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
->has_tls_creds
= true;
922 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
923 params
->has_tls_hostname
= true;
924 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
925 params
->has_tls_authz
= true;
926 params
->tls_authz
= g_strdup(s
->parameters
.tls_authz
?
927 s
->parameters
.tls_authz
: "");
928 params
->has_max_bandwidth
= true;
929 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
930 params
->has_downtime_limit
= true;
931 params
->downtime_limit
= s
->parameters
.downtime_limit
;
932 params
->has_x_checkpoint_delay
= true;
933 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
934 params
->has_block_incremental
= true;
935 params
->block_incremental
= s
->parameters
.block_incremental
;
936 params
->has_multifd_channels
= true;
937 params
->multifd_channels
= s
->parameters
.multifd_channels
;
938 params
->has_multifd_compression
= true;
939 params
->multifd_compression
= s
->parameters
.multifd_compression
;
940 params
->has_multifd_zlib_level
= true;
941 params
->multifd_zlib_level
= s
->parameters
.multifd_zlib_level
;
942 params
->has_multifd_zstd_level
= true;
943 params
->multifd_zstd_level
= s
->parameters
.multifd_zstd_level
;
944 params
->has_xbzrle_cache_size
= true;
945 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
946 params
->has_max_postcopy_bandwidth
= true;
947 params
->max_postcopy_bandwidth
= s
->parameters
.max_postcopy_bandwidth
;
948 params
->has_max_cpu_throttle
= true;
949 params
->max_cpu_throttle
= s
->parameters
.max_cpu_throttle
;
950 params
->has_announce_initial
= true;
951 params
->announce_initial
= s
->parameters
.announce_initial
;
952 params
->has_announce_max
= true;
953 params
->announce_max
= s
->parameters
.announce_max
;
954 params
->has_announce_rounds
= true;
955 params
->announce_rounds
= s
->parameters
.announce_rounds
;
956 params
->has_announce_step
= true;
957 params
->announce_step
= s
->parameters
.announce_step
;
959 if (s
->parameters
.has_block_bitmap_mapping
) {
960 params
->has_block_bitmap_mapping
= true;
961 params
->block_bitmap_mapping
=
962 QAPI_CLONE(BitmapMigrationNodeAliasList
,
963 s
->parameters
.block_bitmap_mapping
);
969 AnnounceParameters
*migrate_announce_params(void)
971 static AnnounceParameters ap
;
973 MigrationState
*s
= migrate_get_current();
975 ap
.initial
= s
->parameters
.announce_initial
;
976 ap
.max
= s
->parameters
.announce_max
;
977 ap
.rounds
= s
->parameters
.announce_rounds
;
978 ap
.step
= s
->parameters
.announce_step
;
984 * Return true if we're already in the middle of a migration
985 * (i.e. any of the active or setup states)
987 bool migration_is_setup_or_active(int state
)
990 case MIGRATION_STATUS_ACTIVE
:
991 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
992 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
993 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
994 case MIGRATION_STATUS_SETUP
:
995 case MIGRATION_STATUS_PRE_SWITCHOVER
:
996 case MIGRATION_STATUS_DEVICE
:
997 case MIGRATION_STATUS_WAIT_UNPLUG
:
998 case MIGRATION_STATUS_COLO
:
1007 bool migration_is_running(int state
)
1010 case MIGRATION_STATUS_ACTIVE
:
1011 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1012 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1013 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1014 case MIGRATION_STATUS_SETUP
:
1015 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1016 case MIGRATION_STATUS_DEVICE
:
1017 case MIGRATION_STATUS_WAIT_UNPLUG
:
1018 case MIGRATION_STATUS_CANCELLING
:
1027 static void populate_time_info(MigrationInfo
*info
, MigrationState
*s
)
1029 info
->has_status
= true;
1030 info
->has_setup_time
= true;
1031 info
->setup_time
= s
->setup_time
;
1032 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
1033 info
->has_total_time
= true;
1034 info
->total_time
= s
->total_time
;
1035 info
->has_downtime
= true;
1036 info
->downtime
= s
->downtime
;
1038 info
->has_total_time
= true;
1039 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) -
1041 info
->has_expected_downtime
= true;
1042 info
->expected_downtime
= s
->expected_downtime
;
1046 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
1048 size_t page_size
= qemu_target_page_size();
1050 info
->has_ram
= true;
1051 info
->ram
= g_malloc0(sizeof(*info
->ram
));
1052 info
->ram
->transferred
= ram_counters
.transferred
;
1053 info
->ram
->total
= ram_bytes_total();
1054 info
->ram
->duplicate
= ram_counters
.duplicate
;
1055 /* legacy value. It is not used anymore */
1056 info
->ram
->skipped
= 0;
1057 info
->ram
->normal
= ram_counters
.normal
;
1058 info
->ram
->normal_bytes
= ram_counters
.normal
* page_size
;
1059 info
->ram
->mbps
= s
->mbps
;
1060 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
1061 info
->ram
->dirty_sync_missed_zero_copy
=
1062 ram_counters
.dirty_sync_missed_zero_copy
;
1063 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
1064 info
->ram
->page_size
= page_size
;
1065 info
->ram
->multifd_bytes
= ram_counters
.multifd_bytes
;
1066 info
->ram
->pages_per_second
= s
->pages_per_second
;
1067 info
->ram
->precopy_bytes
= ram_counters
.precopy_bytes
;
1068 info
->ram
->downtime_bytes
= ram_counters
.downtime_bytes
;
1069 info
->ram
->postcopy_bytes
= ram_counters
.postcopy_bytes
;
1071 if (migrate_use_xbzrle()) {
1072 info
->has_xbzrle_cache
= true;
1073 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
1074 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
1075 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
1076 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
1077 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
1078 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
1079 info
->xbzrle_cache
->encoding_rate
= xbzrle_counters
.encoding_rate
;
1080 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
1083 if (migrate_use_compression()) {
1084 info
->has_compression
= true;
1085 info
->compression
= g_malloc0(sizeof(*info
->compression
));
1086 info
->compression
->pages
= compression_counters
.pages
;
1087 info
->compression
->busy
= compression_counters
.busy
;
1088 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
1089 info
->compression
->compressed_size
=
1090 compression_counters
.compressed_size
;
1091 info
->compression
->compression_rate
=
1092 compression_counters
.compression_rate
;
1095 if (cpu_throttle_active()) {
1096 info
->has_cpu_throttle_percentage
= true;
1097 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
1100 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
1101 info
->ram
->remaining
= ram_bytes_remaining();
1102 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
1106 static void populate_disk_info(MigrationInfo
*info
)
1108 if (blk_mig_active()) {
1109 info
->has_disk
= true;
1110 info
->disk
= g_malloc0(sizeof(*info
->disk
));
1111 info
->disk
->transferred
= blk_mig_bytes_transferred();
1112 info
->disk
->remaining
= blk_mig_bytes_remaining();
1113 info
->disk
->total
= blk_mig_bytes_total();
1117 static void fill_source_migration_info(MigrationInfo
*info
)
1119 MigrationState
*s
= migrate_get_current();
1120 int state
= qatomic_read(&s
->state
);
1121 GSList
*cur_blocker
= migration_blockers
;
1123 info
->blocked_reasons
= NULL
;
1126 * There are two types of reasons a migration might be blocked;
1127 * a) devices marked in VMState as non-migratable, and
1128 * b) Explicit migration blockers
1129 * We need to add both of them here.
1131 qemu_savevm_non_migratable_list(&info
->blocked_reasons
);
1133 while (cur_blocker
) {
1134 QAPI_LIST_PREPEND(info
->blocked_reasons
,
1135 g_strdup(error_get_pretty(cur_blocker
->data
)));
1136 cur_blocker
= g_slist_next(cur_blocker
);
1138 info
->has_blocked_reasons
= info
->blocked_reasons
!= NULL
;
1141 case MIGRATION_STATUS_NONE
:
1142 /* no migration has happened ever */
1143 /* do not overwrite destination migration status */
1145 case MIGRATION_STATUS_SETUP
:
1146 info
->has_status
= true;
1147 info
->has_total_time
= false;
1149 case MIGRATION_STATUS_ACTIVE
:
1150 case MIGRATION_STATUS_CANCELLING
:
1151 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1152 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1153 case MIGRATION_STATUS_DEVICE
:
1154 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1155 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1156 /* TODO add some postcopy stats */
1157 populate_time_info(info
, s
);
1158 populate_ram_info(info
, s
);
1159 populate_disk_info(info
);
1160 populate_vfio_info(info
);
1162 case MIGRATION_STATUS_COLO
:
1163 info
->has_status
= true;
1164 /* TODO: display COLO specific information (checkpoint info etc.) */
1166 case MIGRATION_STATUS_COMPLETED
:
1167 populate_time_info(info
, s
);
1168 populate_ram_info(info
, s
);
1169 populate_vfio_info(info
);
1171 case MIGRATION_STATUS_FAILED
:
1172 info
->has_status
= true;
1174 info
->has_error_desc
= true;
1175 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
1178 case MIGRATION_STATUS_CANCELLED
:
1179 info
->has_status
= true;
1181 case MIGRATION_STATUS_WAIT_UNPLUG
:
1182 info
->has_status
= true;
1185 info
->status
= state
;
1188 typedef enum WriteTrackingSupport
{
1189 WT_SUPPORT_UNKNOWN
= 0,
1191 WT_SUPPORT_AVAILABLE
,
1192 WT_SUPPORT_COMPATIBLE
1193 } WriteTrackingSupport
;
1196 WriteTrackingSupport
migrate_query_write_tracking(void)
1198 /* Check if kernel supports required UFFD features */
1199 if (!ram_write_tracking_available()) {
1200 return WT_SUPPORT_ABSENT
;
1203 * Check if current memory configuration is
1204 * compatible with required UFFD features.
1206 if (!ram_write_tracking_compatible()) {
1207 return WT_SUPPORT_AVAILABLE
;
1210 return WT_SUPPORT_COMPATIBLE
;
1214 * @migration_caps_check - check capability validity
1216 * @cap_list: old capability list, array of bool
1217 * @params: new capabilities to be applied soon
1218 * @errp: set *errp if the check failed, with reason
1220 * Returns true if check passed, otherwise false.
1222 static bool migrate_caps_check(bool *cap_list
,
1223 MigrationCapabilityStatusList
*params
,
1226 MigrationCapabilityStatusList
*cap
;
1227 bool old_postcopy_cap
;
1228 MigrationIncomingState
*mis
= migration_incoming_get_current();
1230 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1232 for (cap
= params
; cap
; cap
= cap
->next
) {
1233 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
1236 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1237 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
1238 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
1240 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
1245 #ifndef CONFIG_REPLICATION
1246 if (cap_list
[MIGRATION_CAPABILITY_X_COLO
]) {
1247 error_setg(errp
, "QEMU compiled without replication module"
1248 " can't enable COLO");
1249 error_append_hint(errp
, "Please enable replication before COLO.\n");
1254 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
1255 /* This check is reasonably expensive, so only when it's being
1256 * set the first time, also it's only the destination that needs
1259 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
1260 !postcopy_ram_supported_by_host(mis
)) {
1261 /* postcopy_ram_supported_by_host will have emitted a more
1264 error_setg(errp
, "Postcopy is not supported");
1268 if (cap_list
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
]) {
1269 error_setg(errp
, "Postcopy is not compatible with ignore-shared");
1274 if (cap_list
[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
]) {
1275 WriteTrackingSupport wt_support
;
1278 * Check if 'background-snapshot' capability is supported by
1279 * host kernel and compatible with guest memory configuration.
1281 wt_support
= migrate_query_write_tracking();
1282 if (wt_support
< WT_SUPPORT_AVAILABLE
) {
1283 error_setg(errp
, "Background-snapshot is not supported by host kernel");
1286 if (wt_support
< WT_SUPPORT_COMPATIBLE
) {
1287 error_setg(errp
, "Background-snapshot is not compatible "
1288 "with guest memory configuration");
1293 * Check if there are any migration capabilities
1294 * incompatible with 'background-snapshot'.
