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 process_incoming_migration_co(void *opaque
)
579 MigrationIncomingState
*mis
= migration_incoming_get_current();
582 Error
*local_err
= NULL
;
584 assert(mis
->from_src_file
);
585 mis
->migration_incoming_co
= qemu_coroutine_self();
586 mis
->largest_page_size
= qemu_ram_pagesize_largest();
587 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
588 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
589 MIGRATION_STATUS_ACTIVE
);
590 ret
= qemu_loadvm_state(mis
->from_src_file
);
592 ps
= postcopy_state_get();
593 trace_process_incoming_migration_co_end(ret
, ps
);
594 if (ps
!= POSTCOPY_INCOMING_NONE
) {
595 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
597 * Where a migration had postcopy enabled (and thus went to advise)
598 * but managed to complete within the precopy period, we can use
601 postcopy_ram_incoming_cleanup(mis
);
602 } else if (ret
>= 0) {
604 * Postcopy was started, cleanup should happen at the end of the
607 trace_process_incoming_migration_co_postcopy_end_main();
610 /* Else if something went wrong then just fall out of the normal exit */
613 /* we get COLO info, and know if we are in COLO mode */
614 if (!ret
&& migration_incoming_colo_enabled()) {
615 /* Make sure all file formats throw away their mutable metadata */
616 bdrv_activate_all(&local_err
);
618 error_report_err(local_err
);
622 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
623 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
624 mis
->have_colo_incoming_thread
= true;
625 qemu_coroutine_yield();
627 qemu_mutex_unlock_iothread();
628 /* Wait checkpoint incoming thread exit before free resource */
629 qemu_thread_join(&mis
->colo_incoming_thread
);
630 qemu_mutex_lock_iothread();
631 /* We hold the global iothread lock, so it is safe here */
632 colo_release_ram_cache();
636 error_report("load of migration failed: %s", strerror(-ret
));
639 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
640 qemu_bh_schedule(mis
->bh
);
641 mis
->migration_incoming_co
= NULL
;
645 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
646 MIGRATION_STATUS_FAILED
);
647 qemu_fclose(mis
->from_src_file
);
648 if (multifd_load_cleanup(&local_err
) != 0) {
649 error_report_err(local_err
);
655 * migration_incoming_setup: Setup incoming migration
656 * @f: file for main migration channel
657 * @errp: where to put errors
659 * Returns: %true on success, %false on error.
661 static bool migration_incoming_setup(QEMUFile
*f
, Error
**errp
)
663 MigrationIncomingState
*mis
= migration_incoming_get_current();
665 if (multifd_load_setup(errp
) != 0) {
669 if (!mis
->from_src_file
) {
670 mis
->from_src_file
= f
;
672 qemu_file_set_blocking(f
, false);
676 void migration_incoming_process(void)
678 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
679 qemu_coroutine_enter(co
);
682 /* Returns true if recovered from a paused migration, otherwise false */
683 static bool postcopy_try_recover(void)
685 MigrationIncomingState
*mis
= migration_incoming_get_current();
687 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
688 /* Resumed from a paused postcopy migration */
690 /* This should be set already in migration_incoming_setup() */
691 assert(mis
->from_src_file
);
692 /* Postcopy has standalone thread to do vm load */
693 qemu_file_set_blocking(mis
->from_src_file
, true);
695 /* Re-configure the return path */
696 mis
->to_src_file
= qemu_file_get_return_path(mis
->from_src_file
);
698 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
699 MIGRATION_STATUS_POSTCOPY_RECOVER
);
702 * Here, we only wake up the main loading thread (while the
703 * rest threads will still be waiting), so that we can receive
704 * commands from source now, and answer it if needed. The
705 * rest threads will be woken up afterwards until we are sure
706 * that source is ready to reply to page requests.
708 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
715 void migration_fd_process_incoming(QEMUFile
*f
, Error
**errp
)
717 if (!migration_incoming_setup(f
, errp
)) {
720 if (postcopy_try_recover()) {
723 migration_incoming_process();
726 static bool migration_needs_multiple_sockets(void)
728 return migrate_use_multifd() || migrate_postcopy_preempt();
731 void migration_ioc_process_incoming(QIOChannel
*ioc
, Error
**errp
)
733 MigrationIncomingState
*mis
= migration_incoming_get_current();
734 Error
*local_err
= NULL
;
735 bool start_migration
;
738 if (!mis
->from_src_file
) {
739 /* The first connection (multifd may have multiple) */
740 f
= qemu_file_new_input(ioc
);
742 if (!migration_incoming_setup(f
, errp
)) {
747 * Common migration only needs one channel, so we can start
748 * right now. Some features need more than one channel, we wait.
750 start_migration
= !migration_needs_multiple_sockets();
752 /* Multiple connections */
753 assert(migration_needs_multiple_sockets());
754 if (migrate_use_multifd()) {
755 start_migration
= multifd_recv_new_channel(ioc
, &local_err
);
757 assert(migrate_postcopy_preempt());
758 f
= qemu_file_new_input(ioc
);
759 start_migration
= postcopy_preempt_new_channel(mis
, f
);
762 error_propagate(errp
, local_err
);
767 if (start_migration
) {
768 /* If it's a recovery, we're done */
769 if (postcopy_try_recover()) {
772 migration_incoming_process();
777 * @migration_has_all_channels: We have received all channels that we need
779 * Returns true when we have got connections to all the channels that
780 * we need for migration.
782 bool migration_has_all_channels(void)
784 MigrationIncomingState
*mis
= migration_incoming_get_current();
786 if (!mis
->from_src_file
) {
790 if (migrate_use_multifd()) {
791 return multifd_recv_all_channels_created();
794 if (migrate_postcopy_preempt()) {
795 return mis
->postcopy_qemufile_dst
!= NULL
;
802 * Send a 'SHUT' message on the return channel with the given value
803 * to indicate that we've finished with the RP. Non-0 value indicates
806 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
811 buf
= cpu_to_be32(value
);
812 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
816 * Send a 'PONG' message on the return channel with the given value
817 * (normally in response to a 'PING')
819 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
824 buf
= cpu_to_be32(value
);
825 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
828 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
836 * First, we send the header part. It contains only the len of
837 * idstr, and the idstr itself.
839 len
= strlen(block_name
);
841 memcpy(buf
+ 1, block_name
, len
);
843 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
844 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
849 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
852 * Next, we dump the received bitmap to the stream.
854 * TODO: currently we are safe since we are the only one that is
855 * using the to_src_file handle (fault thread is still paused),
856 * and it's ok even not taking the mutex. However the best way is
857 * to take the lock before sending the message header, and release
858 * the lock after sending the bitmap.
860 qemu_mutex_lock(&mis
->rp_mutex
);
861 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
862 qemu_mutex_unlock(&mis
->rp_mutex
);
864 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
867 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
871 buf
= cpu_to_be32(value
);
872 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
875 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
877 MigrationCapabilityStatusList
*head
= NULL
, **tail
= &head
;
878 MigrationCapabilityStatus
*caps
;
879 MigrationState
*s
= migrate_get_current();
882 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
883 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
884 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
888 caps
= g_malloc0(sizeof(*caps
));
889 caps
->capability
= i
;
890 caps
->state
= s
->enabled_capabilities
[i
];
891 QAPI_LIST_APPEND(tail
, caps
);
897 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
899 MigrationParameters
*params
;
900 MigrationState
*s
= migrate_get_current();
902 /* TODO use QAPI_CLONE() instead of duplicating it inline */
903 params
= g_malloc0(sizeof(*params
));
904 params
->has_compress_level
= true;
905 params
->compress_level
= s
->parameters
.compress_level
;
906 params
->has_compress_threads
= true;
907 params
->compress_threads
= s
->parameters
.compress_threads
;
908 params
->has_compress_wait_thread
= true;
909 params
->compress_wait_thread
= s
->parameters
.compress_wait_thread
;
910 params
->has_decompress_threads
= true;
911 params
->decompress_threads
= s
->parameters
.decompress_threads
;
912 params
->has_throttle_trigger_threshold
= true;
913 params
->throttle_trigger_threshold
= s
->parameters
.throttle_trigger_threshold
;
914 params
->has_cpu_throttle_initial
= true;
915 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
916 params
->has_cpu_throttle_increment
= true;
917 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
918 params
->has_cpu_throttle_tailslow
= true;
919 params
->cpu_throttle_tailslow
= s
->parameters
.cpu_throttle_tailslow
;
920 params
->has_tls_creds
= true;
921 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
922 params
->has_tls_hostname
= true;
923 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
924 params
->has_tls_authz
= true;
925 params
->tls_authz
= g_strdup(s
->parameters
.tls_authz
?
926 s
->parameters
.tls_authz
: "");
927 params
->has_max_bandwidth
= true;
928 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
929 params
->has_downtime_limit
= true;
930 params
->downtime_limit
= s
->parameters
.downtime_limit
;
931 params
->has_x_checkpoint_delay
= true;
932 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
933 params
->has_block_incremental
= true;
934 params
->block_incremental
= s
->parameters
.block_incremental
;
935 params
->has_multifd_channels
= true;
936 params
->multifd_channels
= s
->parameters
.multifd_channels
;
937 params
->has_multifd_compression
= true;
938 params
->multifd_compression
= s
->parameters
.multifd_compression
;
939 params
->has_multifd_zlib_level
= true;
940 params
->multifd_zlib_level
= s
->parameters
.multifd_zlib_level
;
941 params
->has_multifd_zstd_level
= true;
942 params
->multifd_zstd_level
= s
->parameters
.multifd_zstd_level
;
943 params
->has_xbzrle_cache_size
= true;
944 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
945 params
->has_max_postcopy_bandwidth
= true;
946 params
->max_postcopy_bandwidth
= s
->parameters
.max_postcopy_bandwidth
;
947 params
->has_max_cpu_throttle
= true;
948 params
->max_cpu_throttle
= s
->parameters
.max_cpu_throttle
;
949 params
->has_announce_initial
= true;
950 params
->announce_initial
= s
->parameters
.announce_initial
;
951 params
->has_announce_max
= true;
952 params
->announce_max
= s
->parameters
.announce_max
;
953 params
->has_announce_rounds
= true;
954 params
->announce_rounds
= s
->parameters
.announce_rounds
;
955 params
->has_announce_step
= true;
956 params
->announce_step
= s
->parameters
.announce_step
;
958 if (s
->parameters
.has_block_bitmap_mapping
) {
959 params
->has_block_bitmap_mapping
= true;
960 params
->block_bitmap_mapping
=
961 QAPI_CLONE(BitmapMigrationNodeAliasList
,
962 s
->parameters
.block_bitmap_mapping
);
968 AnnounceParameters
*migrate_announce_params(void)
970 static AnnounceParameters ap
;
972 MigrationState
*s
= migrate_get_current();
974 ap
.initial
= s
->parameters
.announce_initial
;
975 ap
.max
= s
->parameters
.announce_max
;
976 ap
.rounds
= s
->parameters
.announce_rounds
;
977 ap
.step
= s
->parameters
.announce_step
;
983 * Return true if we're already in the middle of a migration
984 * (i.e. any of the active or setup states)
986 bool migration_is_setup_or_active(int state
)
989 case MIGRATION_STATUS_ACTIVE
:
990 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
991 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
992 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
993 case MIGRATION_STATUS_SETUP
:
994 case MIGRATION_STATUS_PRE_SWITCHOVER
:
995 case MIGRATION_STATUS_DEVICE
:
996 case MIGRATION_STATUS_WAIT_UNPLUG
:
997 case MIGRATION_STATUS_COLO
:
1006 bool migration_is_running(int state
)
1009 case MIGRATION_STATUS_ACTIVE
:
1010 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1011 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1012 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1013 case MIGRATION_STATUS_SETUP
:
1014 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1015 case MIGRATION_STATUS_DEVICE
:
1016 case MIGRATION_STATUS_WAIT_UNPLUG
:
1017 case MIGRATION_STATUS_CANCELLING
:
1026 static void populate_time_info(MigrationInfo
*info
, MigrationState
*s
)
1028 info
->has_status
= true;
1029 info
->has_setup_time
= true;
1030 info
->setup_time
= s
->setup_time
;
1031 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
1032 info
->has_total_time
= true;
1033 info
->total_time
= s
->total_time
;
1034 info
->has_downtime
= true;
1035 info
->downtime
= s
->downtime
;
1037 info
->has_total_time
= true;
1038 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) -
1040 info
->has_expected_downtime
= true;
1041 info
->expected_downtime
= s
->expected_downtime
;
1045 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
1047 size_t page_size
= qemu_target_page_size();
1049 info
->has_ram
= true;
1050 info
->ram
= g_malloc0(sizeof(*info
->ram
));
1051 info
->ram
->transferred
= ram_counters
.transferred
;
1052 info
->ram
->total
= ram_bytes_total();
1053 info
->ram
->duplicate
= ram_counters
.duplicate
;
1054 /* legacy value. It is not used anymore */
1055 info
->ram
->skipped
= 0;
1056 info
->ram
->normal
= ram_counters
.normal
;
1057 info
->ram
->normal_bytes
= ram_counters
.normal
* page_size
;
1058 info
->ram
->mbps
= s
->mbps
;
1059 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
1060 info
->ram
->dirty_sync_missed_zero_copy
=
1061 ram_counters
.dirty_sync_missed_zero_copy
;
1062 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
1063 info
->ram
->page_size
= page_size
;
1064 info
->ram
->multifd_bytes
= ram_counters
.multifd_bytes
;
1065 info
->ram
->pages_per_second
= s
->pages_per_second
;
1066 info
->ram
->precopy_bytes
= ram_counters
.precopy_bytes
;
1067 info
->ram
->downtime_bytes
= ram_counters
.downtime_bytes
;
1068 info
->ram
->postcopy_bytes
= ram_counters
.postcopy_bytes
;
1070 if (migrate_use_xbzrle()) {
1071 info
->has_xbzrle_cache
= true;
1072 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
1073 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
1074 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
1075 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
1076 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
1077 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
1078 info
->xbzrle_cache
->encoding_rate
= xbzrle_counters
.encoding_rate
;
1079 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
1082 if (migrate_use_compression()) {
1083 info
->has_compression
= true;
1084 info
->compression
= g_malloc0(sizeof(*info
->compression
));
1085 info
->compression
->pages
= compression_counters
.pages
;
1086 info
->compression
->busy
= compression_counters
.busy
;
1087 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
1088 info
->compression
->compressed_size
=
1089 compression_counters
.compressed_size
;
1090 info
->compression
->compression_rate
=
1091 compression_counters
.compression_rate
;
1094 if (cpu_throttle_active()) {
1095 info
->has_cpu_throttle_percentage
= true;
1096 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
1099 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
1100 info
->ram
->remaining
= ram_bytes_remaining();
1101 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
1105 static void populate_disk_info(MigrationInfo
*info
)
1107 if (blk_mig_active()) {
1108 info
->has_disk
= true;
1109 info
->disk
= g_malloc0(sizeof(*info
->disk
));
1110 info
->disk
->transferred
= blk_mig_bytes_transferred();
1111 info
->disk
->remaining
= blk_mig_bytes_remaining();
1112 info
->disk
->total
= blk_mig_bytes_total();
1116 static void fill_source_migration_info(MigrationInfo
*info
)
1118 MigrationState
*s
= migrate_get_current();
1119 int state
= qatomic_read(&s
->state
);
1120 GSList
*cur_blocker
= migration_blockers
;
1122 info
->blocked_reasons
= NULL
;
1125 * There are two types of reasons a migration might be blocked;
1126 * a) devices marked in VMState as non-migratable, and
1127 * b) Explicit migration blockers
1128 * We need to add both of them here.
