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 error_report_err(local_err
);
445 if (colo_init_ram_cache() < 0) {
446 error_report("Init ram cache failed");
450 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
451 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
452 mis
->have_colo_incoming_thread
= true;
453 qemu_coroutine_yield();
455 /* Wait checkpoint incoming thread exit before free resource */
456 qemu_thread_join(&mis
->colo_incoming_thread
);
457 /* We hold the global iothread lock, so it is safe here */
458 colo_release_ram_cache();
462 error_report("load of migration failed: %s", strerror(-ret
));
465 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
466 qemu_bh_schedule(mis
->bh
);
467 mis
->migration_incoming_co
= NULL
;
471 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
472 MIGRATION_STATUS_FAILED
);
473 qemu_fclose(mis
->from_src_file
);
474 if (multifd_load_cleanup(&local_err
) != 0) {
475 error_report_err(local_err
);
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
, Error
**errp
)
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 Error
*local_err
= NULL
;
567 /* Multiple connections */
568 assert(migrate_use_multifd());
569 start_migration
= multifd_recv_new_channel(ioc
, &local_err
);
571 error_propagate(errp
, local_err
);
576 if (start_migration
) {
577 migration_incoming_process();
582 * @migration_has_all_channels: We have received all channels that we need
584 * Returns true when we have got connections to all the channels that
585 * we need for migration.
587 bool migration_has_all_channels(void)
589 MigrationIncomingState
*mis
= migration_incoming_get_current();
592 all_channels
= multifd_recv_all_channels_created();
594 return all_channels
&& mis
->from_src_file
!= NULL
;
598 * Send a 'SHUT' message on the return channel with the given value
599 * to indicate that we've finished with the RP. Non-0 value indicates
602 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
607 buf
= cpu_to_be32(value
);
608 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
612 * Send a 'PONG' message on the return channel with the given value
613 * (normally in response to a 'PING')
615 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
620 buf
= cpu_to_be32(value
);
621 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
624 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
632 * First, we send the header part. It contains only the len of
633 * idstr, and the idstr itself.
635 len
= strlen(block_name
);
637 memcpy(buf
+ 1, block_name
, len
);
639 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
640 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
645 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
648 * Next, we dump the received bitmap to the stream.
650 * TODO: currently we are safe since we are the only one that is
651 * using the to_src_file handle (fault thread is still paused),
652 * and it's ok even not taking the mutex. However the best way is
653 * to take the lock before sending the message header, and release
654 * the lock after sending the bitmap.
656 qemu_mutex_lock(&mis
->rp_mutex
);
657 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
658 qemu_mutex_unlock(&mis
->rp_mutex
);
660 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
663 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
667 buf
= cpu_to_be32(value
);
668 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
671 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
673 MigrationCapabilityStatusList
*head
= NULL
;
674 MigrationCapabilityStatusList
*caps
;
675 MigrationState
*s
= migrate_get_current();
678 caps
= NULL
; /* silence compiler warning */
679 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
680 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
681 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
686 head
= g_malloc0(sizeof(*caps
));
689 caps
->next
= g_malloc0(sizeof(*caps
));
693 g_malloc(sizeof(*caps
->value
));
694 caps
->value
->capability
= i
;
695 caps
->value
->state
= s
->enabled_capabilities
[i
];
701 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
703 MigrationParameters
*params
;
704 MigrationState
*s
= migrate_get_current();
706 /* TODO use QAPI_CLONE() instead of duplicating it inline */
707 params
= g_malloc0(sizeof(*params
));
708 params
->has_compress_level
= true;
709 params
->compress_level
= s
->parameters
.compress_level
;
710 params
->has_compress_threads
= true;
711 params
->compress_threads
= s
->parameters
.compress_threads
;
712 params
->has_compress_wait_thread
= true;
713 params
->compress_wait_thread
= s
->parameters
.compress_wait_thread
;
714 params
->has_decompress_threads
= true;
715 params
->decompress_threads
= s
->parameters
.decompress_threads
;
716 params
->has_cpu_throttle_initial
= true;
717 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
718 params
->has_cpu_throttle_increment
= true;
719 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
720 params
->has_tls_creds
= true;
721 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
722 params
->has_tls_hostname
= true;
723 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
724 params
->has_max_bandwidth
= true;
725 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
726 params
->has_downtime_limit
= true;
727 params
->downtime_limit
= s
->parameters
.downtime_limit
;
728 params
->has_x_checkpoint_delay
= true;
729 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
730 params
->has_block_incremental
= true;
731 params
->block_incremental
= s
->parameters
.block_incremental
;
732 params
->has_x_multifd_channels
= true;
733 params
->x_multifd_channels
= s
->parameters
.x_multifd_channels
;
734 params
->has_x_multifd_page_count
= true;
735 params
->x_multifd_page_count
= s
->parameters
.x_multifd_page_count
;
736 params
->has_xbzrle_cache_size
= true;
737 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
738 params
->has_max_postcopy_bandwidth
= true;
739 params
->max_postcopy_bandwidth
= s
->parameters
.max_postcopy_bandwidth
;
740 params
->has_max_cpu_throttle
= true;
741 params
->max_cpu_throttle
= s
->parameters
.max_cpu_throttle
;
747 * Return true if we're already in the middle of a migration
748 * (i.e. any of the active or setup states)
750 bool migration_is_setup_or_active(int state
)
753 case MIGRATION_STATUS_ACTIVE
:
754 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
755 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
756 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
757 case MIGRATION_STATUS_SETUP
:
758 case MIGRATION_STATUS_PRE_SWITCHOVER
:
759 case MIGRATION_STATUS_DEVICE
:
768 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
770 info
->has_ram
= true;
771 info
->ram
= g_malloc0(sizeof(*info
->ram
));
772 info
->ram
->transferred
= ram_counters
.transferred
;
773 info
->ram
->total
= ram_bytes_total();
774 info
->ram
->duplicate
= ram_counters
.duplicate
;
775 /* legacy value. It is not used anymore */
776 info
->ram
->skipped
= 0;
777 info
->ram
->normal
= ram_counters
.normal
;
778 info
->ram
->normal_bytes
= ram_counters
.normal
*
779 qemu_target_page_size();
780 info
->ram
->mbps
= s
->mbps
;
781 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
782 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
783 info
->ram
->page_size
= qemu_target_page_size();
784 info
->ram
->multifd_bytes
= ram_counters
.multifd_bytes
;
786 if (migrate_use_xbzrle()) {
787 info
->has_xbzrle_cache
= true;
788 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
789 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
790 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
791 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
792 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
793 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
794 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
797 if (migrate_use_compression()) {
798 info
->has_compression
= true;
799 info
->compression
= g_malloc0(sizeof(*info
->compression
));
800 info
->compression
->pages
= compression_counters
.pages
;
801 info
->compression
->busy
= compression_counters
.busy
;
802 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
803 info
->compression
->compressed_size
=
804 compression_counters
.compressed_size
;
805 info
->compression
->compression_rate
=
806 compression_counters
.compression_rate
;
809 if (cpu_throttle_active()) {
810 info
->has_cpu_throttle_percentage
= true;
811 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
814 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
815 info
->ram
->remaining
= ram_bytes_remaining();
816 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
820 static void populate_disk_info(MigrationInfo
*info
)
822 if (blk_mig_active()) {
823 info
->has_disk
= true;
824 info
->disk
= g_malloc0(sizeof(*info
->disk
));
825 info
->disk
->transferred
= blk_mig_bytes_transferred();
826 info
->disk
->remaining
= blk_mig_bytes_remaining();
827 info
->disk
->total
= blk_mig_bytes_total();
831 static void fill_source_migration_info(MigrationInfo
*info
)
833 MigrationState
*s
= migrate_get_current();
836 case MIGRATION_STATUS_NONE
:
837 /* no migration has happened ever */
838 /* do not overwrite destination migration status */
841 case MIGRATION_STATUS_SETUP
:
842 info
->has_status
= true;
843 info
->has_total_time
= false;
845 case MIGRATION_STATUS_ACTIVE
:
846 case MIGRATION_STATUS_CANCELLING
:
847 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
848 case MIGRATION_STATUS_PRE_SWITCHOVER
:
849 case MIGRATION_STATUS_DEVICE
:
850 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
851 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
852 /* TODO add some postcopy stats */
853 info
->has_status
= true;
854 info
->has_total_time
= true;
855 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
857 info
->has_expected_downtime
= true;
858 info
->expected_downtime
= s
->expected_downtime
;
859 info
->has_setup_time
= true;
860 info
->setup_time
= s
->setup_time
;
862 populate_ram_info(info
, s
);
863 populate_disk_info(info
);
865 case MIGRATION_STATUS_COLO
:
866 info
->has_status
= true;
867 /* TODO: display COLO specific information (checkpoint info etc.) */
869 case MIGRATION_STATUS_COMPLETED
:
870 info
->has_status
= true;
871 info
->has_total_time
= true;
872 info
->total_time
= s
->total_time
;
873 info
->has_downtime
= true;
874 info
->downtime
= s
->downtime
;
875 info
->has_setup_time
= true;
876 info
->setup_time
= s
->setup_time
;
878 populate_ram_info(info
, s
);
880 case MIGRATION_STATUS_FAILED
:
881 info
->has_status
= true;
883 info
->has_error_desc
= true;
884 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
887 case MIGRATION_STATUS_CANCELLED
:
888 info
->has_status
= true;
891 info
->status
= s
->state
;
895 * @migration_caps_check - check capability validity
897 * @cap_list: old capability list, array of bool
898 * @params: new capabilities to be applied soon
899 * @errp: set *errp if the check failed, with reason
901 * Returns true if check passed, otherwise false.
