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"
24 #include "sysemu/runstate.h"
25 #include "sysemu/sysemu.h"
26 #include "sysemu/cpu-throttle.h"
29 #include "ram-compress.h"
30 #include "migration/global_state.h"
31 #include "migration/misc.h"
32 #include "migration.h"
33 #include "migration-stats.h"
35 #include "qemu-file.h"
37 #include "migration/vmstate.h"
38 #include "block/block.h"
39 #include "qapi/error.h"
40 #include "qapi/clone-visitor.h"
41 #include "qapi/qapi-visit-migration.h"
42 #include "qapi/qapi-visit-sockets.h"
43 #include "qapi/qapi-commands-migration.h"
44 #include "qapi/qapi-events-migration.h"
45 #include "qapi/qmp/qerror.h"
46 #include "qapi/qmp/qnull.h"
49 #include "postcopy-ram.h"
50 #include "qemu/thread.h"
52 #include "exec/target_page.h"
53 #include "io/channel-buffer.h"
54 #include "io/channel-tls.h"
55 #include "migration/colo.h"
56 #include "hw/boards.h"
57 #include "monitor/monitor.h"
58 #include "net/announce.h"
59 #include "qemu/queue.h"
61 #include "threadinfo.h"
62 #include "qemu/yank.h"
63 #include "sysemu/cpus.h"
64 #include "yank_functions.h"
65 #include "sysemu/qtest.h"
68 static NotifierList migration_state_notifiers
=
69 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
71 /* Messages sent on the return path from destination to source */
72 enum mig_rp_message_type
{
73 MIG_RP_MSG_INVALID
= 0, /* Must be 0 */
74 MIG_RP_MSG_SHUT
, /* sibling will not send any more RP messages */
75 MIG_RP_MSG_PONG
, /* Response to a PING; data (seq: be32 ) */
77 MIG_RP_MSG_REQ_PAGES_ID
, /* data (start: be64, len: be32, id: string) */
78 MIG_RP_MSG_REQ_PAGES
, /* data (start: be64, len: be32) */
79 MIG_RP_MSG_RECV_BITMAP
, /* send recved_bitmap back to source */
80 MIG_RP_MSG_RESUME_ACK
, /* tell source that we are ready to resume */
85 /* When we add fault tolerance, we could have several
86 migrations at once. For now we don't need to add
87 dynamic creation of migration */
89 static MigrationState
*current_migration
;
90 static MigrationIncomingState
*current_incoming
;
92 static GSList
*migration_blockers
;
94 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
95 static int migration_maybe_pause(MigrationState
*s
,
96 int *current_active_state
,
98 static void migrate_fd_cancel(MigrationState
*s
);
100 static bool migration_needs_multiple_sockets(void)
102 return migrate_multifd() || migrate_postcopy_preempt();
105 static bool uri_supports_multi_channels(const char *uri
)
107 return strstart(uri
, "tcp:", NULL
) || strstart(uri
, "unix:", NULL
) ||
108 strstart(uri
, "vsock:", NULL
);
112 migration_channels_and_uri_compatible(const char *uri
, Error
**errp
)
114 if (migration_needs_multiple_sockets() &&
115 !uri_supports_multi_channels(uri
)) {
116 error_setg(errp
, "Migration requires multi-channel URIs (e.g. tcp)");
123 static gint
page_request_addr_cmp(gconstpointer ap
, gconstpointer bp
)
125 uintptr_t a
= (uintptr_t) ap
, b
= (uintptr_t) bp
;
127 return (a
> b
) - (a
< b
);
130 void migration_object_init(void)
132 /* This can only be called once. */
133 assert(!current_migration
);
134 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
137 * Init the migrate incoming object as well no matter whether
138 * we'll use it or not.
140 assert(!current_incoming
);
141 current_incoming
= g_new0(MigrationIncomingState
, 1);
142 current_incoming
->state
= MIGRATION_STATUS_NONE
;
143 current_incoming
->postcopy_remote_fds
=
144 g_array_new(FALSE
, TRUE
, sizeof(struct PostCopyFD
));
145 qemu_mutex_init(¤t_incoming
->rp_mutex
);
146 qemu_mutex_init(¤t_incoming
->postcopy_prio_thread_mutex
);
147 qemu_event_init(¤t_incoming
->main_thread_load_event
, false);
148 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_dst
, 0);
149 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fault
, 0);
150 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fast_load
, 0);
151 qemu_sem_init(¤t_incoming
->postcopy_qemufile_dst_done
, 0);
153 qemu_mutex_init(¤t_incoming
->page_request_mutex
);
154 current_incoming
->page_requested
= g_tree_new(page_request_addr_cmp
);
156 migration_object_check(current_migration
, &error_fatal
);
160 dirty_bitmap_mig_init();
163 void migration_cancel(const Error
*error
)
166 migrate_set_error(current_migration
, error
);
168 migrate_fd_cancel(current_migration
);
171 void migration_shutdown(void)
174 * When the QEMU main thread exit, the COLO thread
175 * may wait a semaphore. So, we should wakeup the
176 * COLO thread before migration shutdown.
180 * Cancel the current migration - that will (eventually)
181 * stop the migration using this structure
183 migration_cancel(NULL
);
184 object_unref(OBJECT(current_migration
));
187 * Cancel outgoing migration of dirty bitmaps. It should
188 * at least unref used block nodes.
190 dirty_bitmap_mig_cancel_outgoing();
193 * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
194 * are non-critical data, and their loss never considered as
197 dirty_bitmap_mig_cancel_incoming();
201 MigrationState
*migrate_get_current(void)
203 /* This can only be called after the object created. */
204 assert(current_migration
);
205 return current_migration
;
208 MigrationIncomingState
*migration_incoming_get_current(void)
210 assert(current_incoming
);
211 return current_incoming
;
214 void migration_incoming_transport_cleanup(MigrationIncomingState
*mis
)
216 if (mis
->socket_address_list
) {
217 qapi_free_SocketAddressList(mis
->socket_address_list
);
218 mis
->socket_address_list
= NULL
;
221 if (mis
->transport_cleanup
) {
222 mis
->transport_cleanup(mis
->transport_data
);
223 mis
->transport_data
= mis
->transport_cleanup
= NULL
;
227 void migration_incoming_state_destroy(void)
229 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
231 multifd_load_cleanup();
232 compress_threads_load_cleanup();
234 if (mis
->to_src_file
) {
235 /* Tell source that we are done */
236 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
237 qemu_fclose(mis
->to_src_file
);
238 mis
->to_src_file
= NULL
;
241 if (mis
->from_src_file
) {
242 migration_ioc_unregister_yank_from_file(mis
->from_src_file
);
243 qemu_fclose(mis
->from_src_file
);
244 mis
->from_src_file
= NULL
;
246 if (mis
->postcopy_remote_fds
) {
247 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
248 mis
->postcopy_remote_fds
= NULL
;
251 migration_incoming_transport_cleanup(mis
);
252 qemu_event_reset(&mis
->main_thread_load_event
);
254 if (mis
->page_requested
) {
255 g_tree_destroy(mis
->page_requested
);
256 mis
->page_requested
= NULL
;
259 if (mis
->postcopy_qemufile_dst
) {
260 migration_ioc_unregister_yank_from_file(mis
->postcopy_qemufile_dst
);
261 qemu_fclose(mis
->postcopy_qemufile_dst
);
262 mis
->postcopy_qemufile_dst
= NULL
;
265 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
268 static void migrate_generate_event(int new_state
)
270 if (migrate_events()) {
271 qapi_event_send_migration(new_state
);
276 * Send a message on the return channel back to the source
279 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
280 enum mig_rp_message_type message_type
,
281 uint16_t len
, void *data
)
285 trace_migrate_send_rp_message((int)message_type
, len
);
286 QEMU_LOCK_GUARD(&mis
->rp_mutex
);
289 * It's possible that the file handle got lost due to network
292 if (!mis
->to_src_file
) {
297 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
298 qemu_put_be16(mis
->to_src_file
, len
);
299 qemu_put_buffer(mis
->to_src_file
, data
, len
);
300 qemu_fflush(mis
->to_src_file
);
302 /* It's possible that qemu file got error during sending */
303 ret
= qemu_file_get_error(mis
->to_src_file
);
308 /* Request one page from the source VM at the given start address.
309 * rb: the RAMBlock to request the page in
310 * Start: Address offset within the RB
311 * Len: Length in bytes required - must be a multiple of pagesize
313 int migrate_send_rp_message_req_pages(MigrationIncomingState
*mis
,
314 RAMBlock
*rb
, ram_addr_t start
)
316 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
317 size_t msglen
= 12; /* start + len */
318 size_t len
= qemu_ram_pagesize(rb
);
319 enum mig_rp_message_type msg_type
;
323 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
324 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
327 * We maintain the last ramblock that we requested for page. Note that we
328 * don't need locking because this function will only be called within the
329 * postcopy ram fault thread.
331 if (rb
!= mis
->last_rb
) {
334 rbname
= qemu_ram_get_idstr(rb
);
335 rbname_len
= strlen(rbname
);
337 assert(rbname_len
< 256);
339 bufc
[msglen
++] = rbname_len
;
340 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
341 msglen
+= rbname_len
;
342 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
344 msg_type
= MIG_RP_MSG_REQ_PAGES
;
347 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
350 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
,
351 RAMBlock
*rb
, ram_addr_t start
, uint64_t haddr
)
353 void *aligned
= (void *)(uintptr_t)ROUND_DOWN(haddr
, qemu_ram_pagesize(rb
));
354 bool received
= false;
356 WITH_QEMU_LOCK_GUARD(&mis
->page_request_mutex
) {
357 received
= ramblock_recv_bitmap_test_byte_offset(rb
, start
);
358 if (!received
&& !g_tree_lookup(mis
->page_requested
, aligned
)) {
360 * The page has not been received, and it's not yet in the page
361 * request list. Queue it. Set the value of element to 1, so that
362 * things like g_tree_lookup() will return TRUE (1) when found.
364 g_tree_insert(mis
->page_requested
, aligned
, (gpointer
)1);
365 mis
->page_requested_count
++;
366 trace_postcopy_page_req_add(aligned
, mis
->page_requested_count
);
371 * If the page is there, skip sending the message. We don't even need the
372 * lock because as long as the page arrived, it'll be there forever.