1130 qemu_savevm_non_migratable_list(&info
->blocked_reasons
);
1132 while (cur_blocker
) {
1133 QAPI_LIST_PREPEND(info
->blocked_reasons
,
1134 g_strdup(error_get_pretty(cur_blocker
->data
)));
1135 cur_blocker
= g_slist_next(cur_blocker
);
1137 info
->has_blocked_reasons
= info
->blocked_reasons
!= NULL
;
1140 case MIGRATION_STATUS_NONE
:
1141 /* no migration has happened ever */
1142 /* do not overwrite destination migration status */
1144 case MIGRATION_STATUS_SETUP
:
1145 info
->has_status
= true;
1146 info
->has_total_time
= false;
1148 case MIGRATION_STATUS_ACTIVE
:
1149 case MIGRATION_STATUS_CANCELLING
:
1150 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1151 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1152 case MIGRATION_STATUS_DEVICE
:
1153 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1154 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1155 /* TODO add some postcopy stats */
1156 populate_time_info(info
, s
);
1157 populate_ram_info(info
, s
);
1158 populate_disk_info(info
);
1159 populate_vfio_info(info
);
1161 case MIGRATION_STATUS_COLO
:
1162 info
->has_status
= true;
1163 /* TODO: display COLO specific information (checkpoint info etc.) */
1165 case MIGRATION_STATUS_COMPLETED
:
1166 populate_time_info(info
, s
);
1167 populate_ram_info(info
, s
);
1168 populate_vfio_info(info
);
1170 case MIGRATION_STATUS_FAILED
:
1171 info
->has_status
= true;
1173 info
->has_error_desc
= true;
1174 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
1177 case MIGRATION_STATUS_CANCELLED
:
1178 info
->has_status
= true;
1180 case MIGRATION_STATUS_WAIT_UNPLUG
:
1181 info
->has_status
= true;
1184 info
->status
= state
;
1187 typedef enum WriteTrackingSupport
{
1188 WT_SUPPORT_UNKNOWN
= 0,
1190 WT_SUPPORT_AVAILABLE
,
1191 WT_SUPPORT_COMPATIBLE
1192 } WriteTrackingSupport
;
1195 WriteTrackingSupport
migrate_query_write_tracking(void)
1197 /* Check if kernel supports required UFFD features */
1198 if (!ram_write_tracking_available()) {
1199 return WT_SUPPORT_ABSENT
;
1202 * Check if current memory configuration is
1203 * compatible with required UFFD features.
1205 if (!ram_write_tracking_compatible()) {
1206 return WT_SUPPORT_AVAILABLE
;
1209 return WT_SUPPORT_COMPATIBLE
;
1213 * @migration_caps_check - check capability validity
1215 * @cap_list: old capability list, array of bool
1216 * @params: new capabilities to be applied soon
1217 * @errp: set *errp if the check failed, with reason
1219 * Returns true if check passed, otherwise false.
1221 static bool migrate_caps_check(bool *cap_list
,
1222 MigrationCapabilityStatusList
*params
,
1225 MigrationCapabilityStatusList
*cap
;
1226 bool old_postcopy_cap
;
1227 MigrationIncomingState
*mis
= migration_incoming_get_current();
1229 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1231 for (cap
= params
; cap
; cap
= cap
->next
) {
1232 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
1235 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1236 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
1237 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
1239 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
1244 #ifndef CONFIG_REPLICATION
1245 if (cap_list
[MIGRATION_CAPABILITY_X_COLO
]) {
1246 error_setg(errp
, "QEMU compiled without replication module"
1247 " can't enable COLO");
1248 error_append_hint(errp
, "Please enable replication before COLO.\n");
1253 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
1254 /* This check is reasonably expensive, so only when it's being
1255 * set the first time, also it's only the destination that needs
1258 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
1259 !postcopy_ram_supported_by_host(mis
)) {
1260 /* postcopy_ram_supported_by_host will have emitted a more
1263 error_setg(errp
, "Postcopy is not supported");
1267 if (cap_list
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
]) {
1268 error_setg(errp
, "Postcopy is not compatible with ignore-shared");
1273 if (cap_list
[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
]) {
1274 WriteTrackingSupport wt_support
;
1277 * Check if 'background-snapshot' capability is supported by
1278 * host kernel and compatible with guest memory configuration.
1280 wt_support
= migrate_query_write_tracking();
1281 if (wt_support
< WT_SUPPORT_AVAILABLE
) {
1282 error_setg(errp
, "Background-snapshot is not supported by host kernel");
1285 if (wt_support
< WT_SUPPORT_COMPATIBLE
) {
1286 error_setg(errp
, "Background-snapshot is not compatible "
1287 "with guest memory configuration");
1292 * Check if there are any migration capabilities
1293 * incompatible with 'background-snapshot'.
1295 for (idx
= 0; idx
< check_caps_background_snapshot
.size
; idx
++) {
1296 int incomp_cap
= check_caps_background_snapshot
.caps
[idx
];
1297 if (cap_list
[incomp_cap
]) {
1299 "Background-snapshot is not compatible with %s",
1300 MigrationCapability_str(incomp_cap
));
1307 if (cap_list
[MIGRATION_CAPABILITY_ZERO_COPY_SEND
] &&
1308 (!cap_list
[MIGRATION_CAPABILITY_MULTIFD
] ||
1309 cap_list
[MIGRATION_CAPABILITY_COMPRESS
] ||
1310 cap_list
[MIGRATION_CAPABILITY_XBZRLE
] ||
1311 migrate_multifd_compression() ||
1312 migrate_use_tls())) {
1314 "Zero copy only available for non-compressed non-TLS multifd migration");
1318 if (cap_list
[MIGRATION_CAPABILITY_ZERO_COPY_SEND
]) {
1320 "Zero copy currently only available on Linux");
1326 /* incoming side only */
1327 if (runstate_check(RUN_STATE_INMIGRATE
) &&
1328 !migrate_multi_channels_is_allowed() &&
1329 cap_list
[MIGRATION_CAPABILITY_MULTIFD
]) {
1330 error_setg(errp
, "multifd is not supported by current protocol");
1334 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_PREEMPT
]) {
1335 if (!cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
1336 error_setg(errp
, "Postcopy preempt requires postcopy-ram");
1344 static void fill_destination_migration_info(MigrationInfo
*info
)
1346 MigrationIncomingState
*mis
= migration_incoming_get_current();
1348 if (mis
->socket_address_list
) {
1349 info
->has_socket_address
= true;
1350 info
->socket_address
=
1351 QAPI_CLONE(SocketAddressList
, mis
->socket_address_list
);
1354 switch (mis
->state
) {
1355 case MIGRATION_STATUS_NONE
:
1357 case MIGRATION_STATUS_SETUP
:
1358 case MIGRATION_STATUS_CANCELLING
:
1359 case MIGRATION_STATUS_CANCELLED
:
1360 case MIGRATION_STATUS_ACTIVE
:
1361 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1362 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1363 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1364 case MIGRATION_STATUS_FAILED
:
1365 case MIGRATION_STATUS_COLO
:
1366 info
->has_status
= true;
1368 case MIGRATION_STATUS_COMPLETED
:
1369 info
->has_status
= true;
1370 fill_destination_postcopy_migration_info(info
);
1373 info
->status
= mis
->state
;
1376 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1378 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1380 fill_destination_migration_info(info
);
1381 fill_source_migration_info(info
);
1386 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
1389 MigrationState
*s
= migrate_get_current();
1390 MigrationCapabilityStatusList
*cap
;
1391 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
1393 if (migration_is_running(s
->state
)) {
1394 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1398 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
1399 if (!migrate_caps_check(cap_list
, params
, errp
)) {
1403 for (cap
= params
; cap
; cap
= cap
->next
) {
1404 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
1409 * Check whether the parameters are valid. Error will be put into errp
1410 * (if provided). Return true if valid, otherwise false.
