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"
64 #include "ui/qemu-spice.h"
66 #define MAX_THROTTLE (128 << 20) /* Migration transfer speed throttling */
68 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
70 #define BUFFER_DELAY 100
71 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
73 /* Time in milliseconds we are allowed to stop the source,
74 * for sending the last part */
75 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
77 /* Maximum migrate downtime set to 2000 seconds */
78 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
79 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
81 /* Default compression thread count */
82 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
83 /* Default decompression thread count, usually decompression is at
84 * least 4 times as fast as compression.*/
85 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
86 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
87 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
88 /* Define default autoconverge cpu throttle migration parameters */
89 #define DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD 50
90 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
91 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
92 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
94 /* Migration XBZRLE default cache size */
95 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
97 /* The delay time (in ms) between two COLO checkpoints */
98 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
99 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
100 #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE
101 /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */
102 #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
103 /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
104 #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
106 /* Background transfer rate for postcopy, 0 means unlimited, note
107 * that page requests can still exceed this limit.
109 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
112 * Parameters for self_announce_delay giving a stream of RARP/ARP
113 * packets after migration.
115 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
116 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
117 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
118 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
120 static NotifierList migration_state_notifiers
=
121 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
123 /* Messages sent on the return path from destination to source */
124 enum mig_rp_message_type
{
125 MIG_RP_MSG_INVALID
= 0, /* Must be 0 */
126 MIG_RP_MSG_SHUT
, /* sibling will not send any more RP messages */
127 MIG_RP_MSG_PONG
, /* Response to a PING; data (seq: be32 ) */
129 MIG_RP_MSG_REQ_PAGES_ID
, /* data (start: be64, len: be32, id: string) */
130 MIG_RP_MSG_REQ_PAGES
, /* data (start: be64, len: be32) */
131 MIG_RP_MSG_RECV_BITMAP
, /* send recved_bitmap back to source */
132 MIG_RP_MSG_RESUME_ACK
, /* tell source that we are ready to resume */
137 /* Migration capabilities set */
138 struct MigrateCapsSet
{
139 int size
; /* Capability set size */
140 MigrationCapability caps
[]; /* Variadic array of capabilities */
142 typedef struct MigrateCapsSet MigrateCapsSet
;
144 /* Define and initialize MigrateCapsSet */
145 #define INITIALIZE_MIGRATE_CAPS_SET(_name, ...) \
146 MigrateCapsSet _name = { \
147 .size = sizeof((int []) { __VA_ARGS__ }) / sizeof(int), \
148 .caps = { __VA_ARGS__ } \
151 /* Background-snapshot compatibility check list */
153 INITIALIZE_MIGRATE_CAPS_SET(check_caps_background_snapshot
,
154 MIGRATION_CAPABILITY_POSTCOPY_RAM
,
155 MIGRATION_CAPABILITY_DIRTY_BITMAPS
,
156 MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
,
157 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE
,
158 MIGRATION_CAPABILITY_RETURN_PATH
,
159 MIGRATION_CAPABILITY_MULTIFD
,
160 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
,
161 MIGRATION_CAPABILITY_AUTO_CONVERGE
,
162 MIGRATION_CAPABILITY_RELEASE_RAM
,
163 MIGRATION_CAPABILITY_RDMA_PIN_ALL
,
164 MIGRATION_CAPABILITY_COMPRESS
,
165 MIGRATION_CAPABILITY_XBZRLE
,
166 MIGRATION_CAPABILITY_X_COLO
,
167 MIGRATION_CAPABILITY_VALIDATE_UUID
,
168 MIGRATION_CAPABILITY_ZERO_COPY_SEND
);
170 /* When we add fault tolerance, we could have several
171 migrations at once. For now we don't need to add
172 dynamic creation of migration */
174 static MigrationState
*current_migration
;
175 static MigrationIncomingState
*current_incoming
;
177 static GSList
*migration_blockers
;
179 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
180 static int migration_maybe_pause(MigrationState
*s
,
181 int *current_active_state
,
183 static void migrate_fd_cancel(MigrationState
*s
);
185 static bool migrate_allow_multi_channels
= true;
187 void migrate_protocol_allow_multi_channels(bool allow
)
189 migrate_allow_multi_channels
= allow
;
192 bool migrate_multi_channels_is_allowed(void)
194 return migrate_allow_multi_channels
;
197 static gint
page_request_addr_cmp(gconstpointer ap
, gconstpointer bp
)
199 uintptr_t a
= (uintptr_t) ap
, b
= (uintptr_t) bp
;
201 return (a
> b
) - (a
< b
);
204 void migration_object_init(void)
206 /* This can only be called once. */
207 assert(!current_migration
);
208 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
211 * Init the migrate incoming object as well no matter whether
212 * we'll use it or not.
214 assert(!current_incoming
);
215 current_incoming
= g_new0(MigrationIncomingState
, 1);
216 current_incoming
->state
= MIGRATION_STATUS_NONE
;
217 current_incoming
->postcopy_remote_fds
=
218 g_array_new(FALSE
, TRUE
, sizeof(struct PostCopyFD
));
219 qemu_mutex_init(¤t_incoming
->rp_mutex
);
220 qemu_mutex_init(¤t_incoming
->postcopy_prio_thread_mutex
);
221 qemu_event_init(¤t_incoming
->main_thread_load_event
, false);
222 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_dst
, 0);
223 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fault
, 0);
224 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fast_load
, 0);
225 qemu_mutex_init(¤t_incoming
->page_request_mutex
);
226 current_incoming
->page_requested
= g_tree_new(page_request_addr_cmp
);
228 migration_object_check(current_migration
, &error_fatal
);
232 dirty_bitmap_mig_init();
235 void migration_cancel(const Error
*error
)
238 migrate_set_error(current_migration
, error
);
240 migrate_fd_cancel(current_migration
);
243 void migration_shutdown(void)
246 * When the QEMU main thread exit, the COLO thread
247 * may wait a semaphore. So, we should wakeup the
248 * COLO thread before migration shutdown.
252 * Cancel the current migration - that will (eventually)
253 * stop the migration using this structure
255 migration_cancel(NULL
);
256 object_unref(OBJECT(current_migration
));
259 * Cancel outgoing migration of dirty bitmaps. It should
260 * at least unref used block nodes.
