4 * Copyright IBM, Corp. 2008
7 * Anthony Liguori <aliguori@us.ibm.com>
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
16 #include "qemu/osdep.h"
17 #include "qemu/cutils.h"
18 #include "qemu/error-report.h"
19 #include "migration/blocker.h"
25 #include "migration/global_state.h"
26 #include "migration/misc.h"
27 #include "migration.h"
29 #include "qemu-file-channel.h"
30 #include "qemu-file.h"
31 #include "migration/vmstate.h"
32 #include "block/block.h"
33 #include "qapi/error.h"
34 #include "qapi/qapi-commands-migration.h"
35 #include "qapi/qapi-events-migration.h"
36 #include "qapi/qmp/qerror.h"
37 #include "qapi/qmp/qnull.h"
40 #include "postcopy-ram.h"
41 #include "qemu/thread.h"
43 #include "exec/target_page.h"
44 #include "io/channel-buffer.h"
45 #include "migration/colo.h"
46 #include "hw/boards.h"
47 #include "monitor/monitor.h"
48 #include "net/announce.h"
50 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
52 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
54 #define BUFFER_DELAY 100
55 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
57 /* Time in milliseconds we are allowed to stop the source,
58 * for sending the last part */
59 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
61 /* Maximum migrate downtime set to 2000 seconds */
62 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
63 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
65 /* Default compression thread count */
66 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
67 /* Default decompression thread count, usually decompression is at
68 * least 4 times as fast as compression.*/
69 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
70 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
71 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
72 /* Define default autoconverge cpu throttle migration parameters */
73 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
74 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
75 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
77 /* Migration XBZRLE default cache size */
78 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
80 /* The delay time (in ms) between two COLO checkpoints */
81 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
82 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
83 #define DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT 16
85 /* Background transfer rate for postcopy, 0 means unlimited, note
86 * that page requests can still exceed this limit.
88 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
91 * Parameters for self_announce_delay giving a stream of RARP/ARP
92 * packets after migration.
94 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
95 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
96 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
97 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
99 static NotifierList migration_state_notifiers
=
100 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
102 static bool deferred_incoming
;
104 /* Messages sent on the return path from destination to source */
105 enum mig_rp_message_type
{
106 MIG_RP_MSG_INVALID
= 0, /* Must be 0 */
107 MIG_RP_MSG_SHUT
, /* sibling will not send any more RP messages */
108 MIG_RP_MSG_PONG
, /* Response to a PING; data (seq: be32 ) */
110 MIG_RP_MSG_REQ_PAGES_ID
, /* data (start: be64, len: be32, id: string) */
111 MIG_RP_MSG_REQ_PAGES
, /* data (start: be64, len: be32) */
112 MIG_RP_MSG_RECV_BITMAP
, /* send recved_bitmap back to source */
113 MIG_RP_MSG_RESUME_ACK
, /* tell source that we are ready to resume */
118 /* When we add fault tolerance, we could have several
119 migrations at once. For now we don't need to add
120 dynamic creation of migration */
122 static MigrationState
*current_migration
;
123 static MigrationIncomingState
*current_incoming
;
125 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
126 static int migration_maybe_pause(MigrationState
*s
,
127 int *current_active_state
,
130 void migration_object_init(void)
132 MachineState
*ms
= MACHINE(qdev_get_machine());
135 /* This can only be called once. */
136 assert(!current_migration
);
137 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
140 * Init the migrate incoming object as well no matter whether
141 * we'll use it or not.
143 assert(!current_incoming
);
144 current_incoming
= g_new0(MigrationIncomingState
, 1);
145 current_incoming
->state
= MIGRATION_STATUS_NONE
;
146 current_incoming
->postcopy_remote_fds
=
147 g_array_new(FALSE
, TRUE
, sizeof(struct PostCopyFD
));
148 qemu_mutex_init(¤t_incoming
->rp_mutex
);
149 qemu_event_init(¤t_incoming
->main_thread_load_event
, false);
150 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_dst
, 0);
151 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fault
, 0);
153 init_dirty_bitmap_incoming_migration();
155 if (!migration_object_check(current_migration
, &err
)) {
156 error_report_err(err
);
161 * We cannot really do this in migration_instance_init() since at
162 * that time global properties are not yet applied, then this
163 * value will be definitely replaced by something else.
165 if (ms
->enforce_config_section
) {
166 current_migration
->send_configuration
= true;
170 void migration_object_finalize(void)
172 object_unref(OBJECT(current_migration
));
176 MigrationState
*migrate_get_current(void)
178 /* This can only be called after the object created. */
179 assert(current_migration
);
180 return current_migration
;
183 MigrationIncomingState
*migration_incoming_get_current(void)
185 assert(current_incoming
);
186 return current_incoming
;
189 void migration_incoming_state_destroy(void)
191 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
193 if (mis
->to_src_file
) {
194 /* Tell source that we are done */
195 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
196 qemu_fclose(mis
->to_src_file
);
197 mis
->to_src_file
= NULL
;
200 if (mis
->from_src_file
) {
201 qemu_fclose(mis
->from_src_file
);
202 mis
->from_src_file
= NULL
;
204 if (mis
->postcopy_remote_fds
) {
205 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
206 mis
->postcopy_remote_fds
= NULL
;
209 qemu_event_reset(&mis
->main_thread_load_event
);
212 static void migrate_generate_event(int new_state
)
214 if (migrate_use_events()) {
215 qapi_event_send_migration(new_state
);
219 static bool migrate_late_block_activate(void)
223 s
= migrate_get_current();
225 return s
->enabled_capabilities
[
226 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE
];
230 * Called on -incoming with a defer: uri.
231 * The migration can be started later after any parameters have been
234 static void deferred_incoming_migration(Error
**errp
)
236 if (deferred_incoming
) {
237 error_setg(errp
, "Incoming migration already deferred");
239 deferred_incoming
= true;
243 * Send a message on the return channel back to the source
246 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
247 enum mig_rp_message_type message_type
,
248 uint16_t len
, void *data
)
252 trace_migrate_send_rp_message((int)message_type
, len
);
253 qemu_mutex_lock(&mis
->rp_mutex
);
256 * It's possible that the file handle got lost due to network
259 if (!mis
->to_src_file
) {
264 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
265 qemu_put_be16(mis
->to_src_file
, len
);
266 qemu_put_buffer(mis
->to_src_file
, data
, len
);
267 qemu_fflush(mis
->to_src_file
);
269 /* It's possible that qemu file got error during sending */
270 ret
= qemu_file_get_error(mis
->to_src_file
);
273 qemu_mutex_unlock(&mis
->rp_mutex
);
277 /* Request a range of pages from the source VM at the given
279 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
280 * as the last request (a name must have been given previously)
281 * Start: Address offset within the RB
282 * Len: Length in bytes required - must be a multiple of pagesize
284 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
, const char *rbname
,
285 ram_addr_t start
, size_t len
)
287 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
288 size_t msglen
= 12; /* start + len */
289 enum mig_rp_message_type msg_type
;
291 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
292 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
295 int rbname_len
= strlen(rbname
);
296 assert(rbname_len
< 256);
298 bufc
[msglen
++] = rbname_len
;
299 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
300 msglen
+= rbname_len
;
301 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
303 msg_type
= MIG_RP_MSG_REQ_PAGES
;
306 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
309 static bool migration_colo_enabled
;
310 bool migration_incoming_colo_enabled(void)
312 return migration_colo_enabled
;
315 void migration_incoming_disable_colo(void)
317 migration_colo_enabled
= false;
320 void migration_incoming_enable_colo(void)
322 migration_colo_enabled
= true;
325 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
329 qapi_event_send_migration(MIGRATION_STATUS_SETUP
);
330 if (!strcmp(uri
, "defer")) {
331 deferred_incoming_migration(errp
);
332 } else if (strstart(uri
, "tcp:", &p
)) {
333 tcp_start_incoming_migration(p
, errp
);
335 } else if (strstart(uri
, "rdma:", &p
)) {
336 rdma_start_incoming_migration(p
, errp
);
338 } else if (strstart(uri
, "exec:", &p
)) {
339 exec_start_incoming_migration(p
, errp
);
340 } else if (strstart(uri
, "unix:", &p
)) {
341 unix_start_incoming_migration(p
, errp
);
342 } else if (strstart(uri
, "fd:", &p
)) {
343 fd_start_incoming_migration(p
, errp
);
345 error_setg(errp
, "unknown migration protocol: %s", uri
);
349 static void process_incoming_migration_bh(void *opaque
)
351 Error
*local_err
= NULL
;
352 MigrationIncomingState
*mis
= opaque
;
354 /* If capability late_block_activate is set:
355 * Only fire up the block code now if we're going to restart the
356 * VM, else 'cont' will do it.
357 * This causes file locking to happen; so we don't want it to happen
358 * unless we really are starting the VM.
360 if (!migrate_late_block_activate() ||
361 (autostart
&& (!global_state_received() ||
362 global_state_get_runstate() == RUN_STATE_RUNNING
))) {
363 /* Make sure all file formats flush their mutable metadata.
364 * If we get an error here, just don't restart the VM yet. */
365 bdrv_invalidate_cache_all(&local_err
);
367 error_report_err(local_err
);
374 * This must happen after all error conditions are dealt with and
375 * we're sure the VM is going to be running on this host.
