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 "qemu/main-loop.h"
20 #include "migration/blocker.h"
24 #include "sysemu/runstate.h"
25 #include "sysemu/sysemu.h"
26 #include "sysemu/cpu-throttle.h"
29 #include "migration/global_state.h"
30 #include "migration/misc.h"
31 #include "migration.h"
33 #include "qemu-file-channel.h"
34 #include "qemu-file.h"
35 #include "migration/vmstate.h"
36 #include "block/block.h"
37 #include "qapi/error.h"
38 #include "qapi/clone-visitor.h"
39 #include "qapi/qapi-visit-sockets.h"
40 #include "qapi/qapi-commands-migration.h"
41 #include "qapi/qapi-events-migration.h"
42 #include "qapi/qmp/qerror.h"
43 #include "qapi/qmp/qnull.h"
46 #include "postcopy-ram.h"
47 #include "qemu/thread.h"
49 #include "exec/target_page.h"
50 #include "io/channel-buffer.h"
51 #include "migration/colo.h"
52 #include "hw/boards.h"
53 #include "hw/qdev-properties.h"
54 #include "monitor/monitor.h"
55 #include "net/announce.h"
56 #include "qemu/queue.h"
59 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
61 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
63 #define BUFFER_DELAY 100
64 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
66 /* Time in milliseconds we are allowed to stop the source,
67 * for sending the last part */
68 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
70 /* Maximum migrate downtime set to 2000 seconds */
71 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
72 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
74 /* Default compression thread count */
75 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
76 /* Default decompression thread count, usually decompression is at
77 * least 4 times as fast as compression.*/
78 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
79 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
80 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
81 /* Define default autoconverge cpu throttle migration parameters */
82 #define DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD 50
83 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
84 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
85 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
87 /* Migration XBZRLE default cache size */
88 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
90 /* The delay time (in ms) between two COLO checkpoints */
91 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
92 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
93 #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE
94 /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */
95 #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
96 /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
97 #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
99 /* Background transfer rate for postcopy, 0 means unlimited, note
100 * that page requests can still exceed this limit.
102 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
105 * Parameters for self_announce_delay giving a stream of RARP/ARP
106 * packets after migration.
108 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
109 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
110 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
111 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
113 static NotifierList migration_state_notifiers
=
114 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
116 static bool deferred_incoming
;
118 /* Messages sent on the return path from destination to source */
119 enum mig_rp_message_type
{
120 MIG_RP_MSG_INVALID
= 0, /* Must be 0 */
121 MIG_RP_MSG_SHUT
, /* sibling will not send any more RP messages */
122 MIG_RP_MSG_PONG
, /* Response to a PING; data (seq: be32 ) */
124 MIG_RP_MSG_REQ_PAGES_ID
, /* data (start: be64, len: be32, id: string) */
125 MIG_RP_MSG_REQ_PAGES
, /* data (start: be64, len: be32) */
126 MIG_RP_MSG_RECV_BITMAP
, /* send recved_bitmap back to source */
127 MIG_RP_MSG_RESUME_ACK
, /* tell source that we are ready to resume */
132 /* When we add fault tolerance, we could have several
133 migrations at once. For now we don't need to add
134 dynamic creation of migration */
136 static MigrationState
*current_migration
;
137 static MigrationIncomingState
*current_incoming
;
139 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
140 static int migration_maybe_pause(MigrationState
*s
,
141 int *current_active_state
,
143 static void migrate_fd_cancel(MigrationState
*s
);
145 void migration_object_init(void)
147 MachineState
*ms
= MACHINE(qdev_get_machine());
150 /* This can only be called once. */
151 assert(!current_migration
);
152 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
155 * Init the migrate incoming object as well no matter whether
156 * we'll use it or not.
158 assert(!current_incoming
);
159 current_incoming
= g_new0(MigrationIncomingState
, 1);
160 current_incoming
->state
= MIGRATION_STATUS_NONE
;
161 current_incoming
->postcopy_remote_fds
=
162 g_array_new(FALSE
, TRUE
, sizeof(struct PostCopyFD
));
163 qemu_mutex_init(¤t_incoming
->rp_mutex
);
164 qemu_event_init(¤t_incoming
->main_thread_load_event
, false);
165 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_dst
, 0);
166 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fault
, 0);
168 if (!migration_object_check(current_migration
, &err
)) {
169 error_report_err(err
);
174 * We cannot really do this in migration_instance_init() since at
175 * that time global properties are not yet applied, then this
176 * value will be definitely replaced by something else.
178 if (ms
->enforce_config_section
) {
179 current_migration
->send_configuration
= true;
183 void migration_shutdown(void)
186 * Cancel the current migration - that will (eventually)
187 * stop the migration using this structure
189 migrate_fd_cancel(current_migration
);
190 object_unref(OBJECT(current_migration
));
193 * Cancel outgoing migration of dirty bitmaps. It should
194 * at least unref used block nodes.
196 dirty_bitmap_mig_cancel_outgoing();
199 * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
200 * are non-critical data, and their loss never considered as
203 dirty_bitmap_mig_cancel_incoming();
207 MigrationState
*migrate_get_current(void)
209 /* This can only be called after the object created. */
210 assert(current_migration
);
211 return current_migration
;
214 MigrationIncomingState
*migration_incoming_get_current(void)
216 assert(current_incoming
);
217 return current_incoming
;
220 void migration_incoming_state_destroy(void)
222 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
224 if (mis
->to_src_file
) {
225 /* Tell source that we are done */
226 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
227 qemu_fclose(mis
->to_src_file
);
228 mis
->to_src_file
= NULL
;
231 if (mis
->from_src_file
) {
232 qemu_fclose(mis
->from_src_file
);
233 mis
->from_src_file
= NULL
;
235 if (mis
->postcopy_remote_fds
) {
236 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
237 mis
->postcopy_remote_fds
= NULL
;
240 qemu_event_reset(&mis
->main_thread_load_event
);
242 if (mis
->socket_address_list
) {
243 qapi_free_SocketAddressList(mis
->socket_address_list
);
244 mis
->socket_address_list
= NULL
;
248 static void migrate_generate_event(int new_state
)
250 if (migrate_use_events()) {
251 qapi_event_send_migration(new_state
);
255 static bool migrate_late_block_activate(void)
259 s
= migrate_get_current();
261 return s
->enabled_capabilities
[
262 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE
];
266 * Called on -incoming with a defer: uri.
267 * The migration can be started later after any parameters have been
270 static void deferred_incoming_migration(Error
**errp
)
272 if (deferred_incoming
) {
273 error_setg(errp
, "Incoming migration already deferred");
275 deferred_incoming
= true;
279 * Send a message on the return channel back to the source
282 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
283 enum mig_rp_message_type message_type
,
284 uint16_t len
, void *data
)
288 trace_migrate_send_rp_message((int)message_type
, len
);
289 qemu_mutex_lock(&mis
->rp_mutex
);
292 * It's possible that the file handle got lost due to network
295 if (!mis
->to_src_file
) {
300 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
301 qemu_put_be16(mis
->to_src_file
, len
);
302 qemu_put_buffer(mis
->to_src_file
, data
, len
);
303 qemu_fflush(mis
->to_src_file
);
305 /* It's possible that qemu file got error during sending */
306 ret
= qemu_file_get_error(mis
->to_src_file
);
309 qemu_mutex_unlock(&mis
->rp_mutex
);
313 /* Request a range of pages from the source VM at the given
315 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
316 * as the last request (a name must have been given previously)
317 * Start: Address offset within the RB
318 * Len: Length in bytes required - must be a multiple of pagesize
320 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
, const char *rbname
,
321 ram_addr_t start
, size_t len
)
323 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
324 size_t msglen
= 12; /* start + len */
325 enum mig_rp_message_type msg_type
;
327 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
328 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
331 int rbname_len
= strlen(rbname
);
332 assert(rbname_len
< 256);
334 bufc
[msglen
++] = rbname_len
;
335 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
336 msglen
+= rbname_len
;
337 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
339 msg_type
= MIG_RP_MSG_REQ_PAGES
;
342 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
345 static bool migration_colo_enabled
;
346 bool migration_incoming_colo_enabled(void)
348 return migration_colo_enabled
;
351 void migration_incoming_disable_colo(void)
353 ram_block_discard_disable(false);
354 migration_colo_enabled
= false;
357 int migration_incoming_enable_colo(void)
359 if (ram_block_discard_disable(true)) {
360 error_report("COLO: cannot disable RAM discard");
363 migration_colo_enabled
= true;
367 void migrate_add_address(SocketAddress
*address
)
369 MigrationIncomingState
*mis
= migration_incoming_get_current();
370 SocketAddressList
*addrs
;
372 addrs
= g_new0(SocketAddressList
, 1);
373 addrs
->next
= mis
->socket_address_list
;
374 mis
->socket_address_list
= addrs
;
375 addrs
->value
= QAPI_CLONE(SocketAddress
, address
);
378 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
382 qapi_event_send_migration(MIGRATION_STATUS_SETUP
);
383 if (!strcmp(uri
, "defer")) {
384 deferred_incoming_migration(errp
);
385 } else if (strstart(uri
, "tcp:", &p
)) {
386 tcp_start_incoming_migration(p
, errp
);
388 } else if (strstart(uri
, "rdma:", &p
)) {
389 rdma_start_incoming_migration(p
, errp
);
391 } else if (strstart(uri
, "exec:", &p
)) {
392 exec_start_incoming_migration(p
, errp
);
393 } else if (strstart(uri
, "unix:", &p
)) {
394 unix_start_incoming_migration(p
, errp
);
395 } else if (strstart(uri
, "fd:", &p
)) {
396 fd_start_incoming_migration(p
, errp
);
398 error_setg(errp
, "unknown migration protocol: %s", uri
);
402 static void process_incoming_migration_bh(void *opaque
)
404 Error
*local_err
= NULL
;
405 MigrationIncomingState
*mis
= opaque
;
407 /* If capability late_block_activate is set:
408 * Only fire up the block code now if we're going to restart the
409 * VM, else 'cont' will do it.
410 * This causes file locking to happen; so we don't want it to happen
411 * unless we really are starting the VM.
413 if (!migrate_late_block_activate() ||
414 (autostart
&& (!global_state_received() ||
415 global_state_get_runstate() == RUN_STATE_RUNNING
))) {
416 /* Make sure all file formats flush their mutable metadata.
417 * If we get an error here, just don't restart the VM yet. */
418 bdrv_invalidate_cache_all(&local_err
);
420 error_report_err(local_err
);
427 * This must happen after all error conditions are dealt with and
428 * we're sure the VM is going to be running on this host.
