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();
515 assert(mis
->from_src_file
);
517 if (compress_threads_load_setup(mis
->from_src_file
)) {
518 error_report("Failed to setup decompress threads");
522 mis
->largest_page_size
= qemu_ram_pagesize_largest();
523 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
524 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
525 MIGRATION_STATUS_ACTIVE
);
527 mis
->loadvm_co
= qemu_coroutine_self();
528 ret
= qemu_loadvm_state(mis
->from_src_file
);
529 mis
->loadvm_co
= NULL
;
531 ps
= postcopy_state_get();
532 trace_process_incoming_migration_co_end(ret
, ps
);
533 if (ps
!= POSTCOPY_INCOMING_NONE
) {
534 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
536 * Where a migration had postcopy enabled (and thus went to advise)
537 * but managed to complete within the precopy period, we can use
540 postcopy_ram_incoming_cleanup(mis
);
541 } else if (ret
>= 0) {
543 * Postcopy was started, cleanup should happen at the end of the
546 trace_process_incoming_migration_co_postcopy_end_main();
549 /* Else if something went wrong then just fall out of the normal exit */
553 error_report("load of migration failed: %s", strerror(-ret
));
557 if (colo_incoming_co() < 0) {
561 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
562 qemu_bh_schedule(mis
->bh
);
565 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
566 MIGRATION_STATUS_FAILED
);
567 qemu_fclose(mis
->from_src_file
);
569 multifd_load_cleanup();
570 compress_threads_load_cleanup();
576 * migration_incoming_setup: Setup incoming migration
577 * @f: file for main migration channel
578 * @errp: where to put errors
580 * Returns: %true on success, %false on error.
582 static bool migration_incoming_setup(QEMUFile
*f
, Error
**errp
)
584 MigrationIncomingState
*mis
= migration_incoming_get_current();
586 if (!mis
->from_src_file
) {
587 mis
->from_src_file
= f
;
589 qemu_file_set_blocking(f
, false);
593 void migration_incoming_process(void)
595 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
596 qemu_coroutine_enter(co
);
599 /* Returns true if recovered from a paused migration, otherwise false */
600 static bool postcopy_try_recover(void)
602 MigrationIncomingState
*mis
= migration_incoming_get_current();
604 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
605 /* Resumed from a paused postcopy migration */
607 /* This should be set already in migration_incoming_setup() */
608 assert(mis
->from_src_file
);
609 /* Postcopy has standalone thread to do vm load */
610 qemu_file_set_blocking(mis
->from_src_file
, true);
612 /* Re-configure the return path */
613 mis
->to_src_file
= qemu_file_get_return_path(mis
->from_src_file
);
615 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
616 MIGRATION_STATUS_POSTCOPY_RECOVER
);
619 * Here, we only wake up the main loading thread (while the
620 * rest threads will still be waiting), so that we can receive
621 * commands from source now, and answer it if needed. The
622 * rest threads will be woken up afterwards until we are sure
623 * that source is ready to reply to page requests.
625 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
632 void migration_fd_process_incoming(QEMUFile
*f
, Error
**errp
)
634 if (!migration_incoming_setup(f
, errp
)) {
637 if (postcopy_try_recover()) {
640 migration_incoming_process();
644 * Returns true when we want to start a new incoming migration process,
647 static bool migration_should_start_incoming(bool main_channel
)
649 /* Multifd doesn't start unless all channels are established */
650 if (migrate_multifd()) {
651 return migration_has_all_channels();
654 /* Preempt channel only starts when the main channel is created */
655 if (migrate_postcopy_preempt()) {
660 * For all the rest types of migration, we should only reach here when
661 * it's the main channel that's being created, and we should always
662 * proceed with this channel.
664 assert(main_channel
);
668 void migration_ioc_process_incoming(QIOChannel
*ioc
, Error
**errp
)
670 MigrationIncomingState
*mis
= migration_incoming_get_current();
671 Error
*local_err
= NULL
;
673 bool default_channel
= true;
674 uint32_t channel_magic
= 0;
677 if (migrate_multifd() && !migrate_postcopy_ram() &&
678 qio_channel_has_feature(ioc
, QIO_CHANNEL_FEATURE_READ_MSG_PEEK
)) {
680 * With multiple channels, it is possible that we receive channels
681 * out of order on destination side, causing incorrect mapping of
682 * source channels on destination side. Check channel MAGIC to
683 * decide type of channel. Please note this is best effort, postcopy
684 * preempt channel does not send any magic number so avoid it for
685 * postcopy live migration. Also tls live migration already does
686 * tls handshake while initializing main channel so with tls this
687 * issue is not possible.
689 ret
= migration_channel_read_peek(ioc
, (void *)&channel_magic
,
690 sizeof(channel_magic
), &local_err
);
693 error_propagate(errp
, local_err
);
697 default_channel
= (channel_magic
== cpu_to_be32(QEMU_VM_FILE_MAGIC
));
699 default_channel
= !mis
->from_src_file
;
702 if (multifd_load_setup(errp
) != 0) {
703 error_setg(errp
, "Failed to setup multifd channels");
707 if (default_channel
) {
708 f
= qemu_file_new_input(ioc
);
710 if (!migration_incoming_setup(f
, errp
)) {
714 /* Multiple connections */
715 assert(migration_needs_multiple_sockets());
716 if (migrate_multifd()) {
717 multifd_recv_new_channel(ioc
, &local_err
);
719 assert(migrate_postcopy_preempt());
720 f
= qemu_file_new_input(ioc
);
721 postcopy_preempt_new_channel(mis
, f
);
724 error_propagate(errp
, local_err
);
729 if (migration_should_start_incoming(default_channel
)) {
730 /* If it's a recovery, we're done */
731 if (postcopy_try_recover()) {
734 migration_incoming_process();
739 * @migration_has_all_channels: We have received all channels that we need
741 * Returns true when we have got connections to all the channels that
742 * we need for migration.
744 bool migration_has_all_channels(void)
746 MigrationIncomingState
*mis
= migration_incoming_get_current();
748 if (!mis
->from_src_file
) {
752 if (migrate_multifd()) {
753 return multifd_recv_all_channels_created();
756 if (migrate_postcopy_preempt()) {
757 return mis
->postcopy_qemufile_dst
!= NULL
;
764 * Send a 'SHUT' message on the return channel with the given value
765 * to indicate that we've finished with the RP. Non-0 value indicates
768 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
773 buf
= cpu_to_be32(value
);
774 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
778 * Send a 'PONG' message on the return channel with the given value
779 * (normally in response to a 'PING')
781 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
786 buf
= cpu_to_be32(value
);
787 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
790 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
798 * First, we send the header part. It contains only the len of
799 * idstr, and the idstr itself.
801 len
= strlen(block_name
);
803 memcpy(buf
+ 1, block_name
, len
);
805 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
806 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
811 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
814 * Next, we dump the received bitmap to the stream.
816 * TODO: currently we are safe since we are the only one that is
817 * using the to_src_file handle (fault thread is still paused),
818 * and it's ok even not taking the mutex. However the best way is
819 * to take the lock before sending the message header, and release
820 * the lock after sending the bitmap.
