4 * Copyright IBM, Corp. 2008
7 * Anthony Liguori <aliguori@us.ibm.com>
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
16 #include "qemu/osdep.h"
17 #include "qemu/cutils.h"
18 #include "qemu/error-report.h"
19 #include "qemu/main-loop.h"
20 #include "migration/blocker.h"
24 #include "sysemu/runstate.h"
25 #include "sysemu/sysemu.h"
26 #include "sysemu/cpu-throttle.h"
29 #include "migration/global_state.h"
30 #include "migration/misc.h"
31 #include "migration.h"
33 #include "qemu-file.h"
35 #include "migration/vmstate.h"
36 #include "block/block.h"
37 #include "qapi/error.h"
38 #include "qapi/clone-visitor.h"
39 #include "qapi/qapi-visit-migration.h"
40 #include "qapi/qapi-visit-sockets.h"
41 #include "qapi/qapi-commands-migration.h"
42 #include "qapi/qapi-events-migration.h"
43 #include "qapi/qmp/qerror.h"
44 #include "qapi/qmp/qnull.h"
47 #include "postcopy-ram.h"
48 #include "qemu/thread.h"
50 #include "exec/target_page.h"
51 #include "io/channel-buffer.h"
52 #include "io/channel-tls.h"
53 #include "migration/colo.h"
54 #include "hw/boards.h"
55 #include "hw/qdev-properties.h"
56 #include "hw/qdev-properties-system.h"
57 #include "monitor/monitor.h"
58 #include "net/announce.h"
59 #include "qemu/queue.h"
61 #include "threadinfo.h"
62 #include "qemu/yank.h"
63 #include "sysemu/cpus.h"
64 #include "yank_functions.h"
65 #include "sysemu/qtest.h"
68 #define MAX_THROTTLE (128 << 20) /* Migration transfer speed throttling */
70 /* Time in milliseconds we are allowed to stop the source,
71 * for sending the last part */
72 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
74 /* Default compression thread count */
75 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
76 /* Default decompression thread count, usually decompression is at
77 * least 4 times as fast as compression.*/
78 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
79 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
80 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
81 /* Define default autoconverge cpu throttle migration parameters */
82 #define DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD 50
83 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
84 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
85 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
87 /* Migration XBZRLE default cache size */
88 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
90 /* The delay time (in ms) between two COLO checkpoints */
91 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
92 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
93 #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE
94 /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */
95 #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
96 /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
97 #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
99 /* Background transfer rate for postcopy, 0 means unlimited, note
100 * that page requests can still exceed this limit.
102 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
105 * Parameters for self_announce_delay giving a stream of RARP/ARP
106 * packets after migration.
108 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
109 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
110 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
111 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
113 static NotifierList migration_state_notifiers
=
114 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
116 /* Messages sent on the return path from destination to source */
117 enum mig_rp_message_type
{
118 MIG_RP_MSG_INVALID
= 0, /* Must be 0 */
119 MIG_RP_MSG_SHUT
, /* sibling will not send any more RP messages */
120 MIG_RP_MSG_PONG
, /* Response to a PING; data (seq: be32 ) */
122 MIG_RP_MSG_REQ_PAGES_ID
, /* data (start: be64, len: be32, id: string) */
123 MIG_RP_MSG_REQ_PAGES
, /* data (start: be64, len: be32) */
124 MIG_RP_MSG_RECV_BITMAP
, /* send recved_bitmap back to source */
125 MIG_RP_MSG_RESUME_ACK
, /* tell source that we are ready to resume */
130 /* When we add fault tolerance, we could have several
131 migrations at once. For now we don't need to add
132 dynamic creation of migration */
134 static MigrationState
*current_migration
;
135 static MigrationIncomingState
*current_incoming
;
137 static GSList
*migration_blockers
;
139 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
140 static int migration_maybe_pause(MigrationState
*s
,
141 int *current_active_state
,
143 static void migrate_fd_cancel(MigrationState
*s
);
145 static bool migration_needs_multiple_sockets(void)
147 return migrate_multifd() || migrate_postcopy_preempt();
150 static bool uri_supports_multi_channels(const char *uri
)
152 return strstart(uri
, "tcp:", NULL
) || strstart(uri
, "unix:", NULL
) ||
153 strstart(uri
, "vsock:", NULL
);
157 migration_channels_and_uri_compatible(const char *uri
, Error
**errp
)
159 if (migration_needs_multiple_sockets() &&
160 !uri_supports_multi_channels(uri
)) {
161 error_setg(errp
, "Migration requires multi-channel URIs (e.g. tcp)");
168 static gint
page_request_addr_cmp(gconstpointer ap
, gconstpointer bp
)
170 uintptr_t a
= (uintptr_t) ap
, b
= (uintptr_t) bp
;
172 return (a
> b
) - (a
< b
);
175 void migration_object_init(void)
177 /* This can only be called once. */
178 assert(!current_migration
);
179 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
182 * Init the migrate incoming object as well no matter whether
183 * we'll use it or not.
185 assert(!current_incoming
);
186 current_incoming
= g_new0(MigrationIncomingState
, 1);
187 current_incoming
->state
= MIGRATION_STATUS_NONE
;
188 current_incoming
->postcopy_remote_fds
=
189 g_array_new(FALSE
, TRUE
, sizeof(struct PostCopyFD
));
190 qemu_mutex_init(¤t_incoming
->rp_mutex
);
191 qemu_mutex_init(¤t_incoming
->postcopy_prio_thread_mutex
);
192 qemu_event_init(¤t_incoming
->main_thread_load_event
, false);
193 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_dst
, 0);
194 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fault
, 0);
195 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fast_load
, 0);
196 qemu_sem_init(¤t_incoming
->postcopy_qemufile_dst_done
, 0);
198 qemu_mutex_init(¤t_incoming
->page_request_mutex
);
199 current_incoming
->page_requested
= g_tree_new(page_request_addr_cmp
);
201 migration_object_check(current_migration
, &error_fatal
);
205 dirty_bitmap_mig_init();
208 void migration_cancel(const Error
*error
)
211 migrate_set_error(current_migration
, error
);
213 migrate_fd_cancel(current_migration
);
216 void migration_shutdown(void)
219 * When the QEMU main thread exit, the COLO thread
220 * may wait a semaphore. So, we should wakeup the
221 * COLO thread before migration shutdown.
225 * Cancel the current migration - that will (eventually)
226 * stop the migration using this structure
228 migration_cancel(NULL
);
229 object_unref(OBJECT(current_migration
));
232 * Cancel outgoing migration of dirty bitmaps. It should
233 * at least unref used block nodes.
235 dirty_bitmap_mig_cancel_outgoing();
238 * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
239 * are non-critical data, and their loss never considered as
242 dirty_bitmap_mig_cancel_incoming();
246 MigrationState
*migrate_get_current(void)
248 /* This can only be called after the object created. */
249 assert(current_migration
);
250 return current_migration
;
253 MigrationIncomingState
*migration_incoming_get_current(void)
255 assert(current_incoming
);
256 return current_incoming
;
259 void migration_incoming_transport_cleanup(MigrationIncomingState
*mis
)
261 if (mis
->socket_address_list
) {
262 qapi_free_SocketAddressList(mis
->socket_address_list
);
263 mis
->socket_address_list
= NULL
;
266 if (mis
->transport_cleanup
) {
267 mis
->transport_cleanup(mis
->transport_data
);
268 mis
->transport_data
= mis
->transport_cleanup
= NULL
;
272 void migration_incoming_state_destroy(void)
274 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
276 multifd_load_cleanup();
278 if (mis
->to_src_file
) {
279 /* Tell source that we are done */
280 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
281 qemu_fclose(mis
->to_src_file
);
282 mis
->to_src_file
= NULL
;
285 if (mis
->from_src_file
) {
286 migration_ioc_unregister_yank_from_file(mis
->from_src_file
);
287 qemu_fclose(mis
->from_src_file
);
288 mis
->from_src_file
= NULL
;
290 if (mis
->postcopy_remote_fds
) {
291 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
292 mis
->postcopy_remote_fds
= NULL
;
295 migration_incoming_transport_cleanup(mis
);
296 qemu_event_reset(&mis
->main_thread_load_event
);
298 if (mis
->page_requested
) {
299 g_tree_destroy(mis
->page_requested
);
300 mis
->page_requested
= NULL
;
303 if (mis
->postcopy_qemufile_dst
) {
304 migration_ioc_unregister_yank_from_file(mis
->postcopy_qemufile_dst
);
305 qemu_fclose(mis
->postcopy_qemufile_dst
);
306 mis
->postcopy_qemufile_dst
= NULL
;
309 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
312 static void migrate_generate_event(int new_state
)
314 if (migrate_events()) {
315 qapi_event_send_migration(new_state
);
320 * Send a message on the return channel back to the source
323 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
324 enum mig_rp_message_type message_type
,
325 uint16_t len
, void *data
)
329 trace_migrate_send_rp_message((int)message_type
, len
);
330 QEMU_LOCK_GUARD(&mis
->rp_mutex
);
333 * It's possible that the file handle got lost due to network
336 if (!mis
->to_src_file
) {
341 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
342 qemu_put_be16(mis
->to_src_file
, len
);
343 qemu_put_buffer(mis
->to_src_file
, data
, len
);
344 qemu_fflush(mis
->to_src_file
);
346 /* It's possible that qemu file got error during sending */
347 ret
= qemu_file_get_error(mis
->to_src_file
);
352 /* Request one page from the source VM at the given start address.
353 * rb: the RAMBlock to request the page in
354 * Start: Address offset within the RB
355 * Len: Length in bytes required - must be a multiple of pagesize
357 int migrate_send_rp_message_req_pages(MigrationIncomingState
*mis
,
358 RAMBlock
*rb
, ram_addr_t start
)
360 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
361 size_t msglen
= 12; /* start + len */
362 size_t len
= qemu_ram_pagesize(rb
);
363 enum mig_rp_message_type msg_type
;
367 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
368 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
371 * We maintain the last ramblock that we requested for page. Note that we
372 * don't need locking because this function will only be called within the
373 * postcopy ram fault thread.
375 if (rb
!= mis
->last_rb
) {
378 rbname
= qemu_ram_get_idstr(rb
);
379 rbname_len
= strlen(rbname
);
381 assert(rbname_len
< 256);
383 bufc
[msglen
++] = rbname_len
;
384 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
385 msglen
+= rbname_len
;
386 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
388 msg_type
= MIG_RP_MSG_REQ_PAGES
;
391 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
394 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
,
395 RAMBlock
*rb
, ram_addr_t start
, uint64_t haddr
)
397 void *aligned
= (void *)(uintptr_t)ROUND_DOWN(haddr
, qemu_ram_pagesize(rb
));
398 bool received
= false;
400 WITH_QEMU_LOCK_GUARD(&mis
->page_request_mutex
) {
401 received
= ramblock_recv_bitmap_test_byte_offset(rb
, start
);
402 if (!received
&& !g_tree_lookup(mis
->page_requested
, aligned
)) {
404 * The page has not been received, and it's not yet in the page
405 * request list. Queue it. Set the value of element to 1, so that
406 * things like g_tree_lookup() will return TRUE (1) when found.
408 g_tree_insert(mis
->page_requested
, aligned
, (gpointer
)1);
409 mis
->page_requested_count
++;
410 trace_postcopy_page_req_add(aligned
, mis
->page_requested_count
);
415 * If the page is there, skip sending the message. We don't even need the
416 * lock because as long as the page arrived, it'll be there forever.
