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 DEBUG_MIGRATION
31 #ifdef DEBUG_MIGRATION
32 #define DPRINTF(fmt, ...) \
33 do { printf("migration: " fmt, ## __VA_ARGS__); } while (0)
35 #define DPRINTF(fmt, ...) \
49 #define MAX_THROTTLE (32 << 20) /* Migration 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 /* Migration XBZRLE default cache size */
57 #define DEFAULT_MIGRATE_CACHE_SIZE (64 * 1024 * 1024)
59 static NotifierList migration_state_notifiers
=
60 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
62 /* When we add fault tolerance, we could have several
63 migrations at once. For now we don't need to add
64 dynamic creation of migration */
66 MigrationState
*migrate_get_current(void)
68 static MigrationState current_migration
= {
69 .state
= MIG_STATE_NONE
,
70 .bandwidth_limit
= MAX_THROTTLE
,
71 .xbzrle_cache_size
= DEFAULT_MIGRATE_CACHE_SIZE
,
75 return ¤t_migration
;
78 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
82 if (strstart(uri
, "tcp:", &p
))
83 tcp_start_incoming_migration(p
, errp
);
85 else if (strstart(uri
, "x-rdma:", &p
))
86 rdma_start_incoming_migration(p
, errp
);
89 else if (strstart(uri
, "exec:", &p
))
90 exec_start_incoming_migration(p
, errp
);
91 else if (strstart(uri
, "unix:", &p
))
92 unix_start_incoming_migration(p
, errp
);
93 else if (strstart(uri
, "fd:", &p
))
94 fd_start_incoming_migration(p
, errp
);
97 error_setg(errp
, "unknown migration protocol: %s", uri
);
101 static void process_incoming_migration_co(void *opaque
)
103 QEMUFile
*f
= opaque
;
106 ret
= qemu_loadvm_state(f
);
109 fprintf(stderr
, "load of migration failed\n");
112 qemu_announce_self();
113 DPRINTF("successfully loaded vm state\n");
115 bdrv_clear_incoming_migration_all();
116 /* Make sure all file formats flush their mutable metadata */
117 bdrv_invalidate_cache_all();
122 runstate_set(RUN_STATE_PAUSED
);
126 void process_incoming_migration(QEMUFile
*f
)
128 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
);
129 int fd
= qemu_get_fd(f
);
132 qemu_set_nonblock(fd
);
133 qemu_coroutine_enter(co
, f
);
136 /* amount of nanoseconds we are willing to wait for migration to be down.
137 * the choice of nanoseconds is because it is the maximum resolution that
138 * get_clock() can achieve. It is an internal measure. All user-visible
139 * units must be in seconds */
140 static uint64_t max_downtime
= 30000000;
142 uint64_t migrate_max_downtime(void)
147 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
149 MigrationCapabilityStatusList
*head
= NULL
;
150 MigrationCapabilityStatusList
*caps
;
151 MigrationState
*s
= migrate_get_current();
154 caps
= NULL
; /* silence compiler warning */
155 for (i
= 0; i
< MIGRATION_CAPABILITY_MAX
; i
++) {
157 head
= g_malloc0(sizeof(*caps
));
160 caps
->next
= g_malloc0(sizeof(*caps
));
164 g_malloc(sizeof(*caps
->value
));
165 caps
->value
->capability
= i
;
166 caps
->value
->state
= s
->enabled_capabilities
[i
];
172 static void get_xbzrle_cache_stats(MigrationInfo
*info
)
174 if (migrate_use_xbzrle()) {
175 info
->has_xbzrle_cache
= true;
176 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
177 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
178 info
->xbzrle_cache
->bytes
= xbzrle_mig_bytes_transferred();
179 info
->xbzrle_cache
->pages
= xbzrle_mig_pages_transferred();
180 info
->xbzrle_cache
->cache_miss
= xbzrle_mig_pages_cache_miss();
181 info
->xbzrle_cache
->overflow
= xbzrle_mig_pages_overflow();
185 MigrationInfo
*qmp_query_migrate(Error
**errp
)
187 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
188 MigrationState
*s
= migrate_get_current();
192 /* no migration has happened ever */