1296 for (idx
= 0; idx
< check_caps_background_snapshot
.size
; idx
++) {
1297 int incomp_cap
= check_caps_background_snapshot
.caps
[idx
];
1298 if (cap_list
[incomp_cap
]) {
1300 "Background-snapshot is not compatible with %s",
1301 MigrationCapability_str(incomp_cap
));
1308 if (cap_list
[MIGRATION_CAPABILITY_ZERO_COPY_SEND
] &&
1309 (!cap_list
[MIGRATION_CAPABILITY_MULTIFD
] ||
1310 cap_list
[MIGRATION_CAPABILITY_COMPRESS
] ||
1311 cap_list
[MIGRATION_CAPABILITY_XBZRLE
] ||
1312 migrate_multifd_compression() ||
1313 migrate_use_tls())) {
1315 "Zero copy only available for non-compressed non-TLS multifd migration");
1319 if (cap_list
[MIGRATION_CAPABILITY_ZERO_COPY_SEND
]) {
1321 "Zero copy currently only available on Linux");
1327 /* incoming side only */
1328 if (runstate_check(RUN_STATE_INMIGRATE
) &&
1329 !migrate_multi_channels_is_allowed() &&
1330 cap_list
[MIGRATION_CAPABILITY_MULTIFD
]) {
1331 error_setg(errp
, "multifd is not supported by current protocol");
1335 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_PREEMPT
]) {
1336 if (!cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
1337 error_setg(errp
, "Postcopy preempt requires postcopy-ram");
1345 static void fill_destination_migration_info(MigrationInfo
*info
)
1347 MigrationIncomingState
*mis
= migration_incoming_get_current();
1349 if (mis
->socket_address_list
) {
1350 info
->has_socket_address
= true;
1351 info
->socket_address
=
1352 QAPI_CLONE(SocketAddressList
, mis
->socket_address_list
);
1355 switch (mis
->state
) {
1356 case MIGRATION_STATUS_NONE
:
1358 case MIGRATION_STATUS_SETUP
:
1359 case MIGRATION_STATUS_CANCELLING
:
1360 case MIGRATION_STATUS_CANCELLED
:
1361 case MIGRATION_STATUS_ACTIVE
:
1362 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1363 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1364 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1365 case MIGRATION_STATUS_FAILED
:
1366 case MIGRATION_STATUS_COLO
:
1367 info
->has_status
= true;
1369 case MIGRATION_STATUS_COMPLETED
:
1370 info
->has_status
= true;
1371 fill_destination_postcopy_migration_info(info
);
1374 info
->status
= mis
->state
;
1377 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1379 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1381 fill_destination_migration_info(info
);
1382 fill_source_migration_info(info
);
1387 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
1390 MigrationState
*s
= migrate_get_current();
1391 MigrationCapabilityStatusList
*cap
;
1392 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
1394 if (migration_is_running(s
->state
)) {
1395 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1399 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
1400 if (!migrate_caps_check(cap_list
, params
, errp
)) {
1404 for (cap
= params
; cap
; cap
= cap
->next
) {
1405 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
1410 * Check whether the parameters are valid. Error will be put into errp
1411 * (if provided). Return true if valid, otherwise false.
1413 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
1415 if (params
->has_compress_level
&&
1416 (params
->compress_level
> 9)) {
1417 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
1418 "a value between 0 and 9");
1422 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
1423 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1425 "a value between 1 and 255");
1429 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
1430 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1431 "decompress_threads",
1432 "a value between 1 and 255");
1436 if (params
->has_throttle_trigger_threshold
&&
1437 (params
->throttle_trigger_threshold
< 1 ||
1438 params
->throttle_trigger_threshold
> 100)) {
1439 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1440 "throttle_trigger_threshold",
1441 "an integer in the range of 1 to 100");
1445 if (params
->has_cpu_throttle_initial
&&
1446 (params
->cpu_throttle_initial
< 1 ||
1447 params
->cpu_throttle_initial
> 99)) {
1448 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1449 "cpu_throttle_initial",
1450 "an integer in the range of 1 to 99");
1454 if (params
->has_cpu_throttle_increment
&&
1455 (params
->cpu_throttle_increment
< 1 ||
1456 params
->cpu_throttle_increment
> 99)) {
1457 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1458 "cpu_throttle_increment",
1459 "an integer in the range of 1 to 99");
1463 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
1464 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1466 "an integer in the range of 0 to "stringify(SIZE_MAX
)
1471 if (params
->has_downtime_limit
&&
1472 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
1473 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1475 "an integer in the range of 0 to "
1476 stringify(MAX_MIGRATE_DOWNTIME
)" ms");
1480 /* x_checkpoint_delay is now always positive */
1482 if (params
->has_multifd_channels
&& (params
->multifd_channels
< 1)) {
1483 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1485 "a value between 1 and 255");
1489 if (params
->has_multifd_zlib_level
&&
1490 (params
->multifd_zlib_level
> 9)) {
1491 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zlib_level",
1492 "a value between 0 and 9");
1496 if (params
->has_multifd_zstd_level
&&
1497 (params
->multifd_zstd_level
> 20)) {
1498 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zstd_level",
1499 "a value between 0 and 20");
1503 if (params
->has_xbzrle_cache_size
&&
1504 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
1505 !is_power_of_2(params
->xbzrle_cache_size
))) {
1506 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1507 "xbzrle_cache_size",
1508 "a power of two no less than the target page size");
1512 if (params
->has_max_cpu_throttle
&&
1513 (params
->max_cpu_throttle
< params
->cpu_throttle_initial
||
1514 params
->max_cpu_throttle
> 99)) {
1515 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1517 "an integer in the range of cpu_throttle_initial to 99");
1521 if (params
->has_announce_initial
&&
1522 params
->announce_initial
> 100000) {
1523 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1525 "a value between 0 and 100000");
1528 if (params
->has_announce_max
&&
1529 params
->announce_max
> 100000) {
1530 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1532 "a value between 0 and 100000");
1535 if (params
->has_announce_rounds
&&
1536 params
->announce_rounds
> 1000) {
1537 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1539 "a value between 0 and 1000");
1542 if (params
->has_announce_step
&&
1543 (params
->announce_step
< 1 ||
1544 params
->announce_step
> 10000)) {
1545 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1547 "a value between 0 and 10000");
1551 if (params
->has_block_bitmap_mapping
&&
1552 !check_dirty_bitmap_mig_alias_map(params
->block_bitmap_mapping
, errp
)) {
1553 error_prepend(errp
, "Invalid mapping given for block-bitmap-mapping: ");
1558 if (migrate_use_zero_copy_send() &&
1559 ((params
->has_multifd_compression
&& params
->multifd_compression
) ||
1560 (params
->has_tls_creds
&& params
->tls_creds
&& *params
->tls_creds
))) {
1562 "Zero copy only available for non-compressed non-TLS multifd migration");
1570 static void migrate_params_test_apply(MigrateSetParameters
*params
,
1571 MigrationParameters
*dest
)
1573 *dest
= migrate_get_current()->parameters
;
1575 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1577 if (params
->has_compress_level
) {
1578 dest
->compress_level
= params
->compress_level
;
1581 if (params
->has_compress_threads
) {
1582 dest
->compress_threads
= params
->compress_threads
;
1585 if (params
->has_compress_wait_thread
) {
1586 dest
->compress_wait_thread
= params
->compress_wait_thread
;
1589 if (params
->has_decompress_threads
) {
1590 dest
->decompress_threads
= params
->decompress_threads
;
1593 if (params
->has_throttle_trigger_threshold
) {
1594 dest
->throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1597 if (params
->has_cpu_throttle_initial
) {
1598 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
1601 if (params
->has_cpu_throttle_increment
) {
1602 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
1605 if (params
->has_cpu_throttle_tailslow
) {
1606 dest
->cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1609 if (params
->has_tls_creds
) {
1610 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1611 dest
->tls_creds
= params
->tls_creds
->u
.s
;
1614 if (params
->has_tls_hostname
) {
1615 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1616 dest
->tls_hostname
= params
->tls_hostname
->u
.s
;
1619 if (params
->has_max_bandwidth
) {
1620 dest
->max_bandwidth
= params
->max_bandwidth
;
1623 if (params
->has_downtime_limit
) {
1624 dest
->downtime_limit
= params
->downtime_limit
;
1627 if (params
->has_x_checkpoint_delay
) {
1628 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
1631 if (params
->has_block_incremental
) {
1632 dest
->block_incremental
= params
->block_incremental
;
1634 if (params
->has_multifd_channels
) {
1635 dest
->multifd_channels
= params
->multifd_channels
;
1637 if (params
->has_multifd_compression
) {
1638 dest
->multifd_compression
= params
->multifd_compression
;
1640 if (params
->has_xbzrle_cache_size
) {
1641 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
1643 if (params
->has_max_postcopy_bandwidth
) {
1644 dest
->max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1646 if (params
->has_max_cpu_throttle
) {
1647 dest
->max_cpu_throttle
= params
->max_cpu_throttle
;
1649 if (params
->has_announce_initial
) {
1650 dest
->announce_initial
= params
->announce_initial
;
1652 if (params
->has_announce_max
) {
1653 dest
->announce_max
= params
->announce_max
;
1655 if (params
->has_announce_rounds
) {
1656 dest
->announce_rounds
= params
->announce_rounds
;
1658 if (params
->has_announce_step
) {
1659 dest
->announce_step
= params
->announce_step
;
1662 if (params
->has_block_bitmap_mapping
) {
1663 dest
->has_block_bitmap_mapping
= true;
1664 dest
->block_bitmap_mapping
= params
->block_bitmap_mapping
;
1668 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
1670 MigrationState
*s
= migrate_get_current();
1672 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1674 if (params
->has_compress_level
) {
1675 s
->parameters
.compress_level
= params
->compress_level
;
1678 if (params
->has_compress_threads
) {
1679 s
->parameters
.compress_threads
= params
->compress_threads
;
1682 if (params
->has_compress_wait_thread
) {
1683 s
->parameters
.compress_wait_thread
= params
->compress_wait_thread
;
1686 if (params
->has_decompress_threads
) {
1687 s
->parameters
.decompress_threads
= params
->decompress_threads
;
1690 if (params
->has_throttle_trigger_threshold
) {
1691 s
->parameters
.throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1694 if (params
->has_cpu_throttle_initial
) {
1695 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
1698 if (params
->has_cpu_throttle_increment
) {
1699 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
1702 if (params
->has_cpu_throttle_tailslow
) {
1703 s
->parameters
.cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1706 if (params
->has_tls_creds
) {
1707 g_free(s
->parameters
.tls_creds
);
1708 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1709 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1712 if (params
->has_tls_hostname
) {
1713 g_free(s
->parameters
.tls_hostname
);
1714 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1715 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1718 if (params
->has_tls_authz
) {
1719 g_free(s
->parameters
.tls_authz
);
1720 assert(params
->tls_authz
->type
== QTYPE_QSTRING
);
1721 s
->parameters
.tls_authz
= g_strdup(params
->tls_authz
->u
.s
);
1724 if (params
->has_max_bandwidth
) {
1725 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1726 if (s
->to_dst_file
&& !migration_in_postcopy()) {
1727 qemu_file_set_rate_limit(s
->to_dst_file
,
1728 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1732 if (params
->has_downtime_limit
) {
1733 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1736 if (params
->has_x_checkpoint_delay
) {
1737 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1738 if (migration_in_colo_state()) {
1739 colo_checkpoint_notify(s
);
1743 if (params
->has_block_incremental
) {
1744 s
->parameters
.block_incremental
= params
->block_incremental
;
1746 if (params
->has_multifd_channels
) {
1747 s
->parameters
.multifd_channels
= params
->multifd_channels
;
1749 if (params
->has_multifd_compression
) {
1750 s
->parameters
.multifd_compression
= params
->multifd_compression
;
1752 if (params
->has_xbzrle_cache_size
) {
1753 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1754 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1756 if (params
->has_max_postcopy_bandwidth
) {
1757 s
->parameters
.max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1758 if (s
->to_dst_file
&& migration_in_postcopy()) {
1759 qemu_file_set_rate_limit(s
->to_dst_file
,
1760 s
->parameters
.max_postcopy_bandwidth
/ XFER_LIMIT_RATIO
);
1763 if (params
->has_max_cpu_throttle
) {
1764 s
->parameters
.max_cpu_throttle
= params
->max_cpu_throttle
;
1766 if (params
->has_announce_initial
) {
1767 s
->parameters
.announce_initial
= params
->announce_initial
;
1769 if (params
->has_announce_max
) {
1770 s
->parameters
.announce_max
= params
->announce_max
;
1772 if (params
->has_announce_rounds
) {
1773 s
->parameters
.announce_rounds
= params
->announce_rounds
;
1775 if (params
->has_announce_step
) {
1776 s
->parameters
.announce_step
= params
->announce_step
;
1779 if (params
->has_block_bitmap_mapping
) {
1780 qapi_free_BitmapMigrationNodeAliasList(
1781 s
->parameters
.block_bitmap_mapping
);
1783 s
->parameters
.has_block_bitmap_mapping
= true;
1784 s
->parameters
.block_bitmap_mapping
=
1785 QAPI_CLONE(BitmapMigrationNodeAliasList
,
1786 params
->block_bitmap_mapping
);
1790 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1792 MigrationParameters tmp
;
1794 /* TODO Rewrite "" to null instead */
1795 if (params
->has_tls_creds
1796 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1797 qobject_unref(params
->tls_creds
->u
.n
);
1798 params
->tls_creds
->type
= QTYPE_QSTRING
;
1799 params
->tls_creds
->u
.s
= strdup("");
1801 /* TODO Rewrite "" to null instead */
1802 if (params
->has_tls_hostname
1803 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1804 qobject_unref(params
->tls_hostname
->u
.n
);
1805 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1806 params
->tls_hostname
->u
.s
= strdup("");
1809 migrate_params_test_apply(params
, &tmp
);
1811 if (!migrate_params_check(&tmp
, errp
)) {
1812 /* Invalid parameter */
1816 migrate_params_apply(params
, errp
);
1820 void qmp_migrate_start_postcopy(Error
**errp
)
1822 MigrationState
*s
= migrate_get_current();
1824 if (!migrate_postcopy()) {
1825 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1826 " the start of migration");
1830 if (s
->state
== MIGRATION_STATUS_NONE
) {
1831 error_setg(errp
, "Postcopy must be started after migration has been"
1836 * we don't error if migration has finished since that would be racy
1837 * with issuing this command.
