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"
34 #include "migration/vmstate.h"
35 #include "block/block.h"
36 #include "qapi/error.h"
37 #include "qapi/clone-visitor.h"
38 #include "qapi/qapi-visit-migration.h"
39 #include "qapi/qapi-visit-sockets.h"
40 #include "qapi/qapi-commands-migration.h"
41 #include "qapi/qapi-events-migration.h"
42 #include "qapi/qmp/qerror.h"
43 #include "qapi/qmp/qnull.h"
46 #include "postcopy-ram.h"
47 #include "qemu/thread.h"
49 #include "exec/target_page.h"
50 #include "io/channel-buffer.h"
51 #include "io/channel-tls.h"
52 #include "migration/colo.h"
53 #include "hw/boards.h"
54 #include "hw/qdev-properties.h"
55 #include "hw/qdev-properties-system.h"
56 #include "monitor/monitor.h"
57 #include "net/announce.h"
58 #include "qemu/queue.h"
60 #include "qemu/yank.h"
61 #include "sysemu/cpus.h"
62 #include "yank_functions.h"
63 #include "sysemu/qtest.h"
65 #define MAX_THROTTLE (128 << 20) /* Migration transfer speed throttling */
67 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
69 #define BUFFER_DELAY 100
70 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
72 /* Time in milliseconds we are allowed to stop the source,
73 * for sending the last part */
74 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
76 /* Maximum migrate downtime set to 2000 seconds */
77 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
78 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
80 /* Default compression thread count */
81 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
82 /* Default decompression thread count, usually decompression is at
83 * least 4 times as fast as compression.*/
84 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
85 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
86 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
87 /* Define default autoconverge cpu throttle migration parameters */
88 #define DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD 50
89 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
90 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
91 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
93 /* Migration XBZRLE default cache size */
94 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
96 /* The delay time (in ms) between two COLO checkpoints */
97 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
98 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
99 #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE
100 /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */
101 #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
102 /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
103 #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
105 /* Background transfer rate for postcopy, 0 means unlimited, note
106 * that page requests can still exceed this limit.
108 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
111 * Parameters for self_announce_delay giving a stream of RARP/ARP
112 * packets after migration.
114 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
115 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
116 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
117 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
119 static NotifierList migration_state_notifiers
=
120 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
122 /* Messages sent on the return path from destination to source */
123 enum mig_rp_message_type
{
124 MIG_RP_MSG_INVALID
= 0, /* Must be 0 */
125 MIG_RP_MSG_SHUT
, /* sibling will not send any more RP messages */
126 MIG_RP_MSG_PONG
, /* Response to a PING; data (seq: be32 ) */
128 MIG_RP_MSG_REQ_PAGES_ID
, /* data (start: be64, len: be32, id: string) */
129 MIG_RP_MSG_REQ_PAGES
, /* data (start: be64, len: be32) */
130 MIG_RP_MSG_RECV_BITMAP
, /* send recved_bitmap back to source */
131 MIG_RP_MSG_RESUME_ACK
, /* tell source that we are ready to resume */
136 /* Migration capabilities set */
137 struct MigrateCapsSet
{
138 int size
; /* Capability set size */
139 MigrationCapability caps
[]; /* Variadic array of capabilities */
141 typedef struct MigrateCapsSet MigrateCapsSet
;
143 /* Define and initialize MigrateCapsSet */
144 #define INITIALIZE_MIGRATE_CAPS_SET(_name, ...) \
145 MigrateCapsSet _name = { \
146 .size = sizeof((int []) { __VA_ARGS__ }) / sizeof(int), \
147 .caps = { __VA_ARGS__ } \
150 /* Background-snapshot compatibility check list */
152 INITIALIZE_MIGRATE_CAPS_SET(check_caps_background_snapshot
,
153 MIGRATION_CAPABILITY_POSTCOPY_RAM
,
154 MIGRATION_CAPABILITY_DIRTY_BITMAPS
,
155 MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
,
156 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE
,
157 MIGRATION_CAPABILITY_RETURN_PATH
,
158 MIGRATION_CAPABILITY_MULTIFD
,
159 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
,
160 MIGRATION_CAPABILITY_AUTO_CONVERGE
,
161 MIGRATION_CAPABILITY_RELEASE_RAM
,
162 MIGRATION_CAPABILITY_RDMA_PIN_ALL
,
163 MIGRATION_CAPABILITY_COMPRESS
,
164 MIGRATION_CAPABILITY_XBZRLE
,
165 MIGRATION_CAPABILITY_X_COLO
,
166 MIGRATION_CAPABILITY_VALIDATE_UUID
,
167 MIGRATION_CAPABILITY_ZERO_COPY_SEND
);
169 /* When we add fault tolerance, we could have several
170 migrations at once. For now we don't need to add
171 dynamic creation of migration */
173 static MigrationState
*current_migration
;
174 static MigrationIncomingState
*current_incoming
;
176 static GSList
*migration_blockers
;
178 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
179 static int migration_maybe_pause(MigrationState
*s
,
180 int *current_active_state
,
182 static void migrate_fd_cancel(MigrationState
*s
);
184 static bool migrate_allow_multi_channels
= true;
186 void migrate_protocol_allow_multi_channels(bool allow
)
188 migrate_allow_multi_channels
= allow
;
191 bool migrate_multi_channels_is_allowed(void)
193 return migrate_allow_multi_channels
;
196 static gint
page_request_addr_cmp(gconstpointer ap
, gconstpointer bp
)
198 uintptr_t a
= (uintptr_t) ap
, b
= (uintptr_t) bp
;
200 return (a
> b
) - (a
< b
);
203 void migration_object_init(void)
205 /* This can only be called once. */
206 assert(!current_migration
);
207 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
210 * Init the migrate incoming object as well no matter whether
211 * we'll use it or not.
213 assert(!current_incoming
);
214 current_incoming
= g_new0(MigrationIncomingState
, 1);
215 current_incoming
->state
= MIGRATION_STATUS_NONE
;
216 current_incoming
->postcopy_remote_fds
=
217 g_array_new(FALSE
, TRUE
, sizeof(struct PostCopyFD
));
218 qemu_mutex_init(¤t_incoming
->rp_mutex
);
219 qemu_mutex_init(¤t_incoming
->postcopy_prio_thread_mutex
);
220 qemu_event_init(¤t_incoming
->main_thread_load_event
, false);
221 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_dst
, 0);
222 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fault
, 0);
223 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fast_load
, 0);
224 qemu_mutex_init(¤t_incoming
->page_request_mutex
);
225 current_incoming
->page_requested
= g_tree_new(page_request_addr_cmp
);
227 migration_object_check(current_migration
, &error_fatal
);
231 dirty_bitmap_mig_init();
234 void migration_cancel(const Error
*error
)
237 migrate_set_error(current_migration
, error
);
239 migrate_fd_cancel(current_migration
);
242 void migration_shutdown(void)
245 * When the QEMU main thread exit, the COLO thread
246 * may wait a semaphore. So, we should wakeup the
247 * COLO thread before migration shutdown.
251 * Cancel the current migration - that will (eventually)
252 * stop the migration using this structure
254 migration_cancel(NULL
);
255 object_unref(OBJECT(current_migration
));
258 * Cancel outgoing migration of dirty bitmaps. It should
259 * at least unref used block nodes.
261 dirty_bitmap_mig_cancel_outgoing();
264 * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
265 * are non-critical data, and their loss never considered as
268 dirty_bitmap_mig_cancel_incoming();
272 MigrationState
*migrate_get_current(void)
274 /* This can only be called after the object created. */
275 assert(current_migration
);
276 return current_migration
;
279 MigrationIncomingState
*migration_incoming_get_current(void)
281 assert(current_incoming
);
282 return current_incoming
;
285 void migration_incoming_transport_cleanup(MigrationIncomingState
*mis
)
287 if (mis
->socket_address_list
) {
288 qapi_free_SocketAddressList(mis
->socket_address_list
);
289 mis
->socket_address_list
= NULL
;
292 if (mis
->transport_cleanup
) {
293 mis
->transport_cleanup(mis
->transport_data
);
294 mis
->transport_data
= mis
->transport_cleanup
= NULL
;
298 void migration_incoming_state_destroy(void)
300 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
302 if (mis
->to_src_file
) {
303 /* Tell source that we are done */
304 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
305 qemu_fclose(mis
->to_src_file
);
306 mis
->to_src_file
= NULL
;
309 if (mis
->from_src_file
) {
310 migration_ioc_unregister_yank_from_file(mis
->from_src_file
);
311 qemu_fclose(mis
->from_src_file
);
312 mis
->from_src_file
= NULL
;
314 if (mis
->postcopy_remote_fds
) {
315 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
316 mis
->postcopy_remote_fds
= NULL
;
319 migration_incoming_transport_cleanup(mis
);
320 qemu_event_reset(&mis
->main_thread_load_event
);
322 if (mis
->page_requested
) {
323 g_tree_destroy(mis
->page_requested
);
324 mis
->page_requested
= NULL
;
327 if (mis
->postcopy_qemufile_dst
) {
328 migration_ioc_unregister_yank_from_file(mis
->postcopy_qemufile_dst
);
329 qemu_fclose(mis
->postcopy_qemufile_dst
);
330 mis
->postcopy_qemufile_dst
= NULL
;
333 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
336 static void migrate_generate_event(int new_state
)
338 if (migrate_use_events()) {
339 qapi_event_send_migration(new_state
);
343 static bool migrate_late_block_activate(void)
347 s
= migrate_get_current();
349 return s
->enabled_capabilities
[
350 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE
];
354 * Send a message on the return channel back to the source
357 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
358 enum mig_rp_message_type message_type
,
359 uint16_t len
, void *data
)
363 trace_migrate_send_rp_message((int)message_type
, len
);
364 QEMU_LOCK_GUARD(&mis
->rp_mutex
);
367 * It's possible that the file handle got lost due to network
370 if (!mis
->to_src_file
) {
375 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
376 qemu_put_be16(mis
->to_src_file
, len
);
377 qemu_put_buffer(mis
->to_src_file
, data
, len
);
378 qemu_fflush(mis
->to_src_file
);
380 /* It's possible that qemu file got error during sending */
381 ret
= qemu_file_get_error(mis
->to_src_file
);
386 /* Request one page from the source VM at the given start address.
387 * rb: the RAMBlock to request the page in
388 * Start: Address offset within the RB
389 * Len: Length in bytes required - must be a multiple of pagesize
391 int migrate_send_rp_message_req_pages(MigrationIncomingState
*mis
,
392 RAMBlock
*rb
, ram_addr_t start
)
394 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
395 size_t msglen
= 12; /* start + len */
396 size_t len
= qemu_ram_pagesize(rb
);
397 enum mig_rp_message_type msg_type
;
401 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
402 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
405 * We maintain the last ramblock that we requested for page. Note that we
406 * don't need locking because this function will only be called within the
407 * postcopy ram fault thread.
409 if (rb
!= mis
->last_rb
) {
412 rbname
= qemu_ram_get_idstr(rb
);
413 rbname_len
= strlen(rbname
);
415 assert(rbname_len
< 256);
417 bufc
[msglen
++] = rbname_len
;
418 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
419 msglen
+= rbname_len
;
420 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
422 msg_type
= MIG_RP_MSG_REQ_PAGES
;
425 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
428 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
,
429 RAMBlock
*rb
, ram_addr_t start
, uint64_t haddr
)
431 void *aligned
= (void *)(uintptr_t)ROUND_DOWN(haddr
, qemu_ram_pagesize(rb
));
432 bool received
= false;
434 WITH_QEMU_LOCK_GUARD(&mis
->page_request_mutex
) {
435 received
= ramblock_recv_bitmap_test_byte_offset(rb
, start
);
436 if (!received
&& !g_tree_lookup(mis
->page_requested
, aligned
)) {
438 * The page has not been received, and it's not yet in the page
439 * request list. Queue it. Set the value of element to 1, so that
440 * things like g_tree_lookup() will return TRUE (1) when found.
442 g_tree_insert(mis
->page_requested
, aligned
, (gpointer
)1);
443 mis
->page_requested_count
++;
444 trace_postcopy_page_req_add(aligned
, mis
->page_requested_count
);
449 * If the page is there, skip sending the message. We don't even need the
450 * lock because as long as the page arrived, it'll be there forever.
456 return migrate_send_rp_message_req_pages(mis
, rb
, start
);
459 static bool migration_colo_enabled
;
460 bool migration_incoming_colo_enabled(void)
462 return migration_colo_enabled
;
465 void migration_incoming_disable_colo(void)
467 ram_block_discard_disable(false);
468 migration_colo_enabled
= false;
471 int migration_incoming_enable_colo(void)
473 if (ram_block_discard_disable(true)) {
474 error_report("COLO: cannot disable RAM discard");
477 migration_colo_enabled
= true;
481 void migrate_add_address(SocketAddress
*address
)
483 MigrationIncomingState
*mis
= migration_incoming_get_current();
485 QAPI_LIST_PREPEND(mis
->socket_address_list
,
486 QAPI_CLONE(SocketAddress
, address
));
489 static void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
491 const char *p
= NULL
;
493 migrate_protocol_allow_multi_channels(false); /* reset it anyway */
494 qapi_event_send_migration(MIGRATION_STATUS_SETUP
);
495 if (strstart(uri
, "tcp:", &p
) ||
496 strstart(uri
, "unix:", NULL
) ||
497 strstart(uri
, "vsock:", NULL
)) {
498 migrate_protocol_allow_multi_channels(true);
499 socket_start_incoming_migration(p
? p
: uri
, errp
);
501 } else if (strstart(uri
, "rdma:", &p
)) {
502 rdma_start_incoming_migration(p
, errp
);
504 } else if (strstart(uri
, "exec:", &p
)) {
505 exec_start_incoming_migration(p
, errp
);
506 } else if (strstart(uri
, "fd:", &p
)) {
507 fd_start_incoming_migration(p
, errp
);
509 error_setg(errp
, "unknown migration protocol: %s", uri
);
513 static void process_incoming_migration_bh(void *opaque
)
515 Error
*local_err
= NULL
;
516 MigrationIncomingState
*mis
= opaque
;
518 /* If capability late_block_activate is set:
519 * Only fire up the block code now if we're going to restart the
520 * VM, else 'cont' will do it.
521 * This causes file locking to happen; so we don't want it to happen
522 * unless we really are starting the VM.
524 if (!migrate_late_block_activate() ||
525 (autostart
&& (!global_state_received() ||
526 global_state_get_runstate() == RUN_STATE_RUNNING
))) {
527 /* Make sure all file formats throw away their mutable metadata.
528 * If we get an error here, just don't restart the VM yet. */
529 bdrv_activate_all(&local_err
);
531 error_report_err(local_err
);
538 * This must happen after all error conditions are dealt with and
539 * we're sure the VM is going to be running on this host.