378 return migrate_send_rp_message_req_pages(mis
, rb
, start
);
381 static bool migration_colo_enabled
;
382 bool migration_incoming_colo_enabled(void)
384 return migration_colo_enabled
;
387 void migration_incoming_disable_colo(void)
389 ram_block_discard_disable(false);
390 migration_colo_enabled
= false;
393 int migration_incoming_enable_colo(void)
395 if (ram_block_discard_disable(true)) {
396 error_report("COLO: cannot disable RAM discard");
399 migration_colo_enabled
= true;
403 void migrate_add_address(SocketAddress
*address
)
405 MigrationIncomingState
*mis
= migration_incoming_get_current();
407 QAPI_LIST_PREPEND(mis
->socket_address_list
,
408 QAPI_CLONE(SocketAddress
, address
));
411 static void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
413 const char *p
= NULL
;
415 /* URI is not suitable for migration? */
416 if (!migration_channels_and_uri_compatible(uri
, errp
)) {
420 qapi_event_send_migration(MIGRATION_STATUS_SETUP
);
421 if (strstart(uri
, "tcp:", &p
) ||
422 strstart(uri
, "unix:", NULL
) ||
423 strstart(uri
, "vsock:", NULL
)) {
424 socket_start_incoming_migration(p
? p
: uri
, errp
);
426 } else if (strstart(uri
, "rdma:", &p
)) {
427 rdma_start_incoming_migration(p
, errp
);
429 } else if (strstart(uri
, "exec:", &p
)) {
430 exec_start_incoming_migration(p
, errp
);
431 } else if (strstart(uri
, "fd:", &p
)) {
432 fd_start_incoming_migration(p
, errp
);
434 error_setg(errp
, "unknown migration protocol: %s", uri
);
438 static void process_incoming_migration_bh(void *opaque
)
440 Error
*local_err
= NULL
;
441 MigrationIncomingState
*mis
= opaque
;
443 /* If capability late_block_activate is set:
444 * Only fire up the block code now if we're going to restart the
445 * VM, else 'cont' will do it.
446 * This causes file locking to happen; so we don't want it to happen
447 * unless we really are starting the VM.
449 if (!migrate_late_block_activate() ||
450 (autostart
&& (!global_state_received() ||
451 global_state_get_runstate() == RUN_STATE_RUNNING
))) {
452 /* Make sure all file formats throw away their mutable metadata.
453 * If we get an error here, just don't restart the VM yet. */
454 bdrv_activate_all(&local_err
);
456 error_report_err(local_err
);
463 * This must happen after all error conditions are dealt with and
464 * we're sure the VM is going to be running on this host.
466 qemu_announce_self(&mis
->announce_timer
, migrate_announce_params());
468 multifd_load_shutdown();
470 dirty_bitmap_mig_before_vm_start();
472 if (!global_state_received() ||
473 global_state_get_runstate() == RUN_STATE_RUNNING
) {
477 runstate_set(RUN_STATE_PAUSED
);
479 } else if (migration_incoming_colo_enabled()) {
480 migration_incoming_disable_colo();
483 runstate_set(global_state_get_runstate());
486 * This must happen after any state changes since as soon as an external
487 * observer sees this event they might start to prod at the VM assuming
490 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
491 MIGRATION_STATUS_COMPLETED
);
492 qemu_bh_delete(mis
->bh
);
493 migration_incoming_state_destroy();
496 static void coroutine_fn
497 process_incoming_migration_co(void *opaque
)
499 MigrationIncomingState
*mis
= migration_incoming_get_current();
502 Error
*local_err
= NULL
;
504 assert(mis
->from_src_file
);
506 if (compress_threads_load_setup(mis
->from_src_file
)) {
507 error_report("Failed to setup decompress threads");
511 mis
->migration_incoming_co
= qemu_coroutine_self();
512 mis
->largest_page_size
= qemu_ram_pagesize_largest();
513 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
514 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
515 MIGRATION_STATUS_ACTIVE
);
516 ret
= qemu_loadvm_state(mis
->from_src_file
);
518 ps
= postcopy_state_get();
519 trace_process_incoming_migration_co_end(ret
, ps
);
520 if (ps
!= POSTCOPY_INCOMING_NONE
) {
521 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
523 * Where a migration had postcopy enabled (and thus went to advise)
524 * but managed to complete within the precopy period, we can use
527 postcopy_ram_incoming_cleanup(mis
);
528 } else if (ret
>= 0) {
530 * Postcopy was started, cleanup should happen at the end of the
533 trace_process_incoming_migration_co_postcopy_end_main();
536 /* Else if something went wrong then just fall out of the normal exit */
539 /* we get COLO info, and know if we are in COLO mode */
540 if (!ret
&& migration_incoming_colo_enabled()) {
541 /* Make sure all file formats throw away their mutable metadata */
542 bdrv_activate_all(&local_err
);
544 error_report_err(local_err
);
548 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
549 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
550 mis
->have_colo_incoming_thread
= true;
551 qemu_coroutine_yield();
553 qemu_mutex_unlock_iothread();
554 /* Wait checkpoint incoming thread exit before free resource */
555 qemu_thread_join(&mis
->colo_incoming_thread
);
556 qemu_mutex_lock_iothread();
557 /* We hold the global iothread lock, so it is safe here */
558 colo_release_ram_cache();
562 error_report("load of migration failed: %s", strerror(-ret
));
565 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
566 qemu_bh_schedule(mis
->bh
);
567 mis
->migration_incoming_co
= NULL
;
571 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
572 MIGRATION_STATUS_FAILED
);
573 qemu_fclose(mis
->from_src_file
);
575 multifd_load_cleanup();
576 compress_threads_load_cleanup();
582 * migration_incoming_setup: Setup incoming migration
583 * @f: file for main migration channel
584 * @errp: where to put errors
586 * Returns: %true on success, %false on error.
588 static bool migration_incoming_setup(QEMUFile
*f
, Error
**errp
)
590 MigrationIncomingState
*mis
= migration_incoming_get_current();
592 if (!mis
->from_src_file
) {
593 mis
->from_src_file
= f
;
595 qemu_file_set_blocking(f
, false);
599 void migration_incoming_process(void)
601 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
602 qemu_coroutine_enter(co
);
605 /* Returns true if recovered from a paused migration, otherwise false */
606 static bool postcopy_try_recover(void)
608 MigrationIncomingState
*mis
= migration_incoming_get_current();
610 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
611 /* Resumed from a paused postcopy migration */
613 /* This should be set already in migration_incoming_setup() */
614 assert(mis
->from_src_file
);
615 /* Postcopy has standalone thread to do vm load */
616 qemu_file_set_blocking(mis
->from_src_file
, true);
618 /* Re-configure the return path */
619 mis
->to_src_file
= qemu_file_get_return_path(mis
->from_src_file
);
621 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
622 MIGRATION_STATUS_POSTCOPY_RECOVER
);
625 * Here, we only wake up the main loading thread (while the
626 * rest threads will still be waiting), so that we can receive
627 * commands from source now, and answer it if needed. The
628 * rest threads will be woken up afterwards until we are sure
629 * that source is ready to reply to page requests.
631 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
638 void migration_fd_process_incoming(QEMUFile
*f
, Error
**errp
)
640 if (!migration_incoming_setup(f
, errp
)) {
643 if (postcopy_try_recover()) {
646 migration_incoming_process();
650 * Returns true when we want to start a new incoming migration process,
653 static bool migration_should_start_incoming(bool main_channel
)
655 /* Multifd doesn't start unless all channels are established */
656 if (migrate_multifd()) {
657 return migration_has_all_channels();
660 /* Preempt channel only starts when the main channel is created */
661 if (migrate_postcopy_preempt()) {
666 * For all the rest types of migration, we should only reach here when
667 * it's the main channel that's being created, and we should always
668 * proceed with this channel.
670 assert(main_channel
);
674 void migration_ioc_process_incoming(QIOChannel
*ioc
, Error
**errp
)
676 MigrationIncomingState
*mis
= migration_incoming_get_current();
677 Error
*local_err
= NULL
;
679 bool default_channel
= true;
680 uint32_t channel_magic
= 0;
683 if (migrate_multifd() && !migrate_postcopy_ram() &&
684 qio_channel_has_feature(ioc
, QIO_CHANNEL_FEATURE_READ_MSG_PEEK
)) {
686 * With multiple channels, it is possible that we receive channels
687 * out of order on destination side, causing incorrect mapping of
688 * source channels on destination side. Check channel MAGIC to
689 * decide type of channel. Please note this is best effort, postcopy
690 * preempt channel does not send any magic number so avoid it for
691 * postcopy live migration. Also tls live migration already does
692 * tls handshake while initializing main channel so with tls this
693 * issue is not possible.
695 ret
= migration_channel_read_peek(ioc
, (void *)&channel_magic
,
696 sizeof(channel_magic
), &local_err
);
699 error_propagate(errp
, local_err
);
703 default_channel
= (channel_magic
== cpu_to_be32(QEMU_VM_FILE_MAGIC
));
705 default_channel
= !mis
->from_src_file
;
708 if (multifd_load_setup(errp
) != 0) {
709 error_setg(errp
, "Failed to setup multifd channels");
713 if (default_channel
) {
714 f
= qemu_file_new_input(ioc
);
716 if (!migration_incoming_setup(f
, errp
)) {
720 /* Multiple connections */
721 assert(migration_needs_multiple_sockets());
722 if (migrate_multifd()) {
723 multifd_recv_new_channel(ioc
, &local_err
);
725 assert(migrate_postcopy_preempt());
726 f
= qemu_file_new_input(ioc
);
727 postcopy_preempt_new_channel(mis
, f
);
730 error_propagate(errp
, local_err
);
735 if (migration_should_start_incoming(default_channel
)) {
736 /* If it's a recovery, we're done */
737 if (postcopy_try_recover()) {
740 migration_incoming_process();
745 * @migration_has_all_channels: We have received all channels that we need
747 * Returns true when we have got connections to all the channels that
748 * we need for migration.
750 bool migration_has_all_channels(void)
752 MigrationIncomingState
*mis
= migration_incoming_get_current();
754 if (!mis
->from_src_file
) {
758 if (migrate_multifd()) {
759 return multifd_recv_all_channels_created();
762 if (migrate_postcopy_preempt()) {
763 return mis
->postcopy_qemufile_dst
!= NULL
;
770 * Send a 'SHUT' message on the return channel with the given value
771 * to indicate that we've finished with the RP. Non-0 value indicates
774 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
779 buf
= cpu_to_be32(value
);
780 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
784 * Send a 'PONG' message on the return channel with the given value
785 * (normally in response to a 'PING')
787 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
792 buf
= cpu_to_be32(value
);
793 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
796 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
804 * First, we send the header part. It contains only the len of
805 * idstr, and the idstr itself.
807 len
= strlen(block_name
);
809 memcpy(buf
+ 1, block_name
, len
);
811 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
812 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
817 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
820 * Next, we dump the received bitmap to the stream.
822 * TODO: currently we are safe since we are the only one that is
823 * using the to_src_file handle (fault thread is still paused),
824 * and it's ok even not taking the mutex. However the best way is
825 * to take the lock before sending the message header, and release
826 * the lock after sending the bitmap.