377 qemu_announce_self();
379 if (multifd_load_cleanup(&local_err
) != 0) {
380 error_report_err(local_err
);
383 /* If global state section was not received or we are in running
384 state, we need to obey autostart. Any other state is set with
387 dirty_bitmap_mig_before_vm_start();
389 if (!global_state_received() ||
390 global_state_get_runstate() == RUN_STATE_RUNNING
) {
394 runstate_set(RUN_STATE_PAUSED
);
397 runstate_set(global_state_get_runstate());
400 * This must happen after any state changes since as soon as an external
401 * observer sees this event they might start to prod at the VM assuming
404 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
405 MIGRATION_STATUS_COMPLETED
);
406 qemu_bh_delete(mis
->bh
);
407 migration_incoming_state_destroy();
410 static void process_incoming_migration_co(void *opaque
)
412 MigrationIncomingState
*mis
= migration_incoming_get_current();
415 Error
*local_err
= NULL
;
417 assert(mis
->from_src_file
);
418 mis
->migration_incoming_co
= qemu_coroutine_self();
419 mis
->largest_page_size
= qemu_ram_pagesize_largest();
420 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
421 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
422 MIGRATION_STATUS_ACTIVE
);
423 ret
= qemu_loadvm_state(mis
->from_src_file
);
425 ps
= postcopy_state_get();
426 trace_process_incoming_migration_co_end(ret
, ps
);
427 if (ps
!= POSTCOPY_INCOMING_NONE
) {
428 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
430 * Where a migration had postcopy enabled (and thus went to advise)
431 * but managed to complete within the precopy period, we can use
434 postcopy_ram_incoming_cleanup(mis
);
435 } else if (ret
>= 0) {
437 * Postcopy was started, cleanup should happen at the end of the
440 trace_process_incoming_migration_co_postcopy_end_main();
443 /* Else if something went wrong then just fall out of the normal exit */
446 /* we get COLO info, and know if we are in COLO mode */
447 if (!ret
&& migration_incoming_colo_enabled()) {
448 /* Make sure all file formats flush their mutable metadata */
449 bdrv_invalidate_cache_all(&local_err
);
451 error_report_err(local_err
);
455 if (colo_init_ram_cache() < 0) {
456 error_report("Init ram cache failed");
460 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
461 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
462 mis
->have_colo_incoming_thread
= true;
463 qemu_coroutine_yield();
465 /* Wait checkpoint incoming thread exit before free resource */
466 qemu_thread_join(&mis
->colo_incoming_thread
);
467 /* We hold the global iothread lock, so it is safe here */
468 colo_release_ram_cache();
472 error_report("load of migration failed: %s", strerror(-ret
));
475 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
476 qemu_bh_schedule(mis
->bh
);
477 mis
->migration_incoming_co
= NULL
;
481 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
482 MIGRATION_STATUS_FAILED
);
483 qemu_fclose(mis
->from_src_file
);
484 if (multifd_load_cleanup(&local_err
) != 0) {
485 error_report_err(local_err
);
490 static void migration_incoming_setup(QEMUFile
*f
)
492 MigrationIncomingState
*mis
= migration_incoming_get_current();
494 if (multifd_load_setup() != 0) {
495 /* We haven't been able to create multifd threads
496 nothing better to do */
500 if (!mis
->from_src_file
) {
501 mis
->from_src_file
= f
;
503 qemu_file_set_blocking(f
, false);
506 void migration_incoming_process(void)
508 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
509 qemu_coroutine_enter(co
);
512 /* Returns true if recovered from a paused migration, otherwise false */
513 static bool postcopy_try_recover(QEMUFile
*f
)
515 MigrationIncomingState
*mis
= migration_incoming_get_current();
517 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
518 /* Resumed from a paused postcopy migration */
520 mis
->from_src_file
= f
;
521 /* Postcopy has standalone thread to do vm load */
522 qemu_file_set_blocking(f
, true);
524 /* Re-configure the return path */
525 mis
->to_src_file
= qemu_file_get_return_path(f
);
527 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
528 MIGRATION_STATUS_POSTCOPY_RECOVER
);
531 * Here, we only wake up the main loading thread (while the
532 * fault thread will still be waiting), so that we can receive
533 * commands from source now, and answer it if needed. The
534 * fault thread will be woken up afterwards until we are sure
535 * that source is ready to reply to page requests.
537 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
544 void migration_fd_process_incoming(QEMUFile
*f
)
546 if (postcopy_try_recover(f
)) {
550 migration_incoming_setup(f
);
551 migration_incoming_process();
554 void migration_ioc_process_incoming(QIOChannel
*ioc
, Error
**errp
)
556 MigrationIncomingState
*mis
= migration_incoming_get_current();
557 bool start_migration
;
559 if (!mis
->from_src_file
) {
560 /* The first connection (multifd may have multiple) */
561 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
563 /* If it's a recovery, we're done */
564 if (postcopy_try_recover(f
)) {
568 migration_incoming_setup(f
);
571 * Common migration only needs one channel, so we can start
572 * right now. Multifd needs more than one channel, we wait.
574 start_migration
= !migrate_use_multifd();
576 Error
*local_err
= NULL
;
577 /* Multiple connections */
578 assert(migrate_use_multifd());
579 start_migration
= multifd_recv_new_channel(ioc
, &local_err
);
581 error_propagate(errp
, local_err
);
586 if (start_migration
) {
587 migration_incoming_process();
592 * @migration_has_all_channels: We have received all channels that we need
594 * Returns true when we have got connections to all the channels that
595 * we need for migration.
597 bool migration_has_all_channels(void)
599 MigrationIncomingState
*mis
= migration_incoming_get_current();
602 all_channels
= multifd_recv_all_channels_created();
604 return all_channels
&& mis
->from_src_file
!= NULL
;
608 * Send a 'SHUT' message on the return channel with the given value
609 * to indicate that we've finished with the RP. Non-0 value indicates
612 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
617 buf
= cpu_to_be32(value
);
618 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
622 * Send a 'PONG' message on the return channel with the given value
623 * (normally in response to a 'PING')
625 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
630 buf
= cpu_to_be32(value
);
631 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
634 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
642 * First, we send the header part. It contains only the len of
643 * idstr, and the idstr itself.
645 len
= strlen(block_name
);
647 memcpy(buf
+ 1, block_name
, len
);
649 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
650 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
655 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
658 * Next, we dump the received bitmap to the stream.
660 * TODO: currently we are safe since we are the only one that is
661 * using the to_src_file handle (fault thread is still paused),
662 * and it's ok even not taking the mutex. However the best way is
663 * to take the lock before sending the message header, and release
664 * the lock after sending the bitmap.
666 qemu_mutex_lock(&mis
->rp_mutex
);
667 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
668 qemu_mutex_unlock(&mis
->rp_mutex
);
670 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
673 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
677 buf
= cpu_to_be32(value
);
678 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
681 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
683 MigrationCapabilityStatusList
*head
= NULL
;
684 MigrationCapabilityStatusList
*caps
;
685 MigrationState
*s
= migrate_get_current();
688 caps
= NULL
; /* silence compiler warning */
689 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
690 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
691 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
696 head
= g_malloc0(sizeof(*caps
));
699 caps
->next
= g_malloc0(sizeof(*caps
));
703 g_malloc(sizeof(*caps
->value
));
704 caps
->value
->capability
= i
;
705 caps
->value
->state
= s
->enabled_capabilities
[i
];
711 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
713 MigrationParameters
*params
;
714 MigrationState
*s
= migrate_get_current();
716 /* TODO use QAPI_CLONE() instead of duplicating it inline */
717 params
= g_malloc0(sizeof(*params
));
718 params
->has_compress_level
= true;
719 params
->compress_level
= s
->parameters
.compress_level
;
720 params
->has_compress_threads
= true;
721 params
->compress_threads
= s
->parameters
.compress_threads
;
722 params
->has_compress_wait_thread
= true;
723 params
->compress_wait_thread
= s
->parameters
.compress_wait_thread
;
724 params
->has_decompress_threads
= true;
725 params
->decompress_threads
= s
->parameters
.decompress_threads
;
726 params
->has_cpu_throttle_initial
= true;
727 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
728 params
->has_cpu_throttle_increment
= true;
729 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
730 params
->has_tls_creds
= true;
731 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
732 params
->has_tls_hostname
= true;
733 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
734 params
->has_max_bandwidth
= true;
735 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
736 params
->has_downtime_limit
= true;
737 params
->downtime_limit
= s
->parameters
.downtime_limit
;
738 params
->has_x_checkpoint_delay
= true;
739 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
740 params
->has_block_incremental
= true;
741 params
->block_incremental
= s
->parameters
.block_incremental
;
742 params
->has_x_multifd_channels
= true;
743 params
->x_multifd_channels
= s
->parameters
.x_multifd_channels
;
744 params
->has_x_multifd_page_count
= true;
745 params
->x_multifd_page_count
= s
->parameters
.x_multifd_page_count
;
746 params
->has_xbzrle_cache_size
= true;
747 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
748 params
->has_max_postcopy_bandwidth
= true;
749 params
->max_postcopy_bandwidth
= s
->parameters
.max_postcopy_bandwidth
;
750 params
->has_max_cpu_throttle
= true;
751 params
->max_cpu_throttle
= s
->parameters
.max_cpu_throttle
;
752 params
->has_announce_initial
= true;
753 params
->announce_initial
= s
->parameters
.announce_initial
;
754 params
->has_announce_max
= true;
755 params
->announce_max
= s
->parameters
.announce_max
;
756 params
->has_announce_rounds
= true;
757 params
->announce_rounds
= s
->parameters
.announce_rounds
;
758 params
->has_announce_step
= true;
759 params
->announce_step
= s
->parameters
.announce_step
;
764 AnnounceParameters
*migrate_announce_params(void)
766 static AnnounceParameters ap
;
768 MigrationState
*s
= migrate_get_current();
770 ap
.initial
= s
->parameters
.announce_initial
;
771 ap
.max
= s
->parameters
.announce_max
;
772 ap
.rounds
= s
->parameters
.announce_rounds
;
773 ap
.step
= s
->parameters
.announce_step
;
779 * Return true if we're already in the middle of a migration
780 * (i.e. any of the active or setup states)
782 bool migration_is_setup_or_active(int state
)
785 case MIGRATION_STATUS_ACTIVE
:
786 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
787 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
788 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
789 case MIGRATION_STATUS_SETUP
:
790 case MIGRATION_STATUS_PRE_SWITCHOVER
:
791 case MIGRATION_STATUS_DEVICE
:
800 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
802 info
->has_ram
= true;
803 info
->ram
= g_malloc0(sizeof(*info
->ram
));
804 info
->ram
->transferred
= ram_counters
.transferred
;
805 info
->ram
->total
= ram_bytes_total();
806 info
->ram
->duplicate
= ram_counters
.duplicate
;
807 /* legacy value. It is not used anymore */
808 info
->ram
->skipped
= 0;
809 info
->ram
->normal
= ram_counters
.normal
;
810 info
->ram
->normal_bytes
= ram_counters
.normal
*
811 qemu_target_page_size();
812 info
->ram
->mbps
= s
->mbps
;
813 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
814 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
815 info
->ram
->page_size
= qemu_target_page_size();
816 info
->ram
->multifd_bytes
= ram_counters
.multifd_bytes
;
817 info
->ram
->pages_per_second
= s
->pages_per_second
;
819 if (migrate_use_xbzrle()) {
820 info
->has_xbzrle_cache
= true;
821 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
822 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
823 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
824 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
825 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
826 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
827 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
830 if (migrate_use_compression()) {
831 info
->has_compression
= true;
832 info
->compression
= g_malloc0(sizeof(*info
->compression
));
833 info
->compression
->pages
= compression_counters
.pages
;
834 info
->compression
->busy
= compression_counters
.busy
;
835 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
836 info
->compression
->compressed_size
=
837 compression_counters
.compressed_size
;
838 info
->compression
->compression_rate
=
839 compression_counters
.compression_rate
;
842 if (cpu_throttle_active()) {
843 info
->has_cpu_throttle_percentage
= true;
844 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
847 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
848 info
->ram
->remaining
= ram_bytes_remaining();
849 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
853 static void populate_disk_info(MigrationInfo
*info
)
855 if (blk_mig_active()) {
856 info
->has_disk
= true;
857 info
->disk
= g_malloc0(sizeof(*info
->disk
));
858 info
->disk
->transferred
= blk_mig_bytes_transferred();
859 info
->disk
->remaining
= blk_mig_bytes_remaining();
860 info
->disk
->total
= blk_mig_bytes_total();
864 static void fill_source_migration_info(MigrationInfo
*info
)
866 MigrationState
*s
= migrate_get_current();
869 case MIGRATION_STATUS_NONE
:
870 /* no migration has happened ever */
871 /* do not overwrite destination migration status */
874 case MIGRATION_STATUS_SETUP
:
875 info
->has_status
= true;
876 info
->has_total_time
= false;
878 case MIGRATION_STATUS_ACTIVE
:
879 case MIGRATION_STATUS_CANCELLING
:
880 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
881 case MIGRATION_STATUS_PRE_SWITCHOVER
:
882 case MIGRATION_STATUS_DEVICE
:
883 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
884 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
885 /* TODO add some postcopy stats */
886 info
->has_status
= true;
887 info
->has_total_time
= true;
888 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
890 info
->has_expected_downtime
= true;
891 info
->expected_downtime
= s
->expected_downtime
;
892 info
->has_setup_time
= true;
893 info
->setup_time
= s
->setup_time
;
895 populate_ram_info(info
, s
);
896 populate_disk_info(info
);
898 case MIGRATION_STATUS_COLO
:
899 info
->has_status
= true;
900 /* TODO: display COLO specific information (checkpoint info etc.) */
902 case MIGRATION_STATUS_COMPLETED
:
903 info
->has_status
= true;
904 info
->has_total_time
= true;
905 info
->total_time
= s
->total_time
;
906 info
->has_downtime
= true;
907 info
->downtime
= s
->downtime
;
908 info
->has_setup_time
= true;
909 info
->setup_time
= s
->setup_time
;
911 populate_ram_info(info
, s
);
913 case MIGRATION_STATUS_FAILED
:
914 info
->has_status
= true;
916 info
->has_error_desc
= true;
917 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
920 case MIGRATION_STATUS_CANCELLED
:
921 info
->has_status
= true;
924 info
->status
= s
->state
;
928 * @migration_caps_check - check capability validity
930 * @cap_list: old capability list, array of bool
931 * @params: new capabilities to be applied soon
932 * @errp: set *errp if the check failed, with reason
934 * Returns true if check passed, otherwise false.
