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) */
98 MIG_RP_MSG_RECV_BITMAP
, /* send recved_bitmap back to source */
103 /* When we add fault tolerance, we could have several
104 migrations at once. For now we don't need to add
105 dynamic creation of migration */
107 static MigrationState
*current_migration
;
109 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
110 static int migration_maybe_pause(MigrationState
*s
,
111 int *current_active_state
,
114 void migration_object_init(void)
116 MachineState
*ms
= MACHINE(qdev_get_machine());
119 /* This can only be called once. */
120 assert(!current_migration
);
121 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
123 if (!migration_object_check(current_migration
, &err
)) {
124 error_report_err(err
);
129 * We cannot really do this in migration_instance_init() since at
130 * that time global properties are not yet applied, then this
131 * value will be definitely replaced by something else.
133 if (ms
->enforce_config_section
) {
134 current_migration
->send_configuration
= true;
138 void migration_object_finalize(void)
140 object_unref(OBJECT(current_migration
));
144 MigrationState
*migrate_get_current(void)
146 /* This can only be called after the object created. */
147 assert(current_migration
);
148 return current_migration
;
151 MigrationIncomingState
*migration_incoming_get_current(void)
154 static MigrationIncomingState mis_current
;
157 mis_current
.state
= MIGRATION_STATUS_NONE
;
158 memset(&mis_current
, 0, sizeof(MigrationIncomingState
));
159 mis_current
.postcopy_remote_fds
= g_array_new(FALSE
, TRUE
,
160 sizeof(struct PostCopyFD
));
161 qemu_mutex_init(&mis_current
.rp_mutex
);
162 qemu_event_init(&mis_current
.main_thread_load_event
, false);
163 qemu_sem_init(&mis_current
.postcopy_pause_sem_dst
, 0);
164 qemu_sem_init(&mis_current
.postcopy_pause_sem_fault
, 0);
166 init_dirty_bitmap_incoming_migration();
173 void migration_incoming_state_destroy(void)
175 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
177 if (mis
->to_src_file
) {
178 /* Tell source that we are done */
179 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
180 qemu_fclose(mis
->to_src_file
);
181 mis
->to_src_file
= NULL
;
184 if (mis
->from_src_file
) {
185 qemu_fclose(mis
->from_src_file
);
186 mis
->from_src_file
= NULL
;
188 if (mis
->postcopy_remote_fds
) {
189 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
190 mis
->postcopy_remote_fds
= NULL
;
193 qemu_event_reset(&mis
->main_thread_load_event
);
196 static void migrate_generate_event(int new_state
)
198 if (migrate_use_events()) {
199 qapi_event_send_migration(new_state
, &error_abort
);
204 * Called on -incoming with a defer: uri.
205 * The migration can be started later after any parameters have been
208 static void deferred_incoming_migration(Error
**errp
)
210 if (deferred_incoming
) {
211 error_setg(errp
, "Incoming migration already deferred");
213 deferred_incoming
= true;
217 * Send a message on the return channel back to the source
220 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
221 enum mig_rp_message_type message_type
,
222 uint16_t len
, void *data
)
226 trace_migrate_send_rp_message((int)message_type
, len
);
227 qemu_mutex_lock(&mis
->rp_mutex
);
230 * It's possible that the file handle got lost due to network
233 if (!mis
->to_src_file
) {
238 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
239 qemu_put_be16(mis
->to_src_file
, len
);
240 qemu_put_buffer(mis
->to_src_file
, data
, len
);
241 qemu_fflush(mis
->to_src_file
);
243 /* It's possible that qemu file got error during sending */
244 ret
= qemu_file_get_error(mis
->to_src_file
);
247 qemu_mutex_unlock(&mis
->rp_mutex
);
251 /* Request a range of pages from the source VM at the given
253 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
254 * as the last request (a name must have been given previously)
255 * Start: Address offset within the RB
256 * Len: Length in bytes required - must be a multiple of pagesize
258 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
, const char *rbname
,
259 ram_addr_t start
, size_t len
)
261 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
262 size_t msglen
= 12; /* start + len */
263 enum mig_rp_message_type msg_type
;
265 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
266 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
269 int rbname_len
= strlen(rbname
);
270 assert(rbname_len
< 256);
272 bufc
[msglen
++] = rbname_len
;
273 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
274 msglen
+= rbname_len
;
275 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
277 msg_type
= MIG_RP_MSG_REQ_PAGES
;
280 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
283 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
287 qapi_event_send_migration(MIGRATION_STATUS_SETUP
, &error_abort
);
288 if (!strcmp(uri
, "defer")) {
289 deferred_incoming_migration(errp
);
290 } else if (strstart(uri
, "tcp:", &p
)) {
291 tcp_start_incoming_migration(p
, errp
);
293 } else if (strstart(uri
, "rdma:", &p
)) {
294 rdma_start_incoming_migration(p
, errp
);
296 } else if (strstart(uri
, "exec:", &p
)) {
297 exec_start_incoming_migration(p
, errp
);
298 } else if (strstart(uri
, "unix:", &p
)) {
299 unix_start_incoming_migration(p
, errp
);
300 } else if (strstart(uri
, "fd:", &p
)) {
301 fd_start_incoming_migration(p
, errp
);
303 error_setg(errp
, "unknown migration protocol: %s", uri
);
307 static void process_incoming_migration_bh(void *opaque
)
309 Error
*local_err
= NULL
;
310 MigrationIncomingState
*mis
= opaque
;
312 /* Make sure all file formats flush their mutable metadata.
313 * If we get an error here, just don't restart the VM yet. */
314 bdrv_invalidate_cache_all(&local_err
);
316 error_report_err(local_err
);
322 * This must happen after all error conditions are dealt with and
323 * we're sure the VM is going to be running on this host.
325 qemu_announce_self();
327 if (multifd_load_cleanup(&local_err
) != 0) {
328 error_report_err(local_err
);
331 /* If global state section was not received or we are in running
332 state, we need to obey autostart. Any other state is set with
335 dirty_bitmap_mig_before_vm_start();
337 if (!global_state_received() ||
338 global_state_get_runstate() == RUN_STATE_RUNNING
) {
342 runstate_set(RUN_STATE_PAUSED
);
345 runstate_set(global_state_get_runstate());
348 * This must happen after any state changes since as soon as an external
349 * observer sees this event they might start to prod at the VM assuming
352 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
353 MIGRATION_STATUS_COMPLETED
);
354 qemu_bh_delete(mis
->bh
);
355 migration_incoming_state_destroy();
358 static void process_incoming_migration_co(void *opaque
)
360 MigrationIncomingState
*mis
= migration_incoming_get_current();
364 assert(mis
->from_src_file
);
365 mis
->largest_page_size
= qemu_ram_pagesize_largest();
366 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
367 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
368 MIGRATION_STATUS_ACTIVE
);
369 ret
= qemu_loadvm_state(mis
->from_src_file
);
371 ps
= postcopy_state_get();
372 trace_process_incoming_migration_co_end(ret
, ps
);
373 if (ps
!= POSTCOPY_INCOMING_NONE
) {
374 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
376 * Where a migration had postcopy enabled (and thus went to advise)
377 * but managed to complete within the precopy period, we can use
380 postcopy_ram_incoming_cleanup(mis
);
381 } else if (ret
>= 0) {
383 * Postcopy was started, cleanup should happen at the end of the
386 trace_process_incoming_migration_co_postcopy_end_main();
389 /* Else if something went wrong then just fall out of the normal exit */
392 /* we get COLO info, and know if we are in COLO mode */
393 if (!ret
&& migration_incoming_enable_colo()) {
394 mis
->migration_incoming_co
= qemu_coroutine_self();
395 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
396 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
397 mis
->have_colo_incoming_thread
= true;
398 qemu_coroutine_yield();
400 /* Wait checkpoint incoming thread exit before free resource */
401 qemu_thread_join(&mis
->colo_incoming_thread
);
405 Error
*local_err
= NULL
;
407 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
408 MIGRATION_STATUS_FAILED
);
409 error_report("load of migration failed: %s", strerror(-ret
));
410 qemu_fclose(mis
->from_src_file
);
411 if (multifd_load_cleanup(&local_err
) != 0) {
412 error_report_err(local_err
);
416 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
417 qemu_bh_schedule(mis
->bh
);
420 static void migration_incoming_setup(QEMUFile
*f
)
422 MigrationIncomingState
*mis
= migration_incoming_get_current();
424 if (multifd_load_setup() != 0) {
425 /* We haven't been able to create multifd threads
426 nothing better to do */
430 if (!mis
->from_src_file
) {
431 mis
->from_src_file
= f
;
433 qemu_file_set_blocking(f
, false);
436 void migration_incoming_process(void)
438 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
439 qemu_coroutine_enter(co
);
442 void migration_fd_process_incoming(QEMUFile
*f
)
444 MigrationIncomingState
*mis
= migration_incoming_get_current();
446 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
447 /* Resumed from a paused postcopy migration */
449 mis
->from_src_file
= f
;
450 /* Postcopy has standalone thread to do vm load */
451 qemu_file_set_blocking(f
, true);
453 /* Re-configure the return path */
454 mis
->to_src_file
= qemu_file_get_return_path(f
);
456 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
457 MIGRATION_STATUS_POSTCOPY_RECOVER
);
460 * Here, we only wake up the main loading thread (while the
461 * fault thread will still be waiting), so that we can receive
462 * commands from source now, and answer it if needed. The
463 * fault thread will be woken up afterwards until we are sure
464 * that source is ready to reply to page requests.