828 qemu_mutex_lock(&mis
->rp_mutex
);
829 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
830 qemu_mutex_unlock(&mis
->rp_mutex
);
832 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
835 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
839 buf
= cpu_to_be32(value
);
840 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
844 * Return true if we're already in the middle of a migration
845 * (i.e. any of the active or setup states)
847 bool migration_is_setup_or_active(int state
)
850 case MIGRATION_STATUS_ACTIVE
:
851 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
852 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
853 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
854 case MIGRATION_STATUS_SETUP
:
855 case MIGRATION_STATUS_PRE_SWITCHOVER
:
856 case MIGRATION_STATUS_DEVICE
:
857 case MIGRATION_STATUS_WAIT_UNPLUG
:
858 case MIGRATION_STATUS_COLO
:
867 bool migration_is_running(int state
)
870 case MIGRATION_STATUS_ACTIVE
:
871 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
872 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
873 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
874 case MIGRATION_STATUS_SETUP
:
875 case MIGRATION_STATUS_PRE_SWITCHOVER
:
876 case MIGRATION_STATUS_DEVICE
:
877 case MIGRATION_STATUS_WAIT_UNPLUG
:
878 case MIGRATION_STATUS_CANCELLING
:
887 static bool migrate_show_downtime(MigrationState
*s
)
889 return (s
->state
== MIGRATION_STATUS_COMPLETED
) || migration_in_postcopy();
892 static void populate_time_info(MigrationInfo
*info
, MigrationState
*s
)
894 info
->has_status
= true;
895 info
->has_setup_time
= true;
896 info
->setup_time
= s
->setup_time
;
898 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
899 info
->has_total_time
= true;
900 info
->total_time
= s
->total_time
;
902 info
->has_total_time
= true;
903 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) -
907 if (migrate_show_downtime(s
)) {
908 info
->has_downtime
= true;
909 info
->downtime
= s
->downtime
;
911 info
->has_expected_downtime
= true;
912 info
->expected_downtime
= s
->expected_downtime
;
916 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
918 size_t page_size
= qemu_target_page_size();
920 info
->ram
= g_malloc0(sizeof(*info
->ram
));
921 info
->ram
->transferred
= stat64_get(&mig_stats
.transferred
);
922 info
->ram
->total
= ram_bytes_total();
923 info
->ram
->duplicate
= stat64_get(&mig_stats
.zero_pages
);
924 /* legacy value. It is not used anymore */
925 info
->ram
->skipped
= 0;
926 info
->ram
->normal
= stat64_get(&mig_stats
.normal_pages
);
927 info
->ram
->normal_bytes
= info
->ram
->normal
* page_size
;
928 info
->ram
->mbps
= s
->mbps
;
929 info
->ram
->dirty_sync_count
=
930 stat64_get(&mig_stats
.dirty_sync_count
);
931 info
->ram
->dirty_sync_missed_zero_copy
=
932 stat64_get(&mig_stats
.dirty_sync_missed_zero_copy
);
933 info
->ram
->postcopy_requests
=
934 stat64_get(&mig_stats
.postcopy_requests
);
935 info
->ram
->page_size
= page_size
;
936 info
->ram
->multifd_bytes
= stat64_get(&mig_stats
.multifd_bytes
);
937 info
->ram
->pages_per_second
= s
->pages_per_second
;
938 info
->ram
->precopy_bytes
= stat64_get(&mig_stats
.precopy_bytes
);
939 info
->ram
->downtime_bytes
= stat64_get(&mig_stats
.downtime_bytes
);
940 info
->ram
->postcopy_bytes
= stat64_get(&mig_stats
.postcopy_bytes
);
942 if (migrate_xbzrle()) {
943 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
944 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
945 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
946 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
947 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
948 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
949 info
->xbzrle_cache
->encoding_rate
= xbzrle_counters
.encoding_rate
;
950 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
953 if (migrate_compress()) {
954 info
->compression
= g_malloc0(sizeof(*info
->compression
));
955 info
->compression
->pages
= compression_counters
.pages
;
956 info
->compression
->busy
= compression_counters
.busy
;
957 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
958 info
->compression
->compressed_size
=
959 compression_counters
.compressed_size
;
960 info
->compression
->compression_rate
=
961 compression_counters
.compression_rate
;
964 if (cpu_throttle_active()) {
965 info
->has_cpu_throttle_percentage
= true;
966 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
969 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
970 info
->ram
->remaining
= ram_bytes_remaining();
971 info
->ram
->dirty_pages_rate
=
972 stat64_get(&mig_stats
.dirty_pages_rate
);
976 static void populate_disk_info(MigrationInfo
*info
)
978 if (blk_mig_active()) {
979 info
->disk
= g_malloc0(sizeof(*info
->disk
));
980 info
->disk
->transferred
= blk_mig_bytes_transferred();
981 info
->disk
->remaining
= blk_mig_bytes_remaining();
982 info
->disk
->total
= blk_mig_bytes_total();
986 static void fill_source_migration_info(MigrationInfo
*info
)
988 MigrationState
*s
= migrate_get_current();
989 int state
= qatomic_read(&s
->state
);
990 GSList
*cur_blocker
= migration_blockers
;
992 info
->blocked_reasons
= NULL
;
995 * There are two types of reasons a migration might be blocked;
996 * a) devices marked in VMState as non-migratable, and
997 * b) Explicit migration blockers
998 * We need to add both of them here.
1000 qemu_savevm_non_migratable_list(&info
->blocked_reasons
);
1002 while (cur_blocker
) {
1003 QAPI_LIST_PREPEND(info
->blocked_reasons
,
1004 g_strdup(error_get_pretty(cur_blocker
->data
)));
1005 cur_blocker
= g_slist_next(cur_blocker
);
1007 info
->has_blocked_reasons
= info
->blocked_reasons
!= NULL
;
1010 case MIGRATION_STATUS_NONE
:
1011 /* no migration has happened ever */
1012 /* do not overwrite destination migration status */
1014 case MIGRATION_STATUS_SETUP
:
1015 info
->has_status
= true;
1016 info
->has_total_time
= false;
1018 case MIGRATION_STATUS_ACTIVE
:
1019 case MIGRATION_STATUS_CANCELLING
:
1020 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1021 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1022 case MIGRATION_STATUS_DEVICE
:
1023 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1024 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1025 /* TODO add some postcopy stats */
1026 populate_time_info(info
, s
);
1027 populate_ram_info(info
, s
);
1028 populate_disk_info(info
);
1029 populate_vfio_info(info
);
1031 case MIGRATION_STATUS_COLO
:
1032 info
->has_status
= true;
1033 /* TODO: display COLO specific information (checkpoint info etc.) */
1035 case MIGRATION_STATUS_COMPLETED
:
1036 populate_time_info(info
, s
);
1037 populate_ram_info(info
, s
);
1038 populate_vfio_info(info
);
1040 case MIGRATION_STATUS_FAILED
:
1041 info
->has_status
= true;
1043 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
1046 case MIGRATION_STATUS_CANCELLED
:
1047 info
->has_status
= true;
1049 case MIGRATION_STATUS_WAIT_UNPLUG
:
1050 info
->has_status
= true;
1053 info
->status
= state
;
1056 static void fill_destination_migration_info(MigrationInfo
*info
)
1058 MigrationIncomingState
*mis
= migration_incoming_get_current();
1060 if (mis
->socket_address_list
) {
1061 info
->has_socket_address
= true;
1062 info
->socket_address
=
1063 QAPI_CLONE(SocketAddressList
, mis
->socket_address_list
);
1066 switch (mis
->state
) {
1067 case MIGRATION_STATUS_NONE
:
1069 case MIGRATION_STATUS_SETUP
:
1070 case MIGRATION_STATUS_CANCELLING
:
1071 case MIGRATION_STATUS_CANCELLED
:
1072 case MIGRATION_STATUS_ACTIVE
:
1073 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1074 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1075 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1076 case MIGRATION_STATUS_FAILED
:
1077 case MIGRATION_STATUS_COLO
:
1078 info
->has_status
= true;
1080 case MIGRATION_STATUS_COMPLETED
:
1081 info
->has_status
= true;
1082 fill_destination_postcopy_migration_info(info
);
1085 info
->status
= mis
->state
;
1088 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1090 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1092 fill_destination_migration_info(info
);
1093 fill_source_migration_info(info
);
1098 void qmp_migrate_start_postcopy(Error
**errp
)
1100 MigrationState
*s
= migrate_get_current();
1102 if (!migrate_postcopy()) {
1103 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1104 " the start of migration");
1108 if (s
->state
== MIGRATION_STATUS_NONE
) {
1109 error_setg(errp
, "Postcopy must be started after migration has been"
1114 * we don't error if migration has finished since that would be racy
1115 * with issuing this command.
1117 qatomic_set(&s
->start_postcopy
, true);
1120 /* shared migration helpers */
1122 void migrate_set_state(int *state
, int old_state
, int new_state
)
1124 assert(new_state
< MIGRATION_STATUS__MAX
);
1125 if (qatomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1126 trace_migrate_set_state(MigrationStatus_str(new_state
));
1127 migrate_generate_event(new_state
);
1131 static void migrate_fd_cleanup(MigrationState
*s
)
1133 qemu_bh_delete(s
->cleanup_bh
);
1134 s
->cleanup_bh
= NULL
;
1136 g_free(s
->hostname
);
1138 json_writer_free(s
->vmdesc
);
1141 qemu_savevm_state_cleanup();
1143 if (s
->to_dst_file
) {
1146 trace_migrate_fd_cleanup();
1147 qemu_mutex_unlock_iothread();
1148 if (s
->migration_thread_running
) {
1149 qemu_thread_join(&s
->thread
);
1150 s
->migration_thread_running
= false;
1152 qemu_mutex_lock_iothread();
1154 multifd_save_cleanup();
1155 qemu_mutex_lock(&s
->qemu_file_lock
);
1156 tmp
= s
->to_dst_file
;
1157 s
->to_dst_file
= NULL
;
1158 qemu_mutex_unlock(&s
->qemu_file_lock
);
1160 * Close the file handle without the lock to make sure the
1161 * critical section won't block for long.
1163 migration_ioc_unregister_yank_from_file(tmp
);
1167 if (s
->postcopy_qemufile_src
) {
1168 migration_ioc_unregister_yank_from_file(s
->postcopy_qemufile_src
);
1169 qemu_fclose(s
->postcopy_qemufile_src
);
1170 s
->postcopy_qemufile_src
= NULL
;
1173 assert(!migration_is_active(s
));
1175 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1176 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1177 MIGRATION_STATUS_CANCELLED
);
1181 /* It is used on info migrate. We can't free it */
1182 error_report_err(error_copy(s
->error
));
1184 notifier_list_notify(&migration_state_notifiers
, s
);
1185 block_cleanup_parameters();
1186 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1189 static void migrate_fd_cleanup_schedule(MigrationState
*s
)
1192 * Ref the state for bh, because it may be called when
1193 * there're already no other refs
1195 object_ref(OBJECT(s
));
1196 qemu_bh_schedule(s
->cleanup_bh
);
1199 static void migrate_fd_cleanup_bh(void *opaque
)
1201 MigrationState
*s
= opaque
;
1202 migrate_fd_cleanup(s
);
1203 object_unref(OBJECT(s
));
1206 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1208 QEMU_LOCK_GUARD(&s
->error_mutex
);
1210 s
->error
= error_copy(error
);
1214 static void migrate_error_free(MigrationState
*s
)
1216 QEMU_LOCK_GUARD(&s
->error_mutex
);
1218 error_free(s
->error
);
1223 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1225 trace_migrate_fd_error(error_get_pretty(error
));
1226 assert(s
->to_dst_file
== NULL
);
1227 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1228 MIGRATION_STATUS_FAILED
);
1229 migrate_set_error(s
, error
);
1232 static void migrate_fd_cancel(MigrationState
*s
)
1235 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1236 trace_migrate_fd_cancel();
1238 WITH_QEMU_LOCK_GUARD(&s
->qemu_file_lock
) {
1239 if (s
->rp_state
.from_dst_file
) {
1240 /* shutdown the rp socket, so causing the rp thread to shutdown */
1241 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1246 old_state
= s
->state
;
1247 if (!migration_is_running(old_state
)) {
1250 /* If the migration is paused, kick it out of the pause */
1251 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1252 qemu_sem_post(&s
->pause_sem
);
1254 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1255 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1258 * If we're unlucky the migration code might be stuck somewhere in a
1259 * send/write while the network has failed and is waiting to timeout;
1260 * if we've got shutdown(2) available then we can force it to quit.
