4 * Copyright IBM, Corp. 2008
7 * Anthony Liguori <aliguori@us.ibm.com>
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
16 #include "qemu/osdep.h"
17 #include "qemu/cutils.h"
18 #include "qemu/error-report.h"
19 #include "qemu/main-loop.h"
20 #include "migration/blocker.h"
25 #include "sysemu/runstate.h"
26 #include "sysemu/sysemu.h"
27 #include "sysemu/cpu-throttle.h"
30 #include "ram-compress.h"
31 #include "migration/global_state.h"
32 #include "migration/misc.h"
33 #include "migration.h"
34 #include "migration-stats.h"
36 #include "qemu-file.h"
38 #include "migration/vmstate.h"
39 #include "block/block.h"
40 #include "qapi/error.h"
41 #include "qapi/clone-visitor.h"
42 #include "qapi/qapi-visit-migration.h"
43 #include "qapi/qapi-visit-sockets.h"
44 #include "qapi/qapi-commands-migration.h"
45 #include "qapi/qapi-events-migration.h"
46 #include "qapi/qmp/qerror.h"
47 #include "qapi/qmp/qnull.h"
50 #include "postcopy-ram.h"
51 #include "qemu/thread.h"
53 #include "exec/target_page.h"
54 #include "io/channel-buffer.h"
55 #include "io/channel-tls.h"
56 #include "migration/colo.h"
57 #include "hw/boards.h"
58 #include "monitor/monitor.h"
59 #include "net/announce.h"
60 #include "qemu/queue.h"
62 #include "threadinfo.h"
63 #include "qemu/yank.h"
64 #include "sysemu/cpus.h"
65 #include "yank_functions.h"
66 #include "sysemu/qtest.h"
68 #include "sysemu/dirtylimit.h"
70 static NotifierList migration_state_notifiers
=
71 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
73 /* Messages sent on the return path from destination to source */
74 enum mig_rp_message_type
{
75 MIG_RP_MSG_INVALID
= 0, /* Must be 0 */
76 MIG_RP_MSG_SHUT
, /* sibling will not send any more RP messages */
77 MIG_RP_MSG_PONG
, /* Response to a PING; data (seq: be32 ) */
79 MIG_RP_MSG_REQ_PAGES_ID
, /* data (start: be64, len: be32, id: string) */
80 MIG_RP_MSG_REQ_PAGES
, /* data (start: be64, len: be32) */
81 MIG_RP_MSG_RECV_BITMAP
, /* send recved_bitmap back to source */
82 MIG_RP_MSG_RESUME_ACK
, /* tell source that we are ready to resume */
83 MIG_RP_MSG_SWITCHOVER_ACK
, /* Tell source it's OK to do switchover */
88 /* When we add fault tolerance, we could have several
89 migrations at once. For now we don't need to add
90 dynamic creation of migration */
92 static MigrationState
*current_migration
;
93 static MigrationIncomingState
*current_incoming
;
95 static GSList
*migration_blockers
;
97 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
98 static int migration_maybe_pause(MigrationState
*s
,
99 int *current_active_state
,
101 static void migrate_fd_cancel(MigrationState
*s
);
102 static int close_return_path_on_source(MigrationState
*s
);
104 static bool migration_needs_multiple_sockets(void)
106 return migrate_multifd() || migrate_postcopy_preempt();
109 static bool uri_supports_multi_channels(const char *uri
)
111 return strstart(uri
, "tcp:", NULL
) || strstart(uri
, "unix:", NULL
) ||
112 strstart(uri
, "vsock:", NULL
);
116 migration_channels_and_uri_compatible(const char *uri
, Error
**errp
)
118 if (migration_needs_multiple_sockets() &&
119 !uri_supports_multi_channels(uri
)) {
120 error_setg(errp
, "Migration requires multi-channel URIs (e.g. tcp)");
127 static gint
page_request_addr_cmp(gconstpointer ap
, gconstpointer bp
)
129 uintptr_t a
= (uintptr_t) ap
, b
= (uintptr_t) bp
;
131 return (a
> b
) - (a
< b
);
134 void migration_object_init(void)
136 /* This can only be called once. */
137 assert(!current_migration
);
138 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
141 * Init the migrate incoming object as well no matter whether
142 * we'll use it or not.
144 assert(!current_incoming
);
145 current_incoming
= g_new0(MigrationIncomingState
, 1);
146 current_incoming
->state
= MIGRATION_STATUS_NONE
;
147 current_incoming
->postcopy_remote_fds
=
148 g_array_new(FALSE
, TRUE
, sizeof(struct PostCopyFD
));
149 qemu_mutex_init(¤t_incoming
->rp_mutex
);
150 qemu_mutex_init(¤t_incoming
->postcopy_prio_thread_mutex
);
151 qemu_event_init(¤t_incoming
->main_thread_load_event
, false);
152 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_dst
, 0);
153 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fault
, 0);
154 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fast_load
, 0);
155 qemu_sem_init(¤t_incoming
->postcopy_qemufile_dst_done
, 0);
157 qemu_mutex_init(¤t_incoming
->page_request_mutex
);
158 qemu_cond_init(¤t_incoming
->page_request_cond
);
159 current_incoming
->page_requested
= g_tree_new(page_request_addr_cmp
);
161 migration_object_check(current_migration
, &error_fatal
);
165 dirty_bitmap_mig_init();
168 void migration_cancel(const Error
*error
)
171 migrate_set_error(current_migration
, error
);
173 if (migrate_dirty_limit()) {
174 qmp_cancel_vcpu_dirty_limit(false, -1, NULL
);
176 migrate_fd_cancel(current_migration
);
179 void migration_shutdown(void)
182 * When the QEMU main thread exit, the COLO thread
183 * may wait a semaphore. So, we should wakeup the
184 * COLO thread before migration shutdown.
188 * Cancel the current migration - that will (eventually)
189 * stop the migration using this structure
191 migration_cancel(NULL
);
192 object_unref(OBJECT(current_migration
));
195 * Cancel outgoing migration of dirty bitmaps. It should
196 * at least unref used block nodes.
198 dirty_bitmap_mig_cancel_outgoing();
201 * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
202 * are non-critical data, and their loss never considered as
205 dirty_bitmap_mig_cancel_incoming();
209 MigrationState
*migrate_get_current(void)
211 /* This can only be called after the object created. */
212 assert(current_migration
);
213 return current_migration
;
216 MigrationIncomingState
*migration_incoming_get_current(void)
218 assert(current_incoming
);
219 return current_incoming
;
222 void migration_incoming_transport_cleanup(MigrationIncomingState
*mis
)
224 if (mis
->socket_address_list
) {
225 qapi_free_SocketAddressList(mis
->socket_address_list
);
226 mis
->socket_address_list
= NULL
;
229 if (mis
->transport_cleanup
) {
230 mis
->transport_cleanup(mis
->transport_data
);
231 mis
->transport_data
= mis
->transport_cleanup
= NULL
;
235 void migration_incoming_state_destroy(void)
237 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
239 multifd_load_cleanup();
240 compress_threads_load_cleanup();
242 if (mis
->to_src_file
) {
243 /* Tell source that we are done */
244 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
245 qemu_fclose(mis
->to_src_file
);
246 mis
->to_src_file
= NULL
;
249 if (mis
->from_src_file
) {
250 migration_ioc_unregister_yank_from_file(mis
->from_src_file
);
251 qemu_fclose(mis
->from_src_file
);
252 mis
->from_src_file
= NULL
;
254 if (mis
->postcopy_remote_fds
) {
255 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
256 mis
->postcopy_remote_fds
= NULL
;
259 migration_incoming_transport_cleanup(mis
);
260 qemu_event_reset(&mis
->main_thread_load_event
);
262 if (mis
->page_requested
) {
263 g_tree_destroy(mis
->page_requested
);
264 mis
->page_requested
= NULL
;
267 if (mis
->postcopy_qemufile_dst
) {
268 migration_ioc_unregister_yank_from_file(mis
->postcopy_qemufile_dst
);
269 qemu_fclose(mis
->postcopy_qemufile_dst
);
270 mis
->postcopy_qemufile_dst
= NULL
;
273 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
276 static void migrate_generate_event(int new_state
)
278 if (migrate_events()) {
279 qapi_event_send_migration(new_state
);
284 * Send a message on the return channel back to the source
287 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
288 enum mig_rp_message_type message_type
,
289 uint16_t len
, void *data
)
293 trace_migrate_send_rp_message((int)message_type
, len
);
294 QEMU_LOCK_GUARD(&mis
->rp_mutex
);
297 * It's possible that the file handle got lost due to network
300 if (!mis
->to_src_file
) {
305 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
306 qemu_put_be16(mis
->to_src_file
, len
);
307 qemu_put_buffer(mis
->to_src_file
, data
, len
);
308 return qemu_fflush(mis
->to_src_file
);
311 /* Request one page from the source VM at the given start address.
312 * rb: the RAMBlock to request the page in
313 * Start: Address offset within the RB
314 * Len: Length in bytes required - must be a multiple of pagesize
316 int migrate_send_rp_message_req_pages(MigrationIncomingState
*mis
,
317 RAMBlock
*rb
, ram_addr_t start
)
319 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
320 size_t msglen
= 12; /* start + len */
321 size_t len
= qemu_ram_pagesize(rb
);
322 enum mig_rp_message_type msg_type
;
326 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
327 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
330 * We maintain the last ramblock that we requested for page. Note that we
331 * don't need locking because this function will only be called within the
332 * postcopy ram fault thread.
334 if (rb
!= mis
->last_rb
) {
337 rbname
= qemu_ram_get_idstr(rb
);
338 rbname_len
= strlen(rbname
);
340 assert(rbname_len
< 256);
342 bufc
[msglen
++] = rbname_len
;
343 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
344 msglen
+= rbname_len
;
345 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
347 msg_type
= MIG_RP_MSG_REQ_PAGES
;
350 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
353 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
,
354 RAMBlock
*rb
, ram_addr_t start
, uint64_t haddr
)
356 void *aligned
= (void *)(uintptr_t)ROUND_DOWN(haddr
, qemu_ram_pagesize(rb
));
357 bool received
= false;
359 WITH_QEMU_LOCK_GUARD(&mis
->page_request_mutex
) {
360 received
= ramblock_recv_bitmap_test_byte_offset(rb
, start
);
361 if (!received
&& !g_tree_lookup(mis
->page_requested
, aligned
)) {
363 * The page has not been received, and it's not yet in the page
364 * request list. Queue it. Set the value of element to 1, so that
365 * things like g_tree_lookup() will return TRUE (1) when found.
367 g_tree_insert(mis
->page_requested
, aligned
, (gpointer
)1);
368 qatomic_inc(&mis
->page_requested_count
);
369 trace_postcopy_page_req_add(aligned
, mis
->page_requested_count
);
374 * If the page is there, skip sending the message. We don't even need the
375 * lock because as long as the page arrived, it'll be there forever.