422 return migrate_send_rp_message_req_pages(mis
, rb
, start
);
425 static bool migration_colo_enabled
;
426 bool migration_incoming_colo_enabled(void)
428 return migration_colo_enabled
;
431 void migration_incoming_disable_colo(void)
433 ram_block_discard_disable(false);
434 migration_colo_enabled
= false;
437 int migration_incoming_enable_colo(void)
439 if (ram_block_discard_disable(true)) {
440 error_report("COLO: cannot disable RAM discard");
443 migration_colo_enabled
= true;
447 void migrate_add_address(SocketAddress
*address
)
449 MigrationIncomingState
*mis
= migration_incoming_get_current();
451 QAPI_LIST_PREPEND(mis
->socket_address_list
,
452 QAPI_CLONE(SocketAddress
, address
));
455 static void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
457 const char *p
= NULL
;
459 /* URI is not suitable for migration? */
460 if (!migration_channels_and_uri_compatible(uri
, errp
)) {
464 qapi_event_send_migration(MIGRATION_STATUS_SETUP
);
465 if (strstart(uri
, "tcp:", &p
) ||
466 strstart(uri
, "unix:", NULL
) ||
467 strstart(uri
, "vsock:", NULL
)) {
468 socket_start_incoming_migration(p
? p
: uri
, errp
);
470 } else if (strstart(uri
, "rdma:", &p
)) {
471 rdma_start_incoming_migration(p
, errp
);
473 } else if (strstart(uri
, "exec:", &p
)) {
474 exec_start_incoming_migration(p
, errp
);
475 } else if (strstart(uri
, "fd:", &p
)) {
476 fd_start_incoming_migration(p
, errp
);
478 error_setg(errp
, "unknown migration protocol: %s", uri
);
482 static void process_incoming_migration_bh(void *opaque
)
484 Error
*local_err
= NULL
;
485 MigrationIncomingState
*mis
= opaque
;
487 /* If capability late_block_activate is set:
488 * Only fire up the block code now if we're going to restart the
489 * VM, else 'cont' will do it.
490 * This causes file locking to happen; so we don't want it to happen
491 * unless we really are starting the VM.
493 if (!migrate_late_block_activate() ||
494 (autostart
&& (!global_state_received() ||
495 global_state_get_runstate() == RUN_STATE_RUNNING
))) {
496 /* Make sure all file formats throw away their mutable metadata.
497 * If we get an error here, just don't restart the VM yet. */
498 bdrv_activate_all(&local_err
);
500 error_report_err(local_err
);
507 * This must happen after all error conditions are dealt with and
508 * we're sure the VM is going to be running on this host.
510 qemu_announce_self(&mis
->announce_timer
, migrate_announce_params());
512 multifd_load_shutdown();
514 dirty_bitmap_mig_before_vm_start();
516 if (!global_state_received() ||
517 global_state_get_runstate() == RUN_STATE_RUNNING
) {
521 runstate_set(RUN_STATE_PAUSED
);
523 } else if (migration_incoming_colo_enabled()) {
524 migration_incoming_disable_colo();
527 runstate_set(global_state_get_runstate());
530 * This must happen after any state changes since as soon as an external
531 * observer sees this event they might start to prod at the VM assuming
534 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
535 MIGRATION_STATUS_COMPLETED
);
536 qemu_bh_delete(mis
->bh
);
537 migration_incoming_state_destroy();
540 static void coroutine_fn
541 process_incoming_migration_co(void *opaque
)
543 MigrationIncomingState
*mis
= migration_incoming_get_current();
546 Error
*local_err
= NULL
;
548 assert(mis
->from_src_file
);
549 mis
->migration_incoming_co
= qemu_coroutine_self();
550 mis
->largest_page_size
= qemu_ram_pagesize_largest();
551 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
552 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
553 MIGRATION_STATUS_ACTIVE
);
554 ret
= qemu_loadvm_state(mis
->from_src_file
);
556 ps
= postcopy_state_get();
557 trace_process_incoming_migration_co_end(ret
, ps
);
558 if (ps
!= POSTCOPY_INCOMING_NONE
) {
559 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
561 * Where a migration had postcopy enabled (and thus went to advise)
562 * but managed to complete within the precopy period, we can use
565 postcopy_ram_incoming_cleanup(mis
);
566 } else if (ret
>= 0) {
568 * Postcopy was started, cleanup should happen at the end of the
571 trace_process_incoming_migration_co_postcopy_end_main();
574 /* Else if something went wrong then just fall out of the normal exit */
577 /* we get COLO info, and know if we are in COLO mode */
578 if (!ret
&& migration_incoming_colo_enabled()) {
579 /* Make sure all file formats throw away their mutable metadata */
580 bdrv_activate_all(&local_err
);
582 error_report_err(local_err
);
586 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
587 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
588 mis
->have_colo_incoming_thread
= true;
589 qemu_coroutine_yield();
591 qemu_mutex_unlock_iothread();
592 /* Wait checkpoint incoming thread exit before free resource */
593 qemu_thread_join(&mis
->colo_incoming_thread
);
594 qemu_mutex_lock_iothread();
595 /* We hold the global iothread lock, so it is safe here */
596 colo_release_ram_cache();
600 error_report("load of migration failed: %s", strerror(-ret
));
603 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
604 qemu_bh_schedule(mis
->bh
);
605 mis
->migration_incoming_co
= NULL
;
609 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
610 MIGRATION_STATUS_FAILED
);
611 qemu_fclose(mis
->from_src_file
);
613 multifd_load_cleanup();
619 * migration_incoming_setup: Setup incoming migration
620 * @f: file for main migration channel
621 * @errp: where to put errors
623 * Returns: %true on success, %false on error.
625 static bool migration_incoming_setup(QEMUFile
*f
, Error
**errp
)
627 MigrationIncomingState
*mis
= migration_incoming_get_current();
629 if (!mis
->from_src_file
) {
630 mis
->from_src_file
= f
;
632 qemu_file_set_blocking(f
, false);
636 void migration_incoming_process(void)
638 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
639 qemu_coroutine_enter(co
);
642 /* Returns true if recovered from a paused migration, otherwise false */
643 static bool postcopy_try_recover(void)
645 MigrationIncomingState
*mis
= migration_incoming_get_current();
647 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
648 /* Resumed from a paused postcopy migration */
650 /* This should be set already in migration_incoming_setup() */
651 assert(mis
->from_src_file
);
652 /* Postcopy has standalone thread to do vm load */
653 qemu_file_set_blocking(mis
->from_src_file
, true);
655 /* Re-configure the return path */
656 mis
->to_src_file
= qemu_file_get_return_path(mis
->from_src_file
);
658 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
659 MIGRATION_STATUS_POSTCOPY_RECOVER
);
662 * Here, we only wake up the main loading thread (while the
663 * rest threads will still be waiting), so that we can receive
664 * commands from source now, and answer it if needed. The
665 * rest threads will be woken up afterwards until we are sure
666 * that source is ready to reply to page requests.
668 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
675 void migration_fd_process_incoming(QEMUFile
*f
, Error
**errp
)
677 if (!migration_incoming_setup(f
, errp
)) {
680 if (postcopy_try_recover()) {
683 migration_incoming_process();
687 * Returns true when we want to start a new incoming migration process,
690 static bool migration_should_start_incoming(bool main_channel
)
692 /* Multifd doesn't start unless all channels are established */
693 if (migrate_multifd()) {
694 return migration_has_all_channels();
697 /* Preempt channel only starts when the main channel is created */
698 if (migrate_postcopy_preempt()) {
703 * For all the rest types of migration, we should only reach here when
704 * it's the main channel that's being created, and we should always
705 * proceed with this channel.
707 assert(main_channel
);
711 void migration_ioc_process_incoming(QIOChannel
*ioc
, Error
**errp
)
713 MigrationIncomingState
*mis
= migration_incoming_get_current();
714 Error
*local_err
= NULL
;
716 bool default_channel
= true;
717 uint32_t channel_magic
= 0;
720 if (migrate_multifd() && !migrate_postcopy_ram() &&
721 qio_channel_has_feature(ioc
, QIO_CHANNEL_FEATURE_READ_MSG_PEEK
)) {
723 * With multiple channels, it is possible that we receive channels
724 * out of order on destination side, causing incorrect mapping of
725 * source channels on destination side. Check channel MAGIC to
726 * decide type of channel. Please note this is best effort, postcopy
727 * preempt channel does not send any magic number so avoid it for
728 * postcopy live migration. Also tls live migration already does
729 * tls handshake while initializing main channel so with tls this
730 * issue is not possible.
732 ret
= migration_channel_read_peek(ioc
, (void *)&channel_magic
,
733 sizeof(channel_magic
), &local_err
);
736 error_propagate(errp
, local_err
);
740 default_channel
= (channel_magic
== cpu_to_be32(QEMU_VM_FILE_MAGIC
));
742 default_channel
= !mis
->from_src_file
;
745 if (multifd_load_setup(errp
) != 0) {
746 error_setg(errp
, "Failed to setup multifd channels");
750 if (default_channel
) {
751 f
= qemu_file_new_input(ioc
);
753 if (!migration_incoming_setup(f
, errp
)) {
757 /* Multiple connections */
758 assert(migration_needs_multiple_sockets());
759 if (migrate_multifd()) {
760 multifd_recv_new_channel(ioc
, &local_err
);
762 assert(migrate_postcopy_preempt());
763 f
= qemu_file_new_input(ioc
);
764 postcopy_preempt_new_channel(mis
, f
);
767 error_propagate(errp
, local_err
);
772 if (migration_should_start_incoming(default_channel
)) {
773 /* If it's a recovery, we're done */
774 if (postcopy_try_recover()) {
777 migration_incoming_process();
782 * @migration_has_all_channels: We have received all channels that we need
784 * Returns true when we have got connections to all the channels that
785 * we need for migration.
787 bool migration_has_all_channels(void)
789 MigrationIncomingState
*mis
= migration_incoming_get_current();
791 if (!mis
->from_src_file
) {
795 if (migrate_multifd()) {
796 return multifd_recv_all_channels_created();
799 if (migrate_postcopy_preempt()) {
800 return mis
->postcopy_qemufile_dst
!= NULL
;
807 * Send a 'SHUT' message on the return channel with the given value
808 * to indicate that we've finished with the RP. Non-0 value indicates
811 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
816 buf
= cpu_to_be32(value
);
817 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
821 * Send a 'PONG' message on the return channel with the given value
822 * (normally in response to a 'PING')
824 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
829 buf
= cpu_to_be32(value
);
830 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
833 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
841 * First, we send the header part. It contains only the len of
842 * idstr, and the idstr itself.
844 len
= strlen(block_name
);
846 memcpy(buf
+ 1, block_name
, len
);
848 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
849 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
854 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
857 * Next, we dump the received bitmap to the stream.
859 * TODO: currently we are safe since we are the only one that is
860 * using the to_src_file handle (fault thread is still paused),
861 * and it's ok even not taking the mutex. However the best way is
862 * to take the lock before sending the message header, and release
863 * the lock after sending the bitmap.