822 qemu_mutex_lock(&mis
->rp_mutex
);
823 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
824 qemu_mutex_unlock(&mis
->rp_mutex
);
826 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
829 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
833 buf
= cpu_to_be32(value
);
834 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
838 * Return true if we're already in the middle of a migration
839 * (i.e. any of the active or setup states)
841 bool migration_is_setup_or_active(int state
)
844 case MIGRATION_STATUS_ACTIVE
:
845 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
846 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
847 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
848 case MIGRATION_STATUS_SETUP
:
849 case MIGRATION_STATUS_PRE_SWITCHOVER
:
850 case MIGRATION_STATUS_DEVICE
:
851 case MIGRATION_STATUS_WAIT_UNPLUG
:
852 case MIGRATION_STATUS_COLO
:
861 bool migration_is_running(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_CANCELLING
:
881 static bool migrate_show_downtime(MigrationState
*s
)
883 return (s
->state
== MIGRATION_STATUS_COMPLETED
) || migration_in_postcopy();
886 static void populate_time_info(MigrationInfo
*info
, MigrationState
*s
)
888 info
->has_status
= true;
889 info
->has_setup_time
= true;
890 info
->setup_time
= s
->setup_time
;
892 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
893 info
->has_total_time
= true;
894 info
->total_time
= s
->total_time
;
896 info
->has_total_time
= true;
897 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) -
901 if (migrate_show_downtime(s
)) {
902 info
->has_downtime
= true;
903 info
->downtime
= s
->downtime
;
905 info
->has_expected_downtime
= true;
906 info
->expected_downtime
= s
->expected_downtime
;
910 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
912 size_t page_size
= qemu_target_page_size();
914 info
->ram
= g_malloc0(sizeof(*info
->ram
));
915 info
->ram
->transferred
= stat64_get(&mig_stats
.transferred
);
916 info
->ram
->total
= ram_bytes_total();
917 info
->ram
->duplicate
= stat64_get(&mig_stats
.zero_pages
);
918 /* legacy value. It is not used anymore */
919 info
->ram
->skipped
= 0;
920 info
->ram
->normal
= stat64_get(&mig_stats
.normal_pages
);
921 info
->ram
->normal_bytes
= info
->ram
->normal
* page_size
;
922 info
->ram
->mbps
= s
->mbps
;
923 info
->ram
->dirty_sync_count
=
924 stat64_get(&mig_stats
.dirty_sync_count
);
925 info
->ram
->dirty_sync_missed_zero_copy
=
926 stat64_get(&mig_stats
.dirty_sync_missed_zero_copy
);
927 info
->ram
->postcopy_requests
=
928 stat64_get(&mig_stats
.postcopy_requests
);
929 info
->ram
->page_size
= page_size
;
930 info
->ram
->multifd_bytes
= stat64_get(&mig_stats
.multifd_bytes
);
931 info
->ram
->pages_per_second
= s
->pages_per_second
;
932 info
->ram
->precopy_bytes
= stat64_get(&mig_stats
.precopy_bytes
);
933 info
->ram
->downtime_bytes
= stat64_get(&mig_stats
.downtime_bytes
);
934 info
->ram
->postcopy_bytes
= stat64_get(&mig_stats
.postcopy_bytes
);
936 if (migrate_xbzrle()) {
937 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
938 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
939 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
940 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
941 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
942 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
943 info
->xbzrle_cache
->encoding_rate
= xbzrle_counters
.encoding_rate
;
944 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
947 if (migrate_compress()) {
948 info
->compression
= g_malloc0(sizeof(*info
->compression
));
949 info
->compression
->pages
= compression_counters
.pages
;
950 info
->compression
->busy
= compression_counters
.busy
;
951 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
952 info
->compression
->compressed_size
=
953 compression_counters
.compressed_size
;
954 info
->compression
->compression_rate
=
955 compression_counters
.compression_rate
;
958 if (cpu_throttle_active()) {
959 info
->has_cpu_throttle_percentage
= true;
960 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
963 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
964 info
->ram
->remaining
= ram_bytes_remaining();
965 info
->ram
->dirty_pages_rate
=
966 stat64_get(&mig_stats
.dirty_pages_rate
);
970 static void populate_disk_info(MigrationInfo
*info
)
972 if (blk_mig_active()) {
973 info
->disk
= g_malloc0(sizeof(*info
->disk
));
974 info
->disk
->transferred
= blk_mig_bytes_transferred();
975 info
->disk
->remaining
= blk_mig_bytes_remaining();
976 info
->disk
->total
= blk_mig_bytes_total();
980 static void fill_source_migration_info(MigrationInfo
*info
)
982 MigrationState
*s
= migrate_get_current();
983 int state
= qatomic_read(&s
->state
);
984 GSList
*cur_blocker
= migration_blockers
;
986 info
->blocked_reasons
= NULL
;
989 * There are two types of reasons a migration might be blocked;
990 * a) devices marked in VMState as non-migratable, and
991 * b) Explicit migration blockers
992 * We need to add both of them here.
994 qemu_savevm_non_migratable_list(&info
->blocked_reasons
);
996 while (cur_blocker
) {
997 QAPI_LIST_PREPEND(info
->blocked_reasons
,
998 g_strdup(error_get_pretty(cur_blocker
->data
)));
999 cur_blocker
= g_slist_next(cur_blocker
);
1001 info
->has_blocked_reasons
= info
->blocked_reasons
!= NULL
;
1004 case MIGRATION_STATUS_NONE
:
1005 /* no migration has happened ever */
1006 /* do not overwrite destination migration status */
1008 case MIGRATION_STATUS_SETUP
:
1009 info
->has_status
= true;
1010 info
->has_total_time
= false;
1012 case MIGRATION_STATUS_ACTIVE
:
1013 case MIGRATION_STATUS_CANCELLING
:
1014 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1015 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1016 case MIGRATION_STATUS_DEVICE
:
1017 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1018 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1019 /* TODO add some postcopy stats */
1020 populate_time_info(info
, s
);
1021 populate_ram_info(info
, s
);
1022 populate_disk_info(info
);
1023 populate_vfio_info(info
);
1025 case MIGRATION_STATUS_COLO
:
1026 info
->has_status
= true;
1027 /* TODO: display COLO specific information (checkpoint info etc.) */
1029 case MIGRATION_STATUS_COMPLETED
:
1030 populate_time_info(info
, s
);
1031 populate_ram_info(info
, s
);
1032 populate_vfio_info(info
);
1034 case MIGRATION_STATUS_FAILED
:
1035 info
->has_status
= true;
1037 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
1040 case MIGRATION_STATUS_CANCELLED
:
1041 info
->has_status
= true;
1043 case MIGRATION_STATUS_WAIT_UNPLUG
:
1044 info
->has_status
= true;
1047 info
->status
= state
;
1050 static void fill_destination_migration_info(MigrationInfo
*info
)
1052 MigrationIncomingState
*mis
= migration_incoming_get_current();
1054 if (mis
->socket_address_list
) {
1055 info
->has_socket_address
= true;
1056 info
->socket_address
=
1057 QAPI_CLONE(SocketAddressList
, mis
->socket_address_list
);
1060 switch (mis
->state
) {
1061 case MIGRATION_STATUS_NONE
:
1063 case MIGRATION_STATUS_SETUP
:
1064 case MIGRATION_STATUS_CANCELLING
:
1065 case MIGRATION_STATUS_CANCELLED
:
1066 case MIGRATION_STATUS_ACTIVE
:
1067 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1068 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1069 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1070 case MIGRATION_STATUS_FAILED
:
1071 case MIGRATION_STATUS_COLO
:
1072 info
->has_status
= true;
1074 case MIGRATION_STATUS_COMPLETED
:
1075 info
->has_status
= true;
1076 fill_destination_postcopy_migration_info(info
);
1079 info
->status
= mis
->state
;
1082 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1084 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1086 fill_destination_migration_info(info
);
1087 fill_source_migration_info(info
);
1092 void qmp_migrate_start_postcopy(Error
**errp
)
1094 MigrationState
*s
= migrate_get_current();
1096 if (!migrate_postcopy()) {
1097 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1098 " the start of migration");
1102 if (s
->state
== MIGRATION_STATUS_NONE
) {
1103 error_setg(errp
, "Postcopy must be started after migration has been"
1108 * we don't error if migration has finished since that would be racy
1109 * with issuing this command.
1111 qatomic_set(&s
->start_postcopy
, true);
1114 /* shared migration helpers */
1116 void migrate_set_state(int *state
, int old_state
, int new_state
)
1118 assert(new_state
< MIGRATION_STATUS__MAX
);
1119 if (qatomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1120 trace_migrate_set_state(MigrationStatus_str(new_state
));
1121 migrate_generate_event(new_state
);
1125 static void migrate_fd_cleanup(MigrationState
*s
)
1127 qemu_bh_delete(s
->cleanup_bh
);
1128 s
->cleanup_bh
= NULL
;
1130 g_free(s
->hostname
);
1132 json_writer_free(s
->vmdesc
);
1135 qemu_savevm_state_cleanup();
1137 if (s
->to_dst_file
) {
1140 trace_migrate_fd_cleanup();
1141 qemu_mutex_unlock_iothread();
1142 if (s
->migration_thread_running
) {
1143 qemu_thread_join(&s
->thread
);
1144 s
->migration_thread_running
= false;
1146 qemu_mutex_lock_iothread();
1148 multifd_save_cleanup();
1149 qemu_mutex_lock(&s
->qemu_file_lock
);
1150 tmp
= s
->to_dst_file
;
1151 s
->to_dst_file
= NULL
;
1152 qemu_mutex_unlock(&s
->qemu_file_lock
);
1154 * Close the file handle without the lock to make sure the
1155 * critical section won't block for long.
1157 migration_ioc_unregister_yank_from_file(tmp
);
1161 if (s
->postcopy_qemufile_src
) {
1162 migration_ioc_unregister_yank_from_file(s
->postcopy_qemufile_src
);
1163 qemu_fclose(s
->postcopy_qemufile_src
);
1164 s
->postcopy_qemufile_src
= NULL
;
1167 assert(!migration_is_active(s
));
1169 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1170 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1171 MIGRATION_STATUS_CANCELLED
);
1175 /* It is used on info migrate. We can't free it */
1176 error_report_err(error_copy(s
->error
));
1178 notifier_list_notify(&migration_state_notifiers
, s
);
1179 block_cleanup_parameters();
1180 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1183 static void migrate_fd_cleanup_schedule(MigrationState
*s
)
1186 * Ref the state for bh, because it may be called when
1187 * there're already no other refs
1189 object_ref(OBJECT(s
));
1190 qemu_bh_schedule(s
->cleanup_bh
);
1193 static void migrate_fd_cleanup_bh(void *opaque
)
1195 MigrationState
*s
= opaque
;
1196 migrate_fd_cleanup(s
);
1197 object_unref(OBJECT(s
));
1200 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1202 QEMU_LOCK_GUARD(&s
->error_mutex
);
1204 s
->error
= error_copy(error
);
1208 static void migrate_error_free(MigrationState
*s
)
1210 QEMU_LOCK_GUARD(&s
->error_mutex
);
1212 error_free(s
->error
);
1217 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1219 trace_migrate_fd_error(error_get_pretty(error
));
1220 assert(s
->to_dst_file
== NULL
);
1221 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1222 MIGRATION_STATUS_FAILED
);
1223 migrate_set_error(s
, error
);
1226 static void migrate_fd_cancel(MigrationState
*s
)
1229 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1230 trace_migrate_fd_cancel();
1232 WITH_QEMU_LOCK_GUARD(&s
->qemu_file_lock
) {
1233 if (s
->rp_state
.from_dst_file
) {
1234 /* shutdown the rp socket, so causing the rp thread to shutdown */
1235 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1240 old_state
= s
->state
;
1241 if (!migration_is_running(old_state
)) {
1244 /* If the migration is paused, kick it out of the pause */
1245 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1246 qemu_sem_post(&s
->pause_sem
);
1248 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1249 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1252 * If we're unlucky the migration code might be stuck somewhere in a
1253 * send/write while the network has failed and is waiting to timeout;
1254 * if we've got shutdown(2) available then we can force it to quit.
