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 * Note: Please change this default value to 10000 when we support hybrid mode.
82 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY 200
83 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
84 #define DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT 16
86 /* Background transfer rate for postcopy, 0 means unlimited, note
87 * that page requests can still exceed this limit.
89 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
91 static NotifierList migration_state_notifiers
=
92 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
94 static bool deferred_incoming
;
96 /* Messages sent on the return path from destination to source */
97 enum mig_rp_message_type
{
98 MIG_RP_MSG_INVALID
= 0, /* Must be 0 */
99 MIG_RP_MSG_SHUT
, /* sibling will not send any more RP messages */
100 MIG_RP_MSG_PONG
, /* Response to a PING; data (seq: be32 ) */
102 MIG_RP_MSG_REQ_PAGES_ID
, /* data (start: be64, len: be32, id: string) */
103 MIG_RP_MSG_REQ_PAGES
, /* data (start: be64, len: be32) */
104 MIG_RP_MSG_RECV_BITMAP
, /* send recved_bitmap back to source */
105 MIG_RP_MSG_RESUME_ACK
, /* tell source that we are ready to resume */
110 /* When we add fault tolerance, we could have several
111 migrations at once. For now we don't need to add
112 dynamic creation of migration */
114 static MigrationState
*current_migration
;
115 static MigrationIncomingState
*current_incoming
;
117 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
118 static int migration_maybe_pause(MigrationState
*s
,
119 int *current_active_state
,
122 void migration_object_init(void)
124 MachineState
*ms
= MACHINE(qdev_get_machine());
127 /* This can only be called once. */
128 assert(!current_migration
);
129 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
132 * Init the migrate incoming object as well no matter whether
133 * we'll use it or not.
135 assert(!current_incoming
);
136 current_incoming
= g_new0(MigrationIncomingState
, 1);
137 current_incoming
->state
= MIGRATION_STATUS_NONE
;
138 current_incoming
->postcopy_remote_fds
=
139 g_array_new(FALSE
, TRUE
, sizeof(struct PostCopyFD
));
140 qemu_mutex_init(¤t_incoming
->rp_mutex
);
141 qemu_event_init(¤t_incoming
->main_thread_load_event
, false);
142 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_dst
, 0);
143 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fault
, 0);
145 init_dirty_bitmap_incoming_migration();
147 if (!migration_object_check(current_migration
, &err
)) {
148 error_report_err(err
);
153 * We cannot really do this in migration_instance_init() since at
154 * that time global properties are not yet applied, then this
155 * value will be definitely replaced by something else.
157 if (ms
->enforce_config_section
) {
158 current_migration
->send_configuration
= true;
162 void migration_object_finalize(void)
164 object_unref(OBJECT(current_migration
));
168 MigrationState
*migrate_get_current(void)
170 /* This can only be called after the object created. */
171 assert(current_migration
);
172 return current_migration
;
175 MigrationIncomingState
*migration_incoming_get_current(void)
177 assert(current_incoming
);
178 return current_incoming
;
181 void migration_incoming_state_destroy(void)
183 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
185 if (mis
->to_src_file
) {
186 /* Tell source that we are done */
187 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
188 qemu_fclose(mis
->to_src_file
);
189 mis
->to_src_file
= NULL
;
192 if (mis
->from_src_file
) {
193 qemu_fclose(mis
->from_src_file
);
194 mis
->from_src_file
= NULL
;
196 if (mis
->postcopy_remote_fds
) {
197 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
198 mis
->postcopy_remote_fds
= NULL
;
201 qemu_event_reset(&mis
->main_thread_load_event
);
204 static void migrate_generate_event(int new_state
)
206 if (migrate_use_events()) {
207 qapi_event_send_migration(new_state
, &error_abort
);
211 static bool migrate_late_block_activate(void)
215 s
= migrate_get_current();
217 return s
->enabled_capabilities
[
218 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE
];
222 * Called on -incoming with a defer: uri.
223 * The migration can be started later after any parameters have been
226 static void deferred_incoming_migration(Error
**errp
)
228 if (deferred_incoming
) {
229 error_setg(errp
, "Incoming migration already deferred");
231 deferred_incoming
= true;
235 * Send a message on the return channel back to the source
238 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
239 enum mig_rp_message_type message_type
,
240 uint16_t len
, void *data
)
244 trace_migrate_send_rp_message((int)message_type
, len
);
245 qemu_mutex_lock(&mis
->rp_mutex
);
248 * It's possible that the file handle got lost due to network
251 if (!mis
->to_src_file
) {
256 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
257 qemu_put_be16(mis
->to_src_file
, len
);
258 qemu_put_buffer(mis
->to_src_file
, data
, len
);
259 qemu_fflush(mis
->to_src_file
);
261 /* It's possible that qemu file got error during sending */
262 ret
= qemu_file_get_error(mis
->to_src_file
);
265 qemu_mutex_unlock(&mis
->rp_mutex
);
269 /* Request a range of pages from the source VM at the given
271 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
272 * as the last request (a name must have been given previously)
273 * Start: Address offset within the RB
274 * Len: Length in bytes required - must be a multiple of pagesize
276 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
, const char *rbname
,
277 ram_addr_t start
, size_t len
)
279 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
280 size_t msglen
= 12; /* start + len */
281 enum mig_rp_message_type msg_type
;
283 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
284 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
287 int rbname_len
= strlen(rbname
);
288 assert(rbname_len
< 256);
290 bufc
[msglen
++] = rbname_len
;
291 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
292 msglen
+= rbname_len
;
293 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
295 msg_type
= MIG_RP_MSG_REQ_PAGES
;
298 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
301 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
305 qapi_event_send_migration(MIGRATION_STATUS_SETUP
, &error_abort
);
306 if (!strcmp(uri
, "defer")) {
307 deferred_incoming_migration(errp
);
308 } else if (strstart(uri
, "tcp:", &p
)) {
309 tcp_start_incoming_migration(p
, errp
);
311 } else if (strstart(uri
, "rdma:", &p
)) {
312 rdma_start_incoming_migration(p
, errp
);
314 } else if (strstart(uri
, "exec:", &p
)) {
315 exec_start_incoming_migration(p
, errp
);
316 } else if (strstart(uri
, "unix:", &p
)) {
317 unix_start_incoming_migration(p
, errp
);
318 } else if (strstart(uri
, "fd:", &p
)) {
319 fd_start_incoming_migration(p
, errp
);
321 error_setg(errp
, "unknown migration protocol: %s", uri
);
325 static void process_incoming_migration_bh(void *opaque
)
327 Error
*local_err
= NULL
;
328 MigrationIncomingState
*mis
= opaque
;
330 /* If capability late_block_activate is set:
331 * Only fire up the block code now if we're going to restart the
332 * VM, else 'cont' will do it.
333 * This causes file locking to happen; so we don't want it to happen
334 * unless we really are starting the VM.
336 if (!migrate_late_block_activate() ||
337 (autostart
&& (!global_state_received() ||
338 global_state_get_runstate() == RUN_STATE_RUNNING
))) {
339 /* Make sure all file formats flush their mutable metadata.
340 * If we get an error here, just don't restart the VM yet. */
341 bdrv_invalidate_cache_all(&local_err
);
343 error_report_err(local_err
);
350 * This must happen after all error conditions are dealt with and
351 * we're sure the VM is going to be running on this host.
353 qemu_announce_self();
355 if (multifd_load_cleanup(&local_err
) != 0) {
356 error_report_err(local_err
);
359 /* If global state section was not received or we are in running
360 state, we need to obey autostart. Any other state is set with
363 dirty_bitmap_mig_before_vm_start();
365 if (!global_state_received() ||
366 global_state_get_runstate() == RUN_STATE_RUNNING
) {
370 runstate_set(RUN_STATE_PAUSED
);
373 runstate_set(global_state_get_runstate());
376 * This must happen after any state changes since as soon as an external
377 * observer sees this event they might start to prod at the VM assuming
380 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
381 MIGRATION_STATUS_COMPLETED
);
382 qemu_bh_delete(mis
->bh
);
383 migration_incoming_state_destroy();
386 static void process_incoming_migration_co(void *opaque
)
388 MigrationIncomingState
*mis
= migration_incoming_get_current();
392 assert(mis
->from_src_file
);
393 mis
->largest_page_size
= qemu_ram_pagesize_largest();
394 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
395 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
396 MIGRATION_STATUS_ACTIVE
);
397 ret
= qemu_loadvm_state(mis
->from_src_file
);
399 ps
= postcopy_state_get();
400 trace_process_incoming_migration_co_end(ret
, ps
);
401 if (ps
!= POSTCOPY_INCOMING_NONE
) {
402 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
404 * Where a migration had postcopy enabled (and thus went to advise)
405 * but managed to complete within the precopy period, we can use
408 postcopy_ram_incoming_cleanup(mis
);
409 } else if (ret
>= 0) {
411 * Postcopy was started, cleanup should happen at the end of the
414 trace_process_incoming_migration_co_postcopy_end_main();
417 /* Else if something went wrong then just fall out of the normal exit */
420 /* we get COLO info, and know if we are in COLO mode */
421 if (!ret
&& migration_incoming_enable_colo()) {
422 mis
->migration_incoming_co
= qemu_coroutine_self();
423 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
424 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
425 mis
->have_colo_incoming_thread
= true;
426 qemu_coroutine_yield();
428 /* Wait checkpoint incoming thread exit before free resource */
429 qemu_thread_join(&mis
->colo_incoming_thread
);
433 Error
*local_err
= NULL
;
435 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
436 MIGRATION_STATUS_FAILED
);
437 error_report("load of migration failed: %s", strerror(-ret
));
438 qemu_fclose(mis
->from_src_file
);
439 if (multifd_load_cleanup(&local_err
) != 0) {
440 error_report_err(local_err
);
444 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
445 qemu_bh_schedule(mis
->bh
);
448 static void migration_incoming_setup(QEMUFile
*f
)
450 MigrationIncomingState
*mis
= migration_incoming_get_current();
452 if (multifd_load_setup() != 0) {
453 /* We haven't been able to create multifd threads
454 nothing better to do */
458 if (!mis
->from_src_file
) {
459 mis
->from_src_file
= f
;
461 qemu_file_set_blocking(f
, false);
464 void migration_incoming_process(void)
466 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
467 qemu_coroutine_enter(co
);
470 /* Returns true if recovered from a paused migration, otherwise false */
471 static bool postcopy_try_recover(QEMUFile
*f
)
473 MigrationIncomingState
*mis
= migration_incoming_get_current();
475 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
476 /* Resumed from a paused postcopy migration */
478 mis
->from_src_file
= f
;
479 /* Postcopy has standalone thread to do vm load */
480 qemu_file_set_blocking(f
, true);
482 /* Re-configure the return path */
483 mis
->to_src_file
= qemu_file_get_return_path(f
);
485 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
486 MIGRATION_STATUS_POSTCOPY_RECOVER
);
489 * Here, we only wake up the main loading thread (while the
490 * fault thread will still be waiting), so that we can receive
491 * commands from source now, and answer it if needed. The
492 * fault thread will be woken up afterwards until we are sure
493 * that source is ready to reply to page requests.