865 qemu_mutex_lock(&mis
->rp_mutex
);
866 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
867 qemu_mutex_unlock(&mis
->rp_mutex
);
869 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
872 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
876 buf
= cpu_to_be32(value
);
877 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
881 * Return true if we're already in the middle of a migration
882 * (i.e. any of the active or setup states)
884 bool migration_is_setup_or_active(int state
)
887 case MIGRATION_STATUS_ACTIVE
:
888 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
889 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
890 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
891 case MIGRATION_STATUS_SETUP
:
892 case MIGRATION_STATUS_PRE_SWITCHOVER
:
893 case MIGRATION_STATUS_DEVICE
:
894 case MIGRATION_STATUS_WAIT_UNPLUG
:
895 case MIGRATION_STATUS_COLO
:
904 bool migration_is_running(int state
)
907 case MIGRATION_STATUS_ACTIVE
:
908 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
909 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
910 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
911 case MIGRATION_STATUS_SETUP
:
912 case MIGRATION_STATUS_PRE_SWITCHOVER
:
913 case MIGRATION_STATUS_DEVICE
:
914 case MIGRATION_STATUS_WAIT_UNPLUG
:
915 case MIGRATION_STATUS_CANCELLING
:
924 static bool migrate_show_downtime(MigrationState
*s
)
926 return (s
->state
== MIGRATION_STATUS_COMPLETED
) || migration_in_postcopy();
929 static void populate_time_info(MigrationInfo
*info
, MigrationState
*s
)
931 info
->has_status
= true;
932 info
->has_setup_time
= true;
933 info
->setup_time
= s
->setup_time
;
935 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
936 info
->has_total_time
= true;
937 info
->total_time
= s
->total_time
;
939 info
->has_total_time
= true;
940 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) -
944 if (migrate_show_downtime(s
)) {
945 info
->has_downtime
= true;
946 info
->downtime
= s
->downtime
;
948 info
->has_expected_downtime
= true;
949 info
->expected_downtime
= s
->expected_downtime
;
953 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
955 size_t page_size
= qemu_target_page_size();
957 info
->ram
= g_malloc0(sizeof(*info
->ram
));
958 info
->ram
->transferred
= stat64_get(&ram_counters
.transferred
);
959 info
->ram
->total
= ram_bytes_total();
960 info
->ram
->duplicate
= stat64_get(&ram_counters
.zero_pages
);
961 /* legacy value. It is not used anymore */
962 info
->ram
->skipped
= 0;
963 info
->ram
->normal
= stat64_get(&ram_counters
.normal_pages
);
964 info
->ram
->normal_bytes
= info
->ram
->normal
* page_size
;
965 info
->ram
->mbps
= s
->mbps
;
966 info
->ram
->dirty_sync_count
=
967 stat64_get(&ram_counters
.dirty_sync_count
);
968 info
->ram
->dirty_sync_missed_zero_copy
=
969 stat64_get(&ram_counters
.dirty_sync_missed_zero_copy
);
970 info
->ram
->postcopy_requests
=
971 stat64_get(&ram_counters
.postcopy_requests
);
972 info
->ram
->page_size
= page_size
;
973 info
->ram
->multifd_bytes
= stat64_get(&ram_counters
.multifd_bytes
);
974 info
->ram
->pages_per_second
= s
->pages_per_second
;
975 info
->ram
->precopy_bytes
= stat64_get(&ram_counters
.precopy_bytes
);
976 info
->ram
->downtime_bytes
= stat64_get(&ram_counters
.downtime_bytes
);
977 info
->ram
->postcopy_bytes
= stat64_get(&ram_counters
.postcopy_bytes
);
979 if (migrate_xbzrle()) {
980 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
981 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
982 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
983 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
984 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
985 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
986 info
->xbzrle_cache
->encoding_rate
= xbzrle_counters
.encoding_rate
;
987 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
990 if (migrate_compress()) {
991 info
->compression
= g_malloc0(sizeof(*info
->compression
));
992 info
->compression
->pages
= compression_counters
.pages
;
993 info
->compression
->busy
= compression_counters
.busy
;
994 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
995 info
->compression
->compressed_size
=
996 compression_counters
.compressed_size
;
997 info
->compression
->compression_rate
=
998 compression_counters
.compression_rate
;
1001 if (cpu_throttle_active()) {
1002 info
->has_cpu_throttle_percentage
= true;
1003 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
1006 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
1007 info
->ram
->remaining
= ram_bytes_remaining();
1008 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
1012 static void populate_disk_info(MigrationInfo
*info
)
1014 if (blk_mig_active()) {
1015 info
->disk
= g_malloc0(sizeof(*info
->disk
));
1016 info
->disk
->transferred
= blk_mig_bytes_transferred();
1017 info
->disk
->remaining
= blk_mig_bytes_remaining();
1018 info
->disk
->total
= blk_mig_bytes_total();
1022 static void fill_source_migration_info(MigrationInfo
*info
)
1024 MigrationState
*s
= migrate_get_current();
1025 int state
= qatomic_read(&s
->state
);
1026 GSList
*cur_blocker
= migration_blockers
;
1028 info
->blocked_reasons
= NULL
;
1031 * There are two types of reasons a migration might be blocked;
1032 * a) devices marked in VMState as non-migratable, and
1033 * b) Explicit migration blockers
1034 * We need to add both of them here.
1036 qemu_savevm_non_migratable_list(&info
->blocked_reasons
);
1038 while (cur_blocker
) {
1039 QAPI_LIST_PREPEND(info
->blocked_reasons
,
1040 g_strdup(error_get_pretty(cur_blocker
->data
)));
1041 cur_blocker
= g_slist_next(cur_blocker
);
1043 info
->has_blocked_reasons
= info
->blocked_reasons
!= NULL
;
1046 case MIGRATION_STATUS_NONE
:
1047 /* no migration has happened ever */
1048 /* do not overwrite destination migration status */
1050 case MIGRATION_STATUS_SETUP
:
1051 info
->has_status
= true;
1052 info
->has_total_time
= false;
1054 case MIGRATION_STATUS_ACTIVE
:
1055 case MIGRATION_STATUS_CANCELLING
:
1056 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1057 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1058 case MIGRATION_STATUS_DEVICE
:
1059 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1060 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1061 /* TODO add some postcopy stats */
1062 populate_time_info(info
, s
);
1063 populate_ram_info(info
, s
);
1064 populate_disk_info(info
);
1065 populate_vfio_info(info
);
1067 case MIGRATION_STATUS_COLO
:
1068 info
->has_status
= true;
1069 /* TODO: display COLO specific information (checkpoint info etc.) */
1071 case MIGRATION_STATUS_COMPLETED
:
1072 populate_time_info(info
, s
);
1073 populate_ram_info(info
, s
);
1074 populate_vfio_info(info
);
1076 case MIGRATION_STATUS_FAILED
:
1077 info
->has_status
= true;
1079 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
1082 case MIGRATION_STATUS_CANCELLED
:
1083 info
->has_status
= true;
1085 case MIGRATION_STATUS_WAIT_UNPLUG
:
1086 info
->has_status
= true;
1089 info
->status
= state
;
1092 static void fill_destination_migration_info(MigrationInfo
*info
)
1094 MigrationIncomingState
*mis
= migration_incoming_get_current();
1096 if (mis
->socket_address_list
) {
1097 info
->has_socket_address
= true;
1098 info
->socket_address
=
1099 QAPI_CLONE(SocketAddressList
, mis
->socket_address_list
);
1102 switch (mis
->state
) {
1103 case MIGRATION_STATUS_NONE
:
1105 case MIGRATION_STATUS_SETUP
:
1106 case MIGRATION_STATUS_CANCELLING
:
1107 case MIGRATION_STATUS_CANCELLED
:
1108 case MIGRATION_STATUS_ACTIVE
:
1109 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1110 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1111 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1112 case MIGRATION_STATUS_FAILED
:
1113 case MIGRATION_STATUS_COLO
:
1114 info
->has_status
= true;
1116 case MIGRATION_STATUS_COMPLETED
:
1117 info
->has_status
= true;
1118 fill_destination_postcopy_migration_info(info
);
1121 info
->status
= mis
->state
;
1124 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1126 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1128 fill_destination_migration_info(info
);
1129 fill_source_migration_info(info
);
1134 void qmp_migrate_start_postcopy(Error
**errp
)
1136 MigrationState
*s
= migrate_get_current();
1138 if (!migrate_postcopy()) {
1139 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1140 " the start of migration");
1144 if (s
->state
== MIGRATION_STATUS_NONE
) {
1145 error_setg(errp
, "Postcopy must be started after migration has been"
1150 * we don't error if migration has finished since that would be racy
1151 * with issuing this command.
1153 qatomic_set(&s
->start_postcopy
, true);
1156 /* shared migration helpers */
1158 void migrate_set_state(int *state
, int old_state
, int new_state
)
1160 assert(new_state
< MIGRATION_STATUS__MAX
);
1161 if (qatomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1162 trace_migrate_set_state(MigrationStatus_str(new_state
));
1163 migrate_generate_event(new_state
);
1167 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1169 s
->parameters
.block_incremental
= value
;
1172 static void block_cleanup_parameters(MigrationState
*s
)
1174 if (s
->must_remove_block_options
) {
1175 /* setting to false can never fail */
1176 migrate_cap_set(MIGRATION_CAPABILITY_BLOCK
, false, &error_abort
);
1177 migrate_set_block_incremental(s
, false);
1178 s
->must_remove_block_options
= false;
1182 static void migrate_fd_cleanup(MigrationState
*s
)
1184 qemu_bh_delete(s
->cleanup_bh
);
1185 s
->cleanup_bh
= NULL
;
1187 g_free(s
->hostname
);
1189 json_writer_free(s
->vmdesc
);
1192 qemu_savevm_state_cleanup();
1194 if (s
->to_dst_file
) {
1197 trace_migrate_fd_cleanup();
1198 qemu_mutex_unlock_iothread();
1199 if (s
->migration_thread_running
) {
1200 qemu_thread_join(&s
->thread
);
1201 s
->migration_thread_running
= false;
1203 qemu_mutex_lock_iothread();
1205 multifd_save_cleanup();
1206 qemu_mutex_lock(&s
->qemu_file_lock
);
1207 tmp
= s
->to_dst_file
;
1208 s
->to_dst_file
= NULL
;
1209 qemu_mutex_unlock(&s
->qemu_file_lock
);
1211 * Close the file handle without the lock to make sure the
1212 * critical section won't block for long.
1214 migration_ioc_unregister_yank_from_file(tmp
);
1218 if (s
->postcopy_qemufile_src
) {
1219 migration_ioc_unregister_yank_from_file(s
->postcopy_qemufile_src
);
1220 qemu_fclose(s
->postcopy_qemufile_src
);
1221 s
->postcopy_qemufile_src
= NULL
;
1224 assert(!migration_is_active(s
));
1226 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1227 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1228 MIGRATION_STATUS_CANCELLED
);
1232 /* It is used on info migrate. We can't free it */
1233 error_report_err(error_copy(s
->error
));
1235 notifier_list_notify(&migration_state_notifiers
, s
);
1236 block_cleanup_parameters(s
);
1237 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1240 static void migrate_fd_cleanup_schedule(MigrationState
*s
)
1243 * Ref the state for bh, because it may be called when
1244 * there're already no other refs
1246 object_ref(OBJECT(s
));
1247 qemu_bh_schedule(s
->cleanup_bh
);
1250 static void migrate_fd_cleanup_bh(void *opaque
)
1252 MigrationState
*s
= opaque
;
1253 migrate_fd_cleanup(s
);
1254 object_unref(OBJECT(s
));
1257 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1259 QEMU_LOCK_GUARD(&s
->error_mutex
);
1261 s
->error
= error_copy(error
);
1265 static void migrate_error_free(MigrationState
*s
)
1267 QEMU_LOCK_GUARD(&s
->error_mutex
);
1269 error_free(s
->error
);
1274 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1276 trace_migrate_fd_error(error_get_pretty(error
));
1277 assert(s
->to_dst_file
== NULL
);
1278 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1279 MIGRATION_STATUS_FAILED
);
1280 migrate_set_error(s
, error
);
1283 static void migrate_fd_cancel(MigrationState
*s
)
1286 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1287 trace_migrate_fd_cancel();
1289 WITH_QEMU_LOCK_GUARD(&s
->qemu_file_lock
) {
1290 if (s
->rp_state
.from_dst_file
) {
1291 /* shutdown the rp socket, so causing the rp thread to shutdown */
1292 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1297 old_state
= s
->state
;
1298 if (!migration_is_running(old_state
)) {
1301 /* If the migration is paused, kick it out of the pause */
1302 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1303 qemu_sem_post(&s
->pause_sem
);
1305 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1306 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1309 * If we're unlucky the migration code might be stuck somewhere in a
1310 * send/write while the network has failed and is waiting to timeout;
1311 * if we've got shutdown(2) available then we can force it to quit.
1312 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1313 * called in a bh, so there is no race against this cancel.
