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-channel.h"
34 #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 "migration/colo.h"
53 #include "hw/boards.h"
54 #include "hw/qdev-properties.h"
55 #include "monitor/monitor.h"
56 #include "net/announce.h"
57 #include "qemu/queue.h"
60 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
62 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
64 #define BUFFER_DELAY 100
65 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
67 /* Time in milliseconds we are allowed to stop the source,
68 * for sending the last part */
69 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
71 /* Maximum migrate downtime set to 2000 seconds */
72 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
73 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
75 /* Default compression thread count */
76 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
77 /* Default decompression thread count, usually decompression is at
78 * least 4 times as fast as compression.*/
79 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
80 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
81 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
82 /* Define default autoconverge cpu throttle migration parameters */
83 #define DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD 50
84 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
85 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
86 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
88 /* Migration XBZRLE default cache size */
89 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
91 /* The delay time (in ms) between two COLO checkpoints */
92 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
93 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
94 #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE
95 /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */
96 #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
97 /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
98 #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
100 /* Background transfer rate for postcopy, 0 means unlimited, note
101 * that page requests can still exceed this limit.
103 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
106 * Parameters for self_announce_delay giving a stream of RARP/ARP
107 * packets after migration.
109 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
110 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
111 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
112 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
114 static NotifierList migration_state_notifiers
=
115 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
117 static bool deferred_incoming
;
119 /* Messages sent on the return path from destination to source */
120 enum mig_rp_message_type
{
121 MIG_RP_MSG_INVALID
= 0, /* Must be 0 */
122 MIG_RP_MSG_SHUT
, /* sibling will not send any more RP messages */
123 MIG_RP_MSG_PONG
, /* Response to a PING; data (seq: be32 ) */
125 MIG_RP_MSG_REQ_PAGES_ID
, /* data (start: be64, len: be32, id: string) */
126 MIG_RP_MSG_REQ_PAGES
, /* data (start: be64, len: be32) */
127 MIG_RP_MSG_RECV_BITMAP
, /* send recved_bitmap back to source */
128 MIG_RP_MSG_RESUME_ACK
, /* tell source that we are ready to resume */
133 /* When we add fault tolerance, we could have several
134 migrations at once. For now we don't need to add
135 dynamic creation of migration */
137 static MigrationState
*current_migration
;
138 static MigrationIncomingState
*current_incoming
;
140 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
141 static int migration_maybe_pause(MigrationState
*s
,
142 int *current_active_state
,
144 static void migrate_fd_cancel(MigrationState
*s
);
146 void migration_object_init(void)
148 MachineState
*ms
= MACHINE(qdev_get_machine());
151 /* This can only be called once. */
152 assert(!current_migration
);
153 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
156 * Init the migrate incoming object as well no matter whether
157 * we'll use it or not.
159 assert(!current_incoming
);
160 current_incoming
= g_new0(MigrationIncomingState
, 1);
161 current_incoming
->state
= MIGRATION_STATUS_NONE
;
162 current_incoming
->postcopy_remote_fds
=
163 g_array_new(FALSE
, TRUE
, sizeof(struct PostCopyFD
));
164 qemu_mutex_init(¤t_incoming
->rp_mutex
);
165 qemu_event_init(¤t_incoming
->main_thread_load_event
, false);
166 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_dst
, 0);
167 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fault
, 0);
169 if (!migration_object_check(current_migration
, &err
)) {
170 error_report_err(err
);
175 * We cannot really do this in migration_instance_init() since at
176 * that time global properties are not yet applied, then this
177 * value will be definitely replaced by something else.
179 if (ms
->enforce_config_section
) {
180 current_migration
->send_configuration
= true;
184 void migration_shutdown(void)
187 * Cancel the current migration - that will (eventually)
188 * stop the migration using this structure
190 migrate_fd_cancel(current_migration
);
191 object_unref(OBJECT(current_migration
));
194 * Cancel outgoing migration of dirty bitmaps. It should
195 * at least unref used block nodes.
197 dirty_bitmap_mig_cancel_outgoing();
200 * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
201 * are non-critical data, and their loss never considered as
204 dirty_bitmap_mig_cancel_incoming();
208 MigrationState
*migrate_get_current(void)
210 /* This can only be called after the object created. */
211 assert(current_migration
);
212 return current_migration
;
215 MigrationIncomingState
*migration_incoming_get_current(void)
217 assert(current_incoming
);
218 return current_incoming
;
221 void migration_incoming_state_destroy(void)
223 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
225 if (mis
->to_src_file
) {
226 /* Tell source that we are done */
227 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
228 qemu_fclose(mis
->to_src_file
);
229 mis
->to_src_file
= NULL
;
232 if (mis
->from_src_file
) {
233 qemu_fclose(mis
->from_src_file
);
234 mis
->from_src_file
= NULL
;
236 if (mis
->postcopy_remote_fds
) {
237 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
238 mis
->postcopy_remote_fds
= NULL
;
241 if (mis
->socket_address_list
) {
242 qapi_free_SocketAddressList(mis
->socket_address_list
);
243 mis
->socket_address_list
= NULL
;
246 qemu_event_destroy(&mis
->main_thread_load_event
);
247 qemu_sem_destroy(&mis
->postcopy_pause_sem_dst
);
248 qemu_sem_destroy(&mis
->postcopy_pause_sem_fault
);
249 qemu_mutex_destroy(&mis
->rp_mutex
);
252 static void migrate_generate_event(int new_state
)
254 if (migrate_use_events()) {
255 qapi_event_send_migration(new_state
);
259 static bool migrate_late_block_activate(void)
263 s
= migrate_get_current();
265 return s
->enabled_capabilities
[
266 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE
];
270 * Called on -incoming with a defer: uri.
271 * The migration can be started later after any parameters have been
274 static void deferred_incoming_migration(Error
**errp
)
276 if (deferred_incoming
) {
277 error_setg(errp
, "Incoming migration already deferred");
279 deferred_incoming
= true;
283 * Send a message on the return channel back to the source
286 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
287 enum mig_rp_message_type message_type
,
288 uint16_t len
, void *data
)
292 trace_migrate_send_rp_message((int)message_type
, len
);
293 qemu_mutex_lock(&mis
->rp_mutex
);
296 * It's possible that the file handle got lost due to network
299 if (!mis
->to_src_file
) {
304 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
305 qemu_put_be16(mis
->to_src_file
, len
);
306 qemu_put_buffer(mis
->to_src_file
, data
, len
);
307 qemu_fflush(mis
->to_src_file
);
309 /* It's possible that qemu file got error during sending */
310 ret
= qemu_file_get_error(mis
->to_src_file
);
313 qemu_mutex_unlock(&mis
->rp_mutex
);
317 /* Request one page from the source VM at the given start address.
318 * rb: the RAMBlock to request the page in
319 * Start: Address offset within the RB
320 * Len: Length in bytes required - must be a multiple of pagesize
322 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
, RAMBlock
*rb
,
325 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
326 size_t msglen
= 12; /* start + len */
327 size_t len
= qemu_ram_pagesize(rb
);
328 enum mig_rp_message_type msg_type
;
332 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
333 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
336 * We maintain the last ramblock that we requested for page. Note that we
337 * don't need locking because this function will only be called within the
338 * postcopy ram fault thread.
340 if (rb
!= mis
->last_rb
) {
343 rbname
= qemu_ram_get_idstr(rb
);
344 rbname_len
= strlen(rbname
);
346 assert(rbname_len
< 256);
348 bufc
[msglen
++] = rbname_len
;
349 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
350 msglen
+= rbname_len
;
351 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
353 msg_type
= MIG_RP_MSG_REQ_PAGES
;
356 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
359 static bool migration_colo_enabled
;
360 bool migration_incoming_colo_enabled(void)
362 return migration_colo_enabled
;
365 void migration_incoming_disable_colo(void)
367 ram_block_discard_disable(false);
368 migration_colo_enabled
= false;
371 int migration_incoming_enable_colo(void)
373 if (ram_block_discard_disable(true)) {
374 error_report("COLO: cannot disable RAM discard");
377 migration_colo_enabled
= true;
381 void migrate_add_address(SocketAddress
*address
)
383 MigrationIncomingState
*mis
= migration_incoming_get_current();
384 SocketAddressList
*addrs
;
386 addrs
= g_new0(SocketAddressList
, 1);
387 addrs
->next
= mis
->socket_address_list
;
388 mis
->socket_address_list
= addrs
;
389 addrs
->value
= QAPI_CLONE(SocketAddress
, address
);
392 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
394 const char *p
= NULL
;
396 qapi_event_send_migration(MIGRATION_STATUS_SETUP
);
397 if (!strcmp(uri
, "defer")) {
398 deferred_incoming_migration(errp
);
399 } else if (strstart(uri
, "tcp:", &p
) ||
400 strstart(uri
, "unix:", NULL
) ||
401 strstart(uri
, "vsock:", NULL
)) {
402 socket_start_incoming_migration(p
? p
: uri
, errp
);
404 } else if (strstart(uri
, "rdma:", &p
)) {
405 rdma_start_incoming_migration(p
, errp
);
407 } else if (strstart(uri
, "exec:", &p
)) {
408 exec_start_incoming_migration(p
, errp
);
409 } else if (strstart(uri
, "fd:", &p
)) {
410 fd_start_incoming_migration(p
, errp
);
412 error_setg(errp
, "unknown migration protocol: %s", uri
);
416 static void process_incoming_migration_bh(void *opaque
)
418 Error
*local_err
= NULL
;
419 MigrationIncomingState
*mis
= opaque
;
421 /* If capability late_block_activate is set:
422 * Only fire up the block code now if we're going to restart the
423 * VM, else 'cont' will do it.
424 * This causes file locking to happen; so we don't want it to happen
425 * unless we really are starting the VM.
427 if (!migrate_late_block_activate() ||
428 (autostart
&& (!global_state_received() ||
429 global_state_get_runstate() == RUN_STATE_RUNNING
))) {
430 /* Make sure all file formats flush their mutable metadata.
431 * If we get an error here, just don't restart the VM yet. */
432 bdrv_invalidate_cache_all(&local_err
);
434 error_report_err(local_err
);
441 * This must happen after all error conditions are dealt with and
442 * we're sure the VM is going to be running on this host.
