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 #ifndef CONFIG_REPLICATION
396 error_report("ENABLE_COLO command come in migration stream, but COLO "
397 "module is not built in");
401 if (!migrate_colo()) {
402 error_report("ENABLE_COLO command come in migration stream, but c-colo "
403 "capability is not set");
407 if (ram_block_discard_disable(true)) {
408 error_report("COLO: cannot disable RAM discard");
411 migration_colo_enabled
= true;
415 void migrate_add_address(SocketAddress
*address
)
417 MigrationIncomingState
*mis
= migration_incoming_get_current();
419 QAPI_LIST_PREPEND(mis
->socket_address_list
,
420 QAPI_CLONE(SocketAddress
, address
));
423 static void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
425 const char *p
= NULL
;
427 /* URI is not suitable for migration? */
428 if (!migration_channels_and_uri_compatible(uri
, errp
)) {
432 qapi_event_send_migration(MIGRATION_STATUS_SETUP
);
433 if (strstart(uri
, "tcp:", &p
) ||
434 strstart(uri
, "unix:", NULL
) ||
435 strstart(uri
, "vsock:", NULL
)) {
436 socket_start_incoming_migration(p
? p
: uri
, errp
);
438 } else if (strstart(uri
, "rdma:", &p
)) {
439 rdma_start_incoming_migration(p
, errp
);
441 } else if (strstart(uri
, "exec:", &p
)) {
442 exec_start_incoming_migration(p
, errp
);
443 } else if (strstart(uri
, "fd:", &p
)) {
444 fd_start_incoming_migration(p
, errp
);
446 error_setg(errp
, "unknown migration protocol: %s", uri
);
450 static void process_incoming_migration_bh(void *opaque
)
452 Error
*local_err
= NULL
;
453 MigrationIncomingState
*mis
= opaque
;
455 /* If capability late_block_activate is set:
456 * Only fire up the block code now if we're going to restart the
457 * VM, else 'cont' will do it.
458 * This causes file locking to happen; so we don't want it to happen
459 * unless we really are starting the VM.
461 if (!migrate_late_block_activate() ||
462 (autostart
&& (!global_state_received() ||
463 global_state_get_runstate() == RUN_STATE_RUNNING
))) {
464 /* Make sure all file formats throw away their mutable metadata.
465 * If we get an error here, just don't restart the VM yet. */
466 bdrv_activate_all(&local_err
);
468 error_report_err(local_err
);
475 * This must happen after all error conditions are dealt with and
476 * we're sure the VM is going to be running on this host.
478 qemu_announce_self(&mis
->announce_timer
, migrate_announce_params());
480 multifd_load_shutdown();
482 dirty_bitmap_mig_before_vm_start();
484 if (!global_state_received() ||
485 global_state_get_runstate() == RUN_STATE_RUNNING
) {
489 runstate_set(RUN_STATE_PAUSED
);
491 } else if (migration_incoming_colo_enabled()) {
492 migration_incoming_disable_colo();
495 runstate_set(global_state_get_runstate());
498 * This must happen after any state changes since as soon as an external
499 * observer sees this event they might start to prod at the VM assuming
502 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
503 MIGRATION_STATUS_COMPLETED
);
504 qemu_bh_delete(mis
->bh
);
505 migration_incoming_state_destroy();
508 static void coroutine_fn
509 process_incoming_migration_co(void *opaque
)
511 MigrationIncomingState
*mis
= migration_incoming_get_current();
514 Error
*local_err
= NULL
;
516 assert(mis
->from_src_file
);
518 if (compress_threads_load_setup(mis
->from_src_file
)) {
519 error_report("Failed to setup decompress threads");
523 mis
->migration_incoming_co
= qemu_coroutine_self();
524 mis
->largest_page_size
= qemu_ram_pagesize_largest();
525 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
526 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
527 MIGRATION_STATUS_ACTIVE
);
528 ret
= qemu_loadvm_state(mis
->from_src_file
);
530 ps
= postcopy_state_get();
531 trace_process_incoming_migration_co_end(ret
, ps
);
532 if (ps
!= POSTCOPY_INCOMING_NONE
) {
533 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
535 * Where a migration had postcopy enabled (and thus went to advise)
536 * but managed to complete within the precopy period, we can use
539 postcopy_ram_incoming_cleanup(mis
);
540 } else if (ret
>= 0) {
542 * Postcopy was started, cleanup should happen at the end of the
545 trace_process_incoming_migration_co_postcopy_end_main();
548 /* Else if something went wrong then just fall out of the normal exit */
552 error_report("load of migration failed: %s", strerror(-ret
));
556 /* we get COLO info, and know if we are in COLO mode */
557 if (migration_incoming_colo_enabled()) {
558 QemuThread colo_incoming_thread
;
560 /* Make sure all file formats throw away their mutable metadata */
561 bdrv_activate_all(&local_err
);
563 error_report_err(local_err
);
567 qemu_thread_create(&colo_incoming_thread
, "COLO incoming",
568 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
569 qemu_coroutine_yield();
571 qemu_mutex_unlock_iothread();
572 /* Wait checkpoint incoming thread exit before free resource */
573 qemu_thread_join(&colo_incoming_thread
);
574 qemu_mutex_lock_iothread();
575 /* We hold the global iothread lock, so it is safe here */
576 colo_release_ram_cache();
579 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
580 qemu_bh_schedule(mis
->bh
);
581 mis
->migration_incoming_co
= NULL
;
585 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
586 MIGRATION_STATUS_FAILED
);
587 qemu_fclose(mis
->from_src_file
);
589 multifd_load_cleanup();
590 compress_threads_load_cleanup();
596 * migration_incoming_setup: Setup incoming migration
597 * @f: file for main migration channel
598 * @errp: where to put errors
600 * Returns: %true on success, %false on error.
602 static bool migration_incoming_setup(QEMUFile
*f
, Error
**errp
)
604 MigrationIncomingState
*mis
= migration_incoming_get_current();
606 if (!mis
->from_src_file
) {
607 mis
->from_src_file
= f
;
609 qemu_file_set_blocking(f
, false);
613 void migration_incoming_process(void)
615 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
616 qemu_coroutine_enter(co
);
619 /* Returns true if recovered from a paused migration, otherwise false */
620 static bool postcopy_try_recover(void)
622 MigrationIncomingState
*mis
= migration_incoming_get_current();
624 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
625 /* Resumed from a paused postcopy migration */
627 /* This should be set already in migration_incoming_setup() */
628 assert(mis
->from_src_file
);
629 /* Postcopy has standalone thread to do vm load */
630 qemu_file_set_blocking(mis
->from_src_file
, true);
632 /* Re-configure the return path */
633 mis
->to_src_file
= qemu_file_get_return_path(mis
->from_src_file
);
635 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
636 MIGRATION_STATUS_POSTCOPY_RECOVER
);
639 * Here, we only wake up the main loading thread (while the
640 * rest threads will still be waiting), so that we can receive
641 * commands from source now, and answer it if needed. The
642 * rest threads will be woken up afterwards until we are sure
643 * that source is ready to reply to page requests.
645 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
652 void migration_fd_process_incoming(QEMUFile
*f
, Error
**errp
)
654 if (!migration_incoming_setup(f
, errp
)) {
657 if (postcopy_try_recover()) {
660 migration_incoming_process();
664 * Returns true when we want to start a new incoming migration process,
667 static bool migration_should_start_incoming(bool main_channel
)
669 /* Multifd doesn't start unless all channels are established */
670 if (migrate_multifd()) {
671 return migration_has_all_channels();
674 /* Preempt channel only starts when the main channel is created */
675 if (migrate_postcopy_preempt()) {
680 * For all the rest types of migration, we should only reach here when
681 * it's the main channel that's being created, and we should always
682 * proceed with this channel.
684 assert(main_channel
);
688 void migration_ioc_process_incoming(QIOChannel
*ioc
, Error
**errp
)
690 MigrationIncomingState
*mis
= migration_incoming_get_current();
691 Error
*local_err
= NULL
;
693 bool default_channel
= true;
694 uint32_t channel_magic
= 0;
697 if (migrate_multifd() && !migrate_postcopy_ram() &&
698 qio_channel_has_feature(ioc
, QIO_CHANNEL_FEATURE_READ_MSG_PEEK
)) {
700 * With multiple channels, it is possible that we receive channels
701 * out of order on destination side, causing incorrect mapping of
702 * source channels on destination side. Check channel MAGIC to
703 * decide type of channel. Please note this is best effort, postcopy
704 * preempt channel does not send any magic number so avoid it for
705 * postcopy live migration. Also tls live migration already does
706 * tls handshake while initializing main channel so with tls this
707 * issue is not possible.
709 ret
= migration_channel_read_peek(ioc
, (void *)&channel_magic
,
710 sizeof(channel_magic
), &local_err
);
713 error_propagate(errp
, local_err
);
717 default_channel
= (channel_magic
== cpu_to_be32(QEMU_VM_FILE_MAGIC
));
719 default_channel
= !mis
->from_src_file
;
722 if (multifd_load_setup(errp
) != 0) {
723 error_setg(errp
, "Failed to setup multifd channels");
727 if (default_channel
) {
728 f
= qemu_file_new_input(ioc
);
730 if (!migration_incoming_setup(f
, errp
)) {
734 /* Multiple connections */
735 assert(migration_needs_multiple_sockets());
736 if (migrate_multifd()) {
737 multifd_recv_new_channel(ioc
, &local_err
);
739 assert(migrate_postcopy_preempt());
740 f
= qemu_file_new_input(ioc
);
741 postcopy_preempt_new_channel(mis
, f
);
744 error_propagate(errp
, local_err
);
749 if (migration_should_start_incoming(default_channel
)) {
750 /* If it's a recovery, we're done */
751 if (postcopy_try_recover()) {
754 migration_incoming_process();
759 * @migration_has_all_channels: We have received all channels that we need
761 * Returns true when we have got connections to all the channels that
762 * we need for migration.