1315 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1316 qemu_file_shutdown(f
);
1318 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1319 Error
*local_err
= NULL
;
1321 bdrv_activate_all(&local_err
);
1323 error_report_err(local_err
);
1325 s
->block_inactive
= false;
1330 void add_migration_state_change_notifier(Notifier
*notify
)
1332 notifier_list_add(&migration_state_notifiers
, notify
);
1335 void remove_migration_state_change_notifier(Notifier
*notify
)
1337 notifier_remove(notify
);
1340 bool migration_in_setup(MigrationState
*s
)
1342 return s
->state
== MIGRATION_STATUS_SETUP
;
1345 bool migration_has_finished(MigrationState
*s
)
1347 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1350 bool migration_has_failed(MigrationState
*s
)
1352 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1353 s
->state
== MIGRATION_STATUS_FAILED
);
1356 bool migration_in_postcopy(void)
1358 MigrationState
*s
= migrate_get_current();
1361 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1362 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1363 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1370 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1372 return migration_in_postcopy() && s
->postcopy_after_devices
;
1375 bool migration_in_incoming_postcopy(void)
1377 PostcopyState ps
= postcopy_state_get();
1379 return ps
>= POSTCOPY_INCOMING_DISCARD
&& ps
< POSTCOPY_INCOMING_END
;
1382 bool migration_incoming_postcopy_advised(void)
1384 PostcopyState ps
= postcopy_state_get();
1386 return ps
>= POSTCOPY_INCOMING_ADVISE
&& ps
< POSTCOPY_INCOMING_END
;
1389 bool migration_in_bg_snapshot(void)
1391 MigrationState
*s
= migrate_get_current();
1393 return migrate_background_snapshot() &&
1394 migration_is_setup_or_active(s
->state
);
1397 bool migration_is_idle(void)
1399 MigrationState
*s
= current_migration
;
1406 case MIGRATION_STATUS_NONE
:
1407 case MIGRATION_STATUS_CANCELLED
:
1408 case MIGRATION_STATUS_COMPLETED
:
1409 case MIGRATION_STATUS_FAILED
:
1411 case MIGRATION_STATUS_SETUP
:
1412 case MIGRATION_STATUS_CANCELLING
:
1413 case MIGRATION_STATUS_ACTIVE
:
1414 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1415 case MIGRATION_STATUS_COLO
:
1416 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1417 case MIGRATION_STATUS_DEVICE
:
1418 case MIGRATION_STATUS_WAIT_UNPLUG
:
1420 case MIGRATION_STATUS__MAX
:
1421 g_assert_not_reached();
1427 bool migration_is_active(MigrationState
*s
)
1429 return (s
->state
== MIGRATION_STATUS_ACTIVE
||
1430 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1433 void migrate_init(MigrationState
*s
)
1436 * Reinitialise all migration state, except
1437 * parameters/capabilities that the user set, and
1442 s
->to_dst_file
= NULL
;
1443 s
->state
= MIGRATION_STATUS_NONE
;
1444 s
->rp_state
.from_dst_file
= NULL
;
1445 s
->rp_state
.error
= false;
1447 s
->pages_per_second
= 0.0;
1449 s
->expected_downtime
= 0;
1451 s
->start_postcopy
= false;
1452 s
->postcopy_after_devices
= false;
1453 s
->migration_thread_running
= false;
1454 error_free(s
->error
);
1458 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1460 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1462 s
->vm_was_running
= false;
1463 s
->iteration_initial_bytes
= 0;
1464 s
->threshold_size
= 0;
1467 int migrate_add_blocker_internal(Error
*reason
, Error
**errp
)
1469 /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */
1470 if (runstate_check(RUN_STATE_SAVE_VM
) || !migration_is_idle()) {
1471 error_propagate_prepend(errp
, error_copy(reason
),
1472 "disallowing migration blocker "
1473 "(migration/snapshot in progress) for: ");
1477 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1481 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1483 if (only_migratable
) {
1484 error_propagate_prepend(errp
, error_copy(reason
),
1485 "disallowing migration blocker "
1486 "(--only-migratable) for: ");
1490 return migrate_add_blocker_internal(reason
, errp
);
1493 void migrate_del_blocker(Error
*reason
)
1495 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1498 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1500 Error
*local_err
= NULL
;
1501 static bool once
= true;
1504 error_setg(errp
, "The incoming migration has already been started");
1507 if (!runstate_check(RUN_STATE_INMIGRATE
)) {
1508 error_setg(errp
, "'-incoming' was not specified on the command line");
1512 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
1516 qemu_start_incoming_migration(uri
, &local_err
);
1519 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1520 error_propagate(errp
, local_err
);
1527 void qmp_migrate_recover(const char *uri
, Error
**errp
)
1529 MigrationIncomingState
*mis
= migration_incoming_get_current();
1532 * Don't even bother to use ERRP_GUARD() as it _must_ always be set by
1533 * callers (no one should ignore a recover failure); if there is, it's a
1534 * programming error.
1538 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1539 error_setg(errp
, "Migrate recover can only be run "
1540 "when postcopy is paused.");
1544 /* If there's an existing transport, release it */
1545 migration_incoming_transport_cleanup(mis
);
1548 * Note that this call will never start a real migration; it will
1549 * only re-setup the migration stream and poke existing migration
1550 * to continue using that newly established channel.
1552 qemu_start_incoming_migration(uri
, errp
);
1555 void qmp_migrate_pause(Error
**errp
)
1557 MigrationState
*ms
= migrate_get_current();
1558 MigrationIncomingState
*mis
= migration_incoming_get_current();
1561 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1562 /* Source side, during postcopy */
1563 qemu_mutex_lock(&ms
->qemu_file_lock
);
1564 ret
= qemu_file_shutdown(ms
->to_dst_file
);
1565 qemu_mutex_unlock(&ms
->qemu_file_lock
);
1567 error_setg(errp
, "Failed to pause source migration");
1572 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1573 ret
= qemu_file_shutdown(mis
->from_src_file
);
1575 error_setg(errp
, "Failed to pause destination migration");
1580 error_setg(errp
, "migrate-pause is currently only supported "
1581 "during postcopy-active state");
1584 bool migration_is_blocked(Error
**errp
)
1586 if (qemu_savevm_state_blocked(errp
)) {
1590 if (migration_blockers
) {
1591 error_propagate(errp
, error_copy(migration_blockers
->data
));
1598 /* Returns true if continue to migrate, or false if error detected */
1599 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
1600 bool resume
, Error
**errp
)
1602 Error
*local_err
= NULL
;
1605 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1606 error_setg(errp
, "Cannot resume if there is no "
1607 "paused migration");
1612 * Postcopy recovery won't work well with release-ram
1613 * capability since release-ram will drop the page buffer as
1614 * long as the page is put into the send buffer. So if there
1615 * is a network failure happened, any page buffers that have
1616 * not yet reached the destination VM but have already been
1617 * sent from the source VM will be lost forever. Let's refuse
1618 * the client from resuming such a postcopy migration.
1619 * Luckily release-ram was designed to only be used when src
1620 * and destination VMs are on the same host, so it should be
1623 if (migrate_release_ram()) {
1624 error_setg(errp
, "Postcopy recovery cannot work "
1625 "when release-ram capability is set");
1629 /* This is a resume, skip init status */
1633 if (migration_is_running(s
->state
)) {
1634 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1638 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1639 error_setg(errp
, "Guest is waiting for an incoming migration");
1643 if (runstate_check(RUN_STATE_POSTMIGRATE
)) {
1644 error_setg(errp
, "Can't migrate the vm that was paused due to "
1645 "previous migration");
1649 if (migration_is_blocked(errp
)) {
1653 if (blk
|| blk_inc
) {
1654 if (migrate_colo()) {
1655 error_setg(errp
, "No disk migration is required in COLO mode");
1658 if (migrate_block() || migrate_block_incremental()) {
1659 error_setg(errp
, "Command options are incompatible with "
1660 "current migration capabilities");
1663 if (!migrate_cap_set(MIGRATION_CAPABILITY_BLOCK
, true, &local_err
)) {
1664 error_propagate(errp
, local_err
);
1667 s
->must_remove_block_options
= true;
1671 migrate_set_block_incremental(s
, true);
1676 * set ram_counters compression_counters memory to zero for a
1679 memset(&ram_counters
, 0, sizeof(ram_counters
));
1680 memset(&compression_counters
, 0, sizeof(compression_counters
));
1685 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1686 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1687 bool has_resume
, bool resume
, Error
**errp
)
1689 Error
*local_err
= NULL
;
1690 MigrationState
*s
= migrate_get_current();
1691 const char *p
= NULL
;
1693 /* URI is not suitable for migration? */
1694 if (!migration_channels_and_uri_compatible(uri
, errp
)) {
1698 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
1699 has_resume
&& resume
, errp
)) {
1700 /* Error detected, put into errp */
1704 if (!(has_resume
&& resume
)) {
1705 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
1710 if (strstart(uri
, "tcp:", &p
) ||
1711 strstart(uri
, "unix:", NULL
) ||
1712 strstart(uri
, "vsock:", NULL
)) {
1713 socket_start_outgoing_migration(s
, p
? p
: uri
, &local_err
);
1715 } else if (strstart(uri
, "rdma:", &p
)) {
1716 rdma_start_outgoing_migration(s
, p
, &local_err
);
1718 } else if (strstart(uri
, "exec:", &p
)) {
1719 exec_start_outgoing_migration(s
, p
, &local_err
);
1720 } else if (strstart(uri
, "fd:", &p
)) {
1721 fd_start_outgoing_migration(s
, p
, &local_err
);
1723 if (!(has_resume
&& resume
)) {
1724 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1726 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1727 "a valid migration protocol");
1728 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1729 MIGRATION_STATUS_FAILED
);
1730 block_cleanup_parameters(s
);
1735 if (!(has_resume
&& resume
)) {
1736 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1738 migrate_fd_error(s
, local_err
);
1739 error_propagate(errp
, local_err
);
1744 void qmp_migrate_cancel(Error
**errp
)
1746 migration_cancel(NULL
);
1749 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
1751 MigrationState
*s
= migrate_get_current();
1752 if (s
->state
!= state
) {
1753 error_setg(errp
, "Migration not in expected state: %s",
1754 MigrationStatus_str(s
->state
));
1757 qemu_sem_post(&s
->pause_sem
);
1760 /* migration thread support */
1762 * Something bad happened to the RP stream, mark an error
1763 * The caller shall print or trace something to indicate why
1765 static void mark_source_rp_bad(MigrationState
*s
)
1767 s
->rp_state
.error
= true;
1770 static struct rp_cmd_args
{
1771 ssize_t len
; /* -1 = variable */
1774 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
1775 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
1776 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
1777 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
1778 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
1779 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
1780 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
1781 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
1785 * Process a request for pages received on the return path,
1786 * We're allowed to send more than requested (e.g. to round to our page size)
1787 * and we don't need to send pages that have already been sent.
1789 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
1790 ram_addr_t start
, size_t len
)
1792 long our_host_ps
= qemu_real_host_page_size();
1794 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
1797 * Since we currently insist on matching page sizes, just sanity check
1798 * we're being asked for whole host pages.
1800 if (!QEMU_IS_ALIGNED(start
, our_host_ps
) ||
1801 !QEMU_IS_ALIGNED(len
, our_host_ps
)) {
1802 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
1803 " len: %zd", __func__
, start
, len
);
1804 mark_source_rp_bad(ms
);
1808 if (ram_save_queue_pages(rbname
, start
, len
)) {
1809 mark_source_rp_bad(ms
);
1813 /* Return true to retry, false to quit */
1814 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
1816 trace_postcopy_pause_return_path();
1818 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
1820 trace_postcopy_pause_return_path_continued();
1825 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
1827 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
1830 error_report("%s: invalid block name '%s'", __func__
, block_name
);
1834 /* Fetch the received bitmap and refresh the dirty bitmap */
1835 return ram_dirty_bitmap_reload(s
, block
);
1838 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
1840 trace_source_return_path_thread_resume_ack(value
);
1842 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
1843 error_report("%s: illegal resume_ack value %"PRIu32
,
1848 /* Now both sides are active. */
1849 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
1850 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1852 /* Notify send thread that time to continue send pages */
1853 qemu_sem_post(&s
->rp_state
.rp_sem
);
1859 * Release ms->rp_state.from_dst_file (and postcopy_qemufile_src if
1860 * existed) in a safe way.
1862 static void migration_release_dst_files(MigrationState
*ms
)
1866 WITH_QEMU_LOCK_GUARD(&ms
->qemu_file_lock
) {
1868 * Reset the from_dst_file pointer first before releasing it, as we
1869 * can't block within lock section
1871 file
= ms
->rp_state
.from_dst_file
;
1872 ms
->rp_state
.from_dst_file
= NULL
;
1876 * Do the same to postcopy fast path socket too if there is. No
1877 * locking needed because this qemufile should only be managed by
1878 * return path thread.
