4 * Copyright IBM, Corp. 2008
7 * Anthony Liguori <aliguori@us.ibm.com>
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
16 #include "qemu/osdep.h"
17 #include "qemu/cutils.h"
18 #include "qemu/error-report.h"
19 #include "qemu/main-loop.h"
20 #include "migration/blocker.h"
24 #include "sysemu/runstate.h"
25 #include "sysemu/sysemu.h"
26 #include "sysemu/cpu-throttle.h"
29 #include "migration/global_state.h"
30 #include "migration/misc.h"
31 #include "migration.h"
33 #include "qemu-file.h"
35 #include "migration/vmstate.h"
36 #include "block/block.h"
37 #include "qapi/error.h"
38 #include "qapi/clone-visitor.h"
39 #include "qapi/qapi-visit-migration.h"
40 #include "qapi/qapi-visit-sockets.h"
41 #include "qapi/qapi-commands-migration.h"
42 #include "qapi/qapi-events-migration.h"
43 #include "qapi/qmp/qerror.h"
44 #include "qapi/qmp/qnull.h"
47 #include "postcopy-ram.h"
48 #include "qemu/thread.h"
50 #include "exec/target_page.h"
51 #include "io/channel-buffer.h"
52 #include "io/channel-tls.h"
53 #include "migration/colo.h"
54 #include "hw/boards.h"
55 #include "hw/qdev-properties.h"
56 #include "hw/qdev-properties-system.h"
57 #include "monitor/monitor.h"
58 #include "net/announce.h"
59 #include "qemu/queue.h"
61 #include "qemu/yank.h"
62 #include "sysemu/cpus.h"
63 #include "yank_functions.h"
64 #include "sysemu/qtest.h"
65 #include "ui/qemu-spice.h"
67 #define MAX_THROTTLE (128 << 20) /* Migration transfer speed throttling */
69 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
71 #define BUFFER_DELAY 100
72 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
74 /* Time in milliseconds we are allowed to stop the source,
75 * for sending the last part */
76 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
78 /* Maximum migrate downtime set to 2000 seconds */
79 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
80 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
82 /* Default compression thread count */
83 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
84 /* Default decompression thread count, usually decompression is at
85 * least 4 times as fast as compression.*/
86 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
87 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
88 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
89 /* Define default autoconverge cpu throttle migration parameters */
90 #define DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD 50
91 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
92 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
93 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
95 /* Migration XBZRLE default cache size */
96 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
98 /* The delay time (in ms) between two COLO checkpoints */
99 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
100 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
101 #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE
102 /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */
103 #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
104 /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
105 #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
107 /* Background transfer rate for postcopy, 0 means unlimited, note
108 * that page requests can still exceed this limit.
110 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
113 * Parameters for self_announce_delay giving a stream of RARP/ARP
114 * packets after migration.
116 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
117 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
118 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
119 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
121 static NotifierList migration_state_notifiers
=
122 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
124 /* Messages sent on the return path from destination to source */
125 enum mig_rp_message_type
{
126 MIG_RP_MSG_INVALID
= 0, /* Must be 0 */
127 MIG_RP_MSG_SHUT
, /* sibling will not send any more RP messages */
128 MIG_RP_MSG_PONG
, /* Response to a PING; data (seq: be32 ) */
130 MIG_RP_MSG_REQ_PAGES_ID
, /* data (start: be64, len: be32, id: string) */
131 MIG_RP_MSG_REQ_PAGES
, /* data (start: be64, len: be32) */
132 MIG_RP_MSG_RECV_BITMAP
, /* send recved_bitmap back to source */
133 MIG_RP_MSG_RESUME_ACK
, /* tell source that we are ready to resume */
138 /* Migration capabilities set */
139 struct MigrateCapsSet
{
140 int size
; /* Capability set size */
141 MigrationCapability caps
[]; /* Variadic array of capabilities */
143 typedef struct MigrateCapsSet MigrateCapsSet
;
145 /* Define and initialize MigrateCapsSet */
146 #define INITIALIZE_MIGRATE_CAPS_SET(_name, ...) \
147 MigrateCapsSet _name = { \
148 .size = sizeof((int []) { __VA_ARGS__ }) / sizeof(int), \
149 .caps = { __VA_ARGS__ } \
152 /* Background-snapshot compatibility check list */
154 INITIALIZE_MIGRATE_CAPS_SET(check_caps_background_snapshot
,
155 MIGRATION_CAPABILITY_POSTCOPY_RAM
,
156 MIGRATION_CAPABILITY_DIRTY_BITMAPS
,
157 MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
,
158 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE
,
159 MIGRATION_CAPABILITY_RETURN_PATH
,
160 MIGRATION_CAPABILITY_MULTIFD
,
161 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
,
162 MIGRATION_CAPABILITY_AUTO_CONVERGE
,
163 MIGRATION_CAPABILITY_RELEASE_RAM
,
164 MIGRATION_CAPABILITY_RDMA_PIN_ALL
,
165 MIGRATION_CAPABILITY_COMPRESS
,
166 MIGRATION_CAPABILITY_XBZRLE
,
167 MIGRATION_CAPABILITY_X_COLO
,
168 MIGRATION_CAPABILITY_VALIDATE_UUID
,
169 MIGRATION_CAPABILITY_ZERO_COPY_SEND
);
171 /* When we add fault tolerance, we could have several
172 migrations at once. For now we don't need to add
173 dynamic creation of migration */
175 static MigrationState
*current_migration
;
176 static MigrationIncomingState
*current_incoming
;
178 static GSList
*migration_blockers
;
180 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
181 static int migration_maybe_pause(MigrationState
*s
,
182 int *current_active_state
,
184 static void migrate_fd_cancel(MigrationState
*s
);
186 static bool migrate_allow_multi_channels
= true;
188 void migrate_protocol_allow_multi_channels(bool allow
)
190 migrate_allow_multi_channels
= allow
;
193 bool migrate_multi_channels_is_allowed(void)
195 return migrate_allow_multi_channels
;
198 static gint
page_request_addr_cmp(gconstpointer ap
, gconstpointer bp
)
200 uintptr_t a
= (uintptr_t) ap
, b
= (uintptr_t) bp
;
202 return (a
> b
) - (a
< b
);
205 void migration_object_init(void)
207 /* This can only be called once. */
208 assert(!current_migration
);
209 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
212 * Init the migrate incoming object as well no matter whether
213 * we'll use it or not.
215 assert(!current_incoming
);
216 current_incoming
= g_new0(MigrationIncomingState
, 1);
217 current_incoming
->state
= MIGRATION_STATUS_NONE
;
218 current_incoming
->postcopy_remote_fds
=
219 g_array_new(FALSE
, TRUE
, sizeof(struct PostCopyFD
));
220 qemu_mutex_init(¤t_incoming
->rp_mutex
);
221 qemu_mutex_init(¤t_incoming
->postcopy_prio_thread_mutex
);
222 qemu_event_init(¤t_incoming
->main_thread_load_event
, false);
223 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_dst
, 0);
224 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fault
, 0);
225 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fast_load
, 0);
226 qemu_mutex_init(¤t_incoming
->page_request_mutex
);
227 current_incoming
->page_requested
= g_tree_new(page_request_addr_cmp
);
229 migration_object_check(current_migration
, &error_fatal
);
233 dirty_bitmap_mig_init();
236 void migration_cancel(const Error
*error
)
239 migrate_set_error(current_migration
, error
);
241 migrate_fd_cancel(current_migration
);
244 void migration_shutdown(void)
247 * When the QEMU main thread exit, the COLO thread
248 * may wait a semaphore. So, we should wakeup the
249 * COLO thread before migration shutdown.
253 * Cancel the current migration - that will (eventually)
254 * stop the migration using this structure
256 migration_cancel(NULL
);
257 object_unref(OBJECT(current_migration
));
260 * Cancel outgoing migration of dirty bitmaps. It should
261 * at least unref used block nodes.
263 dirty_bitmap_mig_cancel_outgoing();
266 * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
267 * are non-critical data, and their loss never considered as
270 dirty_bitmap_mig_cancel_incoming();
274 MigrationState
*migrate_get_current(void)
276 /* This can only be called after the object created. */
277 assert(current_migration
);
278 return current_migration
;
281 MigrationIncomingState
*migration_incoming_get_current(void)
283 assert(current_incoming
);
284 return current_incoming
;
287 void migration_incoming_transport_cleanup(MigrationIncomingState
*mis
)
289 if (mis
->socket_address_list
) {
290 qapi_free_SocketAddressList(mis
->socket_address_list
);
291 mis
->socket_address_list
= NULL
;
294 if (mis
->transport_cleanup
) {
295 mis
->transport_cleanup(mis
->transport_data
);
296 mis
->transport_data
= mis
->transport_cleanup
= NULL
;
300 void migration_incoming_state_destroy(void)
302 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
304 if (mis
->to_src_file
) {
305 /* Tell source that we are done */
306 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
307 qemu_fclose(mis
->to_src_file
);
308 mis
->to_src_file
= NULL
;
311 if (mis
->from_src_file
) {
312 migration_ioc_unregister_yank_from_file(mis
->from_src_file
);
313 qemu_fclose(mis
->from_src_file
);
314 mis
->from_src_file
= NULL
;
316 if (mis
->postcopy_remote_fds
) {
317 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
318 mis
->postcopy_remote_fds
= NULL
;
321 migration_incoming_transport_cleanup(mis
);
322 qemu_event_reset(&mis
->main_thread_load_event
);
324 if (mis
->page_requested
) {
325 g_tree_destroy(mis
->page_requested
);
326 mis
->page_requested
= NULL
;
329 if (mis
->postcopy_qemufile_dst
) {
330 migration_ioc_unregister_yank_from_file(mis
->postcopy_qemufile_dst
);
331 qemu_fclose(mis
->postcopy_qemufile_dst
);
332 mis
->postcopy_qemufile_dst
= NULL
;
335 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
338 static void migrate_generate_event(int new_state
)
340 if (migrate_use_events()) {
341 qapi_event_send_migration(new_state
);
345 static bool migrate_late_block_activate(void)
349 s
= migrate_get_current();
351 return s
->enabled_capabilities
[
352 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE
];
356 * Send a message on the return channel back to the source
359 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
360 enum mig_rp_message_type message_type
,
361 uint16_t len
, void *data
)
365 trace_migrate_send_rp_message((int)message_type
, len
);
366 QEMU_LOCK_GUARD(&mis
->rp_mutex
);
369 * It's possible that the file handle got lost due to network
372 if (!mis
->to_src_file
) {
377 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
378 qemu_put_be16(mis
->to_src_file
, len
);
379 qemu_put_buffer(mis
->to_src_file
, data
, len
);
380 qemu_fflush(mis
->to_src_file
);
382 /* It's possible that qemu file got error during sending */
383 ret
= qemu_file_get_error(mis
->to_src_file
);
388 /* Request one page from the source VM at the given start address.
389 * rb: the RAMBlock to request the page in
390 * Start: Address offset within the RB
391 * Len: Length in bytes required - must be a multiple of pagesize
393 int migrate_send_rp_message_req_pages(MigrationIncomingState
*mis
,
394 RAMBlock
*rb
, ram_addr_t start
)
396 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
397 size_t msglen
= 12; /* start + len */
398 size_t len
= qemu_ram_pagesize(rb
);
399 enum mig_rp_message_type msg_type
;
403 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
404 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
407 * We maintain the last ramblock that we requested for page. Note that we
408 * don't need locking because this function will only be called within the
409 * postcopy ram fault thread.
411 if (rb
!= mis
->last_rb
) {
414 rbname
= qemu_ram_get_idstr(rb
);
415 rbname_len
= strlen(rbname
);
417 assert(rbname_len
< 256);
419 bufc
[msglen
++] = rbname_len
;
420 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
421 msglen
+= rbname_len
;
422 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
424 msg_type
= MIG_RP_MSG_REQ_PAGES
;
427 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
430 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
,
431 RAMBlock
*rb
, ram_addr_t start
, uint64_t haddr
)
433 void *aligned
= (void *)(uintptr_t)ROUND_DOWN(haddr
, qemu_ram_pagesize(rb
));
434 bool received
= false;
436 WITH_QEMU_LOCK_GUARD(&mis
->page_request_mutex
) {
437 received
= ramblock_recv_bitmap_test_byte_offset(rb
, start
);
438 if (!received
&& !g_tree_lookup(mis
->page_requested
, aligned
)) {
440 * The page has not been received, and it's not yet in the page
441 * request list. Queue it. Set the value of element to 1, so that
442 * things like g_tree_lookup() will return TRUE (1) when found.
444 g_tree_insert(mis
->page_requested
, aligned
, (gpointer
)1);
445 mis
->page_requested_count
++;
446 trace_postcopy_page_req_add(aligned
, mis
->page_requested_count
);
451 * If the page is there, skip sending the message. We don't even need the
452 * lock because as long as the page arrived, it'll be there forever.
458 return migrate_send_rp_message_req_pages(mis
, rb
, start
);
461 static bool migration_colo_enabled
;
462 bool migration_incoming_colo_enabled(void)
464 return migration_colo_enabled
;
467 void migration_incoming_disable_colo(void)
469 ram_block_discard_disable(false);
470 migration_colo_enabled
= false;
473 int migration_incoming_enable_colo(void)
475 if (ram_block_discard_disable(true)) {
476 error_report("COLO: cannot disable RAM discard");
479 migration_colo_enabled
= true;
483 void migrate_add_address(SocketAddress
*address
)
485 MigrationIncomingState
*mis
= migration_incoming_get_current();
487 QAPI_LIST_PREPEND(mis
->socket_address_list
,
488 QAPI_CLONE(SocketAddress
, address
));
491 static void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
493 const char *p
= NULL
;
495 migrate_protocol_allow_multi_channels(false); /* reset it anyway */
496 qapi_event_send_migration(MIGRATION_STATUS_SETUP
);
497 if (strstart(uri
, "tcp:", &p
) ||
498 strstart(uri
, "unix:", NULL
) ||
499 strstart(uri
, "vsock:", NULL
)) {
500 migrate_protocol_allow_multi_channels(true);
501 socket_start_incoming_migration(p
? p
: uri
, errp
);
503 } else if (strstart(uri
, "rdma:", &p
)) {
504 rdma_start_incoming_migration(p
, errp
);
506 } else if (strstart(uri
, "exec:", &p
)) {
507 exec_start_incoming_migration(p
, errp
);
508 } else if (strstart(uri
, "fd:", &p
)) {
509 fd_start_incoming_migration(p
, errp
);
511 error_setg(errp
, "unknown migration protocol: %s", uri
);
515 static void process_incoming_migration_bh(void *opaque
)
517 Error
*local_err
= NULL
;
518 MigrationIncomingState
*mis
= opaque
;
520 /* If capability late_block_activate is set:
521 * Only fire up the block code now if we're going to restart the
522 * VM, else 'cont' will do it.
523 * This causes file locking to happen; so we don't want it to happen
524 * unless we really are starting the VM.
526 if (!migrate_late_block_activate() ||
527 (autostart
&& (!global_state_received() ||
528 global_state_get_runstate() == RUN_STATE_RUNNING
))) {
529 /* Make sure all file formats throw away their mutable metadata.
530 * If we get an error here, just don't restart the VM yet. */
531 bdrv_activate_all(&local_err
);
533 error_report_err(local_err
);
540 * This must happen after all error conditions are dealt with and
541 * we're sure the VM is going to be running on this host.