1412 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
1414 if (params
->has_compress_level
&&
1415 (params
->compress_level
> 9)) {
1416 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
1417 "a value between 0 and 9");
1421 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
1422 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1424 "a value between 1 and 255");
1428 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
1429 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1430 "decompress_threads",
1431 "a value between 1 and 255");
1435 if (params
->has_throttle_trigger_threshold
&&
1436 (params
->throttle_trigger_threshold
< 1 ||
1437 params
->throttle_trigger_threshold
> 100)) {
1438 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1439 "throttle_trigger_threshold",
1440 "an integer in the range of 1 to 100");
1444 if (params
->has_cpu_throttle_initial
&&
1445 (params
->cpu_throttle_initial
< 1 ||
1446 params
->cpu_throttle_initial
> 99)) {
1447 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1448 "cpu_throttle_initial",
1449 "an integer in the range of 1 to 99");
1453 if (params
->has_cpu_throttle_increment
&&
1454 (params
->cpu_throttle_increment
< 1 ||
1455 params
->cpu_throttle_increment
> 99)) {
1456 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1457 "cpu_throttle_increment",
1458 "an integer in the range of 1 to 99");
1462 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
1463 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1465 "an integer in the range of 0 to "stringify(SIZE_MAX
)
1470 if (params
->has_downtime_limit
&&
1471 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
1472 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1474 "an integer in the range of 0 to "
1475 stringify(MAX_MIGRATE_DOWNTIME
)" ms");
1479 /* x_checkpoint_delay is now always positive */
1481 if (params
->has_multifd_channels
&& (params
->multifd_channels
< 1)) {
1482 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1484 "a value between 1 and 255");
1488 if (params
->has_multifd_zlib_level
&&
1489 (params
->multifd_zlib_level
> 9)) {
1490 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zlib_level",
1491 "a value between 0 and 9");
1495 if (params
->has_multifd_zstd_level
&&
1496 (params
->multifd_zstd_level
> 20)) {
1497 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zstd_level",
1498 "a value between 0 and 20");
1502 if (params
->has_xbzrle_cache_size
&&
1503 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
1504 !is_power_of_2(params
->xbzrle_cache_size
))) {
1505 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1506 "xbzrle_cache_size",
1507 "a power of two no less than the target page size");
1511 if (params
->has_max_cpu_throttle
&&
1512 (params
->max_cpu_throttle
< params
->cpu_throttle_initial
||
1513 params
->max_cpu_throttle
> 99)) {
1514 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1516 "an integer in the range of cpu_throttle_initial to 99");
1520 if (params
->has_announce_initial
&&
1521 params
->announce_initial
> 100000) {
1522 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1524 "a value between 0 and 100000");
1527 if (params
->has_announce_max
&&
1528 params
->announce_max
> 100000) {
1529 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1531 "a value between 0 and 100000");
1534 if (params
->has_announce_rounds
&&
1535 params
->announce_rounds
> 1000) {
1536 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1538 "a value between 0 and 1000");
1541 if (params
->has_announce_step
&&
1542 (params
->announce_step
< 1 ||
1543 params
->announce_step
> 10000)) {
1544 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1546 "a value between 0 and 10000");
1550 if (params
->has_block_bitmap_mapping
&&
1551 !check_dirty_bitmap_mig_alias_map(params
->block_bitmap_mapping
, errp
)) {
1552 error_prepend(errp
, "Invalid mapping given for block-bitmap-mapping: ");
1557 if (migrate_use_zero_copy_send() &&
1558 ((params
->has_multifd_compression
&& params
->multifd_compression
) ||
1559 (params
->has_tls_creds
&& params
->tls_creds
&& *params
->tls_creds
))) {
1561 "Zero copy only available for non-compressed non-TLS multifd migration");
1569 static void migrate_params_test_apply(MigrateSetParameters
*params
,
1570 MigrationParameters
*dest
)
1572 *dest
= migrate_get_current()->parameters
;
1574 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1576 if (params
->has_compress_level
) {
1577 dest
->compress_level
= params
->compress_level
;
1580 if (params
->has_compress_threads
) {
1581 dest
->compress_threads
= params
->compress_threads
;
1584 if (params
->has_compress_wait_thread
) {
1585 dest
->compress_wait_thread
= params
->compress_wait_thread
;
1588 if (params
->has_decompress_threads
) {
1589 dest
->decompress_threads
= params
->decompress_threads
;
1592 if (params
->has_throttle_trigger_threshold
) {
1593 dest
->throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1596 if (params
->has_cpu_throttle_initial
) {
1597 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
1600 if (params
->has_cpu_throttle_increment
) {
1601 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
1604 if (params
->has_cpu_throttle_tailslow
) {
1605 dest
->cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1608 if (params
->has_tls_creds
) {
1609 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1610 dest
->tls_creds
= params
->tls_creds
->u
.s
;
1613 if (params
->has_tls_hostname
) {
1614 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1615 dest
->tls_hostname
= params
->tls_hostname
->u
.s
;
1618 if (params
->has_max_bandwidth
) {
1619 dest
->max_bandwidth
= params
->max_bandwidth
;
1622 if (params
->has_downtime_limit
) {
1623 dest
->downtime_limit
= params
->downtime_limit
;
1626 if (params
->has_x_checkpoint_delay
) {
1627 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
1630 if (params
->has_block_incremental
) {
1631 dest
->block_incremental
= params
->block_incremental
;
1633 if (params
->has_multifd_channels
) {
1634 dest
->multifd_channels
= params
->multifd_channels
;
1636 if (params
->has_multifd_compression
) {
1637 dest
->multifd_compression
= params
->multifd_compression
;
1639 if (params
->has_xbzrle_cache_size
) {
1640 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
1642 if (params
->has_max_postcopy_bandwidth
) {
1643 dest
->max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1645 if (params
->has_max_cpu_throttle
) {
1646 dest
->max_cpu_throttle
= params
->max_cpu_throttle
;
1648 if (params
->has_announce_initial
) {
1649 dest
->announce_initial
= params
->announce_initial
;
1651 if (params
->has_announce_max
) {
1652 dest
->announce_max
= params
->announce_max
;
1654 if (params
->has_announce_rounds
) {
1655 dest
->announce_rounds
= params
->announce_rounds
;
1657 if (params
->has_announce_step
) {
1658 dest
->announce_step
= params
->announce_step
;
1661 if (params
->has_block_bitmap_mapping
) {
1662 dest
->has_block_bitmap_mapping
= true;
1663 dest
->block_bitmap_mapping
= params
->block_bitmap_mapping
;
1667 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
1669 MigrationState
*s
= migrate_get_current();
1671 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1673 if (params
->has_compress_level
) {
1674 s
->parameters
.compress_level
= params
->compress_level
;
1677 if (params
->has_compress_threads
) {
1678 s
->parameters
.compress_threads
= params
->compress_threads
;
1681 if (params
->has_compress_wait_thread
) {
1682 s
->parameters
.compress_wait_thread
= params
->compress_wait_thread
;
1685 if (params
->has_decompress_threads
) {
1686 s
->parameters
.decompress_threads
= params
->decompress_threads
;
1689 if (params
->has_throttle_trigger_threshold
) {
1690 s
->parameters
.throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1693 if (params
->has_cpu_throttle_initial
) {
1694 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
1697 if (params
->has_cpu_throttle_increment
) {
1698 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
1701 if (params
->has_cpu_throttle_tailslow
) {
1702 s
->parameters
.cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1705 if (params
->has_tls_creds
) {
1706 g_free(s
->parameters
.tls_creds
);
1707 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1708 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1711 if (params
->has_tls_hostname
) {
1712 g_free(s
->parameters
.tls_hostname
);
1713 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1714 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1717 if (params
->has_tls_authz
) {
1718 g_free(s
->parameters
.tls_authz
);
1719 assert(params
->tls_authz
->type
== QTYPE_QSTRING
);
1720 s
->parameters
.tls_authz
= g_strdup(params
->tls_authz
->u
.s
);
1723 if (params
->has_max_bandwidth
) {
1724 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1725 if (s
->to_dst_file
&& !migration_in_postcopy()) {
1726 qemu_file_set_rate_limit(s
->to_dst_file
,
1727 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1731 if (params
->has_downtime_limit
) {
1732 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1735 if (params
->has_x_checkpoint_delay
) {
1736 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1737 if (migration_in_colo_state()) {
1738 colo_checkpoint_notify(s
);
1742 if (params
->has_block_incremental
) {
1743 s
->parameters
.block_incremental
= params
->block_incremental
;
1745 if (params
->has_multifd_channels
) {
1746 s
->parameters
.multifd_channels
= params
->multifd_channels
;
1748 if (params
->has_multifd_compression
) {
1749 s
->parameters
.multifd_compression
= params
->multifd_compression
;
1751 if (params
->has_xbzrle_cache_size
) {
1752 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1753 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1755 if (params
->has_max_postcopy_bandwidth
) {
1756 s
->parameters
.max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1757 if (s
->to_dst_file
&& migration_in_postcopy()) {
1758 qemu_file_set_rate_limit(s
->to_dst_file
,
1759 s
->parameters
.max_postcopy_bandwidth
/ XFER_LIMIT_RATIO
);
1762 if (params
->has_max_cpu_throttle
) {
1763 s
->parameters
.max_cpu_throttle
= params
->max_cpu_throttle
;
1765 if (params
->has_announce_initial
) {
1766 s
->parameters
.announce_initial
= params
->announce_initial
;
1768 if (params
->has_announce_max
) {
1769 s
->parameters
.announce_max
= params
->announce_max
;
1771 if (params
->has_announce_rounds
) {
1772 s
->parameters
.announce_rounds
= params
->announce_rounds
;
1774 if (params
->has_announce_step
) {
1775 s
->parameters
.announce_step
= params
->announce_step
;
1778 if (params
->has_block_bitmap_mapping
) {
1779 qapi_free_BitmapMigrationNodeAliasList(
1780 s
->parameters
.block_bitmap_mapping
);
1782 s
->parameters
.has_block_bitmap_mapping
= true;
1783 s
->parameters
.block_bitmap_mapping
=
1784 QAPI_CLONE(BitmapMigrationNodeAliasList
,
1785 params
->block_bitmap_mapping
);
1789 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1791 MigrationParameters tmp
;
1793 /* TODO Rewrite "" to null instead */
1794 if (params
->has_tls_creds
1795 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1796 qobject_unref(params
->tls_creds
->u
.n
);
1797 params
->tls_creds
->type
= QTYPE_QSTRING
;
1798 params
->tls_creds
->u
.s
= strdup("");
1800 /* TODO Rewrite "" to null instead */
1801 if (params
->has_tls_hostname
1802 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1803 qobject_unref(params
->tls_hostname
->u
.n
);
1804 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1805 params
->tls_hostname
->u
.s
= strdup("");
1808 migrate_params_test_apply(params
, &tmp
);
1810 if (!migrate_params_check(&tmp
, errp
)) {
1811 /* Invalid parameter */
1815 migrate_params_apply(params
, errp
);
1819 void qmp_migrate_start_postcopy(Error
**errp
)
1821 MigrationState
*s
= migrate_get_current();
1823 if (!migrate_postcopy()) {
1824 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1825 " the start of migration");
1829 if (s
->state
== MIGRATION_STATUS_NONE
) {
1830 error_setg(errp
, "Postcopy must be started after migration has been"
1835 * we don't error if migration has finished since that would be racy
1836 * with issuing this command.
1838 qatomic_set(&s
->start_postcopy
, true);
1841 /* shared migration helpers */
1843 void migrate_set_state(int *state
, int old_state
, int new_state
)
1845 assert(new_state
< MIGRATION_STATUS__MAX
);
1846 if (qatomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1847 trace_migrate_set_state(MigrationStatus_str(new_state
));
1848 migrate_generate_event(new_state
);
1852 static MigrationCapabilityStatus
*migrate_cap_add(MigrationCapability index
,
1855 MigrationCapabilityStatus
*cap
;
1857 cap
= g_new0(MigrationCapabilityStatus
, 1);
1858 cap
->capability
= index
;
1864 void migrate_set_block_enabled(bool value
, Error
**errp
)
1866 MigrationCapabilityStatusList
*cap
= NULL
;
1868 QAPI_LIST_PREPEND(cap
, migrate_cap_add(MIGRATION_CAPABILITY_BLOCK
, value
));
1869 qmp_migrate_set_capabilities(cap
, errp
);
1870 qapi_free_MigrationCapabilityStatusList(cap
);
1873 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1875 s
->parameters
.block_incremental
= value
;
1878 static void block_cleanup_parameters(MigrationState
*s
)
1880 if (s
->must_remove_block_options
) {
1881 /* setting to false can never fail */
1882 migrate_set_block_enabled(false, &error_abort
);
1883 migrate_set_block_incremental(s
, false);
1884 s
->must_remove_block_options
= false;
1888 static void migrate_fd_cleanup(MigrationState
*s
)
1890 qemu_bh_delete(s
->cleanup_bh
);
1891 s
->cleanup_bh
= NULL
;
1893 g_free(s
->hostname
);
1896 qemu_savevm_state_cleanup();
1898 if (s
->to_dst_file
) {
1901 trace_migrate_fd_cleanup();
1902 qemu_mutex_unlock_iothread();
1903 if (s
->migration_thread_running
) {
1904 qemu_thread_join(&s
->thread
);
1905 s
->migration_thread_running
= false;
1907 qemu_mutex_lock_iothread();
1909 multifd_save_cleanup();
1910 qemu_mutex_lock(&s
->qemu_file_lock
);
1911 tmp
= s
->to_dst_file
;
1912 s
->to_dst_file
= NULL
;
1913 qemu_mutex_unlock(&s
->qemu_file_lock
);
1915 * Close the file handle without the lock to make sure the
1916 * critical section won't block for long.
1918 migration_ioc_unregister_yank_from_file(tmp
);
1922 if (s
->postcopy_qemufile_src
) {
1923 migration_ioc_unregister_yank_from_file(s
->postcopy_qemufile_src
);
1924 qemu_fclose(s
->postcopy_qemufile_src
);
1925 s
->postcopy_qemufile_src
= NULL
;
1928 assert(!migration_is_active(s
));
1930 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1931 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1932 MIGRATION_STATUS_CANCELLED
);
1936 /* It is used on info migrate. We can't free it */
1937 error_report_err(error_copy(s
->error
));
1939 notifier_list_notify(&migration_state_notifiers
, s
);
1940 block_cleanup_parameters(s
);
1941 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1944 static void migrate_fd_cleanup_schedule(MigrationState
*s
)
1947 * Ref the state for bh, because it may be called when
1948 * there're already no other refs
1950 object_ref(OBJECT(s
));
1951 qemu_bh_schedule(s
->cleanup_bh
);
1954 static void migrate_fd_cleanup_bh(void *opaque
)
1956 MigrationState
*s
= opaque
;
1957 migrate_fd_cleanup(s
);
1958 object_unref(OBJECT(s
));
1961 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1963 QEMU_LOCK_GUARD(&s
->error_mutex
);
1965 s
->error
= error_copy(error
);
1969 static void migrate_error_free(MigrationState
*s
)
1971 QEMU_LOCK_GUARD(&s
->error_mutex
);
1973 error_free(s
->error
);
1978 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1980 trace_migrate_fd_error(error_get_pretty(error
));
1981 assert(s
->to_dst_file
== NULL
);
1982 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1983 MIGRATION_STATUS_FAILED
);
1984 migrate_set_error(s
, error
);
1987 static void migrate_fd_cancel(MigrationState
*s
)
1990 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1991 trace_migrate_fd_cancel();
1993 WITH_QEMU_LOCK_GUARD(&s
->qemu_file_lock
) {
1994 if (s
->rp_state
.from_dst_file
) {
1995 /* shutdown the rp socket, so causing the rp thread to shutdown */
1996 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
2001 old_state
= s
->state
;
2002 if (!migration_is_running(old_state
)) {
2005 /* If the migration is paused, kick it out of the pause */
2006 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
2007 qemu_sem_post(&s
->pause_sem
);
2009 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
2010 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
2013 * If we're unlucky the migration code might be stuck somewhere in a
2014 * send/write while the network has failed and is waiting to timeout;
2015 * if we've got shutdown(2) available then we can force it to quit.
