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"
25 #include "sysemu/runstate.h"
26 #include "sysemu/sysemu.h"
27 #include "sysemu/cpu-throttle.h"
30 #include "ram-compress.h"
31 #include "migration/global_state.h"
32 #include "migration/misc.h"
33 #include "migration.h"
34 #include "migration-stats.h"
36 #include "qemu-file.h"
38 #include "migration/vmstate.h"
39 #include "block/block.h"
40 #include "qapi/error.h"
41 #include "qapi/clone-visitor.h"
42 #include "qapi/qapi-visit-migration.h"
43 #include "qapi/qapi-visit-sockets.h"
44 #include "qapi/qapi-commands-migration.h"
45 #include "qapi/qapi-events-migration.h"
46 #include "qapi/qmp/qerror.h"
47 #include "qapi/qmp/qnull.h"
50 #include "postcopy-ram.h"
51 #include "qemu/thread.h"
53 #include "exec/target_page.h"
54 #include "io/channel-buffer.h"
55 #include "io/channel-tls.h"
56 #include "migration/colo.h"
57 #include "hw/boards.h"
58 #include "monitor/monitor.h"
59 #include "net/announce.h"
60 #include "qemu/queue.h"
62 #include "threadinfo.h"
63 #include "qemu/yank.h"
64 #include "sysemu/cpus.h"
65 #include "yank_functions.h"
66 #include "sysemu/qtest.h"
68 #include "sysemu/dirtylimit.h"
70 static NotifierList migration_state_notifiers
=
71 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
73 /* Messages sent on the return path from destination to source */
74 enum mig_rp_message_type
{
75 MIG_RP_MSG_INVALID
= 0, /* Must be 0 */
76 MIG_RP_MSG_SHUT
, /* sibling will not send any more RP messages */
77 MIG_RP_MSG_PONG
, /* Response to a PING; data (seq: be32 ) */
79 MIG_RP_MSG_REQ_PAGES_ID
, /* data (start: be64, len: be32, id: string) */
80 MIG_RP_MSG_REQ_PAGES
, /* data (start: be64, len: be32) */
81 MIG_RP_MSG_RECV_BITMAP
, /* send recved_bitmap back to source */
82 MIG_RP_MSG_RESUME_ACK
, /* tell source that we are ready to resume */
83 MIG_RP_MSG_SWITCHOVER_ACK
, /* Tell source it's OK to do switchover */
88 /* When we add fault tolerance, we could have several
89 migrations at once. For now we don't need to add
90 dynamic creation of migration */
92 static MigrationState
*current_migration
;
93 static MigrationIncomingState
*current_incoming
;
95 static GSList
*migration_blockers
;
97 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
98 static int migration_maybe_pause(MigrationState
*s
,
99 int *current_active_state
,
101 static void migrate_fd_cancel(MigrationState
*s
);
102 static int await_return_path_close_on_source(MigrationState
*s
);
104 static bool migration_needs_multiple_sockets(void)
106 return migrate_multifd() || migrate_postcopy_preempt();
109 static bool uri_supports_multi_channels(const char *uri
)
111 return strstart(uri
, "tcp:", NULL
) || strstart(uri
, "unix:", NULL
) ||
112 strstart(uri
, "vsock:", NULL
);
116 migration_channels_and_uri_compatible(const char *uri
, Error
**errp
)
118 if (migration_needs_multiple_sockets() &&
119 !uri_supports_multi_channels(uri
)) {
120 error_setg(errp
, "Migration requires multi-channel URIs (e.g. tcp)");
127 static gint
page_request_addr_cmp(gconstpointer ap
, gconstpointer bp
)
129 uintptr_t a
= (uintptr_t) ap
, b
= (uintptr_t) bp
;
131 return (a
> b
) - (a
< b
);
134 void migration_object_init(void)
136 /* This can only be called once. */
137 assert(!current_migration
);
138 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
141 * Init the migrate incoming object as well no matter whether
142 * we'll use it or not.
144 assert(!current_incoming
);
145 current_incoming
= g_new0(MigrationIncomingState
, 1);
146 current_incoming
->state
= MIGRATION_STATUS_NONE
;
147 current_incoming
->postcopy_remote_fds
=
148 g_array_new(FALSE
, TRUE
, sizeof(struct PostCopyFD
));
149 qemu_mutex_init(¤t_incoming
->rp_mutex
);
150 qemu_mutex_init(¤t_incoming
->postcopy_prio_thread_mutex
);
151 qemu_event_init(¤t_incoming
->main_thread_load_event
, false);
152 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_dst
, 0);
153 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fault
, 0);
154 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fast_load
, 0);
155 qemu_sem_init(¤t_incoming
->postcopy_qemufile_dst_done
, 0);
157 qemu_mutex_init(¤t_incoming
->page_request_mutex
);
158 qemu_cond_init(¤t_incoming
->page_request_cond
);
159 current_incoming
->page_requested
= g_tree_new(page_request_addr_cmp
);
161 migration_object_check(current_migration
, &error_fatal
);
165 dirty_bitmap_mig_init();
168 void migration_cancel(const Error
*error
)
171 migrate_set_error(current_migration
, error
);
173 if (migrate_dirty_limit()) {
174 qmp_cancel_vcpu_dirty_limit(false, -1, NULL
);
176 migrate_fd_cancel(current_migration
);
179 void migration_shutdown(void)
182 * When the QEMU main thread exit, the COLO thread
183 * may wait a semaphore. So, we should wakeup the
184 * COLO thread before migration shutdown.
188 * Cancel the current migration - that will (eventually)
189 * stop the migration using this structure
191 migration_cancel(NULL
);
192 object_unref(OBJECT(current_migration
));
195 * Cancel outgoing migration of dirty bitmaps. It should
196 * at least unref used block nodes.
198 dirty_bitmap_mig_cancel_outgoing();
201 * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
202 * are non-critical data, and their loss never considered as
205 dirty_bitmap_mig_cancel_incoming();
209 MigrationState
*migrate_get_current(void)
211 /* This can only be called after the object created. */
212 assert(current_migration
);
213 return current_migration
;
216 MigrationIncomingState
*migration_incoming_get_current(void)
218 assert(current_incoming
);
219 return current_incoming
;
222 void migration_incoming_transport_cleanup(MigrationIncomingState
*mis
)
224 if (mis
->socket_address_list
) {
225 qapi_free_SocketAddressList(mis
->socket_address_list
);
226 mis
->socket_address_list
= NULL
;
229 if (mis
->transport_cleanup
) {
230 mis
->transport_cleanup(mis
->transport_data
);
231 mis
->transport_data
= mis
->transport_cleanup
= NULL
;
235 void migration_incoming_state_destroy(void)
237 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
239 multifd_load_cleanup();
240 compress_threads_load_cleanup();
242 if (mis
->to_src_file
) {
243 /* Tell source that we are done */
244 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
245 qemu_fclose(mis
->to_src_file
);
246 mis
->to_src_file
= NULL
;
249 if (mis
->from_src_file
) {
250 migration_ioc_unregister_yank_from_file(mis
->from_src_file
);
251 qemu_fclose(mis
->from_src_file
);
252 mis
->from_src_file
= NULL
;
254 if (mis
->postcopy_remote_fds
) {
255 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
256 mis
->postcopy_remote_fds
= NULL
;
259 migration_incoming_transport_cleanup(mis
);
260 qemu_event_reset(&mis
->main_thread_load_event
);
262 if (mis
->page_requested
) {
263 g_tree_destroy(mis
->page_requested
);
264 mis
->page_requested
= NULL
;
267 if (mis
->postcopy_qemufile_dst
) {
268 migration_ioc_unregister_yank_from_file(mis
->postcopy_qemufile_dst
);
269 qemu_fclose(mis
->postcopy_qemufile_dst
);
270 mis
->postcopy_qemufile_dst
= NULL
;
273 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
276 static void migrate_generate_event(int new_state
)
278 if (migrate_events()) {
279 qapi_event_send_migration(new_state
);
284 * Send a message on the return channel back to the source
287 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
288 enum mig_rp_message_type message_type
,
289 uint16_t len
, void *data
)
293 trace_migrate_send_rp_message((int)message_type
, len
);
294 QEMU_LOCK_GUARD(&mis
->rp_mutex
);
297 * It's possible that the file handle got lost due to network
300 if (!mis
->to_src_file
) {
305 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
306 qemu_put_be16(mis
->to_src_file
, len
);
307 qemu_put_buffer(mis
->to_src_file
, data
, len
);
308 qemu_fflush(mis
->to_src_file
);
310 /* It's possible that qemu file got error during sending */
311 ret
= qemu_file_get_error(mis
->to_src_file
);
316 /* Request one page from the source VM at the given start address.
317 * rb: the RAMBlock to request the page in
318 * Start: Address offset within the RB
319 * Len: Length in bytes required - must be a multiple of pagesize
321 int migrate_send_rp_message_req_pages(MigrationIncomingState
*mis
,
322 RAMBlock
*rb
, ram_addr_t start
)
324 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
325 size_t msglen
= 12; /* start + len */
326 size_t len
= qemu_ram_pagesize(rb
);
327 enum mig_rp_message_type msg_type
;
331 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
332 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
335 * We maintain the last ramblock that we requested for page. Note that we
336 * don't need locking because this function will only be called within the
337 * postcopy ram fault thread.
339 if (rb
!= mis
->last_rb
) {
342 rbname
= qemu_ram_get_idstr(rb
);
343 rbname_len
= strlen(rbname
);
345 assert(rbname_len
< 256);
347 bufc
[msglen
++] = rbname_len
;
348 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
349 msglen
+= rbname_len
;
350 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
352 msg_type
= MIG_RP_MSG_REQ_PAGES
;
355 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
358 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
,
359 RAMBlock
*rb
, ram_addr_t start
, uint64_t haddr
)
361 void *aligned
= (void *)(uintptr_t)ROUND_DOWN(haddr
, qemu_ram_pagesize(rb
));
362 bool received
= false;
364 WITH_QEMU_LOCK_GUARD(&mis
->page_request_mutex
) {
365 received
= ramblock_recv_bitmap_test_byte_offset(rb
, start
);
366 if (!received
&& !g_tree_lookup(mis
->page_requested
, aligned
)) {
368 * The page has not been received, and it's not yet in the page
369 * request list. Queue it. Set the value of element to 1, so that
370 * things like g_tree_lookup() will return TRUE (1) when found.
372 g_tree_insert(mis
->page_requested
, aligned
, (gpointer
)1);
373 qatomic_inc(&mis
->page_requested_count
);
374 trace_postcopy_page_req_add(aligned
, mis
->page_requested_count
);
379 * If the page is there, skip sending the message. We don't even need the
380 * lock because as long as the page arrived, it'll be there forever.
