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-common.h"
17 #include "qemu/main-loop.h"
18 #include "migration/migration.h"
19 #include "monitor/monitor.h"
20 #include "migration/qemu-file.h"
21 #include "sysemu/sysemu.h"
22 #include "block/block.h"
23 #include "qemu/sockets.h"
24 #include "migration/block.h"
25 #include "qemu/thread.h"
26 #include "qmp-commands.h"
29 #define MAX_THROTTLE (32 << 20) /* Migration speed throttling */
31 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
33 #define BUFFER_DELAY 100
34 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
36 /* Migration XBZRLE default cache size */
37 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
39 static NotifierList migration_state_notifiers
=
40 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
42 static bool deferred_incoming
;
44 /* When we add fault tolerance, we could have several
45 migrations at once. For now we don't need to add
46 dynamic creation of migration */
48 MigrationState
*migrate_get_current(void)
50 static MigrationState current_migration
= {
51 .state
= MIGRATION_STATUS_NONE
,
52 .bandwidth_limit
= MAX_THROTTLE
,
53 .xbzrle_cache_size
= DEFAULT_MIGRATE_CACHE_SIZE
,
57 return ¤t_migration
;
61 * Called on -incoming with a defer: uri.
62 * The migration can be started later after any parameters have been
65 static void deferred_incoming_migration(Error
**errp
)
67 if (deferred_incoming
) {
68 error_setg(errp
, "Incoming migration already deferred");
70 deferred_incoming
= true;
73 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
77 if (!strcmp(uri
, "defer")) {
78 deferred_incoming_migration(errp
);
79 } else if (strstart(uri
, "tcp:", &p
)) {
80 tcp_start_incoming_migration(p
, errp
);
82 } else if (strstart(uri
, "rdma:", &p
)) {
83 rdma_start_incoming_migration(p
, errp
);
86 } else if (strstart(uri
, "exec:", &p
)) {
87 exec_start_incoming_migration(p
, errp
);
88 } else if (strstart(uri
, "unix:", &p
)) {
89 unix_start_incoming_migration(p
, errp
);
90 } else if (strstart(uri
, "fd:", &p
)) {
91 fd_start_incoming_migration(p
, errp
);
94 error_setg(errp
, "unknown migration protocol: %s", uri
);
98 static void process_incoming_migration_co(void *opaque
)
100 QEMUFile
*f
= opaque
;
101 Error
*local_err
= NULL
;
104 ret
= qemu_loadvm_state(f
);
106 free_xbzrle_decoded_buf();
108 error_report("load of migration failed: %s", strerror(-ret
));
111 qemu_announce_self();
113 /* Make sure all file formats flush their mutable metadata */
114 bdrv_invalidate_cache_all(&local_err
);
116 error_report_err(local_err
);
123 runstate_set(RUN_STATE_PAUSED
);
127 void process_incoming_migration(QEMUFile
*f
)
129 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
);
130 int fd
= qemu_get_fd(f
);
133 qemu_set_nonblock(fd
);
134 qemu_coroutine_enter(co
, f
);
137 /* amount of nanoseconds we are willing to wait for migration to be down.