381 return migrate_send_rp_message_req_pages(mis
, rb
, start
);
384 static bool migration_colo_enabled
;
385 bool migration_incoming_colo_enabled(void)
387 return migration_colo_enabled
;
390 void migration_incoming_disable_colo(void)
392 ram_block_discard_disable(false);
393 migration_colo_enabled
= false;
396 int migration_incoming_enable_colo(void)
398 #ifndef CONFIG_REPLICATION
399 error_report("ENABLE_COLO command come in migration stream, but COLO "
400 "module is not built in");
404 if (!migrate_colo()) {
405 error_report("ENABLE_COLO command come in migration stream, but c-colo "
406 "capability is not set");
410 if (ram_block_discard_disable(true)) {
411 error_report("COLO: cannot disable RAM discard");
414 migration_colo_enabled
= true;
418 void migrate_add_address(SocketAddress
*address
)
420 MigrationIncomingState
*mis
= migration_incoming_get_current();
422 QAPI_LIST_PREPEND(mis
->socket_address_list
,
423 QAPI_CLONE(SocketAddress
, address
));
426 static void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
428 const char *p
= NULL
;
429 MigrationIncomingState
*mis
= migration_incoming_get_current();
431 /* URI is not suitable for migration? */
432 if (!migration_channels_and_uri_compatible(uri
, errp
)) {
436 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
437 MIGRATION_STATUS_SETUP
);
439 if (strstart(uri
, "tcp:", &p
) ||
440 strstart(uri
, "unix:", NULL
) ||
441 strstart(uri
, "vsock:", NULL
)) {
442 socket_start_incoming_migration(p
? p
: uri
, errp
);
444 } else if (strstart(uri
, "rdma:", &p
)) {
445 if (migrate_compress()) {
446 error_setg(errp
, "RDMA and compression can't be used together");
449 if (migrate_xbzrle()) {
450 error_setg(errp
, "RDMA and XBZRLE can't be used together");
453 if (migrate_multifd()) {
454 error_setg(errp
, "RDMA and multifd can't be used together");
457 rdma_start_incoming_migration(p
, errp
);
459 } else if (strstart(uri
, "exec:", &p
)) {
460 exec_start_incoming_migration(p
, errp
);
461 } else if (strstart(uri
, "fd:", &p
)) {
462 fd_start_incoming_migration(p
, errp
);
463 } else if (strstart(uri
, "file:", &p
)) {
464 file_start_incoming_migration(p
, errp
);
466 error_setg(errp
, "unknown migration protocol: %s", uri
);
470 static void process_incoming_migration_bh(void *opaque
)
472 Error
*local_err
= NULL
;
473 MigrationIncomingState
*mis
= opaque
;
475 /* If capability late_block_activate is set:
476 * Only fire up the block code now if we're going to restart the
477 * VM, else 'cont' will do it.
478 * This causes file locking to happen; so we don't want it to happen
479 * unless we really are starting the VM.
481 if (!migrate_late_block_activate() ||
482 (autostart
&& (!global_state_received() ||
483 global_state_get_runstate() == RUN_STATE_RUNNING
))) {
484 /* Make sure all file formats throw away their mutable metadata.
485 * If we get an error here, just don't restart the VM yet. */
486 bdrv_activate_all(&local_err
);
488 error_report_err(local_err
);
495 * This must happen after all error conditions are dealt with and
496 * we're sure the VM is going to be running on this host.
498 qemu_announce_self(&mis
->announce_timer
, migrate_announce_params());
500 multifd_load_shutdown();
502 dirty_bitmap_mig_before_vm_start();
504 if (!global_state_received() ||
505 global_state_get_runstate() == RUN_STATE_RUNNING
) {
509 runstate_set(RUN_STATE_PAUSED
);
511 } else if (migration_incoming_colo_enabled()) {
512 migration_incoming_disable_colo();
515 runstate_set(global_state_get_runstate());
518 * This must happen after any state changes since as soon as an external
519 * observer sees this event they might start to prod at the VM assuming
522 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
523 MIGRATION_STATUS_COMPLETED
);
524 qemu_bh_delete(mis
->bh
);
525 migration_incoming_state_destroy();
528 static void coroutine_fn
529 process_incoming_migration_co(void *opaque
)
531 MigrationIncomingState
*mis
= migration_incoming_get_current();
535 assert(mis
->from_src_file
);
537 if (compress_threads_load_setup(mis
->from_src_file
)) {
538 error_report("Failed to setup decompress threads");
542 mis
->largest_page_size
= qemu_ram_pagesize_largest();
543 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
544 migrate_set_state(&mis
->state
, MIGRATION_STATUS_SETUP
,
545 MIGRATION_STATUS_ACTIVE
);
547 mis
->loadvm_co
= qemu_coroutine_self();
548 ret
= qemu_loadvm_state(mis
->from_src_file
);
549 mis
->loadvm_co
= NULL
;
551 ps
= postcopy_state_get();
552 trace_process_incoming_migration_co_end(ret
, ps
);
553 if (ps
!= POSTCOPY_INCOMING_NONE
) {
554 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
556 * Where a migration had postcopy enabled (and thus went to advise)
557 * but managed to complete within the precopy period, we can use
560 postcopy_ram_incoming_cleanup(mis
);
561 } else if (ret
>= 0) {
563 * Postcopy was started, cleanup should happen at the end of the
566 trace_process_incoming_migration_co_postcopy_end_main();
569 /* Else if something went wrong then just fall out of the normal exit */
573 error_report("load of migration failed: %s", strerror(-ret
));
577 if (colo_incoming_co() < 0) {
581 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
582 qemu_bh_schedule(mis
->bh
);
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
;
783 int migrate_send_rp_switchover_ack(MigrationIncomingState
*mis
)
785 return migrate_send_rp_message(mis
, MIG_RP_MSG_SWITCHOVER_ACK
, 0, NULL
);
789 * Send a 'SHUT' message on the return channel with the given value
790 * to indicate that we've finished with the RP. Non-0 value indicates
793 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
798 buf
= cpu_to_be32(value
);
799 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
803 * Send a 'PONG' message on the return channel with the given value
804 * (normally in response to a 'PING')
806 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
811 buf
= cpu_to_be32(value
);
812 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
815 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
823 * First, we send the header part. It contains only the len of
824 * idstr, and the idstr itself.
826 len
= strlen(block_name
);
828 memcpy(buf
+ 1, block_name
, len
);
830 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
831 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
836 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
839 * Next, we dump the received bitmap to the stream.
841 * TODO: currently we are safe since we are the only one that is
842 * using the to_src_file handle (fault thread is still paused),
843 * and it's ok even not taking the mutex. However the best way is
844 * to take the lock before sending the message header, and release
845 * the lock after sending the bitmap.
847 qemu_mutex_lock(&mis
->rp_mutex
);
848 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
849 qemu_mutex_unlock(&mis
->rp_mutex
);
851 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
854 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
858 buf
= cpu_to_be32(value
);
859 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
863 * Return true if we're already in the middle of a migration
864 * (i.e. any of the active or setup states)
866 bool migration_is_setup_or_active(int state
)
869 case MIGRATION_STATUS_ACTIVE
:
870 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
871 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
872 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
873 case MIGRATION_STATUS_SETUP
:
874 case MIGRATION_STATUS_PRE_SWITCHOVER
:
875 case MIGRATION_STATUS_DEVICE
:
876 case MIGRATION_STATUS_WAIT_UNPLUG
:
877 case MIGRATION_STATUS_COLO
:
886 bool migration_is_running(int state
)
889 case MIGRATION_STATUS_ACTIVE
:
890 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
891 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
892 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
893 case MIGRATION_STATUS_SETUP
:
894 case MIGRATION_STATUS_PRE_SWITCHOVER
:
895 case MIGRATION_STATUS_DEVICE
:
896 case MIGRATION_STATUS_WAIT_UNPLUG
:
897 case MIGRATION_STATUS_CANCELLING
:
906 static bool migrate_show_downtime(MigrationState
*s
)
908 return (s
->state
== MIGRATION_STATUS_COMPLETED
) || migration_in_postcopy();
911 static void populate_time_info(MigrationInfo
*info
, MigrationState
*s
)
913 info
->has_status
= true;
914 info
->has_setup_time
= true;
915 info
->setup_time
= s
->setup_time
;
917 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
918 info
->has_total_time
= true;
919 info
->total_time
= s
->total_time
;
921 info
->has_total_time
= true;
922 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) -
926 if (migrate_show_downtime(s
)) {
927 info
->has_downtime
= true;
928 info
->downtime
= s
->downtime
;
930 info
->has_expected_downtime
= true;
931 info
->expected_downtime
= s
->expected_downtime
;
935 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
937 size_t page_size
= qemu_target_page_size();
939 info
->ram
= g_malloc0(sizeof(*info
->ram
));
940 info
->ram
->transferred
= migration_transferred_bytes();
941 info
->ram
->total
= ram_bytes_total();
942 info
->ram
->duplicate
= stat64_get(&mig_stats
.zero_pages
);
943 /* legacy value. It is not used anymore */
944 info
->ram
->skipped
= 0;
945 info
->ram
->normal
= stat64_get(&mig_stats
.normal_pages
);
946 info
->ram
->normal_bytes
= info
->ram
->normal
* page_size
;
947 info
->ram
->mbps
= s
->mbps
;
948 info
->ram
->dirty_sync_count
=
949 stat64_get(&mig_stats
.dirty_sync_count
);
950 info
->ram
->dirty_sync_missed_zero_copy
=
951 stat64_get(&mig_stats
.dirty_sync_missed_zero_copy
);
952 info
->ram
->postcopy_requests
=
953 stat64_get(&mig_stats
.postcopy_requests
);
954 info
->ram
->page_size
= page_size
;
955 info
->ram
->multifd_bytes
= stat64_get(&mig_stats
.multifd_bytes
);
956 info
->ram
->pages_per_second
= s
->pages_per_second
;
957 info
->ram
->precopy_bytes
= stat64_get(&mig_stats
.precopy_bytes
);
958 info
->ram
->downtime_bytes
= stat64_get(&mig_stats
.downtime_bytes
);
959 info
->ram
->postcopy_bytes
= stat64_get(&mig_stats
.postcopy_bytes
);
961 if (migrate_xbzrle()) {
962 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
963 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
964 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
965 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
966 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
967 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
968 info
->xbzrle_cache
->encoding_rate
= xbzrle_counters
.encoding_rate
;
969 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
972 populate_compress(info
);
974 if (cpu_throttle_active()) {
975 info
->has_cpu_throttle_percentage
= true;
976 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
979 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
980 info
->ram
->remaining
= ram_bytes_remaining();
981 info
->ram
->dirty_pages_rate
=
982 stat64_get(&mig_stats
.dirty_pages_rate
);
985 if (migrate_dirty_limit() && dirtylimit_in_service()) {
986 info
->has_dirty_limit_throttle_time_per_round
= true;
987 info
->dirty_limit_throttle_time_per_round
=
988 dirtylimit_throttle_time_per_round();
990 info
->has_dirty_limit_ring_full_time
= true;
991 info
->dirty_limit_ring_full_time
= dirtylimit_ring_full_time();
995 static void populate_disk_info(MigrationInfo
*info
)
997 if (blk_mig_active()) {
998 info
->disk
= g_malloc0(sizeof(*info
->disk
));
999 info
->disk
->transferred
= blk_mig_bytes_transferred();
1000 info
->disk
->remaining
= blk_mig_bytes_remaining();
1001 info
->disk
->total
= blk_mig_bytes_total();
1005 static void fill_source_migration_info(MigrationInfo
*info
)
1007 MigrationState
*s
= migrate_get_current();
1008 int state
= qatomic_read(&s
->state
);
1009 GSList
*cur_blocker
= migration_blockers
;
1011 info
->blocked_reasons
= NULL
;
1014 * There are two types of reasons a migration might be blocked;
1015 * a) devices marked in VMState as non-migratable, and
1016 * b) Explicit migration blockers
1017 * We need to add both of them here.