2016 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
2017 * called in a bh, so there is no race against this cancel.
2019 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
2020 qemu_file_shutdown(f
);
2022 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
2023 Error
*local_err
= NULL
;
2025 bdrv_activate_all(&local_err
);
2027 error_report_err(local_err
);
2029 s
->block_inactive
= false;
2034 void add_migration_state_change_notifier(Notifier
*notify
)
2036 notifier_list_add(&migration_state_notifiers
, notify
);
2039 void remove_migration_state_change_notifier(Notifier
*notify
)
2041 notifier_remove(notify
);
2044 bool migration_in_setup(MigrationState
*s
)
2046 return s
->state
== MIGRATION_STATUS_SETUP
;
2049 bool migration_has_finished(MigrationState
*s
)
2051 return s
->state
== MIGRATION_STATUS_COMPLETED
;
2054 bool migration_has_failed(MigrationState
*s
)
2056 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
2057 s
->state
== MIGRATION_STATUS_FAILED
);
2060 bool migration_in_postcopy(void)
2062 MigrationState
*s
= migrate_get_current();
2065 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
2066 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
2067 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
2074 bool migration_in_postcopy_after_devices(MigrationState
*s
)
2076 return migration_in_postcopy() && s
->postcopy_after_devices
;
2079 bool migration_in_incoming_postcopy(void)
2081 PostcopyState ps
= postcopy_state_get();
2083 return ps
>= POSTCOPY_INCOMING_DISCARD
&& ps
< POSTCOPY_INCOMING_END
;
2086 bool migration_in_bg_snapshot(void)
2088 MigrationState
*s
= migrate_get_current();
2090 return migrate_background_snapshot() &&
2091 migration_is_setup_or_active(s
->state
);
2094 bool migration_is_idle(void)
2096 MigrationState
*s
= current_migration
;
2103 case MIGRATION_STATUS_NONE
:
2104 case MIGRATION_STATUS_CANCELLED
:
2105 case MIGRATION_STATUS_COMPLETED
:
2106 case MIGRATION_STATUS_FAILED
:
2108 case MIGRATION_STATUS_SETUP
:
2109 case MIGRATION_STATUS_CANCELLING
:
2110 case MIGRATION_STATUS_ACTIVE
:
2111 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
2112 case MIGRATION_STATUS_COLO
:
2113 case MIGRATION_STATUS_PRE_SWITCHOVER
:
2114 case MIGRATION_STATUS_DEVICE
:
2115 case MIGRATION_STATUS_WAIT_UNPLUG
:
2117 case MIGRATION_STATUS__MAX
:
2118 g_assert_not_reached();
2124 bool migration_is_active(MigrationState
*s
)
2126 return (s
->state
== MIGRATION_STATUS_ACTIVE
||
2127 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2130 void migrate_init(MigrationState
*s
)
2133 * Reinitialise all migration state, except
2134 * parameters/capabilities that the user set, and
2139 s
->to_dst_file
= NULL
;
2140 s
->state
= MIGRATION_STATUS_NONE
;
2141 s
->rp_state
.from_dst_file
= NULL
;
2142 s
->rp_state
.error
= false;
2144 s
->pages_per_second
= 0.0;
2146 s
->expected_downtime
= 0;
2148 s
->start_postcopy
= false;
2149 s
->postcopy_after_devices
= false;
2150 s
->migration_thread_running
= false;
2151 error_free(s
->error
);
2155 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
2157 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2159 s
->vm_was_running
= false;
2160 s
->iteration_initial_bytes
= 0;
2161 s
->threshold_size
= 0;
2164 int migrate_add_blocker_internal(Error
*reason
, Error
**errp
)
2166 /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */
2167 if (runstate_check(RUN_STATE_SAVE_VM
) || !migration_is_idle()) {
2168 error_propagate_prepend(errp
, error_copy(reason
),
2169 "disallowing migration blocker "
2170 "(migration/snapshot in progress) for: ");
2174 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
2178 int migrate_add_blocker(Error
*reason
, Error
**errp
)
2180 if (only_migratable
) {
2181 error_propagate_prepend(errp
, error_copy(reason
),
2182 "disallowing migration blocker "
2183 "(--only-migratable) for: ");
2187 return migrate_add_blocker_internal(reason
, errp
);
2190 void migrate_del_blocker(Error
*reason
)
2192 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
2195 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
2197 Error
*local_err
= NULL
;
2198 static bool once
= true;
2201 error_setg(errp
, "The incoming migration has already been started");
2204 if (!runstate_check(RUN_STATE_INMIGRATE
)) {
2205 error_setg(errp
, "'-incoming' was not specified on the command line");
2209 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
2213 qemu_start_incoming_migration(uri
, &local_err
);
2216 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2217 error_propagate(errp
, local_err
);
2224 void qmp_migrate_recover(const char *uri
, Error
**errp
)
2226 MigrationIncomingState
*mis
= migration_incoming_get_current();
2229 * Don't even bother to use ERRP_GUARD() as it _must_ always be set by
2230 * callers (no one should ignore a recover failure); if there is, it's a
2231 * programming error.
2235 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2236 error_setg(errp
, "Migrate recover can only be run "
2237 "when postcopy is paused.");
2241 /* If there's an existing transport, release it */
2242 migration_incoming_transport_cleanup(mis
);
2245 * Note that this call will never start a real migration; it will
2246 * only re-setup the migration stream and poke existing migration
2247 * to continue using that newly established channel.
2249 qemu_start_incoming_migration(uri
, errp
);
2252 void qmp_migrate_pause(Error
**errp
)
2254 MigrationState
*ms
= migrate_get_current();
2255 MigrationIncomingState
*mis
= migration_incoming_get_current();
2258 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2259 /* Source side, during postcopy */
2260 qemu_mutex_lock(&ms
->qemu_file_lock
);
2261 ret
= qemu_file_shutdown(ms
->to_dst_file
);
2262 qemu_mutex_unlock(&ms
->qemu_file_lock
);
2264 error_setg(errp
, "Failed to pause source migration");
2269 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2270 ret
= qemu_file_shutdown(mis
->from_src_file
);
2272 error_setg(errp
, "Failed to pause destination migration");
2277 error_setg(errp
, "migrate-pause is currently only supported "
2278 "during postcopy-active state");
2281 bool migration_is_blocked(Error
**errp
)
2283 if (qemu_savevm_state_blocked(errp
)) {
2287 if (migration_blockers
) {
2288 error_propagate(errp
, error_copy(migration_blockers
->data
));
2295 /* Returns true if continue to migrate, or false if error detected */
2296 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
2297 bool resume
, Error
**errp
)
2299 Error
*local_err
= NULL
;
2302 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2303 error_setg(errp
, "Cannot resume if there is no "
2304 "paused migration");
2309 * Postcopy recovery won't work well with release-ram
2310 * capability since release-ram will drop the page buffer as
2311 * long as the page is put into the send buffer. So if there
2312 * is a network failure happened, any page buffers that have
2313 * not yet reached the destination VM but have already been
2314 * sent from the source VM will be lost forever. Let's refuse
2315 * the client from resuming such a postcopy migration.
2316 * Luckily release-ram was designed to only be used when src
2317 * and destination VMs are on the same host, so it should be
2320 if (migrate_release_ram()) {
2321 error_setg(errp
, "Postcopy recovery cannot work "
2322 "when release-ram capability is set");
2326 /* This is a resume, skip init status */
2330 if (migration_is_running(s
->state
)) {
2331 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
2335 if (runstate_check(RUN_STATE_INMIGRATE
)) {
2336 error_setg(errp
, "Guest is waiting for an incoming migration");
2340 if (runstate_check(RUN_STATE_POSTMIGRATE
)) {
2341 error_setg(errp
, "Can't migrate the vm that was paused due to "
2342 "previous migration");
2346 if (migration_is_blocked(errp
)) {
2350 if (blk
|| blk_inc
) {
2351 if (migrate_colo_enabled()) {
2352 error_setg(errp
, "No disk migration is required in COLO mode");
2355 if (migrate_use_block() || migrate_use_block_incremental()) {
2356 error_setg(errp
, "Command options are incompatible with "
2357 "current migration capabilities");
2360 migrate_set_block_enabled(true, &local_err
);
2362 error_propagate(errp
, local_err
);
2365 s
->must_remove_block_options
= true;
2369 migrate_set_block_incremental(s
, true);
2374 * set ram_counters compression_counters memory to zero for a
2377 memset(&ram_counters
, 0, sizeof(ram_counters
));
2378 memset(&compression_counters
, 0, sizeof(compression_counters
));
2383 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
2384 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
2385 bool has_resume
, bool resume
, Error
**errp
)
2387 Error
*local_err
= NULL
;
2388 MigrationState
*s
= migrate_get_current();
2389 const char *p
= NULL
;
2391 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
2392 has_resume
&& resume
, errp
)) {
2393 /* Error detected, put into errp */
2397 if (!(has_resume
&& resume
)) {
2398 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
2403 migrate_protocol_allow_multi_channels(false);
2404 if (strstart(uri
, "tcp:", &p
) ||
2405 strstart(uri
, "unix:", NULL
) ||
2406 strstart(uri
, "vsock:", NULL
)) {
2407 migrate_protocol_allow_multi_channels(true);
2408 socket_start_outgoing_migration(s
, p
? p
: uri
, &local_err
);
2410 } else if (strstart(uri
, "rdma:", &p
)) {
2411 rdma_start_outgoing_migration(s
, p
, &local_err
);
2413 } else if (strstart(uri
, "exec:", &p
)) {
2414 exec_start_outgoing_migration(s
, p
, &local_err
);
2415 } else if (strstart(uri
, "fd:", &p
)) {
2416 fd_start_outgoing_migration(s
, p
, &local_err
);
2418 if (!(has_resume
&& resume
)) {
2419 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2421 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
2422 "a valid migration protocol");
2423 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2424 MIGRATION_STATUS_FAILED
);
2425 block_cleanup_parameters(s
);
2430 if (!(has_resume
&& resume
)) {
2431 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2433 migrate_fd_error(s
, local_err
);
2434 error_propagate(errp
, local_err
);
2439 void qmp_migrate_cancel(Error
**errp
)
2441 migration_cancel(NULL
);
2444 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
2446 MigrationState
*s
= migrate_get_current();
2447 if (s
->state
!= state
) {
2448 error_setg(errp
, "Migration not in expected state: %s",
2449 MigrationStatus_str(s
->state
));
2452 qemu_sem_post(&s
->pause_sem
);
2455 bool migrate_release_ram(void)
2459 s
= migrate_get_current();
2461 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
2464 bool migrate_postcopy_ram(void)
2468 s
= migrate_get_current();
2470 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
2473 bool migrate_postcopy(void)
2475 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2478 bool migrate_auto_converge(void)
2482 s
= migrate_get_current();
2484 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
2487 bool migrate_zero_blocks(void)
2491 s
= migrate_get_current();
2493 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
2496 bool migrate_postcopy_blocktime(void)
2500 s
= migrate_get_current();
2502 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
2505 bool migrate_use_compression(void)
2509 s
= migrate_get_current();
2511 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
2514 int migrate_compress_level(void)
2518 s
= migrate_get_current();
2520 return s
->parameters
.compress_level
;
2523 int migrate_compress_threads(void)
2527 s
= migrate_get_current();
2529 return s
->parameters
.compress_threads
;
2532 int migrate_compress_wait_thread(void)
2536 s
= migrate_get_current();
2538 return s
->parameters
.compress_wait_thread
;
2541 int migrate_decompress_threads(void)
2545 s
= migrate_get_current();
2547 return s
->parameters
.decompress_threads
;
2550 bool migrate_dirty_bitmaps(void)
2554 s
= migrate_get_current();
2556 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
2559 bool migrate_ignore_shared(void)
2563 s
= migrate_get_current();
2565 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
];
2568 bool migrate_validate_uuid(void)
2572 s
= migrate_get_current();
2574 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_VALIDATE_UUID
];
2577 bool migrate_use_events(void)
2581 s
= migrate_get_current();
2583 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
2586 bool migrate_use_multifd(void)
2590 s
= migrate_get_current();
2592 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_MULTIFD
];
2595 bool migrate_pause_before_switchover(void)
2599 s
= migrate_get_current();
2601 return s
->enabled_capabilities
[
2602 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
2605 int migrate_multifd_channels(void)
2609 s
= migrate_get_current();
2611 return s
->parameters
.multifd_channels
;
2614 MultiFDCompression
migrate_multifd_compression(void)
2618 s
= migrate_get_current();
2620 return s
->parameters
.multifd_compression
;
2623 int migrate_multifd_zlib_level(void)
2627 s
= migrate_get_current();
2629 return s
->parameters
.multifd_zlib_level
;
2632 int migrate_multifd_zstd_level(void)
2636 s
= migrate_get_current();
2638 return s
->parameters
.multifd_zstd_level
;
2642 bool migrate_use_zero_copy_send(void)
2646 s
= migrate_get_current();
2648 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_COPY_SEND
];
2652 int migrate_use_tls(void)
2656 s
= migrate_get_current();
2658 return s
->parameters
.tls_creds
&& *s
->parameters
.tls_creds
;
2661 int migrate_use_xbzrle(void)
2665 s
= migrate_get_current();
2667 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
2670 uint64_t migrate_xbzrle_cache_size(void)
2674 s
= migrate_get_current();
2676 return s
->parameters
.xbzrle_cache_size
;
2679 static int64_t migrate_max_postcopy_bandwidth(void)
2683 s
= migrate_get_current();
2685 return s
->parameters
.max_postcopy_bandwidth
;
2688 bool migrate_use_block(void)
2692 s
= migrate_get_current();
2694 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
2697 bool migrate_use_return_path(void)
2701 s
= migrate_get_current();
2703 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
2706 bool migrate_use_block_incremental(void)
2710 s
= migrate_get_current();
2712 return s
->parameters
.block_incremental
;
2715 bool migrate_background_snapshot(void)
2719 s
= migrate_get_current();
2721 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
];
2724 bool migrate_postcopy_preempt(void)
2728 s
= migrate_get_current();
2730 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_PREEMPT
];
2733 /* migration thread support */
2735 * Something bad happened to the RP stream, mark an error
2736 * The caller shall print or trace something to indicate why
2738 static void mark_source_rp_bad(MigrationState
*s
)
2740 s
->rp_state
.error
= true;
2743 static struct rp_cmd_args
{
2744 ssize_t len
; /* -1 = variable */
2747 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
2748 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
2749 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
2750 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
2751 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
2752 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
2753 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
2754 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
2758 * Process a request for pages received on the return path,
2759 * We're allowed to send more than requested (e.g. to round to our page size)
2760 * and we don't need to send pages that have already been sent.