138 * the choice of nanoseconds is because it is the maximum resolution that
139 * get_clock() can achieve. It is an internal measure. All user-visible
140 * units must be in seconds */
141 static uint64_t max_downtime
= 300000000;
143 uint64_t migrate_max_downtime(void)
148 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
150 MigrationCapabilityStatusList
*head
= NULL
;
151 MigrationCapabilityStatusList
*caps
;
152 MigrationState
*s
= migrate_get_current();
155 caps
= NULL
; /* silence compiler warning */
156 for (i
= 0; i
< MIGRATION_CAPABILITY_MAX
; i
++) {
158 head
= g_malloc0(sizeof(*caps
));
161 caps
->next
= g_malloc0(sizeof(*caps
));
165 g_malloc(sizeof(*caps
->value
));
166 caps
->value
->capability
= i
;
167 caps
->value
->state
= s
->enabled_capabilities
[i
];
173 static void get_xbzrle_cache_stats(MigrationInfo
*info
)
175 if (migrate_use_xbzrle()) {
176 info
->has_xbzrle_cache
= true;
177 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
178 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
179 info
->xbzrle_cache
->bytes
= xbzrle_mig_bytes_transferred();
180 info
->xbzrle_cache
->pages
= xbzrle_mig_pages_transferred();
181 info
->xbzrle_cache
->cache_miss
= xbzrle_mig_pages_cache_miss();
182 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_mig_cache_miss_rate();
183 info
->xbzrle_cache
->overflow
= xbzrle_mig_pages_overflow();
187 MigrationInfo
*qmp_query_migrate(Error
**errp
)
189 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
190 MigrationState
*s
= migrate_get_current();
193 case MIGRATION_STATUS_NONE
:
194 /* no migration has happened ever */
196 case MIGRATION_STATUS_SETUP
:
197 info
->has_status
= true;
198 info
->has_total_time
= false;
200 case MIGRATION_STATUS_ACTIVE
:
201 case MIGRATION_STATUS_CANCELLING
:
202 info
->has_status
= true;
203 info
->has_total_time
= true;
204 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
206 info
->has_expected_downtime
= true;
207 info
->expected_downtime
= s
->expected_downtime
;
208 info
->has_setup_time
= true;
209 info
->setup_time
= s
->setup_time
;
211 info
->has_ram
= true;
212 info
->ram
= g_malloc0(sizeof(*info
->ram
));
213 info
->ram
->transferred
= ram_bytes_transferred();
214 info
->ram
->remaining
= ram_bytes_remaining();
215 info
->ram
->total
= ram_bytes_total();
216 info
->ram
->duplicate
= dup_mig_pages_transferred();
217 info
->ram
->skipped
= skipped_mig_pages_transferred();
218 info
->ram
->normal
= norm_mig_pages_transferred();
219 info
->ram
->normal_bytes
= norm_mig_bytes_transferred();
220 info
->ram
->dirty_pages_rate
= s
->dirty_pages_rate
;
221 info
->ram
->mbps
= s
->mbps
;
222 info
->ram
->dirty_sync_count
= s
->dirty_sync_count
;
224 if (blk_mig_active()) {
225 info
->has_disk
= true;
226 info
->disk
= g_malloc0(sizeof(*info
->disk
));
227 info
->disk
->transferred
= blk_mig_bytes_transferred();
228 info
->disk
->remaining
= blk_mig_bytes_remaining();
229 info
->disk
->total
= blk_mig_bytes_total();
232 get_xbzrle_cache_stats(info
);
234 case MIGRATION_STATUS_COMPLETED
:
235 get_xbzrle_cache_stats(info
);
237 info
->has_status
= true;
238 info
->has_total_time
= true;
239 info
->total_time
= s
->total_time
;
240 info
->has_downtime
= true;
241 info
->downtime
= s
->downtime
;
242 info
->has_setup_time
= true;
243 info
->setup_time
= s
->setup_time
;
245 info
->has_ram
= true;
246 info
->ram
= g_malloc0(sizeof(*info
->ram
));
247 info
->ram
->transferred
= ram_bytes_transferred();
248 info
->ram
->remaining
= 0;
249 info
->ram
->total
= ram_bytes_total();
250 info
->ram
->duplicate
= dup_mig_pages_transferred();
251 info
->ram
->skipped
= skipped_mig_pages_transferred();
252 info
->ram
->normal
= norm_mig_pages_transferred();
253 info
->ram
->normal_bytes
= norm_mig_bytes_transferred();
254 info
->ram
->mbps
= s
->mbps
;
255 info
->ram
->dirty_sync_count
= s