764 bool migration_has_all_channels(void)
766 MigrationIncomingState
*mis
= migration_incoming_get_current();
768 if (!mis
->from_src_file
) {
772 if (migrate_multifd()) {
773 return multifd_recv_all_channels_created();
776 if (migrate_postcopy_preempt()) {
777 return mis
->postcopy_qemufile_dst
!= NULL
;
784 * Send a 'SHUT' message on the return channel with the given value
785 * to indicate that we've finished with the RP. Non-0 value indicates
788 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
793 buf
= cpu_to_be32(value
);
794 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
798 * Send a 'PONG' message on the return channel with the given value
799 * (normally in response to a 'PING')
801 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
806 buf
= cpu_to_be32(value
);
807 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
810 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
818 * First, we send the header part. It contains only the len of
819 * idstr, and the idstr itself.
821 len
= strlen(block_name
);
823 memcpy(buf
+ 1, block_name
, len
);
825 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
826 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
831 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
834 * Next, we dump the received bitmap to the stream.
836 * TODO: currently we are safe since we are the only one that is
837 * using the to_src_file handle (fault thread is still paused),
838 * and it's ok even not taking the mutex. However the best way is
839 * to take the lock before sending the message header, and release
840 * the lock after sending the bitmap.
842 qemu_mutex_lock(&mis
->rp_mutex
);
843 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
844 qemu_mutex_unlock(&mis
->rp_mutex
);
846 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
849 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
853 buf
= cpu_to_be32(value
);
854 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
858 * Return true if we're already in the middle of a migration
859 * (i.e. any of the active or setup states)
861 bool migration_is_setup_or_active(int state
)
864 case MIGRATION_STATUS_ACTIVE
:
865 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
866 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
867 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
868 case MIGRATION_STATUS_SETUP
:
869 case MIGRATION_STATUS_PRE_SWITCHOVER
:
870 case MIGRATION_STATUS_DEVICE
:
871 case MIGRATION_STATUS_WAIT_UNPLUG
:
872 case MIGRATION_STATUS_COLO
:
881 bool migration_is_running(int state
)
884 case MIGRATION_STATUS_ACTIVE
:
885 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
886 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
887 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
888 case MIGRATION_STATUS_SETUP
:
889 case MIGRATION_STATUS_PRE_SWITCHOVER
:
890 case MIGRATION_STATUS_DEVICE
:
891 case MIGRATION_STATUS_WAIT_UNPLUG
:
892 case MIGRATION_STATUS_CANCELLING
:
901 static bool migrate_show_downtime(MigrationState
*s
)
903 return (s
->state
== MIGRATION_STATUS_COMPLETED
) || migration_in_postcopy();
906 static void populate_time_info(MigrationInfo
*info
, MigrationState
*s
)
908 info
->has_status
= true;
909 info
->has_setup_time
= true;
910 info
->setup_time
= s
->setup_time
;
912 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
913 info
->has_total_time
= true;
914 info
->total_time
= s
->total_time
;
916 info
->has_total_time
= true;
917 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) -
921 if (migrate_show_downtime(s
)) {
922 info
->has_downtime
= true;
923 info
->downtime
= s
->downtime
;
925 info
->has_expected_downtime
= true;
926 info
->expected_downtime
= s
->expected_downtime
;
930 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
932 size_t page_size
= qemu_target_page_size();
934 info
->ram
= g_malloc0(sizeof(*info
->ram
));
935 info
->ram
->transferred
= stat64_get(&mig_stats
.transferred
);
936 info
->ram
->total
= ram_bytes_total();
937 info
->ram
->duplicate
= stat64_get(&mig_stats
.zero_pages
);
938 /* legacy value. It is not used anymore */
939 info
->ram
->skipped
= 0;
940 info
->ram
->normal
= stat64_get(&mig_stats
.normal_pages
);
941 info
->ram
->normal_bytes
= info
->ram
->normal
* page_size
;
942 info
->ram
->mbps
= s
->mbps
;
943 info
->ram
->dirty_sync_count
=
944 stat64_get(&mig_stats
.dirty_sync_count
);
945 info
->ram
->dirty_sync_missed_zero_copy
=
946 stat64_get(&mig_stats
.dirty_sync_missed_zero_copy
);
947 info
->ram
->postcopy_requests
=
948 stat64_get(&mig_stats
.postcopy_requests
);
949 info
->ram
->page_size
= page_size
;
950 info
->ram
->multifd_bytes
= stat64_get(&mig_stats
.multifd_bytes
);
951 info
->ram
->pages_per_second
= s
->pages_per_second
;
952 info
->ram
->precopy_bytes
= stat64_get(&mig_stats
.precopy_bytes
);
953 info
->ram
->downtime_bytes
= stat64_get(&mig_stats
.downtime_bytes
);
954 info
->ram
->postcopy_bytes
= stat64_get(&mig_stats
.postcopy_bytes
);
956 if (migrate_xbzrle()) {
957 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
958 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
959 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
960 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
961 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
962 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
963 info
->xbzrle_cache
->encoding_rate
= xbzrle_counters
.encoding_rate
;
964 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
967 if (migrate_compress()) {
968 info
->compression
= g_malloc0(sizeof(*info
->compression
));
969 info
->compression
->pages
= compression_counters
.pages
;
970 info
->compression
->busy
= compression_counters
.busy
;
971 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
972 info
->compression
->compressed_size
=
973 compression_counters
.compressed_size
;
974 info
->compression
->compression_rate
=
975 compression_counters
.compression_rate
;
978 if (cpu_throttle_active()) {
979 info
->has_cpu_throttle_percentage
= true;
980 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
983 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
984 info
->ram
->remaining
= ram_bytes_remaining();
985 info
->ram
->dirty_pages_rate
=
986 stat64_get(&mig_stats
.dirty_pages_rate
);
990 static void populate_disk_info(MigrationInfo
*info
)
992 if (blk_mig_active()) {
993 info
->disk
= g_malloc0(sizeof(*info
->disk
));
994 info
->disk
->transferred
= blk_mig_bytes_transferred();
995 info
->disk
->remaining
= blk_mig_bytes_remaining();
996 info
->disk
->total
= blk_mig_bytes_total();
1000 static void fill_source_migration_info(MigrationInfo
*info
)
1002 MigrationState
*s
= migrate_get_current();
1003 int state
= qatomic_read(&s
->state
);
1004 GSList
*cur_blocker
= migration_blockers
;
1006 info
->blocked_reasons
= NULL
;
1009 * There are two types of reasons a migration might be blocked;
1010 * a) devices marked in VMState as non-migratable, and
1011 * b) Explicit migration blockers
1012 * We need to add both of them here.
1014 qemu_savevm_non_migratable_list(&info
->blocked_reasons
);
1016 while (cur_blocker
) {
1017 QAPI_LIST_PREPEND(info
->blocked_reasons
,
1018 g_strdup(error_get_pretty(cur_blocker
->data
)));
1019 cur_blocker
= g_slist_next(cur_blocker
);
1021 info
->has_blocked_reasons
= info
->blocked_reasons
!= NULL
;
1024 case MIGRATION_STATUS_NONE
:
1025 /* no migration has happened ever */
1026 /* do not overwrite destination migration status */
1028 case MIGRATION_STATUS_SETUP
:
1029 info
->has_status
= true;
1030 info
->has_total_time
= false;
1032 case MIGRATION_STATUS_ACTIVE
:
1033 case MIGRATION_STATUS_CANCELLING
:
1034 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1035 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1036 case MIGRATION_STATUS_DEVICE
:
1037 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1038 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1039 /* TODO add some postcopy stats */
1040 populate_time_info(info
, s
);
1041 populate_ram_info(info
, s
);
1042 populate_disk_info(info
);
1043 populate_vfio_info(info
);
1045 case MIGRATION_STATUS_COLO
:
1046 info
->has_status
= true;
1047 /* TODO: display COLO specific information (checkpoint info etc.) */
1049 case MIGRATION_STATUS_COMPLETED
:
1050 populate_time_info(info
, s
);
1051 populate_ram_info(info
, s
);
1052 populate_vfio_info(info
);
1054 case MIGRATION_STATUS_FAILED
:
1055 info
->has_status
= true;
1057 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
1060 case MIGRATION_STATUS_CANCELLED
:
1061 info
->has_status
= true;
1063 case MIGRATION_STATUS_WAIT_UNPLUG
:
1064 info
->has_status
= true;
1067 info
->status
= state
;
1070 static void fill_destination_migration_info(MigrationInfo
*info
)
1072 MigrationIncomingState
*mis
= migration_incoming_get_current();
1074 if (mis
->socket_address_list
) {
1075 info
->has_socket_address
= true;
1076 info
->socket_address
=
1077 QAPI_CLONE(SocketAddressList
, mis
->socket_address_list
);
1080 switch (mis
->state
) {
1081 case MIGRATION_STATUS_NONE
:
1083 case MIGRATION_STATUS_SETUP
:
1084 case MIGRATION_STATUS_CANCELLING
:
1085 case MIGRATION_STATUS_CANCELLED
:
1086 case MIGRATION_STATUS_ACTIVE
:
1087 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1088 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1089 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1090 case MIGRATION_STATUS_FAILED
:
1091 case MIGRATION_STATUS_COLO
:
1092 info
->has_status
= true;
1094 case MIGRATION_STATUS_COMPLETED
:
1095 info
->has_status
= true;
1096 fill_destination_postcopy_migration_info(info
);
1099 info
->status
= mis
->state
;
1102 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1104 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1106 fill_destination_migration_info(info
);
1107 fill_source_migration_info(info
);
1112 void qmp_migrate_start_postcopy(Error
**errp
)
1114 MigrationState
*s
= migrate_get_current();
1116 if (!migrate_postcopy()) {
1117 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1118 " the start of migration");
1122 if (s
->state
== MIGRATION_STATUS_NONE
) {
1123 error_setg(errp
, "Postcopy must be started after migration has been"
1128 * we don't error if migration has finished since that would be racy
1129 * with issuing this command.