194 case MIG_STATE_SETUP
:
195 info
->has_status
= true;
196 info
->status
= g_strdup("setup");
197 info
->has_total_time
= false;
199 case MIG_STATE_ACTIVE
:
200 case MIG_STATE_CANCELLING
:
201 info
->has_status
= true;
202 info
->status
= g_strdup("active");
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
;
223 if (blk_mig_active()) {
224 info
->has_disk
= true;
225 info
->disk
= g_malloc0(sizeof(*info
->disk
));
226 info
->disk
->transferred
= blk_mig_bytes_transferred();
227 info
->disk
->remaining
= blk_mig_bytes_remaining();
228 info
->disk
->total
= blk_mig_bytes_total();
231 get_xbzrle_cache_stats(info
);
233 case MIG_STATE_COMPLETED
:
234 get_xbzrle_cache_stats(info
);
236 info
->has_status
= true;
237 info
->status
= g_strdup("completed");
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
;
256 case MIG_STATE_ERROR
:
257 info
->has_status
= true;
258 info
->status
= g_strdup("failed");
260 case MIG_STATE_CANCELLED
:
261 info
->has_status
= true;
262 info
->status
= g_strdup("cancelled");
269 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
272 MigrationState
*s
= migrate_get_current();
273 MigrationCapabilityStatusList
*cap
;
275 if (s
->state
== MIG_STATE_ACTIVE
|| s
->state
== MIG_STATE_SETUP
) {
276 error_set(errp
, QERR_MIGRATION_ACTIVE
);
280 for (cap
= params
; cap
; cap
= cap
->next
) {
281 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
285 /* shared migration helpers */
287 static void migrate_set_state(MigrationState
*s
, int old_state
, int new_state
)
289 if (atomic_cmpxchg(&s
->state
, old_state
, new_state
) == new_state
) {
290 trace_migrate_set_state(new_state
);
294 static void migrate_fd_cleanup(void *opaque
)
296 MigrationState
*s
= opaque
;
298 qemu_bh_delete(s
->cleanup_bh
);
299 s
->cleanup_bh
= NULL
;
302 DPRINTF("closing file\n");
303 qemu_mutex_unlock_iothread();
304 qemu_thread_join(&s
->thread
);
305 qemu_mutex_lock_iothread();
307 qemu_fclose(s
->file
);
311 assert(s
->state
!= MIG_STATE_ACTIVE
);
313 if (s
->state
!= MIG_STATE_COMPLETED
) {
314 qemu_savevm_state_cancel();
315 if (s
->state
== MIG_STATE_CANCELLING
) {
316 migrate_set_state(s
, MIG_STATE_CANCELLING
, MIG_STATE_CANCELLED
);
320 notifier_list_notify(&migration_state_notifiers
, s
);
323 void migrate_fd_error(MigrationState
*s
)
325 DPRINTF("setting error state\n");
326 assert(s
->file
== NULL
);
327 s
->state
= MIG_STATE_ERROR
;
328 trace_migrate_set_state(MIG_STATE_ERROR
);
329 notifier_list_notify(&migration_state_notifiers
, s
);
332 static void migrate_fd_cancel(MigrationState
*s
)
335 DPRINTF("cancelling migration\n");
338 old_state
= s
->state
;
339 if (old_state
!= MIG_STATE_SETUP
&& old_state
!= MIG_STATE_ACTIVE
) {
342 migrate_set_state(s
, old_state
, MIG_STATE_CANCELLING
);
343 } while (s
->state
!= MIG_STATE_CANCELLING
);
346 void add_migration_state_change_notifier(Notifier
*notify
)
348 notifier_list_add(&migration_state_notifiers
, notify
);
351 void remove_migration_state_change_notifier(Notifier
*notify
)
353 notifier_remove(notify
);
356 bool migration_in_setup(MigrationState
*s
)
358 return s
->state
== MIG_STATE_SETUP
;
361 bool migration_has_finished(MigrationState
*s
)
363 return s
->state
== MIG_STATE_COMPLETED
;
366 bool migration_has_failed(MigrationState
*s
)
368 return (s
->state
== MIG_STATE_CANCELLED
||
369 s
->state
== MIG_STATE_ERROR
);
372 static MigrationState
*migrate_init(const MigrationParams
*params
)
374 MigrationState
*s
= migrate_get_current();
375 int64_t bandwidth_limit
= s
->bandwidth_limit
;
376 bool enabled_capabilities
[MIGRATION_CAPABILITY_MAX
];
377 int64_t xbzrle_cache_size
= s
->xbzrle_cache_size
;
379 memcpy(enabled_capabilities
, s