466 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
468 /* New incoming migration */
469 migration_incoming_setup(f
);
470 migration_incoming_process();
474 void migration_ioc_process_incoming(QIOChannel
*ioc
)
476 MigrationIncomingState
*mis
= migration_incoming_get_current();
478 if (!mis
->from_src_file
) {
479 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
480 migration_incoming_setup(f
);
483 multifd_recv_new_channel(ioc
);
487 * @migration_has_all_channels: We have received all channels that we need
489 * Returns true when we have got connections to all the channels that
490 * we need for migration.
492 bool migration_has_all_channels(void)
496 all_channels
= multifd_recv_all_channels_created();
502 * Send a 'SHUT' message on the return channel with the given value
503 * to indicate that we've finished with the RP. Non-0 value indicates
506 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
511 buf
= cpu_to_be32(value
);
512 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
516 * Send a 'PONG' message on the return channel with the given value
517 * (normally in response to a 'PING')
519 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
524 buf
= cpu_to_be32(value
);
525 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
528 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
536 * First, we send the header part. It contains only the len of
537 * idstr, and the idstr itself.
539 len
= strlen(block_name
);
541 memcpy(buf
+ 1, block_name
, len
);
543 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
544 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
549 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
552 * Next, we dump the received bitmap to the stream.
554 * TODO: currently we are safe since we are the only one that is
555 * using the to_src_file handle (fault thread is still paused),
556 * and it's ok even not taking the mutex. However the best way is
557 * to take the lock before sending the message header, and release
558 * the lock after sending the bitmap.
560 qemu_mutex_lock(&mis
->rp_mutex
);
561 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
562 qemu_mutex_unlock(&mis
->rp_mutex
);
564 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
567 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
569 MigrationCapabilityStatusList
*head
= NULL
;
570 MigrationCapabilityStatusList
*caps
;
571 MigrationState
*s
= migrate_get_current();
574 caps
= NULL
; /* silence compiler warning */
575 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
576 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
577 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
582 head
= g_malloc0(sizeof(*caps
));
585 caps
->next
= g_malloc0(sizeof(*caps
));
589 g_malloc(sizeof(*caps
->value
));
590 caps
->value
->capability
= i
;
591 caps
->value
->state
= s
->enabled_capabilities
[i
];
597 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
599 MigrationParameters
*params
;
600 MigrationState
*s
= migrate_get_current();
602 /* TODO use QAPI_CLONE() instead of duplicating it inline */
603 params
= g_malloc0(sizeof(*params
));
604 params
->has_compress_level
= true;
605 params
->compress_level
= s
->parameters
.compress_level
;
606 params
->has_compress_threads
= true;
607 params
->compress_threads
= s
->parameters
.compress_threads
;
608 params
->has_decompress_threads
= true;
609 params
->decompress_threads
= s
->parameters
.decompress_threads
;
610 params
->has_cpu_throttle_initial
= true;
611 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
612 params
->has_cpu_throttle_increment
= true;
613 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
614 params
->has_tls_creds
= true;
615 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
616 params
->has_tls_hostname
= true;
617 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
618 params
->has_max_bandwidth
= true;
619 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
620 params
->has_downtime_limit
= true;
621 params
->downtime_limit
= s
->parameters
.downtime_limit
;
622 params
->has_x_checkpoint_delay
= true;
623 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
624 params
->has_block_incremental
= true;
625 params
->block_incremental
= s
->parameters
.block_incremental
;
626 params
->has_x_multifd_channels
= true;
627 params
->x_multifd_channels
= s
->parameters
.x_multifd_channels
;
628 params
->has_x_multifd_page_count
= true;
629 params
->x_multifd_page_count
= s
->parameters
.x_multifd_page_count
;
630 params
->has_xbzrle_cache_size
= true;
631 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
637 * Return true if we're already in the middle of a migration
638 * (i.e. any of the active or setup states)
640 static bool migration_is_setup_or_active(int state
)
643 case MIGRATION_STATUS_ACTIVE
:
644 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
645 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
646 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
647 case MIGRATION_STATUS_SETUP
:
648 case MIGRATION_STATUS_PRE_SWITCHOVER
:
649 case MIGRATION_STATUS_DEVICE
:
658 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
660 info
->has_ram
= true;
661 info
->ram
= g_malloc0(sizeof(*info
->ram
));
662 info
->ram
->transferred
= ram_counters
.transferred
;
663 info
->ram
->total
= ram_bytes_total();
664 info
->ram
->duplicate
= ram_counters
.duplicate
;
665 /* legacy value. It is not used anymore */
666 info
->ram
->skipped
= 0;
667 info
->ram
->normal
= ram_counters
.normal
;
668 info
->ram
->normal_bytes
= ram_counters
.normal
*
669 qemu_target_page_size();
670 info
->ram
->mbps
= s
->mbps
;
671 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
672 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
673 info
->ram
->page_size
= qemu_target_page_size();
675 if (migrate_use_xbzrle()) {
676 info
->has_xbzrle_cache
= true;
677 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
678 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
679 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
680 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
681 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
682 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
683 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
686 if (cpu_throttle_active()) {
687 info
->has_cpu_throttle_percentage
= true;
688 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
691 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
692 info
->ram
->remaining
= ram_bytes_remaining();
693 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
697 static void populate_disk_info(MigrationInfo
*info
)
699 if (blk_mig_active()) {
700 info
->has_disk
= true;
701 info
->disk
= g_malloc0(sizeof(*info
->disk
));
702 info
->disk
->transferred
= blk_mig_bytes_transferred();
703 info
->disk
->remaining
= blk_mig_bytes_remaining();
704 info
->disk
->total
= blk_mig_bytes_total();
708 static void fill_source_migration_info(MigrationInfo
*info
)
710 MigrationState
*s
= migrate_get_current();
713 case MIGRATION_STATUS_NONE
:
714 /* no migration has happened ever */
715 /* do not overwrite destination migration status */
718 case MIGRATION_STATUS_SETUP
:
719 info
->has_status
= true;
720 info
->has_total_time
= false;
722 case MIGRATION_STATUS_ACTIVE
:
723 case MIGRATION_STATUS_CANCELLING
:
724 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
725 case MIGRATION_STATUS_PRE_SWITCHOVER
:
726 case MIGRATION_STATUS_DEVICE
:
727 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
728 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
729 /* TODO add some postcopy stats */
730 info
->has_status
= true;
731 info
->has_total_time
= true;
732 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
734 info
->has_expected_downtime
= true;
735 info
->expected_downtime
= s
->expected_downtime
;
736 info
->has_setup_time
= true;
737 info
->setup_time
= s
->setup_time
;
739 populate_ram_info(info
, s
);
740 populate_disk_info(info
);
742 case MIGRATION_STATUS_COLO
:
743 info
->has_status
= true;
744 /* TODO: display COLO specific information (checkpoint info etc.) */
746 case MIGRATION_STATUS_COMPLETED
:
747 info
->has_status
= true;
748 info
->has_total_time
= true;
749 info
->total_time
= s
->total_time
;
750 info
->has_downtime
= true;
751 info
->downtime
= s
->downtime
;
752 info
->has_setup_time
= true;
753 info
->setup_time
= s
->setup_time
;
755 populate_ram_info(info
, s
);
757 case MIGRATION_STATUS_FAILED
:
758 info
->has_status
= true;
760 info
->has_error_desc
= true;
761 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
764 case MIGRATION_STATUS_CANCELLED
:
765 info
->has_status
= true;
768 info
->status
= s
->state
;
772 * @migration_caps_check - check capability validity
774 * @cap_list: old capability list, array of bool
775 * @params: new capabilities to be applied soon
776 * @errp: set *errp if the check failed, with reason
778 * Returns true if check passed, otherwise false.
