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 "migration/migration.h"
18 #include "monitor/monitor.h"
19 #include "migration/qemu-file.h"
20 #include "sysemu/sysemu.h"
21 #include "block/block.h"
22 #include "qemu/sockets.h"
23 #include "migration/block.h"
24 #include "qemu/thread.h"
25 #include "qmp-commands.h"
27 //#define DEBUG_MIGRATION
29 #ifdef DEBUG_MIGRATION
30 #define DPRINTF(fmt, ...) \
31 do { printf("migration: " fmt, ## __VA_ARGS__); } while (0)
33 #define DPRINTF(fmt, ...) \
45 #define MAX_THROTTLE (32 << 20) /* Migration speed throttling */
47 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
49 #define BUFFER_DELAY 100
50 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
52 /* Migration XBZRLE default cache size */
53 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
55 static NotifierList migration_state_notifiers
=
56 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
58 /* When we add fault tolerance, we could have several
59 migrations at once. For now we don't need to add
60 dynamic creation of migration */
62 MigrationState
*migrate_get_current(void)
64 static MigrationState current_migration
= {
65 .state
= MIG_STATE_SETUP
,
66 .bandwidth_limit
= MAX_THROTTLE
,
67 .xbzrle_cache_size
= DEFAULT_MIGRATE_CACHE_SIZE
,
70 return ¤t_migration
;
73 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
77 if (strstart(uri
, "tcp:", &p
))
78 tcp_start_incoming_migration(p
, errp
);
80 else if (strstart(uri
, "exec:", &p
))
81 exec_start_incoming_migration(p
, errp
);
82 else if (strstart(uri
, "unix:", &p
))
83 unix_start_incoming_migration(p
, errp
);
84 else if (strstart(uri
, "fd:", &p
))
85 fd_start_incoming_migration(p
, errp
);
88 error_setg(errp
, "unknown migration protocol: %s\n", uri
);
92 static void process_incoming_migration_co(void *opaque
)
97 ret
= qemu_loadvm_state(f
);
98 qemu_set_fd_handler(qemu_get_fd(f
), NULL
, NULL
, NULL
);
101 fprintf(stderr
, "load of migration failed\n");
104 qemu_announce_self();
105 DPRINTF("successfully loaded vm state\n");
107 bdrv_clear_incoming_migration_all();
108 /* Make sure all file formats flush their mutable metadata */
109 bdrv_invalidate_cache_all();
114 runstate_set(RUN_STATE_PAUSED
);
118 static void enter_migration_coroutine(void *opaque
)
120 Coroutine
*co
= opaque
;
121 qemu_coroutine_enter(co
, NULL
);
124 void process_incoming_migration(QEMUFile
*f
)
126 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
);
127 int fd
= qemu_get_fd(f
);
130 socket_set_nonblock(fd
);
131 qemu_set_fd_handler(fd
, enter_migration_coroutine
, NULL
, co
);
132 qemu_coroutine_enter(co
, f
);
135 /* amount of nanoseconds we are willing to wait for migration to be down.
