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 /* Background transfer rate for postcopy, 0 means unlimited, note
86 * that page requests can still exceed this limit.
88 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
90 static NotifierList migration_state_notifiers
=
91 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
93 static bool deferred_incoming
;
95 /* Messages sent on the return path from destination to source */
96 enum mig_rp_message_type
{
97 MIG_RP_MSG_INVALID
= 0, /* Must be 0 */
98 MIG_RP_MSG_SHUT
, /* sibling will not send any more RP messages */
99 MIG_RP_MSG_PONG
, /* Response to a PING; data (seq: be32 ) */
101 MIG_RP_MSG_REQ_PAGES_ID
, /* data (start: be64, len: be32, id: string) */
102 MIG_RP_MSG_REQ_PAGES
, /* data (start: be64, len: be32) */
103 MIG_RP_MSG_RECV_BITMAP
, /* send recved_bitmap back to source */
104 MIG_RP_MSG_RESUME_ACK
, /* tell source that we are ready to resume */
109 /* When we add fault tolerance, we could have several
110 migrations at once. For now we don't need to add
111 dynamic creation of migration */
113 static MigrationState
*current_migration
;
114 static MigrationIncomingState
*current_incoming
;
116 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
117 static int migration_maybe_pause(MigrationState
*s
,
118 int *current_active_state
,
121 void migration_object_init(void)
123 MachineState
*ms
= MACHINE(qdev_get_machine());
126 /* This can only be called once. */
127 assert(!current_migration
);
128 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
131 * Init the migrate incoming object as well no matter whether
132 * we'll use it or not.
134 assert(!current_incoming
);
135 current_incoming
= g_new0(MigrationIncomingState
, 1);
136 current_incoming
->state
= MIGRATION_STATUS_NONE
;
137 current_incoming
->postcopy_remote_fds
=
138 g_array_new(FALSE
, TRUE
, sizeof(struct PostCopyFD
));
139 qemu_mutex_init(¤t_incoming
->rp_mutex
);
140 qemu_event_init(¤t_incoming
->main_thread_load_event
, false);
141 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_dst
, 0);
142 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fault
, 0);
144 init_dirty_bitmap_incoming_migration();
146 if (!migration_object_check(current_migration
, &err
)) {
147 error_report_err(err
);
152 * We cannot really do this in migration_instance_init() since at
153 * that time global properties are not yet applied, then this
154 * value will be definitely replaced by something else.
156 if (ms
->enforce_config_section
) {
157 current_migration
->send_configuration
= true;
161 void migration_object_finalize(void)
163 object_unref(OBJECT(current_migration
));
167 MigrationState
*migrate_get_current(void)
169 /* This can only be called after the object created. */
170 assert(current_migration
);
171 return current_migration
;
174 MigrationIncomingState
*migration_incoming_get_current(void)
176 assert(current_incoming
);
177 return current_incoming
;
180 void migration_incoming_state_destroy(void)
182 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
184 if (mis
->to_src_file
) {
185 /* Tell source that we are done */
186 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
187 qemu_fclose(mis
->to_src_file
);
188 mis
->to_src_file
= NULL
;
191 if (mis
->from_src_file
) {
192 qemu_fclose(mis
->from_src_file
);
193 mis
->from_src_file
= NULL
;
195 if (mis
->postcopy_remote_fds
) {
196 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
197 mis
->postcopy_remote_fds
= NULL
;
200 qemu_event_reset(&mis
->main_thread_load_event
);
203 static void migrate_generate_event(int new_state
)
205 if (migrate_use_events()) {
206 qapi_event_send_migration(new_state
, &error_abort
);
210 static bool migrate_late_block_activate(void)
214 s
= migrate_get_current();
216 return s
->enabled_capabilities
[
217 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE
];
221 * Called on -incoming with a defer: uri.
222 * The migration can be started later after any parameters have been
225 static void deferred_incoming_migration(Error
**errp
)
227 if (deferred_incoming
) {
228 error_setg(errp
, "Incoming migration already deferred");
230 deferred_incoming
= true;
234 * Send a message on the return channel back to the source
237 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
238 enum mig_rp_message_type message_type
,
239 uint16_t len
, void *data
)
243 trace_migrate_send_rp_message((int)message_type
, len
);
244 qemu_mutex_lock(&mis
->rp_mutex
);
247 * It's possible that the file handle got lost due to network
250 if (!mis
->to_src_file
) {
255 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
256 qemu_put_be16(mis
->to_src_file
, len
);
257 qemu_put_buffer(mis
->to_src_file
, data
, len
);
258 qemu_fflush(mis
->to_src_file
);
260 /* It's possible that qemu file got error during sending */
261 ret
= qemu_file_get_error(mis
->to_src_file
);
264 qemu_mutex_unlock(&mis
->rp_mutex
);
268 /* Request a range of pages from the source VM at the given
270 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
271 * as the last request (a name must have been given previously)
272 * Start: Address offset within the RB
273 * Len: Length in bytes required - must be a multiple of pagesize
275 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
, const char *rbname
,
276 ram_addr_t start
, size_t len
)
278 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
279 size_t msglen
= 12; /* start + len */
280 enum mig_rp_message_type msg_type
;
282 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
283 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
286 int rbname_len
= strlen(rbname
);
287 assert(rbname_len
< 256);
289 bufc
[msglen
++] = rbname_len
;
290 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
291 msglen
+= rbname_len
;
292 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
294 msg_type
= MIG_RP_MSG_REQ_PAGES
;
297 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
300 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
304 qapi_event_send_migration(MIGRATION_STATUS_SETUP
, &error_abort
);
305 if (!strcmp(uri
, "defer")) {
306 deferred_incoming_migration(errp
);
307 } else if (strstart(uri
, "tcp:", &p
)) {
308 tcp_start_incoming_migration(p
, errp
);
310 } else if (strstart(uri
, "rdma:", &p
)) {
311 rdma_start_incoming_migration(p
, errp
);
313 } else if (strstart(uri
, "exec:", &p
)) {
314 exec_start_incoming_migration(p
, errp
);
315 } else if (strstart(uri
, "unix:", &p
)) {
316 unix_start_incoming_migration(p
, errp
);
317 } else if (strstart(uri
, "fd:", &p
)) {
318 fd_start_incoming_migration(p
, errp
);
320 error_setg(errp
, "unknown migration protocol: %s", uri
);
324 static void process_incoming_migration_bh(void *opaque
)
326 Error
*local_err
= NULL
;
327 MigrationIncomingState
*mis
= opaque
;
329 /* If capability late_block_activate is set:
330 * Only fire up the block code now if we're going to restart the
331 * VM, else 'cont' will do it.
332 * This causes file locking to happen; so we don't want it to happen
333 * unless we really are starting the VM.
335 if (!migrate_late_block_activate() ||
336 (autostart
&& (!global_state_received() ||
337 global_state_get_runstate() == RUN_STATE_RUNNING
))) {
338 /* Make sure all file formats flush their mutable metadata.
339 * If we get an error here, just don't restart the VM yet. */
340 bdrv_invalidate_cache_all(&local_err
);
342 error_report_err(local_err
);
349 * This must happen after all error conditions are dealt with and
350 * we're sure the VM is going to be running on this host.
352 qemu_announce_self();
354 if (multifd_load_cleanup(&local_err
) != 0) {
355 error_report_err(local_err
);
358 /* If global state section was not received or we are in running
359 state, we need to obey autostart. Any other state is set with
362 dirty_bitmap_mig_before_vm_start();
364 if (!global_state_received() ||
365 global_state_get_runstate() == RUN_STATE_RUNNING
) {
369 runstate_set(RUN_STATE_PAUSED
);
372 runstate_set(global_state_get_runstate());
375 * This must happen after any state changes since as soon as an external
376 * observer sees this event they might start to prod at the VM assuming
379 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
380 MIGRATION_STATUS_COMPLETED
);
381 qemu_bh_delete(mis
->bh
);
382 migration_incoming_state_destroy();
385 static void process_incoming_migration_co(void *opaque
)
387 MigrationIncomingState
*mis
= migration_incoming_get_current();
391 assert(mis
->from_src_file
);
392 mis
->largest_page_size
= qemu_ram_pagesize_largest();
393 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
394 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
395 MIGRATION_STATUS_ACTIVE
);
396 ret
= qemu_loadvm_state(mis
->from_src_file
);
398 ps
= postcopy_state_get();
399 trace_process_incoming_migration_co_end(ret
, ps
);
400 if (ps
!= POSTCOPY_INCOMING_NONE
) {
401 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
403 * Where a migration had postcopy enabled (and thus went to advise)
404 * but managed to complete within the precopy period, we can use
407 postcopy_ram_incoming_cleanup(mis
);
408 } else if (ret
>= 0) {
410 * Postcopy was started, cleanup should happen at the end of the
413 trace_process_incoming_migration_co_postcopy_end_main();
416 /* Else if something went wrong then just fall out of the normal exit */
419 /* we get COLO info, and know if we are in COLO mode */
420 if (!ret
&& migration_incoming_enable_colo()) {
421 mis
->migration_incoming_co
= qemu_coroutine_self();
422 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
423 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
424 mis
->have_colo_incoming_thread
= true;
425 qemu_coroutine_yield();
427 /* Wait checkpoint incoming thread exit before free resource */
428 qemu_thread_join(&mis
->colo_incoming_thread
);
432 Error
*local_err
= NULL
;
434 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
435 MIGRATION_STATUS_FAILED
);
436 error_report("load of migration failed: %s", strerror(-ret
));
437 qemu_fclose(mis
->from_src_file
);
438 if (multifd_load_cleanup(&local_err
) != 0) {
439 error_report_err(local_err
);
443 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
444 qemu_bh_schedule(mis
->bh
);
447 static void migration_incoming_setup(QEMUFile
*f
)
449 MigrationIncomingState
*mis
= migration_incoming_get_current();
451 if (multifd_load_setup() != 0) {
452 /* We haven't been able to create multifd threads
453 nothing better to do */
457 if (!mis
->from_src_file
) {
458 mis
->from_src_file
= f
;
460 qemu_file_set_blocking(f
, false);
463 void migration_incoming_process(void)
465 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
466 qemu_coroutine_enter(co
);
469 void migration_fd_process_incoming(QEMUFile
*f
)
471 MigrationIncomingState
*mis
= migration_incoming_get_current();
473 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
474 /* Resumed from a paused postcopy migration */
476 mis
->from_src_file
= f
;
477 /* Postcopy has standalone thread to do vm load */
478 qemu_file_set_blocking(f
, true);
480 /* Re-configure the return path */
481 mis
->to_src_file
= qemu_file_get_return_path(f
);
483 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
484 MIGRATION_STATUS_POSTCOPY_RECOVER
);
487 * Here, we only wake up the main loading thread (while the
488 * fault thread will still be waiting), so that we can receive
489 * commands from source now, and answer it if needed. The
490 * fault thread will be woken up afterwards until we are sure
491 * that source is ready to reply to page requests.
