4 * Copyright IBM, Corp. 2008
7 * Anthony Liguori <aliguori@us.ibm.com>
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
16 #include "qemu/osdep.h"
17 #include "qemu/cutils.h"
18 #include "qemu/error-report.h"
19 #include "migration/blocker.h"
25 #include "migration/global_state.h"
26 #include "migration/misc.h"
27 #include "migration.h"
29 #include "qemu-file-channel.h"
30 #include "qemu-file.h"
31 #include "migration/vmstate.h"
32 #include "block/block.h"
33 #include "qapi/error.h"
34 #include "qapi/qapi-commands-migration.h"
35 #include "qapi/qapi-events-migration.h"
36 #include "qapi/qmp/qerror.h"
37 #include "qapi/qmp/qnull.h"
40 #include "postcopy-ram.h"
41 #include "qemu/thread.h"
43 #include "exec/target_page.h"
44 #include "io/channel-buffer.h"
45 #include "migration/colo.h"
46 #include "hw/boards.h"
47 #include "monitor/monitor.h"
49 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
51 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
53 #define BUFFER_DELAY 100
54 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
56 /* Time in milliseconds we are allowed to stop the source,
57 * for sending the last part */
58 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
60 /* Maximum migrate downtime set to 2000 seconds */
61 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
62 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
64 /* Default compression thread count */
65 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
66 /* Default decompression thread count, usually decompression is at
67 * least 4 times as fast as compression.*/
68 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
69 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
70 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
71 /* Define default autoconverge cpu throttle migration parameters */
72 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
73 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
74 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
76 /* Migration XBZRLE default cache size */
77 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
79 /* The delay time (in ms) between two COLO checkpoints */
80 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
81 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
82 #define DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT 16
84 /* Background transfer rate for postcopy, 0 means unlimited, note
85 * that page requests can still exceed this limit.
87 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
89 static NotifierList migration_state_notifiers
=
90 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
92 static bool deferred_incoming
;
94 /* Messages sent on the return path from destination to source */
95 enum mig_rp_message_type
{
96 MIG_RP_MSG_INVALID
= 0, /* Must be 0 */
97 MIG_RP_MSG_SHUT
, /* sibling will not send any more RP messages */
98 MIG_RP_MSG_PONG
, /* Response to a PING; data (seq: be32 ) */
100 MIG_RP_MSG_REQ_PAGES_ID
, /* data (start: be64, len: be32, id: string) */
101 MIG_RP_MSG_REQ_PAGES
, /* data (start: be64, len: be32) */
102 MIG_RP_MSG_RECV_BITMAP
, /* send recved_bitmap back to source */
103 MIG_RP_MSG_RESUME_ACK
, /* tell source that we are ready to resume */
108 /* When we add fault tolerance, we could have several
109 migrations at once. For now we don't need to add
110 dynamic creation of migration */
112 static MigrationState
*current_migration
;
113 static MigrationIncomingState
*current_incoming
;
115 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
116 static int migration_maybe_pause(MigrationState
*s
,
117 int *current_active_state
,
120 void migration_object_init(void)
122 MachineState
*ms
= MACHINE(qdev_get_machine());
125 /* This can only be called once. */
126 assert(!current_migration
);
127 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
130 * Init the migrate incoming object as well no matter whether
131 * we'll use it or not.
133 assert(!current_incoming
);
134 current_incoming
= g_new0(MigrationIncomingState
, 1);
135 current_incoming
->state
= MIGRATION_STATUS_NONE
;
136 current_incoming
->postcopy_remote_fds
=
137 g_array_new(FALSE
, TRUE
, sizeof(struct PostCopyFD
));
138 qemu_mutex_init(¤t_incoming
->rp_mutex
);
139 qemu_event_init(¤t_incoming
->main_thread_load_event
, false);
140 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_dst
, 0);
141 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fault
, 0);
143 init_dirty_bitmap_incoming_migration();
145 if (!migration_object_check(current_migration
, &err
)) {
146 error_report_err(err
);
151 * We cannot really do this in migration_instance_init() since at
152 * that time global properties are not yet applied, then this
153 * value will be definitely replaced by something else.
155 if (ms
->enforce_config_section
) {
156 current_migration
->send_configuration
= true;
160 void migration_object_finalize(void)
162 object_unref(OBJECT(current_migration
));
166 MigrationState
*migrate_get_current(void)
168 /* This can only be called after the object created. */
169 assert(current_migration
);
170 return current_migration
;
173 MigrationIncomingState
*migration_incoming_get_current(void)
175 assert(current_incoming
);
176 return current_incoming
;
179 void migration_incoming_state_destroy(void)
181 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
183 if (mis
->to_src_file
) {
184 /* Tell source that we are done */
185 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
186 qemu_fclose(mis
->to_src_file
);
187 mis
->to_src_file
= NULL
;
190 if (mis
->from_src_file
) {
191 qemu_fclose(mis
->from_src_file
);
192 mis
->from_src_file
= NULL
;
194 if (mis
->postcopy_remote_fds
) {
195 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
196 mis
->postcopy_remote_fds
= NULL
;
199 qemu_event_reset(&mis
->main_thread_load_event
);
202 static void migrate_generate_event(int new_state
)
204 if (migrate_use_events()) {
205 qapi_event_send_migration(new_state
);
209 static bool migrate_late_block_activate(void)
213 s
= migrate_get_current();
215 return s
->enabled_capabilities
[
216 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE
];
220 * Called on -incoming with a defer: uri.
221 * The migration can be started later after any parameters have been
224 static void deferred_incoming_migration(Error
**errp
)
226 if (deferred_incoming
) {
227 error_setg(errp
, "Incoming migration already deferred");
229 deferred_incoming
= true;
233 * Send a message on the return channel back to the source
236 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
237 enum mig_rp_message_type message_type
,
238 uint16_t len
, void *data
)
242 trace_migrate_send_rp_message((int)message_type
, len
);
243 qemu_mutex_lock(&mis
->rp_mutex
);
246 * It's possible that the file handle got lost due to network
249 if (!mis
->to_src_file
) {
254 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
255 qemu_put_be16(mis
->to_src_file
, len
);
256 qemu_put_buffer(mis
->to_src_file
, data
, len
);
257 qemu_fflush(mis
->to_src_file
);
259 /* It's possible that qemu file got error during sending */
260 ret
= qemu_file_get_error(mis
->to_src_file
);
263 qemu_mutex_unlock(&mis
->rp_mutex
);
267 /* Request a range of pages from the source VM at the given
269 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
270 * as the last request (a name must have been given previously)
271 * Start: Address offset within the RB
272 * Len: Length in bytes required - must be a multiple of pagesize
274 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
, const char *rbname
,
275 ram_addr_t start
, size_t len
)
277 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
278 size_t msglen
= 12; /* start + len */
279 enum mig_rp_message_type msg_type
;
281 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
282 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
285 int rbname_len
= strlen(rbname
);
286 assert(rbname_len
< 256);
288 bufc
[msglen
++] = rbname_len
;
289 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
290 msglen
+= rbname_len
;
291 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
293 msg_type
= MIG_RP_MSG_REQ_PAGES
;
296 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
299 static bool migration_colo_enabled
;
300 bool migration_incoming_colo_enabled(void)
302 return migration_colo_enabled
;
305 void migration_incoming_disable_colo(void)
307 migration_colo_enabled
= false;
310 void migration_incoming_enable_colo(void)
312 migration_colo_enabled
= true;
315 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
319 qapi_event_send_migration(MIGRATION_STATUS_SETUP
);
320 if (!strcmp(uri
, "defer")) {
321 deferred_incoming_migration(errp
);
322 } else if (strstart(uri
, "tcp:", &p
)) {
323 tcp_start_incoming_migration(p
, errp
);
325 } else if (strstart(uri
, "rdma:", &p
)) {
326 rdma_start_incoming_migration(p
, errp
);
328 } else if (strstart(uri
, "exec:", &p
)) {
329 exec_start_incoming_migration(p
, errp
);
330 } else if (strstart(uri
, "unix:", &p
)) {
331 unix_start_incoming_migration(p
, errp
);
332 } else if (strstart(uri
, "fd:", &p
)) {
333 fd_start_incoming_migration(p
, errp
);
335 error_setg(errp
, "unknown migration protocol: %s", uri
);
339 static void process_incoming_migration_bh(void *opaque
)
341 Error
*local_err
= NULL
;
342 MigrationIncomingState
*mis
= opaque
;
344 /* If capability late_block_activate is set:
345 * Only fire up the block code now if we're going to restart the
346 * VM, else 'cont' will do it.
347 * This causes file locking to happen; so we don't want it to happen
348 * unless we really are starting the VM.
350 if (!migrate_late_block_activate() ||
351 (autostart
&& (!global_state_received() ||
352 global_state_get_runstate() == RUN_STATE_RUNNING
))) {
353 /* Make sure all file formats flush their mutable metadata.
354 * If we get an error here, just don't restart the VM yet. */
355 bdrv_invalidate_cache_all(&local_err
);
357 error_report_err(local_err
);
364 * This must happen after all error conditions are dealt with and
365 * we're sure the VM is going to be running on this host.