541 qemu_announce_self(&mis
->announce_timer
, migrate_announce_params());
543 if (multifd_load_cleanup(&local_err
) != 0) {
544 error_report_err(local_err
);
547 /* If global state section was not received or we are in running
548 state, we need to obey autostart. Any other state is set with
551 dirty_bitmap_mig_before_vm_start();
553 if (!global_state_received() ||
554 global_state_get_runstate() == RUN_STATE_RUNNING
) {
558 runstate_set(RUN_STATE_PAUSED
);
560 } else if (migration_incoming_colo_enabled()) {
561 migration_incoming_disable_colo();
564 runstate_set(global_state_get_runstate());
567 * This must happen after any state changes since as soon as an external
568 * observer sees this event they might start to prod at the VM assuming
571 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
572 MIGRATION_STATUS_COMPLETED
);
573 qemu_bh_delete(mis
->bh
);
574 migration_incoming_state_destroy();
577 static void coroutine_fn
578 process_incoming_migration_co(void *opaque
)
580 MigrationIncomingState
*mis
= migration_incoming_get_current();
583 Error
*local_err
= NULL
;
585 assert(mis
->from_src_file
);
586 mis
->migration_incoming_co
= qemu_coroutine_self();
587 mis
->largest_page_size
= qemu_ram_pagesize_largest();
588 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
589 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
590 MIGRATION_STATUS_ACTIVE
);
591 ret
= qemu_loadvm_state(mis
->from_src_file
);
593 ps
= postcopy_state_get();
594 trace_process_incoming_migration_co_end(ret
, ps
);
595 if (ps
!= POSTCOPY_INCOMING_NONE
) {
596 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
598 * Where a migration had postcopy enabled (and thus went to advise)
599 * but managed to complete within the precopy period, we can use
602 postcopy_ram_incoming_cleanup(mis
);
603 } else if (ret
>= 0) {
605 * Postcopy was started, cleanup should happen at the end of the
608 trace_process_incoming_migration_co_postcopy_end_main();
611 /* Else if something went wrong then just fall out of the normal exit */
614 /* we get COLO info, and know if we are in COLO mode */
615 if (!ret
&& migration_incoming_colo_enabled()) {
616 /* Make sure all file formats throw away their mutable metadata */
617 bdrv_activate_all(&local_err
);
619 error_report_err(local_err
);
623 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
624 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
625 mis
->have_colo_incoming_thread
= true;
626 qemu_coroutine_yield();
628 qemu_mutex_unlock_iothread();
629 /* Wait checkpoint incoming thread exit before free resource */
630 qemu_thread_join(&mis
->colo_incoming_thread
);
631 qemu_mutex_lock_iothread();
632 /* We hold the global iothread lock, so it is safe here */
633 colo_release_ram_cache();
637 error_report("load of migration failed: %s", strerror(-ret
));
640 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
641 qemu_bh_schedule(mis
->bh
);
642 mis
->migration_incoming_co
= NULL
;
646 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
647 MIGRATION_STATUS_FAILED
);
648 qemu_fclose(mis
->from_src_file
);
649 if (multifd_load_cleanup(&local_err
) != 0) {
650 error_report_err(local_err
);
656 * migration_incoming_setup: Setup incoming migration
657 * @f: file for main migration channel
658 * @errp: where to put errors
660 * Returns: %true on success, %false on error.
662 static bool migration_incoming_setup(QEMUFile
*f
, Error
**errp
)
664 MigrationIncomingState
*mis
= migration_incoming_get_current();
666 if (multifd_load_setup(errp
) != 0) {
670 if (!mis
->from_src_file
) {
671 mis
->from_src_file
= f
;
673 qemu_file_set_blocking(f
, false);
677 void migration_incoming_process(void)
679 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
680 qemu_coroutine_enter(co
);
683 /* Returns true if recovered from a paused migration, otherwise false */
684 static bool postcopy_try_recover(void)
686 MigrationIncomingState
*mis
= migration_incoming_get_current();
688 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
689 /* Resumed from a paused postcopy migration */
691 /* This should be set already in migration_incoming_setup() */
692 assert(mis
->from_src_file
);
693 /* Postcopy has standalone thread to do vm load */
694 qemu_file_set_blocking(mis
->from_src_file
, true);
696 /* Re-configure the return path */
697 mis
->to_src_file
= qemu_file_get_return_path(mis
->from_src_file
);
699 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
700 MIGRATION_STATUS_POSTCOPY_RECOVER
);
703 * Here, we only wake up the main loading thread (while the
704 * rest threads will still be waiting), so that we can receive
705 * commands from source now, and answer it if needed. The
706 * rest threads will be woken up afterwards until we are sure
707 * that source is ready to reply to page requests.
709 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
716 void migration_fd_process_incoming(QEMUFile
*f
, Error
**errp
)
718 if (!migration_incoming_setup(f
, errp
)) {
721 if (postcopy_try_recover()) {
724 migration_incoming_process();
727 static bool migration_needs_multiple_sockets(void)
729 return migrate_use_multifd() || migrate_postcopy_preempt();
732 void migration_ioc_process_incoming(QIOChannel
*ioc
, Error
**errp
)
734 MigrationIncomingState
*mis
= migration_incoming_get_current();
735 Error
*local_err
= NULL
;
736 bool start_migration
;
739 if (!mis
->from_src_file
) {
740 /* The first connection (multifd may have multiple) */
741 f
= qemu_file_new_input(ioc
);
743 if (!migration_incoming_setup(f
, errp
)) {
748 * Common migration only needs one channel, so we can start
749 * right now. Some features need more than one channel, we wait.
751 start_migration
= !migration_needs_multiple_sockets();
753 /* Multiple connections */
754 assert(migration_needs_multiple_sockets());
755 if (migrate_use_multifd()) {
756 start_migration
= multifd_recv_new_channel(ioc
, &local_err
);
758 assert(migrate_postcopy_preempt());
759 f
= qemu_file_new_input(ioc
);
760 start_migration
= postcopy_preempt_new_channel(mis
, f
);
763 error_propagate(errp
, local_err
);
768 if (start_migration
) {
769 /* If it's a recovery, we're done */
770 if (postcopy_try_recover()) {
773 migration_incoming_process();
778 * @migration_has_all_channels: We have received all channels that we need
780 * Returns true when we have got connections to all the channels that
781 * we need for migration.
783 bool migration_has_all_channels(void)
785 MigrationIncomingState
*mis
= migration_incoming_get_current();
787 if (!mis
->from_src_file
) {
791 if (migrate_use_multifd()) {
792 return multifd_recv_all_channels_created();
795 if (migrate_postcopy_preempt()) {
796 return mis
->postcopy_qemufile_dst
!= NULL
;
803 * Send a 'SHUT' message on the return channel with the given value
804 * to indicate that we've finished with the RP. Non-0 value indicates
807 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
812 buf
= cpu_to_be32(value
);
813 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
817 * Send a 'PONG' message on the return channel with the given value
818 * (normally in response to a 'PING')
820 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
825 buf
= cpu_to_be32(value
);
826 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
829 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
837 * First, we send the header part. It contains only the len of
838 * idstr, and the idstr itself.
840 len
= strlen(block_name
);
842 memcpy(buf
+ 1, block_name
, len
);
844 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
845 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
850 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
853 * Next, we dump the received bitmap to the stream.
855 * TODO: currently we are safe since we are the only one that is
856 * using the to_src_file handle (fault thread is still paused),
857 * and it's ok even not taking the mutex. However the best way is
858 * to take the lock before sending the message header, and release
859 * the lock after sending the bitmap.
861 qemu_mutex_lock(&mis
->rp_mutex
);
862 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
863 qemu_mutex_unlock(&mis
->rp_mutex
);
865 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
868 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
872 buf
= cpu_to_be32(value
);
873 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
876 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
878 MigrationCapabilityStatusList
*head
= NULL
, **tail
= &head
;
879 MigrationCapabilityStatus
*caps
;
880 MigrationState
*s
= migrate_get_current();
883 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
884 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
885 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
889 caps
= g_malloc0(sizeof(*caps
));
890 caps
->capability
= i
;
891 caps
->state
= s
->enabled_capabilities
[i
];
892 QAPI_LIST_APPEND(tail
, caps
);
898 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
900 MigrationParameters
*params
;
901 MigrationState
*s
= migrate_get_current();
903 /* TODO use QAPI_CLONE() instead of duplicating it inline */
904 params
= g_malloc0(sizeof(*params
));
905 params
->has_compress_level
= true;
906 params
->compress_level
= s
->parameters
.compress_level
;
907 params
->has_compress_threads
= true;
908 params
->compress_threads
= s
->parameters
.compress_threads
;
909 params
->has_compress_wait_thread
= true;
910 params
->compress_wait_thread
= s
->parameters
.compress_wait_thread
;
911 params
->has_decompress_threads
= true;
912 params
->decompress_threads
= s
->parameters
.decompress_threads
;
913 params
->has_throttle_trigger_threshold
= true;
914 params
->throttle_trigger_threshold
= s
->parameters
.throttle_trigger_threshold
;
915 params
->has_cpu_throttle_initial
= true;
916 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
917 params
->has_cpu_throttle_increment
= true;
918 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
919 params
->has_cpu_throttle_tailslow
= true;
920 params
->cpu_throttle_tailslow
= s
->parameters
.cpu_throttle_tailslow
;
921 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
922 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
923 params
->tls_authz
= g_strdup(s
->parameters
.tls_authz
?
924 s
->parameters
.tls_authz
: "");
925 params
->has_max_bandwidth
= true;
926 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
927 params
->has_downtime_limit
= true;
928 params
->downtime_limit
= s
->parameters
.downtime_limit
;
929 params
->has_x_checkpoint_delay
= true;
930 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
931 params
->has_block_incremental
= true;
932 params
->block_incremental
= s
->parameters
.block_incremental
;
933 params
->has_multifd_channels
= true;
934 params
->multifd_channels
= s
->parameters
.multifd_channels
;
935 params
->has_multifd_compression
= true;
936 params
->multifd_compression
= s
->parameters
.multifd_compression
;
937 params
->has_multifd_zlib_level
= true;
938 params
->multifd_zlib_level
= s
->parameters
.multifd_zlib_level
;
939 params
->has_multifd_zstd_level
= true;
940 params
->multifd_zstd_level
= s
->parameters
.multifd_zstd_level
;
941 params
->has_xbzrle_cache_size
= true;
942 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
943 params
->has_max_postcopy_bandwidth
= true;
944 params
->max_postcopy_bandwidth
= s
->parameters
.max_postcopy_bandwidth
;
945 params
->has_max_cpu_throttle
= true;
946 params
->max_cpu_throttle
= s
->parameters
.max_cpu_throttle
;
947 params
->has_announce_initial
= true;
948 params
->announce_initial
= s
->parameters
.announce_initial
;
949 params
->has_announce_max
= true;
950 params
->announce_max
= s
->parameters
.announce_max
;
951 params
->has_announce_rounds
= true;
952 params
->announce_rounds
= s
->parameters
.announce_rounds
;
953 params
->has_announce_step
= true;
954 params
->announce_step
= s
->parameters
.announce_step
;
956 if (s
->parameters
.has_block_bitmap_mapping
) {
957 params
->has_block_bitmap_mapping
= true;
958 params
->block_bitmap_mapping
=
959 QAPI_CLONE(BitmapMigrationNodeAliasList
,
960 s
->parameters
.block_bitmap_mapping
);
966 AnnounceParameters
*migrate_announce_params(void)
968 static AnnounceParameters ap
;
970 MigrationState
*s
= migrate_get_current();
972 ap
.initial
= s
->parameters
.announce_initial
;
973 ap
.max
= s
->parameters
.announce_max
;
974 ap
.rounds
= s
->parameters
.announce_rounds
;
975 ap
.step
= s
->parameters
.announce_step
;
981 * Return true if we're already in the middle of a migration
982 * (i.e. any of the active or setup states)
984 bool migration_is_setup_or_active(int state
)
987 case MIGRATION_STATUS_ACTIVE
:
988 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
989 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
990 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
991 case MIGRATION_STATUS_SETUP
:
992 case MIGRATION_STATUS_PRE_SWITCHOVER
:
993 case MIGRATION_STATUS_DEVICE
:
994 case MIGRATION_STATUS_WAIT_UNPLUG
:
995 case MIGRATION_STATUS_COLO
:
1004 bool migration_is_running(int state
)
1007 case MIGRATION_STATUS_ACTIVE
:
1008 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1009 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1010 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1011 case MIGRATION_STATUS_SETUP
:
1012 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1013 case MIGRATION_STATUS_DEVICE
:
1014 case MIGRATION_STATUS_WAIT_UNPLUG
:
1015 case MIGRATION_STATUS_CANCELLING
:
1024 static void populate_time_info(MigrationInfo
*info
, MigrationState
*s
)
1026 info
->has_status
= true;
1027 info
->has_setup_time
= true;
1028 info
->setup_time
= s
->setup_time
;
1029 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
1030 info
->has_total_time
= true;
1031 info
->total_time
= s
->total_time
;
1032 info
->has_downtime
= true;
1033 info
->downtime
= s
->downtime
;
1035 info
->has_total_time
= true;
1036 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) -
1038 info
->has_expected_downtime
= true;
1039 info
->expected_downtime
= s
->expected_downtime
;
1043 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
1045 size_t page_size
= qemu_target_page_size();
1047 info
->ram
= g_malloc0(sizeof(*info
->ram
));
1048 info
->ram
->transferred
= ram_counters
.transferred
;
1049 info
->ram
->total
= ram_bytes_total();
1050 info
->ram
->duplicate
= ram_counters
.duplicate
;
1051 /* legacy value. It is not used anymore */
1052 info
->ram
->skipped
= 0;
1053 info
->ram
->normal
= ram_counters
.normal
;
1054 info
->ram
->normal_bytes
= ram_counters
.normal
* page_size
;
1055 info
->ram
->mbps
= s
->mbps
;
1056 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
1057 info
->ram
->dirty_sync_missed_zero_copy
=
1058 ram_counters
.dirty_sync_missed_zero_copy
;
1059 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
1060 info
->ram
->page_size
= page_size
;
1061 info
->ram
->multifd_bytes
= ram_counters
.multifd_bytes
;
1062 info
->ram
->pages_per_second
= s
->pages_per_second
;
1063 info
->ram
->precopy_bytes
= ram_counters
.precopy_bytes
;
1064 info
->ram
->downtime_bytes
= ram_counters
.downtime_bytes
;
1065 info
->ram
->postcopy_bytes
= ram_counters
.postcopy_bytes
;
1067 if (migrate_use_xbzrle()) {
1068 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
1069 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
1070 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
1071 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
1072 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
1073 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
1074 info
->xbzrle_cache
->encoding_rate
= xbzrle_counters
.encoding_rate
;
1075 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
1078 if (migrate_use_compression()) {
1079 info
->compression
= g_malloc0(sizeof(*info
->compression
));
1080 info
->compression
->pages
= compression_counters
.pages
;
1081 info
->compression
->busy
= compression_counters
.busy
;
1082 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
1083 info
->compression
->compressed_size
=
1084 compression_counters
.compressed_size
;
1085 info
->compression
->compression_rate
=
1086 compression_counters
.compression_rate
;
1089 if (cpu_throttle_active()) {
1090 info
->has_cpu_throttle_percentage
= true;
1091 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
1094 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
1095 info
->ram
->remaining
= ram_bytes_remaining();
1096 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
1100 static void populate_disk_info(MigrationInfo
*info
)
1102 if (blk_mig_active()) {
1103 info
->disk
= g_malloc0(sizeof(*info
->disk
));
1104 info
->disk
->transferred
= blk_mig_bytes_transferred();
1105 info
->disk
->remaining
= blk_mig_bytes_remaining();
1106 info
->disk
->total
= blk_mig_bytes_total();
1110 static void fill_source_migration_info(MigrationInfo
*info
)
1112 MigrationState
*s
= migrate_get_current();
1113 int state
= qatomic_read(&s
->state
);
1114 GSList
*cur_blocker
= migration_blockers
;
1116 info
->blocked_reasons
= NULL
;
1119 * There are two types of reasons a migration might be blocked;
1120 * a) devices marked in VMState as non-migratable, and
1121 * b) Explicit migration blockers
1122 * We need to add both of them here.