493 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
495 /* New incoming migration */
496 migration_incoming_setup(f
);
497 migration_incoming_process();
501 void migration_ioc_process_incoming(QIOChannel
*ioc
)
503 MigrationIncomingState
*mis
= migration_incoming_get_current();
505 if (!mis
->from_src_file
) {
506 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
507 migration_incoming_setup(f
);
510 multifd_recv_new_channel(ioc
);
514 * @migration_has_all_channels: We have received all channels that we need
516 * Returns true when we have got connections to all the channels that
517 * we need for migration.
519 bool migration_has_all_channels(void)
523 all_channels
= multifd_recv_all_channels_created();
529 * Send a 'SHUT' message on the return channel with the given value
530 * to indicate that we've finished with the RP. Non-0 value indicates
533 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
538 buf
= cpu_to_be32(value
);
539 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
543 * Send a 'PONG' message on the return channel with the given value
544 * (normally in response to a 'PING')
546 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
551 buf
= cpu_to_be32(value
);
552 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
555 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
563 * First, we send the header part. It contains only the len of
564 * idstr, and the idstr itself.
566 len
= strlen(block_name
);
568 memcpy(buf
+ 1, block_name
, len
);
570 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
571 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
576 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
579 * Next, we dump the received bitmap to the stream.
581 * TODO: currently we are safe since we are the only one that is
582 * using the to_src_file handle (fault thread is still paused),
583 * and it's ok even not taking the mutex. However the best way is
584 * to take the lock before sending the message header, and release
585 * the lock after sending the bitmap.
587 qemu_mutex_lock(&mis
->rp_mutex
);
588 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
589 qemu_mutex_unlock(&mis
->rp_mutex
);
591 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
594 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
598 buf
= cpu_to_be32(value
);
599 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
602 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
604 MigrationCapabilityStatusList
*head
= NULL
;
605 MigrationCapabilityStatusList
*caps
;
606 MigrationState
*s
= migrate_get_current();
609 caps
= NULL
; /* silence compiler warning */
610 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
611 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
612 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
617 head
= g_malloc0(sizeof(*caps
));
620 caps
->next
= g_malloc0(sizeof(*caps
));
624 g_malloc(sizeof(*caps
->value
));
625 caps
->value
->capability
= i
;
626 caps
->value
->state
= s
->enabled_capabilities
[i
];
632 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
634 MigrationParameters
*params
;
635 MigrationState
*s
= migrate_get_current();
637 /* TODO use QAPI_CLONE() instead of duplicating it inline */
638 params
= g_malloc0(sizeof(*params
));
639 params
->has_compress_level
= true;
640 params
->compress_level
= s
->parameters
.compress_level
;
641 params
->has_compress_threads
= true;
642 params
->compress_threads
= s
->parameters
.compress_threads
;
643 params
->has_decompress_threads
= true;
644 params
->decompress_threads
= s
->parameters
.decompress_threads
;
645 params
->has_cpu_throttle_initial
= true;
646 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
647 params
->has_cpu_throttle_increment
= true;
648 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
649 params
->has_tls_creds
= true;
650 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
651 params
->has_tls_hostname
= true;
652 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
653 params
->has_max_bandwidth
= true;
654 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
655 params
->has_downtime_limit
= true;
656 params
->downtime_limit
= s
->parameters
.downtime_limit
;
657 params
->has_x_checkpoint_delay
= true;
658 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
659 params
->has_block_incremental
= true;
660 params
->block_incremental
= s
->parameters
.block_incremental
;
661 params
->has_x_multifd_channels
= true;
662 params
->x_multifd_channels
= s
->parameters
.x_multifd_channels
;
663 params
->has_x_multifd_page_count
= true;
664 params
->x_multifd_page_count
= s
->parameters
.x_multifd_page_count
;
665 params
->has_xbzrle_cache_size
= true;
666 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
667 params
->has_max_postcopy_bandwidth
= true;
668 params
->max_postcopy_bandwidth
= s
->parameters
.max_postcopy_bandwidth
;
674 * Return true if we're already in the middle of a migration
675 * (i.e. any of the active or setup states)
677 static bool migration_is_setup_or_active(int state
)
680 case MIGRATION_STATUS_ACTIVE
:
681 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
682 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
683 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
684 case MIGRATION_STATUS_SETUP
:
685 case MIGRATION_STATUS_PRE_SWITCHOVER
:
686 case MIGRATION_STATUS_DEVICE
:
695 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
697 info
->has_ram
= true;
698 info
->ram
= g_malloc0(sizeof(*info
->ram
));
699 info
->ram
->transferred
= ram_counters
.transferred
;
700 info
->ram
->total
= ram_bytes_total();
701 info
->ram
->duplicate
= ram_counters
.duplicate
;
702 /* legacy value. It is not used anymore */
703 info
->ram
->skipped
= 0;
704 info
->ram
->normal
= ram_counters
.normal
;
705 info
->ram
->normal_bytes
= ram_counters
.normal
*
706 qemu_target_page_size();
707 info
->ram
->mbps
= s
->mbps
;
708 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
709 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
710 info
->ram
->page_size
= qemu_target_page_size();
711 info
->ram
->multifd_bytes
= ram_counters
.multifd_bytes
;
713 if (migrate_use_xbzrle()) {
714 info
->has_xbzrle_cache
= true;
715 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
716 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
717 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
718 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
719 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
720 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
721 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
724 if (cpu_throttle_active()) {
725 info
->has_cpu_throttle_percentage
= true;
726 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
729 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
730 info
->ram
->remaining
= ram_bytes_remaining();
731 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
735 static void populate_disk_info(MigrationInfo
*info
)
737 if (blk_mig_active()) {
738 info
->has_disk
= true;
739 info
->disk
= g_malloc0(sizeof(*info
->disk
));
740 info
->disk
->transferred
= blk_mig_bytes_transferred();
741 info
->disk
->remaining
= blk_mig_bytes_remaining();
742 info
->disk
->total
= blk_mig_bytes_total();
746 static void fill_source_migration_info(MigrationInfo
*info
)
748 MigrationState
*s
= migrate_get_current();
751 case MIGRATION_STATUS_NONE
:
752 /* no migration has happened ever */
753 /* do not overwrite destination migration status */
756 case MIGRATION_STATUS_SETUP
:
757 info
->has_status
= true;
758 info
->has_total_time
= false;
760 case MIGRATION_STATUS_ACTIVE
:
761 case MIGRATION_STATUS_CANCELLING
:
762 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
763 case MIGRATION_STATUS_PRE_SWITCHOVER
:
764 case MIGRATION_STATUS_DEVICE
:
765 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
766 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
767 /* TODO add some postcopy stats */
768 info
->has_status
= true;
769 info
->has_total_time
= true;
770 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
772 info
->has_expected_downtime
= true;
773 info
->expected_downtime
= s
->expected_downtime
;
774 info
->has_setup_time
= true;
775 info
->setup_time
= s
->setup_time
;
777 populate_ram_info(info
, s
);
778 populate_disk_info(info
);
780 case MIGRATION_STATUS_COLO
:
781 info
->has_status
= true;
782 /* TODO: display COLO specific information (checkpoint info etc.) */
784 case MIGRATION_STATUS_COMPLETED
:
785 info
->has_status
= true;
786 info
->has_total_time
= true;
787 info
->total_time
= s
->total_time
;
788 info
->has_downtime
= true;
789 info
->downtime
= s
->downtime
;
790 info
->has_setup_time
= true;
791 info
->setup_time
= s
->setup_time
;
793 populate_ram_info(info
, s
);
795 case MIGRATION_STATUS_FAILED
:
796 info
->has_status
= true;
798 info
->has_error_desc
= true;
799 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
802 case MIGRATION_STATUS_CANCELLED
:
803 info
->has_status
= true;
806 info
->status
= s
->state
;
810 * @migration_caps_check - check capability validity
812 * @cap_list: old capability list, array of bool
813 * @params: new capabilities to be applied soon
814 * @errp: set *errp if the check failed, with reason
816 * Returns true if check passed, otherwise false.