1131 qatomic_set(&s
->start_postcopy
, true);
1134 /* shared migration helpers */
1136 void migrate_set_state(int *state
, int old_state
, int new_state
)
1138 assert(new_state
< MIGRATION_STATUS__MAX
);
1139 if (qatomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1140 trace_migrate_set_state(MigrationStatus_str(new_state
));
1141 migrate_generate_event(new_state
);
1145 static void migrate_fd_cleanup(MigrationState
*s
)
1147 qemu_bh_delete(s
->cleanup_bh
);
1148 s
->cleanup_bh
= NULL
;
1150 g_free(s
->hostname
);
1152 json_writer_free(s
->vmdesc
);
1155 qemu_savevm_state_cleanup();
1157 if (s
->to_dst_file
) {
1160 trace_migrate_fd_cleanup();
1161 qemu_mutex_unlock_iothread();
1162 if (s
->migration_thread_running
) {
1163 qemu_thread_join(&s
->thread
);
1164 s
->migration_thread_running
= false;
1166 qemu_mutex_lock_iothread();
1168 multifd_save_cleanup();
1169 qemu_mutex_lock(&s
->qemu_file_lock
);
1170 tmp
= s
->to_dst_file
;
1171 s
->to_dst_file
= NULL
;
1172 qemu_mutex_unlock(&s
->qemu_file_lock
);
1174 * Close the file handle without the lock to make sure the
1175 * critical section won't block for long.
1177 migration_ioc_unregister_yank_from_file(tmp
);
1181 if (s
->postcopy_qemufile_src
) {
1182 migration_ioc_unregister_yank_from_file(s
->postcopy_qemufile_src
);
1183 qemu_fclose(s
->postcopy_qemufile_src
);
1184 s
->postcopy_qemufile_src
= NULL
;
1187 assert(!migration_is_active(s
));
1189 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1190 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1191 MIGRATION_STATUS_CANCELLED
);
1195 /* It is used on info migrate. We can't free it */
1196 error_report_err(error_copy(s
->error
));
1198 notifier_list_notify(&migration_state_notifiers
, s
);
1199 block_cleanup_parameters();
1200 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1203 static void migrate_fd_cleanup_schedule(MigrationState
*s
)
1206 * Ref the state for bh, because it may be called when
1207 * there're already no other refs
1209 object_ref(OBJECT(s
));
1210 qemu_bh_schedule(s
->cleanup_bh
);
1213 static void migrate_fd_cleanup_bh(void *opaque
)
1215 MigrationState
*s
= opaque
;
1216 migrate_fd_cleanup(s
);
1217 object_unref(OBJECT(s
));
1220 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1222 QEMU_LOCK_GUARD(&s
->error_mutex
);
1224 s
->error
= error_copy(error
);
1228 static void migrate_error_free(MigrationState
*s
)
1230 QEMU_LOCK_GUARD(&s
->error_mutex
);
1232 error_free(s
->error
);
1237 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1239 trace_migrate_fd_error(error_get_pretty(error
));
1240 assert(s
->to_dst_file
== NULL
);
1241 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1242 MIGRATION_STATUS_FAILED
);
1243 migrate_set_error(s
, error
);
1246 static void migrate_fd_cancel(MigrationState
*s
)
1249 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1250 trace_migrate_fd_cancel();
1252 WITH_QEMU_LOCK_GUARD(&s
->qemu_file_lock
) {
1253 if (s
->rp_state
.from_dst_file
) {
1254 /* shutdown the rp socket, so causing the rp thread to shutdown */
1255 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1260 old_state
= s
->state
;
1261 if (!migration_is_running(old_state
)) {
1264 /* If the migration is paused, kick it out of the pause */
1265 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1266 qemu_sem_post(&s
->pause_sem
);
1268 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1269 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1272 * If we're unlucky the migration code might be stuck somewhere in a
1273 * send/write while the network has failed and is waiting to timeout;
1274 * if we've got shutdown(2) available then we can force it to quit.
1275 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1276 * called in a bh, so there is no race against this cancel.
1278 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1279 qemu_file_shutdown(f
);
1281 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1282 Error
*local_err
= NULL
;
1284 bdrv_activate_all(&local_err
);
1286 error_report_err(local_err
);
1288 s
->block_inactive
= false;
1293 void add_migration_state_change_notifier(Notifier
*notify
)
1295 notifier_list_add(&migration_state_notifiers
, notify
);
1298 void remove_migration_state_change_notifier(Notifier
*notify
)
1300 notifier_remove(notify
);
1303 bool migration_in_setup(MigrationState
*s
)
1305 return s
->state
== MIGRATION_STATUS_SETUP
;
1308 bool migration_has_finished(MigrationState
*s
)
1310 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1313 bool migration_has_failed(MigrationState
*s
)
1315 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1316 s
->state
== MIGRATION_STATUS_FAILED
);
1319 bool migration_in_postcopy(void)
1321 MigrationState
*s
= migrate_get_current();
1324 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1325 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1326 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1333 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1335 return migration_in_postcopy() && s
->postcopy_after_devices
;
1338 bool migration_in_incoming_postcopy(void)
1340 PostcopyState ps
= postcopy_state_get();
1342 return ps
>= POSTCOPY_INCOMING_DISCARD
&& ps
< POSTCOPY_INCOMING_END
;
1345 bool migration_incoming_postcopy_advised(void)
1347 PostcopyState ps
= postcopy_state_get();
1349 return ps
>= POSTCOPY_INCOMING_ADVISE
&& ps
< POSTCOPY_INCOMING_END
;
1352 bool migration_in_bg_snapshot(void)
1354 MigrationState
*s
= migrate_get_current();
1356 return migrate_background_snapshot() &&
1357 migration_is_setup_or_active(s
->state
);
1360 bool migration_is_idle(void)
1362 MigrationState
*s
= current_migration
;
1369 case MIGRATION_STATUS_NONE
:
1370 case MIGRATION_STATUS_CANCELLED
:
1371 case MIGRATION_STATUS_COMPLETED
:
1372 case MIGRATION_STATUS_FAILED
:
1374 case MIGRATION_STATUS_SETUP
:
1375 case MIGRATION_STATUS_CANCELLING
:
1376 case MIGRATION_STATUS_ACTIVE
:
1377 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1378 case MIGRATION_STATUS_COLO
:
1379 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1380 case MIGRATION_STATUS_DEVICE
:
1381 case MIGRATION_STATUS_WAIT_UNPLUG
:
1383 case MIGRATION_STATUS__MAX
:
1384 g_assert_not_reached();
1390 bool migration_is_active(MigrationState
*s
)
1392 return (s
->state
== MIGRATION_STATUS_ACTIVE
||
1393 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1396 void migrate_init(MigrationState
*s
)
1399 * Reinitialise all migration state, except
1400 * parameters/capabilities that the user set, and
1405 s
->to_dst_file
= NULL
;
1406 s
->state
= MIGRATION_STATUS_NONE
;
1407 s
->rp_state
.from_dst_file
= NULL
;
1408 s
->rp_state
.error
= false;
1410 s
->pages_per_second
= 0.0;
1412 s
->expected_downtime
= 0;
1414 s
->start_postcopy
= false;
1415 s
->postcopy_after_devices
= false;
1416 s
->migration_thread_running
= false;
1417 error_free(s
->error
);
1421 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1423 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1425 s
->vm_was_running
= false;
1426 s
->iteration_initial_bytes
= 0;
1427 s
->threshold_size
= 0;
1430 int migrate_add_blocker_internal(Error
*reason
, Error
**errp
)
1432 /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */
1433 if (runstate_check(RUN_STATE_SAVE_VM
) || !migration_is_idle()) {
1434 error_propagate_prepend(errp
, error_copy(reason
),
1435 "disallowing migration blocker "
1436 "(migration/snapshot in progress) for: ");
1440 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1444 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1446 if (only_migratable
) {
1447 error_propagate_prepend(errp
, error_copy(reason
),
1448 "disallowing migration blocker "
1449 "(--only-migratable) for: ");
1453 return migrate_add_blocker_internal(reason
, errp
);
1456 void migrate_del_blocker(Error
*reason
)
1458 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1461 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1463 Error
*local_err
= NULL
;
1464 static bool once
= true;
1467 error_setg(errp
, "The incoming migration has already been started");
1470 if (!runstate_check(RUN_STATE_INMIGRATE
)) {
1471 error_setg(errp
, "'-incoming' was not specified on the command line");
1475 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
1479 qemu_start_incoming_migration(uri
, &local_err
);
1482 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1483 error_propagate(errp
, local_err
);
1490 void qmp_migrate_recover(const char *uri
, Error
**errp
)
1492 MigrationIncomingState
*mis
= migration_incoming_get_current();
1495 * Don't even bother to use ERRP_GUARD() as it _must_ always be set by
1496 * callers (no one should ignore a recover failure); if there is, it's a
1497 * programming error.
1501 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1502 error_setg(errp
, "Migrate recover can only be run "
1503 "when postcopy is paused.");
1507 /* If there's an existing transport, release it */
1508 migration_incoming_transport_cleanup(mis
);
1511 * Note that this call will never start a real migration; it will
1512 * only re-setup the migration stream and poke existing migration
1513 * to continue using that newly established channel.