430 qemu_announce_self(&mis
->announce_timer
, migrate_announce_params());
432 if (multifd_load_cleanup(&local_err
) != 0) {
433 error_report_err(local_err
);
436 /* If global state section was not received or we are in running
437 state, we need to obey autostart. Any other state is set with
440 dirty_bitmap_mig_before_vm_start();
442 if (!global_state_received() ||
443 global_state_get_runstate() == RUN_STATE_RUNNING
) {
447 runstate_set(RUN_STATE_PAUSED
);
449 } else if (migration_incoming_colo_enabled()) {
450 migration_incoming_disable_colo();
453 runstate_set(global_state_get_runstate());
456 * This must happen after any state changes since as soon as an external
457 * observer sees this event they might start to prod at the VM assuming
460 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
461 MIGRATION_STATUS_COMPLETED
);
462 qemu_bh_delete(mis
->bh
);
463 migration_incoming_state_destroy();
466 static void process_incoming_migration_co(void *opaque
)
468 MigrationIncomingState
*mis
= migration_incoming_get_current();
471 Error
*local_err
= NULL
;
473 assert(mis
->from_src_file
);
474 mis
->migration_incoming_co
= qemu_coroutine_self();
475 mis
->largest_page_size
= qemu_ram_pagesize_largest();
476 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
477 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
478 MIGRATION_STATUS_ACTIVE
);
479 ret
= qemu_loadvm_state(mis
->from_src_file
);
481 ps
= postcopy_state_get();
482 trace_process_incoming_migration_co_end(ret
, ps
);
483 if (ps
!= POSTCOPY_INCOMING_NONE
) {
484 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
486 * Where a migration had postcopy enabled (and thus went to advise)
487 * but managed to complete within the precopy period, we can use
490 postcopy_ram_incoming_cleanup(mis
);
491 } else if (ret
>= 0) {
493 * Postcopy was started, cleanup should happen at the end of the
496 trace_process_incoming_migration_co_postcopy_end_main();
499 /* Else if something went wrong then just fall out of the normal exit */
502 /* we get COLO info, and know if we are in COLO mode */
503 if (!ret
&& migration_incoming_colo_enabled()) {
504 /* Make sure all file formats flush their mutable metadata */
505 bdrv_invalidate_cache_all(&local_err
);
507 error_report_err(local_err
);
511 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
512 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
513 mis
->have_colo_incoming_thread
= true;
514 qemu_coroutine_yield();
516 /* Wait checkpoint incoming thread exit before free resource */
517 qemu_thread_join(&mis
->colo_incoming_thread
);
518 /* We hold the global iothread lock, so it is safe here */
519 colo_release_ram_cache();
523 error_report("load of migration failed: %s", strerror(-ret
));
526 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
527 qemu_bh_schedule(mis
->bh
);
528 mis
->migration_incoming_co
= NULL
;
532 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
533 MIGRATION_STATUS_FAILED
);
534 qemu_fclose(mis
->from_src_file
);
535 if (multifd_load_cleanup(&local_err
) != 0) {
536 error_report_err(local_err
);
542 * @migration_incoming_setup: Setup incoming migration
544 * Returns 0 for no error or 1 for error
546 * @f: file for main migration channel
547 * @errp: where to put errors
549 static int migration_incoming_setup(QEMUFile
*f
, Error
**errp
)
551 MigrationIncomingState
*mis
= migration_incoming_get_current();
552 Error
*local_err
= NULL
;
554 if (multifd_load_setup(&local_err
) != 0) {
555 /* We haven't been able to create multifd threads
556 nothing better to do */
557 error_report_err(local_err
);
561 if (!mis
->from_src_file
) {
562 mis
->from_src_file
= f
;
564 qemu_file_set_blocking(f
, false);
568 void migration_incoming_process(void)
570 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
571 qemu_coroutine_enter(co
);
574 /* Returns true if recovered from a paused migration, otherwise false */
575 static bool postcopy_try_recover(QEMUFile
*f
)
577 MigrationIncomingState
*mis
= migration_incoming_get_current();
579 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
580 /* Resumed from a paused postcopy migration */
582 mis
->from_src_file
= f
;
583 /* Postcopy has standalone thread to do vm load */
584 qemu_file_set_blocking(f
, true);
586 /* Re-configure the return path */
587 mis
->to_src_file
= qemu_file_get_return_path(f
);
589 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
590 MIGRATION_STATUS_POSTCOPY_RECOVER
);
593 * Here, we only wake up the main loading thread (while the
594 * fault thread will still be waiting), so that we can receive
595 * commands from source now, and answer it if needed. The
596 * fault thread will be woken up afterwards until we are sure
597 * that source is ready to reply to page requests.
599 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
606 void migration_fd_process_incoming(QEMUFile
*f
, Error
**errp
)
608 Error
*local_err
= NULL
;
610 if (postcopy_try_recover(f
)) {
614 if (migration_incoming_setup(f
, &local_err
)) {
616 error_propagate(errp
, local_err
);
620 migration_incoming_process();
623 void migration_ioc_process_incoming(QIOChannel
*ioc
, Error
**errp
)
625 MigrationIncomingState
*mis
= migration_incoming_get_current();
626 Error
*local_err
= NULL
;
627 bool start_migration
;
629 if (!mis
->from_src_file
) {
630 /* The first connection (multifd may have multiple) */
631 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
633 /* If it's a recovery, we're done */
634 if (postcopy_try_recover(f
)) {
638 if (migration_incoming_setup(f
, &local_err
)) {
640 error_propagate(errp
, local_err
);
646 * Common migration only needs one channel, so we can start
647 * right now. Multifd needs more than one channel, we wait.
649 start_migration
= !migrate_use_multifd();
651 /* Multiple connections */
652 assert(migrate_use_multifd());
653 start_migration
= multifd_recv_new_channel(ioc
, &local_err
);
655 error_propagate(errp
, local_err
);
660 if (start_migration
) {
661 migration_incoming_process();
666 * @migration_has_all_channels: We have received all channels that we need
668 * Returns true when we have got connections to all the channels that
669 * we need for migration.
671 bool migration_has_all_channels(void)
673 MigrationIncomingState
*mis
= migration_incoming_get_current();
676 all_channels
= multifd_recv_all_channels_created();
678 return all_channels
&& mis
->from_src_file
!= NULL
;
682 * Send a 'SHUT' message on the return channel with the given value
683 * to indicate that we've finished with the RP. Non-0 value indicates
686 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
691 buf
= cpu_to_be32(value
);
692 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
696 * Send a 'PONG' message on the return channel with the given value
697 * (normally in response to a 'PING')
699 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
704 buf
= cpu_to_be32(value
);
705 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
708 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
716 * First, we send the header part. It contains only the len of
717 * idstr, and the idstr itself.
719 len
= strlen(block_name
);
721 memcpy(buf
+ 1, block_name
, len
);
723 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
724 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
729 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
732 * Next, we dump the received bitmap to the stream.
734 * TODO: currently we are safe since we are the only one that is
735 * using the to_src_file handle (fault thread is still paused),
736 * and it's ok even not taking the mutex. However the best way is
737 * to take the lock before sending the message header, and release
738 * the lock after sending the bitmap.
740 qemu_mutex_lock(&mis
->rp_mutex
);
741 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
742 qemu_mutex_unlock(&mis
->rp_mutex
);
744 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
747 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
751 buf
= cpu_to_be32(value
);
752 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
755 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
757 MigrationCapabilityStatusList
*head
= NULL
;
758 MigrationCapabilityStatusList
*caps
;
759 MigrationState
*s
= migrate_get_current();
762 caps
= NULL
; /* silence compiler warning */
763 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
764 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
765 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
770 head
= g_malloc0(sizeof(*caps
));
773 caps
->next
= g_malloc0(sizeof(*caps
));
777 g_malloc(sizeof(*caps
->value
));
778 caps
->value
->capability
= i
;
779 caps
->value
->state
= s
->enabled_capabilities
[i
];
785 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
787 MigrationParameters
*params
;
788 MigrationState
*s
= migrate_get_current();
790 /* TODO use QAPI_CLONE() instead of duplicating it inline */
791 params
= g_malloc0(sizeof(*params
));
792 params
->has_compress_level
= true;
793 params
->compress_level
= s
->parameters
.compress_level
;
794 params
->has_compress_threads
= true;
795 params
->compress_threads
= s
->parameters
.compress_threads
;
796 params
->has_compress_wait_thread
= true;
797 params
->compress_wait_thread
= s
->parameters
.compress_wait_thread
;
798 params
->has_decompress_threads
= true;
799 params
->decompress_threads
= s
->parameters
.decompress_threads
;
800 params
->has_throttle_trigger_threshold
= true;
801 params
->throttle_trigger_threshold
= s
->parameters
.throttle_trigger_threshold
;
802 params
->has_cpu_throttle_initial
= true;
803 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
804 params
->has_cpu_throttle_increment
= true;
805 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
806 params
->has_cpu_throttle_tailslow
= true;
807 params
->cpu_throttle_tailslow
= s
->parameters
.cpu_throttle_tailslow
;
808 params
->has_tls_creds
= true;
809 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
810 params
->has_tls_hostname
= true;
811 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
812 params
->has_tls_authz
= true;
813 params
->tls_authz
= g_strdup(s
->parameters
.tls_authz
?
814 s
->parameters
.tls_authz
: "");
815 params
->has_max_bandwidth
= true;
816 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
817 params
->has_downtime_limit
= true;
818 params
->downtime_limit
= s
->parameters
.downtime_limit
;
819 params
->has_x_checkpoint_delay
= true;
820 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
821 params
->has_block_incremental
= true;
822 params
->block_incremental
= s
->parameters
.block_incremental
;
823 params
->has_multifd_channels
= true;
824 params
->multifd_channels
= s
->parameters
.multifd_channels
;
825 params
->has_multifd_compression
= true;
826 params
->multifd_compression
= s
->parameters
.multifd_compression
;
827 params
->has_multifd_zlib_level
= true;
828 params
->multifd_zlib_level
= s
->parameters
.multifd_zlib_level
;
829 params
->has_multifd_zstd_level
= true;
830 params
->multifd_zstd_level
= s
->parameters
.multifd_zstd_level
;
831 params
->has_xbzrle_cache_size
= true;
832 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
833 params
->has_max_postcopy_bandwidth
= true;
834 params
->max_postcopy_bandwidth
= s
->parameters
.max_postcopy_bandwidth
;
835 params
->has_max_cpu_throttle
= true;
836 params
->max_cpu_throttle
= s
->parameters
.max_cpu_throttle
;
837 params
->has_announce_initial
= true;
838 params
->announce_initial
= s
->parameters
.announce_initial
;
839 params
->has_announce_max
= true;
840 params
->announce_max
= s
->parameters
.announce_max
;
841 params
->has_announce_rounds
= true;
842 params
->announce_rounds
= s
->parameters
.announce_rounds
;
843 params
->has_announce_step
= true;
844 params
->announce_step
= s
->parameters
.announce_step
;
849 AnnounceParameters
*migrate_announce_params(void)
851 static AnnounceParameters ap
;
853 MigrationState
*s
= migrate_get_current();
855 ap
.initial
= s
->parameters
.announce_initial
;
856 ap
.max
= s
->parameters
.announce_max
;
857 ap
.rounds
= s
->parameters
.announce_rounds
;
858 ap
.step
= s
->parameters
.announce_step
;
864 * Return true if we're already in the middle of a migration
865 * (i.e. any of the active or setup states)
867 bool migration_is_setup_or_active(int state
)
870 case MIGRATION_STATUS_ACTIVE
:
871 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
872 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
873 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
874 case MIGRATION_STATUS_SETUP
:
875 case MIGRATION_STATUS_PRE_SWITCHOVER
:
876 case MIGRATION_STATUS_DEVICE
:
877 case MIGRATION_STATUS_WAIT_UNPLUG
:
878 case MIGRATION_STATUS_COLO
:
887 bool migration_is_running(int state
)
890 case MIGRATION_STATUS_ACTIVE
:
891 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
892 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
893 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
894 case MIGRATION_STATUS_SETUP
:
895 case MIGRATION_STATUS_PRE_SWITCHOVER
:
896 case MIGRATION_STATUS_DEVICE
:
897 case MIGRATION_STATUS_WAIT_UNPLUG
:
898 case MIGRATION_STATUS_CANCELLING
:
907 static void populate_time_info(MigrationInfo
*info
, MigrationState
*s
)
909 info
->has_status
= true;
910 info
->has_setup_time
= true;
911 info
->setup_time
= s
->setup_time
;
912 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
913 info
->has_total_time
= true;
914 info
->total_time
= s
->total_time
;
915 info
->has_downtime
= true;
916 info
->downtime
= s
->downtime
;
918 info
->has_total_time
= true;
919 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) -
921 info
->has_expected_downtime
= true;
922 info
->expected_downtime
= s
->expected_downtime
;
926 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
928 info
->has_ram
= true;
929 info
->ram
= g_malloc0(sizeof(*info
->ram
));
930 info
->ram
->transferred
= ram_counters
.transferred
;
931 info
->ram
->total
= ram_bytes_total();
932 info
->ram
->duplicate
= ram_counters
.duplicate
;
933 /* legacy value. It is not used anymore */
934 info
->ram
->skipped
= 0;
935 info
->ram
->normal
= ram_counters
.normal
;
936 info
->ram
->normal_bytes
= ram_counters
.normal
*
937 qemu_target_page_size();
938 info
->ram
->mbps
= s
->mbps
;
939 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
940 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
941 info
->ram
->page_size
= qemu_target_page_size();
942 info
->ram
->multifd_bytes
= ram_counters
.multifd_bytes
;
943 info
->ram
->pages_per_second
= s
->pages_per_second
;
945 if (migrate_use_xbzrle()) {
946 info
->has_xbzrle_cache
= true;
947 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
948 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
949 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
950 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
951 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
952 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
953 info
->xbzrle_cache
->encoding_rate
= xbzrle_counters
.encoding_rate
;
954 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
957 if (migrate_use_compression()) {
958 info
->has_compression
= true;
959 info
->compression
= g_malloc0(sizeof(*info
->compression
));
960 info
->compression
->pages
= compression_counters
.pages
;
961 info
->compression
->busy
= compression_counters
.busy
;
962 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
963 info
->compression
->compressed_size
=
964 compression_counters
.compressed_size
;
965 info
->compression
->compression_rate
=
966 compression_counters
.compression_rate
;
969 if (cpu_throttle_active()) {
970 info
->has_cpu_throttle_percentage
= true;
971 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
974 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
975 info
->ram
->remaining
= ram_bytes_remaining();
976 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
980 static void populate_disk_info(MigrationInfo
*info
)
982 if (blk_mig_active()) {
983 info
->has_disk
= true;
984 info
->disk
= g_malloc0(sizeof(*info
->disk
));
985 info
->disk
->transferred
= blk_mig_bytes_transferred();
986 info
->disk
->remaining
= blk_mig_bytes_remaining();
987 info
->disk
->total
= blk_mig_bytes_total();
991 static void fill_source_migration_info(MigrationInfo
*info
)
993 MigrationState
*s
= migrate_get_current();
996 case MIGRATION_STATUS_NONE
:
997 /* no migration has happened ever */
998 /* do not overwrite destination migration status */
1000 case MIGRATION_STATUS_SETUP
:
1001 info
->has_status
= true;
1002 info
->has_total_time
= false;
1004 case MIGRATION_STATUS_ACTIVE
:
1005 case MIGRATION_STATUS_CANCELLING
:
1006 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1007 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1008 case MIGRATION_STATUS_DEVICE
:
1009 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1010 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1011 /* TODO add some postcopy stats */
1012 populate_time_info(info
, s
);
1013 populate_ram_info(info
, s
);
1014 populate_disk_info(info
);
1016 case MIGRATION_STATUS_COLO
:
1017 info
->has_status
= true;
1018 /* TODO: display COLO specific information (checkpoint info etc.) */
1020 case MIGRATION_STATUS_COMPLETED
:
1021 populate_time_info(info
, s
);
1022 populate_ram_info(info
, s
);
1024 case MIGRATION_STATUS_FAILED
:
1025 info
->has_status
= true;
1027 info
->has_error_desc
= true;
1028 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
1031 case MIGRATION_STATUS_CANCELLED
:
1032 info
->has_status
= true;
1034 case MIGRATION_STATUS_WAIT_UNPLUG
:
1035 info
->has_status
= true;
1038 info
->status
= s
->state
;
1042 * @migration_caps_check - check capability validity
1044 * @cap_list: old capability list, array of bool
1045 * @params: new capabilities to be applied soon
1046 * @errp: set *errp if the check failed, with reason
1048 * Returns true if check passed, otherwise false.