1839 qatomic_set(&s
->start_postcopy
, true);
1842 /* shared migration helpers */
1844 void migrate_set_state(int *state
, int old_state
, int new_state
)
1846 assert(new_state
< MIGRATION_STATUS__MAX
);
1847 if (qatomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1848 trace_migrate_set_state(MigrationStatus_str(new_state
));
1849 migrate_generate_event(new_state
);
1853 static MigrationCapabilityStatus
*migrate_cap_add(MigrationCapability index
,
1856 MigrationCapabilityStatus
*cap
;
1858 cap
= g_new0(MigrationCapabilityStatus
, 1);
1859 cap
->capability
= index
;
1865 void migrate_set_block_enabled(bool value
, Error
**errp
)
1867 MigrationCapabilityStatusList
*cap
= NULL
;
1869 QAPI_LIST_PREPEND(cap
, migrate_cap_add(MIGRATION_CAPABILITY_BLOCK
, value
));
1870 qmp_migrate_set_capabilities(cap
, errp
);
1871 qapi_free_MigrationCapabilityStatusList(cap
);
1874 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1876 s
->parameters
.block_incremental
= value
;
1879 static void block_cleanup_parameters(MigrationState
*s
)
1881 if (s
->must_remove_block_options
) {
1882 /* setting to false can never fail */
1883 migrate_set_block_enabled(false, &error_abort
);
1884 migrate_set_block_incremental(s
, false);
1885 s
->must_remove_block_options
= false;
1889 static void migrate_fd_cleanup(MigrationState
*s
)
1891 qemu_bh_delete(s
->cleanup_bh
);
1892 s
->cleanup_bh
= NULL
;
1894 g_free(s
->hostname
);
1897 qemu_savevm_state_cleanup();
1899 if (s
->to_dst_file
) {
1902 trace_migrate_fd_cleanup();
1903 qemu_mutex_unlock_iothread();
1904 if (s
->migration_thread_running
) {
1905 qemu_thread_join(&s
->thread
);
1906 s
->migration_thread_running
= false;
1908 qemu_mutex_lock_iothread();
1910 multifd_save_cleanup();
1911 qemu_mutex_lock(&s
->qemu_file_lock
);
1912 tmp
= s
->to_dst_file
;
1913 s
->to_dst_file
= NULL
;
1914 qemu_mutex_unlock(&s
->qemu_file_lock
);
1916 * Close the file handle without the lock to make sure the
1917 * critical section won't block for long.
1919 migration_ioc_unregister_yank_from_file(tmp
);
1923 if (s
->postcopy_qemufile_src
) {
1924 migration_ioc_unregister_yank_from_file(s
->postcopy_qemufile_src
);
1925 qemu_fclose(s
->postcopy_qemufile_src
);
1926 s
->postcopy_qemufile_src
= NULL
;
1929 assert(!migration_is_active(s
));
1931 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1932 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1933 MIGRATION_STATUS_CANCELLED
);
1937 /* It is used on info migrate. We can't free it */
1938 error_report_err(error_copy(s
->error
));
1940 notifier_list_notify(&migration_state_notifiers
, s
);
1941 block_cleanup_parameters(s
);
1942 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1945 static void migrate_fd_cleanup_schedule(MigrationState
*s
)
1948 * Ref the state for bh, because it may be called when
1949 * there're already no other refs
1951 object_ref(OBJECT(s
));
1952 qemu_bh_schedule(s
->cleanup_bh
);
1955 static void migrate_fd_cleanup_bh(void *opaque
)
1957 MigrationState
*s
= opaque
;
1958 migrate_fd_cleanup(s
);
1959 object_unref(OBJECT(s
));
1962 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1964 QEMU_LOCK_GUARD(&s
->error_mutex
);
1966 s
->error
= error_copy(error
);
1970 static void migrate_error_free(MigrationState
*s
)
1972 QEMU_LOCK_GUARD(&s
->error_mutex
);
1974 error_free(s
->error
);
1979 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1981 trace_migrate_fd_error(error_get_pretty(error
));
1982 assert(s
->to_dst_file
== NULL
);
1983 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1984 MIGRATION_STATUS_FAILED
);
1985 migrate_set_error(s
, error
);
1988 static void migrate_fd_cancel(MigrationState
*s
)
1991 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1992 trace_migrate_fd_cancel();
1994 WITH_QEMU_LOCK_GUARD(&s
->qemu_file_lock
) {
1995 if (s
->rp_state
.from_dst_file
) {
1996 /* shutdown the rp socket, so causing the rp thread to shutdown */
1997 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
2002 old_state
= s
->state
;
2003 if (!migration_is_running(old_state
)) {
2006 /* If the migration is paused, kick it out of the pause */
2007 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
2008 qemu_sem_post(&s
->pause_sem
);
2010 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
2011 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
2014 * If we're unlucky the migration code might be stuck somewhere in a
2015 * send/write while the network has failed and is waiting to timeout;
2016 * if we've got shutdown(2) available then we can force it to quit.
2017 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
2018 * called in a bh, so there is no race against this cancel.
2020 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
2021 qemu_file_shutdown(f
);
2023 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
2024 Error
*local_err
= NULL
;
2026 bdrv_activate_all(&local_err
);
2028 error_report_err(local_err
);
2030 s
->block_inactive
= false;
2035 void add_migration_state_change_notifier(Notifier
*notify
)
2037 notifier_list_add(&migration_state_notifiers
, notify
);
2040 void remove_migration_state_change_notifier(Notifier
*notify
)
2042 notifier_remove(notify
);
2045 bool migration_in_setup(MigrationState
*s
)
2047 return s
->state
== MIGRATION_STATUS_SETUP
;
2050 bool migration_has_finished(MigrationState
*s
)
2052 return s
->state
== MIGRATION_STATUS_COMPLETED
;
2055 bool migration_has_failed(MigrationState
*s
)
2057 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
2058 s
->state
== MIGRATION_STATUS_FAILED
);
2061 bool migration_in_postcopy(void)
2063 MigrationState
*s
= migrate_get_current();
2066 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
2067 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
2068 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
2075 bool migration_in_postcopy_after_devices(MigrationState
*s
)
2077 return migration_in_postcopy() && s
->postcopy_after_devices
;
2080 bool migration_in_incoming_postcopy(void)
2082 PostcopyState ps
= postcopy_state_get();
2084 return ps
>= POSTCOPY_INCOMING_DISCARD
&& ps
< POSTCOPY_INCOMING_END
;
2087 bool migration_in_bg_snapshot(void)
2089 MigrationState
*s
= migrate_get_current();
2091 return migrate_background_snapshot() &&
2092 migration_is_setup_or_active(s
->state
);
2095 bool migration_is_idle(void)
2097 MigrationState
*s
= current_migration
;
2104 case MIGRATION_STATUS_NONE
:
2105 case MIGRATION_STATUS_CANCELLED
:
2106 case MIGRATION_STATUS_COMPLETED
:
2107 case MIGRATION_STATUS_FAILED
:
2109 case MIGRATION_STATUS_SETUP
:
2110 case MIGRATION_STATUS_CANCELLING
:
2111 case MIGRATION_STATUS_ACTIVE
:
2112 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
2113 case MIGRATION_STATUS_COLO
:
2114 case MIGRATION_STATUS_PRE_SWITCHOVER
:
2115 case MIGRATION_STATUS_DEVICE
:
2116 case MIGRATION_STATUS_WAIT_UNPLUG
:
2118 case MIGRATION_STATUS__MAX
:
2119 g_assert_not_reached();
2125 bool migration_is_active(MigrationState
*s
)
2127 return (s
->state
== MIGRATION_STATUS_ACTIVE
||
2128 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2131 void migrate_init(MigrationState
*s
)
2134 * Reinitialise all migration state, except
2135 * parameters/capabilities that the user set, and
2140 s
->to_dst_file
= NULL
;
2141 s
->state
= MIGRATION_STATUS_NONE
;
2142 s
->rp_state
.from_dst_file
= NULL
;
2143 s
->rp_state
.error
= false;
2145 s
->pages_per_second
= 0.0;
2147 s
->expected_downtime
= 0;
2149 s
->start_postcopy
= false;
2150 s
->postcopy_after_devices
= false;
2151 s
->migration_thread_running
= false;
2152 error_free(s
->error
);
2156 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
2158 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2160 s
->vm_was_running
= false;
2161 s
->iteration_initial_bytes
= 0;
2162 s
->threshold_size
= 0;
2165 int migrate_add_blocker_internal(Error
*reason
, Error
**errp
)
2167 /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */
2168 if (runstate_check(RUN_STATE_SAVE_VM
) || !migration_is_idle()) {
2169 error_propagate_prepend(errp
, error_copy(reason
),
2170 "disallowing migration blocker "
2171 "(migration/snapshot in progress) for: ");
2175 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
2179 int migrate_add_blocker(Error
*reason
, Error
**errp
)
2181 if (only_migratable
) {
2182 error_propagate_prepend(errp
, error_copy(reason
),
2183 "disallowing migration blocker "
2184 "(--only-migratable) for: ");
2188 return migrate_add_blocker_internal(reason
, errp
);
2191 void migrate_del_blocker(Error
*reason
)
2193 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
2196 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
2198 Error
*local_err
= NULL
;
2199 static bool once
= true;
2202 error_setg(errp
, "The incoming migration has already been started");
2205 if (!runstate_check(RUN_STATE_INMIGRATE
)) {
2206 error_setg(errp
, "'-incoming' was not specified on the command line");
2210 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
2214 qemu_start_incoming_migration(uri
, &local_err
);
2217 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2218 error_propagate(errp
, local_err
);
2225 void qmp_migrate_recover(const char *uri
, Error
**errp
)
2227 MigrationIncomingState
*mis
= migration_incoming_get_current();
2230 * Don't even bother to use ERRP_GUARD() as it _must_ always be set by
2231 * callers (no one should ignore a recover failure); if there is, it's a
2232 * programming error.
