
690 info
->status
= s
->state
;
696 * @migration_caps_check - check capability validity
698 * @cap_list: old capability list, array of bool
699 * @params: new capabilities to be applied soon
700 * @errp: set *errp if the check failed, with reason
702 * Returns true if check passed, otherwise false.
704 static bool migrate_caps_check(bool *cap_list
,
705 MigrationCapabilityStatusList
*params
,
708 MigrationCapabilityStatusList
*cap
;
709 bool old_postcopy_cap
;
710 MigrationIncomingState
*mis
= migration_incoming_get_current();
712 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
714 for (cap
= params
; cap
; cap
= cap
->next
) {
715 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
718 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
719 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
720 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
722 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
727 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
728 if (cap_list
[MIGRATION_CAPABILITY_COMPRESS
]) {
729 /* The decompression threads asynchronously write into RAM
730 * rather than use the atomic copies needed to avoid
731 * userfaulting. It should be possible to fix the decompression
732 * threads for compatibility in future.
734 error_setg(errp
, "Postcopy is not currently compatible "
739 /* This check is reasonably expensive, so only when it's being
740 * set the first time, also it's only the destination that needs
743 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
744 !postcopy_ram_supported_by_host(mis
)) {
745 /* postcopy_ram_supported_by_host will have emitted a more
748 error_setg(errp
, "Postcopy is not supported");
756 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
759 MigrationState
*s
= migrate_get_current();
760 MigrationCapabilityStatusList
*cap
;
761 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
763 if (migration_is_setup_or_active(s
->state
)) {
764 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
768 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
769 if (!migrate_caps_check(cap_list
, params
, errp
)) {
773 for (cap
= params
; cap
; cap
= cap
->next
) {
774 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
779 * Check whether the parameters are valid. Error will be put into errp
780 * (if provided). Return true if valid, otherwise false.
782 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
784 if (params
->has_compress_level
&&
785 (params
->compress_level
> 9)) {
786 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
787 "is invalid, it should be in the range of 0 to 9");
791 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
792 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
794 "is invalid, it should be in the range of 1 to 255");
798 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
799 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
800 "decompress_threads",
801 "is invalid, it should be in the range of 1 to 255");
805 if (params
->has_cpu_throttle_initial
&&
806 (params
->cpu_throttle_initial
< 1 ||
807 params
->cpu_throttle_initial
> 99)) {
808 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
809 "cpu_throttle_initial",
810 "an integer in the range of 1 to 99");
814 if (params
->has_cpu_throttle_increment
&&
815 (params
->cpu_throttle_increment
< 1 ||
816 params
->cpu_throttle_increment
> 99)) {
817 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
818 "cpu_throttle_increment",
819 "an integer in the range of 1 to 99");
823 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
824 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
825 " range of 0 to %zu bytes/second", SIZE_MAX
);
829 if (params
->has_downtime_limit
&&
830 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
831 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
832 "the range of 0 to %d milliseconds",
833 MAX_MIGRATE_DOWNTIME
);
837 /* x_checkpoint_delay is now always positive */
839 if (params
->has_x_multifd_channels
&& (params
->x_multifd_channels
< 1)) {
840 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
842 "is invalid, it should be in the range of 1 to 255");
845 if (params
->has_x_multifd_page_count
&&
846 (params
->x_multifd_page_count
< 1 ||
847 params
->x_multifd_page_count
> 10000)) {
848 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
849 "multifd_page_count",
850 "is invalid, it should be in the range of 1 to 10000");
854 if (params
->has_xbzrle_cache_size
&&
855 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
856 !is_power_of_2(params
->xbzrle_cache_size
))) {
857 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
859 "is invalid, it should be bigger than target page size"
860 " and a power of two");
867 static void migrate_params_test_apply(MigrateSetParameters
*params
,
868 MigrationParameters
*dest
)
870 *dest
= migrate_get_current()->parameters
;
872 /* TODO use QAPI_CLONE() instead of duplicating it inline */
874 if (params
->has_compress_level
) {
875 dest
->compress_level
= params
->compress_level
;
878 if (params
->has_compress_threads
) {
879 dest
->compress_threads
= params
->compress_threads
;
882 if (params
->has_decompress_threads
) {
883 dest
->decompress_threads
= params
->decompress_threads
;
886 if (params
->has_cpu_throttle_initial
) {
887 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
890 if (params
->has_cpu_throttle_increment
) {
891 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
894 if (params
->has_tls_creds
) {
895 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
896 dest
->tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
899 if (params
->has_tls_hostname
) {
900 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
901 dest
->tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
904 if (params
->has_max_bandwidth
) {
905 dest
->max_bandwidth
= params
->max_bandwidth
;
908 if (params
->has_downtime_limit
) {
909 dest
->downtime_limit
= params
->downtime_limit
;
912 if (params
->has_x_checkpoint_delay
) {
913 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
916 if (params
->has_block_incremental
) {
917 dest
->block_incremental
= params
->block_incremental
;
919 if (params
->has_x_multifd_channels
) {
920 dest
->x_multifd_channels
= params
->x_multifd_channels
;
922 if (params
->has_x_multifd_page_count
) {
923 dest
->x_multifd_page_count
= params
->x_multifd_page_count
;
925 if (params
->has_xbzrle_cache_size
) {
926 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
930 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
932 MigrationState
*s
= migrate_get_current();
934 /* TODO use QAPI_CLONE() instead of duplicating it inline */
936 if (params
->has_compress_level
) {
937 s
->parameters
.compress_level
= params
->compress_level
;
940 if (params
->has_compress_threads
) {
941 s
->parameters
.compress_threads
= params