2762 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
2763 ram_addr_t start
, size_t len
)
2765 long our_host_ps
= qemu_real_host_page_size();
2767 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
2770 * Since we currently insist on matching page sizes, just sanity check
2771 * we're being asked for whole host pages.
2773 if (!QEMU_IS_ALIGNED(start
, our_host_ps
) ||
2774 !QEMU_IS_ALIGNED(len
, our_host_ps
)) {
2775 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2776 " len: %zd", __func__
, start
, len
);
2777 mark_source_rp_bad(ms
);
2781 if (ram_save_queue_pages(rbname
, start
, len
)) {
2782 mark_source_rp_bad(ms
);
2786 /* Return true to retry, false to quit */
2787 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
2789 trace_postcopy_pause_return_path();
2791 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
2793 trace_postcopy_pause_return_path_continued();
2798 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
2800 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
2803 error_report("%s: invalid block name '%s'", __func__
, block_name
);
2807 /* Fetch the received bitmap and refresh the dirty bitmap */
2808 return ram_dirty_bitmap_reload(s
, block
);
2811 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
2813 trace_source_return_path_thread_resume_ack(value
);
2815 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
2816 error_report("%s: illegal resume_ack value %"PRIu32
,
2821 /* Now both sides are active. */
2822 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2823 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2825 /* Notify send thread that time to continue send pages */
2826 qemu_sem_post(&s
->rp_state
.rp_sem
);
2831 /* Release ms->rp_state.from_dst_file in a safe way */
2832 static void migration_release_from_dst_file(MigrationState
*ms
)
2836 WITH_QEMU_LOCK_GUARD(&ms
->qemu_file_lock
) {
2838 * Reset the from_dst_file pointer first before releasing it, as we
2839 * can't block within lock section
2841 file
= ms
->rp_state
.from_dst_file
;
2842 ms
->rp_state
.from_dst_file
= NULL
;
2849 * Handles messages sent on the return path towards the source VM
2852 static void *source_return_path_thread(void *opaque
)
2854 MigrationState
*ms
= opaque
;
2855 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
2856 uint16_t header_len
, header_type
;
2858 uint32_t tmp32
, sibling_error
;
2859 ram_addr_t start
= 0; /* =0 to silence warning */
2860 size_t len
= 0, expected_len
;
2863 trace_source_return_path_thread_entry();
2864 rcu_register_thread();
2867 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
2868 migration_is_setup_or_active(ms
->state
)) {
2869 trace_source_return_path_thread_loop_top();
2870 header_type
= qemu_get_be16(rp
);
2871 header_len
= qemu_get_be16(rp
);
2873 if (qemu_file_get_error(rp
)) {
2874 mark_source_rp_bad(ms
);
2878 if (header_type
>= MIG_RP_MSG_MAX
||
2879 header_type
== MIG_RP_MSG_INVALID
) {
2880 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2881 header_type
, header_len
);
2882 mark_source_rp_bad(ms
);
2886 if ((rp_cmd_args
[header_type
].len
!= -1 &&
2887 header_len
!= rp_cmd_args
[header_type
].len
) ||
2888 header_len
> sizeof(buf
)) {
2889 error_report("RP: Received '%s' message (0x%04x) with"
2890 "incorrect length %d expecting %zu",
2891 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
2892 (size_t)rp_cmd_args
[header_type
].len
);
2893 mark_source_rp_bad(ms
);
2897 /* We know we've got a valid header by this point */
2898 res
= qemu_get_buffer(rp
, buf
, header_len
);
2899 if (res
!= header_len
) {
2900 error_report("RP: Failed reading data for message 0x%04x"
2901 " read %d expected %d",
2902 header_type
, res
, header_len
);
2903 mark_source_rp_bad(ms
);
2907 /* OK, we have the message and the data */
2908 switch (header_type
) {
2909 case MIG_RP_MSG_SHUT
:
2910 sibling_error
= ldl_be_p(buf
);
2911 trace_source_return_path_thread_shut(sibling_error
);
2912 if (sibling_error
) {
2913 error_report("RP: Sibling indicated error %d", sibling_error
);
2914 mark_source_rp_bad(ms
);
2917 * We'll let the main thread deal with closing the RP
2918 * we could do a shutdown(2) on it, but we're the only user
2919 * anyway, so there's nothing gained.
2923 case MIG_RP_MSG_PONG
:
2924 tmp32
= ldl_be_p(buf
);
2925 trace_source_return_path_thread_pong(tmp32
);
2928 case MIG_RP_MSG_REQ_PAGES
:
2929 start
= ldq_be_p(buf
);
2930 len
= ldl_be_p(buf
+ 8);
2931 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
2934 case MIG_RP_MSG_REQ_PAGES_ID
:
2935 expected_len
= 12 + 1; /* header + termination */
2937 if (header_len
>= expected_len
) {
2938 start
= ldq_be_p(buf
);
2939 len
= ldl_be_p(buf
+ 8);
2940 /* Now we expect an idstr */
2941 tmp32
= buf
[12]; /* Length of the following idstr */
2942 buf
[13 + tmp32
] = '\0';
2943 expected_len
+= tmp32
;
2945 if (header_len
!= expected_len
) {
2946 error_report("RP: Req_Page_id with length %d expecting %zd",
2947 header_len
, expected_len
);
2948 mark_source_rp_bad(ms
);
2951 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
2954 case MIG_RP_MSG_RECV_BITMAP
:
2955 if (header_len
< 1) {
2956 error_report("%s: missing block name", __func__
);
2957 mark_source_rp_bad(ms
);
2960 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2961 buf
[buf
[0] + 1] = '\0';
2962 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
2963 mark_source_rp_bad(ms
);
2968 case MIG_RP_MSG_RESUME_ACK
:
2969 tmp32
= ldl_be_p(buf
);
2970 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
2971 mark_source_rp_bad(ms
);
2982 res
= qemu_file_get_error(rp
);
2984 if (res
&& migration_in_postcopy()) {
2986 * Maybe there is something we can do: it looks like a
2987 * network down issue, and we pause for a recovery.
2989 migration_release_from_dst_file(ms
);
2991 if (postcopy_pause_return_path_thread(ms
)) {
2993 * Reload rp, reset the rest. Referencing it is safe since
2994 * it's reset only by us above, or when migration completes
2996 rp
= ms
->rp_state
.from_dst_file
;
2997 ms
->rp_state
.error
= false;
3002 trace_source_return_path_thread_bad_end();
3003 mark_source_rp_bad(ms
);
3006 trace_source_return_path_thread_end();
3007 migration_release_from_dst_file(ms
);
3008 rcu_unregister_thread();
3012 static int open_return_path_on_source(MigrationState
*ms
,
3015 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
3016 if (!ms
->rp_state
.from_dst_file
) {
3020 trace_open_return_path_on_source();
3022 if (!create_thread
) {
3027 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
3028 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
3029 ms
->rp_state
.rp_thread_created
= true;
3031 trace_open_return_path_on_source_continue();
3036 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
3037 static int await_return_path_close_on_source(MigrationState
*ms
)
3040 * If this is a normal exit then the destination will send a SHUT and the
3041 * rp_thread will exit, however if there's an error we need to cause
3044 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
3046 * shutdown(2), if we have it, will cause it to unblock if it's stuck
3047 * waiting for the destination.
3049 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
3050 mark_source_rp_bad(ms
);
3052 trace_await_return_path_close_on_source_joining();
3053 qemu_thread_join(&ms
->rp_state
.rp_thread
);
3054 ms
->rp_state
.rp_thread_created
= false;
3055 trace_await_return_path_close_on_source_close();
3056 return ms
->rp_state
.error
;
3060 * Switch from normal iteration to postcopy
3061 * Returns non-0 on error
3063 static int postcopy_start(MigrationState
*ms
)
3066 QIOChannelBuffer
*bioc
;
3068 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3069 int64_t bandwidth
= migrate_max_postcopy_bandwidth();
3070 bool restart_block
= false;
3071 int cur_state
= MIGRATION_STATUS_ACTIVE
;
3073 if (postcopy_preempt_wait_channel(ms
)) {
3074 migrate_set_state(&ms
->state
, ms
->state
, MIGRATION_STATUS_FAILED
);
3078 if (!migrate_pause_before_switchover()) {
3079 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
3080 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3083 trace_postcopy_start();
3084 qemu_mutex_lock_iothread();
3085 trace_postcopy_start_set_run();
3087 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3088 global_state_store();
3089 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
3094 ret
= migration_maybe_pause(ms
, &cur_state
,
3095 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3100 ret
= bdrv_inactivate_all();
3104 restart_block
= true;
3107 * Cause any non-postcopiable, but iterative devices to
3108 * send out their final data.
3110 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
3113 * in Finish migrate and with the io-lock held everything should
3114 * be quiet, but we've potentially still got dirty pages and we
3115 * need to tell the destination to throw any pages it's already received
3118 if (migrate_postcopy_ram()) {
3119 ram_postcopy_send_discard_bitmap(ms
);
3123 * send rest of state - note things that are doing postcopy
3124 * will notice we're in POSTCOPY_ACTIVE and not actually
3125 * wrap their state up here
3127 /* 0 max-postcopy-bandwidth means unlimited */
3129 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
3131 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
3133 if (migrate_postcopy_ram()) {
3134 /* Ping just for debugging, helps line traces up */
3135 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
3139 * While loading the device state we may trigger page transfer
3140 * requests and the fd must be free to process those, and thus
3141 * the destination must read the whole device state off the fd before
3142 * it starts processing it. Unfortunately the ad-hoc migration format
3143 * doesn't allow the destination to know the size to read without fully
3144 * parsing it through each devices load-state code (especially the open
3145 * coded devices that use get/put).
3146 * So we wrap the device state up in a package with a length at the start;
3147 * to do this we use a qemu_buf to hold the whole of the device state.
3149 bioc
= qio_channel_buffer_new(4096);
3150 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
3151 fb
= qemu_file_new_output(QIO_CHANNEL(bioc
));
3152 object_unref(OBJECT(bioc
));
3155 * Make sure the receiver can get incoming pages before we send the rest
3158 qemu_savevm_send_postcopy_listen(fb
);
3160 qemu_savevm_state_complete_precopy(fb
, false, false);
3161 if (migrate_postcopy_ram()) {
3162 qemu_savevm_send_ping(fb
, 3);
3165 qemu_savevm_send_postcopy_run(fb
);
3167 /* <><> end of stuff going into the package */
3169 /* Last point of recovery; as soon as we send the package the destination
3170 * can open devices and potentially start running.
3171 * Lets just check again we've not got any errors.