903 static bool migrate_caps_check(bool *cap_list
,
904 MigrationCapabilityStatusList
*params
,
907 MigrationCapabilityStatusList
*cap
;
908 bool old_postcopy_cap
;
909 MigrationIncomingState
*mis
= migration_incoming_get_current();
911 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
913 for (cap
= params
; cap
; cap
= cap
->next
) {
914 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
917 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
918 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
919 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
921 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
926 #ifndef CONFIG_REPLICATION
927 if (cap_list
[MIGRATION_CAPABILITY_X_COLO
]) {
928 error_setg(errp
, "QEMU compiled without replication module"
929 " can't enable COLO");
930 error_append_hint(errp
, "Please enable replication before COLO.\n");
935 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
936 if (cap_list
[MIGRATION_CAPABILITY_COMPRESS
]) {
937 /* The decompression threads asynchronously write into RAM
938 * rather than use the atomic copies needed to avoid
939 * userfaulting. It should be possible to fix the decompression
940 * threads for compatibility in future.
942 error_setg(errp
, "Postcopy is not currently compatible "
947 /* This check is reasonably expensive, so only when it's being
948 * set the first time, also it's only the destination that needs
951 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
952 !postcopy_ram_supported_by_host(mis
)) {
953 /* postcopy_ram_supported_by_host will have emitted a more
956 error_setg(errp
, "Postcopy is not supported");
964 static void fill_destination_migration_info(MigrationInfo
*info
)
966 MigrationIncomingState
*mis
= migration_incoming_get_current();
968 switch (mis
->state
) {
969 case MIGRATION_STATUS_NONE
:
972 case MIGRATION_STATUS_SETUP
:
973 case MIGRATION_STATUS_CANCELLING
:
974 case MIGRATION_STATUS_CANCELLED
:
975 case MIGRATION_STATUS_ACTIVE
:
976 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
977 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
978 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
979 case MIGRATION_STATUS_FAILED
:
980 case MIGRATION_STATUS_COLO
:
981 info
->has_status
= true;
983 case MIGRATION_STATUS_COMPLETED
:
984 info
->has_status
= true;
985 fill_destination_postcopy_migration_info(info
);
988 info
->status
= mis
->state
;
991 MigrationInfo
*qmp_query_migrate(Error
**errp
)
993 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
995 fill_destination_migration_info(info
);
996 fill_source_migration_info(info
);
1001 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
1004 MigrationState
*s
= migrate_get_current();
1005 MigrationCapabilityStatusList
*cap
;
1006 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
1008 if (migration_is_setup_or_active(s
->state
)) {
1009 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1013 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
1014 if (!migrate_caps_check(cap_list
, params
, errp
)) {
1018 for (cap
= params
; cap
; cap
= cap
->next
) {
1019 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
1024 * Check whether the parameters are valid. Error will be put into errp
1025 * (if provided). Return true if valid, otherwise false.
1027 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
1029 if (params
->has_compress_level
&&
1030 (params
->compress_level
> 9)) {
1031 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
1032 "is invalid, it should be in the range of 0 to 9");
1036 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
1037 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1039 "is invalid, it should be in the range of 1 to 255");
1043 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
1044 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1045 "decompress_threads",
1046 "is invalid, it should be in the range of 1 to 255");
1050 if (params
->has_cpu_throttle_initial
&&
1051 (params
->cpu_throttle_initial
< 1 ||
1052 params
->cpu_throttle_initial
> 99)) {
1053 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1054 "cpu_throttle_initial",
1055 "an integer in the range of 1 to 99");
1059 if (params
->has_cpu_throttle_increment
&&
1060 (params
->cpu_throttle_increment
< 1 ||
1061 params
->cpu_throttle_increment
> 99)) {
1062 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1063 "cpu_throttle_increment",
1064 "an integer in the range of 1 to 99");
1068 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
1069 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
1070 " range of 0 to %zu bytes/second", SIZE_MAX
);
1074 if (params
->has_downtime_limit
&&
1075 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
1076 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1077 "the range of 0 to %d milliseconds",
1078 MAX_MIGRATE_DOWNTIME
);
1082 /* x_checkpoint_delay is now always positive */
1084 if (params
->has_x_multifd_channels
&& (params
->x_multifd_channels
< 1)) {
1085 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1087 "is invalid, it should be in the range of 1 to 255");
1090 if (params
->has_x_multifd_page_count
&&
1091 (params
->x_multifd_page_count
< 1 ||
1092 params
->x_multifd_page_count
> 10000)) {
1093 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1094 "multifd_page_count",
1095 "is invalid, it should be in the range of 1 to 10000");
1099 if (params
->has_xbzrle_cache_size
&&
1100 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
1101 !is_power_of_2(params
->xbzrle_cache_size
))) {
1102 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1103 "xbzrle_cache_size",
1104 "is invalid, it should be bigger than target page size"
1105 " and a power of two");
1109 if (params
->has_max_cpu_throttle
&&
1110 (params
->max_cpu_throttle
< params
->cpu_throttle_initial
||
1111 params
->max_cpu_throttle
> 99)) {
1112 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1114 "an integer in the range of cpu_throttle_initial to 99");
1121 static void migrate_params_test_apply(MigrateSetParameters
*params
,
1122 MigrationParameters
*dest
)
1124 *dest
= migrate_get_current()->parameters
;
1126 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1128 if (params
->has_compress_level
) {
1129 dest
->compress_level
= params
->compress_level
;
1132 if (params
->has_compress_threads
) {
1133 dest
->compress_threads
= params
->compress_threads
;
1136 if (params
->has_compress_wait_thread
) {
1137 dest
->compress_wait_thread
= params
->compress_wait_thread
;
1140 if (params
->has_decompress_threads
) {
1141 dest
->decompress_threads
= params
->decompress_threads
;
1144 if (params
->has_cpu_throttle_initial
) {
1145 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
1148 if (params
->has_cpu_throttle_increment
) {
1149 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
1152 if (params
->has_tls_creds
) {
1153 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1154 dest
->tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1157 if (params
->has_tls_hostname
) {
1158 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1159 dest
->tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1162 if (params
->has_max_bandwidth
) {
1163 dest
->max_bandwidth
= params
->max_bandwidth
;
1166 if (params
->has_downtime_limit
) {
1167 dest
->downtime_limit
= params
->downtime_limit
;
1170 if (params
->has_x_checkpoint_delay
) {
1171 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
1174 if (params
->has_block_incremental
) {
1175 dest
->block_incremental
= params
->block_incremental
;
1177 if (params
->has_x_multifd_channels
) {
1178 dest
->x_multifd_channels
= params
->x_multifd_channels
;
1180 if (params
->has_x_multifd_page_count
) {
1181 dest
->x_multifd_page_count
= params
->x_multifd_page_count
;
1183 if (params
->has_xbzrle_cache_size
) {
1184 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
1186 if (params
->has_max_postcopy_bandwidth
) {
1187 dest
->max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1189 if (params
->has_max_cpu_throttle
) {
1190 dest
->max_cpu_throttle
= params
->max_cpu_throttle
;
1194 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
1196 MigrationState
*s
= migrate_get_current();
1198 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1200 if (params
->has_compress_level
) {
1201 s
->parameters
.compress_level
= params
->compress_level
;
1204 if (params
->has_compress_threads
) {
1205 s
->parameters
.compress_threads
= params
->compress_threads
;
1208 if (params
->has_compress_wait_thread
) {
1209 s
->parameters
.compress_wait_thread
= params
->compress_wait_thread
;
1212 if (params
->has_decompress_threads
) {
1213 s
->parameters
.decompress_threads
= params
->decompress_threads
;
1216 if (params
->has_cpu_throttle_initial
) {
1217 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
1220 if (params
->has_cpu_throttle_increment
) {
1221 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
1224 if (params
->has_tls_creds
) {
1225 g_free(s
->parameters
.tls_creds
);
1226 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1227 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1230 if (params
->has_tls_hostname
) {
1231 g_free(s
->parameters
.tls_hostname
);
1232 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1233 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1236 if (params
->has_max_bandwidth
) {
1237 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1238 if (s
->to_dst_file
) {
1239 qemu_file_set_rate_limit(s
->to_dst_file
,
1240 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1244 if (params
->has_downtime_limit
) {
1245 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1248 if (params
->has_x_checkpoint_delay
) {
1249 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1250 if (migration_in_colo_state()) {
1251 colo_checkpoint_notify(s
);
1255 if (params
->has_block_incremental
) {
1256 s
->parameters
.block_incremental
= params
->block_incremental
;
1258 if (params
->has_x_multifd_channels
) {
1259 s
->parameters
.x_multifd_channels
= params
->x_multifd_channels
;
1261 if (params
->has_x_multifd_page_count
) {
1262 s
->parameters
.x_multifd_page_count
= params
->x_multifd_page_count
;
1264 if (params
->has_xbzrle_cache_size
) {
1265 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1266 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1268 if (params
->has_max_postcopy_bandwidth
) {
1269 s
->parameters
.max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1271 if (params
->has_max_cpu_throttle
) {
1272 s
->parameters
.max_cpu_throttle
= params
->max_cpu_throttle
;
1276 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1278 MigrationParameters tmp
;
1280 /* TODO Rewrite "" to null instead */
1281 if (params
->has_tls_creds
1282 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1283 qobject_unref(params
->tls_creds
->u
.n
);
1284 params
->tls_creds
->type
= QTYPE_QSTRING
;
1285 params
->tls_creds
->u
.s
= strdup("");
1287 /* TODO Rewrite "" to null instead */
1288 if (params
->has_tls_hostname
1289 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1290 qobject_unref(params
->tls_hostname
->u
.n
);
1291 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1292 params
->tls_hostname
->u
.s
= strdup("");
1295 migrate_params_test_apply(params
, &tmp
);
1297 if (!migrate_params_check(&tmp
, errp
)) {
1298 /* Invalid parameter */
1302 migrate_params_apply(params
, errp
);
1306 void qmp_migrate_start_postcopy(Error
**errp
)
1308 MigrationState
*s
= migrate_get_current();
1310 if (!migrate_postcopy()) {
1311 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1312 " the start of migration");
1316 if (s
->state
== MIGRATION_STATUS_NONE
) {
1317 error_setg(errp
, "Postcopy must be started after migration has been"
1322 * we don't error if migration has finished since that would be racy
1323 * with issuing this command.