780 static bool migrate_caps_check(bool *cap_list
,
781 MigrationCapabilityStatusList
*params
,
784 MigrationCapabilityStatusList
*cap
;
785 bool old_postcopy_cap
;
786 MigrationIncomingState
*mis
= migration_incoming_get_current();
788 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
790 for (cap
= params
; cap
; cap
= cap
->next
) {
791 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
794 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
795 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
796 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
798 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
803 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
804 if (cap_list
[MIGRATION_CAPABILITY_COMPRESS
]) {
805 /* The decompression threads asynchronously write into RAM
806 * rather than use the atomic copies needed to avoid
807 * userfaulting. It should be possible to fix the decompression
808 * threads for compatibility in future.
810 error_setg(errp
, "Postcopy is not currently compatible "
815 /* This check is reasonably expensive, so only when it's being
816 * set the first time, also it's only the destination that needs
819 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
820 !postcopy_ram_supported_by_host(mis
)) {
821 /* postcopy_ram_supported_by_host will have emitted a more
824 error_setg(errp
, "Postcopy is not supported");
832 static void fill_destination_migration_info(MigrationInfo
*info
)
834 MigrationIncomingState
*mis
= migration_incoming_get_current();
836 switch (mis
->state
) {
837 case MIGRATION_STATUS_NONE
:
840 case MIGRATION_STATUS_SETUP
:
841 case MIGRATION_STATUS_CANCELLING
:
842 case MIGRATION_STATUS_CANCELLED
:
843 case MIGRATION_STATUS_ACTIVE
:
844 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
845 case MIGRATION_STATUS_FAILED
:
846 case MIGRATION_STATUS_COLO
:
847 info
->has_status
= true;
849 case MIGRATION_STATUS_COMPLETED
:
850 info
->has_status
= true;
851 fill_destination_postcopy_migration_info(info
);
854 info
->status
= mis
->state
;
857 MigrationInfo
*qmp_query_migrate(Error
**errp
)
859 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
861 fill_destination_migration_info(info
);
862 fill_source_migration_info(info
);
867 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
870 MigrationState
*s
= migrate_get_current();
871 MigrationCapabilityStatusList
*cap
;
872 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
874 if (migration_is_setup_or_active(s
->state
)) {
875 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
879 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
880 if (!migrate_caps_check(cap_list
, params
, errp
)) {
884 for (cap
= params
; cap
; cap
= cap
->next
) {
885 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
890 * Check whether the parameters are valid. Error will be put into errp
891 * (if provided). Return true if valid, otherwise false.
893 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
895 if (params
->has_compress_level
&&
896 (params
->compress_level
> 9)) {
897 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
898 "is invalid, it should be in the range of 0 to 9");
902 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
903 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
905 "is invalid, it should be in the range of 1 to 255");
909 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
910 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
911 "decompress_threads",
912 "is invalid, it should be in the range of 1 to 255");
916 if (params
->has_cpu_throttle_initial
&&
917 (params
->cpu_throttle_initial
< 1 ||
918 params
->cpu_throttle_initial
> 99)) {
919 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
920 "cpu_throttle_initial",
921 "an integer in the range of 1 to 99");
925 if (params
->has_cpu_throttle_increment
&&
926 (params
->cpu_throttle_increment
< 1 ||
927 params
->cpu_throttle_increment
> 99)) {
928 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
929 "cpu_throttle_increment",
930 "an integer in the range of 1 to 99");
934 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
935 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
936 " range of 0 to %zu bytes/second", SIZE_MAX
);
940 if (params
->has_downtime_limit
&&
941 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
942 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
943 "the range of 0 to %d milliseconds",
944 MAX_MIGRATE_DOWNTIME
);
948 /* x_checkpoint_delay is now always positive */
950 if (params
->has_x_multifd_channels
&& (params
->x_multifd_channels
< 1)) {
951 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
953 "is invalid, it should be in the range of 1 to 255");
956 if (params
->has_x_multifd_page_count
&&
957 (params
->x_multifd_page_count
< 1 ||
958 params
->x_multifd_page_count
> 10000)) {
959 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
960 "multifd_page_count",
961 "is invalid, it should be in the range of 1 to 10000");
965 if (params
->has_xbzrle_cache_size
&&
966 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
967 !is_power_of_2(params
->xbzrle_cache_size
))) {
968 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
970 "is invalid, it should be bigger than target page size"
971 " and a power of two");
978 static void migrate_params_test_apply(MigrateSetParameters
*params
,
979 MigrationParameters
*dest
)
981 *dest
= migrate_get_current()->parameters
;
983 /* TODO use QAPI_CLONE() instead of duplicating it inline */
985 if (params
->has_compress_level
) {
986 dest
->compress_level
= params
->compress_level
;
989 if (params
->has_compress_threads
) {
990 dest
->compress_threads
= params
->compress_threads
;
993 if (params
->has_decompress_threads
) {
994 dest
->decompress_threads
= params
->decompress_threads
;
997 if (params
->has_cpu_throttle_initial
) {
998 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
1001 if (params
->has_cpu_throttle_increment
) {
1002 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
1005 if (params
->has_tls_creds
) {
1006 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1007 dest
->tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1010 if (params
->has_tls_hostname
) {
1011 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1012 dest
->tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1015 if (params
->has_max_bandwidth
) {
1016 dest
->max_bandwidth
= params
->max_bandwidth
;
1019 if (params
->has_downtime_limit
) {
1020 dest
->downtime_limit
= params
->downtime_limit
;
1023 if (params
->has_x_checkpoint_delay
) {
1024 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
1027 if (params
->has_block_incremental
) {
1028 dest
->block_incremental
= params
->block_incremental
;
1030 if (params
->has_x_multifd_channels
) {
1031 dest
->x_multifd_channels
= params
->x_multifd_channels
;
1033 if (params
->has_x_multifd_page_count
) {
1034 dest
->x_multifd_page_count
= params
->x_multifd_page_count
;
1036 if (params
->has_xbzrle_cache_size
) {
1037 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
1041 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
1043 MigrationState
*s
= migrate_get_current();
1045 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1047 if (params
->has_compress_level
) {
1048 s
->parameters
.compress_level
= params
->compress_level
;
1051 if (params
->has_compress_threads
) {
1052 s
->parameters
.compress_threads
= params
->compress_threads
;
1055 if (params
->has_decompress_threads
) {
1056 s
->parameters
.decompress_threads
= params
->decompress_threads
;
1059 if (params
->has_cpu_throttle_initial
) {
1060 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
1063 if (params
->has_cpu_throttle_increment
) {
1064 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
1067 if (params
->has_tls_creds
) {
1068 g_free(s
->parameters
.tls_creds
);
1069 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1070 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1073 if (params
->has_tls_hostname
) {
1074 g_free(s
->parameters
.tls_hostname
);
1075 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1076 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1079 if (params
->has_max_bandwidth
) {
1080 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1081 if (s
->to_dst_file
) {
1082 qemu_file_set_rate_limit(s
->to_dst_file
,
1083 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1087 if (params
->has_downtime_limit
) {
1088 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1091 if (params
->has_x_checkpoint_delay
) {
1092 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1093 if (migration_in_colo_state()) {
1094 colo_checkpoint_notify(s
);
1098 if (params
->has_block_incremental
) {
1099 s
->parameters
.block_incremental
= params
->block_incremental
;
1101 if (params
->has_x_multifd_channels
) {
1102 s
->parameters
.x_multifd_channels
= params
->x_multifd_channels
;
1104 if (params
->has_x_multifd_page_count
) {
1105 s
->parameters
.x_multifd_page_count
= params
->x_multifd_page_count
;
1107 if (params
->has_xbzrle_cache_size
) {
1108 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1109 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1113 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1115 MigrationParameters tmp
;
1117 /* TODO Rewrite "" to null instead */
1118 if (params
->has_tls_creds
1119 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1120 qobject_unref(params
->tls_creds
->u
.n
);
1121 params
->tls_creds
->type
= QTYPE_QSTRING
;
1122 params
->tls_creds
->u
.s
= strdup("");
1124 /* TODO Rewrite "" to null instead */
1125 if (params
->has_tls_hostname
1126 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1127 qobject_unref(params
->tls_hostname
->u
.n
);
1128 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1129 params
->tls_hostname
->u
.s
= strdup("");
1132 migrate_params_test_apply(params
, &tmp
);
1134 if (!migrate_params_check(&tmp
, errp
)) {
1135 /* Invalid parameter */
1139 migrate_params_apply(params
, errp
);
1143 void qmp_migrate_start_postcopy(Error
**errp
)
1145 MigrationState
*s
= migrate_get_current();
1147 if (!migrate_postcopy()) {
1148 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1149 " the start of migration");
1153 if (s
->state
== MIGRATION_STATUS_NONE
) {
1154 error_setg(errp
, "Postcopy must be started after migration has been"
1159 * we don't error if migration has finished since that would be racy
1160 * with issuing this command.