386 return migrate_send_rp_message_req_pages(mis
, rb
, start
);
389 static bool migration_colo_enabled
;
390 bool migration_incoming_colo_enabled(void)
392 return migration_colo_enabled
;
395 void migration_incoming_disable_colo(void)
397 ram_block_discard_disable(false);
398 migration_colo_enabled
= false;
401 int migration_incoming_enable_colo(void)
403 #ifndef CONFIG_REPLICATION
404 error_report("ENABLE_COLO command come in migration stream, but COLO "
405 "module is not built in");
409 if (!migrate_colo()) {
410 error_report("ENABLE_COLO command come in migration stream, but c-colo "
411 "capability is not set");
415 if (ram_block_discard_disable(true)) {
416 error_report("COLO: cannot disable RAM discard");
419 migration_colo_enabled
= true;
423 void migrate_add_address(SocketAddress
*address
)
425 MigrationIncomingState
*mis
= migration_incoming_get_current();
427 QAPI_LIST_PREPEND(mis
->socket_address_list
,
428 QAPI_CLONE(SocketAddress
, address
));
431 static void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
433 const char *p
= NULL
;
434 MigrationIncomingState
*mis
= migration_incoming_get_current();
436 /* URI is not suitable for migration? */
437 if (!migration_channels_and_uri_compatible(uri
, errp
)) {
441 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
442 MIGRATION_STATUS_SETUP
);
444 if (strstart(uri
, "tcp:", &p
) ||
445 strstart(uri
, "unix:", NULL
) ||
446 strstart(uri
, "vsock:", NULL
)) {
447 socket_start_incoming_migration(p
? p
: uri
, errp
);
449 } else if (strstart(uri
, "rdma:", &p
)) {
450 rdma_start_incoming_migration(p
, errp
);
452 } else if (strstart(uri
, "exec:", &p
)) {
453 exec_start_incoming_migration(p
, errp
);
454 } else if (strstart(uri
, "fd:", &p
)) {
455 fd_start_incoming_migration(p
, errp
);
456 } else if (strstart(uri
, "file:", &p
)) {
457 file_start_incoming_migration(p
, errp
);
459 error_setg(errp
, "unknown migration protocol: %s", uri
);
463 static void process_incoming_migration_bh(void *opaque
)
465 Error
*local_err
= NULL
;
466 MigrationIncomingState
*mis
= opaque
;
468 /* If capability late_block_activate is set:
469 * Only fire up the block code now if we're going to restart the
470 * VM, else 'cont' will do it.
471 * This causes file locking to happen; so we don't want it to happen
472 * unless we really are starting the VM.
474 if (!migrate_late_block_activate() ||
475 (autostart
&& (!global_state_received() ||
476 global_state_get_runstate() == RUN_STATE_RUNNING
))) {
477 /* Make sure all file formats throw away their mutable metadata.
478 * If we get an error here, just don't restart the VM yet. */
479 bdrv_activate_all(&local_err
);
481 error_report_err(local_err
);
488 * This must happen after all error conditions are dealt with and
489 * we're sure the VM is going to be running on this host.
491 qemu_announce_self(&mis
->announce_timer
, migrate_announce_params());
493 multifd_load_shutdown();
495 dirty_bitmap_mig_before_vm_start();
497 if (!global_state_received() ||
498 global_state_get_runstate() == RUN_STATE_RUNNING
) {
502 runstate_set(RUN_STATE_PAUSED
);
504 } else if (migration_incoming_colo_enabled()) {
505 migration_incoming_disable_colo();
508 runstate_set(global_state_get_runstate());
511 * This must happen after any state changes since as soon as an external
512 * observer sees this event they might start to prod at the VM assuming
515 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
516 MIGRATION_STATUS_COMPLETED
);
517 qemu_bh_delete(mis
->bh
);
518 migration_incoming_state_destroy();
521 static void coroutine_fn
522 process_incoming_migration_co(void *opaque
)
524 MigrationIncomingState
*mis
= migration_incoming_get_current();
528 assert(mis
->from_src_file
);
530 if (compress_threads_load_setup(mis
->from_src_file
)) {
531 error_report("Failed to setup decompress threads");
535 mis
->largest_page_size
= qemu_ram_pagesize_largest();
536 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
537 migrate_set_state(&mis
->state
, MIGRATION_STATUS_SETUP
,
538 MIGRATION_STATUS_ACTIVE
);
540 mis
->loadvm_co
= qemu_coroutine_self();
541 ret
= qemu_loadvm_state(mis
->from_src_file
);
542 mis
->loadvm_co
= NULL
;
544 ps
= postcopy_state_get();
545 trace_process_incoming_migration_co_end(ret
, ps
);
546 if (ps
!= POSTCOPY_INCOMING_NONE
) {
547 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
549 * Where a migration had postcopy enabled (and thus went to advise)
550 * but managed to complete within the precopy period, we can use
553 postcopy_ram_incoming_cleanup(mis
);
554 } else if (ret
>= 0) {
556 * Postcopy was started, cleanup should happen at the end of the
559 trace_process_incoming_migration_co_postcopy_end_main();
562 /* Else if something went wrong then just fall out of the normal exit */
566 error_report("load of migration failed: %s", strerror(-ret
));
570 if (colo_incoming_co() < 0) {
574 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
575 qemu_bh_schedule(mis
->bh
);
578 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
579 MIGRATION_STATUS_FAILED
);
580 qemu_fclose(mis
->from_src_file
);
582 multifd_load_cleanup();
583 compress_threads_load_cleanup();
589 * migration_incoming_setup: Setup incoming migration
590 * @f: file for main migration channel
591 * @errp: where to put errors
593 * Returns: %true on success, %false on error.
595 static bool migration_incoming_setup(QEMUFile
*f
, Error
**errp
)
597 MigrationIncomingState
*mis
= migration_incoming_get_current();
599 if (!mis
->from_src_file
) {
600 mis
->from_src_file
= f
;
602 qemu_file_set_blocking(f
, false);
606 void migration_incoming_process(void)
608 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
609 qemu_coroutine_enter(co
);
612 /* Returns true if recovered from a paused migration, otherwise false */
613 static bool postcopy_try_recover(void)
615 MigrationIncomingState
*mis
= migration_incoming_get_current();
617 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
618 /* Resumed from a paused postcopy migration */
620 /* This should be set already in migration_incoming_setup() */
621 assert(mis
->from_src_file
);
622 /* Postcopy has standalone thread to do vm load */
623 qemu_file_set_blocking(mis
->from_src_file
, true);
625 /* Re-configure the return path */
626 mis
->to_src_file
= qemu_file_get_return_path(mis
->from_src_file
);
628 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
629 MIGRATION_STATUS_POSTCOPY_RECOVER
);
632 * Here, we only wake up the main loading thread (while the
633 * rest threads will still be waiting), so that we can receive
634 * commands from source now, and answer it if needed. The
635 * rest threads will be woken up afterwards until we are sure
636 * that source is ready to reply to page requests.
638 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
645 void migration_fd_process_incoming(QEMUFile
*f
, Error
**errp
)
647 if (!migration_incoming_setup(f
, errp
)) {
650 if (postcopy_try_recover()) {
653 migration_incoming_process();
657 * Returns true when we want to start a new incoming migration process,
660 static bool migration_should_start_incoming(bool main_channel
)
662 /* Multifd doesn't start unless all channels are established */
663 if (migrate_multifd()) {
664 return migration_has_all_channels();
667 /* Preempt channel only starts when the main channel is created */
668 if (migrate_postcopy_preempt()) {
673 * For all the rest types of migration, we should only reach here when
674 * it's the main channel that's being created, and we should always
675 * proceed with this channel.
677 assert(main_channel
);
681 void migration_ioc_process_incoming(QIOChannel
*ioc
, Error
**errp
)
683 MigrationIncomingState
*mis
= migration_incoming_get_current();
684 Error
*local_err
= NULL
;
686 bool default_channel
= true;
687 uint32_t channel_magic
= 0;
690 if (migrate_multifd() && !migrate_postcopy_ram() &&
691 qio_channel_has_feature(ioc
, QIO_CHANNEL_FEATURE_READ_MSG_PEEK
)) {
693 * With multiple channels, it is possible that we receive channels
694 * out of order on destination side, causing incorrect mapping of
695 * source channels on destination side. Check channel MAGIC to
696 * decide type of channel. Please note this is best effort, postcopy
697 * preempt channel does not send any magic number so avoid it for
698 * postcopy live migration. Also tls live migration already does
699 * tls handshake while initializing main channel so with tls this
700 * issue is not possible.
702 ret
= migration_channel_read_peek(ioc
, (void *)&channel_magic
,
703 sizeof(channel_magic
), &local_err
);
706 error_propagate(errp
, local_err
);
710 default_channel
= (channel_magic
== cpu_to_be32(QEMU_VM_FILE_MAGIC
));
712 default_channel
= !mis
->from_src_file
;
715 if (multifd_load_setup(errp
) != 0) {
716 error_setg(errp
, "Failed to setup multifd channels");
720 if (default_channel
) {
721 f
= qemu_file_new_input(ioc
);
723 if (!migration_incoming_setup(f
, errp
)) {
727 /* Multiple connections */
728 assert(migration_needs_multiple_sockets());
729 if (migrate_multifd()) {
730 multifd_recv_new_channel(ioc
, &local_err
);
732 assert(migrate_postcopy_preempt());
733 f
= qemu_file_new_input(ioc
);
734 postcopy_preempt_new_channel(mis
, f
);
737 error_propagate(errp
, local_err
);
742 if (migration_should_start_incoming(default_channel
)) {
743 /* If it's a recovery, we're done */
744 if (postcopy_try_recover()) {
747 migration_incoming_process();
752 * @migration_has_all_channels: We have received all channels that we need
754 * Returns true when we have got connections to all the channels that
755 * we need for migration.
757 bool migration_has_all_channels(void)
759 MigrationIncomingState
*mis
= migration_incoming_get_current();
761 if (!mis
->from_src_file
) {
765 if (migrate_multifd()) {
766 return multifd_recv_all_channels_created();
769 if (migrate_postcopy_preempt()) {
770 return mis
->postcopy_qemufile_dst
!= NULL
;
776 int migrate_send_rp_switchover_ack(MigrationIncomingState
*mis
)
778 return migrate_send_rp_message(mis
, MIG_RP_MSG_SWITCHOVER_ACK
, 0, NULL
);
782 * Send a 'SHUT' message on the return channel with the given value
783 * to indicate that we've finished with the RP. Non-0 value indicates
786 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
791 buf
= cpu_to_be32(value
);
792 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
796 * Send a 'PONG' message on the return channel with the given value
797 * (normally in response to a 'PING')
799 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
804 buf
= cpu_to_be32(value
);
805 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
808 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
816 * First, we send the header part. It contains only the len of
817 * idstr, and the idstr itself.
819 len
= strlen(block_name
);
821 memcpy(buf
+ 1, block_name
, len
);
823 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
824 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
829 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
832 * Next, we dump the received bitmap to the stream.
834 * TODO: currently we are safe since we are the only one that is
835 * using the to_src_file handle (fault thread is still paused),
836 * and it's ok even not taking the mutex. However the best way is
837 * to take the lock before sending the message header, and release
838 * the lock after sending the bitmap.