1515 qemu_start_incoming_migration(uri
, errp
);
1518 void qmp_migrate_pause(Error
**errp
)
1520 MigrationState
*ms
= migrate_get_current();
1521 MigrationIncomingState
*mis
= migration_incoming_get_current();
1524 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1525 /* Source side, during postcopy */
1526 qemu_mutex_lock(&ms
->qemu_file_lock
);
1527 ret
= qemu_file_shutdown(ms
->to_dst_file
);
1528 qemu_mutex_unlock(&ms
->qemu_file_lock
);
1530 error_setg(errp
, "Failed to pause source migration");
1535 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1536 ret
= qemu_file_shutdown(mis
->from_src_file
);
1538 error_setg(errp
, "Failed to pause destination migration");
1543 error_setg(errp
, "migrate-pause is currently only supported "
1544 "during postcopy-active state");
1547 bool migration_is_blocked(Error
**errp
)
1549 if (qemu_savevm_state_blocked(errp
)) {
1553 if (migration_blockers
) {
1554 error_propagate(errp
, error_copy(migration_blockers
->data
));
1561 /* Returns true if continue to migrate, or false if error detected */
1562 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
1563 bool resume
, Error
**errp
)
1565 Error
*local_err
= NULL
;
1568 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1569 error_setg(errp
, "Cannot resume if there is no "
1570 "paused migration");
1575 * Postcopy recovery won't work well with release-ram
1576 * capability since release-ram will drop the page buffer as
1577 * long as the page is put into the send buffer. So if there
1578 * is a network failure happened, any page buffers that have
1579 * not yet reached the destination VM but have already been
1580 * sent from the source VM will be lost forever. Let's refuse
1581 * the client from resuming such a postcopy migration.
1582 * Luckily release-ram was designed to only be used when src
1583 * and destination VMs are on the same host, so it should be
1586 if (migrate_release_ram()) {
1587 error_setg(errp
, "Postcopy recovery cannot work "
1588 "when release-ram capability is set");
1592 /* This is a resume, skip init status */
1596 if (migration_is_running(s
->state
)) {
1597 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1601 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1602 error_setg(errp
, "Guest is waiting for an incoming migration");
1606 if (runstate_check(RUN_STATE_POSTMIGRATE
)) {
1607 error_setg(errp
, "Can't migrate the vm that was paused due to "
1608 "previous migration");
1612 if (migration_is_blocked(errp
)) {
1616 if (blk
|| blk_inc
) {
1617 if (migrate_colo()) {
1618 error_setg(errp
, "No disk migration is required in COLO mode");
1621 if (migrate_block() || migrate_block_incremental()) {
1622 error_setg(errp
, "Command options are incompatible with "
1623 "current migration capabilities");
1626 if (!migrate_cap_set(MIGRATION_CAPABILITY_BLOCK
, true, &local_err
)) {
1627 error_propagate(errp
, local_err
);
1630 s
->must_remove_block_options
= true;
1634 migrate_set_block_incremental(true);
1639 * set mig_stats compression_counters memory to zero for a
1642 memset(&mig_stats
, 0, sizeof(mig_stats
));
1643 memset(&compression_counters
, 0, sizeof(compression_counters
));
1648 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1649 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1650 bool has_resume
, bool resume
, Error
**errp
)
1652 Error
*local_err
= NULL
;
1653 MigrationState
*s
= migrate_get_current();
1654 const char *p
= NULL
;
1656 /* URI is not suitable for migration? */
1657 if (!migration_channels_and_uri_compatible(uri
, errp
)) {
1661 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
1662 has_resume
&& resume
, errp
)) {
1663 /* Error detected, put into errp */
1667 if (!(has_resume
&& resume
)) {
1668 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
1673 if (strstart(uri
, "tcp:", &p
) ||
1674 strstart(uri
, "unix:", NULL
) ||
1675 strstart(uri
, "vsock:", NULL
)) {
1676 socket_start_outgoing_migration(s
, p
? p
: uri
, &local_err
);
1678 } else if (strstart(uri
, "rdma:", &p
)) {
1679 rdma_start_outgoing_migration(s
, p
, &local_err
);
1681 } else if (strstart(uri
, "exec:", &p
)) {
1682 exec_start_outgoing_migration(s
, p
, &local_err
);
1683 } else if (strstart(uri
, "fd:", &p
)) {
1684 fd_start_outgoing_migration(s
, p
, &local_err
);
1686 if (!(has_resume
&& resume
)) {
1687 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1689 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1690 "a valid migration protocol");
1691 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1692 MIGRATION_STATUS_FAILED
);
1693 block_cleanup_parameters();
1698 if (!(has_resume
&& resume
)) {
1699 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1701 migrate_fd_error(s
, local_err
);
1702 error_propagate(errp
, local_err
);
1707 void qmp_migrate_cancel(Error
**errp
)
1709 migration_cancel(NULL
);
1712 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
1714 MigrationState
*s
= migrate_get_current();
1715 if (s
->state
!= state
) {
1716 error_setg(errp
, "Migration not in expected state: %s",
1717 MigrationStatus_str(s
->state
));
1720 qemu_sem_post(&s
->pause_sem
);
1723 /* migration thread support */
1725 * Something bad happened to the RP stream, mark an error
1726 * The caller shall print or trace something to indicate why
1728 static void mark_source_rp_bad(MigrationState
*s
)
1730 s
->rp_state
.error
= true;
1733 static struct rp_cmd_args
{
1734 ssize_t len
; /* -1 = variable */
1737 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1738 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1739 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1740 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1741 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1742 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
1743 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
1744 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1748 * Process a request for pages received on the return path,
1749 * We're allowed to send more than requested (e.g. to round to our page size)
1750 * and we don't need to send pages that have already been sent.
1752 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1753 ram_addr_t start
, size_t len
)
1755 long our_host_ps
= qemu_real_host_page_size();
1757 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1760 * Since we currently insist on matching page sizes, just sanity check
1761 * we're being asked for whole host pages.
1763 if (!QEMU_IS_ALIGNED(start
, our_host_ps
) ||
1764 !QEMU_IS_ALIGNED(len
, our_host_ps
)) {
1765 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1766 " len: %zd", __func__
, start
, len
);
1767 mark_source_rp_bad(ms
);
1771 if (ram_save_queue_pages(rbname
, start
, len
)) {
1772 mark_source_rp_bad(ms
);
1776 /* Return true to retry, false to quit */
1777 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
1779 trace_postcopy_pause_return_path();
1781 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
1783 trace_postcopy_pause_return_path_continued();
1788 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
1790 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
1793 error_report("%s: invalid block name '%s'", __func__
, block_name
);
1797 /* Fetch the received bitmap and refresh the dirty bitmap */
1798 return ram_dirty_bitmap_reload(s
, block
);
1801 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
1803 trace_source_return_path_thread_resume_ack(value
);
1805 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
1806 error_report("%s: illegal resume_ack value %"PRIu32
,
1811 /* Now both sides are active. */
1812 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
1813 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1815 /* Notify send thread that time to continue send pages */
1816 qemu_sem_post(&s
->rp_state
.rp_sem
);
1822 * Release ms->rp_state.from_dst_file (and postcopy_qemufile_src if
1823 * existed) in a safe way.
1825 static void migration_release_dst_files(MigrationState
*ms
)
1829 WITH_QEMU_LOCK_GUARD(&ms
->qemu_file_lock
) {
1831 * Reset the from_dst_file pointer first before releasing it, as we
1832 * can't block within lock section
1834 file
= ms
->rp_state
.from_dst_file
;
1835 ms
->rp_state
.from_dst_file
= NULL
;
1839 * Do the same to postcopy fast path socket too if there is. No
1840 * locking needed because this qemufile should only be managed by
1841 * return path thread.
1843 if (ms
->postcopy_qemufile_src
) {
1844 migration_ioc_unregister_yank_from_file(ms
->postcopy_qemufile_src
);
1845 qemu_file_shutdown(ms
->postcopy_qemufile_src
);
1846 qemu_fclose(ms
->postcopy_qemufile_src
);
1847 ms
->postcopy_qemufile_src
= NULL
;
1854 * Handles messages sent on the return path towards the source VM
1857 static void *source_return_path_thread(void *opaque
)
1859 MigrationState
*ms
= opaque
;
1860 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1861 uint16_t header_len
, header_type
;
1863 uint32_t tmp32
, sibling_error
;
1864 ram_addr_t start
= 0; /* =0 to silence warning */
1865 size_t len
= 0, expected_len
;
1868 trace_source_return_path_thread_entry();
1869 rcu_register_thread();
1872 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1873 migration_is_setup_or_active(ms
->state
)) {
1874 trace_source_return_path_thread_loop_top();
1875 header_type
= qemu_get_be16(rp
);
1876 header_len
= qemu_get_be16(rp
);
1878 if (qemu_file_get_error(rp
)) {
1879 mark_source_rp_bad(ms
);
1883 if (header_type
>= MIG_RP_MSG_MAX
||
1884 header_type
== MIG_RP_MSG_INVALID
) {
1885 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1886 header_type
, header_len
);
1887 mark_source_rp_bad(ms
);
1891 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1892 header_len
!= rp_cmd_args
[header_type
].len
) ||
1893 header_len
> sizeof(buf
)) {
1894 error_report("RP: Received '%s' message (0x%04x) with"
1895 "incorrect length %d expecting %zu",
1896 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1897 (size_t)rp_cmd_args
[header_type
].len
);
1898 mark_source_rp_bad(ms
);
1902 /* We know we've got a valid header by this point */
1903 res
= qemu_get_buffer(rp
, buf
, header_len
);
1904 if (res
!= header_len
) {
1905 error_report("RP: Failed reading data for message 0x%04x"
1906 " read %d expected %d",
1907 header_type
, res
, header_len
);
1908 mark_source_rp_bad(ms
);
1912 /* OK, we have the message and the data */
1913 switch (header_type
) {
1914 case MIG_RP_MSG_SHUT
:
1915 sibling_error
= ldl_be_p(buf
);
1916 trace_source_return_path_thread_shut(sibling_error
);
1917 if (sibling_error
) {
1918 error_report("RP: Sibling indicated error %d", sibling_error
);
1919 mark_source_rp_bad(ms
);
1922 * We'll let the main thread deal with closing the RP
1923 * we could do a shutdown(2) on it, but we're the only user
1924 * anyway, so there's nothing gained.