495 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
502 void migration_fd_process_incoming(QEMUFile
*f
)
504 if (postcopy_try_recover(f
)) {
508 migration_incoming_setup(f
);
509 migration_incoming_process();
512 void migration_ioc_process_incoming(QIOChannel
*ioc
)
514 MigrationIncomingState
*mis
= migration_incoming_get_current();
515 bool start_migration
;
517 if (!mis
->from_src_file
) {
518 /* The first connection (multifd may have multiple) */
519 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
521 /* If it's a recovery, we're done */
522 if (postcopy_try_recover(f
)) {
526 migration_incoming_setup(f
);
529 * Common migration only needs one channel, so we can start
530 * right now. Multifd needs more than one channel, we wait.
532 start_migration
= !migrate_use_multifd();
534 /* Multiple connections */
535 assert(migrate_use_multifd());
536 start_migration
= multifd_recv_new_channel(ioc
);
539 if (start_migration
) {
540 migration_incoming_process();
545 * @migration_has_all_channels: We have received all channels that we need
547 * Returns true when we have got connections to all the channels that
548 * we need for migration.
550 bool migration_has_all_channels(void)
552 MigrationIncomingState
*mis
= migration_incoming_get_current();
555 all_channels
= multifd_recv_all_channels_created();
557 return all_channels
&& mis
->from_src_file
!= NULL
;
561 * Send a 'SHUT' message on the return channel with the given value
562 * to indicate that we've finished with the RP. Non-0 value indicates
565 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
570 buf
= cpu_to_be32(value
);
571 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
575 * Send a 'PONG' message on the return channel with the given value
576 * (normally in response to a 'PING')
578 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
583 buf
= cpu_to_be32(value
);
584 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
587 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
595 * First, we send the header part. It contains only the len of
596 * idstr, and the idstr itself.
598 len
= strlen(block_name
);
600 memcpy(buf
+ 1, block_name
, len
);
602 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
603 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
608 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
611 * Next, we dump the received bitmap to the stream.
613 * TODO: currently we are safe since we are the only one that is
614 * using the to_src_file handle (fault thread is still paused),
615 * and it's ok even not taking the mutex. However the best way is
616 * to take the lock before sending the message header, and release
617 * the lock after sending the bitmap.
619 qemu_mutex_lock(&mis
->rp_mutex
);
620 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
621 qemu_mutex_unlock(&mis
->rp_mutex
);
623 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
626 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
630 buf
= cpu_to_be32(value
);
631 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
634 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
636 MigrationCapabilityStatusList
*head
= NULL
;
637 MigrationCapabilityStatusList
*caps
;
638 MigrationState
*s
= migrate_get_current();
641 caps
= NULL
; /* silence compiler warning */
642 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
643 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
644 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
649 head
= g_malloc0(sizeof(*caps
));
652 caps
->next
= g_malloc0(sizeof(*caps
));
656 g_malloc(sizeof(*caps
->value
));
657 caps
->value
->capability
= i
;
658 caps
->value
->state
= s
->enabled_capabilities
[i
];
664 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
666 MigrationParameters
*params
;
667 MigrationState
*s
= migrate_get_current();
669 /* TODO use QAPI_CLONE() instead of duplicating it inline */
670 params
= g_malloc0(sizeof(*params
));
671 params
->has_compress_level
= true;
672 params
->compress_level
= s
->parameters
.compress_level
;
673 params
->has_compress_threads
= true;
674 params
->compress_threads
= s
->parameters
.compress_threads
;
675 params
->has_decompress_threads
= true;
676 params
->decompress_threads
= s
->parameters
.decompress_threads
;
677 params
->has_cpu_throttle_initial
= true;
678 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
679 params
->has_cpu_throttle_increment
= true;
680 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
681 params
->has_tls_creds
= true;
682 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
683 params
->has_tls_hostname
= true;
684 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
685 params
->has_max_bandwidth
= true;
686 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
687 params
->has_downtime_limit
= true;
688 params
->downtime_limit
= s
->parameters
.downtime_limit
;
689 params
->has_x_checkpoint_delay
= true;
690 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
691 params
->has_block_incremental
= true;
692 params
->block_incremental
= s
->parameters
.block_incremental
;
693 params
->has_x_multifd_channels
= true;
694 params
->x_multifd_channels
= s
->parameters
.x_multifd_channels
;
695 params
->has_x_multifd_page_count
= true;
696 params
->x_multifd_page_count
= s
->parameters
.x_multifd_page_count
;
697 params
->has_xbzrle_cache_size
= true;
698 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
699 params
->has_max_postcopy_bandwidth
= true;
700 params
->max_postcopy_bandwidth
= s
->parameters
.max_postcopy_bandwidth
;
701 params
->has_max_cpu_throttle
= true;
702 params
->max_cpu_throttle
= s
->parameters
.max_cpu_throttle
;
708 * Return true if we're already in the middle of a migration
709 * (i.e. any of the active or setup states)
711 static bool migration_is_setup_or_active(int state
)
714 case MIGRATION_STATUS_ACTIVE
:
715 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
716 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
717 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
718 case MIGRATION_STATUS_SETUP
:
719 case MIGRATION_STATUS_PRE_SWITCHOVER
:
720 case MIGRATION_STATUS_DEVICE
:
729 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
731 info
->has_ram
= true;
732 info
->ram
= g_malloc0(sizeof(*info
->ram
));
733 info
->ram
->transferred
= ram_counters
.transferred
;
734 info
->ram
->total
= ram_bytes_total();
735 info
->ram
->duplicate
= ram_counters
.duplicate
;
736 /* legacy value. It is not used anymore */
737 info
->ram
->skipped
= 0;
738 info
->ram
->normal
= ram_counters
.normal
;
739 info
->ram
->normal_bytes
= ram_counters
.normal
*
740 qemu_target_page_size();
741 info
->ram
->mbps
= s
->mbps
;
742 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
743 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
744 info
->ram
->page_size
= qemu_target_page_size();
745 info
->ram
->multifd_bytes
= ram_counters
.multifd_bytes
;
747 if (migrate_use_xbzrle()) {
748 info
->has_xbzrle_cache
= true;
749 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
750 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
751 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
752 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
753 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
754 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
755 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
758 if (cpu_throttle_active()) {
759 info
->has_cpu_throttle_percentage
= true;
760 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
763 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
764 info
->ram
->remaining
= ram_bytes_remaining();
765 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
769 static void populate_disk_info(MigrationInfo
*info
)
771 if (blk_mig_active()) {
772 info
->has_disk
= true;
773 info
->disk
= g_malloc0(sizeof(*info
->disk
));
774 info
->disk
->transferred
= blk_mig_bytes_transferred();
775 info
->disk
->remaining
= blk_mig_bytes_remaining();
776 info
->disk
->total
= blk_mig_bytes_total();
780 static void fill_source_migration_info(MigrationInfo
*info
)
782 MigrationState
*s
= migrate_get_current();
785 case MIGRATION_STATUS_NONE
:
786 /* no migration has happened ever */
787 /* do not overwrite destination migration status */
790 case MIGRATION_STATUS_SETUP
:
791 info
->has_status
= true;
792 info
->has_total_time
= false;
794 case MIGRATION_STATUS_ACTIVE
:
795 case MIGRATION_STATUS_CANCELLING
:
796 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
797 case MIGRATION_STATUS_PRE_SWITCHOVER
:
798 case MIGRATION_STATUS_DEVICE
:
799 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
800 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
801 /* TODO add some postcopy stats */
802 info
->has_status
= true;
803 info
->has_total_time
= true;
804 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
806 info
->has_expected_downtime
= true;
807 info
->expected_downtime
= s
->expected_downtime
;
808 info
->has_setup_time
= true;
809 info
->setup_time
= s
->setup_time
;
811 populate_ram_info(info
, s
);
812 populate_disk_info(info
);
814 case MIGRATION_STATUS_COLO
:
815 info
->has_status
= true;
816 /* TODO: display COLO specific information (checkpoint info etc.) */
818 case MIGRATION_STATUS_COMPLETED
:
819 info
->has_status
= true;
820 info
->has_total_time
= true;
821 info
->total_time
= s
->total_time
;
822 info
->has_downtime
= true;
823 info
->downtime
= s
->downtime
;
824 info
->has_setup_time
= true;
825 info
->setup_time
= s
->setup_time
;
827 populate_ram_info(info
, s
);
829 case MIGRATION_STATUS_FAILED
:
830 info
->has_status
= true;
832 info
->has_error_desc
= true;
833 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
836 case MIGRATION_STATUS_CANCELLED
:
837 info
->has_status
= true;
840 info
->status
= s
->state
;
844 * @migration_caps_check - check capability validity
846 * @cap_list: old capability list, array of bool
847 * @params: new capabilities to be applied soon
848 * @errp: set *errp if the check failed, with reason
850 * Returns true if check passed, otherwise false.