367 qemu_announce_self();
369 if (multifd_load_cleanup(&local_err
) != 0) {
370 error_report_err(local_err
);
373 /* If global state section was not received or we are in running
374 state, we need to obey autostart. Any other state is set with
377 dirty_bitmap_mig_before_vm_start();
379 if (!global_state_received() ||
380 global_state_get_runstate() == RUN_STATE_RUNNING
) {
384 runstate_set(RUN_STATE_PAUSED
);
387 runstate_set(global_state_get_runstate());
390 * This must happen after any state changes since as soon as an external
391 * observer sees this event they might start to prod at the VM assuming
394 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
395 MIGRATION_STATUS_COMPLETED
);
396 qemu_bh_delete(mis
->bh
);
397 migration_incoming_state_destroy();
400 static void process_incoming_migration_co(void *opaque
)
402 MigrationIncomingState
*mis
= migration_incoming_get_current();
405 Error
*local_err
= NULL
;
407 assert(mis
->from_src_file
);
408 mis
->migration_incoming_co
= qemu_coroutine_self();
409 mis
->largest_page_size
= qemu_ram_pagesize_largest();
410 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
411 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
412 MIGRATION_STATUS_ACTIVE
);
413 ret
= qemu_loadvm_state(mis
->from_src_file
);
415 ps
= postcopy_state_get();
416 trace_process_incoming_migration_co_end(ret
, ps
);
417 if (ps
!= POSTCOPY_INCOMING_NONE
) {
418 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
420 * Where a migration had postcopy enabled (and thus went to advise)
421 * but managed to complete within the precopy period, we can use
424 postcopy_ram_incoming_cleanup(mis
);
425 } else if (ret
>= 0) {
427 * Postcopy was started, cleanup should happen at the end of the
430 trace_process_incoming_migration_co_postcopy_end_main();
433 /* Else if something went wrong then just fall out of the normal exit */
436 /* we get COLO info, and know if we are in COLO mode */
437 if (!ret
&& migration_incoming_colo_enabled()) {
438 /* Make sure all file formats flush their mutable metadata */
439 bdrv_invalidate_cache_all(&local_err
);
441 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
442 MIGRATION_STATUS_FAILED
);
443 error_report_err(local_err
);
447 if (colo_init_ram_cache() < 0) {
448 error_report("Init ram cache failed");
452 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
453 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
454 mis
->have_colo_incoming_thread
= true;
455 qemu_coroutine_yield();
457 /* Wait checkpoint incoming thread exit before free resource */
458 qemu_thread_join(&mis
->colo_incoming_thread
);
459 /* We hold the global iothread lock, so it is safe here */
460 colo_release_ram_cache();
464 Error
*local_err
= NULL
;
466 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
467 MIGRATION_STATUS_FAILED
);
468 error_report("load of migration failed: %s", strerror(-ret
));
469 qemu_fclose(mis
->from_src_file
);
470 if (multifd_load_cleanup(&local_err
) != 0) {
471 error_report_err(local_err
);
475 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
476 qemu_bh_schedule(mis
->bh
);
477 mis
->migration_incoming_co
= NULL
;
480 static void migration_incoming_setup(QEMUFile
*f
)
482 MigrationIncomingState
*mis
= migration_incoming_get_current();
484 if (multifd_load_setup() != 0) {
485 /* We haven't been able to create multifd threads
486 nothing better to do */
490 if (!mis
->from_src_file
) {
491 mis
->from_src_file
= f
;
493 qemu_file_set_blocking(f
, false);
496 void migration_incoming_process(void)
498 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
499 qemu_coroutine_enter(co
);
502 /* Returns true if recovered from a paused migration, otherwise false */
503 static bool postcopy_try_recover(QEMUFile
*f
)
505 MigrationIncomingState
*mis
= migration_incoming_get_current();
507 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
508 /* Resumed from a paused postcopy migration */
510 mis
->from_src_file
= f
;
511 /* Postcopy has standalone thread to do vm load */
512 qemu_file_set_blocking(f
, true);
514 /* Re-configure the return path */
515 mis
->to_src_file
= qemu_file_get_return_path(f
);
517 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
518 MIGRATION_STATUS_POSTCOPY_RECOVER
);
521 * Here, we only wake up the main loading thread (while the
522 * fault thread will still be waiting), so that we can receive
523 * commands from source now, and answer it if needed. The
524 * fault thread will be woken up afterwards until we are sure
525 * that source is ready to reply to page requests.
527 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
534 void migration_fd_process_incoming(QEMUFile
*f
)
536 if (postcopy_try_recover(f
)) {
540 migration_incoming_setup(f
);
541 migration_incoming_process();
544 void migration_ioc_process_incoming(QIOChannel
*ioc
)
546 MigrationIncomingState
*mis
= migration_incoming_get_current();
547 bool start_migration
;
549 if (!mis
->from_src_file
) {
550 /* The first connection (multifd may have multiple) */
551 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
553 /* If it's a recovery, we're done */
554 if (postcopy_try_recover(f
)) {
558 migration_incoming_setup(f
);
561 * Common migration only needs one channel, so we can start
562 * right now. Multifd needs more than one channel, we wait.
564 start_migration
= !migrate_use_multifd();
566 /* Multiple connections */
567 assert(migrate_use_multifd());
568 start_migration
= multifd_recv_new_channel(ioc
);
571 if (start_migration
) {
572 migration_incoming_process();
577 * @migration_has_all_channels: We have received all channels that we need
579 * Returns true when we have got connections to all the channels that
580 * we need for migration.
582 bool migration_has_all_channels(void)
584 MigrationIncomingState
*mis
= migration_incoming_get_current();
587 all_channels
= multifd_recv_all_channels_created();
589 return all_channels
&& mis
->from_src_file
!= NULL
;
593 * Send a 'SHUT' message on the return channel with the given value
594 * to indicate that we've finished with the RP. Non-0 value indicates
597 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
602 buf
= cpu_to_be32(value
);
603 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
607 * Send a 'PONG' message on the return channel with the given value
608 * (normally in response to a 'PING')
610 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
615 buf
= cpu_to_be32(value
);
616 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
619 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
627 * First, we send the header part. It contains only the len of
628 * idstr, and the idstr itself.
630 len
= strlen(block_name
);
632 memcpy(buf
+ 1, block_name
, len
);
634 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
635 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
640 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
643 * Next, we dump the received bitmap to the stream.
645 * TODO: currently we are safe since we are the only one that is
646 * using the to_src_file handle (fault thread is still paused),
647 * and it's ok even not taking the mutex. However the best way is
648 * to take the lock before sending the message header, and release
649 * the lock after sending the bitmap.
651 qemu_mutex_lock(&mis
->rp_mutex
);
652 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
653 qemu_mutex_unlock(&mis
->rp_mutex
);
655 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
658 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
662 buf
= cpu_to_be32(value
);
663 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
666 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
668 MigrationCapabilityStatusList
*head
= NULL
;
669 MigrationCapabilityStatusList
*caps
;
670 MigrationState
*s
= migrate_get_current();
673 caps
= NULL
; /* silence compiler warning */
674 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
675 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
676 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
681 head
= g_malloc0(sizeof(*caps
));
684 caps
->next
= g_malloc0(sizeof(*caps
));
688 g_malloc(sizeof(*caps
->value
));
689 caps
->value
->capability
= i
;
690 caps
->value
->state
= s
->enabled_capabilities
[i
];
696 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
698 MigrationParameters
*params
;
699 MigrationState
*s
= migrate_get_current();
701 /* TODO use QAPI_CLONE() instead of duplicating it inline */
702 params
= g_malloc0(sizeof(*params
));
703 params
->has_compress_level
= true;
704 params
->compress_level
= s
->parameters
.compress_level
;
705 params
->has_compress_threads
= true;
706 params
->compress_threads
= s
->parameters
.compress_threads
;
707 params
->has_compress_wait_thread
= true;
708 params
->compress_wait_thread
= s
->parameters
.compress_wait_thread
;
709 params
->has_decompress_threads
= true;
710 params
->decompress_threads
= s
->parameters
.decompress_threads
;
711 params
->has_cpu_throttle_initial
= true;
712 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
713 params
->has_cpu_throttle_increment
= true;
714 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
715 params
->has_tls_creds
= true;
716 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
717 params
->has_tls_hostname
= true;
718 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
719 params
->has_max_bandwidth
= true;
720 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
721 params
->has_downtime_limit
= true;
722 params
->downtime_limit
= s
->parameters
.downtime_limit
;
723 params
->has_x_checkpoint_delay
= true;
724 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
725 params
->has_block_incremental
= true;
726 params
->block_incremental
= s
->parameters
.block_incremental
;
727 params
->has_x_multifd_channels
= true;
728 params
->x_multifd_channels
= s
->parameters
.x_multifd_channels
;
729 params
->has_x_multifd_page_count
= true;
730 params
->x_multifd_page_count
= s
->parameters
.x_multifd_page_count
;
731 params
->has_xbzrle_cache_size
= true;
732 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
733 params
->has_max_postcopy_bandwidth
= true;
734 params
->max_postcopy_bandwidth
= s
->parameters
.max_postcopy_bandwidth
;
735 params
->has_max_cpu_throttle
= true;
736 params
->max_cpu_throttle
= s
->parameters
.max_cpu_throttle
;
742 * Return true if we're already in the middle of a migration
743 * (i.e. any of the active or setup states)
745 static bool migration_is_setup_or_active(int state
)
748 case MIGRATION_STATUS_ACTIVE
:
749 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
750 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
751 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
752 case MIGRATION_STATUS_SETUP
:
753 case MIGRATION_STATUS_PRE_SWITCHOVER
:
754 case MIGRATION_STATUS_DEVICE
:
763 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
765 info
->has_ram
= true;
766 info
->ram
= g_malloc0(sizeof(*info
->ram
));
767 info
->ram
->transferred
= ram_counters
.transferred
;
768 info
->ram
->total
= ram_bytes_total();
769 info
->ram
->duplicate
= ram_counters
.duplicate
;
770 /* legacy value. It is not used anymore */
771 info
->ram
->skipped
= 0;
772 info
->ram
->normal
= ram_counters
.normal
;
773 info
->ram
->normal_bytes
= ram_counters
.normal
*
774 qemu_target_page_size();
775 info
->ram
->mbps
= s
->mbps
;
776 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
777 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
778 info
->ram
->page_size
= qemu_target_page_size();
779 info
->ram
->multifd_bytes
= ram_counters
.multifd_bytes
;
781 if (migrate_use_xbzrle()) {
782 info
->has_xbzrle_cache
= true;
783 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
784 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
785 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
786 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
787 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
788 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
789 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
792 if (migrate_use_compression()) {
793 info
->has_compression
= true;
794 info
->compression
= g_malloc0(sizeof(*info
->compression
));
795 info
->compression
->pages
= compression_counters
.pages
;
796 info
->compression
->busy
= compression_counters
.busy
;
797 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
798 info
->compression
->compressed_size
=
799 compression_counters
.compressed_size
;
800 info
->compression
->compression_rate
=
801 compression_counters
.compression_rate
;
804 if (cpu_throttle_active()) {
805 info
->has_cpu_throttle_percentage
= true;
806 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
809 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
810 info
->ram
->remaining
= ram_bytes_remaining();
811 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
815 static void populate_disk_info(MigrationInfo
*info
)
817 if (blk_mig_active()) {
818 info
->has_disk
= true;
819 info
->disk
= g_malloc0(sizeof(*info
->disk
));
820 info
->disk
->transferred
= blk_mig_bytes_transferred();
821 info
->disk
->remaining
= blk_mig_bytes_remaining();
822 info
->disk
->total
= blk_mig_bytes_total();
826 static void fill_source_migration_info(MigrationInfo
*info
)
828 MigrationState
*s
= migrate_get_current();
831 case MIGRATION_STATUS_NONE
:
832 /* no migration has happened ever */
833 /* do not overwrite destination migration status */
836 case MIGRATION_STATUS_SETUP
:
837 info
->has_status
= true;
838 info
->has_total_time
= false;
840 case MIGRATION_STATUS_ACTIVE
:
841 case MIGRATION_STATUS_CANCELLING
:
842 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
843 case MIGRATION_STATUS_PRE_SWITCHOVER
:
844 case MIGRATION_STATUS_DEVICE
:
845 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
846 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
847 /* TODO add some postcopy stats */
848 info
->has_status
= true;
849 info
->has_total_time
= true;
850 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
852 info
->has_expected_downtime
= true;
853 info
->expected_downtime
= s
->expected_downtime
;
854 info
->has_setup_time
= true;
855 info
->setup_time
= s
->setup_time
;
857 populate_ram_info(info
, s
);
858 populate_disk_info(info
);
860 case MIGRATION_STATUS_COLO
:
861 info
->has_status
= true;
862 /* TODO: display COLO specific information (checkpoint info etc.) */
864 case MIGRATION_STATUS_COMPLETED
:
865 info
->has_status
= true;
866 info
->has_total_time
= true;
867 info
->total_time
= s
->total_time
;
868 info
->has_downtime
= true;
869 info
->downtime
= s
->downtime
;
870 info
->has_setup_time
= true;
871 info
->setup_time
= s
->setup_time
;
873 populate_ram_info(info
, s
);
875 case MIGRATION_STATUS_FAILED
:
876 info
->has_status
= true;
878 info
->has_error_desc
= true;
879 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
882 case MIGRATION_STATUS_CANCELLED
:
883 info
->has_status
= true;
886 info
->status
= s
->state
;
890 * @migration_caps_check - check capability validity
892 * @cap_list: old capability list, array of bool
893 * @params: new capabilities to be applied soon
894 * @errp: set *errp if the check failed, with reason
896 * Returns true if check passed, otherwise false.