1255 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1256 * called in a bh, so there is no race against this cancel.
1258 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1259 qemu_file_shutdown(f
);
1261 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1262 Error
*local_err
= NULL
;
1264 bdrv_activate_all(&local_err
);
1266 error_report_err(local_err
);
1268 s
->block_inactive
= false;
1273 void add_migration_state_change_notifier(Notifier
*notify
)
1275 notifier_list_add(&migration_state_notifiers
, notify
);
1278 void remove_migration_state_change_notifier(Notifier
*notify
)
1280 notifier_remove(notify
);
1283 bool migration_in_setup(MigrationState
*s
)
1285 return s
->state
== MIGRATION_STATUS_SETUP
;
1288 bool migration_has_finished(MigrationState
*s
)
1290 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1293 bool migration_has_failed(MigrationState
*s
)
1295 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1296 s
->state
== MIGRATION_STATUS_FAILED
);
1299 bool migration_in_postcopy(void)
1301 MigrationState
*s
= migrate_get_current();
1304 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1305 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1306 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1313 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1315 return migration_in_postcopy() && s
->postcopy_after_devices
;
1318 bool migration_in_incoming_postcopy(void)
1320 PostcopyState ps
= postcopy_state_get();
1322 return ps
>= POSTCOPY_INCOMING_DISCARD
&& ps
< POSTCOPY_INCOMING_END
;
1325 bool migration_incoming_postcopy_advised(void)
1327 PostcopyState ps
= postcopy_state_get();
1329 return ps
>= POSTCOPY_INCOMING_ADVISE
&& ps
< POSTCOPY_INCOMING_END
;
1332 bool migration_in_bg_snapshot(void)
1334 MigrationState
*s
= migrate_get_current();
1336 return migrate_background_snapshot() &&
1337 migration_is_setup_or_active(s
->state
);
1340 bool migration_is_idle(void)
1342 MigrationState
*s
= current_migration
;
1349 case MIGRATION_STATUS_NONE
:
1350 case MIGRATION_STATUS_CANCELLED
:
1351 case MIGRATION_STATUS_COMPLETED
:
1352 case MIGRATION_STATUS_FAILED
:
1354 case MIGRATION_STATUS_SETUP
:
1355 case MIGRATION_STATUS_CANCELLING
:
1356 case MIGRATION_STATUS_ACTIVE
:
1357 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1358 case MIGRATION_STATUS_COLO
:
1359 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1360 case MIGRATION_STATUS_DEVICE
:
1361 case MIGRATION_STATUS_WAIT_UNPLUG
:
1363 case MIGRATION_STATUS__MAX
:
1364 g_assert_not_reached();
1370 bool migration_is_active(MigrationState
*s
)
1372 return (s
->state
== MIGRATION_STATUS_ACTIVE
||
1373 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1376 void migrate_init(MigrationState
*s
)
1379 * Reinitialise all migration state, except
1380 * parameters/capabilities that the user set, and
1385 s
->to_dst_file
= NULL
;
1386 s
->state
= MIGRATION_STATUS_NONE
;
1387 s
->rp_state
.from_dst_file
= NULL
;
1388 s
->rp_state
.error
= false;
1390 s
->pages_per_second
= 0.0;
1392 s
->expected_downtime
= 0;
1394 s
->start_postcopy
= false;
1395 s
->postcopy_after_devices
= false;
1396 s
->migration_thread_running
= false;
1397 error_free(s
->error
);
1401 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1403 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1405 s
->vm_was_running
= false;
1406 s
->iteration_initial_bytes
= 0;
1407 s
->threshold_size
= 0;
1410 int migrate_add_blocker_internal(Error
*reason
, Error
**errp
)
1412 /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */
1413 if (runstate_check(RUN_STATE_SAVE_VM
) || !migration_is_idle()) {
1414 error_propagate_prepend(errp
, error_copy(reason
),
1415 "disallowing migration blocker "
1416 "(migration/snapshot in progress) for: ");
1420 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1424 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1426 if (only_migratable
) {
1427 error_propagate_prepend(errp
, error_copy(reason
),
1428 "disallowing migration blocker "
1429 "(--only-migratable) for: ");
1433 return migrate_add_blocker_internal(reason
, errp
);
1436 void migrate_del_blocker(Error
*reason
)
1438 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1441 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1443 Error
*local_err
= NULL
;
1444 static bool once
= true;
1447 error_setg(errp
, "The incoming migration has already been started");
1450 if (!runstate_check(RUN_STATE_INMIGRATE
)) {
1451 error_setg(errp
, "'-incoming' was not specified on the command line");
1455 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
1459 qemu_start_incoming_migration(uri
, &local_err
);
1462 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1463 error_propagate(errp
, local_err
);
1470 void qmp_migrate_recover(const char *uri
, Error
**errp
)
1472 MigrationIncomingState
*mis
= migration_incoming_get_current();
1475 * Don't even bother to use ERRP_GUARD() as it _must_ always be set by
1476 * callers (no one should ignore a recover failure); if there is, it's a
1477 * programming error.
1481 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1482 error_setg(errp
, "Migrate recover can only be run "
1483 "when postcopy is paused.");
1487 /* If there's an existing transport, release it */
1488 migration_incoming_transport_cleanup(mis
);
1491 * Note that this call will never start a real migration; it will
1492 * only re-setup the migration stream and poke existing migration
1493 * to continue using that newly established channel.
1495 qemu_start_incoming_migration(uri
, errp
);
1498 void qmp_migrate_pause(Error
**errp
)
1500 MigrationState
*ms
= migrate_get_current();
1501 MigrationIncomingState
*mis
= migration_incoming_get_current();
1504 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1505 /* Source side, during postcopy */
1506 qemu_mutex_lock(&ms
->qemu_file_lock
);
1507 ret
= qemu_file_shutdown(ms
->to_dst_file
);
1508 qemu_mutex_unlock(&ms
->qemu_file_lock
);
1510 error_setg(errp
, "Failed to pause source migration");
1515 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1516 ret
= qemu_file_shutdown(mis
->from_src_file
);
1518 error_setg(errp
, "Failed to pause destination migration");
1523 error_setg(errp
, "migrate-pause is currently only supported "
1524 "during postcopy-active state");
1527 bool migration_is_blocked(Error
**errp
)
1529 if (qemu_savevm_state_blocked(errp
)) {
1533 if (migration_blockers
) {
1534 error_propagate(errp
, error_copy(migration_blockers
->data
));
1541 /* Returns true if continue to migrate, or false if error detected */
1542 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
1543 bool resume
, Error
**errp
)
1545 Error
*local_err
= NULL
;
1548 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1549 error_setg(errp
, "Cannot resume if there is no "
1550 "paused migration");
1555 * Postcopy recovery won't work well with release-ram
1556 * capability since release-ram will drop the page buffer as
1557 * long as the page is put into the send buffer. So if there
1558 * is a network failure happened, any page buffers that have
1559 * not yet reached the destination VM but have already been
1560 * sent from the source VM will be lost forever. Let's refuse
1561 * the client from resuming such a postcopy migration.