444 qemu_announce_self(&mis
->announce_timer
, migrate_announce_params());
446 if (multifd_load_cleanup(&local_err
) != 0) {
447 error_report_err(local_err
);
450 /* If global state section was not received or we are in running
451 state, we need to obey autostart. Any other state is set with
454 dirty_bitmap_mig_before_vm_start();
456 if (!global_state_received() ||
457 global_state_get_runstate() == RUN_STATE_RUNNING
) {
461 runstate_set(RUN_STATE_PAUSED
);
463 } else if (migration_incoming_colo_enabled()) {
464 migration_incoming_disable_colo();
467 runstate_set(global_state_get_runstate());
470 * This must happen after any state changes since as soon as an external
471 * observer sees this event they might start to prod at the VM assuming
474 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
475 MIGRATION_STATUS_COMPLETED
);
476 qemu_bh_delete(mis
->bh
);
477 migration_incoming_state_destroy();
480 static void process_incoming_migration_co(void *opaque
)
482 MigrationIncomingState
*mis
= migration_incoming_get_current();
485 Error
*local_err
= NULL
;
487 assert(mis
->from_src_file
);
488 mis
->migration_incoming_co
= qemu_coroutine_self();
489 mis
->largest_page_size
= qemu_ram_pagesize_largest();
490 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
491 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
492 MIGRATION_STATUS_ACTIVE
);
493 ret
= qemu_loadvm_state(mis
->from_src_file
);
495 ps
= postcopy_state_get();
496 trace_process_incoming_migration_co_end(ret
, ps
);
497 if (ps
!= POSTCOPY_INCOMING_NONE
) {
498 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
500 * Where a migration had postcopy enabled (and thus went to advise)
501 * but managed to complete within the precopy period, we can use
504 postcopy_ram_incoming_cleanup(mis
);
505 } else if (ret
>= 0) {
507 * Postcopy was started, cleanup should happen at the end of the
510 trace_process_incoming_migration_co_postcopy_end_main();
513 /* Else if something went wrong then just fall out of the normal exit */
516 /* we get COLO info, and know if we are in COLO mode */
517 if (!ret
&& migration_incoming_colo_enabled()) {
518 /* Make sure all file formats flush their mutable metadata */
519 bdrv_invalidate_cache_all(&local_err
);
521 error_report_err(local_err
);
525 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
526 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
527 mis
->have_colo_incoming_thread
= true;
528 qemu_coroutine_yield();
530 /* Wait checkpoint incoming thread exit before free resource */
531 qemu_thread_join(&mis
->colo_incoming_thread
);
532 /* We hold the global iothread lock, so it is safe here */
533 colo_release_ram_cache();
537 error_report("load of migration failed: %s", strerror(-ret
));
540 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
541 qemu_bh_schedule(mis
->bh
);
542 mis
->migration_incoming_co
= NULL
;
546 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
547 MIGRATION_STATUS_FAILED
);
548 qemu_fclose(mis
->from_src_file
);
549 if (multifd_load_cleanup(&local_err
) != 0) {
550 error_report_err(local_err
);
556 * @migration_incoming_setup: Setup incoming migration
558 * Returns 0 for no error or 1 for error
560 * @f: file for main migration channel
561 * @errp: where to put errors
563 static int migration_incoming_setup(QEMUFile
*f
, Error
**errp
)
565 MigrationIncomingState
*mis
= migration_incoming_get_current();
566 Error
*local_err
= NULL
;
568 if (multifd_load_setup(&local_err
) != 0) {
569 /* We haven't been able to create multifd threads
570 nothing better to do */
571 error_report_err(local_err
);
575 if (!mis
->from_src_file
) {
576 mis
->from_src_file
= f
;
578 qemu_file_set_blocking(f
, false);
582 void migration_incoming_process(void)
584 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
585 qemu_coroutine_enter(co
);
588 /* Returns true if recovered from a paused migration, otherwise false */
589 static bool postcopy_try_recover(QEMUFile
*f
)
591 MigrationIncomingState
*mis
= migration_incoming_get_current();
593 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
594 /* Resumed from a paused postcopy migration */
596 mis
->from_src_file
= f
;
597 /* Postcopy has standalone thread to do vm load */
598 qemu_file_set_blocking(f
, true);
600 /* Re-configure the return path */
601 mis
->to_src_file
= qemu_file_get_return_path(f
);
603 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
604 MIGRATION_STATUS_POSTCOPY_RECOVER
);
607 * Here, we only wake up the main loading thread (while the
608 * fault thread will still be waiting), so that we can receive
609 * commands from source now, and answer it if needed. The
610 * fault thread will be woken up afterwards until we are sure
611 * that source is ready to reply to page requests.
613 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
620 void migration_fd_process_incoming(QEMUFile
*f
, Error
**errp
)
622 Error
*local_err
= NULL
;
624 if (postcopy_try_recover(f
)) {
628 if (migration_incoming_setup(f
, &local_err
)) {
630 error_propagate(errp
, local_err
);
634 migration_incoming_process();
637 void migration_ioc_process_incoming(QIOChannel
*ioc
, Error
**errp
)
639 MigrationIncomingState
*mis
= migration_incoming_get_current();
640 Error
*local_err
= NULL
;
641 bool start_migration
;
643 if (!mis
->from_src_file
) {
644 /* The first connection (multifd may have multiple) */
645 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
647 /* If it's a recovery, we're done */
648 if (postcopy_try_recover(f
)) {
652 if (migration_incoming_setup(f
, &local_err
)) {
654 error_propagate(errp
, local_err
);
660 * Common migration only needs one channel, so we can start
661 * right now. Multifd needs more than one channel, we wait.
663 start_migration
= !migrate_use_multifd();
665 /* Multiple connections */
666 assert(migrate_use_multifd());
667 start_migration
= multifd_recv_new_channel(ioc
, &local_err
);
669 error_propagate(errp
, local_err
);
674 if (start_migration
) {
675 migration_incoming_process();
680 * @migration_has_all_channels: We have received all channels that we need
682 * Returns true when we have got connections to all the channels that
683 * we need for migration.
685 bool migration_has_all_channels(void)
687 MigrationIncomingState
*mis
= migration_incoming_get_current();
690 all_channels
= multifd_recv_all_channels_created();
692 return all_channels
&& mis
->from_src_file
!= NULL
;
696 * Send a 'SHUT' message on the return channel with the given value
697 * to indicate that we've finished with the RP. Non-0 value indicates
700 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
705 buf
= cpu_to_be32(value
);
706 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
710 * Send a 'PONG' message on the return channel with the given value
711 * (normally in response to a 'PING')
713 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
718 buf
= cpu_to_be32(value
);
719 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
722 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
730 * First, we send the header part. It contains only the len of
731 * idstr, and the idstr itself.
733 len
= strlen(block_name
);
735 memcpy(buf
+ 1, block_name
, len
);
737 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
738 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
743 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
746 * Next, we dump the received bitmap to the stream.
748 * TODO: currently we are safe since we are the only one that is
749 * using the to_src_file handle (fault thread is still paused),
750 * and it's ok even not taking the mutex. However the best way is
751 * to take the lock before sending the message header, and release
752 * the lock after sending the bitmap.
754 qemu_mutex_lock(&mis
->rp_mutex
);
755 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
756 qemu_mutex_unlock(&mis
->rp_mutex
);
758 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
761 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
765 buf
= cpu_to_be32(value
);
766 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
769 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
771 MigrationCapabilityStatusList
*head
= NULL
;
772 MigrationCapabilityStatusList
*caps
;
773 MigrationState
*s
= migrate_get_current();
776 caps
= NULL
; /* silence compiler warning */
777 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
778 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
779 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
784 head
= g_malloc0(sizeof(*caps
));
787 caps
->next
= g_malloc0(sizeof(*caps
));
791 g_malloc(sizeof(*caps
->value
));
792 caps
->value
->capability
= i
;
793 caps
->value
->state
= s
->enabled_capabilities
[i
];
799 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
801 MigrationParameters
*params
;
802 MigrationState
*s
= migrate_get_current();
804 /* TODO use QAPI_CLONE() instead of duplicating it inline */
805 params
= g_malloc0(sizeof(*params
));
806 params
->has_compress_level
= true;
807 params
->compress_level
= s
->parameters
.compress_level
;
808 params
->has_compress_threads
= true;
809 params
->compress_threads
= s
->parameters
.compress_threads
;
810 params
->has_compress_wait_thread
= true;
811 params
->compress_wait_thread
= s
->parameters
.compress_wait_thread
;
812 params
->has_decompress_threads
= true;
813 params
->decompress_threads
= s
->parameters
.decompress_threads
;
814 params
->has_throttle_trigger_threshold
= true;
815 params
->throttle_trigger_threshold
= s
->parameters
.throttle_trigger_threshold
;
816 params
->has_cpu_throttle_initial
= true;
817 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
818 params
->has_cpu_throttle_increment
= true;
819 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
820 params
->has_cpu_throttle_tailslow
= true;
821 params
->cpu_throttle_tailslow
= s
->parameters
.cpu_throttle_tailslow
;
822 params
->has_tls_creds
= true;
823 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
824 params
->has_tls_hostname
= true;
825 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
826 params
->has_tls_authz
= true;
827 params
->tls_authz
= g_strdup(s
->parameters
.tls_authz
?
828 s
->parameters
.tls_authz
: "");
829 params
->has_max_bandwidth
= true;
830 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
831 params
->has_downtime_limit
= true;
832 params
->downtime_limit
= s
->parameters
.downtime_limit
;
833 params
->has_x_checkpoint_delay
= true;
834 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
835 params
->has_block_incremental
= true;
836 params
->block_incremental
= s
->parameters
.block_incremental
;
837 params
->has_multifd_channels
= true;
838 params
->multifd_channels
= s
->parameters
.multifd_channels
;
839 params
->has_multifd_compression
= true;
840 params
->multifd_compression
= s
->parameters
.multifd_compression
;
841 params
->has_multifd_zlib_level
= true;
842 params
->multifd_zlib_level
= s
->parameters
.multifd_zlib_level
;
843 params
->has_multifd_zstd_level
= true;
844 params
->multifd_zstd_level
= s
->parameters
.multifd_zstd_level
;
845 params
->has_xbzrle_cache_size
= true;
846 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
847 params
->has_max_postcopy_bandwidth
= true;
848 params
->max_postcopy_bandwidth
= s
->parameters
.max_postcopy_bandwidth
;
849 params
->has_max_cpu_throttle
= true;
850 params
->max_cpu_throttle
= s
->parameters
.max_cpu_throttle
;
851 params
->has_announce_initial
= true;
852 params
->announce_initial
= s
->parameters
.announce_initial
;
853 params
->has_announce_max
= true;
854 params
->announce_max
= s
->parameters
.announce_max
;
855 params
->has_announce_rounds
= true;
856 params
->announce_rounds
= s
->parameters
.announce_rounds
;
857 params
->has_announce_step
= true;
858 params
->announce_step
= s
->parameters
.announce_step
;
860 if (s
->parameters
.has_block_bitmap_mapping
) {
861 params
->has_block_bitmap_mapping
= true;
862 params
->block_bitmap_mapping
=
863 QAPI_CLONE(BitmapMigrationNodeAliasList
,
864 s
->parameters
.block_bitmap_mapping
);
870 AnnounceParameters
*migrate_announce_params(void)
872 static AnnounceParameters ap
;
874 MigrationState
*s
= migrate_get_current();
876 ap
.initial
= s
->parameters
.announce_initial
;
877 ap
.max
= s
->parameters
.announce_max
;
878 ap
.rounds
= s
->parameters
.announce_rounds
;
879 ap
.step
= s
->parameters
.announce_step
;
885 * Return true if we're already in the middle of a migration
886 * (i.e. any of the active or setup states)
888 bool migration_is_setup_or_active(int state
)
891 case MIGRATION_STATUS_ACTIVE
:
892 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
893 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
894 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
895 case MIGRATION_STATUS_SETUP
:
896 case MIGRATION_STATUS_PRE_SWITCHOVER
:
897 case MIGRATION_STATUS_DEVICE
:
898 case MIGRATION_STATUS_WAIT_UNPLUG
:
899 case MIGRATION_STATUS_COLO
:
908 bool migration_is_running(int state
)
911 case MIGRATION_STATUS_ACTIVE
:
912 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
913 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
914 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
915 case MIGRATION_STATUS_SETUP
:
916 case MIGRATION_STATUS_PRE_SWITCHOVER
:
917 case MIGRATION_STATUS_DEVICE
:
918 case MIGRATION_STATUS_WAIT_UNPLUG
:
919 case MIGRATION_STATUS_CANCELLING
:
928 static void populate_time_info(MigrationInfo
*info
, MigrationState
*s
)
930 info
->has_status
= true;
931 info
->has_setup_time
= true;
932 info
->setup_time
= s
->setup_time
;
933 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
934 info
->has_total_time
= true;
935 info
->total_time
= s
->total_time
;
936 info
->has_downtime
= true;
937 info
->downtime
= s
->downtime
;
939 info
->has_total_time
= true;
940 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) -
942 info
->has_expected_downtime
= true;
943 info
->expected_downtime
= s
->expected_downtime
;
947 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
949 info
->has_ram
= true;
950 info
->ram
= g_malloc0(sizeof(*info
->ram
));
951 info
->ram
->transferred
= ram_counters
.transferred
;
952 info
->ram
->total
= ram_bytes_total();
953 info
->ram
->duplicate
= ram_counters
.duplicate
;
954 /* legacy value. It is not used anymore */
955 info
->ram
->skipped
= 0;
956 info
->ram
->normal
= ram_counters
.normal
;
957 info
->ram
->normal_bytes
= ram_counters
.normal
*
958 qemu_target_page_size();
959 info
->ram
->mbps
= s
->mbps
;
960 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
961 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
962 info
->ram
->page_size
= qemu_target_page_size();
963 info
->ram
->multifd_bytes
= ram_counters
.multifd_bytes
;
964 info
->ram
->pages_per_second
= s
->pages_per_second
;
966 if (migrate_use_xbzrle()) {
967 info
->has_xbzrle_cache
= true;
968 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
969 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
970 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
971 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
972 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
973 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
974 info
->xbzrle_cache
->encoding_rate
= xbzrle_counters
.encoding_rate
;
975 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
978 if (migrate_use_compression()) {
979 info
->has_compression
= true;
980 info
->compression
= g_malloc0(sizeof(*info
->compression
));
981 info
->compression
->pages
= compression_counters
.pages
;
982 info
->compression
->busy
= compression_counters
.busy
;
983 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
984 info
->compression
->compressed_size
=
985 compression_counters
.compressed_size
;
986 info
->compression
->compression_rate
=
987 compression_counters
.compression_rate
;
990 if (cpu_throttle_active()) {
991 info
->has_cpu_throttle_percentage
= true;
992 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
995 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
996 info
->ram
->remaining
= ram_bytes_remaining();
997 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
1001 static void populate_disk_info(MigrationInfo
*info
)
1003 if (blk_mig_active()) {
1004 info
->has_disk
= true;
1005 info
->disk
= g_malloc0(sizeof(*info
->disk
));
1006 info
->disk
->transferred
= blk_mig_bytes_transferred();
1007 info
->disk
->remaining
= blk_mig_bytes_remaining();
1008 info
->disk
->total
= blk_mig_bytes_total();
1012 static void fill_source_migration_info(MigrationInfo
*info
)
1014 MigrationState
*s
= migrate_get_current();
1017 case MIGRATION_STATUS_NONE
:
1018 /* no migration has happened ever */
1019 /* do not overwrite destination migration status */
1021 case MIGRATION_STATUS_SETUP
:
1022 info
->has_status
= true;
1023 info
->has_total_time
= false;
1025 case MIGRATION_STATUS_ACTIVE
:
1026 case MIGRATION_STATUS_CANCELLING
:
1027 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1028 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1029 case MIGRATION_STATUS_DEVICE
:
1030 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1031 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1032 /* TODO add some postcopy stats */
1033 populate_time_info(info
, s
);
1034 populate_ram_info(info
, s
);
1035 populate_disk_info(info
);
1037 case MIGRATION_STATUS_COLO
:
1038 info
->has_status
= true;
1039 /* TODO: display COLO specific information (checkpoint info etc.) */
1041 case MIGRATION_STATUS_COMPLETED
:
1042 populate_time_info(info
, s
);
1043 populate_ram_info(info
, s
);
1045 case MIGRATION_STATUS_FAILED
:
1046 info
->has_status
= true;
1048 info
->has_error_desc
= true;
1049 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
1052 case MIGRATION_STATUS_CANCELLED
:
1053 info
->has_status
= true;
1055 case MIGRATION_STATUS_WAIT_UNPLUG
:
1056 info
->has_status
= true;
1059 info
->status
= s
->state
;
1063 * @migration_caps_check - check capability validity
1065 * @cap_list: old capability list, array of bool
1066 * @params: new capabilities to be applied soon
1067 * @errp: set *errp if the check failed, with reason
1069 * Returns true if check passed, otherwise false.