1124 qemu_savevm_non_migratable_list(&info
->blocked_reasons
);
1126 while (cur_blocker
) {
1127 QAPI_LIST_PREPEND(info
->blocked_reasons
,
1128 g_strdup(error_get_pretty(cur_blocker
->data
)));
1129 cur_blocker
= g_slist_next(cur_blocker
);
1131 info
->has_blocked_reasons
= info
->blocked_reasons
!= NULL
;
1134 case MIGRATION_STATUS_NONE
:
1135 /* no migration has happened ever */
1136 /* do not overwrite destination migration status */
1138 case MIGRATION_STATUS_SETUP
:
1139 info
->has_status
= true;
1140 info
->has_total_time
= false;
1142 case MIGRATION_STATUS_ACTIVE
:
1143 case MIGRATION_STATUS_CANCELLING
:
1144 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1145 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1146 case MIGRATION_STATUS_DEVICE
:
1147 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1148 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1149 /* TODO add some postcopy stats */
1150 populate_time_info(info
, s
);
1151 populate_ram_info(info
, s
);
1152 populate_disk_info(info
);
1153 populate_vfio_info(info
);
1155 case MIGRATION_STATUS_COLO
:
1156 info
->has_status
= true;
1157 /* TODO: display COLO specific information (checkpoint info etc.) */
1159 case MIGRATION_STATUS_COMPLETED
:
1160 populate_time_info(info
, s
);
1161 populate_ram_info(info
, s
);
1162 populate_vfio_info(info
);
1164 case MIGRATION_STATUS_FAILED
:
1165 info
->has_status
= true;
1167 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
1170 case MIGRATION_STATUS_CANCELLED
:
1171 info
->has_status
= true;
1173 case MIGRATION_STATUS_WAIT_UNPLUG
:
1174 info
->has_status
= true;
1177 info
->status
= state
;
1180 typedef enum WriteTrackingSupport
{
1181 WT_SUPPORT_UNKNOWN
= 0,
1183 WT_SUPPORT_AVAILABLE
,
1184 WT_SUPPORT_COMPATIBLE
1185 } WriteTrackingSupport
;
1188 WriteTrackingSupport
migrate_query_write_tracking(void)
1190 /* Check if kernel supports required UFFD features */
1191 if (!ram_write_tracking_available()) {
1192 return WT_SUPPORT_ABSENT
;
1195 * Check if current memory configuration is
1196 * compatible with required UFFD features.
1198 if (!ram_write_tracking_compatible()) {
1199 return WT_SUPPORT_AVAILABLE
;
1202 return WT_SUPPORT_COMPATIBLE
;
1206 * @migration_caps_check - check capability validity
1208 * @cap_list: old capability list, array of bool
1209 * @params: new capabilities to be applied soon
1210 * @errp: set *errp if the check failed, with reason
1212 * Returns true if check passed, otherwise false.
1214 static bool migrate_caps_check(bool *cap_list
,
1215 MigrationCapabilityStatusList
*params
,
1218 MigrationCapabilityStatusList
*cap
;
1219 bool old_postcopy_cap
;
1220 MigrationIncomingState
*mis
= migration_incoming_get_current();
1222 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1224 for (cap
= params
; cap
; cap
= cap
->next
) {
1225 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
1228 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1229 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
1230 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
1232 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
1237 #ifndef CONFIG_REPLICATION
1238 if (cap_list
[MIGRATION_CAPABILITY_X_COLO
]) {
1239 error_setg(errp
, "QEMU compiled without replication module"
1240 " can't enable COLO");
1241 error_append_hint(errp
, "Please enable replication before COLO.\n");
1246 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
1247 /* This check is reasonably expensive, so only when it's being
1248 * set the first time, also it's only the destination that needs
1251 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
1252 !postcopy_ram_supported_by_host(mis
)) {
1253 /* postcopy_ram_supported_by_host will have emitted a more
1256 error_setg(errp
, "Postcopy is not supported");
1260 if (cap_list
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
]) {
1261 error_setg(errp
, "Postcopy is not compatible with ignore-shared");
1266 if (cap_list
[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
]) {
1267 WriteTrackingSupport wt_support
;
1270 * Check if 'background-snapshot' capability is supported by
1271 * host kernel and compatible with guest memory configuration.
1273 wt_support
= migrate_query_write_tracking();
1274 if (wt_support
< WT_SUPPORT_AVAILABLE
) {
1275 error_setg(errp
, "Background-snapshot is not supported by host kernel");
1278 if (wt_support
< WT_SUPPORT_COMPATIBLE
) {
1279 error_setg(errp
, "Background-snapshot is not compatible "
1280 "with guest memory configuration");
1285 * Check if there are any migration capabilities
1286 * incompatible with 'background-snapshot'.
1288 for (idx
= 0; idx
< check_caps_background_snapshot
.size
; idx
++) {
1289 int incomp_cap
= check_caps_background_snapshot
.caps
[idx
];
1290 if (cap_list
[incomp_cap
]) {
1292 "Background-snapshot is not compatible with %s",
1293 MigrationCapability_str(incomp_cap
));
1300 if (cap_list
[MIGRATION_CAPABILITY_ZERO_COPY_SEND
] &&
1301 (!cap_list
[MIGRATION_CAPABILITY_MULTIFD
] ||
1302 cap_list
[MIGRATION_CAPABILITY_COMPRESS
] ||
1303 cap_list
[MIGRATION_CAPABILITY_XBZRLE
] ||
1304 migrate_multifd_compression() ||
1305 migrate_use_tls())) {
1307 "Zero copy only available for non-compressed non-TLS multifd migration");
1311 if (cap_list
[MIGRATION_CAPABILITY_ZERO_COPY_SEND
]) {
1313 "Zero copy currently only available on Linux");
1319 /* incoming side only */
1320 if (runstate_check(RUN_STATE_INMIGRATE
) &&
1321 !migrate_multi_channels_is_allowed() &&
1322 cap_list
[MIGRATION_CAPABILITY_MULTIFD
]) {
1323 error_setg(errp
, "multifd is not supported by current protocol");
1327 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_PREEMPT
]) {
1328 if (!cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
1329 error_setg(errp
, "Postcopy preempt requires postcopy-ram");
1334 * Preempt mode requires urgent pages to be sent in separate
1335 * channel, OTOH compression logic will disorder all pages into
1336 * different compression channels, which is not compatible with the
1337 * preempt assumptions on channel assignments.
1339 if (cap_list
[MIGRATION_CAPABILITY_COMPRESS
]) {
1340 error_setg(errp
, "Postcopy preempt not compatible with compress");
1345 if (cap_list
[MIGRATION_CAPABILITY_MULTIFD
]) {
1346 if (cap_list
[MIGRATION_CAPABILITY_COMPRESS
]) {
1347 error_setg(errp
, "Multifd is not compatible with compress");
1355 static void fill_destination_migration_info(MigrationInfo
*info
)
1357 MigrationIncomingState
*mis
= migration_incoming_get_current();
1359 if (mis
->socket_address_list
) {
1360 info
->has_socket_address
= true;
1361 info
->socket_address
=
1362 QAPI_CLONE(SocketAddressList
, mis
->socket_address_list
);
1365 switch (mis
->state
) {
1366 case MIGRATION_STATUS_NONE
:
1368 case MIGRATION_STATUS_SETUP
:
1369 case MIGRATION_STATUS_CANCELLING
:
1370 case MIGRATION_STATUS_CANCELLED
:
1371 case MIGRATION_STATUS_ACTIVE
:
1372 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1373 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1374 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1375 case MIGRATION_STATUS_FAILED
:
1376 case MIGRATION_STATUS_COLO
:
1377 info
->has_status
= true;
1379 case MIGRATION_STATUS_COMPLETED
:
1380 info
->has_status
= true;
1381 fill_destination_postcopy_migration_info(info
);
1384 info
->status
= mis
->state
;
1387 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1389 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1391 fill_destination_migration_info(info
);
1392 fill_source_migration_info(info
);
1397 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
1400 MigrationState
*s
= migrate_get_current();
1401 MigrationCapabilityStatusList
*cap
;
1402 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
1404 if (migration_is_running(s
->state
)) {
1405 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1409 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
1410 if (!migrate_caps_check(cap_list
, params
, errp
)) {
1414 for (cap
= params
; cap
; cap
= cap
->next
) {
1415 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
1420 * Check whether the parameters are valid. Error will be put into errp
1421 * (if provided). Return true if valid, otherwise false.
1423 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
1425 if (params
->has_compress_level
&&
1426 (params
->compress_level
> 9)) {
1427 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
1428 "a value between 0 and 9");
1432 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
1433 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1435 "a value between 1 and 255");
1439 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
1440 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1441 "decompress_threads",
1442 "a value between 1 and 255");
1446 if (params
->has_throttle_trigger_threshold
&&
1447 (params
->throttle_trigger_threshold
< 1 ||
1448 params
->throttle_trigger_threshold
> 100)) {
1449 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1450 "throttle_trigger_threshold",
1451 "an integer in the range of 1 to 100");
1455 if (params
->has_cpu_throttle_initial
&&
1456 (params
->cpu_throttle_initial
< 1 ||
1457 params
->cpu_throttle_initial
> 99)) {
1458 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1459 "cpu_throttle_initial",
1460 "an integer in the range of 1 to 99");
1464 if (params
->has_cpu_throttle_increment
&&
1465 (params
->cpu_throttle_increment
< 1 ||
1466 params
->cpu_throttle_increment
> 99)) {
1467 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1468 "cpu_throttle_increment",
1469 "an integer in the range of 1 to 99");
1473 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
1474 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1476 "an integer in the range of 0 to "stringify(SIZE_MAX
)
1481 if (params
->has_downtime_limit
&&
1482 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
1483 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1485 "an integer in the range of 0 to "
1486 stringify(MAX_MIGRATE_DOWNTIME
)" ms");
1490 /* x_checkpoint_delay is now always positive */
1492 if (params
->has_multifd_channels
&& (params
->multifd_channels
< 1)) {
1493 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1495 "a value between 1 and 255");
1499 if (params
->has_multifd_zlib_level
&&
1500 (params
->multifd_zlib_level
> 9)) {
1501 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zlib_level",
1502 "a value between 0 and 9");
1506 if (params
->has_multifd_zstd_level
&&
1507 (params
->multifd_zstd_level
> 20)) {
1508 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zstd_level",
1509 "a value between 0 and 20");
1513 if (params
->has_xbzrle_cache_size
&&
1514 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
1515 !is_power_of_2(params
->xbzrle_cache_size
))) {
1516 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1517 "xbzrle_cache_size",
1518 "a power of two no less than the target page size");
1522 if (params
->has_max_cpu_throttle
&&
1523 (params
->max_cpu_throttle
< params
->cpu_throttle_initial
||
1524 params
->max_cpu_throttle
> 99)) {
1525 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1527 "an integer in the range of cpu_throttle_initial to 99");
1531 if (params
->has_announce_initial
&&
1532 params
->announce_initial
> 100000) {
1533 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1535 "a value between 0 and 100000");
1538 if (params
->has_announce_max
&&
1539 params
->announce_max
> 100000) {
1540 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1542 "a value between 0 and 100000");
1545 if (params
->has_announce_rounds
&&
1546 params
->announce_rounds
> 1000) {
1547 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1549 "a value between 0 and 1000");
1552 if (params
->has_announce_step
&&
1553 (params
->announce_step
< 1 ||
1554 params
->announce_step
> 10000)) {
1555 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1557 "a value between 0 and 10000");
1561 if (params
->has_block_bitmap_mapping
&&
1562 !check_dirty_bitmap_mig_alias_map(params
->block_bitmap_mapping
, errp
)) {
1563 error_prepend(errp
, "Invalid mapping given for block-bitmap-mapping: ");
1568 if (migrate_use_zero_copy_send() &&
1569 ((params
->has_multifd_compression
&& params
->multifd_compression
) ||
1570 (params
->tls_creds
&& *params
->tls_creds
))) {
1572 "Zero copy only available for non-compressed non-TLS multifd migration");
1580 static void migrate_params_test_apply(MigrateSetParameters
*params
,
1581 MigrationParameters
*dest
)
1583 *dest
= migrate_get_current()->parameters
;
1585 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1587 if (params
->has_compress_level
) {
1588 dest
->compress_level
= params
->compress_level
;
1591 if (params
->has_compress_threads
) {
1592 dest
->compress_threads
= params
->compress_threads
;
1595 if (params
->has_compress_wait_thread
) {
1596 dest
->compress_wait_thread
= params
->compress_wait_thread
;
1599 if (params
->has_decompress_threads
) {
1600 dest
->decompress_threads
= params
->decompress_threads
;
1603 if (params
->has_throttle_trigger_threshold
) {
1604 dest
->throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1607 if (params
->has_cpu_throttle_initial
) {
1608 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
1611 if (params
->has_cpu_throttle_increment
) {
1612 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
1615 if (params
->has_cpu_throttle_tailslow
) {
1616 dest
->cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1619 if (params
->tls_creds
) {
1620 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1621 dest
->tls_creds
= params
->tls_creds
->u
.s
;
1624 if (params
->tls_hostname
) {
1625 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1626 dest
->tls_hostname
= params
->tls_hostname
->u
.s
;
1629 if (params
->has_max_bandwidth
) {
1630 dest
->max_bandwidth
= params
->max_bandwidth
;
1633 if (params
->has_downtime_limit
) {
1634 dest
->downtime_limit
= params
->downtime_limit
;
1637 if (params
->has_x_checkpoint_delay
) {
1638 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
1641 if (params
->has_block_incremental
) {
1642 dest
->block_incremental
= params
->block_incremental
;
1644 if (params
->has_multifd_channels
) {
1645 dest
->multifd_channels
= params
->multifd_channels
;
1647 if (params
->has_multifd_compression
) {
1648 dest
->multifd_compression
= params
->multifd_compression
;
1650 if (params
->has_xbzrle_cache_size
) {
1651 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
1653 if (params
->has_max_postcopy_bandwidth
) {
1654 dest
->max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1656 if (params
->has_max_cpu_throttle
) {
1657 dest
->max_cpu_throttle
= params
->max_cpu_throttle
;
1659 if (params
->has_announce_initial
) {
1660 dest
->announce_initial
= params
->announce_initial
;
1662 if (params
->has_announce_max
) {
1663 dest
->announce_max
= params
->announce_max
;
1665 if (params