1562 * Luckily release-ram was designed to only be used when src
1563 * and destination VMs are on the same host, so it should be
1566 if (migrate_release_ram()) {
1567 error_setg(errp
, "Postcopy recovery cannot work "
1568 "when release-ram capability is set");
1572 /* This is a resume, skip init status */
1576 if (migration_is_running(s
->state
)) {
1577 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1581 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1582 error_setg(errp
, "Guest is waiting for an incoming migration");
1586 if (runstate_check(RUN_STATE_POSTMIGRATE
)) {
1587 error_setg(errp
, "Can't migrate the vm that was paused due to "
1588 "previous migration");
1592 if (migration_is_blocked(errp
)) {
1596 if (blk
|| blk_inc
) {
1597 if (migrate_colo()) {
1598 error_setg(errp
, "No disk migration is required in COLO mode");
1601 if (migrate_block() || migrate_block_incremental()) {
1602 error_setg(errp
, "Command options are incompatible with "
1603 "current migration capabilities");
1606 if (!migrate_cap_set(MIGRATION_CAPABILITY_BLOCK
, true, &local_err
)) {
1607 error_propagate(errp
, local_err
);
1610 s
->must_remove_block_options
= true;
1614 migrate_set_block_incremental(true);
1619 * set mig_stats compression_counters memory to zero for a
1622 memset(&mig_stats
, 0, sizeof(mig_stats
));
1623 memset(&compression_counters
, 0, sizeof(compression_counters
));
1628 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1629 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1630 bool has_resume
, bool resume
, Error
**errp
)
1632 Error
*local_err
= NULL
;
1633 MigrationState
*s
= migrate_get_current();
1634 const char *p
= NULL
;
1636 /* URI is not suitable for migration? */
1637 if (!migration_channels_and_uri_compatible(uri
, errp
)) {
1641 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
1642 has_resume
&& resume
, errp
)) {
1643 /* Error detected, put into errp */
1647 if (!(has_resume
&& resume
)) {
1648 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
1653 if (strstart(uri
, "tcp:", &p
) ||
1654 strstart(uri
, "unix:", NULL
) ||
1655 strstart(uri
, "vsock:", NULL
)) {
1656 socket_start_outgoing_migration(s
, p
? p
: uri
, &local_err
);
1658 } else if (strstart(uri
, "rdma:", &p
)) {
1659 rdma_start_outgoing_migration(s
, p
, &local_err
);
1661 } else if (strstart(uri
, "exec:", &p
)) {
1662 exec_start_outgoing_migration(s
, p
, &local_err
);
1663 } else if (strstart(uri
, "fd:", &p
)) {
1664 fd_start_outgoing_migration(s
, p
, &local_err
);
1666 if (!(has_resume
&& resume
)) {
1667 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1669 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1670 "a valid migration protocol");
1671 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1672 MIGRATION_STATUS_FAILED
);
1673 block_cleanup_parameters();
1678 if (!(has_resume
&& resume
)) {
1679 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1681 migrate_fd_error(s
, local_err
);
1682 error_propagate(errp
, local_err
);
1687 void qmp_migrate_cancel(Error
**errp
)
1689 migration_cancel(NULL
);
1692 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
1694 MigrationState
*s
= migrate_get_current();
1695 if (s
->state
!= state
) {
1696 error_setg(errp
, "Migration not in expected state: %s",
1697 MigrationStatus_str(s
->state
));
1700 qemu_sem_post(&s
->pause_sem
);
1703 /* migration thread support */
1705 * Something bad happened to the RP stream, mark an error
1706 * The caller shall print or trace something to indicate why
1708 static void mark_source_rp_bad(MigrationState
*s
)
1710 s
->rp_state
.error
= true;
1713 static struct rp_cmd_args
{
1714 ssize_t len
; /* -1 = variable */
1717 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1718 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1719 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1720 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1721 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1722 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
1723 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
1724 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1728 * Process a request for pages received on the return path,
1729 * We're allowed to send more than requested (e.g. to round to our page size)
1730 * and we don't need to send pages that have already been sent.
1732 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1733 ram_addr_t start
, size_t len
)
1735 long our_host_ps
= qemu_real_host_page_size();
1737 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1740 * Since we currently insist on matching page sizes, just sanity check
1741 * we're being asked for whole host pages.
1743 if (!QEMU_IS_ALIGNED(start
, our_host_ps
) ||
1744 !QEMU_IS_ALIGNED(len
, our_host_ps
)) {
1745 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1746 " len: %zd", __func__
, start
, len
);
1747 mark_source_rp_bad(ms
);
1751 if (ram_save_queue_pages(rbname
, start
, len
)) {
1752 mark_source_rp_bad(ms
);
1756 /* Return true to retry, false to quit */
1757 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
1759 trace_postcopy_pause_return_path();
1761 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
1763 trace_postcopy_pause_return_path_continued();
1768 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
1770 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
1773 error_report("%s: invalid block name '%s'", __func__
, block_name
);
1777 /* Fetch the received bitmap and refresh the dirty bitmap */
1778 return ram_dirty_bitmap_reload(s
, block
);
1781 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
1783 trace_source_return_path_thread_resume_ack(value
);
1785 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
1786 error_report("%s: illegal resume_ack value %"PRIu32
,
1791 /* Now both sides are active. */
1792 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
1793 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1795 /* Notify send thread that time to continue send pages */
1796 qemu_sem_post(&s
->rp_state
.rp_sem
);
1802 * Release ms->rp_state.from_dst_file (and postcopy_qemufile_src if
1803 * existed) in a safe way.
1805 static void migration_release_dst_files(MigrationState
*ms
)
1809 WITH_QEMU_LOCK_GUARD(&ms
->qemu_file_lock
) {
1811 * Reset the from_dst_file pointer first before releasing it, as we
1812 * can't block within lock section
1814 file
= ms
->rp_state
.from_dst_file
;
1815 ms
->rp_state
.from_dst_file
= NULL
;
1819 * Do the same to postcopy fast path socket too if there is. No
1820 * locking needed because this qemufile should only be managed by
1821 * return path thread.
1823 if (ms
->postcopy_qemufile_src
) {
1824 migration_ioc_unregister_yank_from_file(ms
->postcopy_qemufile_src
);
1825 qemu_file_shutdown(ms
->postcopy_qemufile_src
);
1826 qemu_fclose(ms
->postcopy_qemufile_src
);
1827 ms
->postcopy_qemufile_src
= NULL
;
1834 * Handles messages sent on the return path towards the source VM
1837 static void *source_return_path_thread(void *opaque
)
1839 MigrationState
*ms
= opaque
;
1840 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1841 uint16_t header_len
, header_type
;
1843 uint32_t tmp32
, sibling_error
;
1844 ram_addr_t start
= 0; /* =0 to silence warning */
1845 size_t len
= 0, expected_len
;
1848 trace_source_return_path_thread_entry();
1849 rcu_register_thread();
1852 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1853 migration_is_setup_or_active(ms
->state
)) {
1854 trace_source_return_path_thread_loop_top();
1855 header_type
= qemu_get_be16(rp
);
1856 header_len
= qemu_get_be16(rp
);
1858 if (qemu_file_get_error(rp
)) {
1859 mark_source_rp_bad(ms
);
1863 if (header_type
>= MIG_RP_MSG_MAX
||
1864 header_type
== MIG_RP_MSG_INVALID
) {
1865 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1866 header_type
, header_len
);
1867 mark_source_rp_bad(ms
);
1871 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1872 header_len
!= rp_cmd_args
[header_type
].len
) ||
1873 header_len
> sizeof(buf
)) {
1874 error_report("RP: Received '%s' message (0x%04x) with"
1875 "incorrect length %d expecting %zu",
1876 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1877 (size_t)rp_cmd_args
[header_type
].len
);
1878 mark_source_rp_bad(ms
);
1882 /* We know we've got a valid header by this point */
1883 res
= qemu_get_buffer(rp
, buf
, header_len
);
1884 if (res
!= header_len
) {
1885 error_report("RP: Failed reading data for message 0x%04x"
1886 " read %d expected %d",
1887 header_type
, res
, header_len
);
1888 mark_source_rp_bad(ms
);
1892 /* OK, we have the message and the data */
1893 switch (header_type
) {
1894 case MIG_RP_MSG_SHUT
:
1895 sibling_error
= ldl_be_p(buf
);
1896 trace_source_return_path_thread_shut(sibling_error
);
1897 if (sibling_error
) {
1898 error_report("RP: Sibling indicated error %d", sibling_error
);
1899 mark_source_rp_bad(ms
);
1902 * We'll let the main thread deal with closing the RP
1903 * we could do a shutdown(2) on it, but we're the only user
1904 * anyway, so there's nothing gained.
1908 case MIG_RP_MSG_PONG
:
1909 tmp32
= ldl_be_p(buf
);
1910 trace_source_return_path_thread_pong(tmp32
);
1911 qemu_sem_post(&ms
->rp_state
.rp_pong_acks
);
1914 case MIG_RP_MSG_REQ_PAGES
:
1915 start
= ldq_be_p(buf
);
1916 len
= ldl_be_p(buf
+ 8);
1917 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1920 case MIG_RP_MSG_REQ_PAGES_ID
:
1921 expected_len
= 12 + 1; /* header + termination */
1923 if (header_len
>= expected_len
) {
1924 start
= ldq_be_p(buf
);
1925 len
= ldl_be_p(buf
+ 8);
1926 /* Now we expect an idstr */
1927 tmp32
= buf
[12]; /* Length of the following idstr */
1928 buf
[13 + tmp32
] = '\0';
1929 expected_len
+= tmp32
;
1931 if (header_len
!= expected_len
) {
1932 error_report("RP: Req_Page_id with length %d expecting %zd",
1933 header_len
, expected_len
);
1934 mark_source_rp_bad(ms
);
1937 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
1940 case MIG_RP_MSG_RECV_BITMAP
:
1941 if (header_len
< 1) {
1942 error_report("%s: missing block name", __func__
);
1943 mark_source_rp_bad(ms
);
1946 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
1947 buf
[buf
[0] + 1] = '\0';
1948 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
1949 mark_source_rp_bad(ms
);
1954 case MIG_RP_MSG_RESUME_ACK
:
1955 tmp32
= ldl_be_p(buf
);
1956 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
1957 mark_source_rp_bad(ms
);
1968 res
= qemu_file_get_error(rp
);
1970 if (res
&& migration_in_postcopy()) {
1972 * Maybe there is something we can do: it looks like a
1973 * network down issue, and we pause for a recovery.
1975 migration_release_dst_files(ms
);
1977 if (postcopy_pause_return_path_thread(ms
)) {
1979 * Reload rp, reset the rest. Referencing it is safe since
1980 * it's reset only by us above, or when migration completes
1982 rp
= ms
->rp_state
.from_dst_file
;
1983 ms
->rp_state
.error
= false;
1988 trace_source_return_path_thread_bad_end();
1989 mark_source_rp_bad(ms
);
1992 trace_source_return_path_thread_end();
1993 migration_release_dst_files(ms
);
1994 rcu_unregister_thread();
1998 static int open_return_path_on_source(MigrationState
*ms
,
2001 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
2002 if (!ms
->rp_state
.from_dst_file
) {
2006 trace_open_return_path_on_source();
2008 if (!create_thread
) {
2013 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
2014 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
2015 ms
->rp_state
.rp_thread_created
= true;
2017 trace_open_return_path_on_source_continue();
2022 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2023 static int await_return_path_close_on_source(MigrationState
*ms
)
2026 * If this is a normal exit then the destination will send a SHUT and the
2027 * rp_thread will exit, however if there's an error we need to cause
2030 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
2032 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2033 * waiting for the destination.