818 static bool migrate_caps_check(bool *cap_list
,
819 MigrationCapabilityStatusList
*params
,
822 MigrationCapabilityStatusList
*cap
;
823 bool old_postcopy_cap
;
824 MigrationIncomingState
*mis
= migration_incoming_get_current();
826 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
828 for (cap
= params
; cap
; cap
= cap
->next
) {
829 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
832 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
833 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
834 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
836 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
841 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
842 if (cap_list
[MIGRATION_CAPABILITY_COMPRESS
]) {
843 /* The decompression threads asynchronously write into RAM
844 * rather than use the atomic copies needed to avoid
845 * userfaulting. It should be possible to fix the decompression
846 * threads for compatibility in future.
848 error_setg(errp
, "Postcopy is not currently compatible "
853 /* This check is reasonably expensive, so only when it's being
854 * set the first time, also it's only the destination that needs
857 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
858 !postcopy_ram_supported_by_host(mis
)) {
859 /* postcopy_ram_supported_by_host will have emitted a more
862 error_setg(errp
, "Postcopy is not supported");
870 static void fill_destination_migration_info(MigrationInfo
*info
)
872 MigrationIncomingState
*mis
= migration_incoming_get_current();
874 switch (mis
->state
) {
875 case MIGRATION_STATUS_NONE
:
878 case MIGRATION_STATUS_SETUP
:
879 case MIGRATION_STATUS_CANCELLING
:
880 case MIGRATION_STATUS_CANCELLED
:
881 case MIGRATION_STATUS_ACTIVE
:
882 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
883 case MIGRATION_STATUS_FAILED
:
884 case MIGRATION_STATUS_COLO
:
885 info
->has_status
= true;
887 case MIGRATION_STATUS_COMPLETED
:
888 info
->has_status
= true;
889 fill_destination_postcopy_migration_info(info
);
892 info
->status
= mis
->state
;
895 MigrationInfo
*qmp_query_migrate(Error
**errp
)
897 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
899 fill_destination_migration_info(info
);
900 fill_source_migration_info(info
);
905 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
908 MigrationState
*s
= migrate_get_current();
909 MigrationCapabilityStatusList
*cap
;
910 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
912 if (migration_is_setup_or_active(s
->state
)) {
913 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
917 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
918 if (!migrate_caps_check(cap_list
, params
, errp
)) {
922 for (cap
= params
; cap
; cap
= cap
->next
) {
923 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
928 * Check whether the parameters are valid. Error will be put into errp
929 * (if provided). Return true if valid, otherwise false.
931 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
933 if (params
->has_compress_level
&&
934 (params
->compress_level
> 9)) {
935 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
936 "is invalid, it should be in the range of 0 to 9");
940 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
941 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
943 "is invalid, it should be in the range of 1 to 255");
947 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
948 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
949 "decompress_threads",
950 "is invalid, it should be in the range of 1 to 255");
954 if (params
->has_cpu_throttle_initial
&&
955 (params
->cpu_throttle_initial
< 1 ||
956 params
->cpu_throttle_initial
> 99)) {
957 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
958 "cpu_throttle_initial",
959 "an integer in the range of 1 to 99");
963 if (params
->has_cpu_throttle_increment
&&
964 (params
->cpu_throttle_increment
< 1 ||
965 params
->cpu_throttle_increment
> 99)) {
966 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
967 "cpu_throttle_increment",
968 "an integer in the range of 1 to 99");
972 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
973 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
974 " range of 0 to %zu bytes/second", SIZE_MAX
);
978 if (params
->has_downtime_limit
&&
979 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
980 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
981 "the range of 0 to %d milliseconds",
982 MAX_MIGRATE_DOWNTIME
);
986 /* x_checkpoint_delay is now always positive */
988 if (params
->has_x_multifd_channels
&& (params
->x_multifd_channels
< 1)) {
989 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
991 "is invalid, it should be in the range of 1 to 255");
994 if (params
->has_x_multifd_page_count
&&
995 (params
->x_multifd_page_count
< 1 ||
996 params
->x_multifd_page_count
> 10000)) {
997 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
998 "multifd_page_count",
999 "is invalid, it should be in the range of 1 to 10000");
1003 if (params
->has_xbzrle_cache_size
&&
1004 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
1005 !is_power_of_2(params
->xbzrle_cache_size
))) {
1006 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1007 "xbzrle_cache_size",
1008 "is invalid, it should be bigger than target page size"
1009 " and a power of two");
1016 static void migrate_params_test_apply(MigrateSetParameters
*params
,
1017 MigrationParameters
*dest
)
1019 *dest
= migrate_get_current()->parameters
;
1021 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1023 if (params
->has_compress_level
) {
1024 dest
->compress_level
= params
->compress_level
;
1027 if (params
->has_compress_threads
) {
1028 dest
->compress_threads
= params
->compress_threads
;
1031 if (params
->has_decompress_threads
) {
1032 dest
->decompress_threads
= params
->decompress_threads
;
1035 if (params
->has_cpu_throttle_initial
) {
1036 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
1039 if (params
->has_cpu_throttle_increment
) {
1040 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
1043 if (params
->has_tls_creds
) {
1044 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1045 dest
->tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1048 if (params
->has_tls_hostname
) {
1049 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1050 dest
->tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1053 if (params
->has_max_bandwidth
) {
1054 dest
->max_bandwidth
= params
->max_bandwidth
;
1057 if (params
->has_downtime_limit
) {
1058 dest
->downtime_limit
= params
->downtime_limit
;
1061 if (params
->has_x_checkpoint_delay
) {
1062 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
1065 if (params
->has_block_incremental
) {
1066 dest
->block_incremental
= params
->block_incremental
;
1068 if (params
->has_x_multifd_channels
) {
1069 dest
->x_multifd_channels
= params
->x_multifd_channels
;
1071 if (params
->has_x_multifd_page_count
) {
1072 dest
->x_multifd_page_count
= params
->x_multifd_page_count
;
1074 if (params
->has_xbzrle_cache_size
) {
1075 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
1077 if (params
->has_max_postcopy_bandwidth
) {
1078 dest
->max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1082 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
1084 MigrationState
*s
= migrate_get_current();
1086 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1088 if (params
->has_compress_level
) {
1089 s
->parameters
.compress_level
= params
->compress_level
;
1092 if (params
->has_compress_threads
) {
1093 s
->parameters
.compress_threads
= params
->compress_threads
;
1096 if (params
->has_decompress_threads
) {
1097 s
->parameters
.decompress_threads
= params
->decompress_threads
;
1100 if (params
->has_cpu_throttle_initial
) {
1101 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
1104 if (params
->has_cpu_throttle_increment
) {
1105 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
1108 if (params
->has_tls_creds
) {
1109 g_free(s
->parameters
.tls_creds
);
1110 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1111 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1114 if (params
->has_tls_hostname
) {
1115 g_free(s
->parameters
.tls_hostname
);
1116 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1117 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1120 if (params
->has_max_bandwidth
) {
1121 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1122 if (s
->to_dst_file
) {
1123 qemu_file_set_rate_limit(s
->to_dst_file
,
1124 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1128 if (params
->has_downtime_limit
) {
1129 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1132 if (params
->has_x_checkpoint_delay
) {
1133 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1134 if (migration_in_colo_state()) {
1135 colo_checkpoint_notify(s
);
1139 if (params
->has_block_incremental
) {
1140 s
->parameters
.block_incremental
= params
->block_incremental
;
1142 if (params
->has_x_multifd_channels
) {
1143 s
->parameters
.x_multifd_channels
= params
->x_multifd_channels
;
1145 if (params
->has_x_multifd_page_count
) {
1146 s
->parameters
.x_multifd_page_count
= params
->x_multifd_page_count
;
1148 if (params
->has_xbzrle_cache_size
) {
1149 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1150 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1152 if (params
->has_max_postcopy_bandwidth
) {
1153 s
->parameters
.max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1157 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1159 MigrationParameters tmp
;
1161 /* TODO Rewrite "" to null instead */
1162 if (params
->has_tls_creds
1163 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1164 qobject_unref(params
->tls_creds
->u
.n
);
1165 params
->tls_creds
->type
= QTYPE_QSTRING
;
1166 params
->tls_creds
->u
.s
= strdup("");
1168 /* TODO Rewrite "" to null instead */
1169 if (params
->has_tls_hostname
1170 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1171 qobject_unref(params
->tls_hostname
->u
.n
);
1172 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1173 params
->tls_hostname
->u
.s
= strdup("");
1176 migrate_params_test_apply(params
, &tmp
);
1178 if (!migrate_params_check(&tmp
, errp
)) {
1179 /* Invalid parameter */
1183 migrate_params_apply(params
, errp
);
1187 void qmp_migrate_start_postcopy(Error
**errp
)
1189 MigrationState
*s
= migrate_get_current();
1191 if (!migrate_postcopy()) {
1192 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1193 " the start of migration");
1197 if (s
->state
== MIGRATION_STATUS_NONE
) {
1198 error_setg(errp
, "Postcopy must be started after migration has been"
1203 * we don't error if migration has finished since that would be racy
1204 * with issuing this command.