1928 case MIG_RP_MSG_PONG
:
1929 tmp32
= ldl_be_p(buf
);
1930 trace_source_return_path_thread_pong(tmp32
);
1931 qemu_sem_post(&ms
->rp_state
.rp_pong_acks
);
1934 case MIG_RP_MSG_REQ_PAGES
:
1935 start
= ldq_be_p(buf
);
1936 len
= ldl_be_p(buf
+ 8);
1937 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1940 case MIG_RP_MSG_REQ_PAGES_ID
:
1941 expected_len
= 12 + 1; /* header + termination */
1943 if (header_len
>= expected_len
) {
1944 start
= ldq_be_p(buf
);
1945 len
= ldl_be_p(buf
+ 8);
1946 /* Now we expect an idstr */
1947 tmp32
= buf
[12]; /* Length of the following idstr */
1948 buf
[13 + tmp32
] = '\0';
1949 expected_len
+= tmp32
;
1951 if (header_len
!= expected_len
) {
1952 error_report("RP: Req_Page_id with length %d expecting %zd",
1953 header_len
, expected_len
);
1954 mark_source_rp_bad(ms
);
1957 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
1960 case MIG_RP_MSG_RECV_BITMAP
:
1961 if (header_len
< 1) {
1962 error_report("%s: missing block name", __func__
);
1963 mark_source_rp_bad(ms
);
1966 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
1967 buf
[buf
[0] + 1] = '\0';
1968 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
1969 mark_source_rp_bad(ms
);
1974 case MIG_RP_MSG_RESUME_ACK
:
1975 tmp32
= ldl_be_p(buf
);
1976 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
1977 mark_source_rp_bad(ms
);
1988 res
= qemu_file_get_error(rp
);
1990 if (res
&& migration_in_postcopy()) {
1992 * Maybe there is something we can do: it looks like a
1993 * network down issue, and we pause for a recovery.
1995 migration_release_dst_files(ms
);
1997 if (postcopy_pause_return_path_thread(ms
)) {
1999 * Reload rp, reset the rest. Referencing it is safe since
2000 * it's reset only by us above, or when migration completes
2002 rp
= ms
->rp_state
.from_dst_file
;
2003 ms
->rp_state
.error
= false;
2008 trace_source_return_path_thread_bad_end();
2009 mark_source_rp_bad(ms
);
2012 trace_source_return_path_thread_end();
2013 migration_release_dst_files(ms
);
2014 rcu_unregister_thread();
2018 static int open_return_path_on_source(MigrationState
*ms
,
2021 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
2022 if (!ms
->rp_state
.from_dst_file
) {
2026 trace_open_return_path_on_source();
2028 if (!create_thread
) {
2033 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
2034 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
2035 ms
->rp_state
.rp_thread_created
= true;
2037 trace_open_return_path_on_source_continue();
2042 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2043 static int await_return_path_close_on_source(MigrationState
*ms
)
2046 * If this is a normal exit then the destination will send a SHUT and the
2047 * rp_thread will exit, however if there's an error we need to cause
2050 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
2052 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2053 * waiting for the destination.
2055 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
2056 mark_source_rp_bad(ms
);
2058 trace_await_return_path_close_on_source_joining();
2059 qemu_thread_join(&ms
->rp_state
.rp_thread
);
2060 ms
->rp_state
.rp_thread_created
= false;
2061 trace_await_return_path_close_on_source_close();
2062 return ms
->rp_state
.error
;
2066 migration_wait_main_channel(MigrationState
*ms
)
2068 /* Wait until one PONG message received */
2069 qemu_sem_wait(&ms
->rp_state
.rp_pong_acks
);
2073 * Switch from normal iteration to postcopy
2074 * Returns non-0 on error
2076 static int postcopy_start(MigrationState
*ms
)
2079 QIOChannelBuffer
*bioc
;
2081 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2082 uint64_t bandwidth
= migrate_max_postcopy_bandwidth();
2083 bool restart_block
= false;
2084 int cur_state
= MIGRATION_STATUS_ACTIVE
;
2086 if (migrate_postcopy_preempt()) {
2087 migration_wait_main_channel(ms
);
2088 if (postcopy_preempt_establish_channel(ms
)) {
2089 migrate_set_state(&ms
->state
, ms
->state
, MIGRATION_STATUS_FAILED
);
2094 if (!migrate_pause_before_switchover()) {
2095 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
2096 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2099 trace_postcopy_start();
2100 qemu_mutex_lock_iothread();
2101 trace_postcopy_start_set_run();
2103 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2104 global_state_store();
2105 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2110 ret
= migration_maybe_pause(ms
, &cur_state
,
2111 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2116 ret
= bdrv_inactivate_all();
2120 restart_block
= true;
2123 * Cause any non-postcopiable, but iterative devices to
2124 * send out their final data.
2126 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2129 * in Finish migrate and with the io-lock held everything should
2130 * be quiet, but we've potentially still got dirty pages and we
2131 * need to tell the destination to throw any pages it's already received
2134 if (migrate_postcopy_ram()) {
2135 ram_postcopy_send_discard_bitmap(ms
);
2139 * send rest of state - note things that are doing postcopy
2140 * will notice we're in POSTCOPY_ACTIVE and not actually
2141 * wrap their state up here
2143 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
);
2144 if (migrate_postcopy_ram()) {
2145 /* Ping just for debugging, helps line traces up */
2146 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
2150 * While loading the device state we may trigger page transfer
2151 * requests and the fd must be free to process those, and thus
2152 * the destination must read the whole device state off the fd before
2153 * it starts processing it. Unfortunately the ad-hoc migration format
2154 * doesn't allow the destination to know the size to read without fully
2155 * parsing it through each devices load-state code (especially the open
2156 * coded devices that use get/put).
2157 * So we wrap the device state up in a package with a length at the start;
2158 * to do this we use a qemu_buf to hold the whole of the device state.
2160 bioc
= qio_channel_buffer_new(4096);
2161 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
2162 fb
= qemu_file_new_output(QIO_CHANNEL(bioc
));
2163 object_unref(OBJECT(bioc
));
2166 * Make sure the receiver can get incoming pages before we send the rest
2169 qemu_savevm_send_postcopy_listen(fb
);
2171 qemu_savevm_state_complete_precopy(fb
, false, false);
2172 if (migrate_postcopy_ram()) {
2173 qemu_savevm_send_ping(fb
, 3);
2176 qemu_savevm_send_postcopy_run(fb
);
2178 /* <><> end of stuff going into the package */
2180 /* Last point of recovery; as soon as we send the package the destination
2181 * can open devices and potentially start running.
2182 * Lets just check again we've not got any errors.
2184 ret
= qemu_file_get_error(ms
->to_dst_file
);
2186 error_report("postcopy_start: Migration stream errored (pre package)");
2190 restart_block
= false;
2192 /* Now send that blob */
2193 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2198 /* Send a notify to give a chance for anything that needs to happen
2199 * at the transition to postcopy and after the device state; in particular
2200 * spice needs to trigger a transition now
2202 ms
->postcopy_after_devices
= true;
2203 notifier_list_notify(&migration_state_notifiers
, ms
);
2205 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2207 qemu_mutex_unlock_iothread();
2209 if (migrate_postcopy_ram()) {
2211 * Although this ping is just for debug, it could potentially be
2212 * used for getting a better measurement of downtime at the source.
2214 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2217 if (migrate_release_ram()) {
2218 ram_postcopy_migrated_memory_release(ms
);
2221 ret
= qemu_file_get_error(ms
->to_dst_file
);
2223 error_report("postcopy_start: Migration stream errored");
2224 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2225 MIGRATION_STATUS_FAILED
);
2228 trace_postcopy_preempt_enabled(migrate_postcopy_preempt());
2235 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2236 MIGRATION_STATUS_FAILED
);
2237 if (restart_block
) {
2238 /* A failure happened early enough that we know the destination hasn't
2239 * accessed block devices, so we're safe to recover.
2241 Error
*local_err
= NULL
;
2243 bdrv_activate_all(&local_err
);
2245 error_report_err(local_err
);
2248 qemu_mutex_unlock_iothread();
2253 * migration_maybe_pause: Pause if required to by
2254 * migrate_pause_before_switchover called with the iothread locked
2255 * Returns: 0 on success
2257 static int migration_maybe_pause(MigrationState
*s
,
2258 int *current_active_state
,
2261 if (!migrate_pause_before_switchover()) {
2265 /* Since leaving this state is not atomic with posting the semaphore
2266 * it's possible that someone could have issued multiple migrate_continue
2267 * and the semaphore is incorrectly positive at this point;
2268 * the docs say it's undefined to reinit a semaphore that's already
2269 * init'd, so use timedwait to eat up any existing posts.
2271 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2272 /* This block intentionally left blank */
2276 * If the migration is cancelled when it is in the completion phase,
2277 * the migration state is set to MIGRATION_STATUS_CANCELLING.
2278 * So we don't need to wait a semaphore, otherwise we would always
2279 * wait for the 'pause_sem' semaphore.
2281 if (s
->state
!= MIGRATION_STATUS_CANCELLING
) {
2282 qemu_mutex_unlock_iothread();
2283 migrate_set_state(&s
->state
, *current_active_state
,
2284 MIGRATION_STATUS_PRE_SWITCHOVER
);
2285 qemu_sem_wait(&s
->pause_sem
);
2286 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2288 *current_active_state
= new_state
;
2289 qemu_mutex_lock_iothread();
2292 return s
->state
== new_state
? 0 : -EINVAL
;
2296 * migration_completion: Used by migration_thread when there's not much left.