543 qemu_announce_self(&mis
->announce_timer
, migrate_announce_params());
545 if (multifd_load_cleanup(&local_err
) != 0) {
546 error_report_err(local_err
);
549 /* If global state section was not received or we are in running
550 state, we need to obey autostart. Any other state is set with
553 dirty_bitmap_mig_before_vm_start();
555 if (!global_state_received() ||
556 global_state_get_runstate() == RUN_STATE_RUNNING
) {
560 runstate_set(RUN_STATE_PAUSED
);
562 } else if (migration_incoming_colo_enabled()) {
563 migration_incoming_disable_colo();
566 runstate_set(global_state_get_runstate());
569 * This must happen after any state changes since as soon as an external
570 * observer sees this event they might start to prod at the VM assuming
573 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
574 MIGRATION_STATUS_COMPLETED
);
575 qemu_bh_delete(mis
->bh
);
576 migration_incoming_state_destroy();
579 static void coroutine_fn
580 process_incoming_migration_co(void *opaque
)
582 MigrationIncomingState
*mis
= migration_incoming_get_current();
585 Error
*local_err
= NULL
;
587 assert(mis
->from_src_file
);
588 mis
->migration_incoming_co
= qemu_coroutine_self();
589 mis
->largest_page_size
= qemu_ram_pagesize_largest();
590 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
591 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
592 MIGRATION_STATUS_ACTIVE
);
593 ret
= qemu_loadvm_state(mis
->from_src_file
);
595 ps
= postcopy_state_get();
596 trace_process_incoming_migration_co_end(ret
, ps
);
597 if (ps
!= POSTCOPY_INCOMING_NONE
) {
598 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
600 * Where a migration had postcopy enabled (and thus went to advise)
601 * but managed to complete within the precopy period, we can use
604 postcopy_ram_incoming_cleanup(mis
);
605 } else if (ret
>= 0) {
607 * Postcopy was started, cleanup should happen at the end of the
610 trace_process_incoming_migration_co_postcopy_end_main();
613 /* Else if something went wrong then just fall out of the normal exit */
616 /* we get COLO info, and know if we are in COLO mode */
617 if (!ret
&& migration_incoming_colo_enabled()) {
618 /* Make sure all file formats throw away their mutable metadata */
619 bdrv_activate_all(&local_err
);
621 error_report_err(local_err
);
625 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
626 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
627 mis
->have_colo_incoming_thread
= true;
628 qemu_coroutine_yield();
630 qemu_mutex_unlock_iothread();
631 /* Wait checkpoint incoming thread exit before free resource */
632 qemu_thread_join(&mis
->colo_incoming_thread
);
633 qemu_mutex_lock_iothread();
634 /* We hold the global iothread lock, so it is safe here */
635 colo_release_ram_cache();
639 error_report("load of migration failed: %s", strerror(-ret
));
642 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
643 qemu_bh_schedule(mis
->bh
);
644 mis
->migration_incoming_co
= NULL
;
648 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
649 MIGRATION_STATUS_FAILED
);
650 qemu_fclose(mis
->from_src_file
);
651 if (multifd_load_cleanup(&local_err
) != 0) {
652 error_report_err(local_err
);
658 * migration_incoming_setup: Setup incoming migration
659 * @f: file for main migration channel
660 * @errp: where to put errors
662 * Returns: %true on success, %false on error.
664 static bool migration_incoming_setup(QEMUFile
*f
, Error
**errp
)
666 MigrationIncomingState
*mis
= migration_incoming_get_current();
668 if (!mis
->from_src_file
) {
669 mis
->from_src_file
= f
;
671 qemu_file_set_blocking(f
, false);
675 void migration_incoming_process(void)
677 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
678 qemu_coroutine_enter(co
);
681 /* Returns true if recovered from a paused migration, otherwise false */
682 static bool postcopy_try_recover(void)
684 MigrationIncomingState
*mis
= migration_incoming_get_current();
686 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
687 /* Resumed from a paused postcopy migration */
689 /* This should be set already in migration_incoming_setup() */
690 assert(mis
->from_src_file
);
691 /* Postcopy has standalone thread to do vm load */
692 qemu_file_set_blocking(mis
->from_src_file
, true);
694 /* Re-configure the return path */
695 mis
->to_src_file
= qemu_file_get_return_path(mis
->from_src_file
);
697 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
698 MIGRATION_STATUS_POSTCOPY_RECOVER
);
701 * Here, we only wake up the main loading thread (while the
702 * rest threads will still be waiting), so that we can receive
703 * commands from source now, and answer it if needed. The
704 * rest threads will be woken up afterwards until we are sure
705 * that source is ready to reply to page requests.
707 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
714 void migration_fd_process_incoming(QEMUFile
*f
, Error
**errp
)
716 if (!migration_incoming_setup(f
, errp
)) {
719 if (postcopy_try_recover()) {
722 migration_incoming_process();
725 static bool migration_needs_multiple_sockets(void)
727 return migrate_use_multifd() || migrate_postcopy_preempt();
730 void migration_ioc_process_incoming(QIOChannel
*ioc
, Error
**errp
)
732 MigrationIncomingState
*mis
= migration_incoming_get_current();
733 Error
*local_err
= NULL
;
735 bool default_channel
= true;
736 uint32_t channel_magic
= 0;
739 if (migrate_use_multifd() && !migrate_postcopy_ram() &&
740 qio_channel_has_feature(ioc
, QIO_CHANNEL_FEATURE_READ_MSG_PEEK
)) {
742 * With multiple channels, it is possible that we receive channels
743 * out of order on destination side, causing incorrect mapping of
744 * source channels on destination side. Check channel MAGIC to
745 * decide type of channel. Please note this is best effort, postcopy
746 * preempt channel does not send any magic number so avoid it for
747 * postcopy live migration. Also tls live migration already does
748 * tls handshake while initializing main channel so with tls this
749 * issue is not possible.
751 ret
= migration_channel_read_peek(ioc
, (void *)&channel_magic
,
752 sizeof(channel_magic
), &local_err
);
755 error_propagate(errp
, local_err
);
759 default_channel
= (channel_magic
== cpu_to_be32(QEMU_VM_FILE_MAGIC
));
761 default_channel
= !mis
->from_src_file
;
764 if (multifd_load_setup(errp
) != 0) {
765 error_setg(errp
, "Failed to setup multifd channels");
769 if (default_channel
) {
770 f
= qemu_file_new_input(ioc
);
772 if (!migration_incoming_setup(f
, errp
)) {
776 /* Multiple connections */
777 assert(migration_needs_multiple_sockets());
778 if (migrate_use_multifd()) {
779 multifd_recv_new_channel(ioc
, &local_err
);
781 assert(migrate_postcopy_preempt());
782 f
= qemu_file_new_input(ioc
);
783 postcopy_preempt_new_channel(mis
, f
);
786 error_propagate(errp
, local_err
);
791 if (migration_has_all_channels()) {
792 /* If it's a recovery, we're done */
793 if (postcopy_try_recover()) {
796 migration_incoming_process();
801 * @migration_has_all_channels: We have received all channels that we need
803 * Returns true when we have got connections to all the channels that
804 * we need for migration.
806 bool migration_has_all_channels(void)
808 MigrationIncomingState
*mis
= migration_incoming_get_current();
810 if (!mis
->from_src_file
) {
814 if (migrate_use_multifd()) {
815 return multifd_recv_all_channels_created();
818 if (migrate_postcopy_preempt()) {
819 return mis
->postcopy_qemufile_dst
!= NULL
;
826 * Send a 'SHUT' message on the return channel with the given value
827 * to indicate that we've finished with the RP. Non-0 value indicates
830 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
835 buf
= cpu_to_be32(value
);
836 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
840 * Send a 'PONG' message on the return channel with the given value
841 * (normally in response to a 'PING')
843 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
848 buf
= cpu_to_be32(value
);
849 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
852 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
860 * First, we send the header part. It contains only the len of
861 * idstr, and the idstr itself.
863 len
= strlen(block_name
);
865 memcpy(buf
+ 1, block_name
, len
);
867 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
868 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
873 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
876 * Next, we dump the received bitmap to the stream.
878 * TODO: currently we are safe since we are the only one that is
879 * using the to_src_file handle (fault thread is still paused),
880 * and it's ok even not taking the mutex. However the best way is
881 * to take the lock before sending the message header, and release
882 * the lock after sending the bitmap.
884 qemu_mutex_lock(&mis
->rp_mutex
);
885 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
886 qemu_mutex_unlock(&mis
->rp_mutex
);
888 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
891 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
895 buf
= cpu_to_be32(value
);
896 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
899 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
901 MigrationCapabilityStatusList
*head
= NULL
, **tail
= &head
;
902 MigrationCapabilityStatus
*caps
;
903 MigrationState
*s
= migrate_get_current();
906 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
907 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
908 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
912 caps
= g_malloc0(sizeof(*caps
));
913 caps
->capability
= i
;
914 caps
->state
= s
->enabled_capabilities
[i
];
915 QAPI_LIST_APPEND(tail
, caps
);
921 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
923 MigrationParameters
*params
;
924 MigrationState
*s
= migrate_get_current();
926 /* TODO use QAPI_CLONE() instead of duplicating it inline */
927 params
= g_malloc0(sizeof(*params
));
928 params
->has_compress_level
= true;
929 params
->compress_level
= s
->parameters
.compress_level
;
930 params
->has_compress_threads
= true;
931 params
->compress_threads
= s
->parameters
.compress_threads
;
932 params
->has_compress_wait_thread
= true;
933 params
->compress_wait_thread
= s
->parameters
.compress_wait_thread
;
934 params
->has_decompress_threads
= true;
935 params
->decompress_threads
= s
->parameters
.decompress_threads
;
936 params
->has_throttle_trigger_threshold
= true;
937 params
->throttle_trigger_threshold
= s
->parameters
.throttle_trigger_threshold
;
938 params
->has_cpu_throttle_initial
= true;
939 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
940 params
->has_cpu_throttle_increment
= true;
941 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
942 params
->has_cpu_throttle_tailslow
= true;
943 params
->cpu_throttle_tailslow
= s
->parameters
.cpu_throttle_tailslow
;
944 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
945 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
946 params
->tls_authz
= g_strdup(s
->parameters
.tls_authz
?
947 s
->parameters
.tls_authz
: "");
948 params
->has_max_bandwidth
= true;
949 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
950 params
->has_downtime_limit
= true;
951 params
->downtime_limit
= s
->parameters
.downtime_limit
;
952 params
->has_x_checkpoint_delay
= true;
953 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
954 params
->has_block_incremental
= true;
955 params
->block_incremental
= s
->parameters
.block_incremental
;
956 params
->has_multifd_channels
= true;
957 params
->multifd_channels
= s
->parameters
.multifd_channels
;
958 params
->has_multifd_compression
= true;
959 params
->multifd_compression
= s
->parameters
.multifd_compression
;
960 params
->has_multifd_zlib_level
= true;
961 params
->multifd_zlib_level
= s
->parameters
.multifd_zlib_level
;
962 params
->has_multifd_zstd_level
= true;
963 params
->multifd_zstd_level
= s
->parameters
.multifd_zstd_level
;
964 params
->has_xbzrle_cache_size
= true;
965 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
966 params
->has_max_postcopy_bandwidth
= true;
967 params
->max_postcopy_bandwidth
= s
->parameters
.max_postcopy_bandwidth
;
968 params
->has_max_cpu_throttle
= true;
969 params
->max_cpu_throttle
= s
->parameters
.max_cpu_throttle
;
970 params
->has_announce_initial
= true;
971 params
->announce_initial
= s
->parameters
.announce_initial
;
972 params
->has_announce_max
= true;
973 params
->announce_max
= s
->parameters
.announce_max
;
974 params
->has_announce_rounds
= true;
975 params
->announce_rounds
= s
->parameters
.announce_rounds
;
976 params
->has_announce_step
= true;
977 params
->announce_step
= s
->parameters
.announce_step
;
979 if (s
->parameters
.has_block_bitmap_mapping
) {
980 params
->has_block_bitmap_mapping
= true;
981 params
->block_bitmap_mapping
=
982 QAPI_CLONE(BitmapMigrationNodeAliasList
,
983 s
->parameters
.block_bitmap_mapping
);
989 void qmp_client_migrate_info(const char *protocol
, const char *hostname
,
990 bool has_port
, int64_t port
,
991 bool has_tls_port
, int64_t tls_port
,
992 const char *cert_subject
,
995 if (strcmp(protocol
, "spice") == 0) {
996 if (!qemu_using_spice(errp
)) {
1000 if (!has_port
&& !has_tls_port
) {
1001 error_setg(errp
, QERR_MISSING_PARAMETER
, "port/tls-port");
1005 if (qemu_spice
.migrate_info(hostname
,
1006 has_port
? port
: -1,
1007 has_tls_port
? tls_port
: -1,
1009 error_setg(errp
, "Could not set up display for migration");
1015 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "protocol", "'spice'");
1018 AnnounceParameters
*migrate_announce_params(void)
1020 static AnnounceParameters ap
;
1022 MigrationState
*s
= migrate_get_current();
1024 ap
.initial
= s
->parameters
.announce_initial
;
1025 ap
.max
= s
->parameters
.announce_max
;
1026 ap
.rounds
= s
->parameters
.announce_rounds
;
1027 ap
.step
= s
->parameters
.announce_step
;
1033 * Return true if we're already in the middle of a migration
1034 * (i.e. any of the active or setup states)
1036 bool migration_is_setup_or_active(int state
)
1039 case MIGRATION_STATUS_ACTIVE
:
1040 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1041 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1042 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1043 case MIGRATION_STATUS_SETUP
:
1044 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1045 case MIGRATION_STATUS_DEVICE
:
1046 case MIGRATION_STATUS_WAIT_UNPLUG
:
1047 case MIGRATION_STATUS_COLO
:
1056 bool migration_is_running(int state
)
1059 case MIGRATION_STATUS_ACTIVE
:
1060 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1061 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1062 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1063 case MIGRATION_STATUS_SETUP
:
1064 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1065 case MIGRATION_STATUS_DEVICE
:
1066 case MIGRATION_STATUS_WAIT_UNPLUG
:
1067 case MIGRATION_STATUS_CANCELLING
:
1076 static bool migrate_show_downtime(MigrationState
*s
)
1078 return (s
->state
== MIGRATION_STATUS_COMPLETED
) || migration_in_postcopy();
1081 static void populate_time_info(MigrationInfo
*info
, MigrationState
*s
)
1083 info
->has_status
= true;
1084 info
->has_setup_time
= true;
1085 info
->setup_time
= s
->setup_time
;
1087 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
1088 info
->has_total_time
= true;
1089 info
->total_time
= s
->total_time
;
1091 info
->has_total_time
= true;
1092 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) -
1096 if (migrate_show_downtime(s
)) {
1097 info
->has_downtime
= true;
1098 info
->downtime
= s
->downtime
;
1100 info
->has_expected_downtime
= true;
1101 info
->expected_downtime
= s
->expected_downtime
;
1105 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
1107 size_t page_size
= qemu_target_page_size();
1109 info
->ram
= g_malloc0(sizeof(*info
->ram
));
1110 info
->ram
->transferred
= stat64_get(&ram_atomic_counters
.transferred
);
1111 info
->ram
->total
= ram_bytes_total();
1112 info
->ram
->duplicate
= stat64_get(&ram_atomic_counters
.duplicate
);
1113 /* legacy value. It is not used anymore */
1114 info
->ram
->skipped
= 0;
1115 info
->ram
->normal
= stat64_get(&ram_atomic_counters
.normal
);
1116 info
->ram
->normal_bytes
= info
->ram
->normal
* page_size
;
1117 info
->ram
->mbps
= s
->mbps
;
1118 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
1119 info
->ram
->dirty_sync_missed_zero_copy
=
1120 ram_counters
.dirty_sync_missed_zero_copy
;
1121 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
1122 info
->ram
->page_size
= page_size
;
1123 info
->ram
->multifd_bytes
= ram_counters
.multifd_bytes
;
1124 info
->ram
->pages_per_second
= s
->pages_per_second
;
1125 info
->ram
->precopy_bytes
= ram_counters
.precopy_bytes
;
1126 info
->ram
->downtime_bytes
= ram_counters
.downtime_bytes
;
1127 info
->ram
->postcopy_bytes
= stat64_get(&ram_atomic_counters
.postcopy_bytes
);
1129 if (migrate_use_xbzrle()) {
1130 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
1131 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
1132 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
1133 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
1134 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
1135 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
1136 info
->xbzrle_cache
->encoding_rate
= xbzrle_counters
.encoding_rate
;
1137 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
1140 if (migrate_use_compression()) {
1141 info
->compression
= g_malloc0(sizeof(*info
->compression
));
1142 info
->compression
->pages
= compression_counters
.pages
;
1143 info
->compression
->busy
= compression_counters
.busy
;
1144 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
1145 info
->compression
->compressed_size
=
1146 compression_counters
.compressed_size
;
1147 info
->compression
->compression_rate
=
1148 compression_counters
.compression_rate
;
1151 if (cpu_throttle_active()) {
1152 info
->has_cpu_throttle_percentage
= true;
1153 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
1156 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
1157 info
->ram
->remaining
= ram_bytes_remaining();
1158 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
1162 static void populate_disk_info(MigrationInfo
*info
)
1164 if (blk_mig_active()) {
1165 info
->disk
= g_malloc0(sizeof(*info
->disk
));
1166 info
->disk
->transferred
= blk_mig_bytes_transferred();
1167 info
->disk
->remaining
= blk_mig_bytes_remaining();
1168 info
->disk
->total
= blk_mig_bytes_total();
1172 static void fill_source_migration_info(MigrationInfo
*info
)
1174 MigrationState
*s
= migrate_get_current();
1175 int state
= qatomic_read(&s
->state
);
1176 GSList
*cur_blocker
= migration_blockers
;
1178 info
->blocked_reasons
= NULL
;
1181 * There are two types of reasons a migration might be blocked;
1182 * a) devices marked in VMState as non-migratable, and
1183 * b) Explicit migration blockers
1184 * We need to add both of them here.