1206 atomic_set(&s
->start_postcopy
, true);
1209 /* shared migration helpers */
1211 void migrate_set_state(int *state
, int old_state
, int new_state
)
1213 assert(new_state
< MIGRATION_STATUS__MAX
);
1214 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1215 trace_migrate_set_state(MigrationStatus_str(new_state
));
1216 migrate_generate_event(new_state
);
1220 static MigrationCapabilityStatusList
*migrate_cap_add(
1221 MigrationCapabilityStatusList
*list
,
1222 MigrationCapability index
,
1225 MigrationCapabilityStatusList
*cap
;
1227 cap
= g_new0(MigrationCapabilityStatusList
, 1);
1228 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
1229 cap
->value
->capability
= index
;
1230 cap
->value
->state
= state
;
1236 void migrate_set_block_enabled(bool value
, Error
**errp
)
1238 MigrationCapabilityStatusList
*cap
;
1240 cap
= migrate_cap_add(NULL
, MIGRATION_CAPABILITY_BLOCK
, value
);
1241 qmp_migrate_set_capabilities(cap
, errp
);
1242 qapi_free_MigrationCapabilityStatusList(cap
);
1245 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1247 s
->parameters
.block_incremental
= value
;
1250 static void block_cleanup_parameters(MigrationState
*s
)
1252 if (s
->must_remove_block_options
) {
1253 /* setting to false can never fail */
1254 migrate_set_block_enabled(false, &error_abort
);
1255 migrate_set_block_incremental(s
, false);
1256 s
->must_remove_block_options
= false;
1260 static void migrate_fd_cleanup(void *opaque
)
1262 MigrationState
*s
= opaque
;
1264 qemu_bh_delete(s
->cleanup_bh
);
1265 s
->cleanup_bh
= NULL
;
1267 qemu_savevm_state_cleanup();
1269 if (s
->to_dst_file
) {
1270 Error
*local_err
= NULL
;
1273 trace_migrate_fd_cleanup();
1274 qemu_mutex_unlock_iothread();
1275 if (s
->migration_thread_running
) {
1276 qemu_thread_join(&s
->thread
);
1277 s
->migration_thread_running
= false;
1279 qemu_mutex_lock_iothread();
1281 if (multifd_save_cleanup(&local_err
) != 0) {
1282 error_report_err(local_err
);
1284 qemu_mutex_lock(&s
->qemu_file_lock
);
1285 tmp
= s
->to_dst_file
;
1286 s
->to_dst_file
= NULL
;
1287 qemu_mutex_unlock(&s
->qemu_file_lock
);
1289 * Close the file handle without the lock to make sure the
1290 * critical section won't block for long.
1295 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
1296 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
1298 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1299 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1300 MIGRATION_STATUS_CANCELLED
);
1304 /* It is used on info migrate. We can't free it */
1305 error_report_err(error_copy(s
->error
));
1307 notifier_list_notify(&migration_state_notifiers
, s
);
1308 block_cleanup_parameters(s
);
1311 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1313 qemu_mutex_lock(&s
->error_mutex
);
1315 s
->error
= error_copy(error
);
1317 qemu_mutex_unlock(&s
->error_mutex
);
1320 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1322 trace_migrate_fd_error(error_get_pretty(error
));
1323 assert(s
->to_dst_file
== NULL
);
1324 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1325 MIGRATION_STATUS_FAILED
);
1326 migrate_set_error(s
, error
);
1329 static void migrate_fd_cancel(MigrationState
*s
)
1332 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1333 trace_migrate_fd_cancel();
1335 if (s
->rp_state
.from_dst_file
) {
1336 /* shutdown the rp socket, so causing the rp thread to shutdown */
1337 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1341 old_state
= s
->state
;
1342 if (!migration_is_setup_or_active(old_state
)) {
1345 /* If the migration is paused, kick it out of the pause */
1346 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1347 qemu_sem_post(&s
->pause_sem
);
1349 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1350 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1353 * If we're unlucky the migration code might be stuck somewhere in a
1354 * send/write while the network has failed and is waiting to timeout;
1355 * if we've got shutdown(2) available then we can force it to quit.
1356 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1357 * called in a bh, so there is no race against this cancel.
1359 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1360 qemu_file_shutdown(f
);
1362 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1363 Error
*local_err
= NULL
;
1365 bdrv_invalidate_cache_all(&local_err
);
1367 error_report_err(local_err
);
1369 s
->block_inactive
= false;
1374 void add_migration_state_change_notifier(Notifier
*notify
)
1376 notifier_list_add(&migration_state_notifiers
, notify
);
1379 void remove_migration_state_change_notifier(Notifier
*notify
)
1381 notifier_remove(notify
);
1384 bool migration_in_setup(MigrationState
*s
)
1386 return s
->state
== MIGRATION_STATUS_SETUP
;
1389 bool migration_has_finished(MigrationState
*s
)
1391 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1394 bool migration_has_failed(MigrationState
*s
)
1396 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1397 s
->state
== MIGRATION_STATUS_FAILED
);
1400 bool migration_in_postcopy(void)
1402 MigrationState
*s
= migrate_get_current();
1404 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1407 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1409 return migration_in_postcopy() && s
->postcopy_after_devices
;
1412 bool migration_is_idle(void)
1414 MigrationState
*s
= migrate_get_current();
1417 case MIGRATION_STATUS_NONE
:
1418 case MIGRATION_STATUS_CANCELLED
:
1419 case MIGRATION_STATUS_COMPLETED
:
1420 case MIGRATION_STATUS_FAILED
:
1422 case MIGRATION_STATUS_SETUP
:
1423 case MIGRATION_STATUS_CANCELLING
:
1424 case MIGRATION_STATUS_ACTIVE
:
1425 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1426 case MIGRATION_STATUS_COLO
:
1427 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1428 case MIGRATION_STATUS_DEVICE
:
1430 case MIGRATION_STATUS__MAX
:
1431 g_assert_not_reached();
1437 void migrate_init(MigrationState
*s
)
1440 * Reinitialise all migration state, except
1441 * parameters/capabilities that the user set, and
1447 s
->to_dst_file
= NULL
;
1448 s
->state
= MIGRATION_STATUS_NONE
;
1449 s
->rp_state
.from_dst_file
= NULL
;
1450 s
->rp_state
.error
= false;
1453 s
->expected_downtime
= 0;
1455 s
->start_postcopy
= false;
1456 s
->postcopy_after_devices
= false;
1457 s
->migration_thread_running
= false;
1458 error_free(s
->error
);
1461 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1463 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1465 s
->vm_was_running
= false;
1466 s
->iteration_initial_bytes
= 0;
1467 s
->threshold_size
= 0;
1470 static GSList
*migration_blockers
;
1472 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1474 if (migrate_get_current()->only_migratable
) {
1475 error_propagate(errp
, error_copy(reason
));
1476 error_prepend(errp
, "disallowing migration blocker "
1477 "(--only_migratable) for: ");
1481 if (migration_is_idle()) {
1482 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1486 error_propagate(errp
, error_copy(reason
));
1487 error_prepend(errp
, "disallowing migration blocker (migration in "
1492 void migrate_del_blocker(Error
*reason
)
1494 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1497 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1499 Error
*local_err
= NULL
;
1500 static bool once
= true;
1502 if (!deferred_incoming
) {
1503 error_setg(errp
, "For use with '-incoming defer'");
1507 error_setg(errp
, "The incoming migration has already been started");
1510 qemu_start_incoming_migration(uri
, &local_err
);
1513 error_propagate(errp
, local_err
);
1520 void qmp_migrate_recover(const char *uri
, Error
**errp
)
1522 MigrationIncomingState
*mis
= migration_incoming_get_current();
1524 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1525 error_setg(errp
, "Migrate recover can only be run "
1526 "when postcopy is paused.");
1530 if (atomic_cmpxchg(&mis
->postcopy_recover_triggered
,
1531 false, true) == true) {
1532 error_setg(errp
, "Migrate recovery is triggered already");
1537 * Note that this call will never start a real migration; it will
1538 * only re-setup the migration stream and poke existing migration
1539 * to continue using that newly established channel.