1071 static bool migrate_caps_check(bool *cap_list
,
1072 MigrationCapabilityStatusList
*params
,
1075 MigrationCapabilityStatusList
*cap
;
1076 bool old_postcopy_cap
;
1077 MigrationIncomingState
*mis
= migration_incoming_get_current();
1079 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1081 for (cap
= params
; cap
; cap
= cap
->next
) {
1082 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
1085 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1086 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
1087 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
1089 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
1094 #ifndef CONFIG_REPLICATION
1095 if (cap_list
[MIGRATION_CAPABILITY_X_COLO
]) {
1096 error_setg(errp
, "QEMU compiled without replication module"
1097 " can't enable COLO");
1098 error_append_hint(errp
, "Please enable replication before COLO.\n");
1103 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
1104 /* This check is reasonably expensive, so only when it's being
1105 * set the first time, also it's only the destination that needs
1108 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
1109 !postcopy_ram_supported_by_host(mis
)) {
1110 /* postcopy_ram_supported_by_host will have emitted a more
1113 error_setg(errp
, "Postcopy is not supported");
1117 if (cap_list
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
]) {
1118 error_setg(errp
, "Postcopy is not compatible with ignore-shared");
1126 static void fill_destination_migration_info(MigrationInfo
*info
)
1128 MigrationIncomingState
*mis
= migration_incoming_get_current();
1130 if (mis
->socket_address_list
) {
1131 info
->has_socket_address
= true;
1132 info
->socket_address
=
1133 QAPI_CLONE(SocketAddressList
, mis
->socket_address_list
);
1136 switch (mis
->state
) {
1137 case MIGRATION_STATUS_NONE
:
1139 case MIGRATION_STATUS_SETUP
:
1140 case MIGRATION_STATUS_CANCELLING
:
1141 case MIGRATION_STATUS_CANCELLED
:
1142 case MIGRATION_STATUS_ACTIVE
:
1143 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1144 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1145 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1146 case MIGRATION_STATUS_FAILED
:
1147 case MIGRATION_STATUS_COLO
:
1148 info
->has_status
= true;
1150 case MIGRATION_STATUS_COMPLETED
:
1151 info
->has_status
= true;
1152 fill_destination_postcopy_migration_info(info
);
1155 info
->status
= mis
->state
;
1158 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1160 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1162 fill_destination_migration_info(info
);
1163 fill_source_migration_info(info
);
1168 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
1171 MigrationState
*s
= migrate_get_current();
1172 MigrationCapabilityStatusList
*cap
;
1173 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
1175 if (migration_is_running(s
->state
)) {
1176 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1180 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
1181 if (!migrate_caps_check(cap_list
, params
, errp
)) {
1185 for (cap
= params
; cap
; cap
= cap
->next
) {
1186 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
1191 * Check whether the parameters are valid. Error will be put into errp
1192 * (if provided). Return true if valid, otherwise false.
1194 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
1196 if (params
->has_compress_level
&&
1197 (params
->compress_level
> 9)) {
1198 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
1199 "is invalid, it should be in the range of 0 to 9");
1203 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
1204 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1206 "is invalid, it should be in the range of 1 to 255");
1210 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
1211 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1212 "decompress_threads",
1213 "is invalid, it should be in the range of 1 to 255");
1217 if (params
->has_throttle_trigger_threshold
&&
1218 (params
->throttle_trigger_threshold
< 1 ||
1219 params
->throttle_trigger_threshold
> 100)) {
1220 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1221 "throttle_trigger_threshold",
1222 "an integer in the range of 1 to 100");
1226 if (params
->has_cpu_throttle_initial
&&
1227 (params
->cpu_throttle_initial
< 1 ||
1228 params
->cpu_throttle_initial
> 99)) {
1229 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1230 "cpu_throttle_initial",
1231 "an integer in the range of 1 to 99");
1235 if (params
->has_cpu_throttle_increment
&&
1236 (params
->cpu_throttle_increment
< 1 ||
1237 params
->cpu_throttle_increment
> 99)) {
1238 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1239 "cpu_throttle_increment",
1240 "an integer in the range of 1 to 99");
1244 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
1245 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1247 "an integer in the range of 0 to "stringify(SIZE_MAX
)
1252 if (params
->has_downtime_limit
&&
1253 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
1254 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1256 "an integer in the range of 0 to "
1257 stringify(MAX_MIGRATE_DOWNTIME
)" ms");
1261 /* x_checkpoint_delay is now always positive */
1263 if (params
->has_multifd_channels
&& (params
->multifd_channels
< 1)) {
1264 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1266 "is invalid, it should be in the range of 1 to 255");
1270 if (params
->has_multifd_zlib_level
&&
1271 (params
->multifd_zlib_level
> 9)) {
1272 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zlib_level",
1273 "is invalid, it should be in the range of 0 to 9");
1277 if (params
->has_multifd_zstd_level
&&
1278 (params
->multifd_zstd_level
> 20)) {
1279 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zstd_level",
1280 "is invalid, it should be in the range of 0 to 20");
1284 if (params
->has_xbzrle_cache_size
&&
1285 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
1286 !is_power_of_2(params
->xbzrle_cache_size
))) {
1287 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1288 "xbzrle_cache_size",
1289 "is invalid, it should be bigger than target page size"
1290 " and a power of 2");
1294 if (params
->has_max_cpu_throttle
&&
1295 (params
->max_cpu_throttle
< params
->cpu_throttle_initial
||
1296 params
->max_cpu_throttle
> 99)) {
1297 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1299 "an integer in the range of cpu_throttle_initial to 99");
1303 if (params
->has_announce_initial
&&
1304 params
->announce_initial
> 100000) {
1305 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1307 "is invalid, it must be less than 100000 ms");
1310 if (params
->has_announce_max
&&
1311 params
->announce_max
> 100000) {
1312 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1314 "is invalid, it must be less than 100000 ms");
1317 if (params
->has_announce_rounds
&&
1318 params
->announce_rounds
> 1000) {
1319 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1321 "is invalid, it must be in the range of 0 to 1000");
1324 if (params
->has_announce_step
&&
1325 (params
->announce_step
< 1 ||
1326 params
->announce_step
> 10000)) {
1327 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1329 "is invalid, it must be in the range of 1 to 10000 ms");
1333 if (params
->has_block_bitmap_mapping
&&
1334 !check_dirty_bitmap_mig_alias_map(params
->block_bitmap_mapping
, errp
)) {
1335 error_prepend(errp
, "Invalid mapping given for block-bitmap-mapping: ");
1342 static void migrate_params_test_apply(MigrateSetParameters
*params
,
1343 MigrationParameters
*dest
)
1345 *dest
= migrate_get_current()->parameters
;
1347 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1349 if (params
->has_compress_level
) {
1350 dest
->compress_level
= params
->compress_level
;
1353 if (params
->has_compress_threads
) {
1354 dest
->compress_threads
= params
->compress_threads
;
1357 if (params
->has_compress_wait_thread
) {
1358 dest
->compress_wait_thread
= params
->compress_wait_thread
;
1361 if (params
->has_decompress_threads
) {
1362 dest
->decompress_threads
= params
->decompress_threads
;
1365 if (params
->has_throttle_trigger_threshold
) {
1366 dest
->throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1369 if (params
->has_cpu_throttle_initial
) {
1370 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
1373 if (params
->has_cpu_throttle_increment
) {
1374 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
1377 if (params
->has_cpu_throttle_tailslow
) {
1378 dest
->cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1381 if (params
->has_tls_creds
) {
1382 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1383 dest
->tls_creds
= params
->tls_creds
->u
.s
;
1386 if (params
->has_tls_hostname
) {
1387 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1388 dest
->tls_hostname
= params
->tls_hostname
->u
.s
;
1391 if (params
->has_max_bandwidth
) {
1392 dest
->max_bandwidth
= params
->max_bandwidth
;
1395 if (params
->has_downtime_limit
) {
1396 dest
->downtime_limit
= params
->downtime_limit
;
1399 if (params
->has_x_checkpoint_delay
) {
1400 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
1403 if (params
->has_block_incremental
) {
1404 dest
->block_incremental
= params
->block_incremental
;
1406 if (params
->has_multifd_channels
) {
1407 dest
->multifd_channels
= params
->multifd_channels
;
1409 if (params
->has_multifd_compression
) {
1410 dest
->multifd_compression
= params
->multifd_compression
;
1412 if (params
->has_xbzrle_cache_size
) {
1413 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
1415 if (params
->has_max_postcopy_bandwidth
) {
1416 dest
->max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1418 if (params
->has_max_cpu_throttle
) {
1419 dest
->max_cpu_throttle
= params
->max_cpu_throttle
;
1421 if (params
->has_announce_initial
) {
1422 dest
->announce_initial
= params
->announce_initial
;
1424 if (params
->has_announce_max
) {
1425 dest
->announce_max
= params
->announce_max
;
1427 if (params
->has_announce_rounds
) {
1428 dest
->announce_rounds
= params
->announce_rounds
;
1430 if (params
->has_announce_step
) {
1431 dest
->announce_step
= params
->announce_step
;
1434 if (params
->has_block_bitmap_mapping
) {
1435 dest
->has_block_bitmap_mapping
= true;
1436 dest
->block_bitmap_mapping
= params
->block_bitmap_mapping
;
1440 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
1442 MigrationState
*s
= migrate_get_current();
1444 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1446 if (params
->has_compress_level
) {
1447 s
->parameters
.compress_level
= params
->compress_level
;
1450 if (params
->has_compress_threads
) {
1451 s
->parameters
.compress_threads
= params
->compress_threads
;
1454 if (params
->has_compress_wait_thread
) {
1455 s
->parameters
.compress_wait_thread
= params
->compress_wait_thread
;
1458 if (params
->has_decompress_threads
) {
1459 s
->parameters
.decompress_threads
= params
->decompress_threads
;
1462 if (params
->has_throttle_trigger_threshold
) {
1463 s
->parameters
.throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1466 if (params
->has_cpu_throttle_initial
) {
1467 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
1470 if (params
->has_cpu_throttle_increment
) {
1471 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
1474 if (params
->has_cpu_throttle_tailslow
) {
1475 s
->parameters
.cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1478 if (params
->has_tls_creds
) {
1479 g_free(s
->parameters
.tls_creds
);
1480 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1481 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1484 if (params
->has_tls_hostname
) {
1485 g_free(s
->parameters
.tls_hostname
);
1486 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1487 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1490 if (params
->has_tls_authz
) {
1491 g_free(s
->parameters
.tls_authz
);
1492 assert(params
->tls_authz
->type
== QTYPE_QSTRING
);
1493 s
->parameters
.tls_authz
= g_strdup(params
->tls_authz
->u
.s
);
1496 if (params
->has_max_bandwidth
) {
1497 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1498 if (s
->to_dst_file
&& !migration_in_postcopy()) {