1186 qemu_savevm_non_migratable_list(&info
->blocked_reasons
);
1188 while (cur_blocker
) {
1189 QAPI_LIST_PREPEND(info
->blocked_reasons
,
1190 g_strdup(error_get_pretty(cur_blocker
->data
)));
1191 cur_blocker
= g_slist_next(cur_blocker
);
1193 info
->has_blocked_reasons
= info
->blocked_reasons
!= NULL
;
1196 case MIGRATION_STATUS_NONE
:
1197 /* no migration has happened ever */
1198 /* do not overwrite destination migration status */
1200 case MIGRATION_STATUS_SETUP
:
1201 info
->has_status
= true;
1202 info
->has_total_time
= false;
1204 case MIGRATION_STATUS_ACTIVE
:
1205 case MIGRATION_STATUS_CANCELLING
:
1206 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1207 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1208 case MIGRATION_STATUS_DEVICE
:
1209 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1210 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1211 /* TODO add some postcopy stats */
1212 populate_time_info(info
, s
);
1213 populate_ram_info(info
, s
);
1214 populate_disk_info(info
);
1215 populate_vfio_info(info
);
1217 case MIGRATION_STATUS_COLO
:
1218 info
->has_status
= true;
1219 /* TODO: display COLO specific information (checkpoint info etc.) */
1221 case MIGRATION_STATUS_COMPLETED
:
1222 populate_time_info(info
, s
);
1223 populate_ram_info(info
, s
);
1224 populate_vfio_info(info
);
1226 case MIGRATION_STATUS_FAILED
:
1227 info
->has_status
= true;
1229 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
1232 case MIGRATION_STATUS_CANCELLED
:
1233 info
->has_status
= true;
1235 case MIGRATION_STATUS_WAIT_UNPLUG
:
1236 info
->has_status
= true;
1239 info
->status
= state
;
1242 typedef enum WriteTrackingSupport
{
1243 WT_SUPPORT_UNKNOWN
= 0,
1245 WT_SUPPORT_AVAILABLE
,
1246 WT_SUPPORT_COMPATIBLE
1247 } WriteTrackingSupport
;
1250 WriteTrackingSupport
migrate_query_write_tracking(void)
1252 /* Check if kernel supports required UFFD features */
1253 if (!ram_write_tracking_available()) {
1254 return WT_SUPPORT_ABSENT
;
1257 * Check if current memory configuration is
1258 * compatible with required UFFD features.
1260 if (!ram_write_tracking_compatible()) {
1261 return WT_SUPPORT_AVAILABLE
;
1264 return WT_SUPPORT_COMPATIBLE
;
1268 * @migration_caps_check - check capability validity
1270 * @cap_list: old capability list, array of bool
1271 * @params: new capabilities to be applied soon
1272 * @errp: set *errp if the check failed, with reason
1274 * Returns true if check passed, otherwise false.
1276 static bool migrate_caps_check(bool *cap_list
,
1277 MigrationCapabilityStatusList
*params
,
1280 MigrationCapabilityStatusList
*cap
;
1281 bool old_postcopy_cap
;
1282 MigrationIncomingState
*mis
= migration_incoming_get_current();
1284 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1286 for (cap
= params
; cap
; cap
= cap
->next
) {
1287 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
1290 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1291 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
1292 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
1294 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
1299 #ifndef CONFIG_REPLICATION
1300 if (cap_list
[MIGRATION_CAPABILITY_X_COLO
]) {
1301 error_setg(errp
, "QEMU compiled without replication module"
1302 " can't enable COLO");
1303 error_append_hint(errp
, "Please enable replication before COLO.\n");
1308 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
1309 /* This check is reasonably expensive, so only when it's being
1310 * set the first time, also it's only the destination that needs
1313 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
1314 !postcopy_ram_supported_by_host(mis
)) {
1315 /* postcopy_ram_supported_by_host will have emitted a more
1318 error_setg(errp
, "Postcopy is not supported");
1322 if (cap_list
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
]) {
1323 error_setg(errp
, "Postcopy is not compatible with ignore-shared");
1328 if (cap_list
[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
]) {
1329 WriteTrackingSupport wt_support
;
1332 * Check if 'background-snapshot' capability is supported by
1333 * host kernel and compatible with guest memory configuration.
1335 wt_support
= migrate_query_write_tracking();
1336 if (wt_support
< WT_SUPPORT_AVAILABLE
) {
1337 error_setg(errp
, "Background-snapshot is not supported by host kernel");
1340 if (wt_support
< WT_SUPPORT_COMPATIBLE
) {
1341 error_setg(errp
, "Background-snapshot is not compatible "
1342 "with guest memory configuration");
1347 * Check if there are any migration capabilities
1348 * incompatible with 'background-snapshot'.
1350 for (idx
= 0; idx
< check_caps_background_snapshot
.size
; idx
++) {
1351 int incomp_cap
= check_caps_background_snapshot
.caps
[idx
];
1352 if (cap_list
[incomp_cap
]) {
1354 "Background-snapshot is not compatible with %s",
1355 MigrationCapability_str(incomp_cap
));
1362 if (cap_list
[MIGRATION_CAPABILITY_ZERO_COPY_SEND
] &&
1363 (!cap_list
[MIGRATION_CAPABILITY_MULTIFD
] ||
1364 cap_list
[MIGRATION_CAPABILITY_COMPRESS
] ||
1365 cap_list
[MIGRATION_CAPABILITY_XBZRLE
] ||
1366 migrate_multifd_compression() ||
1367 migrate_use_tls())) {
1369 "Zero copy only available for non-compressed non-TLS multifd migration");
1373 if (cap_list
[MIGRATION_CAPABILITY_ZERO_COPY_SEND
]) {
1375 "Zero copy currently only available on Linux");
1381 /* incoming side only */
1382 if (runstate_check(RUN_STATE_INMIGRATE
) &&
1383 !migrate_multi_channels_is_allowed() &&
1384 cap_list
[MIGRATION_CAPABILITY_MULTIFD
]) {
1385 error_setg(errp
, "multifd is not supported by current protocol");
1389 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_PREEMPT
]) {
1390 if (!cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
1391 error_setg(errp
, "Postcopy preempt requires postcopy-ram");
1396 * Preempt mode requires urgent pages to be sent in separate
1397 * channel, OTOH compression logic will disorder all pages into
1398 * different compression channels, which is not compatible with the
1399 * preempt assumptions on channel assignments.
1401 if (cap_list
[MIGRATION_CAPABILITY_COMPRESS
]) {
1402 error_setg(errp
, "Postcopy preempt not compatible with compress");
1407 if (cap_list
[MIGRATION_CAPABILITY_MULTIFD
]) {
1408 if (cap_list
[MIGRATION_CAPABILITY_COMPRESS
]) {
1409 error_setg(errp
, "Multifd is not compatible with compress");
1417 static void fill_destination_migration_info(MigrationInfo
*info
)
1419 MigrationIncomingState
*mis
= migration_incoming_get_current();
1421 if (mis
->socket_address_list
) {
1422 info
->has_socket_address
= true;
1423 info
->socket_address
=
1424 QAPI_CLONE(SocketAddressList
, mis
->socket_address_list
);
1427 switch (mis
->state
) {
1428 case MIGRATION_STATUS_NONE
:
1430 case MIGRATION_STATUS_SETUP
:
1431 case MIGRATION_STATUS_CANCELLING
:
1432 case MIGRATION_STATUS_CANCELLED
:
1433 case MIGRATION_STATUS_ACTIVE
:
1434 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1435 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1436 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1437 case MIGRATION_STATUS_FAILED
:
1438 case MIGRATION_STATUS_COLO
:
1439 info
->has_status
= true;
1441 case MIGRATION_STATUS_COMPLETED
:
1442 info
->has_status
= true;
1443 fill_destination_postcopy_migration_info(info
);
1446 info
->status
= mis
->state
;
1449 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1451 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1453 fill_destination_migration_info(info
);
1454 fill_source_migration_info(info
);
1459 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
1462 MigrationState
*s
= migrate_get_current();
1463 MigrationCapabilityStatusList
*cap
;
1464 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
1466 if (migration_is_running(s
->state
)) {
1467 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1471 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
1472 if (!migrate_caps_check(cap_list
, params
, errp
)) {
1476 for (cap
= params
; cap
; cap
= cap
->next
) {
1477 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
1482 * Check whether the parameters are valid. Error will be put into errp
1483 * (if provided). Return true if valid, otherwise false.