2035 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
2036 mark_source_rp_bad(ms
);
2038 trace_await_return_path_close_on_source_joining();
2039 qemu_thread_join(&ms
->rp_state
.rp_thread
);
2040 ms
->rp_state
.rp_thread_created
= false;
2041 trace_await_return_path_close_on_source_close();
2042 return ms
->rp_state
.error
;
2046 migration_wait_main_channel(MigrationState
*ms
)
2048 /* Wait until one PONG message received */
2049 qemu_sem_wait(&ms
->rp_state
.rp_pong_acks
);
2053 * Switch from normal iteration to postcopy
2054 * Returns non-0 on error
2056 static int postcopy_start(MigrationState
*ms
)
2059 QIOChannelBuffer
*bioc
;
2061 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2062 uint64_t bandwidth
= migrate_max_postcopy_bandwidth();
2063 bool restart_block
= false;
2064 int cur_state
= MIGRATION_STATUS_ACTIVE
;
2066 if (migrate_postcopy_preempt()) {
2067 migration_wait_main_channel(ms
);
2068 if (postcopy_preempt_establish_channel(ms
)) {
2069 migrate_set_state(&ms
->state
, ms
->state
, MIGRATION_STATUS_FAILED
);
2074 if (!migrate_pause_before_switchover()) {
2075 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
2076 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2079 trace_postcopy_start();
2080 qemu_mutex_lock_iothread();
2081 trace_postcopy_start_set_run();
2083 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2084 global_state_store();
2085 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2090 ret
= migration_maybe_pause(ms
, &cur_state
,
2091 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2096 ret
= bdrv_inactivate_all();
2100 restart_block
= true;
2103 * Cause any non-postcopiable, but iterative devices to
2104 * send out their final data.
2106 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2109 * in Finish migrate and with the io-lock held everything should
2110 * be quiet, but we've potentially still got dirty pages and we
2111 * need to tell the destination to throw any pages it's already received
2114 if (migrate_postcopy_ram()) {
2115 ram_postcopy_send_discard_bitmap(ms
);
2119 * send rest of state - note things that are doing postcopy
2120 * will notice we're in POSTCOPY_ACTIVE and not actually
2121 * wrap their state up here
2123 migration_rate_set(bandwidth
);
2124 if (migrate_postcopy_ram()) {
2125 /* Ping just for debugging, helps line traces up */
2126 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
2130 * While loading the device state we may trigger page transfer
2131 * requests and the fd must be free to process those, and thus
2132 * the destination must read the whole device state off the fd before
2133 * it starts processing it. Unfortunately the ad-hoc migration format
2134 * doesn't allow the destination to know the size to read without fully
2135 * parsing it through each devices load-state code (especially the open
2136 * coded devices that use get/put).
2137 * So we wrap the device state up in a package with a length at the start;
2138 * to do this we use a qemu_buf to hold the whole of the device state.
2140 bioc
= qio_channel_buffer_new(4096);
2141 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
2142 fb
= qemu_file_new_output(QIO_CHANNEL(bioc
));
2143 object_unref(OBJECT(bioc
));
2146 * Make sure the receiver can get incoming pages before we send the rest
2149 qemu_savevm_send_postcopy_listen(fb
);
2151 qemu_savevm_state_complete_precopy(fb
, false, false);
2152 if (migrate_postcopy_ram()) {
2153 qemu_savevm_send_ping(fb
, 3);
2156 qemu_savevm_send_postcopy_run(fb
);
2158 /* <><> end of stuff going into the package */
2160 /* Last point of recovery; as soon as we send the package the destination
2161 * can open devices and potentially start running.
2162 * Lets just check again we've not got any errors.
2164 ret
= qemu_file_get_error(ms
->to_dst_file
);
2166 error_report("postcopy_start: Migration stream errored (pre package)");
2170 restart_block
= false;
2172 /* Now send that blob */
2173 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2178 /* Send a notify to give a chance for anything that needs to happen
2179 * at the transition to postcopy and after the device state; in particular
2180 * spice needs to trigger a transition now
2182 ms
->postcopy_after_devices
= true;
2183 notifier_list_notify(&migration_state_notifiers
, ms
);
2185 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2187 qemu_mutex_unlock_iothread();
2189 if (migrate_postcopy_ram()) {
2191 * Although this ping is just for debug, it could potentially be
2192 * used for getting a better measurement of downtime at the source.
2194 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2197 if (migrate_release_ram()) {
2198 ram_postcopy_migrated_memory_release(ms
);
2201 ret
= qemu_file_get_error(ms
->to_dst_file
);
2203 error_report("postcopy_start: Migration stream errored");
2204 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2205 MIGRATION_STATUS_FAILED
);
2208 trace_postcopy_preempt_enabled(migrate_postcopy_preempt());
2215 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2216 MIGRATION_STATUS_FAILED
);
2217 if (restart_block
) {
2218 /* A failure happened early enough that we know the destination hasn't
2219 * accessed block devices, so we're safe to recover.
2221 Error
*local_err
= NULL
;
2223 bdrv_activate_all(&local_err
);
2225 error_report_err(local_err
);
2228 qemu_mutex_unlock_iothread();
2233 * migration_maybe_pause: Pause if required to by
2234 * migrate_pause_before_switchover called with the iothread locked
2235 * Returns: 0 on success
2237 static int migration_maybe_pause(MigrationState
*s
,
2238 int *current_active_state
,
2241 if (!migrate_pause_before_switchover()) {
2245 /* Since leaving this state is not atomic with posting the semaphore
2246 * it's possible that someone could have issued multiple migrate_continue
2247 * and the semaphore is incorrectly positive at this point;
2248 * the docs say it's undefined to reinit a semaphore that's already
2249 * init'd, so use timedwait to eat up any existing posts.
2251 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2252 /* This block intentionally left blank */
2256 * If the migration is cancelled when it is in the completion phase,
2257 * the migration state is set to MIGRATION_STATUS_CANCELLING.
2258 * So we don't need to wait a semaphore, otherwise we would always
2259 * wait for the 'pause_sem' semaphore.
2261 if (s
->state
!= MIGRATION_STATUS_CANCELLING
) {
2262 qemu_mutex_unlock_iothread();
2263 migrate_set_state(&s
->state
, *current_active_state
,
2264 MIGRATION_STATUS_PRE_SWITCHOVER
);
2265 qemu_sem_wait(&s
->pause_sem
);
2266 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2268 *current_active_state
= new_state
;
2269 qemu_mutex_lock_iothread();
2272 return s
->state
== new_state
? 0 : -EINVAL
;
2276 * migration_completion: Used by migration_thread when there's not much left.
2277 * The caller 'breaks' the loop when this returns.
2279 * @s: Current migration state
2281 static void migration_completion(MigrationState
*s
)
2284 int current_active_state
= s
->state
;
2286 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2287 qemu_mutex_lock_iothread();
2288 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2289 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2290 s
->vm_was_running
= runstate_is_running();
2291 ret
= global_state_store();
2294 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2295 trace_migration_completion_vm_stop(ret
);
2297 ret
= migration_maybe_pause(s
, ¤t_active_state
,
2298 MIGRATION_STATUS_DEVICE
);
2302 * Inactivate disks except in COLO, and track that we
2303 * have done so in order to remember to reactivate
2304 * them if migration fails or is cancelled.
2306 s
->block_inactive
= !migrate_colo();
2307 migration_rate_set(RATE_LIMIT_DISABLED
);
2308 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
2312 qemu_mutex_unlock_iothread();
2317 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2318 trace_migration_completion_postcopy_end();
2320 qemu_mutex_lock_iothread();
2321 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
2322 qemu_mutex_unlock_iothread();
2325 * Shutdown the postcopy fast path thread. This is only needed
2326 * when dest QEMU binary is old (7.1/7.2). QEMU 8.0+ doesn't need
2329 if (migrate_postcopy_preempt() && s
->preempt_pre_7_2
) {
2330 postcopy_preempt_shutdown_file(s
);
2333 trace_migration_completion_postcopy_end_after_complete();
2339 * If rp was opened we must clean up the thread before
2340 * cleaning everything else up (since if there are no failures
2341 * it will wait for the destination to send it's status in
2344 if (s
->rp_state
.rp_thread_created
) {
2346 trace_migration_return_path_end_before();
2347 rp_error
= await_return_path_close_on_source(s
);
2348 trace_migration_return_path_end_after(rp_error
);
2354 if (qemu_file_get_error(s
->to_dst_file
)) {
2355 trace_migration_completion_file_err();
2359 if (migrate_colo() && s
->state
== MIGRATION_STATUS_ACTIVE
) {
2360 /* COLO does not support postcopy */
2361 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
2362 MIGRATION_STATUS_COLO
);
2364 migrate_set_state(&s
->state
, current_active_state
,
2365 MIGRATION_STATUS_COMPLETED
);
2371 if (s
->block_inactive
&& (s
->state
== MIGRATION_STATUS_ACTIVE
||
2372 s
->state
== MIGRATION_STATUS_DEVICE
)) {
2374 * If not doing postcopy, vm_start() will be called: let's
2375 * regain control on images.