852 static bool migrate_caps_check(bool *cap_list
,
853 MigrationCapabilityStatusList
*params
,
856 MigrationCapabilityStatusList
*cap
;
857 bool old_postcopy_cap
;
858 MigrationIncomingState
*mis
= migration_incoming_get_current();
860 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
862 for (cap
= params
; cap
; cap
= cap
->next
) {
863 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
866 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
867 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
868 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
870 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
875 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
876 if (cap_list
[MIGRATION_CAPABILITY_COMPRESS
]) {
877 /* The decompression threads asynchronously write into RAM
878 * rather than use the atomic copies needed to avoid
879 * userfaulting. It should be possible to fix the decompression
880 * threads for compatibility in future.
882 error_setg(errp
, "Postcopy is not currently compatible "
887 /* This check is reasonably expensive, so only when it's being
888 * set the first time, also it's only the destination that needs
891 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
892 !postcopy_ram_supported_by_host(mis
)) {
893 /* postcopy_ram_supported_by_host will have emitted a more
896 error_setg(errp
, "Postcopy is not supported");
904 static void fill_destination_migration_info(MigrationInfo
*info
)
906 MigrationIncomingState
*mis
= migration_incoming_get_current();
908 switch (mis
->state
) {
909 case MIGRATION_STATUS_NONE
:
912 case MIGRATION_STATUS_SETUP
:
913 case MIGRATION_STATUS_CANCELLING
:
914 case MIGRATION_STATUS_CANCELLED
:
915 case MIGRATION_STATUS_ACTIVE
:
916 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
917 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
918 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
919 case MIGRATION_STATUS_FAILED
:
920 case MIGRATION_STATUS_COLO
:
921 info
->has_status
= true;
923 case MIGRATION_STATUS_COMPLETED
:
924 info
->has_status
= true;
925 fill_destination_postcopy_migration_info(info
);
928 info
->status
= mis
->state
;
931 MigrationInfo
*qmp_query_migrate(Error
**errp
)
933 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
935 fill_destination_migration_info(info
);
936 fill_source_migration_info(info
);
941 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
944 MigrationState
*s
= migrate_get_current();
945 MigrationCapabilityStatusList
*cap
;
946 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
948 if (migration_is_setup_or_active(s
->state
)) {
949 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
953 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
954 if (!migrate_caps_check(cap_list
, params
, errp
)) {
958 for (cap
= params
; cap
; cap
= cap
->next
) {
959 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
964 * Check whether the parameters are valid. Error will be put into errp
965 * (if provided). Return true if valid, otherwise false.
967 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
969 if (params
->has_compress_level
&&
970 (params
->compress_level
> 9)) {
971 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
972 "is invalid, it should be in the range of 0 to 9");
976 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
977 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
979 "is invalid, it should be in the range of 1 to 255");
983 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
984 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
985 "decompress_threads",
986 "is invalid, it should be in the range of 1 to 255");
990 if (params
->has_cpu_throttle_initial
&&
991 (params
->cpu_throttle_initial
< 1 ||
992 params
->cpu_throttle_initial
> 99)) {
993 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
994 "cpu_throttle_initial",
995 "an integer in the range of 1 to 99");
999 if (params
->has_cpu_throttle_increment
&&
1000 (params
->cpu_throttle_increment
< 1 ||
1001 params
->cpu_throttle_increment
> 99)) {
1002 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1003 "cpu_throttle_increment",
1004 "an integer in the range of 1 to 99");
1008 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
1009 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
1010 " range of 0 to %zu bytes/second", SIZE_MAX
);
1014 if (params
->has_downtime_limit
&&
1015 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
1016 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1017 "the range of 0 to %d milliseconds",
1018 MAX_MIGRATE_DOWNTIME
);
1022 /* x_checkpoint_delay is now always positive */
1024 if (params
->has_x_multifd_channels
&& (params
->x_multifd_channels
< 1)) {
1025 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1027 "is invalid, it should be in the range of 1 to 255");
1030 if (params
->has_x_multifd_page_count
&&
1031 (params
->x_multifd_page_count
< 1 ||
1032 params
->x_multifd_page_count
> 10000)) {
1033 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1034 "multifd_page_count",
1035 "is invalid, it should be in the range of 1 to 10000");
1039 if (params
->has_xbzrle_cache_size
&&
1040 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
1041 !is_power_of_2(params
->xbzrle_cache_size
))) {
1042 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1043 "xbzrle_cache_size",
1044 "is invalid, it should be bigger than target page size"
1045 " and a power of two");
1049 if (params
->has_max_cpu_throttle
&&
1050 (params
->max_cpu_throttle
< params
->cpu_throttle_initial
||
1051 params
->max_cpu_throttle
> 99)) {
1052 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1054 "an integer in the range of cpu_throttle_initial to 99");
1061 static void migrate_params_test_apply(MigrateSetParameters
*params
,
1062 MigrationParameters
*dest
)
1064 *dest
= migrate_get_current()->parameters
;
1066 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1068 if (params
->has_compress_level
) {
1069 dest
->compress_level
= params
->compress_level
;
1072 if (params
->has_compress_threads
) {
1073 dest
->compress_threads
= params
->compress_threads
;
1076 if (params
->has_decompress_threads
) {
1077 dest
->decompress_threads
= params
->decompress_threads
;
1080 if (params
->has_cpu_throttle_initial
) {
1081 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
1084 if (params
->has_cpu_throttle_increment
) {
1085 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
1088 if (params
->has_tls_creds
) {
1089 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1090 dest
->tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1093 if (params
->has_tls_hostname
) {
1094 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1095 dest
->tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1098 if (params
->has_max_bandwidth
) {
1099 dest
->max_bandwidth
= params
->max_bandwidth
;
1102 if (params
->has_downtime_limit
) {
1103 dest
->downtime_limit
= params
->downtime_limit
;
1106 if (params
->has_x_checkpoint_delay
) {
1107 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
1110 if (params
->has_block_incremental
) {
1111 dest
->block_incremental
= params
->block_incremental
;
1113 if (params
->has_x_multifd_channels
) {
1114 dest
->x_multifd_channels
= params
->x_multifd_channels
;
1116 if (params
->has_x_multifd_page_count
) {
1117 dest
->x_multifd_page_count
= params
->x_multifd_page_count
;
1119 if (params
->has_xbzrle_cache_size
) {
1120 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
1122 if (params
->has_max_postcopy_bandwidth
) {
1123 dest
->max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1125 if (params
->has_max_cpu_throttle
) {
1126 dest
->max_cpu_throttle
= params
->max_cpu_throttle
;
1130 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
1132 MigrationState
*s
= migrate_get_current();
1134 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1136 if (params
->has_compress_level
) {
1137 s
->parameters
.compress_level
= params
->compress_level
;
1140 if (params
->has_compress_threads
) {
1141 s
->parameters
.compress_threads
= params
->compress_threads
;
1144 if (params
->has_decompress_threads
) {
1145 s
->parameters
.decompress_threads
= params
->decompress_threads
;
1148 if (params
->has_cpu_throttle_initial
) {
1149 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
1152 if (params
->has_cpu_throttle_increment
) {
1153 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
1156 if (params
->has_tls_creds
) {
1157 g_free(s
->parameters
.tls_creds
);
1158 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1159 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1162 if (params
->has_tls_hostname
) {
1163 g_free(s
->parameters
.tls_hostname
);
1164 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1165 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1168 if (params
->has_max_bandwidth
) {
1169 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1170 if (s
->to_dst_file
) {
1171 qemu_file_set_rate_limit(s
->to_dst_file
,
1172 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1176 if (params
->has_downtime_limit
) {
1177 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1180 if (params
->has_x_checkpoint_delay
) {
1181 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1182 if (migration_in_colo_state()) {
1183 colo_checkpoint_notify(s
);
1187 if (params
->has_block_incremental
) {
1188 s
->parameters
.block_incremental
= params
->block_incremental
;
1190 if (params
->has_x_multifd_channels
) {
1191 s
->parameters
.x_multifd_channels
= params
->x_multifd_channels
;
1193 if (params
->has_x_multifd_page_count
) {
1194 s
->parameters
.x_multifd_page_count
= params
->x_multifd_page_count
;
1196 if (params
->has_xbzrle_cache_size
) {
1197 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1198 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1200 if (params
->has_max_postcopy_bandwidth
) {
1201 s
->parameters
.max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1203 if (params
->has_max_cpu_throttle
) {
1204 s
->parameters
.max_cpu_throttle
= params
->max_cpu_throttle
;
1208 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1210 MigrationParameters tmp
;
1212 /* TODO Rewrite "" to null instead */
1213 if (params
->has_tls_creds
1214 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1215 qobject_unref(params
->tls_creds
->u
.n
);
1216 params
->tls_creds
->type
= QTYPE_QSTRING
;
1217 params
->tls_creds
->u
.s
= strdup("");
1219 /* TODO Rewrite "" to null instead */
1220 if (params
->has_tls_hostname
1221 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1222 qobject_unref(params
->tls_hostname
->u
.n
);
1223 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1224 params
->tls_hostname
->u
.s
= strdup("");
1227 migrate_params_test_apply(params
, &tmp
);
1229 if (!migrate_params_check(&tmp
, errp
)) {
1230 /* Invalid parameter */
1234 migrate_params_apply(params
, errp
);
1238 void qmp_migrate_start_postcopy(Error
**errp
)
1240 MigrationState
*s
= migrate_get_current();
1242 if (!migrate_postcopy()) {
1243 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1244 " the start of migration");
1248 if (s
->state
== MIGRATION_STATUS_NONE
) {
1249 error_setg(errp
, "Postcopy must be started after migration has been"
1254 * we don't error if migration has finished since that would be racy
1255 * with issuing this command.