262 dirty_bitmap_mig_cancel_outgoing();
265 * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
266 * are non-critical data, and their loss never considered as
269 dirty_bitmap_mig_cancel_incoming();
273 MigrationState
*migrate_get_current(void)
275 /* This can only be called after the object created. */
276 assert(current_migration
);
277 return current_migration
;
280 MigrationIncomingState
*migration_incoming_get_current(void)
282 assert(current_incoming
);
283 return current_incoming
;
286 void migration_incoming_transport_cleanup(MigrationIncomingState
*mis
)
288 if (mis
->socket_address_list
) {
289 qapi_free_SocketAddressList(mis
->socket_address_list
);
290 mis
->socket_address_list
= NULL
;
293 if (mis
->transport_cleanup
) {
294 mis
->transport_cleanup(mis
->transport_data
);
295 mis
->transport_data
= mis
->transport_cleanup
= NULL
;
299 void migration_incoming_state_destroy(void)
301 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
303 if (mis
->to_src_file
) {
304 /* Tell source that we are done */
305 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
306 qemu_fclose(mis
->to_src_file
);
307 mis
->to_src_file
= NULL
;
310 if (mis
->from_src_file
) {
311 migration_ioc_unregister_yank_from_file(mis
->from_src_file
);
312 qemu_fclose(mis
->from_src_file
);
313 mis
->from_src_file
= NULL
;
315 if (mis
->postcopy_remote_fds
) {
316 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
317 mis
->postcopy_remote_fds
= NULL
;
320 migration_incoming_transport_cleanup(mis
);
321 qemu_event_reset(&mis
->main_thread_load_event
);
323 if (mis
->page_requested
) {
324 g_tree_destroy(mis
->page_requested
);
325 mis
->page_requested
= NULL
;
328 if (mis
->postcopy_qemufile_dst
) {
329 migration_ioc_unregister_yank_from_file(mis
->postcopy_qemufile_dst
);
330 qemu_fclose(mis
->postcopy_qemufile_dst
);
331 mis
->postcopy_qemufile_dst
= NULL
;
334 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
337 static void migrate_generate_event(int new_state
)
339 if (migrate_use_events()) {
340 qapi_event_send_migration(new_state
);
344 static bool migrate_late_block_activate(void)
348 s
= migrate_get_current();
350 return s
->enabled_capabilities
[
351 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE
];
355 * Send a message on the return channel back to the source
358 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
359 enum mig_rp_message_type message_type
,
360 uint16_t len
, void *data
)
364 trace_migrate_send_rp_message((int)message_type
, len
);
365 QEMU_LOCK_GUARD(&mis
->rp_mutex
);
368 * It's possible that the file handle got lost due to network
371 if (!mis
->to_src_file
) {
376 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
377 qemu_put_be16(mis
->to_src_file
, len
);
378 qemu_put_buffer(mis
->to_src_file
, data
, len
);
379 qemu_fflush(mis
->to_src_file
);
381 /* It's possible that qemu file got error during sending */
382 ret
= qemu_file_get_error(mis
->to_src_file
);
387 /* Request one page from the source VM at the given start address.
388 * rb: the RAMBlock to request the page in
389 * Start: Address offset within the RB
390 * Len: Length in bytes required - must be a multiple of pagesize
392 int migrate_send_rp_message_req_pages(MigrationIncomingState
*mis
,
393 RAMBlock
*rb
, ram_addr_t start
)
395 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
396 size_t msglen
= 12; /* start + len */
397 size_t len
= qemu_ram_pagesize(rb
);
398 enum mig_rp_message_type msg_type
;
402 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
403 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
406 * We maintain the last ramblock that we requested for page. Note that we
407 * don't need locking because this function will only be called within the
408 * postcopy ram fault thread.
410 if (rb
!= mis
->last_rb
) {
413 rbname
= qemu_ram_get_idstr(rb
);
414 rbname_len
= strlen(rbname
);
416 assert(rbname_len
< 256);
418 bufc
[msglen
++] = rbname_len
;
419 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
420 msglen
+= rbname_len
;
421 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
423 msg_type
= MIG_RP_MSG_REQ_PAGES
;
426 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
429 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
,
430 RAMBlock
*rb
, ram_addr_t start
, uint64_t haddr
)
432 void *aligned
= (void *)(uintptr_t)ROUND_DOWN(haddr
, qemu_ram_pagesize(rb
));
433 bool received
= false;
435 WITH_QEMU_LOCK_GUARD(&mis
->page_request_mutex
) {
436 received
= ramblock_recv_bitmap_test_byte_offset(rb
, start
);
437 if (!received
&& !g_tree_lookup(mis
->page_requested
, aligned
)) {
439 * The page has not been received, and it's not yet in the page
440 * request list. Queue it. Set the value of element to 1, so that
441 * things like g_tree_lookup() will return TRUE (1) when found.
443 g_tree_insert(mis
->page_requested
, aligned
, (gpointer
)1);
444 mis
->page_requested_count
++;
445 trace_postcopy_page_req_add(aligned
, mis
->page_requested_count
);
450 * If the page is there, skip sending the message. We don't even need the
451 * lock because as long as the page arrived, it'll be there forever.
457 return migrate_send_rp_message_req_pages(mis
, rb
, start
);
460 static bool migration_colo_enabled
;
461 bool migration_incoming_colo_enabled(void)
463 return migration_colo_enabled
;
466 void migration_incoming_disable_colo(void)
468 ram_block_discard_disable(false);
469 migration_colo_enabled
= false;
472 int migration_incoming_enable_colo(void)
474 if (ram_block_discard_disable(true)) {
475 error_report("COLO: cannot disable RAM discard");
478 migration_colo_enabled
= true;
482 void migrate_add_address(SocketAddress
*address
)
484 MigrationIncomingState
*mis
= migration_incoming_get_current();
486 QAPI_LIST_PREPEND(mis
->socket_address_list
,
487 QAPI_CLONE(SocketAddress
, address
));
490 static void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
492 const char *p
= NULL
;
494 migrate_protocol_allow_multi_channels(false); /* reset it anyway */
495 qapi_event_send_migration(MIGRATION_STATUS_SETUP
);
496 if (strstart(uri
, "tcp:", &p
) ||
497 strstart(uri
, "unix:", NULL
) ||
498 strstart(uri
, "vsock:", NULL
)) {
499 migrate_protocol_allow_multi_channels(true);
500 socket_start_incoming_migration(p
? p
: uri
, errp
);
502 } else if (strstart(uri
, "rdma:", &p
)) {
503 rdma_start_incoming_migration(p
, errp
);
505 } else if (strstart(uri
, "exec:", &p
)) {
506 exec_start_incoming_migration(p
, errp
);
507 } else if (strstart(uri
, "fd:", &p
)) {
508 fd_start_incoming_migration(p
, errp
);
510 error_setg(errp
, "unknown migration protocol: %s", uri
);
514 static void process_incoming_migration_bh(void *opaque
)
516 Error
*local_err
= NULL
;
517 MigrationIncomingState
*mis
= opaque
;
519 /* If capability late_block_activate is set:
520 * Only fire up the block code now if we're going to restart the
521 * VM, else 'cont' will do it.
