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 "qemu/main-loop.h"
20 #include "migration/blocker.h"
26 #include "migration/global_state.h"
27 #include "migration/misc.h"
28 #include "migration.h"
30 #include "qemu-file-channel.h"
31 #include "qemu-file.h"
32 #include "migration/vmstate.h"
33 #include "block/block.h"
34 #include "qapi/error.h"
35 #include "qapi/clone-visitor.h"
36 #include "qapi/qapi-visit-sockets.h"
37 #include "qapi/qapi-commands-migration.h"
38 #include "qapi/qapi-events-migration.h"
39 #include "qapi/qmp/qerror.h"
40 #include "qapi/qmp/qnull.h"
43 #include "postcopy-ram.h"
44 #include "qemu/thread.h"
46 #include "exec/target_page.h"
47 #include "io/channel-buffer.h"
48 #include "migration/colo.h"
49 #include "hw/boards.h"
50 #include "monitor/monitor.h"
51 #include "net/announce.h"
53 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
55 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
57 #define BUFFER_DELAY 100
58 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
60 /* Time in milliseconds we are allowed to stop the source,
61 * for sending the last part */
62 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
64 /* Maximum migrate downtime set to 2000 seconds */
65 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
66 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
68 /* Default compression thread count */
69 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
70 /* Default decompression thread count, usually decompression is at
71 * least 4 times as fast as compression.*/
72 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
73 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
74 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
75 /* Define default autoconverge cpu throttle migration parameters */
76 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
77 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
78 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
80 /* Migration XBZRLE default cache size */
81 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
83 /* The delay time (in ms) between two COLO checkpoints */
84 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
85 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
87 /* Background transfer rate for postcopy, 0 means unlimited, note
88 * that page requests can still exceed this limit.
90 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
93 * Parameters for self_announce_delay giving a stream of RARP/ARP
94 * packets after migration.
96 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
97 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
98 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
99 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
101 static NotifierList migration_state_notifiers
=
102 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
104 static bool deferred_incoming
;
106 /* Messages sent on the return path from destination to source */
107 enum mig_rp_message_type
{
108 MIG_RP_MSG_INVALID
= 0, /* Must be 0 */
109 MIG_RP_MSG_SHUT
, /* sibling will not send any more RP messages */
110 MIG_RP_MSG_PONG
, /* Response to a PING; data (seq: be32 ) */
112 MIG_RP_MSG_REQ_PAGES_ID
, /* data (start: be64, len: be32, id: string) */
113 MIG_RP_MSG_REQ_PAGES
, /* data (start: be64, len: be32) */
114 MIG_RP_MSG_RECV_BITMAP
, /* send recved_bitmap back to source */
115 MIG_RP_MSG_RESUME_ACK
, /* tell source that we are ready to resume */
120 /* When we add fault tolerance, we could have several
121 migrations at once. For now we don't need to add
122 dynamic creation of migration */
124 static MigrationState
*current_migration
;
125 static MigrationIncomingState
*current_incoming
;
127 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
128 static int migration_maybe_pause(MigrationState
*s
,
129 int *current_active_state
,
131 static void migrate_fd_cancel(MigrationState
*s
);
133 void migration_object_init(void)
135 MachineState
*ms
= MACHINE(qdev_get_machine());
138 /* This can only be called once. */
139 assert(!current_migration
);
140 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
143 * Init the migrate incoming object as well no matter whether
144 * we'll use it or not.
146 assert(!current_incoming
);
147 current_incoming
= g_new0(MigrationIncomingState
, 1);
148 current_incoming
->state
= MIGRATION_STATUS_NONE
;
149 current_incoming
->postcopy_remote_fds
=
150 g_array_new(FALSE
, TRUE
, sizeof(struct PostCopyFD
));
151 qemu_mutex_init(¤t_incoming
->rp_mutex
);
152 qemu_event_init(¤t_incoming
->main_thread_load_event
, false);
153 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_dst
, 0);
154 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fault
, 0);
156 init_dirty_bitmap_incoming_migration();
158 if (!migration_object_check(current_migration
, &err
)) {
159 error_report_err(err
);
164 * We cannot really do this in migration_instance_init() since at
165 * that time global properties are not yet applied, then this
166 * value will be definitely replaced by something else.
168 if (ms
->enforce_config_section
) {
169 current_migration
->send_configuration
= true;
173 void migration_shutdown(void)
176 * Cancel the current migration - that will (eventually)
177 * stop the migration using this structure
179 migrate_fd_cancel(current_migration
);
180 object_unref(OBJECT(current_migration
));
184 MigrationState
*migrate_get_current(void)
186 /* This can only be called after the object created. */
187 assert(current_migration
);
188 return current_migration
;
191 MigrationIncomingState
*migration_incoming_get_current(void)
193 assert(current_incoming
);
194 return current_incoming
;
197 void migration_incoming_state_destroy(void)
199 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
201 if (mis
->to_src_file
) {
202 /* Tell source that we are done */
203 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
204 qemu_fclose(mis
->to_src_file
);
205 mis
->to_src_file
= NULL
;
208 if (mis
->from_src_file
) {
209 qemu_fclose(mis
->from_src_file
);
210 mis
->from_src_file
= NULL
;
212 if (mis
->postcopy_remote_fds
) {
213 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
214 mis
->postcopy_remote_fds
= NULL
;
217 qemu_event_reset(&mis
->main_thread_load_event
);
219 if (mis
->socket_address_list
) {
220 qapi_free_SocketAddressList(mis
->socket_address_list
);
221 mis
->socket_address_list
= NULL
;
225 static void migrate_generate_event(int new_state
)
227 if (migrate_use_events()) {
228 qapi_event_send_migration(new_state
);
232 static bool migrate_late_block_activate(void)
236 s
= migrate_get_current();
238 return s
->enabled_capabilities
[
239 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE
];
243 * Called on -incoming with a defer: uri.
244 * The migration can be started later after any parameters have been
247 static void deferred_incoming_migration(Error
**errp
)
249 if (deferred_incoming
) {
250 error_setg(errp
, "Incoming migration already deferred");
252 deferred_incoming
= true;
256 * Send a message on the return channel back to the source
259 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
260 enum mig_rp_message_type message_type
,
261 uint16_t len
, void *data
)
265 trace_migrate_send_rp_message((int)message_type
, len
);
266 qemu_mutex_lock(&mis
->rp_mutex
);
269 * It's possible that the file handle got lost due to network
272 if (!mis
->to_src_file
) {
277 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
278 qemu_put_be16(mis
->to_src_file
, len
);
279 qemu_put_buffer(mis
->to_src_file
, data
, len
);
280 qemu_fflush(mis
->to_src_file
);
282 /* It's possible that qemu file got error during sending */
283 ret
= qemu_file_get_error(mis
->to_src_file
);
286 qemu_mutex_unlock(&mis
->rp_mutex
);
290 /* Request a range of pages from the source VM at the given
292 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
293 * as the last request (a name must have been given previously)
294 * Start: Address offset within the RB
295 * Len: Length in bytes required - must be a multiple of pagesize
297 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
, const char *rbname
,
298 ram_addr_t start
, size_t len
)
300 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
301 size_t msglen
= 12; /* start + len */
302 enum mig_rp_message_type msg_type
;
304 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
305 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
308 int rbname_len
= strlen(rbname
);
309 assert(rbname_len
< 256);
311 bufc
[msglen
++] = rbname_len
;
312 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
313 msglen
+= rbname_len
;
314 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
316 msg_type
= MIG_RP_MSG_REQ_PAGES
;
319 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
322 static bool migration_colo_enabled
;
323 bool migration_incoming_colo_enabled(void)
325 return migration_colo_enabled
;
328 void migration_incoming_disable_colo(void)
330 migration_colo_enabled
= false;
333 void migration_incoming_enable_colo(void)
335 migration_colo_enabled
= true;
338 void migrate_add_address(SocketAddress
*address
)
340 MigrationIncomingState
*mis
= migration_incoming_get_current();
341 SocketAddressList
*addrs
;
343 addrs
= g_new0(SocketAddressList
, 1);
344 addrs
->next
= mis
->socket_address_list
;
345 mis
->socket_address_list
= addrs
;
346 addrs
->value
= QAPI_CLONE(SocketAddress
, address
);
349 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
353 qapi_event_send_migration(MIGRATION_STATUS_SETUP
);
354 if (!strcmp(uri
, "defer")) {
355 deferred_incoming_migration(errp
);
356 } else if (strstart(uri
, "tcp:", &p
)) {
357 tcp_start_incoming_migration(p
, errp
);
359 } else if (strstart(uri
, "rdma:", &p
)) {
360 rdma_start_incoming_migration(p
, errp
);
362 } else if (strstart(uri
, "exec:", &p
)) {
363 exec_start_incoming_migration(p
, errp
);
364 } else if (strstart(uri
, "unix:", &p
)) {
365 unix_start_incoming_migration(p
, errp
);
366 } else if (strstart(uri
, "fd:", &p
)) {
367 fd_start_incoming_migration(p
, errp
);
369 error_setg(errp
, "unknown migration protocol: %s", uri
);
373 static void process_incoming_migration_bh(void *opaque
)
375 Error
*local_err
= NULL
;
376 MigrationIncomingState
*mis
= opaque
;
378 /* If capability late_block_activate is set:
379 * Only fire up the block code now if we're going to restart the
380 * VM, else 'cont' will do it.
381 * This causes file locking to happen; so we don't want it to happen
382 * unless we really are starting the VM.
384 if (!migrate_late_block_activate() ||
385 (autostart
&& (!global_state_received() ||
386 global_state_get_runstate() == RUN_STATE_RUNNING
))) {
387 /* Make sure all file formats flush their mutable metadata.
388 * If we get an error here, just don't restart the VM yet. */
389 bdrv_invalidate_cache_all(&local_err
);
391 error_report_err(local_err
);
398 * This must happen after all error conditions are dealt with and
399 * we're sure the VM is going to be running on this host.