936 static bool migrate_caps_check(bool *cap_list
,
937 MigrationCapabilityStatusList
*params
,
940 MigrationCapabilityStatusList
*cap
;
941 bool old_postcopy_cap
;
942 MigrationIncomingState
*mis
= migration_incoming_get_current();
944 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
946 for (cap
= params
; cap
; cap
= cap
->next
) {
947 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
950 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
951 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
952 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
954 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
959 #ifndef CONFIG_REPLICATION
960 if (cap_list
[MIGRATION_CAPABILITY_X_COLO
]) {
961 error_setg(errp
, "QEMU compiled without replication module"
962 " can't enable COLO");
963 error_append_hint(errp
, "Please enable replication before COLO.\n");
968 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
969 if (cap_list
[MIGRATION_CAPABILITY_COMPRESS
]) {
970 /* The decompression threads asynchronously write into RAM
971 * rather than use the atomic copies needed to avoid
972 * userfaulting. It should be possible to fix the decompression
973 * threads for compatibility in future.
975 error_setg(errp
, "Postcopy is not currently compatible "
980 /* This check is reasonably expensive, so only when it's being
981 * set the first time, also it's only the destination that needs
984 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
985 !postcopy_ram_supported_by_host(mis
)) {
986 /* postcopy_ram_supported_by_host will have emitted a more
989 error_setg(errp
, "Postcopy is not supported");
997 static void fill_destination_migration_info(MigrationInfo
*info
)
999 MigrationIncomingState
*mis
= migration_incoming_get_current();
1001 switch (mis
->state
) {
1002 case MIGRATION_STATUS_NONE
:
1005 case MIGRATION_STATUS_SETUP
:
1006 case MIGRATION_STATUS_CANCELLING
:
1007 case MIGRATION_STATUS_CANCELLED
:
1008 case MIGRATION_STATUS_ACTIVE
:
1009 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1010 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1011 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1012 case MIGRATION_STATUS_FAILED
:
1013 case MIGRATION_STATUS_COLO
:
1014 info
->has_status
= true;
1016 case MIGRATION_STATUS_COMPLETED
:
1017 info
->has_status
= true;
1018 fill_destination_postcopy_migration_info(info
);
1021 info
->status
= mis
->state
;
1024 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1026 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1028 fill_destination_migration_info(info
);
1029 fill_source_migration_info(info
);
1034 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
1037 MigrationState
*s
= migrate_get_current();
1038 MigrationCapabilityStatusList
*cap
;
1039 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
1041 if (migration_is_setup_or_active(s
->state
)) {
1042 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1046 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
1047 if (!migrate_caps_check(cap_list
, params
, errp
)) {
1051 for (cap
= params
; cap
; cap
= cap
->next
) {
1052 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
1057 * Check whether the parameters are valid. Error will be put into errp
1058 * (if provided). Return true if valid, otherwise false.
1060 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
1062 if (params
->has_compress_level
&&
1063 (params
->compress_level
> 9)) {
1064 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
1065 "is invalid, it should be in the range of 0 to 9");
1069 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
1070 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1072 "is invalid, it should be in the range of 1 to 255");
1076 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
1077 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1078 "decompress_threads",
1079 "is invalid, it should be in the range of 1 to 255");
1083 if (params
->has_cpu_throttle_initial
&&
1084 (params
->cpu_throttle_initial
< 1 ||
1085 params
->cpu_throttle_initial
> 99)) {
1086 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1087 "cpu_throttle_initial",
1088 "an integer in the range of 1 to 99");
1092 if (params
->has_cpu_throttle_increment
&&
1093 (params
->cpu_throttle_increment
< 1 ||
1094 params
->cpu_throttle_increment
> 99)) {
1095 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1096 "cpu_throttle_increment",
1097 "an integer in the range of 1 to 99");
1101 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
1102 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
1103 " range of 0 to %zu bytes/second", SIZE_MAX
);
1107 if (params
->has_downtime_limit
&&
1108 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
1109 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1110 "the range of 0 to %d milliseconds",
1111 MAX_MIGRATE_DOWNTIME
);
1115 /* x_checkpoint_delay is now always positive */
1117 if (params
->has_x_multifd_channels
&& (params
->x_multifd_channels
< 1)) {
1118 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1120 "is invalid, it should be in the range of 1 to 255");
1123 if (params
->has_x_multifd_page_count
&&
1124 (params
->x_multifd_page_count
< 1 ||
1125 params
->x_multifd_page_count
> 10000)) {
1126 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1127 "multifd_page_count",
1128 "is invalid, it should be in the range of 1 to 10000");
1132 if (params
->has_xbzrle_cache_size
&&
1133 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
1134 !is_power_of_2(params
->xbzrle_cache_size
))) {
1135 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1136 "xbzrle_cache_size",
1137 "is invalid, it should be bigger than target page size"
1138 " and a power of two");
1142 if (params
->has_max_cpu_throttle
&&
1143 (params
->max_cpu_throttle
< params
->cpu_throttle_initial
||
1144 params
->max_cpu_throttle
> 99)) {
1145 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1147 "an integer in the range of cpu_throttle_initial to 99");
1151 if (params
->has_announce_initial
&&
1152 params
->announce_initial
> 100000) {
1153 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1155 "is invalid, it must be less than 100000 ms");
1158 if (params
->has_announce_max
&&
1159 params
->announce_max
> 100000) {
1160 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1162 "is invalid, it must be less than 100000 ms");
1165 if (params
->has_announce_rounds
&&
1166 params
->announce_rounds
> 1000) {
1167 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1169 "is invalid, it must be in the range of 0 to 1000");
1172 if (params
->has_announce_step
&&
1173 (params
->announce_step
< 1 ||
1174 params
->announce_step
> 10000)) {
1175 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1177 "is invalid, it must be in the range of 1 to 10000 ms");
1183 static void migrate_params_test_apply(MigrateSetParameters
*params
,
1184 MigrationParameters
*dest
)
1186 *dest
= migrate_get_current()->parameters
;
1188 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1190 if (params
->has_compress_level
) {
1191 dest
->compress_level
= params
->compress_level
;
1194 if (params
->has_compress_threads
) {
1195 dest
->compress_threads
= params
->compress_threads
;
1198 if (params
->has_compress_wait_thread
) {
1199 dest
->compress_wait_thread
= params
->compress_wait_thread
;
1202 if (params
->has_decompress_threads
) {
1203 dest
->decompress_threads
= params
->decompress_threads
;
1206 if (params
->has_cpu_throttle_initial
) {
1207 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
1210 if (params
->has_cpu_throttle_increment
) {
1211 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
1214 if (params
->has_tls_creds
) {
1215 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1216 dest
->tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1219 if (params
->has_tls_hostname
) {
1220 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1221 dest
->tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1224 if (params
->has_max_bandwidth
) {
1225 dest
->max_bandwidth
= params
->max_bandwidth
;
1228 if (params
->has_downtime_limit
) {
1229 dest
->downtime_limit
= params
->downtime_limit
;
1232 if (params
->has_x_checkpoint_delay
) {
1233 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
1236 if (params
->has_block_incremental
) {
1237 dest
->block_incremental
= params
->block_incremental
;
1239 if (params
->has_x_multifd_channels
) {
1240 dest
->x_multifd_channels
= params
->x_multifd_channels
;
1242 if (params
->has_x_multifd_page_count
) {
1243 dest
->x_multifd_page_count
= params
->x_multifd_page_count
;
1245 if (params
->has_xbzrle_cache_size
) {
1246 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
1248 if (params
->has_max_postcopy_bandwidth
) {
1249 dest
->max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1251 if (params
->has_max_cpu_throttle
) {
1252 dest
->max_cpu_throttle
= params
->max_cpu_throttle
;
1254 if (params
->has_announce_initial
) {
1255 dest
->announce_initial
= params
->announce_initial
;
1257 if (params
->has_announce_max
) {
1258 dest
->announce_max
= params
->announce_max
;
1260 if (params
->has_announce_rounds
) {
1261 dest
->announce_rounds
= params
->announce_rounds
;
1263 if (params
->has_announce_step
) {
1264 dest
->announce_step
= params
->announce_step
;
1268 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
1270 MigrationState
*s
= migrate_get_current();
1272 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1274 if (params
->has_compress_level
) {
1275 s
->parameters
.compress_level
= params
->compress_level
;
1278 if (params
->has_compress_threads
) {
1279 s
->parameters
.compress_threads
= params
->compress_threads
;
1282 if (params
->has_compress_wait_thread
) {
1283 s
->parameters
.compress_wait_thread
= params
->compress_wait_thread
;
1286 if (params
->has_decompress_threads
) {
1287 s
->parameters
.decompress_threads
= params
->decompress_threads
;
1290 if (params
->has_cpu_throttle_initial
) {
1291 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
1294 if (params
->has_cpu_throttle_increment
) {
1295 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
1298 if (params
->has_tls_creds
) {
1299 g_free(s
->parameters
.tls_creds
);
1300 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1301 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1304 if (params
->has_tls_hostname
) {
1305 g_free(s
->parameters
.tls_hostname
);
1306 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1307 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1310 if (params
->has_max_bandwidth
) {
1311 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1312 if (s
->to_dst_file
) {
1313 qemu_file_set_rate_limit(s
->to_dst_file
,
1314 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1318 if (params
->has_downtime_limit
) {