522 * This causes file locking to happen; so we don't want it to happen
523 * unless we really are starting the VM.
525 if (!migrate_late_block_activate() ||
526 (autostart
&& (!global_state_received() ||
527 global_state_get_runstate() == RUN_STATE_RUNNING
))) {
528 /* Make sure all file formats throw away their mutable metadata.
529 * If we get an error here, just don't restart the VM yet. */
530 bdrv_activate_all(&local_err
);
532 error_report_err(local_err
);
539 * This must happen after all error conditions are dealt with and
540 * we're sure the VM is going to be running on this host.
542 qemu_announce_self(&mis
->announce_timer
, migrate_announce_params());
544 if (multifd_load_cleanup(&local_err
) != 0) {
545 error_report_err(local_err
);
548 /* If global state section was not received or we are in running
549 state, we need to obey autostart. Any other state is set with
552 dirty_bitmap_mig_before_vm_start();
554 if (!global_state_received() ||
555 global_state_get_runstate() == RUN_STATE_RUNNING
) {
559 runstate_set(RUN_STATE_PAUSED
);
561 } else if (migration_incoming_colo_enabled()) {
562 migration_incoming_disable_colo();
565 runstate_set(global_state_get_runstate());
568 * This must happen after any state changes since as soon as an external
569 * observer sees this event they might start to prod at the VM assuming
572 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
573 MIGRATION_STATUS_COMPLETED
);
574 qemu_bh_delete(mis
->bh
);
575 migration_incoming_state_destroy();
578 static void coroutine_fn
579 process_incoming_migration_co(void *opaque
)
581 MigrationIncomingState
*mis
= migration_incoming_get_current();
584 Error
*local_err
= NULL
;
586 assert(mis
->from_src_file
);
587 mis
->migration_incoming_co
= qemu_coroutine_self();
588 mis
->largest_page_size
= qemu_ram_pagesize_largest();
589 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
590 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
591 MIGRATION_STATUS_ACTIVE
);
592 ret
= qemu_loadvm_state(mis
->from_src_file
);
594 ps
= postcopy_state_get();
595 trace_process_incoming_migration_co_end(ret
, ps
);
596 if (ps
!= POSTCOPY_INCOMING_NONE
) {
597 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
599 * Where a migration had postcopy enabled (and thus went to advise)
600 * but managed to complete within the precopy period, we can use
603 postcopy_ram_incoming_cleanup(mis
);
604 } else if (ret
>= 0) {
606 * Postcopy was started, cleanup should happen at the end of the
609 trace_process_incoming_migration_co_postcopy_end_main();
612 /* Else if something went wrong then just fall out of the normal exit */
615 /* we get COLO info, and know if we are in COLO mode */
616 if (!ret
&& migration_incoming_colo_enabled()) {
617 /* Make sure all file formats throw away their mutable metadata */
618 bdrv_activate_all(&local_err
);
620 error_report_err(local_err
);
624 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
625 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
626 mis
->have_colo_incoming_thread
= true;
627 qemu_coroutine_yield();
629 qemu_mutex_unlock_iothread();
630 /* Wait checkpoint incoming thread exit before free resource */
631 qemu_thread_join(&mis
->colo_incoming_thread
);
632 qemu_mutex_lock_iothread();
633 /* We hold the global iothread lock, so it is safe here */
634 colo_release_ram_cache();
638 error_report("load of migration failed: %s", strerror(-ret
));
641 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
642 qemu_bh_schedule(mis
->bh
);
643 mis
->migration_incoming_co
= NULL
;
647 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
648 MIGRATION_STATUS_FAILED
);
649 qemu_fclose(mis
->from_src_file
);
650 if (multifd_load_cleanup(&local_err
) != 0) {
651 error_report_err(local_err
);
657 * migration_incoming_setup: Setup incoming migration
658 * @f: file for main migration channel
659 * @errp: where to put errors
661 * Returns: %true on success, %false on error.
663 static bool migration_incoming_setup(QEMUFile
*f
, Error
**errp
)
665 MigrationIncomingState
*mis
= migration_incoming_get_current();
667 if (multifd_load_setup(errp
) != 0) {
671 if (!mis
->from_src_file
) {
672 mis
->from_src_file
= f
;
674 qemu_file_set_blocking(f
, false);
678 void migration_incoming_process(void)
680 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
681 qemu_coroutine_enter(co
);
684 /* Returns true if recovered from a paused migration, otherwise false */
685 static bool postcopy_try_recover(void)
687 MigrationIncomingState
*mis
= migration_incoming_get_current();
689 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
690 /* Resumed from a paused postcopy migration */
692 /* This should be set already in migration_incoming_setup() */
693 assert(mis
->from_src_file
);
694 /* Postcopy has standalone thread to do vm load */
695 qemu_file_set_blocking(mis
->from_src_file
, true);
697 /* Re-configure the return path */
698 mis
->to_src_file
= qemu_file_get_return_path(mis
->from_src_file
);
700 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
701 MIGRATION_STATUS_POSTCOPY_RECOVER
);
704 * Here, we only wake up the main loading thread (while the
705 * rest threads will still be waiting), so that we can receive
706 * commands from source now, and answer it if needed. The
707 * rest threads will be woken up afterwards until we are sure
708 * that source is ready to reply to page requests.