1499 qemu_file_set_rate_limit(s
->to_dst_file
,
1500 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1504 if (params
->has_downtime_limit
) {
1505 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1508 if (params
->has_x_checkpoint_delay
) {
1509 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1510 if (migration_in_colo_state()) {
1511 colo_checkpoint_notify(s
);
1515 if (params
->has_block_incremental
) {
1516 s
->parameters
.block_incremental
= params
->block_incremental
;
1518 if (params
->has_multifd_channels
) {
1519 s
->parameters
.multifd_channels
= params
->multifd_channels
;
1521 if (params
->has_multifd_compression
) {
1522 s
->parameters
.multifd_compression
= params
->multifd_compression
;
1524 if (params
->has_xbzrle_cache_size
) {
1525 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1526 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1528 if (params
->has_max_postcopy_bandwidth
) {
1529 s
->parameters
.max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1530 if (s
->to_dst_file
&& migration_in_postcopy()) {
1531 qemu_file_set_rate_limit(s
->to_dst_file
,
1532 s
->parameters
.max_postcopy_bandwidth
/ XFER_LIMIT_RATIO
);
1535 if (params
->has_max_cpu_throttle
) {
1536 s
->parameters
.max_cpu_throttle
= params
->max_cpu_throttle
;
1538 if (params
->has_announce_initial
) {
1539 s
->parameters
.announce_initial
= params
->announce_initial
;
1541 if (params
->has_announce_max
) {
1542 s
->parameters
.announce_max
= params
->announce_max
;
1544 if (params
->has_announce_rounds
) {
1545 s
->parameters
.announce_rounds
= params
->announce_rounds
;
1547 if (params
->has_announce_step
) {
1548 s
->parameters
.announce_step
= params
->announce_step
;
1551 if (params
->has_block_bitmap_mapping
) {
1552 qapi_free_BitmapMigrationNodeAliasList(
1553 s
->parameters
.block_bitmap_mapping
);
1555 s
->parameters
.has_block_bitmap_mapping
= true;
1556 s
->parameters
.block_bitmap_mapping
=
1557 QAPI_CLONE(BitmapMigrationNodeAliasList
,
1558 params
->block_bitmap_mapping
);
1562 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1564 MigrationParameters tmp
;
1566 /* TODO Rewrite "" to null instead */
1567 if (params
->has_tls_creds
1568 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1569 qobject_unref(params
->tls_creds
->u
.n
);
1570 params
->tls_creds
->type
= QTYPE_QSTRING
;
1571 params
->tls_creds
->u
.s
= strdup("");
1573 /* TODO Rewrite "" to null instead */
1574 if (params
->has_tls_hostname
1575 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1576 qobject_unref(params
->tls_hostname
->u
.n
);
1577 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1578 params
->tls_hostname
->u
.s
= strdup("");
1581 migrate_params_test_apply(params
, &tmp
);
1583 if (!migrate_params_check(&tmp
, errp
)) {
1584 /* Invalid parameter */
1588 migrate_params_apply(params
, errp
);
1592 void qmp_migrate_start_postcopy(Error
**errp
)
1594 MigrationState
*s
= migrate_get_current();
1596 if (!migrate_postcopy()) {
1597 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1598 " the start of migration");
1602 if (s
->state
== MIGRATION_STATUS_NONE
) {
1603 error_setg(errp
, "Postcopy must be started after migration has been"
1608 * we don't error if migration has finished since that would be racy
1609 * with issuing this command.
1611 qatomic_set(&s
->start_postcopy
, true);
1614 /* shared migration helpers */
1616 void migrate_set_state(int *state
, int old_state
, int new_state
)
1618 assert(new_state
< MIGRATION_STATUS__MAX
);
1619 if (qatomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1620 trace_migrate_set_state(MigrationStatus_str(new_state
));
1621 migrate_generate_event(new_state
);
1625 static MigrationCapabilityStatusList
*migrate_cap_add(
1626 MigrationCapabilityStatusList
*list
,
1627 MigrationCapability index
,
1630 MigrationCapabilityStatusList
*cap
;
1632 cap
= g_new0(MigrationCapabilityStatusList
, 1);
1633 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
1634 cap
->value
->capability
= index
;
1635 cap
->value
->state
= state
;
1641 void migrate_set_block_enabled(bool value
, Error
**errp
)
1643 MigrationCapabilityStatusList
*cap
;
1645 cap
= migrate_cap_add(NULL
, MIGRATION_CAPABILITY_BLOCK
, value
);
1646 qmp_migrate_set_capabilities(cap
, errp
);
1647 qapi_free_MigrationCapabilityStatusList(cap
);
1650 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1652 s
->parameters
.block_incremental
= value
;
1655 static void block_cleanup_parameters(MigrationState
*s
)
1657 if (s
->must_remove_block_options
) {
1658 /* setting to false can never fail */
1659 migrate_set_block_enabled(false, &error_abort
);
1660 migrate_set_block_incremental(s
, false);
1661 s
->must_remove_block_options
= false;
1665 static void migrate_fd_cleanup(MigrationState
*s
)
1667 qemu_bh_delete(s
->cleanup_bh
);
1668 s
->cleanup_bh
= NULL
;
1670 qemu_savevm_state_cleanup();
1672 if (s
->to_dst_file
) {
1675 trace_migrate_fd_cleanup();
1676 qemu_mutex_unlock_iothread();
1677 if (s
->migration_thread_running
) {
1678 qemu_thread_join(&s
->thread
);
1679 s
->migration_thread_running
= false;
1681 qemu_mutex_lock_iothread();
1683 multifd_save_cleanup();
1684 qemu_mutex_lock(&s
->qemu_file_lock
);
1685 tmp
= s
->to_dst_file
;
1686 s
->to_dst_file
= NULL
;
1687 qemu_mutex_unlock(&s
->qemu_file_lock
);
1689 * Close the file handle without the lock to make sure the
1690 * critical section won't block for long.
1695 assert(!migration_is_active(s
));
1697 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1698 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1699 MIGRATION_STATUS_CANCELLED
);
1703 /* It is used on info migrate. We can't free it */
1704 error_report_err(error_copy(s
->error
));
1706 notifier_list_notify(&migration_state_notifiers
, s
);
1707 block_cleanup_parameters(s
);
1710 static void migrate_fd_cleanup_schedule(MigrationState
*s
)
1713 * Ref the state for bh, because it may be called when
1714 * there're already no other refs
1716 object_ref(OBJECT(s
));
1717 qemu_bh_schedule(s
->cleanup_bh
);
1720 static void migrate_fd_cleanup_bh(void *opaque
)
1722 MigrationState
*s
= opaque
;
1723 migrate_fd_cleanup(s
);
1724 object_unref(OBJECT(s
));
1727 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1729 QEMU_LOCK_GUARD(&s
->error_mutex
);
1731 s
->error
= error_copy(error
);
1735 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1737 trace_migrate_fd_error(error_get_pretty(error
));
1738 assert(s
->to_dst_file
== NULL
);
1739 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1740 MIGRATION_STATUS_FAILED
);
1741 migrate_set_error(s
, error
);
1744 static void migrate_fd_cancel(MigrationState
*s
)
1747 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1748 trace_migrate_fd_cancel();
1750 if (s
->rp_state
.from_dst_file
) {
1751 /* shutdown the rp socket, so causing the rp thread to shutdown */
1752 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1756 old_state
= s
->state
;
1757 if (!migration_is_running(old_state
)) {
1760 /* If the migration is paused, kick it out of the pause */
1761 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1762 qemu_sem_post(&s
->pause_sem
);
1764 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1765 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1768 * If we're unlucky the migration code might be stuck somewhere in a
1769 * send/write while the network has failed and is waiting to timeout;
1770 * if we've got shutdown(2) available then we can force it to quit.
1771 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1772 * called in a bh, so there is no race against this cancel.
1774 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1775 qemu_file_shutdown(f
);
1777 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1778 Error
*local_err
= NULL
;
1780 bdrv_invalidate_cache_all(&local_err
);
1782 error_report_err(local_err
);
1784 s
->block_inactive
= false;
1789 void add_migration_state_change_notifier(Notifier
*notify
)
1791 notifier_list_add(&migration_state_notifiers
, notify
);
1794 void remove_migration_state_change_notifier(Notifier
*notify
)
1796 notifier_remove(notify
);
1799 bool migration_in_setup(MigrationState
*s
)
1801 return s
->state
== MIGRATION_STATUS_SETUP
;
1804 bool migration_has_finished(MigrationState
*s
)
1806 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1809 bool migration_has_failed(MigrationState
*s
)
1811 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1812 s
->state
== MIGRATION_STATUS_FAILED
);
1815 bool migration_in_postcopy(void)
1817 MigrationState
*s
= migrate_get_current();
1820 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1821 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1822 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1829 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1831 return migration_in_postcopy() && s
->postcopy_after_devices
;
1834 bool migration_in_incoming_postcopy(void)
1836 PostcopyState ps
= postcopy_state_get();
1838 return ps
>= POSTCOPY_INCOMING_DISCARD
&& ps
< POSTCOPY_INCOMING_END
;
1841 bool migration_is_idle(void)
1843 MigrationState
*s
= current_migration
;
1850 case MIGRATION_STATUS_NONE
:
1851 case MIGRATION_STATUS_CANCELLED
:
1852 case MIGRATION_STATUS_COMPLETED
:
1853 case MIGRATION_STATUS_FAILED
:
1855 case MIGRATION_STATUS_SETUP
:
1856 case MIGRATION_STATUS_CANCELLING
:
1857 case MIGRATION_STATUS_ACTIVE
:
1858 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1859 case MIGRATION_STATUS_COLO
:
1860 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1861 case MIGRATION_STATUS_DEVICE
:
1862 case MIGRATION_STATUS_WAIT_UNPLUG
:
1864 case MIGRATION_STATUS__MAX
:
1865 g_assert_not_reached();
1871 bool migration_is_active(MigrationState
*s
)
1873 return (s
->state
== MIGRATION_STATUS_ACTIVE
||
1874 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1877 void migrate_init(MigrationState
*s
)
1880 * Reinitialise all migration state, except
1881 * parameters/capabilities that the user set, and
1885 s
->to_dst_file
= NULL
;
1886 s
->state
= MIGRATION_STATUS_NONE
;
1887 s
->rp_state
.from_dst_file
= NULL
;
1888 s
->rp_state
.error
= false;
1890 s
->pages_per_second
= 0.0;
1892 s
->expected_downtime
= 0;
1894 s
->start_postcopy
= false;
1895 s
->postcopy_after_devices
= false;
1896 s
->migration_thread_running
= false;
1897 error_free(s
->error
);
1900 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1902 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1904 s
->vm_was_running
= false;
1905 s
->iteration_initial_bytes
= 0;
1906 s
->threshold_size
= 0;
1909 static GSList
*migration_blockers
;
1911 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1913 if (only_migratable
) {
1914 error_propagate_prepend(errp
, error_copy(reason
),
1915 "disallowing migration blocker "
1916 "(--only-migratable) for: ");
1920 if (migration_is_idle()) {
1921 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1925 error_propagate_prepend(errp
, error_copy(reason
),
1926 "disallowing migration blocker "
1927 "(migration in progress) for: ");
1931 void migrate_del_blocker(Error
*reason
)
1933 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1936 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1938 Error
*local_err
= NULL
;
1939 static bool once
= true;
1941 if (!deferred_incoming
) {
1942 error_setg(errp
, "For use with '-incoming defer'");
1946 error_setg(errp
, "The incoming migration has already been started");
1950 qemu_start_incoming_migration(uri
, &local_err
);
1953 error_propagate(errp
, local_err
);
1960 void qmp_migrate_recover(const char *uri
, Error
**errp
)
1962 MigrationIncomingState
*mis
= migration_incoming_get_current();
1964 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1965 error_setg(errp
, "Migrate recover can only be run "
1966 "when postcopy is paused.");
1970 if (qatomic_cmpxchg(&mis
->postcopy_recover_triggered
,
1971 false, true) == true) {
1972 error_setg(errp
, "Migrate recovery is triggered already");
1977 * Note that this call will never start a real migration; it will
1978 * only re-setup the migration stream and poke existing migration
1979 * to continue using that newly established channel.