1319 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1322 if (params
->has_x_checkpoint_delay
) {
1323 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1324 if (migration_in_colo_state()) {
1325 colo_checkpoint_notify(s
);
1329 if (params
->has_block_incremental
) {
1330 s
->parameters
.block_incremental
= params
->block_incremental
;
1332 if (params
->has_x_multifd_channels
) {
1333 s
->parameters
.x_multifd_channels
= params
->x_multifd_channels
;
1335 if (params
->has_x_multifd_page_count
) {
1336 s
->parameters
.x_multifd_page_count
= params
->x_multifd_page_count
;
1338 if (params
->has_xbzrle_cache_size
) {
1339 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1340 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1342 if (params
->has_max_postcopy_bandwidth
) {
1343 s
->parameters
.max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1345 if (params
->has_max_cpu_throttle
) {
1346 s
->parameters
.max_cpu_throttle
= params
->max_cpu_throttle
;
1348 if (params
->has_announce_initial
) {
1349 s
->parameters
.announce_initial
= params
->announce_initial
;
1351 if (params
->has_announce_max
) {
1352 s
->parameters
.announce_max
= params
->announce_max
;
1354 if (params
->has_announce_rounds
) {
1355 s
->parameters
.announce_rounds
= params
->announce_rounds
;
1357 if (params
->has_announce_step
) {
1358 s
->parameters
.announce_step
= params
->announce_step
;
1362 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1364 MigrationParameters tmp
;
1366 /* TODO Rewrite "" to null instead */
1367 if (params
->has_tls_creds
1368 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1369 qobject_unref(params
->tls_creds
->u
.n
);
1370 params
->tls_creds
->type
= QTYPE_QSTRING
;
1371 params
->tls_creds
->u
.s
= strdup("");
1373 /* TODO Rewrite "" to null instead */
1374 if (params
->has_tls_hostname
1375 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1376 qobject_unref(params
->tls_hostname
->u
.n
);
1377 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1378 params
->tls_hostname
->u
.s
= strdup("");
1381 migrate_params_test_apply(params
, &tmp
);
1383 if (!migrate_params_check(&tmp
, errp
)) {
1384 /* Invalid parameter */
1388 migrate_params_apply(params
, errp
);
1392 void qmp_migrate_start_postcopy(Error
**errp
)
1394 MigrationState
*s
= migrate_get_current();
1396 if (!migrate_postcopy()) {
1397 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1398 " the start of migration");
1402 if (s
->state
== MIGRATION_STATUS_NONE
) {
1403 error_setg(errp
, "Postcopy must be started after migration has been"
1408 * we don't error if migration has finished since that would be racy
1409 * with issuing this command.
1411 atomic_set(&s
->start_postcopy
, true);
1414 /* shared migration helpers */
1416 void migrate_set_state(int *state
, int old_state
, int new_state
)
1418 assert(new_state
< MIGRATION_STATUS__MAX
);
1419 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1420 trace_migrate_set_state(MigrationStatus_str(new_state
));
1421 migrate_generate_event(new_state
);
1425 static MigrationCapabilityStatusList
*migrate_cap_add(
1426 MigrationCapabilityStatusList
*list
,
1427 MigrationCapability index
,
1430 MigrationCapabilityStatusList
*cap
;
1432 cap
= g_new0(MigrationCapabilityStatusList
, 1);
1433 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
1434 cap
->value
->capability
= index
;
1435 cap
->value
->state
= state
;
1441 void migrate_set_block_enabled(bool value
, Error
**errp
)
1443 MigrationCapabilityStatusList
*cap
;
1445 cap
= migrate_cap_add(NULL
, MIGRATION_CAPABILITY_BLOCK
, value
);
1446 qmp_migrate_set_capabilities(cap
, errp
);
1447 qapi_free_MigrationCapabilityStatusList(cap
);
1450 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1452 s
->parameters
.block_incremental
= value
;
1455 static void block_cleanup_parameters(MigrationState
*s
)
1457 if (s
->must_remove_block_options
) {
1458 /* setting to false can never fail */
1459 migrate_set_block_enabled(false, &error_abort
);
1460 migrate_set_block_incremental(s
, false);
1461 s
->must_remove_block_options
= false;
1465 static void migrate_fd_cleanup(void *opaque
)
1467 MigrationState
*s
= opaque
;
1469 qemu_bh_delete(s
->cleanup_bh
);
1470 s
->cleanup_bh
= NULL
;
1472 qemu_savevm_state_cleanup();
1474 if (s
->to_dst_file
) {
1477 trace_migrate_fd_cleanup();
1478 qemu_mutex_unlock_iothread();
1479 if (s
->migration_thread_running
) {
1480 qemu_thread_join(&s
->thread
);
1481 s
->migration_thread_running
= false;
1483 qemu_mutex_lock_iothread();
1485 multifd_save_cleanup();
1486 qemu_mutex_lock(&s
->qemu_file_lock
);
1487 tmp
= s
->to_dst_file
;
1488 s
->to_dst_file
= NULL
;
1489 qemu_mutex_unlock(&s
->qemu_file_lock
);
1491 * Close the file handle without the lock to make sure the
1492 * critical section won't block for long.
1497 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
1498 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
1500 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1501 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1502 MIGRATION_STATUS_CANCELLED
);
1506 /* It is used on info migrate. We can't free it */
1507 error_report_err(error_copy(s
->error
));
1509 notifier_list_notify(&migration_state_notifiers
, s
);
1510 block_cleanup_parameters(s
);
1513 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1515 qemu_mutex_lock(&s
->error_mutex
);
1517 s
->error
= error_copy(error
);
1519 qemu_mutex_unlock(&s
->error_mutex
);
1522 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1524 trace_migrate_fd_error(error_get_pretty(error
));
1525 assert(s
->to_dst_file
== NULL
);
1526 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1527 MIGRATION_STATUS_FAILED
);
1528 migrate_set_error(s
, error
);
1531 static void migrate_fd_cancel(MigrationState
*s
)
1534 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1535 trace_migrate_fd_cancel();
1537 if (s
->rp_state
.from_dst_file
) {
1538 /* shutdown the rp socket, so causing the rp thread to shutdown */
1539 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1543 old_state
= s
->state
;
1544 if (!migration_is_setup_or_active(old_state
)) {
1547 /* If the migration is paused, kick it out of the pause */
1548 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1549 qemu_sem_post(&s
->pause_sem
);
1551 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1552 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1555 * If we're unlucky the migration code might be stuck somewhere in a
1556 * send/write while the network has failed and is waiting to timeout;
1557 * if we've got shutdown(2) available then we can force it to quit.
1558 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1559 * called in a bh, so there is no race against this cancel.
1561 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1562 qemu_file_shutdown(f
);
1564 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1565 Error
*local_err
= NULL
;
1567 bdrv_invalidate_cache_all(&local_err
);
1569 error_report_err(local_err
);
1571 s
->block_inactive
= false;
1576 void add_migration_state_change_notifier(Notifier
*notify
)
1578 notifier_list_add(&migration_state_notifiers
, notify
);
1581 void remove_migration_state_change_notifier(Notifier
*notify
)
1583 notifier_remove(notify
);
1586 bool migration_in_setup(MigrationState
*s
)
1588 return s
->state
== MIGRATION_STATUS_SETUP
;
1591 bool migration_has_finished(MigrationState
*s
)
1593 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1596 bool migration_has_failed(MigrationState
*s
)
1598 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1599 s
->state
== MIGRATION_STATUS_FAILED
);
1602 bool migration_in_postcopy(void)
1604 MigrationState
*s
= migrate_get_current();
1606 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1609 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1611 return migration_in_postcopy() && s
->postcopy_after_devices
;
1614 bool migration_is_idle(void)
1616 MigrationState
*s
= migrate_get_current();
1619 case MIGRATION_STATUS_NONE
:
1620 case MIGRATION_STATUS_CANCELLED
:
1621 case MIGRATION_STATUS_COMPLETED
:
1622 case MIGRATION_STATUS_FAILED
:
1624 case MIGRATION_STATUS_SETUP
:
1625 case MIGRATION_STATUS_CANCELLING
:
1626 case MIGRATION_STATUS_ACTIVE
:
1627 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1628 case MIGRATION_STATUS_COLO
:
1629 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1630 case MIGRATION_STATUS_DEVICE
:
1632 case MIGRATION_STATUS__MAX
:
1633 g_assert_not_reached();
1639 void migrate_init(MigrationState
*s
)
1642 * Reinitialise all migration state, except
1643 * parameters/capabilities that the user set, and
1649 s
->to_dst_file
= NULL
;
1650 s
->state
= MIGRATION_STATUS_NONE
;
1651 s
->rp_state
.from_dst_file
= NULL
;
1652 s
->rp_state
.error
= false;
1654 s
->pages_per_second
= 0.0;
1656 s
->expected_downtime
= 0;
1658 s
->start_postcopy
= false;
1659 s
->postcopy_after_devices
= false;
1660 s
->migration_thread_running
= false;
1661 error_free(s
->error
);
1664 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1666 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1668 s
->vm_was_running
= false;
1669 s
->iteration_initial_bytes
= 0;
1670 s
->threshold_size
= 0;
1673 static GSList
*migration_blockers
;
1675 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1677 if (migrate_get_current()->only_migratable
) {
1678 error_propagate_prepend(errp
, error_copy(reason
),
1679 "disallowing migration blocker "
1680 "(--only_migratable) for: ");
1684 if (migration_is_idle()) {
1685 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1689 error_propagate_prepend(errp
, error_copy(reason
),
1690 "disallowing migration blocker "
1691 "(migration in progress) for: ");
1695 void migrate_del_blocker(Error
*reason
)
1697 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1700 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1702 Error
*local_err
= NULL
;
1703 static bool once
= true;
1705 if (!deferred_incoming
) {
1706 error_setg(errp
, "For use with '-incoming defer'");
1710 error_setg(errp
, "The incoming migration has already been started");
1713 qemu_start_incoming_migration(uri
, &local_err
);
1716 error_propagate(errp
, local_err
);
1723 void qmp_migrate_recover(const char *uri
, Error
**errp
)
1725 MigrationIncomingState
*mis
= migration_incoming_get_current();
1727 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1728 error_setg(errp
, "Migrate recover can only be run "
1729 "when postcopy is paused.");
1733 if (atomic_cmpxchg(&mis
->postcopy_recover_triggered
,
1734 false, true) == true) {
1735 error_setg(errp
, "Migrate recovery is triggered already");
1740 * Note that this call will never start a real migration; it will
1741 * only re-setup the migration stream and poke existing migration
1742 * to continue using that newly established channel.