1325 atomic_set(&s
->start_postcopy
, true);
1328 /* shared migration helpers */
1330 void migrate_set_state(int *state
, int old_state
, int new_state
)
1332 assert(new_state
< MIGRATION_STATUS__MAX
);
1333 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1334 trace_migrate_set_state(MigrationStatus_str(new_state
));
1335 migrate_generate_event(new_state
);
1339 static MigrationCapabilityStatusList
*migrate_cap_add(
1340 MigrationCapabilityStatusList
*list
,
1341 MigrationCapability index
,
1344 MigrationCapabilityStatusList
*cap
;
1346 cap
= g_new0(MigrationCapabilityStatusList
, 1);
1347 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
1348 cap
->value
->capability
= index
;
1349 cap
->value
->state
= state
;
1355 void migrate_set_block_enabled(bool value
, Error
**errp
)
1357 MigrationCapabilityStatusList
*cap
;
1359 cap
= migrate_cap_add(NULL
, MIGRATION_CAPABILITY_BLOCK
, value
);
1360 qmp_migrate_set_capabilities(cap
, errp
);
1361 qapi_free_MigrationCapabilityStatusList(cap
);
1364 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1366 s
->parameters
.block_incremental
= value
;
1369 static void block_cleanup_parameters(MigrationState
*s
)
1371 if (s
->must_remove_block_options
) {
1372 /* setting to false can never fail */
1373 migrate_set_block_enabled(false, &error_abort
);
1374 migrate_set_block_incremental(s
, false);
1375 s
->must_remove_block_options
= false;
1379 static void migrate_fd_cleanup(void *opaque
)
1381 MigrationState
*s
= opaque
;
1383 qemu_bh_delete(s
->cleanup_bh
);
1384 s
->cleanup_bh
= NULL
;
1386 qemu_savevm_state_cleanup();
1388 if (s
->to_dst_file
) {
1391 trace_migrate_fd_cleanup();
1392 qemu_mutex_unlock_iothread();
1393 if (s
->migration_thread_running
) {
1394 qemu_thread_join(&s
->thread
);
1395 s
->migration_thread_running
= false;
1397 qemu_mutex_lock_iothread();
1399 multifd_save_cleanup();
1400 qemu_mutex_lock(&s
->qemu_file_lock
);
1401 tmp
= s
->to_dst_file
;
1402 s
->to_dst_file
= NULL
;
1403 qemu_mutex_unlock(&s
->qemu_file_lock
);
1405 * Close the file handle without the lock to make sure the
1406 * critical section won't block for long.
1411 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
1412 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
1414 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1415 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1416 MIGRATION_STATUS_CANCELLED
);
1420 /* It is used on info migrate. We can't free it */
1421 error_report_err(error_copy(s
->error
));
1423 notifier_list_notify(&migration_state_notifiers
, s
);
1424 block_cleanup_parameters(s
);
1427 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1429 qemu_mutex_lock(&s
->error_mutex
);
1431 s
->error
= error_copy(error
);
1433 qemu_mutex_unlock(&s
->error_mutex
);
1436 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1438 trace_migrate_fd_error(error_get_pretty(error
));
1439 assert(s
->to_dst_file
== NULL
);
1440 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1441 MIGRATION_STATUS_FAILED
);
1442 migrate_set_error(s
, error
);
1445 static void migrate_fd_cancel(MigrationState
*s
)
1448 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1449 trace_migrate_fd_cancel();
1451 if (s
->rp_state
.from_dst_file
) {
1452 /* shutdown the rp socket, so causing the rp thread to shutdown */
1453 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1457 old_state
= s
->state
;
1458 if (!migration_is_setup_or_active(old_state
)) {
1461 /* If the migration is paused, kick it out of the pause */
1462 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1463 qemu_sem_post(&s
->pause_sem
);
1465 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1466 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1469 * If we're unlucky the migration code might be stuck somewhere in a
1470 * send/write while the network has failed and is waiting to timeout;
1471 * if we've got shutdown(2) available then we can force it to quit.
1472 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1473 * called in a bh, so there is no race against this cancel.
1475 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1476 qemu_file_shutdown(f
);
1478 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1479 Error
*local_err
= NULL
;
1481 bdrv_invalidate_cache_all(&local_err
);
1483 error_report_err(local_err
);
1485 s
->block_inactive
= false;
1490 void add_migration_state_change_notifier(Notifier
*notify
)
1492 notifier_list_add(&migration_state_notifiers
, notify
);
1495 void remove_migration_state_change_notifier(Notifier
*notify
)
1497 notifier_remove(notify
);
1500 bool migration_in_setup(MigrationState
*s
)
1502 return s
->state
== MIGRATION_STATUS_SETUP
;
1505 bool migration_has_finished(MigrationState
*s
)
1507 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1510 bool migration_has_failed(MigrationState
*s
)
1512 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1513 s
->state
== MIGRATION_STATUS_FAILED
);
1516 bool migration_in_postcopy(void)
1518 MigrationState
*s
= migrate_get_current();
1520 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1523 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1525 return migration_in_postcopy() && s
->postcopy_after_devices
;
1528 bool migration_is_idle(void)
1530 MigrationState
*s
= migrate_get_current();
1533 case MIGRATION_STATUS_NONE
:
1534 case MIGRATION_STATUS_CANCELLED
:
1535 case MIGRATION_STATUS_COMPLETED
:
1536 case MIGRATION_STATUS_FAILED
:
1538 case MIGRATION_STATUS_SETUP
:
1539 case MIGRATION_STATUS_CANCELLING
:
1540 case MIGRATION_STATUS_ACTIVE
:
1541 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1542 case MIGRATION_STATUS_COLO
:
1543 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1544 case MIGRATION_STATUS_DEVICE
:
1546 case MIGRATION_STATUS__MAX
:
1547 g_assert_not_reached();
1553 void migrate_init(MigrationState
*s
)
1556 * Reinitialise all migration state, except
1557 * parameters/capabilities that the user set, and
1563 s
->to_dst_file
= NULL
;
1564 s
->state
= MIGRATION_STATUS_NONE
;
1565 s
->rp_state
.from_dst_file
= NULL
;
1566 s
->rp_state
.error
= false;
1569 s
->expected_downtime
= 0;
1571 s
->start_postcopy
= false;
1572 s
->postcopy_after_devices
= false;
1573 s
->migration_thread_running
= false;
1574 error_free(s
->error
);
1577 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1579 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1581 s
->vm_was_running
= false;
1582 s
->iteration_initial_bytes
= 0;
1583 s
->threshold_size
= 0;
1586 static GSList
*migration_blockers
;
1588 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1590 if (migrate_get_current()->only_migratable
) {
1591 error_propagate_prepend(errp
, error_copy(reason
),
1592 "disallowing migration blocker "
1593 "(--only_migratable) for: ");
1597 if (migration_is_idle()) {
1598 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1602 error_propagate_prepend(errp
, error_copy(reason
),
1603 "disallowing migration blocker "
1604 "(migration in progress) for: ");
1608 void migrate_del_blocker(Error
*reason
)
1610 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1613 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1615 Error
*local_err
= NULL
;
1616 static bool once
= true;
1618 if (!deferred_incoming
) {
1619 error_setg(errp
, "For use with '-incoming defer'");
1623 error_setg(errp
, "The incoming migration has already been started");
1626 qemu_start_incoming_migration(uri
, &local_err
);
1629 error_propagate(errp
, local_err
);
1636 void qmp_migrate_recover(const char *uri
, Error
**errp
)
1638 MigrationIncomingState
*mis
= migration_incoming_get_current();
1640 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1641 error_setg(errp
, "Migrate recover can only be run "
1642 "when postcopy is paused.");
1646 if (atomic_cmpxchg(&mis
->postcopy_recover_triggered
,
1647 false, true) == true) {
1648 error_setg(errp
, "Migrate recovery is triggered already");
1653 * Note that this call will never start a real migration; it will
1654 * only re-setup the migration stream and poke existing migration
1655 * to continue using that newly established channel.