898 static bool migrate_caps_check(bool *cap_list
,
899 MigrationCapabilityStatusList
*params
,
902 MigrationCapabilityStatusList
*cap
;
903 bool old_postcopy_cap
;
904 MigrationIncomingState
*mis
= migration_incoming_get_current();
906 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
908 for (cap
= params
; cap
; cap
= cap
->next
) {
909 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
912 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
913 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
914 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
916 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
921 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
922 if (cap_list
[MIGRATION_CAPABILITY_COMPRESS
]) {
923 /* The decompression threads asynchronously write into RAM
924 * rather than use the atomic copies needed to avoid
925 * userfaulting. It should be possible to fix the decompression
926 * threads for compatibility in future.
928 error_setg(errp
, "Postcopy is not currently compatible "
933 /* This check is reasonably expensive, so only when it's being
934 * set the first time, also it's only the destination that needs
937 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
938 !postcopy_ram_supported_by_host(mis
)) {
939 /* postcopy_ram_supported_by_host will have emitted a more
942 error_setg(errp
, "Postcopy is not supported");
950 static void fill_destination_migration_info(MigrationInfo
*info
)
952 MigrationIncomingState
*mis
= migration_incoming_get_current();
954 switch (mis
->state
) {
955 case MIGRATION_STATUS_NONE
:
958 case MIGRATION_STATUS_SETUP
:
959 case MIGRATION_STATUS_CANCELLING
:
960 case MIGRATION_STATUS_CANCELLED
:
961 case MIGRATION_STATUS_ACTIVE
:
962 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
963 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
964 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
965 case MIGRATION_STATUS_FAILED
:
966 case MIGRATION_STATUS_COLO
:
967 info
->has_status
= true;
969 case MIGRATION_STATUS_COMPLETED
:
970 info
->has_status
= true;
971 fill_destination_postcopy_migration_info(info
);
974 info
->status
= mis
->state
;
977 MigrationInfo
*qmp_query_migrate(Error
**errp
)
979 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
981 fill_destination_migration_info(info
);
982 fill_source_migration_info(info
);
987 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
990 MigrationState
*s
= migrate_get_current();
991 MigrationCapabilityStatusList
*cap
;
992 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
994 if (migration_is_setup_or_active(s
->state
)) {
995 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
999 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
1000 if (!migrate_caps_check(cap_list
, params
, errp
)) {
1004 for (cap
= params
; cap
; cap
= cap
->next
) {
1005 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
1010 * Check whether the parameters are valid. Error will be put into errp
1011 * (if provided). Return true if valid, otherwise false.
1013 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
1015 if (params
->has_compress_level
&&
1016 (params
->compress_level
> 9)) {
1017 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
1018 "is invalid, it should be in the range of 0 to 9");
1022 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
1023 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1025 "is invalid, it should be in the range of 1 to 255");
1029 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
1030 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1031 "decompress_threads",
1032 "is invalid, it should be in the range of 1 to 255");
1036 if (params
->has_cpu_throttle_initial
&&
1037 (params
->cpu_throttle_initial
< 1 ||
1038 params
->cpu_throttle_initial
> 99)) {
1039 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1040 "cpu_throttle_initial",
1041 "an integer in the range of 1 to 99");
1045 if (params
->has_cpu_throttle_increment
&&
1046 (params
->cpu_throttle_increment
< 1 ||
1047 params
->cpu_throttle_increment
> 99)) {
1048 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1049 "cpu_throttle_increment",
1050 "an integer in the range of 1 to 99");
1054 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
1055 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
1056 " range of 0 to %zu bytes/second", SIZE_MAX
);
1060 if (params
->has_downtime_limit
&&
1061 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
1062 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1063 "the range of 0 to %d milliseconds",
1064 MAX_MIGRATE_DOWNTIME
);
1068 /* x_checkpoint_delay is now always positive */
1070 if (params
->has_x_multifd_channels
&& (params
->x_multifd_channels
< 1)) {
1071 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1073 "is invalid, it should be in the range of 1 to 255");
1076 if (params
->has_x_multifd_page_count
&&
1077 (params
->x_multifd_page_count
< 1 ||
1078 params
->x_multifd_page_count
> 10000)) {
1079 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1080 "multifd_page_count",
1081 "is invalid, it should be in the range of 1 to 10000");
1085 if (params
->has_xbzrle_cache_size
&&
1086 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
1087 !is_power_of_2(params
->xbzrle_cache_size
))) {
1088 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1089 "xbzrle_cache_size",
1090 "is invalid, it should be bigger than target page size"
1091 " and a power of two");
1095 if (params
->has_max_cpu_throttle
&&
1096 (params
->max_cpu_throttle
< params
->cpu_throttle_initial
||
1097 params
->max_cpu_throttle
> 99)) {
1098 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1100 "an integer in the range of cpu_throttle_initial to 99");
1107 static void migrate_params_test_apply(MigrateSetParameters
*params
,
1108 MigrationParameters
*dest
)
1110 *dest
= migrate_get_current()->parameters
;
1112 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1114 if (params
->has_compress_level
) {
1115 dest
->compress_level
= params
->compress_level
;
1118 if (params
->has_compress_threads
) {
1119 dest
->compress_threads
= params
->compress_threads
;
1122 if (params
->has_compress_wait_thread
) {
1123 dest
->compress_wait_thread
= params
->compress_wait_thread
;
1126 if (params
->has_decompress_threads
) {
1127 dest
->decompress_threads
= params
->decompress_threads
;
1130 if (params
->has_cpu_throttle_initial
) {
1131 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
1134 if (params
->has_cpu_throttle_increment
) {
1135 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
1138 if (params
->has_tls_creds
) {
1139 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1140 dest
->tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1143 if (params
->has_tls_hostname
) {
1144 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1145 dest
->tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1148 if (params
->has_max_bandwidth
) {
1149 dest
->max_bandwidth
= params
->max_bandwidth
;
1152 if (params
->has_downtime_limit
) {
1153 dest
->downtime_limit
= params
->downtime_limit
;
1156 if (params
->has_x_checkpoint_delay
) {
1157 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
1160 if (params
->has_block_incremental
) {
1161 dest
->block_incremental
= params
->block_incremental
;
1163 if (params
->has_x_multifd_channels
) {
1164 dest
->x_multifd_channels
= params
->x_multifd_channels
;
1166 if (params
->has_x_multifd_page_count
) {
1167 dest
->x_multifd_page_count
= params
->x_multifd_page_count
;
1169 if (params
->has_xbzrle_cache_size
) {
1170 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
1172 if (params
->has_max_postcopy_bandwidth
) {
1173 dest
->max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1175 if (params
->has_max_cpu_throttle
) {
1176 dest
->max_cpu_throttle
= params
->max_cpu_throttle
;
1180 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
1182 MigrationState
*s
= migrate_get_current();
1184 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1186 if (params
->has_compress_level
) {
1187 s
->parameters
.compress_level
= params
->compress_level
;
1190 if (params
->has_compress_threads
) {
1191 s
->parameters
.compress_threads
= params
->compress_threads
;
1194 if (params
->has_compress_wait_thread
) {
1195 s
->parameters
.compress_wait_thread
= params
->compress_wait_thread
;
1198 if (params
->has_decompress_threads
) {
1199 s
->parameters
.decompress_threads
= params
->decompress_threads
;
1202 if (params
->has_cpu_throttle_initial
) {
1203 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
1206 if (params
->has_cpu_throttle_increment
) {
1207 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
1210 if (params
->has_tls_creds
) {
1211 g_free(s
->parameters
.tls_creds
);
1212 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1213 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1216 if (params
->has_tls_hostname
) {
1217 g_free(s
->parameters
.tls_hostname
);
1218 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1219 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1222 if (params
->has_max_bandwidth
) {
1223 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1224 if (s
->to_dst_file
) {
1225 qemu_file_set_rate_limit(s
->to_dst_file
,
1226 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1230 if (params
->has_downtime_limit
) {
1231 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1234 if (params
->has_x_checkpoint_delay
) {
1235 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1236 if (migration_in_colo_state()) {
1237 colo_checkpoint_notify(s
);
1241 if (params
->has_block_incremental
) {
1242 s
->parameters
.block_incremental
= params
->block_incremental
;
1244 if (params
->has_x_multifd_channels
) {
1245 s
->parameters
.x_multifd_channels
= params
->x_multifd_channels
;
1247 if (params
->has_x_multifd_page_count
) {
1248 s
->parameters
.x_multifd_page_count
= params
->x_multifd_page_count
;
1250 if (params
->has_xbzrle_cache_size
) {
1251 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1252 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1254 if (params
->has_max_postcopy_bandwidth
) {
1255 s
->parameters
.max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1257 if (params
->has_max_cpu_throttle
) {
1258 s
->parameters
.max_cpu_throttle
= params
->max_cpu_throttle
;
1262 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1264 MigrationParameters tmp
;
1266 /* TODO Rewrite "" to null instead */
1267 if (params
->has_tls_creds
1268 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1269 qobject_unref(params
->tls_creds
->u
.n
);
1270 params
->tls_creds
->type
= QTYPE_QSTRING
;
1271 params
->tls_creds
->u
.s
= strdup("");
1273 /* TODO Rewrite "" to null instead */
1274 if (params
->has_tls_hostname
1275 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1276 qobject_unref(params
->tls_hostname
->u
.n
);
1277 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1278 params
->tls_hostname
->u
.s
= strdup("");
1281 migrate_params_test_apply(params
, &tmp
);
1283 if (!migrate_params_check(&tmp
, errp
)) {
1284 /* Invalid parameter */
1288 migrate_params_apply(params
, errp
);
1292 void qmp_migrate_start_postcopy(Error
**errp
)
1294 MigrationState
*s
= migrate_get_current();
1296 if (!migrate_postcopy()) {
1297 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1298 " the start of migration");
1302 if (s
->state
== MIGRATION_STATUS_NONE
) {
1303 error_setg(errp
, "Postcopy must be started after migration has been"
1308 * we don't error if migration has finished since that would be racy
1309 * with issuing this command.