->dirty_sync_count
;
257 case MIGRATION_STATUS_FAILED
:
258 info
->has_status
= true;
260 case MIGRATION_STATUS_CANCELLED
:
261 info
->has_status
= true;
264 info
->status
= s
->state
;
269 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
272 MigrationState
*s
= migrate_get_current();
273 MigrationCapabilityStatusList
*cap
;
275 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
276 s
->state
== MIGRATION_STATUS_SETUP
) {
277 error_set(errp
, QERR_MIGRATION_ACTIVE
);
281 for (cap
= params
; cap
; cap
= cap
->next
) {
282 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
286 /* shared migration helpers */
288 static void migrate_set_state(MigrationState
*s
, int old_state
, int new_state
)
290 if (atomic_cmpxchg(&s
->state
, old_state
, new_state
) == new_state
) {
291 trace_migrate_set_state(new_state
);
295 static void migrate_fd_cleanup(void *opaque
)
297 MigrationState
*s
= opaque
;
299 qemu_bh_delete(s
->cleanup_bh
);
300 s
->cleanup_bh
= NULL
;
303 trace_migrate_fd_cleanup();
304 qemu_mutex_unlock_iothread();
305 qemu_thread_join(&s
->thread
);
306 qemu_mutex_lock_iothread();
308 qemu_fclose(s
->file
);
312 assert(s
->state
!= MIGRATION_STATUS_ACTIVE
);
314 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
315 qemu_savevm_state_cancel();
316 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
317 migrate_set_state(s
, MIGRATION_STATUS_CANCELLING
,
318 MIGRATION_STATUS_CANCELLED
);
322 notifier_list_notify(&migration_state_notifiers
, s
);
325 void migrate_fd_error(MigrationState
*s
)
327 trace_migrate_fd_error();
328 assert(s
->file
== NULL
);
329 s
->state
= MIGRATION_STATUS_FAILED
;
330 trace_migrate_set_state(MIGRATION_STATUS_FAILED
);
331 notifier_list_notify(&migration_state_notifiers
, s
);
334 static void migrate_fd_cancel(MigrationState
*s
)
337 QEMUFile
*f
= migrate_get_current()->file
;
338 trace_migrate_fd_cancel();
341 old_state
= s
->state
;
342 if (old_state
!= MIGRATION_STATUS_SETUP
&&
343 old_state
!= MIGRATION_STATUS_ACTIVE
) {
346 migrate_set_state(s
, old_state
, MIGRATION_STATUS_CANCELLING
);
347 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
350 * If we're unlucky the migration code might be stuck somewhere in a
351 * send/write while the network has failed and is waiting to timeout;
352 * if we've got shutdown(2) available then we can force it to quit.
353 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
354 * called in a bh, so there is no race against this cancel.
356 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
357 qemu_file_shutdown(f
);
361 void add_migration_state_change_notifier(Notifier
*notify
)
363 notifier_list_add(&migration_state_notifiers
, notify
);
366 void remove_migration_state_change_notifier(Notifier
*notify
)
368 notifier_remove(notify
);
371 bool migration_in_setup(MigrationState
*s
)
373 return s
->state
== MIGRATION_STATUS_SETUP
;
376 bool migration_has_finished(MigrationState
*s
)
378 return s
->state
== MIGRATION_STATUS_COMPLETED
;
381 bool migration_has_failed(MigrationState
*s
)
383 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
384 s
->state
== MIGRATION_STATUS_FAILED
);
387 static MigrationState
*migrate_init(const MigrationParams
*params
)
389 MigrationState
*s
= migrate_get_current();
390 int64_t bandwidth_limit
= s
->bandwidth_limit
;
391 bool enabled_capabilities
[MIGRATION_CAPABILITY_MAX
];
392 int64_t xbzrle_cache_size
= s
->xbzrle_cache_size
;
394 memcpy(enabled_capabilities
, s
->enabled_capabilities
,
395 sizeof(enabled_capabilities
));
397 memset(s
, 0, sizeof(*s
));
399 memcpy(s
->enabled_capabilities
, enabled_capabilities
,
400 sizeof(enabled_capabilities
));
401 s
->xbzrle_cache_size
= xbzrle_cache_size
;
403 s
->bandwidth_limit
= bandwidth_limit
;
404 s
->state
= MIGRATION_STATUS_SETUP
;
405 trace_migrate_set_state(MIGRATION_STATUS_SETUP
);
407 s