136 * the choice of nanoseconds is because it is the maximum resolution that
137 * get_clock() can achieve. It is an internal measure. All user-visible
138 * units must be in seconds */
139 static uint64_t max_downtime
= 30000000;
141 uint64_t migrate_max_downtime(void)
146 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
148 MigrationCapabilityStatusList
*head
= NULL
;
149 MigrationCapabilityStatusList
*caps
;
150 MigrationState
*s
= migrate_get_current();
153 for (i
= 0; i
< MIGRATION_CAPABILITY_MAX
; i
++) {
155 head
= g_malloc0(sizeof(*caps
));
158 caps
->next
= g_malloc0(sizeof(*caps
));
162 g_malloc(sizeof(*caps
->value
));
163 caps
->value
->capability
= i
;
164 caps
->value
->state
= s
->enabled_capabilities
[i
];
170 static void get_xbzrle_cache_stats(MigrationInfo
*info
)
172 if (migrate_use_xbzrle()) {
173 info
->has_xbzrle_cache
= true;
174 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
175 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
176 info
->xbzrle_cache
->bytes
= xbzrle_mig_bytes_transferred();
177 info
->xbzrle_cache
->pages
= xbzrle_mig_pages_transferred();
178 info
->xbzrle_cache
->cache_miss
= xbzrle_mig_pages_cache_miss();
179 info
->xbzrle_cache
->overflow
= xbzrle_mig_pages_overflow();
183 MigrationInfo
*qmp_query_migrate(Error
**errp
)
185 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
186 MigrationState
*s
= migrate_get_current();
189 case MIG_STATE_SETUP
:
190 /* no migration has happened ever */
192 case MIG_STATE_ACTIVE
:
193 info
->has_status
= true;
194 info
->status
= g_strdup("active");
195 info
->has_total_time
= true;
196 info
->total_time
= qemu_get_clock_ms(rt_clock
)
198 info
->has_expected_downtime
= true;
199 info
->expected_downtime
= s
->expected_downtime
;
201 info
->has_ram
= true;
202 info
->ram
= g_malloc0(sizeof(*info
->ram
));
203 info
->ram
->transferred
= ram_bytes_transferred();
204 info
->ram
->remaining
= ram_bytes_remaining();
205 info
->ram
->total
= ram_bytes_total();
206 info
->ram
->duplicate
= dup_mig_pages_transferred();
207 info
->ram
->normal
= norm_mig_pages_transferred();
208 info
->ram
->normal_bytes
= norm_mig_bytes_transferred();
209 info
->ram
->dirty_pages_rate
= s
->dirty_pages_rate
;
212 if (blk_mig_active()) {
213 info
->has_disk
= true;
214 info
->disk
= g_malloc0(sizeof(*info
->disk
));
215 info
->disk
->transferred
= blk_mig_bytes_transferred();
216 info
->disk
->remaining
= blk_mig_bytes_remaining();
217 info
->disk
->total
= blk_mig_bytes_total();
220 get_xbzrle_cache_stats(info
);
222 case MIG_STATE_COMPLETED
:
223 get_xbzrle_cache_stats(info
);
225 info
->has_status
= true;
226 info
->status
= g_strdup("completed");
227 info
->total_time
= s
->total_time
;
228 info
->has_downtime
= true;
229 info
->downtime
= s
->downtime
;
231 info
->has_ram
= true;
232 info
->ram
= g_malloc0(sizeof(*info
->ram
));
233 info
->ram
->transferred
= ram_bytes_transferred();
234 info
->ram
->remaining
= 0;
235 info
->ram
->total
= ram_bytes_total();
236 info
->ram
->duplicate
= dup_mig_pages_transferred();
237 info
->ram
->normal
= norm_mig_pages_transferred();
238 info
->ram
->normal_bytes
= norm_mig_bytes_transferred();
240 case MIG_STATE_ERROR
:
241 info
->has_status
= true;
242 info
->status
= g_strdup("failed");
244 case MIG_STATE_CANCELLED
:
245 info
->has_status
= true;
246 info
->status
= g_strdup("cancelled");
253 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
256 MigrationState
*s
= migrate_get_current();
257 MigrationCapabilityStatusList
*cap
;
259 if (s
->state
== MIG_STATE_ACTIVE