1162 atomic_set(&s
->start_postcopy
, true);
1165 /* shared migration helpers */
1167 void migrate_set_state(int *state
, int old_state
, int new_state
)
1169 assert(new_state
< MIGRATION_STATUS__MAX
);
1170 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1171 trace_migrate_set_state(MigrationStatus_str(new_state
));
1172 migrate_generate_event(new_state
);
1176 static MigrationCapabilityStatusList
*migrate_cap_add(
1177 MigrationCapabilityStatusList
*list
,
1178 MigrationCapability index
,
1181 MigrationCapabilityStatusList
*cap
;
1183 cap
= g_new0(MigrationCapabilityStatusList
, 1);
1184 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
1185 cap
->value
->capability
= index
;
1186 cap
->value
->state
= state
;
1192 void migrate_set_block_enabled(bool value
, Error
**errp
)
1194 MigrationCapabilityStatusList
*cap
;
1196 cap
= migrate_cap_add(NULL
, MIGRATION_CAPABILITY_BLOCK
, value
);
1197 qmp_migrate_set_capabilities(cap
, errp
);
1198 qapi_free_MigrationCapabilityStatusList(cap
);
1201 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1203 s
->parameters
.block_incremental
= value
;
1206 static void block_cleanup_parameters(MigrationState
*s
)
1208 if (s
->must_remove_block_options
) {
1209 /* setting to false can never fail */
1210 migrate_set_block_enabled(false, &error_abort
);
1211 migrate_set_block_incremental(s
, false);
1212 s
->must_remove_block_options
= false;
1216 static void migrate_fd_cleanup(void *opaque
)
1218 MigrationState
*s
= opaque
;
1220 qemu_bh_delete(s
->cleanup_bh
);
1221 s
->cleanup_bh
= NULL
;
1223 qemu_savevm_state_cleanup();
1225 if (s
->to_dst_file
) {
1226 Error
*local_err
= NULL
;
1228 trace_migrate_fd_cleanup();
1229 qemu_mutex_unlock_iothread();
1230 if (s
->migration_thread_running
) {
1231 qemu_thread_join(&s
->thread
);
1232 s
->migration_thread_running
= false;
1234 qemu_mutex_lock_iothread();
1236 if (multifd_save_cleanup(&local_err
) != 0) {
1237 error_report_err(local_err
);
1239 qemu_fclose(s
->to_dst_file
);
1240 s
->to_dst_file
= NULL
;
1243 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
1244 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
1246 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1247 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1248 MIGRATION_STATUS_CANCELLED
);
1252 /* It is used on info migrate. We can't free it */
1253 error_report_err(error_copy(s
->error
));
1255 notifier_list_notify(&migration_state_notifiers
, s
);
1256 block_cleanup_parameters(s
);
1259 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1261 qemu_mutex_lock(&s
->error_mutex
);
1263 s
->error
= error_copy(error
);
1265 qemu_mutex_unlock(&s
->error_mutex
);
1268 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1270 trace_migrate_fd_error(error_get_pretty(error
));
1271 assert(s
->to_dst_file
== NULL
);
1272 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1273 MIGRATION_STATUS_FAILED
);
1274 migrate_set_error(s
, error
);
1277 static void migrate_fd_cancel(MigrationState
*s
)
1280 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1281 trace_migrate_fd_cancel();
1283 if (s
->rp_state
.from_dst_file
) {
1284 /* shutdown the rp socket, so causing the rp thread to shutdown */
1285 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1289 old_state
= s
->state
;
1290 if (!migration_is_setup_or_active(old_state
)) {
1293 /* If the migration is paused, kick it out of the pause */
1294 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1295 qemu_sem_post(&s
->pause_sem
);
1297 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1298 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1301 * If we're unlucky the migration code might be stuck somewhere in a
1302 * send/write while the network has failed and is waiting to timeout;
1303 * if we've got shutdown(2) available then we can force it to quit.
1304 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1305 * called in a bh, so there is no race against this cancel.
1307 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1308 qemu_file_shutdown(f
);
1310 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1311 Error
*local_err
= NULL
;
1313 bdrv_invalidate_cache_all(&local_err
);
1315 error_report_err(local_err
);
1317 s
->block_inactive
= false;
1322 void add_migration_state_change_notifier(Notifier
*notify
)
1324 notifier_list_add(&migration_state_notifiers
, notify
);
1327 void remove_migration_state_change_notifier(Notifier
*notify
)
1329 notifier_remove(notify
);
1332 bool migration_in_setup(MigrationState
*s
)
1334 return s
->state
== MIGRATION_STATUS_SETUP
;
1337 bool migration_has_finished(MigrationState
*s
)
1339 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1342 bool migration_has_failed(MigrationState
*s
)
1344 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1345 s
->state
== MIGRATION_STATUS_FAILED
);
1348 bool migration_in_postcopy(void)
1350 MigrationState
*s
= migrate_get_current();
1352 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1355 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1357 return migration_in_postcopy() && s
->postcopy_after_devices
;
1360 bool migration_is_idle(void)
1362 MigrationState
*s
= migrate_get_current();
1365 case MIGRATION_STATUS_NONE
:
1366 case MIGRATION_STATUS_CANCELLED
:
1367 case MIGRATION_STATUS_COMPLETED
:
1368 case MIGRATION_STATUS_FAILED
:
1370 case MIGRATION_STATUS_SETUP
:
1371 case MIGRATION_STATUS_CANCELLING
:
1372 case MIGRATION_STATUS_ACTIVE
:
1373 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1374 case MIGRATION_STATUS_COLO
:
1375 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1376 case MIGRATION_STATUS_DEVICE
:
1378 case MIGRATION_STATUS__MAX
:
1379 g_assert_not_reached();
1385 void migrate_init(MigrationState
*s
)
1388 * Reinitialise all migration state, except
1389 * parameters/capabilities that the user set, and
1395 s
->to_dst_file
= NULL
;
1396 s
->state
= MIGRATION_STATUS_NONE
;
1397 s
->rp_state
.from_dst_file
= NULL
;
1398 s
->rp_state
.error
= false;
1401 s
->expected_downtime
= 0;
1403 s
->start_postcopy
= false;
1404 s
->postcopy_after_devices
= false;
1405 s
->migration_thread_running
= false;
1406 error_free(s
->error
);
1409 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1411 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1413 s
->vm_was_running
= false;
1414 s
->iteration_initial_bytes
= 0;
1415 s
->threshold_size
= 0;
1418 static GSList
*migration_blockers
;
1420 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1422 if (migrate_get_current()->only_migratable
) {
1423 error_propagate(errp
, error_copy(reason
));
1424 error_prepend(errp
, "disallowing migration blocker "
1425 "(--only_migratable) for: ");
1429 if (migration_is_idle()) {
1430 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1434 error_propagate(errp
, error_copy(reason
));
1435 error_prepend(errp
, "disallowing migration blocker (migration in "
1440 void migrate_del_blocker(Error
*reason
)
1442 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1445 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1447 Error
*local_err
= NULL
;
1448 static bool once
= true;
1450 if (!deferred_incoming
) {
1451 error_setg(errp
, "For use with '-incoming defer'");
1455 error_setg(errp
, "The incoming migration has already been started");
1458 qemu_start_incoming_migration(uri
, &local_err
);
1461 error_propagate(errp
, local_err
);
1468 bool migration_is_blocked(Error
**errp
)
1470 if (qemu_savevm_state_blocked(errp
)) {
1474 if (migration_blockers
) {
1475 error_propagate(errp
, error_copy(migration_blockers
->data
));
1482 /* Returns true if continue to migrate, or false if error detected */
1483 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
1484 bool resume
, Error
**errp
)
1486 Error
*local_err
= NULL
;
1489 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1490 error_setg(errp
, "Cannot resume if there is no "
1491 "paused migration");
1494 /* This is a resume, skip init status */
1498 if (migration_is_setup_or_active(s
->state
) ||
1499 s
->state
== MIGRATION_STATUS_CANCELLING
||
1500 s
->state
== MIGRATION_STATUS_COLO
) {
1501 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1505 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1506 error_setg(errp
, "Guest is waiting for an incoming migration");
1510 if (migration_is_blocked(errp
)) {
1514 if (blk
|| blk_inc
) {
1515 if (migrate_use_block() || migrate_use_block_incremental()) {
1516 error_setg(errp
, "Command options are incompatible with "
1517 "current migration capabilities");
1520 migrate_set_block_enabled(true, &local_err
);
1522 error_propagate(errp
, local_err
);
1525 s
->must_remove_block_options
= true;
1529 migrate_set_block_incremental(s
, true);
1537 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1538 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1539 bool has_resume
, bool resume
, Error
**errp
)
1541 Error
*local_err
= NULL
;
1542 MigrationState
*s
= migrate_get_current();
1545 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
1546 has_resume
&& resume
, errp
)) {
1547 /* Error detected, put into errp */