840 qemu_mutex_lock(&mis
->rp_mutex
);
841 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
842 qemu_mutex_unlock(&mis
->rp_mutex
);
844 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
847 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
851 buf
= cpu_to_be32(value
);
852 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
856 * Return true if we're already in the middle of a migration
857 * (i.e. any of the active or setup states)
859 bool migration_is_setup_or_active(int state
)
862 case MIGRATION_STATUS_ACTIVE
:
863 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
864 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
865 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
866 case MIGRATION_STATUS_SETUP
:
867 case MIGRATION_STATUS_PRE_SWITCHOVER
:
868 case MIGRATION_STATUS_DEVICE
:
869 case MIGRATION_STATUS_WAIT_UNPLUG
:
870 case MIGRATION_STATUS_COLO
:
879 bool migration_is_running(int state
)
882 case MIGRATION_STATUS_ACTIVE
:
883 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
884 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
885 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
886 case MIGRATION_STATUS_SETUP
:
887 case MIGRATION_STATUS_PRE_SWITCHOVER
:
888 case MIGRATION_STATUS_DEVICE
:
889 case MIGRATION_STATUS_WAIT_UNPLUG
:
890 case MIGRATION_STATUS_CANCELLING
:
899 static bool migrate_show_downtime(MigrationState
*s
)
901 return (s
->state
== MIGRATION_STATUS_COMPLETED
) || migration_in_postcopy();
904 static void populate_time_info(MigrationInfo
*info
, MigrationState
*s
)
906 info
->has_status
= true;
907 info
->has_setup_time
= true;
908 info
->setup_time
= s
->setup_time
;
910 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
911 info
->has_total_time
= true;
912 info
->total_time
= s
->total_time
;
914 info
->has_total_time
= true;
915 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) -
919 if (migrate_show_downtime(s
)) {
920 info
->has_downtime
= true;
921 info
->downtime
= s
->downtime
;
923 info
->has_expected_downtime
= true;
924 info
->expected_downtime
= s
->expected_downtime
;
928 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
930 size_t page_size
= qemu_target_page_size();
932 info
->ram
= g_malloc0(sizeof(*info
->ram
));
933 info
->ram
->transferred
= stat64_get(&mig_stats
.transferred
);
934 info
->ram
->total
= ram_bytes_total();
935 info
->ram
->duplicate
= stat64_get(&mig_stats
.zero_pages
);
936 /* legacy value. It is not used anymore */
937 info
->ram
->skipped
= 0;
938 info
->ram
->normal
= stat64_get(&mig_stats
.normal_pages
);
939 info
->ram
->normal_bytes
= info
->ram
->normal
* page_size
;
940 info
->ram
->mbps
= s
->mbps
;
941 info
->ram
->dirty_sync_count
=
942 stat64_get(&mig_stats
.dirty_sync_count
);
943 info
->ram
->dirty_sync_missed_zero_copy
=
944 stat64_get(&mig_stats
.dirty_sync_missed_zero_copy
);
945 info
->ram
->postcopy_requests
=
946 stat64_get(&mig_stats
.postcopy_requests
);
947 info
->ram
->page_size
= page_size
;
948 info
->ram
->multifd_bytes
= stat64_get(&mig_stats
.multifd_bytes
);
949 info
->ram
->pages_per_second
= s
->pages_per_second
;
950 info
->ram
->precopy_bytes
= stat64_get(&mig_stats
.precopy_bytes
);
951 info
->ram
->downtime_bytes
= stat64_get(&mig_stats
.downtime_bytes
);
952 info
->ram
->postcopy_bytes
= stat64_get(&mig_stats
.postcopy_bytes
);
954 if (migrate_xbzrle()) {
955 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
956 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
957 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
958 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
959 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
960 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
961 info
->xbzrle_cache
->encoding_rate
= xbzrle_counters
.encoding_rate
;
962 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
965 if (migrate_compress()) {
966 info
->compression
= g_malloc0(sizeof(*info
->compression
));
967 info
->compression
->pages
= compression_counters
.pages
;
968 info
->compression
->busy
= compression_counters
.busy
;
969 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
970 info
->compression
->compressed_size
=
971 compression_counters
.compressed_size
;
972 info
->compression
->compression_rate
=
973 compression_counters
.compression_rate
;
976 if (cpu_throttle_active()) {
977 info
->has_cpu_throttle_percentage
= true;
978 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
981 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
982 info
->ram
->remaining
= ram_bytes_remaining();
983 info
->ram
->dirty_pages_rate
=
984 stat64_get(&mig_stats
.dirty_pages_rate
);
987 if (migrate_dirty_limit() && dirtylimit_in_service()) {
988 info
->has_dirty_limit_throttle_time_per_round
= true;
989 info
->dirty_limit_throttle_time_per_round
=
990 dirtylimit_throttle_time_per_round();
992 info
->has_dirty_limit_ring_full_time
= true;
993 info
->dirty_limit_ring_full_time
= dirtylimit_ring_full_time();
997 static void populate_disk_info(MigrationInfo
*info
)
999 if (blk_mig_active()) {
1000 info
->disk
= g_malloc0(sizeof(*info
->disk
));
1001 info
->disk
->transferred
= blk_mig_bytes_transferred();
1002 info
->disk
->remaining
= blk_mig_bytes_remaining();
1003 info
->disk
->total
= blk_mig_bytes_total();
1007 static void fill_source_migration_info(MigrationInfo
*info
)
1009 MigrationState
*s
= migrate_get_current();
1010 int state
= qatomic_read(&s
->state
);
1011 GSList
*cur_blocker
= migration_blockers
;
1013 info
->blocked_reasons
= NULL
;
1016 * There are two types of reasons a migration might be blocked;
1017 * a) devices marked in VMState as non-migratable, and
1018 * b) Explicit migration blockers
1019 * We need to add both of them here.
1021 qemu_savevm_non_migratable_list(&info
->blocked_reasons
);
1023 while (cur_blocker
) {
1024 QAPI_LIST_PREPEND(info
->blocked_reasons
,
1025 g_strdup(error_get_pretty(cur_blocker
->data
)));
1026 cur_blocker
= g_slist_next(cur_blocker
);
1028 info
->has_blocked_reasons
= info
->blocked_reasons
!= NULL
;
1031 case MIGRATION_STATUS_NONE
:
1032 /* no migration has happened ever */
1033 /* do not overwrite destination migration status */
1035 case MIGRATION_STATUS_SETUP
:
1036 info
->has_status
= true;
1037 info
->has_total_time
= false;
1039 case MIGRATION_STATUS_ACTIVE
:
1040 case MIGRATION_STATUS_CANCELLING
:
1041 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1042 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1043 case MIGRATION_STATUS_DEVICE
:
1044 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1045 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1046 /* TODO add some postcopy stats */
1047 populate_time_info(info
, s
);
1048 populate_ram_info(info
, s
);
1049 populate_disk_info(info
);
1050 migration_populate_vfio_info(info
);
1052 case MIGRATION_STATUS_COLO
:
1053 info
->has_status
= true;
1054 /* TODO: display COLO specific information (checkpoint info etc.) */
1056 case MIGRATION_STATUS_COMPLETED
:
1057 populate_time_info(info
, s
);
1058 populate_ram_info(info
, s
);
1059 migration_populate_vfio_info(info
);
1061 case MIGRATION_STATUS_FAILED
:
1062 info
->has_status
= true;
1064 case MIGRATION_STATUS_CANCELLED
:
1065 info
->has_status
= true;
1067 case MIGRATION_STATUS_WAIT_UNPLUG
:
1068 info
->has_status
= true;
1071 info
->status
= state
;
1073 QEMU_LOCK_GUARD(&s
->error_mutex
);
1075 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
1079 static void fill_destination_migration_info(MigrationInfo
*info
)
1081 MigrationIncomingState
*mis
= migration_incoming_get_current();
1083 if (mis
->socket_address_list
) {
1084 info
->has_socket_address
= true;
1085 info
->socket_address
=
1086 QAPI_CLONE(SocketAddressList
, mis
->socket_address_list
);
1089 switch (mis
->state
) {
1090 case MIGRATION_STATUS_NONE
:
1092 case MIGRATION_STATUS_SETUP
:
1093 case MIGRATION_STATUS_CANCELLING
:
1094 case MIGRATION_STATUS_CANCELLED
:
1095 case MIGRATION_STATUS_ACTIVE
:
1096 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1097 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1098 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1099 case MIGRATION_STATUS_FAILED
:
1100 case MIGRATION_STATUS_COLO
:
1101 info
->has_status
= true;
1103 case MIGRATION_STATUS_COMPLETED
:
1104 info
->has_status
= true;
1105 fill_destination_postcopy_migration_info(info
);
1108 info
->status
= mis
->state
;
1111 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1113 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1115 fill_destination_migration_info(info
);
1116 fill_source_migration_info(info
);
1121 void qmp_migrate_start_postcopy(Error
**errp
)
1123 MigrationState
*s
= migrate_get_current();
1125 if (!migrate_postcopy()) {
1126 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1127 " the start of migration");
1131 if (s
->state
== MIGRATION_STATUS_NONE
) {
1132 error_setg(errp
, "Postcopy must be started after migration has been"
1137 * we don't error if migration has finished since that would be racy
1138 * with issuing this command.
1140 qatomic_set(&s
->start_postcopy
, true);
1143 /* shared migration helpers */
1145 void migrate_set_state(int *state
, int old_state
, int new_state
)
1147 assert(new_state
< MIGRATION_STATUS__MAX
);
1148 if (qatomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1149 trace_migrate_set_state(MigrationStatus_str(new_state
));
1150 migrate_generate_event(new_state
);
1154 static void migrate_fd_cleanup(MigrationState
*s
)
1156 qemu_bh_delete(s
->cleanup_bh
);
1157 s
->cleanup_bh
= NULL
;
1159 g_free(s
->hostname
);
1161 json_writer_free(s
->vmdesc
);
1164 qemu_savevm_state_cleanup();
1166 if (s
->to_dst_file
) {
1169 trace_migrate_fd_cleanup();
1170 qemu_mutex_unlock_iothread();
1171 if (s
->migration_thread_running
) {
1172 qemu_thread_join(&s
->thread
);
1173 s
->migration_thread_running
= false;
1175 qemu_mutex_lock_iothread();
1177 multifd_save_cleanup();
1178 qemu_mutex_lock(&s
->qemu_file_lock
);
1179 tmp
= s
->to_dst_file
;
1180 s
->to_dst_file
= NULL
;
1181 qemu_mutex_unlock(&s
->qemu_file_lock
);
1183 * Close the file handle without the lock to make sure the
1184 * critical section won't block for long.
1186 migration_ioc_unregister_yank_from_file(tmp
);
1191 * We already cleaned up to_dst_file, so errors from the return
1192 * path might be due to that, ignore them.
1194 await_return_path_close_on_source(s
);
1196 assert(!migration_is_active(s
));
1198 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1199 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1200 MIGRATION_STATUS_CANCELLED
);
1204 /* It is used on info migrate. We can't free it */
1205 error_report_err(error_copy(s
->error
));
1207 notifier_list_notify(&migration_state_notifiers
, s
);
1208 block_cleanup_parameters();
1209 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1212 static void migrate_fd_cleanup_schedule(MigrationState
*s
)
1215 * Ref the state for bh, because it may be called when
1216 * there're already no other refs
1218 object_ref(OBJECT(s
));
1219 qemu_bh_schedule(s
->cleanup_bh
);
1222 static void migrate_fd_cleanup_bh(void *opaque
)
1224 MigrationState
*s
= opaque
;
1225 migrate_fd_cleanup(s
);
1226 object_unref(OBJECT(s
));
1229 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1231 QEMU_LOCK_GUARD(&s
->error_mutex
);
1233 s
->error
= error_copy(error
);
1237 static void migrate_error_free(MigrationState
*s
)
1239 QEMU_LOCK_GUARD(&s
->error_mutex
);
1241 error_free(s
->error
);
1246 static void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1248 trace_migrate_fd_error(error_get_pretty(error
));
1249 assert(s
->to_dst_file
== NULL
);
1250 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1251 MIGRATION_STATUS_FAILED
);
1252 migrate_set_error(s
, error
);
1255 static void migrate_fd_cancel(MigrationState
*s
)
1259 trace_migrate_fd_cancel();
1261 WITH_QEMU_LOCK_GUARD(&s
->qemu_file_lock
) {
1262 if (s
->rp_state
.from_dst_file
) {
1263 /* shutdown the rp socket, so causing the rp thread to shutdown */
1264 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1269 old_state
= s
->state
;
1270 if (!migration_is_running(old_state
)) {
1273 /* If the migration is paused, kick it out of the pause */
1274 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1275 qemu_sem_post(&s
->pause_sem
);
1277 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1278 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1281 * If we're unlucky the migration code might be stuck somewhere in a
1282 * send/write while the network has failed and is waiting to timeout;
1283 * if we've got shutdown(2) available then we can force it to quit.