1050 static bool migrate_caps_check(bool *cap_list
,
1051 MigrationCapabilityStatusList
*params
,
1054 MigrationCapabilityStatusList
*cap
;
1055 bool old_postcopy_cap
;
1056 MigrationIncomingState
*mis
= migration_incoming_get_current();
1058 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1060 for (cap
= params
; cap
; cap
= cap
->next
) {
1061 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
1064 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1065 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
1066 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
1068 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
1073 #ifndef CONFIG_REPLICATION
1074 if (cap_list
[MIGRATION_CAPABILITY_X_COLO
]) {
1075 error_setg(errp
, "QEMU compiled without replication module"
1076 " can't enable COLO");
1077 error_append_hint(errp
, "Please enable replication before COLO.\n");
1082 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
1083 /* This check is reasonably expensive, so only when it's being
1084 * set the first time, also it's only the destination that needs
1087 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
1088 !postcopy_ram_supported_by_host(mis
)) {
1089 /* postcopy_ram_supported_by_host will have emitted a more
1092 error_setg(errp
, "Postcopy is not supported");
1096 if (cap_list
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
]) {
1097 error_setg(errp
, "Postcopy is not compatible with ignore-shared");
1105 static void fill_destination_migration_info(MigrationInfo
*info
)
1107 MigrationIncomingState
*mis
= migration_incoming_get_current();
1109 if (mis
->socket_address_list
) {
1110 info
->has_socket_address
= true;
1111 info
->socket_address
=
1112 QAPI_CLONE(SocketAddressList
, mis
->socket_address_list
);
1115 switch (mis
->state
) {
1116 case MIGRATION_STATUS_NONE
:
1118 case MIGRATION_STATUS_SETUP
:
1119 case MIGRATION_STATUS_CANCELLING
:
1120 case MIGRATION_STATUS_CANCELLED
:
1121 case MIGRATION_STATUS_ACTIVE
:
1122 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1123 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1124 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1125 case MIGRATION_STATUS_FAILED
:
1126 case MIGRATION_STATUS_COLO
:
1127 info
->has_status
= true;
1129 case MIGRATION_STATUS_COMPLETED
:
1130 info
->has_status
= true;
1131 fill_destination_postcopy_migration_info(info
);
1134 info
->status
= mis
->state
;
1137 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1139 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1141 fill_destination_migration_info(info
);
1142 fill_source_migration_info(info
);
1147 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
1150 MigrationState
*s
= migrate_get_current();
1151 MigrationCapabilityStatusList
*cap
;
1152 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
1154 if (migration_is_running(s
->state
)) {
1155 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1159 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
1160 if (!migrate_caps_check(cap_list
, params
, errp
)) {
1164 for (cap
= params
; cap
; cap
= cap
->next
) {
1165 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
1170 * Check whether the parameters are valid. Error will be put into errp
1171 * (if provided). Return true if valid, otherwise false.
1173 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
1175 if (params
->has_compress_level
&&
1176 (params
->compress_level
> 9)) {
1177 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
1178 "is invalid, it should be in the range of 0 to 9");
1182 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
1183 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1185 "is invalid, it should be in the range of 1 to 255");
1189 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
1190 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1191 "decompress_threads",
1192 "is invalid, it should be in the range of 1 to 255");
1196 if (params
->has_throttle_trigger_threshold
&&
1197 (params
->throttle_trigger_threshold
< 1 ||
1198 params
->throttle_trigger_threshold
> 100)) {
1199 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1200 "throttle_trigger_threshold",
1201 "an integer in the range of 1 to 100");
1205 if (params
->has_cpu_throttle_initial
&&
1206 (params
->cpu_throttle_initial
< 1 ||
1207 params
->cpu_throttle_initial
> 99)) {
1208 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1209 "cpu_throttle_initial",
1210 "an integer in the range of 1 to 99");
1214 if (params
->has_cpu_throttle_increment
&&
1215 (params
->cpu_throttle_increment
< 1 ||
1216 params
->cpu_throttle_increment
> 99)) {
1217 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1218 "cpu_throttle_increment",
1219 "an integer in the range of 1 to 99");
1223 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
1224 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1226 "an integer in the range of 0 to "stringify(SIZE_MAX
)
1231 if (params
->has_downtime_limit
&&
1232 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
1233 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1235 "an integer in the range of 0 to "
1236 stringify(MAX_MIGRATE_DOWNTIME
)" ms");
1240 /* x_checkpoint_delay is now always positive */
1242 if (params
->has_multifd_channels
&& (params
->multifd_channels
< 1)) {
1243 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1245 "is invalid, it should be in the range of 1 to 255");
1249 if (params
->has_multifd_zlib_level
&&
1250 (params
->multifd_zlib_level
> 9)) {
1251 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zlib_level",
1252 "is invalid, it should be in the range of 0 to 9");
1256 if (params
->has_multifd_zstd_level
&&
1257 (params
->multifd_zstd_level
> 20)) {
1258 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zstd_level",
1259 "is invalid, it should be in the range of 0 to 20");
1263 if (params
->has_xbzrle_cache_size
&&
1264 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
1265 !is_power_of_2(params
->xbzrle_cache_size
))) {
1266 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1267 "xbzrle_cache_size",
1268 "is invalid, it should be bigger than target page size"
1269 " and a power of 2");
1273 if (params
->has_max_cpu_throttle
&&
1274 (params
->max_cpu_throttle
< params
->cpu_throttle_initial
||
1275 params
->max_cpu_throttle
> 99)) {
1276 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1278 "an integer in the range of cpu_throttle_initial to 99");
1282 if (params
->has_announce_initial
&&
1283 params
->announce_initial
> 100000) {
1284 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1286 "is invalid, it must be less than 100000 ms");
1289 if (params
->has_announce_max
&&
1290 params
->announce_max
> 100000) {
1291 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1293 "is invalid, it must be less than 100000 ms");
1296 if (params
->has_announce_rounds
&&
1297 params
->announce_rounds
> 1000) {
1298 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1300 "is invalid, it must be in the range of 0 to 1000");
1303 if (params
->has_announce_step
&&
1304 (params
->announce_step
< 1 ||
1305 params
->announce_step
> 10000)) {
1306 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1308 "is invalid, it must be in the range of 1 to 10000 ms");
1314 static void migrate_params_test_apply(MigrateSetParameters
*params
,
1315 MigrationParameters
*dest
)
1317 *dest
= migrate_get_current()->parameters
;
1319 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1321 if (params
->has_compress_level
) {
1322 dest
->compress_level
= params
->compress_level
;
1325 if (params
->has_compress_threads
) {
1326 dest
->compress_threads
= params
->compress_threads
;
1329 if (params
->has_compress_wait_thread
) {
1330 dest
->compress_wait_thread
= params
->compress_wait_thread
;
1333 if (params
->has_decompress_threads
) {
1334 dest
->decompress_threads
= params
->decompress_threads
;
1337 if (params
->has_throttle_trigger_threshold
) {
1338 dest
->throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1341 if (params
->has_cpu_throttle_initial
) {
1342 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
1345 if (params
->has_cpu_throttle_increment
) {
1346 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
1349 if (params
->has_cpu_throttle_tailslow
) {
1350 dest
->cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1353 if (params
->has_tls_creds
) {
1354 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1355 dest
->tls_creds
= params
->tls_creds
->u
.s
;
1358 if (params
->has_tls_hostname
) {
1359 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1360 dest
->tls_hostname
= params
->tls_hostname
->u
.s
;
1363 if (params
->has_max_bandwidth
) {
1364 dest
->max_bandwidth
= params
->max_bandwidth
;
1367 if (params
->has_downtime_limit
) {
1368 dest
->downtime_limit
= params
->downtime_limit
;
1371 if (params
->has_x_checkpoint_delay
) {
1372 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
1375 if (params
->has_block_incremental
) {
1376 dest
->block_incremental
= params
->block_incremental
;
1378 if (params
->has_multifd_channels
) {
1379 dest
->multifd_channels
= params
->multifd_channels
;
1381 if (params
->has_multifd_compression
) {
1382 dest
->multifd_compression
= params
->multifd_compression
;
1384 if (params
->has_xbzrle_cache_size
) {
1385 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
1387 if (params
->has_max_postcopy_bandwidth
) {
1388 dest
->max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1390 if (params
->has_max_cpu_throttle
) {
1391 dest
->max_cpu_throttle
= params
->max_cpu_throttle
;
1393 if (params
->has_announce_initial
) {
1394 dest
->announce_initial
= params
->announce_initial
;
1396 if (params
->has_announce_max
) {
1397 dest
->announce_max
= params
->announce_max
;
1399 if (params
->has_announce_rounds
) {
1400 dest
->announce_rounds
= params
->announce_rounds
;
1402 if (params
->has_announce_step
) {
1403 dest
->announce_step
= params
->announce_step
;
1407 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
1409 MigrationState
*s
= migrate_get_current();
1411 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1413 if (params
->has_compress_level
) {
1414 s
->parameters
.compress_level
= params
->compress_level
;
1417 if (params
->has_compress_threads
) {
1418 s
->parameters
.compress_threads
= params
->compress_threads
;
1421 if (params
->has_compress_wait_thread
) {
1422 s
->parameters
.compress_wait_thread
= params
->compress_wait_thread
;
1425 if (params
->has_decompress_threads
) {
1426 s
->parameters
.decompress_threads
= params
->decompress_threads
;
1429 if (params
->has_throttle_trigger_threshold
) {
1430 s
->parameters
.throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1433 if (params
->has_cpu_throttle_initial
) {
1434 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
1437 if (params
->has_cpu_throttle_increment
) {
1438 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
1441 if (params
->has_cpu_throttle_tailslow
) {
1442 s
->parameters
.cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1445 if (params
->has_tls_creds
) {
1446 g_free(s
->parameters
.tls_creds
);
1447 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1448 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1451 if (params
->has_tls_hostname
) {
1452 g_free(s
->parameters
.tls_hostname
);
1453 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1454 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1457 if (params
->has_tls_authz
) {
1458 g_free(s
->parameters
.tls_authz
);
1459 assert(params
->tls_authz
->type
== QTYPE_QSTRING
);
1460 s
->parameters
.tls_authz
= g_strdup(params
->tls_authz
->u
.s
);
1463 if (params
->has_max_bandwidth
) {
1464 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1465 if (s
->to_dst_file
&& !migration_in_postcopy()) {