2236 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2237 error_setg(errp
, "Migrate recover can only be run "
2238 "when postcopy is paused.");
2242 /* If there's an existing transport, release it */
2243 migration_incoming_transport_cleanup(mis
);
2246 * Note that this call will never start a real migration; it will
2247 * only re-setup the migration stream and poke existing migration
2248 * to continue using that newly established channel.
2250 qemu_start_incoming_migration(uri
, errp
);
2253 void qmp_migrate_pause(Error
**errp
)
2255 MigrationState
*ms
= migrate_get_current();
2256 MigrationIncomingState
*mis
= migration_incoming_get_current();
2259 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2260 /* Source side, during postcopy */
2261 qemu_mutex_lock(&ms
->qemu_file_lock
);
2262 ret
= qemu_file_shutdown(ms
->to_dst_file
);
2263 qemu_mutex_unlock(&ms
->qemu_file_lock
);
2265 error_setg(errp
, "Failed to pause source migration");
2270 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2271 ret
= qemu_file_shutdown(mis
->from_src_file
);
2273 error_setg(errp
, "Failed to pause destination migration");
2278 error_setg(errp
, "migrate-pause is currently only supported "
2279 "during postcopy-active state");
2282 bool migration_is_blocked(Error
**errp
)
2284 if (qemu_savevm_state_blocked(errp
)) {
2288 if (migration_blockers
) {
2289 error_propagate(errp
, error_copy(migration_blockers
->data
));
2296 /* Returns true if continue to migrate, or false if error detected */
2297 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
2298 bool resume
, Error
**errp
)
2300 Error
*local_err
= NULL
;
2303 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2304 error_setg(errp
, "Cannot resume if there is no "
2305 "paused migration");
2310 * Postcopy recovery won't work well with release-ram
2311 * capability since release-ram will drop the page buffer as
2312 * long as the page is put into the send buffer. So if there
2313 * is a network failure happened, any page buffers that have
2314 * not yet reached the destination VM but have already been
2315 * sent from the source VM will be lost forever. Let's refuse
2316 * the client from resuming such a postcopy migration.
2317 * Luckily release-ram was designed to only be used when src
2318 * and destination VMs are on the same host, so it should be
2321 if (migrate_release_ram()) {
2322 error_setg(errp
, "Postcopy recovery cannot work "
2323 "when release-ram capability is set");
2327 /* This is a resume, skip init status */
2331 if (migration_is_running(s
->state
)) {
2332 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
2336 if (runstate_check(RUN_STATE_INMIGRATE
)) {
2337 error_setg(errp
, "Guest is waiting for an incoming migration");
2341 if (runstate_check(RUN_STATE_POSTMIGRATE
)) {
2342 error_setg(errp
, "Can't migrate the vm that was paused due to "
2343 "previous migration");
2347 if (migration_is_blocked(errp
)) {
2351 if (blk
|| blk_inc
) {
2352 if (migrate_colo_enabled()) {
2353 error_setg(errp
, "No disk migration is required in COLO mode");
2356 if (migrate_use_block() || migrate_use_block_incremental()) {
2357 error_setg(errp
, "Command options are incompatible with "
2358 "current migration capabilities");
2361 migrate_set_block_enabled(true, &local_err
);
2363 error_propagate(errp
, local_err
);
2366 s
->must_remove_block_options
= true;
2370 migrate_set_block_incremental(s
, true);
2375 * set ram_counters compression_counters memory to zero for a
2378 memset(&ram_counters
, 0, sizeof(ram_counters
));
2379 memset(&compression_counters
, 0, sizeof(compression_counters
));
2384 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
2385 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
2386 bool has_resume
, bool resume
, Error
**errp
)
2388 Error
*local_err
= NULL
;
2389 MigrationState
*s
= migrate_get_current();
2390 const char *p
= NULL
;
2392 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
2393 has_resume
&& resume
, errp
)) {
2394 /* Error detected, put into errp */
2398 if (!(has_resume
&& resume
)) {
2399 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
2404 migrate_protocol_allow_multi_channels(false);
2405 if (strstart(uri
, "tcp:", &p
) ||
2406 strstart(uri
, "unix:", NULL
) ||
2407 strstart(uri
, "vsock:", NULL
)) {
2408 migrate_protocol_allow_multi_channels(true);
2409 socket_start_outgoing_migration(s
, p
? p
: uri
, &local_err
);
2411 } else if (strstart(uri
, "rdma:", &p
)) {
2412 rdma_start_outgoing_migration(s
, p
, &local_err
);
2414 } else if (strstart(uri
, "exec:", &p
)) {
2415 exec_start_outgoing_migration(s
, p
, &local_err
);
2416 } else if (strstart(uri
, "fd:", &p
)) {
2417 fd_start_outgoing_migration(s
, p
, &local_err
);
2419 if (!(has_resume
&& resume
)) {
2420 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2422 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
2423 "a valid migration protocol");
2424 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2425 MIGRATION_STATUS_FAILED
);
2426 block_cleanup_parameters(s
);
2431 if (!(has_resume
&& resume
)) {
2432 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2434 migrate_fd_error(s
, local_err
);
2435 error_propagate(errp
, local_err
);
2440 void qmp_migrate_cancel(Error
**errp
)
2442 migration_cancel(NULL
);
2445 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
2447 MigrationState
*s
= migrate_get_current();
2448 if (s
->state
!= state
) {
2449 error_setg(errp
, "Migration not in expected state: %s",
2450 MigrationStatus_str(s
->state
));
2453 qemu_sem_post(&s
->pause_sem
);
2456 bool migrate_release_ram(void)
2460 s
= migrate_get_current();
2462 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
2465 bool migrate_postcopy_ram(void)
2469 s
= migrate_get_current();
2471 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
2474 bool migrate_postcopy(void)
2476 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2479 bool migrate_auto_converge(void)
2483 s
= migrate_get_current();
2485 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
2488 bool migrate_zero_blocks(void)
2492 s
= migrate_get_current();
2494 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
2497 bool migrate_postcopy_blocktime(void)
2501 s
= migrate_get_current();
2503 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
2506 bool migrate_use_compression(void)
2510 s
= migrate_get_current();
2512 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
2515 int migrate_compress_level(void)
2519 s
= migrate_get_current();
2521 return s
->parameters
.compress_level
;
2524 int migrate_compress_threads(void)
2528 s
= migrate_get_current();
2530 return s
->parameters
.compress_threads
;
2533 int migrate_compress_wait_thread(void)
2537 s
= migrate_get_current();
2539 return s
->parameters
.compress_wait_thread
;
2542 int migrate_decompress_threads(void)
2546 s
= migrate_get_current();
2548 return s
->parameters
.decompress_threads
;
2551 bool migrate_dirty_bitmaps(void)
2555 s
= migrate_get_current();
2557 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
2560 bool migrate_ignore_shared(void)
2564 s
= migrate_get_current();
2566 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
];
2569 bool migrate_validate_uuid(void)
2573 s
= migrate_get_current();
2575 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_VALIDATE_UUID
];
2578 bool migrate_use_events(void)
2582 s
= migrate_get_current();
2584 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
2587 bool migrate_use_multifd(void)
2591 s
= migrate_get_current();
2593 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_MULTIFD
];
2596 bool migrate_pause_before_switchover(void)
2600 s
= migrate_get_current();
2602 return s
->enabled_capabilities
[
2603 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
2606 int migrate_multifd_channels(void)
2610 s
= migrate_get_current();
2612 return s
->parameters
.multifd_channels
;
2615 MultiFDCompression
migrate_multifd_compression(void)
2619 s
= migrate_get_current();
2621 assert(s
->parameters
.multifd_compression
< MULTIFD_COMPRESSION__MAX
);
2622 return s
->parameters
.multifd_compression
;
2625 int migrate_multifd_zlib_level(void)
2629 s
= migrate_get_current();
2631 return s
->parameters
.multifd_zlib_level
;
2634 int migrate_multifd_zstd_level(void)
2638 s
= migrate_get_current();
2640 return s
->parameters
.multifd_zstd_level
;
2644 bool migrate_use_zero_copy_send(void)
2648 s
= migrate_get_current();
2650 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_COPY_SEND
];
2654 int migrate_use_tls(void)
2658 s
= migrate_get_current();
2660 return s
->parameters
.tls_creds
&& *s
->parameters
.tls_creds
;
2663 int migrate_use_xbzrle(void)
2667 s
= migrate_get_current();
2669 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
2672 uint64_t migrate_xbzrle_cache_size(void)
2676 s
= migrate_get_current();
2678 return s
->parameters
.xbzrle_cache_size
;
2681 static int64_t migrate_max_postcopy_bandwidth(void)
2685 s
= migrate_get_current();
2687 return s
->parameters
.max_postcopy_bandwidth
;
2690 bool migrate_use_block(void)
2694 s
= migrate_get_current();
2696 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
2699 bool migrate_use_return_path(void)
2703 s
= migrate_get_current();
2705 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
2708 bool migrate_use_block_incremental(void)
2712 s
= migrate_get_current();
2714 return s
->parameters
.block_incremental
;
2717 bool migrate_background_snapshot(void)
2721 s
= migrate_get_current();
2723 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
];
2726 bool migrate_postcopy_preempt(void)
2730 s
= migrate_get_current();
2732 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_PREEMPT
];
2735 /* migration thread support */
2737 * Something bad happened to the RP stream, mark an error
2738 * The caller shall print or trace something to indicate why
2740 static void mark_source_rp_bad(MigrationState
*s
)
2742 s
->rp_state
.error
= true;
2745 static struct rp_cmd_args
{
2746 ssize_t len
; /* -1 = variable */
2749 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
2750 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
2751 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
2752 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
2753 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
2754 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
2755 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
2756 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
2760 * Process a request for pages received on the return path,
2761 * We're allowed to send more than requested (e.g. to round to our page size)
2762 * and we don't need to send pages that have already been sent.
2764 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
2765 ram_addr_t start
, size_t len
)
2767 long our_host_ps
= qemu_real_host_page_size();
2769 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
2772 * Since we currently insist on matching page sizes, just sanity check
2773 * we're being asked for whole host pages.