) {
260 error_set(errp
, QERR_MIGRATION_ACTIVE
);
264 for (cap
= params
; cap
; cap
= cap
->next
) {
265 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
269 /* shared migration helpers */
271 static int migrate_fd_cleanup(MigrationState
*s
)
276 DPRINTF("closing file\n");
277 ret
= qemu_fclose(s
->file
);
285 void migrate_fd_error(MigrationState
*s
)
287 DPRINTF("setting error state\n");
288 s
->state
= MIG_STATE_ERROR
;
289 notifier_list_notify(&migration_state_notifiers
, s
);
290 migrate_fd_cleanup(s
);
293 static void migrate_fd_completed(MigrationState
*s
)
295 DPRINTF("setting completed state\n");
296 if (migrate_fd_cleanup(s
) < 0) {
297 s
->state
= MIG_STATE_ERROR
;
299 s
->state
= MIG_STATE_COMPLETED
;
300 runstate_set(RUN_STATE_POSTMIGRATE
);
302 notifier_list_notify(&migration_state_notifiers
, s
);
305 ssize_t
migrate_fd_put_buffer(MigrationState
*s
, const void *data
,
310 if (s
->state
!= MIG_STATE_ACTIVE
) {
315 ret
= s
->write(s
, data
, size
);
316 } while (ret
== -1 && ((s
->get_error(s
)) == EINTR
));
319 ret
= -(s
->get_error(s
));
324 static void migrate_fd_cancel(MigrationState
*s
)
326 if (s
->state
!= MIG_STATE_ACTIVE
)
329 DPRINTF("cancelling migration\n");
331 s
->state
= MIG_STATE_CANCELLED
;
332 notifier_list_notify(&migration_state_notifiers
, s
);
333 qemu_savevm_state_cancel(s
->file
);
335 migrate_fd_cleanup(s
);
338 int migrate_fd_close(MigrationState
*s
)
348 void add_migration_state_change_notifier(Notifier
*notify
)
350 notifier_list_add(&migration_state_notifiers
, notify
);
353 void remove_migration_state_change_notifier(Notifier
*notify
)
355 notifier_remove(notify
);
358 bool migration_is_active(MigrationState
*s
)
360 return s
->state
== MIG_STATE_ACTIVE
;
363 bool migration_has_finished(MigrationState
*s
)
365 return s
->state
== MIG_STATE_COMPLETED
;
368 bool migration_has_failed(MigrationState
*s
)
370 return (s
->state
== MIG_STATE_CANCELLED
||
371 s
->state
== MIG_STATE_ERROR
);
374 static MigrationState
*migrate_init(const MigrationParams
*params
)
376 MigrationState
*s
= migrate_get_current();
377 int64_t bandwidth_limit
= s
->bandwidth_limit
;
378 bool enabled_capabilities
[MIGRATION_CAPABILITY_MAX
];
379 int64_t xbzrle_cache_size
= s
->xbzrle_cache_size
;
381 memcpy(enabled_capabilities
, s
->enabled_capabilities
,
382 sizeof(enabled_capabilities
));
384 memset(s
, 0, sizeof(*s
));
385 s
->bandwidth_limit
= bandwidth_limit
;
387 memcpy(s
->enabled_capabilities
, enabled_capabilities
,
388 sizeof(enabled_capabilities
));
389 s
->xbzrle_cache_size
= xbzrle_cache_size
;
391 s
->bandwidth_limit
= bandwidth_limit
;
392 s
->state
= MIG_STATE_SETUP
;
393 s
->total_time
= qemu_get_clock_ms(rt_clock
);
398 static GSList
*migration_blockers
;
400 void migrate_add_blocker(Error
*reason
)
402 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
405 void migrate_del_blocker(Error
*reason
)
407 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
410 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
411 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
414 Error
*local_err
= NULL
;
415 MigrationState
*s
= migrate_get_current();
416 MigrationParams params
;
422 if (s
->state
== MIG_STATE_ACTIVE
) {
423 error_set(errp
, QERR_MIGRATION_ACTIVE
);
427 if (qemu_savevm_state_blocked(errp
)) {
431 if (migration_blockers
) {
432 *errp
= error_copy(migration_blockers
->data
);
436 s
= migrate_init(¶ms
);
438 if (strstart(uri
, "tcp:", &p
)) {
439 tcp_start_outgoing_migration(s
, p
, &local_err
);
441 } else if (strstart(uri