->compress_threads
;
944 if (params
->has_decompress_threads
) {
945 s
->parameters
.decompress_threads
= params
->decompress_threads
;
948 if (params
->has_cpu_throttle_initial
) {
949 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
952 if (params
->has_cpu_throttle_increment
) {
953 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
956 if (params
->has_tls_creds
) {
957 g_free(s
->parameters
.tls_creds
);
958 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
959 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
962 if (params
->has_tls_hostname
) {
963 g_free(s
->parameters
.tls_hostname
);
964 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
965 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
968 if (params
->has_max_bandwidth
) {
969 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
970 if (s
->to_dst_file
) {
971 qemu_file_set_rate_limit(s
->to_dst_file
,
972 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
976 if (params
->has_downtime_limit
) {
977 s
->parameters
.downtime_limit
= params
->downtime_limit
;
980 if (params
->has_x_checkpoint_delay
) {
981 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
982 if (migration_in_colo_state()) {
983 colo_checkpoint_notify(s
);
987 if (params
->has_block_incremental
) {
988 s
->parameters
.block_incremental
= params
->block_incremental
;
990 if (params
->has_x_multifd_channels
) {
991 s
->parameters
.x_multifd_channels
= params
->x_multifd_channels
;
993 if (params
->has_x_multifd_page_count
) {
994 s
->parameters
.x_multifd_page_count
= params
->x_multifd_page_count
;
996 if (params
->has_xbzrle_cache_size
) {
997 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
998 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1002 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1004 MigrationParameters tmp
;
1006 /* TODO Rewrite "" to null instead */
1007 if (params
->has_tls_creds
1008 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1009 QDECREF(params
->tls_creds
->u
.n
);
1010 params
->tls_creds
->type
= QTYPE_QSTRING
;
1011 params
->tls_creds
->u
.s
= strdup("");
1013 /* TODO Rewrite "" to null instead */
1014 if (params
->has_tls_hostname
1015 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1016 QDECREF(params
->tls_hostname
->u
.n
);
1017 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1018 params
->tls_hostname
->u
.s
= strdup("");
1021 migrate_params_test_apply(params
, &tmp
);
1023 if (!migrate_params_check(&tmp
, errp
)) {
1024 /* Invalid parameter */
1028 migrate_params_apply(params
, errp
);
1032 void qmp_migrate_start_postcopy(Error
**errp
)
1034 MigrationState
*s
= migrate_get_current();
1036 if (!migrate_postcopy()) {
1037 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1038 " the start of migration");
1042 if (s
->state
== MIGRATION_STATUS_NONE
) {
1043 error_setg(errp
, "Postcopy must be started after migration has been"
1048 * we don't error if migration has finished since that would be racy
1049 * with issuing this command.
1051 atomic_set(&s
->start_postcopy
, true);
1054 /* shared migration helpers */
1056 void migrate_set_state(int *state
, int old_state
, int new_state
)
1058 assert(new_state
< MIGRATION_STATUS__MAX
);
1059 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1060 trace_migrate_set_state(MigrationStatus_str(new_state
));
1061 migrate_generate_event(new_state
);
1065 static MigrationCapabilityStatusList
*migrate_cap_add(
1066 MigrationCapabilityStatusList
*list
,
1067 MigrationCapability index
,
1070 MigrationCapabilityStatusList
*cap
;
1072 cap
= g_new0(MigrationCapabilityStatusList
, 1);
1073 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
1074 cap
->value
->capability
= index
;
1075 cap
->value
->state
= state
;
1081 void migrate_set_block_enabled(bool value
, Error
**errp
)
1083 MigrationCapabilityStatusList
*cap
;
1085 cap
= migrate_cap_add(NULL
, MIGRATION_CAPABILITY_BLOCK
, value
);
1086 qmp_migrate_set_capabilities(cap
, errp
);
1087 qapi_free_MigrationCapabilityStatusList(cap
);
1090 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1092 s
->parameters
.block_incremental
= value
;
1095 static void block_cleanup_parameters(MigrationState
*s
)
1097 if (s
->must_remove_block_options
) {
1098 /* setting to false can never fail */
1099 migrate_set_block_enabled(false, &error_abort
);
1100 migrate_set_block_incremental(s
, false);
1101 s
->must_remove_block_options
= false;
1105 static void migrate_fd_cleanup(void *opaque
)
1107 MigrationState
*s
= opaque
;
1109 qemu_bh_delete(s
->cleanup_bh
);
1110 s
->cleanup_bh
= NULL
;
1112 qemu_savevm_state_cleanup();
1114 if (s
->to_dst_file
) {
1115 Error
*local_err
= NULL
;
1117 trace_migrate_fd_cleanup();
1118 qemu_mutex_unlock_iothread();
1119 if (s
->migration_thread_running
) {
1120 qemu_thread_join(&s
->thread
);
1121 s
->migration_thread_running
= false;
1123 qemu_mutex_lock_iothread();
1125 if (multifd_save_cleanup(&local_err
) != 0) {
1126 error_report_err(local_err
);
1128 qemu_fclose(s
->to_dst_file
);
1129 s
->to_dst_file
= NULL
;
1132 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
1133 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
1135 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1136 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1137 MIGRATION_STATUS_CANCELLED
);
1141 /* It is used on info migrate. We can't free it */
1142 error_report_err(error_copy(s
->error
));
1144 notifier_list_notify(&migration_state_notifiers
, s
);
1145 block_cleanup_parameters(s
);
1148 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1150 qemu_mutex_lock(&s
->error_mutex
);
1152 s
->error
= error_copy(error
);
1154 qemu_mutex_unlock(&s
->error_mutex
);
1157 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1159 trace_migrate_fd_error(error_get_pretty(error
));
1160 assert(s
->to_dst_file
== NULL
);
1161 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1162 MIGRATION_STATUS_FAILED
);
1163 migrate_set_error(s
, error
);
1166 static void migrate_fd_cancel(MigrationState
*s
)
1169 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1170 trace_migrate_fd_cancel();
1172 if (s
->rp_state
.from_dst_file
) {
1173 /* shutdown the rp socket, so causing the rp thread to shutdown */
1174 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1178 old_state
= s
->state
;
1179 if (!migration_is_setup_or_active(old_state
)) {
1182 /* If the migration is paused, kick it out of the pause */
1183 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1184 qemu_sem_post(&s
->pause_sem
);
1186 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1187 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1190 * If we're unlucky the migration code might be stuck somewhere in a
1191 * send/write while the network has failed and is waiting to timeout;
1192 * if we've got shutdown(2) available then we can force it to quit.