->has_announce_rounds
) {
1666 dest
->announce_rounds
= params
->announce_rounds
;
1668 if (params
->has_announce_step
) {
1669 dest
->announce_step
= params
->announce_step
;
1672 if (params
->has_block_bitmap_mapping
) {
1673 dest
->has_block_bitmap_mapping
= true;
1674 dest
->block_bitmap_mapping
= params
->block_bitmap_mapping
;
1678 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
1680 MigrationState
*s
= migrate_get_current();
1682 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1684 if (params
->has_compress_level
) {
1685 s
->parameters
.compress_level
= params
->compress_level
;
1688 if (params
->has_compress_threads
) {
1689 s
->parameters
.compress_threads
= params
->compress_threads
;
1692 if (params
->has_compress_wait_thread
) {
1693 s
->parameters
.compress_wait_thread
= params
->compress_wait_thread
;
1696 if (params
->has_decompress_threads
) {
1697 s
->parameters
.decompress_threads
= params
->decompress_threads
;
1700 if (params
->has_throttle_trigger_threshold
) {
1701 s
->parameters
.throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1704 if (params
->has_cpu_throttle_initial
) {
1705 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
1708 if (params
->has_cpu_throttle_increment
) {
1709 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
1712 if (params
->has_cpu_throttle_tailslow
) {
1713 s
->parameters
.cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1716 if (params
->tls_creds
) {
1717 g_free(s
->parameters
.tls_creds
);
1718 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1719 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1722 if (params
->tls_hostname
) {
1723 g_free(s
->parameters
.tls_hostname
);
1724 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1725 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1728 if (params
->tls_authz
) {
1729 g_free(s
->parameters
.tls_authz
);
1730 assert(params
->tls_authz
->type
== QTYPE_QSTRING
);
1731 s
->parameters
.tls_authz
= g_strdup(params
->tls_authz
->u
.s
);
1734 if (params
->has_max_bandwidth
) {
1735 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1736 if (s
->to_dst_file
&& !migration_in_postcopy()) {
1737 qemu_file_set_rate_limit(s
->to_dst_file
,
1738 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1742 if (params
->has_downtime_limit
) {
1743 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1746 if (params
->has_x_checkpoint_delay
) {
1747 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1748 if (migration_in_colo_state()) {
1749 colo_checkpoint_notify(s
);
1753 if (params
->has_block_incremental
) {
1754 s
->parameters
.block_incremental
= params
->block_incremental
;
1756 if (params
->has_multifd_channels
) {
1757 s
->parameters
.multifd_channels
= params
->multifd_channels
;
1759 if (params
->has_multifd_compression
) {
1760 s
->parameters
.multifd_compression
= params
->multifd_compression
;
1762 if (params
->has_xbzrle_cache_size
) {
1763 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1764 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1766 if (params
->has_max_postcopy_bandwidth
) {
1767 s
->parameters
.max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1768 if (s
->to_dst_file
&& migration_in_postcopy()) {
1769 qemu_file_set_rate_limit(s
->to_dst_file
,
1770 s
->parameters
.max_postcopy_bandwidth
/ XFER_LIMIT_RATIO
);
1773 if (params
->has_max_cpu_throttle
) {
1774 s
->parameters
.max_cpu_throttle
= params
->max_cpu_throttle
;
1776 if (params
->has_announce_initial
) {
1777 s
->parameters
.announce_initial
= params
->announce_initial
;
1779 if (params
->has_announce_max
) {
1780 s
->parameters
.announce_max
= params
->announce_max
;
1782 if (params
->has_announce_rounds
) {
1783 s
->parameters
.announce_rounds
= params
->announce_rounds
;
1785 if (params
->has_announce_step
) {
1786 s
->parameters
.announce_step
= params
->announce_step
;
1789 if (params
->has_block_bitmap_mapping
) {
1790 qapi_free_BitmapMigrationNodeAliasList(
1791 s
->parameters
.block_bitmap_mapping
);
1793 s
->parameters
.has_block_bitmap_mapping
= true;
1794 s
->parameters
.block_bitmap_mapping
=
1795 QAPI_CLONE(BitmapMigrationNodeAliasList
,
1796 params
->block_bitmap_mapping
);
1800 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1802 MigrationParameters tmp
;
1804 /* TODO Rewrite "" to null instead */
1805 if (params
->tls_creds
1806 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1807 qobject_unref(params
->tls_creds
->u
.n
);
1808 params
->tls_creds
->type
= QTYPE_QSTRING
;
1809 params
->tls_creds
->u
.s
= strdup("");
1811 /* TODO Rewrite "" to null instead */
1812 if (params
->tls_hostname
1813 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1814 qobject_unref(params
->tls_hostname
->u
.n
);
1815 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1816 params
->tls_hostname
->u
.s
= strdup("");
1819 migrate_params_test_apply(params
, &tmp
);
1821 if (!migrate_params_check(&tmp
, errp
)) {
1822 /* Invalid parameter */
1826 migrate_params_apply(params
, errp
);
1830 void qmp_migrate_start_postcopy(Error
**errp
)
1832 MigrationState
*s
= migrate_get_current();
1834 if (!migrate_postcopy()) {
1835 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1836 " the start of migration");
1840 if (s
->state
== MIGRATION_STATUS_NONE
) {
1841 error_setg(errp
, "Postcopy must be started after migration has been"
1846 * we don't error if migration has finished since that would be racy
1847 * with issuing this command.
1849 qatomic_set(&s
->start_postcopy
, true);
1852 /* shared migration helpers */
1854 void migrate_set_state(int *state
, int old_state
, int new_state
)
1856 assert(new_state
< MIGRATION_STATUS__MAX
);
1857 if (qatomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1858 trace_migrate_set_state(MigrationStatus_str(new_state
));
1859 migrate_generate_event(new_state
);
1863 static MigrationCapabilityStatus
*migrate_cap_add(MigrationCapability index
,
1866 MigrationCapabilityStatus
*cap
;
1868 cap
= g_new0(MigrationCapabilityStatus
, 1);
1869 cap
->capability
= index
;
1875 void migrate_set_block_enabled(bool value
, Error
**errp
)
1877 MigrationCapabilityStatusList
*cap
= NULL
;
1879 QAPI_LIST_PREPEND(cap
, migrate_cap_add(MIGRATION_CAPABILITY_BLOCK
, value
));
1880 qmp_migrate_set_capabilities(cap
, errp
);
1881 qapi_free_MigrationCapabilityStatusList(cap
);
1884 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1886 s
->parameters
.block_incremental
= value
;
1889 static void block_cleanup_parameters(MigrationState
*s
)
1891 if (s
->must_remove_block_options
) {
1892 /* setting to false can never fail */
1893 migrate_set_block_enabled(false, &error_abort
);
1894 migrate_set_block_incremental(s
, false);
1895 s
->must_remove_block_options
= false;
1899 static void migrate_fd_cleanup(MigrationState
*s
)
1901 qemu_bh_delete(s
->cleanup_bh
);
1902 s
->cleanup_bh
= NULL
;
1904 g_free(s
->hostname
);
1907 qemu_savevm_state_cleanup();
1909 if (s
->to_dst_file
) {
1912 trace_migrate_fd_cleanup();
1913 qemu_mutex_unlock_iothread();
1914 if (s
->migration_thread_running
) {
1915 qemu_thread_join(&s
->thread
);
1916 s
->migration_thread_running
= false;
1918 qemu_mutex_lock_iothread();
1920 multifd_save_cleanup();
1921 qemu_mutex_lock(&s
->qemu_file_lock
);
1922 tmp
= s
->to_dst_file
;
1923 s
->to_dst_file
= NULL
;
1924 qemu_mutex_unlock(&s
->qemu_file_lock
);
1926 * Close the file handle without the lock to make sure the
1927 * critical section won't block for long.
1929 migration_ioc_unregister_yank_from_file(tmp
);
1933 if (s
->postcopy_qemufile_src
) {
1934 migration_ioc_unregister_yank_from_file(s
->postcopy_qemufile_src
);
1935 qemu_fclose(s
->postcopy_qemufile_src
);
1936 s
->postcopy_qemufile_src
= NULL
;
1939 assert(!migration_is_active(s
));
1941 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1942 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1943 MIGRATION_STATUS_CANCELLED
);
1947 /* It is used on info migrate. We can't free it */
1948 error_report_err(error_copy(s
->error
));
1950 notifier_list_notify(&migration_state_notifiers
, s
);
1951 block_cleanup_parameters(s
);
1952 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1955 static void migrate_fd_cleanup_schedule(MigrationState
*s
)
1958 * Ref the state for bh, because it may be called when
1959 * there're already no other refs
1961 object_ref(OBJECT(s
));
1962 qemu_bh_schedule(s
->cleanup_bh
);
1965 static void migrate_fd_cleanup_bh(void *opaque
)
1967 MigrationState
*s
= opaque
;
1968 migrate_fd_cleanup(s
);
1969 object_unref(OBJECT(s
));
1972 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1974 QEMU_LOCK_GUARD(&s
->error_mutex
);
1976 s
->error
= error_copy(error
);
1980 static void migrate_error_free(MigrationState
*s
)
1982 QEMU_LOCK_GUARD(&s
->error_mutex
);
1984 error_free(s
->error
);
1989 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1991 trace_migrate_fd_error(error_get_pretty(error
));
1992 assert(s
->to_dst_file
== NULL
);
1993 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1994 MIGRATION_STATUS_FAILED
);
1995 migrate_set_error(s
, error
);
1998 static void migrate_fd_cancel(MigrationState
*s
)
2001 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
2002 trace_migrate_fd_cancel();
2004 WITH_QEMU_LOCK_GUARD(&s
->qemu_file_lock
) {
2005 if (s
->rp_state
.from_dst_file
) {
2006 /* shutdown the rp socket, so causing the rp thread to shutdown */
2007 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
2012 old_state
= s
->state
;
2013 if (!migration_is_running(old_state
)) {
2016 /* If the migration is paused, kick it out of the pause */
2017 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
2018 qemu_sem_post(&s
->pause_sem
);
2020 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
2021 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
2024 * If we're unlucky the migration code might be stuck somewhere in a
2025 * send/write while the network has failed and is waiting to timeout;
2026 * if we've got shutdown(2) available then we can force it to quit.
2027 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
2028 * called in a bh, so there is no race against this cancel.
2030 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
2031 qemu_file_shutdown(f
);
2033 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
2034 Error
*local_err
= NULL
;
2036 bdrv_activate_all(&local_err
);
2038 error_report_err(local_err
);
2040 s
->block_inactive
= false;
2045 void add_migration_state_change_notifier(Notifier
*notify
)
2047 notifier_list_add(&migration_state_notifiers
, notify
);
2050 void remove_migration_state_change_notifier(Notifier
*notify
)
2052 notifier_remove(notify
);
2055 bool migration_in_setup(MigrationState
*s
)
2057 return s
->state
== MIGRATION_STATUS_SETUP
;
2060 bool migration_has_finished(MigrationState
*s
)
2062 return s
->state
== MIGRATION_STATUS_COMPLETED
;
2065 bool migration_has_failed(MigrationState
*s
)
2067 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
2068 s
->state
== MIGRATION_STATUS_FAILED
);
2071 bool migration_in_postcopy(void)
2073 MigrationState
*s
= migrate_get_current();
2076 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
2077 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
2078 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
2085 bool migration_in_postcopy_after_devices(MigrationState
*s
)
2087 return migration_in_postcopy() && s
->postcopy_after_devices
;
2090 bool migration_in_incoming_postcopy(void)
2092 PostcopyState ps
= postcopy_state_get();
2094 return ps
>= POSTCOPY_INCOMING_DISCARD
&& ps
< POSTCOPY_INCOMING_END
;
2097 bool migration_in_bg_snapshot(void)
2099 MigrationState
*s
= migrate_get_current();
2101 return migrate_background_snapshot() &&
2102 migration_is_setup_or_active(s
->state
);
2105 bool migration_is_idle(void)
2107 MigrationState
*s
= current_migration
;
2114 case MIGRATION_STATUS_NONE
:
2115 case MIGRATION_STATUS_CANCELLED
:
2116 case MIGRATION_STATUS_COMPLETED
:
2117 case MIGRATION_STATUS_FAILED
:
2119 case MIGRATION_STATUS_SETUP
:
2120 case MIGRATION_STATUS_CANCELLING
:
2121 case MIGRATION_STATUS_ACTIVE
:
2122 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
2123 case MIGRATION_STATUS_COLO
:
2124 case MIGRATION_STATUS_PRE_SWITCHOVER
:
2125 case MIGRATION_STATUS_DEVICE
:
2126 case MIGRATION_STATUS_WAIT_UNPLUG
:
2128 case MIGRATION_STATUS__MAX
:
2129 g_assert_not_reached();
2135 bool migration_is_active(MigrationState
*s
)
2137 return (s
->state
== MIGRATION_STATUS_ACTIVE
||
2138 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2141 void migrate_init(MigrationState
*s
)
2144 * Reinitialise all migration state, except
2145 * parameters/capabilities that the user set, and
2150 s
->to_dst_file
= NULL
;
2151 s
->state
= MIGRATION_STATUS_NONE
;
2152 s
->rp_state
.from_dst_file
= NULL
;
2153 s
->rp_state
.error
= false;
2155 s
->pages_per_second
= 0.0;
2157 s
->expected_downtime
= 0;
2159 s
->start_postcopy
= false;
2160 s
->postcopy_after_devices
= false;
2161 s
->migration_thread_running
= false;
2162 error_free(s
->error
);
2166 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
2168 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2170 s
->vm_was_running
= false;
2171 s
->iteration_initial_bytes
= 0;
2172 s
->threshold_size
= 0;
2175 int migrate_add_blocker_internal(Error
*reason
, Error
**errp
)
2177 /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */
2178 if (runstate_check(RUN_STATE_SAVE_VM
) || !migration_is_idle()) {
2179 error_propagate_prepend(errp
, error_copy(reason
),
2180 "disallowing migration blocker "
2181 "(migration/snapshot in progress) for: ");
2185 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
2189 int migrate_add_blocker(Error
*reason
, Error
**errp
)
2191 if (only_migratable
) {
2192 error_propagate_prepend(errp
, error_copy(reason
),
2193 "disallowing migration blocker "
2194 "(--only-migratable) for: ");
2198 return migrate_add_blocker_internal(reason
, errp
);
2201 void migrate_del_blocker(Error
*reason
)
2203 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
2206 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
2208 Error
*local_err
= NULL
;
2209 static bool once
= true;
2212 error_setg(errp
, "The incoming migration has already been started");
2215 if (!runstate_check(RUN_STATE_INMIGRATE
)) {
2216 error_setg(errp
, "'-incoming' was not specified on the command line");
2220 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
2224 qemu_start_incoming_migration(uri
, &local_err
);
2227 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2228 error_propagate(errp
, local_err
);
2235 void qmp_migrate_recover(const char *uri
, Error
**errp
)
2237 MigrationIncomingState
*mis
= migration_incoming_get_current();
2240 * Don't even bother to use ERRP_GUARD() as it _must_ always be set by
2241 * callers (no one should ignore a recover failure); if there is, it's a
2242 * programming error.