2377 Error
*local_err
= NULL
;
2379 qemu_mutex_lock_iothread();
2380 bdrv_activate_all(&local_err
);
2382 error_report_err(local_err
);
2384 s
->block_inactive
= false;
2386 qemu_mutex_unlock_iothread();
2389 migrate_set_state(&s
->state
, current_active_state
,
2390 MIGRATION_STATUS_FAILED
);
2394 * bg_migration_completion: Used by bg_migration_thread when after all the
2395 * RAM has been saved. The caller 'breaks' the loop when this returns.
2397 * @s: Current migration state
2399 static void bg_migration_completion(MigrationState
*s
)
2401 int current_active_state
= s
->state
;
2404 * Stop tracking RAM writes - un-protect memory, un-register UFFD
2405 * memory ranges, flush kernel wait queues and wake up threads
2406 * waiting for write fault to be resolved.
2408 ram_write_tracking_stop();
2410 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2412 * By this moment we have RAM content saved into the migration stream.
2413 * The next step is to flush the non-RAM content (device state)
2414 * right after the ram content. The device state has been stored into
2415 * the temporary buffer before RAM saving started.
2417 qemu_put_buffer(s
->to_dst_file
, s
->bioc
->data
, s
->bioc
->usage
);
2418 qemu_fflush(s
->to_dst_file
);
2419 } else if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
2423 if (qemu_file_get_error(s
->to_dst_file
)) {
2424 trace_migration_completion_file_err();
2428 migrate_set_state(&s
->state
, current_active_state
,
2429 MIGRATION_STATUS_COMPLETED
);
2433 migrate_set_state(&s
->state
, current_active_state
,
2434 MIGRATION_STATUS_FAILED
);
2437 typedef enum MigThrError
{
2438 /* No error detected */
2439 MIG_THR_ERR_NONE
= 0,
2440 /* Detected error, but resumed successfully */
2441 MIG_THR_ERR_RECOVERED
= 1,
2442 /* Detected fatal error, need to exit */
2443 MIG_THR_ERR_FATAL
= 2,
2446 static int postcopy_resume_handshake(MigrationState
*s
)
2448 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
2450 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2451 qemu_sem_wait(&s
->rp_state
.rp_sem
);
2454 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2461 /* Return zero if success, or <0 for error */
2462 static int postcopy_do_resume(MigrationState
*s
)
2467 * Call all the resume_prepare() hooks, so that modules can be
2468 * ready for the migration resume.
2470 ret
= qemu_savevm_state_resume_prepare(s
);
2472 error_report("%s: resume_prepare() failure detected: %d",
2478 * If preempt is enabled, re-establish the preempt channel. Note that
2479 * we do it after resume prepare to make sure the main channel will be
2480 * created before the preempt channel. E.g. with weak network, the
2481 * dest QEMU may get messed up with the preempt and main channels on
2482 * the order of connection setup. This guarantees the correct order.
2484 ret
= postcopy_preempt_establish_channel(s
);
2486 error_report("%s: postcopy_preempt_establish_channel(): %d",
2492 * Last handshake with destination on the resume (destination will
2493 * switch to postcopy-active afterwards)
2495 ret
= postcopy_resume_handshake(s
);
2497 error_report("%s: handshake failed: %d", __func__
, ret
);
2505 * We don't return until we are in a safe state to continue current
2506 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2507 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2509 static MigThrError
postcopy_pause(MigrationState
*s
)
2511 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2517 * Current channel is possibly broken. Release it. Note that this is
2518 * guaranteed even without lock because to_dst_file should only be
2519 * modified by the migration thread. That also guarantees that the
2520 * unregister of yank is safe too without the lock. It should be safe
2521 * even to be within the qemu_file_lock, but we didn't do that to avoid
2522 * taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make
2523 * the qemu_file_lock critical section as small as possible.
2525 assert(s
->to_dst_file
);
2526 migration_ioc_unregister_yank_from_file(s
->to_dst_file
);
2527 qemu_mutex_lock(&s
->qemu_file_lock
);
2528 file
= s
->to_dst_file
;
2529 s
->to_dst_file
= NULL
;
2530 qemu_mutex_unlock(&s
->qemu_file_lock
);
2532 qemu_file_shutdown(file
);
2535 migrate_set_state(&s
->state
, s
->state
,
2536 MIGRATION_STATUS_POSTCOPY_PAUSED
);
2538 error_report("Detected IO failure for postcopy. "
2539 "Migration paused.");
2542 * We wait until things fixed up. Then someone will setup the
2543 * status back for us.
2545 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2546 qemu_sem_wait(&s
->postcopy_pause_sem
);
2549 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2550 /* Woken up by a recover procedure. Give it a shot */
2553 * Firstly, let's wake up the return path now, with a new
2554 * return path channel.
2556 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
2558 /* Do the resume logic */
2559 if (postcopy_do_resume(s
) == 0) {
2560 /* Let's continue! */
2561 trace_postcopy_pause_continued();
2562 return MIG_THR_ERR_RECOVERED
;
2565 * Something wrong happened during the recovery, let's
2566 * pause again. Pause is always better than throwing
2572 /* This is not right... Time to quit. */
2573 return MIG_THR_ERR_FATAL
;
2578 static MigThrError
migration_detect_error(MigrationState
*s
)
2581 int state
= s
->state
;
2582 Error
*local_error
= NULL
;
2584 if (state
== MIGRATION_STATUS_CANCELLING
||
2585 state
== MIGRATION_STATUS_CANCELLED
) {
2586 /* End the migration, but don't set the state to failed */
2587 return MIG_THR_ERR_FATAL
;
2591 * Try to detect any file errors. Note that postcopy_qemufile_src will
2592 * be NULL when postcopy preempt is not enabled.
2594 ret
= qemu_file_get_error_obj_any(s
->to_dst_file
,
2595 s
->postcopy_qemufile_src
,
2598 /* Everything is fine */
2599 assert(!local_error
);
2600 return MIG_THR_ERR_NONE
;
2604 migrate_set_error(s
, local_error
);
2605 error_free(local_error
);
2608 if (state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
) {
2610 * For postcopy, we allow the network to be down for a
2611 * while. After that, it can be continued by a
2614 return postcopy_pause(s
);
2617 * For precopy (or postcopy with error outside IO), we fail
2620 migrate_set_state(&s
->state
, state
, MIGRATION_STATUS_FAILED
);
2621 trace_migration_thread_file_err();
2623 /* Time to stop the migration, now. */
2624 return MIG_THR_ERR_FATAL
;
2628 static void migration_calculate_complete(MigrationState
*s
)
2630 uint64_t bytes
= migration_transferred_bytes(s
->to_dst_file
);
2631 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2632 int64_t transfer_time
;
2634 s
->total_time
= end_time
- s
->start_time
;
2637 * It's still not set, so we are precopy migration. For
2638 * postcopy, downtime is calculated during postcopy_start().
2640 s
->downtime
= end_time
- s
->downtime_start
;
2643 transfer_time
= s
->total_time
- s
->setup_time
;
2644 if (transfer_time
) {
2645 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
2649 static void update_iteration_initial_status(MigrationState
*s
)
2652 * Update these three fields at the same time to avoid mismatch info lead
2653 * wrong speed calculation.