401 qemu_announce_self(&mis
->announce_timer
, migrate_announce_params());
403 if (multifd_load_cleanup(&local_err
) != 0) {
404 error_report_err(local_err
);
407 /* If global state section was not received or we are in running
408 state, we need to obey autostart. Any other state is set with
411 dirty_bitmap_mig_before_vm_start();
413 if (!global_state_received() ||
414 global_state_get_runstate() == RUN_STATE_RUNNING
) {
418 runstate_set(RUN_STATE_PAUSED
);
420 } else if (migration_incoming_colo_enabled()) {
421 migration_incoming_disable_colo();
424 runstate_set(global_state_get_runstate());
427 * This must happen after any state changes since as soon as an external
428 * observer sees this event they might start to prod at the VM assuming
431 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
432 MIGRATION_STATUS_COMPLETED
);
433 qemu_bh_delete(mis
->bh
);
434 migration_incoming_state_destroy();
437 static void process_incoming_migration_co(void *opaque
)
439 MigrationIncomingState
*mis
= migration_incoming_get_current();
442 Error
*local_err
= NULL
;
444 assert(mis
->from_src_file
);
445 mis
->migration_incoming_co
= qemu_coroutine_self();
446 mis
->largest_page_size
= qemu_ram_pagesize_largest();
447 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
448 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
449 MIGRATION_STATUS_ACTIVE
);
450 ret
= qemu_loadvm_state(mis
->from_src_file
);
452 ps
= postcopy_state_get();
453 trace_process_incoming_migration_co_end(ret
, ps
);
454 if (ps
!= POSTCOPY_INCOMING_NONE
) {
455 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
457 * Where a migration had postcopy enabled (and thus went to advise)
458 * but managed to complete within the precopy period, we can use
461 postcopy_ram_incoming_cleanup(mis
);
462 } else if (ret
>= 0) {
464 * Postcopy was started, cleanup should happen at the end of the
467 trace_process_incoming_migration_co_postcopy_end_main();
470 /* Else if something went wrong then just fall out of the normal exit */
473 /* we get COLO info, and know if we are in COLO mode */
474 if (!ret
&& migration_incoming_colo_enabled()) {
475 /* Make sure all file formats flush their mutable metadata */
476 bdrv_invalidate_cache_all(&local_err
);
478 error_report_err(local_err
);
482 if (colo_init_ram_cache() < 0) {
483 error_report("Init ram cache failed");
487 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
488 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
489 mis
->have_colo_incoming_thread
= true;
490 qemu_coroutine_yield();
492 /* Wait checkpoint incoming thread exit before free resource */
493 qemu_thread_join(&mis
->colo_incoming_thread
);
494 /* We hold the global iothread lock, so it is safe here */
495 colo_release_ram_cache();
499 error_report("load of migration failed: %s", strerror(-ret
));
502 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
503 qemu_bh_schedule(mis
->bh
);
504 mis
->migration_incoming_co
= NULL
;
508 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
509 MIGRATION_STATUS_FAILED
);
510 qemu_fclose(mis
->from_src_file
);
511 if (multifd_load_cleanup(&local_err
) != 0) {
512 error_report_err(local_err
);
517 static void migration_incoming_setup(QEMUFile
*f
)
519 MigrationIncomingState
*mis
= migration_incoming_get_current();
521 if (multifd_load_setup() != 0) {
522 /* We haven't been able to create multifd threads
523 nothing better to do */
527 if (!mis
->from_src_file
) {
528 mis
->from_src_file
= f
;
530 qemu_file_set_blocking(f
, false);
533 void migration_incoming_process(void)
535 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
536 qemu_coroutine_enter(co
);
539 /* Returns true if recovered from a paused migration, otherwise false */
540 static bool postcopy_try_recover(QEMUFile
*f
)
542 MigrationIncomingState
*mis
= migration_incoming_get_current();
544 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
545 /* Resumed from a paused postcopy migration */
547 mis
->from_src_file
= f
;
548 /* Postcopy has standalone thread to do vm load */
549 qemu_file_set_blocking(f
, true);
551 /* Re-configure the return path */
552 mis
->to_src_file
= qemu_file_get_return_path(f
);
554 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
555 MIGRATION_STATUS_POSTCOPY_RECOVER
);
558 * Here, we only wake up the main loading thread (while the
559 * fault thread will still be waiting), so that we can receive
560 * commands from source now, and answer it if needed. The
561 * fault thread will be woken up afterwards until we are sure
562 * that source is ready to reply to page requests.
564 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
571 void migration_fd_process_incoming(QEMUFile
*f
)
573 if (postcopy_try_recover(f
)) {
577 migration_incoming_setup(f
);
578 migration_incoming_process();
581 void migration_ioc_process_incoming(QIOChannel
*ioc
, Error
**errp
)
583 MigrationIncomingState
*mis
= migration_incoming_get_current();
584 bool start_migration
;
586 if (!mis
->from_src_file
) {
587 /* The first connection (multifd may have multiple) */
588 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
590 /* If it's a recovery, we're done */
591 if (postcopy_try_recover(f
)) {
595 migration_incoming_setup(f
);
598 * Common migration only needs one channel, so we can start
599 * right now. Multifd needs more than one channel, we wait.
601 start_migration
= !migrate_use_multifd();
603 Error
*local_err
= NULL
;
604 /* Multiple connections */
605 assert(migrate_use_multifd());
606 start_migration
= multifd_recv_new_channel(ioc
, &local_err
);
608 error_propagate(errp
, local_err
);
613 if (start_migration
) {
614 migration_incoming_process();
619 * @migration_has_all_channels: We have received all channels that we need
621 * Returns true when we have got connections to all the channels that
622 * we need for migration.
624 bool migration_has_all_channels(void)
626 MigrationIncomingState
*mis
= migration_incoming_get_current();
629 all_channels
= multifd_recv_all_channels_created();
631 return all_channels
&& mis
->from_src_file
!= NULL
;
635 * Send a 'SHUT' message on the return channel with the given value
636 * to indicate that we've finished with the RP. Non-0 value indicates
639 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
644 buf
= cpu_to_be32(value
);
645 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
649 * Send a 'PONG' message on the return channel with the given value
650 * (normally in response to a 'PING')
652 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
657 buf
= cpu_to_be32(value
);
658 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
661 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
669 * First, we send the header part. It contains only the len of
670 * idstr, and the idstr itself.
672 len
= strlen(block_name
);
674 memcpy(buf
+ 1, block_name
, len
);
676 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
677 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
682 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
685 * Next, we dump the received bitmap to the stream.
687 * TODO: currently we are safe since we are the only one that is
688 * using the to_src_file handle (fault thread is still paused),
689 * and it's ok even not taking the mutex. However the best way is
690 * to take the lock before sending the message header, and release
691 * the lock after sending the bitmap.
693 qemu_mutex_lock(&mis
->rp_mutex
);
694 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
695 qemu_mutex_unlock(&mis
->rp_mutex
);
697 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
700 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
704 buf
= cpu_to_be32(value
);
705 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
708 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
710 MigrationCapabilityStatusList
*head
= NULL
;
711 MigrationCapabilityStatusList
*caps
;
712 MigrationState
*s
= migrate_get_current();
715 caps
= NULL
; /* silence compiler warning */
716 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
717 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
718 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
723 head
= g_malloc0(sizeof(*caps
));
726 caps
->next
= g_malloc0(sizeof(*caps
));
730 g_malloc(sizeof(*caps
->value
));
731 caps
->value
->capability
= i
;
732 caps
->value
->state
= s
->enabled_capabilities
[i
];
738 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
740 MigrationParameters
*params
;
741 MigrationState
*s
= migrate_get_current();
743 /* TODO use QAPI_CLONE() instead of duplicating it inline */
744 params
= g_malloc0(sizeof(*params
));
745 params
->has_compress_level
= true;
746 params
->compress_level
= s
->parameters
.compress_level
;
747 params
->has_compress_threads
= true;
748 params
->compress_threads
= s
->parameters
.compress_threads
;
749 params
->has_compress_wait_thread
= true;
750 params
->compress_wait_thread
= s
->parameters
.compress_wait_thread
;
751 params
->has_decompress_threads
= true;
752 params
->decompress_threads
= s
->parameters
.decompress_threads
;
753 params
->has_cpu_throttle_initial
= true;
754 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
755 params
->has_cpu_throttle_increment
= true;
756 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
757 params
->has_tls_creds
= true;
758 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
759 params
->has_tls_hostname
= true;
760 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
761 params
->has_tls_authz
= true;
762 params
->tls_authz
= g_strdup(s
->parameters
.tls_authz
);
763 params
->has_max_bandwidth
= true;
764 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
765 params
->has_downtime_limit
= true;
766 params
->downtime_limit
= s
->parameters
.downtime_limit
;
767 params
->has_x_checkpoint_delay
= true;
768 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
769 params
->has_block_incremental
= true;
770 params
->block_incremental
= s
->parameters
.block_incremental
;
771 params
->has_multifd_channels
= true;
772 params
->multifd_channels
= s
->parameters
.multifd_channels
;
773 params
->has_xbzrle_cache_size
= true;
774 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
775 params
->has_max_postcopy_bandwidth
= true;
776 params
->max_postcopy_bandwidth
= s
->parameters
.max_postcopy_bandwidth
;
777 params
->has_max_cpu_throttle
= true;
778 params
->max_cpu_throttle
= s
->parameters
.max_cpu_throttle
;
779 params
->has_announce_initial
= true;
780 params
->announce_initial
= s
->parameters
.announce_initial
;
781 params
->has_announce_max
= true;
782 params
->announce_max
= s
->parameters
.announce_max
;
783 params
->has_announce_rounds
= true;
784 params
->announce_rounds
= s
->parameters
.announce_rounds
;
785 params
->has_announce_step
= true;
786 params
->announce_step
= s
->parameters
.announce_step
;
791 AnnounceParameters
*migrate_announce_params(void)
793 static AnnounceParameters ap
;
795 MigrationState
*s
= migrate_get_current();
797 ap
.initial
= s
->parameters
.announce_initial
;
798 ap
.max
= s
->parameters
.announce_max
;
799 ap
.rounds
= s
->parameters
.announce_rounds
;
800 ap
.step
= s
->parameters
.announce_step
;
806 * Return true if we're already in the middle of a migration
807 * (i.e. any of the active or setup states)
809 bool migration_is_setup_or_active(int state
)
812 case MIGRATION_STATUS_ACTIVE
:
813 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
814 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
815 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
816 case MIGRATION_STATUS_SETUP
:
817 case MIGRATION_STATUS_PRE_SWITCHOVER
:
818 case MIGRATION_STATUS_DEVICE
:
827 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
829 info
->has_ram
= true;
830 info
->ram
= g_malloc0(sizeof(*info
->ram
));
831 info
->ram
->transferred
= ram_counters
.transferred
;
832 info
->ram
->total
= ram_bytes_total();
833 info
->ram
->duplicate
= ram_counters
.duplicate
;
834 /* legacy value. It is not used anymore */
835 info
->ram
->skipped
= 0;
836 info
->ram
->normal
= ram_counters
.normal
;
837 info
->ram
->normal_bytes
= ram_counters
.normal
*
838 qemu_target_page_size();
839 info
->ram
->mbps
= s
->mbps
;
840 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
841 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
842 info
->ram
->page_size
= qemu_target_page_size();
843 info
->ram
->multifd_bytes
= ram_counters
.multifd_bytes
;
844 info
->ram
->pages_per_second
= s
->pages_per_second
;
846 if (migrate_use_xbzrle()) {
847 info
->has_xbzrle_cache
= true;
848 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
849 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
850 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
851 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
852 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
853 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
854 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
857 if (migrate_use_compression()) {
858 info
->has_compression
= true;
859 info
->compression
= g_malloc0(sizeof(*info
->compression
));
860 info
->compression
->pages
= compression_counters
.pages
;
861 info
->compression
->busy
= compression_counters
.busy
;
862 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
863 info
->compression
->compressed_size
=
864 compression_counters
.compressed_size
;
865 info
->compression
->compression_rate
=
866 compression_counters
.compression_rate
;
869 if (cpu_throttle_active()) {
870 info
->has_cpu_throttle_percentage
= true;
871 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
874 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
875 info
->ram
->remaining
= ram_bytes_remaining();
876 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
880 static void populate_disk_info(MigrationInfo
*info
)
882 if (blk_mig_active()) {
883 info
->has_disk
= true;
884 info
->disk
= g_malloc0(sizeof(*info
->disk
));
885 info
->disk
->transferred
= blk_mig_bytes_transferred();
886 info
->disk
->remaining
= blk_mig_bytes_remaining();
887 info
->disk
->total
= blk_mig_bytes_total();
891 static void fill_source_migration_info(MigrationInfo
*info
)
893 MigrationState
*s
= migrate_get_current();
896 case MIGRATION_STATUS_NONE
:
897 /* no migration has happened ever */
898 /* do not overwrite destination migration status */
901 case MIGRATION_STATUS_SETUP
:
902 info
->has_status
= true;
903 info
->has_total_time
= false;
905 case MIGRATION_STATUS_ACTIVE
:
906 case MIGRATION_STATUS_CANCELLING
:
907 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
908 case MIGRATION_STATUS_PRE_SWITCHOVER
:
909 case MIGRATION_STATUS_DEVICE
:
910 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
911 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
912 /* TODO add some postcopy stats */
913 info
->has_status
= true;
914 info
->has_total_time
= true;
915 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
917 info
->has_expected_downtime
= true;
918 info
->expected_downtime
= s
->expected_downtime
;
919 info
->has_setup_time
= true;
920 info
->setup_time
= s
->setup_time
;
922 populate_ram_info(info
, s
);
923 populate_disk_info(info
);
925 case MIGRATION_STATUS_COLO
:
926 info
->has_status
= true;
927 /* TODO: display COLO specific information (checkpoint info etc.) */
929 case MIGRATION_STATUS_COMPLETED
:
930 info
->has_status
= true;
931 info
->has_total_time
= true;
932 info
->total_time
= s
->total_time
;
933 info
->has_downtime
= true;
934 info
->downtime
= s
->downtime
;
935 info
->has_setup_time
= true;
936 info
->setup_time
= s
->setup_time
;
938 populate_ram_info(info
, s
);
940 case MIGRATION_STATUS_FAILED
:
941 info
->has_status
= true;
943 info
->has_error_desc
= true;
944 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
947 case MIGRATION_STATUS_CANCELLED
:
948 info
->has_status
= true;
951 info
->status
= s
->state
;
955 * @migration_caps_check - check capability validity
957 * @cap_list: old capability list, array of bool
958 * @params: new capabilities to be applied soon
959 * @errp: set *errp if the check failed, with reason
961 * Returns true if check passed, otherwise false.