1551 if (strstart(uri
, "tcp:", &p
)) {
1552 tcp_start_outgoing_migration(s
, p
, &local_err
);
1554 } else if (strstart(uri
, "rdma:", &p
)) {
1555 rdma_start_outgoing_migration(s
, p
, &local_err
);
1557 } else if (strstart(uri
, "exec:", &p
)) {
1558 exec_start_outgoing_migration(s
, p
, &local_err
);
1559 } else if (strstart(uri
, "unix:", &p
)) {
1560 unix_start_outgoing_migration(s
, p
, &local_err
);
1561 } else if (strstart(uri
, "fd:", &p
)) {
1562 fd_start_outgoing_migration(s
, p
, &local_err
);
1564 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1565 "a valid migration protocol");
1566 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1567 MIGRATION_STATUS_FAILED
);
1568 block_cleanup_parameters(s
);
1573 migrate_fd_error(s
, local_err
);
1574 error_propagate(errp
, local_err
);
1579 void qmp_migrate_cancel(Error
**errp
)
1581 migrate_fd_cancel(migrate_get_current());
1584 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
1586 MigrationState
*s
= migrate_get_current();
1587 if (s
->state
!= state
) {
1588 error_setg(errp
, "Migration not in expected state: %s",
1589 MigrationStatus_str(s
->state
));
1592 qemu_sem_post(&s
->pause_sem
);
1595 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1597 MigrateSetParameters p
= {
1598 .has_xbzrle_cache_size
= true,
1599 .xbzrle_cache_size
= value
,
1602 qmp_migrate_set_parameters(&p
, errp
);
1605 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1607 return migrate_xbzrle_cache_size();
1610 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1612 MigrateSetParameters p
= {
1613 .has_max_bandwidth
= true,
1614 .max_bandwidth
= value
,
1617 qmp_migrate_set_parameters(&p
, errp
);
1620 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1622 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
1623 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1624 "the range of 0 to %d seconds",
1625 MAX_MIGRATE_DOWNTIME_SECONDS
);
1629 value
*= 1000; /* Convert to milliseconds */
1630 value
= MAX(0, MIN(INT64_MAX
, value
));
1632 MigrateSetParameters p
= {
1633 .has_downtime_limit
= true,
1634 .downtime_limit
= value
,
1637 qmp_migrate_set_parameters(&p
, errp
);
1640 bool migrate_release_ram(void)
1644 s
= migrate_get_current();
1646 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
1649 bool migrate_postcopy_ram(void)
1653 s
= migrate_get_current();
1655 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1658 bool migrate_postcopy(void)
1660 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
1663 bool migrate_auto_converge(void)
1667 s
= migrate_get_current();
1669 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1672 bool migrate_zero_blocks(void)
1676 s
= migrate_get_current();
1678 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
1681 bool migrate_postcopy_blocktime(void)
1685 s
= migrate_get_current();
1687 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
1690 bool migrate_use_compression(void)
1694 s
= migrate_get_current();
1696 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
1699 int migrate_compress_level(void)
1703 s
= migrate_get_current();
1705 return s
->parameters
.compress_level
;
1708 int migrate_compress_threads(void)
1712 s
= migrate_get_current();
1714 return s
->parameters
.compress_threads
;
1717 int migrate_decompress_threads(void)
1721 s
= migrate_get_current();
1723 return s
->parameters
.decompress_threads
;
1726 bool migrate_dirty_bitmaps(void)
1730 s
= migrate_get_current();
1732 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
1735 bool migrate_use_events(void)
1739 s
= migrate_get_current();
1741 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
1744 bool migrate_use_multifd(void)
1748 s
= migrate_get_current();
1750 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_MULTIFD
];
1753 bool migrate_pause_before_switchover(void)
1757 s
= migrate_get_current();
1759 return s
->enabled_capabilities
[
1760 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
1763 int migrate_multifd_channels(void)
1767 s
= migrate_get_current();
1769 return s
->parameters
.x_multifd_channels
;
1772 int migrate_multifd_page_count(void)
1776 s
= migrate_get_current();
1778 return s
->parameters
.x_multifd_page_count
;
1781 int migrate_use_xbzrle(void)
1785 s
= migrate_get_current();
1787 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
1790 int64_t migrate_xbzrle_cache_size(void)
1794 s
= migrate_get_current();
1796 return s
->parameters
.xbzrle_cache_size
;
1799 bool migrate_use_block(void)
1803 s
= migrate_get_current();
1805 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
1808 bool migrate_use_return_path(void)
1812 s
= migrate_get_current();
1814 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
1817 bool migrate_use_block_incremental(void)
1821 s
= migrate_get_current();
1823 return s
->parameters
.block_incremental
;
1826 /* migration thread support */
1828 * Something bad happened to the RP stream, mark an error
1829 * The caller shall print or trace something to indicate why
1831 static void mark_source_rp_bad(MigrationState
*s
)
1833 s
->rp_state
.error
= true;
1836 static struct rp_cmd_args
{
1837 ssize_t len
; /* -1 = variable */
1840 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1841 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1842 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1843 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1844 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1845 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
1846 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1850 * Process a request for pages received on the return path,
1851 * We're allowed to send more than requested (e.g. to round to our page size)
1852 * and we don't need to send pages that have already been sent.
1854 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1855 ram_addr_t start
, size_t len
)
1857 long our_host_ps
= getpagesize();
1859 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1862 * Since we currently insist on matching page sizes, just sanity check
1863 * we're being asked for whole host pages.
1865 if (start
& (our_host_ps
-1) ||
1866 (len
& (our_host_ps
-1))) {
1867 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1868 " len: %zd", __func__
, start
, len
);
1869 mark_source_rp_bad(ms
);
1873 if (ram_save_queue_pages(rbname
, start
, len
)) {
1874 mark_source_rp_bad(ms
);
1878 /* Return true to retry, false to quit */
1879 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
1881 trace_postcopy_pause_return_path();
1883 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
1885 trace_postcopy_pause_return_path_continued();
1890 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
1892 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
1895 error_report("%s: invalid block name '%s'", __func__
, block_name
);
1899 /* Fetch the received bitmap and refresh the dirty bitmap */
1900 return ram_dirty_bitmap_reload(s
, block
);
1904 * Handles messages sent on the return path towards the source VM
1907 static void *source_return_path_thread(void *opaque
)
1909 MigrationState
*ms
= opaque
;
1910 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1911 uint16_t header_len
, header_type
;
1913 uint32_t tmp32
, sibling_error
;
1914 ram_addr_t start
= 0; /* =0 to silence warning */
1915 size_t len
= 0, expected_len
;
1918 trace_source_return_path_thread_entry();
1921 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1922 migration_is_setup_or_active(ms
->state
)) {
1923 trace_source_return_path_thread_loop_top();
1924 header_type
= qemu_get_be16(rp
);
1925 header_len
= qemu_get_be16(rp
);
1927 if (qemu_file_get_error(rp
)) {
1928 mark_source_rp_bad(ms
);
1932 if (header_type
>= MIG_RP_MSG_MAX
||
1933 header_type
== MIG_RP_MSG_INVALID
) {
1934 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1935 header_type
, header_len
);
1936 mark_source_rp_bad(ms
);
1940 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1941 header_len
!= rp_cmd_args
[header_type
].len
) ||
1942 header_len
> sizeof(buf
)) {
1943 error_report("RP: Received '%s' message (0x%04x) with"
1944 "incorrect length %d expecting %zu",
1945 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1946 (size_t)rp_cmd_args
[header_type
].len
);
1947 mark_source_rp_bad(ms
);
1951 /* We know we've got a valid header by this point */
1952 res
= qemu_get_buffer(rp
, buf
, header_len
);
1953 if (res
!= header_len
) {
1954 error_report("RP: Failed reading data for message 0x%04x"
1955 " read %d expected %d",
1956 header_type
, res
, header_len
);
1957 mark_source_rp_bad(ms
);
1961 /* OK, we have the message and the data */
1962 switch (header_type
) {
1963 case MIG_RP_MSG_SHUT
:
1964 sibling_error
= ldl_be_p(buf
);
1965 trace_source_return_path_thread_shut(sibling_error
);
1966 if (sibling_error
) {
1967 error_report("RP: Sibling indicated error %d", sibling_error
);
1968 mark_source_rp_bad(ms
);
1971 * We'll let the main thread deal with closing the RP
1972 * we could do a shutdown(2) on it, but we're the only user
1973 * anyway, so there's nothing gained.