1466 qemu_file_set_rate_limit(s
->to_dst_file
,
1467 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1471 if (params
->has_downtime_limit
) {
1472 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1475 if (params
->has_x_checkpoint_delay
) {
1476 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1477 if (migration_in_colo_state()) {
1478 colo_checkpoint_notify(s
);
1482 if (params
->has_block_incremental
) {
1483 s
->parameters
.block_incremental
= params
->block_incremental
;
1485 if (params
->has_multifd_channels
) {
1486 s
->parameters
.multifd_channels
= params
->multifd_channels
;
1488 if (params
->has_multifd_compression
) {
1489 s
->parameters
.multifd_compression
= params
->multifd_compression
;
1491 if (params
->has_xbzrle_cache_size
) {
1492 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1493 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1495 if (params
->has_max_postcopy_bandwidth
) {
1496 s
->parameters
.max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1497 if (s
->to_dst_file
&& migration_in_postcopy()) {
1498 qemu_file_set_rate_limit(s
->to_dst_file
,
1499 s
->parameters
.max_postcopy_bandwidth
/ XFER_LIMIT_RATIO
);
1502 if (params
->has_max_cpu_throttle
) {
1503 s
->parameters
.max_cpu_throttle
= params
->max_cpu_throttle
;
1505 if (params
->has_announce_initial
) {
1506 s
->parameters
.announce_initial
= params
->announce_initial
;
1508 if (params
->has_announce_max
) {
1509 s
->parameters
.announce_max
= params
->announce_max
;
1511 if (params
->has_announce_rounds
) {
1512 s
->parameters
.announce_rounds
= params
->announce_rounds
;
1514 if (params
->has_announce_step
) {
1515 s
->parameters
.announce_step
= params
->announce_step
;
1519 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1521 MigrationParameters tmp
;
1523 /* TODO Rewrite "" to null instead */
1524 if (params
->has_tls_creds
1525 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1526 qobject_unref(params
->tls_creds
->u
.n
);
1527 params
->tls_creds
->type
= QTYPE_QSTRING
;
1528 params
->tls_creds
->u
.s
= strdup("");
1530 /* TODO Rewrite "" to null instead */
1531 if (params
->has_tls_hostname
1532 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1533 qobject_unref(params
->tls_hostname
->u
.n
);
1534 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1535 params
->tls_hostname
->u
.s
= strdup("");
1538 migrate_params_test_apply(params
, &tmp
);
1540 if (!migrate_params_check(&tmp
, errp
)) {
1541 /* Invalid parameter */
1545 migrate_params_apply(params
, errp
);
1549 void qmp_migrate_start_postcopy(Error
**errp
)
1551 MigrationState
*s
= migrate_get_current();
1553 if (!migrate_postcopy()) {
1554 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1555 " the start of migration");
1559 if (s
->state
== MIGRATION_STATUS_NONE
) {
1560 error_setg(errp
, "Postcopy must be started after migration has been"
1565 * we don't error if migration has finished since that would be racy
1566 * with issuing this command.
1568 atomic_set(&s
->start_postcopy
, true);
1571 /* shared migration helpers */
1573 void migrate_set_state(int *state
, int old_state
, int new_state
)
1575 assert(new_state
< MIGRATION_STATUS__MAX
);
1576 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1577 trace_migrate_set_state(MigrationStatus_str(new_state
));
1578 migrate_generate_event(new_state
);
1582 static MigrationCapabilityStatusList
*migrate_cap_add(
1583 MigrationCapabilityStatusList
*list
,
1584 MigrationCapability index
,
1587 MigrationCapabilityStatusList
*cap
;
1589 cap
= g_new0(MigrationCapabilityStatusList
, 1);
1590 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
1591 cap
->value
->capability
= index
;
1592 cap
->value
->state
= state
;
1598 void migrate_set_block_enabled(bool value
, Error
**errp
)
1600 MigrationCapabilityStatusList
*cap
;
1602 cap
= migrate_cap_add(NULL
, MIGRATION_CAPABILITY_BLOCK
, value
);
1603 qmp_migrate_set_capabilities(cap
, errp
);
1604 qapi_free_MigrationCapabilityStatusList(cap
);
1607 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1609 s
->parameters
.block_incremental
= value
;
1612 static void block_cleanup_parameters(MigrationState
*s
)
1614 if (s
->must_remove_block_options
) {
1615 /* setting to false can never fail */
1616 migrate_set_block_enabled(false, &error_abort
);
1617 migrate_set_block_incremental(s
, false);
1618 s
->must_remove_block_options
= false;
1622 static void migrate_fd_cleanup(MigrationState
*s
)
1624 qemu_bh_delete(s
->cleanup_bh
);
1625 s
->cleanup_bh
= NULL
;
1627 qemu_savevm_state_cleanup();
1629 if (s
->to_dst_file
) {
1632 trace_migrate_fd_cleanup();
1633 qemu_mutex_unlock_iothread();
1634 if (s
->migration_thread_running
) {
1635 qemu_thread_join(&s
->thread
);
1636 s
->migration_thread_running
= false;
1638 qemu_mutex_lock_iothread();
1640 multifd_save_cleanup();
1641 qemu_mutex_lock(&s
->qemu_file_lock
);
1642 tmp
= s
->to_dst_file
;
1643 s
->to_dst_file
= NULL
;
1644 qemu_mutex_unlock(&s
->qemu_file_lock
);
1646 * Close the file handle without the lock to make sure the
1647 * critical section won't block for long.
1652 assert(!migration_is_active(s
));
1654 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1655 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1656 MIGRATION_STATUS_CANCELLED
);
1660 /* It is used on info migrate. We can't free it */
1661 error_report_err(error_copy(s
->error
));
1663 notifier_list_notify(&migration_state_notifiers
, s
);
1664 block_cleanup_parameters(s
);
1667 static void migrate_fd_cleanup_schedule(MigrationState
*s
)
1670 * Ref the state for bh, because it may be called when
1671 * there're already no other refs
1673 object_ref(OBJECT(s
));
1674 qemu_bh_schedule(s
->cleanup_bh
);
1677 static void migrate_fd_cleanup_bh(void *opaque
)
1679 MigrationState
*s
= opaque
;
1680 migrate_fd_cleanup(s
);
1681 object_unref(OBJECT(s
));
1684 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1686 QEMU_LOCK_GUARD(&s
->error_mutex
);
1688 s
->error
= error_copy(error
);
1692 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1694 trace_migrate_fd_error(error_get_pretty(error
));
1695 assert(s
->to_dst_file
== NULL
);
1696 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1697 MIGRATION_STATUS_FAILED
);
1698 migrate_set_error(s
, error
);
1701 static void migrate_fd_cancel(MigrationState
*s
)
1704 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1705 trace_migrate_fd_cancel();
1707 if (s
->rp_state
.from_dst_file
) {
1708 /* shutdown the rp socket, so causing the rp thread to shutdown */
1709 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1713 old_state
= s
->state
;
1714 if (!migration_is_running(old_state
)) {
1717 /* If the migration is paused, kick it out of the pause */
1718 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1719 qemu_sem_post(&s
->pause_sem
);
1721 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1722 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1725 * If we're unlucky the migration code might be stuck somewhere in a
1726 * send/write while the network has failed and is waiting to timeout;
1727 * if we've got shutdown(2) available then we can force it to quit.
1728 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1729 * called in a bh, so there is no race against this cancel.
1731 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1732 qemu_file_shutdown(f
);
1734 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1735 Error
*local_err
= NULL
;
1737 bdrv_invalidate_cache_all(&local_err
);
1739 error_report_err(local_err
);
1741 s
->block_inactive
= false;
1746 void add_migration_state_change_notifier(Notifier
*notify
)
1748 notifier_list_add(&migration_state_notifiers
, notify
);
1751 void remove_migration_state_change_notifier(Notifier
*notify
)
1753 notifier_remove(notify
);
1756 bool migration_in_setup(MigrationState
*s
)
1758 return s
->state
== MIGRATION_STATUS_SETUP
;
1761 bool migration_has_finished(MigrationState
*s
)
1763 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1766 bool migration_has_failed(MigrationState
*s
)
1768 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1769 s
->state
== MIGRATION_STATUS_FAILED
);
1772 bool migration_in_postcopy(void)
1774 MigrationState
*s
= migrate_get_current();
1777 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1778 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1779 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1786 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1788 return migration_in_postcopy() && s
->postcopy_after_devices
;
1791 bool migration_in_incoming_postcopy(void)
1793 PostcopyState ps
= postcopy_state_get();
1795 return ps
>= POSTCOPY_INCOMING_DISCARD
&& ps
< POSTCOPY_INCOMING_END
;
1798 bool migration_is_idle(void)
1800 MigrationState
*s
= current_migration
;
1807 case MIGRATION_STATUS_NONE
:
1808 case MIGRATION_STATUS_CANCELLED
:
1809 case MIGRATION_STATUS_COMPLETED
:
1810 case MIGRATION_STATUS_FAILED
:
1812 case MIGRATION_STATUS_SETUP
:
1813 case MIGRATION_STATUS_CANCELLING
:
1814 case MIGRATION_STATUS_ACTIVE
:
1815 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1816 case MIGRATION_STATUS_COLO
:
1817 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1818 case MIGRATION_STATUS_DEVICE
:
1819 case MIGRATION_STATUS_WAIT_UNPLUG
:
1821 case MIGRATION_STATUS__MAX
:
1822 g_assert_not_reached();
1828 bool migration_is_active(MigrationState
*s
)
1830 return (s
->state
== MIGRATION_STATUS_ACTIVE
||
1831 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1834 void migrate_init(MigrationState
*s
)
1837 * Reinitialise all migration state, except
1838 * parameters/capabilities that the user set, and
1842 s
->to_dst_file
= NULL
;
1843 s
->state
= MIGRATION_STATUS_NONE
;
1844 s
->rp_state
.from_dst_file
= NULL
;
1845 s
->rp_state
.error
= false;
1847 s
->pages_per_second
= 0.0;
1849 s
->expected_downtime
= 0;
1851 s
->start_postcopy
= false;
1852 s
->postcopy_after_devices
= false;
1853 s
->migration_thread_running
= false;
1854 error_free(s
->error
);
1857 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1859 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1861 s
->vm_was_running
= false;
1862 s
->iteration_initial_bytes
= 0;
1863 s
->threshold_size
= 0;
1866 static GSList
*migration_blockers
;
1868 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1870 if (only_migratable
) {
1871 error_propagate_prepend(errp
, error_copy(reason
),
1872 "disallowing migration blocker "
1873 "(--only-migratable) for: ");
1877 if (migration_is_idle()) {
1878 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1882 error_propagate_prepend(errp
, error_copy(reason
),
1883 "disallowing migration blocker "
1884 "(migration in progress) for: ");
1888 void migrate_del_blocker(Error
*reason
)
1890 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1893 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1895 Error
*local_err
= NULL
;
1896 static bool once
= true;
1898 if (!deferred_incoming
) {
1899 error_setg(errp
, "For use with '-incoming defer'");
1903 error_setg(errp
, "The incoming migration has already been started");
1907 qemu_start_incoming_migration(uri
, &local_err
);
1910 error_propagate(errp
, local_err
);
1917 void qmp_migrate_recover(const char *uri
, Error
**errp
)
1919 MigrationIncomingState
*mis
= migration_incoming_get_current();
1921 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1922 error_setg(errp
, "Migrate recover can only be run "
1923 "when postcopy is paused.");
1927 if (atomic_cmpxchg(&mis
->postcopy_recover_triggered
,
1928 false, true) == true) {
1929 error_setg(errp
, "Migrate recovery is triggered already");
1934 * Note that this call will never start a real migration; it will
1935 * only re-setup the migration stream and poke existing migration
1936 * to continue using that newly established channel.