2246 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2247 error_setg(errp
, "Migrate recover can only be run "
2248 "when postcopy is paused.");
2252 /* If there's an existing transport, release it */
2253 migration_incoming_transport_cleanup(mis
);
2256 * Note that this call will never start a real migration; it will
2257 * only re-setup the migration stream and poke existing migration
2258 * to continue using that newly established channel.
2260 qemu_start_incoming_migration(uri
, errp
);
2263 void qmp_migrate_pause(Error
**errp
)
2265 MigrationState
*ms
= migrate_get_current();
2266 MigrationIncomingState
*mis
= migration_incoming_get_current();
2269 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2270 /* Source side, during postcopy */
2271 qemu_mutex_lock(&ms
->qemu_file_lock
);
2272 ret
= qemu_file_shutdown(ms
->to_dst_file
);
2273 qemu_mutex_unlock(&ms
->qemu_file_lock
);
2275 error_setg(errp
, "Failed to pause source migration");
2280 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2281 ret
= qemu_file_shutdown(mis
->from_src_file
);
2283 error_setg(errp
, "Failed to pause destination migration");
2288 error_setg(errp
, "migrate-pause is currently only supported "
2289 "during postcopy-active state");
2292 bool migration_is_blocked(Error
**errp
)
2294 if (qemu_savevm_state_blocked(errp
)) {
2298 if (migration_blockers
) {
2299 error_propagate(errp
, error_copy(migration_blockers
->data
));
2306 /* Returns true if continue to migrate, or false if error detected */
2307 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
2308 bool resume
, Error
**errp
)
2310 Error
*local_err
= NULL
;
2313 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2314 error_setg(errp
, "Cannot resume if there is no "
2315 "paused migration");
2320 * Postcopy recovery won't work well with release-ram
2321 * capability since release-ram will drop the page buffer as
2322 * long as the page is put into the send buffer. So if there
2323 * is a network failure happened, any page buffers that have
2324 * not yet reached the destination VM but have already been
2325 * sent from the source VM will be lost forever. Let's refuse
2326 * the client from resuming such a postcopy migration.
2327 * Luckily release-ram was designed to only be used when src
2328 * and destination VMs are on the same host, so it should be
2331 if (migrate_release_ram()) {
2332 error_setg(errp
, "Postcopy recovery cannot work "
2333 "when release-ram capability is set");
2337 /* This is a resume, skip init status */
2341 if (migration_is_running(s
->state
)) {
2342 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
2346 if (runstate_check(RUN_STATE_INMIGRATE
)) {
2347 error_setg(errp
, "Guest is waiting for an incoming migration");
2351 if (runstate_check(RUN_STATE_POSTMIGRATE
)) {
2352 error_setg(errp
, "Can't migrate the vm that was paused due to "
2353 "previous migration");
2357 if (migration_is_blocked(errp
)) {
2361 if (blk
|| blk_inc
) {
2362 if (migrate_colo_enabled()) {
2363 error_setg(errp
, "No disk migration is required in COLO mode");
2366 if (migrate_use_block() || migrate_use_block_incremental()) {
2367 error_setg(errp
, "Command options are incompatible with "
2368 "current migration capabilities");
2371 migrate_set_block_enabled(true, &local_err
);
2373 error_propagate(errp
, local_err
);
2376 s
->must_remove_block_options
= true;
2380 migrate_set_block_incremental(s
, true);
2385 * set ram_counters compression_counters memory to zero for a
2388 memset(&ram_counters
, 0, sizeof(ram_counters
));
2389 memset(&compression_counters
, 0, sizeof(compression_counters
));
2394 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
2395 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
2396 bool has_resume
, bool resume
, Error
**errp
)
2398 Error
*local_err
= NULL
;
2399 MigrationState
*s
= migrate_get_current();
2400 const char *p
= NULL
;
2402 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
2403 has_resume
&& resume
, errp
)) {
2404 /* Error detected, put into errp */
2408 if (!(has_resume
&& resume
)) {
2409 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
2414 migrate_protocol_allow_multi_channels(false);
2415 if (strstart(uri
, "tcp:", &p
) ||
2416 strstart(uri
, "unix:", NULL
) ||
2417 strstart(uri
, "vsock:", NULL
)) {
2418 migrate_protocol_allow_multi_channels(true);
2419 socket_start_outgoing_migration(s
, p
? p
: uri
, &local_err
);
2421 } else if (strstart(uri
, "rdma:", &p
)) {
2422 rdma_start_outgoing_migration(s
, p
, &local_err
);
2424 } else if (strstart(uri
, "exec:", &p
)) {
2425 exec_start_outgoing_migration(s
, p
, &local_err
);
2426 } else if (strstart(uri
, "fd:", &p
)) {
2427 fd_start_outgoing_migration(s
, p
, &local_err
);
2429 if (!(has_resume
&& resume
)) {
2430 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2432 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
2433 "a valid migration protocol");
2434 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2435 MIGRATION_STATUS_FAILED
);
2436 block_cleanup_parameters(s
);
2441 if (!(has_resume
&& resume
)) {
2442 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2444 migrate_fd_error(s
, local_err
);
2445 error_propagate(errp
, local_err
);
2450 void qmp_migrate_cancel(Error
**errp
)
2452 migration_cancel(NULL
);
2455 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
2457 MigrationState
*s
= migrate_get_current();
2458 if (s
->state
!= state
) {
2459 error_setg(errp
, "Migration not in expected state: %s",
2460 MigrationStatus_str(s
->state
));
2463 qemu_sem_post(&s
->pause_sem
);
2466 bool migrate_release_ram(void)
2470 s
= migrate_get_current();
2472 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
2475 bool migrate_postcopy_ram(void)
2479 s
= migrate_get_current();
2481 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
2484 bool migrate_postcopy(void)
2486 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2489 bool migrate_auto_converge(void)
2493 s
= migrate_get_current();
2495 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
2498 bool migrate_zero_blocks(void)
2502 s
= migrate_get_current();
2504 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
2507 bool migrate_postcopy_blocktime(void)
2511 s
= migrate_get_current();
2513 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
2516 bool migrate_use_compression(void)
2520 s
= migrate_get_current();
2522 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
2525 int migrate_compress_level(void)
2529 s
= migrate_get_current();
2531 return s
->parameters
.compress_level
;
2534 int migrate_compress_threads(void)
2538 s
= migrate_get_current();
2540 return s
->parameters
.compress_threads
;
2543 int migrate_compress_wait_thread(void)
2547 s
= migrate_get_current();
2549 return s
->parameters
.compress_wait_thread
;
2552 int migrate_decompress_threads(void)
2556 s
= migrate_get_current();
2558 return s
->parameters
.decompress_threads
;
2561 bool migrate_dirty_bitmaps(void)
2565 s
= migrate_get_current();
2567 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
2570 bool migrate_ignore_shared(void)
2574 s
= migrate_get_current();
2576 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
];
2579 bool migrate_validate_uuid(void)
2583 s
= migrate_get_current();
2585 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_VALIDATE_UUID
];
2588 bool migrate_use_events(void)
2592 s
= migrate_get_current();
2594 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
2597 bool migrate_use_multifd(void)
2601 s
= migrate_get_current();
2603 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_MULTIFD
];
2606 bool migrate_pause_before_switchover(void)
2610 s
= migrate_get_current();
2612 return s
->enabled_capabilities
[
2613 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
2616 int migrate_multifd_channels(void)
2620 s
= migrate_get_current();
2622 return s
->parameters
.multifd_channels
;
2625 MultiFDCompression
migrate_multifd_compression(void)
2629 s
= migrate_get_current();
2631 assert(s
->parameters
.multifd_compression
< MULTIFD_COMPRESSION__MAX
);
2632 return s
->parameters
.multifd_compression
;
2635 int migrate_multifd_zlib_level(void)
2639 s
= migrate_get_current();
2641 return s
->parameters
.multifd_zlib_level
;
2644 int migrate_multifd_zstd_level(void)
2648 s
= migrate_get_current();
2650 return s
->parameters
.multifd_zstd_level
;
2654 bool migrate_use_zero_copy_send(void)
2658 s
= migrate_get_current();
2660 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_COPY_SEND
];
2664 int migrate_use_tls(void)
2668 s
= migrate_get_current();
2670 return s
->parameters
.tls_creds
&& *s
->parameters
.tls_creds
;
2673 int migrate_use_xbzrle(void)
2677 s
= migrate_get_current();
2679 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
2682 uint64_t migrate_xbzrle_cache_size(void)
2686 s
= migrate_get_current();
2688 return s
->parameters
.xbzrle_cache_size
;
2691 static int64_t migrate_max_postcopy_bandwidth(void)
2695 s
= migrate_get_current();
2697 return s
->parameters
.max_postcopy_bandwidth
;
2700 bool migrate_use_block(void)
2704 s
= migrate_get_current();
2706 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
2709 bool migrate_use_return_path(void)
2713 s
= migrate_get_current();
2715 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
2718 bool migrate_use_block_incremental(void)
2722 s
= migrate_get_current();
2724 return s
->parameters
.block_incremental
;
2727 bool migrate_background_snapshot(void)
2731 s
= migrate_get_current();
2733 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
];
2736 bool migrate_postcopy_preempt(void)
2740 s
= migrate_get_current();
2742 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_PREEMPT
];
2745 /* migration thread support */
2747 * Something bad happened to the RP stream, mark an error
2748 * The caller shall print or trace something to indicate why
2750 static void mark_source_rp_bad(MigrationState
*s
)
2752 s
->rp_state
.error
= true;
2755 static struct rp_cmd_args
{
2756 ssize_t len
; /* -1 = variable */
2759 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
2760 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
2761 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
2762 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
2763 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
2764 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
2765 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
2766 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
2770 * Process a request for pages received on the return path,
2771 * We're allowed to send more than requested (e.g. to round to our page size)
2772 * and we don't need to send pages that have already been sent.
2774 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
2775 ram_addr_t start
, size_t len
)
2777 long our_host_ps
= qemu_real_host_page_size();
2779 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
2782 * Since we currently insist on matching page sizes, just sanity check
2783 * we're being asked for whole host pages.
2785 if (!QEMU_IS_ALIGNED(start
, our_host_ps
) ||
2786 !QEMU_IS_ALIGNED(len
, our_host_ps
)) {
2787 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2788 " len: %zd", __func__
, start
, len
);
2789 mark_source_rp_bad(ms
);
2793 if (ram_save_queue_pages(rbname
, start
, len
)) {
2794 mark_source_rp_bad(ms
);
2798 /* Return true to retry, false to quit */
2799 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
2801 trace_postcopy_pause_return_path();
2803 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
2805 trace_postcopy_pause_return_path_continued();
2810 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
2812 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
2815 error_report("%s: invalid block name '%s'", __func__
, block_name
);
2819 /* Fetch the received bitmap and refresh the dirty bitmap */
2820 return ram_dirty_bitmap_reload(s
, block
);
2823 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
2825 trace_source_return_path_thread_resume_ack(value
);
2827 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
2828 error_report("%s: illegal resume_ack value %"PRIu32
,
2833 /* Now both sides are active. */
2834 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2835 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2837 /* Notify send thread that time to continue send pages */
2838 qemu_sem_post(&s
->rp_state
.rp_sem
);
2843 /* Release ms->rp_state.from_dst_file in a safe way */
2844 static void migration_release_from_dst_file(MigrationState
*ms
)
2848 WITH_QEMU_LOCK_GUARD(&ms
->qemu_file_lock
) {
2850 * Reset the from_dst_file pointer first before releasing it, as we
2851 * can't block within lock section
2853 file
= ms
->rp_state
.from_dst_file
;
2854 ms
->rp_state
.from_dst_file
= NULL
;
2861 * Handles messages sent on the return path towards the source VM
2864 static void *source_return_path_thread(void *opaque
)
2866 MigrationState
*ms
= opaque
;
2867 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
2868 uint16_t header_len
, header_type
;
2870 uint32_t tmp32
, sibling_error
;
2871 ram_addr_t start
= 0; /* =0 to silence warning */
2872 size_t len
= 0, expected_len
;
2875 trace_source_return_path_thread_entry();
2876 rcu_register_thread();
2879 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
2880 migration_is_setup_or_active(ms
->state
)) {
2881 trace_source_return_path_thread_loop_top();
2882 header_type
= qemu_get_be16(rp
);
2883 header_len
= qemu_get_be16(rp
);
2885 if (qemu_file_get_error(rp
)) {
2886 mark_source_rp_bad(ms
);
2890 if (header_type
>= MIG_RP_MSG_MAX
||
2891 header_type
== MIG_RP_MSG_INVALID
) {
2892 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2893 header_type
, header_len
);
2894 mark_source_rp_bad(ms
);
2898 if ((rp_cmd_args
[header_type
].len
!= -1 &&
2899 header_len
!= rp_cmd_args
[header_type
].len
) ||
2900 header_len
> sizeof(buf
)) {
2901 error_report("RP: Received '%s' message (0x%04x) with"
2902 "incorrect length %d expecting %zu",
2903 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
2904 (size_t)rp_cmd_args
[header_type
].len
);
2905 mark_source_rp_bad(ms
);
2909 /* We know we've got a valid header by this point */
2910 res
= qemu_get_buffer(rp
, buf
, header_len
);
2911 if (res
!= header_len
) {
2912 error_report("RP: Failed reading data for message 0x%04x"
2913 " read %d expected %d",
2914 header_type
, res
, header_len
);
2915 mark_source_rp_bad(ms
);
2919 /* OK, we have the message and the data */
2920 switch (header_type
) {
2921 case MIG_RP_MSG_SHUT
:
2922 sibling_error
= ldl_be_p(buf
);
2923 trace_source_return_path_thread_shut(sibling_error
);
2924 if (sibling_error
) {
2925 error_report("RP: Sibling indicated error %d", sibling_error
);
2926 mark_source_rp_bad(ms
);
2929 * We'll let the main thread deal with closing the RP
2930 * we could do a shutdown(2) on it, but we're the only user
2931 * anyway, so there's nothing gained.