1744 qemu_start_incoming_migration(uri
, errp
);
1747 void qmp_migrate_pause(Error
**errp
)
1749 MigrationState
*ms
= migrate_get_current();
1750 MigrationIncomingState
*mis
= migration_incoming_get_current();
1753 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1754 /* Source side, during postcopy */
1755 qemu_mutex_lock(&ms
->qemu_file_lock
);
1756 ret
= qemu_file_shutdown(ms
->to_dst_file
);
1757 qemu_mutex_unlock(&ms
->qemu_file_lock
);
1759 error_setg(errp
, "Failed to pause source migration");
1764 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1765 ret
= qemu_file_shutdown(mis
->from_src_file
);
1767 error_setg(errp
, "Failed to pause destination migration");
1772 error_setg(errp
, "migrate-pause is currently only supported "
1773 "during postcopy-active state");
1776 bool migration_is_blocked(Error
**errp
)
1778 if (qemu_savevm_state_blocked(errp
)) {
1782 if (migration_blockers
) {
1783 error_propagate(errp
, error_copy(migration_blockers
->data
));
1790 /* Returns true if continue to migrate, or false if error detected */
1791 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
1792 bool resume
, Error
**errp
)
1794 Error
*local_err
= NULL
;
1797 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1798 error_setg(errp
, "Cannot resume if there is no "
1799 "paused migration");
1804 * Postcopy recovery won't work well with release-ram
1805 * capability since release-ram will drop the page buffer as
1806 * long as the page is put into the send buffer. So if there
1807 * is a network failure happened, any page buffers that have
1808 * not yet reached the destination VM but have already been
1809 * sent from the source VM will be lost forever. Let's refuse
1810 * the client from resuming such a postcopy migration.
1811 * Luckily release-ram was designed to only be used when src
1812 * and destination VMs are on the same host, so it should be
1815 if (migrate_release_ram()) {
1816 error_setg(errp
, "Postcopy recovery cannot work "
1817 "when release-ram capability is set");
1821 /* This is a resume, skip init status */
1825 if (migration_is_setup_or_active(s
->state
) ||
1826 s
->state
== MIGRATION_STATUS_CANCELLING
||
1827 s
->state
== MIGRATION_STATUS_COLO
) {
1828 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1832 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1833 error_setg(errp
, "Guest is waiting for an incoming migration");
1837 if (migration_is_blocked(errp
)) {
1841 if (blk
|| blk_inc
) {
1842 if (migrate_use_block() || migrate_use_block_incremental()) {
1843 error_setg(errp
, "Command options are incompatible with "
1844 "current migration capabilities");
1847 migrate_set_block_enabled(true, &local_err
);
1849 error_propagate(errp
, local_err
);
1852 s
->must_remove_block_options
= true;
1856 migrate_set_block_incremental(s
, true);
1864 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1865 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1866 bool has_resume
, bool resume
, Error
**errp
)
1868 Error
*local_err
= NULL
;
1869 MigrationState
*s
= migrate_get_current();
1872 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
1873 has_resume
&& resume
, errp
)) {
1874 /* Error detected, put into errp */
1878 if (strstart(uri
, "tcp:", &p
)) {
1879 tcp_start_outgoing_migration(s
, p
, &local_err
);
1881 } else if (strstart(uri
, "rdma:", &p
)) {
1882 rdma_start_outgoing_migration(s
, p
, &local_err
);
1884 } else if (strstart(uri
, "exec:", &p
)) {
1885 exec_start_outgoing_migration(s
, p
, &local_err
);
1886 } else if (strstart(uri
, "unix:", &p
)) {
1887 unix_start_outgoing_migration(s
, p
, &local_err
);
1888 } else if (strstart(uri
, "fd:", &p
)) {
1889 fd_start_outgoing_migration(s
, p
, &local_err
);
1891 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1892 "a valid migration protocol");
1893 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1894 MIGRATION_STATUS_FAILED
);
1895 block_cleanup_parameters(s
);
1900 migrate_fd_error(s
, local_err
);
1901 error_propagate(errp
, local_err
);
1906 void qmp_migrate_cancel(Error
**errp
)
1908 migrate_fd_cancel(migrate_get_current());
1911 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
1913 MigrationState
*s
= migrate_get_current();
1914 if (s
->state
!= state
) {
1915 error_setg(errp
, "Migration not in expected state: %s",
1916 MigrationStatus_str(s
->state
));
1919 qemu_sem_post(&s
->pause_sem
);
1922 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1924 MigrateSetParameters p
= {
1925 .has_xbzrle_cache_size
= true,
1926 .xbzrle_cache_size
= value
,
1929 qmp_migrate_set_parameters(&p
, errp
);
1932 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1934 return migrate_xbzrle_cache_size();
1937 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1939 MigrateSetParameters p
= {
1940 .has_max_bandwidth
= true,
1941 .max_bandwidth
= value
,
1944 qmp_migrate_set_parameters(&p
, errp
);
1947 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1949 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
1950 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1951 "the range of 0 to %d seconds",
1952 MAX_MIGRATE_DOWNTIME_SECONDS
);
1956 value
*= 1000; /* Convert to milliseconds */
1957 value
= MAX(0, MIN(INT64_MAX
, value
));
1959 MigrateSetParameters p
= {
1960 .has_downtime_limit
= true,
1961 .downtime_limit
= value
,
1964 qmp_migrate_set_parameters(&p
, errp
);
1967 bool migrate_release_ram(void)
1971 s
= migrate_get_current();
1973 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
1976 bool migrate_postcopy_ram(void)
1980 s
= migrate_get_current();
1982 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1985 bool migrate_postcopy(void)
1987 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
1990 bool migrate_auto_converge(void)
1994 s
= migrate_get_current();
1996 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1999 bool migrate_zero_blocks(void)
2003 s
= migrate_get_current();
2005 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
2008 bool migrate_postcopy_blocktime(void)
2012 s
= migrate_get_current();
2014 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
2017 bool migrate_use_compression(void)
2021 s
= migrate_get_current();
2023 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
2026 int migrate_compress_level(void)
2030 s
= migrate_get_current();
2032 return s
->parameters
.compress_level
;
2035 int migrate_compress_threads(void)
2039 s
= migrate_get_current();
2041 return s
->parameters
.compress_threads
;
2044 int migrate_compress_wait_thread(void)
2048 s
= migrate_get_current();
2050 return s
->parameters
.compress_wait_thread
;
2053 int migrate_decompress_threads(void)
2057 s
= migrate_get_current();
2059 return s
->parameters
.decompress_threads
;
2062 bool migrate_dirty_bitmaps(void)
2066 s
= migrate_get_current();
2068 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
2071 bool migrate_use_events(void)
2075 s
= migrate_get_current();
2077 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
2080 bool migrate_use_multifd(void)
2084 s
= migrate_get_current();
2086 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_MULTIFD
];
2089 bool migrate_pause_before_switchover(void)
2093 s
= migrate_get_current();
2095 return s
->enabled_capabilities
[
2096 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
2099 int migrate_multifd_channels(void)
2103 s
= migrate_get_current();
2105 return s
->parameters
.x_multifd_channels
;
2108 int migrate_multifd_page_count(void)
2112 s
= migrate_get_current();
2114 return s
->parameters
.x_multifd_page_count
;
2117 int migrate_use_xbzrle(void)
2121 s
= migrate_get_current();
2123 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
2126 int64_t migrate_xbzrle_cache_size(void)
2130 s
= migrate_get_current();
2132 return s
->parameters
.xbzrle_cache_size
;
2135 static int64_t migrate_max_postcopy_bandwidth(void)
2139 s
= migrate_get_current();
2141 return s
->parameters
.max_postcopy_bandwidth
;
2144 bool migrate_use_block(void)
2148 s
= migrate_get_current();
2150 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
2153 bool migrate_use_return_path(void)
2157 s
= migrate_get_current();
2159 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
2162 bool migrate_use_block_incremental(void)
2166 s
= migrate_get_current();
2168 return s
->parameters
.block_incremental
;
2171 /* migration thread support */
2173 * Something bad happened to the RP stream, mark an error
2174 * The caller shall print or trace something to indicate why
2176 static void mark_source_rp_bad(MigrationState
*s
)
2178 s
->rp_state
.error
= true;
2181 static struct rp_cmd_args
{
2182 ssize_t len
; /* -1 = variable */
2185 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
2186 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
2187 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
2188 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
2189 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
2190 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
2191 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
2192 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
2196 * Process a request for pages received on the return path,
2197 * We're allowed to send more than requested (e.g. to round to our page size)
2198 * and we don't need to send pages that have already been sent.
2200 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
2201 ram_addr_t start
, size_t len
)
2203 long our_host_ps
= getpagesize();
2205 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
2208 * Since we currently insist on matching page sizes, just sanity check
2209 * we're being asked for whole host pages.