1541 qemu_start_incoming_migration(uri
, errp
);
1544 void qmp_migrate_pause(Error
**errp
)
1546 MigrationState
*ms
= migrate_get_current();
1547 MigrationIncomingState
*mis
= migration_incoming_get_current();
1550 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1551 /* Source side, during postcopy */
1552 qemu_mutex_lock(&ms
->qemu_file_lock
);
1553 ret
= qemu_file_shutdown(ms
->to_dst_file
);
1554 qemu_mutex_unlock(&ms
->qemu_file_lock
);
1556 error_setg(errp
, "Failed to pause source migration");
1561 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1562 ret
= qemu_file_shutdown(mis
->from_src_file
);
1564 error_setg(errp
, "Failed to pause destination migration");
1569 error_setg(errp
, "migrate-pause is currently only supported "
1570 "during postcopy-active state");
1573 bool migration_is_blocked(Error
**errp
)
1575 if (qemu_savevm_state_blocked(errp
)) {
1579 if (migration_blockers
) {
1580 error_propagate(errp
, error_copy(migration_blockers
->data
));
1587 /* Returns true if continue to migrate, or false if error detected */
1588 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
1589 bool resume
, Error
**errp
)
1591 Error
*local_err
= NULL
;
1594 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1595 error_setg(errp
, "Cannot resume if there is no "
1596 "paused migration");
1599 /* This is a resume, skip init status */
1603 if (migration_is_setup_or_active(s
->state
) ||
1604 s
->state
== MIGRATION_STATUS_CANCELLING
||
1605 s
->state
== MIGRATION_STATUS_COLO
) {
1606 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1610 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1611 error_setg(errp
, "Guest is waiting for an incoming migration");
1615 if (migration_is_blocked(errp
)) {
1619 if (blk
|| blk_inc
) {
1620 if (migrate_use_block() || migrate_use_block_incremental()) {
1621 error_setg(errp
, "Command options are incompatible with "
1622 "current migration capabilities");
1625 migrate_set_block_enabled(true, &local_err
);
1627 error_propagate(errp
, local_err
);
1630 s
->must_remove_block_options
= true;
1634 migrate_set_block_incremental(s
, true);
1642 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1643 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1644 bool has_resume
, bool resume
, Error
**errp
)
1646 Error
*local_err
= NULL
;
1647 MigrationState
*s
= migrate_get_current();
1650 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
1651 has_resume
&& resume
, errp
)) {
1652 /* Error detected, put into errp */
1656 if (strstart(uri
, "tcp:", &p
)) {
1657 tcp_start_outgoing_migration(s
, p
, &local_err
);
1659 } else if (strstart(uri
, "rdma:", &p
)) {
1660 rdma_start_outgoing_migration(s
, p
, &local_err
);
1662 } else if (strstart(uri
, "exec:", &p
)) {
1663 exec_start_outgoing_migration(s
, p
, &local_err
);
1664 } else if (strstart(uri
, "unix:", &p
)) {
1665 unix_start_outgoing_migration(s
, p
, &local_err
);
1666 } else if (strstart(uri
, "fd:", &p
)) {
1667 fd_start_outgoing_migration(s
, p
, &local_err
);
1669 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1670 "a valid migration protocol");
1671 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1672 MIGRATION_STATUS_FAILED
);
1673 block_cleanup_parameters(s
);
1678 migrate_fd_error(s
, local_err
);
1679 error_propagate(errp
, local_err
);
1684 void qmp_migrate_cancel(Error
**errp
)
1686 migrate_fd_cancel(migrate_get_current());
1689 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
1691 MigrationState
*s
= migrate_get_current();
1692 if (s
->state
!= state
) {
1693 error_setg(errp
, "Migration not in expected state: %s",
1694 MigrationStatus_str(s
->state
));
1697 qemu_sem_post(&s
->pause_sem
);
1700 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1702 MigrateSetParameters p
= {
1703 .has_xbzrle_cache_size
= true,
1704 .xbzrle_cache_size
= value
,
1707 qmp_migrate_set_parameters(&p
, errp
);
1710 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1712 return migrate_xbzrle_cache_size();
1715 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1717 MigrateSetParameters p
= {
1718 .has_max_bandwidth
= true,
1719 .max_bandwidth
= value
,
1722 qmp_migrate_set_parameters(&p
, errp
);
1725 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1727 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
1728 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1729 "the range of 0 to %d seconds",
1730 MAX_MIGRATE_DOWNTIME_SECONDS
);
1734 value
*= 1000; /* Convert to milliseconds */
1735 value
= MAX(0, MIN(INT64_MAX
, value
));
1737 MigrateSetParameters p
= {
1738 .has_downtime_limit
= true,
1739 .downtime_limit
= value
,
1742 qmp_migrate_set_parameters(&p
, errp
);
1745 bool migrate_release_ram(void)
1749 s
= migrate_get_current();
1751 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
1754 bool migrate_postcopy_ram(void)
1758 s
= migrate_get_current();
1760 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1763 bool migrate_postcopy(void)
1765 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
1768 bool migrate_auto_converge(void)
1772 s
= migrate_get_current();
1774 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1777 bool migrate_zero_blocks(void)
1781 s
= migrate_get_current();
1783 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
1786 bool migrate_postcopy_blocktime(void)
1790 s
= migrate_get_current();
1792 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
1795 bool migrate_use_compression(void)
1799 s
= migrate_get_current();
1801 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
1804 int migrate_compress_level(void)
1808 s
= migrate_get_current();
1810 return s
->parameters
.compress_level
;
1813 int migrate_compress_threads(void)
1817 s
= migrate_get_current();
1819 return s
->parameters
.compress_threads
;
1822 int migrate_decompress_threads(void)
1826 s
= migrate_get_current();
1828 return s
->parameters
.decompress_threads
;
1831 bool migrate_dirty_bitmaps(void)
1835 s
= migrate_get_current();
1837 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
1840 bool migrate_use_events(void)
1844 s
= migrate_get_current();
1846 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
1849 bool migrate_use_multifd(void)
1853 s
= migrate_get_current();
1855 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_MULTIFD
];
1858 bool migrate_pause_before_switchover(void)
1862 s
= migrate_get_current();
1864 return s
->enabled_capabilities
[
1865 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
1868 int migrate_multifd_channels(void)
1872 s
= migrate_get_current();
1874 return s
->parameters
.x_multifd_channels
;
1877 int migrate_multifd_page_count(void)
1881 s
= migrate_get_current();
1883 return s
->parameters
.x_multifd_page_count
;
1886 int migrate_use_xbzrle(void)
1890 s
= migrate_get_current();
1892 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
1895 int64_t migrate_xbzrle_cache_size(void)
1899 s
= migrate_get_current();
1901 return s
->parameters
.xbzrle_cache_size
;
1904 static int64_t migrate_max_postcopy_bandwidth(void)
1908 s
= migrate_get_current();
1910 return s
->parameters
.max_postcopy_bandwidth
;
1914 bool migrate_use_block(void)
1918 s
= migrate_get_current();
1920 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
1923 bool migrate_use_return_path(void)
1927 s
= migrate_get_current();
1929 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
1932 bool migrate_use_block_incremental(void)
1936 s
= migrate_get_current();
1938 return s
->parameters
.block_incremental
;
1941 /* migration thread support */
1943 * Something bad happened to the RP stream, mark an error
1944 * The caller shall print or trace something to indicate why
1946 static void mark_source_rp_bad(MigrationState
*s
)
1948 s
->rp_state
.error
= true;
1951 static struct rp_cmd_args
{
1952 ssize_t len
; /* -1 = variable */
1955 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1956 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1957 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1958 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1959 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1960 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
1961 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
1962 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1966 * Process a request for pages received on the return path,
1967 * We're allowed to send more than requested (e.g. to round to our page size)
1968 * and we don't need to send pages that have already been sent.
1970 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1971 ram_addr_t start
, size_t len
)
1973 long our_host_ps
= getpagesize();
1975 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1978 * Since we currently insist on matching page sizes, just sanity check
1979 * we're being asked for whole host pages.
1981 if (start
& (our_host_ps
-1) ||
1982 (len
& (our_host_ps
-1))) {
1983 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1984 " len: %zd", __func__
, start
, len
);
1985 mark_source_rp_bad(ms
);
1989 if (ram_save_queue_pages(rbname
, start
, len
)) {
1990 mark_source_rp_bad(ms
);
1994 /* Return true to retry, false to quit */
1995 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
1997 trace_postcopy_pause_return_path();
1999 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
2001 trace_postcopy_pause_return_path_continued();
2006 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
2008 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
2011 error_report("%s: invalid block name '%s'", __func__
, block_name
);
2015 /* Fetch the received bitmap and refresh the dirty bitmap */
2016 return ram_dirty_bitmap_reload(s
, block
);
2019 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
2021 trace_source_return_path_thread_resume_ack(value
);
2023 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
2024 error_report("%s: illegal resume_ack value %"PRIu32
,
2029 /* Now both sides are active. */
2030 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2031 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2033 /* Notify send thread that time to continue send pages */
2034 qemu_sem_post(&s
->rp_state
.rp_sem
);
2040 * Handles messages sent on the return path towards the source VM
2043 static void *source_return_path_thread(void *opaque
)
2045 MigrationState
*ms
= opaque
;
2046 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
2047 uint16_t header_len
, header_type
;
2049 uint32_t tmp32
, sibling_error
;
2050 ram_addr_t start
= 0; /* =0 to silence warning */
2051 size_t len
= 0, expected_len
;
2054 trace_source_return_path_thread_entry();
2057 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
2058 migration_is_setup_or_active(ms
->state
)) {
2059 trace_source_return_path_thread_loop_top();
2060 header_type
= qemu_get_be16(rp
);
2061 header_len
= qemu_get_be16(rp
);
2063 if (qemu_file_get_error(rp
)) {
2064 mark_source_rp_bad(ms
);
2068 if (header_type
>= MIG_RP_MSG_MAX
||
2069 header_type
== MIG_RP_MSG_INVALID
) {
2070 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2071 header_type
, header_len
);
2072 mark_source_rp_bad(ms
);
2076 if ((rp_cmd_args
[header_type
].len
!= -1 &&
2077 header_len
!= rp_cmd_args
[header_type
].len
) ||
2078 header_len
> sizeof(buf
)) {
2079 error_report("RP: Received '%s' message (0x%04x) with"
2080 "incorrect length %d expecting %zu",
2081 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
2082 (size_t)rp_cmd_args
[header_type
].len
);
2083 mark_source_rp_bad(ms
);
2087 /* We know we've got a valid header by this point */
2088 res
= qemu_get_buffer(rp
, buf
, header_len
);
2089 if (res
!= header_len
) {
2090 error_report("RP: Failed reading data for message 0x%04x"
2091 " read %d expected %d",
2092 header_type
, res
, header_len
);
2093 mark_source_rp_bad(ms
);
2097 /* OK, we have the message and the data */
2098 switch (header_type
) {
2099 case MIG_RP_MSG_SHUT
:
2100 sibling_error
= ldl_be_p(buf
);
2101 trace_source_return_path_thread_shut(sibling_error
);
2102 if (sibling_error
) {
2103 error_report("RP: Sibling indicated error %d", sibling_error
);
2104 mark_source_rp_bad(ms
);
2107 * We'll let the main thread deal with closing the RP
2108 * we could do a shutdown(2) on it, but we're the only user
2109 * anyway, so there's nothing gained.