3173 ret
= qemu_file_get_error(ms
->to_dst_file
);
3175 error_report("postcopy_start: Migration stream errored (pre package)");
3179 restart_block
= false;
3181 /* Now send that blob */
3182 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
3187 /* Send a notify to give a chance for anything that needs to happen
3188 * at the transition to postcopy and after the device state; in particular
3189 * spice needs to trigger a transition now
3191 ms
->postcopy_after_devices
= true;
3192 notifier_list_notify(&migration_state_notifiers
, ms
);
3194 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
3196 qemu_mutex_unlock_iothread();
3198 if (migrate_postcopy_ram()) {
3200 * Although this ping is just for debug, it could potentially be
3201 * used for getting a better measurement of downtime at the source.
3203 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
3206 if (migrate_release_ram()) {
3207 ram_postcopy_migrated_memory_release(ms
);
3210 ret
= qemu_file_get_error(ms
->to_dst_file
);
3212 error_report("postcopy_start: Migration stream errored");
3213 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
3214 MIGRATION_STATUS_FAILED
);
3217 trace_postcopy_preempt_enabled(migrate_postcopy_preempt());
3224 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
3225 MIGRATION_STATUS_FAILED
);
3226 if (restart_block
) {
3227 /* A failure happened early enough that we know the destination hasn't
3228 * accessed block devices, so we're safe to recover.
3230 Error
*local_err
= NULL
;
3232 bdrv_activate_all(&local_err
);
3234 error_report_err(local_err
);
3237 qemu_mutex_unlock_iothread();
3242 * migration_maybe_pause: Pause if required to by
3243 * migrate_pause_before_switchover called with the iothread locked
3244 * Returns: 0 on success
3246 static int migration_maybe_pause(MigrationState
*s
,
3247 int *current_active_state
,
3250 if (!migrate_pause_before_switchover()) {
3254 /* Since leaving this state is not atomic with posting the semaphore
3255 * it's possible that someone could have issued multiple migrate_continue
3256 * and the semaphore is incorrectly positive at this point;
3257 * the docs say it's undefined to reinit a semaphore that's already
3258 * init'd, so use timedwait to eat up any existing posts.
3260 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
3261 /* This block intentionally left blank */
3265 * If the migration is cancelled when it is in the completion phase,
3266 * the migration state is set to MIGRATION_STATUS_CANCELLING.
3267 * So we don't need to wait a semaphore, otherwise we would always
3268 * wait for the 'pause_sem' semaphore.
3270 if (s
->state
!= MIGRATION_STATUS_CANCELLING
) {
3271 qemu_mutex_unlock_iothread();
3272 migrate_set_state(&s
->state
, *current_active_state
,
3273 MIGRATION_STATUS_PRE_SWITCHOVER
);
3274 qemu_sem_wait(&s
->pause_sem
);
3275 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
3277 *current_active_state
= new_state
;
3278 qemu_mutex_lock_iothread();
3281 return s
->state
== new_state
? 0 : -EINVAL
;
3285 * migration_completion: Used by migration_thread when there's not much left.
3286 * The caller 'breaks' the loop when this returns.
3288 * @s: Current migration state
3290 static void migration_completion(MigrationState
*s
)
3293 int current_active_state
= s
->state
;
3295 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
3296 qemu_mutex_lock_iothread();
3297 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3298 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3299 s
->vm_was_running
= runstate_is_running();
3300 ret
= global_state_store();
3303 bool inactivate
= !migrate_colo_enabled();
3304 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
3305 trace_migration_completion_vm_stop(ret
);
3307 ret
= migration_maybe_pause(s
, ¤t_active_state
,
3308 MIGRATION_STATUS_DEVICE
);
3311 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
3312 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
3315 if (inactivate
&& ret
>= 0) {
3316 s
->block_inactive
= true;
3319 qemu_mutex_unlock_iothread();
3324 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3325 trace_migration_completion_postcopy_end();
3327 qemu_mutex_lock_iothread();
3328 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
3329 qemu_mutex_unlock_iothread();
3331 /* Shutdown the postcopy fast path thread */
3332 if (migrate_postcopy_preempt()) {
3333 postcopy_preempt_shutdown_file(s
);
3336 trace_migration_completion_postcopy_end_after_complete();
3342 * If rp was opened we must clean up the thread before
3343 * cleaning everything else up (since if there are no failures
3344 * it will wait for the destination to send it's status in
3347 if (s
->rp_state
.rp_thread_created
) {
3349 trace_migration_return_path_end_before();
3350 rp_error
= await_return_path_close_on_source(s
);
3351 trace_migration_return_path_end_after(rp_error
);
3353 goto fail_invalidate
;
3357 if (qemu_file_get_error(s
->to_dst_file
)) {
3358 trace_migration_completion_file_err();
3359 goto fail_invalidate
;
3362 if (migrate_colo_enabled() && s
->state
== MIGRATION_STATUS_ACTIVE
) {
3363 /* COLO does not support postcopy */
3364 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
3365 MIGRATION_STATUS_COLO
);
3367 migrate_set_state(&s
->state
, current_active_state
,
3368 MIGRATION_STATUS_COMPLETED
);
3374 /* If not doing postcopy, vm_start() will be called: let's regain
3375 * control on images.
3377 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
3378 s
->state
== MIGRATION_STATUS_DEVICE
) {
3379 Error
*local_err
= NULL
;
3381 qemu_mutex_lock_iothread();
3382 bdrv_activate_all(&local_err
);
3384 error_report_err(local_err
);
3386 s
->block_inactive
= false;
3388 qemu_mutex_unlock_iothread();
3392 migrate_set_state(&s
->state
, current_active_state
,
3393 MIGRATION_STATUS_FAILED
);
3397 * bg_migration_completion: Used by bg_migration_thread when after all the
3398 * RAM has been saved. The caller 'breaks' the loop when this returns.
3400 * @s: Current migration state
3402 static void bg_migration_completion(MigrationState
*s
)
3404 int current_active_state
= s
->state
;
3407 * Stop tracking RAM writes - un-protect memory, un-register UFFD
3408 * memory ranges, flush kernel wait queues and wake up threads
3409 * waiting for write fault to be resolved.
3411 ram_write_tracking_stop();
3413 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
3415 * By this moment we have RAM content saved into the migration stream.
3416 * The next step is to flush the non-RAM content (device state)
3417 * right after the ram content. The device state has been stored into
3418 * the temporary buffer before RAM saving started.
3420 qemu_put_buffer(s
->to_dst_file
, s
->bioc
->data
, s
->bioc
->usage
);
3421 qemu_fflush(s
->to_dst_file
);
3422 } else if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
3426 if (qemu_file_get_error(s
->to_dst_file
)) {
3427 trace_migration_completion_file_err();
3431 migrate_set_state(&s
->state
, current_active_state
,
3432 MIGRATION_STATUS_COMPLETED
);
3436 migrate_set_state(&s
->state
, current_active_state
,
3437 MIGRATION_STATUS_FAILED
);
3440 bool migrate_colo_enabled(void)
3442 MigrationState
*s
= migrate_get_current();
3443 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
3446 typedef enum MigThrError
{
3447 /* No error detected */
3448 MIG_THR_ERR_NONE
= 0,
3449 /* Detected error, but resumed successfully */
3450 MIG_THR_ERR_RECOVERED
= 1,
3451 /* Detected fatal error, need to exit */
3452 MIG_THR_ERR_FATAL
= 2,
3455 static int postcopy_resume_handshake(MigrationState
*s
)
3457 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
3459 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3460 qemu_sem_wait(&s
->rp_state
.rp_sem
);
3463 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3470 /* Return zero if success, or <0 for error */
3471 static int postcopy_do_resume(MigrationState
*s
)
3476 * Call all the resume_prepare() hooks, so that modules can be
3477 * ready for the migration resume.
3479 ret
= qemu_savevm_state_resume_prepare(s
);
3481 error_report("%s: resume_prepare() failure detected: %d",
3487 * Last handshake with destination on the resume (destination will
3488 * switch to postcopy-active afterwards)
3490 ret
= postcopy_resume_handshake(s
);
3492 error_report("%s: handshake failed: %d", __func__
, ret
);
3500 * We don't return until we are in a safe state to continue current
3501 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
3502 * MIG_THR_ERR_FATAL if unrecovery failure happened.
3504 static MigThrError
postcopy_pause(MigrationState
*s
)
3506 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3512 * Current channel is possibly broken. Release it. Note that this is
3513 * guaranteed even without lock because to_dst_file should only be
3514 * modified by the migration thread. That also guarantees that the
3515 * unregister of yank is safe too without the lock. It should be safe
3516 * even to be within the qemu_file_lock, but we didn't do that to avoid
3517 * taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make
3518 * the qemu_file_lock critical section as small as possible.
3520 assert(s
->to_dst_file
);
3521 migration_ioc_unregister_yank_from_file(s
->to_dst_file
);
3522 qemu_mutex_lock(&s
->qemu_file_lock
);
3523 file
= s
->to_dst_file
;
3524 s
->to_dst_file
= NULL
;
3525 qemu_mutex_unlock(&s
->qemu_file_lock
);
3527 qemu_file_shutdown(file
);
3531 * Do the same to postcopy fast path socket too if there is. No
3532 * locking needed because no racer as long as we do this before setting
3535 if (s
->postcopy_qemufile_src
) {
3536 migration_ioc_unregister_yank_from_file(s
->postcopy_qemufile_src
);
3537 qemu_file_shutdown(s
->postcopy_qemufile_src
);
3538 qemu_fclose(s
->postcopy_qemufile_src
);
3539 s
->postcopy_qemufile_src
= NULL
;
3542 migrate_set_state(&s
->state
, s
->state
,
3543 MIGRATION_STATUS_POSTCOPY_PAUSED
);
3545 error_report("Detected IO failure for postcopy. "
3546 "Migration paused.");
3549 * We wait until things fixed up. Then someone will setup the
3550 * status back for us.
3552 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
3553 qemu_sem_wait(&s
->postcopy_pause_sem
);
3556 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3557 /* Woken up by a recover procedure. Give it a shot */
3559 if (postcopy_preempt_wait_channel(s
)) {
3561 * Preempt enabled, and new channel create failed; loop
3562 * back to wait for another recovery.
3568 * Firstly, let's wake up the return path now, with a new
3569 * return path channel.
3571 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
3573 /* Do the resume logic */
3574 if (postcopy_do_resume(s
) == 0) {
3575 /* Let's continue! */
3576 trace_postcopy_pause_continued();
3577 return MIG_THR_ERR_RECOVERED
;
3580 * Something wrong happened during the recovery, let's
3581 * pause again. Pause is always better than throwing
3587 /* This is not right... Time to quit. */
3588 return MIG_THR_ERR_FATAL
;
3593 static MigThrError
migration_detect_error(MigrationState
*s
)
3596 int state
= s
->state
;
3597 Error
*local_error
= NULL
;
3599 if (state
== MIGRATION_STATUS_CANCELLING
||
3600 state
== MIGRATION_STATUS_CANCELLED
) {
3601 /* End the migration, but don't set the state to failed */
3602 return MIG_THR_ERR_FATAL
;
3606 * Try to detect any file errors. Note that postcopy_qemufile_src will
3607 * be NULL when postcopy preempt is not enabled.
3609 ret
= qemu_file_get_error_obj_any(s
->to_dst_file
,
3610 s
->postcopy_qemufile_src
,
3613 /* Everything is fine */
3614 assert(!local_error
);
3615 return MIG_THR_ERR_NONE
;
3619 migrate_set_error(s
, local_error
);
3620 error_free(local_error
);
3623 if (state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
) {
3625 * For postcopy, we allow the network to be down for a
3626 * while. After that, it can be continued by a
3629 return postcopy_pause(s
);
3632 * For precopy (or postcopy with error outside IO), we fail
3635 migrate_set_state(&s
->state
, state
, MIGRATION_STATUS_FAILED
);
3636 trace_migration_thread_file_err();
3638 /* Time to stop the migration, now. */
3639 return MIG_THR_ERR_FATAL
;
3643 /* How many bytes have we transferred since the beginning of the migration */
3644 static uint64_t migration_total_bytes(MigrationState
*s
)
3646 return qemu_file_total_transferred(s
->to_dst_file
) +
3647 ram_counters
.multifd_bytes
;
3650 static void migration_calculate_complete(MigrationState
*s
)
3652 uint64_t bytes
= migration_total_bytes(s
);
3653 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3654 int64_t transfer_time
;
3656 s
->total_time
= end_time
- s
->start_time
;
3659 * It's still not set, so we are precopy migration. For
3660 * postcopy, downtime is calculated during postcopy_start().
3662 s
->downtime
= end_time
- s
->downtime_start
;
3665 transfer_time
= s
->total_time
- s
->setup_time
;
3666 if (transfer_time
) {
3667 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
3671 static void update_iteration_initial_status(MigrationState
*s
)
3674 * Update these three fields at the same time to avoid mismatch info lead
3675 * wrong speed calculation.