1485 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
1487 if (params
->has_compress_level
&&
1488 (params
->compress_level
> 9)) {
1489 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
1490 "a value between 0 and 9");
1494 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
1495 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1497 "a value between 1 and 255");
1501 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
1502 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1503 "decompress_threads",
1504 "a value between 1 and 255");
1508 if (params
->has_throttle_trigger_threshold
&&
1509 (params
->throttle_trigger_threshold
< 1 ||
1510 params
->throttle_trigger_threshold
> 100)) {
1511 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1512 "throttle_trigger_threshold",
1513 "an integer in the range of 1 to 100");
1517 if (params
->has_cpu_throttle_initial
&&
1518 (params
->cpu_throttle_initial
< 1 ||
1519 params
->cpu_throttle_initial
> 99)) {
1520 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1521 "cpu_throttle_initial",
1522 "an integer in the range of 1 to 99");
1526 if (params
->has_cpu_throttle_increment
&&
1527 (params
->cpu_throttle_increment
< 1 ||
1528 params
->cpu_throttle_increment
> 99)) {
1529 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1530 "cpu_throttle_increment",
1531 "an integer in the range of 1 to 99");
1535 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
1536 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1538 "an integer in the range of 0 to "stringify(SIZE_MAX
)
1543 if (params
->has_downtime_limit
&&
1544 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
1545 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1547 "an integer in the range of 0 to "
1548 stringify(MAX_MIGRATE_DOWNTIME
)" ms");
1552 /* x_checkpoint_delay is now always positive */
1554 if (params
->has_multifd_channels
&& (params
->multifd_channels
< 1)) {
1555 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1557 "a value between 1 and 255");
1561 if (params
->has_multifd_zlib_level
&&
1562 (params
->multifd_zlib_level
> 9)) {
1563 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zlib_level",
1564 "a value between 0 and 9");
1568 if (params
->has_multifd_zstd_level
&&
1569 (params
->multifd_zstd_level
> 20)) {
1570 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zstd_level",
1571 "a value between 0 and 20");
1575 if (params
->has_xbzrle_cache_size
&&
1576 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
1577 !is_power_of_2(params
->xbzrle_cache_size
))) {
1578 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1579 "xbzrle_cache_size",
1580 "a power of two no less than the target page size");
1584 if (params
->has_max_cpu_throttle
&&
1585 (params
->max_cpu_throttle
< params
->cpu_throttle_initial
||
1586 params
->max_cpu_throttle
> 99)) {
1587 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1589 "an integer in the range of cpu_throttle_initial to 99");
1593 if (params
->has_announce_initial
&&
1594 params
->announce_initial
> 100000) {
1595 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1597 "a value between 0 and 100000");
1600 if (params
->has_announce_max
&&
1601 params
->announce_max
> 100000) {
1602 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1604 "a value between 0 and 100000");
1607 if (params
->has_announce_rounds
&&
1608 params
->announce_rounds
> 1000) {
1609 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1611 "a value between 0 and 1000");
1614 if (params
->has_announce_step
&&
1615 (params
->announce_step
< 1 ||
1616 params
->announce_step
> 10000)) {
1617 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1619 "a value between 0 and 10000");
1623 if (params
->has_block_bitmap_mapping
&&
1624 !check_dirty_bitmap_mig_alias_map(params
->block_bitmap_mapping
, errp
)) {
1625 error_prepend(errp
, "Invalid mapping given for block-bitmap-mapping: ");
1630 if (migrate_use_zero_copy_send() &&
1631 ((params
->has_multifd_compression
&& params
->multifd_compression
) ||
1632 (params
->tls_creds
&& *params
->tls_creds
))) {
1634 "Zero copy only available for non-compressed non-TLS multifd migration");
1642 static void migrate_params_test_apply(MigrateSetParameters
*params
,
1643 MigrationParameters
*dest
)
1645 *dest
= migrate_get_current()->parameters
;
1647 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1649 if (params
->has_compress_level
) {
1650 dest
->compress_level
= params
->compress_level
;
1653 if (params
->has_compress_threads
) {
1654 dest
->compress_threads
= params
->compress_threads
;
1657 if (params
->has_compress_wait_thread
) {
1658 dest
->compress_wait_thread
= params
->compress_wait_thread
;
1661 if (params
->has_decompress_threads
) {
1662 dest
->decompress_threads
= params
->decompress_threads
;
1665 if (params
->has_throttle_trigger_threshold
) {
1666 dest
->throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1669 if (params
->has_cpu_throttle_initial
) {
1670 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
1673 if (params
->has_cpu_throttle_increment
) {
1674 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
1677 if (params
->has_cpu_throttle_tailslow
) {
1678 dest
->cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1681 if (params
->tls_creds
) {
1682 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1683 dest
->tls_creds
= params
->tls_creds
->u
.s
;
1686 if (params
->tls_hostname
) {
1687 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1688 dest
->tls_hostname
= params
->tls_hostname
->u
.s
;
1691 if (params
->has_max_bandwidth
) {
1692 dest
->max_bandwidth
= params
->max_bandwidth
;
1695 if (params
->has_downtime_limit
) {
1696 dest
->downtime_limit
= params
->downtime_limit
;
1699 if (params
->has_x_checkpoint_delay
) {
1700 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
1703 if (params
->has_block_incremental
) {
1704 dest
->block_incremental
= params
->block_incremental
;
1706 if (params
->has_multifd_channels
) {
1707 dest
->multifd_channels
= params
->multifd_channels
;
1709 if (params
->has_multifd_compression
) {
1710 dest
->multifd_compression
= params
->multifd_compression
;
1712 if (params
->has_xbzrle_cache_size
) {
1713 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
1715 if (params
->has_max_postcopy_bandwidth
) {
1716 dest
->max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1718 if (params
->has_max_cpu_throttle
) {
1719 dest
->max_cpu_throttle
= params
->max_cpu_throttle
;
1721 if (params
->has_announce_initial
) {
1722 dest
->announce_initial
= params
->announce_initial
;
1724 if (params
->has_announce_max
) {
1725 dest
->announce_max
= params
->announce_max
;
1727 if (params
->has_announce_rounds
) {
1728 dest
->announce_rounds
= params
->announce_rounds
;
1730 if (params
->has_announce_step
) {
1731 dest
->announce_step
= params
->announce_step
;
1734 if (params
->has_block_bitmap_mapping
) {
1735 dest
->has_block_bitmap_mapping
= true;
1736 dest
->block_bitmap_mapping
= params
->block_bitmap_mapping
;
1740 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
1742 MigrationState
*s
= migrate_get_current();
1744 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1746 if (params
->has_compress_level
) {
1747 s
->parameters
.compress_level
= params
->compress_level
;
1750 if (params
->has_compress_threads
) {
1751 s
->parameters
.compress_threads
= params
->compress_threads
;
1754 if (params
->has_compress_wait_thread
) {
1755 s
->parameters
.compress_wait_thread
= params
->compress_wait_thread
;
1758 if (params
->has_decompress_threads
) {
1759 s
->parameters
.decompress_threads
= params
->decompress_threads
;
1762 if (params
->has_throttle_trigger_threshold
) {
1763 s
->parameters
.throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1766 if (params
->has_cpu_throttle_initial
) {
1767 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
1770 if (params
->has_cpu_throttle_increment
) {
1771 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
1774 if (params
->has_cpu_throttle_tailslow
) {
1775 s
->parameters
.cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1778 if (params
->tls_creds
) {
1779 g_free(s
->parameters
.tls_creds
);
1780 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1781 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1784 if (params
->tls_hostname
) {
1785 g_free(s
->parameters
.tls_hostname
);
1786 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1787 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1790 if (params
->tls_authz
) {
1791 g_free(s
->parameters
.tls_authz
);
1792 assert(params
->tls_authz
->type
== QTYPE_QSTRING
);
1793 s
->parameters
.tls_authz
= g_strdup(params
->tls_authz
->u
.s
);
1796 if (params
->has_max_bandwidth
) {
1797 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1798 if (s
->to_dst_file
&& !migration_in_postcopy()) {
1799 qemu_file_set_rate_limit(s
->to_dst_file
,
1800 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1804 if (params
->has_downtime_limit
) {
1805 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1808 if (params
->has_x_checkpoint_delay
) {
1809 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1810 if (migration_in_colo_state()) {
1811 colo_checkpoint_notify(s
);
1815 if (params
->has_block_incremental
) {
1816 s
->parameters
.block_incremental
= params
->block_incremental
;
1818 if (params
->has_multifd_channels
) {
1819 s
->parameters
.multifd_channels
= params
->multifd_channels
;
1821 if (params
->has_multifd_compression
) {
1822 s
->parameters
.multifd_compression
= params
->multifd_compression
;
1824 if (params
->has_xbzrle_cache_size
) {
1825 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1826 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1828 if (params
->has_max_postcopy_bandwidth
) {
1829 s
->parameters
.max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1830 if (s
->to_dst_file
&& migration_in_postcopy()) {
1831 qemu_file_set_rate_limit(s
->to_dst_file
,
1832 s
->parameters
.max_postcopy_bandwidth
/ XFER_LIMIT_RATIO
);
1835 if (params
->has_max_cpu_throttle
) {
1836 s
->parameters
.max_cpu_throttle
= params
->max_cpu_throttle
;
1838 if (params
->has_announce_initial
) {
1839 s
->parameters
.announce_initial
= params
->announce_initial
;
1841 if (params
->has_announce_max
) {
1842 s
->parameters
.announce_max
= params
->announce_max
;
1844 if (params
->has_announce_rounds
) {
1845 s
->parameters
.announce_rounds
= params
->announce_rounds
;
1847 if (params
->has_announce_step
) {
1848 s
->parameters
.announce_step
= params
->announce_step
;
1851 if (params
->has_block_bitmap_mapping
) {
1852 qapi_free_BitmapMigrationNodeAliasList(
1853 s
->parameters
.block_bitmap_mapping
);
1855 s
->parameters
.has_block_bitmap_mapping
= true;
1856 s
->parameters
.block_bitmap_mapping
=
1857 QAPI_CLONE(BitmapMigrationNodeAliasList
,
1858 params
->block_bitmap_mapping
);
1862 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1864 MigrationParameters tmp
;
1866 /* TODO Rewrite "" to null instead */
1867 if (params
->tls_creds
1868 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1869 qobject_unref(params
->tls_creds
->u
.n
);
1870 params
->tls_creds
->type
= QTYPE_QSTRING
;
1871 params
->tls_creds
->u
.s
= strdup("");
1873 /* TODO Rewrite "" to null instead */
1874 if (params
->tls_hostname
1875 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1876 qobject_unref(params
->tls_hostname
->u
.n
);
1877 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1878 params
->tls_hostname
->u
.s
= strdup("");
1881 migrate_params_test_apply(params
, &tmp
);
1883 if (!migrate_params_check(&tmp
, errp
)) {
1884 /* Invalid parameter */
1888 migrate_params_apply(params
, errp
);
1892 void qmp_migrate_start_postcopy(Error
**errp
)
1894 MigrationState
*s
= migrate_get_current();
1896 if (!migrate_postcopy()) {
1897 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1898 " the start of migration");
1902 if (s
->state
== MIGRATION_STATUS_NONE
) {
1903 error_setg(errp
, "Postcopy must be started after migration has been"
1908 * we don't error if migration has finished since that would be racy
1909 * with issuing this command.
1911 qatomic_set(&s
->start_postcopy
, true);
1914 /* shared migration helpers */
1916 void migrate_set_state(int *state
, int old_state
, int new_state
)
1918 assert(new_state
< MIGRATION_STATUS__MAX
);
1919 if (qatomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1920 trace_migrate_set_state(MigrationStatus_str(new_state
));
1921 migrate_generate_event(new_state
);
1925 static MigrationCapabilityStatus
*migrate_cap_add(MigrationCapability index
,
1928 MigrationCapabilityStatus
*cap
;
1930 cap
= g_new0(MigrationCapabilityStatus
, 1);
1931 cap
->capability
= index
;
1937 void migrate_set_block_enabled(bool value
, Error
**errp
)
1939 MigrationCapabilityStatusList
*cap
= NULL
;
1941 QAPI_LIST_PREPEND(cap
, migrate_cap_add(MIGRATION_CAPABILITY_BLOCK
, value
));
1942 qmp_migrate_set_capabilities(cap
, errp
);
1943 qapi_free_MigrationCapabilityStatusList(cap
);
1946 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1948 s
->parameters
.block_incremental
= value
;
1951 static void block_cleanup_parameters(MigrationState
*s
)
1953 if (s
->must_remove_block_options
) {
1954 /* setting to false can never fail */
1955 migrate_set_block_enabled(false, &error_abort
);
1956 migrate_set_block_incremental(s
, false);
1957 s
->must_remove_block_options
= false;
1961 static void migrate_fd_cleanup(MigrationState
*s
)
1963 qemu_bh_delete(s
->cleanup_bh
);
1964 s
->cleanup_bh
= NULL
;
1966 g_free(s
->hostname
);
1968 json_writer_free(s
->vmdesc
);
1971 qemu_savevm_state_cleanup();
1973 if (s
->to_dst_file
) {
1976 trace_migrate_fd_cleanup();
1977 qemu_mutex_unlock_iothread();
1978 if (s
->migration_thread_running
) {
1979 qemu_thread_join(&s
->thread
);
1980 s
->migration_thread_running
= false;
1982 qemu_mutex_lock_iothread();
1984 multifd_save_cleanup();
1985 qemu_mutex_lock(&s
->qemu_file_lock
);
1986 tmp
= s
->to_dst_file
;
1987 s
->to_dst_file
= NULL
;
1988 qemu_mutex_unlock(&s
->qemu_file_lock
);
1990 * Close the file handle without the lock to make sure the
1991 * critical section won't block for long.
1993 migration_ioc_unregister_yank_from_file(tmp
);
1997 if (s
->postcopy_qemufile_src
) {
1998 migration_ioc_unregister_yank_from_file(s
->postcopy_qemufile_src
);
1999 qemu_fclose(s
->postcopy_qemufile_src
);
2000 s
->postcopy_qemufile_src
= NULL
;
2003 assert(!migration_is_active(s
));
2005 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
2006 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
2007 MIGRATION_STATUS_CANCELLED
);
2011 /* It is used on info migrate. We can't free it */
2012 error_report_err(error_copy(s
->error
));
2014 notifier_list_notify(&migration_state_notifiers
, s
);
2015 block_cleanup_parameters(s
);
2016 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2019 static void migrate_fd_cleanup_schedule(MigrationState
*s
)
2022 * Ref the state for bh, because it may be called when
2023 * there're already no other refs
2025 object_ref(OBJECT(s
));
2026 qemu_bh_schedule(s
->cleanup_bh
);
2029 static void migrate_fd_cleanup_bh(void *opaque
)
2031 MigrationState
*s
= opaque
;
2032 migrate_fd_cleanup(s
);
2033 object_unref(OBJECT(s
));
2036 void migrate_set_error(MigrationState
*s
, const Error
*error
)
2038 QEMU_LOCK_GUARD(&s
->error_mutex
);
2040 s
->error
= error_copy(error
);
2044 static void migrate_error_free(MigrationState
*s
)
2046 QEMU_LOCK_GUARD(&s
->error_mutex
);
2048 error_free(s
->error
);
2053 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
2055 trace_migrate_fd_error(error_get_pretty(error
));
2056 assert(s
->to_dst_file
== NULL
);
2057 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2058 MIGRATION_STATUS_FAILED
);
2059 migrate_set_error(s
, error
);
2062 static void migrate_fd_cancel(MigrationState
*s
)
2065 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
2066 trace_migrate_fd_cancel();
2068 WITH_QEMU_LOCK_GUARD(&s
->qemu_file_lock
) {
2069 if (s
->rp_state
.from_dst_file
) {
2070 /* shutdown the rp socket, so causing the rp thread to shutdown */
2071 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
2076 old_state
= s
->state
;
2077 if (!migration_is_running(old_state
)) {
2080 /* If the migration is paused, kick it out of the pause */
2081 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
2082 qemu_sem_post(&s
->pause_sem
);
2084 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
2085 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
2088 * If we're unlucky the migration code might be stuck somewhere in a
2089 * send/write while the network has failed and is waiting to timeout;
2090 * if we've got shutdown(2) available then we can force it to quit.
2091 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
2092 * called in a bh, so there is no race against this cancel.
2094 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
2095 qemu_file_shutdown(f
);
2097 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
2098 Error
*local_err
= NULL
;
2100 bdrv_activate_all(&local_err
);
2102 error_report_err(local_err
);
2104 s
->block_inactive
= false;
2109 void add_migration_state_change_notifier(Notifier
*notify
)
2111 notifier_list_add(&migration_state_notifiers
, notify
);
2114 void remove_migration_state_change_notifier(Notifier
*notify
)
2116 notifier_remove(notify
);
2119 bool migration_in_setup(MigrationState
*s
)
2121 return s
->state
== MIGRATION_STATUS_SETUP
;
2124 bool migration_has_finished(MigrationState
*s
)
2126 return s
->state
== MIGRATION_STATUS_COMPLETED
;
2129 bool migration_has_failed(MigrationState
*s
)
2131 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
2132 s
->state
== MIGRATION_STATUS_FAILED
);
2135 bool migration_in_postcopy(void)
2137 MigrationState
*s
= migrate_get_current();
2140 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
2141 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
2142 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
2149 bool migration_in_postcopy_after_devices(MigrationState
*s
)
2151 return migration_in_postcopy() && s
->postcopy_after_devices
;
2154 bool migration_in_incoming_postcopy(void)
2156 PostcopyState ps
= postcopy_state_get();
2158 return ps
>= POSTCOPY_INCOMING_DISCARD
&& ps
< POSTCOPY_INCOMING_END
;
2161 bool migration_incoming_postcopy_advised(void)
2163 PostcopyState ps
= postcopy_state_get();
2165 return ps
>= POSTCOPY_INCOMING_ADVISE
&& ps
< POSTCOPY_INCOMING_END
;
2168 bool migration_in_bg_snapshot(void)
2170 MigrationState
*s
= migrate_get_current();
2172 return migrate_background_snapshot() &&
2173 migration_is_setup_or_active(s
->state
);
2176 bool migration_is_idle(void)
2178 MigrationState
*s
= current_migration
;
2185 case MIGRATION_STATUS_NONE
:
2186 case MIGRATION_STATUS_CANCELLED
:
2187 case MIGRATION_STATUS_COMPLETED
:
2188 case MIGRATION_STATUS_FAILED
:
2190 case MIGRATION_STATUS_SETUP
:
2191 case MIGRATION_STATUS_CANCELLING
:
2192 case MIGRATION_STATUS_ACTIVE
:
2193 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
2194 case MIGRATION_STATUS_COLO
:
2195 case MIGRATION_STATUS_PRE_SWITCHOVER
:
2196 case MIGRATION_STATUS_DEVICE
:
2197 case MIGRATION_STATUS_WAIT_UNPLUG
:
2199 case MIGRATION_STATUS__MAX
:
2200 g_assert_not_reached();
2206 bool migration_is_active(MigrationState
*s
)
2208 return (s
->state
== MIGRATION_STATUS_ACTIVE
||
2209 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2212 void migrate_init(MigrationState
*s
)
2215 * Reinitialise all migration state, except
2216 * parameters/capabilities that the user set, and
2221 s
->to_dst_file
= NULL
;
2222 s
->state
= MIGRATION_STATUS_NONE
;
2223 s
->rp_state
.from_dst_file
= NULL
;
2224 s
->rp_state
.error
= false;
2226 s
->pages_per_second
= 0.0;
2228 s
->expected_downtime
= 0;
2230 s
->start_postcopy
= false;
2231 s
->postcopy_after_devices
= false;
2232 s
->migration_thread_running
= false;
2233 error_free(s
->error
);
2237 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
2239 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2241 s
->vm_was_running
= false;
2242 s
->iteration_initial_bytes
= 0;
2243 s
->threshold_size
= 0;
2246 int migrate_add_blocker_internal(Error
*reason
, Error
**errp
)
2248 /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */
2249 if (runstate_check(RUN_STATE_SAVE_VM
) || !migration_is_idle()) {
2250 error_propagate_prepend(errp
, error_copy(reason
),
2251 "disallowing migration blocker "
2252 "(migration/snapshot in progress) for: ");
2256 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
2260 int migrate_add_blocker(Error
*reason
, Error
**errp
)
2262 if (only_migratable
) {
2263 error_propagate_prepend(errp
, error_copy(reason
),
2264 "disallowing migration blocker "
2265 "(--only-migratable) for: ");
2269 return migrate_add_blocker_internal(reason
, errp
);
2272 void migrate_del_blocker(Error
*reason
)
2274 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
2277 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
2279 Error
*local_err
= NULL
;
2280 static bool once
= true;
2283 error_setg(errp
, "The incoming migration has already been started");
2286 if (!runstate_check(RUN_STATE_INMIGRATE
)) {
2287 error_setg(errp
, "'-incoming' was not specified on the command line");
2291 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
2295 qemu_start_incoming_migration(uri
, &local_err
);
2298 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2299 error_propagate(errp
, local_err
);
2306 void qmp_migrate_recover(const char *uri
, Error
**errp
)
2308 MigrationIncomingState
*mis
= migration_incoming_get_current();
2311 * Don't even bother to use ERRP_GUARD() as it _must_ always be set by
2312 * callers (no one should ignore a recover failure); if there is, it's a
2313 * programming error.