1019 qemu_savevm_non_migratable_list(&info
->blocked_reasons
);
1021 while (cur_blocker
) {
1022 QAPI_LIST_PREPEND(info
->blocked_reasons
,
1023 g_strdup(error_get_pretty(cur_blocker
->data
)));
1024 cur_blocker
= g_slist_next(cur_blocker
);
1026 info
->has_blocked_reasons
= info
->blocked_reasons
!= NULL
;
1029 case MIGRATION_STATUS_NONE
:
1030 /* no migration has happened ever */
1031 /* do not overwrite destination migration status */
1033 case MIGRATION_STATUS_SETUP
:
1034 info
->has_status
= true;
1035 info
->has_total_time
= false;
1037 case MIGRATION_STATUS_ACTIVE
:
1038 case MIGRATION_STATUS_CANCELLING
:
1039 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1040 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1041 case MIGRATION_STATUS_DEVICE
:
1042 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1043 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1044 /* TODO add some postcopy stats */
1045 populate_time_info(info
, s
);
1046 populate_ram_info(info
, s
);
1047 populate_disk_info(info
);
1048 migration_populate_vfio_info(info
);
1050 case MIGRATION_STATUS_COLO
:
1051 info
->has_status
= true;
1052 /* TODO: display COLO specific information (checkpoint info etc.) */
1054 case MIGRATION_STATUS_COMPLETED
:
1055 populate_time_info(info
, s
);
1056 populate_ram_info(info
, s
);
1057 migration_populate_vfio_info(info
);
1059 case MIGRATION_STATUS_FAILED
:
1060 info
->has_status
= true;
1062 case MIGRATION_STATUS_CANCELLED
:
1063 info
->has_status
= true;
1065 case MIGRATION_STATUS_WAIT_UNPLUG
:
1066 info
->has_status
= true;
1069 info
->status
= state
;
1071 QEMU_LOCK_GUARD(&s
->error_mutex
);
1073 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
1077 static void fill_destination_migration_info(MigrationInfo
*info
)
1079 MigrationIncomingState
*mis
= migration_incoming_get_current();
1081 if (mis
->socket_address_list
) {
1082 info
->has_socket_address
= true;
1083 info
->socket_address
=
1084 QAPI_CLONE(SocketAddressList
, mis
->socket_address_list
);
1087 switch (mis
->state
) {
1088 case MIGRATION_STATUS_NONE
:
1090 case MIGRATION_STATUS_SETUP
:
1091 case MIGRATION_STATUS_CANCELLING
:
1092 case MIGRATION_STATUS_CANCELLED
:
1093 case MIGRATION_STATUS_ACTIVE
:
1094 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1095 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1096 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1097 case MIGRATION_STATUS_FAILED
:
1098 case MIGRATION_STATUS_COLO
:
1099 info
->has_status
= true;
1101 case MIGRATION_STATUS_COMPLETED
:
1102 info
->has_status
= true;
1103 fill_destination_postcopy_migration_info(info
);
1106 info
->status
= mis
->state
;
1109 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1111 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1113 fill_destination_migration_info(info
);
1114 fill_source_migration_info(info
);
1119 void qmp_migrate_start_postcopy(Error
**errp
)
1121 MigrationState
*s
= migrate_get_current();
1123 if (!migrate_postcopy()) {
1124 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1125 " the start of migration");
1129 if (s
->state
== MIGRATION_STATUS_NONE
) {
1130 error_setg(errp
, "Postcopy must be started after migration has been"
1135 * we don't error if migration has finished since that would be racy
1136 * with issuing this command.
1138 qatomic_set(&s
->start_postcopy
, true);
1141 /* shared migration helpers */
1143 void migrate_set_state(int *state
, int old_state
, int new_state
)
1145 assert(new_state
< MIGRATION_STATUS__MAX
);
1146 if (qatomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1147 trace_migrate_set_state(MigrationStatus_str(new_state
));
1148 migrate_generate_event(new_state
);
1152 static void migrate_fd_cleanup(MigrationState
*s
)
1154 qemu_bh_delete(s
->cleanup_bh
);
1155 s
->cleanup_bh
= NULL
;
1157 g_free(s
->hostname
);
1159 json_writer_free(s
->vmdesc
);
1162 qemu_savevm_state_cleanup();
1164 if (s
->to_dst_file
) {
1167 trace_migrate_fd_cleanup();
1168 qemu_mutex_unlock_iothread();
1169 if (s
->migration_thread_running
) {
1170 qemu_thread_join(&s
->thread
);
1171 s
->migration_thread_running
= false;
1173 qemu_mutex_lock_iothread();
1175 multifd_save_cleanup();
1176 qemu_mutex_lock(&s
->qemu_file_lock
);
1177 tmp
= s
->to_dst_file
;
1178 s
->to_dst_file
= NULL
;
1179 qemu_mutex_unlock(&s
->qemu_file_lock
);
1181 * Close the file handle without the lock to make sure the
1182 * critical section won't block for long.
1184 migration_ioc_unregister_yank_from_file(tmp
);
1189 * We already cleaned up to_dst_file, so errors from the return
1190 * path might be due to that, ignore them.
1192 close_return_path_on_source(s
);
1194 assert(!migration_is_active(s
));
1196 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1197 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1198 MIGRATION_STATUS_CANCELLED
);
1202 /* It is used on info migrate. We can't free it */
1203 error_report_err(error_copy(s
->error
));
1205 migration_call_notifiers(s
);
1206 block_cleanup_parameters();
1207 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1210 static void migrate_fd_cleanup_schedule(MigrationState
*s
)
1213 * Ref the state for bh, because it may be called when
1214 * there're already no other refs
1216 object_ref(OBJECT(s
));
1217 qemu_bh_schedule(s
->cleanup_bh
);
1220 static void migrate_fd_cleanup_bh(void *opaque
)
1222 MigrationState
*s
= opaque
;
1223 migrate_fd_cleanup(s
);
1224 object_unref(OBJECT(s
));
1227 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1229 QEMU_LOCK_GUARD(&s
->error_mutex
);
1231 s
->error
= error_copy(error
);
1235 bool migrate_has_error(MigrationState
*s
)
1237 /* The lock is not helpful here, but still follow the rule */
1238 QEMU_LOCK_GUARD(&s
->error_mutex
);
1239 return qatomic_read(&s
->error
);
1242 static void migrate_error_free(MigrationState
*s
)
1244 QEMU_LOCK_GUARD(&s
->error_mutex
);
1246 error_free(s
->error
);
1251 static void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1253 trace_migrate_fd_error(error_get_pretty(error
));
1254 assert(s
->to_dst_file
== NULL
);
1255 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1256 MIGRATION_STATUS_FAILED
);
1257 migrate_set_error(s
, error
);
1260 static void migrate_fd_cancel(MigrationState
*s
)
1264 trace_migrate_fd_cancel();
1266 WITH_QEMU_LOCK_GUARD(&s
->qemu_file_lock
) {
1267 if (s
->rp_state
.from_dst_file
) {
1268 /* shutdown the rp socket, so causing the rp thread to shutdown */
1269 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1274 old_state
= s
->state
;
1275 if (!migration_is_running(old_state
)) {
1278 /* If the migration is paused, kick it out of the pause */
1279 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1280 qemu_sem_post(&s
->pause_sem
);
1282 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1283 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1286 * If we're unlucky the migration code might be stuck somewhere in a
1287 * send/write while the network has failed and is waiting to timeout;
1288 * if we've got shutdown(2) available then we can force it to quit.
1290 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1291 WITH_QEMU_LOCK_GUARD(&s
->qemu_file_lock
) {
1292 if (s
->to_dst_file
) {
1293 qemu_file_shutdown(s
->to_dst_file
);
1297 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1298 Error
*local_err
= NULL
;
1300 bdrv_activate_all(&local_err
);
1302 error_report_err(local_err
);
1304 s
->block_inactive
= false;
1309 void migration_add_notifier(Notifier
*notify
,
1310 void (*func
)(Notifier
*notifier
, void *data
))
1312 notify
->notify
= func
;
1313 notifier_list_add(&migration_state_notifiers
, notify
);
1316 void migration_remove_notifier(Notifier
*notify
)
1318 if (notify
->notify
) {
1319 notifier_remove(notify
);
1320 notify
->notify
= NULL
;
1324 void migration_call_notifiers(MigrationState
*s
)
1326 notifier_list_notify(&migration_state_notifiers
, s
);
1329 bool migration_in_setup(MigrationState
*s
)
1331 return s
->state
== MIGRATION_STATUS_SETUP
;
1334 bool migration_has_finished(MigrationState
*s
)
1336 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1339 bool migration_has_failed(MigrationState
*s
)
1341 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1342 s
->state
== MIGRATION_STATUS_FAILED
);
1345 bool migration_in_postcopy(void)
1347 MigrationState
*s
= migrate_get_current();
1350 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1351 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1352 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1359 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1361 return migration_in_postcopy() && s
->postcopy_after_devices
;
1364 bool migration_in_incoming_postcopy(void)
1366 PostcopyState ps
= postcopy_state_get();
1368 return ps
>= POSTCOPY_INCOMING_DISCARD
&& ps
< POSTCOPY_INCOMING_END
;
1371 bool migration_incoming_postcopy_advised(void)
1373 PostcopyState ps
= postcopy_state_get();
1375 return ps
>= POSTCOPY_INCOMING_ADVISE
&& ps
< POSTCOPY_INCOMING_END
;
1378 bool migration_in_bg_snapshot(void)
1380 MigrationState
*s
= migrate_get_current();
1382 return migrate_background_snapshot() &&
1383 migration_is_setup_or_active(s
->state
);
1386 bool migration_is_idle(void)
1388 MigrationState
*s
= current_migration
;
1395 case MIGRATION_STATUS_NONE
:
1396 case MIGRATION_STATUS_CANCELLED
:
1397 case MIGRATION_STATUS_COMPLETED
:
1398 case MIGRATION_STATUS_FAILED
:
1400 case MIGRATION_STATUS_SETUP
:
1401 case MIGRATION_STATUS_CANCELLING
:
1402 case MIGRATION_STATUS_ACTIVE
:
1403 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1404 case MIGRATION_STATUS_COLO
:
1405 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1406 case MIGRATION_STATUS_DEVICE
:
1407 case MIGRATION_STATUS_WAIT_UNPLUG
:
1409 case MIGRATION_STATUS__MAX
:
1410 g_assert_not_reached();
1416 bool migration_is_active(MigrationState
*s
)
1418 return (s
->state
== MIGRATION_STATUS_ACTIVE
||
1419 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1422 int migrate_init(MigrationState
*s
, Error
**errp
)
1426 ret
= qemu_savevm_state_prepare(errp
);
1432 * Reinitialise all migration state, except
1433 * parameters/capabilities that the user set, and
1438 s
->to_dst_file
= NULL
;
1439 s
->state
= MIGRATION_STATUS_NONE
;
1440 s
->rp_state
.from_dst_file
= NULL
;
1441 s
->rp_state
.error
= false;
1443 s
->pages_per_second
= 0.0;
1445 s
->expected_downtime
= 0;
1447 s
->start_postcopy
= false;
1448 s
->postcopy_after_devices
= false;
1449 s
->migration_thread_running
= false;
1450 error_free(s
->error
);
1455 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1457 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1459 s
->vm_old_state
= -1;
1460 s
->iteration_initial_bytes
= 0;
1461 s
->threshold_size
= 0;
1462 s
->switchover_acked
= false;
1463 s
->rdma_migration
= false;
1465 * set mig_stats memory to zero for a new migration
1467 memset(&mig_stats
, 0, sizeof(mig_stats
));
1468 migration_reset_vfio_bytes_transferred();
1473 int migrate_add_blocker_internal(Error
**reasonp
, Error
**errp
)
1475 /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */
1476 if (runstate_check(RUN_STATE_SAVE_VM
) || !migration_is_idle()) {
1477 error_propagate_prepend(errp
, *reasonp
,
1478 "disallowing migration blocker "
1479 "(migration/snapshot in progress) for: ");
1484 migration_blockers
= g_slist_prepend(migration_blockers
, *reasonp
);
1488 int migrate_add_blocker(Error
**reasonp
, Error
**errp
)
1490 if (only_migratable
) {
1491 error_propagate_prepend(errp
, *reasonp
,
1492 "disallowing migration blocker "
1493 "(--only-migratable) for: ");
1498 return migrate_add_blocker_internal(reasonp
, errp
);
1501 void migrate_del_blocker(Error
**reasonp
)
1504 migration_blockers
= g_slist_remove(migration_blockers
, *reasonp
);
1505 error_free(*reasonp
);
1510 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1512 Error
*local_err
= NULL
;
1513 static bool once
= true;
1516 error_setg(errp
, "The incoming migration has already been started");
1519 if (!runstate_check(RUN_STATE_INMIGRATE
)) {
1520 error_setg(errp
, "'-incoming' was not specified on the command line");
1524 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
1528 qemu_start_incoming_migration(uri
, &local_err
);
1531 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1532 error_propagate(errp
, local_err
);
1539 void qmp_migrate_recover(const char *uri
, Error
**errp
)
1541 MigrationIncomingState
*mis
= migration_incoming_get_current();
1544 * Don't even bother to use ERRP_GUARD() as it _must_ always be set by
1545 * callers (no one should ignore a recover failure); if there is, it's a
1546 * programming error.