1261 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1262 * called in a bh, so there is no race against this cancel.
1264 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1265 qemu_file_shutdown(f
);
1267 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1268 Error
*local_err
= NULL
;
1270 bdrv_activate_all(&local_err
);
1272 error_report_err(local_err
);
1274 s
->block_inactive
= false;
1279 void add_migration_state_change_notifier(Notifier
*notify
)
1281 notifier_list_add(&migration_state_notifiers
, notify
);
1284 void remove_migration_state_change_notifier(Notifier
*notify
)
1286 notifier_remove(notify
);
1289 bool migration_in_setup(MigrationState
*s
)
1291 return s
->state
== MIGRATION_STATUS_SETUP
;
1294 bool migration_has_finished(MigrationState
*s
)
1296 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1299 bool migration_has_failed(MigrationState
*s
)
1301 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1302 s
->state
== MIGRATION_STATUS_FAILED
);
1305 bool migration_in_postcopy(void)
1307 MigrationState
*s
= migrate_get_current();
1310 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1311 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1312 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1319 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1321 return migration_in_postcopy() && s
->postcopy_after_devices
;
1324 bool migration_in_incoming_postcopy(void)
1326 PostcopyState ps
= postcopy_state_get();
1328 return ps
>= POSTCOPY_INCOMING_DISCARD
&& ps
< POSTCOPY_INCOMING_END
;
1331 bool migration_incoming_postcopy_advised(void)
1333 PostcopyState ps
= postcopy_state_get();
1335 return ps
>= POSTCOPY_INCOMING_ADVISE
&& ps
< POSTCOPY_INCOMING_END
;
1338 bool migration_in_bg_snapshot(void)
1340 MigrationState
*s
= migrate_get_current();
1342 return migrate_background_snapshot() &&
1343 migration_is_setup_or_active(s
->state
);
1346 bool migration_is_idle(void)
1348 MigrationState
*s
= current_migration
;
1355 case MIGRATION_STATUS_NONE
:
1356 case MIGRATION_STATUS_CANCELLED
:
1357 case MIGRATION_STATUS_COMPLETED
:
1358 case MIGRATION_STATUS_FAILED
:
1360 case MIGRATION_STATUS_SETUP
:
1361 case MIGRATION_STATUS_CANCELLING
:
1362 case MIGRATION_STATUS_ACTIVE
:
1363 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1364 case MIGRATION_STATUS_COLO
:
1365 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1366 case MIGRATION_STATUS_DEVICE
:
1367 case MIGRATION_STATUS_WAIT_UNPLUG
:
1369 case MIGRATION_STATUS__MAX
:
1370 g_assert_not_reached();
1376 bool migration_is_active(MigrationState
*s
)
1378 return (s
->state
== MIGRATION_STATUS_ACTIVE
||
1379 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1382 void migrate_init(MigrationState
*s
)
1385 * Reinitialise all migration state, except
1386 * parameters/capabilities that the user set, and
1391 s
->to_dst_file
= NULL
;
1392 s
->state
= MIGRATION_STATUS_NONE
;
1393 s
->rp_state
.from_dst_file
= NULL
;
1394 s
->rp_state
.error
= false;
1396 s
->pages_per_second
= 0.0;
1398 s
->expected_downtime
= 0;
1400 s
->start_postcopy
= false;
1401 s
->postcopy_after_devices
= false;
1402 s
->migration_thread_running
= false;
1403 error_free(s
->error
);
1407 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1409 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1411 s
->vm_was_running
= false;
1412 s
->iteration_initial_bytes
= 0;
1413 s
->threshold_size
= 0;
1416 int migrate_add_blocker_internal(Error
*reason
, Error
**errp
)
1418 /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */
1419 if (runstate_check(RUN_STATE_SAVE_VM
) || !migration_is_idle()) {
1420 error_propagate_prepend(errp
, error_copy(reason
),
1421 "disallowing migration blocker "
1422 "(migration/snapshot in progress) for: ");
1426 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1430 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1432 if (only_migratable
) {
1433 error_propagate_prepend(errp
, error_copy(reason
),
1434 "disallowing migration blocker "
1435 "(--only-migratable) for: ");
1439 return migrate_add_blocker_internal(reason
, errp
);
1442 void migrate_del_blocker(Error
*reason
)
1444 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1447 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1449 Error
*local_err
= NULL
;
1450 static bool once
= true;
1453 error_setg(errp
, "The incoming migration has already been started");
1456 if (!runstate_check(RUN_STATE_INMIGRATE
)) {
1457 error_setg(errp
, "'-incoming' was not specified on the command line");
1461 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
1465 qemu_start_incoming_migration(uri
, &local_err
);
1468 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1469 error_propagate(errp
, local_err
);
1476 void qmp_migrate_recover(const char *uri
, Error
**errp
)
1478 MigrationIncomingState
*mis
= migration_incoming_get_current();
1481 * Don't even bother to use ERRP_GUARD() as it _must_ always be set by
1482 * callers (no one should ignore a recover failure); if there is, it's a
1483 * programming error.
1487 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1488 error_setg(errp
, "Migrate recover can only be run "
1489 "when postcopy is paused.");
1493 /* If there's an existing transport, release it */
1494 migration_incoming_transport_cleanup(mis
);
1497 * Note that this call will never start a real migration; it will
1498 * only re-setup the migration stream and poke existing migration
1499 * to continue using that newly established channel.
1501 qemu_start_incoming_migration(uri
, errp
);
1504 void qmp_migrate_pause(Error
**errp
)
1506 MigrationState
*ms
= migrate_get_current();
1507 MigrationIncomingState
*mis
= migration_incoming_get_current();
1510 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1511 /* Source side, during postcopy */
1512 qemu_mutex_lock(&ms
->qemu_file_lock
);
1513 ret
= qemu_file_shutdown(ms
->to_dst_file
);
1514 qemu_mutex_unlock(&ms
->qemu_file_lock
);
1516 error_setg(errp
, "Failed to pause source migration");
1521 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1522 ret
= qemu_file_shutdown(mis
->from_src_file
);
1524 error_setg(errp
, "Failed to pause destination migration");
1529 error_setg(errp
, "migrate-pause is currently only supported "
1530 "during postcopy-active state");
1533 bool migration_is_blocked(Error
**errp
)
1535 if (qemu_savevm_state_blocked(errp
)) {
1539 if (migration_blockers
) {
1540 error_propagate(errp
, error_copy(migration_blockers
->data
));
1547 /* Returns true if continue to migrate, or false if error detected */
1548 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
1549 bool resume
, Error
**errp
)
1551 Error
*local_err
= NULL
;
1554 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1555 error_setg(errp
, "Cannot resume if there is no "
1556 "paused migration");
1561 * Postcopy recovery won't work well with release-ram
1562 * capability since release-ram will drop the page buffer as
1563 * long as the page is put into the send buffer. So if there
1564 * is a network failure happened, any page buffers that have
1565 * not yet reached the destination VM but have already been
1566 * sent from the source VM will be lost forever. Let's refuse
1567 * the client from resuming such a postcopy migration.
1568 * Luckily release-ram was designed to only be used when src
1569 * and destination VMs are on the same host, so it should be
1572 if (migrate_release_ram()) {
1573 error_setg(errp
, "Postcopy recovery cannot work "
1574 "when release-ram capability is set");
1578 /* This is a resume, skip init status */
1582 if (migration_is_running(s
->state
)) {
1583 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1587 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1588 error_setg(errp
, "Guest is waiting for an incoming migration");
1592 if (runstate_check(RUN_STATE_POSTMIGRATE
)) {
1593 error_setg(errp
, "Can't migrate the vm that was paused due to "
1594 "previous migration");
1598 if (migration_is_blocked(errp
)) {
1602 if (blk
|| blk_inc
) {
1603 if (migrate_colo()) {
1604 error_setg(errp
, "No disk migration is required in COLO mode");
1607 if (migrate_block() || migrate_block_incremental()) {
1608 error_setg(errp
, "Command options are incompatible with "
1609 "current migration capabilities");
1612 if (!migrate_cap_set(MIGRATION_CAPABILITY_BLOCK
, true, &local_err
)) {
1613 error_propagate(errp
, local_err
);
1616 s
->must_remove_block_options
= true;
1620 migrate_set_block_incremental(true);
1625 * set mig_stats compression_counters memory to zero for a
1628 memset(&mig_stats
, 0, sizeof(mig_stats
));
1629 memset(&compression_counters
, 0, sizeof(compression_counters
));
1634 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1635 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1636 bool has_resume
, bool resume
, Error
**errp
)
1638 Error
*local_err
= NULL
;
1639 MigrationState
*s
= migrate_get_current();
1640 const char *p
= NULL
;
1642 /* URI is not suitable for migration? */
1643 if (!migration_channels_and_uri_compatible(uri
, errp
)) {
1647 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
1648 has_resume
&& resume
, errp
)) {
1649 /* Error detected, put into errp */
1653 if (!(has_resume
&& resume
)) {
1654 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
1659 if (strstart(uri
, "tcp:", &p
) ||
1660 strstart(uri
, "unix:", NULL
) ||
1661 strstart(uri
, "vsock:", NULL
)) {
1662 socket_start_outgoing_migration(s
, p
? p
: uri
, &local_err
);
1664 } else if (strstart(uri
, "rdma:", &p
)) {
1665 rdma_start_outgoing_migration(s
, p
, &local_err
);
1667 } else if (strstart(uri
, "exec:", &p
)) {
1668 exec_start_outgoing_migration(s
, p
, &local_err
);
1669 } else if (strstart(uri
, "fd:", &p
)) {
1670 fd_start_outgoing_migration(s
, p
, &local_err
);
1672 if (!(has_resume
&& resume
)) {
1673 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1675 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1676 "a valid migration protocol");
1677 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1678 MIGRATION_STATUS_FAILED
);
1679 block_cleanup_parameters();
1684 if (!(has_resume
&& resume
)) {
1685 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1687 migrate_fd_error(s
, local_err
);
1688 error_propagate(errp
, local_err
);
1693 void qmp_migrate_cancel(Error
**errp
)
1695 migration_cancel(NULL
);
1698 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
1700 MigrationState
*s
= migrate_get_current();
1701 if (s
->state
!= state
) {
1702 error_setg(errp
, "Migration not in expected state: %s",
1703 MigrationStatus_str(s
->state
));
1706 qemu_sem_post(&s
->pause_sem
);
1709 /* migration thread support */
1711 * Something bad happened to the RP stream, mark an error
1712 * The caller shall print or trace something to indicate why
1714 static void mark_source_rp_bad(MigrationState
*s
)
1716 s
->rp_state
.error
= true;
1719 static struct rp_cmd_args
{
1720 ssize_t len
; /* -1 = variable */
1723 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1724 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1725 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1726 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1727 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1728 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
1729 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
1730 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1734 * Process a request for pages received on the return path,
1735 * We're allowed to send more than requested (e.g. to round to our page size)
1736 * and we don't need to send pages that have already been sent.