710 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
717 void migration_fd_process_incoming(QEMUFile
*f
, Error
**errp
)
719 if (!migration_incoming_setup(f
, errp
)) {
722 if (postcopy_try_recover()) {
725 migration_incoming_process();
728 static bool migration_needs_multiple_sockets(void)
730 return migrate_use_multifd() || migrate_postcopy_preempt();
733 void migration_ioc_process_incoming(QIOChannel
*ioc
, Error
**errp
)
735 MigrationIncomingState
*mis
= migration_incoming_get_current();
736 Error
*local_err
= NULL
;
737 bool start_migration
;
740 if (!mis
->from_src_file
) {
741 /* The first connection (multifd may have multiple) */
742 f
= qemu_file_new_input(ioc
);
744 if (!migration_incoming_setup(f
, errp
)) {
749 * Common migration only needs one channel, so we can start
750 * right now. Some features need more than one channel, we wait.
752 start_migration
= !migration_needs_multiple_sockets();
754 /* Multiple connections */
755 assert(migration_needs_multiple_sockets());
756 if (migrate_use_multifd()) {
757 start_migration
= multifd_recv_new_channel(ioc
, &local_err
);
759 assert(migrate_postcopy_preempt());
760 f
= qemu_file_new_input(ioc
);
761 start_migration
= postcopy_preempt_new_channel(mis
, f
);
764 error_propagate(errp
, local_err
);
769 if (start_migration
) {
770 /* If it's a recovery, we're done */
771 if (postcopy_try_recover()) {
774 migration_incoming_process();
779 * @migration_has_all_channels: We have received all channels that we need
781 * Returns true when we have got connections to all the channels that
782 * we need for migration.
784 bool migration_has_all_channels(void)
786 MigrationIncomingState
*mis
= migration_incoming_get_current();
788 if (!mis
->from_src_file
) {
792 if (migrate_use_multifd()) {
793 return multifd_recv_all_channels_created();
796 if (migrate_postcopy_preempt()) {
797 return mis
->postcopy_qemufile_dst
!= NULL
;
804 * Send a 'SHUT' message on the return channel with the given value
805 * to indicate that we've finished with the RP. Non-0 value indicates
808 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
813 buf
= cpu_to_be32(value
);
814 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
818 * Send a 'PONG' message on the return channel with the given value
819 * (normally in response to a 'PING')
821 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
826 buf
= cpu_to_be32(value
);
827 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
830 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
838 * First, we send the header part. It contains only the len of
839 * idstr, and the idstr itself.
841 len
= strlen(block_name
);
843 memcpy(buf
+ 1, block_name
, len
);
845 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
846 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
851 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
854 * Next, we dump the received bitmap to the stream.
856 * TODO: currently we are safe since we are the only one that is
857 * using the to_src_file handle (fault thread is still paused),
858 * and it's ok even not taking the mutex. However the best way is
859 * to take the lock before sending the message header, and release
860 * the lock after sending the bitmap.
862 qemu_mutex_lock(&mis
->rp_mutex
);
863 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
864 qemu_mutex_unlock(&mis
->rp_mutex
);
866 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
869 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
873 buf
= cpu_to_be32(value
);
874 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
877 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
879 MigrationCapabilityStatusList
*head
= NULL
, **tail
= &head
;
880 MigrationCapabilityStatus
*caps
;
881 MigrationState
*s
= migrate_get_current();
884 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
885 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
886 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
890 caps
= g_malloc0(sizeof(*caps
));
891 caps
->capability
= i
;
892 caps
->state
= s
->enabled_capabilities
[i
];
893 QAPI_LIST_APPEND(tail
, caps
);
899 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
901 MigrationParameters
*params
;
902 MigrationState
*s
= migrate_get_current();
904 /* TODO use QAPI_CLONE() instead of duplicating it inline */
905 params
= g_malloc0(sizeof(*params
));
906 params
->has_compress_level
= true;
907 params
->compress_level
= s
->parameters
.compress_level
;
908 params
->has_compress_threads
= true;
909 params
->compress_threads
= s
->parameters
.compress_threads
;
910 params
->has_compress_wait_thread
= true;
911 params
->compress_wait_thread
= s
->parameters
.compress_wait_thread
;
912 params
->has_decompress_threads
= true;
913 params
->decompress_threads
= s
->parameters
.decompress_threads
;
914 params
->has_throttle_trigger_threshold
= true;
915 params
->throttle_trigger_threshold
= s
->parameters
.throttle_trigger_threshold
;
916 params
->has_cpu_throttle_initial
= true;
917 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
918 params
->has_cpu_throttle_increment
= true;
919 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
920 params
->has_cpu_throttle_tailslow
= true;
921 params
->cpu_throttle_tailslow
= s
->parameters
.cpu_throttle_tailslow
;
922 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
923 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
924 params
->tls_authz
= g_strdup(s
->parameters
.tls_authz
?