1285 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1286 WITH_QEMU_LOCK_GUARD(&s
->qemu_file_lock
) {
1287 if (s
->to_dst_file
) {
1288 qemu_file_shutdown(s
->to_dst_file
);
1292 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1293 Error
*local_err
= NULL
;
1295 bdrv_activate_all(&local_err
);
1297 error_report_err(local_err
);
1299 s
->block_inactive
= false;
1304 void add_migration_state_change_notifier(Notifier
*notify
)
1306 notifier_list_add(&migration_state_notifiers
, notify
);
1309 void remove_migration_state_change_notifier(Notifier
*notify
)
1311 notifier_remove(notify
);
1314 bool migration_in_setup(MigrationState
*s
)
1316 return s
->state
== MIGRATION_STATUS_SETUP
;
1319 bool migration_has_finished(MigrationState
*s
)
1321 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1324 bool migration_has_failed(MigrationState
*s
)
1326 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1327 s
->state
== MIGRATION_STATUS_FAILED
);
1330 bool migration_in_postcopy(void)
1332 MigrationState
*s
= migrate_get_current();
1335 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1336 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1337 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1344 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1346 return migration_in_postcopy() && s
->postcopy_after_devices
;
1349 bool migration_in_incoming_postcopy(void)
1351 PostcopyState ps
= postcopy_state_get();
1353 return ps
>= POSTCOPY_INCOMING_DISCARD
&& ps
< POSTCOPY_INCOMING_END
;
1356 bool migration_incoming_postcopy_advised(void)
1358 PostcopyState ps
= postcopy_state_get();
1360 return ps
>= POSTCOPY_INCOMING_ADVISE
&& ps
< POSTCOPY_INCOMING_END
;
1363 bool migration_in_bg_snapshot(void)
1365 MigrationState
*s
= migrate_get_current();
1367 return migrate_background_snapshot() &&
1368 migration_is_setup_or_active(s
->state
);
1371 bool migration_is_idle(void)
1373 MigrationState
*s
= current_migration
;
1380 case MIGRATION_STATUS_NONE
:
1381 case MIGRATION_STATUS_CANCELLED
:
1382 case MIGRATION_STATUS_COMPLETED
:
1383 case MIGRATION_STATUS_FAILED
:
1385 case MIGRATION_STATUS_SETUP
:
1386 case MIGRATION_STATUS_CANCELLING
:
1387 case MIGRATION_STATUS_ACTIVE
:
1388 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1389 case MIGRATION_STATUS_COLO
:
1390 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1391 case MIGRATION_STATUS_DEVICE
:
1392 case MIGRATION_STATUS_WAIT_UNPLUG
:
1394 case MIGRATION_STATUS__MAX
:
1395 g_assert_not_reached();
1401 bool migration_is_active(MigrationState
*s
)
1403 return (s
->state
== MIGRATION_STATUS_ACTIVE
||
1404 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1407 int migrate_init(MigrationState
*s
, Error
**errp
)
1411 ret
= qemu_savevm_state_prepare(errp
);
1417 * Reinitialise all migration state, except
1418 * parameters/capabilities that the user set, and
1423 s
->to_dst_file
= NULL
;
1424 s
->state
= MIGRATION_STATUS_NONE
;
1425 s
->rp_state
.from_dst_file
= NULL
;
1426 s
->rp_state
.error
= false;
1428 s
->pages_per_second
= 0.0;
1430 s
->expected_downtime
= 0;
1432 s
->start_postcopy
= false;
1433 s
->postcopy_after_devices
= false;
1434 s
->migration_thread_running
= false;
1435 error_free(s
->error
);
1439 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1441 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1443 s
->vm_old_state
= -1;
1444 s
->iteration_initial_bytes
= 0;
1445 s
->threshold_size
= 0;
1446 s
->switchover_acked
= false;
1448 * set mig_stats compression_counters memory to zero for a
1451 memset(&mig_stats
, 0, sizeof(mig_stats
));
1452 memset(&compression_counters
, 0, sizeof(compression_counters
));
1453 migration_reset_vfio_bytes_transferred();
1458 int migrate_add_blocker_internal(Error
*reason
, Error
**errp
)
1460 /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */
1461 if (runstate_check(RUN_STATE_SAVE_VM
) || !migration_is_idle()) {
1462 error_propagate_prepend(errp
, error_copy(reason
),
1463 "disallowing migration blocker "
1464 "(migration/snapshot in progress) for: ");
1468 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1472 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1474 if (only_migratable
) {
1475 error_propagate_prepend(errp
, error_copy(reason
),
1476 "disallowing migration blocker "
1477 "(--only-migratable) for: ");
1481 return migrate_add_blocker_internal(reason
, errp
);
1484 void migrate_del_blocker(Error
*reason
)
1486 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1489 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1491 Error
*local_err
= NULL
;
1492 static bool once
= true;
1495 error_setg(errp
, "The incoming migration has already been started");
1498 if (!runstate_check(RUN_STATE_INMIGRATE
)) {
1499 error_setg(errp
, "'-incoming' was not specified on the command line");
1503 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
1507 qemu_start_incoming_migration(uri
, &local_err
);
1510 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1511 error_propagate(errp
, local_err
);
1518 void qmp_migrate_recover(const char *uri
, Error
**errp
)
1520 MigrationIncomingState
*mis
= migration_incoming_get_current();
1523 * Don't even bother to use ERRP_GUARD() as it _must_ always be set by
1524 * callers (no one should ignore a recover failure); if there is, it's a
1525 * programming error.
1529 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1530 error_setg(errp
, "Migrate recover can only be run "
1531 "when postcopy is paused.");
1535 /* If there's an existing transport, release it */
1536 migration_incoming_transport_cleanup(mis
);
1539 * Note that this call will never start a real migration; it will
1540 * only re-setup the migration stream and poke existing migration
1541 * to continue using that newly established channel.
1543 qemu_start_incoming_migration(uri
, errp
);
1546 void qmp_migrate_pause(Error
**errp
)
1548 MigrationState
*ms
= migrate_get_current();
1549 MigrationIncomingState
*mis
= migration_incoming_get_current();
1552 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1553 /* Source side, during postcopy */
1554 qemu_mutex_lock(&ms
->qemu_file_lock
);
1555 if (ms
->to_dst_file
) {
1556 ret
= qemu_file_shutdown(ms
->to_dst_file
);
1558 qemu_mutex_unlock(&ms
->qemu_file_lock
);
1560 error_setg(errp
, "Failed to pause source migration");
1565 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1566 ret
= qemu_file_shutdown(mis
->from_src_file
);
1568 error_setg(errp
, "Failed to pause destination migration");
1573 error_setg(errp
, "migrate-pause is currently only supported "
1574 "during postcopy-active state");
1577 bool migration_is_blocked(Error
**errp
)
1579 if (qemu_savevm_state_blocked(errp
)) {
1583 if (migration_blockers
) {
1584 error_propagate(errp
, error_copy(migration_blockers
->data
));
1591 /* Returns true if continue to migrate, or false if error detected */
1592 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
1593 bool resume
, Error
**errp
)
1595 Error
*local_err
= NULL
;
1598 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1599 error_setg(errp
, "Cannot resume if there is no "
1600 "paused migration");
1605 * Postcopy recovery won't work well with release-ram
1606 * capability since release-ram will drop the page buffer as
1607 * long as the page is put into the send buffer. So if there
1608 * is a network failure happened, any page buffers that have
1609 * not yet reached the destination VM but have already been
1610 * sent from the source VM will be lost forever. Let's refuse
1611 * the client from resuming such a postcopy migration.
1612 * Luckily release-ram was designed to only be used when src
1613 * and destination VMs are on the same host, so it should be
1616 if (migrate_release_ram()) {
1617 error_setg(errp
, "Postcopy recovery cannot work "
1618 "when release-ram capability is set");
1622 /* This is a resume, skip init status */
1626 if (migration_is_running(s
->state
)) {
1627 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1631 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1632 error_setg(errp
, "Guest is waiting for an incoming migration");
1636 if (runstate_check(RUN_STATE_POSTMIGRATE
)) {
1637 error_setg(errp
, "Can't migrate the vm that was paused due to "
1638 "previous migration");
1642 if (migration_is_blocked(errp
)) {
1646 if (blk
|| blk_inc
) {
1647 if (migrate_colo()) {
1648 error_setg(errp
, "No disk migration is required in COLO mode");
1651 if (migrate_block() || migrate_block_incremental()) {
1652 error_setg(errp
, "Command options are incompatible with "
1653 "current migration capabilities");
1656 if (!migrate_cap_set(MIGRATION_CAPABILITY_BLOCK
, true, &local_err
)) {
1657 error_propagate(errp
, local_err
);
1660 s
->must_remove_block_options
= true;
1664 migrate_set_block_incremental(true);
1667 if (migrate_init(s
, errp
)) {
1674 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1675 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1676 bool has_resume
, bool resume
, Error
**errp
)
1678 bool resume_requested
;
1679 Error
*local_err
= NULL
;
1680 MigrationState
*s
= migrate_get_current();
1681 const char *p
= NULL
;
1683 /* URI is not suitable for migration? */
1684 if (!migration_channels_and_uri_compatible(uri
, errp
)) {
1688 resume_requested
= has_resume
&& resume
;
1689 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
1690 resume_requested
, errp
)) {
1691 /* Error detected, put into errp */
1695 if (!resume_requested
) {
1696 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
1701 if (strstart(uri
, "tcp:", &p
) ||
1702 strstart(uri
, "unix:", NULL
) ||
1703 strstart(uri
, "vsock:", NULL
)) {
1704 socket_start_outgoing_migration(s
, p
? p
: uri
, &local_err
);
1706 } else if (strstart(uri
, "rdma:", &p
)) {
1707 rdma_start_outgoing_migration(s
, p
, &local_err
);
1709 } else if (strstart(uri
, "exec:", &p
)) {
1710 exec_start_outgoing_migration(s
, p
, &local_err
);
1711 } else if (strstart(uri
, "fd:", &p
)) {
1712 fd_start_outgoing_migration(s
, p
, &local_err
);
1713 } else if (strstart(uri
, "file:", &p
)) {
1714 file_start_outgoing_migration(s
, p
, &local_err
);
1716 error_setg(&local_err
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1717 "a valid migration protocol");
1718 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1719 MIGRATION_STATUS_FAILED
);
1720 block_cleanup_parameters();
1724 if (!resume_requested
) {
1725 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1727 migrate_fd_error(s
, local_err
);
1728 error_propagate(errp
, local_err
);
1733 void qmp_migrate_cancel(Error
**errp
)
1735 migration_cancel(NULL
);
1738 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
1740 MigrationState
*s
= migrate_get_current();
1741 if (s
->state
!= state
) {
1742 error_setg(errp
, "Migration not in expected state: %s",
1743 MigrationStatus_str(s
->state
));
1746 qemu_sem_post(&s
->pause_sem
);
1749 /* migration thread support */
1751 * Something bad happened to the RP stream, mark an error
1752 * The caller shall print or trace something to indicate why
1754 static void mark_source_rp_bad(MigrationState
*s
)
1756 s
->rp_state
.error
= true;
1759 static struct rp_cmd_args
{
1760 ssize_t len
; /* -1 = variable */
1763 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1764 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1765 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1766 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1767 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1768 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
1769 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
1770 [MIG_RP_MSG_SWITCHOVER_ACK
] = { .len
= 0, .name
= "SWITCHOVER_ACK" },
1771 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1775 * Process a request for pages received on the return path,
1776 * We're allowed to send more than requested (e.g. to round to our page size)
1777 * and we don't need to send pages that have already been sent.