1311 atomic_set(&s
->start_postcopy
, true);
1314 /* shared migration helpers */
1316 void migrate_set_state(int *state
, int old_state
, int new_state
)
1318 assert(new_state
< MIGRATION_STATUS__MAX
);
1319 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1320 trace_migrate_set_state(MigrationStatus_str(new_state
));
1321 migrate_generate_event(new_state
);
1325 static MigrationCapabilityStatusList
*migrate_cap_add(
1326 MigrationCapabilityStatusList
*list
,
1327 MigrationCapability index
,
1330 MigrationCapabilityStatusList
*cap
;
1332 cap
= g_new0(MigrationCapabilityStatusList
, 1);
1333 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
1334 cap
->value
->capability
= index
;
1335 cap
->value
->state
= state
;
1341 void migrate_set_block_enabled(bool value
, Error
**errp
)
1343 MigrationCapabilityStatusList
*cap
;
1345 cap
= migrate_cap_add(NULL
, MIGRATION_CAPABILITY_BLOCK
, value
);
1346 qmp_migrate_set_capabilities(cap
, errp
);
1347 qapi_free_MigrationCapabilityStatusList(cap
);
1350 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1352 s
->parameters
.block_incremental
= value
;
1355 static void block_cleanup_parameters(MigrationState
*s
)
1357 if (s
->must_remove_block_options
) {
1358 /* setting to false can never fail */
1359 migrate_set_block_enabled(false, &error_abort
);
1360 migrate_set_block_incremental(s
, false);
1361 s
->must_remove_block_options
= false;
1365 static void migrate_fd_cleanup(void *opaque
)
1367 MigrationState
*s
= opaque
;
1369 qemu_bh_delete(s
->cleanup_bh
);
1370 s
->cleanup_bh
= NULL
;
1372 qemu_savevm_state_cleanup();
1374 if (s
->to_dst_file
) {
1375 Error
*local_err
= NULL
;
1378 trace_migrate_fd_cleanup();
1379 qemu_mutex_unlock_iothread();
1380 if (s
->migration_thread_running
) {
1381 qemu_thread_join(&s
->thread
);
1382 s
->migration_thread_running
= false;
1384 qemu_mutex_lock_iothread();
1386 if (multifd_save_cleanup(&local_err
) != 0) {
1387 error_report_err(local_err
);
1389 qemu_mutex_lock(&s
->qemu_file_lock
);
1390 tmp
= s
->to_dst_file
;
1391 s
->to_dst_file
= NULL
;
1392 qemu_mutex_unlock(&s
->qemu_file_lock
);
1394 * Close the file handle without the lock to make sure the
1395 * critical section won't block for long.
1400 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
1401 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
1403 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1404 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1405 MIGRATION_STATUS_CANCELLED
);
1409 /* It is used on info migrate. We can't free it */
1410 error_report_err(error_copy(s
->error
));
1412 notifier_list_notify(&migration_state_notifiers
, s
);
1413 block_cleanup_parameters(s
);
1416 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1418 qemu_mutex_lock(&s
->error_mutex
);
1420 s
->error
= error_copy(error
);
1422 qemu_mutex_unlock(&s
->error_mutex
);
1425 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1427 trace_migrate_fd_error(error_get_pretty(error
));
1428 assert(s
->to_dst_file
== NULL
);
1429 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1430 MIGRATION_STATUS_FAILED
);
1431 migrate_set_error(s
, error
);
1434 static void migrate_fd_cancel(MigrationState
*s
)
1437 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1438 trace_migrate_fd_cancel();
1440 if (s
->rp_state
.from_dst_file
) {
1441 /* shutdown the rp socket, so causing the rp thread to shutdown */
1442 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1446 old_state
= s
->state
;
1447 if (!migration_is_setup_or_active(old_state
)) {
1450 /* If the migration is paused, kick it out of the pause */
1451 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1452 qemu_sem_post(&s
->pause_sem
);
1454 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1455 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1458 * If we're unlucky the migration code might be stuck somewhere in a
1459 * send/write while the network has failed and is waiting to timeout;
1460 * if we've got shutdown(2) available then we can force it to quit.
1461 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1462 * called in a bh, so there is no race against this cancel.
1464 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1465 qemu_file_shutdown(f
);
1467 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1468 Error
*local_err
= NULL
;
1470 bdrv_invalidate_cache_all(&local_err
);
1472 error_report_err(local_err
);
1474 s
->block_inactive
= false;
1479 void add_migration_state_change_notifier(Notifier
*notify
)
1481 notifier_list_add(&migration_state_notifiers
, notify
);
1484 void remove_migration_state_change_notifier(Notifier
*notify
)
1486 notifier_remove(notify
);
1489 bool migration_in_setup(MigrationState
*s
)
1491 return s
->state
== MIGRATION_STATUS_SETUP
;
1494 bool migration_has_finished(MigrationState
*s
)
1496 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1499 bool migration_has_failed(MigrationState
*s
)
1501 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1502 s
->state
== MIGRATION_STATUS_FAILED
);
1505 bool migration_in_postcopy(void)
1507 MigrationState
*s
= migrate_get_current();
1509 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1512 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1514 return migration_in_postcopy() && s
->postcopy_after_devices
;
1517 bool migration_is_idle(void)
1519 MigrationState
*s
= migrate_get_current();
1522 case MIGRATION_STATUS_NONE
:
1523 case MIGRATION_STATUS_CANCELLED
:
1524 case MIGRATION_STATUS_COMPLETED
:
1525 case MIGRATION_STATUS_FAILED
:
1527 case MIGRATION_STATUS_SETUP
:
1528 case MIGRATION_STATUS_CANCELLING
:
1529 case MIGRATION_STATUS_ACTIVE
:
1530 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1531 case MIGRATION_STATUS_COLO
:
1532 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1533 case MIGRATION_STATUS_DEVICE
:
1535 case MIGRATION_STATUS__MAX
:
1536 g_assert_not_reached();
1542 void migrate_init(MigrationState
*s
)
1545 * Reinitialise all migration state, except
1546 * parameters/capabilities that the user set, and
1552 s
->to_dst_file
= NULL
;
1553 s
->state
= MIGRATION_STATUS_NONE
;
1554 s
->rp_state
.from_dst_file
= NULL
;
1555 s
->rp_state
.error
= false;
1558 s
->expected_downtime
= 0;
1560 s
->start_postcopy
= false;
1561 s
->postcopy_after_devices
= false;
1562 s
->migration_thread_running
= false;
1563 error_free(s
->error
);
1566 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1568 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1570 s
->vm_was_running
= false;
1571 s
->iteration_initial_bytes
= 0;
1572 s
->threshold_size
= 0;
1575 static GSList
*migration_blockers
;
1577 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1579 if (migrate_get_current()->only_migratable
) {
1580 error_propagate_prepend(errp
, error_copy(reason
),
1581 "disallowing migration blocker "
1582 "(--only_migratable) for: ");
1586 if (migration_is_idle()) {
1587 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1591 error_propagate_prepend(errp
, error_copy(reason
),
1592 "disallowing migration blocker "
1593 "(migration in progress) for: ");
1597 void migrate_del_blocker(Error
*reason
)
1599 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1602 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1604 Error
*local_err
= NULL
;
1605 static bool once
= true;
1607 if (!deferred_incoming
) {
1608 error_setg(errp
, "For use with '-incoming defer'");
1612 error_setg(errp
, "The incoming migration has already been started");
1615 qemu_start_incoming_migration(uri
, &local_err
);
1618 error_propagate(errp
, local_err
);
1625 void qmp_migrate_recover(const char *uri
, Error
**errp
)
1627 MigrationIncomingState
*mis
= migration_incoming_get_current();
1629 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1630 error_setg(errp
, "Migrate recover can only be run "
1631 "when postcopy is paused.");
1635 if (atomic_cmpxchg(&mis
->postcopy_recover_triggered
,
1636 false, true) == true) {
1637 error_setg(errp
, "Migrate recovery is triggered already");
1642 * Note that this call will never start a real migration; it will
1643 * only re-setup the migration stream and poke existing migration
1644 * to continue using that newly established channel.