1938 qemu_start_incoming_migration(uri
, errp
);
1941 void qmp_migrate_pause(Error
**errp
)
1943 MigrationState
*ms
= migrate_get_current();
1944 MigrationIncomingState
*mis
= migration_incoming_get_current();
1947 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1948 /* Source side, during postcopy */
1949 qemu_mutex_lock(&ms
->qemu_file_lock
);
1950 ret
= qemu_file_shutdown(ms
->to_dst_file
);
1951 qemu_mutex_unlock(&ms
->qemu_file_lock
);
1953 error_setg(errp
, "Failed to pause source migration");
1958 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1959 ret
= qemu_file_shutdown(mis
->from_src_file
);
1961 error_setg(errp
, "Failed to pause destination migration");
1966 error_setg(errp
, "migrate-pause is currently only supported "
1967 "during postcopy-active state");
1970 bool migration_is_blocked(Error
**errp
)
1972 if (qemu_savevm_state_blocked(errp
)) {
1976 if (migration_blockers
) {
1977 error_propagate(errp
, error_copy(migration_blockers
->data
));
1984 /* Returns true if continue to migrate, or false if error detected */
1985 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
1986 bool resume
, Error
**errp
)
1988 Error
*local_err
= NULL
;
1991 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1992 error_setg(errp
, "Cannot resume if there is no "
1993 "paused migration");
1998 * Postcopy recovery won't work well with release-ram
1999 * capability since release-ram will drop the page buffer as
2000 * long as the page is put into the send buffer. So if there
2001 * is a network failure happened, any page buffers that have
2002 * not yet reached the destination VM but have already been
2003 * sent from the source VM will be lost forever. Let's refuse
2004 * the client from resuming such a postcopy migration.
2005 * Luckily release-ram was designed to only be used when src
2006 * and destination VMs are on the same host, so it should be
2009 if (migrate_release_ram()) {
2010 error_setg(errp
, "Postcopy recovery cannot work "
2011 "when release-ram capability is set");
2015 /* This is a resume, skip init status */
2019 if (migration_is_running(s
->state
)) {
2020 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
2024 if (runstate_check(RUN_STATE_INMIGRATE
)) {
2025 error_setg(errp
, "Guest is waiting for an incoming migration");
2029 if (migration_is_blocked(errp
)) {
2033 if (blk
|| blk_inc
) {
2034 if (migrate_use_block() || migrate_use_block_incremental()) {
2035 error_setg(errp
, "Command options are incompatible with "
2036 "current migration capabilities");
2039 migrate_set_block_enabled(true, &local_err
);
2041 error_propagate(errp
, local_err
);
2044 s
->must_remove_block_options
= true;
2048 migrate_set_block_incremental(s
, true);
2053 * set ram_counters memory to zero for a
2056 memset(&ram_counters
, 0, sizeof(ram_counters
));
2061 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
2062 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
2063 bool has_resume
, bool resume
, Error
**errp
)
2065 Error
*local_err
= NULL
;
2066 MigrationState
*s
= migrate_get_current();
2069 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
2070 has_resume
&& resume
, errp
)) {
2071 /* Error detected, put into errp */
2075 if (strstart(uri
, "tcp:", &p
)) {
2076 tcp_start_outgoing_migration(s
, p
, &local_err
);
2078 } else if (strstart(uri
, "rdma:", &p
)) {
2079 rdma_start_outgoing_migration(s
, p
, &local_err
);
2081 } else if (strstart(uri
, "exec:", &p
)) {
2082 exec_start_outgoing_migration(s
, p
, &local_err
);
2083 } else if (strstart(uri
, "unix:", &p
)) {
2084 unix_start_outgoing_migration(s
, p
, &local_err
);
2085 } else if (strstart(uri
, "fd:", &p
)) {
2086 fd_start_outgoing_migration(s
, p
, &local_err
);
2088 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
2089 "a valid migration protocol");
2090 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2091 MIGRATION_STATUS_FAILED
);
2092 block_cleanup_parameters(s
);
2097 migrate_fd_error(s
, local_err
);
2098 error_propagate(errp
, local_err
);
2103 void qmp_migrate_cancel(Error
**errp
)
2105 migrate_fd_cancel(migrate_get_current());
2108 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
2110 MigrationState
*s
= migrate_get_current();
2111 if (s
->state
!= state
) {
2112 error_setg(errp
, "Migration not in expected state: %s",
2113 MigrationStatus_str(s
->state
));
2116 qemu_sem_post(&s
->pause_sem
);
2119 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
2121 MigrateSetParameters p
= {
2122 .has_xbzrle_cache_size
= true,
2123 .xbzrle_cache_size
= value
,
2126 qmp_migrate_set_parameters(&p
, errp
);
2129 int64_t qmp_query_migrate_cache_size(Error
**errp
)
2131 return migrate_xbzrle_cache_size();
2134 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
2136 MigrateSetParameters p
= {
2137 .has_max_bandwidth
= true,
2138 .max_bandwidth
= value
,
2141 qmp_migrate_set_parameters(&p
, errp
);
2144 void qmp_migrate_set_downtime(double value
, Error
**errp
)
2146 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
2147 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
2149 "an integer in the range of 0 to "
2150 stringify(MAX_MIGRATE_DOWNTIME_SECONDS
)" seconds");
2154 value
*= 1000; /* Convert to milliseconds */
2156 MigrateSetParameters p
= {
2157 .has_downtime_limit
= true,
2158 .downtime_limit
= (int64_t)value
,
2161 qmp_migrate_set_parameters(&p
, errp
);
2164 bool migrate_release_ram(void)
2168 s
= migrate_get_current();
2170 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
2173 bool migrate_postcopy_ram(void)
2177 s
= migrate_get_current();
2179 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
2182 bool migrate_postcopy(void)
2184 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2187 bool migrate_auto_converge(void)
2191 s
= migrate_get_current();
2193 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
2196 bool migrate_zero_blocks(void)
2200 s
= migrate_get_current();
2202 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
2205 bool migrate_postcopy_blocktime(void)
2209 s
= migrate_get_current();
2211 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
2214 bool migrate_use_compression(void)
2218 s
= migrate_get_current();
2220 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
2223 int migrate_compress_level(void)
2227 s
= migrate_get_current();
2229 return s
->parameters
.compress_level
;
2232 int migrate_compress_threads(void)
2236 s
= migrate_get_current();
2238 return s
->parameters
.compress_threads
;
2241 int migrate_compress_wait_thread(void)
2245 s
= migrate_get_current();
2247 return s
->parameters
.compress_wait_thread
;
2250 int migrate_decompress_threads(void)
2254 s
= migrate_get_current();
2256 return s
->parameters
.decompress_threads
;
2259 bool migrate_dirty_bitmaps(void)
2263 s
= migrate_get_current();
2265 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
2268 bool migrate_ignore_shared(void)
2272 s
= migrate_get_current();
2274 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
];
2277 bool migrate_validate_uuid(void)
2281 s
= migrate_get_current();
2283 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_VALIDATE_UUID
];
2286 bool migrate_use_events(void)
2290 s
= migrate_get_current();
2292 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
2295 bool migrate_use_multifd(void)
2299 s
= migrate_get_current();
2301 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_MULTIFD
];
2304 bool migrate_pause_before_switchover(void)
2308 s
= migrate_get_current();
2310 return s
->enabled_capabilities
[
2311 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
2314 int migrate_multifd_channels(void)
2318 s
= migrate_get_current();
2320 return s
->parameters
.multifd_channels
;
2323 MultiFDCompression
migrate_multifd_compression(void)
2327 s
= migrate_get_current();
2329 return s
->parameters
.multifd_compression
;
2332 int migrate_multifd_zlib_level(void)
2336 s
= migrate_get_current();
2338 return s
->parameters
.multifd_zlib_level
;
2341 int migrate_multifd_zstd_level(void)
2345 s
= migrate_get_current();
2347 return s
->parameters
.multifd_zstd_level
;
2350 int migrate_use_xbzrle(void)
2354 s
= migrate_get_current();
2356 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
2359 int64_t migrate_xbzrle_cache_size(void)
2363 s
= migrate_get_current();
2365 return s
->parameters
.xbzrle_cache_size
;
2368 static int64_t migrate_max_postcopy_bandwidth(void)
2372 s
= migrate_get_current();
2374 return s
->parameters
.max_postcopy_bandwidth
;
2377 bool migrate_use_block(void)
2381 s
= migrate_get_current();
2383 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
2386 bool migrate_use_return_path(void)
2390 s
= migrate_get_current();
2392 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
2395 bool migrate_use_block_incremental(void)
2399 s
= migrate_get_current();
2401 return s
->parameters
.block_incremental
;
2404 /* migration thread support */
2406 * Something bad happened to the RP stream, mark an error
2407 * The caller shall print or trace something to indicate why
2409 static void mark_source_rp_bad(MigrationState
*s
)
2411 s
->rp_state
.error
= true;
2414 static struct rp_cmd_args
{
2415 ssize_t len
; /* -1 = variable */
2418 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
2419 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
2420 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
2421 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
2422 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
2423 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
2424 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
2425 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
2429 * Process a request for pages received on the return path,
2430 * We're allowed to send more than requested (e.g. to round to our page size)
2431 * and we don't need to send pages that have already been sent.
2433 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
2434 ram_addr_t start
, size_t len
)
2436 long our_host_ps
= qemu_real_host_page_size
;
2438 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
2441 * Since we currently insist on matching page sizes, just sanity check
2442 * we're being asked for whole host pages.