1550 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1551 error_setg(errp
, "Migrate recover can only be run "
1552 "when postcopy is paused.");
1556 /* If there's an existing transport, release it */
1557 migration_incoming_transport_cleanup(mis
);
1560 * Note that this call will never start a real migration; it will
1561 * only re-setup the migration stream and poke existing migration
1562 * to continue using that newly established channel.
1564 qemu_start_incoming_migration(uri
, errp
);
1567 void qmp_migrate_pause(Error
**errp
)
1569 MigrationState
*ms
= migrate_get_current();
1570 MigrationIncomingState
*mis
= migration_incoming_get_current();
1573 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1574 /* Source side, during postcopy */
1575 qemu_mutex_lock(&ms
->qemu_file_lock
);
1576 if (ms
->to_dst_file
) {
1577 ret
= qemu_file_shutdown(ms
->to_dst_file
);
1579 qemu_mutex_unlock(&ms
->qemu_file_lock
);
1581 error_setg(errp
, "Failed to pause source migration");
1586 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1587 ret
= qemu_file_shutdown(mis
->from_src_file
);
1589 error_setg(errp
, "Failed to pause destination migration");
1594 error_setg(errp
, "migrate-pause is currently only supported "
1595 "during postcopy-active state");
1598 bool migration_is_blocked(Error
**errp
)
1600 if (qemu_savevm_state_blocked(errp
)) {
1604 if (migration_blockers
) {
1605 error_propagate(errp
, error_copy(migration_blockers
->data
));
1612 /* Returns true if continue to migrate, or false if error detected */
1613 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
1614 bool resume
, Error
**errp
)
1616 Error
*local_err
= NULL
;
1619 warn_report("parameter 'inc' is deprecated;"
1620 " use blockdev-mirror with NBD instead");
1624 warn_report("parameter 'blk' is deprecated;"
1625 " use blockdev-mirror with NBD instead");
1629 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1630 error_setg(errp
, "Cannot resume if there is no "
1631 "paused migration");
1636 * Postcopy recovery won't work well with release-ram
1637 * capability since release-ram will drop the page buffer as
1638 * long as the page is put into the send buffer. So if there
1639 * is a network failure happened, any page buffers that have
1640 * not yet reached the destination VM but have already been
1641 * sent from the source VM will be lost forever. Let's refuse
1642 * the client from resuming such a postcopy migration.
1643 * Luckily release-ram was designed to only be used when src
1644 * and destination VMs are on the same host, so it should be
1647 if (migrate_release_ram()) {
1648 error_setg(errp
, "Postcopy recovery cannot work "
1649 "when release-ram capability is set");
1653 /* This is a resume, skip init status */
1657 if (migration_is_running(s
->state
)) {
1658 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1662 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1663 error_setg(errp
, "Guest is waiting for an incoming migration");
1667 if (runstate_check(RUN_STATE_POSTMIGRATE
)) {
1668 error_setg(errp
, "Can't migrate the vm that was paused due to "
1669 "previous migration");
1673 if (migration_is_blocked(errp
)) {
1677 if (blk
|| blk_inc
) {
1678 if (migrate_colo()) {
1679 error_setg(errp
, "No disk migration is required in COLO mode");
1682 if (migrate_block() || migrate_block_incremental()) {
1683 error_setg(errp
, "Command options are incompatible with "
1684 "current migration capabilities");
1687 if (!migrate_cap_set(MIGRATION_CAPABILITY_BLOCK
, true, &local_err
)) {
1688 error_propagate(errp
, local_err
);
1691 s
->must_remove_block_options
= true;
1695 migrate_set_block_incremental(true);
1698 if (migrate_init(s
, errp
)) {
1705 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1706 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1707 bool has_resume
, bool resume
, Error
**errp
)
1709 bool resume_requested
;
1710 Error
*local_err
= NULL
;
1711 MigrationState
*s
= migrate_get_current();
1712 const char *p
= NULL
;
1714 /* URI is not suitable for migration? */
1715 if (!migration_channels_and_uri_compatible(uri
, errp
)) {
1719 resume_requested
= has_resume
&& resume
;
1720 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
1721 resume_requested
, errp
)) {
1722 /* Error detected, put into errp */
1726 if (!resume_requested
) {
1727 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
1732 if (strstart(uri
, "tcp:", &p
) ||
1733 strstart(uri
, "unix:", NULL
) ||
1734 strstart(uri
, "vsock:", NULL
)) {
1735 socket_start_outgoing_migration(s
, p
? p
: uri
, &local_err
);
1737 } else if (strstart(uri
, "rdma:", &p
)) {
1738 rdma_start_outgoing_migration(s
, p
, &local_err
);
1740 } else if (strstart(uri
, "exec:", &p
)) {
1741 exec_start_outgoing_migration(s
, p
, &local_err
);
1742 } else if (strstart(uri
, "fd:", &p
)) {
1743 fd_start_outgoing_migration(s
, p
, &local_err
);
1744 } else if (strstart(uri
, "file:", &p
)) {
1745 file_start_outgoing_migration(s
, p
, &local_err
);
1747 error_setg(&local_err
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1748 "a valid migration protocol");
1749 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1750 MIGRATION_STATUS_FAILED
);
1751 block_cleanup_parameters();
1755 if (!resume_requested
) {
1756 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1758 migrate_fd_error(s
, local_err
);
1759 error_propagate(errp
, local_err
);
1764 void qmp_migrate_cancel(Error
**errp
)
1766 migration_cancel(NULL
);
1769 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
1771 MigrationState
*s
= migrate_get_current();
1772 if (s
->state
!= state
) {
1773 error_setg(errp
, "Migration not in expected state: %s",
1774 MigrationStatus_str(s
->state
));
1777 qemu_sem_post(&s
->pause_sem
);
1780 /* migration thread support */
1782 * Something bad happened to the RP stream, mark an error
1783 * The caller shall print or trace something to indicate why
1785 static void mark_source_rp_bad(MigrationState
*s
)
1787 s
->rp_state
.error
= true;
1790 void migration_rp_wait(MigrationState
*s
)
1792 qemu_sem_wait(&s
->rp_state
.rp_sem
);
1795 void migration_rp_kick(MigrationState
*s
)
1797 qemu_sem_post(&s
->rp_state
.rp_sem
);
1800 static struct rp_cmd_args
{
1801 ssize_t len
; /* -1 = variable */
1804 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1805 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1806 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1807 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1808 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1809 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
1810 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
1811 [MIG_RP_MSG_SWITCHOVER_ACK
] = { .len
= 0, .name
= "SWITCHOVER_ACK" },
1812 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1816 * Process a request for pages received on the return path,
1817 * We're allowed to send more than requested (e.g. to round to our page size)
1818 * and we don't need to send pages that have already been sent.
1820 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1821 ram_addr_t start
, size_t len
)
1823 long our_host_ps
= qemu_real_host_page_size();
1825 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1828 * Since we currently insist on matching page sizes, just sanity check
1829 * we're being asked for whole host pages.
1831 if (!QEMU_IS_ALIGNED(start
, our_host_ps
) ||
1832 !QEMU_IS_ALIGNED(len
, our_host_ps
)) {
1833 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1834 " len: %zd", __func__
, start
, len
);
1835 mark_source_rp_bad(ms
);
1839 if (ram_save_queue_pages(rbname
, start
, len
)) {
1840 mark_source_rp_bad(ms
);
1844 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
1846 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
1849 error_report("%s: invalid block name '%s'", __func__
, block_name
);
1853 /* Fetch the received bitmap and refresh the dirty bitmap */
1854 return ram_dirty_bitmap_reload(s
, block
);
1857 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
1859 trace_source_return_path_thread_resume_ack(value
);
1861 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
1862 error_report("%s: illegal resume_ack value %"PRIu32
,
1867 /* Now both sides are active. */
1868 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
1869 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1871 /* Notify send thread that time to continue send pages */
1872 migration_rp_kick(s
);
1878 * Release ms->rp_state.from_dst_file (and postcopy_qemufile_src if
1879 * existed) in a safe way.
1881 static void migration_release_dst_files(MigrationState
*ms
)
1885 WITH_QEMU_LOCK_GUARD(&ms
->qemu_file_lock
) {
1887 * Reset the from_dst_file pointer first before releasing it, as we
1888 * can't block within lock section
1890 file
= ms
->rp_state
.from_dst_file
;
1891 ms
->rp_state
.from_dst_file
= NULL
;
1895 * Do the same to postcopy fast path socket too if there is. No
1896 * locking needed because this qemufile should only be managed by
1897 * return path thread.