2317 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2318 error_setg(errp
, "Migrate recover can only be run "
2319 "when postcopy is paused.");
2323 /* If there's an existing transport, release it */
2324 migration_incoming_transport_cleanup(mis
);
2327 * Note that this call will never start a real migration; it will
2328 * only re-setup the migration stream and poke existing migration
2329 * to continue using that newly established channel.
2331 qemu_start_incoming_migration(uri
, errp
);
2334 void qmp_migrate_pause(Error
**errp
)
2336 MigrationState
*ms
= migrate_get_current();
2337 MigrationIncomingState
*mis
= migration_incoming_get_current();
2340 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2341 /* Source side, during postcopy */
2342 qemu_mutex_lock(&ms
->qemu_file_lock
);
2343 ret
= qemu_file_shutdown(ms
->to_dst_file
);
2344 qemu_mutex_unlock(&ms
->qemu_file_lock
);
2346 error_setg(errp
, "Failed to pause source migration");
2351 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2352 ret
= qemu_file_shutdown(mis
->from_src_file
);
2354 error_setg(errp
, "Failed to pause destination migration");
2359 error_setg(errp
, "migrate-pause is currently only supported "
2360 "during postcopy-active state");
2363 bool migration_is_blocked(Error
**errp
)
2365 if (qemu_savevm_state_blocked(errp
)) {
2369 if (migration_blockers
) {
2370 error_propagate(errp
, error_copy(migration_blockers
->data
));
2377 /* Returns true if continue to migrate, or false if error detected */
2378 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
2379 bool resume
, Error
**errp
)
2381 Error
*local_err
= NULL
;
2384 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2385 error_setg(errp
, "Cannot resume if there is no "
2386 "paused migration");
2391 * Postcopy recovery won't work well with release-ram
2392 * capability since release-ram will drop the page buffer as
2393 * long as the page is put into the send buffer. So if there
2394 * is a network failure happened, any page buffers that have
2395 * not yet reached the destination VM but have already been
2396 * sent from the source VM will be lost forever. Let's refuse
2397 * the client from resuming such a postcopy migration.
2398 * Luckily release-ram was designed to only be used when src
2399 * and destination VMs are on the same host, so it should be
2402 if (migrate_release_ram()) {
2403 error_setg(errp
, "Postcopy recovery cannot work "
2404 "when release-ram capability is set");
2408 /* This is a resume, skip init status */
2412 if (migration_is_running(s
->state
)) {
2413 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
2417 if (runstate_check(RUN_STATE_INMIGRATE
)) {
2418 error_setg(errp
, "Guest is waiting for an incoming migration");
2422 if (runstate_check(RUN_STATE_POSTMIGRATE
)) {
2423 error_setg(errp
, "Can't migrate the vm that was paused due to "
2424 "previous migration");
2428 if (migration_is_blocked(errp
)) {
2432 if (blk
|| blk_inc
) {
2433 if (migrate_colo_enabled()) {
2434 error_setg(errp
, "No disk migration is required in COLO mode");
2437 if (migrate_use_block() || migrate_use_block_incremental()) {
2438 error_setg(errp
, "Command options are incompatible with "
2439 "current migration capabilities");
2442 migrate_set_block_enabled(true, &local_err
);
2444 error_propagate(errp
, local_err
);
2447 s
->must_remove_block_options
= true;
2451 migrate_set_block_incremental(s
, true);
2456 * set ram_counters compression_counters memory to zero for a
2459 memset(&ram_counters
, 0, sizeof(ram_counters
));
2460 memset(&compression_counters
, 0, sizeof(compression_counters
));
2465 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
2466 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
2467 bool has_resume
, bool resume
, Error
**errp
)
2469 Error
*local_err
= NULL
;
2470 MigrationState
*s
= migrate_get_current();
2471 const char *p
= NULL
;
2473 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
2474 has_resume
&& resume
, errp
)) {
2475 /* Error detected, put into errp */
2479 if (!(has_resume
&& resume
)) {
2480 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
2485 migrate_protocol_allow_multi_channels(false);
2486 if (strstart(uri
, "tcp:", &p
) ||
2487 strstart(uri
, "unix:", NULL
) ||
2488 strstart(uri
, "vsock:", NULL
)) {
2489 migrate_protocol_allow_multi_channels(true);
2490 socket_start_outgoing_migration(s
, p
? p
: uri
, &local_err
);
2492 } else if (strstart(uri
, "rdma:", &p
)) {
2493 rdma_start_outgoing_migration(s
, p
, &local_err
);
2495 } else if (strstart(uri
, "exec:", &p
)) {
2496 exec_start_outgoing_migration(s
, p
, &local_err
);
2497 } else if (strstart(uri
, "fd:", &p
)) {
2498 fd_start_outgoing_migration(s
, p
, &local_err
);
2500 if (!(has_resume
&& resume
)) {
2501 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2503 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
2504 "a valid migration protocol");
2505 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2506 MIGRATION_STATUS_FAILED
);
2507 block_cleanup_parameters(s
);
2512 if (!(has_resume
&& resume
)) {
2513 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2515 migrate_fd_error(s
, local_err
);
2516 error_propagate(errp
, local_err
);
2521 void qmp_migrate_cancel(Error
**errp
)
2523 migration_cancel(NULL
);
2526 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
2528 MigrationState
*s
= migrate_get_current();
2529 if (s
->state
!= state
) {
2530 error_setg(errp
, "Migration not in expected state: %s",
2531 MigrationStatus_str(s
->state
));
2534 qemu_sem_post(&s
->pause_sem
);
2537 bool migrate_release_ram(void)
2541 s
= migrate_get_current();
2543 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
2546 bool migrate_postcopy_ram(void)
2550 s
= migrate_get_current();
2552 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
2555 bool migrate_postcopy(void)
2557 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2560 bool migrate_auto_converge(void)
2564 s
= migrate_get_current();
2566 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
2569 bool migrate_zero_blocks(void)
2573 s
= migrate_get_current();
2575 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
2578 bool migrate_postcopy_blocktime(void)
2582 s
= migrate_get_current();
2584 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
2587 bool migrate_use_compression(void)
2591 s
= migrate_get_current();
2593 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
2596 int migrate_compress_level(void)
2600 s
= migrate_get_current();
2602 return s
->parameters
.compress_level
;
2605 int migrate_compress_threads(void)
2609 s
= migrate_get_current();
2611 return s
->parameters
.compress_threads
;
2614 int migrate_compress_wait_thread(void)
2618 s
= migrate_get_current();
2620 return s
->parameters
.compress_wait_thread
;
2623 int migrate_decompress_threads(void)
2627 s
= migrate_get_current();
2629 return s
->parameters
.decompress_threads
;
2632 bool migrate_dirty_bitmaps(void)
2636 s
= migrate_get_current();
2638 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
2641 bool migrate_ignore_shared(void)
2645 s
= migrate_get_current();
2647 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
];
2650 bool migrate_validate_uuid(void)
2654 s
= migrate_get_current();
2656 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_VALIDATE_UUID
];
2659 bool migrate_use_events(void)
2663 s
= migrate_get_current();
2665 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
2668 bool migrate_use_multifd(void)
2672 s
= migrate_get_current();
2674 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_MULTIFD
];
2677 bool migrate_pause_before_switchover(void)
2681 s
= migrate_get_current();
2683 return s
->enabled_capabilities
[
2684 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
2687 int migrate_multifd_channels(void)
2691 s
= migrate_get_current();
2693 return s
->parameters
.multifd_channels
;
2696 MultiFDCompression
migrate_multifd_compression(void)
2700 s
= migrate_get_current();
2702 assert(s
->parameters
.multifd_compression
< MULTIFD_COMPRESSION__MAX
);
2703 return s
->parameters
.multifd_compression
;
2706 int migrate_multifd_zlib_level(void)
2710 s
= migrate_get_current();
2712 return s
->parameters
.multifd_zlib_level
;
2715 int migrate_multifd_zstd_level(void)
2719 s
= migrate_get_current();
2721 return s
->parameters
.multifd_zstd_level
;
2725 bool migrate_use_zero_copy_send(void)
2729 s
= migrate_get_current();
2731 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_COPY_SEND
];
2735 int migrate_use_tls(void)
2739 s
= migrate_get_current();
2741 return s
->parameters
.tls_creds
&& *s
->parameters
.tls_creds
;
2744 int migrate_use_xbzrle(void)
2748 s
= migrate_get_current();
2750 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
2753 uint64_t migrate_xbzrle_cache_size(void)
2757 s
= migrate_get_current();
2759 return s
->parameters
.xbzrle_cache_size
;
2762 static int64_t migrate_max_postcopy_bandwidth(void)
2766 s
= migrate_get_current();
2768 return s
->parameters
.max_postcopy_bandwidth
;
2771 bool migrate_use_block(void)
2775 s
= migrate_get_current();
2777 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
2780 bool migrate_use_return_path(void)
2784 s
= migrate_get_current();
2786 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
2789 bool migrate_use_block_incremental(void)
2793 s
= migrate_get_current();
2795 return s
->parameters
.block_incremental
;
2798 bool migrate_background_snapshot(void)
2802 s
= migrate_get_current();
2804 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
];
2807 bool migrate_postcopy_preempt(void)
2811 s
= migrate_get_current();
2813 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_PREEMPT
];
2816 /* migration thread support */
2818 * Something bad happened to the RP stream, mark an error
2819 * The caller shall print or trace something to indicate why
2821 static void mark_source_rp_bad(MigrationState
*s
)
2823 s
->rp_state
.error
= true;
2826 static struct rp_cmd_args
{
2827 ssize_t len
; /* -1 = variable */
2830 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
2831 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
2832 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
2833 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
2834 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
2835 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
2836 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
2837 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
2841 * Process a request for pages received on the return path,
2842 * We're allowed to send more than requested (e.g. to round to our page size)
2843 * and we don't need to send pages that have already been sent.
2845 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
2846 ram_addr_t start
, size_t len
)
2848 long our_host_ps
= qemu_real_host_page_size();
2850 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
2853 * Since we currently insist on matching page sizes, just sanity check
2854 * we're being asked for whole host pages.
2856 if (!QEMU_IS_ALIGNED(start
, our_host_ps
) ||
2857 !QEMU_IS_ALIGNED(len
, our_host_ps
)) {
2858 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2859 " len: %zd", __func__
, start
, len
);
2860 mark_source_rp_bad(ms
);
2864 if (ram_save_queue_pages(rbname
, start
, len
)) {
2865 mark_source_rp_bad(ms
);
2869 /* Return true to retry, false to quit */
2870 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
2872 trace_postcopy_pause_return_path();
2874 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
2876 trace_postcopy_pause_return_path_continued();
2881 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
2883 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
2886 error_report("%s: invalid block name '%s'", __func__
, block_name
);
2890 /* Fetch the received bitmap and refresh the dirty bitmap */
2891 return ram_dirty_bitmap_reload(s
, block
);
2894 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
2896 trace_source_return_path_thread_resume_ack(value
);
2898 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
2899 error_report("%s: illegal resume_ack value %"PRIu32
,
2904 /* Now both sides are active. */
2905 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2906 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2908 /* Notify send thread that time to continue send pages */
2909 qemu_sem_post(&s
->rp_state
.rp_sem
);
2915 * Release ms->rp_state.from_dst_file (and postcopy_qemufile_src if
2916 * existed) in a safe way.
2918 static void migration_release_dst_files(MigrationState
*ms
)
2922 WITH_QEMU_LOCK_GUARD(&ms
->qemu_file_lock
) {
2924 * Reset the from_dst_file pointer first before releasing it, as we
2925 * can't block within lock section
2927 file
= ms
->rp_state
.from_dst_file
;
2928 ms
->rp_state
.from_dst_file
= NULL
;
2932 * Do the same to postcopy fast path socket too if there is. No
2933 * locking needed because this qemufile should only be managed by
2934 * return path thread.
2936 if (ms
->postcopy_qemufile_src
) {
2937 migration_ioc_unregister_yank_from_file(ms
->postcopy_qemufile_src
);
2938 qemu_file_shutdown(ms
->postcopy_qemufile_src
);
2939 qemu_fclose(ms
->postcopy_qemufile_src
);
2940 ms
->postcopy_qemufile_src
= NULL
;
2947 * Handles messages sent on the return path towards the source VM
2950 static void *source_return_path_thread(void *opaque
)
2952 MigrationState
*ms
= opaque
;
2953 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
2954 uint16_t header_len
, header_type
;
2956 uint32_t tmp32
, sibling_error
;
2957 ram_addr_t start
= 0; /* =0 to silence warning */
2958 size_t len
= 0, expected_len
;
2961 trace_source_return_path_thread_entry();
2962 rcu_register_thread();
2965 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
2966 migration_is_setup_or_active(ms
->state
)) {
2967 trace_source_return_path_thread_loop_top();
2968 header_type
= qemu_get_be16(rp
);
2969 header_len
= qemu_get_be16(rp
);
2971 if (qemu_file_get_error(rp
)) {
2972 mark_source_rp_bad(ms
);
2976 if (header_type
>= MIG_RP_MSG_MAX
||
2977 header_type
== MIG_RP_MSG_INVALID
) {
2978 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2979 header_type
, header_len
);
2980 mark_source_rp_bad(ms
);
2984 if ((rp_cmd_args
[header_type
].len
!= -1 &&
2985 header_len
!= rp_cmd_args
[header_type
].len
) ||
2986 header_len
> sizeof(buf
)) {
2987 error_report("RP: Received '%s' message (0x%04x) with"
2988 "incorrect length %d expecting %zu",
2989 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
2990 (size_t)rp_cmd_args
[header_type
].len
);
2991 mark_source_rp_bad(ms
);
2995 /* We know we've got a valid header by this point */
2996 res
= qemu_get_buffer(rp
, buf
, header_len
);
2997 if (res
!= header_len
) {
2998 error_report("RP: Failed reading data for message 0x%04x"
2999 " read %d expected %d",
3000 header_type
, res
, header_len
);
3001 mark_source_rp_bad(ms
);
3005 /* OK, we have the message and the data */
3006 switch (header_type
) {
3007 case MIG_RP_MSG_SHUT
:
3008 sibling_error
= ldl_be_p(buf
);
3009 trace_source_return_path_thread_shut(sibling_error
);
3010 if (sibling_error
) {
3011 error_report("RP: Sibling indicated error %d", sibling_error
);
3012 mark_source_rp_bad(ms
);
3015 * We'll let the main thread deal with closing the RP
3016 * we could do a shutdown(2) on it, but we're the only user
3017 * anyway, so there's nothing gained.