2297 * The caller 'breaks' the loop when this returns.
2299 * @s: Current migration state
2301 static void migration_completion(MigrationState
*s
)
2304 int current_active_state
= s
->state
;
2306 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2307 qemu_mutex_lock_iothread();
2308 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2309 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2310 s
->vm_was_running
= runstate_is_running();
2311 ret
= global_state_store();
2314 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2315 trace_migration_completion_vm_stop(ret
);
2317 ret
= migration_maybe_pause(s
, ¤t_active_state
,
2318 MIGRATION_STATUS_DEVICE
);
2322 * Inactivate disks except in COLO, and track that we
2323 * have done so in order to remember to reactivate
2324 * them if migration fails or is cancelled.
2326 s
->block_inactive
= !migrate_colo();
2327 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
2328 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
2332 qemu_mutex_unlock_iothread();
2337 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2338 trace_migration_completion_postcopy_end();
2340 qemu_mutex_lock_iothread();
2341 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
2342 qemu_mutex_unlock_iothread();
2345 * Shutdown the postcopy fast path thread. This is only needed
2346 * when dest QEMU binary is old (7.1/7.2). QEMU 8.0+ doesn't need
2349 if (migrate_postcopy_preempt() && s
->preempt_pre_7_2
) {
2350 postcopy_preempt_shutdown_file(s
);
2353 trace_migration_completion_postcopy_end_after_complete();
2359 * If rp was opened we must clean up the thread before
2360 * cleaning everything else up (since if there are no failures
2361 * it will wait for the destination to send it's status in
2364 if (s
->rp_state
.rp_thread_created
) {
2366 trace_migration_return_path_end_before();
2367 rp_error
= await_return_path_close_on_source(s
);
2368 trace_migration_return_path_end_after(rp_error
);
2374 if (qemu_file_get_error(s
->to_dst_file
)) {
2375 trace_migration_completion_file_err();
2379 if (migrate_colo() && s
->state
== MIGRATION_STATUS_ACTIVE
) {
2380 /* COLO does not support postcopy */
2381 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
2382 MIGRATION_STATUS_COLO
);
2384 migrate_set_state(&s
->state
, current_active_state
,
2385 MIGRATION_STATUS_COMPLETED
);
2391 if (s
->block_inactive
&& (s
->state
== MIGRATION_STATUS_ACTIVE
||
2392 s
->state
== MIGRATION_STATUS_DEVICE
)) {
2394 * If not doing postcopy, vm_start() will be called: let's
2395 * regain control on images.
2397 Error
*local_err
= NULL
;
2399 qemu_mutex_lock_iothread();
2400 bdrv_activate_all(&local_err
);
2402 error_report_err(local_err
);
2404 s
->block_inactive
= false;
2406 qemu_mutex_unlock_iothread();
2409 migrate_set_state(&s
->state
, current_active_state
,
2410 MIGRATION_STATUS_FAILED
);
2414 * bg_migration_completion: Used by bg_migration_thread when after all the
2415 * RAM has been saved. The caller 'breaks' the loop when this returns.
2417 * @s: Current migration state
2419 static void bg_migration_completion(MigrationState
*s
)
2421 int current_active_state
= s
->state
;
2424 * Stop tracking RAM writes - un-protect memory, un-register UFFD
2425 * memory ranges, flush kernel wait queues and wake up threads
2426 * waiting for write fault to be resolved.
2428 ram_write_tracking_stop();
2430 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2432 * By this moment we have RAM content saved into the migration stream.
2433 * The next step is to flush the non-RAM content (device state)
2434 * right after the ram content. The device state has been stored into
2435 * the temporary buffer before RAM saving started.
2437 qemu_put_buffer(s
->to_dst_file
, s
->bioc
->data
, s
->bioc
->usage
);
2438 qemu_fflush(s
->to_dst_file
);
2439 } else if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
2443 if (qemu_file_get_error(s
->to_dst_file
)) {
2444 trace_migration_completion_file_err();
2448 migrate_set_state(&s
->state
, current_active_state
,
2449 MIGRATION_STATUS_COMPLETED
);
2453 migrate_set_state(&s
->state
, current_active_state
,
2454 MIGRATION_STATUS_FAILED
);
2457 typedef enum MigThrError
{
2458 /* No error detected */
2459 MIG_THR_ERR_NONE
= 0,
2460 /* Detected error, but resumed successfully */
2461 MIG_THR_ERR_RECOVERED
= 1,
2462 /* Detected fatal error, need to exit */
2463 MIG_THR_ERR_FATAL
= 2,
2466 static int postcopy_resume_handshake(MigrationState
*s
)
2468 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
2470 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2471 qemu_sem_wait(&s
->rp_state
.rp_sem
);
2474 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2481 /* Return zero if success, or <0 for error */
2482 static int postcopy_do_resume(MigrationState
*s
)
2487 * Call all the resume_prepare() hooks, so that modules can be
2488 * ready for the migration resume.
2490 ret
= qemu_savevm_state_resume_prepare(s
);
2492 error_report("%s: resume_prepare() failure detected: %d",
2498 * If preempt is enabled, re-establish the preempt channel. Note that
2499 * we do it after resume prepare to make sure the main channel will be
2500 * created before the preempt channel. E.g. with weak network, the
2501 * dest QEMU may get messed up with the preempt and main channels on
2502 * the order of connection setup. This guarantees the correct order.
2504 ret
= postcopy_preempt_establish_channel(s
);
2506 error_report("%s: postcopy_preempt_establish_channel(): %d",
2512 * Last handshake with destination on the resume (destination will
2513 * switch to postcopy-active afterwards)
2515 ret
= postcopy_resume_handshake(s
);
2517 error_report("%s: handshake failed: %d", __func__
, ret
);
2525 * We don't return until we are in a safe state to continue current
2526 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2527 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2529 static MigThrError
postcopy_pause(MigrationState
*s
)
2531 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2537 * Current channel is possibly broken. Release it. Note that this is
2538 * guaranteed even without lock because to_dst_file should only be
2539 * modified by the migration thread. That also guarantees that the
2540 * unregister of yank is safe too without the lock. It should be safe
2541 * even to be within the qemu_file_lock, but we didn't do that to avoid
2542 * taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make
2543 * the qemu_file_lock critical section as small as possible.
2545 assert(s
->to_dst_file
);
2546 migration_ioc_unregister_yank_from_file(s
->to_dst_file
);
2547 qemu_mutex_lock(&s
->qemu_file_lock
);
2548 file
= s
->to_dst_file
;
2549 s
->to_dst_file
= NULL
;
2550 qemu_mutex_unlock(&s
->qemu_file_lock
);
2552 qemu_file_shutdown(file
);
2555 migrate_set_state(&s
->state
, s
->state
,
2556 MIGRATION_STATUS_POSTCOPY_PAUSED
);
2558 error_report("Detected IO failure for postcopy. "
2559 "Migration paused.");
2562 * We wait until things fixed up. Then someone will setup the
2563 * status back for us.
2565 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2566 qemu_sem_wait(&s
->postcopy_pause_sem
);
2569 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2570 /* Woken up by a recover procedure. Give it a shot */
2573 * Firstly, let's wake up the return path now, with a new
2574 * return path channel.
2576 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
2578 /* Do the resume logic */
2579 if (postcopy_do_resume(s
) == 0) {
2580 /* Let's continue! */
2581 trace_postcopy_pause_continued();
2582 return MIG_THR_ERR_RECOVERED
;
2585 * Something wrong happened during the recovery, let's
2586 * pause again. Pause is always better than throwing
2592 /* This is not right... Time to quit. */
2593 return MIG_THR_ERR_FATAL
;
2598 static MigThrError
migration_detect_error(MigrationState
*s
)
2601 int state
= s
->state
;
2602 Error
*local_error
= NULL
;
2604 if (state
== MIGRATION_STATUS_CANCELLING
||
2605 state
== MIGRATION_STATUS_CANCELLED
) {
2606 /* End the migration, but don't set the state to failed */
2607 return MIG_THR_ERR_FATAL
;
2611 * Try to detect any file errors. Note that postcopy_qemufile_src will
2612 * be NULL when postcopy preempt is not enabled.
2614 ret
= qemu_file_get_error_obj_any(s
->to_dst_file
,
2615 s
->postcopy_qemufile_src
,
2618 /* Everything is fine */
2619 assert(!local_error
);
2620 return MIG_THR_ERR_NONE
;
2624 migrate_set_error(s
, local_error
);
2625 error_free(local_error
);
2628 if (state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
) {
2630 * For postcopy, we allow the network to be down for a
2631 * while. After that, it can be continued by a
2634 return postcopy_pause(s
);
2637 * For precopy (or postcopy with error outside IO), we fail
2640 migrate_set_state(&s
->state
, state
, MIGRATION_STATUS_FAILED
);
2641 trace_migration_thread_file_err();
2643 /* Time to stop the migration, now. */
2644 return MIG_THR_ERR_FATAL
;
2648 /* How many bytes have we transferred since the beginning of the migration */
2649 static uint64_t migration_total_bytes(MigrationState
*s
)
2651 return qemu_file_transferred(s
->to_dst_file
) +
2652 stat64_get(&mig_stats
.multifd_bytes
);
2655 static void migration_calculate_complete(MigrationState
*s
)
2657 uint64_t bytes
= migration_total_bytes(s
);
2658 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2659 int64_t transfer_time
;
2661 s
->total_time
= end_time
- s
->start_time
;
2664 * It's still not set, so we are precopy migration. For
2665 * postcopy, downtime is calculated during postcopy_start().
2667 s
->downtime
= end_time
- s
->downtime_start
;
2670 transfer_time
= s
->total_time
- s
->setup_time
;
2671 if (transfer_time
) {
2672 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
2676 static void update_iteration_initial_status(MigrationState
*s
)
2679 * Update these three fields at the same time to avoid mismatch info lead
2680 * wrong speed calculation.