925 s
->parameters
.tls_authz
: "");
926 params
->has_max_bandwidth
= true;
927 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
928 params
->has_downtime_limit
= true;
929 params
->downtime_limit
= s
->parameters
.downtime_limit
;
930 params
->has_x_checkpoint_delay
= true;
931 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
932 params
->has_block_incremental
= true;
933 params
->block_incremental
= s
->parameters
.block_incremental
;
934 params
->has_multifd_channels
= true;
935 params
->multifd_channels
= s
->parameters
.multifd_channels
;
936 params
->has_multifd_compression
= true;
937 params
->multifd_compression
= s
->parameters
.multifd_compression
;
938 params
->has_multifd_zlib_level
= true;
939 params
->multifd_zlib_level
= s
->parameters
.multifd_zlib_level
;
940 params
->has_multifd_zstd_level
= true;
941 params
->multifd_zstd_level
= s
->parameters
.multifd_zstd_level
;
942 params
->has_xbzrle_cache_size
= true;
943 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
944 params
->has_max_postcopy_bandwidth
= true;
945 params
->max_postcopy_bandwidth
= s
->parameters
.max_postcopy_bandwidth
;
946 params
->has_max_cpu_throttle
= true;
947 params
->max_cpu_throttle
= s
->parameters
.max_cpu_throttle
;
948 params
->has_announce_initial
= true;
949 params
->announce_initial
= s
->parameters
.announce_initial
;
950 params
->has_announce_max
= true;
951 params
->announce_max
= s
->parameters
.announce_max
;
952 params
->has_announce_rounds
= true;
953 params
->announce_rounds
= s
->parameters
.announce_rounds
;
954 params
->has_announce_step
= true;
955 params
->announce_step
= s
->parameters
.announce_step
;
957 if (s
->parameters
.has_block_bitmap_mapping
) {
958 params
->has_block_bitmap_mapping
= true;
959 params
->block_bitmap_mapping
=
960 QAPI_CLONE(BitmapMigrationNodeAliasList
,
961 s
->parameters
.block_bitmap_mapping
);
967 void qmp_client_migrate_info(const char *protocol
, const char *hostname
,
968 bool has_port
, int64_t port
,
969 bool has_tls_port
, int64_t tls_port
,
970 const char *cert_subject
,
973 if (strcmp(protocol
, "spice") == 0) {
974 if (!qemu_using_spice(errp
)) {
978 if (!has_port
&& !has_tls_port
) {
979 error_setg(errp
, QERR_MISSING_PARAMETER
, "port/tls-port");
983 if (qemu_spice
.migrate_info(hostname
,
984 has_port
? port
: -1,
985 has_tls_port
? tls_port
: -1,
987 error_setg(errp
, "Could not set up display for migration");
993 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "protocol", "'spice'");
996 AnnounceParameters
*migrate_announce_params(void)
998 static AnnounceParameters ap
;
1000 MigrationState
*s
= migrate_get_current();
1002 ap
.initial
= s
->parameters
.announce_initial
;
1003 ap
.max
= s
->parameters
.announce_max
;
1004 ap
.rounds
= s
->parameters
.announce_rounds
;
1005 ap
.step
= s
->parameters
.announce_step
;
1011 * Return true if we're already in the middle of a migration
1012 * (i.e. any of the active or setup states)
1014 bool migration_is_setup_or_active(int state
)
1017 case MIGRATION_STATUS_ACTIVE
:
1018 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1019 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1020 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1021 case MIGRATION_STATUS_SETUP
:
1022 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1023 case MIGRATION_STATUS_DEVICE
:
1024 case MIGRATION_STATUS_WAIT_UNPLUG
:
1025 case MIGRATION_STATUS_COLO
:
1034 bool migration_is_running(int state
)
1037 case MIGRATION_STATUS_ACTIVE
:
1038 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1039 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1040 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1041 case MIGRATION_STATUS_SETUP
:
1042 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1043 case MIGRATION_STATUS_DEVICE
:
1044 case MIGRATION_STATUS_WAIT_UNPLUG
:
1045 case MIGRATION_STATUS_CANCELLING
:
1054 static void populate_time_info(MigrationInfo
*info
, MigrationState
*s
)
1056 info
->has_status
= true;
1057 info
->has_setup_time
= true;
1058 info
->setup_time
= s
->setup_time
;
1059 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
1060 info
->has_total_time
= true;
1061 info
->total_time
= s
->total_time
;
1062 info
->has_downtime
= true;
1063 info
->downtime
= s
->downtime
;
1065 info
->has_total_time
= true;
1066 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) -
1068 info
->has_expected_downtime
= true;
1069 info
->expected_downtime
= s
->expected_downtime
;
1073 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
1075 size_t page_size
= qemu_target_page_size();
1077 info
->ram
= g_malloc0(sizeof(*info
->ram
));
1078 info
->ram
->transferred
= stat64_get(&ram_atomic_counters
.transferred
);
1079 info
->ram
->total
= ram_bytes_total();
1080 info
->ram
->duplicate
= stat64_get(&ram_atomic_counters
.duplicate
);
1081 /* legacy value. It is not used anymore */
1082 info
->ram
->skipped
= 0;
1083 info
->ram
->normal
= stat64_get(&ram_atomic_counters
.normal
);
1084 info
->ram
->normal_bytes
= info
->ram
->normal
* page_size
;
1085 info
->ram
->mbps
= s
->mbps
;
1086 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
1087 info
->ram
->dirty_sync_missed_zero_copy
=
1088 ram_counters
.dirty_sync_missed_zero_copy
;
1089 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
1090 info
->ram
->page_size
= page_size
;
1091 info
->ram
->multifd_bytes
= ram_counters
.multifd_bytes
;
1092 info
->ram
->pages_per_second
= s
->pages_per_second
;
1093 info
->ram
->precopy_bytes
= ram_counters
.precopy_bytes
;
1094 info
->ram
->downtime_bytes
= ram_counters
.downtime_bytes
;
1095 info
->ram
->postcopy_bytes
= stat64_get(&ram_atomic_counters
.postcopy_bytes
);
1097 if (migrate_use_xbzrle()) {
1098 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
1099 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
1100 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
1101 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
1102 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
1103 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
1104 info
->xbzrle_cache
->encoding_rate
= xbzrle_counters
.encoding_rate
;
1105 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
1108 if (migrate_use_compression()) {
1109 info
->compression
= g_malloc0(sizeof(*info
->compression
));
1110 info
->compression
->pages
= compression_counters
.pages
;
1111 info
->compression
->busy
= compression_counters
.busy
;
1112 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
1113 info
->compression
->compressed_size
=
1114 compression_counters
.compressed_size
;
1115 info
->compression
->compression_rate
=
1116 compression_counters
.compression_rate
;
1119 if (cpu_throttle_active()) {
1120 info
->has_cpu_throttle_percentage
= true;
1121 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
1124 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
1125 info
->ram
->remaining
= ram_bytes_remaining();
1126 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
1130 static void populate_disk_info(MigrationInfo
*info
)
1132 if (blk_mig_active()) {
1133 info
->disk
= g_malloc0(sizeof(*info
->disk
));
1134 info
->disk
->transferred
= blk_mig_bytes_transferred();
1135 info
->disk
->remaining
= blk_mig_bytes_remaining();
1136 info
->disk
->total
= blk_mig_bytes_total();
1140 static void fill_source_migration_info(MigrationInfo
*info
)
1142 MigrationState
*s
= migrate_get_current();
1143 int state
= qatomic_read(&s
->state
);
1144 GSList
*cur_blocker
= migration_blockers
;
1146 info
->blocked_reasons
= NULL
;
1149 * There are two types of reasons a migration might be blocked;
1150 * a) devices marked in VMState as non-migratable, and
1151 * b) Explicit migration blockers
1152 * We need to add both of them here.