2655 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2656 s
->iteration_initial_bytes
= migration_transferred_bytes(s
->to_dst_file
);
2657 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
2660 static void migration_update_counters(MigrationState
*s
,
2661 int64_t current_time
)
2663 uint64_t transferred
, transferred_pages
, time_spent
;
2664 uint64_t current_bytes
; /* bytes transferred since the beginning */
2667 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
2671 current_bytes
= migration_transferred_bytes(s
->to_dst_file
);
2672 transferred
= current_bytes
- s
->iteration_initial_bytes
;
2673 time_spent
= current_time
- s
->iteration_start_time
;
2674 bandwidth
= (double)transferred
/ time_spent
;
2675 s
->threshold_size
= bandwidth
* migrate_downtime_limit();
2677 s
->mbps
= (((double) transferred
* 8.0) /
2678 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2680 transferred_pages
= ram_get_total_transferred_pages() -
2681 s
->iteration_initial_pages
;
2682 s
->pages_per_second
= (double) transferred_pages
/
2683 (((double) time_spent
/ 1000.0));
2686 * if we haven't sent anything, we don't want to
2687 * recalculate. 10000 is a small enough number for our purposes
2689 if (stat64_get(&mig_stats
.dirty_pages_rate
) &&
2690 transferred
> 10000) {
2691 s
->expected_downtime
=
2692 stat64_get(&mig_stats
.dirty_bytes_last_sync
) / bandwidth
;
2695 migration_rate_reset(s
->to_dst_file
);
2697 update_iteration_initial_status(s
);
2699 trace_migrate_transferred(transferred
, time_spent
,
2700 bandwidth
, s
->threshold_size
);
2703 /* Migration thread iteration status */
2705 MIG_ITERATE_RESUME
, /* Resume current iteration */
2706 MIG_ITERATE_SKIP
, /* Skip current iteration */
2707 MIG_ITERATE_BREAK
, /* Break the loop */
2711 * Return true if continue to the next iteration directly, false
2714 static MigIterateState
migration_iteration_run(MigrationState
*s
)
2716 uint64_t must_precopy
, can_postcopy
;
2717 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
2719 qemu_savevm_state_pending_estimate(&must_precopy
, &can_postcopy
);
2720 uint64_t pending_size
= must_precopy
+ can_postcopy
;
2722 trace_migrate_pending_estimate(pending_size
, must_precopy
, can_postcopy
);
2724 if (must_precopy
<= s
->threshold_size
) {
2725 qemu_savevm_state_pending_exact(&must_precopy
, &can_postcopy
);
2726 pending_size
= must_precopy
+ can_postcopy
;
2727 trace_migrate_pending_exact(pending_size
, must_precopy
, can_postcopy
);
2730 if (!pending_size
|| pending_size
< s
->threshold_size
) {
2731 trace_migration_thread_low_pending(pending_size
);
2732 migration_completion(s
);
2733 return MIG_ITERATE_BREAK
;
2736 /* Still a significant amount to transfer */
2737 if (!in_postcopy
&& must_precopy
<= s
->threshold_size
&&
2738 qatomic_read(&s
->start_postcopy
)) {
2739 if (postcopy_start(s
)) {
2740 error_report("%s: postcopy failed to start", __func__
);
2742 return MIG_ITERATE_SKIP
;
2745 /* Just another iteration step */
2746 qemu_savevm_state_iterate(s
->to_dst_file
, in_postcopy
);
2747 return MIG_ITERATE_RESUME
;
2750 static void migration_iteration_finish(MigrationState
*s
)
2752 /* If we enabled cpu throttling for auto-converge, turn it off. */
2753 cpu_throttle_stop();
2755 qemu_mutex_lock_iothread();
2757 case MIGRATION_STATUS_COMPLETED
:
2758 migration_calculate_complete(s
);
2759 runstate_set(RUN_STATE_POSTMIGRATE
);
2761 case MIGRATION_STATUS_COLO
:
2762 assert(migrate_colo());
2763 migrate_start_colo_process(s
);
2764 s
->vm_was_running
= true;
2766 case MIGRATION_STATUS_FAILED
:
2767 case MIGRATION_STATUS_CANCELLED
:
2768 case MIGRATION_STATUS_CANCELLING
:
2769 if (s
->vm_was_running
) {
2770 if (!runstate_check(RUN_STATE_SHUTDOWN
)) {
2774 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2775 runstate_set(RUN_STATE_POSTMIGRATE
);
2781 /* Should not reach here, but if so, forgive the VM. */
2782 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
2785 migrate_fd_cleanup_schedule(s
);
2786 qemu_mutex_unlock_iothread();
2789 static void bg_migration_iteration_finish(MigrationState
*s
)
2791 qemu_mutex_lock_iothread();
2793 case MIGRATION_STATUS_COMPLETED
:
2794 migration_calculate_complete(s
);
2797 case MIGRATION_STATUS_ACTIVE
:
2798 case MIGRATION_STATUS_FAILED
:
2799 case MIGRATION_STATUS_CANCELLED
:
2800 case MIGRATION_STATUS_CANCELLING
:
2804 /* Should not reach here, but if so, forgive the VM. */
2805 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
2809 migrate_fd_cleanup_schedule(s
);
2810 qemu_mutex_unlock_iothread();
2814 * Return true if continue to the next iteration directly, false
2817 static MigIterateState
bg_migration_iteration_run(MigrationState
*s
)
2821 res
= qemu_savevm_state_iterate(s
->to_dst_file
, false);
2823 bg_migration_completion(s
);
2824 return MIG_ITERATE_BREAK
;
2827 return MIG_ITERATE_RESUME
;
2830 void migration_make_urgent_request(void)
2832 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
2835 void migration_consume_urgent_request(void)
2837 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
2840 /* Returns true if the rate limiting was broken by an urgent request */
2841 bool migration_rate_limit(void)
2843 int64_t now
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2844 MigrationState
*s
= migrate_get_current();
2846 bool urgent
= false;
2847 migration_update_counters(s
, now
);
2848 if (migration_rate_exceeded(s
->to_dst_file
)) {
2850 if (qemu_file_get_error(s
->to_dst_file
)) {
2854 * Wait for a delay to do rate limiting OR
2855 * something urgent to post the semaphore.
2857 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- now
;
2858 trace_migration_rate_limit_pre(ms
);
2859 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
2861 * We were woken by one or more urgent things but
2862 * the timedwait will have consumed one of them.
2863 * The service routine for the urgent wake will dec
2864 * the semaphore itself for each item it consumes,
2865 * so add this one we just eat back.
2867 qemu_sem_post(&s
->rate_limit_sem
);
2870 trace_migration_rate_limit_post(urgent
);
2876 * if failover devices are present, wait they are completely
2880 static void qemu_savevm_wait_unplug(MigrationState
*s
, int old_state
,
2883 if (qemu_savevm_state_guest_unplug_pending()) {
2884 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_WAIT_UNPLUG
);
2886 while (s
->state
== MIGRATION_STATUS_WAIT_UNPLUG
&&
2887 qemu_savevm_state_guest_unplug_pending()) {
2888 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
2890 if (s
->state
!= MIGRATION_STATUS_WAIT_UNPLUG
) {
2891 int timeout
= 120; /* 30 seconds */
2893 * migration has been canceled
2894 * but as we have started an unplug we must wait the end
2895 * to be able to plug back the card
2897 while (timeout
-- && qemu_savevm_state_guest_unplug_pending()) {
2898 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
2900 if (qemu_savevm_state_guest_unplug_pending() &&
2902 warn_report("migration: partially unplugged device on "
2907 migrate_set_state(&s
->state
, MIGRATION_STATUS_WAIT_UNPLUG
, new_state
);
2909 migrate_set_state(&s
->state
, old_state
, new_state
);
2914 * Master migration thread on the source VM.
2915 * It drives the migration and pumps the data down the outgoing channel.
2917 static void *migration_thread(void *opaque
)
2919 MigrationState
*s
= opaque
;
2920 MigrationThread
*thread
= NULL
;
2921 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
2922 MigThrError thr_error
;
2923 bool urgent
= false;
2925 thread
= MigrationThreadAdd("live_migration", qemu_get_thread_id());
2927 rcu_register_thread();
2929 object_ref(OBJECT(s
));
2930 update_iteration_initial_status(s
);
2932 qemu_savevm_state_header(s
->to_dst_file
);
2935 * If we opened the return path, we need to make sure dst has it
2938 if (s
->rp_state
.rp_thread_created
) {
2939 /* Now tell the dest that it should open its end so it can reply */
2940 qemu_savevm_send_open_return_path(s
->to_dst_file
);
2942 /* And do a ping that will make stuff easier to debug */
2943 qemu_savevm_send_ping(s
->to_dst_file
, 1);
2946 if (migrate_postcopy()) {
2948 * Tell the destination that we *might* want to do postcopy later;
2949 * if the other end can't do postcopy it should fail now, nice and
2952 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
2955 if (migrate_colo()) {
2956 /* Notify migration destination that we enable COLO */
2957 qemu_savevm_send_colo_enable(s
->to_dst_file
);
2960 qemu_savevm_state_setup(s
->to_dst_file
);
2962 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
2963 MIGRATION_STATUS_ACTIVE
);
2965 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
2967 trace_migration_thread_setup_complete();
2969 while (migration_is_active(s
)) {
2970 if (urgent
|| !migration_rate_exceeded(s
->to_dst_file
)) {
2971 MigIterateState iter_state
= migration_iteration_run(s
);
2972 if (iter_state
== MIG_ITERATE_SKIP
) {
2974 } else if (iter_state
== MIG_ITERATE_BREAK
) {
2980 * Try to detect any kind of failures, and see whether we
2981 * should stop the migration now.
2983 thr_error
= migration_detect_error(s
);
2984 if (thr_error
== MIG_THR_ERR_FATAL
) {
2985 /* Stop migration */
2987 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
2989 * Just recovered from a e.g. network failure, reset all
2990 * the local variables. This is important to avoid
2991 * breaking transferred_bytes and bandwidth calculation
2993 update_iteration_initial_status(s
);
2996 urgent
= migration_rate_limit();
2999 trace_migration_thread_after_loop();
3000 migration_iteration_finish(s
);
3001 object_unref(OBJECT(s
));
3002 rcu_unregister_thread();
3003 MigrationThreadDel(thread
);
3007 static void bg_migration_vm_start_bh(void *opaque
)
3009 MigrationState
*s
= opaque
;
3011 qemu_bh_delete(s
->vm_start_bh
);
3012 s
->vm_start_bh
= NULL
;
3015 s
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - s
->downtime_start
;
3019 * Background snapshot thread, based on live migration code.
3020 * This is an alternative implementation of live migration mechanism
3021 * introduced specifically to support background snapshots.
3023 * It takes advantage of userfault_fd write protection mechanism introduced
3024 * in v5.7 kernel. Compared to existing dirty page logging migration much
3025 * lesser stream traffic is produced resulting in smaller snapshot images,
3026 * simply cause of no page duplicates can get into the stream.
3028 * Another key point is that generated vmstate stream reflects machine state
3029 * 'frozen' at the beginning of snapshot creation compared to dirty page logging
3030 * mechanism, which effectively results in that saved snapshot is the state of VM
3031 * at the end of the process.