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
411 static GSList
*migration_blockers
;
413 void migrate_add_blocker(Error
*reason
)
415 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
418 void migrate_del_blocker(Error
*reason
)
420 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
423 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
425 Error
*local_err
= NULL
;
426 static bool once
= true;
428 if (!deferred_incoming
) {
429 error_setg(errp
, "For use with '-incoming defer'");
433 error_setg(errp
, "The incoming migration has already been started");
436 qemu_start_incoming_migration(uri
, &local_err
);
439 error_propagate(errp
, local_err
);
446 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
447 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
450 Error
*local_err
= NULL
;
451 MigrationState
*s
= migrate_get_current();
452 MigrationParams params
;
455 params
.blk
= has_blk
&& blk
;
456 params
.shared
= has_inc
&& inc
;
458 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
459 s
->state
== MIGRATION_STATUS_SETUP
||
460 s
->state
== MIGRATION_STATUS_CANCELLING
) {
461 error_set(errp
, QERR_MIGRATION_ACTIVE
);
465 if (runstate_check(RUN_STATE_INMIGRATE
)) {
466 error_setg(errp
, "Guest is waiting for an incoming migration");
470 if (qemu_savevm_state_blocked(errp
)) {
474 if (migration_blockers
) {
475 *errp
= error_copy(migration_blockers
->data
);
479 s
= migrate_init(¶ms
);
481 if (strstart(uri
, "tcp:", &p
)) {
482 tcp_start_outgoing_migration(s
, p
, &local_err
);
484 } else if (strstart(uri
, "rdma:", &p
)) {
485 rdma_start_outgoing_migration(s
, p
, &local_err
);
488 } else if (strstart(uri
, "exec:", &p
)) {
489 exec_start_outgoing_migration(s
, p
, &local_err
);
490 } else if (strstart(uri
, "unix:", &p
)) {
491 unix_start_outgoing_migration(s
, p
, &local_err
);
492 } else if (strstart(uri
, "fd:", &p
)) {
493 fd_start_outgoing_migration(s
, p
, &local_err
);
496 error_set(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri", "a valid migration protocol");
497 s
->state
= MIGRATION_STATUS_FAILED
;
503 error_propagate(errp
, local_err
);
508 void qmp_migrate_cancel(Error
**errp
)
510 migrate_fd_cancel(migrate_get_current());
513 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
515 MigrationState
*s
= migrate_get_current();
518 /* Check for truncation */
519 if (value
!= (size_t)value
) {
520 error_set(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
521 "exceeding address space");
525 /* Cache should not be larger than guest ram size */
526 if (value
> ram_bytes_total()) {
527 error_set(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
528 "exceeds guest ram size ");
532 new_size
= xbzrle_cache_resize(value
);
534 error_set(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
535 "is smaller than page size");
539 s
->xbzrle_cache_size
= new_size
;
542 int64_t qmp_query_migrate_cache_size(Error
**errp
)
544 return migrate_xbzrle_cache_size();
547 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
554 if (value
> SIZE_MAX
) {
558 s
= migrate_get_current();
559 s
->bandwidth_limit
= value
;
561 qemu_file_set_rate_limit(s
->file
, s
->bandwidth_limit
/ XFER_LIMIT_RATIO
);
565 void qmp_migrate_set_downtime(double value
, Error
**errp
)
568 value
= MAX(0, MIN(UINT64_MAX
, value
));
569 max_downtime
= (uint64_t)value
;
572 bool migrate_auto_converge(void)
576 s
= migrate_get_current();
578 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
581 bool migrate_zero_blocks(void)
585 s
= migrate_get_current();
587 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
590 int migrate_use_xbzrle(void)
594 s
= migrate_get_current();
596 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
599 int64_t migrate_xbzrle_cache_size(void)
603 s
= migrate_get_current();
605 return s
->xbzrle_cache_size
;
608 /* migration thread support */