2211 if (start
& (our_host_ps
-1) ||
2212 (len
& (our_host_ps
-1))) {
2213 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2214 " len: %zd", __func__
, start
, len
);
2215 mark_source_rp_bad(ms
);
2219 if (ram_save_queue_pages(rbname
, start
, len
)) {
2220 mark_source_rp_bad(ms
);
2224 /* Return true to retry, false to quit */
2225 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
2227 trace_postcopy_pause_return_path();
2229 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
2231 trace_postcopy_pause_return_path_continued();
2236 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
2238 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
2241 error_report("%s: invalid block name '%s'", __func__
, block_name
);
2245 /* Fetch the received bitmap and refresh the dirty bitmap */
2246 return ram_dirty_bitmap_reload(s
, block
);
2249 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
2251 trace_source_return_path_thread_resume_ack(value
);
2253 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
2254 error_report("%s: illegal resume_ack value %"PRIu32
,
2259 /* Now both sides are active. */
2260 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2261 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2263 /* Notify send thread that time to continue send pages */
2264 qemu_sem_post(&s
->rp_state
.rp_sem
);
2270 * Handles messages sent on the return path towards the source VM
2273 static void *source_return_path_thread(void *opaque
)
2275 MigrationState
*ms
= opaque
;
2276 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
2277 uint16_t header_len
, header_type
;
2279 uint32_t tmp32
, sibling_error
;
2280 ram_addr_t start
= 0; /* =0 to silence warning */
2281 size_t len
= 0, expected_len
;
2284 trace_source_return_path_thread_entry();
2285 rcu_register_thread();
2288 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
2289 migration_is_setup_or_active(ms
->state
)) {
2290 trace_source_return_path_thread_loop_top();
2291 header_type
= qemu_get_be16(rp
);
2292 header_len
= qemu_get_be16(rp
);
2294 if (qemu_file_get_error(rp
)) {
2295 mark_source_rp_bad(ms
);
2299 if (header_type
>= MIG_RP_MSG_MAX
||
2300 header_type
== MIG_RP_MSG_INVALID
) {
2301 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2302 header_type
, header_len
);
2303 mark_source_rp_bad(ms
);
2307 if ((rp_cmd_args
[header_type
].len
!= -1 &&
2308 header_len
!= rp_cmd_args
[header_type
].len
) ||
2309 header_len
> sizeof(buf
)) {
2310 error_report("RP: Received '%s' message (0x%04x) with"
2311 "incorrect length %d expecting %zu",
2312 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
2313 (size_t)rp_cmd_args
[header_type
].len
);
2314 mark_source_rp_bad(ms
);
2318 /* We know we've got a valid header by this point */
2319 res
= qemu_get_buffer(rp
, buf
, header_len
);
2320 if (res
!= header_len
) {
2321 error_report("RP: Failed reading data for message 0x%04x"
2322 " read %d expected %d",
2323 header_type
, res
, header_len
);
2324 mark_source_rp_bad(ms
);
2328 /* OK, we have the message and the data */
2329 switch (header_type
) {
2330 case MIG_RP_MSG_SHUT
:
2331 sibling_error
= ldl_be_p(buf
);
2332 trace_source_return_path_thread_shut(sibling_error
);
2333 if (sibling_error
) {
2334 error_report("RP: Sibling indicated error %d", sibling_error
);
2335 mark_source_rp_bad(ms
);
2338 * We'll let the main thread deal with closing the RP
2339 * we could do a shutdown(2) on it, but we're the only user
2340 * anyway, so there's nothing gained.
2344 case MIG_RP_MSG_PONG
:
2345 tmp32
= ldl_be_p(buf
);
2346 trace_source_return_path_thread_pong(tmp32
);
2349 case MIG_RP_MSG_REQ_PAGES
:
2350 start
= ldq_be_p(buf
);
2351 len
= ldl_be_p(buf
+ 8);
2352 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
2355 case MIG_RP_MSG_REQ_PAGES_ID
:
2356 expected_len
= 12 + 1; /* header + termination */
2358 if (header_len
>= expected_len
) {
2359 start
= ldq_be_p(buf
);
2360 len
= ldl_be_p(buf
+ 8);
2361 /* Now we expect an idstr */
2362 tmp32
= buf
[12]; /* Length of the following idstr */
2363 buf
[13 + tmp32
] = '\0';
2364 expected_len
+= tmp32
;
2366 if (header_len
!= expected_len
) {
2367 error_report("RP: Req_Page_id with length %d expecting %zd",
2368 header_len
, expected_len
);
2369 mark_source_rp_bad(ms
);
2372 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
2375 case MIG_RP_MSG_RECV_BITMAP
:
2376 if (header_len
< 1) {
2377 error_report("%s: missing block name", __func__
);
2378 mark_source_rp_bad(ms
);
2381 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2382 buf
[buf
[0] + 1] = '\0';
2383 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
2384 mark_source_rp_bad(ms
);
2389 case MIG_RP_MSG_RESUME_ACK
:
2390 tmp32
= ldl_be_p(buf
);
2391 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
2392 mark_source_rp_bad(ms
);
2403 res
= qemu_file_get_error(rp
);
2407 * Maybe there is something we can do: it looks like a
2408 * network down issue, and we pause for a recovery.
2410 if (postcopy_pause_return_path_thread(ms
)) {
2411 /* Reload rp, reset the rest */
2412 if (rp
!= ms
->rp_state
.from_dst_file
) {
2414 rp
= ms
->rp_state
.from_dst_file
;
2416 ms
->rp_state
.error
= false;
2421 trace_source_return_path_thread_bad_end();
2422 mark_source_rp_bad(ms
);
2425 trace_source_return_path_thread_end();
2426 ms
->rp_state
.from_dst_file
= NULL
;
2428 rcu_unregister_thread();
2432 static int open_return_path_on_source(MigrationState
*ms
,
2436 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
2437 if (!ms
->rp_state
.from_dst_file
) {
2441 trace_open_return_path_on_source();
2443 if (!create_thread
) {
2448 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
2449 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
2451 trace_open_return_path_on_source_continue();
2456 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2457 static int await_return_path_close_on_source(MigrationState
*ms
)
2460 * If this is a normal exit then the destination will send a SHUT and the
2461 * rp_thread will exit, however if there's an error we need to cause
2464 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
2466 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2467 * waiting for the destination.
2469 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
2470 mark_source_rp_bad(ms
);
2472 trace_await_return_path_close_on_source_joining();
2473 qemu_thread_join(&ms
->rp_state
.rp_thread
);
2474 trace_await_return_path_close_on_source_close();
2475 return ms
->rp_state
.error
;
2479 * Switch from normal iteration to postcopy
2480 * Returns non-0 on error
2482 static int postcopy_start(MigrationState
*ms
)
2485 QIOChannelBuffer
*bioc
;
2487 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2488 int64_t bandwidth
= migrate_max_postcopy_bandwidth();
2489 bool restart_block
= false;
2490 int cur_state
= MIGRATION_STATUS_ACTIVE
;
2491 if (!migrate_pause_before_switchover()) {
2492 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
2493 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2496 trace_postcopy_start();
2497 qemu_mutex_lock_iothread();
2498 trace_postcopy_start_set_run();
2500 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2501 global_state_store();
2502 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2507 ret
= migration_maybe_pause(ms
, &cur_state
,
2508 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2513 ret
= bdrv_inactivate_all();
2517 restart_block
= true;
2520 * Cause any non-postcopiable, but iterative devices to
2521 * send out their final data.
2523 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2526 * in Finish migrate and with the io-lock held everything should
2527 * be quiet, but we've potentially still got dirty pages and we
2528 * need to tell the destination to throw any pages it's already received
2531 if (migrate_postcopy_ram()) {
2532 if (ram_postcopy_send_discard_bitmap(ms
)) {
2533 error_report("postcopy send discard bitmap failed");
2539 * send rest of state - note things that are doing postcopy
2540 * will notice we're in POSTCOPY_ACTIVE and not actually
2541 * wrap their state up here
2543 /* 0 max-postcopy-bandwidth means unlimited */
2545 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
2547 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
2549 if (migrate_postcopy_ram()) {
2550 /* Ping just for debugging, helps line traces up */
2551 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
2555 * While loading the device state we may trigger page transfer
2556 * requests and the fd must be free to process those, and thus
2557 * the destination must read the whole device state off the fd before
2558 * it starts processing it. Unfortunately the ad-hoc migration format
2559 * doesn't allow the destination to know the size to read without fully
2560 * parsing it through each devices load-state code (especially the open
2561 * coded devices that use get/put).
2562 * So we wrap the device state up in a package with a length at the start;
2563 * to do this we use a qemu_buf to hold the whole of the device state.
2565 bioc
= qio_channel_buffer_new(4096);
2566 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
2567 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
2568 object_unref(OBJECT(bioc
));
2571 * Make sure the receiver can get incoming pages before we send the rest
2574 qemu_savevm_send_postcopy_listen(fb
);
2576 qemu_savevm_state_complete_precopy(fb
, false, false);
2577 if (migrate_postcopy_ram()) {
2578 qemu_savevm_send_ping(fb
, 3);
2581 qemu_savevm_send_postcopy_run(fb
);
2583 /* <><> end of stuff going into the package */
2585 /* Last point of recovery; as soon as we send the package the destination
2586 * can open devices and potentially start running.
2587 * Lets just check again we've not got any errors.
2589 ret
= qemu_file_get_error(ms
->to_dst_file
);
2591 error_report("postcopy_start: Migration stream errored (pre package)");
2595 restart_block
= false;
2597 /* Now send that blob */
2598 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2603 /* Send a notify to give a chance for anything that needs to happen
2604 * at the transition to postcopy and after the device state; in particular
2605 * spice needs to trigger a transition now
2607 ms
->postcopy_after_devices
= true;
2608 notifier_list_notify(&migration_state_notifiers
, ms
);
2610 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2612 qemu_mutex_unlock_iothread();
2614 if (migrate_postcopy_ram()) {
2616 * Although this ping is just for debug, it could potentially be
2617 * used for getting a better measurement of downtime at the source.
2619 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2622 if (migrate_release_ram()) {
2623 ram_postcopy_migrated_memory_release(ms
);
2626 ret
= qemu_file_get_error(ms
->to_dst_file
);
2628 error_report("postcopy_start: Migration stream errored");
2629 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2630 MIGRATION_STATUS_FAILED
);
2638 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2639 MIGRATION_STATUS_FAILED
);
2640 if (restart_block
) {
2641 /* A failure happened early enough that we know the destination hasn't
2642 * accessed block devices, so we're safe to recover.
2644 Error
*local_err
= NULL
;
2646 bdrv_invalidate_cache_all(&local_err
);
2648 error_report_err(local_err
);
2651 qemu_mutex_unlock_iothread();
2656 * migration_maybe_pause: Pause if required to by
2657 * migrate_pause_before_switchover called with the iothread locked
2658 * Returns: 0 on success
2660 static int migration_maybe_pause(MigrationState
*s
,
2661 int *current_active_state
,
2664 if (!migrate_pause_before_switchover()) {
2668 /* Since leaving this state is not atomic with posting the semaphore
2669 * it's possible that someone could have issued multiple migrate_continue
2670 * and the semaphore is incorrectly positive at this point;
2671 * the docs say it's undefined to reinit a semaphore that's already
2672 * init'd, so use timedwait to eat up any existing posts.
2674 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2675 /* This block intentionally left blank */
2678 qemu_mutex_unlock_iothread();
2679 migrate_set_state(&s
->state
, *current_active_state
,
2680 MIGRATION_STATUS_PRE_SWITCHOVER
);
2681 qemu_sem_wait(&s
->pause_sem
);
2682 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2684 *current_active_state
= new_state
;
2685 qemu_mutex_lock_iothread();
2687 return s
->state
== new_state
? 0 : -EINVAL
;
2691 * migration_completion: Used by migration_thread when there's not much left.
2692 * The caller 'breaks' the loop when this returns.