2682 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2683 s
->iteration_initial_bytes
= migration_total_bytes(s
);
2684 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
2687 static void migration_update_counters(MigrationState
*s
,
2688 int64_t current_time
)
2690 uint64_t transferred
, transferred_pages
, time_spent
;
2691 uint64_t current_bytes
; /* bytes transferred since the beginning */
2694 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
2698 current_bytes
= migration_total_bytes(s
);
2699 transferred
= current_bytes
- s
->iteration_initial_bytes
;
2700 time_spent
= current_time
- s
->iteration_start_time
;
2701 bandwidth
= (double)transferred
/ time_spent
;
2702 s
->threshold_size
= bandwidth
* migrate_downtime_limit();
2704 s
->mbps
= (((double) transferred
* 8.0) /
2705 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2707 transferred_pages
= ram_get_total_transferred_pages() -
2708 s
->iteration_initial_pages
;
2709 s
->pages_per_second
= (double) transferred_pages
/
2710 (((double) time_spent
/ 1000.0));
2713 * if we haven't sent anything, we don't want to
2714 * recalculate. 10000 is a small enough number for our purposes
2716 if (stat64_get(&mig_stats
.dirty_pages_rate
) &&
2717 transferred
> 10000) {
2718 s
->expected_downtime
=
2719 stat64_get(&mig_stats
.dirty_bytes_last_sync
) / bandwidth
;
2722 qemu_file_reset_rate_limit(s
->to_dst_file
);
2724 update_iteration_initial_status(s
);
2726 trace_migrate_transferred(transferred
, time_spent
,
2727 bandwidth
, s
->threshold_size
);
2730 /* Migration thread iteration status */
2732 MIG_ITERATE_RESUME
, /* Resume current iteration */
2733 MIG_ITERATE_SKIP
, /* Skip current iteration */
2734 MIG_ITERATE_BREAK
, /* Break the loop */
2738 * Return true if continue to the next iteration directly, false
2741 static MigIterateState
migration_iteration_run(MigrationState
*s
)
2743 uint64_t must_precopy
, can_postcopy
;
2744 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
2746 qemu_savevm_state_pending_estimate(&must_precopy
, &can_postcopy
);
2747 uint64_t pending_size
= must_precopy
+ can_postcopy
;
2749 trace_migrate_pending_estimate(pending_size
, must_precopy
, can_postcopy
);
2751 if (must_precopy
<= s
->threshold_size
) {
2752 qemu_savevm_state_pending_exact(&must_precopy
, &can_postcopy
);
2753 pending_size
= must_precopy
+ can_postcopy
;
2754 trace_migrate_pending_exact(pending_size
, must_precopy
, can_postcopy
);
2757 if (!pending_size
|| pending_size
< s
->threshold_size
) {
2758 trace_migration_thread_low_pending(pending_size
);
2759 migration_completion(s
);
2760 return MIG_ITERATE_BREAK
;
2763 /* Still a significant amount to transfer */
2764 if (!in_postcopy
&& must_precopy
<= s
->threshold_size
&&
2765 qatomic_read(&s
->start_postcopy
)) {
2766 if (postcopy_start(s
)) {
2767 error_report("%s: postcopy failed to start", __func__
);
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_was_running
= true;
2793 case MIGRATION_STATUS_FAILED
:
2794 case MIGRATION_STATUS_CANCELLED
:
2795 case MIGRATION_STATUS_CANCELLING
:
2796 if (s
->vm_was_running
) {
2797 if (!runstate_check(RUN_STATE_SHUTDOWN
)) {
2801 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2802 runstate_set(RUN_STATE_POSTMIGRATE
);
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
)
2818 qemu_mutex_lock_iothread();
2820 case MIGRATION_STATUS_COMPLETED
:
2821 migration_calculate_complete(s
);
2824 case MIGRATION_STATUS_ACTIVE
:
2825 case MIGRATION_STATUS_FAILED
:
2826 case MIGRATION_STATUS_CANCELLED
:
2827 case MIGRATION_STATUS_CANCELLING
:
2831 /* Should not reach here, but if so, forgive the VM. */
2832 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
2836 migrate_fd_cleanup_schedule(s
);
2837 qemu_mutex_unlock_iothread();
2841 * Return true if continue to the next iteration directly, false
2844 static MigIterateState
bg_migration_iteration_run(MigrationState
*s
)
2848 res
= qemu_savevm_state_iterate(s
->to_dst_file
, false);
2850 bg_migration_completion(s
);
2851 return MIG_ITERATE_BREAK
;
2854 return MIG_ITERATE_RESUME
;
2857 void migration_make_urgent_request(void)
2859 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
2862 void migration_consume_urgent_request(void)
2864 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
2867 /* Returns true if the rate limiting was broken by an urgent request */
2868 bool migration_rate_limit(void)
2870 int64_t now
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2871 MigrationState
*s
= migrate_get_current();
2873 bool urgent
= false;
2874 migration_update_counters(s
, now
);
2875 if (qemu_file_rate_limit(s
->to_dst_file
)) {
2877 if (qemu_file_get_error(s
->to_dst_file
)) {
2881 * Wait for a delay to do rate limiting OR
2882 * something urgent to post the semaphore.
2884 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- now
;
2885 trace_migration_rate_limit_pre(ms
);
2886 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
2888 * We were woken by one or more urgent things but
2889 * the timedwait will have consumed one of them.
2890 * The service routine for the urgent wake will dec
2891 * the semaphore itself for each item it consumes,
2892 * so add this one we just eat back.
2894 qemu_sem_post(&s
->rate_limit_sem
);
2897 trace_migration_rate_limit_post(urgent
);
2903 * if failover devices are present, wait they are completely
2907 static void qemu_savevm_wait_unplug(MigrationState
*s
, int old_state
,
2910 if (qemu_savevm_state_guest_unplug_pending()) {
2911 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_WAIT_UNPLUG
);
2913 while (s
->state
== MIGRATION_STATUS_WAIT_UNPLUG
&&
2914 qemu_savevm_state_guest_unplug_pending()) {
2915 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
2917 if (s
->state
!= MIGRATION_STATUS_WAIT_UNPLUG
) {
2918 int timeout
= 120; /* 30 seconds */
2920 * migration has been canceled
2921 * but as we have started an unplug we must wait the end
2922 * to be able to plug back the card
2924 while (timeout
-- && qemu_savevm_state_guest_unplug_pending()) {
2925 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
2927 if (qemu_savevm_state_guest_unplug_pending() &&
2929 warn_report("migration: partially unplugged device on "
2934 migrate_set_state(&s
->state
, MIGRATION_STATUS_WAIT_UNPLUG
, new_state
);
2936 migrate_set_state(&s
->state
, old_state
, new_state
);
2941 * Master migration thread on the source VM.
2942 * It drives the migration and pumps the data down the outgoing channel.
2944 static void *migration_thread(void *opaque
)
2946 MigrationState
*s
= opaque
;
2947 MigrationThread
*thread
= NULL
;
2948 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
2949 MigThrError thr_error
;
2950 bool urgent
= false;
2952 thread
= MigrationThreadAdd("live_migration", qemu_get_thread_id());
2954 rcu_register_thread();
2956 object_ref(OBJECT(s
));
2957 update_iteration_initial_status(s
);
2959 qemu_savevm_state_header(s
->to_dst_file
);
2962 * If we opened the return path, we need to make sure dst has it
2965 if (s
->rp_state
.rp_thread_created
) {
2966 /* Now tell the dest that it should open its end so it can reply */
2967 qemu_savevm_send_open_return_path(s
->to_dst_file
);
2969 /* And do a ping that will make stuff easier to debug */
2970 qemu_savevm_send_ping(s
->to_dst_file
, 1);
2973 if (migrate_postcopy()) {
2975 * Tell the destination that we *might* want to do postcopy later;
2976 * if the other end can't do postcopy it should fail now, nice and
2979 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
2982 if (migrate_colo()) {
2983 /* Notify migration destination that we enable COLO */
2984 qemu_savevm_send_colo_enable(s
->to_dst_file
);
2987 qemu_savevm_state_setup(s
->to_dst_file
);
2989 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
2990 MIGRATION_STATUS_ACTIVE
);
2992 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
2994 trace_migration_thread_setup_complete();
2996 while (migration_is_active(s
)) {
2997 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
2998 MigIterateState iter_state
= migration_iteration_run(s
);
2999 if (iter_state
== MIG_ITERATE_SKIP
) {
3001 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3007 * Try to detect any kind of failures, and see whether we
3008 * should stop the migration now.
3010 thr_error
= migration_detect_error(s
);
3011 if (thr_error
== MIG_THR_ERR_FATAL
) {
3012 /* Stop migration */
3014 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
3016 * Just recovered from a e.g. network failure, reset all
3017 * the local variables. This is important to avoid
3018 * breaking transferred_bytes and bandwidth calculation
3020 update_iteration_initial_status(s
);
3023 urgent
= migration_rate_limit();
3026 trace_migration_thread_after_loop();
3027 migration_iteration_finish(s
);
3028 object_unref(OBJECT(s
));
3029 rcu_unregister_thread();
3030 MigrationThreadDel(thread
);
3034 static void bg_migration_vm_start_bh(void *opaque
)
3036 MigrationState
*s
= opaque
;
3038 qemu_bh_delete(s
->vm_start_bh
);
3039 s
->vm_start_bh
= NULL
;
3042 s
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - s
->downtime_start
;
3046 * Background snapshot thread, based on live migration code.
3047 * This is an alternative implementation of live migration mechanism
3048 * introduced specifically to support background snapshots.
3050 * It takes advantage of userfault_fd write protection mechanism introduced
3051 * in v5.7 kernel. Compared to existing dirty page logging migration much
3052 * lesser stream traffic is produced resulting in smaller snapshot images,
3053 * simply cause of no page duplicates can get into the stream.