1657 qemu_start_incoming_migration(uri
, errp
);
1660 void qmp_migrate_pause(Error
**errp
)
1662 MigrationState
*ms
= migrate_get_current();
1663 MigrationIncomingState
*mis
= migration_incoming_get_current();
1666 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1667 /* Source side, during postcopy */
1668 qemu_mutex_lock(&ms
->qemu_file_lock
);
1669 ret
= qemu_file_shutdown(ms
->to_dst_file
);
1670 qemu_mutex_unlock(&ms
->qemu_file_lock
);
1672 error_setg(errp
, "Failed to pause source migration");
1677 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1678 ret
= qemu_file_shutdown(mis
->from_src_file
);
1680 error_setg(errp
, "Failed to pause destination migration");
1685 error_setg(errp
, "migrate-pause is currently only supported "
1686 "during postcopy-active state");
1689 bool migration_is_blocked(Error
**errp
)
1691 if (qemu_savevm_state_blocked(errp
)) {
1695 if (migration_blockers
) {
1696 error_propagate(errp
, error_copy(migration_blockers
->data
));
1703 /* Returns true if continue to migrate, or false if error detected */
1704 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
1705 bool resume
, Error
**errp
)
1707 Error
*local_err
= NULL
;
1710 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1711 error_setg(errp
, "Cannot resume if there is no "
1712 "paused migration");
1717 * Postcopy recovery won't work well with release-ram
1718 * capability since release-ram will drop the page buffer as
1719 * long as the page is put into the send buffer. So if there
1720 * is a network failure happened, any page buffers that have
1721 * not yet reached the destination VM but have already been
1722 * sent from the source VM will be lost forever. Let's refuse
1723 * the client from resuming such a postcopy migration.
1724 * Luckily release-ram was designed to only be used when src
1725 * and destination VMs are on the same host, so it should be
1728 if (migrate_release_ram()) {
1729 error_setg(errp
, "Postcopy recovery cannot work "
1730 "when release-ram capability is set");
1734 /* This is a resume, skip init status */
1738 if (migration_is_setup_or_active(s
->state
) ||
1739 s
->state
== MIGRATION_STATUS_CANCELLING
||
1740 s
->state
== MIGRATION_STATUS_COLO
) {
1741 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1745 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1746 error_setg(errp
, "Guest is waiting for an incoming migration");
1750 if (migration_is_blocked(errp
)) {
1754 if (blk
|| blk_inc
) {
1755 if (migrate_use_block() || migrate_use_block_incremental()) {
1756 error_setg(errp
, "Command options are incompatible with "
1757 "current migration capabilities");
1760 migrate_set_block_enabled(true, &local_err
);
1762 error_propagate(errp
, local_err
);
1765 s
->must_remove_block_options
= true;
1769 migrate_set_block_incremental(s
, true);
1777 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1778 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1779 bool has_resume
, bool resume
, Error
**errp
)
1781 Error
*local_err
= NULL
;
1782 MigrationState
*s
= migrate_get_current();
1785 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
1786 has_resume
&& resume
, errp
)) {
1787 /* Error detected, put into errp */
1791 if (strstart(uri
, "tcp:", &p
)) {
1792 tcp_start_outgoing_migration(s
, p
, &local_err
);
1794 } else if (strstart(uri
, "rdma:", &p
)) {
1795 rdma_start_outgoing_migration(s
, p
, &local_err
);
1797 } else if (strstart(uri
, "exec:", &p
)) {
1798 exec_start_outgoing_migration(s
, p
, &local_err
);
1799 } else if (strstart(uri
, "unix:", &p
)) {
1800 unix_start_outgoing_migration(s
, p
, &local_err
);
1801 } else if (strstart(uri
, "fd:", &p
)) {
1802 fd_start_outgoing_migration(s
, p
, &local_err
);
1804 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1805 "a valid migration protocol");
1806 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1807 MIGRATION_STATUS_FAILED
);
1808 block_cleanup_parameters(s
);
1813 migrate_fd_error(s
, local_err
);
1814 error_propagate(errp
, local_err
);
1819 void qmp_migrate_cancel(Error
**errp
)
1821 migrate_fd_cancel(migrate_get_current());
1824 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
1826 MigrationState
*s
= migrate_get_current();
1827 if (s
->state
!= state
) {
1828 error_setg(errp
, "Migration not in expected state: %s",
1829 MigrationStatus_str(s
->state
));
1832 qemu_sem_post(&s
->pause_sem
);
1835 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1837 MigrateSetParameters p
= {
1838 .has_xbzrle_cache_size
= true,
1839 .xbzrle_cache_size
= value
,
1842 qmp_migrate_set_parameters(&p
, errp
);
1845 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1847 return migrate_xbzrle_cache_size();
1850 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1852 MigrateSetParameters p
= {
1853 .has_max_bandwidth
= true,
1854 .max_bandwidth
= value
,
1857 qmp_migrate_set_parameters(&p
, errp
);
1860 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1862 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
1863 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1864 "the range of 0 to %d seconds",
1865 MAX_MIGRATE_DOWNTIME_SECONDS
);
1869 value
*= 1000; /* Convert to milliseconds */
1870 value
= MAX(0, MIN(INT64_MAX
, value
));
1872 MigrateSetParameters p
= {
1873 .has_downtime_limit
= true,
1874 .downtime_limit
= value
,
1877 qmp_migrate_set_parameters(&p
, errp
);
1880 bool migrate_release_ram(void)
1884 s
= migrate_get_current();
1886 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
1889 bool migrate_postcopy_ram(void)
1893 s
= migrate_get_current();
1895 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1898 bool migrate_postcopy(void)
1900 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
1903 bool migrate_auto_converge(void)
1907 s
= migrate_get_current();
1909 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1912 bool migrate_zero_blocks(void)
1916 s
= migrate_get_current();
1918 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
1921 bool migrate_postcopy_blocktime(void)
1925 s
= migrate_get_current();
1927 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
1930 bool migrate_use_compression(void)
1934 s
= migrate_get_current();
1936 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
1939 int migrate_compress_level(void)
1943 s
= migrate_get_current();
1945 return s
->parameters
.compress_level
;
1948 int migrate_compress_threads(void)
1952 s
= migrate_get_current();
1954 return s
->parameters
.compress_threads
;
1957 int migrate_compress_wait_thread(void)
1961 s
= migrate_get_current();
1963 return s
->parameters
.compress_wait_thread
;
1966 int migrate_decompress_threads(void)
1970 s
= migrate_get_current();
1972 return s
->parameters
.decompress_threads
;
1975 bool migrate_dirty_bitmaps(void)
1979 s
= migrate_get_current();
1981 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
1984 bool migrate_use_events(void)
1988 s
= migrate_get_current();
1990 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
1993 bool migrate_use_multifd(void)
1997 s
= migrate_get_current();
1999 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_MULTIFD
];
2002 bool migrate_pause_before_switchover(void)
2006 s
= migrate_get_current();
2008 return s
->enabled_capabilities
[
2009 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
2012 int migrate_multifd_channels(void)
2016 s
= migrate_get_current();
2018 return s
->parameters
.x_multifd_channels
;
2021 int migrate_multifd_page_count(void)
2025 s
= migrate_get_current();
2027 return s
->parameters
.x_multifd_page_count
;
2030 int migrate_use_xbzrle(void)
2034 s
= migrate_get_current();
2036 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
2039 int64_t migrate_xbzrle_cache_size(void)
2043 s
= migrate_get_current();
2045 return s
->parameters
.xbzrle_cache_size
;
2048 static int64_t migrate_max_postcopy_bandwidth(void)
2052 s
= migrate_get_current();
2054 return s
->parameters
.max_postcopy_bandwidth
;
2057 bool migrate_use_block(void)
2061 s
= migrate_get_current();
2063 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
2066 bool migrate_use_return_path(void)
2070 s
= migrate_get_current();
2072 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
2075 bool migrate_use_block_incremental(void)
2079 s
= migrate_get_current();
2081 return s
->parameters
.block_incremental
;
2084 /* migration thread support */
2086 * Something bad happened to the RP stream, mark an error
2087 * The caller shall print or trace something to indicate why
2089 static void mark_source_rp_bad(MigrationState
*s
)
2091 s
->rp_state
.error
= true;
2094 static struct rp_cmd_args
{
2095 ssize_t len
; /* -1 = variable */
2098 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
2099 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
2100 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
2101 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
2102 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
2103 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
2104 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
2105 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
2109 * Process a request for pages received on the return path,
2110 * We're allowed to send more than requested (e.g. to round to our page size)
2111 * and we don't need to send pages that have already been sent.