1899 if (ms
->postcopy_qemufile_src
) {
1900 migration_ioc_unregister_yank_from_file(ms
->postcopy_qemufile_src
);
1901 qemu_file_shutdown(ms
->postcopy_qemufile_src
);
1902 qemu_fclose(ms
->postcopy_qemufile_src
);
1903 ms
->postcopy_qemufile_src
= NULL
;
1910 * Handles messages sent on the return path towards the source VM
1913 static void *source_return_path_thread(void *opaque
)
1915 MigrationState
*ms
= opaque
;
1916 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1917 uint16_t header_len
, header_type
;
1919 uint32_t tmp32
, sibling_error
;
1920 ram_addr_t start
= 0; /* =0 to silence warning */
1921 size_t len
= 0, expected_len
;
1924 trace_source_return_path_thread_entry();
1925 rcu_register_thread();
1927 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1928 migration_is_setup_or_active(ms
->state
)) {
1929 trace_source_return_path_thread_loop_top();
1930 header_type
= qemu_get_be16(rp
);
1931 header_len
= qemu_get_be16(rp
);
1933 if (qemu_file_get_error(rp
)) {
1934 mark_source_rp_bad(ms
);
1938 if (header_type
>= MIG_RP_MSG_MAX
||
1939 header_type
== MIG_RP_MSG_INVALID
) {
1940 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1941 header_type
, header_len
);
1942 mark_source_rp_bad(ms
);
1946 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1947 header_len
!= rp_cmd_args
[header_type
].len
) ||
1948 header_len
> sizeof(buf
)) {
1949 error_report("RP: Received '%s' message (0x%04x) with"
1950 "incorrect length %d expecting %zu",
1951 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1952 (size_t)rp_cmd_args
[header_type
].len
);
1953 mark_source_rp_bad(ms
);
1957 /* We know we've got a valid header by this point */
1958 res
= qemu_get_buffer(rp
, buf
, header_len
);
1959 if (res
!= header_len
) {
1960 error_report("RP: Failed reading data for message 0x%04x"
1961 " read %d expected %d",
1962 header_type
, res
, header_len
);
1963 mark_source_rp_bad(ms
);
1967 /* OK, we have the message and the data */
1968 switch (header_type
) {
1969 case MIG_RP_MSG_SHUT
:
1970 sibling_error
= ldl_be_p(buf
);
1971 trace_source_return_path_thread_shut(sibling_error
);
1972 if (sibling_error
) {
1973 error_report("RP: Sibling indicated error %d", sibling_error
);
1974 mark_source_rp_bad(ms
);
1977 * We'll let the main thread deal with closing the RP
1978 * we could do a shutdown(2) on it, but we're the only user
1979 * anyway, so there's nothing gained.
1983 case MIG_RP_MSG_PONG
:
1984 tmp32
= ldl_be_p(buf
);
1985 trace_source_return_path_thread_pong(tmp32
);
1986 qemu_sem_post(&ms
->rp_state
.rp_pong_acks
);
1989 case MIG_RP_MSG_REQ_PAGES
:
1990 start
= ldq_be_p(buf
);
1991 len
= ldl_be_p(buf
+ 8);
1992 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1995 case MIG_RP_MSG_REQ_PAGES_ID
:
1996 expected_len
= 12 + 1; /* header + termination */
1998 if (header_len
>= expected_len
) {
1999 start
= ldq_be_p(buf
);
2000 len
= ldl_be_p(buf
+ 8);
2001 /* Now we expect an idstr */
2002 tmp32
= buf
[12]; /* Length of the following idstr */
2003 buf
[13 + tmp32
] = '\0';
2004 expected_len
+= tmp32
;
2006 if (header_len
!= expected_len
) {
2007 error_report("RP: Req_Page_id with length %d expecting %zd",
2008 header_len
, expected_len
);
2009 mark_source_rp_bad(ms
);
2012 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
2015 case MIG_RP_MSG_RECV_BITMAP
:
2016 if (header_len
< 1) {
2017 error_report("%s: missing block name", __func__
);
2018 mark_source_rp_bad(ms
);
2021 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2022 buf
[buf
[0] + 1] = '\0';
2023 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
2024 mark_source_rp_bad(ms
);
2029 case MIG_RP_MSG_RESUME_ACK
:
2030 tmp32
= ldl_be_p(buf
);
2031 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
2032 mark_source_rp_bad(ms
);
2037 case MIG_RP_MSG_SWITCHOVER_ACK
:
2038 ms
->switchover_acked
= true;
2039 trace_source_return_path_thread_switchover_acked();
2048 if (qemu_file_get_error(rp
)) {
2049 trace_source_return_path_thread_bad_end();
2050 mark_source_rp_bad(ms
);
2053 trace_source_return_path_thread_end();
2054 rcu_unregister_thread();
2058 static int open_return_path_on_source(MigrationState
*ms
)
2060 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
2061 if (!ms
->rp_state
.from_dst_file
) {
2065 trace_open_return_path_on_source();
2067 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
2068 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
2069 ms
->rp_state
.rp_thread_created
= true;
2071 trace_open_return_path_on_source_continue();
2076 static int close_return_path_on_source(MigrationState
*ms
)
2080 if (!ms
->rp_state
.rp_thread_created
) {
2084 trace_migration_return_path_end_before();
2087 * If this is a normal exit then the destination will send a SHUT
2088 * and the rp_thread will exit, however if there's an error we
2089 * need to cause it to exit. shutdown(2), if we have it, will
2090 * cause it to unblock if it's stuck waiting for the destination.
2092 WITH_QEMU_LOCK_GUARD(&ms
->qemu_file_lock
) {
2093 if (ms
->to_dst_file
&& ms
->rp_state
.from_dst_file
&&
2094 qemu_file_get_error(ms
->to_dst_file
)) {
2095 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
2099 trace_await_return_path_close_on_source_joining();
2100 qemu_thread_join(&ms
->rp_state
.rp_thread
);
2101 ms
->rp_state
.rp_thread_created
= false;
2102 trace_await_return_path_close_on_source_close();
2104 ret
= ms
->rp_state
.error
;
2105 ms
->rp_state
.error
= false;
2107 migration_release_dst_files(ms
);
2109 trace_migration_return_path_end_after(ret
);
2114 migration_wait_main_channel(MigrationState
*ms
)
2116 /* Wait until one PONG message received */
2117 qemu_sem_wait(&ms
->rp_state
.rp_pong_acks
);
2121 * Switch from normal iteration to postcopy
2122 * Returns non-0 on error
2124 static int postcopy_start(MigrationState
*ms
, Error
**errp
)
2127 QIOChannelBuffer
*bioc
;
2129 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2130 uint64_t bandwidth
= migrate_max_postcopy_bandwidth();
2131 bool restart_block
= false;
2132 int cur_state
= MIGRATION_STATUS_ACTIVE
;
2134 if (migrate_postcopy_preempt()) {
2135 migration_wait_main_channel(ms
);
2136 if (postcopy_preempt_establish_channel(ms
)) {
2137 migrate_set_state(&ms
->state
, ms
->state
, MIGRATION_STATUS_FAILED
);
2142 if (!migrate_pause_before_switchover()) {
2143 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
2144 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2147 trace_postcopy_start();
2148 qemu_mutex_lock_iothread();
2149 trace_postcopy_start_set_run();
2151 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2152 global_state_store();
2153 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2158 ret
= migration_maybe_pause(ms
, &cur_state
,
2159 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2164 ret
= bdrv_inactivate_all();
2168 restart_block
= true;
2171 * Cause any non-postcopiable, but iterative devices to
2172 * send out their final data.
2174 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2177 * in Finish migrate and with the io-lock held everything should
2178 * be quiet, but we've potentially still got dirty pages and we
2179 * need to tell the destination to throw any pages it's already received
2182 if (migrate_postcopy_ram()) {
2183 ram_postcopy_send_discard_bitmap(ms
);
2187 * send rest of state - note things that are doing postcopy
2188 * will notice we're in POSTCOPY_ACTIVE and not actually
2189 * wrap their state up here
2191 migration_rate_set(bandwidth
);
2192 if (migrate_postcopy_ram()) {
2193 /* Ping just for debugging, helps line traces up */
2194 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
2198 * While loading the device state we may trigger page transfer
2199 * requests and the fd must be free to process those, and thus
2200 * the destination must read the whole device state off the fd before
2201 * it starts processing it. Unfortunately the ad-hoc migration format
2202 * doesn't allow the destination to know the size to read without fully
2203 * parsing it through each devices load-state code (especially the open
2204 * coded devices that use get/put).
2205 * So we wrap the device state up in a package with a length at the start;
2206 * to do this we use a qemu_buf to hold the whole of the device state.
2208 bioc
= qio_channel_buffer_new(4096);
2209 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
2210 fb
= qemu_file_new_output(QIO_CHANNEL(bioc
));
2211 object_unref(OBJECT(bioc
));
2214 * Make sure the receiver can get incoming pages before we send the rest
2217 qemu_savevm_send_postcopy_listen(fb
);
2219 qemu_savevm_state_complete_precopy(fb
, false, false);
2220 if (migrate_postcopy_ram()) {
2221 qemu_savevm_send_ping(fb
, 3);
2224 qemu_savevm_send_postcopy_run(fb
);
2226 /* <><> end of stuff going into the package */
2228 /* Last point of recovery; as soon as we send the package the destination
2229 * can open devices and potentially start running.
2230 * Lets just check again we've not got any errors.
2232 ret
= qemu_file_get_error(ms
->to_dst_file
);
2234 error_setg(errp
, "postcopy_start: Migration stream errored (pre package)");
2238 restart_block
= false;
2240 /* Now send that blob */
2241 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2246 /* Send a notify to give a chance for anything that needs to happen
2247 * at the transition to postcopy and after the device state; in particular
2248 * spice needs to trigger a transition now
2250 ms
->postcopy_after_devices
= true;
2251 migration_call_notifiers(ms
);
2253 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2255 qemu_mutex_unlock_iothread();
2257 if (migrate_postcopy_ram()) {
2259 * Although this ping is just for debug, it could potentially be
2260 * used for getting a better measurement of downtime at the source.
2262 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2265 if (migrate_release_ram()) {
2266 ram_postcopy_migrated_memory_release(ms
);
2269 ret
= qemu_file_get_error(ms
->to_dst_file
);
2271 error_setg(errp
, "postcopy_start: Migration stream errored");
2272 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2273 MIGRATION_STATUS_FAILED
);
2276 trace_postcopy_preempt_enabled(migrate_postcopy_preempt());
2283 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2284 MIGRATION_STATUS_FAILED
);
2285 if (restart_block
) {
2286 /* A failure happened early enough that we know the destination hasn't
2287 * accessed block devices, so we're safe to recover.