963 static bool migrate_caps_check(bool *cap_list
,
964 MigrationCapabilityStatusList
*params
,
967 MigrationCapabilityStatusList
*cap
;
968 bool old_postcopy_cap
;
969 MigrationIncomingState
*mis
= migration_incoming_get_current();
971 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
973 for (cap
= params
; cap
; cap
= cap
->next
) {
974 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
977 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
978 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
979 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
981 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
986 #ifndef CONFIG_REPLICATION
987 if (cap_list
[MIGRATION_CAPABILITY_X_COLO
]) {
988 error_setg(errp
, "QEMU compiled without replication module"
989 " can't enable COLO");
990 error_append_hint(errp
, "Please enable replication before COLO.\n");
995 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
996 if (cap_list
[MIGRATION_CAPABILITY_COMPRESS
]) {
997 /* The decompression threads asynchronously write into RAM
998 * rather than use the atomic copies needed to avoid
999 * userfaulting. It should be possible to fix the decompression
1000 * threads for compatibility in future.
1002 error_setg(errp
, "Postcopy is not currently compatible "
1003 "with compression");
1007 /* This check is reasonably expensive, so only when it's being
1008 * set the first time, also it's only the destination that needs
1011 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
1012 !postcopy_ram_supported_by_host(mis
)) {
1013 /* postcopy_ram_supported_by_host will have emitted a more
1016 error_setg(errp
, "Postcopy is not supported");
1020 if (cap_list
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
]) {
1021 error_setg(errp
, "Postcopy is not compatible with ignore-shared");
1029 static void fill_destination_migration_info(MigrationInfo
*info
)
1031 MigrationIncomingState
*mis
= migration_incoming_get_current();
1033 if (mis
->socket_address_list
) {
1034 info
->has_socket_address
= true;
1035 info
->socket_address
=
1036 QAPI_CLONE(SocketAddressList
, mis
->socket_address_list
);
1039 switch (mis
->state
) {
1040 case MIGRATION_STATUS_NONE
:
1043 case MIGRATION_STATUS_SETUP
:
1044 case MIGRATION_STATUS_CANCELLING
:
1045 case MIGRATION_STATUS_CANCELLED
:
1046 case MIGRATION_STATUS_ACTIVE
:
1047 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1048 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1049 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1050 case MIGRATION_STATUS_FAILED
:
1051 case MIGRATION_STATUS_COLO
:
1052 info
->has_status
= true;
1054 case MIGRATION_STATUS_COMPLETED
:
1055 info
->has_status
= true;
1056 fill_destination_postcopy_migration_info(info
);
1059 info
->status
= mis
->state
;
1062 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1064 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1066 fill_destination_migration_info(info
);
1067 fill_source_migration_info(info
);
1072 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
1075 MigrationState
*s
= migrate_get_current();
1076 MigrationCapabilityStatusList
*cap
;
1077 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
1079 if (migration_is_setup_or_active(s
->state
)) {
1080 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1084 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
1085 if (!migrate_caps_check(cap_list
, params
, errp
)) {
1089 for (cap
= params
; cap
; cap
= cap
->next
) {
1090 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
1095 * Check whether the parameters are valid. Error will be put into errp
1096 * (if provided). Return true if valid, otherwise false.
1098 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
1100 if (params
->has_compress_level
&&
1101 (params
->compress_level
> 9)) {
1102 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
1103 "is invalid, it should be in the range of 0 to 9");
1107 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
1108 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1110 "is invalid, it should be in the range of 1 to 255");
1114 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
1115 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1116 "decompress_threads",
1117 "is invalid, it should be in the range of 1 to 255");
1121 if (params
->has_cpu_throttle_initial
&&
1122 (params
->cpu_throttle_initial
< 1 ||
1123 params
->cpu_throttle_initial
> 99)) {
1124 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1125 "cpu_throttle_initial",
1126 "an integer in the range of 1 to 99");
1130 if (params
->has_cpu_throttle_increment
&&
1131 (params
->cpu_throttle_increment
< 1 ||
1132 params
->cpu_throttle_increment
> 99)) {
1133 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1134 "cpu_throttle_increment",
1135 "an integer in the range of 1 to 99");
1139 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
1140 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
1141 " range of 0 to %zu bytes/second", SIZE_MAX
);
1145 if (params
->has_downtime_limit
&&
1146 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
1147 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1148 "the range of 0 to %d milliseconds",
1149 MAX_MIGRATE_DOWNTIME
);
1153 /* x_checkpoint_delay is now always positive */
1155 if (params
->has_multifd_channels
&& (params
->multifd_channels
< 1)) {
1156 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1158 "is invalid, it should be in the range of 1 to 255");
1162 if (params
->has_xbzrle_cache_size
&&
1163 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
1164 !is_power_of_2(params
->xbzrle_cache_size
))) {
1165 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1166 "xbzrle_cache_size",
1167 "is invalid, it should be bigger than target page size"
1168 " and a power of two");
1172 if (params
->has_max_cpu_throttle
&&
1173 (params
->max_cpu_throttle
< params
->cpu_throttle_initial
||
1174 params
->max_cpu_throttle
> 99)) {
1175 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1177 "an integer in the range of cpu_throttle_initial to 99");
1181 if (params
->has_announce_initial
&&
1182 params
->announce_initial
> 100000) {
1183 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1185 "is invalid, it must be less than 100000 ms");
1188 if (params
->has_announce_max
&&
1189 params
->announce_max
> 100000) {
1190 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1192 "is invalid, it must be less than 100000 ms");
1195 if (params
->has_announce_rounds
&&
1196 params
->announce_rounds
> 1000) {
1197 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1199 "is invalid, it must be in the range of 0 to 1000");
1202 if (params
->has_announce_step
&&
1203 (params
->announce_step
< 1 ||
1204 params
->announce_step
> 10000)) {
1205 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1207 "is invalid, it must be in the range of 1 to 10000 ms");
1213 static void migrate_params_test_apply(MigrateSetParameters
*params
,
1214 MigrationParameters
*dest
)
1216 *dest
= migrate_get_current()->parameters
;
1218 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1220 if (params
->has_compress_level
) {
1221 dest
->compress_level
= params
->compress_level
;
1224 if (params
->has_compress_threads
) {
1225 dest
->compress_threads
= params
->compress_threads
;
1228 if (params
->has_compress_wait_thread
) {
1229 dest
->compress_wait_thread
= params
->compress_wait_thread
;
1232 if (params
->has_decompress_threads
) {
1233 dest
->decompress_threads
= params
->decompress_threads
;
1236 if (params
->has_cpu_throttle_initial
) {
1237 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
1240 if (params
->has_cpu_throttle_increment
) {
1241 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
1244 if (params
->has_tls_creds
) {
1245 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1246 dest
->tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1249 if (params
->has_tls_hostname
) {
1250 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1251 dest
->tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1254 if (params
->has_max_bandwidth
) {
1255 dest
->max_bandwidth
= params
->max_bandwidth
;
1258 if (params
->has_downtime_limit
) {
1259 dest
->downtime_limit
= params
->downtime_limit
;
1262 if (params
->has_x_checkpoint_delay
) {
1263 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
1266 if (params
->has_block_incremental
) {
1267 dest
->block_incremental
= params
->block_incremental
;
1269 if (params
->has_multifd_channels
) {
1270 dest
->multifd_channels
= params
->multifd_channels
;
1272 if (params
->has_xbzrle_cache_size
) {
1273 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
1275 if (params
->has_max_postcopy_bandwidth
) {
1276 dest
->max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1278 if (params
->has_max_cpu_throttle
) {
1279 dest
->max_cpu_throttle
= params
->max_cpu_throttle
;
1281 if (params
->has_announce_initial
) {
1282 dest
->announce_initial
= params
->announce_initial
;
1284 if (params
->has_announce_max
) {
1285 dest
->announce_max
= params
->announce_max
;
1287 if (params
->has_announce_rounds
) {
1288 dest
->announce_rounds
= params
->announce_rounds
;
1290 if (params
->has_announce_step
) {
1291 dest
->announce_step
= params
->announce_step
;
1295 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
1297 MigrationState
*s
= migrate_get_current();
1299 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1301 if (params
->has_compress_level
) {
1302 s
->parameters
.compress_level
= params
->compress_level
;
1305 if (params
->has_compress_threads
) {
1306 s
->parameters
.compress_threads
= params
->compress_threads
;
1309 if (params
->has_compress_wait_thread
) {
1310 s
->parameters
.compress_wait_thread
= params
->compress_wait_thread
;
1313 if (params
->has_decompress_threads
) {
1314 s
->parameters
.decompress_threads
= params
->decompress_threads
;
1317 if (params
->has_cpu_throttle_initial
) {
1318 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
1321 if (params
->has_cpu_throttle_increment
) {
1322 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
1325 if (params
->has_tls_creds
) {
1326 g_free(s
->parameters
.tls_creds
);
1327 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1328 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1331 if (params
->has_tls_hostname
) {
1332 g_free(s
->parameters
.tls_hostname
);
1333 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1334 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1337 if (params
->has_tls_authz
) {
1338 g_free(s
->parameters
.tls_authz
);
1339 assert(params
->tls_authz
->type
== QTYPE_QSTRING
);
1340 s
->parameters
.tls_authz
= g_strdup(params
->tls_authz
->u
.s
);
1343 if (params
->has_max_bandwidth
) {
1344 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1345 if (s
->to_dst_file
&& !migration_in_postcopy()) {
1346 qemu_file_set_rate_limit(s
->to_dst_file
,
1347 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1351 if (params
->has_downtime_limit
) {
1352 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1355 if (params
->has_x_checkpoint_delay