2444 if (start
& (our_host_ps
-1) ||
2445 (len
& (our_host_ps
-1))) {
2446 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2447 " len: %zd", __func__
, start
, len
);
2448 mark_source_rp_bad(ms
);
2452 if (ram_save_queue_pages(rbname
, start
, len
)) {
2453 mark_source_rp_bad(ms
);
2457 /* Return true to retry, false to quit */
2458 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
2460 trace_postcopy_pause_return_path();
2462 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
2464 trace_postcopy_pause_return_path_continued();
2469 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
2471 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
2474 error_report("%s: invalid block name '%s'", __func__
, block_name
);
2478 /* Fetch the received bitmap and refresh the dirty bitmap */
2479 return ram_dirty_bitmap_reload(s
, block
);
2482 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
2484 trace_source_return_path_thread_resume_ack(value
);
2486 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
2487 error_report("%s: illegal resume_ack value %"PRIu32
,
2492 /* Now both sides are active. */
2493 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2494 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2496 /* Notify send thread that time to continue send pages */
2497 qemu_sem_post(&s
->rp_state
.rp_sem
);
2503 * Handles messages sent on the return path towards the source VM
2506 static void *source_return_path_thread(void *opaque
)
2508 MigrationState
*ms
= opaque
;
2509 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
2510 uint16_t header_len
, header_type
;
2512 uint32_t tmp32
, sibling_error
;
2513 ram_addr_t start
= 0; /* =0 to silence warning */
2514 size_t len
= 0, expected_len
;
2517 trace_source_return_path_thread_entry();
2518 rcu_register_thread();
2521 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
2522 migration_is_setup_or_active(ms
->state
)) {
2523 trace_source_return_path_thread_loop_top();
2524 header_type
= qemu_get_be16(rp
);
2525 header_len
= qemu_get_be16(rp
);
2527 if (qemu_file_get_error(rp
)) {
2528 mark_source_rp_bad(ms
);
2532 if (header_type
>= MIG_RP_MSG_MAX
||
2533 header_type
== MIG_RP_MSG_INVALID
) {
2534 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2535 header_type
, header_len
);
2536 mark_source_rp_bad(ms
);
2540 if ((rp_cmd_args
[header_type
].len
!= -1 &&
2541 header_len
!= rp_cmd_args
[header_type
].len
) ||
2542 header_len
> sizeof(buf
)) {
2543 error_report("RP: Received '%s' message (0x%04x) with"
2544 "incorrect length %d expecting %zu",
2545 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
2546 (size_t)rp_cmd_args
[header_type
].len
);
2547 mark_source_rp_bad(ms
);
2551 /* We know we've got a valid header by this point */
2552 res
= qemu_get_buffer(rp
, buf
, header_len
);
2553 if (res
!= header_len
) {
2554 error_report("RP: Failed reading data for message 0x%04x"
2555 " read %d expected %d",
2556 header_type
, res
, header_len
);
2557 mark_source_rp_bad(ms
);
2561 /* OK, we have the message and the data */
2562 switch (header_type
) {
2563 case MIG_RP_MSG_SHUT
:
2564 sibling_error
= ldl_be_p(buf
);
2565 trace_source_return_path_thread_shut(sibling_error
);
2566 if (sibling_error
) {
2567 error_report("RP: Sibling indicated error %d", sibling_error
);
2568 mark_source_rp_bad(ms
);
2571 * We'll let the main thread deal with closing the RP
2572 * we could do a shutdown(2) on it, but we're the only user
2573 * anyway, so there's nothing gained.
2577 case MIG_RP_MSG_PONG
:
2578 tmp32
= ldl_be_p(buf
);
2579 trace_source_return_path_thread_pong(tmp32
);
2582 case MIG_RP_MSG_REQ_PAGES
:
2583 start
= ldq_be_p(buf
);
2584 len
= ldl_be_p(buf
+ 8);
2585 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
2588 case MIG_RP_MSG_REQ_PAGES_ID
:
2589 expected_len
= 12 + 1; /* header + termination */
2591 if (header_len
>= expected_len
) {
2592 start
= ldq_be_p(buf
);
2593 len
= ldl_be_p(buf
+ 8);
2594 /* Now we expect an idstr */
2595 tmp32
= buf
[12]; /* Length of the following idstr */
2596 buf
[13 + tmp32
] = '\0';
2597 expected_len
+= tmp32
;
2599 if (header_len
!= expected_len
) {
2600 error_report("RP: Req_Page_id with length %d expecting %zd",
2601 header_len
, expected_len
);
2602 mark_source_rp_bad(ms
);
2605 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
2608 case MIG_RP_MSG_RECV_BITMAP
:
2609 if (header_len
< 1) {
2610 error_report("%s: missing block name", __func__
);
2611 mark_source_rp_bad(ms
);
2614 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2615 buf
[buf
[0] + 1] = '\0';
2616 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
2617 mark_source_rp_bad(ms
);
2622 case MIG_RP_MSG_RESUME_ACK
:
2623 tmp32
= ldl_be_p(buf
);
2624 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
2625 mark_source_rp_bad(ms
);
2636 res
= qemu_file_get_error(rp
);
2638 if (res
== -EIO
&& migration_in_postcopy()) {
2640 * Maybe there is something we can do: it looks like a
2641 * network down issue, and we pause for a recovery.
2643 if (postcopy_pause_return_path_thread(ms
)) {
2644 /* Reload rp, reset the rest */
2645 if (rp
!= ms
->rp_state
.from_dst_file
) {
2647 rp
= ms
->rp_state
.from_dst_file
;
2649 ms
->rp_state
.error
= false;
2654 trace_source_return_path_thread_bad_end();
2655 mark_source_rp_bad(ms
);
2658 trace_source_return_path_thread_end();
2659 ms
->rp_state
.from_dst_file
= NULL
;
2661 rcu_unregister_thread();
2665 static int open_return_path_on_source(MigrationState
*ms
,
2669 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
2670 if (!ms
->rp_state
.from_dst_file
) {
2674 trace_open_return_path_on_source();
2676 if (!create_thread
) {
2681 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
2682 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
2684 trace_open_return_path_on_source_continue();
2689 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2690 static int await_return_path_close_on_source(MigrationState
*ms
)
2693 * If this is a normal exit then the destination will send a SHUT and the
2694 * rp_thread will exit, however if there's an error we need to cause
2697 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
2699 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2700 * waiting for the destination.
2702 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
2703 mark_source_rp_bad(ms
);
2705 trace_await_return_path_close_on_source_joining();
2706 qemu_thread_join(&ms
->rp_state
.rp_thread
);
2707 trace_await_return_path_close_on_source_close();
2708 return ms
->rp_state
.error
;
2712 * Switch from normal iteration to postcopy
2713 * Returns non-0 on error
2715 static int postcopy_start(MigrationState
*ms
)
2718 QIOChannelBuffer
*bioc
;
2720 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2721 int64_t bandwidth
= migrate_max_postcopy_bandwidth();
2722 bool restart_block
= false;
2723 int cur_state
= MIGRATION_STATUS_ACTIVE
;
2724 if (!migrate_pause_before_switchover()) {
2725 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
2726 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2729 trace_postcopy_start();
2730 qemu_mutex_lock_iothread();
2731 trace_postcopy_start_set_run();
2733 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2734 global_state_store();
2735 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2740 ret
= migration_maybe_pause(ms
, &cur_state
,
2741 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2746 ret
= bdrv_inactivate_all();
2750 restart_block
= true;
2753 * Cause any non-postcopiable, but iterative devices to
2754 * send out their final data.
2756 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2759 * in Finish migrate and with the io-lock held everything should
2760 * be quiet, but we've potentially still got dirty pages and we
2761 * need to tell the destination to throw any pages it's already received
2764 if (migrate_postcopy_ram()) {
2765 if (ram_postcopy_send_discard_bitmap(ms
)) {
2766 error_report("postcopy send discard bitmap failed");
2772 * send rest of state - note things that are doing postcopy
2773 * will notice we're in POSTCOPY_ACTIVE and not actually
2774 * wrap their state up here
2776 /* 0 max-postcopy-bandwidth means unlimited */
2778 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
2780 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
2782 if (migrate_postcopy_ram()) {
2783 /* Ping just for debugging, helps line traces up */
2784 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
2788 * While loading the device state we may trigger page transfer
2789 * requests and the fd must be free to process those, and thus
2790 * the destination must read the whole device state off the fd before
2791 * it starts processing it. Unfortunately the ad-hoc migration format
2792 * doesn't allow the destination to know the size to read without fully
2793 * parsing it through each devices load-state code (especially the open
2794 * coded devices that use get/put).
2795 * So we wrap the device state up in a package with a length at the start;
2796 * to do this we use a qemu_buf to hold the whole of the device state.
2798 bioc
= qio_channel_buffer_new(4096);
2799 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
2800 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
2801 object_unref(OBJECT(bioc
));
2804 * Make sure the receiver can get incoming pages before we send the rest
2807 qemu_savevm_send_postcopy_listen(fb
);
2809 qemu_savevm_state_complete_precopy(fb
, false, false);
2810 if (migrate_postcopy_ram()) {
2811 qemu_savevm_send_ping(fb
, 3);
2814 qemu_savevm_send_postcopy_run(fb
);
2816 /* <><> end of stuff going into the package */
2818 /* Last point of recovery; as soon as we send the package the destination
2819 * can open devices and potentially start running.
2820 * Lets just check again we've not got any errors.
2822 ret
= qemu_file_get_error(ms
->to_dst_file
);
2824 error_report("postcopy_start: Migration stream errored (pre package)");
2828 restart_block
= false;
2830 /* Now send that blob */
2831 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2836 /* Send a notify to give a chance for anything that needs to happen
2837 * at the transition to postcopy and after the device state; in particular
2838 * spice needs to trigger a transition now
2840 ms
->postcopy_after_devices
= true;
2841 notifier_list_notify(&migration_state_notifiers
, ms
);
2843 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2845 qemu_mutex_unlock_iothread();
2847 if (migrate_postcopy_ram()) {
2849 * Although this ping is just for debug, it could potentially be
2850 * used for getting a better measurement of downtime at the source.
2852 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2855 if (migrate_release_ram()) {
2856 ram_postcopy_migrated_memory_release(ms
);
2859 ret
= qemu_file_get_error(ms
->to_dst_file
);
2861 error_report("postcopy_start: Migration stream errored");
2862 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2863 MIGRATION_STATUS_FAILED
);
2871 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2872 MIGRATION_STATUS_FAILED
);
2873 if (restart_block
) {
2874 /* A failure happened early enough that we know the destination hasn't
2875 * accessed block devices, so we're safe to recover.
2877 Error
*local_err
= NULL
;
2879 bdrv_invalidate_cache_all(&local_err
);
2881 error_report_err(local_err
);
2884 qemu_mutex_unlock_iothread();
2889 * migration_maybe_pause: Pause if required to by
2890 * migrate_pause_before_switchover called with the iothread locked
2891 * Returns: 0 on success
2893 static int migration_maybe_pause(MigrationState
*s
,
2894 int *current_active_state
,
2897 if (!migrate_pause_before_switchover()) {
2901 /* Since leaving this state is not atomic with posting the semaphore
2902 * it's possible that someone could have issued multiple migrate_continue
2903 * and the semaphore is incorrectly positive at this point;
2904 * the docs say it's undefined to reinit a semaphore that's already
2905 * init'd, so use timedwait to eat up any existing posts.
2907 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2908 /* This block intentionally left blank */
2912 * If the migration is cancelled when it is in the completion phase,
2913 * the migration state is set to MIGRATION_STATUS_CANCELLING.
2914 * So we don't need to wait a semaphore, otherwise we would always
2915 * wait for the 'pause_sem' semaphore.
2917 if (s
->state
!= MIGRATION_STATUS_CANCELLING
) {
2918 qemu_mutex_unlock_iothread();
2919 migrate_set_state(&s
->state
, *current_active_state
,
2920 MIGRATION_STATUS_PRE_SWITCHOVER
);
2921 qemu_sem_wait(&s
->pause_sem
);
2922 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2924 *current_active_state
= new_state
;
2925 qemu_mutex_lock_iothread();
2928 return s
->state
== new_state
? 0 : -EINVAL
;
2932 * migration_completion: Used by migration_thread when there's not much left.
2933 * The caller 'breaks' the loop when this returns.
2935 * @s: Current migration state
2937 static void migration_completion(MigrationState
*s
)
2940 int current_active_state
= s
->state
;
2942 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2943 qemu_mutex_lock_iothread();
2944 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2945 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2946 s
->vm_was_running
= runstate_is_running();
2947 ret
= global_state_store();
2950 bool inactivate
= !migrate_colo_enabled();
2951 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2953 ret
= migration_maybe_pause(s
, ¤t_active_state
,
2954 MIGRATION_STATUS_DEVICE
);
2957 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
2958 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
2961 if (inactivate
&& ret
>= 0) {
2962 s
->block_inactive
= true;
2965 qemu_mutex_unlock_iothread();
2970 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2971 trace_migration_completion_postcopy_end();
2973 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
2974 trace_migration_completion_postcopy_end_after_complete();
2978 * If rp was opened we must clean up the thread before
2979 * cleaning everything else up (since if there are no failures
2980 * it will wait for the destination to send it's status in
2983 if (s
->rp_state
.from_dst_file
) {
2985 trace_migration_return_path_end_before();
2986 rp_error
= await_return_path_close_on_source(s
);
2987 trace_migration_return_path_end_after(rp_error
);
2989 goto fail_invalidate
;
2993 if (qemu_file_get_error(s
->to_dst_file
)) {
2994 trace_migration_completion_file_err();
2995 goto fail_invalidate
;
2998 if (!migrate_colo_enabled()) {
2999 migrate_set_state(&s
->state
, current_active_state
,
3000 MIGRATION_STATUS_COMPLETED
);
3006 /* If not doing postcopy, vm_start() will be called: let's regain
3007 * control on images.