1779 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1780 ram_addr_t start
, size_t len
)
1782 long our_host_ps
= qemu_real_host_page_size();
1784 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1787 * Since we currently insist on matching page sizes, just sanity check
1788 * we're being asked for whole host pages.
1790 if (!QEMU_IS_ALIGNED(start
, our_host_ps
) ||
1791 !QEMU_IS_ALIGNED(len
, our_host_ps
)) {
1792 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1793 " len: %zd", __func__
, start
, len
);
1794 mark_source_rp_bad(ms
);
1798 if (ram_save_queue_pages(rbname
, start
, len
)) {
1799 mark_source_rp_bad(ms
);
1803 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
1805 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
1808 error_report("%s: invalid block name '%s'", __func__
, block_name
);
1812 /* Fetch the received bitmap and refresh the dirty bitmap */
1813 return ram_dirty_bitmap_reload(s
, block
);
1816 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
1818 trace_source_return_path_thread_resume_ack(value
);
1820 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
1821 error_report("%s: illegal resume_ack value %"PRIu32
,
1826 /* Now both sides are active. */
1827 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
1828 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1830 /* Notify send thread that time to continue send pages */
1831 qemu_sem_post(&s
->rp_state
.rp_sem
);
1837 * Release ms->rp_state.from_dst_file (and postcopy_qemufile_src if
1838 * existed) in a safe way.
1840 static void migration_release_dst_files(MigrationState
*ms
)
1844 WITH_QEMU_LOCK_GUARD(&ms
->qemu_file_lock
) {
1846 * Reset the from_dst_file pointer first before releasing it, as we
1847 * can't block within lock section
1849 file
= ms
->rp_state
.from_dst_file
;
1850 ms
->rp_state
.from_dst_file
= NULL
;
1854 * Do the same to postcopy fast path socket too if there is. No
1855 * locking needed because this qemufile should only be managed by
1856 * return path thread.
1858 if (ms
->postcopy_qemufile_src
) {
1859 migration_ioc_unregister_yank_from_file(ms
->postcopy_qemufile_src
);
1860 qemu_file_shutdown(ms
->postcopy_qemufile_src
);
1861 qemu_fclose(ms
->postcopy_qemufile_src
);
1862 ms
->postcopy_qemufile_src
= NULL
;
1869 * Handles messages sent on the return path towards the source VM
1872 static void *source_return_path_thread(void *opaque
)
1874 MigrationState
*ms
= opaque
;
1875 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1876 uint16_t header_len
, header_type
;
1878 uint32_t tmp32
, sibling_error
;
1879 ram_addr_t start
= 0; /* =0 to silence warning */
1880 size_t len
= 0, expected_len
;
1883 trace_source_return_path_thread_entry();
1884 rcu_register_thread();
1886 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1887 migration_is_setup_or_active(ms
->state
)) {
1888 trace_source_return_path_thread_loop_top();
1889 header_type
= qemu_get_be16(rp
);
1890 header_len
= qemu_get_be16(rp
);
1892 if (qemu_file_get_error(rp
)) {
1893 mark_source_rp_bad(ms
);
1897 if (header_type
>= MIG_RP_MSG_MAX
||
1898 header_type
== MIG_RP_MSG_INVALID
) {
1899 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1900 header_type
, header_len
);
1901 mark_source_rp_bad(ms
);
1905 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1906 header_len
!= rp_cmd_args
[header_type
].len
) ||
1907 header_len
> sizeof(buf
)) {
1908 error_report("RP: Received '%s' message (0x%04x) with"
1909 "incorrect length %d expecting %zu",
1910 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1911 (size_t)rp_cmd_args
[header_type
].len
);
1912 mark_source_rp_bad(ms
);
1916 /* We know we've got a valid header by this point */
1917 res
= qemu_get_buffer(rp
, buf
, header_len
);
1918 if (res
!= header_len
) {
1919 error_report("RP: Failed reading data for message 0x%04x"
1920 " read %d expected %d",
1921 header_type
, res
, header_len
);
1922 mark_source_rp_bad(ms
);
1926 /* OK, we have the message and the data */
1927 switch (header_type
) {
1928 case MIG_RP_MSG_SHUT
:
1929 sibling_error
= ldl_be_p(buf
);
1930 trace_source_return_path_thread_shut(sibling_error
);
1931 if (sibling_error
) {
1932 error_report("RP: Sibling indicated error %d", sibling_error
);
1933 mark_source_rp_bad(ms
);
1936 * We'll let the main thread deal with closing the RP
1937 * we could do a shutdown(2) on it, but we're the only user
1938 * anyway, so there's nothing gained.
1942 case MIG_RP_MSG_PONG
:
1943 tmp32
= ldl_be_p(buf
);
1944 trace_source_return_path_thread_pong(tmp32
);
1945 qemu_sem_post(&ms
->rp_state
.rp_pong_acks
);
1948 case MIG_RP_MSG_REQ_PAGES
:
1949 start
= ldq_be_p(buf
);
1950 len
= ldl_be_p(buf
+ 8);
1951 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1954 case MIG_RP_MSG_REQ_PAGES_ID
:
1955 expected_len
= 12 + 1; /* header + termination */
1957 if (header_len
>= expected_len
) {
1958 start
= ldq_be_p(buf
);
1959 len
= ldl_be_p(buf
+ 8);
1960 /* Now we expect an idstr */
1961 tmp32
= buf
[12]; /* Length of the following idstr */
1962 buf
[13 + tmp32
] = '\0';
1963 expected_len
+= tmp32
;
1965 if (header_len
!= expected_len
) {
1966 error_report("RP: Req_Page_id with length %d expecting %zd",
1967 header_len
, expected_len
);
1968 mark_source_rp_bad(ms
);
1971 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
1974 case MIG_RP_MSG_RECV_BITMAP
:
1975 if (header_len
< 1) {
1976 error_report("%s: missing block name", __func__
);
1977 mark_source_rp_bad(ms
);
1980 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
1981 buf
[buf
[0] + 1] = '\0';
1982 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
1983 mark_source_rp_bad(ms
);
1988 case MIG_RP_MSG_RESUME_ACK
:
1989 tmp32
= ldl_be_p(buf
);
1990 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
1991 mark_source_rp_bad(ms
);
1996 case MIG_RP_MSG_SWITCHOVER_ACK
:
1997 ms
->switchover_acked
= true;
1998 trace_source_return_path_thread_switchover_acked();
2007 if (qemu_file_get_error(rp
)) {
2008 trace_source_return_path_thread_bad_end();
2009 mark_source_rp_bad(ms
);
2012 trace_source_return_path_thread_end();
2013 rcu_unregister_thread();
2017 static int open_return_path_on_source(MigrationState
*ms
)
2019 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
2020 if (!ms
->rp_state
.from_dst_file
) {
2024 trace_open_return_path_on_source();
2026 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
2027 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
2028 ms
->rp_state
.rp_thread_created
= true;
2030 trace_open_return_path_on_source_continue();
2035 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2036 static int await_return_path_close_on_source(MigrationState
*ms
)
2040 if (!ms
->rp_state
.rp_thread_created
) {
2044 trace_migration_return_path_end_before();
2047 * If this is a normal exit then the destination will send a SHUT
2048 * and the rp_thread will exit, however if there's an error we
2049 * need to cause it to exit. shutdown(2), if we have it, will
2050 * cause it to unblock if it's stuck waiting for the destination.
2052 WITH_QEMU_LOCK_GUARD(&ms
->qemu_file_lock
) {
2053 if (ms
->to_dst_file
&& ms
->rp_state
.from_dst_file
&&
2054 qemu_file_get_error(ms
->to_dst_file
)) {
2055 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
2059 trace_await_return_path_close_on_source_joining();
2060 qemu_thread_join(&ms
->rp_state
.rp_thread
);
2061 ms
->rp_state
.rp_thread_created
= false;
2062 trace_await_return_path_close_on_source_close();
2064 ret
= ms
->rp_state
.error
;
2065 ms
->rp_state
.error
= false;
2067 migration_release_dst_files(ms
);
2069 trace_migration_return_path_end_after(ret
);
2074 migration_wait_main_channel(MigrationState
*ms
)
2076 /* Wait until one PONG message received */
2077 qemu_sem_wait(&ms
->rp_state
.rp_pong_acks
);
2081 * Switch from normal iteration to postcopy
2082 * Returns non-0 on error
2084 static int postcopy_start(MigrationState
*ms
, Error
**errp
)
2087 QIOChannelBuffer
*bioc
;
2089 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2090 uint64_t bandwidth
= migrate_max_postcopy_bandwidth();
2091 bool restart_block
= false;
2092 int cur_state
= MIGRATION_STATUS_ACTIVE
;
2094 if (migrate_postcopy_preempt()) {
2095 migration_wait_main_channel(ms
);
2096 if (postcopy_preempt_establish_channel(ms
)) {
2097 migrate_set_state(&ms
->state
, ms
->state
, MIGRATION_STATUS_FAILED
);
2102 if (!migrate_pause_before_switchover()) {
2103 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
2104 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2107 trace_postcopy_start();
2108 qemu_mutex_lock_iothread();
2109 trace_postcopy_start_set_run();
2111 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2112 global_state_store();
2113 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2118 ret
= migration_maybe_pause(ms
, &cur_state
,
2119 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2124 ret
= bdrv_inactivate_all();
2128 restart_block
= true;
2131 * Cause any non-postcopiable, but iterative devices to
2132 * send out their final data.
2134 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2137 * in Finish migrate and with the io-lock held everything should
2138 * be quiet, but we've potentially still got dirty pages and we
2139 * need to tell the destination to throw any pages it's already received
2142 if (migrate_postcopy_ram()) {
2143 ram_postcopy_send_discard_bitmap(ms
);
2147 * send rest of state - note things that are doing postcopy
2148 * will notice we're in POSTCOPY_ACTIVE and not actually
2149 * wrap their state up here
2151 migration_rate_set(bandwidth
);
2152 if (migrate_postcopy_ram()) {
2153 /* Ping just for debugging, helps line traces up */
2154 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
2158 * While loading the device state we may trigger page transfer
2159 * requests and the fd must be free to process those, and thus
2160 * the destination must read the whole device state off the fd before
2161 * it starts processing it. Unfortunately the ad-hoc migration format
2162 * doesn't allow the destination to know the size to read without fully
2163 * parsing it through each devices load-state code (especially the open
2164 * coded devices that use get/put).
2165 * So we wrap the device state up in a package with a length at the start;
2166 * to do this we use a qemu_buf to hold the whole of the device state.