1880 if (ms
->postcopy_qemufile_src
) {
1881 migration_ioc_unregister_yank_from_file(ms
->postcopy_qemufile_src
);
1882 qemu_file_shutdown(ms
->postcopy_qemufile_src
);
1883 qemu_fclose(ms
->postcopy_qemufile_src
);
1884 ms
->postcopy_qemufile_src
= NULL
;
1891 * Handles messages sent on the return path towards the source VM
1894 static void *source_return_path_thread(void *opaque
)
1896 MigrationState
*ms
= opaque
;
1897 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
1898 uint16_t header_len
, header_type
;
1900 uint32_t tmp32
, sibling_error
;
1901 ram_addr_t start
= 0; /* =0 to silence warning */
1902 size_t len
= 0, expected_len
;
1905 trace_source_return_path_thread_entry();
1906 rcu_register_thread();
1909 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
1910 migration_is_setup_or_active(ms
->state
)) {
1911 trace_source_return_path_thread_loop_top();
1912 header_type
= qemu_get_be16(rp
);
1913 header_len
= qemu_get_be16(rp
);
1915 if (qemu_file_get_error(rp
)) {
1916 mark_source_rp_bad(ms
);
1920 if (header_type
>= MIG_RP_MSG_MAX
||
1921 header_type
== MIG_RP_MSG_INVALID
) {
1922 error_report("RP: Received invalid message 0x%04x length 0x%04x",
1923 header_type
, header_len
);
1924 mark_source_rp_bad(ms
);
1928 if ((rp_cmd_args
[header_type
].len
!= -1 &&
1929 header_len
!= rp_cmd_args
[header_type
].len
) ||
1930 header_len
> sizeof(buf
)) {
1931 error_report("RP: Received '%s' message (0x%04x) with"
1932 "incorrect length %d expecting %zu",
1933 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
1934 (size_t)rp_cmd_args
[header_type
].len
);
1935 mark_source_rp_bad(ms
);
1939 /* We know we've got a valid header by this point */
1940 res
= qemu_get_buffer(rp
, buf
, header_len
);
1941 if (res
!= header_len
) {
1942 error_report("RP: Failed reading data for message 0x%04x"
1943 " read %d expected %d",
1944 header_type
, res
, header_len
);
1945 mark_source_rp_bad(ms
);
1949 /* OK, we have the message and the data */
1950 switch (header_type
) {
1951 case MIG_RP_MSG_SHUT
:
1952 sibling_error
= ldl_be_p(buf
);
1953 trace_source_return_path_thread_shut(sibling_error
);
1954 if (sibling_error
) {
1955 error_report("RP: Sibling indicated error %d", sibling_error
);
1956 mark_source_rp_bad(ms
);
1959 * We'll let the main thread deal with closing the RP
1960 * we could do a shutdown(2) on it, but we're the only user
1961 * anyway, so there's nothing gained.
1965 case MIG_RP_MSG_PONG
:
1966 tmp32
= ldl_be_p(buf
);
1967 trace_source_return_path_thread_pong(tmp32
);
1968 qemu_sem_post(&ms
->rp_state
.rp_pong_acks
);
1971 case MIG_RP_MSG_REQ_PAGES
:
1972 start
= ldq_be_p(buf
);
1973 len
= ldl_be_p(buf
+ 8);
1974 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
1977 case MIG_RP_MSG_REQ_PAGES_ID
:
1978 expected_len
= 12 + 1; /* header + termination */
1980 if (header_len
>= expected_len
) {
1981 start
= ldq_be_p(buf
);
1982 len
= ldl_be_p(buf
+ 8);
1983 /* Now we expect an idstr */
1984 tmp32
= buf
[12]; /* Length of the following idstr */
1985 buf
[13 + tmp32
] = '\0';
1986 expected_len
+= tmp32
;
1988 if (header_len
!= expected_len
) {
1989 error_report("RP: Req_Page_id with length %d expecting %zd",
1990 header_len
, expected_len
);
1991 mark_source_rp_bad(ms
);
1994 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
1997 case MIG_RP_MSG_RECV_BITMAP
:
1998 if (header_len
< 1) {
1999 error_report("%s: missing block name", __func__
);
2000 mark_source_rp_bad(ms
);
2003 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2004 buf
[buf
[0] + 1] = '\0';
2005 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
2006 mark_source_rp_bad(ms
);
2011 case MIG_RP_MSG_RESUME_ACK
:
2012 tmp32
= ldl_be_p(buf
);
2013 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
2014 mark_source_rp_bad(ms
);
2025 res
= qemu_file_get_error(rp
);
2027 if (res
&& migration_in_postcopy()) {
2029 * Maybe there is something we can do: it looks like a
2030 * network down issue, and we pause for a recovery.
2032 migration_release_dst_files(ms
);
2034 if (postcopy_pause_return_path_thread(ms
)) {
2036 * Reload rp, reset the rest. Referencing it is safe since
2037 * it's reset only by us above, or when migration completes
2039 rp
= ms
->rp_state
.from_dst_file
;
2040 ms
->rp_state
.error
= false;
2045 trace_source_return_path_thread_bad_end();
2046 mark_source_rp_bad(ms
);
2049 trace_source_return_path_thread_end();
2050 migration_release_dst_files(ms
);
2051 rcu_unregister_thread();
2055 static int open_return_path_on_source(MigrationState
*ms
,
2058 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
2059 if (!ms
->rp_state
.from_dst_file
) {
2063 trace_open_return_path_on_source();
2065 if (!create_thread
) {
2070 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
2071 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
2072 ms
->rp_state
.rp_thread_created
= true;
2074 trace_open_return_path_on_source_continue();
2079 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2080 static int await_return_path_close_on_source(MigrationState
*ms
)
2083 * If this is a normal exit then the destination will send a SHUT and the
2084 * rp_thread will exit, however if there's an error we need to cause
2087 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
2089 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2090 * waiting for the destination.
2092 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
2093 mark_source_rp_bad(ms
);
2095 trace_await_return_path_close_on_source_joining();
2096 qemu_thread_join(&ms
->rp_state
.rp_thread
);
2097 ms
->rp_state
.rp_thread_created
= false;
2098 trace_await_return_path_close_on_source_close();
2099 return ms
->rp_state
.error
;
2103 migration_wait_main_channel(MigrationState
*ms
)
2105 /* Wait until one PONG message received */
2106 qemu_sem_wait(&ms
->rp_state
.rp_pong_acks
);
2110 * Switch from normal iteration to postcopy
2111 * Returns non-0 on error
2113 static int postcopy_start(MigrationState
*ms
)
2116 QIOChannelBuffer
*bioc
;
2118 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2119 int64_t bandwidth
= migrate_max_postcopy_bandwidth();
2120 bool restart_block
= false;
2121 int cur_state
= MIGRATION_STATUS_ACTIVE
;
2123 if (migrate_postcopy_preempt()) {
2124 migration_wait_main_channel(ms
);
2125 if (postcopy_preempt_establish_channel(ms
)) {
2126 migrate_set_state(&ms
->state
, ms
->state
, MIGRATION_STATUS_FAILED
);
2131 if (!migrate_pause_before_switchover()) {
2132 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
2133 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2136 trace_postcopy_start();
2137 qemu_mutex_lock_iothread();
2138 trace_postcopy_start_set_run();
2140 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2141 global_state_store();
2142 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2147 ret
= migration_maybe_pause(ms
, &cur_state
,
2148 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2153 ret
= bdrv_inactivate_all();
2157 restart_block
= true;
2160 * Cause any non-postcopiable, but iterative devices to
2161 * send out their final data.
2163 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2166 * in Finish migrate and with the io-lock held everything should
2167 * be quiet, but we've potentially still got dirty pages and we
2168 * need to tell the destination to throw any pages it's already received
2171 if (migrate_postcopy_ram()) {
2172 ram_postcopy_send_discard_bitmap(ms
);
2176 * send rest of state - note things that are doing postcopy
2177 * will notice we're in POSTCOPY_ACTIVE and not actually
2178 * wrap their state up here
2180 /* 0 max-postcopy-bandwidth means unlimited */
2182 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
2184 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
2186 if (migrate_postcopy_ram()) {
2187 /* Ping just for debugging, helps line traces up */
2188 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
2192 * While loading the device state we may trigger page transfer
2193 * requests and the fd must be free to process those, and thus
2194 * the destination must read the whole device state off the fd before
2195 * it starts processing it. Unfortunately the ad-hoc migration format
2196 * doesn't allow the destination to know the size to read without fully
2197 * parsing it through each devices load-state code (especially the open
2198 * coded devices that use get/put).
2199 * So we wrap the device state up in a package with a length at the start;
2200 * to do this we use a qemu_buf to hold the whole of the device state.
2202 bioc
= qio_channel_buffer_new(4096);
2203 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
2204 fb
= qemu_file_new_output(QIO_CHANNEL(bioc
));
2205 object_unref(OBJECT(bioc
));
2208 * Make sure the receiver can get incoming pages before we send the rest
2211 qemu_savevm_send_postcopy_listen(fb
);
2213 qemu_savevm_state_complete_precopy(fb
, false, false);
2214 if (migrate_postcopy_ram()) {
2215 qemu_savevm_send_ping(fb
, 3);
2218 qemu_savevm_send_postcopy_run(fb
);
2220 /* <><> end of stuff going into the package */
2222 /* Last point of recovery; as soon as we send the package the destination
2223 * can open devices and potentially start running.
2224 * Lets just check again we've not got any errors.
2226 ret
= qemu_file_get_error(ms
->to_dst_file
);
2228 error_report("postcopy_start: Migration stream errored (pre package)");
2232 restart_block
= false;
2234 /* Now send that blob */
2235 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2240 /* Send a notify to give a chance for anything that needs to happen
2241 * at the transition to postcopy and after the device state; in particular
2242 * spice needs to trigger a transition now
2244 ms
->postcopy_after_devices
= true;
2245 notifier_list_notify(&migration_state_notifiers
, ms
);
2247 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2249 qemu_mutex_unlock_iothread();
2251 if (migrate_postcopy_ram()) {
2253 * Although this ping is just for debug, it could potentially be
2254 * used for getting a better measurement of downtime at the source.
2256 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2259 if (migrate_release_ram()) {
2260 ram_postcopy_migrated_memory_release(ms
);
2263 ret
= qemu_file_get_error(ms
->to_dst_file
);
2265 error_report("postcopy_start: Migration stream errored");
2266 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2267 MIGRATION_STATUS_FAILED
);
2270 trace_postcopy_preempt_enabled(migrate_postcopy_preempt());
2277 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2278 MIGRATION_STATUS_FAILED
);
2279 if (restart_block
) {
2280 /* A failure happened early enough that we know the destination hasn't
2281 * accessed block devices, so we're safe to recover.
2283 Error
*local_err
= NULL
;
2285 bdrv_activate_all(&local_err
);
2287 error_report_err(local_err
);
2290 qemu_mutex_unlock_iothread();
2295 * migration_maybe_pause: Pause if required to by
2296 * migrate_pause_before_switchover called with the iothread locked
2297 * Returns: 0 on success
2299 static int migration_maybe_pause(MigrationState
*s
,
2300 int *current_active_state
,
2303 if (!migrate_pause_before_switchover()) {
2307 /* Since leaving this state is not atomic with posting the semaphore
2308 * it's possible that someone could have issued multiple migrate_continue
2309 * and the semaphore is incorrectly positive at this point;
2310 * the docs say it's undefined to reinit a semaphore that's already
2311 * init'd, so use timedwait to eat up any existing posts.
2313 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2314 /* This block intentionally left blank */
2318 * If the migration is cancelled when it is in the completion phase,
2319 * the migration state is set to MIGRATION_STATUS_CANCELLING.
2320 * So we don't need to wait a semaphore, otherwise we would always
2321 * wait for the 'pause_sem' semaphore.