2289 Error
*local_err
= NULL
;
2291 bdrv_activate_all(&local_err
);
2293 error_report_err(local_err
);
2296 qemu_mutex_unlock_iothread();
2301 * migration_maybe_pause: Pause if required to by
2302 * migrate_pause_before_switchover called with the iothread locked
2303 * Returns: 0 on success
2305 static int migration_maybe_pause(MigrationState
*s
,
2306 int *current_active_state
,
2309 if (!migrate_pause_before_switchover()) {
2313 /* Since leaving this state is not atomic with posting the semaphore
2314 * it's possible that someone could have issued multiple migrate_continue
2315 * and the semaphore is incorrectly positive at this point;
2316 * the docs say it's undefined to reinit a semaphore that's already
2317 * init'd, so use timedwait to eat up any existing posts.
2319 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2320 /* This block intentionally left blank */
2324 * If the migration is cancelled when it is in the completion phase,
2325 * the migration state is set to MIGRATION_STATUS_CANCELLING.
2326 * So we don't need to wait a semaphore, otherwise we would always
2327 * wait for the 'pause_sem' semaphore.
2329 if (s
->state
!= MIGRATION_STATUS_CANCELLING
) {
2330 qemu_mutex_unlock_iothread();
2331 migrate_set_state(&s
->state
, *current_active_state
,
2332 MIGRATION_STATUS_PRE_SWITCHOVER
);
2333 qemu_sem_wait(&s
->pause_sem
);
2334 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2336 *current_active_state
= new_state
;
2337 qemu_mutex_lock_iothread();
2340 return s
->state
== new_state
? 0 : -EINVAL
;
2343 static int migration_completion_precopy(MigrationState
*s
,
2344 int *current_active_state
)
2348 qemu_mutex_lock_iothread();
2349 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2350 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2352 s
->vm_old_state
= runstate_get();
2353 global_state_store();
2355 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2356 trace_migration_completion_vm_stop(ret
);
2361 ret
= migration_maybe_pause(s
, current_active_state
,
2362 MIGRATION_STATUS_DEVICE
);
2368 * Inactivate disks except in COLO, and track that we have done so in order
2369 * to remember to reactivate them if migration fails or is cancelled.
2371 s
->block_inactive
= !migrate_colo();
2372 migration_rate_set(RATE_LIMIT_DISABLED
);
2373 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
2376 qemu_mutex_unlock_iothread();
2380 static void migration_completion_postcopy(MigrationState
*s
)
2382 trace_migration_completion_postcopy_end();
2384 qemu_mutex_lock_iothread();
2385 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
2386 qemu_mutex_unlock_iothread();
2389 * Shutdown the postcopy fast path thread. This is only needed when dest
2390 * QEMU binary is old (7.1/7.2). QEMU 8.0+ doesn't need this.
2392 if (migrate_postcopy_preempt() && s
->preempt_pre_7_2
) {
2393 postcopy_preempt_shutdown_file(s
);
2396 trace_migration_completion_postcopy_end_after_complete();
2399 static void migration_completion_failed(MigrationState
*s
,
2400 int current_active_state
)
2402 if (s
->block_inactive
&& (s
->state
== MIGRATION_STATUS_ACTIVE
||
2403 s
->state
== MIGRATION_STATUS_DEVICE
)) {
2405 * If not doing postcopy, vm_start() will be called: let's
2406 * regain control on images.
2408 Error
*local_err
= NULL
;
2410 qemu_mutex_lock_iothread();
2411 bdrv_activate_all(&local_err
);
2413 error_report_err(local_err
);
2415 s
->block_inactive
= false;
2417 qemu_mutex_unlock_iothread();
2420 migrate_set_state(&s
->state
, current_active_state
,
2421 MIGRATION_STATUS_FAILED
);
2425 * migration_completion: Used by migration_thread when there's not much left.
2426 * The caller 'breaks' the loop when this returns.
2428 * @s: Current migration state
2430 static void migration_completion(MigrationState
*s
)
2433 int current_active_state
= s
->state
;
2435 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2436 ret
= migration_completion_precopy(s
, ¤t_active_state
);
2437 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2438 migration_completion_postcopy(s
);
2447 if (close_return_path_on_source(s
)) {
2451 if (qemu_file_get_error(s
->to_dst_file
)) {
2452 trace_migration_completion_file_err();
2456 if (migrate_colo() && s
->state
== MIGRATION_STATUS_ACTIVE
) {
2457 /* COLO does not support postcopy */
2458 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
2459 MIGRATION_STATUS_COLO
);
2461 migrate_set_state(&s
->state
, current_active_state
,
2462 MIGRATION_STATUS_COMPLETED
);
2468 migration_completion_failed(s
, current_active_state
);
2472 * bg_migration_completion: Used by bg_migration_thread when after all the
2473 * RAM has been saved. The caller 'breaks' the loop when this returns.
2475 * @s: Current migration state
2477 static void bg_migration_completion(MigrationState
*s
)
2479 int current_active_state
= s
->state
;
2481 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2483 * By this moment we have RAM content saved into the migration stream.
2484 * The next step is to flush the non-RAM content (device state)
2485 * right after the ram content. The device state has been stored into
2486 * the temporary buffer before RAM saving started.
2488 qemu_put_buffer(s
->to_dst_file
, s
->bioc
->data
, s
->bioc
->usage
);
2489 qemu_fflush(s
->to_dst_file
);
2490 } else if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
2494 if (qemu_file_get_error(s
->to_dst_file
)) {
2495 trace_migration_completion_file_err();
2499 migrate_set_state(&s
->state
, current_active_state
,
2500 MIGRATION_STATUS_COMPLETED
);
2504 migrate_set_state(&s
->state
, current_active_state
,
2505 MIGRATION_STATUS_FAILED
);
2508 typedef enum MigThrError
{
2509 /* No error detected */
2510 MIG_THR_ERR_NONE
= 0,
2511 /* Detected error, but resumed successfully */
2512 MIG_THR_ERR_RECOVERED
= 1,
2513 /* Detected fatal error, need to exit */
2514 MIG_THR_ERR_FATAL
= 2,
2517 static int postcopy_resume_handshake(MigrationState
*s
)
2519 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
2521 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2522 migration_rp_wait(s
);
2525 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2532 /* Return zero if success, or <0 for error */
2533 static int postcopy_do_resume(MigrationState
*s
)
2538 * Call all the resume_prepare() hooks, so that modules can be
2539 * ready for the migration resume.
2541 ret
= qemu_savevm_state_resume_prepare(s
);
2543 error_report("%s: resume_prepare() failure detected: %d",
2549 * If preempt is enabled, re-establish the preempt channel. Note that
2550 * we do it after resume prepare to make sure the main channel will be
2551 * created before the preempt channel. E.g. with weak network, the
2552 * dest QEMU may get messed up with the preempt and main channels on
2553 * the order of connection setup. This guarantees the correct order.
2555 ret
= postcopy_preempt_establish_channel(s
);
2557 error_report("%s: postcopy_preempt_establish_channel(): %d",
2563 * Last handshake with destination on the resume (destination will
2564 * switch to postcopy-active afterwards)
2566 ret
= postcopy_resume_handshake(s
);
2568 error_report("%s: handshake failed: %d", __func__
, ret
);
2576 * We don't return until we are in a safe state to continue current
2577 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2578 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2580 static MigThrError
postcopy_pause(MigrationState
*s
)
2582 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2588 * Current channel is possibly broken. Release it. Note that this is
2589 * guaranteed even without lock because to_dst_file should only be
2590 * modified by the migration thread. That also guarantees that the
2591 * unregister of yank is safe too without the lock. It should be safe
2592 * even to be within the qemu_file_lock, but we didn't do that to avoid
2593 * taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make
2594 * the qemu_file_lock critical section as small as possible.
2596 assert(s
->to_dst_file
);
2597 migration_ioc_unregister_yank_from_file(s
->to_dst_file
);
2598 qemu_mutex_lock(&s
->qemu_file_lock
);
2599 file
= s
->to_dst_file
;
2600 s
->to_dst_file
= NULL
;
2601 qemu_mutex_unlock(&s
->qemu_file_lock
);
2603 qemu_file_shutdown(file
);
2607 * We're already pausing, so ignore any errors on the return
2608 * path and just wait for the thread to finish. It will be
2609 * re-created when we resume.
2611 close_return_path_on_source(s
);
2613 migrate_set_state(&s
->state
, s
->state
,
2614 MIGRATION_STATUS_POSTCOPY_PAUSED
);
2616 error_report("Detected IO failure for postcopy. "
2617 "Migration paused.");
2620 * We wait until things fixed up. Then someone will setup the
2621 * status back for us.
2623 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2624 qemu_sem_wait(&s
->postcopy_pause_sem
);
2627 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2628 /* Woken up by a recover procedure. Give it a shot */
2630 /* Do the resume logic */
2631 if (postcopy_do_resume(s
) == 0) {
2632 /* Let's continue! */
2633 trace_postcopy_pause_continued();
2634 return MIG_THR_ERR_RECOVERED
;
2637 * Something wrong happened during the recovery, let's
2638 * pause again. Pause is always better than throwing
2644 /* This is not right... Time to quit. */
2645 return MIG_THR_ERR_FATAL
;
2650 static MigThrError
migration_detect_error(MigrationState
*s
)
2653 int state
= s
->state
;
2654 Error
*local_error
= NULL
;
2656 if (state
== MIGRATION_STATUS_CANCELLING
||
2657 state
== MIGRATION_STATUS_CANCELLED
) {
2658 /* End the migration, but don't set the state to failed */
2659 return MIG_THR_ERR_FATAL
;
2663 * Try to detect any file errors. Note that postcopy_qemufile_src will
2664 * be NULL when postcopy preempt is not enabled.
2666 ret
= qemu_file_get_error_obj_any(s
->to_dst_file
,
2667 s
->postcopy_qemufile_src
,
2670 /* Everything is fine */
2671 assert(!local_error
);
2672 return MIG_THR_ERR_NONE
;
2676 migrate_set_error(s
, local_error
);
2677 error_free(local_error
);
2680 if (state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
) {
2682 * For postcopy, we allow the network to be down for a
2683 * while. After that, it can be continued by a
2686 return postcopy_pause(s
);
2689 * For precopy (or postcopy with error outside IO), we fail
2692 migrate_set_state(&s
->state
, state
, MIGRATION_STATUS_FAILED
);
2693 trace_migration_thread_file_err();
2695 /* Time to stop the migration, now. */
2696 return MIG_THR_ERR_FATAL
;
2700 static void migration_calculate_complete(MigrationState
*s
)
2702 uint64_t bytes
= migration_transferred_bytes();
2703 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2704 int64_t transfer_time
;
2706 s
->total_time
= end_time
- s
->start_time
;
2709 * It's still not set, so we are precopy migration. For
2710 * postcopy, downtime is calculated during postcopy_start().
2712 s
->downtime
= end_time
- s
->downtime_start
;
2715 transfer_time
= s
->total_time
- s
->setup_time
;
2716 if (transfer_time
) {
2717 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
2721 static void update_iteration_initial_status(MigrationState
*s
)
2724 * Update these three fields at the same time to avoid mismatch info lead
2725 * wrong speed calculation.