2935 case MIG_RP_MSG_PONG
:
2936 tmp32
= ldl_be_p(buf
);
2937 trace_source_return_path_thread_pong(tmp32
);
2940 case MIG_RP_MSG_REQ_PAGES
:
2941 start
= ldq_be_p(buf
);
2942 len
= ldl_be_p(buf
+ 8);
2943 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
2946 case MIG_RP_MSG_REQ_PAGES_ID
:
2947 expected_len
= 12 + 1; /* header + termination */
2949 if (header_len
>= expected_len
) {
2950 start
= ldq_be_p(buf
);
2951 len
= ldl_be_p(buf
+ 8);
2952 /* Now we expect an idstr */
2953 tmp32
= buf
[12]; /* Length of the following idstr */
2954 buf
[13 + tmp32
] = '\0';
2955 expected_len
+= tmp32
;
2957 if (header_len
!= expected_len
) {
2958 error_report("RP: Req_Page_id with length %d expecting %zd",
2959 header_len
, expected_len
);
2960 mark_source_rp_bad(ms
);
2963 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
2966 case MIG_RP_MSG_RECV_BITMAP
:
2967 if (header_len
< 1) {
2968 error_report("%s: missing block name", __func__
);
2969 mark_source_rp_bad(ms
);
2972 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2973 buf
[buf
[0] + 1] = '\0';
2974 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
2975 mark_source_rp_bad(ms
);
2980 case MIG_RP_MSG_RESUME_ACK
:
2981 tmp32
= ldl_be_p(buf
);
2982 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
2983 mark_source_rp_bad(ms
);
2994 res
= qemu_file_get_error(rp
);
2996 if (res
&& migration_in_postcopy()) {
2998 * Maybe there is something we can do: it looks like a
2999 * network down issue, and we pause for a recovery.
3001 migration_release_from_dst_file(ms
);
3003 if (postcopy_pause_return_path_thread(ms
)) {
3005 * Reload rp, reset the rest. Referencing it is safe since
3006 * it's reset only by us above, or when migration completes
3008 rp
= ms
->rp_state
.from_dst_file
;
3009 ms
->rp_state
.error
= false;
3014 trace_source_return_path_thread_bad_end();
3015 mark_source_rp_bad(ms
);
3018 trace_source_return_path_thread_end();
3019 migration_release_from_dst_file(ms
);
3020 rcu_unregister_thread();
3024 static int open_return_path_on_source(MigrationState
*ms
,
3027 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
3028 if (!ms
->rp_state
.from_dst_file
) {
3032 trace_open_return_path_on_source();
3034 if (!create_thread
) {
3039 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
3040 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
3041 ms
->rp_state
.rp_thread_created
= true;
3043 trace_open_return_path_on_source_continue();
3048 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
3049 static int await_return_path_close_on_source(MigrationState
*ms
)
3052 * If this is a normal exit then the destination will send a SHUT and the
3053 * rp_thread will exit, however if there's an error we need to cause
3056 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
3058 * shutdown(2), if we have it, will cause it to unblock if it's stuck
3059 * waiting for the destination.
3061 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
3062 mark_source_rp_bad(ms
);
3064 trace_await_return_path_close_on_source_joining();
3065 qemu_thread_join(&ms
->rp_state
.rp_thread
);
3066 ms
->rp_state
.rp_thread_created
= false;
3067 trace_await_return_path_close_on_source_close();
3068 return ms
->rp_state
.error
;
3072 * Switch from normal iteration to postcopy
3073 * Returns non-0 on error
3075 static int postcopy_start(MigrationState
*ms
)
3078 QIOChannelBuffer
*bioc
;
3080 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3081 int64_t bandwidth
= migrate_max_postcopy_bandwidth();
3082 bool restart_block
= false;
3083 int cur_state
= MIGRATION_STATUS_ACTIVE
;
3085 if (postcopy_preempt_wait_channel(ms
)) {
3086 migrate_set_state(&ms
->state
, ms
->state
, MIGRATION_STATUS_FAILED
);
3090 if (!migrate_pause_before_switchover()) {
3091 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
3092 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3095 trace_postcopy_start();
3096 qemu_mutex_lock_iothread();
3097 trace_postcopy_start_set_run();
3099 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3100 global_state_store();
3101 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
3106 ret
= migration_maybe_pause(ms
, &cur_state
,
3107 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3112 ret
= bdrv_inactivate_all();
3116 restart_block
= true;
3119 * Cause any non-postcopiable, but iterative devices to
3120 * send out their final data.
3122 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
3125 * in Finish migrate and with the io-lock held everything should
3126 * be quiet, but we've potentially still got dirty pages and we
3127 * need to tell the destination to throw any pages it's already received
3130 if (migrate_postcopy_ram()) {
3131 ram_postcopy_send_discard_bitmap(ms
);
3135 * send rest of state - note things that are doing postcopy
3136 * will notice we're in POSTCOPY_ACTIVE and not actually
3137 * wrap their state up here
3139 /* 0 max-postcopy-bandwidth means unlimited */
3141 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
3143 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
3145 if (migrate_postcopy_ram()) {
3146 /* Ping just for debugging, helps line traces up */
3147 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
3151 * While loading the device state we may trigger page transfer
3152 * requests and the fd must be free to process those, and thus
3153 * the destination must read the whole device state off the fd before
3154 * it starts processing it. Unfortunately the ad-hoc migration format
3155 * doesn't allow the destination to know the size to read without fully
3156 * parsing it through each devices load-state code (especially the open
3157 * coded devices that use get/put).
3158 * So we wrap the device state up in a package with a length at the start;
3159 * to do this we use a qemu_buf to hold the whole of the device state.
3161 bioc
= qio_channel_buffer_new(4096);
3162 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
3163 fb
= qemu_file_new_output(QIO_CHANNEL(bioc
));
3164 object_unref(OBJECT(bioc
));
3167 * Make sure the receiver can get incoming pages before we send the rest
3170 qemu_savevm_send_postcopy_listen(fb
);
3172 qemu_savevm_state_complete_precopy(fb
, false, false);
3173 if (migrate_postcopy_ram()) {
3174 qemu_savevm_send_ping(fb
, 3);
3177 qemu_savevm_send_postcopy_run(fb
);
3179 /* <><> end of stuff going into the package */
3181 /* Last point of recovery; as soon as we send the package the destination
3182 * can open devices and potentially start running.
3183 * Lets just check again we've not got any errors.
3185 ret
= qemu_file_get_error(ms
->to_dst_file
);
3187 error_report("postcopy_start: Migration stream errored (pre package)");
3191 restart_block
= false;
3193 /* Now send that blob */
3194 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
3199 /* Send a notify to give a chance for anything that needs to happen
3200 * at the transition to postcopy and after the device state; in particular
3201 * spice needs to trigger a transition now
3203 ms
->postcopy_after_devices
= true;
3204 notifier_list_notify(&migration_state_notifiers
, ms
);
3206 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
3208 qemu_mutex_unlock_iothread();
3210 if (migrate_postcopy_ram()) {
3212 * Although this ping is just for debug, it could potentially be
3213 * used for getting a better measurement of downtime at the source.
3215 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
3218 if (migrate_release_ram()) {
3219 ram_postcopy_migrated_memory_release(ms
);
3222 ret
= qemu_file_get_error(ms
->to_dst_file
);
3224 error_report("postcopy_start: Migration stream errored");
3225 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
3226 MIGRATION_STATUS_FAILED
);
3229 trace_postcopy_preempt_enabled(migrate_postcopy_preempt());
3236 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
3237 MIGRATION_STATUS_FAILED
);
3238 if (restart_block
) {
3239 /* A failure happened early enough that we know the destination hasn't
3240 * accessed block devices, so we're safe to recover.
3242 Error
*local_err
= NULL
;
3244 bdrv_activate_all(&local_err
);
3246 error_report_err(local_err
);
3249 qemu_mutex_unlock_iothread();
3254 * migration_maybe_pause: Pause if required to by
3255 * migrate_pause_before_switchover called with the iothread locked
3256 * Returns: 0 on success
3258 static int migration_maybe_pause(MigrationState
*s
,
3259 int *current_active_state
,
3262 if (!migrate_pause_before_switchover()) {
3266 /* Since leaving this state is not atomic with posting the semaphore
3267 * it's possible that someone could have issued multiple migrate_continue
3268 * and the semaphore is incorrectly positive at this point;
3269 * the docs say it's undefined to reinit a semaphore that's already
3270 * init'd, so use timedwait to eat up any existing posts.
3272 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
3273 /* This block intentionally left blank */
3277 * If the migration is cancelled when it is in the completion phase,
3278 * the migration state is set to MIGRATION_STATUS_CANCELLING.
3279 * So we don't need to wait a semaphore, otherwise we would always
3280 * wait for the 'pause_sem' semaphore.
3282 if (s
->state
!= MIGRATION_STATUS_CANCELLING
) {
3283 qemu_mutex_unlock_iothread();
3284 migrate_set_state(&s
->state
, *current_active_state
,
3285 MIGRATION_STATUS_PRE_SWITCHOVER
);
3286 qemu_sem_wait(&s
->pause_sem
);
3287 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
3289 *current_active_state
= new_state
;
3290 qemu_mutex_lock_iothread();
3293 return s
->state
== new_state
? 0 : -EINVAL
;
3297 * migration_completion: Used by migration_thread when there's not much left.
3298 * The caller 'breaks' the loop when this returns.
3300 * @s: Current migration state
3302 static void migration_completion(MigrationState
*s
)
3305 int current_active_state
= s
->state
;
3307 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
3308 qemu_mutex_lock_iothread();
3309 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3310 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3311 s
->vm_was_running
= runstate_is_running();
3312 ret
= global_state_store();
3315 bool inactivate
= !migrate_colo_enabled();
3316 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
3317 trace_migration_completion_vm_stop(ret
);
3319 ret
= migration_maybe_pause(s
, ¤t_active_state
,
3320 MIGRATION_STATUS_DEVICE
);
3323 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
3324 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
3327 if (inactivate
&& ret
>= 0) {
3328 s
->block_inactive
= true;
3331 qemu_mutex_unlock_iothread();
3336 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3337 trace_migration_completion_postcopy_end();
3339 qemu_mutex_lock_iothread();
3340 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
3341 qemu_mutex_unlock_iothread();
3343 /* Shutdown the postcopy fast path thread */
3344 if (migrate_postcopy_preempt()) {
3345 postcopy_preempt_shutdown_file(s
);
3348 trace_migration_completion_postcopy_end_after_complete();
3354 * If rp was opened we must clean up the thread before
3355 * cleaning everything else up (since if there are no failures
3356 * it will wait for the destination to send it's status in
3359 if (s
->rp_state
.rp_thread_created
) {
3361 trace_migration_return_path_end_before();
3362 rp_error
= await_return_path_close_on_source(s
);
3363 trace_migration_return_path_end_after(rp_error
);
3365 goto fail_invalidate
;
3369 if (qemu_file_get_error(s
->to_dst_file
)) {
3370 trace_migration_completion_file_err();
3371 goto fail_invalidate
;
3374 if (migrate_colo_enabled() && s
->state
== MIGRATION_STATUS_ACTIVE
) {
3375 /* COLO does not support postcopy */
3376 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
3377 MIGRATION_STATUS_COLO
);
3379 migrate_set_state(&s
->state
, current_active_state
,
3380 MIGRATION_STATUS_COMPLETED
);
3386 /* If not doing postcopy, vm_start() will be called: let's regain
3387 * control on images.
3389 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
3390 s
->state
== MIGRATION_STATUS_DEVICE
) {
3391 Error
*local_err
= NULL
;
3393 qemu_mutex_lock_iothread();
3394 bdrv_activate_all(&local_err
);
3396 error_report_err(local_err
);
3398 s
->block_inactive
= false;
3400 qemu_mutex_unlock_iothread();
3404 migrate_set_state(&s
->state
, current_active_state
,
3405 MIGRATION_STATUS_FAILED
);
3409 * bg_migration_completion: Used by bg_migration_thread when after all the
3410 * RAM has been saved. The caller 'breaks' the loop when this returns.
3412 * @s: Current migration state
3414 static void bg_migration_completion(MigrationState
*s
)
3416 int current_active_state
= s
->state
;
3419 * Stop tracking RAM writes - un-protect memory, un-register UFFD
3420 * memory ranges, flush kernel wait queues and wake up threads
3421 * waiting for write fault to be resolved.
3423 ram_write_tracking_stop();
3425 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
3427 * By this moment we have RAM content saved into the migration stream.
3428 * The next step is to flush the non-RAM content (device state)
3429 * right after the ram content. The device state has been stored into
3430 * the temporary buffer before RAM saving started.
3432 qemu_put_buffer(s
->to_dst_file
, s
->bioc
->data
, s
->bioc
->usage
);
3433 qemu_fflush(s
->to_dst_file
);
3434 } else if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
3438 if (qemu_file_get_error(s
->to_dst_file
)) {
3439 trace_migration_completion_file_err();
3443 migrate_set_state(&s
->state
, current_active_state
,
3444 MIGRATION_STATUS_COMPLETED
);
3448 migrate_set_state(&s
->state
, current_active_state
,
3449 MIGRATION_STATUS_FAILED
);
3452 bool migrate_colo_enabled(void)
3454 MigrationState
*s
= migrate_get_current();
3455 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
3458 typedef enum MigThrError
{
3459 /* No error detected */
3460 MIG_THR_ERR_NONE
= 0,
3461 /* Detected error, but resumed successfully */
3462 MIG_THR_ERR_RECOVERED
= 1,
3463 /* Detected fatal error, need to exit */
3464 MIG_THR_ERR_FATAL
= 2,
3467 static int postcopy_resume_handshake(MigrationState
*s
)
3469 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
3471 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3472 qemu_sem_wait(&s
->rp_state
.rp_sem
);
3475 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3482 /* Return zero if success, or <0 for error */
3483 static int postcopy_do_resume(MigrationState
*s
)
3488 * Call all the resume_prepare() hooks, so that modules can be
3489 * ready for the migration resume.
3491 ret
= qemu_savevm_state_resume_prepare(s
);
3493 error_report("%s: resume_prepare() failure detected: %d",
3499 * Last handshake with destination on the resume (destination will
3500 * switch to postcopy-active afterwards)
3502 ret
= postcopy_resume_handshake(s
);
3504 error_report("%s: handshake failed: %d", __func__
, ret
);
3512 * We don't return until we are in a safe state to continue current
3513 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
3514 * MIG_THR_ERR_FATAL if unrecovery failure happened.
3516 static MigThrError
postcopy_pause(MigrationState
*s
)
3518 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3524 * Current channel is possibly broken. Release it. Note that this is
3525 * guaranteed even without lock because to_dst_file should only be
3526 * modified by the migration thread. That also guarantees that the
3527 * unregister of yank is safe too without the lock. It should be safe
3528 * even to be within the qemu_file_lock, but we didn't do that to avoid
3529 * taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make
3530 * the qemu_file_lock critical section as small as possible.
3532 assert(s
->to_dst_file
);
3533 migration_ioc_unregister_yank_from_file(s
->to_dst_file
);
3534 qemu_mutex_lock(&s
->qemu_file_lock
);
3535 file
= s
->to_dst_file
;
3536 s
->to_dst_file
= NULL
;
3537 qemu_mutex_unlock(&s
->qemu_file_lock
);
3539 qemu_file_shutdown(file
);
3543 * Do the same to postcopy fast path socket too if there is. No
3544 * locking needed because no racer as long as we do this before setting
3547 if (s
->postcopy_qemufile_src
) {
3548 migration_ioc_unregister_yank_from_file(s
->postcopy_qemufile_src
);
3549 qemu_file_shutdown(s
->postcopy_qemufile_src
);
3550 qemu_fclose(s
->postcopy_qemufile_src
);
3551 s
->postcopy_qemufile_src
= NULL
;
3554 migrate_set_state(&s
->state
, s
->state
,
3555 MIGRATION_STATUS_POSTCOPY_PAUSED
);
3557 error_report("Detected IO failure for postcopy. "
3558 "Migration paused.");
3561 * We wait until things fixed up. Then someone will setup the
3562 * status back for us.