1257 atomic_set(&s
->start_postcopy
, true);
1260 /* shared migration helpers */
1262 void migrate_set_state(int *state
, int old_state
, int new_state
)
1264 assert(new_state
< MIGRATION_STATUS__MAX
);
1265 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1266 trace_migrate_set_state(MigrationStatus_str(new_state
));
1267 migrate_generate_event(new_state
);
1271 static MigrationCapabilityStatusList
*migrate_cap_add(
1272 MigrationCapabilityStatusList
*list
,
1273 MigrationCapability index
,
1276 MigrationCapabilityStatusList
*cap
;
1278 cap
= g_new0(MigrationCapabilityStatusList
, 1);
1279 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
1280 cap
->value
->capability
= index
;
1281 cap
->value
->state
= state
;
1287 void migrate_set_block_enabled(bool value
, Error
**errp
)
1289 MigrationCapabilityStatusList
*cap
;
1291 cap
= migrate_cap_add(NULL
, MIGRATION_CAPABILITY_BLOCK
, value
);
1292 qmp_migrate_set_capabilities(cap
, errp
);
1293 qapi_free_MigrationCapabilityStatusList(cap
);
1296 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1298 s
->parameters
.block_incremental
= value
;
1301 static void block_cleanup_parameters(MigrationState
*s
)
1303 if (s
->must_remove_block_options
) {
1304 /* setting to false can never fail */
1305 migrate_set_block_enabled(false, &error_abort
);
1306 migrate_set_block_incremental(s
, false);
1307 s
->must_remove_block_options
= false;
1311 static void migrate_fd_cleanup(void *opaque
)
1313 MigrationState
*s
= opaque
;
1315 qemu_bh_delete(s
->cleanup_bh
);
1316 s
->cleanup_bh
= NULL
;
1318 qemu_savevm_state_cleanup();
1320 if (s
->to_dst_file
) {
1321 Error
*local_err
= NULL
;
1324 trace_migrate_fd_cleanup();
1325 qemu_mutex_unlock_iothread();
1326 if (s
->migration_thread_running
) {
1327 qemu_thread_join(&s
->thread
);
1328 s
->migration_thread_running
= false;
1330 qemu_mutex_lock_iothread();
1332 if (multifd_save_cleanup(&local_err
) != 0) {
1333 error_report_err(local_err
);
1335 qemu_mutex_lock(&s
->qemu_file_lock
);
1336 tmp
= s
->to_dst_file
;
1337 s
->to_dst_file
= NULL
;
1338 qemu_mutex_unlock(&s
->qemu_file_lock
);
1340 * Close the file handle without the lock to make sure the
1341 * critical section won't block for long.
1346 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
1347 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
1349 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1350 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1351 MIGRATION_STATUS_CANCELLED
);
1355 /* It is used on info migrate. We can't free it */
1356 error_report_err(error_copy(s
->error
));
1358 notifier_list_notify(&migration_state_notifiers
, s
);
1359 block_cleanup_parameters(s
);
1362 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1364 qemu_mutex_lock(&s
->error_mutex
);
1366 s
->error
= error_copy(error
);
1368 qemu_mutex_unlock(&s
->error_mutex
);
1371 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1373 trace_migrate_fd_error(error_get_pretty(error
));
1374 assert(s
->to_dst_file
== NULL
);
1375 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1376 MIGRATION_STATUS_FAILED
);
1377 migrate_set_error(s
, error
);
1380 static void migrate_fd_cancel(MigrationState
*s
)
1383 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1384 trace_migrate_fd_cancel();
1386 if (s
->rp_state
.from_dst_file
) {
1387 /* shutdown the rp socket, so causing the rp thread to shutdown */
1388 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1392 old_state
= s
->state
;
1393 if (!migration_is_setup_or_active(old_state
)) {
1396 /* If the migration is paused, kick it out of the pause */
1397 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1398 qemu_sem_post(&s
->pause_sem
);
1400 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1401 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1404 * If we're unlucky the migration code might be stuck somewhere in a
1405 * send/write while the network has failed and is waiting to timeout;
1406 * if we've got shutdown(2) available then we can force it to quit.
1407 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1408 * called in a bh, so there is no race against this cancel.
1410 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1411 qemu_file_shutdown(f
);
1413 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1414 Error
*local_err
= NULL
;
1416 bdrv_invalidate_cache_all(&local_err
);
1418 error_report_err(local_err
);
1420 s
->block_inactive
= false;
1425 void add_migration_state_change_notifier(Notifier
*notify
)
1427 notifier_list_add(&migration_state_notifiers
, notify
);
1430 void remove_migration_state_change_notifier(Notifier
*notify
)
1432 notifier_remove(notify
);
1435 bool migration_in_setup(MigrationState
*s
)
1437 return s
->state
== MIGRATION_STATUS_SETUP
;
1440 bool migration_has_finished(MigrationState
*s
)
1442 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1445 bool migration_has_failed(MigrationState
*s
)
1447 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1448 s
->state
== MIGRATION_STATUS_FAILED
);
1451 bool migration_in_postcopy(void)
1453 MigrationState
*s
= migrate_get_current();
1455 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1458 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1460 return migration_in_postcopy() && s
->postcopy_after_devices
;
1463 bool migration_is_idle(void)
1465 MigrationState
*s
= migrate_get_current();
1468 case MIGRATION_STATUS_NONE
:
1469 case MIGRATION_STATUS_CANCELLED
:
1470 case MIGRATION_STATUS_COMPLETED
:
1471 case MIGRATION_STATUS_FAILED
:
1473 case MIGRATION_STATUS_SETUP
:
1474 case MIGRATION_STATUS_CANCELLING
:
1475 case MIGRATION_STATUS_ACTIVE
:
1476 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1477 case MIGRATION_STATUS_COLO
:
1478 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1479 case MIGRATION_STATUS_DEVICE
:
1481 case MIGRATION_STATUS__MAX
:
1482 g_assert_not_reached();
1488 void migrate_init(MigrationState
*s
)
1491 * Reinitialise all migration state, except
1492 * parameters/capabilities that the user set, and
1498 s
->to_dst_file
= NULL
;
1499 s
->state
= MIGRATION_STATUS_NONE
;
1500 s
->rp_state
.from_dst_file
= NULL
;
1501 s
->rp_state
.error
= false;
1504 s
->expected_downtime
= 0;
1506 s
->start_postcopy
= false;
1507 s
->postcopy_after_devices
= false;
1508 s
->migration_thread_running
= false;
1509 error_free(s
->error
);
1512 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1514 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1516 s
->vm_was_running
= false;
1517 s
->iteration_initial_bytes
= 0;
1518 s
->threshold_size
= 0;
1521 static GSList
*migration_blockers
;
1523 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1525 if (migrate_get_current()->only_migratable
) {
1526 error_propagate(errp
, error_copy(reason
));
1527 error_prepend(errp
, "disallowing migration blocker "
1528 "(--only_migratable) for: ");
1532 if (migration_is_idle()) {
1533 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1537 error_propagate(errp
, error_copy(reason
));
1538 error_prepend(errp
, "disallowing migration blocker (migration in "
1543 void migrate_del_blocker(Error
*reason
)
1545 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1548 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1550 Error
*local_err
= NULL
;
1551 static bool once
= true;
1553 if (!deferred_incoming
) {
1554 error_setg(errp
, "For use with '-incoming defer'");
1558 error_setg(errp
, "The incoming migration has already been started");
1561 qemu_start_incoming_migration(uri
, &local_err
);
1564 error_propagate(errp
, local_err
);
1571 void qmp_migrate_recover(const char *uri
, Error
**errp
)
1573 MigrationIncomingState
*mis
= migration_incoming_get_current();
1575 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1576 error_setg(errp
, "Migrate recover can only be run "
1577 "when postcopy is paused.");
1581 if (atomic_cmpxchg(&mis
->postcopy_recover_triggered
,
1582 false, true) == true) {
1583 error_setg(errp
, "Migrate recovery is triggered already");
1588 * Note that this call will never start a real migration; it will
1589 * only re-setup the migration stream and poke existing migration
1590 * to continue using that newly established channel.
1592 qemu_start_incoming_migration(uri
, errp
);
1595 void qmp_migrate_pause(Error
**errp
)
1597 MigrationState
*ms
= migrate_get_current();
1598 MigrationIncomingState
*mis
= migration_incoming_get_current();
1601 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1602 /* Source side, during postcopy */
1603 qemu_mutex_lock(&ms
->qemu_file_lock
);
1604 ret
= qemu_file_shutdown(ms
->to_dst_file
);
1605 qemu_mutex_unlock(&ms
->qemu_file_lock
);
1607 error_setg(errp
, "Failed to pause source migration");
1612 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1613 ret
= qemu_file_shutdown(mis
->from_src_file
);
1615 error_setg(errp
, "Failed to pause destination migration");
1620 error_setg(errp
, "migrate-pause is currently only supported "
1621 "during postcopy-active state");
1624 bool migration_is_blocked(Error
**errp
)
1626 if (qemu_savevm_state_blocked(errp
)) {
1630 if (migration_blockers
) {
1631 error_propagate(errp
, error_copy(migration_blockers
->data
));
1638 /* Returns true if continue to migrate, or false if error detected */
1639 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
1640 bool resume
, Error
**errp
)
1642 Error
*local_err
= NULL
;
1645 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1646 error_setg(errp
, "Cannot resume if there is no "
1647 "paused migration");
1652 * Postcopy recovery won't work well with release-ram
1653 * capability since release-ram will drop the page buffer as
1654 * long as the page is put into the send buffer. So if there
1655 * is a network failure happened, any page buffers that have
1656 * not yet reached the destination VM but have already been
1657 * sent from the source VM will be lost forever. Let's refuse
1658 * the client from resuming such a postcopy migration.