3677 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3678 s
->iteration_initial_bytes
= migration_total_bytes(s
);
3679 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
3682 static void migration_update_counters(MigrationState
*s
,
3683 int64_t current_time
)
3685 uint64_t transferred
, transferred_pages
, time_spent
;
3686 uint64_t current_bytes
; /* bytes transferred since the beginning */
3689 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
3693 current_bytes
= migration_total_bytes(s
);
3694 transferred
= current_bytes
- s
->iteration_initial_bytes
;
3695 time_spent
= current_time
- s
->iteration_start_time
;
3696 bandwidth
= (double)transferred
/ time_spent
;
3697 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
3699 s
->mbps
= (((double) transferred
* 8.0) /
3700 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
3702 transferred_pages
= ram_get_total_transferred_pages() -
3703 s
->iteration_initial_pages
;
3704 s
->pages_per_second
= (double) transferred_pages
/
3705 (((double) time_spent
/ 1000.0));
3708 * if we haven't sent anything, we don't want to
3709 * recalculate. 10000 is a small enough number for our purposes
3711 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
3712 s
->expected_downtime
= ram_counters
.remaining
/ bandwidth
;
3715 qemu_file_reset_rate_limit(s
->to_dst_file
);
3717 update_iteration_initial_status(s
);
3719 trace_migrate_transferred(transferred
, time_spent
,
3720 bandwidth
, s
->threshold_size
);
3723 /* Migration thread iteration status */
3725 MIG_ITERATE_RESUME
, /* Resume current iteration */
3726 MIG_ITERATE_SKIP
, /* Skip current iteration */
3727 MIG_ITERATE_BREAK
, /* Break the loop */
3731 * Return true if continue to the next iteration directly, false
3734 static MigIterateState
migration_iteration_run(MigrationState
*s
)
3736 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
3737 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
3739 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
3740 &pend_compat
, &pend_post
);
3741 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
3743 trace_migrate_pending(pending_size
, s
->threshold_size
,
3744 pend_pre
, pend_compat
, pend_post
);
3746 if (pending_size
&& pending_size
>= s
->threshold_size
) {
3747 /* Still a significant amount to transfer */
3748 if (!in_postcopy
&& pend_pre
<= s
->threshold_size
&&
3749 qatomic_read(&s
->start_postcopy
)) {
3750 if (postcopy_start(s
)) {
3751 error_report("%s: postcopy failed to start", __func__
);
3753 return MIG_ITERATE_SKIP
;
3755 /* Just another iteration step */
3756 qemu_savevm_state_iterate(s
->to_dst_file
, in_postcopy
);
3758 trace_migration_thread_low_pending(pending_size
);
3759 migration_completion(s
);
3760 return MIG_ITERATE_BREAK
;
3763 return MIG_ITERATE_RESUME
;
3766 static void migration_iteration_finish(MigrationState
*s
)
3768 /* If we enabled cpu throttling for auto-converge, turn it off. */
3769 cpu_throttle_stop();
3771 qemu_mutex_lock_iothread();
3773 case MIGRATION_STATUS_COMPLETED
:
3774 migration_calculate_complete(s
);
3775 runstate_set(RUN_STATE_POSTMIGRATE
);
3777 case MIGRATION_STATUS_COLO
:
3778 if (!migrate_colo_enabled()) {
3779 error_report("%s: critical error: calling COLO code without "
3780 "COLO enabled", __func__
);
3782 migrate_start_colo_process(s
);
3783 s
->vm_was_running
= true;
3785 case MIGRATION_STATUS_FAILED
:
3786 case MIGRATION_STATUS_CANCELLED
:
3787 case MIGRATION_STATUS_CANCELLING
:
3788 if (s
->vm_was_running
) {
3789 if (!runstate_check(RUN_STATE_SHUTDOWN
)) {
3793 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
3794 runstate_set(RUN_STATE_POSTMIGRATE
);
3800 /* Should not reach here, but if so, forgive the VM. */
3801 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3804 migrate_fd_cleanup_schedule(s
);
3805 qemu_mutex_unlock_iothread();
3808 static void bg_migration_iteration_finish(MigrationState
*s
)
3810 qemu_mutex_lock_iothread();
3812 case MIGRATION_STATUS_COMPLETED
:
3813 migration_calculate_complete(s
);
3816 case MIGRATION_STATUS_ACTIVE
:
3817 case MIGRATION_STATUS_FAILED
:
3818 case MIGRATION_STATUS_CANCELLED
:
3819 case MIGRATION_STATUS_CANCELLING
:
3823 /* Should not reach here, but if so, forgive the VM. */
3824 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3828 migrate_fd_cleanup_schedule(s
);
3829 qemu_mutex_unlock_iothread();
3833 * Return true if continue to the next iteration directly, false
3836 static MigIterateState
bg_migration_iteration_run(MigrationState
*s
)
3840 res
= qemu_savevm_state_iterate(s
->to_dst_file
, false);
3842 bg_migration_completion(s
);
3843 return MIG_ITERATE_BREAK
;
3846 return MIG_ITERATE_RESUME
;
3849 void migration_make_urgent_request(void)
3851 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
3854 void migration_consume_urgent_request(void)
3856 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
3859 /* Returns true if the rate limiting was broken by an urgent request */
3860 bool migration_rate_limit(void)
3862 int64_t now
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3863 MigrationState
*s
= migrate_get_current();
3865 bool urgent
= false;
3866 migration_update_counters(s
, now
);
3867 if (qemu_file_rate_limit(s
->to_dst_file
)) {
3869 if (qemu_file_get_error(s
->to_dst_file
)) {
3873 * Wait for a delay to do rate limiting OR
3874 * something urgent to post the semaphore.
3876 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- now
;
3877 trace_migration_rate_limit_pre(ms
);
3878 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
3880 * We were woken by one or more urgent things but
3881 * the timedwait will have consumed one of them.
3882 * The service routine for the urgent wake will dec
3883 * the semaphore itself for each item it consumes,
3884 * so add this one we just eat back.
3886 qemu_sem_post(&s
->rate_limit_sem
);
3889 trace_migration_rate_limit_post(urgent
);
3895 * if failover devices are present, wait they are completely
3899 static void qemu_savevm_wait_unplug(MigrationState
*s
, int old_state
,
3902 if (qemu_savevm_state_guest_unplug_pending()) {
3903 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_WAIT_UNPLUG
);
3905 while (s
->state
== MIGRATION_STATUS_WAIT_UNPLUG
&&
3906 qemu_savevm_state_guest_unplug_pending()) {
3907 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
3909 if (s
->state
!= MIGRATION_STATUS_WAIT_UNPLUG
) {
3910 int timeout
= 120; /* 30 seconds */
3912 * migration has been canceled
3913 * but as we have started an unplug we must wait the end
3914 * to be able to plug back the card
3916 while (timeout
-- && qemu_savevm_state_guest_unplug_pending()) {
3917 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
3919 if (qemu_savevm_state_guest_unplug_pending() &&
3921 warn_report("migration: partially unplugged device on "
3926 migrate_set_state(&s
->state
, MIGRATION_STATUS_WAIT_UNPLUG
, new_state
);
3928 migrate_set_state(&s
->state
, old_state
, new_state
);
3933 * Master migration thread on the source VM.
3934 * It drives the migration and pumps the data down the outgoing channel.
3936 static void *migration_thread(void *opaque
)
3938 MigrationState
*s
= opaque
;
3939 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3940 MigThrError thr_error
;
3941 bool urgent
= false;
3943 rcu_register_thread();
3945 object_ref(OBJECT(s
));
3946 update_iteration_initial_status(s
);
3948 qemu_savevm_state_header(s
->to_dst_file
);
3951 * If we opened the return path, we need to make sure dst has it
3954 if (s
->rp_state
.rp_thread_created
) {
3955 /* Now tell the dest that it should open its end so it can reply */
3956 qemu_savevm_send_open_return_path(s
->to_dst_file
);
3958 /* And do a ping that will make stuff easier to debug */
3959 qemu_savevm_send_ping(s
->to_dst_file
, 1);
3962 if (migrate_postcopy()) {
3964 * Tell the destination that we *might* want to do postcopy later;
3965 * if the other end can't do postcopy it should fail now, nice and
3968 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
3971 if (migrate_colo_enabled()) {
3972 /* Notify migration destination that we enable COLO */
3973 qemu_savevm_send_colo_enable(s
->to_dst_file
);
3976 qemu_savevm_state_setup(s
->to_dst_file
);
3978 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
3979 MIGRATION_STATUS_ACTIVE
);
3981 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3983 trace_migration_thread_setup_complete();
3985 while (migration_is_active(s
)) {
3986 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
3987 MigIterateState iter_state
= migration_iteration_run(s
);
3988 if (iter_state
== MIG_ITERATE_SKIP
) {
3990 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3996 * Try to detect any kind of failures, and see whether we
3997 * should stop the migration now.
3999 thr_error
= migration_detect_error(s
);
4000 if (thr_error
== MIG_THR_ERR_FATAL
) {
4001 /* Stop migration */
4003 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
4005 * Just recovered from a e.g. network failure, reset all
4006 * the local variables. This is important to avoid
4007 * breaking transferred_bytes and bandwidth calculation
4009 update_iteration_initial_status(s
);
4012 urgent
= migration_rate_limit();
4015 trace_migration_thread_after_loop();
4016 migration_iteration_finish(s
);
4017 object_unref(OBJECT(s
));
4018 rcu_unregister_thread();
4022 static void bg_migration_vm_start_bh(void *opaque
)
4024 MigrationState
*s
= opaque
;
4026 qemu_bh_delete(s
->vm_start_bh
);
4027 s
->vm_start_bh
= NULL
;
4030 s
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - s
->downtime_start
;
4034 * Background snapshot thread, based on live migration code.
4035 * This is an alternative implementation of live migration mechanism
4036 * introduced specifically to support background snapshots.
4038 * It takes advantage of userfault_fd write protection mechanism introduced
4039 * in v5.7 kernel. Compared to existing dirty page logging migration much
4040 * lesser stream traffic is produced resulting in smaller snapshot images,
4041 * simply cause of no page duplicates can get into the stream.
4043 * Another key point is that generated vmstate stream reflects machine state
4044 * 'frozen' at the beginning of snapshot creation compared to dirty page logging
4045 * mechanism, which effectively results in that saved snapshot is the state of VM
4046 * at the end of the process.
4048 static void *bg_migration_thread(void *opaque
)
4050 MigrationState
*s
= opaque
;
4051 int64_t setup_start
;
4052 MigThrError thr_error
;
4054 bool early_fail
= true;
4056 rcu_register_thread();
4057 object_ref(OBJECT(s
));
4059 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
4061 setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
4063 * We want to save vmstate for the moment when migration has been
4064 * initiated but also we want to save RAM content while VM is running.
4065 * The RAM content should appear first in the vmstate. So, we first
4066 * stash the non-RAM part of the vmstate to the temporary buffer,
4067 * then write RAM part of the vmstate to the migration stream
4068 * with vCPUs running and, finally, write stashed non-RAM part of
4069 * the vmstate from the buffer to the migration stream.
4071 s
->bioc
= qio_channel_buffer_new(512 * 1024);
4072 qio_channel_set_name(QIO_CHANNEL(s
->bioc
), "vmstate-buffer");
4073 fb
= qemu_file_new_output(QIO_CHANNEL(s
->bioc
));
4074 object_unref(OBJECT(s
->bioc
));
4076 update_iteration_initial_status(s
);
4079 * Prepare for tracking memory writes with UFFD-WP - populate
4080 * RAM pages before protecting.
4083 ram_write_tracking_prepare();
4086 qemu_savevm_state_header(s
->to_dst_file
);
4087 qemu_savevm_state_setup(s
->to_dst_file
);
4089 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
4090 MIGRATION_STATUS_ACTIVE
);
4092 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
4094 trace_migration_thread_setup_complete();
4095 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
4097 qemu_mutex_lock_iothread();
4100 * If VM is currently in suspended state, then, to make a valid runstate
4101 * transition in vm_stop_force_state() we need to wakeup it up.
4103 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
4104 s
->vm_was_running
= runstate_is_running();
4106 if (global_state_store()) {
4109 /* Forcibly stop VM before saving state of vCPUs and devices */
4110 if (vm_stop_force_state(RUN_STATE_PAUSED
)) {
4114 * Put vCPUs in sync with shadow context structures, then
4115 * save their state to channel-buffer along with devices.
4117 cpu_synchronize_all_states();
4118 if (qemu_savevm_state_complete_precopy_non_iterable(fb
, false, false)) {
4122 * Since we are going to get non-iterable state data directly
4123 * from s->bioc->data, explicit flush is needed here.
4127 /* Now initialize UFFD context and start tracking RAM writes */
4128 if (ram_write_tracking_start()) {
4134 * Start VM from BH handler to avoid write-fault lock here.
4135 * UFFD-WP protection for the whole RAM is already enabled so
4136 * calling VM state change notifiers from vm_start() would initiate
4137 * writes to virtio VQs memory which is in write-protected region.
4139 s
->vm_start_bh
= qemu_bh_new(bg_migration_vm_start_bh
, s
);
4140 qemu_bh_schedule(s
->vm_start_bh
);
4142 qemu_mutex_unlock_iothread();
4144 while (migration_is_active(s
)) {
4145 MigIterateState iter_state
= bg_migration_iteration_run(s
);
4146 if (iter_state
== MIG_ITERATE_SKIP
) {
4148 } else if (iter_state
== MIG_ITERATE_BREAK
) {
4153 * Try to detect any kind of failures, and see whether we
4154 * should stop the migration now.