2727 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2728 s
->iteration_initial_bytes
= migration_transferred_bytes();
2729 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
2732 static void migration_update_counters(MigrationState
*s
,
2733 int64_t current_time
)
2735 uint64_t transferred
, transferred_pages
, time_spent
;
2736 uint64_t current_bytes
; /* bytes transferred since the beginning */
2737 uint64_t switchover_bw
;
2738 /* Expected bandwidth when switching over to destination QEMU */
2739 double expected_bw_per_ms
;
2742 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
2746 switchover_bw
= migrate_avail_switchover_bandwidth();
2747 current_bytes
= migration_transferred_bytes();
2748 transferred
= current_bytes
- s
->iteration_initial_bytes
;
2749 time_spent
= current_time
- s
->iteration_start_time
;
2750 bandwidth
= (double)transferred
/ time_spent
;
2752 if (switchover_bw
) {
2754 * If the user specified a switchover bandwidth, let's trust the
2755 * user so that can be more accurate than what we estimated.
2757 expected_bw_per_ms
= switchover_bw
/ 1000;
2759 /* If the user doesn't specify bandwidth, we use the estimated */
2760 expected_bw_per_ms
= bandwidth
;
2763 s
->threshold_size
= expected_bw_per_ms
* migrate_downtime_limit();
2765 s
->mbps
= (((double) transferred
* 8.0) /
2766 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2768 transferred_pages
= ram_get_total_transferred_pages() -
2769 s
->iteration_initial_pages
;
2770 s
->pages_per_second
= (double) transferred_pages
/
2771 (((double) time_spent
/ 1000.0));
2774 * if we haven't sent anything, we don't want to
2775 * recalculate. 10000 is a small enough number for our purposes
2777 if (stat64_get(&mig_stats
.dirty_pages_rate
) &&
2778 transferred
> 10000) {
2779 s
->expected_downtime
=
2780 stat64_get(&mig_stats
.dirty_bytes_last_sync
) / expected_bw_per_ms
;
2783 migration_rate_reset();
2785 update_iteration_initial_status(s
);
2787 trace_migrate_transferred(transferred
, time_spent
,
2788 /* Both in unit bytes/ms */
2789 bandwidth
, switchover_bw
/ 1000,
2793 static bool migration_can_switchover(MigrationState
*s
)
2795 if (!migrate_switchover_ack()) {
2799 /* No reason to wait for switchover ACK if VM is stopped */
2800 if (!runstate_is_running()) {
2804 return s
->switchover_acked
;
2807 /* Migration thread iteration status */
2809 MIG_ITERATE_RESUME
, /* Resume current iteration */
2810 MIG_ITERATE_SKIP
, /* Skip current iteration */
2811 MIG_ITERATE_BREAK
, /* Break the loop */
2815 * Return true if continue to the next iteration directly, false
2818 static MigIterateState
migration_iteration_run(MigrationState
*s
)
2820 uint64_t must_precopy
, can_postcopy
;
2821 Error
*local_err
= NULL
;
2822 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
2823 bool can_switchover
= migration_can_switchover(s
);
2825 qemu_savevm_state_pending_estimate(&must_precopy
, &can_postcopy
);
2826 uint64_t pending_size
= must_precopy
+ can_postcopy
;
2828 trace_migrate_pending_estimate(pending_size
, must_precopy
, can_postcopy
);
2830 if (must_precopy
<= s
->threshold_size
) {
2831 qemu_savevm_state_pending_exact(&must_precopy
, &can_postcopy
);
2832 pending_size
= must_precopy
+ can_postcopy
;
2833 trace_migrate_pending_exact(pending_size
, must_precopy
, can_postcopy
);
2836 if ((!pending_size
|| pending_size
< s
->threshold_size
) && can_switchover
) {
2837 trace_migration_thread_low_pending(pending_size
);
2838 migration_completion(s
);
2839 return MIG_ITERATE_BREAK
;
2842 /* Still a significant amount to transfer */
2843 if (!in_postcopy
&& must_precopy
<= s
->threshold_size
&& can_switchover
&&
2844 qatomic_read(&s
->start_postcopy
)) {
2845 if (postcopy_start(s
, &local_err
)) {
2846 migrate_set_error(s
, local_err
);
2847 error_report_err(local_err
);
2849 return MIG_ITERATE_SKIP
;
2852 /* Just another iteration step */
2853 qemu_savevm_state_iterate(s
->to_dst_file
, in_postcopy
);
2854 return MIG_ITERATE_RESUME
;
2857 static void migration_iteration_finish(MigrationState
*s
)
2859 /* If we enabled cpu throttling for auto-converge, turn it off. */
2860 cpu_throttle_stop();
2862 qemu_mutex_lock_iothread();
2864 case MIGRATION_STATUS_COMPLETED
:
2865 migration_calculate_complete(s
);
2866 runstate_set(RUN_STATE_POSTMIGRATE
);
2868 case MIGRATION_STATUS_COLO
:
2869 assert(migrate_colo());
2870 migrate_start_colo_process(s
);
2871 s
->vm_old_state
= RUN_STATE_RUNNING
;
2873 case MIGRATION_STATUS_FAILED
:
2874 case MIGRATION_STATUS_CANCELLED
:
2875 case MIGRATION_STATUS_CANCELLING
:
2876 if (s
->vm_old_state
== RUN_STATE_RUNNING
) {
2877 if (!runstate_check(RUN_STATE_SHUTDOWN
)) {
2881 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2882 runstate_set(s
->vm_old_state
);
2888 /* Should not reach here, but if so, forgive the VM. */
2889 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
2892 migrate_fd_cleanup_schedule(s
);
2893 qemu_mutex_unlock_iothread();
2896 static void bg_migration_iteration_finish(MigrationState
*s
)
2899 * Stop tracking RAM writes - un-protect memory, un-register UFFD
2900 * memory ranges, flush kernel wait queues and wake up threads
2901 * waiting for write fault to be resolved.
2903 ram_write_tracking_stop();
2905 qemu_mutex_lock_iothread();
2907 case MIGRATION_STATUS_COMPLETED
:
2908 migration_calculate_complete(s
);
2911 case MIGRATION_STATUS_ACTIVE
:
2912 case MIGRATION_STATUS_FAILED
:
2913 case MIGRATION_STATUS_CANCELLED
:
2914 case MIGRATION_STATUS_CANCELLING
:
2918 /* Should not reach here, but if so, forgive the VM. */
2919 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
2923 migrate_fd_cleanup_schedule(s
);
2924 qemu_mutex_unlock_iothread();
2928 * Return true if continue to the next iteration directly, false
2931 static MigIterateState
bg_migration_iteration_run(MigrationState
*s
)
2935 res
= qemu_savevm_state_iterate(s
->to_dst_file
, false);
2937 bg_migration_completion(s
);
2938 return MIG_ITERATE_BREAK
;
2941 return MIG_ITERATE_RESUME
;
2944 void migration_make_urgent_request(void)
2946 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
2949 void migration_consume_urgent_request(void)
2951 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
2954 /* Returns true if the rate limiting was broken by an urgent request */
2955 bool migration_rate_limit(void)
2957 int64_t now
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2958 MigrationState
*s
= migrate_get_current();
2960 bool urgent
= false;
2961 migration_update_counters(s
, now
);
2962 if (migration_rate_exceeded(s
->to_dst_file
)) {
2964 if (qemu_file_get_error(s
->to_dst_file
)) {
2968 * Wait for a delay to do rate limiting OR
2969 * something urgent to post the semaphore.
2971 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- now
;
2972 trace_migration_rate_limit_pre(ms
);
2973 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
2975 * We were woken by one or more urgent things but
2976 * the timedwait will have consumed one of them.
2977 * The service routine for the urgent wake will dec
2978 * the semaphore itself for each item it consumes,
2979 * so add this one we just eat back.
2981 qemu_sem_post(&s
->rate_limit_sem
);
2984 trace_migration_rate_limit_post(urgent
);
2990 * if failover devices are present, wait they are completely
2994 static void qemu_savevm_wait_unplug(MigrationState
*s
, int old_state
,
2997 if (qemu_savevm_state_guest_unplug_pending()) {
2998 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_WAIT_UNPLUG
);
3000 while (s
->state
== MIGRATION_STATUS_WAIT_UNPLUG
&&
3001 qemu_savevm_state_guest_unplug_pending()) {
3002 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
3004 if (s
->state
!= MIGRATION_STATUS_WAIT_UNPLUG
) {
3005 int timeout
= 120; /* 30 seconds */
3007 * migration has been canceled
3008 * but as we have started an unplug we must wait the end
3009 * to be able to plug back the card
3011 while (timeout
-- && qemu_savevm_state_guest_unplug_pending()) {
3012 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
3014 if (qemu_savevm_state_guest_unplug_pending() &&
3016 warn_report("migration: partially unplugged device on "
3021 migrate_set_state(&s
->state
, MIGRATION_STATUS_WAIT_UNPLUG
, new_state
);
3023 migrate_set_state(&s
->state
, old_state
, new_state
);
3028 * Master migration thread on the source VM.
3029 * It drives the migration and pumps the data down the outgoing channel.
3031 static void *migration_thread(void *opaque
)
3033 MigrationState
*s
= opaque
;
3034 MigrationThread
*thread
= NULL
;
3035 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3036 MigThrError thr_error
;
3037 bool urgent
= false;
3039 thread
= migration_threads_add("live_migration", qemu_get_thread_id());
3041 rcu_register_thread();
3043 object_ref(OBJECT(s
));
3044 update_iteration_initial_status(s
);
3046 qemu_mutex_lock_iothread();
3047 qemu_savevm_state_header(s
->to_dst_file
);
3048 qemu_mutex_unlock_iothread();
3051 * If we opened the return path, we need to make sure dst has it
3054 if (s
->rp_state
.rp_thread_created
) {
3055 /* Now tell the dest that it should open its end so it can reply */
3056 qemu_savevm_send_open_return_path(s
->to_dst_file
);
3058 /* And do a ping that will make stuff easier to debug */
3059 qemu_savevm_send_ping(s
->to_dst_file
, 1);
3062 if (migrate_postcopy()) {
3064 * Tell the destination that we *might* want to do postcopy later;
3065 * if the other end can't do postcopy it should fail now, nice and
3068 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
3071 if (migrate_colo()) {
3072 /* Notify migration destination that we enable COLO */
3073 qemu_savevm_send_colo_enable(s
->to_dst_file
);
3076 qemu_mutex_lock_iothread();
3077 qemu_savevm_state_setup(s
->to_dst_file
);
3078 qemu_mutex_unlock_iothread();
3080 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
3081 MIGRATION_STATUS_ACTIVE
);
3083 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3085 trace_migration_thread_setup_complete();
3087 while (migration_is_active(s
)) {
3088 if (urgent
|| !migration_rate_exceeded(s
->to_dst_file
)) {
3089 MigIterateState iter_state
= migration_iteration_run(s
);
3090 if (iter_state
== MIG_ITERATE_SKIP
) {
3092 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3098 * Try to detect any kind of failures, and see whether we
3099 * should stop the migration now.