1981 qemu_start_incoming_migration(uri
, errp
);
1984 void qmp_migrate_pause(Error
**errp
)
1986 MigrationState
*ms
= migrate_get_current();
1987 MigrationIncomingState
*mis
= migration_incoming_get_current();
1990 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1991 /* Source side, during postcopy */
1992 qemu_mutex_lock(&ms
->qemu_file_lock
);
1993 ret
= qemu_file_shutdown(ms
->to_dst_file
);
1994 qemu_mutex_unlock(&ms
->qemu_file_lock
);
1996 error_setg(errp
, "Failed to pause source migration");
2001 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2002 ret
= qemu_file_shutdown(mis
->from_src_file
);
2004 error_setg(errp
, "Failed to pause destination migration");
2009 error_setg(errp
, "migrate-pause is currently only supported "
2010 "during postcopy-active state");
2013 bool migration_is_blocked(Error
**errp
)
2015 if (qemu_savevm_state_blocked(errp
)) {
2019 if (migration_blockers
) {
2020 error_propagate(errp
, error_copy(migration_blockers
->data
));
2027 /* Returns true if continue to migrate, or false if error detected */
2028 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
2029 bool resume
, Error
**errp
)
2031 Error
*local_err
= NULL
;
2034 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2035 error_setg(errp
, "Cannot resume if there is no "
2036 "paused migration");
2041 * Postcopy recovery won't work well with release-ram
2042 * capability since release-ram will drop the page buffer as
2043 * long as the page is put into the send buffer. So if there
2044 * is a network failure happened, any page buffers that have
2045 * not yet reached the destination VM but have already been
2046 * sent from the source VM will be lost forever. Let's refuse
2047 * the client from resuming such a postcopy migration.
2048 * Luckily release-ram was designed to only be used when src
2049 * and destination VMs are on the same host, so it should be
2052 if (migrate_release_ram()) {
2053 error_setg(errp
, "Postcopy recovery cannot work "
2054 "when release-ram capability is set");
2058 /* This is a resume, skip init status */
2062 if (migration_is_running(s
->state
)) {
2063 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
2067 if (runstate_check(RUN_STATE_INMIGRATE
)) {
2068 error_setg(errp
, "Guest is waiting for an incoming migration");
2072 if (migration_is_blocked(errp
)) {
2076 if (blk
|| blk_inc
) {
2077 if (migrate_use_block() || migrate_use_block_incremental()) {
2078 error_setg(errp
, "Command options are incompatible with "
2079 "current migration capabilities");
2082 migrate_set_block_enabled(true, &local_err
);
2084 error_propagate(errp
, local_err
);
2087 s
->must_remove_block_options
= true;
2091 migrate_set_block_incremental(s
, true);
2096 * set ram_counters memory to zero for a
2099 memset(&ram_counters
, 0, sizeof(ram_counters
));
2104 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
2105 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
2106 bool has_resume
, bool resume
, Error
**errp
)
2108 Error
*local_err
= NULL
;
2109 MigrationState
*s
= migrate_get_current();
2110 const char *p
= NULL
;
2112 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
2113 has_resume
&& resume
, errp
)) {
2114 /* Error detected, put into errp */
2118 if (strstart(uri
, "tcp:", &p
) ||
2119 strstart(uri
, "unix:", NULL
) ||
2120 strstart(uri
, "vsock:", NULL
)) {
2121 socket_start_outgoing_migration(s
, p
? p
: uri
, &local_err
);
2123 } else if (strstart(uri
, "rdma:", &p
)) {
2124 rdma_start_outgoing_migration(s
, p
, &local_err
);
2126 } else if (strstart(uri
, "exec:", &p
)) {
2127 exec_start_outgoing_migration(s
, p
, &local_err
);
2128 } else if (strstart(uri
, "fd:", &p
)) {
2129 fd_start_outgoing_migration(s
, p
, &local_err
);
2131 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
2132 "a valid migration protocol");
2133 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2134 MIGRATION_STATUS_FAILED
);
2135 block_cleanup_parameters(s
);
2140 migrate_fd_error(s
, local_err
);
2141 error_propagate(errp
, local_err
);
2146 void qmp_migrate_cancel(Error
**errp
)
2148 migrate_fd_cancel(migrate_get_current());
2151 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
2153 MigrationState
*s
= migrate_get_current();
2154 if (s
->state
!= state
) {
2155 error_setg(errp
, "Migration not in expected state: %s",
2156 MigrationStatus_str(s
->state
));
2159 qemu_sem_post(&s
->pause_sem
);
2162 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
2164 MigrateSetParameters p
= {
2165 .has_xbzrle_cache_size
= true,
2166 .xbzrle_cache_size
= value
,
2169 qmp_migrate_set_parameters(&p
, errp
);
2172 int64_t qmp_query_migrate_cache_size(Error
**errp
)
2174 return migrate_xbzrle_cache_size();
2177 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
2179 MigrateSetParameters p
= {
2180 .has_max_bandwidth
= true,
2181 .max_bandwidth
= value
,
2184 qmp_migrate_set_parameters(&p
, errp
);
2187 void qmp_migrate_set_downtime(double value
, Error
**errp
)
2189 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
2190 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
2192 "an integer in the range of 0 to "
2193 stringify(MAX_MIGRATE_DOWNTIME_SECONDS
)" seconds");
2197 value
*= 1000; /* Convert to milliseconds */
2199 MigrateSetParameters p
= {
2200 .has_downtime_limit
= true,
2201 .downtime_limit
= (int64_t)value
,
2204 qmp_migrate_set_parameters(&p
, errp
);
2207 bool migrate_release_ram(void)
2211 s
= migrate_get_current();
2213 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
2216 bool migrate_postcopy_ram(void)
2220 s
= migrate_get_current();
2222 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
2225 bool migrate_postcopy(void)
2227 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2230 bool migrate_auto_converge(void)
2234 s
= migrate_get_current();
2236 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
2239 bool migrate_zero_blocks(void)
2243 s
= migrate_get_current();
2245 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
2248 bool migrate_postcopy_blocktime(void)
2252 s
= migrate_get_current();
2254 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
2257 bool migrate_use_compression(void)
2261 s
= migrate_get_current();
2263 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
2266 int migrate_compress_level(void)
2270 s
= migrate_get_current();
2272 return s
->parameters
.compress_level
;
2275 int migrate_compress_threads(void)
2279 s
= migrate_get_current();
2281 return s
->parameters
.compress_threads
;
2284 int migrate_compress_wait_thread(void)
2288 s
= migrate_get_current();
2290 return s
->parameters
.compress_wait_thread
;
2293 int migrate_decompress_threads(void)
2297 s
= migrate_get_current();
2299 return s
->parameters
.decompress_threads
;
2302 bool migrate_dirty_bitmaps(void)
2306 s
= migrate_get_current();
2308 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
2311 bool migrate_ignore_shared(void)
2315 s
= migrate_get_current();
2317 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
];
2320 bool migrate_validate_uuid(void)
2324 s
= migrate_get_current();
2326 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_VALIDATE_UUID
];
2329 bool migrate_use_events(void)
2333 s
= migrate_get_current();
2335 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
2338 bool migrate_use_multifd(void)
2342 s
= migrate_get_current();
2344 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_MULTIFD
];
2347 bool migrate_pause_before_switchover(void)
2351 s
= migrate_get_current();
2353 return s
->enabled_capabilities
[
2354 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
2357 int migrate_multifd_channels(void)
2361 s
= migrate_get_current();
2363 return s
->parameters
.multifd_channels
;
2366 MultiFDCompression
migrate_multifd_compression(void)
2370 s
= migrate_get_current();
2372 return s
->parameters
.multifd_compression
;
2375 int migrate_multifd_zlib_level(void)
2379 s
= migrate_get_current();
2381 return s
->parameters
.multifd_zlib_level
;
2384 int migrate_multifd_zstd_level(void)
2388 s
= migrate_get_current();
2390 return s
->parameters
.multifd_zstd_level
;
2393 int migrate_use_xbzrle(void)
2397 s
= migrate_get_current();
2399 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
2402 int64_t migrate_xbzrle_cache_size(void)
2406 s
= migrate_get_current();
2408 return s
->parameters
.xbzrle_cache_size
;
2411 static int64_t migrate_max_postcopy_bandwidth(void)
2415 s
= migrate_get_current();
2417 return s
->parameters
.max_postcopy_bandwidth
;
2420 bool migrate_use_block(void)
2424 s
= migrate_get_current();
2426 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
2429 bool migrate_use_return_path(void)
2433 s
= migrate_get_current();
2435 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
2438 bool migrate_use_block_incremental(void)
2442 s
= migrate_get_current();
2444 return s
->parameters
.block_incremental
;
2447 /* migration thread support */
2449 * Something bad happened to the RP stream, mark an error
2450 * The caller shall print or trace something to indicate why
2452 static void mark_source_rp_bad(MigrationState
*s
)
2454 s
->rp_state
.error
= true;
2457 static struct rp_cmd_args
{
2458 ssize_t len
; /* -1 = variable */
2461 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
2462 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
2463 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
2464 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
2465 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
2466 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
2467 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
2468 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
2472 * Process a request for pages received on the return path,
2473 * We're allowed to send more than requested (e.g. to round to our page size)
2474 * and we don't need to send pages that have already been sent.
2476 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
2477 ram_addr_t start
, size_t len
)
2479 long our_host_ps
= qemu_real_host_page_size
;
2481 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
2484 * Since we currently insist on matching page sizes, just sanity check
2485 * we're being asked for whole host pages.