2168 bioc
= qio_channel_buffer_new(4096);
2169 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
2170 fb
= qemu_file_new_output(QIO_CHANNEL(bioc
));
2171 object_unref(OBJECT(bioc
));
2174 * Make sure the receiver can get incoming pages before we send the rest
2177 qemu_savevm_send_postcopy_listen(fb
);
2179 qemu_savevm_state_complete_precopy(fb
, false, false);
2180 if (migrate_postcopy_ram()) {
2181 qemu_savevm_send_ping(fb
, 3);
2184 qemu_savevm_send_postcopy_run(fb
);
2186 /* <><> end of stuff going into the package */
2188 /* Last point of recovery; as soon as we send the package the destination
2189 * can open devices and potentially start running.
2190 * Lets just check again we've not got any errors.
2192 ret
= qemu_file_get_error(ms
->to_dst_file
);
2194 error_setg(errp
, "postcopy_start: Migration stream errored (pre package)");
2198 restart_block
= false;
2200 /* Now send that blob */
2201 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2206 /* Send a notify to give a chance for anything that needs to happen
2207 * at the transition to postcopy and after the device state; in particular
2208 * spice needs to trigger a transition now
2210 ms
->postcopy_after_devices
= true;
2211 notifier_list_notify(&migration_state_notifiers
, ms
);
2213 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2215 qemu_mutex_unlock_iothread();
2217 if (migrate_postcopy_ram()) {
2219 * Although this ping is just for debug, it could potentially be
2220 * used for getting a better measurement of downtime at the source.
2222 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2225 if (migrate_release_ram()) {
2226 ram_postcopy_migrated_memory_release(ms
);
2229 ret
= qemu_file_get_error(ms
->to_dst_file
);
2231 error_setg(errp
, "postcopy_start: Migration stream errored");
2232 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2233 MIGRATION_STATUS_FAILED
);
2236 trace_postcopy_preempt_enabled(migrate_postcopy_preempt());
2243 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2244 MIGRATION_STATUS_FAILED
);
2245 if (restart_block
) {
2246 /* A failure happened early enough that we know the destination hasn't
2247 * accessed block devices, so we're safe to recover.
2249 Error
*local_err
= NULL
;
2251 bdrv_activate_all(&local_err
);
2253 error_report_err(local_err
);
2256 qemu_mutex_unlock_iothread();
2261 * migration_maybe_pause: Pause if required to by
2262 * migrate_pause_before_switchover called with the iothread locked
2263 * Returns: 0 on success
2265 static int migration_maybe_pause(MigrationState
*s
,
2266 int *current_active_state
,
2269 if (!migrate_pause_before_switchover()) {
2273 /* Since leaving this state is not atomic with posting the semaphore
2274 * it's possible that someone could have issued multiple migrate_continue
2275 * and the semaphore is incorrectly positive at this point;
2276 * the docs say it's undefined to reinit a semaphore that's already
2277 * init'd, so use timedwait to eat up any existing posts.
2279 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2280 /* This block intentionally left blank */
2284 * If the migration is cancelled when it is in the completion phase,
2285 * the migration state is set to MIGRATION_STATUS_CANCELLING.
2286 * So we don't need to wait a semaphore, otherwise we would always
2287 * wait for the 'pause_sem' semaphore.
2289 if (s
->state
!= MIGRATION_STATUS_CANCELLING
) {
2290 qemu_mutex_unlock_iothread();
2291 migrate_set_state(&s
->state
, *current_active_state
,
2292 MIGRATION_STATUS_PRE_SWITCHOVER
);
2293 qemu_sem_wait(&s
->pause_sem
);
2294 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2296 *current_active_state
= new_state
;
2297 qemu_mutex_lock_iothread();
2300 return s
->state
== new_state
? 0 : -EINVAL
;
2304 * migration_completion: Used by migration_thread when there's not much left.
2305 * The caller 'breaks' the loop when this returns.
2307 * @s: Current migration state
2309 static void migration_completion(MigrationState
*s
)
2312 int current_active_state
= s
->state
;
2314 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2315 qemu_mutex_lock_iothread();
2316 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2317 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2319 s
->vm_old_state
= runstate_get();
2320 global_state_store();
2322 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2323 trace_migration_completion_vm_stop(ret
);
2325 ret
= migration_maybe_pause(s
, ¤t_active_state
,
2326 MIGRATION_STATUS_DEVICE
);
2330 * Inactivate disks except in COLO, and track that we
2331 * have done so in order to remember to reactivate
2332 * them if migration fails or is cancelled.
2334 s
->block_inactive
= !migrate_colo();
2335 migration_rate_set(RATE_LIMIT_DISABLED
);
2336 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
2340 qemu_mutex_unlock_iothread();
2345 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2346 trace_migration_completion_postcopy_end();
2348 qemu_mutex_lock_iothread();
2349 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
2350 qemu_mutex_unlock_iothread();
2353 * Shutdown the postcopy fast path thread. This is only needed
2354 * when dest QEMU binary is old (7.1/7.2). QEMU 8.0+ doesn't need
2357 if (migrate_postcopy_preempt() && s
->preempt_pre_7_2
) {
2358 postcopy_preempt_shutdown_file(s
);
2361 trace_migration_completion_postcopy_end_after_complete();
2366 if (await_return_path_close_on_source(s
)) {
2370 if (qemu_file_get_error(s
->to_dst_file
)) {
2371 trace_migration_completion_file_err();
2375 if (migrate_colo() && s
->state
== MIGRATION_STATUS_ACTIVE
) {
2376 /* COLO does not support postcopy */
2377 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
2378 MIGRATION_STATUS_COLO
);
2380 migrate_set_state(&s
->state
, current_active_state
,
2381 MIGRATION_STATUS_COMPLETED
);
2387 if (s
->block_inactive
&& (s
->state
== MIGRATION_STATUS_ACTIVE
||
2388 s
->state
== MIGRATION_STATUS_DEVICE
)) {
2390 * If not doing postcopy, vm_start() will be called: let's
2391 * regain control on images.
2393 Error
*local_err
= NULL
;
2395 qemu_mutex_lock_iothread();
2396 bdrv_activate_all(&local_err
);
2398 error_report_err(local_err
);
2400 s
->block_inactive
= false;
2402 qemu_mutex_unlock_iothread();
2405 migrate_set_state(&s
->state
, current_active_state
,
2406 MIGRATION_STATUS_FAILED
);
2410 * bg_migration_completion: Used by bg_migration_thread when after all the
2411 * RAM has been saved. The caller 'breaks' the loop when this returns.
2413 * @s: Current migration state
2415 static void bg_migration_completion(MigrationState
*s
)
2417 int current_active_state
= s
->state
;
2419 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2421 * By this moment we have RAM content saved into the migration stream.
2422 * The next step is to flush the non-RAM content (device state)
2423 * right after the ram content. The device state has been stored into
2424 * the temporary buffer before RAM saving started.
2426 qemu_put_buffer(s
->to_dst_file
, s
->bioc
->data
, s
->bioc
->usage
);
2427 qemu_fflush(s
->to_dst_file
);
2428 } else if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
2432 if (qemu_file_get_error(s
->to_dst_file
)) {
2433 trace_migration_completion_file_err();
2437 migrate_set_state(&s
->state
, current_active_state
,
2438 MIGRATION_STATUS_COMPLETED
);
2442 migrate_set_state(&s
->state
, current_active_state
,
2443 MIGRATION_STATUS_FAILED
);
2446 typedef enum MigThrError
{
2447 /* No error detected */
2448 MIG_THR_ERR_NONE
= 0,
2449 /* Detected error, but resumed successfully */
2450 MIG_THR_ERR_RECOVERED
= 1,
2451 /* Detected fatal error, need to exit */
2452 MIG_THR_ERR_FATAL
= 2,
2455 static int postcopy_resume_handshake(MigrationState
*s
)
2457 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
2459 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2460 qemu_sem_wait(&s
->rp_state
.rp_sem
);
2463 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2470 /* Return zero if success, or <0 for error */
2471 static int postcopy_do_resume(MigrationState
*s
)
2476 * Call all the resume_prepare() hooks, so that modules can be
2477 * ready for the migration resume.
2479 ret
= qemu_savevm_state_resume_prepare(s
);
2481 error_report("%s: resume_prepare() failure detected: %d",
2487 * If preempt is enabled, re-establish the preempt channel. Note that
2488 * we do it after resume prepare to make sure the main channel will be
2489 * created before the preempt channel. E.g. with weak network, the
2490 * dest QEMU may get messed up with the preempt and main channels on
2491 * the order of connection setup. This guarantees the correct order.
2493 ret
= postcopy_preempt_establish_channel(s
);
2495 error_report("%s: postcopy_preempt_establish_channel(): %d",
2501 * Last handshake with destination on the resume (destination will
2502 * switch to postcopy-active afterwards)
2504 ret
= postcopy_resume_handshake(s
);
2506 error_report("%s: handshake failed: %d", __func__
, ret
);
2514 * We don't return until we are in a safe state to continue current
2515 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2516 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2518 static MigThrError
postcopy_pause(MigrationState
*s
)
2520 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2526 * Current channel is possibly broken. Release it. Note that this is
2527 * guaranteed even without lock because to_dst_file should only be
2528 * modified by the migration thread. That also guarantees that the
2529 * unregister of yank is safe too without the lock. It should be safe
2530 * even to be within the qemu_file_lock, but we didn't do that to avoid
2531 * taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make
2532 * the qemu_file_lock critical section as small as possible.
2534 assert(s
->to_dst_file
);
2535 migration_ioc_unregister_yank_from_file(s
->to_dst_file
);
2536 qemu_mutex_lock(&s
->qemu_file_lock
);
2537 file
= s
->to_dst_file
;
2538 s
->to_dst_file
= NULL
;
2539 qemu_mutex_unlock(&s
->qemu_file_lock
);
2541 qemu_file_shutdown(file
);
2545 * We're already pausing, so ignore any errors on the return
2546 * path and just wait for the thread to finish. It will be
2547 * re-created when we resume.
2549 await_return_path_close_on_source(s
);
2551 migrate_set_state(&s
->state
, s
->state
,
2552 MIGRATION_STATUS_POSTCOPY_PAUSED
);
2554 error_report("Detected IO failure for postcopy. "
2555 "Migration paused.");
2558 * We wait until things fixed up. Then someone will setup the
2559 * status back for us.
2561 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2562 qemu_sem_wait(&s
->postcopy_pause_sem
);
2565 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2566 /* Woken up by a recover procedure. Give it a shot */
2568 /* Do the resume logic */
2569 if (postcopy_do_resume(s
) == 0) {
2570 /* Let's continue! */
2571 trace_postcopy_pause_continued();
2572 return MIG_THR_ERR_RECOVERED
;
2575 * Something wrong happened during the recovery, let's
2576 * pause again. Pause is always better than throwing
2582 /* This is not right... Time to quit. */
2583 return MIG_THR_ERR_FATAL
;
2588 static MigThrError
migration_detect_error(MigrationState
*s
)
2591 int state
= s
->state
;
2592 Error
*local_error
= NULL
;
2594 if (state
== MIGRATION_STATUS_CANCELLING
||
2595 state
== MIGRATION_STATUS_CANCELLED
) {
2596 /* End the migration, but don't set the state to failed */
2597 return MIG_THR_ERR_FATAL
;
2601 * Try to detect any file errors. Note that postcopy_qemufile_src will
2602 * be NULL when postcopy preempt is not enabled.