610 static void *migration_thread(void *opaque
)
612 MigrationState
*s
= opaque
;
613 int64_t initial_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
614 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
615 int64_t initial_bytes
= 0;
616 int64_t max_size
= 0;
617 int64_t start_time
= initial_time
;
618 bool old_vm_running
= false;
620 qemu_savevm_state_begin(s
->file
, &s
->params
);
622 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
623 migrate_set_state(s
, MIGRATION_STATUS_SETUP
, MIGRATION_STATUS_ACTIVE
);
625 while (s
->state
== MIGRATION_STATUS_ACTIVE
) {
626 int64_t current_time
;
627 uint64_t pending_size
;
629 if (!qemu_file_rate_limit(s
->file
)) {
630 pending_size
= qemu_savevm_state_pending(s
->file
, max_size
);
631 trace_migrate_pending(pending_size
, max_size
);
632 if (pending_size
&& pending_size
>= max_size
) {
633 qemu_savevm_state_iterate(s
->file
);
637 qemu_mutex_lock_iothread();
638 start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
639 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
640 old_vm_running
= runstate_is_running();
642 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
644 qemu_file_set_rate_limit(s
->file
, INT64_MAX
);
645 qemu_savevm_state_complete(s
->file
);
647 qemu_mutex_unlock_iothread();
650 migrate_set_state(s
, MIGRATION_STATUS_ACTIVE
,
651 MIGRATION_STATUS_FAILED
);
655 if (!qemu_file_get_error(s
->file
)) {
656 migrate_set_state(s
, MIGRATION_STATUS_ACTIVE
,
657 MIGRATION_STATUS_COMPLETED
);
663 if (qemu_file_get_error(s
->file
)) {
664 migrate_set_state(s
, MIGRATION_STATUS_ACTIVE
,
665 MIGRATION_STATUS_FAILED
);
668 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
669 if (current_time
>= initial_time
+ BUFFER_DELAY
) {
670 uint64_t transferred_bytes
= qemu_ftell(s
->file
) - initial_bytes
;
671 uint64_t time_spent
= current_time
- initial_time
;
672 double bandwidth
= transferred_bytes
/ time_spent
;
673 max_size
= bandwidth
* migrate_max_downtime() / 1000000;
675 s
->mbps
= time_spent
? (((double) transferred_bytes
* 8.0) /
676 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0 : -1;
678 trace_migrate_transferred(transferred_bytes
, time_spent
,
679 bandwidth
, max_size
);
680 /* if we haven't sent anything, we don't want to recalculate
681 10000 is a small enough number for our purposes */
682 if (s
->dirty_bytes_rate
&& transferred_bytes
> 10000) {
683 s
->expected_downtime
= s
->dirty_bytes_rate
/ bandwidth
;
686 qemu_file_reset_rate_limit(s
->file
);
687 initial_time
= current_time
;
688 initial_bytes
= qemu_ftell(s
->file
);
690 if (qemu_file_rate_limit(s
->file
)) {
691 /* usleep expects microseconds */
692 g_usleep((initial_time
+ BUFFER_DELAY
- current_time
)*1000);
696 qemu_mutex_lock_iothread();
697 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
698 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
699 uint64_t transferred_bytes
= qemu_ftell(s
->file
);
700 s
->total_time
= end_time
- s
->total_time
;
701 s
->downtime
= end_time
- start_time
;
703 s
->mbps
= (((double) transferred_bytes
* 8.0) /
704 ((double) s
->total_time
)) / 1000;
706 runstate_set(RUN_STATE_POSTMIGRATE
);
708 if (old_vm_running
) {
712 qemu_bh_schedule(s
->cleanup_bh
);
713 qemu_mutex_unlock_iothread();
718 void migrate_fd_connect(MigrationState
*s
)
720 s
->state
= MIGRATION_STATUS_SETUP
;
721 trace_migrate_set_state(MIGRATION_STATUS_SETUP
);
723 /* This is a best 1st approximation. ns to ms */
724 s
->expected_downtime
= max_downtime
/1000000;
725 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
727 qemu_file_set_rate_limit(s
->file
,
728 s
->bandwidth_limit
/ XFER_LIMIT_RATIO
);
730 /* Notify before starting migration thread */
731 notifier_list_notify(&migration_state_notifiers
, s
);
733 qemu_thread_create(&s
->thread
, "migration", migration_thread
, s
,
734 QEMU_THREAD_JOINABLE
);