1154 qemu_savevm_non_migratable_list(&info
->blocked_reasons
);
1156 while (cur_blocker
) {
1157 QAPI_LIST_PREPEND(info
->blocked_reasons
,
1158 g_strdup(error_get_pretty(cur_blocker
->data
)));
1159 cur_blocker
= g_slist_next(cur_blocker
);
1161 info
->has_blocked_reasons
= info
->blocked_reasons
!= NULL
;
1164 case MIGRATION_STATUS_NONE
:
1165 /* no migration has happened ever */
1166 /* do not overwrite destination migration status */
1168 case MIGRATION_STATUS_SETUP
:
1169 info
->has_status
= true;
1170 info
->has_total_time
= false;
1172 case MIGRATION_STATUS_ACTIVE
:
1173 case MIGRATION_STATUS_CANCELLING
:
1174 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1175 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1176 case MIGRATION_STATUS_DEVICE
:
1177 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1178 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1179 /* TODO add some postcopy stats */
1180 populate_time_info(info
, s
);
1181 populate_ram_info(info
, s
);
1182 populate_disk_info(info
);
1183 populate_vfio_info(info
);
1185 case MIGRATION_STATUS_COLO
:
1186 info
->has_status
= true;
1187 /* TODO: display COLO specific information (checkpoint info etc.) */
1189 case MIGRATION_STATUS_COMPLETED
:
1190 populate_time_info(info
, s
);
1191 populate_ram_info(info
, s
);
1192 populate_vfio_info(info
);
1194 case MIGRATION_STATUS_FAILED
:
1195 info
->has_status
= true;
1197 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
1200 case MIGRATION_STATUS_CANCELLED
:
1201 info
->has_status
= true;
1203 case MIGRATION_STATUS_WAIT_UNPLUG
:
1204 info
->has_status
= true;
1207 info
->status
= state
;
1210 typedef enum WriteTrackingSupport
{
1211 WT_SUPPORT_UNKNOWN
= 0,
1213 WT_SUPPORT_AVAILABLE
,
1214 WT_SUPPORT_COMPATIBLE
1215 } WriteTrackingSupport
;
1218 WriteTrackingSupport
migrate_query_write_tracking(void)
1220 /* Check if kernel supports required UFFD features */
1221 if (!ram_write_tracking_available()) {
1222 return WT_SUPPORT_ABSENT
;
1225 * Check if current memory configuration is
1226 * compatible with required UFFD features.
1228 if (!ram_write_tracking_compatible()) {
1229 return WT_SUPPORT_AVAILABLE
;
1232 return WT_SUPPORT_COMPATIBLE
;
1236 * @migration_caps_check - check capability validity
1238 * @cap_list: old capability list, array of bool
1239 * @params: new capabilities to be applied soon
1240 * @errp: set *errp if the check failed, with reason
1242 * Returns true if check passed, otherwise false.
1244 static bool migrate_caps_check(bool *cap_list
,
1245 MigrationCapabilityStatusList
*params
,
1248 MigrationCapabilityStatusList
*cap
;
1249 bool old_postcopy_cap
;
1250 MigrationIncomingState
*mis
= migration_incoming_get_current();
1252 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1254 for (cap
= params
; cap
; cap
= cap
->next
) {
1255 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
1258 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1259 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
1260 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
1262 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
1267 #ifndef CONFIG_REPLICATION
1268 if (cap_list
[MIGRATION_CAPABILITY_X_COLO
]) {
1269 error_setg(errp
, "QEMU compiled without replication module"
1270 " can't enable COLO");
1271 error_append_hint(errp
, "Please enable replication before COLO.\n");
1276 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
1277 /* This check is reasonably expensive, so only when it's being
1278 * set the first time, also it's only the destination that needs
1281 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
1282 !postcopy_ram_supported_by_host(mis
)) {
1283 /* postcopy_ram_supported_by_host will have emitted a more
1286 error_setg(errp
, "Postcopy is not supported");
1290 if (cap_list
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
]) {
1291 error_setg(errp
, "Postcopy is not compatible with ignore-shared");
1296 if (cap_list
[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
]) {
1297 WriteTrackingSupport wt_support
;
1300 * Check if 'background-snapshot' capability is supported by
1301 * host kernel and compatible with guest memory configuration.
1303 wt_support
= migrate_query_write_tracking();
1304 if (wt_support
< WT_SUPPORT_AVAILABLE
) {
1305 error_setg(errp
, "Background-snapshot is not supported by host kernel");
1308 if (wt_support
< WT_SUPPORT_COMPATIBLE
) {
1309 error_setg(errp
, "Background-snapshot is not compatible "
1310 "with guest memory configuration");
1315 * Check if there are any migration capabilities
1316 * incompatible with 'background-snapshot'.
1318 for (idx
= 0; idx
< check_caps_background_snapshot
.size
; idx
++) {
1319 int incomp_cap
= check_caps_background_snapshot
.caps
[idx
];
1320 if (cap_list
[incomp_cap
]) {
1322 "Background-snapshot is not compatible with %s",
1323 MigrationCapability_str(incomp_cap
));
1330 if (cap_list
[MIGRATION_CAPABILITY_ZERO_COPY_SEND
] &&
1331 (!cap_list
[MIGRATION_CAPABILITY_MULTIFD
] ||
1332 cap_list
[MIGRATION_CAPABILITY_COMPRESS
] ||
1333 cap_list
[MIGRATION_CAPABILITY_XBZRLE
] ||
1334 migrate_multifd_compression() ||
1335 migrate_use_tls())) {
1337 "Zero copy only available for non-compressed non-TLS multifd migration");
1341 if (cap_list
[MIGRATION_CAPABILITY_ZERO_COPY_SEND
]) {
1343 "Zero copy currently only available on Linux");
1349 /* incoming side only */
1350 if (runstate_check(RUN_STATE_INMIGRATE
) &&
1351 !migrate_multi_channels_is_allowed() &&
1352 cap_list
[MIGRATION_CAPABILITY_MULTIFD
]) {
1353 error_setg(errp
, "multifd is not supported by current protocol");
1357 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_PREEMPT
]) {
1358 if (!cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
1359 error_setg(errp
, "Postcopy preempt requires postcopy-ram");
1364 * Preempt mode requires urgent pages to be sent in separate
1365 * channel, OTOH compression logic will disorder all pages into
1366 * different compression channels, which is not compatible with the
1367 * preempt assumptions on channel assignments.