3055 * Another key point is that generated vmstate stream reflects machine state
3056 * 'frozen' at the beginning of snapshot creation compared to dirty page logging
3057 * mechanism, which effectively results in that saved snapshot is the state of VM
3058 * at the end of the process.
3060 static void *bg_migration_thread(void *opaque
)
3062 MigrationState
*s
= opaque
;
3063 int64_t setup_start
;
3064 MigThrError thr_error
;
3066 bool early_fail
= true;
3068 rcu_register_thread();
3069 object_ref(OBJECT(s
));
3071 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
3073 setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3075 * We want to save vmstate for the moment when migration has been
3076 * initiated but also we want to save RAM content while VM is running.
3077 * The RAM content should appear first in the vmstate. So, we first
3078 * stash the non-RAM part of the vmstate to the temporary buffer,
3079 * then write RAM part of the vmstate to the migration stream
3080 * with vCPUs running and, finally, write stashed non-RAM part of
3081 * the vmstate from the buffer to the migration stream.
3083 s
->bioc
= qio_channel_buffer_new(512 * 1024);
3084 qio_channel_set_name(QIO_CHANNEL(s
->bioc
), "vmstate-buffer");
3085 fb
= qemu_file_new_output(QIO_CHANNEL(s
->bioc
));
3086 object_unref(OBJECT(s
->bioc
));
3088 update_iteration_initial_status(s
);
3091 * Prepare for tracking memory writes with UFFD-WP - populate
3092 * RAM pages before protecting.
3095 ram_write_tracking_prepare();
3098 qemu_savevm_state_header(s
->to_dst_file
);
3099 qemu_savevm_state_setup(s
->to_dst_file
);
3101 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
3102 MIGRATION_STATUS_ACTIVE
);
3104 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3106 trace_migration_thread_setup_complete();
3107 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3109 qemu_mutex_lock_iothread();
3112 * If VM is currently in suspended state, then, to make a valid runstate
3113 * transition in vm_stop_force_state() we need to wakeup it up.
3115 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3116 s
->vm_was_running
= runstate_is_running();
3118 if (global_state_store()) {
3121 /* Forcibly stop VM before saving state of vCPUs and devices */
3122 if (vm_stop_force_state(RUN_STATE_PAUSED
)) {
3126 * Put vCPUs in sync with shadow context structures, then
3127 * save their state to channel-buffer along with devices.
3129 cpu_synchronize_all_states();
3130 if (qemu_savevm_state_complete_precopy_non_iterable(fb
, false, false)) {
3134 * Since we are going to get non-iterable state data directly
3135 * from s->bioc->data, explicit flush is needed here.
3139 /* Now initialize UFFD context and start tracking RAM writes */
3140 if (ram_write_tracking_start()) {
3146 * Start VM from BH handler to avoid write-fault lock here.
3147 * UFFD-WP protection for the whole RAM is already enabled so
3148 * calling VM state change notifiers from vm_start() would initiate
3149 * writes to virtio VQs memory which is in write-protected region.
3151 s
->vm_start_bh
= qemu_bh_new(bg_migration_vm_start_bh
, s
);
3152 qemu_bh_schedule(s
->vm_start_bh
);
3154 qemu_mutex_unlock_iothread();
3156 while (migration_is_active(s
)) {
3157 MigIterateState iter_state
= bg_migration_iteration_run(s
);
3158 if (iter_state
== MIG_ITERATE_SKIP
) {
3160 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3165 * Try to detect any kind of failures, and see whether we
3166 * should stop the migration now.
3168 thr_error
= migration_detect_error(s
);
3169 if (thr_error
== MIG_THR_ERR_FATAL
) {
3170 /* Stop migration */
3174 migration_update_counters(s
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
));
3177 trace_migration_thread_after_loop();
3181 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
3182 MIGRATION_STATUS_FAILED
);
3183 qemu_mutex_unlock_iothread();
3186 bg_migration_iteration_finish(s
);
3189 object_unref(OBJECT(s
));
3190 rcu_unregister_thread();
3195 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
3197 Error
*local_err
= NULL
;
3198 uint64_t rate_limit
;
3199 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
3202 * If there's a previous error, free it and prepare for another one.
3203 * Meanwhile if migration completes successfully, there won't have an error
3204 * dumped when calling migrate_fd_cleanup().
3206 migrate_error_free(s
);
3208 s
->expected_downtime
= migrate_downtime_limit();
3210 assert(s
->cleanup_bh
);
3212 assert(!s
->cleanup_bh
);
3213 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup_bh
, s
);
3216 migrate_fd_error(s
, error_in
);
3219 * Don't do cleanup for resume if channel is invalid, but only dump
3220 * the error. We wait for another channel connect from the user.
3221 * The error_report still gives HMP user a hint on what failed.
3222 * It's normally done in migrate_fd_cleanup(), but call it here
3225 error_report_err(error_copy(s
->error
));
3227 migrate_fd_cleanup(s
);
3233 /* This is a resumed migration */
3234 rate_limit
= migrate_max_postcopy_bandwidth();
3236 /* This is a fresh new migration */
3237 rate_limit
= migrate_max_bandwidth();
3239 /* Notify before starting migration thread */
3240 notifier_list_notify(&migration_state_notifiers
, s
);
3243 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
3244 qemu_file_set_blocking(s
->to_dst_file
, true);
3247 * Open the return path. For postcopy, it is used exclusively. For
3248 * precopy, only if user specified "return-path" capability would
3249 * QEMU uses the return path.
3251 if (migrate_postcopy_ram() || migrate_return_path()) {
3252 if (open_return_path_on_source(s
, !resume
)) {
3253 error_report("Unable to open return-path for postcopy");
3254 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
3255 migrate_fd_cleanup(s
);
3261 * This needs to be done before resuming a postcopy. Note: for newer
3262 * QEMUs we will delay the channel creation until postcopy_start(), to
3263 * avoid disorder of channel creations.
3265 if (migrate_postcopy_preempt() && s
->preempt_pre_7_2
) {
3266 postcopy_preempt_setup(s
);
3270 /* Wakeup the main migration thread to do the recovery */
3271 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
3272 MIGRATION_STATUS_POSTCOPY_RECOVER
);
3273 qemu_sem_post(&s
->postcopy_pause_sem
);
3277 if (multifd_save_setup(&local_err
) != 0) {
3278 error_report_err(local_err
);
3279 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3280 MIGRATION_STATUS_FAILED
);
3281 migrate_fd_cleanup(s
);
3285 if (migrate_background_snapshot()) {
3286 qemu_thread_create(&s
->thread
, "bg_snapshot",
3287 bg_migration_thread
, s
, QEMU_THREAD_JOINABLE
);
3289 qemu_thread_create(&s
->thread
, "live_migration",
3290 migration_thread
, s
, QEMU_THREAD_JOINABLE
);
3292 s
->migration_thread_running
= true;
3295 static void migration_class_init(ObjectClass
*klass
, void *data
)
3297 DeviceClass
*dc
= DEVICE_CLASS(klass
);
3299 dc
->user_creatable
= false;
3300 device_class_set_props(dc
, migration_properties
);
3303 static void migration_instance_finalize(Object
*obj
)
3305 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3307 qemu_mutex_destroy(&ms
->error_mutex
);
3308 qemu_mutex_destroy(&ms
->qemu_file_lock
);
3309 qemu_sem_destroy(&ms
->wait_unplug_sem
);
3310 qemu_sem_destroy(&ms
->rate_limit_sem
);
3311 qemu_sem_destroy(&ms
->pause_sem
);
3312 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
3313 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
3314 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
3315 qemu_sem_destroy(&ms
->rp_state
.rp_pong_acks
);
3316 qemu_sem_destroy(&ms
->postcopy_qemufile_src_sem
);
3317 error_free(ms
->error
);
3320 static void migration_instance_init(Object
*obj
)
3322 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3324 ms
->state
= MIGRATION_STATUS_NONE
;
3326 ms
->pages_per_second
= -1;
3327 qemu_sem_init(&ms
->pause_sem
, 0);
3328 qemu_mutex_init(&ms
->error_mutex
);
3330 migrate_params_init(&ms
->parameters
);
3332 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
3333 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
3334 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
3335 qemu_sem_init(&ms
->rp_state
.rp_pong_acks
, 0);
3336 qemu_sem_init(&ms
->rate_limit_sem
, 0);
3337 qemu_sem_init(&ms
->wait_unplug_sem
, 0);
3338 qemu_sem_init(&ms
->postcopy_qemufile_src_sem
, 0);
3339 qemu_mutex_init(&ms
->qemu_file_lock
);
3343 * Return true if check pass, false otherwise. Error will be put
3344 * inside errp if provided.
3346 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
3348 /* Assuming all off */
3349 bool old_caps
[MIGRATION_CAPABILITY__MAX
] = { 0 };
3351 if (!migrate_params_check(&ms
->parameters
, errp
)) {
3355 return migrate_caps_check(old_caps
, ms
->capabilities
, errp
);
3358 static const TypeInfo migration_type
= {
3359 .name
= TYPE_MIGRATION
,
3361 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3362 * not created using qdev_new(), it is not attached to the qdev
3363 * device tree, and it is never realized.
3365 * TODO: Make this TYPE_OBJECT once QOM provides something like
3366 * TYPE_DEVICE's "-global" properties.
3368 .parent
= TYPE_DEVICE
,
3369 .class_init
= migration_class_init
,
3370 .class_size
= sizeof(MigrationClass
),
3371 .instance_size
= sizeof(MigrationState
),
3372 .instance_init
= migration_instance_init
,
3373 .instance_finalize
= migration_instance_finalize
,
3376 static void register_migration_types(void)
3378 type_register_static(&migration_type
);
3381 type_init(register_migration_types
);