4156 thr_error
= migration_detect_error(s
);
4157 if (thr_error
== MIG_THR_ERR_FATAL
) {
4158 /* Stop migration */
4162 migration_update_counters(s
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
));
4165 trace_migration_thread_after_loop();
4169 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
4170 MIGRATION_STATUS_FAILED
);
4171 qemu_mutex_unlock_iothread();
4174 bg_migration_iteration_finish(s
);
4177 object_unref(OBJECT(s
));
4178 rcu_unregister_thread();
4183 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
4185 Error
*local_err
= NULL
;
4187 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
4190 * If there's a previous error, free it and prepare for another one.
4191 * Meanwhile if migration completes successfully, there won't have an error
4192 * dumped when calling migrate_fd_cleanup().
4194 migrate_error_free(s
);
4196 s
->expected_downtime
= s
->parameters
.downtime_limit
;
4198 assert(s
->cleanup_bh
);
4200 assert(!s
->cleanup_bh
);
4201 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup_bh
, s
);
4204 migrate_fd_error(s
, error_in
);
4207 * Don't do cleanup for resume if channel is invalid, but only dump
4208 * the error. We wait for another channel connect from the user.
4209 * The error_report still gives HMP user a hint on what failed.
4210 * It's normally done in migrate_fd_cleanup(), but call it here
4213 error_report_err(error_copy(s
->error
));
4215 migrate_fd_cleanup(s
);
4221 /* This is a resumed migration */
4222 rate_limit
= s
->parameters
.max_postcopy_bandwidth
/
4225 /* This is a fresh new migration */
4226 rate_limit
= s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
;
4228 /* Notify before starting migration thread */
4229 notifier_list_notify(&migration_state_notifiers
, s
);
4232 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
4233 qemu_file_set_blocking(s
->to_dst_file
, true);
4236 * Open the return path. For postcopy, it is used exclusively. For
4237 * precopy, only if user specified "return-path" capability would
4238 * QEMU uses the return path.
4240 if (migrate_postcopy_ram() || migrate_use_return_path()) {
4241 if (open_return_path_on_source(s
, !resume
)) {
4242 error_report("Unable to open return-path for postcopy");
4243 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
4244 migrate_fd_cleanup(s
);
4249 /* This needs to be done before resuming a postcopy */
4250 if (postcopy_preempt_setup(s
, &local_err
)) {
4251 error_report_err(local_err
);
4252 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
4253 MIGRATION_STATUS_FAILED
);
4254 migrate_fd_cleanup(s
);
4259 /* Wakeup the main migration thread to do the recovery */
4260 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
4261 MIGRATION_STATUS_POSTCOPY_RECOVER
);
4262 qemu_sem_post(&s
->postcopy_pause_sem
);
4266 if (multifd_save_setup(&local_err
) != 0) {
4267 error_report_err(local_err
);
4268 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
4269 MIGRATION_STATUS_FAILED
);
4270 migrate_fd_cleanup(s
);
4274 if (migrate_background_snapshot()) {
4275 qemu_thread_create(&s
->thread
, "bg_snapshot",
4276 bg_migration_thread
, s
, QEMU_THREAD_JOINABLE
);
4278 qemu_thread_create(&s
->thread
, "live_migration",
4279 migration_thread
, s
, QEMU_THREAD_JOINABLE
);
4281 s
->migration_thread_running
= true;
4284 void migration_global_dump(Monitor
*mon
)
4286 MigrationState
*ms
= migrate_get_current();
4288 monitor_printf(mon
, "globals:\n");
4289 monitor_printf(mon
, "store-global-state: %s\n",
4290 ms
->store_global_state
? "on" : "off");
4291 monitor_printf(mon
, "only-migratable: %s\n",
4292 only_migratable
? "on" : "off");
4293 monitor_printf(mon
, "send-configuration: %s\n",
4294 ms
->send_configuration
? "on" : "off");
4295 monitor_printf(mon
, "send-section-footer: %s\n",
4296 ms
->send_section_footer
? "on" : "off");
4297 monitor_printf(mon
, "decompress-error-check: %s\n",
4298 ms
->decompress_error_check
? "on" : "off");
4299 monitor_printf(mon
, "clear-bitmap-shift: %u\n",
4300 ms
->clear_bitmap_shift
);
4303 #define DEFINE_PROP_MIG_CAP(name, x) \
4304 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
4306 static Property migration_properties
[] = {
4307 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
4308 store_global_state
, true),
4309 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
4310 send_configuration
, true),
4311 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
4312 send_section_footer
, true),
4313 DEFINE_PROP_BOOL("decompress-error-check", MigrationState
,
4314 decompress_error_check
, true),
4315 DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState
,
4316 clear_bitmap_shift
, CLEAR_BITMAP_SHIFT_DEFAULT
),
4318 /* Migration parameters */
4319 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
4320 parameters
.compress_level
,
4321 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
4322 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
4323 parameters
.compress_threads
,
4324 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
4325 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState
,
4326 parameters
.compress_wait_thread
, true),
4327 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
4328 parameters
.decompress_threads
,
4329 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
4330 DEFINE_PROP_UINT8("x-throttle-trigger-threshold", MigrationState
,
4331 parameters
.throttle_trigger_threshold
,
4332 DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD
),
4333 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
4334 parameters
.cpu_throttle_initial
,
4335 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
4336 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
4337 parameters
.cpu_throttle_increment
,
4338 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
4339 DEFINE_PROP_BOOL("x-cpu-throttle-tailslow", MigrationState
,
4340 parameters
.cpu_throttle_tailslow
, false),
4341 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
4342 parameters
.max_bandwidth
, MAX_THROTTLE
),
4343 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
4344 parameters
.downtime_limit
,
4345 DEFAULT_MIGRATE_SET_DOWNTIME
),
4346 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
4347 parameters
.x_checkpoint_delay
,
4348 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
4349 DEFINE_PROP_UINT8("multifd-channels", MigrationState
,
4350 parameters
.multifd_channels
,
4351 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
4352 DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState
,
4353 parameters
.multifd_compression
,
4354 DEFAULT_MIGRATE_MULTIFD_COMPRESSION
),
4355 DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState
,
4356 parameters
.multifd_zlib_level
,
4357 DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL
),
4358 DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState
,
4359 parameters
.multifd_zstd_level
,
4360 DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL
),
4361 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
4362 parameters
.xbzrle_cache_size
,
4363 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
4364 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState
,
4365 parameters
.max_postcopy_bandwidth
,
4366 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH
),
4367 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState
,
4368 parameters
.max_cpu_throttle
,
4369 DEFAULT_MIGRATE_MAX_CPU_THROTTLE
),
4370 DEFINE_PROP_SIZE("announce-initial", MigrationState
,
4371 parameters
.announce_initial
,
4372 DEFAULT_MIGRATE_ANNOUNCE_INITIAL
),
4373 DEFINE_PROP_SIZE("announce-max", MigrationState
,
4374 parameters
.announce_max
,
4375 DEFAULT_MIGRATE_ANNOUNCE_MAX
),
4376 DEFINE_PROP_SIZE("announce-rounds", MigrationState
,
4377 parameters
.announce_rounds
,
4378 DEFAULT_MIGRATE_ANNOUNCE_ROUNDS
),
4379 DEFINE_PROP_SIZE("announce-step", MigrationState
,
4380 parameters
.announce_step
,
4381 DEFAULT_MIGRATE_ANNOUNCE_STEP
),
4382 DEFINE_PROP_BOOL("x-postcopy-preempt-break-huge", MigrationState
,
4383 postcopy_preempt_break_huge
, true),
4384 DEFINE_PROP_STRING("tls-creds", MigrationState
, parameters
.tls_creds
),
4385 DEFINE_PROP_STRING("tls-hostname", MigrationState
, parameters
.tls_hostname
),
4386 DEFINE_PROP_STRING("tls-authz", MigrationState
, parameters
.tls_authz
),
4388 /* Migration capabilities */
4389 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
4390 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
4391 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
4392 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
4393 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
4394 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
4395 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
4396 DEFINE_PROP_MIG_CAP("x-postcopy-preempt",
4397 MIGRATION_CAPABILITY_POSTCOPY_PREEMPT
),
4398 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
4399 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
4400 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
4401 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
4402 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD
),
4403 DEFINE_PROP_MIG_CAP("x-background-snapshot",
4404 MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
),
4406 DEFINE_PROP_MIG_CAP("x-zero-copy-send",
4407 MIGRATION_CAPABILITY_ZERO_COPY_SEND
),
4410 DEFINE_PROP_END_OF_LIST(),
4413 static void migration_class_init(ObjectClass
*klass
, void *data
)
4415 DeviceClass
*dc
= DEVICE_CLASS(klass
);
4417 dc
->user_creatable
= false;
4418 device_class_set_props(dc
, migration_properties
);
4421 static void migration_instance_finalize(Object
*obj
)
4423 MigrationState
*ms
= MIGRATION_OBJ(obj
);
4425 qemu_mutex_destroy(&ms
->error_mutex
);
4426 qemu_mutex_destroy(&ms
->qemu_file_lock
);
4427 qemu_sem_destroy(&ms
->wait_unplug_sem
);
4428 qemu_sem_destroy(&ms
->rate_limit_sem
);
4429 qemu_sem_destroy(&ms
->pause_sem
);
4430 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
4431 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
4432 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
4433 qemu_sem_destroy(&ms
->postcopy_qemufile_src_sem
);
4434 error_free(ms
->error
);
4437 static void migration_instance_init(Object
*obj
)
4439 MigrationState
*ms
= MIGRATION_OBJ(obj
);
4440 MigrationParameters
*params
= &ms
->parameters
;
4442 ms
->state
= MIGRATION_STATUS_NONE
;
4444 ms
->pages_per_second
= -1;
4445 qemu_sem_init(&ms
->pause_sem
, 0);
4446 qemu_mutex_init(&ms
->error_mutex
);
4448 params
->tls_hostname
= g_strdup("");
4449 params
->tls_creds
= g_strdup("");
4451 /* Set has_* up only for parameter checks */
4452 params
->has_compress_level
= true;
4453 params
->has_compress_threads
= true;
4454 params
->has_decompress_threads
= true;
4455 params
->has_throttle_trigger_threshold
= true;
4456 params
->has_cpu_throttle_initial
= true;
4457 params
->has_cpu_throttle_increment
= true;
4458 params
->has_cpu_throttle_tailslow
= true;
4459 params
->has_max_bandwidth
= true;
4460 params
->has_downtime_limit
= true;
4461 params
->has_x_checkpoint_delay
= true;
4462 params
->has_block_incremental
= true;
4463 params
->has_multifd_channels
= true;
4464 params
->has_multifd_compression
= true;
4465 params
->has_multifd_zlib_level
= true;
4466 params
->has_multifd_zstd_level
= true;
4467 params
->has_xbzrle_cache_size
= true;
4468 params
->has_max_postcopy_bandwidth
= true;
4469 params
->has_max_cpu_throttle
= true;
4470 params
->has_announce_initial
= true;
4471 params
->has_announce_max
= true;
4472 params
->has_announce_rounds
= true;
4473 params
->has_announce_step
= true;
4475 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
4476 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
4477 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
4478 qemu_sem_init(&ms
->rate_limit_sem
, 0);
4479 qemu_sem_init(&ms
->wait_unplug_sem
, 0);
4480 qemu_sem_init(&ms
->postcopy_qemufile_src_sem
, 0);
4481 qemu_mutex_init(&ms
->qemu_file_lock
);
4485 * Return true if check pass, false otherwise. Error will be put
4486 * inside errp if provided.
4488 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
4490 MigrationCapabilityStatusList
*head
= NULL
;
4491 /* Assuming all off */
4492 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
4495 if (!migrate_params_check(&ms
->parameters
, errp
)) {
4499 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
4500 if (ms
->enabled_capabilities
[i
]) {
4501 QAPI_LIST_PREPEND(head
, migrate_cap_add(i
, true));
4505 ret
= migrate_caps_check(cap_list
, head
, errp
);
4507 /* It works with head == NULL */
4508 qapi_free_MigrationCapabilityStatusList(head
);
4513 static const TypeInfo migration_type
= {
4514 .name
= TYPE_MIGRATION
,
4516 * NOTE: TYPE_MIGRATION is not really a device, as the object is
4517 * not created using qdev_new(), it is not attached to the qdev
4518 * device tree, and it is never realized.
4520 * TODO: Make this TYPE_OBJECT once QOM provides something like
4521 * TYPE_DEVICE's "-global" properties.
4523 .parent
= TYPE_DEVICE
,
4524 .class_init
= migration_class_init
,
4525 .class_size
= sizeof(MigrationClass
),
4526 .instance_size
= sizeof(MigrationState
),
4527 .instance_init
= migration_instance_init
,
4528 .instance_finalize
= migration_instance_finalize
,
4531 static void register_migration_types(void)
4533 type_register_static(&migration_type
);
4536 type_init(register_migration_types
);