2775 if (!QEMU_IS_ALIGNED(start
, our_host_ps
) ||
2776 !QEMU_IS_ALIGNED(len
, our_host_ps
)) {
2777 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2778 " len: %zd", __func__
, start
, len
);
2779 mark_source_rp_bad(ms
);
2783 if (ram_save_queue_pages(rbname
, start
, len
)) {
2784 mark_source_rp_bad(ms
);
2788 /* Return true to retry, false to quit */
2789 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
2791 trace_postcopy_pause_return_path();
2793 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
2795 trace_postcopy_pause_return_path_continued();
2800 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
2802 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
2805 error_report("%s: invalid block name '%s'", __func__
, block_name
);
2809 /* Fetch the received bitmap and refresh the dirty bitmap */
2810 return ram_dirty_bitmap_reload(s
, block
);
2813 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
2815 trace_source_return_path_thread_resume_ack(value
);
2817 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
2818 error_report("%s: illegal resume_ack value %"PRIu32
,
2823 /* Now both sides are active. */
2824 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2825 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2827 /* Notify send thread that time to continue send pages */
2828 qemu_sem_post(&s
->rp_state
.rp_sem
);
2833 /* Release ms->rp_state.from_dst_file in a safe way */
2834 static void migration_release_from_dst_file(MigrationState
*ms
)
2838 WITH_QEMU_LOCK_GUARD(&ms
->qemu_file_lock
) {
2840 * Reset the from_dst_file pointer first before releasing it, as we
2841 * can't block within lock section
2843 file
= ms
->rp_state
.from_dst_file
;
2844 ms
->rp_state
.from_dst_file
= NULL
;
2851 * Handles messages sent on the return path towards the source VM
2854 static void *source_return_path_thread(void *opaque
)
2856 MigrationState
*ms
= opaque
;
2857 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
2858 uint16_t header_len
, header_type
;
2860 uint32_t tmp32
, sibling_error
;
2861 ram_addr_t start
= 0; /* =0 to silence warning */
2862 size_t len
= 0, expected_len
;
2865 trace_source_return_path_thread_entry();
2866 rcu_register_thread();
2869 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
2870 migration_is_setup_or_active(ms
->state
)) {
2871 trace_source_return_path_thread_loop_top();
2872 header_type
= qemu_get_be16(rp
);
2873 header_len
= qemu_get_be16(rp
);
2875 if (qemu_file_get_error(rp
)) {
2876 mark_source_rp_bad(ms
);
2880 if (header_type
>= MIG_RP_MSG_MAX
||
2881 header_type
== MIG_RP_MSG_INVALID
) {
2882 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2883 header_type
, header_len
);
2884 mark_source_rp_bad(ms
);
2888 if ((rp_cmd_args
[header_type
].len
!= -1 &&
2889 header_len
!= rp_cmd_args
[header_type
].len
) ||
2890 header_len
> sizeof(buf
)) {
2891 error_report("RP: Received '%s' message (0x%04x) with"
2892 "incorrect length %d expecting %zu",
2893 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
2894 (size_t)rp_cmd_args
[header_type
].len
);
2895 mark_source_rp_bad(ms
);
2899 /* We know we've got a valid header by this point */
2900 res
= qemu_get_buffer(rp
, buf
, header_len
);
2901 if (res
!= header_len
) {
2902 error_report("RP: Failed reading data for message 0x%04x"
2903 " read %d expected %d",
2904 header_type
, res
, header_len
);
2905 mark_source_rp_bad(ms
);
2909 /* OK, we have the message and the data */
2910 switch (header_type
) {
2911 case MIG_RP_MSG_SHUT
:
2912 sibling_error
= ldl_be_p(buf
);
2913 trace_source_return_path_thread_shut(sibling_error
);
2914 if (sibling_error
) {
2915 error_report("RP: Sibling indicated error %d", sibling_error
);
2916 mark_source_rp_bad(ms
);
2919 * We'll let the main thread deal with closing the RP
2920 * we could do a shutdown(2) on it, but we're the only user
2921 * anyway, so there's nothing gained.
2925 case MIG_RP_MSG_PONG
:
2926 tmp32
= ldl_be_p(buf
);
2927 trace_source_return_path_thread_pong(tmp32
);
2930 case MIG_RP_MSG_REQ_PAGES
:
2931 start
= ldq_be_p(buf
);
2932 len
= ldl_be_p(buf
+ 8);
2933 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
2936 case MIG_RP_MSG_REQ_PAGES_ID
:
2937 expected_len
= 12 + 1; /* header + termination */
2939 if (header_len
>= expected_len
) {
2940 start
= ldq_be_p(buf
);
2941 len
= ldl_be_p(buf
+ 8);
2942 /* Now we expect an idstr */
2943 tmp32
= buf
[12]; /* Length of the following idstr */
2944 buf
[13 + tmp32
] = '\0';
2945 expected_len
+= tmp32
;
2947 if (header_len
!= expected_len
) {
2948 error_report("RP: Req_Page_id with length %d expecting %zd",
2949 header_len
, expected_len
);
2950 mark_source_rp_bad(ms
);
2953 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
2956 case MIG_RP_MSG_RECV_BITMAP
:
2957 if (header_len
< 1) {
2958 error_report("%s: missing block name", __func__
);
2959 mark_source_rp_bad(ms
);
2962 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2963 buf
[buf
[0] + 1] = '\0';
2964 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
2965 mark_source_rp_bad(ms
);
2970 case MIG_RP_MSG_RESUME_ACK
:
2971 tmp32
= ldl_be_p(buf
);
2972 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
2973 mark_source_rp_bad(ms
);
2984 res
= qemu_file_get_error(rp
);
2986 if (res
&& migration_in_postcopy()) {
2988 * Maybe there is something we can do: it looks like a
2989 * network down issue, and we pause for a recovery.
2991 migration_release_from_dst_file(ms
);
2993 if (postcopy_pause_return_path_thread(ms
)) {
2995 * Reload rp, reset the rest. Referencing it is safe since
2996 * it's reset only by us above, or when migration completes
2998 rp
= ms
->rp_state
.from_dst_file
;
2999 ms
->rp_state
.error
= false;
3004 trace_source_return_path_thread_bad_end();
3005 mark_source_rp_bad(ms
);
3008 trace_source_return_path_thread_end();
3009 migration_release_from_dst_file(ms
);
3010 rcu_unregister_thread();
3014 static int open_return_path_on_source(MigrationState
*ms
,
3017 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
3018 if (!ms
->rp_state
.from_dst_file
) {
3022 trace_open_return_path_on_source();
3024 if (!create_thread
) {
3029 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
3030 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
3031 ms
->rp_state
.rp_thread_created
= true;
3033 trace_open_return_path_on_source_continue();
3038 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
3039 static int await_return_path_close_on_source(MigrationState
*ms
)
3042 * If this is a normal exit then the destination will send a SHUT and the
3043 * rp_thread will exit, however if there's an error we need to cause
3046 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
3048 * shutdown(2), if we have it, will cause it to unblock if it's stuck
3049 * waiting for the destination.
3051 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
3052 mark_source_rp_bad(ms
);
3054 trace_await_return_path_close_on_source_joining();
3055 qemu_thread_join(&ms
->rp_state
.rp_thread
);
3056 ms
->rp_state
.rp_thread_created
= false;
3057 trace_await_return_path_close_on_source_close();
3058 return ms
->rp_state
.error
;
3062 * Switch from normal iteration to postcopy
3063 * Returns non-0 on error
3065 static int postcopy_start(MigrationState
*ms
)
3068 QIOChannelBuffer
*bioc
;
3070 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3071 int64_t bandwidth
= migrate_max_postcopy_bandwidth();
3072 bool restart_block
= false;
3073 int cur_state
= MIGRATION_STATUS_ACTIVE
;
3075 if (postcopy_preempt_wait_channel(ms
)) {
3076 migrate_set_state(&ms
->state
, ms
->state
, MIGRATION_STATUS_FAILED
);
3080 if (!migrate_pause_before_switchover()) {
3081 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
3082 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3085 trace_postcopy_start();
3086 qemu_mutex_lock_iothread();
3087 trace_postcopy_start_set_run();
3089 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3090 global_state_store();
3091 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
3096 ret
= migration_maybe_pause(ms
, &cur_state
,
3097 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3102 ret
= bdrv_inactivate_all();
3106 restart_block
= true;
3109 * Cause any non-postcopiable, but iterative devices to
3110 * send out their final data.
3112 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
3115 * in Finish migrate and with the io-lock held everything should
3116 * be quiet, but we've potentially still got dirty pages and we
3117 * need to tell the destination to throw any pages it's already received
3120 if (migrate_postcopy_ram()) {
3121 ram_postcopy_send_discard_bitmap(ms
);
3125 * send rest of state - note things that are doing postcopy
3126 * will notice we're in POSTCOPY_ACTIVE and not actually
3127 * wrap their state up here
3129 /* 0 max-postcopy-bandwidth means unlimited */
3131 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
3133 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
3135 if (migrate_postcopy_ram()) {
3136 /* Ping just for debugging, helps line traces up */
3137 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
3141 * While loading the device state we may trigger page transfer
3142 * requests and the fd must be free to process those, and thus
3143 * the destination must read the whole device state off the fd before
3144 * it starts processing it. Unfortunately the ad-hoc migration format
3145 * doesn't allow the destination to know the size to read without fully
3146 * parsing it through each devices load-state code (especially the open
3147 * coded devices that use get/put).
3148 * So we wrap the device state up in a package with a length at the start;
3149 * to do this we use a qemu_buf to hold the whole of the device state.
3151 bioc
= qio_channel_buffer_new(4096);
3152 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
3153 fb
= qemu_file_new_output(QIO_CHANNEL(bioc
));
3154 object_unref(OBJECT(bioc
));
3157 * Make sure the receiver can get incoming pages before we send the rest
3160 qemu_savevm_send_postcopy_listen(fb
);
3162 qemu_savevm_state_complete_precopy(fb
, false, false);
3163 if (migrate_postcopy_ram()) {
3164 qemu_savevm_send_ping(fb
, 3);
3167 qemu_savevm_send_postcopy_run(fb
);
3169 /* <><> end of stuff going into the package */
3171 /* Last point of recovery; as soon as we send the package the destination
3172 * can open devices and potentially start running.
3173 * Lets just check again we've not got any errors.
3175 ret
= qemu_file_get_error(ms
->to_dst_file
);
3177 error_report("postcopy_start: Migration stream errored (pre package)");
3181 restart_block
= false;
3183 /* Now send that blob */
3184 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
3189 /* Send a notify to give a chance for anything that needs to happen
3190 * at the transition to postcopy and after the device state; in particular
3191 * spice needs to trigger a transition now
3193 ms
->postcopy_after_devices
= true;
3194 notifier_list_notify(&migration_state_notifiers
, ms
);
3196 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
3198 qemu_mutex_unlock_iothread();
3200 if (migrate_postcopy_ram()) {
3202 * Although this ping is just for debug, it could potentially be
3203 * used for getting a better measurement of downtime at the source.
3205 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
3208 if (migrate_release_ram()) {
3209 ram_postcopy_migrated_memory_release(ms
);
3212 ret
= qemu_file_get_error(ms
->to_dst_file
);
3214 error_report("postcopy_start: Migration stream errored");
3215 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
3216 MIGRATION_STATUS_FAILED
);
3219 trace_postcopy_preempt_enabled(migrate_postcopy_preempt());
3226 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
3227 MIGRATION_STATUS_FAILED
);
3228 if (restart_block
) {
3229 /* A failure happened early enough that we know the destination hasn't
3230 * accessed block devices, so we're safe to recover.
3232 Error
*local_err
= NULL
;
3234 bdrv_activate_all(&local_err
);
3236 error_report_err(local_err
);
3239 qemu_mutex_unlock_iothread();
3244 * migration_maybe_pause: Pause if required to by
3245 * migrate_pause_before_switchover called with the iothread locked
3246 * Returns: 0 on success
3248 static int migration_maybe_pause(MigrationState
*s
,
3249 int *current_active_state
,
3252 if (!migrate_pause_before_switchover()) {
3256 /* Since leaving this state is not atomic with posting the semaphore
3257 * it's possible that someone could have issued multiple migrate_continue
3258 * and the semaphore is incorrectly positive at this point;
3259 * the docs say it's undefined to reinit a semaphore that's already
3260 * init'd, so use timedwait to eat up any existing posts.
3262 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
3263 /* This block intentionally left blank */
3267 * If the migration is cancelled when it is in the completion phase,
3268 * the migration state is set to MIGRATION_STATUS_CANCELLING.
3269 * So we don't need to wait a semaphore, otherwise we would always
3270 * wait for the 'pause_sem' semaphore.
3272 if (s
->state
!= MIGRATION_STATUS_CANCELLING
) {
3273 qemu_mutex_unlock_iothread();
3274 migrate_set_state(&s
->state
, *current_active_state
,
3275 MIGRATION_STATUS_PRE_SWITCHOVER
);
3276 qemu_sem_wait(&s
->pause_sem
);
3277 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
3279 *current_active_state
= new_state
;
3280 qemu_mutex_lock_iothread();
3283 return s
->state
== new_state
? 0 : -EINVAL
;
3287 * migration_completion: Used by migration_thread when there's not much left.
3288 * The caller 'breaks' the loop when this returns.