1659 * Luckily release-ram was designed to only be used when src
1660 * and destination VMs are on the same host, so it should be
1663 if (migrate_release_ram()) {
1664 error_setg(errp
, "Postcopy recovery cannot work "
1665 "when release-ram capability is set");
1669 /* This is a resume, skip init status */
1673 if (migration_is_setup_or_active(s
->state
) ||
1674 s
->state
== MIGRATION_STATUS_CANCELLING
||
1675 s
->state
== MIGRATION_STATUS_COLO
) {
1676 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1680 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1681 error_setg(errp
, "Guest is waiting for an incoming migration");
1685 if (migration_is_blocked(errp
)) {
1689 if (blk
|| blk_inc
) {
1690 if (migrate_use_block() || migrate_use_block_incremental()) {
1691 error_setg(errp
, "Command options are incompatible with "
1692 "current migration capabilities");
1695 migrate_set_block_enabled(true, &local_err
);
1697 error_propagate(errp
, local_err
);
1700 s
->must_remove_block_options
= true;
1704 migrate_set_block_incremental(s
, true);
1712 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1713 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1714 bool has_resume
, bool resume
, Error
**errp
)
1716 Error
*local_err
= NULL
;
1717 MigrationState
*s
= migrate_get_current();
1720 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
1721 has_resume
&& resume
, errp
)) {
1722 /* Error detected, put into errp */
1726 if (strstart(uri
, "tcp:", &p
)) {
1727 tcp_start_outgoing_migration(s
, p
, &local_err
);
1729 } else if (strstart(uri
, "rdma:", &p
)) {
1730 rdma_start_outgoing_migration(s
, p
, &local_err
);
1732 } else if (strstart(uri
, "exec:", &p
)) {
1733 exec_start_outgoing_migration(s
, p
, &local_err
);
1734 } else if (strstart(uri
, "unix:", &p
)) {
1735 unix_start_outgoing_migration(s
, p
, &local_err
);
1736 } else if (strstart(uri
, "fd:", &p
)) {
1737 fd_start_outgoing_migration(s
, p
, &local_err
);
1739 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1740 "a valid migration protocol");
1741 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1742 MIGRATION_STATUS_FAILED
);
1743 block_cleanup_parameters(s
);
1748 migrate_fd_error(s
, local_err
);
1749 error_propagate(errp
, local_err
);
1754 void qmp_migrate_cancel(Error
**errp
)
1756 migrate_fd_cancel(migrate_get_current());
1759 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
1761 MigrationState
*s
= migrate_get_current();
1762 if (s
->state
!= state
) {
1763 error_setg(errp
, "Migration not in expected state: %s",
1764 MigrationStatus_str(s
->state
));
1767 qemu_sem_post(&s
->pause_sem
);
1770 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1772 MigrateSetParameters p
= {
1773 .has_xbzrle_cache_size
= true,
1774 .xbzrle_cache_size
= value
,
1777 qmp_migrate_set_parameters(&p
, errp
);
1780 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1782 return migrate_xbzrle_cache_size();
1785 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1787 MigrateSetParameters p
= {
1788 .has_max_bandwidth
= true,
1789 .max_bandwidth
= value
,
1792 qmp_migrate_set_parameters(&p
, errp
);
1795 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1797 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
1798 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1799 "the range of 0 to %d seconds",
1800 MAX_MIGRATE_DOWNTIME_SECONDS
);
1804 value
*= 1000; /* Convert to milliseconds */
1805 value
= MAX(0, MIN(INT64_MAX
, value
));
1807 MigrateSetParameters p
= {
1808 .has_downtime_limit
= true,
1809 .downtime_limit
= value
,
1812 qmp_migrate_set_parameters(&p
, errp
);
1815 bool migrate_release_ram(void)
1819 s
= migrate_get_current();
1821 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
1824 bool migrate_postcopy_ram(void)
1828 s
= migrate_get_current();
1830 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1833 bool migrate_postcopy(void)
1835 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
1838 bool migrate_auto_converge(void)
1842 s
= migrate_get_current();
1844 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
1847 bool migrate_zero_blocks(void)
1851 s
= migrate_get_current();
1853 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
1856 bool migrate_postcopy_blocktime(void)
1860 s
= migrate_get_current();
1862 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
1865 bool migrate_use_compression(void)
1869 s
= migrate_get_current();
1871 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
1874 int migrate_compress_level(void)
1878 s
= migrate_get_current();
1880 return s
->parameters
.compress_level
;
1883 int migrate_compress_threads(void)
1887 s
= migrate_get_current();
1889 return s
->parameters
.compress_threads
;
1892 int migrate_decompress_threads(void)
1896 s
= migrate_get_current();
1898 return s
->parameters
.decompress_threads
;
1901 bool migrate_dirty_bitmaps(void)
1905 s
= migrate_get_current();
1907 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
1910 bool migrate_use_events(void)
1914 s
= migrate_get_current();
1916 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
1919 bool migrate_use_multifd(void)
1923 s
= migrate_get_current();
1925 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_MULTIFD
];
1928 bool migrate_pause_before_switchover(void)
1932 s
= migrate_get_current();
1934 return s
->enabled_capabilities
[
1935 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
1938 int migrate_multifd_channels(void)
1942 s
= migrate_get_current();
1944 return s
->parameters
.x_multifd_channels
;
1947 int migrate_multifd_page_count(void)
1951 s
= migrate_get_current();
1953 return s
->parameters
.x_multifd_page_count
;
1956 int migrate_use_xbzrle(void)
1960 s
= migrate_get_current();
1962 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
1965 int64_t migrate_xbzrle_cache_size(void)
1969 s
= migrate_get_current();
1971 return s
->parameters
.xbzrle_cache_size
;
1974 static int64_t migrate_max_postcopy_bandwidth(void)
1978 s
= migrate_get_current();
1980 return s
->parameters
.max_postcopy_bandwidth
;
1983 bool migrate_use_block(void)
1987 s
= migrate_get_current();
1989 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
1992 bool migrate_use_return_path(void)
1996 s
= migrate_get_current();
1998 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
2001 bool migrate_use_block_incremental(void)
2005 s
= migrate_get_current();
2007 return s
->parameters
.block_incremental
;
2010 /* migration thread support */
2012 * Something bad happened to the RP stream, mark an error
2013 * The caller shall print or trace something to indicate why
2015 static void mark_source_rp_bad(MigrationState
*s
)
2017 s
->rp_state
.error
= true;
2020 static struct rp_cmd_args
{
2021 ssize_t len
; /* -1 = variable */
2024 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
2025 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
2026 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
2027 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
2028 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
2029 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
2030 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
2031 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
2035 * Process a request for pages received on the return path,
2036 * We're allowed to send more than requested (e.g. to round to our page size)
2037 * and we don't need to send pages that have already been sent.
2039 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
2040 ram_addr_t start
, size_t len
)
2042 long our_host_ps
= getpagesize();
2044 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
2047 * Since we currently insist on matching page sizes, just sanity check
2048 * we're being asked for whole host pages.
2050 if (start
& (our_host_ps
-1) ||
2051 (len
& (our_host_ps
-1))) {
2052 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2053 " len: %zd", __func__
, start
, len
);
2054 mark_source_rp_bad(ms
);
2058 if (ram_save_queue_pages(rbname
, start
, len
)) {
2059 mark_source_rp_bad(ms
);
2063 /* Return true to retry, false to quit */
2064 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
2066 trace_postcopy_pause_return_path();
2068 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
2070 trace_postcopy_pause_return_path_continued();
2075 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
2077 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
2080 error_report("%s: invalid block name '%s'", __func__
, block_name
);
2084 /* Fetch the received bitmap and refresh the dirty bitmap */
2085 return ram_dirty_bitmap_reload(s
, block
);
2088 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
2090 trace_source_return_path_thread_resume_ack(value
);
2092 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
2093 error_report("%s: illegal resume_ack value %"PRIu32
,
2098 /* Now both sides are active. */
2099 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2100 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2102 /* Notify send thread that time to continue send pages */
2103 qemu_sem_post(&s
->rp_state
.rp_sem
);
2109 * Handles messages sent on the return path towards the source VM
2112 static void *source_return_path_thread(void *opaque
)
2114 MigrationState
*ms
= opaque
;
2115 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
2116 uint16_t header_len
, header_type
;
2118 uint32_t tmp32
, sibling_error
;
2119 ram_addr_t start
= 0; /* =0 to silence warning */
2120 size_t len
= 0, expected_len
;
2123 trace_source_return_path_thread_entry();
2124 rcu_register_thread();
2127 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
2128 migration_is_setup_or_active(ms
->state
)) {
2129 trace_source_return_path_thread_loop_top();
2130 header_type
= qemu_get_be16(rp
);
2131 header_len
= qemu_get_be16(rp
);
2133 if (qemu_file_get_error(rp
)) {
2134 mark_source_rp_bad(ms
);
2138 if (header_type
>= MIG_RP_MSG_MAX
||
2139 header_type
== MIG_RP_MSG_INVALID
) {
2140 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2141 header_type
, header_len
);
2142 mark_source_rp_bad(ms
);
2146 if ((rp_cmd_args
[header_type
].len
!= -1 &&
2147 header_len
!= rp_cmd_args
[header_type
].len
) ||
2148 header_len
> sizeof(buf
)) {
2149 error_report("RP: Received '%s' message (0x%04x) with"
2150 "incorrect length %d expecting %zu",
2151 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
2152 (size_t)rp_cmd_args
[header_type
].len
);
2153 mark_source_rp_bad(ms
);
2157 /* We know we've got a valid header by this point */
2158 res
= qemu_get_buffer(rp
, buf
, header_len
);
2159 if (res
!= header_len
) {
2160 error_report("RP: Failed reading data for message 0x%04x"
2161 " read %d expected %d",
2162 header_type
, res
, header_len
);
2163 mark_source_rp_bad(ms
);
2167 /* OK, we have the message and the data */
2168 switch (header_type
) {
2169 case MIG_RP_MSG_SHUT
:
2170 sibling_error
= ldl_be_p(buf
);
2171 trace_source_return_path_thread_shut(sibling_error
);
2172 if (sibling_error
) {
2173 error_report("RP: Sibling indicated error %d", sibling_error
);
2174 mark_source_rp_bad(ms
);
2177 * We'll let the main thread deal with closing the RP
2178 * we could do a shutdown(2) on it, but we're the only user
2179 * anyway, so there's nothing gained.