1977 case MIG_RP_MSG_PONG
:
1978 tmp32
= ldl_be_p(buf
);
1979 trace_source_return_path_thread_pong(tmp32
);
1982 case MIG_RP_MSG_REQ_PAGES
:
1983 start
= ldq_be_p(buf
);
1984 len
= ldl_be_p(buf
+ 8);
1985 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1988 case MIG_RP_MSG_REQ_PAGES_ID
:
1989 expected_len
= 12 + 1; /* header + termination */
1991 if (header_len
>= expected_len
) {
1992 start
= ldq_be_p(buf
);
1993 len
= ldl_be_p(buf
+ 8);
1994 /* Now we expect an idstr */
1995 tmp32
= buf
[12]; /* Length of the following idstr */
1996 buf
[13 + tmp32
] = '\0';
1997 expected_len
+= tmp32
;
1999 if (header_len
!= expected_len
) {
2000 error_report("RP: Req_Page_id with length %d expecting %zd",
2001 header_len
, expected_len
);
2002 mark_source_rp_bad(ms
);
2005 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
2008 case MIG_RP_MSG_RECV_BITMAP
:
2009 if (header_len
< 1) {
2010 error_report("%s: missing block name", __func__
);
2011 mark_source_rp_bad(ms
);
2014 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2015 buf
[buf
[0] + 1] = '\0';
2016 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
2017 mark_source_rp_bad(ms
);
2028 res
= qemu_file_get_error(rp
);
2032 * Maybe there is something we can do: it looks like a
2033 * network down issue, and we pause for a recovery.
2035 if (postcopy_pause_return_path_thread(ms
)) {
2036 /* Reload rp, reset the rest */
2037 rp
= ms
->rp_state
.from_dst_file
;
2038 ms
->rp_state
.error
= false;
2043 trace_source_return_path_thread_bad_end();
2044 mark_source_rp_bad(ms
);
2047 trace_source_return_path_thread_end();
2048 ms
->rp_state
.from_dst_file
= NULL
;
2053 static int open_return_path_on_source(MigrationState
*ms
,
2057 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
2058 if (!ms
->rp_state
.from_dst_file
) {
2062 trace_open_return_path_on_source();
2064 if (!create_thread
) {
2069 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
2070 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
2072 trace_open_return_path_on_source_continue();
2077 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2078 static int await_return_path_close_on_source(MigrationState
*ms
)
2081 * If this is a normal exit then the destination will send a SHUT and the
2082 * rp_thread will exit, however if there's an error we need to cause
2085 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
2087 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2088 * waiting for the destination.
2090 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
2091 mark_source_rp_bad(ms
);
2093 trace_await_return_path_close_on_source_joining();
2094 qemu_thread_join(&ms
->rp_state
.rp_thread
);
2095 trace_await_return_path_close_on_source_close();
2096 return ms
->rp_state
.error
;
2100 * Switch from normal iteration to postcopy
2101 * Returns non-0 on error
2103 static int postcopy_start(MigrationState
*ms
)
2106 QIOChannelBuffer
*bioc
;
2108 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2109 bool restart_block
= false;
2110 int cur_state
= MIGRATION_STATUS_ACTIVE
;
2111 if (!migrate_pause_before_switchover()) {
2112 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
2113 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2116 trace_postcopy_start();
2117 qemu_mutex_lock_iothread();
2118 trace_postcopy_start_set_run();
2120 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
2121 global_state_store();
2122 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2127 ret
= migration_maybe_pause(ms
, &cur_state
,
2128 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2133 ret
= bdrv_inactivate_all();
2137 restart_block
= true;
2140 * Cause any non-postcopiable, but iterative devices to
2141 * send out their final data.
2143 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2146 * in Finish migrate and with the io-lock held everything should
2147 * be quiet, but we've potentially still got dirty pages and we
2148 * need to tell the destination to throw any pages it's already received
2151 if (migrate_postcopy_ram()) {
2152 if (ram_postcopy_send_discard_bitmap(ms
)) {
2153 error_report("postcopy send discard bitmap failed");
2159 * send rest of state - note things that are doing postcopy
2160 * will notice we're in POSTCOPY_ACTIVE and not actually
2161 * wrap their state up here
2163 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
2164 if (migrate_postcopy_ram()) {
2165 /* Ping just for debugging, helps line traces up */
2166 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
2170 * While loading the device state we may trigger page transfer
2171 * requests and the fd must be free to process those, and thus
2172 * the destination must read the whole device state off the fd before
2173 * it starts processing it. Unfortunately the ad-hoc migration format
2174 * doesn't allow the destination to know the size to read without fully
2175 * parsing it through each devices load-state code (especially the open
2176 * coded devices that use get/put).
2177 * So we wrap the device state up in a package with a length at the start;
2178 * to do this we use a qemu_buf to hold the whole of the device state.
2180 bioc
= qio_channel_buffer_new(4096);
2181 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
2182 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
2183 object_unref(OBJECT(bioc
));
2186 * Make sure the receiver can get incoming pages before we send the rest
2189 qemu_savevm_send_postcopy_listen(fb
);
2191 qemu_savevm_state_complete_precopy(fb
, false, false);
2192 if (migrate_postcopy_ram()) {
2193 qemu_savevm_send_ping(fb
, 3);
2196 qemu_savevm_send_postcopy_run(fb
);
2198 /* <><> end of stuff going into the package */
2200 /* Last point of recovery; as soon as we send the package the destination
2201 * can open devices and potentially start running.
2202 * Lets just check again we've not got any errors.
2204 ret
= qemu_file_get_error(ms
->to_dst_file
);
2206 error_report("postcopy_start: Migration stream errored (pre package)");
2210 restart_block
= false;
2212 /* Now send that blob */
2213 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2218 /* Send a notify to give a chance for anything that needs to happen
2219 * at the transition to postcopy and after the device state; in particular
2220 * spice needs to trigger a transition now
2222 ms
->postcopy_after_devices
= true;
2223 notifier_list_notify(&migration_state_notifiers
, ms
);
2225 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2227 qemu_mutex_unlock_iothread();
2229 if (migrate_postcopy_ram()) {
2231 * Although this ping is just for debug, it could potentially be
2232 * used for getting a better measurement of downtime at the source.
2234 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2237 if (migrate_release_ram()) {
2238 ram_postcopy_migrated_memory_release(ms
);
2241 ret
= qemu_file_get_error(ms
->to_dst_file
);
2243 error_report("postcopy_start: Migration stream errored");
2244 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2245 MIGRATION_STATUS_FAILED
);
2253 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2254 MIGRATION_STATUS_FAILED
);
2255 if (restart_block
) {
2256 /* A failure happened early enough that we know the destination hasn't
2257 * accessed block devices, so we're safe to recover.
2259 Error
*local_err
= NULL
;
2261 bdrv_invalidate_cache_all(&local_err
);
2263 error_report_err(local_err
);
2266 qemu_mutex_unlock_iothread();
2271 * migration_maybe_pause: Pause if required to by
2272 * migrate_pause_before_switchover called with the iothread locked
2273 * Returns: 0 on success
2275 static int migration_maybe_pause(MigrationState
*s
,
2276 int *current_active_state
,
2279 if (!migrate_pause_before_switchover()) {
2283 /* Since leaving this state is not atomic with posting the semaphore
2284 * it's possible that someone could have issued multiple migrate_continue
2285 * and the semaphore is incorrectly positive at this point;
2286 * the docs say it's undefined to reinit a semaphore that's already
2287 * init'd, so use timedwait to eat up any existing posts.
2289 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2290 /* This block intentionally left blank */
2293 qemu_mutex_unlock_iothread();
2294 migrate_set_state(&s
->state
, *current_active_state
,
2295 MIGRATION_STATUS_PRE_SWITCHOVER
);
2296 qemu_sem_wait(&s
->pause_sem
);
2297 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2299 *current_active_state
= new_state
;
2300 qemu_mutex_lock_iothread();
2302 return s
->state
== new_state
? 0 : -EINVAL
;
2306 * migration_completion: Used by migration_thread when there's not much left.
2307 * The caller 'breaks' the loop when this returns.
2309 * @s: Current migration state
2311 static void migration_completion(MigrationState
*s
)
2314 int current_active_state
= s
->state
;
2316 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2317 qemu_mutex_lock_iothread();
2318 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2319 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
2320 s
->vm_was_running
= runstate_is_running();
2321 ret
= global_state_store();
2324 bool inactivate
= !migrate_colo_enabled();
2325 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2327 ret
= migration_maybe_pause(s
, ¤t_active_state
,
2328 MIGRATION_STATUS_DEVICE
);
2331 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
2332 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
2335 if (inactivate
&& ret
>= 0) {
2336 s
->block_inactive
= true;
2339 qemu_mutex_unlock_iothread();
2344 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2345 trace_migration_completion_postcopy_end();
2347 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
2348 trace_migration_completion_postcopy_end_after_complete();
2352 * If rp was opened we must clean up the thread before
2353 * cleaning everything else up (since if there are no failures
2354 * it will wait for the destination to send it's status in
2357 if (s
->rp_state
.from_dst_file
) {
2359 trace_migration_return_path_end_before();
2360 rp_error
= await_return_path_close_on_source(s
);
2361 trace_migration_return_path_end_after(rp_error
);
2363 goto fail_invalidate
;
2367 if (qemu_file_get_error(s
->to_dst_file
)) {
2368 trace_migration_completion_file_err();
2369 goto fail_invalidate
;
2372 if (!migrate_colo_enabled()) {
2373 migrate_set_state(&s
->state
, current_active_state
,
2374 MIGRATION_STATUS_COMPLETED
);
2380 /* If not doing postcopy, vm_start() will be called: let's regain
2381 * control on images.