3290 * @s: Current migration state
3292 static void migration_completion(MigrationState
*s
)
3295 int current_active_state
= s
->state
;
3297 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
3298 qemu_mutex_lock_iothread();
3299 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3300 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3301 s
->vm_was_running
= runstate_is_running();
3302 ret
= global_state_store();
3305 bool inactivate
= !migrate_colo_enabled();
3306 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
3307 trace_migration_completion_vm_stop(ret
);
3309 ret
= migration_maybe_pause(s
, ¤t_active_state
,
3310 MIGRATION_STATUS_DEVICE
);
3313 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
3314 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
3317 if (inactivate
&& ret
>= 0) {
3318 s
->block_inactive
= true;
3321 qemu_mutex_unlock_iothread();
3326 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3327 trace_migration_completion_postcopy_end();
3329 qemu_mutex_lock_iothread();
3330 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
3331 qemu_mutex_unlock_iothread();
3333 /* Shutdown the postcopy fast path thread */
3334 if (migrate_postcopy_preempt()) {
3335 postcopy_preempt_shutdown_file(s
);
3338 trace_migration_completion_postcopy_end_after_complete();
3344 * If rp was opened we must clean up the thread before
3345 * cleaning everything else up (since if there are no failures
3346 * it will wait for the destination to send it's status in
3349 if (s
->rp_state
.rp_thread_created
) {
3351 trace_migration_return_path_end_before();
3352 rp_error
= await_return_path_close_on_source(s
);
3353 trace_migration_return_path_end_after(rp_error
);
3355 goto fail_invalidate
;
3359 if (qemu_file_get_error(s
->to_dst_file
)) {
3360 trace_migration_completion_file_err();
3361 goto fail_invalidate
;
3364 if (migrate_colo_enabled() && s
->state
== MIGRATION_STATUS_ACTIVE
) {
3365 /* COLO does not support postcopy */
3366 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
3367 MIGRATION_STATUS_COLO
);
3369 migrate_set_state(&s
->state
, current_active_state
,
3370 MIGRATION_STATUS_COMPLETED
);
3376 /* If not doing postcopy, vm_start() will be called: let's regain
3377 * control on images.
3379 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
3380 s
->state
== MIGRATION_STATUS_DEVICE
) {
3381 Error
*local_err
= NULL
;
3383 qemu_mutex_lock_iothread();
3384 bdrv_activate_all(&local_err
);
3386 error_report_err(local_err
);
3388 s
->block_inactive
= false;
3390 qemu_mutex_unlock_iothread();
3394 migrate_set_state(&s
->state
, current_active_state
,
3395 MIGRATION_STATUS_FAILED
);
3399 * bg_migration_completion: Used by bg_migration_thread when after all the
3400 * RAM has been saved. The caller 'breaks' the loop when this returns.
3402 * @s: Current migration state
3404 static void bg_migration_completion(MigrationState
*s
)
3406 int current_active_state
= s
->state
;
3409 * Stop tracking RAM writes - un-protect memory, un-register UFFD
3410 * memory ranges, flush kernel wait queues and wake up threads
3411 * waiting for write fault to be resolved.
3413 ram_write_tracking_stop();
3415 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
3417 * By this moment we have RAM content saved into the migration stream.
3418 * The next step is to flush the non-RAM content (device state)
3419 * right after the ram content. The device state has been stored into
3420 * the temporary buffer before RAM saving started.
3422 qemu_put_buffer(s
->to_dst_file
, s
->bioc
->data
, s
->bioc
->usage
);
3423 qemu_fflush(s
->to_dst_file
);
3424 } else if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
3428 if (qemu_file_get_error(s
->to_dst_file
)) {
3429 trace_migration_completion_file_err();
3433 migrate_set_state(&s
->state
, current_active_state
,
3434 MIGRATION_STATUS_COMPLETED
);
3438 migrate_set_state(&s
->state
, current_active_state
,
3439 MIGRATION_STATUS_FAILED
);
3442 bool migrate_colo_enabled(void)
3444 MigrationState
*s
= migrate_get_current();
3445 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
3448 typedef enum MigThrError
{
3449 /* No error detected */
3450 MIG_THR_ERR_NONE
= 0,
3451 /* Detected error, but resumed successfully */
3452 MIG_THR_ERR_RECOVERED
= 1,
3453 /* Detected fatal error, need to exit */
3454 MIG_THR_ERR_FATAL
= 2,
3457 static int postcopy_resume_handshake(MigrationState
*s
)
3459 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
3461 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3462 qemu_sem_wait(&s
->rp_state
.rp_sem
);
3465 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3472 /* Return zero if success, or <0 for error */
3473 static int postcopy_do_resume(MigrationState
*s
)
3478 * Call all the resume_prepare() hooks, so that modules can be
3479 * ready for the migration resume.
3481 ret
= qemu_savevm_state_resume_prepare(s
);
3483 error_report("%s: resume_prepare() failure detected: %d",
3489 * Last handshake with destination on the resume (destination will
3490 * switch to postcopy-active afterwards)
3492 ret
= postcopy_resume_handshake(s
);
3494 error_report("%s: handshake failed: %d", __func__
, ret
);
3502 * We don't return until we are in a safe state to continue current
3503 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
3504 * MIG_THR_ERR_FATAL if unrecovery failure happened.
3506 static MigThrError
postcopy_pause(MigrationState
*s
)
3508 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3514 * Current channel is possibly broken. Release it. Note that this is
3515 * guaranteed even without lock because to_dst_file should only be
3516 * modified by the migration thread. That also guarantees that the
3517 * unregister of yank is safe too without the lock. It should be safe
3518 * even to be within the qemu_file_lock, but we didn't do that to avoid
3519 * taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make
3520 * the qemu_file_lock critical section as small as possible.
3522 assert(s
->to_dst_file
);
3523 migration_ioc_unregister_yank_from_file(s
->to_dst_file
);
3524 qemu_mutex_lock(&s
->qemu_file_lock
);
3525 file
= s
->to_dst_file
;
3526 s
->to_dst_file
= NULL
;
3527 qemu_mutex_unlock(&s
->qemu_file_lock
);
3529 qemu_file_shutdown(file
);
3533 * Do the same to postcopy fast path socket too if there is. No
3534 * locking needed because no racer as long as we do this before setting
3537 if (s
->postcopy_qemufile_src
) {
3538 migration_ioc_unregister_yank_from_file(s
->postcopy_qemufile_src
);
3539 qemu_file_shutdown(s
->postcopy_qemufile_src
);
3540 qemu_fclose(s
->postcopy_qemufile_src
);
3541 s
->postcopy_qemufile_src
= NULL
;
3544 migrate_set_state(&s
->state
, s
->state
,
3545 MIGRATION_STATUS_POSTCOPY_PAUSED
);
3547 error_report("Detected IO failure for postcopy. "
3548 "Migration paused.");
3551 * We wait until things fixed up. Then someone will setup the
3552 * status back for us.
3554 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
3555 qemu_sem_wait(&s
->postcopy_pause_sem
);
3558 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3559 /* Woken up by a recover procedure. Give it a shot */
3561 if (postcopy_preempt_wait_channel(s
)) {
3563 * Preempt enabled, and new channel create failed; loop
3564 * back to wait for another recovery.
3570 * Firstly, let's wake up the return path now, with a new
3571 * return path channel.
3573 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
3575 /* Do the resume logic */
3576 if (postcopy_do_resume(s
) == 0) {
3577 /* Let's continue! */
3578 trace_postcopy_pause_continued();
3579 return MIG_THR_ERR_RECOVERED
;
3582 * Something wrong happened during the recovery, let's
3583 * pause again. Pause is always better than throwing
3589 /* This is not right... Time to quit. */
3590 return MIG_THR_ERR_FATAL
;
3595 static MigThrError
migration_detect_error(MigrationState
*s
)
3598 int state
= s
->state
;
3599 Error
*local_error
= NULL
;
3601 if (state
== MIGRATION_STATUS_CANCELLING
||
3602 state
== MIGRATION_STATUS_CANCELLED
) {
3603 /* End the migration, but don't set the state to failed */
3604 return MIG_THR_ERR_FATAL
;
3608 * Try to detect any file errors. Note that postcopy_qemufile_src will
3609 * be NULL when postcopy preempt is not enabled.
3611 ret
= qemu_file_get_error_obj_any(s
->to_dst_file
,
3612 s
->postcopy_qemufile_src
,
3615 /* Everything is fine */
3616 assert(!local_error
);
3617 return MIG_THR_ERR_NONE
;
3621 migrate_set_error(s
, local_error
);
3622 error_free(local_error
);
3625 if (state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
) {
3627 * For postcopy, we allow the network to be down for a
3628 * while. After that, it can be continued by a
3631 return postcopy_pause(s
);
3634 * For precopy (or postcopy with error outside IO), we fail
3637 migrate_set_state(&s
->state
, state
, MIGRATION_STATUS_FAILED
);
3638 trace_migration_thread_file_err();
3640 /* Time to stop the migration, now. */
3641 return MIG_THR_ERR_FATAL
;
3645 /* How many bytes have we transferred since the beginning of the migration */
3646 static uint64_t migration_total_bytes(MigrationState
*s
)
3648 return qemu_file_total_transferred(s
->to_dst_file
) +
3649 ram_counters
.multifd_bytes
;
3652 static void migration_calculate_complete(MigrationState
*s
)
3654 uint64_t bytes
= migration_total_bytes(s
);
3655 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3656 int64_t transfer_time
;
3658 s
->total_time
= end_time
- s
->start_time
;
3661 * It's still not set, so we are precopy migration. For
3662 * postcopy, downtime is calculated during postcopy_start().
3664 s
->downtime
= end_time
- s
->downtime_start
;
3667 transfer_time
= s
->total_time
- s
->setup_time
;
3668 if (transfer_time
) {
3669 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
3673 static void update_iteration_initial_status(MigrationState
*s
)
3676 * Update these three fields at the same time to avoid mismatch info lead
3677 * wrong speed calculation.