2694 * @s: Current migration state
2696 static void migration_completion(MigrationState
*s
)
2699 int current_active_state
= s
->state
;
2701 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2702 qemu_mutex_lock_iothread();
2703 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2704 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2705 s
->vm_was_running
= runstate_is_running();
2706 ret
= global_state_store();
2709 bool inactivate
= !migrate_colo_enabled();
2710 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2712 ret
= migration_maybe_pause(s
, ¤t_active_state
,
2713 MIGRATION_STATUS_DEVICE
);
2716 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
2717 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
2720 if (inactivate
&& ret
>= 0) {
2721 s
->block_inactive
= true;
2724 qemu_mutex_unlock_iothread();
2729 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2730 trace_migration_completion_postcopy_end();
2732 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
2733 trace_migration_completion_postcopy_end_after_complete();
2737 * If rp was opened we must clean up the thread before
2738 * cleaning everything else up (since if there are no failures
2739 * it will wait for the destination to send it's status in
2742 if (s
->rp_state
.from_dst_file
) {
2744 trace_migration_return_path_end_before();
2745 rp_error
= await_return_path_close_on_source(s
);
2746 trace_migration_return_path_end_after(rp_error
);
2748 goto fail_invalidate
;
2752 if (qemu_file_get_error(s
->to_dst_file
)) {
2753 trace_migration_completion_file_err();
2754 goto fail_invalidate
;
2757 if (!migrate_colo_enabled()) {
2758 migrate_set_state(&s
->state
, current_active_state
,
2759 MIGRATION_STATUS_COMPLETED
);
2765 /* If not doing postcopy, vm_start() will be called: let's regain
2766 * control on images.
2768 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
2769 s
->state
== MIGRATION_STATUS_DEVICE
) {
2770 Error
*local_err
= NULL
;
2772 qemu_mutex_lock_iothread();
2773 bdrv_invalidate_cache_all(&local_err
);
2775 error_report_err(local_err
);
2777 s
->block_inactive
= false;
2779 qemu_mutex_unlock_iothread();
2783 migrate_set_state(&s
->state
, current_active_state
,
2784 MIGRATION_STATUS_FAILED
);
2787 bool migrate_colo_enabled(void)
2789 MigrationState
*s
= migrate_get_current();
2790 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
2793 typedef enum MigThrError
{
2794 /* No error detected */
2795 MIG_THR_ERR_NONE
= 0,
2796 /* Detected error, but resumed successfully */
2797 MIG_THR_ERR_RECOVERED
= 1,
2798 /* Detected fatal error, need to exit */
2799 MIG_THR_ERR_FATAL
= 2,
2802 static int postcopy_resume_handshake(MigrationState
*s
)
2804 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
2806 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2807 qemu_sem_wait(&s
->rp_state
.rp_sem
);
2810 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2817 /* Return zero if success, or <0 for error */
2818 static int postcopy_do_resume(MigrationState
*s
)
2823 * Call all the resume_prepare() hooks, so that modules can be
2824 * ready for the migration resume.
2826 ret
= qemu_savevm_state_resume_prepare(s
);
2828 error_report("%s: resume_prepare() failure detected: %d",
2834 * Last handshake with destination on the resume (destination will
2835 * switch to postcopy-active afterwards)
2837 ret
= postcopy_resume_handshake(s
);
2839 error_report("%s: handshake failed: %d", __func__
, ret
);
2847 * We don't return until we are in a safe state to continue current
2848 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2849 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2851 static MigThrError
postcopy_pause(MigrationState
*s
)
2853 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2858 migrate_set_state(&s
->state
, s
->state
,
2859 MIGRATION_STATUS_POSTCOPY_PAUSED
);
2861 /* Current channel is possibly broken. Release it. */
2862 assert(s
->to_dst_file
);
2863 qemu_mutex_lock(&s
->qemu_file_lock
);
2864 file
= s
->to_dst_file
;
2865 s
->to_dst_file
= NULL
;
2866 qemu_mutex_unlock(&s
->qemu_file_lock
);
2868 qemu_file_shutdown(file
);
2871 error_report("Detected IO failure for postcopy. "
2872 "Migration paused.");
2875 * We wait until things fixed up. Then someone will setup the
2876 * status back for us.
2878 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2879 qemu_sem_wait(&s
->postcopy_pause_sem
);
2882 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2883 /* Woken up by a recover procedure. Give it a shot */
2886 * Firstly, let's wake up the return path now, with a new
2887 * return path channel.
2889 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
2891 /* Do the resume logic */
2892 if (postcopy_do_resume(s
) == 0) {
2893 /* Let's continue! */
2894 trace_postcopy_pause_continued();
2895 return MIG_THR_ERR_RECOVERED
;
2898 * Something wrong happened during the recovery, let's
2899 * pause again. Pause is always better than throwing
2905 /* This is not right... Time to quit. */
2906 return MIG_THR_ERR_FATAL
;
2911 static MigThrError
migration_detect_error(MigrationState
*s
)
2915 /* Try to detect any file errors */
2916 ret
= qemu_file_get_error(s
->to_dst_file
);
2919 /* Everything is fine */
2920 return MIG_THR_ERR_NONE
;
2923 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
== -EIO
) {
2925 * For postcopy, we allow the network to be down for a
2926 * while. After that, it can be continued by a
2929 return postcopy_pause(s
);
2932 * For precopy (or postcopy with error outside IO), we fail
2935 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
2936 trace_migration_thread_file_err();
2938 /* Time to stop the migration, now. */
2939 return MIG_THR_ERR_FATAL
;
2943 /* How many bytes have we transferred since the beggining of the migration */
2944 static uint64_t migration_total_bytes(MigrationState
*s
)
2946 return qemu_ftell(s
->to_dst_file
) + ram_counters
.multifd_bytes
;
2949 static void migration_calculate_complete(MigrationState
*s
)
2951 uint64_t bytes
= migration_total_bytes(s
);
2952 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2953 int64_t transfer_time
;
2955 s
->total_time
= end_time
- s
->start_time
;
2958 * It's still not set, so we are precopy migration. For
2959 * postcopy, downtime is calculated during postcopy_start().
2961 s
->downtime
= end_time
- s
->downtime_start
;
2964 transfer_time
= s
->total_time
- s
->setup_time
;
2965 if (transfer_time
) {
2966 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
2970 static void migration_update_counters(MigrationState
*s
,
2971 int64_t current_time
)
2973 uint64_t transferred
, transferred_pages
, time_spent
;
2974 uint64_t current_bytes
; /* bytes transferred since the beginning */
2977 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
2981 current_bytes
= migration_total_bytes(s
);
2982 transferred
= current_bytes
- s
->iteration_initial_bytes
;
2983 time_spent
= current_time
- s
->iteration_start_time
;
2984 bandwidth
= (double)transferred
/ time_spent
;
2985 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
2987 s
->mbps
= (((double) transferred
* 8.0) /
2988 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2990 transferred_pages
= ram_get_total_transferred_pages() -
2991 s
->iteration_initial_pages
;
2992 s
->pages_per_second
= (double) transferred_pages
/
2993 (((double) time_spent
/ 1000.0));
2996 * if we haven't sent anything, we don't want to
2997 * recalculate. 10000 is a small enough number for our purposes
2999 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
3000 s
->expected_downtime
= ram_counters
.remaining
/ bandwidth
;
3003 qemu_file_reset_rate_limit(s
->to_dst_file
);
3005 s
->iteration_start_time
= current_time
;
3006 s
->iteration_initial_bytes
= current_bytes
;
3007 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
3009 trace_migrate_transferred(transferred
, time_spent
,
3010 bandwidth
, s
->threshold_size
);
3013 /* Migration thread iteration status */
3015 MIG_ITERATE_RESUME
, /* Resume current iteration */
3016 MIG_ITERATE_SKIP
, /* Skip current iteration */
3017 MIG_ITERATE_BREAK
, /* Break the loop */
3021 * Return true if continue to the next iteration directly, false
3024 static MigIterateState
migration_iteration_run(MigrationState
*s
)
3026 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
3027 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
3029 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
3030 &pend_compat
, &pend_post
);
3031 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
3033 trace_migrate_pending(pending_size
, s
->threshold_size
,
3034 pend_pre
, pend_compat
, pend_post
);
3036 if (pending_size
&& pending_size
>= s
->threshold_size
) {
3037 /* Still a significant amount to transfer */
3038 if (migrate_postcopy() && !in_postcopy
&&
3039 pend_pre
<= s
->threshold_size
&&
3040 atomic_read(&s
->start_postcopy
)) {
3041 if (postcopy_start(s
)) {
3042 error_report("%s: postcopy failed to start", __func__
);
3044 return MIG_ITERATE_SKIP
;
3046 /* Just another iteration step */
3047 qemu_savevm_state_iterate(s
->to_dst_file
,
3048 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3050 trace_migration_thread_low_pending(pending_size
);
3051 migration_completion(s
);
3052 return MIG_ITERATE_BREAK
;
3055 return MIG_ITERATE_RESUME
;
3058 static void migration_iteration_finish(MigrationState
*s
)
3060 /* If we enabled cpu throttling for auto-converge, turn it off. */
3061 cpu_throttle_stop();
3063 qemu_mutex_lock_iothread();
3065 case MIGRATION_STATUS_COMPLETED
:
3066 migration_calculate_complete(s
);
3067 runstate_set(RUN_STATE_POSTMIGRATE
);
3070 case MIGRATION_STATUS_ACTIVE
:
3072 * We should really assert here, but since it's during
3073 * migration, let's try to reduce the usage of assertions.
3075 if (!migrate_colo_enabled()) {
3076 error_report("%s: critical error: calling COLO code without "
3077 "COLO enabled", __func__
);
3079 migrate_start_colo_process(s
);
3081 * Fixme: we will run VM in COLO no matter its old running state.
3082 * After exited COLO, we will keep running.
3084 s
->vm_was_running
= true;
3086 case MIGRATION_STATUS_FAILED
:
3087 case MIGRATION_STATUS_CANCELLED
:
3088 case MIGRATION_STATUS_CANCELLING
:
3089 if (s
->vm_was_running
) {
3092 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
3093 runstate_set(RUN_STATE_POSTMIGRATE
);
3099 /* Should not reach here, but if so, forgive the VM. */
3100 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3103 qemu_bh_schedule(s
->cleanup_bh
);
3104 qemu_mutex_unlock_iothread();
3107 void migration_make_urgent_request(void)
3109 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
3112 void migration_consume_urgent_request(void)
3114 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
3118 * Master migration thread on the source VM.
3119 * It drives the migration and pumps the data down the outgoing channel.
3121 static void *migration_thread(void *opaque
)
3123 MigrationState
*s
= opaque
;
3124 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3125 MigThrError thr_error
;
3126 bool urgent
= false;
3128 rcu_register_thread();
3130 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3132 qemu_savevm_state_header(s
->to_dst_file
);
3135 * If we opened the return path, we need to make sure dst has it
3138 if (s
->rp_state
.from_dst_file
) {
3139 /* Now tell the dest that it should open its end so it can reply */
3140 qemu_savevm_send_open_return_path(s
->to_dst_file
);
3142 /* And do a ping that will make stuff easier to debug */
3143 qemu_savevm_send_ping(s
->to_dst_file
, 1);
3146 if (migrate_postcopy()) {
3148 * Tell the destination that we *might* want to do postcopy later;
3149 * if the other end can't do postcopy it should fail now, nice and
3152 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
3155 if (migrate_colo_enabled()) {
3156 /* Notify migration destination that we enable COLO */
3157 qemu_savevm_send_colo_enable(s
->to_dst_file
);
3160 qemu_savevm_state_setup(s
->to_dst_file
);
3162 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3163 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3164 MIGRATION_STATUS_ACTIVE
);
3166 trace_migration_thread_setup_complete();
3168 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
3169 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3170 int64_t current_time
;
3172 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
3173 MigIterateState iter_state
= migration_iteration_run(s
);
3174 if (iter_state
== MIG_ITERATE_SKIP
) {
3176 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3182 * Try to detect any kind of failures, and see whether we
3183 * should stop the migration now.