1193 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1194 * called in a bh, so there is no race against this cancel.
1196 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1197 qemu_file_shutdown(f
);
1199 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1200 Error
*local_err
= NULL
;
1202 bdrv_invalidate_cache_all(&local_err
);
1204 error_report_err(local_err
);
1206 s
->block_inactive
= false;
1211 void add_migration_state_change_notifier(Notifier
*notify
)
1213 notifier_list_add(&migration_state_notifiers
, notify
);
1216 void remove_migration_state_change_notifier(Notifier
*notify
)
1218 notifier_remove(notify
);
1221 bool migration_in_setup(MigrationState
*s
)
1223 return s
->state
== MIGRATION_STATUS_SETUP
;
1226 bool migration_has_finished(MigrationState
*s
)
1228 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1231 bool migration_has_failed(MigrationState
*s
)
1233 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1234 s
->state
== MIGRATION_STATUS_FAILED
);
1237 bool migration_in_postcopy(void)
1239 MigrationState
*s
= migrate_get_current();
1241 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1244 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1246 return migration_in_postcopy() && s
->postcopy_after_devices
;
1249 bool migration_is_idle(void)
1251 MigrationState
*s
= migrate_get_current();
1254 case MIGRATION_STATUS_NONE
:
1255 case MIGRATION_STATUS_CANCELLED
:
1256 case MIGRATION_STATUS_COMPLETED
:
1257 case MIGRATION_STATUS_FAILED
:
1259 case MIGRATION_STATUS_SETUP
:
1260 case MIGRATION_STATUS_CANCELLING
:
1261 case MIGRATION_STATUS_ACTIVE
:
1262 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1263 case MIGRATION_STATUS_COLO
:
1264 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1265 case MIGRATION_STATUS_DEVICE
:
1267 case MIGRATION_STATUS__MAX
:
1268 g_assert_not_reached();
1274 void migrate_init(MigrationState
*s
)
1277 * Reinitialise all migration state, except
1278 * parameters/capabilities that the user set, and
1284 s
->to_dst_file
= NULL
;
1285 s
->state
= MIGRATION_STATUS_NONE
;
1286 s
->rp_state
.from_dst_file
= NULL
;
1287 s
->rp_state
.error
= false;
1290 s
->expected_downtime
= 0;
1292 s
->start_postcopy
= false;
1293 s
->postcopy_after_devices
= false;
1294 s
->migration_thread_running
= false;
1295 error_free(s
->error
);
1298 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1300 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1302 s
->vm_was_running
= false;
1303 s
->iteration_initial_bytes
= 0;
1304 s
->threshold_size
= 0;
1307 static GSList
*migration_blockers
;
1309 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1311 if (migrate_get_current()->only_migratable
) {
1312 error_propagate(errp
, error_copy(reason
));
1313 error_prepend(errp
, "disallowing migration blocker "
1314 "(--only_migratable) for: ");
1318 if (migration_is_idle()) {
1319 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1323 error_propagate(errp
, error_copy(reason
));
1324 error_prepend(errp
, "disallowing migration blocker (migration in "
1329 void migrate_del_blocker(Error
*reason
)
1331 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1334 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1336 Error
*local_err
= NULL
;
1337 static bool once
= true;
1339 if (!deferred_incoming
) {
1340 error_setg(errp
, "For use with '-incoming defer'");
1344 error_setg(errp
, "The incoming migration has already been started");
1347 qemu_start_incoming_migration(uri
, &local_err
);
1350 error_propagate(errp
, local_err
);
1357 bool migration_is_blocked(Error
**errp
)
1359 if (qemu_savevm_state_blocked(errp
)) {
1363 if (migration_blockers
) {
1364 error_propagate(errp
, error_copy(migration_blockers
->data
));
1371 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1372 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1375 Error
*local_err
= NULL
;
1376 MigrationState
*s
= migrate_get_current();
1379 if (migration_is_setup_or_active(s
->state
) ||
1380 s
->state
== MIGRATION_STATUS_CANCELLING
||
1381 s
->state
== MIGRATION_STATUS_COLO
) {
1382 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1385 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1386 error_setg(errp
, "Guest is waiting for an incoming migration");
1390 if (migration_is_blocked(errp
)) {
1394 if ((has_blk
&& blk
) || (has_inc
&& inc
)) {
1395 if (migrate_use_block() || migrate_use_block_incremental()) {
1396 error_setg(errp
, "Command options are incompatible with "
1397 "current migration capabilities");
1400 migrate_set_block_enabled(true, &local_err
);
1402 error_propagate(errp
, local_err
);
1405 s
->must_remove_block_options
= true;
1408 if (has_inc
&& inc
) {
1409 migrate_set_block_incremental(s
, true);
1414 if (strstart(uri
, "tcp:", &p
)) {
1415 tcp_start_outgoing_migration(s
, p
, &local_err
);
1417 } else if (strstart(uri
, "rdma:", &p
)) {
1418 rdma_start_outgoing_migration(s
, p
, &local_err
);
1420 } else if (strstart(uri
, "exec:", &p
)) {
1421 exec_start_outgoing_migration(s
, p
, &local_err
);
1422 } else if (strstart(uri
, "unix:", &p
)) {
1423 unix_start_outgoing_migration(s
, p
, &local_err
);
1424 } else if (strstart(uri
, "fd:", &p
)) {
1425 fd_start_outgoing_migration(s
, p
, &local_err
);
1427 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1428 "a valid migration protocol");
1429 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1430 MIGRATION_STATUS_FAILED
);
1435 migrate_fd_error(s
, local_err
);
1436 error_propagate(errp
, local_err
);
1441 void qmp_migrate_cancel(Error
**errp
)
1443 migrate_fd_cancel(migrate_get_current());
1446 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
1448 MigrationState
*s
= migrate_get_current();
1449 if (s
->state
!= state
) {
1450 error_setg(errp
, "Migration not in expected state: %s",
1451 MigrationStatus_str(s
->state
));
1454 qemu_sem_post(&s
->pause_sem
);
1457 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1459 MigrateSetParameters p
= {
1460 .has_xbzrle_cache_size
= true,
1461 .xbzrle_cache_size
= value
,
1464 qmp_migrate_set_parameters(&p
, errp
);
1467 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1469 return migrate_xbzrle_cache_size();
1472 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1474 MigrateSetParameters p
= {
1475 .has_max_bandwidth
= true,
1476 .max_bandwidth
= value
,
1479 qmp_migrate_set_parameters(&p
, errp
);
1482 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1484 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