3021 case MIG_RP_MSG_PONG
:
3022 tmp32
= ldl_be_p(buf
);
3023 trace_source_return_path_thread_pong(tmp32
);
3026 case MIG_RP_MSG_REQ_PAGES
:
3027 start
= ldq_be_p(buf
);
3028 len
= ldl_be_p(buf
+ 8);
3029 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
3032 case MIG_RP_MSG_REQ_PAGES_ID
:
3033 expected_len
= 12 + 1; /* header + termination */
3035 if (header_len
>= expected_len
) {
3036 start
= ldq_be_p(buf
);
3037 len
= ldl_be_p(buf
+ 8);
3038 /* Now we expect an idstr */
3039 tmp32
= buf
[12]; /* Length of the following idstr */
3040 buf
[13 + tmp32
] = '\0';
3041 expected_len
+= tmp32
;
3043 if (header_len
!= expected_len
) {
3044 error_report("RP: Req_Page_id with length %d expecting %zd",
3045 header_len
, expected_len
);
3046 mark_source_rp_bad(ms
);
3049 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
3052 case MIG_RP_MSG_RECV_BITMAP
:
3053 if (header_len
< 1) {
3054 error_report("%s: missing block name", __func__
);
3055 mark_source_rp_bad(ms
);
3058 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
3059 buf
[buf
[0] + 1] = '\0';
3060 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
3061 mark_source_rp_bad(ms
);
3066 case MIG_RP_MSG_RESUME_ACK
:
3067 tmp32
= ldl_be_p(buf
);
3068 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
3069 mark_source_rp_bad(ms
);
3080 res
= qemu_file_get_error(rp
);
3082 if (res
&& migration_in_postcopy()) {
3084 * Maybe there is something we can do: it looks like a
3085 * network down issue, and we pause for a recovery.
3087 migration_release_dst_files(ms
);
3089 if (postcopy_pause_return_path_thread(ms
)) {
3091 * Reload rp, reset the rest. Referencing it is safe since
3092 * it's reset only by us above, or when migration completes
3094 rp
= ms
->rp_state
.from_dst_file
;
3095 ms
->rp_state
.error
= false;
3100 trace_source_return_path_thread_bad_end();
3101 mark_source_rp_bad(ms
);
3104 trace_source_return_path_thread_end();
3105 migration_release_dst_files(ms
);
3106 rcu_unregister_thread();
3110 static int open_return_path_on_source(MigrationState
*ms
,
3113 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
3114 if (!ms
->rp_state
.from_dst_file
) {
3118 trace_open_return_path_on_source();
3120 if (!create_thread
) {
3125 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
3126 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
3127 ms
->rp_state
.rp_thread_created
= true;
3129 trace_open_return_path_on_source_continue();
3134 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
3135 static int await_return_path_close_on_source(MigrationState
*ms
)
3138 * If this is a normal exit then the destination will send a SHUT and the
3139 * rp_thread will exit, however if there's an error we need to cause
3142 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
3144 * shutdown(2), if we have it, will cause it to unblock if it's stuck
3145 * waiting for the destination.
3147 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
3148 mark_source_rp_bad(ms
);
3150 trace_await_return_path_close_on_source_joining();
3151 qemu_thread_join(&ms
->rp_state
.rp_thread
);
3152 ms
->rp_state
.rp_thread_created
= false;
3153 trace_await_return_path_close_on_source_close();
3154 return ms
->rp_state
.error
;
3158 * Switch from normal iteration to postcopy
3159 * Returns non-0 on error
3161 static int postcopy_start(MigrationState
*ms
)
3164 QIOChannelBuffer
*bioc
;
3166 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3167 int64_t bandwidth
= migrate_max_postcopy_bandwidth();
3168 bool restart_block
= false;
3169 int cur_state
= MIGRATION_STATUS_ACTIVE
;
3171 if (postcopy_preempt_wait_channel(ms
)) {
3172 migrate_set_state(&ms
->state
, ms
->state
, MIGRATION_STATUS_FAILED
);
3176 if (!migrate_pause_before_switchover()) {
3177 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
3178 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3181 trace_postcopy_start();
3182 qemu_mutex_lock_iothread();
3183 trace_postcopy_start_set_run();
3185 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3186 global_state_store();
3187 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
3192 ret
= migration_maybe_pause(ms
, &cur_state
,
3193 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3198 ret
= bdrv_inactivate_all();
3202 restart_block
= true;
3205 * Cause any non-postcopiable, but iterative devices to
3206 * send out their final data.
3208 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
3211 * in Finish migrate and with the io-lock held everything should
3212 * be quiet, but we've potentially still got dirty pages and we
3213 * need to tell the destination to throw any pages it's already received
3216 if (migrate_postcopy_ram()) {
3217 ram_postcopy_send_discard_bitmap(ms
);
3221 * send rest of state - note things that are doing postcopy
3222 * will notice we're in POSTCOPY_ACTIVE and not actually
3223 * wrap their state up here
3225 /* 0 max-postcopy-bandwidth means unlimited */
3227 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
3229 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
3231 if (migrate_postcopy_ram()) {
3232 /* Ping just for debugging, helps line traces up */
3233 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
3237 * While loading the device state we may trigger page transfer
3238 * requests and the fd must be free to process those, and thus
3239 * the destination must read the whole device state off the fd before
3240 * it starts processing it. Unfortunately the ad-hoc migration format
3241 * doesn't allow the destination to know the size to read without fully
3242 * parsing it through each devices load-state code (especially the open
3243 * coded devices that use get/put).
3244 * So we wrap the device state up in a package with a length at the start;
3245 * to do this we use a qemu_buf to hold the whole of the device state.
3247 bioc
= qio_channel_buffer_new(4096);
3248 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
3249 fb
= qemu_file_new_output(QIO_CHANNEL(bioc
));
3250 object_unref(OBJECT(bioc
));
3253 * Make sure the receiver can get incoming pages before we send the rest
3256 qemu_savevm_send_postcopy_listen(fb
);
3258 qemu_savevm_state_complete_precopy(fb
, false, false);
3259 if (migrate_postcopy_ram()) {
3260 qemu_savevm_send_ping(fb
, 3);
3263 qemu_savevm_send_postcopy_run(fb
);
3265 /* <><> end of stuff going into the package */
3267 /* Last point of recovery; as soon as we send the package the destination
3268 * can open devices and potentially start running.
3269 * Lets just check again we've not got any errors.
3271 ret
= qemu_file_get_error(ms
->to_dst_file
);
3273 error_report("postcopy_start: Migration stream errored (pre package)");
3277 restart_block
= false;
3279 /* Now send that blob */
3280 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
3285 /* Send a notify to give a chance for anything that needs to happen
3286 * at the transition to postcopy and after the device state; in particular
3287 * spice needs to trigger a transition now
3289 ms
->postcopy_after_devices
= true;
3290 notifier_list_notify(&migration_state_notifiers
, ms
);
3292 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
3294 qemu_mutex_unlock_iothread();
3296 if (migrate_postcopy_ram()) {
3298 * Although this ping is just for debug, it could potentially be
3299 * used for getting a better measurement of downtime at the source.
3301 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
3304 if (migrate_release_ram()) {
3305 ram_postcopy_migrated_memory_release(ms
);
3308 ret
= qemu_file_get_error(ms
->to_dst_file
);
3310 error_report("postcopy_start: Migration stream errored");
3311 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
3312 MIGRATION_STATUS_FAILED
);
3315 trace_postcopy_preempt_enabled(migrate_postcopy_preempt());
3322 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
3323 MIGRATION_STATUS_FAILED
);
3324 if (restart_block
) {
3325 /* A failure happened early enough that we know the destination hasn't
3326 * accessed block devices, so we're safe to recover.
3328 Error
*local_err
= NULL
;
3330 bdrv_activate_all(&local_err
);
3332 error_report_err(local_err
);
3335 qemu_mutex_unlock_iothread();
3340 * migration_maybe_pause: Pause if required to by
3341 * migrate_pause_before_switchover called with the iothread locked
3342 * Returns: 0 on success
3344 static int migration_maybe_pause(MigrationState
*s
,
3345 int *current_active_state
,
3348 if (!migrate_pause_before_switchover()) {
3352 /* Since leaving this state is not atomic with posting the semaphore
3353 * it's possible that someone could have issued multiple migrate_continue
3354 * and the semaphore is incorrectly positive at this point;
3355 * the docs say it's undefined to reinit a semaphore that's already
3356 * init'd, so use timedwait to eat up any existing posts.
3358 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
3359 /* This block intentionally left blank */
3363 * If the migration is cancelled when it is in the completion phase,
3364 * the migration state is set to MIGRATION_STATUS_CANCELLING.
3365 * So we don't need to wait a semaphore, otherwise we would always
3366 * wait for the 'pause_sem' semaphore.
3368 if (s
->state
!= MIGRATION_STATUS_CANCELLING
) {
3369 qemu_mutex_unlock_iothread();
3370 migrate_set_state(&s
->state
, *current_active_state
,
3371 MIGRATION_STATUS_PRE_SWITCHOVER
);
3372 qemu_sem_wait(&s
->pause_sem
);
3373 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
3375 *current_active_state
= new_state
;
3376 qemu_mutex_lock_iothread();
3379 return s
->state
== new_state
? 0 : -EINVAL
;
3383 * migration_completion: Used by migration_thread when there's not much left.
3384 * The caller 'breaks' the loop when this returns.
3386 * @s: Current migration state
3388 static void migration_completion(MigrationState
*s
)
3391 int current_active_state
= s
->state
;
3393 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
3394 qemu_mutex_lock_iothread();
3395 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3396 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3397 s
->vm_was_running
= runstate_is_running();
3398 ret
= global_state_store();
3401 bool inactivate
= !migrate_colo_enabled();
3402 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
3403 trace_migration_completion_vm_stop(ret
);
3405 ret
= migration_maybe_pause(s
, ¤t_active_state
,
3406 MIGRATION_STATUS_DEVICE
);
3409 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
3410 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
3413 if (inactivate
&& ret
>= 0) {
3414 s
->block_inactive
= true;
3417 qemu_mutex_unlock_iothread();
3422 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3423 trace_migration_completion_postcopy_end();
3425 qemu_mutex_lock_iothread();
3426 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
3427 qemu_mutex_unlock_iothread();
3429 /* Shutdown the postcopy fast path thread */
3430 if (migrate_postcopy_preempt()) {
3431 postcopy_preempt_shutdown_file(s
);
3434 trace_migration_completion_postcopy_end_after_complete();
3440 * If rp was opened we must clean up the thread before
3441 * cleaning everything else up (since if there are no failures
3442 * it will wait for the destination to send it's status in
3445 if (s
->rp_state
.rp_thread_created
) {
3447 trace_migration_return_path_end_before();
3448 rp_error
= await_return_path_close_on_source(s
);
3449 trace_migration_return_path_end_after(rp_error
);
3451 goto fail_invalidate
;
3455 if (qemu_file_get_error(s
->to_dst_file
)) {
3456 trace_migration_completion_file_err();
3457 goto fail_invalidate
;
3460 if (migrate_colo_enabled() && s
->state
== MIGRATION_STATUS_ACTIVE
) {
3461 /* COLO does not support postcopy */
3462 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
3463 MIGRATION_STATUS_COLO
);
3465 migrate_set_state(&s
->state
, current_active_state
,
3466 MIGRATION_STATUS_COMPLETED
);
3472 /* If not doing postcopy, vm_start() will be called: let's regain
3473 * control on images.
3475 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
3476 s
->state
== MIGRATION_STATUS_DEVICE
) {
3477 Error
*local_err
= NULL
;
3479 qemu_mutex_lock_iothread();
3480 bdrv_activate_all(&local_err
);
3482 error_report_err(local_err
);
3484 s
->block_inactive
= false;
3486 qemu_mutex_unlock_iothread();
3490 migrate_set_state(&s
->state
, current_active_state
,
3491 MIGRATION_STATUS_FAILED
);
3495 * bg_migration_completion: Used by bg_migration_thread when after all the
3496 * RAM has been saved. The caller 'breaks' the loop when this returns.
3498 * @s: Current migration state
3500 static void bg_migration_completion(MigrationState
*s
)
3502 int current_active_state
= s
->state
;
3505 * Stop tracking RAM writes - un-protect memory, un-register UFFD
3506 * memory ranges, flush kernel wait queues and wake up threads
3507 * waiting for write fault to be resolved.
3509 ram_write_tracking_stop();
3511 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
3513 * By this moment we have RAM content saved into the migration stream.
3514 * The next step is to flush the non-RAM content (device state)
3515 * right after the ram content. The device state has been stored into
3516 * the temporary buffer before RAM saving started.
3518 qemu_put_buffer(s
->to_dst_file
, s
->bioc
->data
, s
->bioc
->usage
);
3519 qemu_fflush(s
->to_dst_file
);
3520 } else if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
3524 if (qemu_file_get_error(s
->to_dst_file
)) {
3525 trace_migration_completion_file_err();
3529 migrate_set_state(&s
->state
, current_active_state
,
3530 MIGRATION_STATUS_COMPLETED
);
3534 migrate_set_state(&s
->state
, current_active_state
,
3535 MIGRATION_STATUS_FAILED
);
3538 bool migrate_colo_enabled(void)
3540 MigrationState
*s
= migrate_get_current();
3541 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
3544 typedef enum MigThrError
{
3545 /* No error detected */
3546 MIG_THR_ERR_NONE
= 0,
3547 /* Detected error, but resumed successfully */
3548 MIG_THR_ERR_RECOVERED
= 1,
3549 /* Detected fatal error, need to exit */
3550 MIG_THR_ERR_FATAL
= 2,
3553 static int postcopy_resume_handshake(MigrationState
*s
)
3555 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
3557 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3558 qemu_sem_wait(&s
->rp_state
.rp_sem
);
3561 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3568 /* Return zero if success, or <0 for error */
3569 static int postcopy_do_resume(MigrationState
*s
)
3574 * Call all the resume_prepare() hooks, so that modules can be
3575 * ready for the migration resume.
3577 ret
= qemu_savevm_state_resume_prepare(s
);
3579 error_report("%s: resume_prepare() failure detected: %d",
3585 * Last handshake with destination on the resume (destination will
3586 * switch to postcopy-active afterwards)
3588 ret
= postcopy_resume_handshake(s
);
3590 error_report("%s: handshake failed: %d", __func__
, ret
);
3598 * We don't return until we are in a safe state to continue current
3599 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
3600 * MIG_THR_ERR_FATAL if unrecovery failure happened.
3602 static MigThrError
postcopy_pause(MigrationState
*s
)
3604 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3610 * Current channel is possibly broken. Release it. Note that this is
3611 * guaranteed even without lock because to_dst_file should only be
3612 * modified by the migration thread. That also guarantees that the
3613 * unregister of yank is safe too without the lock. It should be safe
3614 * even to be within the qemu_file_lock, but we didn't do that to avoid
3615 * taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make
3616 * the qemu_file_lock critical section as small as possible.