2604 ret
= qemu_file_get_error_obj_any(s
->to_dst_file
,
2605 s
->postcopy_qemufile_src
,
2608 /* Everything is fine */
2609 assert(!local_error
);
2610 return MIG_THR_ERR_NONE
;
2614 migrate_set_error(s
, local_error
);
2615 error_free(local_error
);
2618 if (state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
) {
2620 * For postcopy, we allow the network to be down for a
2621 * while. After that, it can be continued by a
2624 return postcopy_pause(s
);
2627 * For precopy (or postcopy with error outside IO), we fail
2630 migrate_set_state(&s
->state
, state
, MIGRATION_STATUS_FAILED
);
2631 trace_migration_thread_file_err();
2633 /* Time to stop the migration, now. */
2634 return MIG_THR_ERR_FATAL
;
2638 static void migration_calculate_complete(MigrationState
*s
)
2640 uint64_t bytes
= migration_transferred_bytes(s
->to_dst_file
);
2641 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2642 int64_t transfer_time
;
2644 s
->total_time
= end_time
- s
->start_time
;
2647 * It's still not set, so we are precopy migration. For
2648 * postcopy, downtime is calculated during postcopy_start().
2650 s
->downtime
= end_time
- s
->downtime_start
;
2653 transfer_time
= s
->total_time
- s
->setup_time
;
2654 if (transfer_time
) {
2655 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
2659 static void update_iteration_initial_status(MigrationState
*s
)
2662 * Update these three fields at the same time to avoid mismatch info lead
2663 * wrong speed calculation.
2665 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2666 s
->iteration_initial_bytes
= migration_transferred_bytes(s
->to_dst_file
);
2667 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
2670 static void migration_update_counters(MigrationState
*s
,
2671 int64_t current_time
)
2673 uint64_t transferred
, transferred_pages
, time_spent
;
2674 uint64_t current_bytes
; /* bytes transferred since the beginning */
2677 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
2681 current_bytes
= migration_transferred_bytes(s
->to_dst_file
);
2682 transferred
= current_bytes
- s
->iteration_initial_bytes
;
2683 time_spent
= current_time
- s
->iteration_start_time
;
2684 bandwidth
= (double)transferred
/ time_spent
;
2685 s
->threshold_size
= bandwidth
* migrate_downtime_limit();
2687 s
->mbps
= (((double) transferred
* 8.0) /
2688 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2690 transferred_pages
= ram_get_total_transferred_pages() -
2691 s
->iteration_initial_pages
;
2692 s
->pages_per_second
= (double) transferred_pages
/
2693 (((double) time_spent
/ 1000.0));
2696 * if we haven't sent anything, we don't want to
2697 * recalculate. 10000 is a small enough number for our purposes
2699 if (stat64_get(&mig_stats
.dirty_pages_rate
) &&
2700 transferred
> 10000) {
2701 s
->expected_downtime
=
2702 stat64_get(&mig_stats
.dirty_bytes_last_sync
) / bandwidth
;
2705 migration_rate_reset(s
->to_dst_file
);
2707 update_iteration_initial_status(s
);
2709 trace_migrate_transferred(transferred
, time_spent
,
2710 bandwidth
, s
->threshold_size
);
2713 static bool migration_can_switchover(MigrationState
*s
)
2715 if (!migrate_switchover_ack()) {
2719 /* No reason to wait for switchover ACK if VM is stopped */
2720 if (!runstate_is_running()) {
2724 return s
->switchover_acked
;
2727 /* Migration thread iteration status */
2729 MIG_ITERATE_RESUME
, /* Resume current iteration */
2730 MIG_ITERATE_SKIP
, /* Skip current iteration */
2731 MIG_ITERATE_BREAK
, /* Break the loop */
2735 * Return true if continue to the next iteration directly, false
2738 static MigIterateState
migration_iteration_run(MigrationState
*s
)
2740 uint64_t must_precopy
, can_postcopy
;
2741 Error
*local_err
= NULL
;
2742 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
2743 bool can_switchover
= migration_can_switchover(s
);
2745 qemu_savevm_state_pending_estimate(&must_precopy
, &can_postcopy
);
2746 uint64_t pending_size
= must_precopy
+ can_postcopy
;
2748 trace_migrate_pending_estimate(pending_size
, must_precopy
, can_postcopy
);
2750 if (must_precopy
<= s
->threshold_size
) {
2751 qemu_savevm_state_pending_exact(&must_precopy
, &can_postcopy
);
2752 pending_size
= must_precopy
+ can_postcopy
;
2753 trace_migrate_pending_exact(pending_size
, must_precopy
, can_postcopy
);
2756 if ((!pending_size
|| pending_size
< s
->threshold_size
) && can_switchover
) {
2757 trace_migration_thread_low_pending(pending_size
);
2758 migration_completion(s
);
2759 return MIG_ITERATE_BREAK
;
2762 /* Still a significant amount to transfer */
2763 if (!in_postcopy
&& must_precopy
<= s
->threshold_size
&& can_switchover
&&
2764 qatomic_read(&s
->start_postcopy
)) {
2765 if (postcopy_start(s
, &local_err
)) {
2766 migrate_set_error(s
, local_err
);
2767 error_report_err(local_err
);
2769 return MIG_ITERATE_SKIP
;
2772 /* Just another iteration step */
2773 qemu_savevm_state_iterate(s
->to_dst_file
, in_postcopy
);
2774 return MIG_ITERATE_RESUME
;
2777 static void migration_iteration_finish(MigrationState
*s
)
2779 /* If we enabled cpu throttling for auto-converge, turn it off. */
2780 cpu_throttle_stop();
2782 qemu_mutex_lock_iothread();
2784 case MIGRATION_STATUS_COMPLETED
:
2785 migration_calculate_complete(s
);
2786 runstate_set(RUN_STATE_POSTMIGRATE
);
2788 case MIGRATION_STATUS_COLO
:
2789 assert(migrate_colo());
2790 migrate_start_colo_process(s
);
2791 s
->vm_old_state
= RUN_STATE_RUNNING
;
2793 case MIGRATION_STATUS_FAILED
:
2794 case MIGRATION_STATUS_CANCELLED
:
2795 case MIGRATION_STATUS_CANCELLING
:
2796 if (s
->vm_old_state
== RUN_STATE_RUNNING
) {
2797 if (!runstate_check(RUN_STATE_SHUTDOWN
)) {
2801 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2802 runstate_set(s
->vm_old_state
);
2808 /* Should not reach here, but if so, forgive the VM. */
2809 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
2812 migrate_fd_cleanup_schedule(s
);
2813 qemu_mutex_unlock_iothread();
2816 static void bg_migration_iteration_finish(MigrationState
*s
)
2819 * Stop tracking RAM writes - un-protect memory, un-register UFFD
2820 * memory ranges, flush kernel wait queues and wake up threads
2821 * waiting for write fault to be resolved.
2823 ram_write_tracking_stop();
2825 qemu_mutex_lock_iothread();
2827 case MIGRATION_STATUS_COMPLETED
:
2828 migration_calculate_complete(s
);
2831 case MIGRATION_STATUS_ACTIVE
:
2832 case MIGRATION_STATUS_FAILED
:
2833 case MIGRATION_STATUS_CANCELLED
:
2834 case MIGRATION_STATUS_CANCELLING
:
2838 /* Should not reach here, but if so, forgive the VM. */
2839 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
2843 migrate_fd_cleanup_schedule(s
);
2844 qemu_mutex_unlock_iothread();
2848 * Return true if continue to the next iteration directly, false
2851 static MigIterateState
bg_migration_iteration_run(MigrationState
*s
)
2855 res
= qemu_savevm_state_iterate(s
->to_dst_file
, false);
2857 bg_migration_completion(s
);
2858 return MIG_ITERATE_BREAK
;
2861 return MIG_ITERATE_RESUME
;
2864 void migration_make_urgent_request(void)
2866 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
2869 void migration_consume_urgent_request(void)
2871 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
2874 /* Returns true if the rate limiting was broken by an urgent request */
2875 bool migration_rate_limit(void)
2877 int64_t now
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2878 MigrationState
*s
= migrate_get_current();
2880 bool urgent
= false;
2881 migration_update_counters(s
, now
);
2882 if (migration_rate_exceeded(s
->to_dst_file
)) {
2884 if (qemu_file_get_error(s
->to_dst_file
)) {
2888 * Wait for a delay to do rate limiting OR
2889 * something urgent to post the semaphore.
2891 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- now
;
2892 trace_migration_rate_limit_pre(ms
);
2893 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
2895 * We were woken by one or more urgent things but
2896 * the timedwait will have consumed one of them.
2897 * The service routine for the urgent wake will dec
2898 * the semaphore itself for each item it consumes,
2899 * so add this one we just eat back.
2901 qemu_sem_post(&s
->rate_limit_sem
);
2904 trace_migration_rate_limit_post(urgent
);
2910 * if failover devices are present, wait they are completely
2914 static void qemu_savevm_wait_unplug(MigrationState
*s
, int old_state
,
2917 if (qemu_savevm_state_guest_unplug_pending()) {
2918 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_WAIT_UNPLUG
);
2920 while (s
->state
== MIGRATION_STATUS_WAIT_UNPLUG
&&
2921 qemu_savevm_state_guest_unplug_pending()) {
2922 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
2924 if (s
->state
!= MIGRATION_STATUS_WAIT_UNPLUG
) {
2925 int timeout
= 120; /* 30 seconds */
2927 * migration has been canceled
2928 * but as we have started an unplug we must wait the end
2929 * to be able to plug back the card
2931 while (timeout
-- && qemu_savevm_state_guest_unplug_pending()) {
2932 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
2934 if (qemu_savevm_state_guest_unplug_pending() &&
2936 warn_report("migration: partially unplugged device on "
2941 migrate_set_state(&s
->state
, MIGRATION_STATUS_WAIT_UNPLUG
, new_state
);
2943 migrate_set_state(&s
->state
, old_state
, new_state
);
2948 * Master migration thread on the source VM.
2949 * It drives the migration and pumps the data down the outgoing channel.
2951 static void *migration_thread(void *opaque
)
2953 MigrationState
*s
= opaque
;
2954 MigrationThread
*thread
= NULL
;
2955 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
2956 MigThrError thr_error
;
2957 bool urgent
= false;
2959 thread
= migration_threads_add("live_migration", qemu_get_thread_id());
2961 rcu_register_thread();
2963 object_ref(OBJECT(s
));
2964 update_iteration_initial_status(s
);
2966 qemu_savevm_state_header(s
->to_dst_file
);
2969 * If we opened the return path, we need to make sure dst has it
2972 if (s
->rp_state
.rp_thread_created
) {
2973 /* Now tell the dest that it should open its end so it can reply */
2974 qemu_savevm_send_open_return_path(s
->to_dst_file
);
2976 /* And do a ping that will make stuff easier to debug */
2977 qemu_savevm_send_ping(s
->to_dst_file
, 1);
2980 if (migrate_postcopy()) {
2982 * Tell the destination that we *might* want to do postcopy later;
2983 * if the other end can't do postcopy it should fail now, nice and
2986 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
2989 if (migrate_colo()) {
2990 /* Notify migration destination that we enable COLO */
2991 qemu_savevm_send_colo_enable(s
->to_dst_file
);
2994 qemu_savevm_state_setup(s
->to_dst_file
);
2996 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
2997 MIGRATION_STATUS_ACTIVE
);
2999 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3001 trace_migration_thread_setup_complete();
3003 while (migration_is_active(s
)) {
3004 if (urgent
|| !migration_rate_exceeded(s
->to_dst_file
)) {
3005 MigIterateState iter_state
= migration_iteration_run(s
);
3006 if (iter_state
== MIG_ITERATE_SKIP
) {
3008 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3014 * Try to detect any kind of failures, and see whether we
3015 * should stop the migration now.