3679 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3680 s
->iteration_initial_bytes
= migration_total_bytes(s
);
3681 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
3684 static void migration_update_counters(MigrationState
*s
,
3685 int64_t current_time
)
3687 uint64_t transferred
, transferred_pages
, time_spent
;
3688 uint64_t current_bytes
; /* bytes transferred since the beginning */
3691 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
3695 current_bytes
= migration_total_bytes(s
);
3696 transferred
= current_bytes
- s
->iteration_initial_bytes
;
3697 time_spent
= current_time
- s
->iteration_start_time
;
3698 bandwidth
= (double)transferred
/ time_spent
;
3699 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
3701 s
->mbps
= (((double) transferred
* 8.0) /
3702 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
3704 transferred_pages
= ram_get_total_transferred_pages() -
3705 s
->iteration_initial_pages
;
3706 s
->pages_per_second
= (double) transferred_pages
/
3707 (((double) time_spent
/ 1000.0));
3710 * if we haven't sent anything, we don't want to
3711 * recalculate. 10000 is a small enough number for our purposes
3713 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
3714 s
->expected_downtime
= ram_counters
.remaining
/ bandwidth
;
3717 qemu_file_reset_rate_limit(s
->to_dst_file
);
3719 update_iteration_initial_status(s
);
3721 trace_migrate_transferred(transferred
, time_spent
,
3722 bandwidth
, s
->threshold_size
);
3725 /* Migration thread iteration status */
3727 MIG_ITERATE_RESUME
, /* Resume current iteration */
3728 MIG_ITERATE_SKIP
, /* Skip current iteration */
3729 MIG_ITERATE_BREAK
, /* Break the loop */
3733 * Return true if continue to the next iteration directly, false
3736 static MigIterateState
migration_iteration_run(MigrationState
*s
)
3738 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
3739 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
3741 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
3742 &pend_compat
, &pend_post
);
3743 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
3745 trace_migrate_pending(pending_size
, s
->threshold_size
,
3746 pend_pre
, pend_compat
, pend_post
);
3748 if (pending_size
&& pending_size
>= s
->threshold_size
) {
3749 /* Still a significant amount to transfer */
3750 if (!in_postcopy
&& pend_pre
<= s
->threshold_size
&&
3751 qatomic_read(&s
->start_postcopy
)) {
3752 if (postcopy_start(s
)) {
3753 error_report("%s: postcopy failed to start", __func__
);
3755 return MIG_ITERATE_SKIP
;
3757 /* Just another iteration step */
3758 qemu_savevm_state_iterate(s
->to_dst_file
, in_postcopy
);
3760 trace_migration_thread_low_pending(pending_size
);
3761 migration_completion(s
);
3762 return MIG_ITERATE_BREAK
;
3765 return MIG_ITERATE_RESUME
;
3768 static void migration_iteration_finish(MigrationState
*s
)
3770 /* If we enabled cpu throttling for auto-converge, turn it off. */
3771 cpu_throttle_stop();
3773 qemu_mutex_lock_iothread();
3775 case MIGRATION_STATUS_COMPLETED
:
3776 migration_calculate_complete(s
);
3777 runstate_set(RUN_STATE_POSTMIGRATE
);
3779 case MIGRATION_STATUS_COLO
:
3780 if (!migrate_colo_enabled()) {
3781 error_report("%s: critical error: calling COLO code without "
3782 "COLO enabled", __func__
);
3784 migrate_start_colo_process(s
);
3785 s
->vm_was_running
= true;
3787 case MIGRATION_STATUS_FAILED
:
3788 case MIGRATION_STATUS_CANCELLED
:
3789 case MIGRATION_STATUS_CANCELLING
:
3790 if (s
->vm_was_running
) {
3791 if (!runstate_check(RUN_STATE_SHUTDOWN
)) {
3795 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
3796 runstate_set(RUN_STATE_POSTMIGRATE
);
3802 /* Should not reach here, but if so, forgive the VM. */
3803 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3806 migrate_fd_cleanup_schedule(s
);
3807 qemu_mutex_unlock_iothread();
3810 static void bg_migration_iteration_finish(MigrationState
*s
)
3812 qemu_mutex_lock_iothread();
3814 case MIGRATION_STATUS_COMPLETED
:
3815 migration_calculate_complete(s
);
3818 case MIGRATION_STATUS_ACTIVE
:
3819 case MIGRATION_STATUS_FAILED
:
3820 case MIGRATION_STATUS_CANCELLED
:
3821 case MIGRATION_STATUS_CANCELLING
:
3825 /* Should not reach here, but if so, forgive the VM. */
3826 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3830 migrate_fd_cleanup_schedule(s
);
3831 qemu_mutex_unlock_iothread();
3835 * Return true if continue to the next iteration directly, false
3838 static MigIterateState
bg_migration_iteration_run(MigrationState
*s
)
3842 res
= qemu_savevm_state_iterate(s
->to_dst_file
, false);
3844 bg_migration_completion(s
);
3845 return MIG_ITERATE_BREAK
;
3848 return MIG_ITERATE_RESUME
;
3851 void migration_make_urgent_request(void)
3853 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
3856 void migration_consume_urgent_request(void)
3858 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
3861 /* Returns true if the rate limiting was broken by an urgent request */
3862 bool migration_rate_limit(void)
3864 int64_t now
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3865 MigrationState
*s
= migrate_get_current();
3867 bool urgent
= false;
3868 migration_update_counters(s
, now
);
3869 if (qemu_file_rate_limit(s
->to_dst_file
)) {
3871 if (qemu_file_get_error(s
->to_dst_file
)) {
3875 * Wait for a delay to do rate limiting OR
3876 * something urgent to post the semaphore.
3878 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- now
;
3879 trace_migration_rate_limit_pre(ms
);
3880 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
3882 * We were woken by one or more urgent things but
3883 * the timedwait will have consumed one of them.
3884 * The service routine for the urgent wake will dec
3885 * the semaphore itself for each item it consumes,
3886 * so add this one we just eat back.
3888 qemu_sem_post(&s
->rate_limit_sem
);
3891 trace_migration_rate_limit_post(urgent
);
3897 * if failover devices are present, wait they are completely
3901 static void qemu_savevm_wait_unplug(MigrationState
*s
, int old_state
,
3904 if (qemu_savevm_state_guest_unplug_pending()) {
3905 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_WAIT_UNPLUG
);
3907 while (s
->state
== MIGRATION_STATUS_WAIT_UNPLUG
&&
3908 qemu_savevm_state_guest_unplug_pending()) {
3909 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
3911 if (s
->state
!= MIGRATION_STATUS_WAIT_UNPLUG
) {
3912 int timeout
= 120; /* 30 seconds */
3914 * migration has been canceled
3915 * but as we have started an unplug we must wait the end
3916 * to be able to plug back the card
3918 while (timeout
-- && qemu_savevm_state_guest_unplug_pending()) {
3919 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
3921 if (qemu_savevm_state_guest_unplug_pending() &&
3923 warn_report("migration: partially unplugged device on "
3928 migrate_set_state(&s
->state
, MIGRATION_STATUS_WAIT_UNPLUG
, new_state
);
3930 migrate_set_state(&s
->state
, old_state
, new_state
);
3935 * Master migration thread on the source VM.
3936 * It drives the migration and pumps the data down the outgoing channel.
3938 static void *migration_thread(void *opaque
)
3940 MigrationState
*s
= opaque
;
3941 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3942 MigThrError thr_error
;
3943 bool urgent
= false;
3945 rcu_register_thread();
3947 object_ref(OBJECT(s
));
3948 update_iteration_initial_status(s
);
3950 qemu_savevm_state_header(s
->to_dst_file
);
3953 * If we opened the return path, we need to make sure dst has it
3956 if (s
->rp_state
.rp_thread_created
) {
3957 /* Now tell the dest that it should open its end so it can reply */
3958 qemu_savevm_send_open_return_path(s
->to_dst_file
);
3960 /* And do a ping that will make stuff easier to debug */
3961 qemu_savevm_send_ping(s
->to_dst_file
, 1);
3964 if (migrate_postcopy()) {
3966 * Tell the destination that we *might* want to do postcopy later;
3967 * if the other end can't do postcopy it should fail now, nice and
3970 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
3973 if (migrate_colo_enabled()) {
3974 /* Notify migration destination that we enable COLO */
3975 qemu_savevm_send_colo_enable(s
->to_dst_file
);
3978 qemu_savevm_state_setup(s
->to_dst_file
);
3980 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
3981 MIGRATION_STATUS_ACTIVE
);
3983 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3985 trace_migration_thread_setup_complete();
3987 while (migration_is_active(s
)) {
3988 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
3989 MigIterateState iter_state
= migration_iteration_run(s
);
3990 if (iter_state
== MIG_ITERATE_SKIP
) {
3992 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3998 * Try to detect any kind of failures, and see whether we
3999 * should stop the migration now.
4001 thr_error
= migration_detect_error(s
);
4002 if (thr_error
== MIG_THR_ERR_FATAL
) {
4003 /* Stop migration */
4005 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
4007 * Just recovered from a e.g. network failure, reset all
4008 * the local variables. This is important to avoid
4009 * breaking transferred_bytes and bandwidth calculation
4011 update_iteration_initial_status(s
);
4014 urgent
= migration_rate_limit();
4017 trace_migration_thread_after_loop();
4018 migration_iteration_finish(s
);
4019 object_unref(OBJECT(s
));
4020 rcu_unregister_thread();
4024 static void bg_migration_vm_start_bh(void *opaque
)
4026 MigrationState
*s
= opaque
;
4028 qemu_bh_delete(s
->vm_start_bh
);
4029 s
->vm_start_bh
= NULL
;
4032 s
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - s
->downtime_start
;
4036 * Background snapshot thread, based on live migration code.
4037 * This is an alternative implementation of live migration mechanism
4038 * introduced specifically to support background snapshots.
4040 * It takes advantage of userfault_fd write protection mechanism introduced
4041 * in v5.7 kernel. Compared to existing dirty page logging migration much
4042 * lesser stream traffic is produced resulting in smaller snapshot images,
4043 * simply cause of no page duplicates can get into the stream.
4045 * Another key point is that generated vmstate stream reflects machine state
4046 * 'frozen' at the beginning of snapshot creation compared to dirty page logging
4047 * mechanism, which effectively results in that saved snapshot is the state of VM
4048 * at the end of the process.
4050 static void *bg_migration_thread(void *opaque
)
4052 MigrationState
*s
= opaque
;
4053 int64_t setup_start
;
4054 MigThrError thr_error
;
4056 bool early_fail
= true;
4058 rcu_register_thread();
4059 object_ref(OBJECT(s
));
4061 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
4063 setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
4065 * We want to save vmstate for the moment when migration has been
4066 * initiated but also we want to save RAM content while VM is running.
4067 * The RAM content should appear first in the vmstate. So, we first
4068 * stash the non-RAM part of the vmstate to the temporary buffer,
4069 * then write RAM part of the vmstate to the migration stream
4070 * with vCPUs running and, finally, write stashed non-RAM part of
4071 * the vmstate from the buffer to the migration stream.
4073 s
->bioc
= qio_channel_buffer_new(512 * 1024);
4074 qio_channel_set_name(QIO_CHANNEL(s
->bioc
), "vmstate-buffer");
4075 fb
= qemu_file_new_output(QIO_CHANNEL(s
->bioc
));
4076 object_unref(OBJECT(s
->bioc
));
4078 update_iteration_initial_status(s
);
4081 * Prepare for tracking memory writes with UFFD-WP - populate
4082 * RAM pages before protecting.
4085 ram_write_tracking_prepare();
4088 qemu_savevm_state_header(s
->to_dst_file
);
4089 qemu_savevm_state_setup(s
->to_dst_file
);
4091 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
4092 MIGRATION_STATUS_ACTIVE
);
4094 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
4096 trace_migration_thread_setup_complete();
4097 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
4099 qemu_mutex_lock_iothread();
4102 * If VM is currently in suspended state, then, to make a valid runstate
4103 * transition in vm_stop_force_state() we need to wakeup it up.
4105 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
4106 s
->vm_was_running
= runstate_is_running();
4108 if (global_state_store()) {
4111 /* Forcibly stop VM before saving state of vCPUs and devices */
4112 if (vm_stop_force_state(RUN_STATE_PAUSED
)) {
4116 * Put vCPUs in sync with shadow context structures, then
4117 * save their state to channel-buffer along with devices.
4119 cpu_synchronize_all_states();
4120 if (qemu_savevm_state_complete_precopy_non_iterable(fb
, false, false)) {
4124 * Since we are going to get non-iterable state data directly
4125 * from s->bioc->data, explicit flush is needed here.
4129 /* Now initialize UFFD context and start tracking RAM writes */
4130 if (ram_write_tracking_start()) {
4136 * Start VM from BH handler to avoid write-fault lock here.
4137 * UFFD-WP protection for the whole RAM is already enabled so
4138 * calling VM state change notifiers from vm_start() would initiate
4139 * writes to virtio VQs memory which is in write-protected region.
4141 s
->vm_start_bh
= qemu_bh_new(bg_migration_vm_start_bh
, s
);
4142 qemu_bh_schedule(s
->vm_start_bh
);
4144 qemu_mutex_unlock_iothread();
4146 while (migration_is_active(s
)) {
4147 MigIterateState iter_state
= bg_migration_iteration_run(s
);
4148 if (iter_state
== MIG_ITERATE_SKIP
) {
4150 } else if (iter_state
== MIG_ITERATE_BREAK
) {
4155 * Try to detect any kind of failures, and see whether we
4156 * should stop the migration now.
4158 thr_error
= migration_detect_error(s
);
4159 if (thr_error
== MIG_THR_ERR_FATAL
) {
4160 /* Stop migration */
4164 migration_update_counters(s
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
));
4167 trace_migration_thread_after_loop();
4171 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
4172 MIGRATION_STATUS_FAILED
);
4173 qemu_mutex_unlock_iothread();
4176 bg_migration_iteration_finish(s
);
4179 object_unref(OBJECT(s
));
4180 rcu_unregister_thread();
4185 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
4187 Error
*local_err
= NULL
;
4189 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
4192 * If there's a previous error, free it and prepare for another one.
4193 * Meanwhile if migration completes successfully, there won't have an error
4194 * dumped when calling migrate_fd_cleanup().
4196 migrate_error_free(s
);
4198 s
->expected_downtime
= s
->parameters
.downtime_limit
;
4200 assert(s
->cleanup_bh
);
4202 assert(!s
->cleanup_bh
);
4203 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup_bh
, s
);
4206 migrate_fd_error(s
, error_in
);
4209 * Don't do cleanup for resume if channel is invalid, but only dump
4210 * the error. We wait for another channel connect from the user.