1485 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1486 "the range of 0 to %d seconds",
1487 MAX_MIGRATE_DOWNTIME_SECONDS
);
1491 value
*= 1000; /* Convert to milliseconds */
1492 value
= MAX(0, MIN(INT64_MAX
, value
));
1494 MigrateSetParameters p
= {
1495 .has_downtime_limit
= true,
1496 .downtime_limit
= value
,
1499 qmp_migrate_set_parameters(&p
, errp
);
1502 bool migrate_release_ram(void)
1506 s
= migrate_get_current();
1508 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
1511 bool migrate_postcopy_ram(void)
1515 s
= migrate_get_current();
1517 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1520 bool migrate_postcopy(void)
1522 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
1525 bool migrate_auto_converge(void)
1529 s
= migrate_get_current();
1531 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1534 bool migrate_zero_blocks(void)
1538 s
= migrate_get_current();
1540 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
1543 bool migrate_use_compression(void)
1547 s
= migrate_get_current();
1549 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
1552 int migrate_compress_level(void)
1556 s
= migrate_get_current();
1558 return s
->parameters
.compress_level
;
1561 int migrate_compress_threads(void)
1565 s
= migrate_get_current();
1567 return s
->parameters
.compress_threads
;
1570 int migrate_decompress_threads(void)
1574 s
= migrate_get_current();
1576 return s
->parameters
.decompress_threads
;
1579 bool migrate_dirty_bitmaps(void)
1583 s
= migrate_get_current();
1585 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
1588 bool migrate_use_events(void)
1592 s
= migrate_get_current();
1594 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
1597 bool migrate_use_multifd(void)
1601 s
= migrate_get_current();
1603 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_MULTIFD
];
1606 bool migrate_pause_before_switchover(void)
1610 s
= migrate_get_current();
1612 return s
->enabled_capabilities
[
1613 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
1616 int migrate_multifd_channels(void)
1620 s
= migrate_get_current();
1622 return s
->parameters
.x_multifd_channels
;
1625 int migrate_multifd_page_count(void)
1629 s
= migrate_get_current();
1631 return s
->parameters
.x_multifd_page_count
;
1634 int migrate_use_xbzrle(void)
1638 s
= migrate_get_current();
1640 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
1643 int64_t migrate_xbzrle_cache_size(void)
1647 s
= migrate_get_current();
1649 return s
->parameters
.xbzrle_cache_size
;
1652 bool migrate_use_block(void)
1656 s
= migrate_get_current();
1658 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
1661 bool migrate_use_return_path(void)
1665 s
= migrate_get_current();
1667 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
1670 bool migrate_use_block_incremental(void)
1674 s
= migrate_get_current();
1676 return s
->parameters
.block_incremental
;
1679 /* migration thread support */
1681 * Something bad happened to the RP stream, mark an error
1682 * The caller shall print or trace something to indicate why
1684 static void mark_source_rp_bad(MigrationState
*s
)
1686 s
->rp_state
.error
= true;
1689 static struct rp_cmd_args
{
1690 ssize_t len
; /* -1 = variable */
1693 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1694 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1695 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1696 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1697 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1698 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1702 * Process a request for pages received on the return path,
1703 * We're allowed to send more than requested (e.g. to round to our page size)
1704 * and we don't need to send pages that have already been sent.
1706 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1707 ram_addr_t start
, size_t len
)
1709 long our_host_ps
= getpagesize();
1711 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1714 * Since we currently insist on matching page sizes, just sanity check
1715 * we're being asked for whole host pages.
1717 if (start
& (our_host_ps
-1) ||
1718 (len
& (our_host_ps
-1))) {
1719 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1720 " len: %zd", __func__
, start
, len
);
1721 mark_source_rp_bad(ms
);
1725 if (ram_save_queue_pages(rbname
, start
, len
)) {
1726 mark_source_rp_bad(ms
);
1731 * Handles messages sent on the return path towards the source VM
1734 static void *source_return_path_thread(void *opaque
)
1736 MigrationState
*ms
= opaque
;
1737 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1738 uint16_t header_len
, header_type
;
1740 uint32_t tmp32
, sibling_error
;
1741 ram_addr_t start
= 0; /* =0 to silence warning */
1742 size_t len
= 0, expected_len
;
1745 trace_source_return_path_thread_entry();
1746 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1747 migration_is_setup_or_active(ms
->state
)) {
1748 trace_source_return_path_thread_loop_top();
1749 header_type
= qemu_get_be16(rp
);
1750 header_len
= qemu_get_be16(rp
);
1752 if (qemu_file_get_error(rp
)) {
1753 mark_source_rp_bad(ms
);
1757 if (header_type
>= MIG_RP_MSG_MAX
||
1758 header_type
== MIG_RP_MSG_INVALID
) {
1759 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1760 header_type
, header_len
);
1761 mark_source_rp_bad(ms
);
1765 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1766 header_len
!= rp_cmd_args
[header_type
].len
) ||
1767 header_len
> sizeof(buf
)) {
1768 error_report("RP: Received '%s' message (0x%04x) with"
1769 "incorrect length %d expecting %zu",
1770 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1771 (size_t)rp_cmd_args
[header_type
].len
);
1772 mark_source_rp_bad(ms
);
1776 /* We know we've got a valid header by this point */
1777 res
= qemu_get_buffer(rp
, buf
, header_len
);
1778 if (res
!= header_len
) {
1779 error_report("RP: Failed reading data for message 0x%04x"
1780 " read %d expected %d",
1781 header_type
, res
, header_len
);
1782 mark_source_rp_bad(ms
);
1786 /* OK, we have the message and the data */
1787 switch (header_type
) {
1788 case MIG_RP_MSG_SHUT
:
1789 sibling_error
= ldl_be_p(buf
);
1790 trace_source_return_path_thread_shut(sibling_error
);
1791 if (sibling_error
) {
1792 error_report("RP: Sibling indicated error %d", sibling_error
);
1793 mark_source_rp_bad(ms
);
1796 * We'll let the main thread deal with closing the RP
1797 * we could do a shutdown(2) on it, but we're the only user
1798 * anyway, so there's nothing gained.