2323 if (s
->state
!= MIGRATION_STATUS_CANCELLING
) {
2324 qemu_mutex_unlock_iothread();
2325 migrate_set_state(&s
->state
, *current_active_state
,
2326 MIGRATION_STATUS_PRE_SWITCHOVER
);
2327 qemu_sem_wait(&s
->pause_sem
);
2328 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2330 *current_active_state
= new_state
;
2331 qemu_mutex_lock_iothread();
2334 return s
->state
== new_state
? 0 : -EINVAL
;
2338 * migration_completion: Used by migration_thread when there's not much left.
2339 * The caller 'breaks' the loop when this returns.
2341 * @s: Current migration state
2343 static void migration_completion(MigrationState
*s
)
2346 int current_active_state
= s
->state
;
2348 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2349 qemu_mutex_lock_iothread();
2350 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2351 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2352 s
->vm_was_running
= runstate_is_running();
2353 ret
= global_state_store();
2356 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2357 trace_migration_completion_vm_stop(ret
);
2359 ret
= migration_maybe_pause(s
, ¤t_active_state
,
2360 MIGRATION_STATUS_DEVICE
);
2363 s
->block_inactive
= !migrate_colo();
2364 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
2365 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
2369 qemu_mutex_unlock_iothread();
2374 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2375 trace_migration_completion_postcopy_end();
2377 qemu_mutex_lock_iothread();
2378 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
2379 qemu_mutex_unlock_iothread();
2382 * Shutdown the postcopy fast path thread. This is only needed
2383 * when dest QEMU binary is old (7.1/7.2). QEMU 8.0+ doesn't need
2386 if (migrate_postcopy_preempt() && s
->preempt_pre_7_2
) {
2387 postcopy_preempt_shutdown_file(s
);
2390 trace_migration_completion_postcopy_end_after_complete();
2396 * If rp was opened we must clean up the thread before
2397 * cleaning everything else up (since if there are no failures
2398 * it will wait for the destination to send it's status in
2401 if (s
->rp_state
.rp_thread_created
) {
2403 trace_migration_return_path_end_before();
2404 rp_error
= await_return_path_close_on_source(s
);
2405 trace_migration_return_path_end_after(rp_error
);
2407 goto fail_invalidate
;
2411 if (qemu_file_get_error(s
->to_dst_file
)) {
2412 trace_migration_completion_file_err();
2413 goto fail_invalidate
;
2416 if (migrate_colo() && s
->state
== MIGRATION_STATUS_ACTIVE
) {
2417 /* COLO does not support postcopy */
2418 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
2419 MIGRATION_STATUS_COLO
);
2421 migrate_set_state(&s
->state
, current_active_state
,
2422 MIGRATION_STATUS_COMPLETED
);
2428 /* If not doing postcopy, vm_start() will be called: let's regain
2429 * control on images.
2431 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
2432 s
->state
== MIGRATION_STATUS_DEVICE
) {
2433 Error
*local_err
= NULL
;
2435 qemu_mutex_lock_iothread();
2436 bdrv_activate_all(&local_err
);
2438 error_report_err(local_err
);
2439 s
->block_inactive
= true;
2441 s
->block_inactive
= false;
2443 qemu_mutex_unlock_iothread();
2447 migrate_set_state(&s
->state
, current_active_state
,
2448 MIGRATION_STATUS_FAILED
);
2452 * bg_migration_completion: Used by bg_migration_thread when after all the
2453 * RAM has been saved. The caller 'breaks' the loop when this returns.
2455 * @s: Current migration state
2457 static void bg_migration_completion(MigrationState
*s
)
2459 int current_active_state
= s
->state
;
2462 * Stop tracking RAM writes - un-protect memory, un-register UFFD
2463 * memory ranges, flush kernel wait queues and wake up threads
2464 * waiting for write fault to be resolved.
2466 ram_write_tracking_stop();
2468 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2470 * By this moment we have RAM content saved into the migration stream.
2471 * The next step is to flush the non-RAM content (device state)
2472 * right after the ram content. The device state has been stored into
2473 * the temporary buffer before RAM saving started.
2475 qemu_put_buffer(s
->to_dst_file
, s
->bioc
->data
, s
->bioc
->usage
);
2476 qemu_fflush(s
->to_dst_file
);
2477 } else if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
2481 if (qemu_file_get_error(s
->to_dst_file
)) {
2482 trace_migration_completion_file_err();
2486 migrate_set_state(&s
->state
, current_active_state
,
2487 MIGRATION_STATUS_COMPLETED
);
2491 migrate_set_state(&s
->state
, current_active_state
,
2492 MIGRATION_STATUS_FAILED
);
2495 typedef enum MigThrError
{
2496 /* No error detected */
2497 MIG_THR_ERR_NONE
= 0,
2498 /* Detected error, but resumed successfully */
2499 MIG_THR_ERR_RECOVERED
= 1,
2500 /* Detected fatal error, need to exit */
2501 MIG_THR_ERR_FATAL
= 2,
2504 static int postcopy_resume_handshake(MigrationState
*s
)
2506 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
2508 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2509 qemu_sem_wait(&s
->rp_state
.rp_sem
);
2512 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2519 /* Return zero if success, or <0 for error */
2520 static int postcopy_do_resume(MigrationState
*s
)
2525 * Call all the resume_prepare() hooks, so that modules can be
2526 * ready for the migration resume.
2528 ret
= qemu_savevm_state_resume_prepare(s
);
2530 error_report("%s: resume_prepare() failure detected: %d",
2536 * If preempt is enabled, re-establish the preempt channel. Note that
2537 * we do it after resume prepare to make sure the main channel will be
2538 * created before the preempt channel. E.g. with weak network, the
2539 * dest QEMU may get messed up with the preempt and main channels on
2540 * the order of connection setup. This guarantees the correct order.
2542 ret
= postcopy_preempt_establish_channel(s
);
2544 error_report("%s: postcopy_preempt_establish_channel(): %d",
2550 * Last handshake with destination on the resume (destination will
2551 * switch to postcopy-active afterwards)
2553 ret
= postcopy_resume_handshake(s
);
2555 error_report("%s: handshake failed: %d", __func__
, ret
);
2563 * We don't return until we are in a safe state to continue current
2564 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2565 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2567 static MigThrError
postcopy_pause(MigrationState
*s
)
2569 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2575 * Current channel is possibly broken. Release it. Note that this is
2576 * guaranteed even without lock because to_dst_file should only be
2577 * modified by the migration thread. That also guarantees that the
2578 * unregister of yank is safe too without the lock. It should be safe
2579 * even to be within the qemu_file_lock, but we didn't do that to avoid
2580 * taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make
2581 * the qemu_file_lock critical section as small as possible.
2583 assert(s
->to_dst_file
);
2584 migration_ioc_unregister_yank_from_file(s
->to_dst_file
);
2585 qemu_mutex_lock(&s
->qemu_file_lock
);
2586 file
= s
->to_dst_file
;
2587 s
->to_dst_file
= NULL
;
2588 qemu_mutex_unlock(&s
->qemu_file_lock
);
2590 qemu_file_shutdown(file
);
2593 migrate_set_state(&s
->state
, s
->state
,
2594 MIGRATION_STATUS_POSTCOPY_PAUSED
);
2596 error_report("Detected IO failure for postcopy. "
2597 "Migration paused.");
2600 * We wait until things fixed up. Then someone will setup the
2601 * status back for us.
2603 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2604 qemu_sem_wait(&s
->postcopy_pause_sem
);
2607 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2608 /* Woken up by a recover procedure. Give it a shot */
2611 * Firstly, let's wake up the return path now, with a new
2612 * return path channel.
2614 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
2616 /* Do the resume logic */
2617 if (postcopy_do_resume(s
) == 0) {
2618 /* Let's continue! */
2619 trace_postcopy_pause_continued();
2620 return MIG_THR_ERR_RECOVERED
;
2623 * Something wrong happened during the recovery, let's
2624 * pause again. Pause is always better than throwing
2630 /* This is not right... Time to quit. */
2631 return MIG_THR_ERR_FATAL
;
2636 static MigThrError
migration_detect_error(MigrationState
*s
)
2639 int state
= s
->state
;
2640 Error
*local_error
= NULL
;
2642 if (state
== MIGRATION_STATUS_CANCELLING
||
2643 state
== MIGRATION_STATUS_CANCELLED
) {
2644 /* End the migration, but don't set the state to failed */
2645 return MIG_THR_ERR_FATAL
;
2649 * Try to detect any file errors. Note that postcopy_qemufile_src will
2650 * be NULL when postcopy preempt is not enabled.
2652 ret
= qemu_file_get_error_obj_any(s
->to_dst_file
,
2653 s
->postcopy_qemufile_src
,
2656 /* Everything is fine */
2657 assert(!local_error
);
2658 return MIG_THR_ERR_NONE
;
2662 migrate_set_error(s
, local_error
);
2663 error_free(local_error
);
2666 if (state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
) {
2668 * For postcopy, we allow the network to be down for a
2669 * while. After that, it can be continued by a
2672 return postcopy_pause(s
);
2675 * For precopy (or postcopy with error outside IO), we fail
2678 migrate_set_state(&s
->state
, state
, MIGRATION_STATUS_FAILED
);
2679 trace_migration_thread_file_err();
2681 /* Time to stop the migration, now. */
2682 return MIG_THR_ERR_FATAL
;
2686 /* How many bytes have we transferred since the beginning of the migration */
2687 static uint64_t migration_total_bytes(MigrationState
*s
)
2689 return qemu_file_total_transferred(s
->to_dst_file
) +
2690 stat64_get(&ram_counters
.multifd_bytes
);
2693 static void migration_calculate_complete(MigrationState
*s
)
2695 uint64_t bytes
= migration_total_bytes(s
);
2696 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2697 int64_t transfer_time
;
2699 s
->total_time
= end_time
- s
->start_time
;
2702 * It's still not set, so we are precopy migration. For
2703 * postcopy, downtime is calculated during postcopy_start().
2705 s
->downtime
= end_time
- s
->downtime_start
;
2708 transfer_time
= s
->total_time
- s
->setup_time
;
2709 if (transfer_time
) {
2710 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
2714 static void update_iteration_initial_status(MigrationState
*s
)
2717 * Update these three fields at the same time to avoid mismatch info lead
2718 * wrong speed calculation.