1646 qemu_start_incoming_migration(uri
, errp
);
1649 void qmp_migrate_pause(Error
**errp
)
1651 MigrationState
*ms
= migrate_get_current();
1652 MigrationIncomingState
*mis
= migration_incoming_get_current();
1655 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1656 /* Source side, during postcopy */
1657 qemu_mutex_lock(&ms
->qemu_file_lock
);
1658 ret
= qemu_file_shutdown(ms
->to_dst_file
);
1659 qemu_mutex_unlock(&ms
->qemu_file_lock
);
1661 error_setg(errp
, "Failed to pause source migration");
1666 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1667 ret
= qemu_file_shutdown(mis
->from_src_file
);
1669 error_setg(errp
, "Failed to pause destination migration");
1674 error_setg(errp
, "migrate-pause is currently only supported "
1675 "during postcopy-active state");
1678 bool migration_is_blocked(Error
**errp
)
1680 if (qemu_savevm_state_blocked(errp
)) {
1684 if (migration_blockers
) {
1685 error_propagate(errp
, error_copy(migration_blockers
->data
));
1692 /* Returns true if continue to migrate, or false if error detected */
1693 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
1694 bool resume
, Error
**errp
)
1696 Error
*local_err
= NULL
;
1699 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1700 error_setg(errp
, "Cannot resume if there is no "
1701 "paused migration");
1706 * Postcopy recovery won't work well with release-ram
1707 * capability since release-ram will drop the page buffer as
1708 * long as the page is put into the send buffer. So if there
1709 * is a network failure happened, any page buffers that have
1710 * not yet reached the destination VM but have already been
1711 * sent from the source VM will be lost forever. Let's refuse
1712 * the client from resuming such a postcopy migration.
1713 * Luckily release-ram was designed to only be used when src
1714 * and destination VMs are on the same host, so it should be
1717 if (migrate_release_ram()) {
1718 error_setg(errp
, "Postcopy recovery cannot work "
1719 "when release-ram capability is set");
1723 /* This is a resume, skip init status */
1727 if (migration_is_setup_or_active(s
->state
) ||
1728 s
->state
== MIGRATION_STATUS_CANCELLING
||
1729 s
->state
== MIGRATION_STATUS_COLO
) {
1730 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1734 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1735 error_setg(errp
, "Guest is waiting for an incoming migration");
1739 if (migration_is_blocked(errp
)) {
1743 if (blk
|| blk_inc
) {
1744 if (migrate_use_block() || migrate_use_block_incremental()) {
1745 error_setg(errp
, "Command options are incompatible with "
1746 "current migration capabilities");
1749 migrate_set_block_enabled(true, &local_err
);
1751 error_propagate(errp
, local_err
);
1754 s
->must_remove_block_options
= true;
1758 migrate_set_block_incremental(s
, true);
1766 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1767 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1768 bool has_resume
, bool resume
, Error
**errp
)
1770 Error
*local_err
= NULL
;
1771 MigrationState
*s
= migrate_get_current();
1774 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
1775 has_resume
&& resume
, errp
)) {
1776 /* Error detected, put into errp */
1780 if (strstart(uri
, "tcp:", &p
)) {
1781 tcp_start_outgoing_migration(s
, p
, &local_err
);
1783 } else if (strstart(uri
, "rdma:", &p
)) {
1784 rdma_start_outgoing_migration(s
, p
, &local_err
);
1786 } else if (strstart(uri
, "exec:", &p
)) {
1787 exec_start_outgoing_migration(s
, p
, &local_err
);
1788 } else if (strstart(uri
, "unix:", &p
)) {
1789 unix_start_outgoing_migration(s
, p
, &local_err
);
1790 } else if (strstart(uri
, "fd:", &p
)) {
1791 fd_start_outgoing_migration(s
, p
, &local_err
);
1793 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1794 "a valid migration protocol");
1795 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1796 MIGRATION_STATUS_FAILED
);
1797 block_cleanup_parameters(s
);
1802 migrate_fd_error(s
, local_err
);
1803 error_propagate(errp
, local_err
);
1808 void qmp_migrate_cancel(Error
**errp
)
1810 migrate_fd_cancel(migrate_get_current());
1813 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
1815 MigrationState
*s
= migrate_get_current();
1816 if (s
->state
!= state
) {
1817 error_setg(errp
, "Migration not in expected state: %s",
1818 MigrationStatus_str(s
->state
));
1821 qemu_sem_post(&s
->pause_sem
);
1824 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1826 MigrateSetParameters p
= {
1827 .has_xbzrle_cache_size
= true,
1828 .xbzrle_cache_size
= value
,
1831 qmp_migrate_set_parameters(&p
, errp
);
1834 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1836 return migrate_xbzrle_cache_size();
1839 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1841 MigrateSetParameters p
= {
1842 .has_max_bandwidth
= true,
1843 .max_bandwidth
= value
,
1846 qmp_migrate_set_parameters(&p
, errp
);
1849 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1851 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
1852 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1853 "the range of 0 to %d seconds",
1854 MAX_MIGRATE_DOWNTIME_SECONDS
);
1858 value
*= 1000; /* Convert to milliseconds */
1859 value
= MAX(0, MIN(INT64_MAX
, value
));
1861 MigrateSetParameters p
= {
1862 .has_downtime_limit
= true,
1863 .downtime_limit
= value
,
1866 qmp_migrate_set_parameters(&p
, errp
);
1869 bool migrate_release_ram(void)
1873 s
= migrate_get_current();
1875 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
1878 bool migrate_postcopy_ram(void)
1882 s
= migrate_get_current();
1884 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1887 bool migrate_postcopy(void)
1889 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
1892 bool migrate_auto_converge(void)
1896 s
= migrate_get_current();
1898 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1901 bool migrate_zero_blocks(void)
1905 s
= migrate_get_current();
1907 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
1910 bool migrate_postcopy_blocktime(void)
1914 s
= migrate_get_current();
1916 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
1919 bool migrate_use_compression(void)
1923 s
= migrate_get_current();
1925 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
1928 int migrate_compress_level(void)
1932 s
= migrate_get_current();
1934 return s
->parameters
.compress_level
;
1937 int migrate_compress_threads(void)
1941 s
= migrate_get_current();
1943 return s
->parameters
.compress_threads
;
1946 int migrate_compress_wait_thread(void)
1950 s
= migrate_get_current();
1952 return s
->parameters
.compress_wait_thread
;
1955 int migrate_decompress_threads(void)
1959 s
= migrate_get_current();
1961 return s
->parameters
.decompress_threads
;
1964 bool migrate_dirty_bitmaps(void)
1968 s
= migrate_get_current();
1970 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
1973 bool migrate_use_events(void)
1977 s
= migrate_get_current();
1979 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
1982 bool migrate_use_multifd(void)
1986 s
= migrate_get_current();
1988 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_MULTIFD
];
1991 bool migrate_pause_before_switchover(void)
1995 s
= migrate_get_current();
1997 return s
->enabled_capabilities
[
1998 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
2001 int migrate_multifd_channels(void)
2005 s
= migrate_get_current();
2007 return s
->parameters
.x_multifd_channels
;
2010 int migrate_multifd_page_count(void)
2014 s
= migrate_get_current();
2016 return s
->parameters
.x_multifd_page_count
;
2019 int migrate_use_xbzrle(void)
2023 s
= migrate_get_current();
2025 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
2028 int64_t migrate_xbzrle_cache_size(void)
2032 s
= migrate_get_current();
2034 return s
->parameters
.xbzrle_cache_size
;
2037 static int64_t migrate_max_postcopy_bandwidth(void)
2041 s
= migrate_get_current();
2043 return s
->parameters
.max_postcopy_bandwidth
;
2046 bool migrate_use_block(void)
2050 s
= migrate_get_current();
2052 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
2055 bool migrate_use_return_path(void)
2059 s
= migrate_get_current();
2061 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
2064 bool migrate_use_block_incremental(void)
2068 s
= migrate_get_current();
2070 return s
->parameters
.block_incremental
;
2073 /* migration thread support */
2075 * Something bad happened to the RP stream, mark an error
2076 * The caller shall print or trace something to indicate why
2078 static void mark_source_rp_bad(MigrationState
*s
)
2080 s
->rp_state
.error
= true;
2083 static struct rp_cmd_args
{
2084 ssize_t len
; /* -1 = variable */
2087 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
2088 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
2089 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
2090 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
2091 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
2092 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
2093 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
2094 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
2098 * Process a request for pages received on the return path,
2099 * We're allowed to send more than requested (e.g. to round to our page size)
2100 * and we don't need to send pages that have already been sent.
2102 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
2103 ram_addr_t start
, size_t len
)
2105 long our_host_ps
= getpagesize();
2107 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
2110 * Since we currently insist on matching page sizes, just sanity check
2111 * we're being asked for whole host pages.
2113 if (start
& (our_host_ps
-1) ||
2114 (len
& (our_host_ps
-1))) {
2115 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2116 " len: %zd", __func__
, start
, len
);
2117 mark_source_rp_bad(ms
);
2121 if (ram_save_queue_pages(rbname
, start
, len
)) {
2122 mark_source_rp_bad(ms
);
2126 /* Return true to retry, false to quit */
2127 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
2129 trace_postcopy_pause_return_path();
2131 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
2133 trace_postcopy_pause_return_path_continued();
2138 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
2140 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
2143 error_report("%s: invalid block name '%s'", __func__
, block_name
);
2147 /* Fetch the received bitmap and refresh the dirty bitmap */
2148 return ram_dirty_bitmap_reload(s
, block
);
2151 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
2153 trace_source_return_path_thread_resume_ack(value
);
2155 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
2156 error_report("%s: illegal resume_ack value %"PRIu32
,
2161 /* Now both sides are active. */
2162 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2163 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2165 /* Notify send thread that time to continue send pages */
2166 qemu_sem_post(&s
->rp_state
.rp_sem
);
2172 * Handles messages sent on the return path towards the source VM
2175 static void *source_return_path_thread(void *opaque
)
2177 MigrationState
*ms
= opaque
;
2178 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
2179 uint16_t header_len
, header_type
;
2181 uint32_t tmp32
, sibling_error
;
2182 ram_addr_t start
= 0; /* =0 to silence warning */
2183 size_t len
= 0, expected_len
;
2186 trace_source_return_path_thread_entry();
2187 rcu_register_thread();
2190 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
2191 migration_is_setup_or_active(ms
->state
)) {
2192 trace_source_return_path_thread_loop_top();
2193 header_type
= qemu_get_be16(rp
);
2194 header_len
= qemu_get_be16(rp
);
2196 if (qemu_file_get_error(rp
)) {
2197 mark_source_rp_bad(ms
);
2201 if (header_type
>= MIG_RP_MSG_MAX
||
2202 header_type
== MIG_RP_MSG_INVALID
) {
2203 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2204 header_type
, header_len
);
2205 mark_source_rp_bad(ms
);
2209 if ((rp_cmd_args
[header_type
].len
!= -1 &&
2210 header_len
!= rp_cmd_args
[header_type
].len
) ||
2211 header_len
> sizeof(buf
)) {
2212 error_report("RP: Received '%s' message (0x%04x) with"
2213 "incorrect length %d expecting %zu",
2214 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
2215 (size_t)rp_cmd_args
[header_type
].len
);
2216 mark_source_rp_bad(ms
);
2220 /* We know we've got a valid header by this point */
2221 res
= qemu_get_buffer(rp
, buf
, header_len
);
2222 if (res
!= header_len
) {
2223 error_report("RP: Failed reading data for message 0x%04x"
2224 " read %d expected %d",
2225 header_type
, res
, header_len
);
2226 mark_source_rp_bad(ms
);
2230 /* OK, we have the message and the data */
2231 switch (header_type
) {
2232 case MIG_RP_MSG_SHUT
:
2233 sibling_error
= ldl_be_p(buf
);
2234 trace_source_return_path_thread_shut(sibling_error
);
2235 if (sibling_error
) {
2236 error_report("RP: Sibling indicated error %d", sibling_error
);
2237 mark_source_rp_bad(ms
);
2240 * We'll let the main thread deal with closing the RP
2241 * we could do a shutdown(2) on it, but we're the only user
2242 * anyway, so there's nothing gained.