1802 case MIG_RP_MSG_PONG
:
1803 tmp32
= ldl_be_p(buf
);
1804 trace_source_return_path_thread_pong(tmp32
);
1807 case MIG_RP_MSG_REQ_PAGES
:
1808 start
= ldq_be_p(buf
);
1809 len
= ldl_be_p(buf
+ 8);
1810 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1813 case MIG_RP_MSG_REQ_PAGES_ID
:
1814 expected_len
= 12 + 1; /* header + termination */
1816 if (header_len
>= expected_len
) {
1817 start
= ldq_be_p(buf
);
1818 len
= ldl_be_p(buf
+ 8);
1819 /* Now we expect an idstr */
1820 tmp32
= buf
[12]; /* Length of the following idstr */
1821 buf
[13 + tmp32
] = '\0';
1822 expected_len
+= tmp32
;
1824 if (header_len
!= expected_len
) {
1825 error_report("RP: Req_Page_id with length %d expecting %zd",
1826 header_len
, expected_len
);
1827 mark_source_rp_bad(ms
);
1830 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
1837 if (qemu_file_get_error(rp
)) {
1838 trace_source_return_path_thread_bad_end();
1839 mark_source_rp_bad(ms
);
1842 trace_source_return_path_thread_end();
1844 ms
->rp_state
.from_dst_file
= NULL
;
1849 static int open_return_path_on_source(MigrationState
*ms
)
1852 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
1853 if (!ms
->rp_state
.from_dst_file
) {
1857 trace_open_return_path_on_source();
1858 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
1859 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
1861 trace_open_return_path_on_source_continue();
1866 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
1867 static int await_return_path_close_on_source(MigrationState
*ms
)
1870 * If this is a normal exit then the destination will send a SHUT and the
1871 * rp_thread will exit, however if there's an error we need to cause
1874 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
1876 * shutdown(2), if we have it, will cause it to unblock if it's stuck
1877 * waiting for the destination.
1879 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
1880 mark_source_rp_bad(ms
);
1882 trace_await_return_path_close_on_source_joining();
1883 qemu_thread_join(&ms
->rp_state
.rp_thread
);
1884 trace_await_return_path_close_on_source_close();
1885 return ms
->rp_state
.error
;
1889 * Switch from normal iteration to postcopy
1890 * Returns non-0 on error
1892 static int postcopy_start(MigrationState
*ms
)
1895 QIOChannelBuffer
*bioc
;
1897 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1898 bool restart_block
= false;
1899 int cur_state
= MIGRATION_STATUS_ACTIVE
;
1900 if (!migrate_pause_before_switchover()) {
1901 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
1902 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1905 trace_postcopy_start();
1906 qemu_mutex_lock_iothread();
1907 trace_postcopy_start_set_run();
1909 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
1910 global_state_store();
1911 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
1916 ret
= migration_maybe_pause(ms
, &cur_state
,
1917 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1922 ret
= bdrv_inactivate_all();
1926 restart_block
= true;
1929 * Cause any non-postcopiable, but iterative devices to
1930 * send out their final data.
1932 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
1935 * in Finish migrate and with the io-lock held everything should
1936 * be quiet, but we've potentially still got dirty pages and we
1937 * need to tell the destination to throw any pages it's already received
1940 if (migrate_postcopy_ram()) {
1941 if (ram_postcopy_send_discard_bitmap(ms
)) {
1942 error_report("postcopy send discard bitmap failed");
1948 * send rest of state - note things that are doing postcopy
1949 * will notice we're in POSTCOPY_ACTIVE and not actually
1950 * wrap their state up here
1952 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
1953 if (migrate_postcopy_ram()) {
1954 /* Ping just for debugging, helps line traces up */
1955 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
1959 * While loading the device state we may trigger page transfer
1960 * requests and the fd must be free to process those, and thus
1961 * the destination must read the whole device state off the fd before
1962 * it starts processing it. Unfortunately the ad-hoc migration format
1963 * doesn't allow the destination to know the size to read without fully
1964 * parsing it through each devices load-state code (especially the open
1965 * coded devices that use get/put).
1966 * So we wrap the device state up in a package with a length at the start;
1967 * to do this we use a qemu_buf to hold the whole of the device state.
1969 bioc
= qio_channel_buffer_new(4096);
1970 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
1971 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
1972 object_unref(OBJECT(bioc
));
1975 * Make sure the receiver can get incoming pages before we send the rest
1978 qemu_savevm_send_postcopy_listen(fb
);
1980 qemu_savevm_state_complete_precopy(fb
, false, false);
1981 if (migrate_postcopy_ram()) {
1982 qemu_savevm_send_ping(fb
, 3);
1985 qemu_savevm_send_postcopy_run(fb
);
1987 /* <><> end of stuff going into the package */
1989 /* Last point of recovery; as soon as we send the package the destination
1990 * can open devices and potentially start running.
1991 * Lets just check again we've not got any errors.