2113 case MIG_RP_MSG_PONG
:
2114 tmp32
= ldl_be_p(buf
);
2115 trace_source_return_path_thread_pong(tmp32
);
2118 case MIG_RP_MSG_REQ_PAGES
:
2119 start
= ldq_be_p(buf
);
2120 len
= ldl_be_p(buf
+ 8);
2121 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
2124 case MIG_RP_MSG_REQ_PAGES_ID
:
2125 expected_len
= 12 + 1; /* header + termination */
2127 if (header_len
>= expected_len
) {
2128 start
= ldq_be_p(buf
);
2129 len
= ldl_be_p(buf
+ 8);
2130 /* Now we expect an idstr */
2131 tmp32
= buf
[12]; /* Length of the following idstr */
2132 buf
[13 + tmp32
] = '\0';
2133 expected_len
+= tmp32
;
2135 if (header_len
!= expected_len
) {
2136 error_report("RP: Req_Page_id with length %d expecting %zd",
2137 header_len
, expected_len
);
2138 mark_source_rp_bad(ms
);
2141 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
2144 case MIG_RP_MSG_RECV_BITMAP
:
2145 if (header_len
< 1) {
2146 error_report("%s: missing block name", __func__
);
2147 mark_source_rp_bad(ms
);
2150 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2151 buf
[buf
[0] + 1] = '\0';
2152 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
2153 mark_source_rp_bad(ms
);
2158 case MIG_RP_MSG_RESUME_ACK
:
2159 tmp32
= ldl_be_p(buf
);
2160 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
2161 mark_source_rp_bad(ms
);
2172 res
= qemu_file_get_error(rp
);
2176 * Maybe there is something we can do: it looks like a
2177 * network down issue, and we pause for a recovery.
2179 if (postcopy_pause_return_path_thread(ms
)) {
2180 /* Reload rp, reset the rest */
2181 rp
= ms
->rp_state
.from_dst_file
;
2182 ms
->rp_state
.error
= false;
2187 trace_source_return_path_thread_bad_end();
2188 mark_source_rp_bad(ms
);
2191 trace_source_return_path_thread_end();
2192 ms
->rp_state
.from_dst_file
= NULL
;
2197 static int open_return_path_on_source(MigrationState
*ms
,
2201 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
2202 if (!ms
->rp_state
.from_dst_file
) {
2206 trace_open_return_path_on_source();
2208 if (!create_thread
) {
2213 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
2214 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
2216 trace_open_return_path_on_source_continue();
2221 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2222 static int await_return_path_close_on_source(MigrationState
*ms
)
2225 * If this is a normal exit then the destination will send a SHUT and the
2226 * rp_thread will exit, however if there's an error we need to cause
2229 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
2231 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2232 * waiting for the destination.
2234 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
2235 mark_source_rp_bad(ms
);
2237 trace_await_return_path_close_on_source_joining();
2238 qemu_thread_join(&ms
->rp_state
.rp_thread
);
2239 trace_await_return_path_close_on_source_close();
2240 return ms
->rp_state
.error
;
2244 * Switch from normal iteration to postcopy
2245 * Returns non-0 on error
2247 static int postcopy_start(MigrationState
*ms
)
2250 QIOChannelBuffer
*bioc
;
2252 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2253 int64_t bandwidth
= migrate_max_postcopy_bandwidth();
2254 bool restart_block
= false;
2255 int cur_state
= MIGRATION_STATUS_ACTIVE
;
2256 if (!migrate_pause_before_switchover()) {
2257 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
2258 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2261 trace_postcopy_start();
2262 qemu_mutex_lock_iothread();
2263 trace_postcopy_start_set_run();
2265 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
2266 global_state_store();
2267 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2272 ret
= migration_maybe_pause(ms
, &cur_state
,
2273 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2278 ret
= bdrv_inactivate_all();
2282 restart_block
= true;
2285 * Cause any non-postcopiable, but iterative devices to
2286 * send out their final data.
2288 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2291 * in Finish migrate and with the io-lock held everything should
2292 * be quiet, but we've potentially still got dirty pages and we
2293 * need to tell the destination to throw any pages it's already received
2296 if (migrate_postcopy_ram()) {
2297 if (ram_postcopy_send_discard_bitmap(ms
)) {
2298 error_report("postcopy send discard bitmap failed");
2304 * send rest of state - note things that are doing postcopy
2305 * will notice we're in POSTCOPY_ACTIVE and not actually
2306 * wrap their state up here
2308 /* 0 max-postcopy-bandwidth means unlimited */
2310 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
2312 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
2314 if (migrate_postcopy_ram()) {
2315 /* Ping just for debugging, helps line traces up */
2316 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
2320 * While loading the device state we may trigger page transfer
2321 * requests and the fd must be free to process those, and thus
2322 * the destination must read the whole device state off the fd before
2323 * it starts processing it. Unfortunately the ad-hoc migration format
2324 * doesn't allow the destination to know the size to read without fully
2325 * parsing it through each devices load-state code (especially the open
2326 * coded devices that use get/put).
2327 * So we wrap the device state up in a package with a length at the start;
2328 * to do this we use a qemu_buf to hold the whole of the device state.
2330 bioc
= qio_channel_buffer_new(4096);
2331 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
2332 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
2333 object_unref(OBJECT(bioc
));
2336 * Make sure the receiver can get incoming pages before we send the rest
2339 qemu_savevm_send_postcopy_listen(fb
);
2341 qemu_savevm_state_complete_precopy(fb
, false, false);
2342 if (migrate_postcopy_ram()) {
2343 qemu_savevm_send_ping(fb
, 3);
2346 qemu_savevm_send_postcopy_run(fb
);
2348 /* <><> end of stuff going into the package */
2350 /* Last point of recovery; as soon as we send the package the destination
2351 * can open devices and potentially start running.
2352 * Lets just check again we've not got any errors.
2354 ret
= qemu_file_get_error(ms
->to_dst_file
);
2356 error_report("postcopy_start: Migration stream errored (pre package)");
2360 restart_block
= false;
2362 /* Now send that blob */
2363 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2368 /* Send a notify to give a chance for anything that needs to happen
2369 * at the transition to postcopy and after the device state; in particular
2370 * spice needs to trigger a transition now
2372 ms
->postcopy_after_devices
= true;
2373 notifier_list_notify(&migration_state_notifiers
, ms
);
2375 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2377 qemu_mutex_unlock_iothread();
2379 if (migrate_postcopy_ram()) {
2381 * Although this ping is just for debug, it could potentially be
2382 * used for getting a better measurement of downtime at the source.
2384 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2387 if (migrate_release_ram()) {
2388 ram_postcopy_migrated_memory_release(ms
);
2391 ret
= qemu_file_get_error(ms
->to_dst_file
);
2393 error_report("postcopy_start: Migration stream errored");
2394 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2395 MIGRATION_STATUS_FAILED
);
2403 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2404 MIGRATION_STATUS_FAILED
);
2405 if (restart_block
) {
2406 /* A failure happened early enough that we know the destination hasn't
2407 * accessed block devices, so we're safe to recover.
2409 Error
*local_err
= NULL
;
2411 bdrv_invalidate_cache_all(&local_err
);
2413 error_report_err(local_err
);
2416 qemu_mutex_unlock_iothread();
2421 * migration_maybe_pause: Pause if required to by
2422 * migrate_pause_before_switchover called with the iothread locked
2423 * Returns: 0 on success
2425 static int migration_maybe_pause(MigrationState
*s
,
2426 int *current_active_state
,
2429 if (!migrate_pause_before_switchover()) {
2433 /* Since leaving this state is not atomic with posting the semaphore
2434 * it's possible that someone could have issued multiple migrate_continue
2435 * and the semaphore is incorrectly positive at this point;
2436 * the docs say it's undefined to reinit a semaphore that's already
2437 * init'd, so use timedwait to eat up any existing posts.
2439 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2440 /* This block intentionally left blank */
2443 qemu_mutex_unlock_iothread();
2444 migrate_set_state(&s
->state
, *current_active_state
,
2445 MIGRATION_STATUS_PRE_SWITCHOVER
);
2446 qemu_sem_wait(&s
->pause_sem
);
2447 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2449 *current_active_state
= new_state
;
2450 qemu_mutex_lock_iothread();
2452 return s
->state
== new_state
? 0 : -EINVAL
;
2456 * migration_completion: Used by migration_thread when there's not much left.
2457 * The caller 'breaks' the loop when this returns.
2459 * @s: Current migration state
2461 static void migration_completion(MigrationState
*s
)
2464 int current_active_state
= s
->state
;
2466 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2467 qemu_mutex_lock_iothread();
2468 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2469 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
2470 s
->vm_was_running
= runstate_is_running();
2471 ret
= global_state_store();
2474 bool inactivate
= !migrate_colo_enabled();
2475 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2477 ret
= migration_maybe_pause(s
, ¤t_active_state
,
2478 MIGRATION_STATUS_DEVICE
);
2481 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
2482 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
2485 if (inactivate
&& ret
>= 0) {
2486 s
->block_inactive
= true;
2489 qemu_mutex_unlock_iothread();
2494 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2495 trace_migration_completion_postcopy_end();
2497 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
2498 trace_migration_completion_postcopy_end_after_complete();
2502 * If rp was opened we must clean up the thread before
2503 * cleaning everything else up (since if there are no failures
2504 * it will wait for the destination to send it's status in
2507 if (s
->rp_state
.from_dst_file
) {
2509 trace_migration_return_path_end_before();
2510 rp_error
= await_return_path_close_on_source(s
);
2511 trace_migration_return_path_end_after(rp_error
);
2513 goto fail_invalidate
;
2517 if (qemu_file_get_error(s
->to_dst_file
)) {
2518 trace_migration_completion_file_err();
2519 goto fail_invalidate
;
2522 if (!migrate_colo_enabled()) {
2523 migrate_set_state(&s
->state
, current_active_state
,
2524 MIGRATION_STATUS_COMPLETED
);
2530 /* If not doing postcopy, vm_start() will be called: let's regain
2531 * control on images.