2113 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
2114 ram_addr_t start
, size_t len
)
2116 long our_host_ps
= getpagesize();
2118 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
2121 * Since we currently insist on matching page sizes, just sanity check
2122 * we're being asked for whole host pages.
2124 if (start
& (our_host_ps
-1) ||
2125 (len
& (our_host_ps
-1))) {
2126 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2127 " len: %zd", __func__
, start
, len
);
2128 mark_source_rp_bad(ms
);
2132 if (ram_save_queue_pages(rbname
, start
, len
)) {
2133 mark_source_rp_bad(ms
);
2137 /* Return true to retry, false to quit */
2138 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
2140 trace_postcopy_pause_return_path();
2142 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
2144 trace_postcopy_pause_return_path_continued();
2149 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
2151 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
2154 error_report("%s: invalid block name '%s'", __func__
, block_name
);
2158 /* Fetch the received bitmap and refresh the dirty bitmap */
2159 return ram_dirty_bitmap_reload(s
, block
);
2162 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
2164 trace_source_return_path_thread_resume_ack(value
);
2166 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
2167 error_report("%s: illegal resume_ack value %"PRIu32
,
2172 /* Now both sides are active. */
2173 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2174 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2176 /* Notify send thread that time to continue send pages */
2177 qemu_sem_post(&s
->rp_state
.rp_sem
);
2183 * Handles messages sent on the return path towards the source VM
2186 static void *source_return_path_thread(void *opaque
)
2188 MigrationState
*ms
= opaque
;
2189 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
2190 uint16_t header_len
, header_type
;
2192 uint32_t tmp32
, sibling_error
;
2193 ram_addr_t start
= 0; /* =0 to silence warning */
2194 size_t len
= 0, expected_len
;
2197 trace_source_return_path_thread_entry();
2198 rcu_register_thread();
2201 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
2202 migration_is_setup_or_active(ms
->state
)) {
2203 trace_source_return_path_thread_loop_top();
2204 header_type
= qemu_get_be16(rp
);
2205 header_len
= qemu_get_be16(rp
);
2207 if (qemu_file_get_error(rp
)) {
2208 mark_source_rp_bad(ms
);
2212 if (header_type
>= MIG_RP_MSG_MAX
||
2213 header_type
== MIG_RP_MSG_INVALID
) {
2214 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2215 header_type
, header_len
);
2216 mark_source_rp_bad(ms
);
2220 if ((rp_cmd_args
[header_type
].len
!= -1 &&
2221 header_len
!= rp_cmd_args
[header_type
].len
) ||
2222 header_len
> sizeof(buf
)) {
2223 error_report("RP: Received '%s' message (0x%04x) with"
2224 "incorrect length %d expecting %zu",
2225 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
2226 (size_t)rp_cmd_args
[header_type
].len
);
2227 mark_source_rp_bad(ms
);
2231 /* We know we've got a valid header by this point */
2232 res
= qemu_get_buffer(rp
, buf
, header_len
);
2233 if (res
!= header_len
) {
2234 error_report("RP: Failed reading data for message 0x%04x"
2235 " read %d expected %d",
2236 header_type
, res
, header_len
);
2237 mark_source_rp_bad(ms
);
2241 /* OK, we have the message and the data */
2242 switch (header_type
) {
2243 case MIG_RP_MSG_SHUT
:
2244 sibling_error
= ldl_be_p(buf
);
2245 trace_source_return_path_thread_shut(sibling_error
);
2246 if (sibling_error
) {
2247 error_report("RP: Sibling indicated error %d", sibling_error
);
2248 mark_source_rp_bad(ms
);
2251 * We'll let the main thread deal with closing the RP
2252 * we could do a shutdown(2) on it, but we're the only user
2253 * anyway, so there's nothing gained.
2257 case MIG_RP_MSG_PONG
:
2258 tmp32
= ldl_be_p(buf
);
2259 trace_source_return_path_thread_pong(tmp32
);
2262 case MIG_RP_MSG_REQ_PAGES
:
2263 start
= ldq_be_p(buf
);
2264 len
= ldl_be_p(buf
+ 8);
2265 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
2268 case MIG_RP_MSG_REQ_PAGES_ID
:
2269 expected_len
= 12 + 1; /* header + termination */
2271 if (header_len
>= expected_len
) {
2272 start
= ldq_be_p(buf
);
2273 len
= ldl_be_p(buf
+ 8);
2274 /* Now we expect an idstr */
2275 tmp32
= buf
[12]; /* Length of the following idstr */
2276 buf
[13 + tmp32
] = '\0';
2277 expected_len
+= tmp32
;
2279 if (header_len
!= expected_len
) {
2280 error_report("RP: Req_Page_id with length %d expecting %zd",
2281 header_len
, expected_len
);
2282 mark_source_rp_bad(ms
);
2285 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
2288 case MIG_RP_MSG_RECV_BITMAP
:
2289 if (header_len
< 1) {
2290 error_report("%s: missing block name", __func__
);
2291 mark_source_rp_bad(ms
);
2294 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2295 buf
[buf
[0] + 1] = '\0';
2296 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
2297 mark_source_rp_bad(ms
);
2302 case MIG_RP_MSG_RESUME_ACK
:
2303 tmp32
= ldl_be_p(buf
);
2304 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
2305 mark_source_rp_bad(ms
);
2316 res
= qemu_file_get_error(rp
);
2320 * Maybe there is something we can do: it looks like a
2321 * network down issue, and we pause for a recovery.
2323 if (postcopy_pause_return_path_thread(ms
)) {
2324 /* Reload rp, reset the rest */
2325 if (rp
!= ms
->rp_state
.from_dst_file
) {
2327 rp
= ms
->rp_state
.from_dst_file
;
2329 ms
->rp_state
.error
= false;
2334 trace_source_return_path_thread_bad_end();
2335 mark_source_rp_bad(ms
);
2338 trace_source_return_path_thread_end();
2339 ms
->rp_state
.from_dst_file
= NULL
;
2341 rcu_unregister_thread();
2345 static int open_return_path_on_source(MigrationState
*ms
,
2349 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
2350 if (!ms
->rp_state
.from_dst_file
) {
2354 trace_open_return_path_on_source();
2356 if (!create_thread
) {
2361 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
2362 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
2364 trace_open_return_path_on_source_continue();
2369 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2370 static int await_return_path_close_on_source(MigrationState
*ms
)
2373 * If this is a normal exit then the destination will send a SHUT and the
2374 * rp_thread will exit, however if there's an error we need to cause
2377 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
2379 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2380 * waiting for the destination.