1993 ret
= qemu_file_get_error(ms
->to_dst_file
);
1995 error_report("postcopy_start: Migration stream errored (pre package)");
1999 restart_block
= false;
2001 /* Now send that blob */
2002 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2007 /* Send a notify to give a chance for anything that needs to happen
2008 * at the transition to postcopy and after the device state; in particular
2009 * spice needs to trigger a transition now
2011 ms
->postcopy_after_devices
= true;
2012 notifier_list_notify(&migration_state_notifiers
, ms
);
2014 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2016 qemu_mutex_unlock_iothread();
2018 if (migrate_postcopy_ram()) {
2020 * Although this ping is just for debug, it could potentially be
2021 * used for getting a better measurement of downtime at the source.
2023 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2026 if (migrate_release_ram()) {
2027 ram_postcopy_migrated_memory_release(ms
);
2030 ret
= qemu_file_get_error(ms
->to_dst_file
);
2032 error_report("postcopy_start: Migration stream errored");
2033 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2034 MIGRATION_STATUS_FAILED
);
2042 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2043 MIGRATION_STATUS_FAILED
);
2044 if (restart_block
) {
2045 /* A failure happened early enough that we know the destination hasn't
2046 * accessed block devices, so we're safe to recover.
2048 Error
*local_err
= NULL
;
2050 bdrv_invalidate_cache_all(&local_err
);
2052 error_report_err(local_err
);
2055 qemu_mutex_unlock_iothread();
2060 * migration_maybe_pause: Pause if required to by
2061 * migrate_pause_before_switchover called with the iothread locked
2062 * Returns: 0 on success
2064 static int migration_maybe_pause(MigrationState
*s
,
2065 int *current_active_state
,
2068 if (!migrate_pause_before_switchover()) {
2072 /* Since leaving this state is not atomic with posting the semaphore
2073 * it's possible that someone could have issued multiple migrate_continue
2074 * and the semaphore is incorrectly positive at this point;
2075 * the docs say it's undefined to reinit a semaphore that's already
2076 * init'd, so use timedwait to eat up any existing posts.
2078 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2079 /* This block intentionally left blank */
2082 qemu_mutex_unlock_iothread();
2083 migrate_set_state(&s
->state
, *current_active_state
,
2084 MIGRATION_STATUS_PRE_SWITCHOVER
);
2085 qemu_sem_wait(&s
->pause_sem
);
2086 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2088 *current_active_state
= new_state
;
2089 qemu_mutex_lock_iothread();
2091 return s
->state
== new_state
? 0 : -EINVAL
;
2095 * migration_completion: Used by migration_thread when there's not much left.
2096 * The caller 'breaks' the loop when this returns.
2098 * @s: Current migration state
2100 static void migration_completion(MigrationState
*s
)
2103 int current_active_state
= s
->state
;
2105 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2106 qemu_mutex_lock_iothread();
2107 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2108 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
2109 s
->vm_was_running
= runstate_is_running();
2110 ret
= global_state_store();
2113 bool inactivate
= !migrate_colo_enabled();
2114 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2116 ret
= migration_maybe_pause(s
, ¤t_active_state
,
2117 MIGRATION_STATUS_DEVICE
);
2120 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
2121 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
2124 if (inactivate
&& ret
>= 0) {
2125 s
->block_inactive
= true;
2128 qemu_mutex_unlock_iothread();
2133 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2134 trace_migration_completion_postcopy_end();
2136 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
2137 trace_migration_completion_postcopy_end_after_complete();
2141 * If rp was opened we must clean up the thread before
2142 * cleaning everything else up (since if there are no failures
2143 * it will wait for the destination to send it's status in
2146 if (s
->rp_state
.from_dst_file
) {
2148 trace_migration_return_path_end_before();
2149 rp_error
= await_return_path_close_on_source(s
);
2150 trace_migration_return_path_end_after(rp_error
);
2152 goto fail_invalidate
;
2156 if (qemu_file_get_error(s
->to_dst_file
)) {
2157 trace_migration_completion_file_err();
2158 goto fail_invalidate
;
2161 if (!migrate_colo_enabled()) {
2162 migrate_set_state(&s
->state
, current_active_state
,
2163 MIGRATION_STATUS_COMPLETED
);
2169 /* If not doing postcopy, vm_start() will be called: let's regain
2170 * control on images.
2172 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
2173 s
->state
== MIGRATION_STATUS_DEVICE
) {
2174 Error
*local_err
= NULL
;
2176 qemu_mutex_lock_iothread();
2177 bdrv_invalidate_cache_all(&local_err
);
2179 error_report_err(local_err
);
2181 s
->block_inactive
= false;
2183 qemu_mutex_unlock_iothread();
2187 migrate_set_state(&s
->state
, current_active_state
,
2188 MIGRATION_STATUS_FAILED
);
2191 bool migrate_colo_enabled(void)
2193 MigrationState
*s
= migrate_get_current();
2194 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
2197 static void migration_calculate_complete(MigrationState
*s
)
2199 uint64_t bytes
= qemu_ftell(s
->to_dst_file
);
2200 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2202 s
->total_time
= end_time
- s
->start_time
;
2205 * It's still not set, so we are precopy migration. For
2206 * postcopy, downtime is calculated during postcopy_start().