2183 case MIG_RP_MSG_PONG
:
2184 tmp32
= ldl_be_p(buf
);
2185 trace_source_return_path_thread_pong(tmp32
);
2188 case MIG_RP_MSG_REQ_PAGES
:
2189 start
= ldq_be_p(buf
);
2190 len
= ldl_be_p(buf
+ 8);
2191 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
2194 case MIG_RP_MSG_REQ_PAGES_ID
:
2195 expected_len
= 12 + 1; /* header + termination */
2197 if (header_len
>= expected_len
) {
2198 start
= ldq_be_p(buf
);
2199 len
= ldl_be_p(buf
+ 8);
2200 /* Now we expect an idstr */
2201 tmp32
= buf
[12]; /* Length of the following idstr */
2202 buf
[13 + tmp32
] = '\0';
2203 expected_len
+= tmp32
;
2205 if (header_len
!= expected_len
) {
2206 error_report("RP: Req_Page_id with length %d expecting %zd",
2207 header_len
, expected_len
);
2208 mark_source_rp_bad(ms
);
2211 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
2214 case MIG_RP_MSG_RECV_BITMAP
:
2215 if (header_len
< 1) {
2216 error_report("%s: missing block name", __func__
);
2217 mark_source_rp_bad(ms
);
2220 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2221 buf
[buf
[0] + 1] = '\0';
2222 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
2223 mark_source_rp_bad(ms
);
2228 case MIG_RP_MSG_RESUME_ACK
:
2229 tmp32
= ldl_be_p(buf
);
2230 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
2231 mark_source_rp_bad(ms
);
2242 res
= qemu_file_get_error(rp
);
2246 * Maybe there is something we can do: it looks like a
2247 * network down issue, and we pause for a recovery.
2249 if (postcopy_pause_return_path_thread(ms
)) {
2250 /* Reload rp, reset the rest */
2251 rp
= ms
->rp_state
.from_dst_file
;
2252 ms
->rp_state
.error
= false;
2257 trace_source_return_path_thread_bad_end();
2258 mark_source_rp_bad(ms
);
2261 trace_source_return_path_thread_end();
2262 ms
->rp_state
.from_dst_file
= NULL
;
2264 rcu_unregister_thread();
2268 static int open_return_path_on_source(MigrationState
*ms
,
2272 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
2273 if (!ms
->rp_state
.from_dst_file
) {
2277 trace_open_return_path_on_source();
2279 if (!create_thread
) {
2284 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
2285 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
2287 trace_open_return_path_on_source_continue();
2292 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2293 static int await_return_path_close_on_source(MigrationState
*ms
)
2296 * If this is a normal exit then the destination will send a SHUT and the
2297 * rp_thread will exit, however if there's an error we need to cause
2300 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
2302 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2303 * waiting for the destination.
2305 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
2306 mark_source_rp_bad(ms
);
2308 trace_await_return_path_close_on_source_joining();
2309 qemu_thread_join(&ms
->rp_state
.rp_thread
);
2310 trace_await_return_path_close_on_source_close();
2311 return ms
->rp_state
.error
;
2315 * Switch from normal iteration to postcopy
2316 * Returns non-0 on error
2318 static int postcopy_start(MigrationState
*ms
)
2321 QIOChannelBuffer
*bioc
;
2323 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2324 int64_t bandwidth
= migrate_max_postcopy_bandwidth();
2325 bool restart_block
= false;
2326 int cur_state
= MIGRATION_STATUS_ACTIVE
;
2327 if (!migrate_pause_before_switchover()) {
2328 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
2329 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2332 trace_postcopy_start();
2333 qemu_mutex_lock_iothread();
2334 trace_postcopy_start_set_run();
2336 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
2337 global_state_store();
2338 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2343 ret
= migration_maybe_pause(ms
, &cur_state
,
2344 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2349 ret
= bdrv_inactivate_all();
2353 restart_block
= true;
2356 * Cause any non-postcopiable, but iterative devices to
2357 * send out their final data.
2359 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2362 * in Finish migrate and with the io-lock held everything should
2363 * be quiet, but we've potentially still got dirty pages and we
2364 * need to tell the destination to throw any pages it's already received
2367 if (migrate_postcopy_ram()) {
2368 if (ram_postcopy_send_discard_bitmap(ms
)) {
2369 error_report("postcopy send discard bitmap failed");
2375 * send rest of state - note things that are doing postcopy
2376 * will notice we're in POSTCOPY_ACTIVE and not actually
2377 * wrap their state up here
2379 /* 0 max-postcopy-bandwidth means unlimited */
2381 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
2383 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
2385 if (migrate_postcopy_ram()) {
2386 /* Ping just for debugging, helps line traces up */
2387 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
2391 * While loading the device state we may trigger page transfer
2392 * requests and the fd must be free to process those, and thus
2393 * the destination must read the whole device state off the fd before
2394 * it starts processing it. Unfortunately the ad-hoc migration format
2395 * doesn't allow the destination to know the size to read without fully
2396 * parsing it through each devices load-state code (especially the open
2397 * coded devices that use get/put).
2398 * So we wrap the device state up in a package with a length at the start;
2399 * to do this we use a qemu_buf to hold the whole of the device state.
2401 bioc
= qio_channel_buffer_new(4096);
2402 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
2403 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
2404 object_unref(OBJECT(bioc
));
2407 * Make sure the receiver can get incoming pages before we send the rest
2410 qemu_savevm_send_postcopy_listen(fb
);
2412 qemu_savevm_state_complete_precopy(fb
, false, false);
2413 if (migrate_postcopy_ram()) {
2414 qemu_savevm_send_ping(fb
, 3);
2417 qemu_savevm_send_postcopy_run(fb
);
2419 /* <><> end of stuff going into the package */
2421 /* Last point of recovery; as soon as we send the package the destination
2422 * can open devices and potentially start running.
2423 * Lets just check again we've not got any errors.
2425 ret
= qemu_file_get_error(ms
->to_dst_file
);
2427 error_report("postcopy_start: Migration stream errored (pre package)");
2431 restart_block
= false;
2433 /* Now send that blob */
2434 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2439 /* Send a notify to give a chance for anything that needs to happen
2440 * at the transition to postcopy and after the device state; in particular
2441 * spice needs to trigger a transition now
2443 ms
->postcopy_after_devices
= true;
2444 notifier_list_notify(&migration_state_notifiers
, ms
);
2446 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2448 qemu_mutex_unlock_iothread();
2450 if (migrate_postcopy_ram()) {
2452 * Although this ping is just for debug, it could potentially be
2453 * used for getting a better measurement of downtime at the source.
2455 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2458 if (migrate_release_ram()) {
2459 ram_postcopy_migrated_memory_release(ms
);
2462 ret
= qemu_file_get_error(ms
->to_dst_file
);
2464 error_report("postcopy_start: Migration stream errored");
2465 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2466 MIGRATION_STATUS_FAILED
);
2474 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2475 MIGRATION_STATUS_FAILED
);
2476 if (restart_block
) {
2477 /* A failure happened early enough that we know the destination hasn't
2478 * accessed block devices, so we're safe to recover.
2480 Error
*local_err
= NULL
;
2482 bdrv_invalidate_cache_all(&local_err
);
2484 error_report_err(local_err
);
2487 qemu_mutex_unlock_iothread();
2492 * migration_maybe_pause: Pause if required to by
2493 * migrate_pause_before_switchover called with the iothread locked
2494 * Returns: 0 on success
2496 static int migration_maybe_pause(MigrationState
*s
,
2497 int *current_active_state
,
2500 if (!migrate_pause_before_switchover()) {
2504 /* Since leaving this state is not atomic with posting the semaphore
2505 * it's possible that someone could have issued multiple migrate_continue
2506 * and the semaphore is incorrectly positive at this point;
2507 * the docs say it's undefined to reinit a semaphore that's already
2508 * init'd, so use timedwait to eat up any existing posts.
2510 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2511 /* This block intentionally left blank */
2514 qemu_mutex_unlock_iothread();
2515 migrate_set_state(&s
->state
, *current_active_state
,
2516 MIGRATION_STATUS_PRE_SWITCHOVER
);
2517 qemu_sem_wait(&s
->pause_sem
);
2518 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2520 *current_active_state
= new_state
;
2521 qemu_mutex_lock_iothread();
2523 return s
->state
== new_state
? 0 : -EINVAL
;
2527 * migration_completion: Used by migration_thread when there's not much left.
2528 * The caller 'breaks' the loop when this returns.
2530 * @s: Current migration state
2532 static void migration_completion(MigrationState
*s
)
2535 int current_active_state
= s
->state
;
2537 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2538 qemu_mutex_lock_iothread();
2539 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2540 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
2541 s
->vm_was_running
= runstate_is_running();
2542 ret
= global_state_store();
2545 bool inactivate
= !migrate_colo_enabled();
2546 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2548 ret
= migration_maybe_pause(s
, ¤t_active_state
,
2549 MIGRATION_STATUS_DEVICE
);
2552 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
2553 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
2556 if (inactivate
&& ret
>= 0) {
2557 s
->block_inactive
= true;
2560 qemu_mutex_unlock_iothread();
2565 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2566 trace_migration_completion_postcopy_end();
2568 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
2569 trace_migration_completion_postcopy_end_after_complete();
2573 * If rp was opened we must clean up the thread before
2574 * cleaning everything else up (since if there are no failures
2575 * it will wait for the destination to send it's status in
2578 if (s
->rp_state
.from_dst_file
) {
2580 trace_migration_return_path_end_before();
2581 rp_error
= await_return_path_close_on_source(s
);
2582 trace_migration_return_path_end_after(rp_error
);
2584 goto fail_invalidate
;
2588 if (qemu_file_get_error(s
->to_dst_file
)) {
2589 trace_migration_completion_file_err();
2590 goto fail_invalidate
;
2593 if (!migrate_colo_enabled()) {
2594 migrate_set_state(&s
->state
, current_active_state
,
2595 MIGRATION_STATUS_COMPLETED
);
2601 /* If not doing postcopy, vm_start() will be called: let's regain
2602 * control on images.