3564 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
3565 qemu_sem_wait(&s
->postcopy_pause_sem
);
3568 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3569 /* Woken up by a recover procedure. Give it a shot */
3571 if (postcopy_preempt_wait_channel(s
)) {
3573 * Preempt enabled, and new channel create failed; loop
3574 * back to wait for another recovery.
3580 * Firstly, let's wake up the return path now, with a new
3581 * return path channel.
3583 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
3585 /* Do the resume logic */
3586 if (postcopy_do_resume(s
) == 0) {
3587 /* Let's continue! */
3588 trace_postcopy_pause_continued();
3589 return MIG_THR_ERR_RECOVERED
;
3592 * Something wrong happened during the recovery, let's
3593 * pause again. Pause is always better than throwing
3599 /* This is not right... Time to quit. */
3600 return MIG_THR_ERR_FATAL
;
3605 static MigThrError
migration_detect_error(MigrationState
*s
)
3608 int state
= s
->state
;
3609 Error
*local_error
= NULL
;
3611 if (state
== MIGRATION_STATUS_CANCELLING
||
3612 state
== MIGRATION_STATUS_CANCELLED
) {
3613 /* End the migration, but don't set the state to failed */
3614 return MIG_THR_ERR_FATAL
;
3618 * Try to detect any file errors. Note that postcopy_qemufile_src will
3619 * be NULL when postcopy preempt is not enabled.
3621 ret
= qemu_file_get_error_obj_any(s
->to_dst_file
,
3622 s
->postcopy_qemufile_src
,
3625 /* Everything is fine */
3626 assert(!local_error
);
3627 return MIG_THR_ERR_NONE
;
3631 migrate_set_error(s
, local_error
);
3632 error_free(local_error
);
3635 if (state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
) {
3637 * For postcopy, we allow the network to be down for a
3638 * while. After that, it can be continued by a
3641 return postcopy_pause(s
);
3644 * For precopy (or postcopy with error outside IO), we fail
3647 migrate_set_state(&s
->state
, state
, MIGRATION_STATUS_FAILED
);
3648 trace_migration_thread_file_err();
3650 /* Time to stop the migration, now. */
3651 return MIG_THR_ERR_FATAL
;
3655 /* How many bytes have we transferred since the beginning of the migration */
3656 static uint64_t migration_total_bytes(MigrationState
*s
)
3658 return qemu_file_total_transferred(s
->to_dst_file
) +
3659 ram_counters
.multifd_bytes
;
3662 static void migration_calculate_complete(MigrationState
*s
)
3664 uint64_t bytes
= migration_total_bytes(s
);
3665 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3666 int64_t transfer_time
;
3668 s
->total_time
= end_time
- s
->start_time
;
3671 * It's still not set, so we are precopy migration. For
3672 * postcopy, downtime is calculated during postcopy_start().
3674 s
->downtime
= end_time
- s
->downtime_start
;
3677 transfer_time
= s
->total_time
- s
->setup_time
;
3678 if (transfer_time
) {
3679 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
3683 static void update_iteration_initial_status(MigrationState
*s
)
3686 * Update these three fields at the same time to avoid mismatch info lead
3687 * wrong speed calculation.
3689 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3690 s
->iteration_initial_bytes
= migration_total_bytes(s
);
3691 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
3694 static void migration_update_counters(MigrationState
*s
,
3695 int64_t current_time
)
3697 uint64_t transferred
, transferred_pages
, time_spent
;
3698 uint64_t current_bytes
; /* bytes transferred since the beginning */
3701 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
3705 current_bytes
= migration_total_bytes(s
);
3706 transferred
= current_bytes
- s
->iteration_initial_bytes
;
3707 time_spent
= current_time
- s
->iteration_start_time
;
3708 bandwidth
= (double)transferred
/ time_spent
;
3709 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
3711 s
->mbps
= (((double) transferred
* 8.0) /
3712 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
3714 transferred_pages
= ram_get_total_transferred_pages() -
3715 s
->iteration_initial_pages
;
3716 s
->pages_per_second
= (double) transferred_pages
/
3717 (((double) time_spent
/ 1000.0));
3720 * if we haven't sent anything, we don't want to
3721 * recalculate. 10000 is a small enough number for our purposes
3723 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
3724 s
->expected_downtime
= ram_counters
.remaining
/ bandwidth
;
3727 qemu_file_reset_rate_limit(s
->to_dst_file
);
3729 update_iteration_initial_status(s
);
3731 trace_migrate_transferred(transferred
, time_spent
,
3732 bandwidth
, s
->threshold_size
);
3735 /* Migration thread iteration status */
3737 MIG_ITERATE_RESUME
, /* Resume current iteration */
3738 MIG_ITERATE_SKIP
, /* Skip current iteration */
3739 MIG_ITERATE_BREAK
, /* Break the loop */
3743 * Return true if continue to the next iteration directly, false
3746 static MigIterateState
migration_iteration_run(MigrationState
*s
)
3748 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
3749 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
3751 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
3752 &pend_compat
, &pend_post
);
3753 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
3755 trace_migrate_pending(pending_size
, s
->threshold_size
,
3756 pend_pre
, pend_compat
, pend_post
);
3758 if (pending_size
&& pending_size
>= s
->threshold_size
) {
3759 /* Still a significant amount to transfer */
3760 if (!in_postcopy
&& pend_pre
<= s
->threshold_size
&&
3761 qatomic_read(&s
->start_postcopy
)) {
3762 if (postcopy_start(s
)) {
3763 error_report("%s: postcopy failed to start", __func__
);
3765 return MIG_ITERATE_SKIP
;
3767 /* Just another iteration step */
3768 qemu_savevm_state_iterate(s
->to_dst_file
, in_postcopy
);
3770 trace_migration_thread_low_pending(pending_size
);
3771 migration_completion(s
);
3772 return MIG_ITERATE_BREAK
;
3775 return MIG_ITERATE_RESUME
;
3778 static void migration_iteration_finish(MigrationState
*s
)
3780 /* If we enabled cpu throttling for auto-converge, turn it off. */
3781 cpu_throttle_stop();
3783 qemu_mutex_lock_iothread();
3785 case MIGRATION_STATUS_COMPLETED
:
3786 migration_calculate_complete(s
);
3787 runstate_set(RUN_STATE_POSTMIGRATE
);
3789 case MIGRATION_STATUS_COLO
:
3790 if (!migrate_colo_enabled()) {
3791 error_report("%s: critical error: calling COLO code without "
3792 "COLO enabled", __func__
);
3794 migrate_start_colo_process(s
);
3795 s
->vm_was_running
= true;
3797 case MIGRATION_STATUS_FAILED
:
3798 case MIGRATION_STATUS_CANCELLED
:
3799 case MIGRATION_STATUS_CANCELLING
:
3800 if (s
->vm_was_running
) {
3801 if (!runstate_check(RUN_STATE_SHUTDOWN
)) {
3805 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
3806 runstate_set(RUN_STATE_POSTMIGRATE
);
3812 /* Should not reach here, but if so, forgive the VM. */
3813 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3816 migrate_fd_cleanup_schedule(s
);
3817 qemu_mutex_unlock_iothread();
3820 static void bg_migration_iteration_finish(MigrationState
*s
)
3822 qemu_mutex_lock_iothread();
3824 case MIGRATION_STATUS_COMPLETED
:
3825 migration_calculate_complete(s
);
3828 case MIGRATION_STATUS_ACTIVE
:
3829 case MIGRATION_STATUS_FAILED
:
3830 case MIGRATION_STATUS_CANCELLED
:
3831 case MIGRATION_STATUS_CANCELLING
:
3835 /* Should not reach here, but if so, forgive the VM. */
3836 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3840 migrate_fd_cleanup_schedule(s
);
3841 qemu_mutex_unlock_iothread();
3845 * Return true if continue to the next iteration directly, false
3848 static MigIterateState
bg_migration_iteration_run(MigrationState
*s
)
3852 res
= qemu_savevm_state_iterate(s
->to_dst_file
, false);
3854 bg_migration_completion(s
);
3855 return MIG_ITERATE_BREAK
;
3858 return MIG_ITERATE_RESUME
;
3861 void migration_make_urgent_request(void)
3863 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
3866 void migration_consume_urgent_request(void)
3868 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
3871 /* Returns true if the rate limiting was broken by an urgent request */
3872 bool migration_rate_limit(void)
3874 int64_t now
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3875 MigrationState
*s
= migrate_get_current();
3877 bool urgent
= false;
3878 migration_update_counters(s
, now
);
3879 if (qemu_file_rate_limit(s
->to_dst_file
)) {
3881 if (qemu_file_get_error(s
->to_dst_file
)) {
3885 * Wait for a delay to do rate limiting OR
3886 * something urgent to post the semaphore.
3888 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- now
;
3889 trace_migration_rate_limit_pre(ms
);
3890 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
3892 * We were woken by one or more urgent things but
3893 * the timedwait will have consumed one of them.
3894 * The service routine for the urgent wake will dec
3895 * the semaphore itself for each item it consumes,
3896 * so add this one we just eat back.
3898 qemu_sem_post(&s
->rate_limit_sem
);
3901 trace_migration_rate_limit_post(urgent
);
3907 * if failover devices are present, wait they are completely
3911 static void qemu_savevm_wait_unplug(MigrationState
*s
, int old_state
,
3914 if (qemu_savevm_state_guest_unplug_pending()) {
3915 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_WAIT_UNPLUG
);
3917 while (s
->state
== MIGRATION_STATUS_WAIT_UNPLUG
&&
3918 qemu_savevm_state_guest_unplug_pending()) {
3919 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
3921 if (s
->state
!= MIGRATION_STATUS_WAIT_UNPLUG
) {
3922 int timeout
= 120; /* 30 seconds */
3924 * migration has been canceled
3925 * but as we have started an unplug we must wait the end
3926 * to be able to plug back the card
3928 while (timeout
-- && qemu_savevm_state_guest_unplug_pending()) {
3929 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
3931 if (qemu_savevm_state_guest_unplug_pending() &&
3933 warn_report("migration: partially unplugged device on "
3938 migrate_set_state(&s
->state
, MIGRATION_STATUS_WAIT_UNPLUG
, new_state
);
3940 migrate_set_state(&s
->state
, old_state
, new_state
);
3945 * Master migration thread on the source VM.
3946 * It drives the migration and pumps the data down the outgoing channel.
3948 static void *migration_thread(void *opaque
)
3950 MigrationState
*s
= opaque
;
3951 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3952 MigThrError thr_error
;
3953 bool urgent
= false;
3955 rcu_register_thread();
3957 object_ref(OBJECT(s
));
3958 update_iteration_initial_status(s
);
3960 qemu_savevm_state_header(s
->to_dst_file
);
3963 * If we opened the return path, we need to make sure dst has it
3966 if (s
->rp_state
.rp_thread_created
) {
3967 /* Now tell the dest that it should open its end so it can reply */
3968 qemu_savevm_send_open_return_path(s
->to_dst_file
);
3970 /* And do a ping that will make stuff easier to debug */
3971 qemu_savevm_send_ping(s
->to_dst_file
, 1);
3974 if (migrate_postcopy()) {
3976 * Tell the destination that we *might* want to do postcopy later;
3977 * if the other end can't do postcopy it should fail now, nice and
3980 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
3983 if (migrate_colo_enabled()) {
3984 /* Notify migration destination that we enable COLO */
3985 qemu_savevm_send_colo_enable(s
->to_dst_file
);
3988 qemu_savevm_state_setup(s
->to_dst_file
);
3990 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
3991 MIGRATION_STATUS_ACTIVE
);
3993 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3995 trace_migration_thread_setup_complete();
3997 while (migration_is_active(s
)) {
3998 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
3999 MigIterateState iter_state
= migration_iteration_run(s
);
4000 if (iter_state
== MIG_ITERATE_SKIP
) {
4002 } else if (iter_state
== MIG_ITERATE_BREAK
) {
4008 * Try to detect any kind of failures, and see whether we
4009 * should stop the migration now.
4011 thr_error
= migration_detect_error(s
);
4012 if (thr_error
== MIG_THR_ERR_FATAL
) {
4013 /* Stop migration */
4015 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
4017 * Just recovered from a e.g. network failure, reset all
4018 * the local variables. This is important to avoid
4019 * breaking transferred_bytes and bandwidth calculation
4021 update_iteration_initial_status(s
);
4024 urgent
= migration_rate_limit();
4027 trace_migration_thread_after_loop();
4028 migration_iteration_finish(s
);
4029 object_unref(OBJECT(s
));
4030 rcu_unregister_thread();
4034 static void bg_migration_vm_start_bh(void *opaque
)
4036 MigrationState
*s
= opaque
;
4038 qemu_bh_delete(s
->vm_start_bh
);
4039 s
->vm_start_bh
= NULL
;
4042 s
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - s
->downtime_start
;
4046 * Background snapshot thread, based on live migration code.
4047 * This is an alternative implementation of live migration mechanism
4048 * introduced specifically to support background snapshots.
4050 * It takes advantage of userfault_fd write protection mechanism introduced
4051 * in v5.7 kernel. Compared to existing dirty page logging migration much
4052 * lesser stream traffic is produced resulting in smaller snapshot images,
4053 * simply cause of no page duplicates can get into the stream.
4055 * Another key point is that generated vmstate stream reflects machine state
4056 * 'frozen' at the beginning of snapshot creation compared to dirty page logging
4057 * mechanism, which effectively results in that saved snapshot is the state of VM
4058 * at the end of the process.
4060 static void *bg_migration_thread(void *opaque
)
4062 MigrationState
*s
= opaque
;
4063 int64_t setup_start
;
4064 MigThrError thr_error
;
4066 bool early_fail
= true;
4068 rcu_register_thread();
4069 object_ref(OBJECT(s
));
4071 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
4073 setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
4075 * We want to save vmstate for the moment when migration has been
4076 * initiated but also we want to save RAM content while VM is running.
4077 * The RAM content should appear first in the vmstate. So, we first
4078 * stash the non-RAM part of the vmstate to the temporary buffer,
4079 * then write RAM part of the vmstate to the migration stream
4080 * with vCPUs running and, finally, write stashed non-RAM part of
4081 * the vmstate from the buffer to the migration stream.
4083 s
->bioc
= qio_channel_buffer_new(512 * 1024);
4084 qio_channel_set_name(QIO_CHANNEL(s
->bioc
), "vmstate-buffer");
4085 fb
= qemu_file_new_output(QIO_CHANNEL(s
->bioc
));
4086 object_unref(OBJECT(s
->bioc
));
4088 update_iteration_initial_status(s
);
4091 * Prepare for tracking memory writes with UFFD-WP - populate
4092 * RAM pages before protecting.