) {
1356 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1357 if (migration_in_colo_state()) {
1358 colo_checkpoint_notify(s
);
1362 if (params
->has_block_incremental
) {
1363 s
->parameters
.block_incremental
= params
->block_incremental
;
1365 if (params
->has_multifd_channels
) {
1366 s
->parameters
.multifd_channels
= params
->multifd_channels
;
1368 if (params
->has_xbzrle_cache_size
) {
1369 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1370 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1372 if (params
->has_max_postcopy_bandwidth
) {
1373 s
->parameters
.max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1374 if (s
->to_dst_file
&& migration_in_postcopy()) {
1375 qemu_file_set_rate_limit(s
->to_dst_file
,
1376 s
->parameters
.max_postcopy_bandwidth
/ XFER_LIMIT_RATIO
);
1379 if (params
->has_max_cpu_throttle
) {
1380 s
->parameters
.max_cpu_throttle
= params
->max_cpu_throttle
;
1382 if (params
->has_announce_initial
) {
1383 s
->parameters
.announce_initial
= params
->announce_initial
;
1385 if (params
->has_announce_max
) {
1386 s
->parameters
.announce_max
= params
->announce_max
;
1388 if (params
->has_announce_rounds
) {
1389 s
->parameters
.announce_rounds
= params
->announce_rounds
;
1391 if (params
->has_announce_step
) {
1392 s
->parameters
.announce_step
= params
->announce_step
;
1396 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1398 MigrationParameters tmp
;
1400 /* TODO Rewrite "" to null instead */
1401 if (params
->has_tls_creds
1402 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1403 qobject_unref(params
->tls_creds
->u
.n
);
1404 params
->tls_creds
->type
= QTYPE_QSTRING
;
1405 params
->tls_creds
->u
.s
= strdup("");
1407 /* TODO Rewrite "" to null instead */
1408 if (params
->has_tls_hostname
1409 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1410 qobject_unref(params
->tls_hostname
->u
.n
);
1411 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1412 params
->tls_hostname
->u
.s
= strdup("");
1415 migrate_params_test_apply(params
, &tmp
);
1417 if (!migrate_params_check(&tmp
, errp
)) {
1418 /* Invalid parameter */
1422 migrate_params_apply(params
, errp
);
1426 void qmp_migrate_start_postcopy(Error
**errp
)
1428 MigrationState
*s
= migrate_get_current();
1430 if (!migrate_postcopy()) {
1431 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1432 " the start of migration");
1436 if (s
->state
== MIGRATION_STATUS_NONE
) {
1437 error_setg(errp
, "Postcopy must be started after migration has been"
1442 * we don't error if migration has finished since that would be racy
1443 * with issuing this command.
1445 atomic_set(&s
->start_postcopy
, true);
1448 /* shared migration helpers */
1450 void migrate_set_state(int *state
, int old_state
, int new_state
)
1452 assert(new_state
< MIGRATION_STATUS__MAX
);
1453 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1454 trace_migrate_set_state(MigrationStatus_str(new_state
));
1455 migrate_generate_event(new_state
);
1459 static MigrationCapabilityStatusList
*migrate_cap_add(
1460 MigrationCapabilityStatusList
*list
,
1461 MigrationCapability index
,
1464 MigrationCapabilityStatusList
*cap
;
1466 cap
= g_new0(MigrationCapabilityStatusList
, 1);
1467 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
1468 cap
->value
->capability
= index
;
1469 cap
->value
->state
= state
;
1475 void migrate_set_block_enabled(bool value
, Error
**errp
)
1477 MigrationCapabilityStatusList
*cap
;
1479 cap
= migrate_cap_add(NULL
, MIGRATION_CAPABILITY_BLOCK
, value
);
1480 qmp_migrate_set_capabilities(cap
, errp
);
1481 qapi_free_MigrationCapabilityStatusList(cap
);
1484 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1486 s
->parameters
.block_incremental
= value
;
1489 static void block_cleanup_parameters(MigrationState
*s
)
1491 if (s
->must_remove_block_options
) {
1492 /* setting to false can never fail */
1493 migrate_set_block_enabled(false, &error_abort
);
1494 migrate_set_block_incremental(s
, false);
1495 s
->must_remove_block_options
= false;
1499 static void migrate_fd_cleanup(MigrationState
*s
)
1501 qemu_bh_delete(s
->cleanup_bh
);
1502 s
->cleanup_bh
= NULL
;
1504 qemu_savevm_state_cleanup();
1506 if (s
->to_dst_file
) {
1509 trace_migrate_fd_cleanup();
1510 qemu_mutex_unlock_iothread();
1511 if (s
->migration_thread_running
) {
1512 qemu_thread_join(&s
->thread
);
1513 s
->migration_thread_running
= false;
1515 qemu_mutex_lock_iothread();
1517 multifd_save_cleanup();
1518 qemu_mutex_lock(&s
->qemu_file_lock
);
1519 tmp
= s
->to_dst_file
;
1520 s
->to_dst_file
= NULL
;
1521 qemu_mutex_unlock(&s
->qemu_file_lock
);
1523 * Close the file handle without the lock to make sure the
1524 * critical section won't block for long.
1529 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
1530 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
1532 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1533 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1534 MIGRATION_STATUS_CANCELLED
);
1538 /* It is used on info migrate. We can't free it */
1539 error_report_err(error_copy(s
->error
));
1541 notifier_list_notify(&migration_state_notifiers
, s
);
1542 block_cleanup_parameters(s
);
1545 static void migrate_fd_cleanup_schedule(MigrationState
*s
)
1548 * Ref the state for bh, because it may be called when
1549 * there're already no other refs
1551 object_ref(OBJECT(s
));
1552 qemu_bh_schedule(s
->cleanup_bh
);
1555 static void migrate_fd_cleanup_bh(void *opaque
)
1557 MigrationState
*s
= opaque
;
1558 migrate_fd_cleanup(s
);
1559 object_unref(OBJECT(s
));
1562 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1564 qemu_mutex_lock(&s
->error_mutex
);
1566 s
->error
= error_copy(error
);
1568 qemu_mutex_unlock(&s
->error_mutex
);
1571 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1573 trace_migrate_fd_error(error_get_pretty(error
));
1574 assert(s
->to_dst_file
== NULL
);
1575 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1576 MIGRATION_STATUS_FAILED
);
1577 migrate_set_error(s
, error
);
1580 static void migrate_fd_cancel(MigrationState
*s
)
1583 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1584 trace_migrate_fd_cancel();
1586 if (s
->rp_state
.from_dst_file
) {
1587 /* shutdown the rp socket, so causing the rp thread to shutdown */
1588 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1592 old_state
= s
->state
;
1593 if (!migration_is_setup_or_active(old_state
)) {
1596 /* If the migration is paused, kick it out of the pause */
1597 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1598 qemu_sem_post(&s
->pause_sem
);
1600 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1601 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1604 * If we're unlucky the migration code might be stuck somewhere in a
1605 * send/write while the network has failed and is waiting to timeout;
1606 * if we've got shutdown(2) available then we can force it to quit.
1607 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1608 * called in a bh, so there is no race against this cancel.
1610 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1611 qemu_file_shutdown(f
);
1613 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1614 Error
*local_err
= NULL
;
1616 bdrv_invalidate_cache_all(&local_err
);
1618 error_report_err(local_err
);
1620 s
->block_inactive
= false;
1625 void add_migration_state_change_notifier(Notifier
*notify
)
1627 notifier_list_add(&migration_state_notifiers
, notify
);
1630 void remove_migration_state_change_notifier(Notifier
*notify
)
1632 notifier_remove(notify
);
1635 bool migration_in_setup(MigrationState
*s
)
1637 return s
->state
== MIGRATION_STATUS_SETUP
;
1640 bool migration_has_finished(MigrationState
*s
)
1642 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1645 bool migration_has_failed(MigrationState
*s
)
1647 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1648 s
->state
== MIGRATION_STATUS_FAILED
);
1651 bool migration_in_postcopy(void)
1653 MigrationState
*s
= migrate_get_current();
1655 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1658 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1660 return migration_in_postcopy() && s
->postcopy_after_devices
;
1663 bool migration_is_idle(void)
1665 MigrationState
*s
= current_migration
;
1672 case MIGRATION_STATUS_NONE
:
1673 case MIGRATION_STATUS_CANCELLED
:
1674 case MIGRATION_STATUS_COMPLETED
:
1675 case MIGRATION_STATUS_FAILED
:
1677 case MIGRATION_STATUS_SETUP
:
1678 case MIGRATION_STATUS_CANCELLING
:
1679 case MIGRATION_STATUS_ACTIVE
:
1680 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1681 case MIGRATION_STATUS_COLO
:
1682 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1683 case MIGRATION_STATUS_DEVICE
:
1685 case MIGRATION_STATUS__MAX
:
1686 g_assert_not_reached();
1692 void migrate_init(MigrationState
*s
)
1695 * Reinitialise all migration state, except
1696 * parameters/capabilities that the user set, and
1701 s
->to_dst_file
= NULL
;
1702 s
->state
= MIGRATION_STATUS_NONE
;
1703 s
->rp_state
.from_dst_file
= NULL
;
1704 s
->rp_state
.error
= false;
1706 s
->pages_per_second
= 0.0;
1708 s
->expected_downtime
= 0;
1710 s
->start_postcopy
= false;
1711 s
->postcopy_after_devices
= false;
1712 s
->migration_thread_running
= false;
1713 error_free(s
->error
);
1716 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1718 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1720 s
->vm_was_running
= false;
1721 s
->iteration_initial_bytes
= 0;
1722 s
->threshold_size
= 0;
1725 static GSList
*migration_blockers
;
1727 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1729 if (only_migratable
) {
1730 error_propagate_prepend(errp
, error_copy(reason
),
1731 "disallowing migration blocker "
1732 "(--only-migratable) for: ");
1736 if (migration_is_idle()) {
1737 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1741 error_propagate_prepend(errp
, error_copy(reason
),
1742 "disallowing migration blocker "
1743 "(migration in progress) for: ");
1747 void migrate_del_blocker(Error
*reason
)
1749 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1752 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1754 Error
*local_err
= NULL
;
1755 static bool once
= true;
1757 if (!deferred_incoming
) {
1758 error_setg(errp
, "For use with '-incoming defer'");
1762 error_setg(errp
, "The incoming migration has already been started");
1765 qemu_start_incoming_migration(uri
, &local_err
);
1768 error_propagate(errp
, local_err
);
1775 void qmp_migrate_recover(const char *uri
, Error
**errp
)
1777 MigrationIncomingState
*mis
= migration_incoming_get_current();
1779 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1780 error_setg(errp
, "Migrate recover can only be run "
1781 "when postcopy is paused.");
1785 if (atomic_cmpxchg(&mis
->postcopy_recover_triggered
,
1786 false, true) == true) {
1787 error_setg(errp
, "Migrate recovery is triggered already");
1792 * Note that this call will never start a real migration; it will
1793 * only re-setup the migration stream and poke existing migration
1794 * to continue using that newly established channel.