3618 assert(s
->to_dst_file
);
3619 migration_ioc_unregister_yank_from_file(s
->to_dst_file
);
3620 qemu_mutex_lock(&s
->qemu_file_lock
);
3621 file
= s
->to_dst_file
;
3622 s
->to_dst_file
= NULL
;
3623 qemu_mutex_unlock(&s
->qemu_file_lock
);
3625 qemu_file_shutdown(file
);
3628 migrate_set_state(&s
->state
, s
->state
,
3629 MIGRATION_STATUS_POSTCOPY_PAUSED
);
3631 error_report("Detected IO failure for postcopy. "
3632 "Migration paused.");
3635 * We wait until things fixed up. Then someone will setup the
3636 * status back for us.
3638 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
3639 qemu_sem_wait(&s
->postcopy_pause_sem
);
3642 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3643 /* Woken up by a recover procedure. Give it a shot */
3645 if (postcopy_preempt_wait_channel(s
)) {
3647 * Preempt enabled, and new channel create failed; loop
3648 * back to wait for another recovery.
3654 * Firstly, let's wake up the return path now, with a new
3655 * return path channel.
3657 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
3659 /* Do the resume logic */
3660 if (postcopy_do_resume(s
) == 0) {
3661 /* Let's continue! */
3662 trace_postcopy_pause_continued();
3663 return MIG_THR_ERR_RECOVERED
;
3666 * Something wrong happened during the recovery, let's
3667 * pause again. Pause is always better than throwing
3673 /* This is not right... Time to quit. */
3674 return MIG_THR_ERR_FATAL
;
3679 static MigThrError
migration_detect_error(MigrationState
*s
)
3682 int state
= s
->state
;
3683 Error
*local_error
= NULL
;
3685 if (state
== MIGRATION_STATUS_CANCELLING
||
3686 state
== MIGRATION_STATUS_CANCELLED
) {
3687 /* End the migration, but don't set the state to failed */
3688 return MIG_THR_ERR_FATAL
;
3692 * Try to detect any file errors. Note that postcopy_qemufile_src will
3693 * be NULL when postcopy preempt is not enabled.
3695 ret
= qemu_file_get_error_obj_any(s
->to_dst_file
,
3696 s
->postcopy_qemufile_src
,
3699 /* Everything is fine */
3700 assert(!local_error
);
3701 return MIG_THR_ERR_NONE
;
3705 migrate_set_error(s
, local_error
);
3706 error_free(local_error
);
3709 if (state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
) {
3711 * For postcopy, we allow the network to be down for a
3712 * while. After that, it can be continued by a
3715 return postcopy_pause(s
);
3718 * For precopy (or postcopy with error outside IO), we fail
3721 migrate_set_state(&s
->state
, state
, MIGRATION_STATUS_FAILED
);
3722 trace_migration_thread_file_err();
3724 /* Time to stop the migration, now. */
3725 return MIG_THR_ERR_FATAL
;
3729 /* How many bytes have we transferred since the beginning of the migration */
3730 static uint64_t migration_total_bytes(MigrationState
*s
)
3732 return qemu_file_total_transferred(s
->to_dst_file
) +
3733 ram_counters
.multifd_bytes
;
3736 static void migration_calculate_complete(MigrationState
*s
)
3738 uint64_t bytes
= migration_total_bytes(s
);
3739 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3740 int64_t transfer_time
;
3742 s
->total_time
= end_time
- s
->start_time
;
3745 * It's still not set, so we are precopy migration. For
3746 * postcopy, downtime is calculated during postcopy_start().
3748 s
->downtime
= end_time
- s
->downtime_start
;
3751 transfer_time
= s
->total_time
- s
->setup_time
;
3752 if (transfer_time
) {
3753 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
3757 static void update_iteration_initial_status(MigrationState
*s
)
3760 * Update these three fields at the same time to avoid mismatch info lead
3761 * wrong speed calculation.
3763 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3764 s
->iteration_initial_bytes
= migration_total_bytes(s
);
3765 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
3768 static void migration_update_counters(MigrationState
*s
,
3769 int64_t current_time
)
3771 uint64_t transferred
, transferred_pages
, time_spent
;
3772 uint64_t current_bytes
; /* bytes transferred since the beginning */
3775 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
3779 current_bytes
= migration_total_bytes(s
);
3780 transferred
= current_bytes
- s
->iteration_initial_bytes
;
3781 time_spent
= current_time
- s
->iteration_start_time
;
3782 bandwidth
= (double)transferred
/ time_spent
;
3783 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
3785 s
->mbps
= (((double) transferred
* 8.0) /
3786 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
3788 transferred_pages
= ram_get_total_transferred_pages() -
3789 s
->iteration_initial_pages
;
3790 s
->pages_per_second
= (double) transferred_pages
/
3791 (((double) time_spent
/ 1000.0));
3794 * if we haven't sent anything, we don't want to
3795 * recalculate. 10000 is a small enough number for our purposes
3797 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
3798 s
->expected_downtime
= ram_counters
.remaining
/ bandwidth
;
3801 qemu_file_reset_rate_limit(s
->to_dst_file
);
3803 update_iteration_initial_status(s
);
3805 trace_migrate_transferred(transferred
, time_spent
,
3806 bandwidth
, s
->threshold_size
);
3809 /* Migration thread iteration status */
3811 MIG_ITERATE_RESUME
, /* Resume current iteration */
3812 MIG_ITERATE_SKIP
, /* Skip current iteration */
3813 MIG_ITERATE_BREAK
, /* Break the loop */
3817 * Return true if continue to the next iteration directly, false
3820 static MigIterateState
migration_iteration_run(MigrationState
*s
)
3822 uint64_t pend_pre
, pend_compat
, pend_post
;
3823 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
3825 qemu_savevm_state_pending_estimate(&pend_pre
, &pend_compat
, &pend_post
);
3826 uint64_t pending_size
= pend_pre
+ pend_compat
+ pend_post
;
3828 trace_migrate_pending_estimate(pending_size
,
3829 pend_pre
, pend_compat
, pend_post
);
3831 if (pend_pre
+ pend_compat
<= s
->threshold_size
) {
3832 qemu_savevm_state_pending_exact(&pend_pre
, &pend_compat
, &pend_post
);
3833 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
3834 trace_migrate_pending_exact(pending_size
,
3835 pend_pre
, pend_compat
, pend_post
);
3838 if (!pending_size
|| pending_size
< s
->threshold_size
) {
3839 trace_migration_thread_low_pending(pending_size
);
3840 migration_completion(s
);
3841 return MIG_ITERATE_BREAK
;
3844 /* Still a significant amount to transfer */
3845 if (!in_postcopy
&& pend_pre
<= s
->threshold_size
&&
3846 qatomic_read(&s
->start_postcopy
)) {
3847 if (postcopy_start(s
)) {
3848 error_report("%s: postcopy failed to start", __func__
);
3850 return MIG_ITERATE_SKIP
;
3853 /* Just another iteration step */
3854 qemu_savevm_state_iterate(s
->to_dst_file
, in_postcopy
);
3855 return MIG_ITERATE_RESUME
;
3858 static void migration_iteration_finish(MigrationState
*s
)
3860 /* If we enabled cpu throttling for auto-converge, turn it off. */
3861 cpu_throttle_stop();
3863 qemu_mutex_lock_iothread();
3865 case MIGRATION_STATUS_COMPLETED
:
3866 migration_calculate_complete(s
);
3867 runstate_set(RUN_STATE_POSTMIGRATE
);
3869 case MIGRATION_STATUS_COLO
:
3870 if (!migrate_colo_enabled()) {
3871 error_report("%s: critical error: calling COLO code without "
3872 "COLO enabled", __func__
);
3874 migrate_start_colo_process(s
);
3875 s
->vm_was_running
= true;
3877 case MIGRATION_STATUS_FAILED
:
3878 case MIGRATION_STATUS_CANCELLED
:
3879 case MIGRATION_STATUS_CANCELLING
:
3880 if (s
->vm_was_running
) {
3881 if (!runstate_check(RUN_STATE_SHUTDOWN
)) {
3885 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
3886 runstate_set(RUN_STATE_POSTMIGRATE
);
3892 /* Should not reach here, but if so, forgive the VM. */
3893 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3896 migrate_fd_cleanup_schedule(s
);
3897 qemu_mutex_unlock_iothread();
3900 static void bg_migration_iteration_finish(MigrationState
*s
)
3902 qemu_mutex_lock_iothread();
3904 case MIGRATION_STATUS_COMPLETED
:
3905 migration_calculate_complete(s
);
3908 case MIGRATION_STATUS_ACTIVE
:
3909 case MIGRATION_STATUS_FAILED
:
3910 case MIGRATION_STATUS_CANCELLED
:
3911 case MIGRATION_STATUS_CANCELLING
:
3915 /* Should not reach here, but if so, forgive the VM. */
3916 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3920 migrate_fd_cleanup_schedule(s
);
3921 qemu_mutex_unlock_iothread();
3925 * Return true if continue to the next iteration directly, false
3928 static MigIterateState
bg_migration_iteration_run(MigrationState
*s
)
3932 res
= qemu_savevm_state_iterate(s
->to_dst_file
, false);
3934 bg_migration_completion(s
);
3935 return MIG_ITERATE_BREAK
;
3938 return MIG_ITERATE_RESUME
;
3941 void migration_make_urgent_request(void)
3943 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
3946 void migration_consume_urgent_request(void)
3948 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
3951 /* Returns true if the rate limiting was broken by an urgent request */
3952 bool migration_rate_limit(void)
3954 int64_t now
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3955 MigrationState
*s
= migrate_get_current();
3957 bool urgent
= false;
3958 migration_update_counters(s
, now
);
3959 if (qemu_file_rate_limit(s
->to_dst_file
)) {
3961 if (qemu_file_get_error(s
->to_dst_file
)) {
3965 * Wait for a delay to do rate limiting OR
3966 * something urgent to post the semaphore.
3968 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- now
;
3969 trace_migration_rate_limit_pre(ms
);
3970 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
3972 * We were woken by one or more urgent things but
3973 * the timedwait will have consumed one of them.
3974 * The service routine for the urgent wake will dec
3975 * the semaphore itself for each item it consumes,
3976 * so add this one we just eat back.
3978 qemu_sem_post(&s
->rate_limit_sem
);
3981 trace_migration_rate_limit_post(urgent
);
3987 * if failover devices are present, wait they are completely
3991 static void qemu_savevm_wait_unplug(MigrationState
*s
, int old_state
,
3994 if (qemu_savevm_state_guest_unplug_pending()) {
3995 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_WAIT_UNPLUG
);
3997 while (s
->state
== MIGRATION_STATUS_WAIT_UNPLUG
&&
3998 qemu_savevm_state_guest_unplug_pending()) {
3999 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
4001 if (s
->state
!= MIGRATION_STATUS_WAIT_UNPLUG
) {
4002 int timeout
= 120; /* 30 seconds */
4004 * migration has been canceled
4005 * but as we have started an unplug we must wait the end
4006 * to be able to plug back the card
4008 while (timeout
-- && qemu_savevm_state_guest_unplug_pending()) {
4009 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
4011 if (qemu_savevm_state_guest_unplug_pending() &&
4013 warn_report("migration: partially unplugged device on "
4018 migrate_set_state(&s
->state
, MIGRATION_STATUS_WAIT_UNPLUG
, new_state
);
4020 migrate_set_state(&s
->state
, old_state
, new_state
);
4025 * Master migration thread on the source VM.
4026 * It drives the migration and pumps the data down the outgoing channel.
4028 static void *migration_thread(void *opaque
)
4030 MigrationState
*s
= opaque
;
4031 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
4032 MigThrError thr_error
;
4033 bool urgent
= false;
4035 rcu_register_thread();
4037 object_ref(OBJECT(s
));
4038 update_iteration_initial_status(s
);
4040 qemu_savevm_state_header(s
->to_dst_file
);
4043 * If we opened the return path, we need to make sure dst has it
4046 if (s
->rp_state
.rp_thread_created
) {
4047 /* Now tell the dest that it should open its end so it can reply */
4048 qemu_savevm_send_open_return_path(s
->to_dst_file
);
4050 /* And do a ping that will make stuff easier to debug */
4051 qemu_savevm_send_ping(s
->to_dst_file
, 1);
4054 if (migrate_postcopy()) {
4056 * Tell the destination that we *might* want to do postcopy later;
4057 * if the other end can't do postcopy it should fail now, nice and
4060 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
4063 if (migrate_colo_enabled()) {
4064 /* Notify migration destination that we enable COLO */
4065 qemu_savevm_send_colo_enable(s
->to_dst_file
);
4068 qemu_savevm_state_setup(s
->to_dst_file
);
4070 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
4071 MIGRATION_STATUS_ACTIVE
);
4073 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
4075 trace_migration_thread_setup_complete();
4077 while (migration_is_active(s
)) {
4078 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
4079 MigIterateState iter_state
= migration_iteration_run(s
);
4080 if (iter_state
== MIG_ITERATE_SKIP
) {
4082 } else if (iter_state
== MIG_ITERATE_BREAK
) {
4088 * Try to detect any kind of failures, and see whether we
4089 * should stop the migration now.
4091 thr_error
= migration_detect_error(s
);
4092 if (thr_error
== MIG_THR_ERR_FATAL
) {
4093 /* Stop migration */
4095 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
4097 * Just recovered from a e.g. network failure, reset all
4098 * the local variables. This is important to avoid
4099 * breaking transferred_bytes and bandwidth calculation
4101 update_iteration_initial_status(s
);
4104 urgent
= migration_rate_limit();
4107 trace_migration_thread_after_loop();
4108 migration_iteration_finish(s
);
4109 object_unref(OBJECT(s
));
4110 rcu_unregister_thread();
4114 static void bg_migration_vm_start_bh(void *opaque
)
4116 MigrationState
*s
= opaque
;
4118 qemu_bh_delete(s
->vm_start_bh
);
4119 s
->vm_start_bh
= NULL
;
4122 s
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - s
->downtime_start
;
4126 * Background snapshot thread, based on live migration code.
4127 * This is an alternative implementation of live migration mechanism
4128 * introduced specifically to support background snapshots.
4130 * It takes advantage of userfault_fd write protection mechanism introduced
4131 * in v5.7 kernel. Compared to existing dirty page logging migration much
4132 * lesser stream traffic is produced resulting in smaller snapshot images,
4133 * simply cause of no page duplicates can get into the stream.
4135 * Another key point is that generated vmstate stream reflects machine state
4136 * 'frozen' at the beginning of snapshot creation compared to dirty page logging
4137 * mechanism, which effectively results in that saved snapshot is the state of VM
4138 * at the end of the process.
4140 static void *bg_migration_thread(void *opaque
)
4142 MigrationState
*s
= opaque
;
4143 int64_t setup_start
;
4144 MigThrError thr_error
;
4146 bool early_fail
= true;
4148 rcu_register_thread();
4149 object_ref(OBJECT(s
));
4151 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
4153 setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
4155 * We want to save vmstate for the moment when migration has been
4156 * initiated but also we want to save RAM content while VM is running.
4157 * The RAM content should appear first in the vmstate. So, we first
4158 * stash the non-RAM part of the vmstate to the temporary buffer,
4159 * then write RAM part of the vmstate to the migration stream
4160 * with vCPUs running and, finally, write stashed non-RAM part of
4161 * the vmstate from the buffer to the migration stream.