3017 thr_error
= migration_detect_error(s
);
3018 if (thr_error
== MIG_THR_ERR_FATAL
) {
3019 /* Stop migration */
3021 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
3023 * Just recovered from a e.g. network failure, reset all
3024 * the local variables. This is important to avoid
3025 * breaking transferred_bytes and bandwidth calculation
3027 update_iteration_initial_status(s
);
3030 urgent
= migration_rate_limit();
3033 trace_migration_thread_after_loop();
3034 migration_iteration_finish(s
);
3035 object_unref(OBJECT(s
));
3036 rcu_unregister_thread();
3037 migration_threads_remove(thread
);
3041 static void bg_migration_vm_start_bh(void *opaque
)
3043 MigrationState
*s
= opaque
;
3045 qemu_bh_delete(s
->vm_start_bh
);
3046 s
->vm_start_bh
= NULL
;
3049 s
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - s
->downtime_start
;
3053 * Background snapshot thread, based on live migration code.
3054 * This is an alternative implementation of live migration mechanism
3055 * introduced specifically to support background snapshots.
3057 * It takes advantage of userfault_fd write protection mechanism introduced
3058 * in v5.7 kernel. Compared to existing dirty page logging migration much
3059 * lesser stream traffic is produced resulting in smaller snapshot images,
3060 * simply cause of no page duplicates can get into the stream.
3062 * Another key point is that generated vmstate stream reflects machine state
3063 * 'frozen' at the beginning of snapshot creation compared to dirty page logging
3064 * mechanism, which effectively results in that saved snapshot is the state of VM
3065 * at the end of the process.
3067 static void *bg_migration_thread(void *opaque
)
3069 MigrationState
*s
= opaque
;
3070 int64_t setup_start
;
3071 MigThrError thr_error
;
3073 bool early_fail
= true;
3075 rcu_register_thread();
3076 object_ref(OBJECT(s
));
3078 migration_rate_set(RATE_LIMIT_DISABLED
);
3080 setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3082 * We want to save vmstate for the moment when migration has been
3083 * initiated but also we want to save RAM content while VM is running.
3084 * The RAM content should appear first in the vmstate. So, we first
3085 * stash the non-RAM part of the vmstate to the temporary buffer,
3086 * then write RAM part of the vmstate to the migration stream
3087 * with vCPUs running and, finally, write stashed non-RAM part of
3088 * the vmstate from the buffer to the migration stream.
3090 s
->bioc
= qio_channel_buffer_new(512 * 1024);
3091 qio_channel_set_name(QIO_CHANNEL(s
->bioc
), "vmstate-buffer");
3092 fb
= qemu_file_new_output(QIO_CHANNEL(s
->bioc
));
3093 object_unref(OBJECT(s
->bioc
));
3095 update_iteration_initial_status(s
);
3098 * Prepare for tracking memory writes with UFFD-WP - populate
3099 * RAM pages before protecting.
3102 ram_write_tracking_prepare();
3105 qemu_savevm_state_header(s
->to_dst_file
);
3106 qemu_savevm_state_setup(s
->to_dst_file
);
3108 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
3109 MIGRATION_STATUS_ACTIVE
);
3111 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3113 trace_migration_thread_setup_complete();
3114 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3116 qemu_mutex_lock_iothread();
3119 * If VM is currently in suspended state, then, to make a valid runstate
3120 * transition in vm_stop_force_state() we need to wakeup it up.
3122 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3123 s
->vm_old_state
= runstate_get();
3125 global_state_store();
3126 /* Forcibly stop VM before saving state of vCPUs and devices */
3127 if (vm_stop_force_state(RUN_STATE_PAUSED
)) {
3131 * Put vCPUs in sync with shadow context structures, then
3132 * save their state to channel-buffer along with devices.
3134 cpu_synchronize_all_states();
3135 if (qemu_savevm_state_complete_precopy_non_iterable(fb
, false, false)) {
3139 * Since we are going to get non-iterable state data directly
3140 * from s->bioc->data, explicit flush is needed here.
3144 /* Now initialize UFFD context and start tracking RAM writes */
3145 if (ram_write_tracking_start()) {
3151 * Start VM from BH handler to avoid write-fault lock here.
3152 * UFFD-WP protection for the whole RAM is already enabled so
3153 * calling VM state change notifiers from vm_start() would initiate
3154 * writes to virtio VQs memory which is in write-protected region.
3156 s
->vm_start_bh
= qemu_bh_new(bg_migration_vm_start_bh
, s
);
3157 qemu_bh_schedule(s
->vm_start_bh
);
3159 qemu_mutex_unlock_iothread();
3161 while (migration_is_active(s
)) {
3162 MigIterateState iter_state
= bg_migration_iteration_run(s
);
3163 if (iter_state
== MIG_ITERATE_SKIP
) {
3165 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3170 * Try to detect any kind of failures, and see whether we
3171 * should stop the migration now.
3173 thr_error
= migration_detect_error(s
);
3174 if (thr_error
== MIG_THR_ERR_FATAL
) {
3175 /* Stop migration */
3179 migration_update_counters(s
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
));
3182 trace_migration_thread_after_loop();
3186 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
3187 MIGRATION_STATUS_FAILED
);
3188 qemu_mutex_unlock_iothread();
3191 bg_migration_iteration_finish(s
);
3194 object_unref(OBJECT(s
));
3195 rcu_unregister_thread();
3200 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
3202 Error
*local_err
= NULL
;
3203 uint64_t rate_limit
;
3204 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
3207 * If there's a previous error, free it and prepare for another one.
3208 * Meanwhile if migration completes successfully, there won't have an error
3209 * dumped when calling migrate_fd_cleanup().
3211 migrate_error_free(s
);
3213 s
->expected_downtime
= migrate_downtime_limit();
3215 assert(s
->cleanup_bh
);
3217 assert(!s
->cleanup_bh
);
3218 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup_bh
, s
);
3221 migrate_fd_error(s
, error_in
);
3224 * Don't do cleanup for resume if channel is invalid, but only dump
3225 * the error. We wait for another channel connect from the user.
3226 * The error_report still gives HMP user a hint on what failed.
3227 * It's normally done in migrate_fd_cleanup(), but call it here
3230 error_report_err(error_copy(s
->error
));
3232 migrate_fd_cleanup(s
);
3238 /* This is a resumed migration */
3239 rate_limit
= migrate_max_postcopy_bandwidth();
3241 /* This is a fresh new migration */
3242 rate_limit
= migrate_max_bandwidth();
3244 /* Notify before starting migration thread */
3245 notifier_list_notify(&migration_state_notifiers
, s
);
3248 migration_rate_set(rate_limit
);
3249 qemu_file_set_blocking(s
->to_dst_file
, true);
3252 * Open the return path. For postcopy, it is used exclusively. For
3253 * precopy, only if user specified "return-path" capability would
3254 * QEMU uses the return path.
3256 if (migrate_postcopy_ram() || migrate_return_path()) {
3257 if (open_return_path_on_source(s
)) {
3258 error_setg(&local_err
, "Unable to open return-path for postcopy");
3259 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
3260 migrate_set_error(s
, local_err
);
3261 error_report_err(local_err
);
3262 migrate_fd_cleanup(s
);
3268 * This needs to be done before resuming a postcopy. Note: for newer
3269 * QEMUs we will delay the channel creation until postcopy_start(), to
3270 * avoid disorder of channel creations.
3272 if (migrate_postcopy_preempt() && s
->preempt_pre_7_2
) {
3273 postcopy_preempt_setup(s
);
3277 /* Wakeup the main migration thread to do the recovery */
3278 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
3279 MIGRATION_STATUS_POSTCOPY_RECOVER
);
3280 qemu_sem_post(&s
->postcopy_pause_sem
);
3284 if (multifd_save_setup(&local_err
) != 0) {
3285 migrate_set_error(s
, local_err
);
3286 error_report_err(local_err
);
3287 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3288 MIGRATION_STATUS_FAILED
);
3289 migrate_fd_cleanup(s
);
3293 if (migrate_background_snapshot()) {
3294 qemu_thread_create(&s
->thread
, "bg_snapshot",
3295 bg_migration_thread
, s
, QEMU_THREAD_JOINABLE
);
3297 qemu_thread_create(&s
->thread
, "live_migration",
3298 migration_thread
, s
, QEMU_THREAD_JOINABLE
);
3300 s
->migration_thread_running
= true;
3303 static void migration_class_init(ObjectClass
*klass
, void *data
)
3305 DeviceClass
*dc
= DEVICE_CLASS(klass
);
3307 dc
->user_creatable
= false;
3308 device_class_set_props(dc
, migration_properties
);
3311 static void migration_instance_finalize(Object
*obj
)
3313 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3315 qemu_mutex_destroy(&ms
->error_mutex
);
3316 qemu_mutex_destroy(&ms
->qemu_file_lock
);
3317 qemu_sem_destroy(&ms
->wait_unplug_sem
);
3318 qemu_sem_destroy(&ms
->rate_limit_sem
);
3319 qemu_sem_destroy(&ms
->pause_sem
);
3320 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
3321 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
3322 qemu_sem_destroy(&ms
->rp_state
.rp_pong_acks
);
3323 qemu_sem_destroy(&ms
->postcopy_qemufile_src_sem
);
3324 error_free(ms
->error
);
3327 static void migration_instance_init(Object
*obj
)
3329 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3331 ms
->state
= MIGRATION_STATUS_NONE
;
3333 ms
->pages_per_second
= -1;
3334 qemu_sem_init(&ms
->pause_sem
, 0);
3335 qemu_mutex_init(&ms
->error_mutex
);
3337 migrate_params_init(&ms
->parameters
);
3339 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
3340 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
3341 qemu_sem_init(&ms
->rp_state
.rp_pong_acks
, 0);
3342 qemu_sem_init(&ms
->rate_limit_sem
, 0);
3343 qemu_sem_init(&ms
->wait_unplug_sem
, 0);
3344 qemu_sem_init(&ms
->postcopy_qemufile_src_sem
, 0);
3345 qemu_mutex_init(&ms
->qemu_file_lock
);
3349 * Return true if check pass, false otherwise. Error will be put
3350 * inside errp if provided.
3352 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
3354 /* Assuming all off */
3355 bool old_caps
[MIGRATION_CAPABILITY__MAX
] = { 0 };
3357 if (!migrate_params_check(&ms
->parameters
, errp
)) {
3361 return migrate_caps_check(old_caps
, ms
->capabilities
, errp
);
3364 static const TypeInfo migration_type
= {
3365 .name
= TYPE_MIGRATION
,
3367 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3368 * not created using qdev_new(), it is not attached to the qdev
3369 * device tree, and it is never realized.
3371 * TODO: Make this TYPE_OBJECT once QOM provides something like
3372 * TYPE_DEVICE's "-global" properties.
3374 .parent
= TYPE_DEVICE
,
3375 .class_init
= migration_class_init
,
3376 .class_size
= sizeof(MigrationClass
),
3377 .instance_size
= sizeof(MigrationState
),
3378 .instance_init
= migration_instance_init
,
3379 .instance_finalize
= migration_instance_finalize
,
3382 static void register_migration_types(void)
3384 type_register_static(&migration_type
);
3387 type_init(register_migration_types
);