1796 qemu_start_incoming_migration(uri
, errp
);
1799 void qmp_migrate_pause(Error
**errp
)
1801 MigrationState
*ms
= migrate_get_current();
1802 MigrationIncomingState
*mis
= migration_incoming_get_current();
1805 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1806 /* Source side, during postcopy */
1807 qemu_mutex_lock(&ms
->qemu_file_lock
);
1808 ret
= qemu_file_shutdown(ms
->to_dst_file
);
1809 qemu_mutex_unlock(&ms
->qemu_file_lock
);
1811 error_setg(errp
, "Failed to pause source migration");
1816 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1817 ret
= qemu_file_shutdown(mis
->from_src_file
);
1819 error_setg(errp
, "Failed to pause destination migration");
1824 error_setg(errp
, "migrate-pause is currently only supported "
1825 "during postcopy-active state");
1828 bool migration_is_blocked(Error
**errp
)
1830 if (qemu_savevm_state_blocked(errp
)) {
1834 if (migration_blockers
) {
1835 error_propagate(errp
, error_copy(migration_blockers
->data
));
1842 /* Returns true if continue to migrate, or false if error detected */
1843 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
1844 bool resume
, Error
**errp
)
1846 Error
*local_err
= NULL
;
1849 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1850 error_setg(errp
, "Cannot resume if there is no "
1851 "paused migration");
1856 * Postcopy recovery won't work well with release-ram
1857 * capability since release-ram will drop the page buffer as
1858 * long as the page is put into the send buffer. So if there
1859 * is a network failure happened, any page buffers that have
1860 * not yet reached the destination VM but have already been
1861 * sent from the source VM will be lost forever. Let's refuse
1862 * the client from resuming such a postcopy migration.
1863 * Luckily release-ram was designed to only be used when src
1864 * and destination VMs are on the same host, so it should be
1867 if (migrate_release_ram()) {
1868 error_setg(errp
, "Postcopy recovery cannot work "
1869 "when release-ram capability is set");
1873 /* This is a resume, skip init status */
1877 if (migration_is_setup_or_active(s
->state
) ||
1878 s
->state
== MIGRATION_STATUS_CANCELLING
||
1879 s
->state
== MIGRATION_STATUS_COLO
) {
1880 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1884 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1885 error_setg(errp
, "Guest is waiting for an incoming migration");
1889 if (migration_is_blocked(errp
)) {
1893 if (blk
|| blk_inc
) {
1894 if (migrate_use_block() || migrate_use_block_incremental()) {
1895 error_setg(errp
, "Command options are incompatible with "
1896 "current migration capabilities");
1899 migrate_set_block_enabled(true, &local_err
);
1901 error_propagate(errp
, local_err
);
1904 s
->must_remove_block_options
= true;
1908 migrate_set_block_incremental(s
, true);
1916 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1917 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1918 bool has_resume
, bool resume
, Error
**errp
)
1920 Error
*local_err
= NULL
;
1921 MigrationState
*s
= migrate_get_current();
1924 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
1925 has_resume
&& resume
, errp
)) {
1926 /* Error detected, put into errp */
1930 if (strstart(uri
, "tcp:", &p
)) {
1931 tcp_start_outgoing_migration(s
, p
, &local_err
);
1933 } else if (strstart(uri
, "rdma:", &p
)) {
1934 rdma_start_outgoing_migration(s
, p
, &local_err
);
1936 } else if (strstart(uri
, "exec:", &p
)) {
1937 exec_start_outgoing_migration(s
, p
, &local_err
);
1938 } else if (strstart(uri
, "unix:", &p
)) {
1939 unix_start_outgoing_migration(s
, p
, &local_err
);
1940 } else if (strstart(uri
, "fd:", &p
)) {
1941 fd_start_outgoing_migration(s
, p
, &local_err
);
1943 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1944 "a valid migration protocol");
1945 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1946 MIGRATION_STATUS_FAILED
);
1947 block_cleanup_parameters(s
);
1952 migrate_fd_error(s
, local_err
);
1953 error_propagate(errp
, local_err
);
1958 void qmp_migrate_cancel(Error
**errp
)
1960 migrate_fd_cancel(migrate_get_current());
1963 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
1965 MigrationState
*s
= migrate_get_current();
1966 if (s
->state
!= state
) {
1967 error_setg(errp
, "Migration not in expected state: %s",
1968 MigrationStatus_str(s
->state
));
1971 qemu_sem_post(&s
->pause_sem
);
1974 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1976 MigrateSetParameters p
= {
1977 .has_xbzrle_cache_size
= true,
1978 .xbzrle_cache_size
= value
,
1981 qmp_migrate_set_parameters(&p
, errp
);
1984 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1986 return migrate_xbzrle_cache_size();
1989 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1991 MigrateSetParameters p
= {
1992 .has_max_bandwidth
= true,
1993 .max_bandwidth
= value
,
1996 qmp_migrate_set_parameters(&p
, errp
);
1999 void qmp_migrate_set_downtime(double value
, Error
**errp
)
2001 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
2002 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
2003 "the range of 0 to %d seconds",
2004 MAX_MIGRATE_DOWNTIME_SECONDS
);
2008 value
*= 1000; /* Convert to milliseconds */
2009 value
= MAX(0, MIN(INT64_MAX
, value
));
2011 MigrateSetParameters p
= {
2012 .has_downtime_limit
= true,
2013 .downtime_limit
= value
,
2016 qmp_migrate_set_parameters(&p
, errp
);
2019 bool migrate_release_ram(void)
2023 s
= migrate_get_current();
2025 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
2028 bool migrate_postcopy_ram(void)
2032 s
= migrate_get_current();
2034 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
2037 bool migrate_postcopy(void)
2039 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2042 bool migrate_auto_converge(void)
2046 s
= migrate_get_current();
2048 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
2051 bool migrate_zero_blocks(void)
2055 s
= migrate_get_current();
2057 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
2060 bool migrate_postcopy_blocktime(void)
2064 s
= migrate_get_current();
2066 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
2069 bool migrate_use_compression(void)
2073 s
= migrate_get_current();
2075 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
2078 int migrate_compress_level(void)
2082 s
= migrate_get_current();
2084 return s
->parameters
.compress_level
;
2087 int migrate_compress_threads(void)
2091 s
= migrate_get_current();
2093 return s
->parameters
.compress_threads
;
2096 int migrate_compress_wait_thread(void)
2100 s
= migrate_get_current();
2102 return s
->parameters
.compress_wait_thread
;
2105 int migrate_decompress_threads(void)
2109 s
= migrate_get_current();
2111 return s
->parameters
.decompress_threads
;
2114 bool migrate_dirty_bitmaps(void)
2118 s
= migrate_get_current();
2120 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
2123 bool migrate_ignore_shared(void)
2127 s
= migrate_get_current();
2129 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
];
2132 bool migrate_use_events(void)
2136 s
= migrate_get_current();
2138 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
2141 bool migrate_use_multifd(void)
2145 s
= migrate_get_current();
2147 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_MULTIFD
];
2150 bool migrate_pause_before_switchover(void)
2154 s
= migrate_get_current();
2156 return s
->enabled_capabilities
[
2157 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
2160 int migrate_multifd_channels(void)
2164 s
= migrate_get_current();
2166 return s
->parameters
.multifd_channels
;
2169 int migrate_use_xbzrle(void)
2173 s
= migrate_get_current();
2175 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
2178 int64_t migrate_xbzrle_cache_size(void)
2182 s
= migrate_get_current();
2184 return s
->parameters
.xbzrle_cache_size
;
2187 static int64_t migrate_max_postcopy_bandwidth(void)
2191 s
= migrate_get_current();
2193 return s
->parameters
.max_postcopy_bandwidth
;
2196 bool migrate_use_block(void)
2200 s
= migrate_get_current();
2202 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
2205 bool migrate_use_return_path(void)
2209 s
= migrate_get_current();
2211 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
2214 bool migrate_use_block_incremental(void)
2218 s
= migrate_get_current();
2220 return s
->parameters
.block_incremental
;
2223 /* migration thread support */
2225 * Something bad happened to the RP stream, mark an error
2226 * The caller shall print or trace something to indicate why
2228 static void mark_source_rp_bad(MigrationState
*s
)
2230 s
->rp_state
.error
= true;
2233 static struct rp_cmd_args
{
2234 ssize_t len
; /* -1 = variable */
2237 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
2238 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
2239 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
2240 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
2241 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
2242 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
2243 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
2244 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
2248 * Process a request for pages received on the return path,
2249 * We're allowed to send more than requested (e.g. to round to our page size)
2250 * and we don't need to send pages that have already been sent.
2252 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
2253 ram_addr_t start
, size_t len
)
2255 long our_host_ps
= getpagesize();
2257 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
2260 * Since we currently insist on matching page sizes, just sanity check
2261 * we're being asked for whole host pages.