2604 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
2605 s
->state
== MIGRATION_STATUS_DEVICE
) {
2606 Error
*local_err
= NULL
;
2608 qemu_mutex_lock_iothread();
2609 bdrv_invalidate_cache_all(&local_err
);
2611 error_report_err(local_err
);
2613 s
->block_inactive
= false;
2615 qemu_mutex_unlock_iothread();
2619 migrate_set_state(&s
->state
, current_active_state
,
2620 MIGRATION_STATUS_FAILED
);
2623 bool migrate_colo_enabled(void)
2625 MigrationState
*s
= migrate_get_current();
2626 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
2629 typedef enum MigThrError
{
2630 /* No error detected */
2631 MIG_THR_ERR_NONE
= 0,
2632 /* Detected error, but resumed successfully */
2633 MIG_THR_ERR_RECOVERED
= 1,
2634 /* Detected fatal error, need to exit */
2635 MIG_THR_ERR_FATAL
= 2,
2638 static int postcopy_resume_handshake(MigrationState
*s
)
2640 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
2642 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2643 qemu_sem_wait(&s
->rp_state
.rp_sem
);
2646 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2653 /* Return zero if success, or <0 for error */
2654 static int postcopy_do_resume(MigrationState
*s
)
2659 * Call all the resume_prepare() hooks, so that modules can be
2660 * ready for the migration resume.
2662 ret
= qemu_savevm_state_resume_prepare(s
);
2664 error_report("%s: resume_prepare() failure detected: %d",
2670 * Last handshake with destination on the resume (destination will
2671 * switch to postcopy-active afterwards)
2673 ret
= postcopy_resume_handshake(s
);
2675 error_report("%s: handshake failed: %d", __func__
, ret
);
2683 * We don't return until we are in a safe state to continue current
2684 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2685 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2687 static MigThrError
postcopy_pause(MigrationState
*s
)
2689 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2694 migrate_set_state(&s
->state
, s
->state
,
2695 MIGRATION_STATUS_POSTCOPY_PAUSED
);
2697 /* Current channel is possibly broken. Release it. */
2698 assert(s
->to_dst_file
);
2699 qemu_mutex_lock(&s
->qemu_file_lock
);
2700 file
= s
->to_dst_file
;
2701 s
->to_dst_file
= NULL
;
2702 qemu_mutex_unlock(&s
->qemu_file_lock
);
2704 qemu_file_shutdown(file
);
2707 error_report("Detected IO failure for postcopy. "
2708 "Migration paused.");
2711 * We wait until things fixed up. Then someone will setup the
2712 * status back for us.
2714 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2715 qemu_sem_wait(&s
->postcopy_pause_sem
);
2718 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2719 /* Woken up by a recover procedure. Give it a shot */
2722 * Firstly, let's wake up the return path now, with a new
2723 * return path channel.
2725 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
2727 /* Do the resume logic */
2728 if (postcopy_do_resume(s
) == 0) {
2729 /* Let's continue! */
2730 trace_postcopy_pause_continued();
2731 return MIG_THR_ERR_RECOVERED
;
2734 * Something wrong happened during the recovery, let's
2735 * pause again. Pause is always better than throwing
2741 /* This is not right... Time to quit. */
2742 return MIG_THR_ERR_FATAL
;
2747 static MigThrError
migration_detect_error(MigrationState
*s
)
2751 /* Try to detect any file errors */
2752 ret
= qemu_file_get_error(s
->to_dst_file
);
2755 /* Everything is fine */
2756 return MIG_THR_ERR_NONE
;
2759 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
== -EIO
) {
2761 * For postcopy, we allow the network to be down for a
2762 * while. After that, it can be continued by a
2765 return postcopy_pause(s
);
2768 * For precopy (or postcopy with error outside IO), we fail
2771 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
2772 trace_migration_thread_file_err();
2774 /* Time to stop the migration, now. */
2775 return MIG_THR_ERR_FATAL
;
2779 /* How many bytes have we transferred since the beggining of the migration */
2780 static uint64_t migration_total_bytes(MigrationState
*s
)
2782 return qemu_ftell(s
->to_dst_file
) + ram_counters
.multifd_bytes
;
2785 static void migration_calculate_complete(MigrationState
*s
)
2787 uint64_t bytes
= migration_total_bytes(s
);
2788 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2789 int64_t transfer_time
;
2791 s
->total_time
= end_time
- s
->start_time
;
2794 * It's still not set, so we are precopy migration. For
2795 * postcopy, downtime is calculated during postcopy_start().
2797 s
->downtime
= end_time
- s
->downtime_start
;
2800 transfer_time
= s
->total_time
- s
->setup_time
;
2801 if (transfer_time
) {
2802 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
2806 static void migration_update_counters(MigrationState
*s
,
2807 int64_t current_time
)
2809 uint64_t transferred
, time_spent
;
2810 uint64_t current_bytes
; /* bytes transferred since the beginning */
2813 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
2817 current_bytes
= migration_total_bytes(s
);
2818 transferred
= current_bytes
- s
->iteration_initial_bytes
;
2819 time_spent
= current_time
- s
->iteration_start_time
;
2820 bandwidth
= (double)transferred
/ time_spent
;
2821 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
2823 s
->mbps
= (((double) transferred
* 8.0) /
2824 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2827 * if we haven't sent anything, we don't want to
2828 * recalculate. 10000 is a small enough number for our purposes
2830 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
2831 s
->expected_downtime
= ram_counters
.remaining
/ bandwidth
;
2834 qemu_file_reset_rate_limit(s
->to_dst_file
);
2836 s
->iteration_start_time
= current_time
;
2837 s
->iteration_initial_bytes
= current_bytes
;
2839 trace_migrate_transferred(transferred
, time_spent
,
2840 bandwidth
, s
->threshold_size
);
2843 /* Migration thread iteration status */
2845 MIG_ITERATE_RESUME
, /* Resume current iteration */
2846 MIG_ITERATE_SKIP
, /* Skip current iteration */
2847 MIG_ITERATE_BREAK
, /* Break the loop */
2851 * Return true if continue to the next iteration directly, false
2854 static MigIterateState
migration_iteration_run(MigrationState
*s
)
2856 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
2857 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
2859 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
2860 &pend_compat
, &pend_post
);
2861 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
2863 trace_migrate_pending(pending_size
, s
->threshold_size
,
2864 pend_pre
, pend_compat
, pend_post
);
2866 if (pending_size
&& pending_size
>= s
->threshold_size
) {
2867 /* Still a significant amount to transfer */
2868 if (migrate_postcopy() && !in_postcopy
&&
2869 pend_pre
<= s
->threshold_size
&&
2870 atomic_read(&s
->start_postcopy
)) {
2871 if (postcopy_start(s
)) {
2872 error_report("%s: postcopy failed to start", __func__
);
2874 return MIG_ITERATE_SKIP
;
2876 /* Just another iteration step */
2877 qemu_savevm_state_iterate(s
->to_dst_file
,
2878 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2880 trace_migration_thread_low_pending(pending_size
);
2881 migration_completion(s
);
2882 return MIG_ITERATE_BREAK
;
2885 return MIG_ITERATE_RESUME
;
2888 static void migration_iteration_finish(MigrationState
*s
)
2890 /* If we enabled cpu throttling for auto-converge, turn it off. */
2891 cpu_throttle_stop();
2893 qemu_mutex_lock_iothread();
2895 case MIGRATION_STATUS_COMPLETED
:
2896 migration_calculate_complete(s
);
2897 runstate_set(RUN_STATE_POSTMIGRATE
);
2900 case MIGRATION_STATUS_ACTIVE
:
2902 * We should really assert here, but since it's during
2903 * migration, let's try to reduce the usage of assertions.
2905 if (!migrate_colo_enabled()) {
2906 error_report("%s: critical error: calling COLO code without "
2907 "COLO enabled", __func__
);
2909 migrate_start_colo_process(s
);
2911 * Fixme: we will run VM in COLO no matter its old running state.
2912 * After exited COLO, we will keep running.
2914 s
->vm_was_running
= true;
2916 case MIGRATION_STATUS_FAILED
:
2917 case MIGRATION_STATUS_CANCELLED
:
2918 case MIGRATION_STATUS_CANCELLING
:
2919 if (s
->vm_was_running
) {
2922 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2923 runstate_set(RUN_STATE_POSTMIGRATE
);
2929 /* Should not reach here, but if so, forgive the VM. */
2930 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
2933 qemu_bh_schedule(s
->cleanup_bh
);
2934 qemu_mutex_unlock_iothread();
2937 void migration_make_urgent_request(void)
2939 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
2942 void migration_consume_urgent_request(void)
2944 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
2948 * Master migration thread on the source VM.
2949 * It drives the migration and pumps the data down the outgoing channel.