, "exec:", &p
)) {
442 exec_start_outgoing_migration(s
, p
, &local_err
);
443 } else if (strstart(uri
, "unix:", &p
)) {
444 unix_start_outgoing_migration(s
, p
, &local_err
);
445 } else if (strstart(uri
, "fd:", &p
)) {
446 fd_start_outgoing_migration(s
, p
, &local_err
);
449 error_set(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri", "a valid migration protocol");
455 error_propagate(errp
, local_err
);
460 void qmp_migrate_cancel(Error
**errp
)
462 migrate_fd_cancel(migrate_get_current());
465 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
467 MigrationState
*s
= migrate_get_current();
469 /* Check for truncation */
470 if (value
!= (size_t)value
) {
471 error_set(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
472 "exceeding address space");
476 s
->xbzrle_cache_size
= xbzrle_cache_resize(value
);
479 int64_t qmp_query_migrate_cache_size(Error
**errp
)
481 return migrate_xbzrle_cache_size();
484 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
492 s
= migrate_get_current();
493 s
->bandwidth_limit
= value
;
494 qemu_file_set_rate_limit(s
->file
, s
->bandwidth_limit
);
497 void qmp_migrate_set_downtime(double value
, Error
**errp
)
500 value
= MAX(0, MIN(UINT64_MAX
, value
));
501 max_downtime
= (uint64_t)value
;
504 int migrate_use_xbzrle(void)
508 s
= migrate_get_current();
510 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
513 int64_t migrate_xbzrle_cache_size(void)
517 s
= migrate_get_current();
519 return s
->xbzrle_cache_size
;
522 /* migration thread support */
525 static ssize_t
buffered_flush(MigrationState
*s
)
530 DPRINTF("flushing %zu byte(s) of data\n", s
->buffer_size
);
532 while (s
->bytes_xfer
< s
->xfer_limit
&& offset
< s
->buffer_size
) {
533 size_t to_send
= MIN(s
->buffer_size
- offset
, s
->xfer_limit
- s
->bytes_xfer
);
534 ret
= migrate_fd_put_buffer(s
, s
->buffer
+ offset
, to_send
);
536 DPRINTF("error flushing data, %zd\n", ret
);
539 DPRINTF("flushed %zd byte(s)\n", ret
);
541 s
->bytes_xfer
+= ret
;
545 DPRINTF("flushed %zu of %zu byte(s)\n", offset
, s
->buffer_size
);
546 memmove(s
->buffer
, s
->buffer
+ offset
, s
->buffer_size
- offset
);
547 s
->buffer_size
-= offset
;
555 static int buffered_put_buffer(void *opaque
, const uint8_t *buf
,
556 int64_t pos
, int size
)
558 MigrationState
*s
= opaque
;
561 DPRINTF("putting %d bytes at %" PRId64
"\n", size
, pos
);
563 error
= qemu_file_get_error(s
->file
);
565 DPRINTF("flush when error, bailing: %s\n", strerror(-error
));
573 if (size
> (s
->buffer_capacity
- s
->buffer_size
)) {
574 DPRINTF("increasing buffer capacity from %zu by %zu\n",
575 s
->buffer_capacity
, size
+ 1024);
577 s
->buffer_capacity
+= size
+ 1024;
579 s
->buffer
= g_realloc(s
->buffer
, s
->buffer_capacity
);
582 memcpy(s
->buffer
+ s
->buffer_size
, buf
, size
);
583 s
->buffer_size
+= size
;
588 static int buffered_close(void *opaque
)
590 MigrationState
*s
= opaque
;
594 DPRINTF("closing\n");
596 s
->xfer_limit
= INT_MAX
;
597 while (!qemu_file_get_error(s
->file
) && s
->buffer_size
) {
598 ret
= buffered_flush(s
);
604 ret2
= migrate_fd_close(s
);
608 ret
= migrate_fd_close(s
);
613 static int buffered_get_fd(void *opaque
)
615 MigrationState
*s
= opaque
;
621 * The meaning of the return values is:
622 * 0: We can continue sending
624 * negative: There has been an error
626 static int buffered_rate_limit(void *opaque
)
628 MigrationState
*s
= opaque
;
631 ret
= qemu_file_get_error(s
->file
);
636 if (s
->bytes_xfer
> s
->xfer_limit
) {
643 static int64_t buffered_set_rate_limit(void *opaque