2263 if (start
& (our_host_ps
-1) ||
2264 (len
& (our_host_ps
-1))) {
2265 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2266 " len: %zd", __func__
, start
, len
);
2267 mark_source_rp_bad(ms
);
2271 if (ram_save_queue_pages(rbname
, start
, len
)) {
2272 mark_source_rp_bad(ms
);
2276 /* Return true to retry, false to quit */
2277 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
2279 trace_postcopy_pause_return_path();
2281 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
2283 trace_postcopy_pause_return_path_continued();
2288 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
2290 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
2293 error_report("%s: invalid block name '%s'", __func__
, block_name
);
2297 /* Fetch the received bitmap and refresh the dirty bitmap */
2298 return ram_dirty_bitmap_reload(s
, block
);
2301 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
2303 trace_source_return_path_thread_resume_ack(value
);
2305 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
2306 error_report("%s: illegal resume_ack value %"PRIu32
,
2311 /* Now both sides are active. */
2312 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2313 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2315 /* Notify send thread that time to continue send pages */
2316 qemu_sem_post(&s
->rp_state
.rp_sem
);
2322 * Handles messages sent on the return path towards the source VM
2325 static void *source_return_path_thread(void *opaque
)
2327 MigrationState
*ms
= opaque
;
2328 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
2329 uint16_t header_len
, header_type
;
2331 uint32_t tmp32
, sibling_error
;
2332 ram_addr_t start
= 0; /* =0 to silence warning */
2333 size_t len
= 0, expected_len
;
2336 trace_source_return_path_thread_entry();
2337 rcu_register_thread();
2340 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
2341 migration_is_setup_or_active(ms
->state
)) {
2342 trace_source_return_path_thread_loop_top();
2343 header_type
= qemu_get_be16(rp
);
2344 header_len
= qemu_get_be16(rp
);
2346 if (qemu_file_get_error(rp
)) {
2347 mark_source_rp_bad(ms
);
2351 if (header_type
>= MIG_RP_MSG_MAX
||
2352 header_type
== MIG_RP_MSG_INVALID
) {
2353 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2354 header_type
, header_len
);
2355 mark_source_rp_bad(ms
);
2359 if ((rp_cmd_args
[header_type
].len
!= -1 &&
2360 header_len
!= rp_cmd_args
[header_type
].len
) ||
2361 header_len
> sizeof(buf
)) {
2362 error_report("RP: Received '%s' message (0x%04x) with"
2363 "incorrect length %d expecting %zu",
2364 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
2365 (size_t)rp_cmd_args
[header_type
].len
);
2366 mark_source_rp_bad(ms
);
2370 /* We know we've got a valid header by this point */
2371 res
= qemu_get_buffer(rp
, buf
, header_len
);
2372 if (res
!= header_len
) {
2373 error_report("RP: Failed reading data for message 0x%04x"
2374 " read %d expected %d",
2375 header_type
, res
, header_len
);
2376 mark_source_rp_bad(ms
);
2380 /* OK, we have the message and the data */
2381 switch (header_type
) {
2382 case MIG_RP_MSG_SHUT
:
2383 sibling_error
= ldl_be_p(buf
);
2384 trace_source_return_path_thread_shut(sibling_error
);
2385 if (sibling_error
) {
2386 error_report("RP: Sibling indicated error %d", sibling_error
);
2387 mark_source_rp_bad(ms
);
2390 * We'll let the main thread deal with closing the RP
2391 * we could do a shutdown(2) on it, but we're the only user
2392 * anyway, so there's nothing gained.
2396 case MIG_RP_MSG_PONG
:
2397 tmp32
= ldl_be_p(buf
);
2398 trace_source_return_path_thread_pong(tmp32
);
2401 case MIG_RP_MSG_REQ_PAGES
:
2402 start
= ldq_be_p(buf
);
2403 len
= ldl_be_p(buf
+ 8);
2404 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
2407 case MIG_RP_MSG_REQ_PAGES_ID
:
2408 expected_len
= 12 + 1; /* header + termination */
2410 if (header_len
>= expected_len
) {
2411 start
= ldq_be_p(buf
);
2412 len
= ldl_be_p(buf
+ 8);
2413 /* Now we expect an idstr */
2414 tmp32
= buf
[12]; /* Length of the following idstr */
2415 buf
[13 + tmp32
] = '\0';
2416 expected_len
+= tmp32
;
2418 if (header_len
!= expected_len
) {
2419 error_report("RP: Req_Page_id with length %d expecting %zd",
2420 header_len
, expected_len
);
2421 mark_source_rp_bad(ms
);
2424 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
2427 case MIG_RP_MSG_RECV_BITMAP
:
2428 if (header_len
< 1) {
2429 error_report("%s: missing block name", __func__
);
2430 mark_source_rp_bad(ms
);
2433 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2434 buf
[buf
[0] + 1] = '\0';
2435 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
2436 mark_source_rp_bad(ms
);
2441 case MIG_RP_MSG_RESUME_ACK
:
2442 tmp32
= ldl_be_p(buf
);
2443 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
2444 mark_source_rp_bad(ms
);
2455 res
= qemu_file_get_error(rp
);
2459 * Maybe there is something we can do: it looks like a
2460 * network down issue, and we pause for a recovery.
2462 if (postcopy_pause_return_path_thread(ms
)) {
2463 /* Reload rp, reset the rest */
2464 if (rp
!= ms
->rp_state
.from_dst_file
) {
2466 rp
= ms
->rp_state
.from_dst_file
;
2468 ms
->rp_state
.error
= false;
2473 trace_source_return_path_thread_bad_end();
2474 mark_source_rp_bad(ms
);
2477 trace_source_return_path_thread_end();
2478 ms
->rp_state
.from_dst_file
= NULL
;
2480 rcu_unregister_thread();
2484 static int open_return_path_on_source(MigrationState
*ms
,
2488 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
2489 if (!ms
->rp_state
.from_dst_file
) {
2493 trace_open_return_path_on_source();
2495 if (!create_thread
) {
2500 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
2501 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
2503 trace_open_return_path_on_source_continue();
2508 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2509 static int await_return_path_close_on_source(MigrationState
*ms
)
2512 * If this is a normal exit then the destination will send a SHUT and the
2513 * rp_thread will exit, however if there's an error we need to cause
2516 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
2518 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2519 * waiting for the destination.
2521 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
2522 mark_source_rp_bad(ms
);
2524 trace_await_return_path_close_on_source_joining();
2525 qemu_thread_join(&ms
->rp_state
.rp_thread
);
2526 trace_await_return_path_close_on_source_close();
2527 return ms
->rp_state
.error
;
2531 * Switch from normal iteration to postcopy
2532 * Returns non-0 on error
2534 static int postcopy_start(MigrationState
*ms
)
2537 QIOChannelBuffer
*bioc
;
2539 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2540 int64_t bandwidth
= migrate_max_postcopy_bandwidth();
2541 bool restart_block
= false;
2542 int cur_state
= MIGRATION_STATUS_ACTIVE
;
2543 if (!migrate_pause_before_switchover()) {
2544 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
2545 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2548 trace_postcopy_start();
2549 qemu_mutex_lock_iothread();
2550 trace_postcopy_start_set_run();
2552 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2553 global_state_store();
2554 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2559 ret
= migration_maybe_pause(ms
, &cur_state
,
2560 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2565 ret
= bdrv_inactivate_all();
2569 restart_block
= true;
2572 * Cause any non-postcopiable, but iterative devices to
2573 * send out their final data.
2575 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2578 * in Finish migrate and with the io-lock held everything should
2579 * be quiet, but we've potentially still got dirty pages and we
2580 * need to tell the destination to throw any pages it's already received
2583 if (migrate_postcopy_ram()) {
2584 if (ram_postcopy_send_discard_bitmap(ms
)) {
2585 error_report("postcopy send discard bitmap failed");
2591 * send rest of state - note things that are doing postcopy
2592 * will notice we're in POSTCOPY_ACTIVE and not actually
2593 * wrap their state up here
2595 /* 0 max-postcopy-bandwidth means unlimited */
2597 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
2599 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
2601 if (migrate_postcopy_ram()) {
2602 /* Ping just for debugging, helps line traces up */
2603 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
2607 * While loading the device state we may trigger page transfer
2608 * requests and the fd must be free to process those, and thus
2609 * the destination must read the whole device state off the fd before
2610 * it starts processing it. Unfortunately the ad-hoc migration format
2611 * doesn't allow the destination to know the size to read without fully
2612 * parsing it through each devices load-state code (especially the open
2613 * coded devices that use get/put).
2614 * So we wrap the device state up in a package with a length at the start;
2615 * to do this we use a qemu_buf to hold the whole of the device state.
2617 bioc
= qio_channel_buffer_new(4096);
2618 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
2619 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
2620 object_unref(OBJECT(bioc
));
2623 * Make sure the receiver can get incoming pages before we send the rest
2626 qemu_savevm_send_postcopy_listen(fb
);
2628 qemu_savevm_state_complete_precopy(fb
, false, false);
2629 if (migrate_postcopy_ram()) {
2630 qemu_savevm_send_ping(fb
, 3);
2633 qemu_savevm_send_postcopy_run(fb
);
2635 /* <><> end of stuff going into the package */
2637 /* Last point of recovery; as soon as we send the package the destination
2638 * can open devices and potentially start running.
2639 * Lets just check again we've not got any errors.
2641 ret
= qemu_file_get_error(ms
->to_dst_file
);
2643 error_report("postcopy_start: Migration stream errored (pre package)");
2647 restart_block
= false;
2649 /* Now send that blob */
2650 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2655 /* Send a notify to give a chance for anything that needs to happen
2656 * at the transition to postcopy and after the device state; in particular
2657 * spice needs to trigger a transition now
2659 ms
->postcopy_after_devices
= true;
2660 notifier_list_notify(&migration_state_notifiers
, ms
);
2662 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2664 qemu_mutex_unlock_iothread();
2666 if (migrate_postcopy_ram()) {
2668 * Although this ping is just for debug, it could potentially be
2669 * used for getting a better measurement of downtime at the source.
2671 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2674 if (migrate_release_ram()) {
2675 ram_postcopy_migrated_memory_release(ms
);
2678 ret
= qemu_file_get_error(ms
->to_dst_file
);
2680 error_report("postcopy_start: Migration stream errored");
2681 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2682 MIGRATION_STATUS_FAILED
);
2690 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2691 MIGRATION_STATUS_FAILED
);
2692 if (restart_block
) {
2693 /* A failure happened early enough that we know the destination hasn't
2694 * accessed block devices, so we're safe to recover.
2696 Error
*local_err
= NULL
;
2698 bdrv_invalidate_cache_all(&local_err
);
2700 error_report_err(local_err
);
2703 qemu_mutex_unlock_iothread();
2708 * migration_maybe_pause: Pause if required to by
2709 * migrate_pause_before_switchover called with the iothread locked
2710 * Returns: 0 on success
2712 static int migration_maybe_pause(MigrationState
*s
,
2713 int *current_active_state
,
2716 if (!migrate_pause_before_switchover()) {
2720 /* Since leaving this state is not atomic with posting the semaphore
2721 * it's possible that someone could have issued multiple migrate_continue
2722 * and the semaphore is incorrectly positive at this point;
2723 * the docs say it's undefined to reinit a semaphore that's already
2724 * init'd, so use timedwait to eat up any existing posts.
2726 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2727 /* This block intentionally left blank */
2730 qemu_mutex_unlock_iothread();
2731 migrate_set_state(&s
->state
, *current_active_state
,
2732 MIGRATION_STATUS_PRE_SWITCHOVER
);
2733 qemu_sem_wait(&s
->pause_sem
);
2734 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2736 *current_active_state
= new_state
;
2737 qemu_mutex_lock_iothread();
2739 return s
->state
== new_state
? 0 : -EINVAL
;
2743 * migration_completion: Used by migration_thread when there's not much left.
2744 * The caller 'breaks' the loop when this returns.