4095 ram_write_tracking_prepare();
4098 qemu_savevm_state_header(s
->to_dst_file
);
4099 qemu_savevm_state_setup(s
->to_dst_file
);
4101 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
4102 MIGRATION_STATUS_ACTIVE
);
4104 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
4106 trace_migration_thread_setup_complete();
4107 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
4109 qemu_mutex_lock_iothread();
4112 * If VM is currently in suspended state, then, to make a valid runstate
4113 * transition in vm_stop_force_state() we need to wakeup it up.
4115 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
4116 s
->vm_was_running
= runstate_is_running();
4118 if (global_state_store()) {
4121 /* Forcibly stop VM before saving state of vCPUs and devices */
4122 if (vm_stop_force_state(RUN_STATE_PAUSED
)) {
4126 * Put vCPUs in sync with shadow context structures, then
4127 * save their state to channel-buffer along with devices.
4129 cpu_synchronize_all_states();
4130 if (qemu_savevm_state_complete_precopy_non_iterable(fb
, false, false)) {
4134 * Since we are going to get non-iterable state data directly
4135 * from s->bioc->data, explicit flush is needed here.
4139 /* Now initialize UFFD context and start tracking RAM writes */
4140 if (ram_write_tracking_start()) {
4146 * Start VM from BH handler to avoid write-fault lock here.
4147 * UFFD-WP protection for the whole RAM is already enabled so
4148 * calling VM state change notifiers from vm_start() would initiate
4149 * writes to virtio VQs memory which is in write-protected region.
4151 s
->vm_start_bh
= qemu_bh_new(bg_migration_vm_start_bh
, s
);
4152 qemu_bh_schedule(s
->vm_start_bh
);
4154 qemu_mutex_unlock_iothread();
4156 while (migration_is_active(s
)) {
4157 MigIterateState iter_state
= bg_migration_iteration_run(s
);
4158 if (iter_state
== MIG_ITERATE_SKIP
) {
4160 } else if (iter_state
== MIG_ITERATE_BREAK
) {
4165 * Try to detect any kind of failures, and see whether we
4166 * should stop the migration now.
4168 thr_error
= migration_detect_error(s
);
4169 if (thr_error
== MIG_THR_ERR_FATAL
) {
4170 /* Stop migration */
4174 migration_update_counters(s
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
));
4177 trace_migration_thread_after_loop();
4181 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
4182 MIGRATION_STATUS_FAILED
);
4183 qemu_mutex_unlock_iothread();
4186 bg_migration_iteration_finish(s
);
4189 object_unref(OBJECT(s
));
4190 rcu_unregister_thread();
4195 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
4197 Error
*local_err
= NULL
;
4199 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
4202 * If there's a previous error, free it and prepare for another one.
4203 * Meanwhile if migration completes successfully, there won't have an error
4204 * dumped when calling migrate_fd_cleanup().
4206 migrate_error_free(s
);
4208 s
->expected_downtime
= s
->parameters
.downtime_limit
;
4210 assert(s
->cleanup_bh
);
4212 assert(!s
->cleanup_bh
);
4213 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup_bh
, s
);
4216 migrate_fd_error(s
, error_in
);
4219 * Don't do cleanup for resume if channel is invalid, but only dump
4220 * the error. We wait for another channel connect from the user.
4221 * The error_report still gives HMP user a hint on what failed.
4222 * It's normally done in migrate_fd_cleanup(), but call it here
4225 error_report_err(error_copy(s
->error
));
4227 migrate_fd_cleanup(s
);
4233 /* This is a resumed migration */
4234 rate_limit
= s
->parameters
.max_postcopy_bandwidth
/
4237 /* This is a fresh new migration */
4238 rate_limit
= s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
;
4240 /* Notify before starting migration thread */
4241 notifier_list_notify(&migration_state_notifiers
, s
);
4244 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
4245 qemu_file_set_blocking(s
->to_dst_file
, true);
4248 * Open the return path. For postcopy, it is used exclusively. For
4249 * precopy, only if user specified "return-path" capability would
4250 * QEMU uses the return path.
4252 if (migrate_postcopy_ram() || migrate_use_return_path()) {
4253 if (open_return_path_on_source(s
, !resume
)) {
4254 error_report("Unable to open return-path for postcopy");
4255 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
4256 migrate_fd_cleanup(s
);
4261 /* This needs to be done before resuming a postcopy */
4262 if (postcopy_preempt_setup(s
, &local_err
)) {
4263 error_report_err(local_err
);
4264 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
4265 MIGRATION_STATUS_FAILED
);
4266 migrate_fd_cleanup(s
);
4271 /* Wakeup the main migration thread to do the recovery */
4272 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
4273 MIGRATION_STATUS_POSTCOPY_RECOVER
);
4274 qemu_sem_post(&s
->postcopy_pause_sem
);
4278 if (multifd_save_setup(&local_err
) != 0) {
4279 error_report_err(local_err
);
4280 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
4281 MIGRATION_STATUS_FAILED
);
4282 migrate_fd_cleanup(s
);
4286 if (migrate_background_snapshot()) {
4287 qemu_thread_create(&s
->thread
, "bg_snapshot",
4288 bg_migration_thread
, s
, QEMU_THREAD_JOINABLE
);
4290 qemu_thread_create(&s
->thread
, "live_migration",
4291 migration_thread
, s
, QEMU_THREAD_JOINABLE
);
4293 s
->migration_thread_running
= true;
4296 void migration_global_dump(Monitor
*mon
)
4298 MigrationState
*ms
= migrate_get_current();
4300 monitor_printf(mon
, "globals:\n");
4301 monitor_printf(mon
, "store-global-state: %s\n",
4302 ms
->store_global_state
? "on" : "off");
4303 monitor_printf(mon
, "only-migratable: %s\n",
4304 only_migratable
? "on" : "off");
4305 monitor_printf(mon
, "send-configuration: %s\n",
4306 ms
->send_configuration
? "on" : "off");
4307 monitor_printf(mon
, "send-section-footer: %s\n",
4308 ms
->send_section_footer
? "on" : "off");
4309 monitor_printf(mon
, "decompress-error-check: %s\n",
4310 ms
->decompress_error_check
? "on" : "off");
4311 monitor_printf(mon
, "clear-bitmap-shift: %u\n",
4312 ms
->clear_bitmap_shift
);
4315 #define DEFINE_PROP_MIG_CAP(name, x) \
4316 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
4318 static Property migration_properties
[] = {
4319 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
4320 store_global_state
, true),
4321 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
4322 send_configuration
, true),
4323 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
4324 send_section_footer
, true),
4325 DEFINE_PROP_BOOL("decompress-error-check", MigrationState
,
4326 decompress_error_check
, true),
4327 DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState
,
4328 clear_bitmap_shift
, CLEAR_BITMAP_SHIFT_DEFAULT
),
4330 /* Migration parameters */
4331 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
4332 parameters
.compress_level
,
4333 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
4334 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
4335 parameters
.compress_threads
,
4336 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
4337 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState
,
4338 parameters
.compress_wait_thread
, true),
4339 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
4340 parameters
.decompress_threads
,
4341 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
4342 DEFINE_PROP_UINT8("x-throttle-trigger-threshold", MigrationState
,
4343 parameters
.throttle_trigger_threshold
,
4344 DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD
),
4345 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
4346 parameters
.cpu_throttle_initial
,
4347 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
4348 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
4349 parameters
.cpu_throttle_increment
,
4350 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
4351 DEFINE_PROP_BOOL("x-cpu-throttle-tailslow", MigrationState
,
4352 parameters
.cpu_throttle_tailslow
, false),
4353 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
4354 parameters
.max_bandwidth
, MAX_THROTTLE
),
4355 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
4356 parameters
.downtime_limit
,
4357 DEFAULT_MIGRATE_SET_DOWNTIME
),
4358 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
4359 parameters
.x_checkpoint_delay
,
4360 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
4361 DEFINE_PROP_UINT8("multifd-channels", MigrationState
,
4362 parameters
.multifd_channels
,
4363 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
4364 DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState
,
4365 parameters
.multifd_compression
,
4366 DEFAULT_MIGRATE_MULTIFD_COMPRESSION
),
4367 DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState
,
4368 parameters
.multifd_zlib_level
,
4369 DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL
),
4370 DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState
,
4371 parameters
.multifd_zstd_level
,
4372 DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL
),
4373 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
4374 parameters
.xbzrle_cache_size
,
4375 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
4376 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState
,
4377 parameters
.max_postcopy_bandwidth
,
4378 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH
),
4379 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState
,
4380 parameters
.max_cpu_throttle
,
4381 DEFAULT_MIGRATE_MAX_CPU_THROTTLE
),
4382 DEFINE_PROP_SIZE("announce-initial", MigrationState
,
4383 parameters
.announce_initial
,
4384 DEFAULT_MIGRATE_ANNOUNCE_INITIAL
),
4385 DEFINE_PROP_SIZE("announce-max", MigrationState
,
4386 parameters
.announce_max
,
4387 DEFAULT_MIGRATE_ANNOUNCE_MAX
),
4388 DEFINE_PROP_SIZE("announce-rounds", MigrationState
,
4389 parameters
.announce_rounds
,
4390 DEFAULT_MIGRATE_ANNOUNCE_ROUNDS
),
4391 DEFINE_PROP_SIZE("announce-step", MigrationState
,
4392 parameters
.announce_step
,
4393 DEFAULT_MIGRATE_ANNOUNCE_STEP
),
4394 DEFINE_PROP_BOOL("x-postcopy-preempt-break-huge", MigrationState
,
4395 postcopy_preempt_break_huge
, true),
4396 DEFINE_PROP_STRING("tls-creds", MigrationState
, parameters
.tls_creds
),
4397 DEFINE_PROP_STRING("tls-hostname", MigrationState
, parameters
.tls_hostname
),
4398 DEFINE_PROP_STRING("tls-authz", MigrationState
, parameters
.tls_authz
),
4400 /* Migration capabilities */
4401 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
4402 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
4403 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
4404 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
4405 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
4406 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
4407 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
4408 DEFINE_PROP_MIG_CAP("x-postcopy-preempt",
4409 MIGRATION_CAPABILITY_POSTCOPY_PREEMPT
),
4410 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
4411 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
4412 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
4413 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
4414 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD
),
4415 DEFINE_PROP_MIG_CAP("x-background-snapshot",
4416 MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
),
4418 DEFINE_PROP_MIG_CAP("x-zero-copy-send",
4419 MIGRATION_CAPABILITY_ZERO_COPY_SEND
),
4422 DEFINE_PROP_END_OF_LIST(),
4425 static void migration_class_init(ObjectClass
*klass
, void *data
)
4427 DeviceClass
*dc
= DEVICE_CLASS(klass
);
4429 dc
->user_creatable
= false;
4430 device_class_set_props(dc
, migration_properties
);
4433 static void migration_instance_finalize(Object
*obj
)
4435 MigrationState
*ms
= MIGRATION_OBJ(obj
);
4437 qemu_mutex_destroy(&ms
->error_mutex
);
4438 qemu_mutex_destroy(&ms
->qemu_file_lock
);
4439 qemu_sem_destroy(&ms
->wait_unplug_sem
);
4440 qemu_sem_destroy(&ms
->rate_limit_sem
);
4441 qemu_sem_destroy(&ms
->pause_sem
);
4442 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
4443 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
4444 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
4445 qemu_sem_destroy(&ms
->postcopy_qemufile_src_sem
);
4446 error_free(ms
->error
);
4449 static void migration_instance_init(Object
*obj
)
4451 MigrationState
*ms
= MIGRATION_OBJ(obj
);
4452 MigrationParameters
*params
= &ms
->parameters
;
4454 ms
->state
= MIGRATION_STATUS_NONE
;
4456 ms
->pages_per_second
= -1;
4457 qemu_sem_init(&ms
->pause_sem
, 0);
4458 qemu_mutex_init(&ms
->error_mutex
);
4460 params
->tls_hostname
= g_strdup("");
4461 params
->tls_creds
= g_strdup("");
4463 /* Set has_* up only for parameter checks */
4464 params
->has_compress_level
= true;
4465 params
->has_compress_threads
= true;
4466 params
->has_compress_wait_thread
= true;
4467 params
->has_decompress_threads
= true;
4468 params
->has_throttle_trigger_threshold
= true;
4469 params
->has_cpu_throttle_initial
= true;
4470 params
->has_cpu_throttle_increment
= true;
4471 params
->has_cpu_throttle_tailslow
= true;
4472 params
->has_max_bandwidth
= true;
4473 params
->has_downtime_limit
= true;
4474 params
->has_x_checkpoint_delay
= true;
4475 params
->has_block_incremental
= true;
4476 params
->has_multifd_channels
= true;
4477 params
->has_multifd_compression
= true;
4478 params
->has_multifd_zlib_level
= true;
4479 params
->has_multifd_zstd_level
= true;
4480 params
->has_xbzrle_cache_size
= true;
4481 params
->has_max_postcopy_bandwidth
= true;
4482 params
->has_max_cpu_throttle
= true;
4483 params
->has_announce_initial
= true;
4484 params
->has_announce_max
= true;
4485 params
->has_announce_rounds
= true;
4486 params
->has_announce_step
= true;
4488 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
4489 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
4490 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
4491 qemu_sem_init(&ms
->rate_limit_sem
, 0);
4492 qemu_sem_init(&ms
->wait_unplug_sem
, 0);
4493 qemu_sem_init(&ms
->postcopy_qemufile_src_sem
, 0);
4494 qemu_mutex_init(&ms
->qemu_file_lock
);
4498 * Return true if check pass, false otherwise. Error will be put
4499 * inside errp if provided.
4501 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
4503 MigrationCapabilityStatusList
*head
= NULL
;
4504 /* Assuming all off */
4505 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
4508 if (!migrate_params_check(&ms
->parameters
, errp
)) {
4512 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
4513 if (ms
->enabled_capabilities
[i
]) {
4514 QAPI_LIST_PREPEND(head
, migrate_cap_add(i
, true));
4518 ret
= migrate_caps_check(cap_list
, head
, errp
);
4520 /* It works with head == NULL */
4521 qapi_free_MigrationCapabilityStatusList(head
);
4526 static const TypeInfo migration_type
= {
4527 .name
= TYPE_MIGRATION
,
4529 * NOTE: TYPE_MIGRATION is not really a device, as the object is
4530 * not created using qdev_new(), it is not attached to the qdev
4531 * device tree, and it is never realized.
4533 * TODO: Make this TYPE_OBJECT once QOM provides something like
4534 * TYPE_DEVICE's "-global" properties.
4536 .parent
= TYPE_DEVICE
,
4537 .class_init
= migration_class_init
,
4538 .class_size
= sizeof(MigrationClass
),
4539 .instance_size
= sizeof(MigrationState
),
4540 .instance_init
= migration_instance_init
,
4541 .instance_finalize
= migration_instance_finalize
,
4544 static void register_migration_types(void)
4546 type_register_static(&migration_type
);
4549 type_init(register_migration_types
);