2382 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
2383 mark_source_rp_bad(ms
);
2385 trace_await_return_path_close_on_source_joining();
2386 qemu_thread_join(&ms
->rp_state
.rp_thread
);
2387 trace_await_return_path_close_on_source_close();
2388 return ms
->rp_state
.error
;
2392 * Switch from normal iteration to postcopy
2393 * Returns non-0 on error
2395 static int postcopy_start(MigrationState
*ms
)
2398 QIOChannelBuffer
*bioc
;
2400 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2401 int64_t bandwidth
= migrate_max_postcopy_bandwidth();
2402 bool restart_block
= false;
2403 int cur_state
= MIGRATION_STATUS_ACTIVE
;
2404 if (!migrate_pause_before_switchover()) {
2405 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
2406 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2409 trace_postcopy_start();
2410 qemu_mutex_lock_iothread();
2411 trace_postcopy_start_set_run();
2413 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2414 global_state_store();
2415 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2420 ret
= migration_maybe_pause(ms
, &cur_state
,
2421 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2426 ret
= bdrv_inactivate_all();
2430 restart_block
= true;
2433 * Cause any non-postcopiable, but iterative devices to
2434 * send out their final data.
2436 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2439 * in Finish migrate and with the io-lock held everything should
2440 * be quiet, but we've potentially still got dirty pages and we
2441 * need to tell the destination to throw any pages it's already received
2444 if (migrate_postcopy_ram()) {
2445 if (ram_postcopy_send_discard_bitmap(ms
)) {
2446 error_report("postcopy send discard bitmap failed");
2452 * send rest of state - note things that are doing postcopy
2453 * will notice we're in POSTCOPY_ACTIVE and not actually
2454 * wrap their state up here
2456 /* 0 max-postcopy-bandwidth means unlimited */
2458 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
2460 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
2462 if (migrate_postcopy_ram()) {
2463 /* Ping just for debugging, helps line traces up */
2464 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
2468 * While loading the device state we may trigger page transfer
2469 * requests and the fd must be free to process those, and thus
2470 * the destination must read the whole device state off the fd before
2471 * it starts processing it. Unfortunately the ad-hoc migration format
2472 * doesn't allow the destination to know the size to read without fully
2473 * parsing it through each devices load-state code (especially the open
2474 * coded devices that use get/put).
2475 * So we wrap the device state up in a package with a length at the start;
2476 * to do this we use a qemu_buf to hold the whole of the device state.
2478 bioc
= qio_channel_buffer_new(4096);
2479 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
2480 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
2481 object_unref(OBJECT(bioc
));
2484 * Make sure the receiver can get incoming pages before we send the rest
2487 qemu_savevm_send_postcopy_listen(fb
);
2489 qemu_savevm_state_complete_precopy(fb
, false, false);
2490 if (migrate_postcopy_ram()) {
2491 qemu_savevm_send_ping(fb
, 3);
2494 qemu_savevm_send_postcopy_run(fb
);
2496 /* <><> end of stuff going into the package */
2498 /* Last point of recovery; as soon as we send the package the destination
2499 * can open devices and potentially start running.
2500 * Lets just check again we've not got any errors.
2502 ret
= qemu_file_get_error(ms
->to_dst_file
);
2504 error_report("postcopy_start: Migration stream errored (pre package)");
2508 restart_block
= false;
2510 /* Now send that blob */
2511 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2516 /* Send a notify to give a chance for anything that needs to happen
2517 * at the transition to postcopy and after the device state; in particular
2518 * spice needs to trigger a transition now
2520 ms
->postcopy_after_devices
= true;
2521 notifier_list_notify(&migration_state_notifiers
, ms
);
2523 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2525 qemu_mutex_unlock_iothread();
2527 if (migrate_postcopy_ram()) {
2529 * Although this ping is just for debug, it could potentially be
2530 * used for getting a better measurement of downtime at the source.
2532 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2535 if (migrate_release_ram()) {
2536 ram_postcopy_migrated_memory_release(ms
);
2539 ret
= qemu_file_get_error(ms
->to_dst_file
);
2541 error_report("postcopy_start: Migration stream errored");
2542 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2543 MIGRATION_STATUS_FAILED
);
2551 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2552 MIGRATION_STATUS_FAILED
);
2553 if (restart_block
) {
2554 /* A failure happened early enough that we know the destination hasn't
2555 * accessed block devices, so we're safe to recover.
2557 Error
*local_err
= NULL
;
2559 bdrv_invalidate_cache_all(&local_err
);
2561 error_report_err(local_err
);
2564 qemu_mutex_unlock_iothread();
2569 * migration_maybe_pause: Pause if required to by
2570 * migrate_pause_before_switchover called with the iothread locked
2571 * Returns: 0 on success
2573 static int migration_maybe_pause(MigrationState
*s
,
2574 int *current_active_state
,
2577 if (!migrate_pause_before_switchover()) {
2581 /* Since leaving this state is not atomic with posting the semaphore
2582 * it's possible that someone could have issued multiple migrate_continue
2583 * and the semaphore is incorrectly positive at this point;
2584 * the docs say it's undefined to reinit a semaphore that's already
2585 * init'd, so use timedwait to eat up any existing posts.
2587 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2588 /* This block intentionally left blank */
2591 qemu_mutex_unlock_iothread();
2592 migrate_set_state(&s
->state
, *current_active_state
,
2593 MIGRATION_STATUS_PRE_SWITCHOVER
);
2594 qemu_sem_wait(&s
->pause_sem
);
2595 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2597 *current_active_state
= new_state
;
2598 qemu_mutex_lock_iothread();
2600 return s
->state
== new_state
? 0 : -EINVAL
;
2604 * migration_completion: Used by migration_thread when there's not much left.
2605 * The caller 'breaks' the loop when this returns.
2607 * @s: Current migration state
2609 static void migration_completion(MigrationState
*s
)
2612 int current_active_state
= s
->state
;
2614 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2615 qemu_mutex_lock_iothread();
2616 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2617 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2618 s
->vm_was_running
= runstate_is_running();
2619 ret
= global_state_store();
2622 bool inactivate
= !migrate_colo_enabled();
2623 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2625 ret
= migration_maybe_pause(s
, ¤t_active_state
,
2626 MIGRATION_STATUS_DEVICE
);
2629 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
2630 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
2633 if (inactivate
&& ret
>= 0) {
2634 s
->block_inactive
= true;
2637 qemu_mutex_unlock_iothread();
2642 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2643 trace_migration_completion_postcopy_end();
2645 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
2646 trace_migration_completion_postcopy_end_after_complete();
2650 * If rp was opened we must clean up the thread before
2651 * cleaning everything else up (since if there are no failures
2652 * it will wait for the destination to send it's status in
2655 if (s
->rp_state
.from_dst_file
) {
2657 trace_migration_return_path_end_before();
2658 rp_error
= await_return_path_close_on_source(s
);
2659 trace_migration_return_path_end_after(rp_error
);
2661 goto fail_invalidate
;
2665 if (qemu_file_get_error(s
->to_dst_file
)) {
2666 trace_migration_completion_file_err();
2667 goto fail_invalidate
;
2670 if (!migrate_colo_enabled()) {
2671 migrate_set_state(&s
->state
, current_active_state
,
2672 MIGRATION_STATUS_COMPLETED
);
2678 /* If not doing postcopy, vm_start() will be called: let's regain
2679 * control on images.
2681 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
2682 s
->state
== MIGRATION_STATUS_DEVICE
) {
2683 Error
*local_err
= NULL
;
2685 qemu_mutex_lock_iothread();
2686 bdrv_invalidate_cache_all(&local_err
);
2688 error_report_err(local_err
);
2690 s
->block_inactive
= false;
2692 qemu_mutex_unlock_iothread();
2696 migrate_set_state(&s
->state
, current_active_state
,
2697 MIGRATION_STATUS_FAILED
);
2700 bool migrate_colo_enabled(void)
2702 MigrationState
*s
= migrate_get_current();
2703 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
2706 typedef enum MigThrError
{
2707 /* No error detected */
2708 MIG_THR_ERR_NONE
= 0,
2709 /* Detected error, but resumed successfully */
2710 MIG_THR_ERR_RECOVERED
= 1,
2711 /* Detected fatal error, need to exit */
2712 MIG_THR_ERR_FATAL
= 2,
2715 static int postcopy_resume_handshake(MigrationState
*s
)
2717 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
2719 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2720 qemu_sem_wait(&s
->rp_state
.rp_sem
);
2723 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2730 /* Return zero if success, or <0 for error */
2731 static int postcopy_do_resume(MigrationState
*s
)
2736 * Call all the resume_prepare() hooks, so that modules can be
2737 * ready for the migration resume.