2746 * @s: Current migration state
2748 static void migration_completion(MigrationState
*s
)
2751 int current_active_state
= s
->state
;
2753 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2754 qemu_mutex_lock_iothread();
2755 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2756 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2757 s
->vm_was_running
= runstate_is_running();
2758 ret
= global_state_store();
2761 bool inactivate
= !migrate_colo_enabled();
2762 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2764 ret
= migration_maybe_pause(s
, ¤t_active_state
,
2765 MIGRATION_STATUS_DEVICE
);
2768 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
2769 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
2772 if (inactivate
&& ret
>= 0) {
2773 s
->block_inactive
= true;
2776 qemu_mutex_unlock_iothread();
2781 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2782 trace_migration_completion_postcopy_end();
2784 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
2785 trace_migration_completion_postcopy_end_after_complete();
2789 * If rp was opened we must clean up the thread before
2790 * cleaning everything else up (since if there are no failures
2791 * it will wait for the destination to send it's status in
2794 if (s
->rp_state
.from_dst_file
) {
2796 trace_migration_return_path_end_before();
2797 rp_error
= await_return_path_close_on_source(s
);
2798 trace_migration_return_path_end_after(rp_error
);
2800 goto fail_invalidate
;
2804 if (qemu_file_get_error(s
->to_dst_file
)) {
2805 trace_migration_completion_file_err();
2806 goto fail_invalidate
;
2809 if (!migrate_colo_enabled()) {
2810 migrate_set_state(&s
->state
, current_active_state
,
2811 MIGRATION_STATUS_COMPLETED
);
2817 /* If not doing postcopy, vm_start() will be called: let's regain
2818 * control on images.
2820 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
2821 s
->state
== MIGRATION_STATUS_DEVICE
) {
2822 Error
*local_err
= NULL
;
2824 qemu_mutex_lock_iothread();
2825 bdrv_invalidate_cache_all(&local_err
);
2827 error_report_err(local_err
);
2829 s
->block_inactive
= false;
2831 qemu_mutex_unlock_iothread();
2835 migrate_set_state(&s
->state
, current_active_state
,
2836 MIGRATION_STATUS_FAILED
);
2839 bool migrate_colo_enabled(void)
2841 MigrationState
*s
= migrate_get_current();
2842 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
2845 typedef enum MigThrError
{
2846 /* No error detected */
2847 MIG_THR_ERR_NONE
= 0,
2848 /* Detected error, but resumed successfully */
2849 MIG_THR_ERR_RECOVERED
= 1,
2850 /* Detected fatal error, need to exit */
2851 MIG_THR_ERR_FATAL
= 2,
2854 static int postcopy_resume_handshake(MigrationState
*s
)
2856 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
2858 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2859 qemu_sem_wait(&s
->rp_state
.rp_sem
);
2862 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2869 /* Return zero if success, or <0 for error */
2870 static int postcopy_do_resume(MigrationState
*s
)
2875 * Call all the resume_prepare() hooks, so that modules can be
2876 * ready for the migration resume.
2878 ret
= qemu_savevm_state_resume_prepare(s
);
2880 error_report("%s: resume_prepare() failure detected: %d",
2886 * Last handshake with destination on the resume (destination will
2887 * switch to postcopy-active afterwards)
2889 ret
= postcopy_resume_handshake(s
);
2891 error_report("%s: handshake failed: %d", __func__
, ret
);
2899 * We don't return until we are in a safe state to continue current
2900 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2901 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2903 static MigThrError
postcopy_pause(MigrationState
*s
)
2905 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2910 migrate_set_state(&s
->state
, s
->state
,
2911 MIGRATION_STATUS_POSTCOPY_PAUSED
);
2913 /* Current channel is possibly broken. Release it. */
2914 assert(s
->to_dst_file
);
2915 qemu_mutex_lock(&s
->qemu_file_lock
);
2916 file
= s
->to_dst_file
;
2917 s
->to_dst_file
= NULL
;
2918 qemu_mutex_unlock(&s
->qemu_file_lock
);
2920 qemu_file_shutdown(file
);
2923 error_report("Detected IO failure for postcopy. "
2924 "Migration paused.");
2927 * We wait until things fixed up. Then someone will setup the
2928 * status back for us.
2930 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2931 qemu_sem_wait(&s
->postcopy_pause_sem
);
2934 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2935 /* Woken up by a recover procedure. Give it a shot */
2938 * Firstly, let's wake up the return path now, with a new
2939 * return path channel.
2941 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
2943 /* Do the resume logic */
2944 if (postcopy_do_resume(s
) == 0) {
2945 /* Let's continue! */
2946 trace_postcopy_pause_continued();
2947 return MIG_THR_ERR_RECOVERED
;
2950 * Something wrong happened during the recovery, let's
2951 * pause again. Pause is always better than throwing
2957 /* This is not right... Time to quit. */
2958 return MIG_THR_ERR_FATAL
;
2963 static MigThrError
migration_detect_error(MigrationState
*s
)
2966 int state
= s
->state
;
2968 if (state
== MIGRATION_STATUS_CANCELLING
||
2969 state
== MIGRATION_STATUS_CANCELLED
) {
2970 /* End the migration, but don't set the state to failed */
2971 return MIG_THR_ERR_FATAL
;
2974 /* Try to detect any file errors */
2975 ret
= qemu_file_get_error(s
->to_dst_file
);
2978 /* Everything is fine */
2979 return MIG_THR_ERR_NONE
;
2982 if (state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
== -EIO
) {
2984 * For postcopy, we allow the network to be down for a
2985 * while. After that, it can be continued by a
2988 return postcopy_pause(s
);
2991 * For precopy (or postcopy with error outside IO), we fail
2994 migrate_set_state(&s
->state
, state
, MIGRATION_STATUS_FAILED
);
2995 trace_migration_thread_file_err();
2997 /* Time to stop the migration, now. */
2998 return MIG_THR_ERR_FATAL
;
3002 /* How many bytes have we transferred since the beggining of the migration */
3003 static uint64_t migration_total_bytes(MigrationState
*s
)
3005 return qemu_ftell(s
->to_dst_file
) + ram_counters
.multifd_bytes
;
3008 static void migration_calculate_complete(MigrationState
*s
)
3010 uint64_t bytes
= migration_total_bytes(s
);
3011 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3012 int64_t transfer_time
;
3014 s
->total_time
= end_time
- s
->start_time
;
3017 * It's still not set, so we are precopy migration. For
3018 * postcopy, downtime is calculated during postcopy_start().
3020 s
->downtime
= end_time
- s
->downtime_start
;
3023 transfer_time
= s
->total_time
- s
->setup_time
;
3024 if (transfer_time
) {
3025 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
3029 static void migration_update_counters(MigrationState
*s
,
3030 int64_t current_time
)
3032 uint64_t transferred
, transferred_pages
, time_spent
;
3033 uint64_t current_bytes
; /* bytes transferred since the beginning */
3036 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
3040 current_bytes
= migration_total_bytes(s
);
3041 transferred
= current_bytes
- s
->iteration_initial_bytes
;
3042 time_spent
= current_time
- s
->iteration_start_time
;
3043 bandwidth
= (double)transferred
/ time_spent
;
3044 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
3046 s
->mbps
= (((double) transferred
* 8.0) /
3047 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
3049 transferred_pages
= ram_get_total_transferred_pages() -
3050 s
->iteration_initial_pages
;
3051 s
->pages_per_second
= (double) transferred_pages
/
3052 (((double) time_spent
/ 1000.0));
3055 * if we haven't sent anything, we don't want to
3056 * recalculate. 10000 is a small enough number for our purposes
3058 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
3059 s
->expected_downtime
= ram_counters
.remaining
/ bandwidth
;
3062 qemu_file_reset_rate_limit(s
->to_dst_file
);
3064 s
->iteration_start_time
= current_time
;
3065 s
->iteration_initial_bytes
= current_bytes
;
3066 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
3068 trace_migrate_transferred(transferred
, time_spent
,
3069 bandwidth
, s
->threshold_size
);
3072 /* Migration thread iteration status */
3074 MIG_ITERATE_RESUME
, /* Resume current iteration */
3075 MIG_ITERATE_SKIP
, /* Skip current iteration */
3076 MIG_ITERATE_BREAK
, /* Break the loop */
3080 * Return true if continue to the next iteration directly, false
3083 static MigIterateState
migration_iteration_run(MigrationState
*s
)
3085 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
3086 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
3088 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
3089 &pend_compat
, &pend_post
);
3090 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
3092 trace_migrate_pending(pending_size
, s
->threshold_size
,
3093 pend_pre
, pend_compat
, pend_post
);
3095 if (pending_size
&& pending_size
>= s
->threshold_size
) {
3096 /* Still a significant amount to transfer */
3097 if (migrate_postcopy() && !in_postcopy
&&
3098 pend_pre
<= s
->threshold_size
&&
3099 atomic_read(&s
->start_postcopy
)) {
3100 if (postcopy_start(s
)) {
3101 error_report("%s: postcopy failed to start", __func__
);
3103 return MIG_ITERATE_SKIP
;
3105 /* Just another iteration step */
3106 qemu_savevm_state_iterate(s
->to_dst_file
,
3107 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3109 trace_migration_thread_low_pending(pending_size
);
3110 migration_completion(s
);
3111 return MIG_ITERATE_BREAK
;
3114 return MIG_ITERATE_RESUME
;
3117 static void migration_iteration_finish(MigrationState
*s
)
3119 /* If we enabled cpu throttling for auto-converge, turn it off. */
3120 cpu_throttle_stop();
3122 qemu_mutex_lock_iothread();
3124 case MIGRATION_STATUS_COMPLETED
:
3125 migration_calculate_complete(s
);
3126 runstate_set(RUN_STATE_POSTMIGRATE
);
3129 case MIGRATION_STATUS_ACTIVE
:
3131 * We should really assert here, but since it's during
3132 * migration, let's try to reduce the usage of assertions.
3134 if (!migrate_colo_enabled()) {
3135 error_report("%s: critical error: calling COLO code without "
3136 "COLO enabled", __func__
);
3138 migrate_start_colo_process(s
);
3140 * Fixme: we will run VM in COLO no matter its old running state.
3141 * After exited COLO, we will keep running.
3143 s
->vm_was_running
= true;
3145 case MIGRATION_STATUS_FAILED
:
3146 case MIGRATION_STATUS_CANCELLED
:
3147 case MIGRATION_STATUS_CANCELLING
:
3148 if (s
->vm_was_running
) {
3151 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
3152 runstate_set(RUN_STATE_POSTMIGRATE
);
3158 /* Should not reach here, but if so, forgive the VM. */
3159 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3162 migrate_fd_cleanup_schedule(s
);
3163 qemu_mutex_unlock_iothread();
3166 void migration_make_urgent_request(void)
3168 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
3171 void migration_consume_urgent_request(void)
3173 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
3177 * Master migration thread on the source VM.
3178 * It drives the migration and pumps the data down the outgoing channel.