->enabled_capabilities
,
380 sizeof(enabled_capabilities
));
382 memset(s
, 0, sizeof(*s
));
384 memcpy(s
->enabled_capabilities
, enabled_capabilities
,
385 sizeof(enabled_capabilities
));
386 s
->xbzrle_cache_size
= xbzrle_cache_size
;
388 s
->bandwidth_limit
= bandwidth_limit
;
389 s
->state
= MIG_STATE_SETUP
;
390 trace_migrate_set_state(MIG_STATE_SETUP
);
392 s
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
396 static GSList
*migration_blockers
;
398 void migrate_add_blocker(Error
*reason
)
400 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
403 void migrate_del_blocker(Error
*reason
)
405 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
408 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
409 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
412 Error
*local_err
= NULL
;
413 MigrationState
*s
= migrate_get_current();
414 MigrationParams params
;
417 params
.blk
= has_blk
&& blk
;
418 params
.shared
= has_inc
&& inc
;
420 if (s
->state
== MIG_STATE_ACTIVE
|| s
->state
== MIG_STATE_SETUP
||
421 s
->state
== MIG_STATE_CANCELLING
) {
422 error_set(errp
, QERR_MIGRATION_ACTIVE
);
426 if (qemu_savevm_state_blocked(errp
)) {
430 if (migration_blockers
) {
431 *errp
= error_copy(migration_blockers
->data
);
435 s
= migrate_init(¶ms
);
437 if (strstart(uri
, "tcp:", &p
)) {
438 tcp_start_outgoing_migration(s
, p
, &local_err
);
440 } else if (strstart(uri
, "x-rdma:", &p
)) {
441 rdma_start_outgoing_migration(s
, p
, &local_err
);
444 } else if (strstart(uri
, "exec:", &p
)) {
445 exec_start_outgoing_migration(s
, p
, &local_err
);
446 } else if (strstart(uri
, "unix:", &p
)) {
447 unix_start_outgoing_migration(s
, p
, &local_err
);
448 } else if (strstart(uri
, "fd:", &p
)) {
449 fd_start_outgoing_migration(s
, p
, &local_err
);
452 error_set(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri", "a valid migration protocol");
453 s
->state
= MIG_STATE_ERROR
;
459 error_propagate(errp
, local_err
);
464 void qmp_migrate_cancel(Error
**errp
)
466 migrate_fd_cancel(migrate_get_current());
469 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
471 MigrationState
*s
= migrate_get_current();
473 /* Check for truncation */
474 if (value
!= (size_t)value
) {
475 error_set(errp
, QERR_INVALID_PARAMETER_VALUE
, "cache size",
476 "exceeding address space");
480 s
->xbzrle_cache_size
= xbzrle_cache_resize(value
);
483 int64_t qmp_query_migrate_cache_size(Error
**errp
)
485 return migrate_xbzrle_cache_size();
488 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
495 if (value
> SIZE_MAX
) {
499 s
= migrate_get_current();
500 s
->bandwidth_limit
= value
;
502 qemu_file_set_rate_limit(s
->file
, s
->bandwidth_limit
/ XFER_LIMIT_RATIO
);
506 void qmp_migrate_set_downtime(double value
, Error
**errp
)
509 value
= MAX(0, MIN(UINT64_MAX
, value
));
510 max_downtime
= (uint64_t)value
;
513 bool migrate_rdma_pin_all(void)
517 s
= migrate_get_current();
519 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_RDMA_PIN_ALL
];
522 bool migrate_auto_converge(void)
526 s
= migrate_get_current();
528 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
531 bool migrate_zero_blocks(void)
535 s
= migrate_get_current();
537 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
540 int migrate_use_xbzrle(void)
544 s
= migrate_get_current();
546 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
549 int64_t migrate_xbzrle_cache_size(void)
553 s
= migrate_get_current();
555 return s
->xbzrle_cache_size
;
558 /* migration thread support */
560 static void *migration_thread(void *opaque
)
562 MigrationState
*s
= opaque
;
563 int64_t initial_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