2951 static void *migration_thread(void *opaque
)
2953 MigrationState
*s
= opaque
;
2954 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
2955 MigThrError thr_error
;
2956 bool urgent
= false;
2958 rcu_register_thread();
2960 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2962 qemu_savevm_state_header(s
->to_dst_file
);
2965 * If we opened the return path, we need to make sure dst has it
2968 if (s
->rp_state
.from_dst_file
) {
2969 /* Now tell the dest that it should open its end so it can reply */
2970 qemu_savevm_send_open_return_path(s
->to_dst_file
);
2972 /* And do a ping that will make stuff easier to debug */
2973 qemu_savevm_send_ping(s
->to_dst_file
, 1);
2976 if (migrate_postcopy()) {
2978 * Tell the destination that we *might* want to do postcopy later;
2979 * if the other end can't do postcopy it should fail now, nice and
2982 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
2985 qemu_savevm_state_setup(s
->to_dst_file
);
2987 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
2988 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2989 MIGRATION_STATUS_ACTIVE
);
2991 trace_migration_thread_setup_complete();
2993 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
2994 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2995 int64_t current_time
;
2997 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
2998 MigIterateState iter_state
= migration_iteration_run(s
);
2999 if (iter_state
== MIG_ITERATE_SKIP
) {
3001 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3007 * Try to detect any kind of failures, and see whether we
3008 * should stop the migration now.
3010 thr_error
= migration_detect_error(s
);
3011 if (thr_error
== MIG_THR_ERR_FATAL
) {
3012 /* Stop migration */
3014 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
3016 * Just recovered from a e.g. network failure, reset all
3017 * the local variables. This is important to avoid
3018 * breaking transferred_bytes and bandwidth calculation
3020 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3021 s
->iteration_initial_bytes
= 0;
3024 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3026 migration_update_counters(s
, current_time
);
3029 if (qemu_file_rate_limit(s
->to_dst_file
)) {
3030 /* Wait for a delay to do rate limiting OR
3031 * something urgent to post the semaphore.
3033 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- current_time
;
3034 trace_migration_thread_ratelimit_pre(ms
);
3035 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
3036 /* We were worken by one or more urgent things but
3037 * the timedwait will have consumed one of them.
3038 * The service routine for the urgent wake will dec
3039 * the semaphore itself for each item it consumes,
3040 * so add this one we just eat back.
3042 qemu_sem_post(&s
->rate_limit_sem
);
3045 trace_migration_thread_ratelimit_post(urgent
);
3049 trace_migration_thread_after_loop();
3050 migration_iteration_finish(s
);
3051 rcu_unregister_thread();
3055 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
3058 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
3060 s
->expected_downtime
= s
->parameters
.downtime_limit
;
3061 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
3063 migrate_fd_error(s
, error_in
);
3064 migrate_fd_cleanup(s
);
3069 /* This is a resumed migration */
3070 rate_limit
= INT64_MAX
;
3072 /* This is a fresh new migration */
3073 rate_limit
= s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
;
3075 /* Notify before starting migration thread */
3076 notifier_list_notify(&migration_state_notifiers
, s
);
3079 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
3080 qemu_file_set_blocking(s
->to_dst_file
, true);
3083 * Open the return path. For postcopy, it is used exclusively. For
3084 * precopy, only if user specified "return-path" capability would
3085 * QEMU uses the return path.
3087 if (migrate_postcopy_ram() || migrate_use_return_path()) {
3088 if (open_return_path_on_source(s
, !resume
)) {
3089 error_report("Unable to open return-path for postcopy");
3090 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
3091 migrate_fd_cleanup(s
);
3097 /* Wakeup the main migration thread to do the recovery */
3098 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
3099 MIGRATION_STATUS_POSTCOPY_RECOVER
);
3100 qemu_sem_post(&s
->postcopy_pause_sem
);
3104 if (multifd_save_setup() != 0) {
3105 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3106 MIGRATION_STATUS_FAILED
);
3107 migrate_fd_cleanup(s
);
3110 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
3111 QEMU_THREAD_JOINABLE
);
3112 s
->migration_thread_running
= true;
3115 void migration_global_dump(Monitor
*mon
)
3117 MigrationState
*ms
= migrate_get_current();
3119 monitor_printf(mon
, "globals:\n");
3120 monitor_printf(mon
, "store-global-state: %s\n",
3121 ms
->store_global_state
? "on" : "off");
3122 monitor_printf(mon
, "only-migratable: %s\n",
3123 ms
->only_migratable
? "on" : "off");
3124 monitor_printf(mon
, "send-configuration: %s\n",
3125 ms
->send_configuration
? "on" : "off");
3126 monitor_printf(mon
, "send-section-footer: %s\n",
3127 ms
->send_section_footer
? "on" : "off");
3128 monitor_printf(mon
, "decompress-error-check: %s\n",
3129 ms
->decompress_error_check
? "on" : "off");
3132 #define DEFINE_PROP_MIG_CAP(name, x) \
3133 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3135 static Property migration_properties
[] = {
3136 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
3137 store_global_state
, true),
3138 DEFINE_PROP_BOOL("only-migratable", MigrationState
, only_migratable
, false),
3139 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
3140 send_configuration
, true),
3141 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
3142 send_section_footer
, true),
3143 DEFINE_PROP_BOOL("decompress-error-check", MigrationState
,
3144 decompress_error_check
, true),
3146 /* Migration parameters */
3147 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
3148 parameters
.compress_level
,
3149 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
3150 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
3151 parameters
.compress_threads
,
3152 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
3153 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
3154 parameters
.decompress_threads
,
3155 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
3156 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
3157 parameters
.cpu_throttle_initial
,
3158 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
3159 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
3160 parameters
.cpu_throttle_increment
,
3161 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
3162 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
3163 parameters
.max_bandwidth
, MAX_THROTTLE
),
3164 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
3165 parameters
.downtime_limit
,
3166 DEFAULT_MIGRATE_SET_DOWNTIME
),
3167 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
3168 parameters
.x_checkpoint_delay
,
3169 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
3170 DEFINE_PROP_UINT8("x-multifd-channels", MigrationState
,
3171 parameters
.x_multifd_channels
,
3172 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
3173 DEFINE_PROP_UINT32("x-multifd-page-count", MigrationState
,
3174 parameters
.x_multifd_page_count
,
3175 DEFAULT_MIGRATE_MULTIFD_PAGE_COUNT
),
3176 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
3177 parameters
.xbzrle_cache_size
,
3178 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
3179 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState
,
3180 parameters
.max_postcopy_bandwidth
,
3181 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH
),
3182 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState
,
3183 parameters
.max_cpu_throttle
,
3184 DEFAULT_MIGRATE_MAX_CPU_THROTTLE
),
3186 /* Migration capabilities */
3187 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
3188 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
3189 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
3190 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
3191 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
3192 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
3193 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
3194 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
3195 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
3196 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
3197 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
3198 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD
),
3200 DEFINE_PROP_END_OF_LIST(),
3203 static void migration_class_init(ObjectClass
*klass
, void *data
)
3205 DeviceClass
*dc
= DEVICE_CLASS(klass
);
3207 dc
->user_creatable
= false;
3208 dc
->props
= migration_properties
;
3211 static void migration_instance_finalize(Object
*obj
)
3213 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3214 MigrationParameters
*params
= &ms
->parameters
;
3216 qemu_mutex_destroy(&ms
->error_mutex
);
3217 qemu_mutex_destroy(&ms
->qemu_file_lock
);
3218 g_free(params
->tls_hostname
);
3219 g_free(params
->tls_creds
);
3220 qemu_sem_destroy(&ms
->rate_limit_sem
);
3221 qemu_sem_destroy(&ms
->pause_sem
);
3222 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
3223 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
3224 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
3225 error_free(ms
->error
);
3228 static void migration_instance_init(Object
*obj
)
3230 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3231 MigrationParameters
*params
= &ms
->parameters
;
3233 ms
->state
= MIGRATION_STATUS_NONE
;
3235 qemu_sem_init(&ms
->pause_sem
, 0);
3236 qemu_mutex_init(&ms
->error_mutex
);
3238 params
->tls_hostname
= g_strdup("");
3239 params
->tls_creds
= g_strdup("");
3241 /* Set has_* up only for parameter checks */
3242 params
->has_compress_level
= true;
3243 params
->has_compress_threads
= true;
3244 params
->has_decompress_threads
= true;
3245 params
->has_cpu_throttle_initial
= true;
3246 params
->has_cpu_throttle_increment
= true;
3247 params
->has_max_bandwidth
= true;
3248 params
->has_downtime_limit
= true;
3249 params
->has_x_checkpoint_delay
= true;
3250 params
->has_block_incremental
= true;
3251 params
->has_x_multifd_channels
= true;
3252 params
->has_x_multifd_page_count
= true;
3253 params
->has_xbzrle_cache_size
= true;
3254 params
->has_max_postcopy_bandwidth
= true;
3255 params
->has_max_cpu_throttle
= true;
3257 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
3258 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
3259 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
3260 qemu_sem_init(&ms
->rate_limit_sem
, 0);
3261 qemu_mutex_init(&ms
->qemu_file_lock
);
3265 * Return true if check pass, false otherwise. Error will be put
3266 * inside errp if provided.
3268 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
3270 MigrationCapabilityStatusList
*head
= NULL
;
3271 /* Assuming all off */
3272 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
3275 if (!migrate_params_check(&ms
->parameters
, errp
)) {
3279 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
3280 if (ms
->enabled_capabilities
[i
]) {
3281 head
= migrate_cap_add(head
, i
, true);
3285 ret
= migrate_caps_check(cap_list
, head
, errp
);
3287 /* It works with head == NULL */
3288 qapi_free_MigrationCapabilityStatusList(head
);
3293 static const TypeInfo migration_type
= {
3294 .name
= TYPE_MIGRATION
,
3296 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3297 * not created using qdev_create(), it is not attached to the qdev
3298 * device tree, and it is never realized.
3300 * TODO: Make this TYPE_OBJECT once QOM provides something like
3301 * TYPE_DEVICE's "-global" properties.
3303 .parent
= TYPE_DEVICE
,
3304 .class_init
= migration_class_init
,
3305 .class_size
= sizeof(MigrationClass
),
3306 .instance_size
= sizeof(MigrationState
),
3307 .instance_init
= migration_instance_init
,
3308 .instance_finalize
= migration_instance_finalize
,
3311 static void register_migration_types(void)
3313 type_register_static(&migration_type
);
3316 type_init(register_migration_types
);