2739 ret
= qemu_savevm_state_resume_prepare(s
);
2741 error_report("%s: resume_prepare() failure detected: %d",
2747 * Last handshake with destination on the resume (destination will
2748 * switch to postcopy-active afterwards)
2750 ret
= postcopy_resume_handshake(s
);
2752 error_report("%s: handshake failed: %d", __func__
, ret
);
2760 * We don't return until we are in a safe state to continue current
2761 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2762 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2764 static MigThrError
postcopy_pause(MigrationState
*s
)
2766 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2771 migrate_set_state(&s
->state
, s
->state
,
2772 MIGRATION_STATUS_POSTCOPY_PAUSED
);
2774 /* Current channel is possibly broken. Release it. */
2775 assert(s
->to_dst_file
);
2776 qemu_mutex_lock(&s
->qemu_file_lock
);
2777 file
= s
->to_dst_file
;
2778 s
->to_dst_file
= NULL
;
2779 qemu_mutex_unlock(&s
->qemu_file_lock
);
2781 qemu_file_shutdown(file
);
2784 error_report("Detected IO failure for postcopy. "
2785 "Migration paused.");
2788 * We wait until things fixed up. Then someone will setup the
2789 * status back for us.
2791 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2792 qemu_sem_wait(&s
->postcopy_pause_sem
);
2795 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2796 /* Woken up by a recover procedure. Give it a shot */
2799 * Firstly, let's wake up the return path now, with a new
2800 * return path channel.
2802 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
2804 /* Do the resume logic */
2805 if (postcopy_do_resume(s
) == 0) {
2806 /* Let's continue! */
2807 trace_postcopy_pause_continued();
2808 return MIG_THR_ERR_RECOVERED
;
2811 * Something wrong happened during the recovery, let's
2812 * pause again. Pause is always better than throwing
2818 /* This is not right... Time to quit. */
2819 return MIG_THR_ERR_FATAL
;
2824 static MigThrError
migration_detect_error(MigrationState
*s
)
2828 /* Try to detect any file errors */
2829 ret
= qemu_file_get_error(s
->to_dst_file
);
2832 /* Everything is fine */
2833 return MIG_THR_ERR_NONE
;
2836 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
== -EIO
) {
2838 * For postcopy, we allow the network to be down for a
2839 * while. After that, it can be continued by a
2842 return postcopy_pause(s
);
2845 * For precopy (or postcopy with error outside IO), we fail
2848 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
2849 trace_migration_thread_file_err();
2851 /* Time to stop the migration, now. */
2852 return MIG_THR_ERR_FATAL
;
2856 /* How many bytes have we transferred since the beggining of the migration */
2857 static uint64_t migration_total_bytes(MigrationState
*s
)
2859 return qemu_ftell(s
->to_dst_file
) + ram_counters
.multifd_bytes
;
2862 static void migration_calculate_complete(MigrationState
*s
)
2864 uint64_t bytes
= migration_total_bytes(s
);
2865 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2866 int64_t transfer_time
;
2868 s
->total_time
= end_time
- s
->start_time
;
2871 * It's still not set, so we are precopy migration. For
2872 * postcopy, downtime is calculated during postcopy_start().
2874 s
->downtime
= end_time
- s
->downtime_start
;
2877 transfer_time
= s
->total_time
- s
->setup_time
;
2878 if (transfer_time
) {
2879 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
2883 static void migration_update_counters(MigrationState
*s
,
2884 int64_t current_time
)
2886 uint64_t transferred
, time_spent
;
2887 uint64_t current_bytes
; /* bytes transferred since the beginning */
2890 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
2894 current_bytes
= migration_total_bytes(s
);
2895 transferred
= current_bytes
- s
->iteration_initial_bytes
;
2896 time_spent
= current_time
- s
->iteration_start_time
;
2897 bandwidth
= (double)transferred
/ time_spent
;
2898 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
2900 s
->mbps
= (((double) transferred
* 8.0) /
2901 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2904 * if we haven't sent anything, we don't want to
2905 * recalculate. 10000 is a small enough number for our purposes
2907 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
2908 s
->expected_downtime
= ram_counters
.remaining
/ bandwidth
;
2911 qemu_file_reset_rate_limit(s
->to_dst_file
);
2913 s
->iteration_start_time
= current_time
;
2914 s
->iteration_initial_bytes
= current_bytes
;
2916 trace_migrate_transferred(transferred
, time_spent
,
2917 bandwidth
, s
->threshold_size
);
2920 /* Migration thread iteration status */
2922 MIG_ITERATE_RESUME
, /* Resume current iteration */
2923 MIG_ITERATE_SKIP
, /* Skip current iteration */
2924 MIG_ITERATE_BREAK
, /* Break the loop */
2928 * Return true if continue to the next iteration directly, false
2931 static MigIterateState
migration_iteration_run(MigrationState
*s
)
2933 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
2934 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
2936 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
2937 &pend_compat
, &pend_post
);
2938 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
2940 trace_migrate_pending(pending_size
, s
->threshold_size
,
2941 pend_pre
, pend_compat
, pend_post
);
2943 if (pending_size
&& pending_size
>= s
->threshold_size
) {
2944 /* Still a significant amount to transfer */
2945 if (migrate_postcopy() && !in_postcopy
&&
2946 pend_pre
<= s
->threshold_size
&&
2947 atomic_read(&s
->start_postcopy
)) {
2948 if (postcopy_start(s
)) {
2949 error_report("%s: postcopy failed to start", __func__
);
2951 return MIG_ITERATE_SKIP
;
2953 /* Just another iteration step */
2954 qemu_savevm_state_iterate(s
->to_dst_file
,
2955 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2957 trace_migration_thread_low_pending(pending_size
);
2958 migration_completion(s
);
2959 return MIG_ITERATE_BREAK
;
2962 return MIG_ITERATE_RESUME
;
2965 static void migration_iteration_finish(MigrationState
*s
)
2967 /* If we enabled cpu throttling for auto-converge, turn it off. */
2968 cpu_throttle_stop();
2970 qemu_mutex_lock_iothread();
2972 case MIGRATION_STATUS_COMPLETED
:
2973 migration_calculate_complete(s
);
2974 runstate_set(RUN_STATE_POSTMIGRATE
);
2977 case MIGRATION_STATUS_ACTIVE
:
2979 * We should really assert here, but since it's during
2980 * migration, let's try to reduce the usage of assertions.
2982 if (!migrate_colo_enabled()) {
2983 error_report("%s: critical error: calling COLO code without "
2984 "COLO enabled", __func__
);
2986 migrate_start_colo_process(s
);
2988 * Fixme: we will run VM in COLO no matter its old running state.
2989 * After exited COLO, we will keep running.
2991 s
->vm_was_running
= true;
2993 case MIGRATION_STATUS_FAILED
:
2994 case MIGRATION_STATUS_CANCELLED
:
2995 case MIGRATION_STATUS_CANCELLING
:
2996 if (s
->vm_was_running
) {
2999 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
3000 runstate_set(RUN_STATE_POSTMIGRATE
);
3006 /* Should not reach here, but if so, forgive the VM. */
3007 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3010 qemu_bh_schedule(s
->cleanup_bh
);
3011 qemu_mutex_unlock_iothread();
3014 void migration_make_urgent_request(void)
3016 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
3019 void migration_consume_urgent_request(void)
3021 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
3025 * Master migration thread on the source VM.
3026 * It drives the migration and pumps the data down the outgoing channel.
3028 static void *migration_thread(void *opaque
)
3030 MigrationState
*s
= opaque
;
3031 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3032 MigThrError thr_error
;
3033 bool urgent
= false;
3035 rcu_register_thread();
3037 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3039 qemu_savevm_state_header(s
->to_dst_file
);
3042 * If we opened the return path, we need to make sure dst has it
3045 if (s
->rp_state
.from_dst_file
) {
3046 /* Now tell the dest that it should open its end so it can reply */
3047 qemu_savevm_send_open_return_path(s
->to_dst_file
);
3049 /* And do a ping that will make stuff easier to debug */
3050 qemu_savevm_send_ping(s
->to_dst_file
, 1);
3053 if (migrate_postcopy()) {
3055 * Tell the destination that we *might* want to do postcopy later;
3056 * if the other end can't do postcopy it should fail now, nice and
3059 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
3062 if (migrate_colo_enabled()) {
3063 /* Notify migration destination that we enable COLO */
3064 qemu_savevm_send_colo_enable(s
->to_dst_file
);
3067 qemu_savevm_state_setup(s
->to_dst_file
);
3069 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3070 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3071 MIGRATION_STATUS_ACTIVE
);
3073 trace_migration_thread_setup_complete();
3075 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
3076 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3077 int64_t current_time
;
3079 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
3080 MigIterateState iter_state
= migration_iteration_run(s
);
3081 if (iter_state
== MIG_ITERATE_SKIP
) {
3083 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3089 * Try to detect any kind of failures, and see whether we
3090 * should stop the migration now.
3092 thr_error
= migration_detect_error(s
);
3093 if (thr_error
== MIG_THR_ERR_FATAL
) {
3094 /* Stop migration */
3096 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
3098 * Just recovered from a e.g. network failure, reset all
3099 * the local variables. This is important to avoid
3100 * breaking transferred_bytes and bandwidth calculation
3102 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3103 s
->iteration_initial_bytes
= 0;
3106 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3108 migration_update_counters(s
, current_time
);
3111 if (qemu_file_rate_limit(s
->to_dst_file
)) {
3112 /* Wait for a delay to do rate limiting OR
3113 * something urgent to post the semaphore.