4211 * The error_report still gives HMP user a hint on what failed.
4212 * It's normally done in migrate_fd_cleanup(), but call it here
4215 error_report_err(error_copy(s
->error
));
4217 migrate_fd_cleanup(s
);
4223 /* This is a resumed migration */
4224 rate_limit
= s
->parameters
.max_postcopy_bandwidth
/
4227 /* This is a fresh new migration */
4228 rate_limit
= s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
;
4230 /* Notify before starting migration thread */
4231 notifier_list_notify(&migration_state_notifiers
, s
);
4234 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
4235 qemu_file_set_blocking(s
->to_dst_file
, true);
4238 * Open the return path. For postcopy, it is used exclusively. For
4239 * precopy, only if user specified "return-path" capability would
4240 * QEMU uses the return path.
4242 if (migrate_postcopy_ram() || migrate_use_return_path()) {
4243 if (open_return_path_on_source(s
, !resume
)) {
4244 error_report("Unable to open return-path for postcopy");
4245 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
4246 migrate_fd_cleanup(s
);
4251 /* This needs to be done before resuming a postcopy */
4252 if (postcopy_preempt_setup(s
, &local_err
)) {
4253 error_report_err(local_err
);
4254 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
4255 MIGRATION_STATUS_FAILED
);
4256 migrate_fd_cleanup(s
);
4261 /* Wakeup the main migration thread to do the recovery */
4262 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
4263 MIGRATION_STATUS_POSTCOPY_RECOVER
);
4264 qemu_sem_post(&s
->postcopy_pause_sem
);
4268 if (multifd_save_setup(&local_err
) != 0) {
4269 error_report_err(local_err
);
4270 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
4271 MIGRATION_STATUS_FAILED
);
4272 migrate_fd_cleanup(s
);
4276 if (migrate_background_snapshot()) {
4277 qemu_thread_create(&s
->thread
, "bg_snapshot",
4278 bg_migration_thread
, s
, QEMU_THREAD_JOINABLE
);
4280 qemu_thread_create(&s
->thread
, "live_migration",
4281 migration_thread
, s
, QEMU_THREAD_JOINABLE
);
4283 s
->migration_thread_running
= true;
4286 void migration_global_dump(Monitor
*mon
)
4288 MigrationState
*ms
= migrate_get_current();
4290 monitor_printf(mon
, "globals:\n");
4291 monitor_printf(mon
, "store-global-state: %s\n",
4292 ms
->store_global_state
? "on" : "off");
4293 monitor_printf(mon
, "only-migratable: %s\n",
4294 only_migratable
? "on" : "off");
4295 monitor_printf(mon
, "send-configuration: %s\n",
4296 ms
->send_configuration
? "on" : "off");
4297 monitor_printf(mon
, "send-section-footer: %s\n",
4298 ms
->send_section_footer
? "on" : "off");
4299 monitor_printf(mon
, "decompress-error-check: %s\n",
4300 ms
->decompress_error_check
? "on" : "off");
4301 monitor_printf(mon
, "clear-bitmap-shift: %u\n",
4302 ms
->clear_bitmap_shift
);
4305 #define DEFINE_PROP_MIG_CAP(name, x) \
4306 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
4308 static Property migration_properties
[] = {
4309 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
4310 store_global_state
, true),
4311 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
4312 send_configuration
, true),
4313 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
4314 send_section_footer
, true),
4315 DEFINE_PROP_BOOL("decompress-error-check", MigrationState
,
4316 decompress_error_check
, true),
4317 DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState
,
4318 clear_bitmap_shift
, CLEAR_BITMAP_SHIFT_DEFAULT
),
4320 /* Migration parameters */
4321 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
4322 parameters
.compress_level
,
4323 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
4324 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
4325 parameters
.compress_threads
,
4326 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
4327 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState
,
4328 parameters
.compress_wait_thread
, true),
4329 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
4330 parameters
.decompress_threads
,
4331 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
4332 DEFINE_PROP_UINT8("x-throttle-trigger-threshold", MigrationState
,
4333 parameters
.throttle_trigger_threshold
,
4334 DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD
),
4335 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
4336 parameters
.cpu_throttle_initial
,
4337 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
4338 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
4339 parameters
.cpu_throttle_increment
,
4340 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
4341 DEFINE_PROP_BOOL("x-cpu-throttle-tailslow", MigrationState
,
4342 parameters
.cpu_throttle_tailslow
, false),
4343 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
4344 parameters
.max_bandwidth
, MAX_THROTTLE
),
4345 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
4346 parameters
.downtime_limit
,
4347 DEFAULT_MIGRATE_SET_DOWNTIME
),
4348 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
4349 parameters
.x_checkpoint_delay
,
4350 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
4351 DEFINE_PROP_UINT8("multifd-channels", MigrationState
,
4352 parameters
.multifd_channels
,
4353 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
4354 DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState
,
4355 parameters
.multifd_compression
,
4356 DEFAULT_MIGRATE_MULTIFD_COMPRESSION
),
4357 DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState
,
4358 parameters
.multifd_zlib_level
,
4359 DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL
),
4360 DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState
,
4361 parameters
.multifd_zstd_level
,
4362 DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL
),
4363 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
4364 parameters
.xbzrle_cache_size
,
4365 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
4366 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState
,
4367 parameters
.max_postcopy_bandwidth
,
4368 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH
),
4369 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState
,
4370 parameters
.max_cpu_throttle
,
4371 DEFAULT_MIGRATE_MAX_CPU_THROTTLE
),
4372 DEFINE_PROP_SIZE("announce-initial", MigrationState
,
4373 parameters
.announce_initial
,
4374 DEFAULT_MIGRATE_ANNOUNCE_INITIAL
),
4375 DEFINE_PROP_SIZE("announce-max", MigrationState
,
4376 parameters
.announce_max
,
4377 DEFAULT_MIGRATE_ANNOUNCE_MAX
),
4378 DEFINE_PROP_SIZE("announce-rounds", MigrationState
,
4379 parameters
.announce_rounds
,
4380 DEFAULT_MIGRATE_ANNOUNCE_ROUNDS
),
4381 DEFINE_PROP_SIZE("announce-step", MigrationState
,
4382 parameters
.announce_step
,
4383 DEFAULT_MIGRATE_ANNOUNCE_STEP
),
4384 DEFINE_PROP_BOOL("x-postcopy-preempt-break-huge", MigrationState
,
4385 postcopy_preempt_break_huge
, true),
4386 DEFINE_PROP_STRING("tls-creds", MigrationState
, parameters
.tls_creds
),
4387 DEFINE_PROP_STRING("tls-hostname", MigrationState
, parameters
.tls_hostname
),
4388 DEFINE_PROP_STRING("tls-authz", MigrationState
, parameters
.tls_authz
),
4390 /* Migration capabilities */
4391 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
4392 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
4393 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
4394 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
4395 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
4396 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
4397 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
4398 DEFINE_PROP_MIG_CAP("x-postcopy-preempt",
4399 MIGRATION_CAPABILITY_POSTCOPY_PREEMPT
),
4400 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
4401 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
4402 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
4403 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
4404 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD
),
4405 DEFINE_PROP_MIG_CAP("x-background-snapshot",
4406 MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
),
4408 DEFINE_PROP_MIG_CAP("x-zero-copy-send",
4409 MIGRATION_CAPABILITY_ZERO_COPY_SEND
),
4412 DEFINE_PROP_END_OF_LIST(),
4415 static void migration_class_init(ObjectClass
*klass
, void *data
)
4417 DeviceClass
*dc
= DEVICE_CLASS(klass
);
4419 dc
->user_creatable
= false;
4420 device_class_set_props(dc
, migration_properties
);
4423 static void migration_instance_finalize(Object
*obj
)
4425 MigrationState
*ms
= MIGRATION_OBJ(obj
);
4427 qemu_mutex_destroy(&ms
->error_mutex
);
4428 qemu_mutex_destroy(&ms
->qemu_file_lock
);
4429 qemu_sem_destroy(&ms
->wait_unplug_sem
);
4430 qemu_sem_destroy(&ms
->rate_limit_sem
);
4431 qemu_sem_destroy(&ms
->pause_sem
);
4432 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
4433 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
4434 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
4435 qemu_sem_destroy(&ms
->postcopy_qemufile_src_sem
);
4436 error_free(ms
->error
);
4439 static void migration_instance_init(Object
*obj
)
4441 MigrationState
*ms
= MIGRATION_OBJ(obj
);
4442 MigrationParameters
*params
= &ms
->parameters
;
4444 ms
->state
= MIGRATION_STATUS_NONE
;
4446 ms
->pages_per_second
= -1;
4447 qemu_sem_init(&ms
->pause_sem
, 0);
4448 qemu_mutex_init(&ms
->error_mutex
);
4450 params
->tls_hostname
= g_strdup("");
4451 params
->tls_creds
= g_strdup("");
4453 /* Set has_* up only for parameter checks */
4454 params
->has_compress_level
= true;
4455 params
->has_compress_threads
= true;
4456 params
->has_compress_wait_thread
= true;
4457 params
->has_decompress_threads
= true;
4458 params
->has_throttle_trigger_threshold
= true;
4459 params
->has_cpu_throttle_initial
= true;
4460 params
->has_cpu_throttle_increment
= true;
4461 params
->has_cpu_throttle_tailslow
= true;
4462 params
->has_max_bandwidth
= true;
4463 params
->has_downtime_limit
= true;
4464 params
->has_x_checkpoint_delay
= true;
4465 params
->has_block_incremental
= true;
4466 params
->has_multifd_channels
= true;
4467 params
->has_multifd_compression
= true;
4468 params
->has_multifd_zlib_level
= true;
4469 params
->has_multifd_zstd_level
= true;
4470 params
->has_xbzrle_cache_size
= true;
4471 params
->has_max_postcopy_bandwidth
= true;
4472 params
->has_max_cpu_throttle
= true;
4473 params
->has_announce_initial
= true;
4474 params
->has_announce_max
= true;
4475 params
->has_announce_rounds
= true;
4476 params
->has_announce_step
= true;
4477 params
->has_tls_creds
= true;
4478 params
->has_tls_hostname
= true;
4479 params
->has_tls_authz
= true;
4481 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
4482 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
4483 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
4484 qemu_sem_init(&ms
->rate_limit_sem
, 0);
4485 qemu_sem_init(&ms
->wait_unplug_sem
, 0);
4486 qemu_sem_init(&ms
->postcopy_qemufile_src_sem
, 0);
4487 qemu_mutex_init(&ms
->qemu_file_lock
);
4491 * Return true if check pass, false otherwise. Error will be put
4492 * inside errp if provided.
4494 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
4496 MigrationCapabilityStatusList
*head
= NULL
;
4497 /* Assuming all off */
4498 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
4501 if (!migrate_params_check(&ms
->parameters
, errp
)) {
4505 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
4506 if (ms
->enabled_capabilities
[i
]) {
4507 QAPI_LIST_PREPEND(head
, migrate_cap_add(i
, true));
4511 ret
= migrate_caps_check(cap_list
, head
, errp
);
4513 /* It works with head == NULL */
4514 qapi_free_MigrationCapabilityStatusList(head
);
4519 static const TypeInfo migration_type
= {
4520 .name
= TYPE_MIGRATION
,
4522 * NOTE: TYPE_MIGRATION is not really a device, as the object is
4523 * not created using qdev_new(), it is not attached to the qdev
4524 * device tree, and it is never realized.
4526 * TODO: Make this TYPE_OBJECT once QOM provides something like
4527 * TYPE_DEVICE's "-global" properties.
4529 .parent
= TYPE_DEVICE
,
4530 .class_init
= migration_class_init
,
4531 .class_size
= sizeof(MigrationClass
),
4532 .instance_size
= sizeof(MigrationState
),
4533 .instance_init
= migration_instance_init
,
4534 .instance_finalize
= migration_instance_finalize
,
4537 static void register_migration_types(void)
4539 type_register_static(&migration_type
);
4542 type_init(register_migration_types
);