3185 thr_error
= migration_detect_error(s
);
3186 if (thr_error
== MIG_THR_ERR_FATAL
) {
3187 /* Stop migration */
3189 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
3191 * Just recovered from a e.g. network failure, reset all
3192 * the local variables. This is important to avoid
3193 * breaking transferred_bytes and bandwidth calculation
3195 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3196 s
->iteration_initial_bytes
= 0;
3199 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3201 migration_update_counters(s
, current_time
);
3204 if (qemu_file_rate_limit(s
->to_dst_file
)) {
3205 /* Wait for a delay to do rate limiting OR
3206 * something urgent to post the semaphore.
3208 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- current_time
;
3209 trace_migration_thread_ratelimit_pre(ms
);
3210 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
3211 /* We were worken by one or more urgent things but
3212 * the timedwait will have consumed one of them.
3213 * The service routine for the urgent wake will dec
3214 * the semaphore itself for each item it consumes,
3215 * so add this one we just eat back.
3217 qemu_sem_post(&s
->rate_limit_sem
);
3220 trace_migration_thread_ratelimit_post(urgent
);
3224 trace_migration_thread_after_loop();
3225 migration_iteration_finish(s
);
3226 rcu_unregister_thread();
3230 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
3233 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
3235 s
->expected_downtime
= s
->parameters
.downtime_limit
;
3236 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
3238 migrate_fd_error(s
, error_in
);
3239 migrate_fd_cleanup(s
);
3244 /* This is a resumed migration */
3245 rate_limit
= INT64_MAX
;
3247 /* This is a fresh new migration */
3248 rate_limit
= s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
;
3250 /* Notify before starting migration thread */
3251 notifier_list_notify(&migration_state_notifiers
, s
);
3254 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
3255 qemu_file_set_blocking(s
->to_dst_file
, true);
3258 * Open the return path. For postcopy, it is used exclusively. For
3259 * precopy, only if user specified "return-path" capability would
3260 * QEMU uses the return path.
3262 if (migrate_postcopy_ram() || migrate_use_return_path()) {
3263 if (open_return_path_on_source(s
, !resume
)) {
3264 error_report("Unable to open return-path for postcopy");
3265 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
3266 migrate_fd_cleanup(s
);
3272 /* Wakeup the main migration thread to do the recovery */
3273 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
3274 MIGRATION_STATUS_POSTCOPY_RECOVER
);
3275 qemu_sem_post(&s
->postcopy_pause_sem
);
3279 if (multifd_save_setup() != 0) {
3280 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3281 MIGRATION_STATUS_FAILED
);
3282 migrate_fd_cleanup(s
);
3285 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
3286 QEMU_THREAD_JOINABLE
);
3287 s
->migration_thread_running
= true;
3290 void migration_global_dump(Monitor
*mon
)
3292 MigrationState
*ms
= migrate_get_current();
3294 monitor_printf(mon
, "globals:\n");
3295 monitor_printf(mon
, "store-global-state: %s\n",
3296 ms
->store_global_state
? "on" : "off");
3297 monitor_printf(mon
, "only-migratable: %s\n",
3298 ms
->only_migratable
? "on" : "off");
3299 monitor_printf(mon
, "send-configuration: %s\n",
3300 ms
->send_configuration
? "on" : "off");
3301 monitor_printf(mon
, "send-section-footer: %s\n",
3302 ms
->send_section_footer
? "on" : "off");
3303 monitor_printf(mon
, "decompress-error-check: %s\n",
3304 ms
->decompress_error_check
? "on" : "off");
3307 #define DEFINE_PROP_MIG_CAP(name, x) \
3308 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3310 static Property migration_properties
[] = {
3311 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
3312 store_global_state
, true),
3313 DEFINE_PROP_BOOL("only-migratable", MigrationState
, only_migratable
, false),
3314 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
3315 send_configuration
, true),
3316 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
3317 send_section_footer
, true),
3318 DEFINE_PROP_BOOL("decompress-error-check", MigrationState
,
3319 decompress_error_check
, true),
3321 /* Migration parameters */
3322 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
3323 parameters
.compress_level
,
3324 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
3325 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
3326 parameters
.compress_threads
,
3327 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
3328 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState
,
3329 parameters
.compress_wait_thread
, true),
3330 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
3331 parameters
.decompress_threads
,
3332 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
3333 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
3334 parameters
.cpu_throttle_initial
,
3335 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
3336 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
3337 parameters
.cpu_throttle_increment
,
3338 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
3339 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
3340 parameters
.max_bandwidth
, MAX_THROTTLE
),
3341 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
3342 parameters
.downtime_limit
,
3343 DEFAULT_MIGRATE_SET_DOWNTIME
),
3344 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
3345 parameters
.x_checkpoint_delay
,
3346 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
3347 DEFINE_PROP_UINT8("x-multifd-channels", MigrationState
,
3348 parameters
.x_multifd_channels
,
3349 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
3350 DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState
,
3351 parameters
.x_multifd_page_count
,
3352 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT
),
3353 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
3354 parameters
.xbzrle_cache_size
,
3355 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
3356 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState
,
3357 parameters
.max_postcopy_bandwidth
,
3358 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH
),
3359 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState
,
3360 parameters
.max_cpu_throttle
,
3361 DEFAULT_MIGRATE_MAX_CPU_THROTTLE
),
3362 DEFINE_PROP_SIZE("announce-initial", MigrationState
,
3363 parameters
.announce_initial
,
3364 DEFAULT_MIGRATE_ANNOUNCE_INITIAL
),
3365 DEFINE_PROP_SIZE("announce-max", MigrationState
,
3366 parameters
.announce_max
,
3367 DEFAULT_MIGRATE_ANNOUNCE_MAX
),
3368 DEFINE_PROP_SIZE("announce-rounds", MigrationState
,
3369 parameters
.announce_rounds
,
3370 DEFAULT_MIGRATE_ANNOUNCE_ROUNDS
),
3371 DEFINE_PROP_SIZE("announce-step", MigrationState
,
3372 parameters
.announce_step
,
3373 DEFAULT_MIGRATE_ANNOUNCE_STEP
),
3375 /* Migration capabilities */
3376 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
3377 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
3378 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
3379 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
3380 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
3381 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
3382 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
3383 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
3384 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
3385 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
3386 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
3387 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD
),
3389 DEFINE_PROP_END_OF_LIST(),
3392 static void migration_class_init(ObjectClass
*klass
, void *data
)
3394 DeviceClass
*dc
= DEVICE_CLASS(klass
);
3396 dc
->user_creatable
= false;
3397 dc
->props
= migration_properties
;
3400 static void migration_instance_finalize(Object
*obj
)
3402 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3403 MigrationParameters
*params
= &ms
->parameters
;
3405 qemu_mutex_destroy(&ms
->error_mutex
);
3406 qemu_mutex_destroy(&ms
->qemu_file_lock
);
3407 g_free(params
->tls_hostname
);
3408 g_free(params
->tls_creds
);
3409 qemu_sem_destroy(&ms
->rate_limit_sem
);
3410 qemu_sem_destroy(&ms
->pause_sem
);
3411 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
3412 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
3413 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
3414 error_free(ms
->error
);
3417 static void migration_instance_init(Object
*obj
)
3419 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3420 MigrationParameters
*params
= &ms
->parameters
;
3422 ms
->state
= MIGRATION_STATUS_NONE
;
3424 ms
->pages_per_second
= -1;
3425 qemu_sem_init(&ms
->pause_sem
, 0);
3426 qemu_mutex_init(&ms
->error_mutex
);
3428 params
->tls_hostname
= g_strdup("");
3429 params
->tls_creds
= g_strdup("");
3431 /* Set has_* up only for parameter checks */
3432 params
->has_compress_level
= true;
3433 params
->has_compress_threads
= true;
3434 params
->has_decompress_threads
= true;
3435 params
->has_cpu_throttle_initial
= true;
3436 params
->has_cpu_throttle_increment
= true;
3437 params
->has_max_bandwidth
= true;
3438 params
->has_downtime_limit
= true;
3439 params
->has_x_checkpoint_delay
= true;
3440 params
->has_block_incremental
= true;
3441 params
->has_x_multifd_channels
= true;
3442 params
->has_x_multifd_page_count
= true;
3443 params
->has_xbzrle_cache_size
= true;
3444 params
->has_max_postcopy_bandwidth
= true;
3445 params
->has_max_cpu_throttle
= true;
3446 params
->has_announce_initial
= true;
3447 params
->has_announce_max
= true;
3448 params
->has_announce_rounds
= true;
3449 params
->has_announce_step
= true;
3451 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
3452 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
3453 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
3454 qemu_sem_init(&ms
->rate_limit_sem
, 0);
3455 qemu_mutex_init(&ms
->qemu_file_lock
);
3459 * Return true if check pass, false otherwise. Error will be put
3460 * inside errp if provided.
3462 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
3464 MigrationCapabilityStatusList
*head
= NULL
;
3465 /* Assuming all off */
3466 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
3469 if (!migrate_params_check(&ms
->parameters
, errp
)) {
3473 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
3474 if (ms
->enabled_capabilities
[i
]) {
3475 head
= migrate_cap_add(head
, i
, true);
3479 ret
= migrate_caps_check(cap_list
, head
, errp
);
3481 /* It works with head == NULL */
3482 qapi_free_MigrationCapabilityStatusList(head
);
3487 static const TypeInfo migration_type
= {
3488 .name
= TYPE_MIGRATION
,
3490 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3491 * not created using qdev_create(), it is not attached to the qdev
3492 * device tree, and it is never realized.
3494 * TODO: Make this TYPE_OBJECT once QOM provides something like
3495 * TYPE_DEVICE's "-global" properties.
3497 .parent
= TYPE_DEVICE
,
3498 .class_init
= migration_class_init
,
3499 .class_size
= sizeof(MigrationClass
),
3500 .instance_size
= sizeof(MigrationState
),
3501 .instance_init
= migration_instance_init
,
3502 .instance_finalize
= migration_instance_finalize
,
3505 static void register_migration_types(void)
3507 type_register_static(&migration_type
);
3510 type_init(register_migration_types
);