564 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
565 int64_t initial_bytes
= 0;
566 int64_t max_size
= 0;
567 int64_t start_time
= initial_time
;
568 bool old_vm_running
= false;
570 DPRINTF("beginning savevm\n");
571 qemu_savevm_state_begin(s
->file
, &s
->params
);
573 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
574 migrate_set_state(s
, MIG_STATE_SETUP
, MIG_STATE_ACTIVE
);
576 DPRINTF("setup complete\n");
578 while (s
->state
== MIG_STATE_ACTIVE
) {
579 int64_t current_time
;
580 uint64_t pending_size
;
582 if (!qemu_file_rate_limit(s
->file
)) {
583 DPRINTF("iterate\n");
584 pending_size
= qemu_savevm_state_pending(s
->file
, max_size
);
585 DPRINTF("pending size %" PRIu64
" max %" PRIu64
"\n",
586 pending_size
, max_size
);
587 if (pending_size
&& pending_size
>= max_size
) {
588 qemu_savevm_state_iterate(s
->file
);
592 DPRINTF("done iterating\n");
593 qemu_mutex_lock_iothread();
594 start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
595 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
);
596 old_vm_running
= runstate_is_running();
598 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
600 qemu_file_set_rate_limit(s
->file
, INT64_MAX
);
601 qemu_savevm_state_complete(s
->file
);
603 qemu_mutex_unlock_iothread();
606 migrate_set_state(s
, MIG_STATE_ACTIVE
, MIG_STATE_ERROR
);
610 if (!qemu_file_get_error(s
->file
)) {
611 migrate_set_state(s
, MIG_STATE_ACTIVE
, MIG_STATE_COMPLETED
);
617 if (qemu_file_get_error(s
->file
)) {
618 migrate_set_state(s
, MIG_STATE_ACTIVE
, MIG_STATE_ERROR
);
621 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
622 if (current_time
>= initial_time
+ BUFFER_DELAY
) {
623 uint64_t transferred_bytes
= qemu_ftell(s
->file
) - initial_bytes
;
624 uint64_t time_spent
= current_time
- initial_time
;
625 double bandwidth
= transferred_bytes
/ time_spent
;
626 max_size
= bandwidth
* migrate_max_downtime() / 1000000;
628 s
->mbps
= time_spent
? (((double) transferred_bytes
* 8.0) /
629 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0 : -1;
631 DPRINTF("transferred %" PRIu64
" time_spent %" PRIu64
632 " bandwidth %g max_size %" PRId64
"\n",
633 transferred_bytes
, time_spent
, bandwidth
, max_size
);
634 /* if we haven't sent anything, we don't want to recalculate
635 10000 is a small enough number for our purposes */
636 if (s
->dirty_bytes_rate
&& transferred_bytes
> 10000) {
637 s
->expected_downtime
= s
->dirty_bytes_rate
/ bandwidth
;
640 qemu_file_reset_rate_limit(s
->file
);
641 initial_time
= current_time
;
642 initial_bytes
= qemu_ftell(s
->file
);
644 if (qemu_file_rate_limit(s
->file
)) {
645 /* usleep expects microseconds */
646 g_usleep((initial_time
+ BUFFER_DELAY
- current_time
)*1000);
650 qemu_mutex_lock_iothread();
651 if (s
->state
== MIG_STATE_COMPLETED
) {
652 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
653 s
->total_time
= end_time
- s
->total_time
;
654 s
->downtime
= end_time
- start_time
;
655 runstate_set(RUN_STATE_POSTMIGRATE
);
657 if (old_vm_running
) {
661 qemu_bh_schedule(s
->cleanup_bh
);
662 qemu_mutex_unlock_iothread();
667 void migrate_fd_connect(MigrationState
*s
)
669 s
->state
= MIG_STATE_SETUP
;
670 trace_migrate_set_state(MIG_STATE_SETUP
);
672 /* This is a best 1st approximation. ns to ms */
673 s
->expected_downtime
= max_downtime
/1000000;
674 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
676 qemu_file_set_rate_limit(s
->file
,
677 s
->bandwidth_limit
/ XFER_LIMIT_RATIO
);
679 /* Notify before starting migration thread */
680 notifier_list_notify(&migration_state_notifiers
, s
);
682 qemu_thread_create(&s
->thread
, migration_thread
, s
,
683 QEMU_THREAD_JOINABLE
);