1738 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1739 ram_addr_t start
, size_t len
)
1741 long our_host_ps
= qemu_real_host_page_size();
1743 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1746 * Since we currently insist on matching page sizes, just sanity check
1747 * we're being asked for whole host pages.
1749 if (!QEMU_IS_ALIGNED(start
, our_host_ps
) ||
1750 !QEMU_IS_ALIGNED(len
, our_host_ps
)) {
1751 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1752 " len: %zd", __func__
, start
, len
);
1753 mark_source_rp_bad(ms
);
1757 if (ram_save_queue_pages(rbname
, start
, len
)) {
1758 mark_source_rp_bad(ms
);
1762 /* Return true to retry, false to quit */
1763 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
1765 trace_postcopy_pause_return_path();
1767 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
1769 trace_postcopy_pause_return_path_continued();
1774 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
1776 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
1779 error_report("%s: invalid block name '%s'", __func__
, block_name
);
1783 /* Fetch the received bitmap and refresh the dirty bitmap */
1784 return ram_dirty_bitmap_reload(s
, block
);
1787 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
1789 trace_source_return_path_thread_resume_ack(value
);
1791 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
1792 error_report("%s: illegal resume_ack value %"PRIu32
,
1797 /* Now both sides are active. */
1798 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
1799 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1801 /* Notify send thread that time to continue send pages */
1802 qemu_sem_post(&s
->rp_state
.rp_sem
);
1808 * Release ms->rp_state.from_dst_file (and postcopy_qemufile_src if
1809 * existed) in a safe way.
1811 static void migration_release_dst_files(MigrationState
*ms
)
1815 WITH_QEMU_LOCK_GUARD(&ms
->qemu_file_lock
) {
1817 * Reset the from_dst_file pointer first before releasing it, as we
1818 * can't block within lock section
1820 file
= ms
->rp_state
.from_dst_file
;
1821 ms
->rp_state
.from_dst_file
= NULL
;
1825 * Do the same to postcopy fast path socket too if there is. No
1826 * locking needed because this qemufile should only be managed by
1827 * return path thread.
1829 if (ms
->postcopy_qemufile_src
) {
1830 migration_ioc_unregister_yank_from_file(ms
->postcopy_qemufile_src
);
1831 qemu_file_shutdown(ms
->postcopy_qemufile_src
);
1832 qemu_fclose(ms
->postcopy_qemufile_src
);
1833 ms
->postcopy_qemufile_src
= NULL
;
1840 * Handles messages sent on the return path towards the source VM
1843 static void *source_return_path_thread(void *opaque
)
1845 MigrationState
*ms
= opaque
;
1846 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1847 uint16_t header_len
, header_type
;
1849 uint32_t tmp32
, sibling_error
;
1850 ram_addr_t start
= 0; /* =0 to silence warning */
1851 size_t len
= 0, expected_len
;
1854 trace_source_return_path_thread_entry();
1855 rcu_register_thread();
1858 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1859 migration_is_setup_or_active(ms
->state
)) {
1860 trace_source_return_path_thread_loop_top();
1861 header_type
= qemu_get_be16(rp
);
1862 header_len
= qemu_get_be16(rp
);
1864 if (qemu_file_get_error(rp
)) {
1865 mark_source_rp_bad(ms
);
1869 if (header_type
>= MIG_RP_MSG_MAX
||
1870 header_type
== MIG_RP_MSG_INVALID
) {
1871 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1872 header_type
, header_len
);
1873 mark_source_rp_bad(ms
);
1877 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1878 header_len
!= rp_cmd_args
[header_type
].len
) ||
1879 header_len
> sizeof(buf
)) {
1880 error_report("RP: Received '%s' message (0x%04x) with"
1881 "incorrect length %d expecting %zu",
1882 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1883 (size_t)rp_cmd_args
[header_type
].len
);
1884 mark_source_rp_bad(ms
);
1888 /* We know we've got a valid header by this point */
1889 res
= qemu_get_buffer(rp
, buf
, header_len
);
1890 if (res
!= header_len
) {
1891 error_report("RP: Failed reading data for message 0x%04x"
1892 " read %d expected %d",
1893 header_type
, res
, header_len
);
1894 mark_source_rp_bad(ms
);
1898 /* OK, we have the message and the data */
1899 switch (header_type
) {
1900 case MIG_RP_MSG_SHUT
:
1901 sibling_error
= ldl_be_p(buf
);
1902 trace_source_return_path_thread_shut(sibling_error
);
1903 if (sibling_error
) {
1904 error_report("RP: Sibling indicated error %d", sibling_error
);
1905 mark_source_rp_bad(ms
);
1908 * We'll let the main thread deal with closing the RP
1909 * we could do a shutdown(2) on it, but we're the only user
1910 * anyway, so there's nothing gained.
1914 case MIG_RP_MSG_PONG
:
1915 tmp32
= ldl_be_p(buf
);
1916 trace_source_return_path_thread_pong(tmp32
);
1917 qemu_sem_post(&ms
->rp_state
.rp_pong_acks
);
1920 case MIG_RP_MSG_REQ_PAGES
:
1921 start
= ldq_be_p(buf
);
1922 len
= ldl_be_p(buf
+ 8);
1923 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1926 case MIG_RP_MSG_REQ_PAGES_ID
:
1927 expected_len
= 12 + 1; /* header + termination */
1929 if (header_len
>= expected_len
) {
1930 start
= ldq_be_p(buf
);
1931 len
= ldl_be_p(buf
+ 8);
1932 /* Now we expect an idstr */
1933 tmp32
= buf
[12]; /* Length of the following idstr */
1934 buf
[13 + tmp32
] = '\0';
1935 expected_len
+= tmp32
;
1937 if (header_len
!= expected_len
) {
1938 error_report("RP: Req_Page_id with length %d expecting %zd",
1939 header_len
, expected_len
);
1940 mark_source_rp_bad(ms
);
1943 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
1946 case MIG_RP_MSG_RECV_BITMAP
:
1947 if (header_len
< 1) {
1948 error_report("%s: missing block name", __func__
);
1949 mark_source_rp_bad(ms
);
1952 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
1953 buf
[buf
[0] + 1] = '\0';
1954 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
1955 mark_source_rp_bad(ms
);
1960 case MIG_RP_MSG_RESUME_ACK
:
1961 tmp32
= ldl_be_p(buf
);
1962 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
1963 mark_source_rp_bad(ms
);
1974 res
= qemu_file_get_error(rp
);
1976 if (res
&& migration_in_postcopy()) {
1978 * Maybe there is something we can do: it looks like a
1979 * network down issue, and we pause for a recovery.
1981 migration_release_dst_files(ms
);
1983 if (postcopy_pause_return_path_thread(ms
)) {
1985 * Reload rp, reset the rest. Referencing it is safe since
1986 * it's reset only by us above, or when migration completes
1988 rp
= ms
->rp_state
.from_dst_file
;
1989 ms
->rp_state
.error
= false;
1994 trace_source_return_path_thread_bad_end();
1995 mark_source_rp_bad(ms
);
1998 trace_source_return_path_thread_end();
1999 migration_release_dst_files(ms
);
2000 rcu_unregister_thread();
2004 static int open_return_path_on_source(MigrationState
*ms
,
2007 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
2008 if (!ms
->rp_state
.from_dst_file
) {
2012 trace_open_return_path_on_source();
2014 if (!create_thread
) {
2019 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
2020 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
2021 ms
->rp_state
.rp_thread_created
= true;
2023 trace_open_return_path_on_source_continue();
2028 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2029 static int await_return_path_close_on_source(MigrationState
*ms
)
2032 * If this is a normal exit then the destination will send a SHUT and the
2033 * rp_thread will exit, however if there's an error we need to cause
2036 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
2038 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2039 * waiting for the destination.
2041 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
2042 mark_source_rp_bad(ms
);
2044 trace_await_return_path_close_on_source_joining();
2045 qemu_thread_join(&ms
->rp_state
.rp_thread
);
2046 ms
->rp_state
.rp_thread_created
= false;
2047 trace_await_return_path_close_on_source_close();
2048 return ms
->rp_state
.error
;
2052 migration_wait_main_channel(MigrationState
*ms
)
2054 /* Wait until one PONG message received */
2055 qemu_sem_wait(&ms
->rp_state
.rp_pong_acks
);
2059 * Switch from normal iteration to postcopy
2060 * Returns non-0 on error
2062 static int postcopy_start(MigrationState
*ms
)
2065 QIOChannelBuffer
*bioc
;
2067 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2068 uint64_t bandwidth
= migrate_max_postcopy_bandwidth();
2069 bool restart_block
= false;
2070 int cur_state
= MIGRATION_STATUS_ACTIVE
;
2072 if (migrate_postcopy_preempt()) {
2073 migration_wait_main_channel(ms
);
2074 if (postcopy_preempt_establish_channel(ms
)) {
2075 migrate_set_state(&ms
->state
, ms
->state
, MIGRATION_STATUS_FAILED
);
2080 if (!migrate_pause_before_switchover()) {
2081 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
2082 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2085 trace_postcopy_start();
2086 qemu_mutex_lock_iothread();
2087 trace_postcopy_start_set_run();
2089 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2090 global_state_store();
2091 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2096 ret
= migration_maybe_pause(ms
, &cur_state
,
2097 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2102 ret
= bdrv_inactivate_all();
2106 restart_block
= true;
2109 * Cause any non-postcopiable, but iterative devices to
2110 * send out their final data.
2112 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2115 * in Finish migrate and with the io-lock held everything should
2116 * be quiet, but we've potentially still got dirty pages and we
2117 * need to tell the destination to throw any pages it's already received
2120 if (migrate_postcopy_ram()) {
2121 ram_postcopy_send_discard_bitmap(ms
);
2125 * send rest of state - note things that are doing postcopy
2126 * will notice we're in POSTCOPY_ACTIVE and not actually
2127 * wrap their state up here
2129 /* 0 max-postcopy-bandwidth means unlimited */
2131 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
2133 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
2135 if (migrate_postcopy_ram()) {
2136 /* Ping just for debugging, helps line traces up */
2137 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
2141 * While loading the device state we may trigger page transfer
2142 * requests and the fd must be free to process those, and thus
2143 * the destination must read the whole device state off the fd before
2144 * it starts processing it. Unfortunately the ad-hoc migration format
2145 * doesn't allow the destination to know the size to read without fully
2146 * parsing it through each devices load-state code (especially the open
2147 * coded devices that use get/put).
2148 * So we wrap the device state up in a package with a length at the start;
2149 * to do this we use a qemu_buf to hold the whole of the device state.
2151 bioc
= qio_channel_buffer_new(4096);
2152 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
2153 fb
= qemu_file_new_output(QIO_CHANNEL(bioc
));
2154 object_unref(OBJECT(bioc
));
2157 * Make sure the receiver can get incoming pages before we send the rest
2160 qemu_savevm_send_postcopy_listen(fb
);
2162 qemu_savevm_state_complete_precopy(fb
, false, false);
2163 if (migrate_postcopy_ram()) {
2164 qemu_savevm_send_ping(fb
, 3);
2167 qemu_savevm_send_postcopy_run(fb
);
2169 /* <><> end of stuff going into the package */
2171 /* Last point of recovery; as soon as we send the package the destination
2172 * can open devices and potentially start running.