4163 s
->bioc
= qio_channel_buffer_new(512 * 1024);
4164 qio_channel_set_name(QIO_CHANNEL(s
->bioc
), "vmstate-buffer");
4165 fb
= qemu_file_new_output(QIO_CHANNEL(s
->bioc
));
4166 object_unref(OBJECT(s
->bioc
));
4168 update_iteration_initial_status(s
);
4171 * Prepare for tracking memory writes with UFFD-WP - populate
4172 * RAM pages before protecting.
4175 ram_write_tracking_prepare();
4178 qemu_savevm_state_header(s
->to_dst_file
);
4179 qemu_savevm_state_setup(s
->to_dst_file
);
4181 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
4182 MIGRATION_STATUS_ACTIVE
);
4184 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
4186 trace_migration_thread_setup_complete();
4187 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
4189 qemu_mutex_lock_iothread();
4192 * If VM is currently in suspended state, then, to make a valid runstate
4193 * transition in vm_stop_force_state() we need to wakeup it up.
4195 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
4196 s
->vm_was_running
= runstate_is_running();
4198 if (global_state_store()) {
4201 /* Forcibly stop VM before saving state of vCPUs and devices */
4202 if (vm_stop_force_state(RUN_STATE_PAUSED
)) {
4206 * Put vCPUs in sync with shadow context structures, then
4207 * save their state to channel-buffer along with devices.
4209 cpu_synchronize_all_states();
4210 if (qemu_savevm_state_complete_precopy_non_iterable(fb
, false, false)) {
4214 * Since we are going to get non-iterable state data directly
4215 * from s->bioc->data, explicit flush is needed here.
4219 /* Now initialize UFFD context and start tracking RAM writes */
4220 if (ram_write_tracking_start()) {
4226 * Start VM from BH handler to avoid write-fault lock here.
4227 * UFFD-WP protection for the whole RAM is already enabled so
4228 * calling VM state change notifiers from vm_start() would initiate
4229 * writes to virtio VQs memory which is in write-protected region.
4231 s
->vm_start_bh
= qemu_bh_new(bg_migration_vm_start_bh
, s
);
4232 qemu_bh_schedule(s
->vm_start_bh
);
4234 qemu_mutex_unlock_iothread();
4236 while (migration_is_active(s
)) {
4237 MigIterateState iter_state
= bg_migration_iteration_run(s
);
4238 if (iter_state
== MIG_ITERATE_SKIP
) {
4240 } else if (iter_state
== MIG_ITERATE_BREAK
) {
4245 * Try to detect any kind of failures, and see whether we
4246 * should stop the migration now.
4248 thr_error
= migration_detect_error(s
);
4249 if (thr_error
== MIG_THR_ERR_FATAL
) {
4250 /* Stop migration */
4254 migration_update_counters(s
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
));
4257 trace_migration_thread_after_loop();
4261 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
4262 MIGRATION_STATUS_FAILED
);
4263 qemu_mutex_unlock_iothread();
4266 bg_migration_iteration_finish(s
);
4269 object_unref(OBJECT(s
));
4270 rcu_unregister_thread();
4275 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
4277 Error
*local_err
= NULL
;
4279 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
4282 * If there's a previous error, free it and prepare for another one.
4283 * Meanwhile if migration completes successfully, there won't have an error
4284 * dumped when calling migrate_fd_cleanup().
4286 migrate_error_free(s
);
4288 s
->expected_downtime
= s
->parameters
.downtime_limit
;
4290 assert(s
->cleanup_bh
);
4292 assert(!s
->cleanup_bh
);
4293 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup_bh
, s
);
4296 migrate_fd_error(s
, error_in
);
4299 * Don't do cleanup for resume if channel is invalid, but only dump
4300 * the error. We wait for another channel connect from the user.
4301 * The error_report still gives HMP user a hint on what failed.
4302 * It's normally done in migrate_fd_cleanup(), but call it here
4305 error_report_err(error_copy(s
->error
));
4307 migrate_fd_cleanup(s
);
4313 /* This is a resumed migration */
4314 rate_limit
= s
->parameters
.max_postcopy_bandwidth
/
4317 /* This is a fresh new migration */
4318 rate_limit
= s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
;
4320 /* Notify before starting migration thread */
4321 notifier_list_notify(&migration_state_notifiers
, s
);
4324 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
4325 qemu_file_set_blocking(s
->to_dst_file
, true);
4328 * Open the return path. For postcopy, it is used exclusively. For
4329 * precopy, only if user specified "return-path" capability would
4330 * QEMU uses the return path.
4332 if (migrate_postcopy_ram() || migrate_use_return_path()) {
4333 if (open_return_path_on_source(s
, !resume
)) {
4334 error_report("Unable to open return-path for postcopy");
4335 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
4336 migrate_fd_cleanup(s
);
4341 /* This needs to be done before resuming a postcopy */
4342 if (postcopy_preempt_setup(s
, &local_err
)) {
4343 error_report_err(local_err
);
4344 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
4345 MIGRATION_STATUS_FAILED
);
4346 migrate_fd_cleanup(s
);
4351 /* Wakeup the main migration thread to do the recovery */
4352 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
4353 MIGRATION_STATUS_POSTCOPY_RECOVER
);
4354 qemu_sem_post(&s
->postcopy_pause_sem
);
4358 if (multifd_save_setup(&local_err
) != 0) {
4359 error_report_err(local_err
);
4360 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
4361 MIGRATION_STATUS_FAILED
);
4362 migrate_fd_cleanup(s
);
4366 if (migrate_background_snapshot()) {
4367 qemu_thread_create(&s
->thread
, "bg_snapshot",
4368 bg_migration_thread
, s
, QEMU_THREAD_JOINABLE
);
4370 qemu_thread_create(&s
->thread
, "live_migration",
4371 migration_thread
, s
, QEMU_THREAD_JOINABLE
);
4373 s
->migration_thread_running
= true;
4376 void migration_global_dump(Monitor
*mon
)
4378 MigrationState
*ms
= migrate_get_current();
4380 monitor_printf(mon
, "globals:\n");
4381 monitor_printf(mon
, "store-global-state: %s\n",
4382 ms
->store_global_state
? "on" : "off");
4383 monitor_printf(mon
, "only-migratable: %s\n",
4384 only_migratable
? "on" : "off");
4385 monitor_printf(mon
, "send-configuration: %s\n",
4386 ms
->send_configuration
? "on" : "off");
4387 monitor_printf(mon
, "send-section-footer: %s\n",
4388 ms
->send_section_footer
? "on" : "off");
4389 monitor_printf(mon
, "decompress-error-check: %s\n",
4390 ms
->decompress_error_check
? "on" : "off");
4391 monitor_printf(mon
, "clear-bitmap-shift: %u\n",
4392 ms
->clear_bitmap_shift
);
4395 #define DEFINE_PROP_MIG_CAP(name, x) \
4396 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
4398 static Property migration_properties
[] = {
4399 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
4400 store_global_state
, true),
4401 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
4402 send_configuration
, true),
4403 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
4404 send_section_footer
, true),
4405 DEFINE_PROP_BOOL("decompress-error-check", MigrationState
,
4406 decompress_error_check
, true),
4407 DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState
,
4408 clear_bitmap_shift
, CLEAR_BITMAP_SHIFT_DEFAULT
),
4410 /* Migration parameters */
4411 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
4412 parameters
.compress_level
,
4413 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
4414 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
4415 parameters
.compress_threads
,
4416 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
4417 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState
,
4418 parameters
.compress_wait_thread
, true),
4419 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
4420 parameters
.decompress_threads
,
4421 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
4422 DEFINE_PROP_UINT8("x-throttle-trigger-threshold", MigrationState
,
4423 parameters
.throttle_trigger_threshold
,
4424 DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD
),
4425 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
4426 parameters
.cpu_throttle_initial
,
4427 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
4428 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
4429 parameters
.cpu_throttle_increment
,
4430 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
4431 DEFINE_PROP_BOOL("x-cpu-throttle-tailslow", MigrationState
,
4432 parameters
.cpu_throttle_tailslow
, false),
4433 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
4434 parameters
.max_bandwidth
, MAX_THROTTLE
),
4435 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
4436 parameters
.downtime_limit
,
4437 DEFAULT_MIGRATE_SET_DOWNTIME
),
4438 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
4439 parameters
.x_checkpoint_delay
,
4440 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
4441 DEFINE_PROP_UINT8("multifd-channels", MigrationState
,
4442 parameters
.multifd_channels
,
4443 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
4444 DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState
,
4445 parameters
.multifd_compression
,
4446 DEFAULT_MIGRATE_MULTIFD_COMPRESSION
),
4447 DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState
,
4448 parameters
.multifd_zlib_level
,
4449 DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL
),
4450 DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState
,
4451 parameters
.multifd_zstd_level
,
4452 DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL
),
4453 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
4454 parameters
.xbzrle_cache_size
,
4455 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
4456 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState
,
4457 parameters
.max_postcopy_bandwidth
,
4458 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH
),
4459 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState
,
4460 parameters
.max_cpu_throttle
,
4461 DEFAULT_MIGRATE_MAX_CPU_THROTTLE
),
4462 DEFINE_PROP_SIZE("announce-initial", MigrationState
,
4463 parameters
.announce_initial
,
4464 DEFAULT_MIGRATE_ANNOUNCE_INITIAL
),
4465 DEFINE_PROP_SIZE("announce-max", MigrationState
,
4466 parameters
.announce_max
,
4467 DEFAULT_MIGRATE_ANNOUNCE_MAX
),
4468 DEFINE_PROP_SIZE("announce-rounds", MigrationState
,
4469 parameters
.announce_rounds
,
4470 DEFAULT_MIGRATE_ANNOUNCE_ROUNDS
),
4471 DEFINE_PROP_SIZE("announce-step", MigrationState
,
4472 parameters
.announce_step
,
4473 DEFAULT_MIGRATE_ANNOUNCE_STEP
),
4474 DEFINE_PROP_STRING("tls-creds", MigrationState
, parameters
.tls_creds
),
4475 DEFINE_PROP_STRING("tls-hostname", MigrationState
, parameters
.tls_hostname
),
4476 DEFINE_PROP_STRING("tls-authz", MigrationState
, parameters
.tls_authz
),
4478 /* Migration capabilities */
4479 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
4480 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
4481 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
4482 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
4483 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
4484 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
4485 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
4486 DEFINE_PROP_MIG_CAP("x-postcopy-preempt",
4487 MIGRATION_CAPABILITY_POSTCOPY_PREEMPT
),
4488 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
4489 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
4490 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
4491 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
4492 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD
),
4493 DEFINE_PROP_MIG_CAP("x-background-snapshot",
4494 MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
),
4496 DEFINE_PROP_MIG_CAP("x-zero-copy-send",
4497 MIGRATION_CAPABILITY_ZERO_COPY_SEND
),
4500 DEFINE_PROP_END_OF_LIST(),
4503 static void migration_class_init(ObjectClass
*klass
, void *data
)
4505 DeviceClass
*dc
= DEVICE_CLASS(klass
);
4507 dc
->user_creatable
= false;
4508 device_class_set_props(dc
, migration_properties
);
4511 static void migration_instance_finalize(Object
*obj
)
4513 MigrationState
*ms
= MIGRATION_OBJ(obj
);
4515 qemu_mutex_destroy(&ms
->error_mutex
);
4516 qemu_mutex_destroy(&ms
->qemu_file_lock
);
4517 qemu_sem_destroy(&ms
->wait_unplug_sem
);
4518 qemu_sem_destroy(&ms
->rate_limit_sem
);
4519 qemu_sem_destroy(&ms
->pause_sem
);
4520 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
4521 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
4522 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
4523 qemu_sem_destroy(&ms
->postcopy_qemufile_src_sem
);
4524 error_free(ms
->error
);
4527 static void migration_instance_init(Object
*obj
)
4529 MigrationState
*ms
= MIGRATION_OBJ(obj
);
4530 MigrationParameters
*params
= &ms
->parameters
;
4532 ms
->state
= MIGRATION_STATUS_NONE
;
4534 ms
->pages_per_second
= -1;
4535 qemu_sem_init(&ms
->pause_sem
, 0);
4536 qemu_mutex_init(&ms
->error_mutex
);
4538 params
->tls_hostname
= g_strdup("");
4539 params
->tls_creds
= g_strdup("");
4541 /* Set has_* up only for parameter checks */
4542 params
->has_compress_level
= true;
4543 params
->has_compress_threads
= true;
4544 params
->has_compress_wait_thread
= true;
4545 params
->has_decompress_threads
= true;
4546 params
->has_throttle_trigger_threshold
= true;
4547 params
->has_cpu_throttle_initial
= true;
4548 params
->has_cpu_throttle_increment
= true;
4549 params
->has_cpu_throttle_tailslow
= true;
4550 params
->has_max_bandwidth
= true;
4551 params
->has_downtime_limit
= true;
4552 params
->has_x_checkpoint_delay
= true;
4553 params
->has_block_incremental
= true;
4554 params
->has_multifd_channels
= true;
4555 params
->has_multifd_compression
= true;
4556 params
->has_multifd_zlib_level
= true;
4557 params
->has_multifd_zstd_level
= true;
4558 params
->has_xbzrle_cache_size
= true;
4559 params
->has_max_postcopy_bandwidth
= true;
4560 params
->has_max_cpu_throttle
= true;
4561 params
->has_announce_initial
= true;
4562 params
->has_announce_max
= true;
4563 params
->has_announce_rounds
= true;
4564 params
->has_announce_step
= true;
4566 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
4567 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
4568 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
4569 qemu_sem_init(&ms
->rate_limit_sem
, 0);
4570 qemu_sem_init(&ms
->wait_unplug_sem
, 0);
4571 qemu_sem_init(&ms
->postcopy_qemufile_src_sem
, 0);
4572 qemu_mutex_init(&ms
->qemu_file_lock
);
4576 * Return true if check pass, false otherwise. Error will be put
4577 * inside errp if provided.
4579 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
4581 MigrationCapabilityStatusList
*head
= NULL
;
4582 /* Assuming all off */
4583 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
4586 if (!migrate_params_check(&ms
->parameters
, errp
)) {
4590 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
4591 if (ms
->enabled_capabilities
[i
]) {
4592 QAPI_LIST_PREPEND(head
, migrate_cap_add(i
, true));
4596 ret
= migrate_caps_check(cap_list
, head
, errp
);
4598 /* It works with head == NULL */
4599 qapi_free_MigrationCapabilityStatusList(head
);
4604 static const TypeInfo migration_type
= {
4605 .name
= TYPE_MIGRATION
,
4607 * NOTE: TYPE_MIGRATION is not really a device, as the object is
4608 * not created using qdev_new(), it is not attached to the qdev
4609 * device tree, and it is never realized.
4611 * TODO: Make this TYPE_OBJECT once QOM provides something like
4612 * TYPE_DEVICE's "-global" properties.
4614 .parent
= TYPE_DEVICE
,
4615 .class_init
= migration_class_init
,
4616 .class_size
= sizeof(MigrationClass
),
4617 .instance_size
= sizeof(MigrationState
),
4618 .instance_init
= migration_instance_init
,
4619 .instance_finalize
= migration_instance_finalize
,
4622 static void register_migration_types(void)
4624 type_register_static(&migration_type
);
4627 type_init(register_migration_types
);