2533 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
2534 s
->state
== MIGRATION_STATUS_DEVICE
) {
2535 Error
*local_err
= NULL
;
2537 qemu_mutex_lock_iothread();
2538 bdrv_invalidate_cache_all(&local_err
);
2540 error_report_err(local_err
);
2542 s
->block_inactive
= false;
2544 qemu_mutex_unlock_iothread();
2548 migrate_set_state(&s
->state
, current_active_state
,
2549 MIGRATION_STATUS_FAILED
);
2552 bool migrate_colo_enabled(void)
2554 MigrationState
*s
= migrate_get_current();
2555 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
2558 typedef enum MigThrError
{
2559 /* No error detected */
2560 MIG_THR_ERR_NONE
= 0,
2561 /* Detected error, but resumed successfully */
2562 MIG_THR_ERR_RECOVERED
= 1,
2563 /* Detected fatal error, need to exit */
2564 MIG_THR_ERR_FATAL
= 2,
2567 static int postcopy_resume_handshake(MigrationState
*s
)
2569 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
2571 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2572 qemu_sem_wait(&s
->rp_state
.rp_sem
);
2575 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2582 /* Return zero if success, or <0 for error */
2583 static int postcopy_do_resume(MigrationState
*s
)
2588 * Call all the resume_prepare() hooks, so that modules can be
2589 * ready for the migration resume.
2591 ret
= qemu_savevm_state_resume_prepare(s
);
2593 error_report("%s: resume_prepare() failure detected: %d",
2599 * Last handshake with destination on the resume (destination will
2600 * switch to postcopy-active afterwards)
2602 ret
= postcopy_resume_handshake(s
);
2604 error_report("%s: handshake failed: %d", __func__
, ret
);
2612 * We don't return until we are in a safe state to continue current
2613 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2614 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2616 static MigThrError
postcopy_pause(MigrationState
*s
)
2618 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2623 migrate_set_state(&s
->state
, s
->state
,
2624 MIGRATION_STATUS_POSTCOPY_PAUSED
);
2626 /* Current channel is possibly broken. Release it. */
2627 assert(s
->to_dst_file
);
2628 qemu_mutex_lock(&s
->qemu_file_lock
);
2629 file
= s
->to_dst_file
;
2630 s
->to_dst_file
= NULL
;
2631 qemu_mutex_unlock(&s
->qemu_file_lock
);
2633 qemu_file_shutdown(file
);
2636 error_report("Detected IO failure for postcopy. "
2637 "Migration paused.");
2640 * We wait until things fixed up. Then someone will setup the
2641 * status back for us.
2643 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2644 qemu_sem_wait(&s
->postcopy_pause_sem
);
2647 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2648 /* Woken up by a recover procedure. Give it a shot */
2651 * Firstly, let's wake up the return path now, with a new
2652 * return path channel.
2654 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
2656 /* Do the resume logic */
2657 if (postcopy_do_resume(s
) == 0) {
2658 /* Let's continue! */
2659 trace_postcopy_pause_continued();
2660 return MIG_THR_ERR_RECOVERED
;
2663 * Something wrong happened during the recovery, let's
2664 * pause again. Pause is always better than throwing
2670 /* This is not right... Time to quit. */
2671 return MIG_THR_ERR_FATAL
;
2676 static MigThrError
migration_detect_error(MigrationState
*s
)
2680 /* Try to detect any file errors */
2681 ret
= qemu_file_get_error(s
->to_dst_file
);
2684 /* Everything is fine */
2685 return MIG_THR_ERR_NONE
;
2688 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
== -EIO
) {
2690 * For postcopy, we allow the network to be down for a
2691 * while. After that, it can be continued by a
2694 return postcopy_pause(s
);
2697 * For precopy (or postcopy with error outside IO), we fail
2700 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
2701 trace_migration_thread_file_err();
2703 /* Time to stop the migration, now. */
2704 return MIG_THR_ERR_FATAL
;
2708 /* How many bytes have we transferred since the beggining of the migration */
2709 static uint64_t migration_total_bytes(MigrationState
*s
)
2711 return qemu_ftell(s
->to_dst_file
) + ram_counters
.multifd_bytes
;
2714 static void migration_calculate_complete(MigrationState
*s
)
2716 uint64_t bytes
= migration_total_bytes(s
);
2717 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2718 int64_t transfer_time
;
2720 s
->total_time
= end_time
- s
->start_time
;
2723 * It's still not set, so we are precopy migration. For
2724 * postcopy, downtime is calculated during postcopy_start().
2726 s
->downtime
= end_time
- s
->downtime_start
;
2729 transfer_time
= s
->total_time
- s
->setup_time
;
2730 if (transfer_time
) {
2731 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
2735 static void migration_update_counters(MigrationState
*s
,
2736 int64_t current_time
)
2738 uint64_t transferred
, time_spent
;
2739 uint64_t current_bytes
; /* bytes transferred since the beginning */
2742 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
2746 current_bytes
= migration_total_bytes(s
);
2747 transferred
= current_bytes
- s
->iteration_initial_bytes
;
2748 time_spent
= current_time
- s
->iteration_start_time
;
2749 bandwidth
= (double)transferred
/ time_spent
;
2750 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
2752 s
->mbps
= (((double) transferred
* 8.0) /
2753 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2756 * if we haven't sent anything, we don't want to
2757 * recalculate. 10000 is a small enough number for our purposes
2759 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
2760 s
->expected_downtime
= ram_counters
.remaining
/ bandwidth
;
2763 qemu_file_reset_rate_limit(s
->to_dst_file
);
2765 s
->iteration_start_time
= current_time
;
2766 s
->iteration_initial_bytes
= current_bytes
;
2768 trace_migrate_transferred(transferred
, time_spent
,
2769 bandwidth
, s
->threshold_size
);
2772 /* Migration thread iteration status */
2774 MIG_ITERATE_RESUME
, /* Resume current iteration */
2775 MIG_ITERATE_SKIP
, /* Skip current iteration */
2776 MIG_ITERATE_BREAK
, /* Break the loop */
2780 * Return true if continue to the next iteration directly, false
2783 static MigIterateState
migration_iteration_run(MigrationState
*s
)
2785 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
2786 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
2788 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
2789 &pend_compat
, &pend_post
);
2790 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
2792 trace_migrate_pending(pending_size
, s
->threshold_size
,
2793 pend_pre
, pend_compat
, pend_post
);
2795 if (pending_size
&& pending_size
>= s
->threshold_size
) {
2796 /* Still a significant amount to transfer */
2797 if (migrate_postcopy() && !in_postcopy
&&
2798 pend_pre
<= s
->threshold_size
&&
2799 atomic_read(&s
->start_postcopy
)) {
2800 if (postcopy_start(s
)) {
2801 error_report("%s: postcopy failed to start", __func__
);
2803 return MIG_ITERATE_SKIP
;
2805 /* Just another iteration step */
2806 qemu_savevm_state_iterate(s
->to_dst_file
,
2807 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2809 trace_migration_thread_low_pending(pending_size
);
2810 migration_completion(s
);
2811 return MIG_ITERATE_BREAK
;
2814 return MIG_ITERATE_RESUME
;
2817 static void migration_iteration_finish(MigrationState
*s
)
2819 /* If we enabled cpu throttling for auto-converge, turn it off. */
2820 cpu_throttle_stop();
2822 qemu_mutex_lock_iothread();
2824 case MIGRATION_STATUS_COMPLETED
:
2825 migration_calculate_complete(s
);
2826 runstate_set(RUN_STATE_POSTMIGRATE
);
2829 case MIGRATION_STATUS_ACTIVE
:
2831 * We should really assert here, but since it's during
2832 * migration, let's try to reduce the usage of assertions.
2834 if (!migrate_colo_enabled()) {
2835 error_report("%s: critical error: calling COLO code without "
2836 "COLO enabled", __func__
);
2838 migrate_start_colo_process(s
);
2840 * Fixme: we will run VM in COLO no matter its old running state.
2841 * After exited COLO, we will keep running.
2843 s
->vm_was_running
= true;
2845 case MIGRATION_STATUS_FAILED
:
2846 case MIGRATION_STATUS_CANCELLED
:
2847 if (s
->vm_was_running
) {
2850 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2851 runstate_set(RUN_STATE_POSTMIGRATE
);
2857 /* Should not reach here, but if so, forgive the VM. */
2858 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
2861 qemu_bh_schedule(s
->cleanup_bh
);
2862 qemu_mutex_unlock_iothread();
2865 void migration_make_urgent_request(void)
2867 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
2870 void migration_consume_urgent_request(void)
2872 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
2876 * Master migration thread on the source VM.
2877 * It drives the migration and pumps the data down the outgoing channel.
2879 static void *migration_thread(void *opaque
)
2881 MigrationState
*s
= opaque
;
2882 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
2883 MigThrError thr_error
;
2884 bool urgent
= false;
2886 rcu_register_thread();
2888 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2890 qemu_savevm_state_header(s
->to_dst_file
);
2893 * If we opened the return path, we need to make sure dst has it
2896 if (s
->rp_state
.from_dst_file
) {
2897 /* Now tell the dest that it should open its end so it can reply */
2898 qemu_savevm_send_open_return_path(s
->to_dst_file
);
2900 /* And do a ping that will make stuff easier to debug */
2901 qemu_savevm_send_ping(s
->to_dst_file
, 1);
2904 if (migrate_postcopy()) {
2906 * Tell the destination that we *might* want to do postcopy later;
2907 * if the other end can't do postcopy it should fail now, nice and
2910 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
2913 qemu_savevm_state_setup(s
->to_dst_file
);
2915 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
2916 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2917 MIGRATION_STATUS_ACTIVE
);
2919 trace_migration_thread_setup_complete();
2921 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
2922 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2923 int64_t current_time
;
2925 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
2926 MigIterateState iter_state
= migration_iteration_run(s
);
2927 if (iter_state
== MIG_ITERATE_SKIP
) {
2929 } else if (iter_state
== MIG_ITERATE_BREAK
) {
2935 * Try to detect any kind of failures, and see whether we
2936 * should stop the migration now.