3009 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
3010 s
->state
== MIGRATION_STATUS_DEVICE
) {
3011 Error
*local_err
= NULL
;
3013 qemu_mutex_lock_iothread();
3014 bdrv_invalidate_cache_all(&local_err
);
3016 error_report_err(local_err
);
3018 s
->block_inactive
= false;
3020 qemu_mutex_unlock_iothread();
3024 migrate_set_state(&s
->state
, current_active_state
,
3025 MIGRATION_STATUS_FAILED
);
3028 bool migrate_colo_enabled(void)
3030 MigrationState
*s
= migrate_get_current();
3031 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
3034 typedef enum MigThrError
{
3035 /* No error detected */
3036 MIG_THR_ERR_NONE
= 0,
3037 /* Detected error, but resumed successfully */
3038 MIG_THR_ERR_RECOVERED
= 1,
3039 /* Detected fatal error, need to exit */
3040 MIG_THR_ERR_FATAL
= 2,
3043 static int postcopy_resume_handshake(MigrationState
*s
)
3045 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
3047 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3048 qemu_sem_wait(&s
->rp_state
.rp_sem
);
3051 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3058 /* Return zero if success, or <0 for error */
3059 static int postcopy_do_resume(MigrationState
*s
)
3064 * Call all the resume_prepare() hooks, so that modules can be
3065 * ready for the migration resume.
3067 ret
= qemu_savevm_state_resume_prepare(s
);
3069 error_report("%s: resume_prepare() failure detected: %d",
3075 * Last handshake with destination on the resume (destination will
3076 * switch to postcopy-active afterwards)
3078 ret
= postcopy_resume_handshake(s
);
3080 error_report("%s: handshake failed: %d", __func__
, ret
);
3088 * We don't return until we are in a safe state to continue current
3089 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
3090 * MIG_THR_ERR_FATAL if unrecovery failure happened.
3092 static MigThrError
postcopy_pause(MigrationState
*s
)
3094 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3099 migrate_set_state(&s
->state
, s
->state
,
3100 MIGRATION_STATUS_POSTCOPY_PAUSED
);
3102 /* Current channel is possibly broken. Release it. */
3103 assert(s
->to_dst_file
);
3104 qemu_mutex_lock(&s
->qemu_file_lock
);
3105 file
= s
->to_dst_file
;
3106 s
->to_dst_file
= NULL
;
3107 qemu_mutex_unlock(&s
->qemu_file_lock
);
3109 qemu_file_shutdown(file
);
3112 error_report("Detected IO failure for postcopy. "
3113 "Migration paused.");
3116 * We wait until things fixed up. Then someone will setup the
3117 * status back for us.
3119 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
3120 qemu_sem_wait(&s
->postcopy_pause_sem
);
3123 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3124 /* Woken up by a recover procedure. Give it a shot */
3127 * Firstly, let's wake up the return path now, with a new
3128 * return path channel.
3130 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
3132 /* Do the resume logic */
3133 if (postcopy_do_resume(s
) == 0) {
3134 /* Let's continue! */
3135 trace_postcopy_pause_continued();
3136 return MIG_THR_ERR_RECOVERED
;
3139 * Something wrong happened during the recovery, let's
3140 * pause again. Pause is always better than throwing
3146 /* This is not right... Time to quit. */
3147 return MIG_THR_ERR_FATAL
;
3152 static MigThrError
migration_detect_error(MigrationState
*s
)
3155 int state
= s
->state
;
3156 Error
*local_error
= NULL
;
3158 if (state
== MIGRATION_STATUS_CANCELLING
||
3159 state
== MIGRATION_STATUS_CANCELLED
) {
3160 /* End the migration, but don't set the state to failed */
3161 return MIG_THR_ERR_FATAL
;
3164 /* Try to detect any file errors */
3165 ret
= qemu_file_get_error_obj(s
->to_dst_file
, &local_error
);
3167 /* Everything is fine */
3168 assert(!local_error
);
3169 return MIG_THR_ERR_NONE
;
3173 migrate_set_error(s
, local_error
);
3174 error_free(local_error
);
3177 if (state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
== -EIO
) {
3179 * For postcopy, we allow the network to be down for a
3180 * while. After that, it can be continued by a
3183 return postcopy_pause(s
);
3186 * For precopy (or postcopy with error outside IO), we fail
3189 migrate_set_state(&s
->state
, state
, MIGRATION_STATUS_FAILED
);
3190 trace_migration_thread_file_err();
3192 /* Time to stop the migration, now. */
3193 return MIG_THR_ERR_FATAL
;
3197 /* How many bytes have we transferred since the beginning of the migration */
3198 static uint64_t migration_total_bytes(MigrationState
*s
)
3200 return qemu_ftell(s
->to_dst_file
) + ram_counters
.multifd_bytes
;
3203 static void migration_calculate_complete(MigrationState
*s
)
3205 uint64_t bytes
= migration_total_bytes(s
);
3206 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3207 int64_t transfer_time
;
3209 s
->total_time
= end_time
- s
->start_time
;
3212 * It's still not set, so we are precopy migration. For
3213 * postcopy, downtime is calculated during postcopy_start().
3215 s
->downtime
= end_time
- s
->downtime_start
;
3218 transfer_time
= s
->total_time
- s
->setup_time
;
3219 if (transfer_time
) {
3220 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
3224 static void update_iteration_initial_status(MigrationState
*s
)
3227 * Update these three fields at the same time to avoid mismatch info lead
3228 * wrong speed calculation.
3230 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3231 s
->iteration_initial_bytes
= migration_total_bytes(s
);
3232 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
3235 static void migration_update_counters(MigrationState
*s
,
3236 int64_t current_time
)
3238 uint64_t transferred
, transferred_pages
, time_spent
;
3239 uint64_t current_bytes
; /* bytes transferred since the beginning */
3242 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
3246 current_bytes
= migration_total_bytes(s
);
3247 transferred
= current_bytes
- s
->iteration_initial_bytes
;
3248 time_spent
= current_time
- s
->iteration_start_time
;
3249 bandwidth
= (double)transferred
/ time_spent
;
3250 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
3252 s
->mbps
= (((double) transferred
* 8.0) /
3253 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
3255 transferred_pages
= ram_get_total_transferred_pages() -
3256 s
->iteration_initial_pages
;
3257 s
->pages_per_second
= (double) transferred_pages
/
3258 (((double) time_spent
/ 1000.0));
3261 * if we haven't sent anything, we don't want to
3262 * recalculate. 10000 is a small enough number for our purposes
3264 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
3265 s
->expected_downtime
= ram_counters
.remaining
/ bandwidth
;
3268 qemu_file_reset_rate_limit(s
->to_dst_file
);
3270 update_iteration_initial_status(s
);
3272 trace_migrate_transferred(transferred
, time_spent
,
3273 bandwidth
, s
->threshold_size
);
3276 /* Migration thread iteration status */
3278 MIG_ITERATE_RESUME
, /* Resume current iteration */
3279 MIG_ITERATE_SKIP
, /* Skip current iteration */
3280 MIG_ITERATE_BREAK
, /* Break the loop */
3284 * Return true if continue to the next iteration directly, false
3287 static MigIterateState
migration_iteration_run(MigrationState
*s
)
3289 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
3290 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
3292 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
3293 &pend_compat
, &pend_post
);
3294 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
3296 trace_migrate_pending(pending_size
, s
->threshold_size
,
3297 pend_pre
, pend_compat
, pend_post
);
3299 if (pending_size
&& pending_size
>= s
->threshold_size
) {
3300 /* Still a significant amount to transfer */
3301 if (!in_postcopy
&& pend_pre
<= s
->threshold_size
&&
3302 atomic_read(&s
->start_postcopy
)) {
3303 if (postcopy_start(s
)) {
3304 error_report("%s: postcopy failed to start", __func__
);
3306 return MIG_ITERATE_SKIP
;
3308 /* Just another iteration step */
3309 qemu_savevm_state_iterate(s
->to_dst_file
, in_postcopy
);
3311 trace_migration_thread_low_pending(pending_size
);
3312 migration_completion(s
);
3313 return MIG_ITERATE_BREAK
;
3316 return MIG_ITERATE_RESUME
;
3319 static void migration_iteration_finish(MigrationState
*s
)
3321 /* If we enabled cpu throttling for auto-converge, turn it off. */
3322 cpu_throttle_stop();
3324 qemu_mutex_lock_iothread();
3326 case MIGRATION_STATUS_COMPLETED
:
3327 migration_calculate_complete(s
);
3328 runstate_set(RUN_STATE_POSTMIGRATE
);
3331 case MIGRATION_STATUS_ACTIVE
:
3333 * We should really assert here, but since it's during
3334 * migration, let's try to reduce the usage of assertions.
3336 if (!migrate_colo_enabled()) {
3337 error_report("%s: critical error: calling COLO code without "
3338 "COLO enabled", __func__
);
3340 migrate_start_colo_process(s
);
3342 * Fixme: we will run VM in COLO no matter its old running state.
3343 * After exited COLO, we will keep running.
3345 s
->vm_was_running
= true;
3347 case MIGRATION_STATUS_FAILED
:
3348 case MIGRATION_STATUS_CANCELLED
:
3349 case MIGRATION_STATUS_CANCELLING
:
3350 if (s
->vm_was_running
) {
3353 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
3354 runstate_set(RUN_STATE_POSTMIGRATE
);
3360 /* Should not reach here, but if so, forgive the VM. */
3361 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3364 migrate_fd_cleanup_schedule(s
);
3365 qemu_mutex_unlock_iothread();
3368 void migration_make_urgent_request(void)
3370 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
3373 void migration_consume_urgent_request(void)
3375 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
3378 /* Returns true if the rate limiting was broken by an urgent request */
3379 bool migration_rate_limit(void)
3381 int64_t now
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3382 MigrationState
*s
= migrate_get_current();
3384 bool urgent
= false;
3385 migration_update_counters(s
, now
);
3386 if (qemu_file_rate_limit(s
->to_dst_file
)) {
3388 if (qemu_file_get_error(s
->to_dst_file
)) {
3392 * Wait for a delay to do rate limiting OR
3393 * something urgent to post the semaphore.
3395 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- now
;
3396 trace_migration_rate_limit_pre(ms
);
3397 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
3399 * We were woken by one or more urgent things but
3400 * the timedwait will have consumed one of them.
3401 * The service routine for the urgent wake will dec
3402 * the semaphore itself for each item it consumes,
3403 * so add this one we just eat back.
3405 qemu_sem_post(&s
->rate_limit_sem
);
3408 trace_migration_rate_limit_post(urgent
);
3414 * Master migration thread on the source VM.
3415 * It drives the migration and pumps the data down the outgoing channel.