2246 case MIG_RP_MSG_PONG
:
2247 tmp32
= ldl_be_p(buf
);
2248 trace_source_return_path_thread_pong(tmp32
);
2251 case MIG_RP_MSG_REQ_PAGES
:
2252 start
= ldq_be_p(buf
);
2253 len
= ldl_be_p(buf
+ 8);
2254 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
2257 case MIG_RP_MSG_REQ_PAGES_ID
:
2258 expected_len
= 12 + 1; /* header + termination */
2260 if (header_len
>= expected_len
) {
2261 start
= ldq_be_p(buf
);
2262 len
= ldl_be_p(buf
+ 8);
2263 /* Now we expect an idstr */
2264 tmp32
= buf
[12]; /* Length of the following idstr */
2265 buf
[13 + tmp32
] = '\0';
2266 expected_len
+= tmp32
;
2268 if (header_len
!= expected_len
) {
2269 error_report("RP: Req_Page_id with length %d expecting %zd",
2270 header_len
, expected_len
);
2271 mark_source_rp_bad(ms
);
2274 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
2277 case MIG_RP_MSG_RECV_BITMAP
:
2278 if (header_len
< 1) {
2279 error_report("%s: missing block name", __func__
);
2280 mark_source_rp_bad(ms
);
2283 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2284 buf
[buf
[0] + 1] = '\0';
2285 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
2286 mark_source_rp_bad(ms
);
2291 case MIG_RP_MSG_RESUME_ACK
:
2292 tmp32
= ldl_be_p(buf
);
2293 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
2294 mark_source_rp_bad(ms
);
2305 res
= qemu_file_get_error(rp
);
2309 * Maybe there is something we can do: it looks like a
2310 * network down issue, and we pause for a recovery.
2312 if (postcopy_pause_return_path_thread(ms
)) {
2313 /* Reload rp, reset the rest */
2314 if (rp
!= ms
->rp_state
.from_dst_file
) {
2316 rp
= ms
->rp_state
.from_dst_file
;
2318 ms
->rp_state
.error
= false;
2323 trace_source_return_path_thread_bad_end();
2324 mark_source_rp_bad(ms
);
2327 trace_source_return_path_thread_end();
2328 ms
->rp_state
.from_dst_file
= NULL
;
2330 rcu_unregister_thread();
2334 static int open_return_path_on_source(MigrationState
*ms
,
2338 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
2339 if (!ms
->rp_state
.from_dst_file
) {
2343 trace_open_return_path_on_source();
2345 if (!create_thread
) {
2350 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
2351 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
2353 trace_open_return_path_on_source_continue();
2358 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2359 static int await_return_path_close_on_source(MigrationState
*ms
)
2362 * If this is a normal exit then the destination will send a SHUT and the
2363 * rp_thread will exit, however if there's an error we need to cause
2366 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
2368 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2369 * waiting for the destination.
2371 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
2372 mark_source_rp_bad(ms
);
2374 trace_await_return_path_close_on_source_joining();
2375 qemu_thread_join(&ms
->rp_state
.rp_thread
);
2376 trace_await_return_path_close_on_source_close();
2377 return ms
->rp_state
.error
;
2381 * Switch from normal iteration to postcopy
2382 * Returns non-0 on error
2384 static int postcopy_start(MigrationState
*ms
)
2387 QIOChannelBuffer
*bioc
;
2389 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2390 int64_t bandwidth
= migrate_max_postcopy_bandwidth();
2391 bool restart_block
= false;
2392 int cur_state
= MIGRATION_STATUS_ACTIVE
;
2393 if (!migrate_pause_before_switchover()) {
2394 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
2395 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2398 trace_postcopy_start();
2399 qemu_mutex_lock_iothread();
2400 trace_postcopy_start_set_run();
2402 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
2403 global_state_store();
2404 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2409 ret
= migration_maybe_pause(ms
, &cur_state
,
2410 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2415 ret
= bdrv_inactivate_all();
2419 restart_block
= true;
2422 * Cause any non-postcopiable, but iterative devices to
2423 * send out their final data.
2425 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2428 * in Finish migrate and with the io-lock held everything should
2429 * be quiet, but we've potentially still got dirty pages and we
2430 * need to tell the destination to throw any pages it's already received
2433 if (migrate_postcopy_ram()) {
2434 if (ram_postcopy_send_discard_bitmap(ms
)) {
2435 error_report("postcopy send discard bitmap failed");
2441 * send rest of state - note things that are doing postcopy
2442 * will notice we're in POSTCOPY_ACTIVE and not actually
2443 * wrap their state up here
2445 /* 0 max-postcopy-bandwidth means unlimited */
2447 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
2449 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
2451 if (migrate_postcopy_ram()) {
2452 /* Ping just for debugging, helps line traces up */
2453 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
2457 * While loading the device state we may trigger page transfer
2458 * requests and the fd must be free to process those, and thus
2459 * the destination must read the whole device state off the fd before
2460 * it starts processing it. Unfortunately the ad-hoc migration format
2461 * doesn't allow the destination to know the size to read without fully
2462 * parsing it through each devices load-state code (especially the open
2463 * coded devices that use get/put).
2464 * So we wrap the device state up in a package with a length at the start;
2465 * to do this we use a qemu_buf to hold the whole of the device state.
2467 bioc
= qio_channel_buffer_new(4096);
2468 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
2469 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
2470 object_unref(OBJECT(bioc
));
2473 * Make sure the receiver can get incoming pages before we send the rest
2476 qemu_savevm_send_postcopy_listen(fb
);
2478 qemu_savevm_state_complete_precopy(fb
, false, false);
2479 if (migrate_postcopy_ram()) {
2480 qemu_savevm_send_ping(fb
, 3);
2483 qemu_savevm_send_postcopy_run(fb
);
2485 /* <><> end of stuff going into the package */
2487 /* Last point of recovery; as soon as we send the package the destination
2488 * can open devices and potentially start running.
2489 * Lets just check again we've not got any errors.
2491 ret
= qemu_file_get_error(ms
->to_dst_file
);
2493 error_report("postcopy_start: Migration stream errored (pre package)");
2497 restart_block
= false;
2499 /* Now send that blob */
2500 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2505 /* Send a notify to give a chance for anything that needs to happen
2506 * at the transition to postcopy and after the device state; in particular
2507 * spice needs to trigger a transition now
2509 ms
->postcopy_after_devices
= true;
2510 notifier_list_notify(&migration_state_notifiers
, ms
);
2512 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2514 qemu_mutex_unlock_iothread();
2516 if (migrate_postcopy_ram()) {
2518 * Although this ping is just for debug, it could potentially be
2519 * used for getting a better measurement of downtime at the source.
2521 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2524 if (migrate_release_ram()) {
2525 ram_postcopy_migrated_memory_release(ms
);
2528 ret
= qemu_file_get_error(ms
->to_dst_file
);
2530 error_report("postcopy_start: Migration stream errored");
2531 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2532 MIGRATION_STATUS_FAILED
);
2540 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2541 MIGRATION_STATUS_FAILED
);
2542 if (restart_block
) {
2543 /* A failure happened early enough that we know the destination hasn't
2544 * accessed block devices, so we're safe to recover.
2546 Error
*local_err
= NULL
;
2548 bdrv_invalidate_cache_all(&local_err
);
2550 error_report_err(local_err
);
2553 qemu_mutex_unlock_iothread();
2558 * migration_maybe_pause: Pause if required to by
2559 * migrate_pause_before_switchover called with the iothread locked
2560 * Returns: 0 on success
2562 static int migration_maybe_pause(MigrationState
*s
,
2563 int *current_active_state
,
2566 if (!migrate_pause_before_switchover()) {
2570 /* Since leaving this state is not atomic with posting the semaphore
2571 * it's possible that someone could have issued multiple migrate_continue
2572 * and the semaphore is incorrectly positive at this point;
2573 * the docs say it's undefined to reinit a semaphore that's already
2574 * init'd, so use timedwait to eat up any existing posts.
2576 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2577 /* This block intentionally left blank */
2580 qemu_mutex_unlock_iothread();
2581 migrate_set_state(&s
->state
, *current_active_state
,
2582 MIGRATION_STATUS_PRE_SWITCHOVER
);
2583 qemu_sem_wait(&s
->pause_sem
);
2584 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2586 *current_active_state
= new_state
;
2587 qemu_mutex_lock_iothread();
2589 return s
->state
== new_state
? 0 : -EINVAL
;
2593 * migration_completion: Used by migration_thread when there's not much left.
2594 * The caller 'breaks' the loop when this returns.