1369 if (cap_list
[MIGRATION_CAPABILITY_COMPRESS
]) {
1370 error_setg(errp
, "Postcopy preempt not compatible with compress");
1375 if (cap_list
[MIGRATION_CAPABILITY_MULTIFD
]) {
1376 if (cap_list
[MIGRATION_CAPABILITY_COMPRESS
]) {
1377 error_setg(errp
, "Multifd is not compatible with compress");
1385 static void fill_destination_migration_info(MigrationInfo
*info
)
1387 MigrationIncomingState
*mis
= migration_incoming_get_current();
1389 if (mis
->socket_address_list
) {
1390 info
->has_socket_address
= true;
1391 info
->socket_address
=
1392 QAPI_CLONE(SocketAddressList
, mis
->socket_address_list
);
1395 switch (mis
->state
) {
1396 case MIGRATION_STATUS_NONE
:
1398 case MIGRATION_STATUS_SETUP
:
1399 case MIGRATION_STATUS_CANCELLING
:
1400 case MIGRATION_STATUS_CANCELLED
:
1401 case MIGRATION_STATUS_ACTIVE
:
1402 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1403 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1404 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1405 case MIGRATION_STATUS_FAILED
:
1406 case MIGRATION_STATUS_COLO
:
1407 info
->has_status
= true;
1409 case MIGRATION_STATUS_COMPLETED
:
1410 info
->has_status
= true;
1411 fill_destination_postcopy_migration_info(info
);
1414 info
->status
= mis
->state
;
1417 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1419 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1421 fill_destination_migration_info(info
);
1422 fill_source_migration_info(info
);
1427 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
1430 MigrationState
*s
= migrate_get_current();
1431 MigrationCapabilityStatusList
*cap
;
1432 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
1434 if (migration_is_running(s
->state
)) {
1435 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1439 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
1440 if (!migrate_caps_check(cap_list
, params
, errp
)) {
1444 for (cap
= params
; cap
; cap
= cap
->next
) {
1445 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
1450 * Check whether the parameters are valid. Error will be put into errp
1451 * (if provided). Return true if valid, otherwise false.
1453 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
1455 if (params
->has_compress_level
&&
1456 (params
->compress_level
> 9)) {
1457 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
1458 "a value between 0 and 9");
1462 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
1463 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1465 "a value between 1 and 255");
1469 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
1470 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1471 "decompress_threads",
1472 "a value between 1 and 255");
1476 if (params
->has_throttle_trigger_threshold
&&
1477 (params
->throttle_trigger_threshold
< 1 ||
1478 params
->throttle_trigger_threshold
> 100)) {
1479 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1480 "throttle_trigger_threshold",
1481 "an integer in the range of 1 to 100");
1485 if (params
->has_cpu_throttle_initial
&&
1486 (params
->cpu_throttle_initial
< 1 ||
1487 params
->cpu_throttle_initial
> 99)) {
1488 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1489 "cpu_throttle_initial",
1490 "an integer in the range of 1 to 99");
1494 if (params
->has_cpu_throttle_increment
&&
1495 (params
->cpu_throttle_increment
< 1 ||
1496 params
->cpu_throttle_increment
> 99)) {
1497 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1498 "cpu_throttle_increment",
1499 "an integer in the range of 1 to 99");
1503 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
1504 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1506 "an integer in the range of 0 to "stringify(SIZE_MAX
)
1511 if (params
->has_downtime_limit
&&
1512 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
1513 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1515 "an integer in the range of 0 to "
1516 stringify(MAX_MIGRATE_DOWNTIME
)" ms");
1520 /* x_checkpoint_delay is now always positive */
1522 if (params
->has_multifd_channels
&& (params
->multifd_channels
< 1)) {
1523 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1525 "a value between 1 and 255");
1529 if (params
->has_multifd_zlib_level
&&
1530 (params
->multifd_zlib_level
> 9)) {
1531 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zlib_level",
1532 "a value between 0 and 9");
1536 if (params
->has_multifd_zstd_level
&&
1537 (params
->multifd_zstd_level
> 20)) {
1538 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zstd_level",
1539 "a value between 0 and 20");
1543 if (params
->has_xbzrle_cache_size
&&
1544 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
1545 !is_power_of_2(params
->xbzrle_cache_size
))) {
1546 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1547 "xbzrle_cache_size",
1548 "a power of two no less than the target page size");
1552 if (params
->has_max_cpu_throttle
&&
1553 (params
->max_cpu_throttle
< params
->cpu_throttle_initial
||
1554 params
->max_cpu_throttle
> 99)) {
1555 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1557 "an integer in the range of cpu_throttle_initial to 99");
1561 if (params
->has_announce_initial
&&
1562 params
->announce_initial
> 100000) {
1563 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1565 "a value between 0 and 100000");
1568 if (params
->has_announce_max
&&
1569 params
->announce_max
> 100000) {
1570 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1572 "a value between 0 and 100000");
1575 if (params
->has_announce_rounds
&&
1576 params
->announce_rounds
> 1000) {
1577 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1579 "a value between 0 and 1000");
1582 if (params
->has_announce_step
&&
1583 (params
->announce_step
< 1 ||
1584 params
->announce_step
> 10000)) {
1585 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1587 "a value between 0 and 10000");
1591 if (params
->has_block_bitmap_mapping
&&
1592 !check_dirty_bitmap_mig_alias_map(params
->block_bitmap_mapping
, errp
)) {
1593 error_prepend(errp
, "Invalid mapping given for block-bitmap-mapping: ");
1598 if (migrate_use_zero_copy_send() &&
1599 ((params
->has_multifd_compression
&& params
->multifd_compression
) ||
1600 (params
->tls_creds
&& *params
->tls_creds
))) {
1602 "Zero copy only available for non-compressed non-TLS multifd migration");
1610 static void migrate_params_test_apply(MigrateSetParameters
*params
,
1611 MigrationParameters
*dest
)
1613 *dest
= migrate_get_current()->parameters
;