3417 static void *migration_thread(void *opaque
)
3419 MigrationState
*s
= opaque
;
3420 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3421 MigThrError thr_error
;
3422 bool urgent
= false;
3424 rcu_register_thread();
3426 object_ref(OBJECT(s
));
3427 update_iteration_initial_status(s
);
3429 qemu_savevm_state_header(s
->to_dst_file
);
3432 * If we opened the return path, we need to make sure dst has it
3435 if (s
->rp_state
.from_dst_file
) {
3436 /* Now tell the dest that it should open its end so it can reply */
3437 qemu_savevm_send_open_return_path(s
->to_dst_file
);
3439 /* And do a ping that will make stuff easier to debug */
3440 qemu_savevm_send_ping(s
->to_dst_file
, 1);
3443 if (migrate_postcopy()) {
3445 * Tell the destination that we *might* want to do postcopy later;
3446 * if the other end can't do postcopy it should fail now, nice and
3449 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
3452 if (migrate_colo_enabled()) {
3453 /* Notify migration destination that we enable COLO */
3454 qemu_savevm_send_colo_enable(s
->to_dst_file
);
3457 qemu_savevm_state_setup(s
->to_dst_file
);
3459 if (qemu_savevm_state_guest_unplug_pending()) {
3460 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3461 MIGRATION_STATUS_WAIT_UNPLUG
);
3463 while (s
->state
== MIGRATION_STATUS_WAIT_UNPLUG
&&
3464 qemu_savevm_state_guest_unplug_pending()) {
3465 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
3468 migrate_set_state(&s
->state
, MIGRATION_STATUS_WAIT_UNPLUG
,
3469 MIGRATION_STATUS_ACTIVE
);
3472 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3473 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3474 MIGRATION_STATUS_ACTIVE
);
3476 trace_migration_thread_setup_complete();
3478 while (migration_is_active(s
)) {
3479 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
3480 MigIterateState iter_state
= migration_iteration_run(s
);
3481 if (iter_state
== MIG_ITERATE_SKIP
) {
3483 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3489 * Try to detect any kind of failures, and see whether we
3490 * should stop the migration now.
3492 thr_error
= migration_detect_error(s
);
3493 if (thr_error
== MIG_THR_ERR_FATAL
) {
3494 /* Stop migration */
3496 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
3498 * Just recovered from a e.g. network failure, reset all
3499 * the local variables. This is important to avoid
3500 * breaking transferred_bytes and bandwidth calculation
3502 update_iteration_initial_status(s
);
3505 urgent
= migration_rate_limit();
3508 trace_migration_thread_after_loop();
3509 migration_iteration_finish(s
);
3510 object_unref(OBJECT(s
));
3511 rcu_unregister_thread();
3515 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
3517 Error
*local_err
= NULL
;
3519 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
3521 s
->expected_downtime
= s
->parameters
.downtime_limit
;
3523 assert(s
->cleanup_bh
);
3525 assert(!s
->cleanup_bh
);
3526 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup_bh
, s
);
3529 migrate_fd_error(s
, error_in
);
3530 migrate_fd_cleanup(s
);
3535 /* This is a resumed migration */
3536 rate_limit
= s
->parameters
.max_postcopy_bandwidth
/
3539 /* This is a fresh new migration */
3540 rate_limit
= s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
;
3542 /* Notify before starting migration thread */
3543 notifier_list_notify(&migration_state_notifiers
, s
);
3546 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
3547 qemu_file_set_blocking(s
->to_dst_file
, true);
3550 * Open the return path. For postcopy, it is used exclusively. For
3551 * precopy, only if user specified "return-path" capability would
3552 * QEMU uses the return path.
3554 if (migrate_postcopy_ram() || migrate_use_return_path()) {
3555 if (open_return_path_on_source(s
, !resume
)) {
3556 error_report("Unable to open return-path for postcopy");
3557 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
3558 migrate_fd_cleanup(s
);
3564 /* Wakeup the main migration thread to do the recovery */
3565 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
3566 MIGRATION_STATUS_POSTCOPY_RECOVER
);
3567 qemu_sem_post(&s
->postcopy_pause_sem
);
3571 if (multifd_save_setup(&local_err
) != 0) {
3572 error_report_err(local_err
);
3573 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3574 MIGRATION_STATUS_FAILED
);
3575 migrate_fd_cleanup(s
);
3578 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
3579 QEMU_THREAD_JOINABLE
);
3580 s
->migration_thread_running
= true;
3583 void migration_global_dump(Monitor
*mon
)
3585 MigrationState
*ms
= migrate_get_current();
3587 monitor_printf(mon
, "globals:\n");
3588 monitor_printf(mon
, "store-global-state: %s\n",
3589 ms
->store_global_state
? "on" : "off");
3590 monitor_printf(mon
, "only-migratable: %s\n",
3591 only_migratable
? "on" : "off");
3592 monitor_printf(mon
, "send-configuration: %s\n",
3593 ms
->send_configuration
? "on" : "off");
3594 monitor_printf(mon
, "send-section-footer: %s\n",
3595 ms
->send_section_footer
? "on" : "off");
3596 monitor_printf(mon
, "decompress-error-check: %s\n",
3597 ms
->decompress_error_check
? "on" : "off");
3598 monitor_printf(mon
, "clear-bitmap-shift: %u\n",
3599 ms
->clear_bitmap_shift
);
3602 #define DEFINE_PROP_MIG_CAP(name, x) \
3603 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3605 static Property migration_properties
[] = {
3606 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
3607 store_global_state
, true),
3608 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
3609 send_configuration
, true),
3610 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
3611 send_section_footer
, true),
3612 DEFINE_PROP_BOOL("decompress-error-check", MigrationState
,
3613 decompress_error_check
, true),
3614 DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState
,
3615 clear_bitmap_shift
, CLEAR_BITMAP_SHIFT_DEFAULT
),
3617 /* Migration parameters */
3618 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
3619 parameters
.compress_level
,
3620 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
3621 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
3622 parameters
.compress_threads
,
3623 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
3624 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState
,
3625 parameters
.compress_wait_thread
, true),
3626 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
3627 parameters
.decompress_threads
,
3628 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
3629 DEFINE_PROP_UINT8("x-throttle-trigger-threshold", MigrationState
,
3630 parameters
.throttle_trigger_threshold
,
3631 DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD
),
3632 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
3633 parameters
.cpu_throttle_initial
,
3634 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
3635 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
3636 parameters
.cpu_throttle_increment
,
3637 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
3638 DEFINE_PROP_BOOL("x-cpu-throttle-tailslow", MigrationState
,
3639 parameters
.cpu_throttle_tailslow
, false),
3640 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
3641 parameters
.max_bandwidth
, MAX_THROTTLE
),
3642 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
3643 parameters
.downtime_limit
,
3644 DEFAULT_MIGRATE_SET_DOWNTIME
),
3645 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
3646 parameters
.x_checkpoint_delay
,
3647 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
3648 DEFINE_PROP_UINT8("multifd-channels", MigrationState
,
3649 parameters
.multifd_channels
,
3650 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
3651 DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState
,
3652 parameters
.multifd_compression
,
3653 DEFAULT_MIGRATE_MULTIFD_COMPRESSION
),
3654 DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState
,
3655 parameters
.multifd_zlib_level
,
3656 DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL
),
3657 DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState
,
3658 parameters
.multifd_zstd_level
,
3659 DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL
),
3660 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
3661 parameters
.xbzrle_cache_size
,
3662 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
3663 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState
,
3664 parameters
.max_postcopy_bandwidth
,
3665 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH
),
3666 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState
,
3667 parameters
.max_cpu_throttle
,
3668 DEFAULT_MIGRATE_MAX_CPU_THROTTLE
),
3669 DEFINE_PROP_SIZE("announce-initial", MigrationState
,
3670 parameters
.announce_initial
,
3671 DEFAULT_MIGRATE_ANNOUNCE_INITIAL
),
3672 DEFINE_PROP_SIZE("announce-max", MigrationState
,
3673 parameters
.announce_max
,
3674 DEFAULT_MIGRATE_ANNOUNCE_MAX
),
3675 DEFINE_PROP_SIZE("announce-rounds", MigrationState
,
3676 parameters
.announce_rounds
,
3677 DEFAULT_MIGRATE_ANNOUNCE_ROUNDS
),
3678 DEFINE_PROP_SIZE("announce-step", MigrationState
,
3679 parameters
.announce_step
,
3680 DEFAULT_MIGRATE_ANNOUNCE_STEP
),
3682 /* Migration capabilities */
3683 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
3684 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
3685 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
3686 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
3687 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
3688 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
3689 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
3690 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
3691 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
3692 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
3693 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
3694 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD
),
3696 DEFINE_PROP_END_OF_LIST(),
3699 static void migration_class_init(ObjectClass
*klass
, void *data
)
3701 DeviceClass
*dc
= DEVICE_CLASS(klass
);
3703 dc
->user_creatable
= false;
3704 device_class_set_props(dc
, migration_properties
);
3707 static void migration_instance_finalize(Object
*obj
)
3709 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3710 MigrationParameters
*params
= &ms
->parameters
;
3712 qemu_mutex_destroy(&ms
->error_mutex
);
3713 qemu_mutex_destroy(&ms
->qemu_file_lock
);
3714 g_free(params
->tls_hostname
);
3715 g_free(params
->tls_creds
);
3716 qemu_sem_destroy(&ms
->wait_unplug_sem
);
3717 qemu_sem_destroy(&ms
->rate_limit_sem
);
3718 qemu_sem_destroy(&ms
->pause_sem
);
3719 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
3720 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
3721 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
3722 error_free(ms
->error
);
3725 static void migration_instance_init(Object
*obj
)
3727 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3728 MigrationParameters
*params
= &ms
->parameters
;
3730 ms
->state
= MIGRATION_STATUS_NONE
;
3732 ms
->pages_per_second
= -1;
3733 qemu_sem_init(&ms
->pause_sem
, 0);
3734 qemu_mutex_init(&ms
->error_mutex
);
3736 params
->tls_hostname
= g_strdup("");
3737 params
->tls_creds
= g_strdup("");
3739 /* Set has_* up only for parameter checks */
3740 params
->has_compress_level
= true;
3741 params
->has_compress_threads
= true;
3742 params
->has_decompress_threads
= true;
3743 params
->has_throttle_trigger_threshold
= true;
3744 params
->has_cpu_throttle_initial
= true;
3745 params
->has_cpu_throttle_increment
= true;
3746 params
->has_cpu_throttle_tailslow
= true;
3747 params
->has_max_bandwidth
= true;
3748 params
->has_downtime_limit
= true;
3749 params
->has_x_checkpoint_delay
= true;
3750 params
->has_block_incremental
= true;
3751 params
->has_multifd_channels
= true;
3752 params
->has_multifd_compression
= true;
3753 params
->has_multifd_zlib_level
= true;
3754 params
->has_multifd_zstd_level
= true;
3755 params
->has_xbzrle_cache_size
= true;
3756 params
->has_max_postcopy_bandwidth
= true;
3757 params
->has_max_cpu_throttle
= true;
3758 params
->has_announce_initial
= true;
3759 params
->has_announce_max
= true;
3760 params
->has_announce_rounds
= true;
3761 params
->has_announce_step
= true;
3763 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
3764 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
3765 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
3766 qemu_sem_init(&ms
->rate_limit_sem
, 0);
3767 qemu_sem_init(&ms
->wait_unplug_sem
, 0);
3768 qemu_mutex_init(&ms
->qemu_file_lock
);
3772 * Return true if check pass, false otherwise. Error will be put
3773 * inside errp if provided.
3775 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
3777 MigrationCapabilityStatusList
*head
= NULL
;
3778 /* Assuming all off */
3779 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
3782 if (!migrate_params_check(&ms
->parameters
, errp
)) {
3786 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
3787 if (ms
->enabled_capabilities
[i
]) {
3788 head
= migrate_cap_add(head
, i
, true);
3792 ret
= migrate_caps_check(cap_list
, head
, errp
);
3794 /* It works with head == NULL */
3795 qapi_free_MigrationCapabilityStatusList(head
);
3800 static const TypeInfo migration_type
= {
3801 .name
= TYPE_MIGRATION
,
3803 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3804 * not created using qdev_new(), it is not attached to the qdev
3805 * device tree, and it is never realized.
3807 * TODO: Make this TYPE_OBJECT once QOM provides something like
3808 * TYPE_DEVICE's "-global" properties.
3810 .parent
= TYPE_DEVICE
,
3811 .class_init
= migration_class_init
,
3812 .class_size
= sizeof(MigrationClass
),
3813 .instance_size
= sizeof(MigrationState
),
3814 .instance_init
= migration_instance_init
,
3815 .instance_finalize
= migration_instance_finalize
,
3818 static void register_migration_types(void)
3820 type_register_static(&migration_type
);
3823 type_init(register_migration_types
);