2596 * @s: Current migration state
2598 static void migration_completion(MigrationState
*s
)
2601 int current_active_state
= s
->state
;
2603 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2604 qemu_mutex_lock_iothread();
2605 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2606 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
2607 s
->vm_was_running
= runstate_is_running();
2608 ret
= global_state_store();
2611 bool inactivate
= !migrate_colo_enabled();
2612 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2614 ret
= migration_maybe_pause(s
, ¤t_active_state
,
2615 MIGRATION_STATUS_DEVICE
);
2618 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
2619 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
2622 if (inactivate
&& ret
>= 0) {
2623 s
->block_inactive
= true;
2626 qemu_mutex_unlock_iothread();
2631 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2632 trace_migration_completion_postcopy_end();
2634 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
2635 trace_migration_completion_postcopy_end_after_complete();
2639 * If rp was opened we must clean up the thread before
2640 * cleaning everything else up (since if there are no failures
2641 * it will wait for the destination to send it's status in
2644 if (s
->rp_state
.from_dst_file
) {
2646 trace_migration_return_path_end_before();
2647 rp_error
= await_return_path_close_on_source(s
);
2648 trace_migration_return_path_end_after(rp_error
);
2650 goto fail_invalidate
;
2654 if (qemu_file_get_error(s
->to_dst_file
)) {
2655 trace_migration_completion_file_err();
2656 goto fail_invalidate
;
2659 if (!migrate_colo_enabled()) {
2660 migrate_set_state(&s
->state
, current_active_state
,
2661 MIGRATION_STATUS_COMPLETED
);
2667 /* If not doing postcopy, vm_start() will be called: let's regain
2668 * control on images.
2670 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
2671 s
->state
== MIGRATION_STATUS_DEVICE
) {
2672 Error
*local_err
= NULL
;
2674 qemu_mutex_lock_iothread();
2675 bdrv_invalidate_cache_all(&local_err
);
2677 error_report_err(local_err
);
2679 s
->block_inactive
= false;
2681 qemu_mutex_unlock_iothread();
2685 migrate_set_state(&s
->state
, current_active_state
,
2686 MIGRATION_STATUS_FAILED
);
2689 bool migrate_colo_enabled(void)
2691 MigrationState
*s
= migrate_get_current();
2692 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
2695 typedef enum MigThrError
{
2696 /* No error detected */
2697 MIG_THR_ERR_NONE
= 0,
2698 /* Detected error, but resumed successfully */
2699 MIG_THR_ERR_RECOVERED
= 1,
2700 /* Detected fatal error, need to exit */
2701 MIG_THR_ERR_FATAL
= 2,
2704 static int postcopy_resume_handshake(MigrationState
*s
)
2706 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
2708 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2709 qemu_sem_wait(&s
->rp_state
.rp_sem
);
2712 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2719 /* Return zero if success, or <0 for error */
2720 static int postcopy_do_resume(MigrationState
*s
)
2725 * Call all the resume_prepare() hooks, so that modules can be
2726 * ready for the migration resume.
2728 ret
= qemu_savevm_state_resume_prepare(s
);
2730 error_report("%s: resume_prepare() failure detected: %d",
2736 * Last handshake with destination on the resume (destination will
2737 * switch to postcopy-active afterwards)
2739 ret
= postcopy_resume_handshake(s
);
2741 error_report("%s: handshake failed: %d", __func__
, ret
);
2749 * We don't return until we are in a safe state to continue current
2750 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2751 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2753 static MigThrError
postcopy_pause(MigrationState
*s
)
2755 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2760 migrate_set_state(&s
->state
, s
->state
,
2761 MIGRATION_STATUS_POSTCOPY_PAUSED
);
2763 /* Current channel is possibly broken. Release it. */
2764 assert(s
->to_dst_file
);
2765 qemu_mutex_lock(&s
->qemu_file_lock
);
2766 file
= s
->to_dst_file
;
2767 s
->to_dst_file
= NULL
;
2768 qemu_mutex_unlock(&s
->qemu_file_lock
);
2770 qemu_file_shutdown(file
);
2773 error_report("Detected IO failure for postcopy. "
2774 "Migration paused.");
2777 * We wait until things fixed up. Then someone will setup the
2778 * status back for us.
2780 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2781 qemu_sem_wait(&s
->postcopy_pause_sem
);
2784 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2785 /* Woken up by a recover procedure. Give it a shot */
2788 * Firstly, let's wake up the return path now, with a new
2789 * return path channel.
2791 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
2793 /* Do the resume logic */
2794 if (postcopy_do_resume(s
) == 0) {
2795 /* Let's continue! */
2796 trace_postcopy_pause_continued();
2797 return MIG_THR_ERR_RECOVERED
;
2800 * Something wrong happened during the recovery, let's
2801 * pause again. Pause is always better than throwing
2807 /* This is not right... Time to quit. */
2808 return MIG_THR_ERR_FATAL
;
2813 static MigThrError
migration_detect_error(MigrationState
*s
)
2817 /* Try to detect any file errors */
2818 ret
= qemu_file_get_error(s
->to_dst_file
);
2821 /* Everything is fine */
2822 return MIG_THR_ERR_NONE
;
2825 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
== -EIO
) {
2827 * For postcopy, we allow the network to be down for a
2828 * while. After that, it can be continued by a
2831 return postcopy_pause(s
);
2834 * For precopy (or postcopy with error outside IO), we fail
2837 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
2838 trace_migration_thread_file_err();
2840 /* Time to stop the migration, now. */
2841 return MIG_THR_ERR_FATAL
;
2845 /* How many bytes have we transferred since the beggining of the migration */
2846 static uint64_t migration_total_bytes(MigrationState
*s
)
2848 return qemu_ftell(s
->to_dst_file
) + ram_counters
.multifd_bytes
;
2851 static void migration_calculate_complete(MigrationState
*s
)
2853 uint64_t bytes
= migration_total_bytes(s
);
2854 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2855 int64_t transfer_time
;
2857 s
->total_time
= end_time
- s
->start_time
;
2860 * It's still not set, so we are precopy migration. For
2861 * postcopy, downtime is calculated during postcopy_start().
2863 s
->downtime
= end_time
- s
->downtime_start
;
2866 transfer_time
= s
->total_time
- s
->setup_time
;
2867 if (transfer_time
) {
2868 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
2872 static void migration_update_counters(MigrationState
*s
,
2873 int64_t current_time
)
2875 uint64_t transferred
, time_spent
;
2876 uint64_t current_bytes
; /* bytes transferred since the beginning */
2879 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
2883 current_bytes
= migration_total_bytes(s
);
2884 transferred
= current_bytes
- s
->iteration_initial_bytes
;
2885 time_spent
= current_time
- s
->iteration_start_time
;
2886 bandwidth
= (double)transferred
/ time_spent
;
2887 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
2889 s
->mbps
= (((double) transferred
* 8.0) /
2890 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2893 * if we haven't sent anything, we don't want to
2894 * recalculate. 10000 is a small enough number for our purposes
2896 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
2897 s
->expected_downtime
= ram_counters
.remaining
/ bandwidth
;
2900 qemu_file_reset_rate_limit(s
->to_dst_file
);
2902 s
->iteration_start_time
= current_time
;
2903 s
->iteration_initial_bytes
= current_bytes
;
2905 trace_migrate_transferred(transferred
, time_spent
,
2906 bandwidth
, s
->threshold_size
);
2909 /* Migration thread iteration status */
2911 MIG_ITERATE_RESUME
, /* Resume current iteration */
2912 MIG_ITERATE_SKIP
, /* Skip current iteration */
2913 MIG_ITERATE_BREAK
, /* Break the loop */
2917 * Return true if continue to the next iteration directly, false
2920 static MigIterateState
migration_iteration_run(MigrationState
*s
)
2922 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
2923 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
2925 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
2926 &pend_compat
, &pend_post
);
2927 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
2929 trace_migrate_pending(pending_size
, s
->threshold_size
,
2930 pend_pre
, pend_compat
, pend_post
);
2932 if (pending_size
&& pending_size
>= s
->threshold_size
) {
2933 /* Still a significant amount to transfer */
2934 if (migrate_postcopy() && !in_postcopy
&&
2935 pend_pre
<= s
->threshold_size
&&
2936 atomic_read(&s
->start_postcopy
)) {
2937 if (postcopy_start(s
)) {
2938 error_report("%s: postcopy failed to start", __func__
);
2940 return MIG_ITERATE_SKIP
;
2942 /* Just another iteration step */
2943 qemu_savevm_state_iterate(s
->to_dst_file
,
2944 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2946 trace_migration_thread_low_pending(pending_size
);
2947 migration_completion(s
);
2948 return MIG_ITERATE_BREAK
;
2951 return MIG_ITERATE_RESUME
;
2954 static void migration_iteration_finish(MigrationState
*s
)
2956 /* If we enabled cpu throttling for auto-converge, turn it off. */
2957 cpu_throttle_stop();
2959 qemu_mutex_lock_iothread();
2961 case MIGRATION_STATUS_COMPLETED
:
2962 migration_calculate_complete(s
);
2963 runstate_set(RUN_STATE_POSTMIGRATE
);
2966 case MIGRATION_STATUS_ACTIVE
:
2968 * We should really assert here, but since it's during
2969 * migration, let's try to reduce the usage of assertions.
2971 if (!migrate_colo_enabled()) {
2972 error_report("%s: critical error: calling COLO code without "
2973 "COLO enabled", __func__
);
2975 migrate_start_colo_process(s
);
2977 * Fixme: we will run VM in COLO no matter its old running state.
2978 * After exited COLO, we will keep running.
2980 s
->vm_was_running
= true;
2982 case MIGRATION_STATUS_FAILED
:
2983 case MIGRATION_STATUS_CANCELLED
:
2984 case MIGRATION_STATUS_CANCELLING
:
2985 if (s
->vm_was_running
) {
2988 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2989 runstate_set(RUN_STATE_POSTMIGRATE
);
2995 /* Should not reach here, but if so, forgive the VM. */
2996 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
2999 qemu_bh_schedule(s
->cleanup_bh
);
3000 qemu_mutex_unlock_iothread();
3003 void migration_make_urgent_request(void)
3005 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
3008 void migration_consume_urgent_request(void)
3010 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
3014 * Master migration thread on the source VM.
3015 * It drives the migration and pumps the data down the outgoing channel.