3101 thr_error
= migration_detect_error(s
);
3102 if (thr_error
== MIG_THR_ERR_FATAL
) {
3103 /* Stop migration */
3105 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
3107 * Just recovered from a e.g. network failure, reset all
3108 * the local variables. This is important to avoid
3109 * breaking transferred_bytes and bandwidth calculation
3111 update_iteration_initial_status(s
);
3114 urgent
= migration_rate_limit();
3117 trace_migration_thread_after_loop();
3118 migration_iteration_finish(s
);
3119 object_unref(OBJECT(s
));
3120 rcu_unregister_thread();
3121 migration_threads_remove(thread
);
3125 static void bg_migration_vm_start_bh(void *opaque
)
3127 MigrationState
*s
= opaque
;
3129 qemu_bh_delete(s
->vm_start_bh
);
3130 s
->vm_start_bh
= NULL
;
3133 s
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - s
->downtime_start
;
3137 * Background snapshot thread, based on live migration code.
3138 * This is an alternative implementation of live migration mechanism
3139 * introduced specifically to support background snapshots.
3141 * It takes advantage of userfault_fd write protection mechanism introduced
3142 * in v5.7 kernel. Compared to existing dirty page logging migration much
3143 * lesser stream traffic is produced resulting in smaller snapshot images,
3144 * simply cause of no page duplicates can get into the stream.
3146 * Another key point is that generated vmstate stream reflects machine state
3147 * 'frozen' at the beginning of snapshot creation compared to dirty page logging
3148 * mechanism, which effectively results in that saved snapshot is the state of VM
3149 * at the end of the process.
3151 static void *bg_migration_thread(void *opaque
)
3153 MigrationState
*s
= opaque
;
3154 int64_t setup_start
;
3155 MigThrError thr_error
;
3157 bool early_fail
= true;
3159 rcu_register_thread();
3160 object_ref(OBJECT(s
));
3162 migration_rate_set(RATE_LIMIT_DISABLED
);
3164 setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3166 * We want to save vmstate for the moment when migration has been
3167 * initiated but also we want to save RAM content while VM is running.
3168 * The RAM content should appear first in the vmstate. So, we first
3169 * stash the non-RAM part of the vmstate to the temporary buffer,
3170 * then write RAM part of the vmstate to the migration stream
3171 * with vCPUs running and, finally, write stashed non-RAM part of
3172 * the vmstate from the buffer to the migration stream.
3174 s
->bioc
= qio_channel_buffer_new(512 * 1024);
3175 qio_channel_set_name(QIO_CHANNEL(s
->bioc
), "vmstate-buffer");
3176 fb
= qemu_file_new_output(QIO_CHANNEL(s
->bioc
));
3177 object_unref(OBJECT(s
->bioc
));
3179 update_iteration_initial_status(s
);
3182 * Prepare for tracking memory writes with UFFD-WP - populate
3183 * RAM pages before protecting.
3186 ram_write_tracking_prepare();
3189 qemu_mutex_lock_iothread();
3190 qemu_savevm_state_header(s
->to_dst_file
);
3191 qemu_savevm_state_setup(s
->to_dst_file
);
3192 qemu_mutex_unlock_iothread();
3194 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
3195 MIGRATION_STATUS_ACTIVE
);
3197 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3199 trace_migration_thread_setup_complete();
3200 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3202 qemu_mutex_lock_iothread();
3205 * If VM is currently in suspended state, then, to make a valid runstate
3206 * transition in vm_stop_force_state() we need to wakeup it up.
3208 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3209 s
->vm_old_state
= runstate_get();
3211 global_state_store();
3212 /* Forcibly stop VM before saving state of vCPUs and devices */
3213 if (vm_stop_force_state(RUN_STATE_PAUSED
)) {
3217 * Put vCPUs in sync with shadow context structures, then
3218 * save their state to channel-buffer along with devices.
3220 cpu_synchronize_all_states();
3221 if (qemu_savevm_state_complete_precopy_non_iterable(fb
, false, false)) {
3225 * Since we are going to get non-iterable state data directly
3226 * from s->bioc->data, explicit flush is needed here.
3230 /* Now initialize UFFD context and start tracking RAM writes */
3231 if (ram_write_tracking_start()) {
3237 * Start VM from BH handler to avoid write-fault lock here.
3238 * UFFD-WP protection for the whole RAM is already enabled so
3239 * calling VM state change notifiers from vm_start() would initiate
3240 * writes to virtio VQs memory which is in write-protected region.
3242 s
->vm_start_bh
= qemu_bh_new(bg_migration_vm_start_bh
, s
);
3243 qemu_bh_schedule(s
->vm_start_bh
);
3245 qemu_mutex_unlock_iothread();
3247 while (migration_is_active(s
)) {
3248 MigIterateState iter_state
= bg_migration_iteration_run(s
);
3249 if (iter_state
== MIG_ITERATE_SKIP
) {
3251 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3256 * Try to detect any kind of failures, and see whether we
3257 * should stop the migration now.
3259 thr_error
= migration_detect_error(s
);
3260 if (thr_error
== MIG_THR_ERR_FATAL
) {
3261 /* Stop migration */
3265 migration_update_counters(s
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
));
3268 trace_migration_thread_after_loop();
3272 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
3273 MIGRATION_STATUS_FAILED
);
3274 qemu_mutex_unlock_iothread();
3277 bg_migration_iteration_finish(s
);
3280 object_unref(OBJECT(s
));
3281 rcu_unregister_thread();
3286 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
3288 Error
*local_err
= NULL
;
3289 uint64_t rate_limit
;
3290 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
3293 * If there's a previous error, free it and prepare for another one.
3294 * Meanwhile if migration completes successfully, there won't have an error
3295 * dumped when calling migrate_fd_cleanup().
3297 migrate_error_free(s
);
3299 s
->expected_downtime
= migrate_downtime_limit();
3301 assert(s
->cleanup_bh
);
3303 assert(!s
->cleanup_bh
);
3304 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup_bh
, s
);
3307 migrate_fd_error(s
, error_in
);
3310 * Don't do cleanup for resume if channel is invalid, but only dump
3311 * the error. We wait for another channel connect from the user.
3312 * The error_report still gives HMP user a hint on what failed.
3313 * It's normally done in migrate_fd_cleanup(), but call it here
3316 error_report_err(error_copy(s
->error
));
3318 migrate_fd_cleanup(s
);
3324 /* This is a resumed migration */
3325 rate_limit
= migrate_max_postcopy_bandwidth();
3327 /* This is a fresh new migration */
3328 rate_limit
= migrate_max_bandwidth();
3330 /* Notify before starting migration thread */
3331 migration_call_notifiers(s
);
3334 migration_rate_set(rate_limit
);
3335 qemu_file_set_blocking(s
->to_dst_file
, true);
3338 * Open the return path. For postcopy, it is used exclusively. For
3339 * precopy, only if user specified "return-path" capability would
3340 * QEMU uses the return path.
3342 if (migrate_postcopy_ram() || migrate_return_path()) {
3343 if (open_return_path_on_source(s
)) {
3344 error_setg(&local_err
, "Unable to open return-path for postcopy");
3345 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
3346 migrate_set_error(s
, local_err
);
3347 error_report_err(local_err
);
3348 migrate_fd_cleanup(s
);
3354 * This needs to be done before resuming a postcopy. Note: for newer
3355 * QEMUs we will delay the channel creation until postcopy_start(), to
3356 * avoid disorder of channel creations.
3358 if (migrate_postcopy_preempt() && s
->preempt_pre_7_2
) {
3359 postcopy_preempt_setup(s
);
3363 /* Wakeup the main migration thread to do the recovery */
3364 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
3365 MIGRATION_STATUS_POSTCOPY_RECOVER
);
3366 qemu_sem_post(&s
->postcopy_pause_sem
);
3370 if (multifd_save_setup(&local_err
) != 0) {
3371 migrate_set_error(s
, local_err
);
3372 error_report_err(local_err
);
3373 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3374 MIGRATION_STATUS_FAILED
);
3375 migrate_fd_cleanup(s
);
3379 if (migrate_background_snapshot()) {
3380 qemu_thread_create(&s
->thread
, "bg_snapshot",
3381 bg_migration_thread
, s
, QEMU_THREAD_JOINABLE
);
3383 qemu_thread_create(&s
->thread
, "live_migration",
3384 migration_thread
, s
, QEMU_THREAD_JOINABLE
);
3386 s
->migration_thread_running
= true;
3389 static void migration_class_init(ObjectClass
*klass
, void *data
)
3391 DeviceClass
*dc
= DEVICE_CLASS(klass
);
3393 dc
->user_creatable
= false;
3394 device_class_set_props(dc
, migration_properties
);
3397 static void migration_instance_finalize(Object
*obj
)
3399 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3401 qemu_mutex_destroy(&ms
->error_mutex
);
3402 qemu_mutex_destroy(&ms
->qemu_file_lock
);
3403 qemu_sem_destroy(&ms
->wait_unplug_sem
);
3404 qemu_sem_destroy(&ms
->rate_limit_sem
);
3405 qemu_sem_destroy(&ms
->pause_sem
);
3406 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
3407 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
3408 qemu_sem_destroy(&ms
->rp_state
.rp_pong_acks
);
3409 qemu_sem_destroy(&ms
->postcopy_qemufile_src_sem
);
3410 error_free(ms
->error
);
3413 static void migration_instance_init(Object
*obj
)
3415 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3417 ms
->state
= MIGRATION_STATUS_NONE
;
3419 ms
->pages_per_second
= -1;
3420 qemu_sem_init(&ms
->pause_sem
, 0);
3421 qemu_mutex_init(&ms
->error_mutex
);
3423 migrate_params_init(&ms
->parameters
);
3425 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
3426 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
3427 qemu_sem_init(&ms
->rp_state
.rp_pong_acks
, 0);
3428 qemu_sem_init(&ms
->rate_limit_sem
, 0);
3429 qemu_sem_init(&ms
->wait_unplug_sem
, 0);
3430 qemu_sem_init(&ms
->postcopy_qemufile_src_sem
, 0);
3431 qemu_mutex_init(&ms
->qemu_file_lock
);
3435 * Return true if check pass, false otherwise. Error will be put
3436 * inside errp if provided.
3438 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
3440 /* Assuming all off */
3441 bool old_caps
[MIGRATION_CAPABILITY__MAX
] = { 0 };
3443 if (!migrate_params_check(&ms
->parameters
, errp
)) {
3447 return migrate_caps_check(old_caps
, ms
->capabilities
, errp
);
3450 static const TypeInfo migration_type
= {
3451 .name
= TYPE_MIGRATION
,
3453 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3454 * not created using qdev_new(), it is not attached to the qdev
3455 * device tree, and it is never realized.
3457 * TODO: Make this TYPE_OBJECT once QOM provides something like
3458 * TYPE_DEVICE's "-global" properties.
3460 .parent
= TYPE_DEVICE
,
3461 .class_init
= migration_class_init
,
3462 .class_size
= sizeof(MigrationClass
),
3463 .instance_size
= sizeof(MigrationState
),
3464 .instance_init
= migration_instance_init
,
3465 .instance_finalize
= migration_instance_finalize
,
3468 static void register_migration_types(void)
3470 type_register_static(&migration_type
);
3473 type_init(register_migration_types
);