2383 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
2384 s
->state
== MIGRATION_STATUS_DEVICE
) {
2385 Error
*local_err
= NULL
;
2387 qemu_mutex_lock_iothread();
2388 bdrv_invalidate_cache_all(&local_err
);
2390 error_report_err(local_err
);
2392 s
->block_inactive
= false;
2394 qemu_mutex_unlock_iothread();
2398 migrate_set_state(&s
->state
, current_active_state
,
2399 MIGRATION_STATUS_FAILED
);
2402 bool migrate_colo_enabled(void)
2404 MigrationState
*s
= migrate_get_current();
2405 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
2408 typedef enum MigThrError
{
2409 /* No error detected */
2410 MIG_THR_ERR_NONE
= 0,
2411 /* Detected error, but resumed successfully */
2412 MIG_THR_ERR_RECOVERED
= 1,
2413 /* Detected fatal error, need to exit */
2414 MIG_THR_ERR_FATAL
= 2,
2417 /* Return zero if success, or <0 for error */
2418 static int postcopy_do_resume(MigrationState
*s
)
2420 /* TODO: do the resume logic */
2425 * We don't return until we are in a safe state to continue current
2426 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2427 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2429 static MigThrError
postcopy_pause(MigrationState
*s
)
2431 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2434 migrate_set_state(&s
->state
, s
->state
,
2435 MIGRATION_STATUS_POSTCOPY_PAUSED
);
2437 /* Current channel is possibly broken. Release it. */
2438 assert(s
->to_dst_file
);
2439 qemu_file_shutdown(s
->to_dst_file
);
2440 qemu_fclose(s
->to_dst_file
);
2441 s
->to_dst_file
= NULL
;
2443 error_report("Detected IO failure for postcopy. "
2444 "Migration paused.");
2447 * We wait until things fixed up. Then someone will setup the
2448 * status back for us.
2450 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2451 qemu_sem_wait(&s
->postcopy_pause_sem
);
2454 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2455 /* Woken up by a recover procedure. Give it a shot */
2458 * Firstly, let's wake up the return path now, with a new
2459 * return path channel.
2461 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
2463 /* Do the resume logic */
2464 if (postcopy_do_resume(s
) == 0) {
2465 /* Let's continue! */
2466 trace_postcopy_pause_continued();
2467 return MIG_THR_ERR_RECOVERED
;
2470 * Something wrong happened during the recovery, let's
2471 * pause again. Pause is always better than throwing
2477 /* This is not right... Time to quit. */
2478 return MIG_THR_ERR_FATAL
;
2483 static MigThrError
migration_detect_error(MigrationState
*s
)
2487 /* Try to detect any file errors */
2488 ret
= qemu_file_get_error(s
->to_dst_file
);
2491 /* Everything is fine */
2492 return MIG_THR_ERR_NONE
;
2495 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
== -EIO
) {
2497 * For postcopy, we allow the network to be down for a
2498 * while. After that, it can be continued by a
2501 return postcopy_pause(s
);
2504 * For precopy (or postcopy with error outside IO), we fail
2507 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
2508 trace_migration_thread_file_err();
2510 /* Time to stop the migration, now. */
2511 return MIG_THR_ERR_FATAL
;
2515 static void migration_calculate_complete(MigrationState
*s
)
2517 uint64_t bytes
= qemu_ftell(s
->to_dst_file
);
2518 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2520 s
->total_time
= end_time
- s
->start_time
;
2523 * It's still not set, so we are precopy migration. For
2524 * postcopy, downtime is calculated during postcopy_start().
2526 s
->downtime
= end_time
- s
->downtime_start
;
2529 if (s
->total_time
) {
2530 s
->mbps
= ((double) bytes
* 8.0) / s
->total_time
/ 1000;
2534 static void migration_update_counters(MigrationState
*s
,
2535 int64_t current_time
)
2537 uint64_t transferred
, time_spent
;
2540 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
2544 transferred
= qemu_ftell(s
->to_dst_file
) - s
->iteration_initial_bytes
;
2545 time_spent
= current_time
- s
->iteration_start_time
;
2546 bandwidth
= (double)transferred
/ time_spent
;
2547 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
2549 s
->mbps
= (((double) transferred
* 8.0) /
2550 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2553 * if we haven't sent anything, we don't want to
2554 * recalculate. 10000 is a small enough number for our purposes
2556 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
2557 s
->expected_downtime
= ram_counters
.dirty_pages_rate
*
2558 qemu_target_page_size() / bandwidth
;
2561 qemu_file_reset_rate_limit(s
->to_dst_file
);
2563 s
->iteration_start_time
= current_time
;
2564 s
->iteration_initial_bytes
= qemu_ftell(s
->to_dst_file
);
2566 trace_migrate_transferred(transferred
, time_spent
,
2567 bandwidth
, s
->threshold_size
);
2570 /* Migration thread iteration status */
2572 MIG_ITERATE_RESUME
, /* Resume current iteration */
2573 MIG_ITERATE_SKIP
, /* Skip current iteration */
2574 MIG_ITERATE_BREAK
, /* Break the loop */
2578 * Return true if continue to the next iteration directly, false
2581 static MigIterateState
migration_iteration_run(MigrationState
*s
)
2583 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
2584 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
2586 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
2587 &pend_compat
, &pend_post
);
2588 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
2590 trace_migrate_pending(pending_size
, s
->threshold_size
,
2591 pend_pre
, pend_compat
, pend_post
);
2593 if (pending_size
&& pending_size
>= s
->threshold_size
) {
2594 /* Still a significant amount to transfer */
2595 if (migrate_postcopy() && !in_postcopy
&&
2596 pend_pre
<= s
->threshold_size
&&
2597 atomic_read(&s
->start_postcopy
)) {
2598 if (postcopy_start(s
)) {
2599 error_report("%s: postcopy failed to start", __func__
);
2601 return MIG_ITERATE_SKIP
;
2603 /* Just another iteration step */
2604 qemu_savevm_state_iterate(s
->to_dst_file
,
2605 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2607 trace_migration_thread_low_pending(pending_size
);
2608 migration_completion(s
);
2609 return MIG_ITERATE_BREAK
;
2612 return MIG_ITERATE_RESUME
;
2615 static void migration_iteration_finish(MigrationState
*s
)
2617 /* If we enabled cpu throttling for auto-converge, turn it off. */
2618 cpu_throttle_stop();
2620 qemu_mutex_lock_iothread();
2622 case MIGRATION_STATUS_COMPLETED
:
2623 migration_calculate_complete(s
);
2624 runstate_set(RUN_STATE_POSTMIGRATE
);
2627 case MIGRATION_STATUS_ACTIVE
:
2629 * We should really assert here, but since it's during
2630 * migration, let's try to reduce the usage of assertions.
2632 if (!migrate_colo_enabled()) {
2633 error_report("%s: critical error: calling COLO code without "
2634 "COLO enabled", __func__
);
2636 migrate_start_colo_process(s
);
2638 * Fixme: we will run VM in COLO no matter its old running state.
2639 * After exited COLO, we will keep running.
2641 s
->vm_was_running
= true;
2643 case MIGRATION_STATUS_FAILED
:
2644 case MIGRATION_STATUS_CANCELLED
:
2645 if (s
->vm_was_running
) {
2648 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2649 runstate_set(RUN_STATE_POSTMIGRATE
);
2655 /* Should not reach here, but if so, forgive the VM. */
2656 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
2659 qemu_bh_schedule(s
->cleanup_bh
);
2660 qemu_mutex_unlock_iothread();
2664 * Master migration thread on the source VM.
2665 * It drives the migration and pumps the data down the outgoing channel.