2173 * Lets just check again we've not got any errors.
2175 ret
= qemu_file_get_error(ms
->to_dst_file
);
2177 error_report("postcopy_start: Migration stream errored (pre package)");
2181 restart_block
= false;
2183 /* Now send that blob */
2184 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2189 /* Send a notify to give a chance for anything that needs to happen
2190 * at the transition to postcopy and after the device state; in particular
2191 * spice needs to trigger a transition now
2193 ms
->postcopy_after_devices
= true;
2194 notifier_list_notify(&migration_state_notifiers
, ms
);
2196 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2198 qemu_mutex_unlock_iothread();
2200 if (migrate_postcopy_ram()) {
2202 * Although this ping is just for debug, it could potentially be
2203 * used for getting a better measurement of downtime at the source.
2205 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2208 if (migrate_release_ram()) {
2209 ram_postcopy_migrated_memory_release(ms
);
2212 ret
= qemu_file_get_error(ms
->to_dst_file
);
2214 error_report("postcopy_start: Migration stream errored");
2215 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2216 MIGRATION_STATUS_FAILED
);
2219 trace_postcopy_preempt_enabled(migrate_postcopy_preempt());
2226 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2227 MIGRATION_STATUS_FAILED
);
2228 if (restart_block
) {
2229 /* A failure happened early enough that we know the destination hasn't
2230 * accessed block devices, so we're safe to recover.
2232 Error
*local_err
= NULL
;
2234 bdrv_activate_all(&local_err
);
2236 error_report_err(local_err
);
2239 qemu_mutex_unlock_iothread();
2244 * migration_maybe_pause: Pause if required to by
2245 * migrate_pause_before_switchover called with the iothread locked
2246 * Returns: 0 on success
2248 static int migration_maybe_pause(MigrationState
*s
,
2249 int *current_active_state
,
2252 if (!migrate_pause_before_switchover()) {
2256 /* Since leaving this state is not atomic with posting the semaphore
2257 * it's possible that someone could have issued multiple migrate_continue
2258 * and the semaphore is incorrectly positive at this point;
2259 * the docs say it's undefined to reinit a semaphore that's already
2260 * init'd, so use timedwait to eat up any existing posts.
2262 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2263 /* This block intentionally left blank */
2267 * If the migration is cancelled when it is in the completion phase,
2268 * the migration state is set to MIGRATION_STATUS_CANCELLING.
2269 * So we don't need to wait a semaphore, otherwise we would always
2270 * wait for the 'pause_sem' semaphore.
2272 if (s
->state
!= MIGRATION_STATUS_CANCELLING
) {
2273 qemu_mutex_unlock_iothread();
2274 migrate_set_state(&s
->state
, *current_active_state
,
2275 MIGRATION_STATUS_PRE_SWITCHOVER
);
2276 qemu_sem_wait(&s
->pause_sem
);
2277 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2279 *current_active_state
= new_state
;
2280 qemu_mutex_lock_iothread();
2283 return s
->state
== new_state
? 0 : -EINVAL
;
2287 * migration_completion: Used by migration_thread when there's not much left.
2288 * The caller 'breaks' the loop when this returns.
2290 * @s: Current migration state
2292 static void migration_completion(MigrationState
*s
)
2295 int current_active_state
= s
->state
;
2297 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2298 qemu_mutex_lock_iothread();
2299 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2300 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2301 s
->vm_was_running
= runstate_is_running();
2302 ret
= global_state_store();
2305 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2306 trace_migration_completion_vm_stop(ret
);
2308 ret
= migration_maybe_pause(s
, ¤t_active_state
,
2309 MIGRATION_STATUS_DEVICE
);
2313 * Inactivate disks except in COLO, and track that we
2314 * have done so in order to remember to reactivate
2315 * them if migration fails or is cancelled.
2317 s
->block_inactive
= !migrate_colo();
2318 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
2319 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
2323 qemu_mutex_unlock_iothread();
2328 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2329 trace_migration_completion_postcopy_end();
2331 qemu_mutex_lock_iothread();
2332 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
2333 qemu_mutex_unlock_iothread();
2336 * Shutdown the postcopy fast path thread. This is only needed
2337 * when dest QEMU binary is old (7.1/7.2). QEMU 8.0+ doesn't need
2340 if (migrate_postcopy_preempt() && s
->preempt_pre_7_2
) {
2341 postcopy_preempt_shutdown_file(s
);
2344 trace_migration_completion_postcopy_end_after_complete();
2350 * If rp was opened we must clean up the thread before
2351 * cleaning everything else up (since if there are no failures
2352 * it will wait for the destination to send it's status in
2355 if (s
->rp_state
.rp_thread_created
) {
2357 trace_migration_return_path_end_before();
2358 rp_error
= await_return_path_close_on_source(s
);
2359 trace_migration_return_path_end_after(rp_error
);
2365 if (qemu_file_get_error(s
->to_dst_file
)) {
2366 trace_migration_completion_file_err();
2370 if (migrate_colo() && s
->state
== MIGRATION_STATUS_ACTIVE
) {
2371 /* COLO does not support postcopy */
2372 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
2373 MIGRATION_STATUS_COLO
);
2375 migrate_set_state(&s
->state
, current_active_state
,
2376 MIGRATION_STATUS_COMPLETED
);
2382 if (s
->block_inactive
&& (s
->state
== MIGRATION_STATUS_ACTIVE
||
2383 s
->state
== MIGRATION_STATUS_DEVICE
)) {
2385 * If not doing postcopy, vm_start() will be called: let's
2386 * regain control on images.
2388 Error
*local_err
= NULL
;
2390 qemu_mutex_lock_iothread();
2391 bdrv_activate_all(&local_err
);
2393 error_report_err(local_err
);
2395 s
->block_inactive
= false;
2397 qemu_mutex_unlock_iothread();
2400 migrate_set_state(&s
->state
, current_active_state
,
2401 MIGRATION_STATUS_FAILED
);
2405 * bg_migration_completion: Used by bg_migration_thread when after all the
2406 * RAM has been saved. The caller 'breaks' the loop when this returns.
2408 * @s: Current migration state
2410 static void bg_migration_completion(MigrationState
*s
)
2412 int current_active_state
= s
->state
;
2415 * Stop tracking RAM writes - un-protect memory, un-register UFFD
2416 * memory ranges, flush kernel wait queues and wake up threads
2417 * waiting for write fault to be resolved.
2419 ram_write_tracking_stop();
2421 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2423 * By this moment we have RAM content saved into the migration stream.
2424 * The next step is to flush the non-RAM content (device state)
2425 * right after the ram content. The device state has been stored into
2426 * the temporary buffer before RAM saving started.
2428 qemu_put_buffer(s
->to_dst_file
, s
->bioc
->data
, s
->bioc
->usage
);
2429 qemu_fflush(s
->to_dst_file
);
2430 } else if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
2434 if (qemu_file_get_error(s
->to_dst_file
)) {
2435 trace_migration_completion_file_err();
2439 migrate_set_state(&s
->state
, current_active_state
,
2440 MIGRATION_STATUS_COMPLETED
);
2444 migrate_set_state(&s
->state
, current_active_state
,
2445 MIGRATION_STATUS_FAILED
);
2448 typedef enum MigThrError
{
2449 /* No error detected */
2450 MIG_THR_ERR_NONE
= 0,
2451 /* Detected error, but resumed successfully */
2452 MIG_THR_ERR_RECOVERED
= 1,
2453 /* Detected fatal error, need to exit */
2454 MIG_THR_ERR_FATAL
= 2,
2457 static int postcopy_resume_handshake(MigrationState
*s
)
2459 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
2461 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2462 qemu_sem_wait(&s
->rp_state
.rp_sem
);
2465 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2472 /* Return zero if success, or <0 for error */
2473 static int postcopy_do_resume(MigrationState
*s
)
2478 * Call all the resume_prepare() hooks, so that modules can be
2479 * ready for the migration resume.
2481 ret
= qemu_savevm_state_resume_prepare(s
);
2483 error_report("%s: resume_prepare() failure detected: %d",
2489 * If preempt is enabled, re-establish the preempt channel. Note that
2490 * we do it after resume prepare to make sure the main channel will be
2491 * created before the preempt channel. E.g. with weak network, the
2492 * dest QEMU may get messed up with the preempt and main channels on
2493 * the order of connection setup. This guarantees the correct order.
2495 ret
= postcopy_preempt_establish_channel(s
);
2497 error_report("%s: postcopy_preempt_establish_channel(): %d",
2503 * Last handshake with destination on the resume (destination will
2504 * switch to postcopy-active afterwards)
2506 ret
= postcopy_resume_handshake(s
);
2508 error_report("%s: handshake failed: %d", __func__
, ret
);
2516 * We don't return until we are in a safe state to continue current
2517 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2518 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2520 static MigThrError
postcopy_pause(MigrationState
*s
)
2522 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2528 * Current channel is possibly broken. Release it. Note that this is
2529 * guaranteed even without lock because to_dst_file should only be
2530 * modified by the migration thread. That also guarantees that the
2531 * unregister of yank is safe too without the lock. It should be safe
2532 * even to be within the qemu_file_lock, but we didn't do that to avoid
2533 * taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make
2534 * the qemu_file_lock critical section as small as possible.
2536 assert(s
->to_dst_file
);
2537 migration_ioc_unregister_yank_from_file(s
->to_dst_file
);
2538 qemu_mutex_lock(&s
->qemu_file_lock
);
2539 file
= s
->to_dst_file
;
2540 s
->to_dst_file
= NULL
;
2541 qemu_mutex_unlock(&s
->qemu_file_lock
);
2543 qemu_file_shutdown(file
);
2546 migrate_set_state(&s
->state
, s
->state
,
2547 MIGRATION_STATUS_POSTCOPY_PAUSED
);
2549 error_report("Detected IO failure for postcopy. "
2550 "Migration paused.");
2553 * We wait until things fixed up. Then someone will setup the
2554 * status back for us.
2556 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2557 qemu_sem_wait(&s
->postcopy_pause_sem
);
2560 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2561 /* Woken up by a recover procedure. Give it a shot */
2564 * Firstly, let's wake up the return path now, with a new
2565 * return path channel.
2567 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
2569 /* Do the resume logic */
2570 if (postcopy_do_resume(s
) == 0) {
2571 /* Let's continue! */
2572 trace_postcopy_pause_continued();
2573 return MIG_THR_ERR_RECOVERED
;
2576 * Something wrong happened during the recovery, let's
2577 * pause again. Pause is always better than throwing
2583 /* This is not right... Time to quit. */
2584 return MIG_THR_ERR_FATAL
;
2589 static MigThrError
migration_detect_error(MigrationState
*s
)
2592 int state
= s
->state
;
2593 Error
*local_error
= NULL
;
2595 if (state
== MIGRATION_STATUS_CANCELLING
||
2596 state
== MIGRATION_STATUS_CANCELLED
) {
2597 /* End the migration, but don't set the state to failed */
2598 return MIG_THR_ERR_FATAL
;
2602 * Try to detect any file errors. Note that postcopy_qemufile_src will
2603 * be NULL when postcopy preempt is not enabled.