2720 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2721 s
->iteration_initial_bytes
= migration_total_bytes(s
);
2722 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
2725 static void migration_update_counters(MigrationState
*s
,
2726 int64_t current_time
)
2728 uint64_t transferred
, transferred_pages
, time_spent
;
2729 uint64_t current_bytes
; /* bytes transferred since the beginning */
2732 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
2736 current_bytes
= migration_total_bytes(s
);
2737 transferred
= current_bytes
- s
->iteration_initial_bytes
;
2738 time_spent
= current_time
- s
->iteration_start_time
;
2739 bandwidth
= (double)transferred
/ time_spent
;
2740 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
2742 s
->mbps
= (((double) transferred
* 8.0) /
2743 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
2745 transferred_pages
= ram_get_total_transferred_pages() -
2746 s
->iteration_initial_pages
;
2747 s
->pages_per_second
= (double) transferred_pages
/
2748 (((double) time_spent
/ 1000.0));
2751 * if we haven't sent anything, we don't want to
2752 * recalculate. 10000 is a small enough number for our purposes
2754 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
2755 s
->expected_downtime
= ram_counters
.remaining
/ bandwidth
;
2758 qemu_file_reset_rate_limit(s
->to_dst_file
);
2760 update_iteration_initial_status(s
);
2762 trace_migrate_transferred(transferred
, time_spent
,
2763 bandwidth
, s
->threshold_size
);
2766 /* Migration thread iteration status */
2768 MIG_ITERATE_RESUME
, /* Resume current iteration */
2769 MIG_ITERATE_SKIP
, /* Skip current iteration */
2770 MIG_ITERATE_BREAK
, /* Break the loop */
2774 * Return true if continue to the next iteration directly, false
2777 static MigIterateState
migration_iteration_run(MigrationState
*s
)
2779 uint64_t must_precopy
, can_postcopy
;
2780 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
2782 qemu_savevm_state_pending_estimate(&must_precopy
, &can_postcopy
);
2783 uint64_t pending_size
= must_precopy
+ can_postcopy
;
2785 trace_migrate_pending_estimate(pending_size
, must_precopy
, can_postcopy
);
2787 if (must_precopy
<= s
->threshold_size
) {
2788 qemu_savevm_state_pending_exact(&must_precopy
, &can_postcopy
);
2789 pending_size
= must_precopy
+ can_postcopy
;
2790 trace_migrate_pending_exact(pending_size
, must_precopy
, can_postcopy
);
2793 if (!pending_size
|| pending_size
< s
->threshold_size
) {
2794 trace_migration_thread_low_pending(pending_size
);
2795 migration_completion(s
);
2796 return MIG_ITERATE_BREAK
;
2799 /* Still a significant amount to transfer */
2800 if (!in_postcopy
&& must_precopy
<= s
->threshold_size
&&
2801 qatomic_read(&s
->start_postcopy
)) {
2802 if (postcopy_start(s
)) {
2803 error_report("%s: postcopy failed to start", __func__
);
2805 return MIG_ITERATE_SKIP
;
2808 /* Just another iteration step */
2809 qemu_savevm_state_iterate(s
->to_dst_file
, in_postcopy
);
2810 return MIG_ITERATE_RESUME
;
2813 static void migration_iteration_finish(MigrationState
*s
)
2815 /* If we enabled cpu throttling for auto-converge, turn it off. */
2816 cpu_throttle_stop();
2818 qemu_mutex_lock_iothread();
2820 case MIGRATION_STATUS_COMPLETED
:
2821 migration_calculate_complete(s
);
2822 runstate_set(RUN_STATE_POSTMIGRATE
);
2824 case MIGRATION_STATUS_COLO
:
2825 if (!migrate_colo()) {
2826 error_report("%s: critical error: calling COLO code without "
2827 "COLO enabled", __func__
);
2829 migrate_start_colo_process(s
);
2830 s
->vm_was_running
= true;
2832 case MIGRATION_STATUS_FAILED
:
2833 case MIGRATION_STATUS_CANCELLED
:
2834 case MIGRATION_STATUS_CANCELLING
:
2835 if (s
->vm_was_running
) {
2836 if (!runstate_check(RUN_STATE_SHUTDOWN
)) {
2840 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
2841 runstate_set(RUN_STATE_POSTMIGRATE
);
2847 /* Should not reach here, but if so, forgive the VM. */
2848 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
2851 migrate_fd_cleanup_schedule(s
);
2852 qemu_mutex_unlock_iothread();
2855 static void bg_migration_iteration_finish(MigrationState
*s
)
2857 qemu_mutex_lock_iothread();
2859 case MIGRATION_STATUS_COMPLETED
:
2860 migration_calculate_complete(s
);
2863 case MIGRATION_STATUS_ACTIVE
:
2864 case MIGRATION_STATUS_FAILED
:
2865 case MIGRATION_STATUS_CANCELLED
:
2866 case MIGRATION_STATUS_CANCELLING
:
2870 /* Should not reach here, but if so, forgive the VM. */
2871 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
2875 migrate_fd_cleanup_schedule(s
);
2876 qemu_mutex_unlock_iothread();
2880 * Return true if continue to the next iteration directly, false
2883 static MigIterateState
bg_migration_iteration_run(MigrationState
*s
)
2887 res
= qemu_savevm_state_iterate(s
->to_dst_file
, false);
2889 bg_migration_completion(s
);
2890 return MIG_ITERATE_BREAK
;
2893 return MIG_ITERATE_RESUME
;
2896 void migration_make_urgent_request(void)
2898 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
2901 void migration_consume_urgent_request(void)
2903 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
2906 /* Returns true if the rate limiting was broken by an urgent request */
2907 bool migration_rate_limit(void)
2909 int64_t now
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2910 MigrationState
*s
= migrate_get_current();
2912 bool urgent
= false;
2913 migration_update_counters(s
, now
);
2914 if (qemu_file_rate_limit(s
->to_dst_file
)) {
2916 if (qemu_file_get_error(s
->to_dst_file
)) {
2920 * Wait for a delay to do rate limiting OR
2921 * something urgent to post the semaphore.
2923 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- now
;
2924 trace_migration_rate_limit_pre(ms
);
2925 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
2927 * We were woken by one or more urgent things but
2928 * the timedwait will have consumed one of them.
2929 * The service routine for the urgent wake will dec
2930 * the semaphore itself for each item it consumes,
2931 * so add this one we just eat back.
2933 qemu_sem_post(&s
->rate_limit_sem
);
2936 trace_migration_rate_limit_post(urgent
);
2942 * if failover devices are present, wait they are completely
2946 static void qemu_savevm_wait_unplug(MigrationState
*s
, int old_state
,
2949 if (qemu_savevm_state_guest_unplug_pending()) {
2950 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_WAIT_UNPLUG
);
2952 while (s
->state
== MIGRATION_STATUS_WAIT_UNPLUG
&&
2953 qemu_savevm_state_guest_unplug_pending()) {
2954 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
2956 if (s
->state
!= MIGRATION_STATUS_WAIT_UNPLUG
) {
2957 int timeout
= 120; /* 30 seconds */
2959 * migration has been canceled
2960 * but as we have started an unplug we must wait the end
2961 * to be able to plug back the card
2963 while (timeout
-- && qemu_savevm_state_guest_unplug_pending()) {
2964 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
2966 if (qemu_savevm_state_guest_unplug_pending() &&
2968 warn_report("migration: partially unplugged device on "
2973 migrate_set_state(&s
->state
, MIGRATION_STATUS_WAIT_UNPLUG
, new_state
);
2975 migrate_set_state(&s
->state
, old_state
, new_state
);
2980 * Master migration thread on the source VM.
2981 * It drives the migration and pumps the data down the outgoing channel.
2983 static void *migration_thread(void *opaque
)
2985 MigrationState
*s
= opaque
;
2986 MigrationThread
*thread
= NULL
;
2987 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
2988 MigThrError thr_error
;
2989 bool urgent
= false;
2991 thread
= MigrationThreadAdd("live_migration", qemu_get_thread_id());
2993 rcu_register_thread();
2995 object_ref(OBJECT(s
));
2996 update_iteration_initial_status(s
);
2998 qemu_savevm_state_header(s
->to_dst_file
);
3001 * If we opened the return path, we need to make sure dst has it
3004 if (s
->rp_state
.rp_thread_created
) {
3005 /* Now tell the dest that it should open its end so it can reply */
3006 qemu_savevm_send_open_return_path(s
->to_dst_file
);
3008 /* And do a ping that will make stuff easier to debug */
3009 qemu_savevm_send_ping(s
->to_dst_file
, 1);
3012 if (migrate_postcopy()) {
3014 * Tell the destination that we *might* want to do postcopy later;
3015 * if the other end can't do postcopy it should fail now, nice and
3018 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
3021 if (migrate_colo()) {
3022 /* Notify migration destination that we enable COLO */
3023 qemu_savevm_send_colo_enable(s
->to_dst_file
);
3026 qemu_savevm_state_setup(s
->to_dst_file
);
3028 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
3029 MIGRATION_STATUS_ACTIVE
);
3031 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3033 trace_migration_thread_setup_complete();
3035 while (migration_is_active(s
)) {
3036 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
3037 MigIterateState iter_state
= migration_iteration_run(s
);
3038 if (iter_state
== MIG_ITERATE_SKIP
) {
3040 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3046 * Try to detect any kind of failures, and see whether we
3047 * should stop the migration now.
3049 thr_error
= migration_detect_error(s
);
3050 if (thr_error
== MIG_THR_ERR_FATAL
) {
3051 /* Stop migration */
3053 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
3055 * Just recovered from a e.g. network failure, reset all
3056 * the local variables. This is important to avoid
3057 * breaking transferred_bytes and bandwidth calculation
3059 update_iteration_initial_status(s
);
3062 urgent
= migration_rate_limit();
3065 trace_migration_thread_after_loop();
3066 migration_iteration_finish(s
);
3067 object_unref(OBJECT(s
));
3068 rcu_unregister_thread();
3069 MigrationThreadDel(thread
);
3073 static void bg_migration_vm_start_bh(void *opaque
)
3075 MigrationState
*s
= opaque
;
3077 qemu_bh_delete(s
->vm_start_bh
);
3078 s
->vm_start_bh
= NULL
;
3081 s
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - s
->downtime_start
;
3085 * Background snapshot thread, based on live migration code.
3086 * This is an alternative implementation of live migration mechanism
3087 * introduced specifically to support background snapshots.
3089 * It takes advantage of userfault_fd write protection mechanism introduced
3090 * in v5.7 kernel. Compared to existing dirty page logging migration much
3091 * lesser stream traffic is produced resulting in smaller snapshot images,
3092 * simply cause of no page duplicates can get into the stream.
3094 * Another key point is that generated vmstate stream reflects machine state
3095 * 'frozen' at the beginning of snapshot creation compared to dirty page logging
3096 * mechanism, which effectively results in that saved snapshot is the state of VM
3097 * at the end of the process.
3099 static void *bg_migration_thread(void *opaque
)
3101 MigrationState
*s
= opaque
;
3102 int64_t setup_start
;
3103 MigThrError thr_error
;
3105 bool early_fail
= true;
3107 rcu_register_thread();
3108 object_ref(OBJECT(s
));
3110 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
3112 setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3114 * We want to save vmstate for the moment when migration has been
3115 * initiated but also we want to save RAM content while VM is running.
3116 * The RAM content should appear first in the vmstate. So, we first
3117 * stash the non-RAM part of the vmstate to the temporary buffer,
3118 * then write RAM part of the vmstate to the migration stream
3119 * with vCPUs running and, finally, write stashed non-RAM part of
3120 * the vmstate from the buffer to the migration stream.
3122 s
->bioc
= qio_channel_buffer_new(512 * 1024);
3123 qio_channel_set_name(QIO_CHANNEL(s
->bioc
), "vmstate-buffer");
3124 fb
= qemu_file_new_output(QIO_CHANNEL(s
->bioc
));
3125 object_unref(OBJECT(s
->bioc
));
3127 update_iteration_initial_status(s
);
3130 * Prepare for tracking memory writes with UFFD-WP - populate
3131 * RAM pages before protecting.
3134 ram_write_tracking_prepare();
3137 qemu_savevm_state_header(s
->to_dst_file
);
3138 qemu_savevm_state_setup(s
->to_dst_file
);
3140 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
3141 MIGRATION_STATUS_ACTIVE
);
3143 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3145 trace_migration_thread_setup_complete();
3146 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3148 qemu_mutex_lock_iothread();
3151 * If VM is currently in suspended state, then, to make a valid runstate
3152 * transition in vm_stop_force_state() we need to wakeup it up.
3154 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3155 s
->vm_was_running
= runstate_is_running();
3157 if (global_state_store()) {
3160 /* Forcibly stop VM before saving state of vCPUs and devices */
3161 if (vm_stop_force_state(RUN_STATE_PAUSED
)) {
3165 * Put vCPUs in sync with shadow context structures, then
3166 * save their state to channel-buffer along with devices.
3168 cpu_synchronize_all_states();
3169 if (qemu_savevm_state_complete_precopy_non_iterable(fb
, false, false)) {
3173 * Since we are going to get non-iterable state data directly
3174 * from s->bioc->data, explicit flush is needed here.
3178 /* Now initialize UFFD context and start tracking RAM writes */
3179 if (ram_write_tracking_start()) {
3185 * Start VM from BH handler to avoid write-fault lock here.
3186 * UFFD-WP protection for the whole RAM is already enabled so
3187 * calling VM state change notifiers from vm_start() would initiate
3188 * writes to virtio VQs memory which is in write-protected region.
3190 s
->vm_start_bh
= qemu_bh_new(bg_migration_vm_start_bh
, s
);
3191 qemu_bh_schedule(s
->vm_start_bh
);
3193 qemu_mutex_unlock_iothread();
3195 while (migration_is_active(s
)) {
3196 MigIterateState iter_state
= bg_migration_iteration_run(s
);
3197 if (iter_state
== MIG_ITERATE_SKIP
) {
3199 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3204 * Try to detect any kind of failures, and see whether we
3205 * should stop the migration now.