2487 if (start
& (our_host_ps
-1) ||
2488 (len
& (our_host_ps
-1))) {
2489 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2490 " len: %zd", __func__
, start
, len
);
2491 mark_source_rp_bad(ms
);
2495 if (ram_save_queue_pages(rbname
, start
, len
)) {
2496 mark_source_rp_bad(ms
);
2500 /* Return true to retry, false to quit */
2501 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
2503 trace_postcopy_pause_return_path();
2505 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
2507 trace_postcopy_pause_return_path_continued();
2512 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
2514 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
2517 error_report("%s: invalid block name '%s'", __func__
, block_name
);
2521 /* Fetch the received bitmap and refresh the dirty bitmap */
2522 return ram_dirty_bitmap_reload(s
, block
);
2525 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
2527 trace_source_return_path_thread_resume_ack(value
);
2529 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
2530 error_report("%s: illegal resume_ack value %"PRIu32
,
2535 /* Now both sides are active. */
2536 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2537 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2539 /* Notify send thread that time to continue send pages */
2540 qemu_sem_post(&s
->rp_state
.rp_sem
);
2546 * Handles messages sent on the return path towards the source VM
2549 static void *source_return_path_thread(void *opaque
)
2551 MigrationState
*ms
= opaque
;
2552 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
2553 uint16_t header_len
, header_type
;
2555 uint32_t tmp32
, sibling_error
;
2556 ram_addr_t start
= 0; /* =0 to silence warning */
2557 size_t len
= 0, expected_len
;
2560 trace_source_return_path_thread_entry();
2561 rcu_register_thread();
2564 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
2565 migration_is_setup_or_active(ms
->state
)) {
2566 trace_source_return_path_thread_loop_top();
2567 header_type
= qemu_get_be16(rp
);
2568 header_len
= qemu_get_be16(rp
);
2570 if (qemu_file_get_error(rp
)) {
2571 mark_source_rp_bad(ms
);
2575 if (header_type
>= MIG_RP_MSG_MAX
||
2576 header_type
== MIG_RP_MSG_INVALID
) {
2577 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2578 header_type
, header_len
);
2579 mark_source_rp_bad(ms
);
2583 if ((rp_cmd_args
[header_type
].len
!= -1 &&
2584 header_len
!= rp_cmd_args
[header_type
].len
) ||
2585 header_len
> sizeof(buf
)) {
2586 error_report("RP: Received '%s' message (0x%04x) with"
2587 "incorrect length %d expecting %zu",
2588 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
2589 (size_t)rp_cmd_args
[header_type
].len
);
2590 mark_source_rp_bad(ms
);
2594 /* We know we've got a valid header by this point */
2595 res
= qemu_get_buffer(rp
, buf
, header_len
);
2596 if (res
!= header_len
) {
2597 error_report("RP: Failed reading data for message 0x%04x"
2598 " read %d expected %d",
2599 header_type
, res
, header_len
);
2600 mark_source_rp_bad(ms
);
2604 /* OK, we have the message and the data */
2605 switch (header_type
) {
2606 case MIG_RP_MSG_SHUT
:
2607 sibling_error
= ldl_be_p(buf
);
2608 trace_source_return_path_thread_shut(sibling_error
);
2609 if (sibling_error
) {
2610 error_report("RP: Sibling indicated error %d", sibling_error
);
2611 mark_source_rp_bad(ms
);
2614 * We'll let the main thread deal with closing the RP
2615 * we could do a shutdown(2) on it, but we're the only user
2616 * anyway, so there's nothing gained.
2620 case MIG_RP_MSG_PONG
:
2621 tmp32
= ldl_be_p(buf
);
2622 trace_source_return_path_thread_pong(tmp32
);
2625 case MIG_RP_MSG_REQ_PAGES
:
2626 start
= ldq_be_p(buf
);
2627 len
= ldl_be_p(buf
+ 8);
2628 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
2631 case MIG_RP_MSG_REQ_PAGES_ID
:
2632 expected_len
= 12 + 1; /* header + termination */
2634 if (header_len
>= expected_len
) {
2635 start
= ldq_be_p(buf
);
2636 len
= ldl_be_p(buf
+ 8);
2637 /* Now we expect an idstr */
2638 tmp32
= buf
[12]; /* Length of the following idstr */
2639 buf
[13 + tmp32
] = '\0';
2640 expected_len
+= tmp32
;
2642 if (header_len
!= expected_len
) {
2643 error_report("RP: Req_Page_id with length %d expecting %zd",
2644 header_len
, expected_len
);
2645 mark_source_rp_bad(ms
);
2648 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
2651 case MIG_RP_MSG_RECV_BITMAP
:
2652 if (header_len
< 1) {
2653 error_report("%s: missing block name", __func__
);
2654 mark_source_rp_bad(ms
);
2657 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2658 buf
[buf
[0] + 1] = '\0';
2659 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
2660 mark_source_rp_bad(ms
);
2665 case MIG_RP_MSG_RESUME_ACK
:
2666 tmp32
= ldl_be_p(buf
);
2667 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
2668 mark_source_rp_bad(ms
);
2679 res
= qemu_file_get_error(rp
);
2681 if (res
== -EIO
&& migration_in_postcopy()) {
2683 * Maybe there is something we can do: it looks like a
2684 * network down issue, and we pause for a recovery.
2686 if (postcopy_pause_return_path_thread(ms
)) {
2687 /* Reload rp, reset the rest */
2688 if (rp
!= ms
->rp_state
.from_dst_file
) {
2690 rp
= ms
->rp_state
.from_dst_file
;
2692 ms
->rp_state
.error
= false;
2697 trace_source_return_path_thread_bad_end();
2698 mark_source_rp_bad(ms
);
2701 trace_source_return_path_thread_end();
2702 ms
->rp_state
.from_dst_file
= NULL
;
2704 rcu_unregister_thread();
2708 static int open_return_path_on_source(MigrationState
*ms
,
2712 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
2713 if (!ms
->rp_state
.from_dst_file
) {
2717 trace_open_return_path_on_source();
2719 if (!create_thread
) {
2724 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
2725 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
2727 trace_open_return_path_on_source_continue();
2732 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2733 static int await_return_path_close_on_source(MigrationState
*ms
)
2736 * If this is a normal exit then the destination will send a SHUT and the
2737 * rp_thread will exit, however if there's an error we need to cause
2740 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
2742 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2743 * waiting for the destination.
2745 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
2746 mark_source_rp_bad(ms
);
2748 trace_await_return_path_close_on_source_joining();
2749 qemu_thread_join(&ms
->rp_state
.rp_thread
);
2750 trace_await_return_path_close_on_source_close();
2751 return ms
->rp_state
.error
;
2755 * Switch from normal iteration to postcopy
2756 * Returns non-0 on error
2758 static int postcopy_start(MigrationState
*ms
)
2761 QIOChannelBuffer
*bioc
;
2763 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2764 int64_t bandwidth
= migrate_max_postcopy_bandwidth();
2765 bool restart_block
= false;
2766 int cur_state
= MIGRATION_STATUS_ACTIVE
;
2767 if (!migrate_pause_before_switchover()) {
2768 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
2769 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2772 trace_postcopy_start();
2773 qemu_mutex_lock_iothread();
2774 trace_postcopy_start_set_run();
2776 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2777 global_state_store();
2778 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2783 ret
= migration_maybe_pause(ms
, &cur_state
,
2784 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2789 ret
= bdrv_inactivate_all();
2793 restart_block
= true;
2796 * Cause any non-postcopiable, but iterative devices to
2797 * send out their final data.
2799 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2802 * in Finish migrate and with the io-lock held everything should
2803 * be quiet, but we've potentially still got dirty pages and we
2804 * need to tell the destination to throw any pages it's already received
2807 if (migrate_postcopy_ram()) {
2808 if (ram_postcopy_send_discard_bitmap(ms
)) {
2809 error_report("postcopy send discard bitmap failed");
2815 * send rest of state - note things that are doing postcopy
2816 * will notice we're in POSTCOPY_ACTIVE and not actually
2817 * wrap their state up here
2819 /* 0 max-postcopy-bandwidth means unlimited */
2821 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
2823 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
2825 if (migrate_postcopy_ram()) {
2826 /* Ping just for debugging, helps line traces up */
2827 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
2831 * While loading the device state we may trigger page transfer
2832 * requests and the fd must be free to process those, and thus
2833 * the destination must read the whole device state off the fd before
2834 * it starts processing it. Unfortunately the ad-hoc migration format
2835 * doesn't allow the destination to know the size to read without fully
2836 * parsing it through each devices load-state code (especially the open
2837 * coded devices that use get/put).
2838 * So we wrap the device state up in a package with a length at the start;
2839 * to do this we use a qemu_buf to hold the whole of the device state.
2841 bioc
= qio_channel_buffer_new(4096);
2842 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
2843 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
2844 object_unref(OBJECT(bioc
));
2847 * Make sure the receiver can get incoming pages before we send the rest
2850 qemu_savevm_send_postcopy_listen(fb
);
2852 qemu_savevm_state_complete_precopy(fb
, false, false);
2853 if (migrate_postcopy_ram()) {
2854 qemu_savevm_send_ping(fb
, 3);
2857 qemu_savevm_send_postcopy_run(fb
);
2859 /* <><> end of stuff going into the package */
2861 /* Last point of recovery; as soon as we send the package the destination
2862 * can open devices and potentially start running.
2863 * Lets just check again we've not got any errors.
2865 ret
= qemu_file_get_error(ms
->to_dst_file
);
2867 error_report("postcopy_start: Migration stream errored (pre package)");
2871 restart_block
= false;
2873 /* Now send that blob */
2874 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2879 /* Send a notify to give a chance for anything that needs to happen
2880 * at the transition to postcopy and after the device state; in particular
2881 * spice needs to trigger a transition now
2883 ms
->postcopy_after_devices
= true;
2884 notifier_list_notify(&migration_state_notifiers
, ms
);
2886 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2888 qemu_mutex_unlock_iothread();
2890 if (migrate_postcopy_ram()) {
2892 * Although this ping is just for debug, it could potentially be
2893 * used for getting a better measurement of downtime at the source.
2895 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2898 if (migrate_release_ram()) {
2899 ram_postcopy_migrated_memory_release(ms
);
2902 ret
= qemu_file_get_error(ms
->to_dst_file
);
2904 error_report("postcopy_start: Migration stream errored");
2905 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2906 MIGRATION_STATUS_FAILED
);
2914 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2915 MIGRATION_STATUS_FAILED
);
2916 if (restart_block
) {
2917 /* A failure happened early enough that we know the destination hasn't
2918 * accessed block devices, so we're safe to recover.
2920 Error
*local_err
= NULL
;
2922 bdrv_invalidate_cache_all(&local_err
);
2924 error_report_err(local_err
);
2927 qemu_mutex_unlock_iothread();
2932 * migration_maybe_pause: Pause if required to by
2933 * migrate_pause_before_switchover called with the iothread locked
2934 * Returns: 0 on success
2936 static int migration_maybe_pause(MigrationState
*s
,
2937 int *current_active_state
,
2940 if (!migrate_pause_before_switchover()) {
2944 /* Since leaving this state is not atomic with posting the semaphore
2945 * it's possible that someone could have issued multiple migrate_continue
2946 * and the semaphore is incorrectly positive at this point;
2947 * the docs say it's undefined to reinit a semaphore that's already
2948 * init'd, so use timedwait to eat up any existing posts.
2950 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2951 /* This block intentionally left blank */
2955 * If the migration is cancelled when it is in the completion phase,
2956 * the migration state is set to MIGRATION_STATUS_CANCELLING.
2957 * So we don't need to wait a semaphore, otherwise we would always
2958 * wait for the 'pause_sem' semaphore.
2960 if (s
->state
!= MIGRATION_STATUS_CANCELLING
) {
2961 qemu_mutex_unlock_iothread();
2962 migrate_set_state(&s
->state
, *current_active_state
,
2963 MIGRATION_STATUS_PRE_SWITCHOVER
);
2964 qemu_sem_wait(&s
->pause_sem
);
2965 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2967 *current_active_state
= new_state
;
2968 qemu_mutex_lock_iothread();
2971 return s
->state
== new_state
? 0 : -EINVAL
;
2975 * migration_completion: Used by migration_thread when there's not much left.
2976 * The caller 'breaks' the loop when this returns.
2978 * @s: Current migration state
2980 static void migration_completion(MigrationState
*s
)
2983 int current_active_state
= s
->state
;
2985 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2986 qemu_mutex_lock_iothread();
2987 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2988 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2989 s
->vm_was_running
= runstate_is_running();
2990 ret
= global_state_store();
2993 bool inactivate
= !migrate_colo_enabled();
2994 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2996 ret
= migration_maybe_pause(s
, ¤t_active_state
,
2997 MIGRATION_STATUS_DEVICE
);
3000 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
3001 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
3004 if (inactivate
&& ret
>= 0) {
3005 s
->block_inactive
= true;
3008 qemu_mutex_unlock_iothread();
3013 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3014 trace_migration_completion_postcopy_end();
3016 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
3017 trace_migration_completion_postcopy_end_after_complete();
3021 * If rp was opened we must clean up the thread before
3022 * cleaning everything else up (since if there are no failures
3023 * it will wait for the destination to send it's status in
3026 if (s
->rp_state
.from_dst_file
) {
3028 trace_migration_return_path_end_before();
3029 rp_error
= await_return_path_close_on_source(s
);
3030 trace_migration_return_path_end_after(rp_error
);
3032 goto fail_invalidate
;
3036 if (qemu_file_get_error(s
->to_dst_file
)) {
3037 trace_migration_completion_file_err();
3038 goto fail_invalidate
;
3041 if (!migrate_colo_enabled()) {
3042 migrate_set_state(&s
->state
, current_active_state
,
3043 MIGRATION_STATUS_COMPLETED
);
3049 /* If not doing postcopy, vm_start() will be called: let's regain
3050 * control on images.