, int64_t new_rate
)
645 MigrationState
*s
= opaque
;
646 if (qemu_file_get_error(s
->file
)) {
649 if (new_rate
> SIZE_MAX
) {
653 s
->xfer_limit
= new_rate
/ 10;
656 return s
->xfer_limit
;
659 static int64_t buffered_get_rate_limit(void *opaque
)
661 MigrationState
*s
= opaque
;
663 return s
->xfer_limit
;
666 static bool migrate_fd_put_ready(MigrationState
*s
, uint64_t max_size
)
669 uint64_t pending_size
;
670 bool last_round
= false;
672 qemu_mutex_lock_iothread();
673 if (s
->state
!= MIG_STATE_ACTIVE
) {
674 DPRINTF("put_ready returning because of non-active state\n");
675 qemu_mutex_unlock_iothread();
679 s
->first_time
= false;
680 DPRINTF("beginning savevm\n");
681 ret
= qemu_savevm_state_begin(s
->file
, &s
->params
);
683 DPRINTF("failed, %d\n", ret
);
685 qemu_mutex_unlock_iothread();
690 DPRINTF("iterate\n");
691 pending_size
= qemu_savevm_state_pending(s
->file
, max_size
);
692 DPRINTF("pending size %lu max %lu\n", pending_size
, max_size
);
693 if (pending_size
>= max_size
) {
694 ret
= qemu_savevm_state_iterate(s
->file
);
699 int old_vm_running
= runstate_is_running();
700 int64_t start_time
, end_time
;
702 DPRINTF("done iterating\n");
703 start_time
= qemu_get_clock_ms(rt_clock
);
704 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
705 if (old_vm_running
) {
706 vm_stop(RUN_STATE_FINISH_MIGRATE
);
708 vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
711 if (qemu_savevm_state_complete(s
->file
) < 0) {
714 migrate_fd_completed(s
);
716 end_time
= qemu_get_clock_ms(rt_clock
);
717 s
->total_time
= end_time
- s
->total_time
;
718 s
->downtime
= end_time
- start_time
;
719 if (s
->state
!= MIG_STATE_COMPLETED
) {
720 if (old_vm_running
) {
726 qemu_mutex_unlock_iothread();
731 static void *buffered_file_thread(void *opaque
)
733 MigrationState
*s
= opaque
;
734 int64_t initial_time
= qemu_get_clock_ms(rt_clock
);
735 int64_t max_size
= 0;
736 bool last_round
= false;
739 int64_t current_time
= qemu_get_clock_ms(rt_clock
);
744 if (current_time
>= initial_time
+ BUFFER_DELAY
) {
745 uint64_t transferred_bytes
= s
->bytes_xfer
;
746 uint64_t time_spent
= current_time
- initial_time
;
747 double bandwidth
= transferred_bytes
/ time_spent
;
748 max_size
= bandwidth
* migrate_max_downtime() / 1000000;
750 DPRINTF("transferred %" PRIu64
" time_spent %" PRIu64
751 " bandwidth %g max_size %" PRId64
"\n",
752 transferred_bytes
, time_spent
, bandwidth
, max_size
);
755 initial_time
= current_time
;
757 if (!last_round
&& (s
->bytes_xfer
>= s
->xfer_limit
)) {
758 /* usleep expects microseconds */
759 g_usleep((initial_time
+ BUFFER_DELAY
- current_time
)*1000);
761 if (buffered_flush(s
) < 0) {
765 DPRINTF("file is ready\n");
766 if (s
->bytes_xfer
< s
->xfer_limit
) {
767 DPRINTF("notifying client\n");
768 last_round
= migrate_fd_put_ready(s
, max_size
);
776 static const QEMUFileOps buffered_file_ops
= {
777 .get_fd
= buffered_get_fd
,
778 .put_buffer
= buffered_put_buffer
,
779 .close
= buffered_close
,
780 .rate_limit
= buffered_rate_limit
,
781 .get_rate_limit
= buffered_get_rate_limit
,
782 .set_rate_limit
= buffered_set_rate_limit
,
785 void migrate_fd_connect(MigrationState
*s
)
787 s
->state
= MIG_STATE_ACTIVE
;
791 s
->buffer_capacity
= 0;
793 s
->first_time
= true;
795 s
->xfer_limit
= s
->bandwidth_limit
/ XFER_LIMIT_RATIO
;
798 s
->file
= qemu_fopen_ops(s
, &buffered_file_ops
);
800 qemu_thread_create(&s
->thread
, buffered_file_thread
, s
,
801 QEMU_THREAD_DETACHED
);
802 notifier_list_notify(&migration_state_notifiers
, s
);