3207 thr_error
= migration_detect_error(s
);
3208 if (thr_error
== MIG_THR_ERR_FATAL
) {
3209 /* Stop migration */
3213 migration_update_counters(s
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
));
3216 trace_migration_thread_after_loop();
3220 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
3221 MIGRATION_STATUS_FAILED
);
3222 qemu_mutex_unlock_iothread();
3225 bg_migration_iteration_finish(s
);
3228 object_unref(OBJECT(s
));
3229 rcu_unregister_thread();
3234 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
3236 Error
*local_err
= NULL
;
3238 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
3241 * If there's a previous error, free it and prepare for another one.
3242 * Meanwhile if migration completes successfully, there won't have an error
3243 * dumped when calling migrate_fd_cleanup().
3245 migrate_error_free(s
);
3247 s
->expected_downtime
= s
->parameters
.downtime_limit
;
3249 assert(s
->cleanup_bh
);
3251 assert(!s
->cleanup_bh
);
3252 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup_bh
, s
);
3255 migrate_fd_error(s
, error_in
);
3258 * Don't do cleanup for resume if channel is invalid, but only dump
3259 * the error. We wait for another channel connect from the user.
3260 * The error_report still gives HMP user a hint on what failed.
3261 * It's normally done in migrate_fd_cleanup(), but call it here
3264 error_report_err(error_copy(s
->error
));
3266 migrate_fd_cleanup(s
);
3272 /* This is a resumed migration */
3273 rate_limit
= migrate_max_postcopy_bandwidth() /
3276 /* This is a fresh new migration */
3277 rate_limit
= migrate_max_bandwidth() / XFER_LIMIT_RATIO
;
3279 /* Notify before starting migration thread */
3280 notifier_list_notify(&migration_state_notifiers
, s
);
3283 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
3284 qemu_file_set_blocking(s
->to_dst_file
, true);
3287 * Open the return path. For postcopy, it is used exclusively. For
3288 * precopy, only if user specified "return-path" capability would
3289 * QEMU uses the return path.
3291 if (migrate_postcopy_ram() || migrate_return_path()) {
3292 if (open_return_path_on_source(s
, !resume
)) {
3293 error_report("Unable to open return-path for postcopy");
3294 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
3295 migrate_fd_cleanup(s
);
3301 * This needs to be done before resuming a postcopy. Note: for newer
3302 * QEMUs we will delay the channel creation until postcopy_start(), to
3303 * avoid disorder of channel creations.
3305 if (migrate_postcopy_preempt() && s
->preempt_pre_7_2
) {
3306 postcopy_preempt_setup(s
);
3310 /* Wakeup the main migration thread to do the recovery */
3311 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
3312 MIGRATION_STATUS_POSTCOPY_RECOVER
);
3313 qemu_sem_post(&s
->postcopy_pause_sem
);
3317 if (multifd_save_setup(&local_err
) != 0) {
3318 error_report_err(local_err
);
3319 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3320 MIGRATION_STATUS_FAILED
);
3321 migrate_fd_cleanup(s
);
3325 if (migrate_background_snapshot()) {
3326 qemu_thread_create(&s
->thread
, "bg_snapshot",
3327 bg_migration_thread
, s
, QEMU_THREAD_JOINABLE
);
3329 qemu_thread_create(&s
->thread
, "live_migration",
3330 migration_thread
, s
, QEMU_THREAD_JOINABLE
);
3332 s
->migration_thread_running
= true;
3335 #define DEFINE_PROP_MIG_CAP(name, x) \
3336 DEFINE_PROP_BOOL(name, MigrationState, capabilities[x], false)
3338 static Property migration_properties
[] = {
3339 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
3340 store_global_state
, true),
3341 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
3342 send_configuration
, true),
3343 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
3344 send_section_footer
, true),
3345 DEFINE_PROP_BOOL("decompress-error-check", MigrationState
,
3346 decompress_error_check
, true),
3347 DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState
,
3348 clear_bitmap_shift
, CLEAR_BITMAP_SHIFT_DEFAULT
),
3349 DEFINE_PROP_BOOL("x-preempt-pre-7-2", MigrationState
,
3350 preempt_pre_7_2
, false),
3352 /* Migration parameters */
3353 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
3354 parameters
.compress_level
,
3355 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
3356 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
3357 parameters
.compress_threads
,
3358 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
3359 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState
,
3360 parameters
.compress_wait_thread
, true),
3361 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
3362 parameters
.decompress_threads
,
3363 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
3364 DEFINE_PROP_UINT8("x-throttle-trigger-threshold", MigrationState
,
3365 parameters
.throttle_trigger_threshold
,
3366 DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD
),
3367 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
3368 parameters
.cpu_throttle_initial
,
3369 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
3370 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
3371 parameters
.cpu_throttle_increment
,
3372 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
3373 DEFINE_PROP_BOOL("x-cpu-throttle-tailslow", MigrationState
,
3374 parameters
.cpu_throttle_tailslow
, false),
3375 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
3376 parameters
.max_bandwidth
, MAX_THROTTLE
),
3377 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
3378 parameters
.downtime_limit
,
3379 DEFAULT_MIGRATE_SET_DOWNTIME
),
3380 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
3381 parameters
.x_checkpoint_delay
,
3382 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
3383 DEFINE_PROP_UINT8("multifd-channels", MigrationState
,
3384 parameters
.multifd_channels
,
3385 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
3386 DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState
,
3387 parameters
.multifd_compression
,
3388 DEFAULT_MIGRATE_MULTIFD_COMPRESSION
),
3389 DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState
,
3390 parameters
.multifd_zlib_level
,
3391 DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL
),
3392 DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState
,
3393 parameters
.multifd_zstd_level
,
3394 DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL
),
3395 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
3396 parameters
.xbzrle_cache_size
,
3397 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
3398 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState
,
3399 parameters
.max_postcopy_bandwidth
,
3400 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH
),
3401 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState
,
3402 parameters
.max_cpu_throttle
,
3403 DEFAULT_MIGRATE_MAX_CPU_THROTTLE
),
3404 DEFINE_PROP_SIZE("announce-initial", MigrationState
,
3405 parameters
.announce_initial
,
3406 DEFAULT_MIGRATE_ANNOUNCE_INITIAL
),
3407 DEFINE_PROP_SIZE("announce-max", MigrationState
,
3408 parameters
.announce_max
,
3409 DEFAULT_MIGRATE_ANNOUNCE_MAX
),
3410 DEFINE_PROP_SIZE("announce-rounds", MigrationState
,
3411 parameters
.announce_rounds
,
3412 DEFAULT_MIGRATE_ANNOUNCE_ROUNDS
),
3413 DEFINE_PROP_SIZE("announce-step", MigrationState
,
3414 parameters
.announce_step
,
3415 DEFAULT_MIGRATE_ANNOUNCE_STEP
),
3416 DEFINE_PROP_STRING("tls-creds", MigrationState
, parameters
.tls_creds
),
3417 DEFINE_PROP_STRING("tls-hostname", MigrationState
, parameters
.tls_hostname
),
3418 DEFINE_PROP_STRING("tls-authz", MigrationState
, parameters
.tls_authz
),
3420 /* Migration capabilities */
3421 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
3422 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
3423 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
3424 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
3425 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
3426 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
3427 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
3428 DEFINE_PROP_MIG_CAP("x-postcopy-preempt",
3429 MIGRATION_CAPABILITY_POSTCOPY_PREEMPT
),
3430 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
3431 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
3432 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
3433 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
3434 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD
),
3435 DEFINE_PROP_MIG_CAP("x-background-snapshot",
3436 MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
),
3438 DEFINE_PROP_MIG_CAP("x-zero-copy-send",
3439 MIGRATION_CAPABILITY_ZERO_COPY_SEND
),
3442 DEFINE_PROP_END_OF_LIST(),
3445 static void migration_class_init(ObjectClass
*klass
, void *data
)
3447 DeviceClass
*dc
= DEVICE_CLASS(klass
);
3449 dc
->user_creatable
= false;
3450 device_class_set_props(dc
, migration_properties
);
3453 static void migration_instance_finalize(Object
*obj
)
3455 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3457 qemu_mutex_destroy(&ms
->error_mutex
);
3458 qemu_mutex_destroy(&ms
->qemu_file_lock
);
3459 qemu_sem_destroy(&ms
->wait_unplug_sem
);
3460 qemu_sem_destroy(&ms
->rate_limit_sem
);
3461 qemu_sem_destroy(&ms
->pause_sem
);
3462 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
3463 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
3464 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
3465 qemu_sem_destroy(&ms
->rp_state
.rp_pong_acks
);
3466 qemu_sem_destroy(&ms
->postcopy_qemufile_src_sem
);
3467 error_free(ms
->error
);
3470 static void migration_instance_init(Object
*obj
)
3472 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3473 MigrationParameters
*params
= &ms
->parameters
;
3475 ms
->state
= MIGRATION_STATUS_NONE
;
3477 ms
->pages_per_second
= -1;
3478 qemu_sem_init(&ms
->pause_sem
, 0);
3479 qemu_mutex_init(&ms
->error_mutex
);
3481 params
->tls_hostname
= g_strdup("");
3482 params
->tls_creds
= g_strdup("");
3484 /* Set has_* up only for parameter checks */
3485 params
->has_compress_level
= true;
3486 params
->has_compress_threads
= true;
3487 params
->has_compress_wait_thread
= true;
3488 params
->has_decompress_threads
= true;
3489 params
->has_throttle_trigger_threshold
= true;
3490 params
->has_cpu_throttle_initial
= true;
3491 params
->has_cpu_throttle_increment
= true;
3492 params
->has_cpu_throttle_tailslow
= true;
3493 params
->has_max_bandwidth
= true;
3494 params
->has_downtime_limit
= true;
3495 params
->has_x_checkpoint_delay
= true;
3496 params
->has_block_incremental
= true;
3497 params
->has_multifd_channels
= true;
3498 params
->has_multifd_compression
= true;
3499 params
->has_multifd_zlib_level
= true;
3500 params
->has_multifd_zstd_level
= true;
3501 params
->has_xbzrle_cache_size
= true;
3502 params
->has_max_postcopy_bandwidth
= true;
3503 params
->has_max_cpu_throttle
= true;
3504 params
->has_announce_initial
= true;
3505 params
->has_announce_max
= true;
3506 params
->has_announce_rounds
= true;
3507 params
->has_announce_step
= true;
3509 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
3510 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
3511 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
3512 qemu_sem_init(&ms
->rp_state
.rp_pong_acks
, 0);
3513 qemu_sem_init(&ms
->rate_limit_sem
, 0);
3514 qemu_sem_init(&ms
->wait_unplug_sem
, 0);
3515 qemu_sem_init(&ms
->postcopy_qemufile_src_sem
, 0);
3516 qemu_mutex_init(&ms
->qemu_file_lock
);
3520 * Return true if check pass, false otherwise. Error will be put
3521 * inside errp if provided.
3523 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
3525 /* Assuming all off */
3526 bool old_caps
[MIGRATION_CAPABILITY__MAX
] = { 0 };
3528 if (!migrate_params_check(&ms
->parameters
, errp
)) {
3532 return migrate_caps_check(old_caps
, ms
->capabilities
, errp
);
3535 static const TypeInfo migration_type
= {
3536 .name
= TYPE_MIGRATION
,
3538 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3539 * not created using qdev_new(), it is not attached to the qdev
3540 * device tree, and it is never realized.
3542 * TODO: Make this TYPE_OBJECT once QOM provides something like
3543 * TYPE_DEVICE's "-global" properties.
3545 .parent
= TYPE_DEVICE
,
3546 .class_init
= migration_class_init
,
3547 .class_size
= sizeof(MigrationClass
),
3548 .instance_size
= sizeof(MigrationState
),
3549 .instance_init
= migration_instance_init
,
3550 .instance_finalize
= migration_instance_finalize
,
3553 static void register_migration_types(void)
3555 type_register_static(&migration_type
);
3558 type_init(register_migration_types
);