3052 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
3053 s
->state
== MIGRATION_STATUS_DEVICE
) {
3054 Error
*local_err
= NULL
;
3056 qemu_mutex_lock_iothread();
3057 bdrv_invalidate_cache_all(&local_err
);
3059 error_report_err(local_err
);
3061 s
->block_inactive
= false;
3063 qemu_mutex_unlock_iothread();
3067 migrate_set_state(&s
->state
, current_active_state
,
3068 MIGRATION_STATUS_FAILED
);
3071 bool migrate_colo_enabled(void)
3073 MigrationState
*s
= migrate_get_current();
3074 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
3077 typedef enum MigThrError
{
3078 /* No error detected */
3079 MIG_THR_ERR_NONE
= 0,
3080 /* Detected error, but resumed successfully */
3081 MIG_THR_ERR_RECOVERED
= 1,
3082 /* Detected fatal error, need to exit */
3083 MIG_THR_ERR_FATAL
= 2,
3086 static int postcopy_resume_handshake(MigrationState
*s
)
3088 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
3090 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3091 qemu_sem_wait(&s
->rp_state
.rp_sem
);
3094 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3101 /* Return zero if success, or <0 for error */
3102 static int postcopy_do_resume(MigrationState
*s
)
3107 * Call all the resume_prepare() hooks, so that modules can be
3108 * ready for the migration resume.
3110 ret
= qemu_savevm_state_resume_prepare(s
);
3112 error_report("%s: resume_prepare() failure detected: %d",
3118 * Last handshake with destination on the resume (destination will
3119 * switch to postcopy-active afterwards)
3121 ret
= postcopy_resume_handshake(s
);
3123 error_report("%s: handshake failed: %d", __func__
, ret
);
3131 * We don't return until we are in a safe state to continue current
3132 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
3133 * MIG_THR_ERR_FATAL if unrecovery failure happened.
3135 static MigThrError
postcopy_pause(MigrationState
*s
)
3137 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3142 migrate_set_state(&s
->state
, s
->state
,
3143 MIGRATION_STATUS_POSTCOPY_PAUSED
);
3145 /* Current channel is possibly broken. Release it. */
3146 assert(s
->to_dst_file
);
3147 qemu_mutex_lock(&s
->qemu_file_lock
);
3148 file
= s
->to_dst_file
;
3149 s
->to_dst_file
= NULL
;
3150 qemu_mutex_unlock(&s
->qemu_file_lock
);
3152 qemu_file_shutdown(file
);
3155 error_report("Detected IO failure for postcopy. "
3156 "Migration paused.");
3159 * We wait until things fixed up. Then someone will setup the
3160 * status back for us.
3162 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
3163 qemu_sem_wait(&s
->postcopy_pause_sem
);
3166 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3167 /* Woken up by a recover procedure. Give it a shot */
3170 * Firstly, let's wake up the return path now, with a new
3171 * return path channel.
3173 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
3175 /* Do the resume logic */
3176 if (postcopy_do_resume(s
) == 0) {
3177 /* Let's continue! */
3178 trace_postcopy_pause_continued();
3179 return MIG_THR_ERR_RECOVERED
;
3182 * Something wrong happened during the recovery, let's
3183 * pause again. Pause is always better than throwing
3189 /* This is not right... Time to quit. */
3190 return MIG_THR_ERR_FATAL
;
3195 static MigThrError
migration_detect_error(MigrationState
*s
)
3198 int state
= s
->state
;
3199 Error
*local_error
= NULL
;
3201 if (state
== MIGRATION_STATUS_CANCELLING
||
3202 state
== MIGRATION_STATUS_CANCELLED
) {
3203 /* End the migration, but don't set the state to failed */
3204 return MIG_THR_ERR_FATAL
;
3207 /* Try to detect any file errors */
3208 ret
= qemu_file_get_error_obj(s
->to_dst_file
, &local_error
);
3210 /* Everything is fine */
3211 assert(!local_error
);
3212 return MIG_THR_ERR_NONE
;
3216 migrate_set_error(s
, local_error
);
3217 error_free(local_error
);
3220 if (state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
== -EIO
) {
3222 * For postcopy, we allow the network to be down for a
3223 * while. After that, it can be continued by a
3226 return postcopy_pause(s
);
3229 * For precopy (or postcopy with error outside IO), we fail
3232 migrate_set_state(&s
->state
, state
, MIGRATION_STATUS_FAILED
);
3233 trace_migration_thread_file_err();
3235 /* Time to stop the migration, now. */
3236 return MIG_THR_ERR_FATAL
;
3240 /* How many bytes have we transferred since the beginning of the migration */
3241 static uint64_t migration_total_bytes(MigrationState
*s
)
3243 return qemu_ftell(s
->to_dst_file
) + ram_counters
.multifd_bytes
;
3246 static void migration_calculate_complete(MigrationState
*s
)
3248 uint64_t bytes
= migration_total_bytes(s
);
3249 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3250 int64_t transfer_time
;
3252 s
->total_time
= end_time
- s
->start_time
;
3255 * It's still not set, so we are precopy migration. For
3256 * postcopy, downtime is calculated during postcopy_start().
3258 s
->downtime
= end_time
- s
->downtime_start
;
3261 transfer_time
= s
->total_time
- s
->setup_time
;
3262 if (transfer_time
) {
3263 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
3267 static void update_iteration_initial_status(MigrationState
*s
)
3270 * Update these three fields at the same time to avoid mismatch info lead
3271 * wrong speed calculation.
3273 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3274 s
->iteration_initial_bytes
= migration_total_bytes(s
);
3275 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
3278 static void migration_update_counters(MigrationState
*s
,
3279 int64_t current_time
)
3281 uint64_t transferred
, transferred_pages
, time_spent
;
3282 uint64_t current_bytes
; /* bytes transferred since the beginning */
3285 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
3289 current_bytes
= migration_total_bytes(s
);
3290 transferred
= current_bytes
- s
->iteration_initial_bytes
;
3291 time_spent
= current_time
- s
->iteration_start_time
;
3292 bandwidth
= (double)transferred
/ time_spent
;
3293 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
3295 s
->mbps
= (((double) transferred
* 8.0) /
3296 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
3298 transferred_pages
= ram_get_total_transferred_pages() -
3299 s
->iteration_initial_pages
;
3300 s
->pages_per_second
= (double) transferred_pages
/
3301 (((double) time_spent
/ 1000.0));
3304 * if we haven't sent anything, we don't want to
3305 * recalculate. 10000 is a small enough number for our purposes
3307 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
3308 s
->expected_downtime
= ram_counters
.remaining
/ bandwidth
;
3311 qemu_file_reset_rate_limit(s
->to_dst_file
);
3313 update_iteration_initial_status(s
);
3315 trace_migrate_transferred(transferred
, time_spent
,
3316 bandwidth
, s
->threshold_size
);
3319 /* Migration thread iteration status */
3321 MIG_ITERATE_RESUME
, /* Resume current iteration */
3322 MIG_ITERATE_SKIP
, /* Skip current iteration */
3323 MIG_ITERATE_BREAK
, /* Break the loop */
3327 * Return true if continue to the next iteration directly, false
3330 static MigIterateState
migration_iteration_run(MigrationState
*s
)
3332 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
3333 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
3335 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
3336 &pend_compat
, &pend_post
);
3337 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
3339 trace_migrate_pending(pending_size
, s
->threshold_size
,
3340 pend_pre
, pend_compat
, pend_post
);
3342 if (pending_size
&& pending_size
>= s
->threshold_size
) {
3343 /* Still a significant amount to transfer */
3344 if (!in_postcopy
&& pend_pre
<= s
->threshold_size
&&
3345 qatomic_read(&s
->start_postcopy
)) {
3346 if (postcopy_start(s
)) {
3347 error_report("%s: postcopy failed to start", __func__
);
3349 return MIG_ITERATE_SKIP
;
3351 /* Just another iteration step */
3352 qemu_savevm_state_iterate(s
->to_dst_file
, in_postcopy
);
3354 trace_migration_thread_low_pending(pending_size
);
3355 migration_completion(s
);
3356 return MIG_ITERATE_BREAK
;
3359 return MIG_ITERATE_RESUME
;
3362 static void migration_iteration_finish(MigrationState
*s
)
3364 /* If we enabled cpu throttling for auto-converge, turn it off. */
3365 cpu_throttle_stop();
3367 qemu_mutex_lock_iothread();
3369 case MIGRATION_STATUS_COMPLETED
:
3370 migration_calculate_complete(s
);
3371 runstate_set(RUN_STATE_POSTMIGRATE
);
3374 case MIGRATION_STATUS_ACTIVE
:
3376 * We should really assert here, but since it's during
3377 * migration, let's try to reduce the usage of assertions.
3379 if (!migrate_colo_enabled()) {
3380 error_report("%s: critical error: calling COLO code without "
3381 "COLO enabled", __func__
);
3383 migrate_start_colo_process(s
);
3385 * Fixme: we will run VM in COLO no matter its old running state.
3386 * After exited COLO, we will keep running.
3388 s
->vm_was_running
= true;
3390 case MIGRATION_STATUS_FAILED
:
3391 case MIGRATION_STATUS_CANCELLED
:
3392 case MIGRATION_STATUS_CANCELLING
:
3393 if (s
->vm_was_running
) {
3396 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
3397 runstate_set(RUN_STATE_POSTMIGRATE
);
3403 /* Should not reach here, but if so, forgive the VM. */
3404 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3407 migrate_fd_cleanup_schedule(s
);
3408 qemu_mutex_unlock_iothread();
3411 void migration_make_urgent_request(void)
3413 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
3416 void migration_consume_urgent_request(void)
3418 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
3421 /* Returns true if the rate limiting was broken by an urgent request */
3422 bool migration_rate_limit(void)
3424 int64_t now
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3425 MigrationState
*s
= migrate_get_current();
3427 bool urgent
= false;
3428 migration_update_counters(s
, now
);
3429 if (qemu_file_rate_limit(s
->to_dst_file
)) {
3431 if (qemu_file_get_error(s
->to_dst_file
)) {
3435 * Wait for a delay to do rate limiting OR
3436 * something urgent to post the semaphore.
3438 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- now
;
3439 trace_migration_rate_limit_pre(ms
);
3440 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
3442 * We were woken by one or more urgent things but
3443 * the timedwait will have consumed one of them.
3444 * The service routine for the urgent wake will dec
3445 * the semaphore itself for each item it consumes,
3446 * so add this one we just eat back.
3448 qemu_sem_post(&s
->rate_limit_sem
);
3451 trace_migration_rate_limit_post(urgent
);
3457 * Master migration thread on the source VM.
3458 * It drives the migration and pumps the data down the outgoing channel.