3033 static void *bg_migration_thread(void *opaque
)
3035 MigrationState
*s
= opaque
;
3036 int64_t setup_start
;
3037 MigThrError thr_error
;
3039 bool early_fail
= true;
3041 rcu_register_thread();
3042 object_ref(OBJECT(s
));
3044 migration_rate_set(RATE_LIMIT_DISABLED
);
3046 setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3048 * We want to save vmstate for the moment when migration has been
3049 * initiated but also we want to save RAM content while VM is running.
3050 * The RAM content should appear first in the vmstate. So, we first
3051 * stash the non-RAM part of the vmstate to the temporary buffer,
3052 * then write RAM part of the vmstate to the migration stream
3053 * with vCPUs running and, finally, write stashed non-RAM part of
3054 * the vmstate from the buffer to the migration stream.
3056 s
->bioc
= qio_channel_buffer_new(512 * 1024);
3057 qio_channel_set_name(QIO_CHANNEL(s
->bioc
), "vmstate-buffer");
3058 fb
= qemu_file_new_output(QIO_CHANNEL(s
->bioc
));
3059 object_unref(OBJECT(s
->bioc
));
3061 update_iteration_initial_status(s
);
3064 * Prepare for tracking memory writes with UFFD-WP - populate
3065 * RAM pages before protecting.
3068 ram_write_tracking_prepare();
3071 qemu_savevm_state_header(s
->to_dst_file
);
3072 qemu_savevm_state_setup(s
->to_dst_file
);
3074 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
3075 MIGRATION_STATUS_ACTIVE
);
3077 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3079 trace_migration_thread_setup_complete();
3080 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3082 qemu_mutex_lock_iothread();
3085 * If VM is currently in suspended state, then, to make a valid runstate
3086 * transition in vm_stop_force_state() we need to wakeup it up.
3088 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3089 s
->vm_was_running
= runstate_is_running();
3091 if (global_state_store()) {
3094 /* Forcibly stop VM before saving state of vCPUs and devices */
3095 if (vm_stop_force_state(RUN_STATE_PAUSED
)) {
3099 * Put vCPUs in sync with shadow context structures, then
3100 * save their state to channel-buffer along with devices.
3102 cpu_synchronize_all_states();
3103 if (qemu_savevm_state_complete_precopy_non_iterable(fb
, false, false)) {
3107 * Since we are going to get non-iterable state data directly
3108 * from s->bioc->data, explicit flush is needed here.
3112 /* Now initialize UFFD context and start tracking RAM writes */
3113 if (ram_write_tracking_start()) {
3119 * Start VM from BH handler to avoid write-fault lock here.
3120 * UFFD-WP protection for the whole RAM is already enabled so
3121 * calling VM state change notifiers from vm_start() would initiate
3122 * writes to virtio VQs memory which is in write-protected region.
3124 s
->vm_start_bh
= qemu_bh_new(bg_migration_vm_start_bh
, s
);
3125 qemu_bh_schedule(s
->vm_start_bh
);
3127 qemu_mutex_unlock_iothread();
3129 while (migration_is_active(s
)) {
3130 MigIterateState iter_state
= bg_migration_iteration_run(s
);
3131 if (iter_state
== MIG_ITERATE_SKIP
) {
3133 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3138 * Try to detect any kind of failures, and see whether we
3139 * should stop the migration now.
3141 thr_error
= migration_detect_error(s
);
3142 if (thr_error
== MIG_THR_ERR_FATAL
) {
3143 /* Stop migration */
3147 migration_update_counters(s
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
));
3150 trace_migration_thread_after_loop();
3154 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
3155 MIGRATION_STATUS_FAILED
);
3156 qemu_mutex_unlock_iothread();
3159 bg_migration_iteration_finish(s
);
3162 object_unref(OBJECT(s
));
3163 rcu_unregister_thread();
3168 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
3170 Error
*local_err
= NULL
;
3171 uint64_t rate_limit
;
3172 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
3175 * If there's a previous error, free it and prepare for another one.
3176 * Meanwhile if migration completes successfully, there won't have an error
3177 * dumped when calling migrate_fd_cleanup().
3179 migrate_error_free(s
);
3181 s
->expected_downtime
= migrate_downtime_limit();
3183 assert(s
->cleanup_bh
);
3185 assert(!s
->cleanup_bh
);
3186 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup_bh
, s
);
3189 migrate_fd_error(s
, error_in
);
3192 * Don't do cleanup for resume if channel is invalid, but only dump
3193 * the error. We wait for another channel connect from the user.
3194 * The error_report still gives HMP user a hint on what failed.
3195 * It's normally done in migrate_fd_cleanup(), but call it here
3198 error_report_err(error_copy(s
->error
));
3200 migrate_fd_cleanup(s
);
3206 /* This is a resumed migration */
3207 rate_limit
= migrate_max_postcopy_bandwidth();
3209 /* This is a fresh new migration */
3210 rate_limit
= migrate_max_bandwidth();
3212 /* Notify before starting migration thread */
3213 notifier_list_notify(&migration_state_notifiers
, s
);
3216 migration_rate_set(rate_limit
);
3217 qemu_file_set_blocking(s
->to_dst_file
, true);
3220 * Open the return path. For postcopy, it is used exclusively. For
3221 * precopy, only if user specified "return-path" capability would
3222 * QEMU uses the return path.
3224 if (migrate_postcopy_ram() || migrate_return_path()) {
3225 if (open_return_path_on_source(s
, !resume
)) {
3226 error_report("Unable to open return-path for postcopy");
3227 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
3228 migrate_fd_cleanup(s
);
3234 * This needs to be done before resuming a postcopy. Note: for newer
3235 * QEMUs we will delay the channel creation until postcopy_start(), to
3236 * avoid disorder of channel creations.
3238 if (migrate_postcopy_preempt() && s
->preempt_pre_7_2
) {
3239 postcopy_preempt_setup(s
);
3243 /* Wakeup the main migration thread to do the recovery */
3244 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
3245 MIGRATION_STATUS_POSTCOPY_RECOVER
);
3246 qemu_sem_post(&s
->postcopy_pause_sem
);
3250 if (multifd_save_setup(&local_err
) != 0) {
3251 error_report_err(local_err
);
3252 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3253 MIGRATION_STATUS_FAILED
);
3254 migrate_fd_cleanup(s
);
3258 if (migrate_background_snapshot()) {
3259 qemu_thread_create(&s
->thread
, "bg_snapshot",
3260 bg_migration_thread
, s
, QEMU_THREAD_JOINABLE
);
3262 qemu_thread_create(&s
->thread
, "live_migration",
3263 migration_thread
, s
, QEMU_THREAD_JOINABLE
);
3265 s
->migration_thread_running
= true;
3268 static void migration_class_init(ObjectClass
*klass
, void *data
)
3270 DeviceClass
*dc
= DEVICE_CLASS(klass
);
3272 dc
->user_creatable
= false;
3273 device_class_set_props(dc
, migration_properties
);
3276 static void migration_instance_finalize(Object
*obj
)
3278 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3280 qemu_mutex_destroy(&ms
->error_mutex
);
3281 qemu_mutex_destroy(&ms
->qemu_file_lock
);
3282 qemu_sem_destroy(&ms
->wait_unplug_sem
);
3283 qemu_sem_destroy(&ms
->rate_limit_sem
);
3284 qemu_sem_destroy(&ms
->pause_sem
);
3285 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
3286 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
3287 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
3288 qemu_sem_destroy(&ms
->rp_state
.rp_pong_acks
);
3289 qemu_sem_destroy(&ms
->postcopy_qemufile_src_sem
);
3290 error_free(ms
->error
);
3293 static void migration_instance_init(Object
*obj
)
3295 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3297 ms
->state
= MIGRATION_STATUS_NONE
;
3299 ms
->pages_per_second
= -1;
3300 qemu_sem_init(&ms
->pause_sem
, 0);
3301 qemu_mutex_init(&ms
->error_mutex
);
3303 migrate_params_init(&ms
->parameters
);
3305 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
3306 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
3307 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
3308 qemu_sem_init(&ms
->rp_state
.rp_pong_acks
, 0);
3309 qemu_sem_init(&ms
->rate_limit_sem
, 0);
3310 qemu_sem_init(&ms
->wait_unplug_sem
, 0);
3311 qemu_sem_init(&ms
->postcopy_qemufile_src_sem
, 0);
3312 qemu_mutex_init(&ms
->qemu_file_lock
);
3316 * Return true if check pass, false otherwise. Error will be put
3317 * inside errp if provided.
3319 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
3321 /* Assuming all off */
3322 bool old_caps
[MIGRATION_CAPABILITY__MAX
] = { 0 };
3324 if (!migrate_params_check(&ms
->parameters
, errp
)) {
3328 return migrate_caps_check(old_caps
, ms
->capabilities
, errp
);
3331 static const TypeInfo migration_type
= {
3332 .name
= TYPE_MIGRATION
,
3334 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3335 * not created using qdev_new(), it is not attached to the qdev
3336 * device tree, and it is never realized.
3338 * TODO: Make this TYPE_OBJECT once QOM provides something like
3339 * TYPE_DEVICE's "-global" properties.
3341 .parent
= TYPE_DEVICE
,
3342 .class_init
= migration_class_init
,
3343 .class_size
= sizeof(MigrationClass
),
3344 .instance_size
= sizeof(MigrationState
),
3345 .instance_init
= migration_instance_init
,
3346 .instance_finalize
= migration_instance_finalize
,
3349 static void register_migration_types(void)
3351 type_register_static(&migration_type
);
3354 type_init(register_migration_types
);