2667 static void *migration_thread(void *opaque
)
2669 MigrationState
*s
= opaque
;
2670 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
2671 MigThrError thr_error
;
2673 rcu_register_thread();
2675 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2677 qemu_savevm_state_header(s
->to_dst_file
);
2680 * If we opened the return path, we need to make sure dst has it
2683 if (s
->rp_state
.from_dst_file
) {
2684 /* Now tell the dest that it should open its end so it can reply */
2685 qemu_savevm_send_open_return_path(s
->to_dst_file
);
2687 /* And do a ping that will make stuff easier to debug */
2688 qemu_savevm_send_ping(s
->to_dst_file
, 1);
2691 if (migrate_postcopy()) {
2693 * Tell the destination that we *might* want to do postcopy later;
2694 * if the other end can't do postcopy it should fail now, nice and
2697 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
2700 qemu_savevm_state_setup(s
->to_dst_file
);
2702 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
2703 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2704 MIGRATION_STATUS_ACTIVE
);
2706 trace_migration_thread_setup_complete();
2708 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
2709 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2710 int64_t current_time
;
2712 if (!qemu_file_rate_limit(s
->to_dst_file
)) {
2713 MigIterateState iter_state
= migration_iteration_run(s
);
2714 if (iter_state
== MIG_ITERATE_SKIP
) {
2716 } else if (iter_state
== MIG_ITERATE_BREAK
) {
2722 * Try to detect any kind of failures, and see whether we
2723 * should stop the migration now.
2725 thr_error
= migration_detect_error(s
);
2726 if (thr_error
== MIG_THR_ERR_FATAL
) {
2727 /* Stop migration */
2729 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
2731 * Just recovered from a e.g. network failure, reset all
2732 * the local variables. This is important to avoid
2733 * breaking transferred_bytes and bandwidth calculation
2735 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2736 s
->iteration_initial_bytes
= 0;
2739 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2741 migration_update_counters(s
, current_time
);
2743 if (qemu_file_rate_limit(s
->to_dst_file
)) {
2744 /* usleep expects microseconds */
2745 g_usleep((s
->iteration_start_time
+ BUFFER_DELAY
-
2746 current_time
) * 1000);
2750 trace_migration_thread_after_loop();
2751 migration_iteration_finish(s
);
2752 rcu_unregister_thread();
2756 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
2759 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
2761 s
->expected_downtime
= s
->parameters
.downtime_limit
;
2762 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
2764 migrate_fd_error(s
, error_in
);
2765 migrate_fd_cleanup(s
);
2770 /* This is a resumed migration */
2771 rate_limit
= INT64_MAX
;
2773 /* This is a fresh new migration */
2774 rate_limit
= s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
;
2775 s
->expected_downtime
= s
->parameters
.downtime_limit
;
2776 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
2778 /* Notify before starting migration thread */
2779 notifier_list_notify(&migration_state_notifiers
, s
);
2782 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
2783 qemu_file_set_blocking(s
->to_dst_file
, true);
2786 * Open the return path. For postcopy, it is used exclusively. For
2787 * precopy, only if user specified "return-path" capability would
2788 * QEMU uses the return path.
2790 if (migrate_postcopy_ram() || migrate_use_return_path()) {
2791 if (open_return_path_on_source(s
, !resume
)) {
2792 error_report("Unable to open return-path for postcopy");
2793 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
2794 migrate_fd_cleanup(s
);
2800 /* Wakeup the main migration thread to do the recovery */
2801 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
2802 MIGRATION_STATUS_POSTCOPY_RECOVER
);
2803 qemu_sem_post(&s
->postcopy_pause_sem
);
2807 if (multifd_save_setup() != 0) {
2808 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2809 MIGRATION_STATUS_FAILED
);
2810 migrate_fd_cleanup(s
);
2813 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
2814 QEMU_THREAD_JOINABLE
);
2815 s
->migration_thread_running
= true;
2818 void migration_global_dump(Monitor
*mon
)
2820 MigrationState
*ms
= migrate_get_current();
2822 monitor_printf(mon
, "globals:\n");
2823 monitor_printf(mon
, "store-global-state: %s\n",
2824 ms
->store_global_state
? "on" : "off");
2825 monitor_printf(mon
, "only-migratable: %s\n",
2826 ms
->only_migratable
? "on" : "off");
2827 monitor_printf(mon
, "send-configuration: %s\n",
2828 ms
->send_configuration
? "on" : "off");
2829 monitor_printf(mon
, "send-section-footer: %s\n",
2830 ms
->send_section_footer
? "on" : "off");
2833 #define DEFINE_PROP_MIG_CAP(name, x) \
2834 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
2836 static Property migration_properties
[] = {
2837 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
2838 store_global_state
, true),
2839 DEFINE_PROP_BOOL("only-migratable", MigrationState
, only_migratable
, false),
2840 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
2841 send_configuration
, true),
2842 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
2843 send_section_footer
, true),
2845 /* Migration parameters */
2846 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
2847 parameters
.compress_level
,
2848 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
2849 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
2850 parameters
.compress_threads
,
2851 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
2852 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
2853 parameters
.decompress_threads
,
2854 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
2855 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
2856 parameters
.cpu_throttle_initial
,
2857 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
2858 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
2859 parameters
.cpu_throttle_increment
,
2860 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
2861 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
2862 parameters
.max_bandwidth
, MAX_THROTTLE
),
2863 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
2864 parameters
.downtime_limit
,
2865 DEFAULT_MIGRATE_SET_DOWNTIME
),
2866 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
2867 parameters
.x_checkpoint_delay
,
2868 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
2869 DEFINE_PROP_UINT8("x-multifd-channels", MigrationState
,
2870 parameters
.x_multifd_channels
,
2871 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
2872 DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState
,
2873 parameters
.x_multifd_page_count
,
2874 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT
),
2875 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
2876 parameters
.xbzrle_cache_size
,
2877 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
2879 /* Migration capabilities */
2880 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
2881 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
2882 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
2883 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
2884 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
2885 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
2886 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
2887 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
2888 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
2889 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
2890 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
2891 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD
),
2893 DEFINE_PROP_END_OF_LIST(),
2896 static void migration_class_init(ObjectClass
*klass
, void *data
)
2898 DeviceClass
*dc
= DEVICE_CLASS(klass
);
2900 dc
->user_creatable
= false;
2901 dc
->props
= migration_properties
;
2904 static void migration_instance_finalize(Object
*obj
)
2906 MigrationState
*ms
= MIGRATION_OBJ(obj
);
2907 MigrationParameters
*params
= &ms
->parameters
;
2909 qemu_mutex_destroy(&ms
->error_mutex
);
2910 g_free(params
->tls_hostname
);
2911 g_free(params
->tls_creds
);
2912 qemu_sem_destroy(&ms
->pause_sem
);
2913 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
2914 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
2915 error_free(ms
->error
);
2918 static void migration_instance_init(Object
*obj
)
2920 MigrationState
*ms
= MIGRATION_OBJ(obj
);
2921 MigrationParameters
*params
= &ms
->parameters
;
2923 ms
->state
= MIGRATION_STATUS_NONE
;
2925 qemu_sem_init(&ms
->pause_sem
, 0);
2926 qemu_mutex_init(&ms
->error_mutex
);
2928 params
->tls_hostname
= g_strdup("");
2929 params
->tls_creds
= g_strdup("");
2931 /* Set has_* up only for parameter checks */
2932 params
->has_compress_level
= true;
2933 params
->has_compress_threads
= true;
2934 params
->has_decompress_threads
= true;
2935 params
->has_cpu_throttle_initial
= true;
2936 params
->has_cpu_throttle_increment
= true;
2937 params
->has_max_bandwidth
= true;
2938 params
->has_downtime_limit
= true;
2939 params
->has_x_checkpoint_delay
= true;
2940 params
->has_block_incremental
= true;
2941 params
->has_x_multifd_channels
= true;
2942 params
->has_x_multifd_page_count
= true;
2943 params
->has_xbzrle_cache_size
= true;
2945 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
2946 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
2950 * Return true if check pass, false otherwise. Error will be put
2951 * inside errp if provided.
2953 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
2955 MigrationCapabilityStatusList
*head
= NULL
;
2956 /* Assuming all off */
2957 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
2960 if (!migrate_params_check(&ms
->parameters
, errp
)) {
2964 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
2965 if (ms
->enabled_capabilities
[i
]) {
2966 head
= migrate_cap_add(head
, i
, true);
2970 ret
= migrate_caps_check(cap_list
, head
, errp
);
2972 /* It works with head == NULL */
2973 qapi_free_MigrationCapabilityStatusList(head
);
2978 static const TypeInfo migration_type
= {
2979 .name
= TYPE_MIGRATION
,
2981 * NOTE: TYPE_MIGRATION is not really a device, as the object is
2982 * not created using qdev_create(), it is not attached to the qdev
2983 * device tree, and it is never realized.
2985 * TODO: Make this TYPE_OBJECT once QOM provides something like
2986 * TYPE_DEVICE's "-global" properties.
2988 .parent
= TYPE_DEVICE
,
2989 .class_init
= migration_class_init
,
2990 .class_size
= sizeof(MigrationClass
),
2991 .instance_size
= sizeof(MigrationState
),
2992 .instance_init
= migration_instance_init
,
2993 .instance_finalize
= migration_instance_finalize
,
2996 static void register_migration_types(void)
2998 type_register_static(&migration_type
);
3001 type_init(register_migration_types
);