3460 static void *migration_thread(void *opaque
)
3462 MigrationState
*s
= opaque
;
3463 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3464 MigThrError thr_error
;
3465 bool urgent
= false;
3467 rcu_register_thread();
3469 object_ref(OBJECT(s
));
3470 update_iteration_initial_status(s
);
3472 qemu_savevm_state_header(s
->to_dst_file
);
3475 * If we opened the return path, we need to make sure dst has it
3478 if (s
->rp_state
.from_dst_file
) {
3479 /* Now tell the dest that it should open its end so it can reply */
3480 qemu_savevm_send_open_return_path(s
->to_dst_file
);
3482 /* And do a ping that will make stuff easier to debug */
3483 qemu_savevm_send_ping(s
->to_dst_file
, 1);
3486 if (migrate_postcopy()) {
3488 * Tell the destination that we *might* want to do postcopy later;
3489 * if the other end can't do postcopy it should fail now, nice and
3492 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
3495 if (migrate_colo_enabled()) {
3496 /* Notify migration destination that we enable COLO */
3497 qemu_savevm_send_colo_enable(s
->to_dst_file
);
3500 qemu_savevm_state_setup(s
->to_dst_file
);
3502 if (qemu_savevm_state_guest_unplug_pending()) {
3503 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3504 MIGRATION_STATUS_WAIT_UNPLUG
);
3506 while (s
->state
== MIGRATION_STATUS_WAIT_UNPLUG
&&
3507 qemu_savevm_state_guest_unplug_pending()) {
3508 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
3511 migrate_set_state(&s
->state
, MIGRATION_STATUS_WAIT_UNPLUG
,
3512 MIGRATION_STATUS_ACTIVE
);
3515 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3516 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3517 MIGRATION_STATUS_ACTIVE
);
3519 trace_migration_thread_setup_complete();
3521 while (migration_is_active(s
)) {
3522 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
3523 MigIterateState iter_state
= migration_iteration_run(s
);
3524 if (iter_state
== MIG_ITERATE_SKIP
) {
3526 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3532 * Try to detect any kind of failures, and see whether we
3533 * should stop the migration now.
3535 thr_error
= migration_detect_error(s
);
3536 if (thr_error
== MIG_THR_ERR_FATAL
) {
3537 /* Stop migration */
3539 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
3541 * Just recovered from a e.g. network failure, reset all
3542 * the local variables. This is important to avoid
3543 * breaking transferred_bytes and bandwidth calculation
3545 update_iteration_initial_status(s
);
3548 urgent
= migration_rate_limit();
3551 trace_migration_thread_after_loop();
3552 migration_iteration_finish(s
);
3553 object_unref(OBJECT(s
));
3554 rcu_unregister_thread();
3558 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
3560 Error
*local_err
= NULL
;
3562 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
3564 s
->expected_downtime
= s
->parameters
.downtime_limit
;
3566 assert(s
->cleanup_bh
);
3568 assert(!s
->cleanup_bh
);
3569 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup_bh
, s
);
3572 migrate_fd_error(s
, error_in
);
3573 migrate_fd_cleanup(s
);
3578 /* This is a resumed migration */
3579 rate_limit
= s
->parameters
.max_postcopy_bandwidth
/
3582 /* This is a fresh new migration */
3583 rate_limit
= s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
;
3585 /* Notify before starting migration thread */
3586 notifier_list_notify(&migration_state_notifiers
, s
);
3589 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
3590 qemu_file_set_blocking(s
->to_dst_file
, true);
3593 * Open the return path. For postcopy, it is used exclusively. For
3594 * precopy, only if user specified "return-path" capability would
3595 * QEMU uses the return path.
3597 if (migrate_postcopy_ram() || migrate_use_return_path()) {
3598 if (open_return_path_on_source(s
, !resume
)) {
3599 error_report("Unable to open return-path for postcopy");
3600 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
3601 migrate_fd_cleanup(s
);
3607 /* Wakeup the main migration thread to do the recovery */
3608 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
3609 MIGRATION_STATUS_POSTCOPY_RECOVER
);
3610 qemu_sem_post(&s
->postcopy_pause_sem
);
3614 if (multifd_save_setup(&local_err
) != 0) {
3615 error_report_err(local_err
);
3616 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3617 MIGRATION_STATUS_FAILED
);
3618 migrate_fd_cleanup(s
);
3621 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
3622 QEMU_THREAD_JOINABLE
);
3623 s
->migration_thread_running
= true;
3626 void migration_global_dump(Monitor
*mon
)
3628 MigrationState
*ms
= migrate_get_current();
3630 monitor_printf(mon
, "globals:\n");
3631 monitor_printf(mon
, "store-global-state: %s\n",
3632 ms
->store_global_state
? "on" : "off");
3633 monitor_printf(mon
, "only-migratable: %s\n",
3634 only_migratable
? "on" : "off");
3635 monitor_printf(mon
, "send-configuration: %s\n",
3636 ms
->send_configuration
? "on" : "off");
3637 monitor_printf(mon
, "send-section-footer: %s\n",
3638 ms
->send_section_footer
? "on" : "off");
3639 monitor_printf(mon
, "decompress-error-check: %s\n",
3640 ms
->decompress_error_check
? "on" : "off");
3641 monitor_printf(mon
, "clear-bitmap-shift: %u\n",
3642 ms
->clear_bitmap_shift
);
3645 #define DEFINE_PROP_MIG_CAP(name, x) \
3646 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3648 static Property migration_properties
[] = {
3649 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
3650 store_global_state
, true),
3651 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
3652 send_configuration
, true),
3653 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
3654 send_section_footer
, true),
3655 DEFINE_PROP_BOOL("decompress-error-check", MigrationState
,
3656 decompress_error_check
, true),
3657 DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState
,
3658 clear_bitmap_shift
, CLEAR_BITMAP_SHIFT_DEFAULT
),
3660 /* Migration parameters */
3661 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
3662 parameters
.compress_level
,
3663 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
3664 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
3665 parameters
.compress_threads
,
3666 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
3667 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState
,
3668 parameters
.compress_wait_thread
, true),
3669 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
3670 parameters
.decompress_threads
,
3671 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
3672 DEFINE_PROP_UINT8("x-throttle-trigger-threshold", MigrationState
,
3673 parameters
.throttle_trigger_threshold
,
3674 DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD
),
3675 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
3676 parameters
.cpu_throttle_initial
,
3677 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
3678 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
3679 parameters
.cpu_throttle_increment
,
3680 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
3681 DEFINE_PROP_BOOL("x-cpu-throttle-tailslow", MigrationState
,
3682 parameters
.cpu_throttle_tailslow
, false),
3683 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
3684 parameters
.max_bandwidth
, MAX_THROTTLE
),
3685 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
3686 parameters
.downtime_limit
,
3687 DEFAULT_MIGRATE_SET_DOWNTIME
),
3688 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
3689 parameters
.x_checkpoint_delay
,
3690 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
3691 DEFINE_PROP_UINT8("multifd-channels", MigrationState
,
3692 parameters
.multifd_channels
,
3693 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
3694 DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState
,
3695 parameters
.multifd_compression
,
3696 DEFAULT_MIGRATE_MULTIFD_COMPRESSION
),
3697 DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState
,
3698 parameters
.multifd_zlib_level
,
3699 DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL
),
3700 DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState
,
3701 parameters
.multifd_zstd_level
,
3702 DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL
),
3703 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
3704 parameters
.xbzrle_cache_size
,
3705 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
3706 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState
,
3707 parameters
.max_postcopy_bandwidth
,
3708 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH
),
3709 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState
,
3710 parameters
.max_cpu_throttle
,
3711 DEFAULT_MIGRATE_MAX_CPU_THROTTLE
),
3712 DEFINE_PROP_SIZE("announce-initial", MigrationState
,
3713 parameters
.announce_initial
,
3714 DEFAULT_MIGRATE_ANNOUNCE_INITIAL
),
3715 DEFINE_PROP_SIZE("announce-max", MigrationState
,
3716 parameters
.announce_max
,
3717 DEFAULT_MIGRATE_ANNOUNCE_MAX
),
3718 DEFINE_PROP_SIZE("announce-rounds", MigrationState
,
3719 parameters
.announce_rounds
,
3720 DEFAULT_MIGRATE_ANNOUNCE_ROUNDS
),
3721 DEFINE_PROP_SIZE("announce-step", MigrationState
,
3722 parameters
.announce_step
,
3723 DEFAULT_MIGRATE_ANNOUNCE_STEP
),
3725 /* Migration capabilities */
3726 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
3727 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
3728 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
3729 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
3730 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
3731 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
3732 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
3733 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
3734 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
3735 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
3736 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
3737 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD
),
3739 DEFINE_PROP_END_OF_LIST(),
3742 static void migration_class_init(ObjectClass
*klass
, void *data
)
3744 DeviceClass
*dc
= DEVICE_CLASS(klass
);
3746 dc
->user_creatable
= false;
3747 device_class_set_props(dc
, migration_properties
);
3750 static void migration_instance_finalize(Object
*obj
)
3752 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3753 MigrationParameters
*params
= &ms
->parameters
;
3755 qemu_mutex_destroy(&ms
->error_mutex
);
3756 qemu_mutex_destroy(&ms
->qemu_file_lock
);
3757 g_free(params
->tls_hostname
);
3758 g_free(params
->tls_creds
);
3759 qemu_sem_destroy(&ms
->wait_unplug_sem
);
3760 qemu_sem_destroy(&ms
->rate_limit_sem
);
3761 qemu_sem_destroy(&ms
->pause_sem
);
3762 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
3763 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
3764 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
3765 error_free(ms
->error
);
3768 static void migration_instance_init(Object
*obj
)
3770 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3771 MigrationParameters
*params
= &ms
->parameters
;
3773 ms
->state
= MIGRATION_STATUS_NONE
;
3775 ms
->pages_per_second
= -1;
3776 qemu_sem_init(&ms
->pause_sem
, 0);
3777 qemu_mutex_init(&ms
->error_mutex
);
3779 params
->tls_hostname
= g_strdup("");
3780 params
->tls_creds
= g_strdup("");
3782 /* Set has_* up only for parameter checks */
3783 params
->has_compress_level
= true;
3784 params
->has_compress_threads
= true;
3785 params
->has_decompress_threads
= true;
3786 params
->has_throttle_trigger_threshold
= true;
3787 params
->has_cpu_throttle_initial
= true;
3788 params
->has_cpu_throttle_increment
= true;
3789 params
->has_cpu_throttle_tailslow
= true;
3790 params
->has_max_bandwidth
= true;
3791 params
->has_downtime_limit
= true;
3792 params
->has_x_checkpoint_delay
= true;
3793 params
->has_block_incremental
= true;
3794 params
->has_multifd_channels
= true;
3795 params
->has_multifd_compression
= true;
3796 params
->has_multifd_zlib_level
= true;
3797 params
->has_multifd_zstd_level
= true;
3798 params
->has_xbzrle_cache_size
= true;
3799 params
->has_max_postcopy_bandwidth
= true;
3800 params
->has_max_cpu_throttle
= true;
3801 params
->has_announce_initial
= true;
3802 params
->has_announce_max
= true;
3803 params
->has_announce_rounds
= true;
3804 params
->has_announce_step
= true;
3806 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
3807 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
3808 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
3809 qemu_sem_init(&ms
->rate_limit_sem
, 0);
3810 qemu_sem_init(&ms
->wait_unplug_sem
, 0);
3811 qemu_mutex_init(&ms
->qemu_file_lock
);
3815 * Return true if check pass, false otherwise. Error will be put
3816 * inside errp if provided.
3818 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
3820 MigrationCapabilityStatusList
*head
= NULL
;
3821 /* Assuming all off */
3822 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
3825 if (!migrate_params_check(&ms
->parameters
, errp
)) {
3829 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
3830 if (ms
->enabled_capabilities
[i
]) {
3831 head
= migrate_cap_add(head
, i
, true);
3835 ret
= migrate_caps_check(cap_list
, head
, errp
);
3837 /* It works with head == NULL */
3838 qapi_free_MigrationCapabilityStatusList(head
);
3843 static const TypeInfo migration_type
= {
3844 .name
= TYPE_MIGRATION
,
3846 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3847 * not created using qdev_new(), it is not attached to the qdev
3848 * device tree, and it is never realized.
3850 * TODO: Make this TYPE_OBJECT once QOM provides something like
3851 * TYPE_DEVICE's "-global" properties.
3853 .parent
= TYPE_DEVICE
,
3854 .class_init
= migration_class_init
,
3855 .class_size
= sizeof(MigrationClass
),
3856 .instance_size
= sizeof(MigrationState
),
3857 .instance_init
= migration_instance_init
,
3858 .instance_finalize
= migration_instance_finalize
,
3861 static void register_migration_types(void)
3863 type_register_static(&migration_type
);
3866 type_init(register_migration_types
);