2605 ret
= qemu_file_get_error_obj_any(s
->to_dst_file
,
2606 s
->postcopy_qemufile_src
,
2609 /* Everything is fine */
2610 assert(!local_error
);
2611 return MIG_THR_ERR_NONE
;
2615 migrate_set_error(s
, local_error
);
2616 error_free(local_error
);
2619 if (state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
) {
2621 * For postcopy, we allow the network to be down for a
2622 * while. After that, it can be continued by a
2625 return postcopy_pause(s
);
2628 * For precopy (or postcopy with error outside IO), we fail
2631 migrate_set_state(&s
->state
, state
, MIGRATION_STATUS_FAILED
);
2632 trace_migration_thread_file_err();
2634 /* Time to stop the migration, now. */
2635 return MIG_THR_ERR_FATAL
;
2639 /* How many bytes have we transferred since the beginning of the migration */
2640 static uint64_t migration_total_bytes(MigrationState
*s
)
2642 return qemu_file_total_transferred(s
->to_dst_file
) +
2643 stat64_get(&mig_stats
.multifd_bytes
);
2646 static void migration_calculate_complete(MigrationState
*s
)
2648 uint64_t bytes
= migration_total_bytes(s
);
2649 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2650 int64_t transfer_time
;
2652 s
->total_time
= end_time
- s
->start_time
;
2655 * It's still not set, so we are precopy migration. For
2656 * postcopy, downtime is calculated during postcopy_start().
2658 s
->downtime
= end_time
- s
->downtime_start
;
2661 transfer_time
= s
->total_time
- s
->setup_time
;
2662 if (transfer_time
) {
2663 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
2667 static void update_iteration_initial_status(MigrationState
*s
)
2670 * Update these three fields at the same time to avoid mismatch info lead
2671 * wrong speed calculation.
2673 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2674 s
->iteration_initial_bytes
= migration_total_bytes(s
);
2675 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
2678 static void migration_update_counters(MigrationState
*s
,
2679 int64_t current_time
)
2681 uint64_t transferred
, transferred_pages
, time_spent
;
2682 uint64_t current_bytes
; /* bytes transferred since the beginning */
2685 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
2689 current_bytes
= migration_total_bytes(s
);
2690 transferred
= current_bytes
- s
->iteration_initial_bytes
;
2691 time_spent
= current_time
- s
->iteration_start_time
;
2692 bandwidth
= (double)transferred
/ time_spent
;
2693 s
->threshold_size
= bandwidth
* migrate_downtime_limit();
2695 s
->mbps
= (((double) transferred
* 8.0) /
2696 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2698 transferred_pages
= ram_get_total_transferred_pages() -
2699 s
->iteration_initial_pages
;
2700 s
->pages_per_second
= (double) transferred_pages
/
2701 (((double) time_spent
/ 1000.0));
2704 * if we haven't sent anything, we don't want to
2705 * recalculate. 10000 is a small enough number for our purposes
2707 if (stat64_get(&mig_stats
.dirty_pages_rate
) &&
2708 transferred
> 10000) {
2709 s
->expected_downtime
=
2710 stat64_get(&mig_stats
.dirty_bytes_last_sync
) / bandwidth
;
2713 qemu_file_reset_rate_limit(s
->to_dst_file
);
2715 update_iteration_initial_status(s
);
2717 trace_migrate_transferred(transferred
, time_spent
,
2718 bandwidth
, s
->threshold_size
);
2721 /* Migration thread iteration status */
2723 MIG_ITERATE_RESUME
, /* Resume current iteration */
2724 MIG_ITERATE_SKIP
, /* Skip current iteration */
2725 MIG_ITERATE_BREAK
, /* Break the loop */
2729 * Return true if continue to the next iteration directly, false
2732 static MigIterateState
migration_iteration_run(MigrationState
*s
)
2734 uint64_t must_precopy
, can_postcopy
;
2735 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
2737 qemu_savevm_state_pending_estimate(&must_precopy
, &can_postcopy
);
2738 uint64_t pending_size
= must_precopy
+ can_postcopy
;
2740 trace_migrate_pending_estimate(pending_size
, must_precopy
, can_postcopy
);
2742 if (must_precopy
<= s
->threshold_size
) {
2743 qemu_savevm_state_pending_exact(&must_precopy
, &can_postcopy
);
2744 pending_size
= must_precopy
+ can_postcopy
;
2745 trace_migrate_pending_exact(pending_size
, must_precopy
, can_postcopy
);
2748 if (!pending_size
|| pending_size
< s
->threshold_size
) {
2749 trace_migration_thread_low_pending(pending_size
);
2750 migration_completion(s
);
2751 return MIG_ITERATE_BREAK
;
2754 /* Still a significant amount to transfer */
2755 if (!in_postcopy
&& must_precopy
<= s
->threshold_size
&&
2756 qatomic_read(&s
->start_postcopy
)) {
2757 if (postcopy_start(s
)) {
2758 error_report("%s: postcopy failed to start", __func__
);
2760 return MIG_ITERATE_SKIP
;
2763 /* Just another iteration step */
2764 qemu_savevm_state_iterate(s
->to_dst_file
, in_postcopy
);
2765 return MIG_ITERATE_RESUME
;
2768 static void migration_iteration_finish(MigrationState
*s
)
2770 /* If we enabled cpu throttling for auto-converge, turn it off. */
2771 cpu_throttle_stop();
2773 qemu_mutex_lock_iothread();
2775 case MIGRATION_STATUS_COMPLETED
:
2776 migration_calculate_complete(s
);
2777 runstate_set(RUN_STATE_POSTMIGRATE
);
2779 case MIGRATION_STATUS_COLO
:
2780 if (!migrate_colo()) {
2781 error_report("%s: critical error: calling COLO code without "
2782 "COLO enabled", __func__
);
2784 migrate_start_colo_process(s
);
2785 s
->vm_was_running
= true;
2787 case MIGRATION_STATUS_FAILED
:
2788 case MIGRATION_STATUS_CANCELLED
:
2789 case MIGRATION_STATUS_CANCELLING
:
2790 if (s
->vm_was_running
) {
2791 if (!runstate_check(RUN_STATE_SHUTDOWN
)) {
2795 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2796 runstate_set(RUN_STATE_POSTMIGRATE
);
2802 /* Should not reach here, but if so, forgive the VM. */
2803 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
2806 migrate_fd_cleanup_schedule(s
);
2807 qemu_mutex_unlock_iothread();
2810 static void bg_migration_iteration_finish(MigrationState
*s
)
2812 qemu_mutex_lock_iothread();
2814 case MIGRATION_STATUS_COMPLETED
:
2815 migration_calculate_complete(s
);
2818 case MIGRATION_STATUS_ACTIVE
:
2819 case MIGRATION_STATUS_FAILED
:
2820 case MIGRATION_STATUS_CANCELLED
:
2821 case MIGRATION_STATUS_CANCELLING
:
2825 /* Should not reach here, but if so, forgive the VM. */
2826 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
2830 migrate_fd_cleanup_schedule(s
);
2831 qemu_mutex_unlock_iothread();
2835 * Return true if continue to the next iteration directly, false
2838 static MigIterateState
bg_migration_iteration_run(MigrationState
*s
)
2842 res
= qemu_savevm_state_iterate(s
->to_dst_file
, false);
2844 bg_migration_completion(s
);
2845 return MIG_ITERATE_BREAK
;
2848 return MIG_ITERATE_RESUME
;
2851 void migration_make_urgent_request(void)
2853 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
2856 void migration_consume_urgent_request(void)
2858 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
2861 /* Returns true if the rate limiting was broken by an urgent request */
2862 bool migration_rate_limit(void)
2864 int64_t now
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2865 MigrationState
*s
= migrate_get_current();
2867 bool urgent
= false;
2868 migration_update_counters(s
, now
);
2869 if (qemu_file_rate_limit(s
->to_dst_file
)) {
2871 if (qemu_file_get_error(s
->to_dst_file
)) {
2875 * Wait for a delay to do rate limiting OR
2876 * something urgent to post the semaphore.
2878 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- now
;
2879 trace_migration_rate_limit_pre(ms
);
2880 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
2882 * We were woken by one or more urgent things but
2883 * the timedwait will have consumed one of them.
2884 * The service routine for the urgent wake will dec
2885 * the semaphore itself for each item it consumes,
2886 * so add this one we just eat back.
2888 qemu_sem_post(&s
->rate_limit_sem
);
2891 trace_migration_rate_limit_post(urgent
);
2897 * if failover devices are present, wait they are completely
2901 static void qemu_savevm_wait_unplug(MigrationState
*s
, int old_state
,
2904 if (qemu_savevm_state_guest_unplug_pending()) {
2905 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_WAIT_UNPLUG
);
2907 while (s
->state
== MIGRATION_STATUS_WAIT_UNPLUG
&&
2908 qemu_savevm_state_guest_unplug_pending()) {
2909 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
2911 if (s
->state
!= MIGRATION_STATUS_WAIT_UNPLUG
) {
2912 int timeout
= 120; /* 30 seconds */
2914 * migration has been canceled
2915 * but as we have started an unplug we must wait the end
2916 * to be able to plug back the card
2918 while (timeout
-- && qemu_savevm_state_guest_unplug_pending()) {
2919 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
2921 if (qemu_savevm_state_guest_unplug_pending() &&
2923 warn_report("migration: partially unplugged device on "
2928 migrate_set_state(&s
->state
, MIGRATION_STATUS_WAIT_UNPLUG
, new_state
);
2930 migrate_set_state(&s
->state
, old_state
, new_state
);
2935 * Master migration thread on the source VM.
2936 * It drives the migration and pumps the data down the outgoing channel.
2938 static void *migration_thread(void *opaque
)
2940 MigrationState
*s
= opaque
;
2941 MigrationThread
*thread
= NULL
;
2942 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
2943 MigThrError thr_error
;
2944 bool urgent
= false;
2946 thread
= MigrationThreadAdd("live_migration", qemu_get_thread_id());
2948 rcu_register_thread();
2950 object_ref(OBJECT(s
));
2951 update_iteration_initial_status(s
);
2953 qemu_savevm_state_header(s
->to_dst_file
);
2956 * If we opened the return path, we need to make sure dst has it
2959 if (s
->rp_state
.rp_thread_created
) {
2960 /* Now tell the dest that it should open its end so it can reply */
2961 qemu_savevm_send_open_return_path(s
->to_dst_file
);
2963 /* And do a ping that will make stuff easier to debug */
2964 qemu_savevm_send_ping(s
->to_dst_file
, 1);
2967 if (migrate_postcopy()) {
2969 * Tell the destination that we *might* want to do postcopy later;
2970 * if the other end can't do postcopy it should fail now, nice and
2973 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
2976 if (migrate_colo()) {
2977 /* Notify migration destination that we enable COLO */
2978 qemu_savevm_send_colo_enable(s
->to_dst_file
);
2981 qemu_savevm_state_setup(s
->to_dst_file
);
2983 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
2984 MIGRATION_STATUS_ACTIVE
);
2986 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
2988 trace_migration_thread_setup_complete();
2990 while (migration_is_active(s
)) {
2991 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
2992 MigIterateState iter_state
= migration_iteration_run(s
);
2993 if (iter_state
== MIG_ITERATE_SKIP
) {
2995 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3001 * Try to detect any kind of failures, and see whether we
3002 * should stop the migration now.