3017 static void *migration_thread(void *opaque
)
3019 MigrationState
*s
= opaque
;
3020 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3021 MigThrError thr_error
;
3022 bool urgent
= false;
3024 rcu_register_thread();
3026 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3028 qemu_savevm_state_header(s
->to_dst_file
);
3031 * If we opened the return path, we need to make sure dst has it
3034 if (s
->rp_state
.from_dst_file
) {
3035 /* Now tell the dest that it should open its end so it can reply */
3036 qemu_savevm_send_open_return_path(s
->to_dst_file
);
3038 /* And do a ping that will make stuff easier to debug */
3039 qemu_savevm_send_ping(s
->to_dst_file
, 1);
3042 if (migrate_postcopy()) {
3044 * Tell the destination that we *might* want to do postcopy later;
3045 * if the other end can't do postcopy it should fail now, nice and
3048 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
3051 if (migrate_colo_enabled()) {
3052 /* Notify migration destination that we enable COLO */
3053 qemu_savevm_send_colo_enable(s
->to_dst_file
);
3056 qemu_savevm_state_setup(s
->to_dst_file
);
3058 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3059 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3060 MIGRATION_STATUS_ACTIVE
);
3062 trace_migration_thread_setup_complete();
3064 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
3065 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3066 int64_t current_time
;
3068 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
3069 MigIterateState iter_state
= migration_iteration_run(s
);
3070 if (iter_state
== MIG_ITERATE_SKIP
) {
3072 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3078 * Try to detect any kind of failures, and see whether we
3079 * should stop the migration now.
3081 thr_error
= migration_detect_error(s
);
3082 if (thr_error
== MIG_THR_ERR_FATAL
) {
3083 /* Stop migration */
3085 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
3087 * Just recovered from a e.g. network failure, reset all
3088 * the local variables. This is important to avoid
3089 * breaking transferred_bytes and bandwidth calculation
3091 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3092 s
->iteration_initial_bytes
= 0;
3095 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3097 migration_update_counters(s
, current_time
);
3100 if (qemu_file_rate_limit(s
->to_dst_file
)) {
3101 /* Wait for a delay to do rate limiting OR
3102 * something urgent to post the semaphore.
3104 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- current_time
;
3105 trace_migration_thread_ratelimit_pre(ms
);
3106 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
3107 /* We were worken by one or more urgent things but
3108 * the timedwait will have consumed one of them.
3109 * The service routine for the urgent wake will dec
3110 * the semaphore itself for each item it consumes,
3111 * so add this one we just eat back.
3113 qemu_sem_post(&s
->rate_limit_sem
);
3116 trace_migration_thread_ratelimit_post(urgent
);
3120 trace_migration_thread_after_loop();
3121 migration_iteration_finish(s
);
3122 rcu_unregister_thread();
3126 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
3129 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
3131 s
->expected_downtime
= s
->parameters
.downtime_limit
;
3132 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
3134 migrate_fd_error(s
, error_in
);
3135 migrate_fd_cleanup(s
);
3140 /* This is a resumed migration */
3141 rate_limit
= INT64_MAX
;
3143 /* This is a fresh new migration */
3144 rate_limit
= s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
;
3146 /* Notify before starting migration thread */
3147 notifier_list_notify(&migration_state_notifiers
, s
);
3150 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
3151 qemu_file_set_blocking(s
->to_dst_file
, true);
3154 * Open the return path. For postcopy, it is used exclusively. For
3155 * precopy, only if user specified "return-path" capability would
3156 * QEMU uses the return path.
3158 if (migrate_postcopy_ram() || migrate_use_return_path()) {
3159 if (open_return_path_on_source(s
, !resume
)) {
3160 error_report("Unable to open return-path for postcopy");
3161 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
3162 migrate_fd_cleanup(s
);
3168 /* Wakeup the main migration thread to do the recovery */
3169 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
3170 MIGRATION_STATUS_POSTCOPY_RECOVER
);
3171 qemu_sem_post(&s
->postcopy_pause_sem
);
3175 if (multifd_save_setup() != 0) {
3176 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3177 MIGRATION_STATUS_FAILED
);
3178 migrate_fd_cleanup(s
);
3181 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
3182 QEMU_THREAD_JOINABLE
);
3183 s
->migration_thread_running
= true;
3186 void migration_global_dump(Monitor
*mon
)
3188 MigrationState
*ms
= migrate_get_current();
3190 monitor_printf(mon
, "globals:\n");
3191 monitor_printf(mon
, "store-global-state: %s\n",
3192 ms
->store_global_state
? "on" : "off");
3193 monitor_printf(mon
, "only-migratable: %s\n",
3194 ms
->only_migratable
? "on" : "off");
3195 monitor_printf(mon
, "send-configuration: %s\n",
3196 ms
->send_configuration
? "on" : "off");
3197 monitor_printf(mon
, "send-section-footer: %s\n",
3198 ms
->send_section_footer
? "on" : "off");
3199 monitor_printf(mon
, "decompress-error-check: %s\n",
3200 ms
->decompress_error_check
? "on" : "off");
3203 #define DEFINE_PROP_MIG_CAP(name, x) \
3204 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3206 static Property migration_properties
[] = {
3207 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
3208 store_global_state
, true),
3209 DEFINE_PROP_BOOL("only-migratable", MigrationState
, only_migratable
, false),
3210 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
3211 send_configuration
, true),
3212 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
3213 send_section_footer
, true),
3214 DEFINE_PROP_BOOL("decompress-error-check", MigrationState
,
3215 decompress_error_check
, true),
3217 /* Migration parameters */
3218 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
3219 parameters
.compress_level
,
3220 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
3221 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
3222 parameters
.compress_threads
,
3223 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
3224 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState
,
3225 parameters
.compress_wait_thread
, true),
3226 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
3227 parameters
.decompress_threads
,
3228 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
3229 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
3230 parameters
.cpu_throttle_initial
,
3231 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
3232 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
3233 parameters
.cpu_throttle_increment
,
3234 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
3235 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
3236 parameters
.max_bandwidth
, MAX_THROTTLE
),
3237 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
3238 parameters
.downtime_limit
,
3239 DEFAULT_MIGRATE_SET_DOWNTIME
),
3240 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
3241 parameters
.x_checkpoint_delay
,
3242 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
3243 DEFINE_PROP_UINT8("x-multifd-channels", MigrationState
,
3244 parameters
.x_multifd_channels
,
3245 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
3246 DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState
,
3247 parameters
.x_multifd_page_count
,
3248 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT
),
3249 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
3250 parameters
.xbzrle_cache_size
,
3251 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
3252 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState
,
3253 parameters
.max_postcopy_bandwidth
,
3254 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH
),
3255 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState
,
3256 parameters
.max_cpu_throttle
,
3257 DEFAULT_MIGRATE_MAX_CPU_THROTTLE
),
3259 /* Migration capabilities */
3260 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
3261 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
3262 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
3263 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
3264 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
3265 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
3266 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
3267 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
3268 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
3269 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
3270 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
3271 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD
),
3273 DEFINE_PROP_END_OF_LIST(),
3276 static void migration_class_init(ObjectClass
*klass
, void *data
)
3278 DeviceClass
*dc
= DEVICE_CLASS(klass
);
3280 dc
->user_creatable
= false;
3281 dc
->props
= migration_properties
;
3284 static void migration_instance_finalize(Object
*obj
)
3286 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3287 MigrationParameters
*params
= &ms
->parameters
;
3289 qemu_mutex_destroy(&ms
->error_mutex
);
3290 qemu_mutex_destroy(&ms
->qemu_file_lock
);
3291 g_free(params
->tls_hostname
);
3292 g_free(params
->tls_creds
);
3293 qemu_sem_destroy(&ms
->rate_limit_sem
);
3294 qemu_sem_destroy(&ms
->pause_sem
);
3295 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
3296 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
3297 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
3298 error_free(ms
->error
);
3301 static void migration_instance_init(Object
*obj
)
3303 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3304 MigrationParameters
*params
= &ms
->parameters
;
3306 ms
->state
= MIGRATION_STATUS_NONE
;
3308 qemu_sem_init(&ms
->pause_sem
, 0);
3309 qemu_mutex_init(&ms
->error_mutex
);
3311 params
->tls_hostname
= g_strdup("");
3312 params
->tls_creds
= g_strdup("");
3314 /* Set has_* up only for parameter checks */
3315 params
->has_compress_level
= true;
3316 params
->has_compress_threads
= true;
3317 params
->has_decompress_threads
= true;
3318 params
->has_cpu_throttle_initial
= true;
3319 params
->has_cpu_throttle_increment
= true;
3320 params
->has_max_bandwidth
= true;
3321 params
->has_downtime_limit
= true;
3322 params
->has_x_checkpoint_delay
= true;
3323 params
->has_block_incremental
= true;
3324 params
->has_x_multifd_channels
= true;
3325 params
->has_x_multifd_page_count
= true;
3326 params
->has_xbzrle_cache_size
= true;
3327 params
->has_max_postcopy_bandwidth
= true;
3328 params
->has_max_cpu_throttle
= true;
3330 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
3331 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
3332 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
3333 qemu_sem_init(&ms
->rate_limit_sem
, 0);
3334 qemu_mutex_init(&ms
->qemu_file_lock
);
3338 * Return true if check pass, false otherwise. Error will be put
3339 * inside errp if provided.
3341 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
3343 MigrationCapabilityStatusList
*head
= NULL
;
3344 /* Assuming all off */
3345 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
3348 if (!migrate_params_check(&ms
->parameters
, errp
)) {
3352 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
3353 if (ms
->enabled_capabilities
[i
]) {
3354 head
= migrate_cap_add(head
, i
, true);
3358 ret
= migrate_caps_check(cap_list
, head
, errp
);
3360 /* It works with head == NULL */
3361 qapi_free_MigrationCapabilityStatusList(head
);
3366 static const TypeInfo migration_type
= {
3367 .name
= TYPE_MIGRATION
,
3369 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3370 * not created using qdev_create(), it is not attached to the qdev
3371 * device tree, and it is never realized.
3373 * TODO: Make this TYPE_OBJECT once QOM provides something like
3374 * TYPE_DEVICE's "-global" properties.
3376 .parent
= TYPE_DEVICE
,
3377 .class_init
= migration_class_init
,
3378 .class_size
= sizeof(MigrationClass
),
3379 .instance_size
= sizeof(MigrationState
),
3380 .instance_init
= migration_instance_init
,
3381 .instance_finalize
= migration_instance_finalize
,
3384 static void register_migration_types(void)
3386 type_register_static(&migration_type
);
3389 type_init(register_migration_types
);