3180 static void *migration_thread(void *opaque
)
3182 MigrationState
*s
= opaque
;
3183 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3184 MigThrError thr_error
;
3185 bool urgent
= false;
3187 rcu_register_thread();
3189 object_ref(OBJECT(s
));
3190 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3192 qemu_savevm_state_header(s
->to_dst_file
);
3195 * If we opened the return path, we need to make sure dst has it
3198 if (s
->rp_state
.from_dst_file
) {
3199 /* Now tell the dest that it should open its end so it can reply */
3200 qemu_savevm_send_open_return_path(s
->to_dst_file
);
3202 /* And do a ping that will make stuff easier to debug */
3203 qemu_savevm_send_ping(s
->to_dst_file
, 1);
3206 if (migrate_postcopy()) {
3208 * Tell the destination that we *might* want to do postcopy later;
3209 * if the other end can't do postcopy it should fail now, nice and
3212 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
3215 if (migrate_colo_enabled()) {
3216 /* Notify migration destination that we enable COLO */
3217 qemu_savevm_send_colo_enable(s
->to_dst_file
);
3220 qemu_savevm_state_setup(s
->to_dst_file
);
3222 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3223 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3224 MIGRATION_STATUS_ACTIVE
);
3226 trace_migration_thread_setup_complete();
3228 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
3229 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3230 int64_t current_time
;
3232 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
3233 MigIterateState iter_state
= migration_iteration_run(s
);
3234 if (iter_state
== MIG_ITERATE_SKIP
) {
3236 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3242 * Try to detect any kind of failures, and see whether we
3243 * should stop the migration now.
3245 thr_error
= migration_detect_error(s
);
3246 if (thr_error
== MIG_THR_ERR_FATAL
) {
3247 /* Stop migration */
3249 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
3251 * Just recovered from a e.g. network failure, reset all
3252 * the local variables. This is important to avoid
3253 * breaking transferred_bytes and bandwidth calculation
3255 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3256 s
->iteration_initial_bytes
= 0;
3259 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3261 migration_update_counters(s
, current_time
);
3264 if (qemu_file_rate_limit(s
->to_dst_file
)) {
3265 /* Wait for a delay to do rate limiting OR
3266 * something urgent to post the semaphore.
3268 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- current_time
;
3269 trace_migration_thread_ratelimit_pre(ms
);
3270 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
3271 /* We were worken by one or more urgent things but
3272 * the timedwait will have consumed one of them.
3273 * The service routine for the urgent wake will dec
3274 * the semaphore itself for each item it consumes,
3275 * so add this one we just eat back.
3277 qemu_sem_post(&s
->rate_limit_sem
);
3280 trace_migration_thread_ratelimit_post(urgent
);
3284 trace_migration_thread_after_loop();
3285 migration_iteration_finish(s
);
3286 object_unref(OBJECT(s
));
3287 rcu_unregister_thread();
3291 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
3294 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
3296 s
->expected_downtime
= s
->parameters
.downtime_limit
;
3297 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup_bh
, s
);
3299 migrate_fd_error(s
, error_in
);
3300 migrate_fd_cleanup(s
);
3305 /* This is a resumed migration */
3306 rate_limit
= INT64_MAX
;
3308 /* This is a fresh new migration */
3309 rate_limit
= s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
;
3311 /* Notify before starting migration thread */
3312 notifier_list_notify(&migration_state_notifiers
, s
);
3315 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
3316 qemu_file_set_blocking(s
->to_dst_file
, true);
3319 * Open the return path. For postcopy, it is used exclusively. For
3320 * precopy, only if user specified "return-path" capability would
3321 * QEMU uses the return path.
3323 if (migrate_postcopy_ram() || migrate_use_return_path()) {
3324 if (open_return_path_on_source(s
, !resume
)) {
3325 error_report("Unable to open return-path for postcopy");
3326 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
3327 migrate_fd_cleanup(s
);
3333 /* Wakeup the main migration thread to do the recovery */
3334 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
3335 MIGRATION_STATUS_POSTCOPY_RECOVER
);
3336 qemu_sem_post(&s
->postcopy_pause_sem
);
3340 if (multifd_save_setup() != 0) {
3341 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3342 MIGRATION_STATUS_FAILED
);
3343 migrate_fd_cleanup(s
);
3346 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
3347 QEMU_THREAD_JOINABLE
);
3348 s
->migration_thread_running
= true;
3351 void migration_global_dump(Monitor
*mon
)
3353 MigrationState
*ms
= migrate_get_current();
3355 monitor_printf(mon
, "globals:\n");
3356 monitor_printf(mon
, "store-global-state: %s\n",
3357 ms
->store_global_state
? "on" : "off");
3358 monitor_printf(mon
, "only-migratable: %s\n",
3359 only_migratable
? "on" : "off");
3360 monitor_printf(mon
, "send-configuration: %s\n",
3361 ms
->send_configuration
? "on" : "off");
3362 monitor_printf(mon
, "send-section-footer: %s\n",
3363 ms
->send_section_footer
? "on" : "off");
3364 monitor_printf(mon
, "decompress-error-check: %s\n",
3365 ms
->decompress_error_check
? "on" : "off");
3366 monitor_printf(mon
, "clear-bitmap-shift: %u\n",
3367 ms
->clear_bitmap_shift
);
3370 #define DEFINE_PROP_MIG_CAP(name, x) \
3371 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3373 static Property migration_properties
[] = {
3374 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
3375 store_global_state
, true),
3376 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
3377 send_configuration
, true),
3378 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
3379 send_section_footer
, true),
3380 DEFINE_PROP_BOOL("decompress-error-check", MigrationState
,
3381 decompress_error_check
, true),
3382 DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState
,
3383 clear_bitmap_shift
, CLEAR_BITMAP_SHIFT_DEFAULT
),
3385 /* Migration parameters */
3386 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
3387 parameters
.compress_level
,
3388 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
3389 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
3390 parameters
.compress_threads
,
3391 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
3392 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState
,
3393 parameters
.compress_wait_thread
, true),
3394 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
3395 parameters
.decompress_threads
,
3396 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
3397 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
3398 parameters
.cpu_throttle_initial
,
3399 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
3400 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
3401 parameters
.cpu_throttle_increment
,
3402 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
3403 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
3404 parameters
.max_bandwidth
, MAX_THROTTLE
),
3405 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
3406 parameters
.downtime_limit
,
3407 DEFAULT_MIGRATE_SET_DOWNTIME
),
3408 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
3409 parameters
.x_checkpoint_delay
,
3410 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
3411 DEFINE_PROP_UINT8("multifd-channels", MigrationState
,
3412 parameters
.multifd_channels
,
3413 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
3414 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
3415 parameters
.xbzrle_cache_size
,
3416 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
3417 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState
,
3418 parameters
.max_postcopy_bandwidth
,
3419 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH
),
3420 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState
,
3421 parameters
.max_cpu_throttle
,
3422 DEFAULT_MIGRATE_MAX_CPU_THROTTLE
),
3423 DEFINE_PROP_SIZE("announce-initial", MigrationState
,
3424 parameters
.announce_initial
,
3425 DEFAULT_MIGRATE_ANNOUNCE_INITIAL
),
3426 DEFINE_PROP_SIZE("announce-max", MigrationState
,
3427 parameters
.announce_max
,
3428 DEFAULT_MIGRATE_ANNOUNCE_MAX
),
3429 DEFINE_PROP_SIZE("announce-rounds", MigrationState
,
3430 parameters
.announce_rounds
,
3431 DEFAULT_MIGRATE_ANNOUNCE_ROUNDS
),
3432 DEFINE_PROP_SIZE("announce-step", MigrationState
,
3433 parameters
.announce_step
,
3434 DEFAULT_MIGRATE_ANNOUNCE_STEP
),
3436 /* Migration capabilities */
3437 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
3438 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
3439 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
3440 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
3441 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
3442 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
3443 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
3444 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
3445 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
3446 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
3447 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
3448 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD
),
3450 DEFINE_PROP_END_OF_LIST(),
3453 static void migration_class_init(ObjectClass
*klass
, void *data
)
3455 DeviceClass
*dc
= DEVICE_CLASS(klass
);
3457 dc
->user_creatable
= false;
3458 dc
->props
= migration_properties
;
3461 static void migration_instance_finalize(Object
*obj
)
3463 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3464 MigrationParameters
*params
= &ms
->parameters
;
3466 qemu_mutex_destroy(&ms
->error_mutex
);
3467 qemu_mutex_destroy(&ms
->qemu_file_lock
);
3468 g_free(params
->tls_hostname
);
3469 g_free(params
->tls_creds
);
3470 qemu_sem_destroy(&ms
->rate_limit_sem
);
3471 qemu_sem_destroy(&ms
->pause_sem
);
3472 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
3473 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
3474 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
3475 error_free(ms
->error
);
3478 static void migration_instance_init(Object
*obj
)
3480 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3481 MigrationParameters
*params
= &ms
->parameters
;
3483 ms
->state
= MIGRATION_STATUS_NONE
;
3485 ms
->pages_per_second
= -1;
3486 qemu_sem_init(&ms
->pause_sem
, 0);
3487 qemu_mutex_init(&ms
->error_mutex
);
3489 params
->tls_hostname
= g_strdup("");
3490 params
->tls_creds
= g_strdup("");
3492 /* Set has_* up only for parameter checks */
3493 params
->has_compress_level
= true;
3494 params
->has_compress_threads
= true;
3495 params
->has_decompress_threads
= true;
3496 params
->has_cpu_throttle_initial
= true;
3497 params
->has_cpu_throttle_increment
= true;
3498 params
->has_max_bandwidth
= true;
3499 params
->has_downtime_limit
= true;
3500 params
->has_x_checkpoint_delay
= true;
3501 params
->has_block_incremental
= true;
3502 params
->has_multifd_channels
= true;
3503 params
->has_xbzrle_cache_size
= true;
3504 params
->has_max_postcopy_bandwidth
= true;
3505 params
->has_max_cpu_throttle
= true;
3506 params
->has_announce_initial
= true;
3507 params
->has_announce_max
= true;
3508 params
->has_announce_rounds
= true;
3509 params
->has_announce_step
= true;
3511 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
3512 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
3513 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
3514 qemu_sem_init(&ms
->rate_limit_sem
, 0);
3515 qemu_mutex_init(&ms
->qemu_file_lock
);
3519 * Return true if check pass, false otherwise. Error will be put
3520 * inside errp if provided.
3522 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
3524 MigrationCapabilityStatusList
*head
= NULL
;
3525 /* Assuming all off */
3526 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
3529 if (!migrate_params_check(&ms
->parameters
, errp
)) {
3533 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
3534 if (ms
->enabled_capabilities
[i
]) {
3535 head
= migrate_cap_add(head
, i
, true);
3539 ret
= migrate_caps_check(cap_list
, head
, errp
);
3541 /* It works with head == NULL */
3542 qapi_free_MigrationCapabilityStatusList(head
);
3547 static const TypeInfo migration_type
= {
3548 .name
= TYPE_MIGRATION
,
3550 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3551 * not created using qdev_create(), it is not attached to the qdev
3552 * device tree, and it is never realized.
3554 * TODO: Make this TYPE_OBJECT once QOM provides something like
3555 * TYPE_DEVICE's "-global" properties.
3557 .parent
= TYPE_DEVICE
,
3558 .class_init
= migration_class_init
,
3559 .class_size
= sizeof(MigrationClass
),
3560 .instance_size
= sizeof(MigrationState
),
3561 .instance_init
= migration_instance_init
,
3562 .instance_finalize
= migration_instance_finalize
,
3565 static void register_migration_types(void)
3567 type_register_static(&migration_type
);
3570 type_init(register_migration_types
);