2208 s
->downtime
= end_time
- s
->downtime_start
;
2211 if (s
->total_time
) {
2212 s
->mbps
= ((double) bytes
* 8.0) / s
->total_time
/ 1000;
2216 static void migration_update_counters(MigrationState
*s
,
2217 int64_t current_time
)
2219 uint64_t transferred
, time_spent
;
2222 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
2226 transferred
= qemu_ftell(s
->to_dst_file
) - s
->iteration_initial_bytes
;
2227 time_spent
= current_time
- s
->iteration_start_time
;
2228 bandwidth
= (double)transferred
/ time_spent
;
2229 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
2231 s
->mbps
= (((double) transferred
* 8.0) /
2232 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2235 * if we haven't sent anything, we don't want to
2236 * recalculate. 10000 is a small enough number for our purposes
2238 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
2239 s
->expected_downtime
= ram_counters
.dirty_pages_rate
*
2240 qemu_target_page_size() / bandwidth
;
2243 qemu_file_reset_rate_limit(s
->to_dst_file
);
2245 s
->iteration_start_time
= current_time
;
2246 s
->iteration_initial_bytes
= qemu_ftell(s
->to_dst_file
);
2248 trace_migrate_transferred(transferred
, time_spent
,
2249 bandwidth
, s
->threshold_size
);
2252 /* Migration thread iteration status */
2254 MIG_ITERATE_RESUME
, /* Resume current iteration */
2255 MIG_ITERATE_SKIP
, /* Skip current iteration */
2256 MIG_ITERATE_BREAK
, /* Break the loop */
2260 * Return true if continue to the next iteration directly, false
2263 static MigIterateState
migration_iteration_run(MigrationState
*s
)
2265 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
2266 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
2268 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
2269 &pend_compat
, &pend_post
);
2270 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
2272 trace_migrate_pending(pending_size
, s
->threshold_size
,
2273 pend_pre
, pend_compat
, pend_post
);
2275 if (pending_size
&& pending_size
>= s
->threshold_size
) {
2276 /* Still a significant amount to transfer */
2277 if (migrate_postcopy() && !in_postcopy
&&
2278 pend_pre
<= s
->threshold_size
&&
2279 atomic_read(&s
->start_postcopy
)) {
2280 if (postcopy_start(s
)) {
2281 error_report("%s: postcopy failed to start", __func__
);
2283 return MIG_ITERATE_SKIP
;
2285 /* Just another iteration step */
2286 qemu_savevm_state_iterate(s
->to_dst_file
,
2287 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2289 trace_migration_thread_low_pending(pending_size
);
2290 migration_completion(s
);
2291 return MIG_ITERATE_BREAK
;
2294 return MIG_ITERATE_RESUME
;
2297 static void migration_iteration_finish(MigrationState
*s
)
2299 /* If we enabled cpu throttling for auto-converge, turn it off. */
2300 cpu_throttle_stop();
2302 qemu_mutex_lock_iothread();
2304 case MIGRATION_STATUS_COMPLETED
:
2305 migration_calculate_complete(s
);
2306 runstate_set(RUN_STATE_POSTMIGRATE
);
2309 case MIGRATION_STATUS_ACTIVE
:
2311 * We should really assert here, but since it's during
2312 * migration, let's try to reduce the usage of assertions.
2314 if (!migrate_colo_enabled()) {
2315 error_report("%s: critical error: calling COLO code without "
2316 "COLO enabled", __func__
);
2318 migrate_start_colo_process(s
);
2320 * Fixme: we will run VM in COLO no matter its old running state.
2321 * After exited COLO, we will keep running.
2323 s
->vm_was_running
= true;
2325 case MIGRATION_STATUS_FAILED
:
2326 case MIGRATION_STATUS_CANCELLED
:
2327 if (s
->vm_was_running
) {
2330 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2331 runstate_set(RUN_STATE_POSTMIGRATE
);
2337 /* Should not reach here, but if so, forgive the VM. */
2338 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
2341 qemu_bh_schedule(s
->cleanup_bh
);
2342 qemu_mutex_unlock_iothread();
2346 * Master migration thread on the source VM.
2347 * It drives the migration and pumps the data down the outgoing channel.
2349 static void *migration_thread(void *opaque
)
2351 MigrationState
*s
= opaque
;
2352 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
2354 rcu_register_thread();
2356 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2358 qemu_savevm_state_header(s
->to_dst_file
);
2361 * If we opened the return path, we need to make sure dst has it
2364 if (s
->rp_state
.from_dst_file
) {
2365 /* Now tell the dest that it should open its end so it can reply */
2366 qemu_savevm_send_open_return_path(s
->to_dst_file
);
2368 /* And do a ping that will make stuff easier to debug */
2369 qemu_savevm_send_ping(s
->to_dst_file
, 1);
2372 if (migrate_postcopy()) {
2374 * Tell the destination that we *might* want to do postcopy later;
2375 * if the other end can't do postcopy it should fail now, nice and
2378 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
2381 qemu_savevm_state_setup(s
->to_dst_file
);
2383 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
2384 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2385 MIGRATION_STATUS_ACTIVE
);
2387 trace_migration_thread_setup_complete();
2389 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
2390 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2391 int64_t current_time
;
2393 if (!qemu_file_rate_limit(s
->to_dst_file
)) {
2394 MigIterateState iter_state
= migration_iteration_run(s
);
2395 if (iter_state
== MIG_ITERATE_SKIP
) {
2397 } else if (iter_state
== MIG_ITERATE_BREAK
) {
2402 if (qemu_file_get_error(s
->to_dst_file
)) {
2403 if (migration_is_setup_or_active(s
->state
)) {
2404 migrate_set_state(&s
->state
, s
->state
,
2405 MIGRATION_STATUS_FAILED
);
2407 trace_migration_thread_file_err();
2411 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2413 migration_update_counters(s
, current_time
);
2415 if (qemu_file_rate_limit(s
->to_dst_file
)) {
2416 /* usleep expects microseconds */
2417 g_usleep((s
->iteration_start_time
+ BUFFER_DELAY
-
2418 current_time
) * 1000);
2422 trace_migration_thread_after_loop();
2423 migration_iteration_finish(s
);
2424 rcu_unregister_thread();
2428 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
2430 s
->expected_downtime
= s
->parameters
.downtime_limit
;
2431 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
2433 migrate_fd_error(s
, error_in
);
2434 migrate_fd_cleanup(s
);
2438 qemu_file_set_blocking(s
->to_dst_file
, true);
2439 qemu_file_set_rate_limit(s
->to_dst_file
,
2440 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
2442 /* Notify before starting migration thread */
2443 notifier_list_notify(&migration_state_notifiers
, s
);
2446 * Open the return path. For postcopy, it is used exclusively. For
2447 * precopy, only if user specified "return-path" capability would
2448 * QEMU uses the return path.