3004 thr_error
= migration_detect_error(s
);
3005 if (thr_error
== MIG_THR_ERR_FATAL
) {
3006 /* Stop migration */
3008 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
3010 * Just recovered from a e.g. network failure, reset all
3011 * the local variables. This is important to avoid
3012 * breaking transferred_bytes and bandwidth calculation
3014 update_iteration_initial_status(s
);
3017 urgent
= migration_rate_limit();
3020 trace_migration_thread_after_loop();
3021 migration_iteration_finish(s
);
3022 object_unref(OBJECT(s
));
3023 rcu_unregister_thread();
3024 MigrationThreadDel(thread
);
3028 static void bg_migration_vm_start_bh(void *opaque
)
3030 MigrationState
*s
= opaque
;
3032 qemu_bh_delete(s
->vm_start_bh
);
3033 s
->vm_start_bh
= NULL
;
3036 s
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - s
->downtime_start
;
3040 * Background snapshot thread, based on live migration code.
3041 * This is an alternative implementation of live migration mechanism
3042 * introduced specifically to support background snapshots.
3044 * It takes advantage of userfault_fd write protection mechanism introduced
3045 * in v5.7 kernel. Compared to existing dirty page logging migration much
3046 * lesser stream traffic is produced resulting in smaller snapshot images,
3047 * simply cause of no page duplicates can get into the stream.
3049 * Another key point is that generated vmstate stream reflects machine state
3050 * 'frozen' at the beginning of snapshot creation compared to dirty page logging
3051 * mechanism, which effectively results in that saved snapshot is the state of VM
3052 * at the end of the process.
3054 static void *bg_migration_thread(void *opaque
)
3056 MigrationState
*s
= opaque
;
3057 int64_t setup_start
;
3058 MigThrError thr_error
;
3060 bool early_fail
= true;
3062 rcu_register_thread();
3063 object_ref(OBJECT(s
));
3065 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
3067 setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3069 * We want to save vmstate for the moment when migration has been
3070 * initiated but also we want to save RAM content while VM is running.
3071 * The RAM content should appear first in the vmstate. So, we first
3072 * stash the non-RAM part of the vmstate to the temporary buffer,
3073 * then write RAM part of the vmstate to the migration stream
3074 * with vCPUs running and, finally, write stashed non-RAM part of
3075 * the vmstate from the buffer to the migration stream.
3077 s
->bioc
= qio_channel_buffer_new(512 * 1024);
3078 qio_channel_set_name(QIO_CHANNEL(s
->bioc
), "vmstate-buffer");
3079 fb
= qemu_file_new_output(QIO_CHANNEL(s
->bioc
));
3080 object_unref(OBJECT(s
->bioc
));
3082 update_iteration_initial_status(s
);
3085 * Prepare for tracking memory writes with UFFD-WP - populate
3086 * RAM pages before protecting.
3089 ram_write_tracking_prepare();
3092 qemu_savevm_state_header(s
->to_dst_file
);
3093 qemu_savevm_state_setup(s
->to_dst_file
);
3095 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
3096 MIGRATION_STATUS_ACTIVE
);
3098 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3100 trace_migration_thread_setup_complete();
3101 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3103 qemu_mutex_lock_iothread();
3106 * If VM is currently in suspended state, then, to make a valid runstate
3107 * transition in vm_stop_force_state() we need to wakeup it up.
3109 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3110 s
->vm_was_running
= runstate_is_running();
3112 if (global_state_store()) {
3115 /* Forcibly stop VM before saving state of vCPUs and devices */
3116 if (vm_stop_force_state(RUN_STATE_PAUSED
)) {
3120 * Put vCPUs in sync with shadow context structures, then
3121 * save their state to channel-buffer along with devices.
3123 cpu_synchronize_all_states();
3124 if (qemu_savevm_state_complete_precopy_non_iterable(fb
, false, false)) {
3128 * Since we are going to get non-iterable state data directly
3129 * from s->bioc->data, explicit flush is needed here.
3133 /* Now initialize UFFD context and start tracking RAM writes */
3134 if (ram_write_tracking_start()) {
3140 * Start VM from BH handler to avoid write-fault lock here.
3141 * UFFD-WP protection for the whole RAM is already enabled so
3142 * calling VM state change notifiers from vm_start() would initiate
3143 * writes to virtio VQs memory which is in write-protected region.
3145 s
->vm_start_bh
= qemu_bh_new(bg_migration_vm_start_bh
, s
);
3146 qemu_bh_schedule(s
->vm_start_bh
);
3148 qemu_mutex_unlock_iothread();
3150 while (migration_is_active(s
)) {
3151 MigIterateState iter_state
= bg_migration_iteration_run(s
);
3152 if (iter_state
== MIG_ITERATE_SKIP
) {
3154 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3159 * Try to detect any kind of failures, and see whether we
3160 * should stop the migration now.
3162 thr_error
= migration_detect_error(s
);
3163 if (thr_error
== MIG_THR_ERR_FATAL
) {
3164 /* Stop migration */
3168 migration_update_counters(s
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
));
3171 trace_migration_thread_after_loop();
3175 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
3176 MIGRATION_STATUS_FAILED
);
3177 qemu_mutex_unlock_iothread();
3180 bg_migration_iteration_finish(s
);
3183 object_unref(OBJECT(s
));
3184 rcu_unregister_thread();
3189 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
3191 Error
*local_err
= NULL
;
3192 uint64_t rate_limit
;
3193 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
3196 * If there's a previous error, free it and prepare for another one.
3197 * Meanwhile if migration completes successfully, there won't have an error
3198 * dumped when calling migrate_fd_cleanup().
3200 migrate_error_free(s
);
3202 s
->expected_downtime
= migrate_downtime_limit();
3204 assert(s
->cleanup_bh
);
3206 assert(!s
->cleanup_bh
);
3207 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup_bh
, s
);
3210 migrate_fd_error(s
, error_in
);
3213 * Don't do cleanup for resume if channel is invalid, but only dump
3214 * the error. We wait for another channel connect from the user.
3215 * The error_report still gives HMP user a hint on what failed.
3216 * It's normally done in migrate_fd_cleanup(), but call it here
3219 error_report_err(error_copy(s
->error
));
3221 migrate_fd_cleanup(s
);
3227 /* This is a resumed migration */
3228 rate_limit
= migrate_max_postcopy_bandwidth() /
3231 /* This is a fresh new migration */
3232 rate_limit
= migrate_max_bandwidth() / XFER_LIMIT_RATIO
;
3234 /* Notify before starting migration thread */
3235 notifier_list_notify(&migration_state_notifiers
, s
);
3238 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
3239 qemu_file_set_blocking(s
->to_dst_file
, true);
3242 * Open the return path. For postcopy, it is used exclusively. For
3243 * precopy, only if user specified "return-path" capability would
3244 * QEMU uses the return path.
3246 if (migrate_postcopy_ram() || migrate_return_path()) {
3247 if (open_return_path_on_source(s
, !resume
)) {
3248 error_report("Unable to open return-path for postcopy");
3249 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
3250 migrate_fd_cleanup(s
);
3256 * This needs to be done before resuming a postcopy. Note: for newer
3257 * QEMUs we will delay the channel creation until postcopy_start(), to
3258 * avoid disorder of channel creations.
3260 if (migrate_postcopy_preempt() && s
->preempt_pre_7_2
) {
3261 postcopy_preempt_setup(s
);
3265 /* Wakeup the main migration thread to do the recovery */
3266 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
3267 MIGRATION_STATUS_POSTCOPY_RECOVER
);
3268 qemu_sem_post(&s
->postcopy_pause_sem
);
3272 if (multifd_save_setup(&local_err
) != 0) {
3273 error_report_err(local_err
);
3274 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3275 MIGRATION_STATUS_FAILED
);
3276 migrate_fd_cleanup(s
);
3280 if (migrate_background_snapshot()) {
3281 qemu_thread_create(&s
->thread
, "bg_snapshot",
3282 bg_migration_thread
, s
, QEMU_THREAD_JOINABLE
);
3284 qemu_thread_create(&s
->thread
, "live_migration",
3285 migration_thread
, s
, QEMU_THREAD_JOINABLE
);
3287 s
->migration_thread_running
= true;
3290 static void migration_class_init(ObjectClass
*klass
, void *data
)
3292 DeviceClass
*dc
= DEVICE_CLASS(klass
);
3294 dc
->user_creatable
= false;
3295 device_class_set_props(dc
, migration_properties
);
3298 static void migration_instance_finalize(Object
*obj
)
3300 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3302 qemu_mutex_destroy(&ms
->error_mutex
);
3303 qemu_mutex_destroy(&ms
->qemu_file_lock
);
3304 qemu_sem_destroy(&ms
->wait_unplug_sem
);
3305 qemu_sem_destroy(&ms
->rate_limit_sem
);
3306 qemu_sem_destroy(&ms
->pause_sem
);
3307 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
3308 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
3309 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
3310 qemu_sem_destroy(&ms
->rp_state
.rp_pong_acks
);
3311 qemu_sem_destroy(&ms
->postcopy_qemufile_src_sem
);
3312 error_free(ms
->error
);
3315 static void migration_instance_init(Object
*obj
)
3317 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3319 ms
->state
= MIGRATION_STATUS_NONE
;
3321 ms
->pages_per_second
= -1;
3322 qemu_sem_init(&ms
->pause_sem
, 0);
3323 qemu_mutex_init(&ms
->error_mutex
);
3325 migrate_params_init(&ms
->parameters
);
3327 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
3328 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
3329 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
3330 qemu_sem_init(&ms
->rp_state
.rp_pong_acks
, 0);
3331 qemu_sem_init(&ms
->rate_limit_sem
, 0);
3332 qemu_sem_init(&ms
->wait_unplug_sem
, 0);
3333 qemu_sem_init(&ms
->postcopy_qemufile_src_sem
, 0);
3334 qemu_mutex_init(&ms
->qemu_file_lock
);
3338 * Return true if check pass, false otherwise. Error will be put
3339 * inside errp if provided.
3341 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
3343 /* Assuming all off */
3344 bool old_caps
[MIGRATION_CAPABILITY__MAX
] = { 0 };
3346 if (!migrate_params_check(&ms
->parameters
, errp
)) {
3350 return migrate_caps_check(old_caps
, ms
->capabilities
, errp
);
3353 static const TypeInfo migration_type
= {
3354 .name
= TYPE_MIGRATION
,
3356 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3357 * not created using qdev_new(), it is not attached to the qdev
3358 * device tree, and it is never realized.
3360 * TODO: Make this TYPE_OBJECT once QOM provides something like
3361 * TYPE_DEVICE's "-global" properties.
3363 .parent
= TYPE_DEVICE
,
3364 .class_init
= migration_class_init
,
3365 .class_size
= sizeof(MigrationClass
),
3366 .instance_size
= sizeof(MigrationState
),
3367 .instance_init
= migration_instance_init
,
3368 .instance_finalize
= migration_instance_finalize
,
3371 static void register_migration_types(void)
3373 type_register_static(&migration_type
);
3376 type_init(register_migration_types
);