3115 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- current_time
;
3116 trace_migration_thread_ratelimit_pre(ms
);
3117 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
3118 /* We were worken by one or more urgent things but
3119 * the timedwait will have consumed one of them.
3120 * The service routine for the urgent wake will dec
3121 * the semaphore itself for each item it consumes,
3122 * so add this one we just eat back.
3124 qemu_sem_post(&s
->rate_limit_sem
);
3127 trace_migration_thread_ratelimit_post(urgent
);
3131 trace_migration_thread_after_loop();
3132 migration_iteration_finish(s
);
3133 rcu_unregister_thread();
3137 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
3140 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
3142 s
->expected_downtime
= s
->parameters
.downtime_limit
;
3143 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
3145 migrate_fd_error(s
, error_in
);
3146 migrate_fd_cleanup(s
);
3151 /* This is a resumed migration */
3152 rate_limit
= INT64_MAX
;
3154 /* This is a fresh new migration */
3155 rate_limit
= s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
;
3157 /* Notify before starting migration thread */
3158 notifier_list_notify(&migration_state_notifiers
, s
);
3161 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
3162 qemu_file_set_blocking(s
->to_dst_file
, true);
3165 * Open the return path. For postcopy, it is used exclusively. For
3166 * precopy, only if user specified "return-path" capability would
3167 * QEMU uses the return path.
3169 if (migrate_postcopy_ram() || migrate_use_return_path()) {
3170 if (open_return_path_on_source(s
, !resume
)) {
3171 error_report("Unable to open return-path for postcopy");
3172 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
3173 migrate_fd_cleanup(s
);
3179 /* Wakeup the main migration thread to do the recovery */
3180 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
3181 MIGRATION_STATUS_POSTCOPY_RECOVER
);
3182 qemu_sem_post(&s
->postcopy_pause_sem
);
3186 if (multifd_save_setup() != 0) {
3187 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3188 MIGRATION_STATUS_FAILED
);
3189 migrate_fd_cleanup(s
);
3192 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
3193 QEMU_THREAD_JOINABLE
);
3194 s
->migration_thread_running
= true;
3197 void migration_global_dump(Monitor
*mon
)
3199 MigrationState
*ms
= migrate_get_current();
3201 monitor_printf(mon
, "globals:\n");
3202 monitor_printf(mon
, "store-global-state: %s\n",
3203 ms
->store_global_state
? "on" : "off");
3204 monitor_printf(mon
, "only-migratable: %s\n",
3205 ms
->only_migratable
? "on" : "off");
3206 monitor_printf(mon
, "send-configuration: %s\n",
3207 ms
->send_configuration
? "on" : "off");
3208 monitor_printf(mon
, "send-section-footer: %s\n",
3209 ms
->send_section_footer
? "on" : "off");
3210 monitor_printf(mon
, "decompress-error-check: %s\n",
3211 ms
->decompress_error_check
? "on" : "off");
3214 #define DEFINE_PROP_MIG_CAP(name, x) \
3215 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3217 static Property migration_properties
[] = {
3218 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
3219 store_global_state
, true),
3220 DEFINE_PROP_BOOL("only-migratable", MigrationState
, only_migratable
, false),
3221 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
3222 send_configuration
, true),
3223 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
3224 send_section_footer
, true),
3225 DEFINE_PROP_BOOL("decompress-error-check", MigrationState
,
3226 decompress_error_check
, true),
3228 /* Migration parameters */
3229 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
3230 parameters
.compress_level
,
3231 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
3232 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
3233 parameters
.compress_threads
,
3234 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
3235 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState
,
3236 parameters
.compress_wait_thread
, true),
3237 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
3238 parameters
.decompress_threads
,
3239 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
3240 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
3241 parameters
.cpu_throttle_initial
,
3242 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
3243 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
3244 parameters
.cpu_throttle_increment
,
3245 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
3246 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
3247 parameters
.max_bandwidth
, MAX_THROTTLE
),
3248 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
3249 parameters
.downtime_limit
,
3250 DEFAULT_MIGRATE_SET_DOWNTIME
),
3251 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
3252 parameters
.x_checkpoint_delay
,
3253 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
3254 DEFINE_PROP_UINT8("x-multifd-channels", MigrationState
,
3255 parameters
.x_multifd_channels
,
3256 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
3257 DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState
,
3258 parameters
.x_multifd_page_count
,
3259 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT
),
3260 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
3261 parameters
.xbzrle_cache_size
,
3262 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
3263 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState
,
3264 parameters
.max_postcopy_bandwidth
,
3265 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH
),
3266 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState
,
3267 parameters
.max_cpu_throttle
,
3268 DEFAULT_MIGRATE_MAX_CPU_THROTTLE
),
3270 /* Migration capabilities */
3271 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
3272 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
3273 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
3274 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
3275 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
3276 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
3277 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
3278 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
3279 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
3280 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
3281 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
3282 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD
),
3284 DEFINE_PROP_END_OF_LIST(),
3287 static void migration_class_init(ObjectClass
*klass
, void *data
)
3289 DeviceClass
*dc
= DEVICE_CLASS(klass
);
3291 dc
->user_creatable
= false;
3292 dc
->props
= migration_properties
;
3295 static void migration_instance_finalize(Object
*obj
)
3297 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3298 MigrationParameters
*params
= &ms
->parameters
;
3300 qemu_mutex_destroy(&ms
->error_mutex
);
3301 qemu_mutex_destroy(&ms
->qemu_file_lock
);
3302 g_free(params
->tls_hostname
);
3303 g_free(params
->tls_creds
);
3304 qemu_sem_destroy(&ms
->rate_limit_sem
);
3305 qemu_sem_destroy(&ms
->pause_sem
);
3306 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
3307 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
3308 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
3309 error_free(ms
->error
);
3312 static void migration_instance_init(Object
*obj
)
3314 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3315 MigrationParameters
*params
= &ms
->parameters
;
3317 ms
->state
= MIGRATION_STATUS_NONE
;
3319 qemu_sem_init(&ms
->pause_sem
, 0);
3320 qemu_mutex_init(&ms
->error_mutex
);
3322 params
->tls_hostname
= g_strdup("");
3323 params
->tls_creds
= g_strdup("");
3325 /* Set has_* up only for parameter checks */
3326 params
->has_compress_level
= true;
3327 params
->has_compress_threads
= true;
3328 params
->has_decompress_threads
= true;
3329 params
->has_cpu_throttle_initial
= true;
3330 params
->has_cpu_throttle_increment
= true;
3331 params
->has_max_bandwidth
= true;
3332 params
->has_downtime_limit
= true;
3333 params
->has_x_checkpoint_delay
= true;
3334 params
->has_block_incremental
= true;
3335 params
->has_x_multifd_channels
= true;
3336 params
->has_x_multifd_page_count
= true;
3337 params
->has_xbzrle_cache_size
= true;
3338 params
->has_max_postcopy_bandwidth
= true;
3339 params
->has_max_cpu_throttle
= true;
3341 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
3342 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
3343 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
3344 qemu_sem_init(&ms
->rate_limit_sem
, 0);
3345 qemu_mutex_init(&ms
->qemu_file_lock
);
3349 * Return true if check pass, false otherwise. Error will be put
3350 * inside errp if provided.
3352 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
3354 MigrationCapabilityStatusList
*head
= NULL
;
3355 /* Assuming all off */
3356 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
3359 if (!migrate_params_check(&ms
->parameters
, errp
)) {
3363 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
3364 if (ms
->enabled_capabilities
[i
]) {
3365 head
= migrate_cap_add(head
, i
, true);
3369 ret
= migrate_caps_check(cap_list
, head
, errp
);
3371 /* It works with head == NULL */
3372 qapi_free_MigrationCapabilityStatusList(head
);
3377 static const TypeInfo migration_type
= {
3378 .name
= TYPE_MIGRATION
,
3380 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3381 * not created using qdev_create(), it is not attached to the qdev
3382 * device tree, and it is never realized.
3384 * TODO: Make this TYPE_OBJECT once QOM provides something like
3385 * TYPE_DEVICE's "-global" properties.
3387 .parent
= TYPE_DEVICE
,
3388 .class_init
= migration_class_init
,
3389 .class_size
= sizeof(MigrationClass
),
3390 .instance_size
= sizeof(MigrationState
),
3391 .instance_init
= migration_instance_init
,
3392 .instance_finalize
= migration_instance_finalize
,
3395 static void register_migration_types(void)
3397 type_register_static(&migration_type
);
3400 type_init(register_migration_types
);