2450 if (migrate_postcopy_ram() || migrate_use_return_path()) {
2451 if (open_return_path_on_source(s
)) {
2452 error_report("Unable to open return-path for postcopy");
2453 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2454 MIGRATION_STATUS_FAILED
);
2455 migrate_fd_cleanup(s
);
2460 if (multifd_save_setup() != 0) {
2461 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2462 MIGRATION_STATUS_FAILED
);
2463 migrate_fd_cleanup(s
);
2466 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
2467 QEMU_THREAD_JOINABLE
);
2468 s
->migration_thread_running
= true;
2471 void migration_global_dump(Monitor
*mon
)
2473 MigrationState
*ms
= migrate_get_current();
2475 monitor_printf(mon
, "globals:\n");
2476 monitor_printf(mon
, "store-global-state: %s\n",
2477 ms
->store_global_state
? "on" : "off");
2478 monitor_printf(mon
, "only-migratable: %s\n",
2479 ms
->only_migratable
? "on" : "off");
2480 monitor_printf(mon
, "send-configuration: %s\n",
2481 ms
->send_configuration
? "on" : "off");
2482 monitor_printf(mon
, "send-section-footer: %s\n",
2483 ms
->send_section_footer
? "on" : "off");
2486 #define DEFINE_PROP_MIG_CAP(name, x) \
2487 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
2489 static Property migration_properties
[] = {
2490 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
2491 store_global_state
, true),
2492 DEFINE_PROP_BOOL("only-migratable", MigrationState
, only_migratable
, false),
2493 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
2494 send_configuration
, true),
2495 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
2496 send_section_footer
, true),
2498 /* Migration parameters */
2499 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
2500 parameters
.compress_level
,
2501 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
2502 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
2503 parameters
.compress_threads
,
2504 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
2505 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
2506 parameters
.decompress_threads
,
2507 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
2508 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
2509 parameters
.cpu_throttle_initial
,
2510 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
2511 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
2512 parameters
.cpu_throttle_increment
,
2513 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
2514 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
2515 parameters
.max_bandwidth
, MAX_THROTTLE
),
2516 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
2517 parameters
.downtime_limit
,
2518 DEFAULT_MIGRATE_SET_DOWNTIME
),
2519 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
2520 parameters
.x_checkpoint_delay
,
2521 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
2522 DEFINE_PROP_UINT8("x-multifd-channels", MigrationState
,
2523 parameters
.x_multifd_channels
,
2524 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
2525 DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState
,
2526 parameters
.x_multifd_page_count
,
2527 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT
),
2528 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
2529 parameters
.xbzrle_cache_size
,
2530 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
2532 /* Migration capabilities */
2533 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
2534 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
2535 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
2536 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
2537 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
2538 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
2539 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
2540 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
2541 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
2542 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
2543 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
2544 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD
),
2546 DEFINE_PROP_END_OF_LIST(),
2549 static void migration_class_init(ObjectClass
*klass
, void *data
)
2551 DeviceClass
*dc
= DEVICE_CLASS(klass
);
2553 dc
->user_creatable
= false;
2554 dc
->props
= migration_properties
;
2557 static void migration_instance_finalize(Object
*obj
)
2559 MigrationState
*ms
= MIGRATION_OBJ(obj
);
2560 MigrationParameters
*params
= &ms
->parameters
;
2562 qemu_mutex_destroy(&ms
->error_mutex
);
2563 g_free(params
->tls_hostname
);
2564 g_free(params
->tls_creds
);
2565 qemu_sem_destroy(&ms
->pause_sem
);
2566 error_free(ms
->error
);
2569 static void migration_instance_init(Object
*obj
)
2571 MigrationState
*ms
= MIGRATION_OBJ(obj
);
2572 MigrationParameters
*params
= &ms
->parameters
;
2574 ms
->state
= MIGRATION_STATUS_NONE
;
2576 qemu_sem_init(&ms
->pause_sem
, 0);
2577 qemu_mutex_init(&ms
->error_mutex
);
2579 params
->tls_hostname
= g_strdup("");
2580 params
->tls_creds
= g_strdup("");
2582 /* Set has_* up only for parameter checks */
2583 params
->has_compress_level
= true;
2584 params
->has_compress_threads
= true;
2585 params
->has_decompress_threads
= true;
2586 params
->has_cpu_throttle_initial
= true;
2587 params
->has_cpu_throttle_increment
= true;
2588 params
->has_max_bandwidth
= true;
2589 params
->has_downtime_limit
= true;
2590 params
->has_x_checkpoint_delay
= true;
2591 params
->has_block_incremental
= true;
2592 params
->has_x_multifd_channels
= true;
2593 params
->has_x_multifd_page_count
= true;
2594 params
->has_xbzrle_cache_size
= true;
2598 * Return true if check pass, false otherwise. Error will be put
2599 * inside errp if provided.
2601 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
2603 MigrationCapabilityStatusList
*head
= NULL
;
2604 /* Assuming all off */
2605 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
2608 if (!migrate_params_check(&ms
->parameters
, errp
)) {
2612 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
2613 if (ms
->enabled_capabilities
[i
]) {
2614 head
= migrate_cap_add(head
, i
, true);
2618 ret
= migrate_caps_check(cap_list
, head
, errp
);
2620 /* It works with head == NULL */
2621 qapi_free_MigrationCapabilityStatusList(head
);
2626 static const TypeInfo migration_type
= {
2627 .name
= TYPE_MIGRATION
,
2629 * NOTE: TYPE_MIGRATION is not really a device, as the object is
2630 * not created using qdev_create(), it is not attached to the qdev
2631 * device tree, and it is never realized.
2633 * TODO: Make this TYPE_OBJECT once QOM provides something like
2634 * TYPE_DEVICE's "-global" properties.
2636 .parent
= TYPE_DEVICE
,
2637 .class_init
= migration_class_init
,
2638 .class_size
= sizeof(MigrationClass
),
2639 .instance_size
= sizeof(MigrationState
),
2640 .instance_init
= migration_instance_init
,
2641 .instance_finalize
= migration_instance_finalize
,
2644 static void register_migration_types(void)
2646 type_register_static(&migration_type
);
2649 type_init(register_migration_types
);