2938 thr_error
= migration_detect_error(s
);
2939 if (thr_error
== MIG_THR_ERR_FATAL
) {
2940 /* Stop migration */
2942 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
2944 * Just recovered from a e.g. network failure, reset all
2945 * the local variables. This is important to avoid
2946 * breaking transferred_bytes and bandwidth calculation
2948 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2949 s
->iteration_initial_bytes
= 0;
2952 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2954 migration_update_counters(s
, current_time
);
2957 if (qemu_file_rate_limit(s
->to_dst_file
)) {
2958 /* Wait for a delay to do rate limiting OR
2959 * something urgent to post the semaphore.
2961 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- current_time
;
2962 trace_migration_thread_ratelimit_pre(ms
);
2963 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
2964 /* We were worken by one or more urgent things but
2965 * the timedwait will have consumed one of them.
2966 * The service routine for the urgent wake will dec
2967 * the semaphore itself for each item it consumes,
2968 * so add this one we just eat back.
2970 qemu_sem_post(&s
->rate_limit_sem
);
2973 trace_migration_thread_ratelimit_post(urgent
);
2977 trace_migration_thread_after_loop();
2978 migration_iteration_finish(s
);
2979 rcu_unregister_thread();
2983 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
2986 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
2988 s
->expected_downtime
= s
->parameters
.downtime_limit
;
2989 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
2991 migrate_fd_error(s
, error_in
);
2992 migrate_fd_cleanup(s
);
2997 /* This is a resumed migration */
2998 rate_limit
= INT64_MAX
;
3000 /* This is a fresh new migration */
3001 rate_limit
= s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
;
3002 s
->expected_downtime
= s
->parameters
.downtime_limit
;
3003 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
3005 /* Notify before starting migration thread */
3006 notifier_list_notify(&migration_state_notifiers
, s
);
3009 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
3010 qemu_file_set_blocking(s
->to_dst_file
, true);
3013 * Open the return path. For postcopy, it is used exclusively. For
3014 * precopy, only if user specified "return-path" capability would
3015 * QEMU uses the return path.
3017 if (migrate_postcopy_ram() || migrate_use_return_path()) {
3018 if (open_return_path_on_source(s
, !resume
)) {
3019 error_report("Unable to open return-path for postcopy");
3020 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
3021 migrate_fd_cleanup(s
);
3027 /* Wakeup the main migration thread to do the recovery */
3028 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
3029 MIGRATION_STATUS_POSTCOPY_RECOVER
);
3030 qemu_sem_post(&s
->postcopy_pause_sem
);
3034 if (multifd_save_setup() != 0) {
3035 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3036 MIGRATION_STATUS_FAILED
);
3037 migrate_fd_cleanup(s
);
3040 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
3041 QEMU_THREAD_JOINABLE
);
3042 s
->migration_thread_running
= true;
3045 void migration_global_dump(Monitor
*mon
)
3047 MigrationState
*ms
= migrate_get_current();
3049 monitor_printf(mon
, "globals:\n");
3050 monitor_printf(mon
, "store-global-state: %s\n",
3051 ms
->store_global_state
? "on" : "off");
3052 monitor_printf(mon
, "only-migratable: %s\n",
3053 ms
->only_migratable
? "on" : "off");
3054 monitor_printf(mon
, "send-configuration: %s\n",
3055 ms
->send_configuration
? "on" : "off");
3056 monitor_printf(mon
, "send-section-footer: %s\n",
3057 ms
->send_section_footer
? "on" : "off");
3058 monitor_printf(mon
, "decompress-error-check: %s\n",
3059 ms
->decompress_error_check
? "on" : "off");
3062 #define DEFINE_PROP_MIG_CAP(name, x) \
3063 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3065 static Property migration_properties
[] = {
3066 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
3067 store_global_state
, true),
3068 DEFINE_PROP_BOOL("only-migratable", MigrationState
, only_migratable
, false),
3069 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
3070 send_configuration
, true),
3071 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
3072 send_section_footer
, true),
3073 DEFINE_PROP_BOOL("decompress-error-check", MigrationState
,
3074 decompress_error_check
, true),
3076 /* Migration parameters */
3077 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
3078 parameters
.compress_level
,
3079 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
3080 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
3081 parameters
.compress_threads
,
3082 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
3083 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
3084 parameters
.decompress_threads
,
3085 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
3086 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
3087 parameters
.cpu_throttle_initial
,
3088 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
3089 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
3090 parameters
.cpu_throttle_increment
,
3091 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
3092 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
3093 parameters
.max_bandwidth
, MAX_THROTTLE
),
3094 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
3095 parameters
.downtime_limit
,
3096 DEFAULT_MIGRATE_SET_DOWNTIME
),
3097 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
3098 parameters
.x_checkpoint_delay
,
3099 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
3100 DEFINE_PROP_UINT8("x-multifd-channels", MigrationState
,
3101 parameters
.x_multifd_channels
,
3102 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
3103 DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState
,
3104 parameters
.x_multifd_page_count
,
3105 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT
),
3106 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
3107 parameters
.xbzrle_cache_size
,
3108 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
3109 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState
,
3110 parameters
.max_postcopy_bandwidth
,
3111 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH
),
3113 /* Migration capabilities */
3114 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
3115 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
3116 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
3117 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
3118 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
3119 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
3120 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
3121 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
3122 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
3123 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
3124 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
3125 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD
),
3127 DEFINE_PROP_END_OF_LIST(),
3130 static void migration_class_init(ObjectClass
*klass
, void *data
)
3132 DeviceClass
*dc
= DEVICE_CLASS(klass
);
3134 dc
->user_creatable
= false;
3135 dc
->props
= migration_properties
;
3138 static void migration_instance_finalize(Object
*obj
)
3140 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3141 MigrationParameters
*params
= &ms
->parameters
;
3143 qemu_mutex_destroy(&ms
->error_mutex
);
3144 qemu_mutex_destroy(&ms
->qemu_file_lock
);
3145 g_free(params
->tls_hostname
);
3146 g_free(params
->tls_creds
);
3147 qemu_sem_destroy(&ms
->rate_limit_sem
);
3148 qemu_sem_destroy(&ms
->pause_sem
);
3149 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
3150 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
3151 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
3152 error_free(ms
->error
);
3155 static void migration_instance_init(Object
*obj
)
3157 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3158 MigrationParameters
*params
= &ms
->parameters
;
3160 ms
->state
= MIGRATION_STATUS_NONE
;
3162 qemu_sem_init(&ms
->pause_sem
, 0);
3163 qemu_mutex_init(&ms
->error_mutex
);
3165 params
->tls_hostname
= g_strdup("");
3166 params
->tls_creds
= g_strdup("");
3168 /* Set has_* up only for parameter checks */
3169 params
->has_compress_level
= true;
3170 params
->has_compress_threads
= true;
3171 params
->has_decompress_threads
= true;
3172 params
->has_cpu_throttle_initial
= true;
3173 params
->has_cpu_throttle_increment
= true;
3174 params
->has_max_bandwidth
= true;
3175 params
->has_downtime_limit
= true;
3176 params
->has_x_checkpoint_delay
= true;
3177 params
->has_block_incremental
= true;
3178 params
->has_x_multifd_channels
= true;
3179 params
->has_x_multifd_page_count
= true;
3180 params
->has_xbzrle_cache_size
= true;
3181 params
->has_max_postcopy_bandwidth
= true;
3183 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
3184 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
3185 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
3186 qemu_sem_init(&ms
->rate_limit_sem
, 0);
3187 qemu_mutex_init(&ms
->qemu_file_lock
);
3191 * Return true if check pass, false otherwise. Error will be put
3192 * inside errp if provided.
3194 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
3196 MigrationCapabilityStatusList
*head
= NULL
;
3197 /* Assuming all off */
3198 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
3201 if (!migrate_params_check(&ms
->parameters
, errp
)) {
3205 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
3206 if (ms
->enabled_capabilities
[i
]) {
3207 head
= migrate_cap_add(head
, i
, true);
3211 ret
= migrate_caps_check(cap_list
, head
, errp
);
3213 /* It works with head == NULL */
3214 qapi_free_MigrationCapabilityStatusList(head
);
3219 static const TypeInfo migration_type
= {
3220 .name
= TYPE_MIGRATION
,
3222 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3223 * not created using qdev_create(), it is not attached to the qdev
3224 * device tree, and it is never realized.
3226 * TODO: Make this TYPE_OBJECT once QOM provides something like
3227 * TYPE_DEVICE's "-global" properties.
3229 .parent
= TYPE_DEVICE
,
3230 .class_init
= migration_class_init
,
3231 .class_size
= sizeof(MigrationClass
),
3232 .instance_size
= sizeof(MigrationState
),
3233 .instance_init
= migration_instance_init
,
3234 .instance_finalize
= migration_instance_finalize
,
3237 static void register_migration_types(void)
3239 type_register_static(&migration_type
);
3242 type_init(register_migration_types
);