1615 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1617 if (params
->has_compress_level
) {
1618 dest
->compress_level
= params
->compress_level
;
1621 if (params
->has_compress_threads
) {
1622 dest
->compress_threads
= params
->compress_threads
;
1625 if (params
->has_compress_wait_thread
) {
1626 dest
->compress_wait_thread
= params
->compress_wait_thread
;
1629 if (params
->has_decompress_threads
) {
1630 dest
->decompress_threads
= params
->decompress_threads
;
1633 if (params
->has_throttle_trigger_threshold
) {
1634 dest
->throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1637 if (params
->has_cpu_throttle_initial
) {
1638 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
1641 if (params
->has_cpu_throttle_increment
) {
1642 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
1645 if (params
->has_cpu_throttle_tailslow
) {
1646 dest
->cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1649 if (params
->tls_creds
) {
1650 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1651 dest
->tls_creds
= params
->tls_creds
->u
.s
;
1654 if (params
->tls_hostname
) {
1655 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1656 dest
->tls_hostname
= params
->tls_hostname
->u
.s
;
1659 if (params
->has_max_bandwidth
) {
1660 dest
->max_bandwidth
= params
->max_bandwidth
;
1663 if (params
->has_downtime_limit
) {
1664 dest
->downtime_limit
= params
->downtime_limit
;
1667 if (params
->has_x_checkpoint_delay
) {
1668 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
1671 if (params
->has_block_incremental
) {
1672 dest
->block_incremental
= params
->block_incremental
;
1674 if (params
->has_multifd_channels
) {
1675 dest
->multifd_channels
= params
->multifd_channels
;
1677 if (params
->has_multifd_compression
) {
1678 dest
->multifd_compression
= params
->multifd_compression
;
1680 if (params
->has_xbzrle_cache_size
) {
1681 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
1683 if (params
->has_max_postcopy_bandwidth
) {
1684 dest
->max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1686 if (params
->has_max_cpu_throttle
) {
1687 dest
->max_cpu_throttle
= params
->max_cpu_throttle
;
1689 if (params
->has_announce_initial
) {
1690 dest
->announce_initial
= params
->announce_initial
;
1692 if (params
->has_announce_max
) {
1693 dest
->announce_max
= params
->announce_max
;
1695 if (params
->has_announce_rounds
) {
1696 dest
->announce_rounds
= params
->announce_rounds
;
1698 if (params
->has_announce_step
) {
1699 dest
->announce_step
= params
->announce_step
;
1702 if (params
->has_block_bitmap_mapping
) {
1703 dest
->has_block_bitmap_mapping
= true;
1704 dest
->block_bitmap_mapping
= params
->block_bitmap_mapping
;
1708 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
1710 MigrationState
*s
= migrate_get_current();
1712 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1714 if (params
->has_compress_level
) {
1715 s
->parameters
.compress_level
= params
->compress_level
;
1718 if (params
->has_compress_threads
) {
1719 s
->parameters
.compress_threads
= params
->compress_threads
;
1722 if (params
->has_compress_wait_thread
) {
1723 s
->parameters
.compress_wait_thread
= params
->compress_wait_thread
;
1726 if (params
->has_decompress_threads
) {
1727 s
->parameters
.decompress_threads
= params
->decompress_threads
;
1730 if (params
->has_throttle_trigger_threshold
) {
1731 s
->parameters
.throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1734 if (params
->has_cpu_throttle_initial
) {
1735 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
1738 if (params
->has_cpu_throttle_increment
) {
1739 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
1742 if (params
->has_cpu_throttle_tailslow
) {
1743 s
->parameters
.cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1746 if (params
->tls_creds
) {
1747 g_free(s
->parameters
.tls_creds
);
1748 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1749 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1752 if (params
->tls_hostname
) {
1753 g_free(s
->parameters
.tls_hostname
);
1754 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1755 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1758 if (params
->tls_authz
) {
1759 g_free(s
->parameters
.tls_authz
);
1760 assert(params
->tls_authz
->type
== QTYPE_QSTRING
);
1761 s
->parameters
.tls_authz
= g_strdup(params
->tls_authz
->u
.s
);
1764 if (params
->has_max_bandwidth
) {
1765 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1766 if (s
->to_dst_file
&& !migration_in_postcopy()) {
1767 qemu_file_set_rate_limit(s
->to_dst_file
,
1768 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1772 if (params
->has_downtime_limit
) {
1773 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1776 if (params
->has_x_checkpoint_delay
) {
1777 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1778 if (migration_in_colo_state()) {
1779 colo_checkpoint_notify(s
);
1783 if (params
->has_block_incremental
) {
1784 s
->parameters
.block_incremental
= params
->block_incremental
;
1786 if (params
->has_multifd_channels
) {
1787 s
->parameters
.multifd_channels
= params
->multifd_channels
;
1789 if (params
->has_multifd_compression
) {
1790 s
->parameters
.multifd_compression
= params
->multifd_compression
;
1792 if (params
->has_xbzrle_cache_size
) {
1793 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1794 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1796 if (params
->has_max_postcopy_bandwidth
) {
1797 s
->parameters
.max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1798 if (s
->to_dst_file
&& migration_in_postcopy()) {
1799 qemu_file_set_rate_limit(s
->to_dst_file
,
1800 s
->parameters
.max_postcopy_bandwidth
/ XFER_LIMIT_RATIO
);
1803 if (params
->has_max_cpu_throttle
) {
1804 s
->parameters
.max_cpu_throttle
= params
->max_cpu_throttle
;
1806 if (params
->has_announce_initial
) {
1807 s
->parameters
.announce_initial
= params
->announce_initial
;
1809 if (params
->has_announce_max
) {
1810 s
->parameters
.announce_max
= params
->announce_max
;
1812 if (params
->has_announce_rounds
) {
1813 s
->parameters
.announce_rounds
= params
->announce_rounds
;
1815 if (params
->has_announce_step
) {
1816 s
->parameters
.announce_step
= params
->announce_step
;
1819 if (params
->has_block_bitmap_mapping
) {
1820 qapi_free_BitmapMigrationNodeAliasList(
1821 s
->parameters
.block_bitmap_mapping
);
1823 s
->parameters
.has_block_bitmap_mapping
= true;
1824 s
->parameters
.block_bitmap_mapping
=
1825 QAPI_CLONE(BitmapMigrationNodeAliasList
,
1826 params
->block_bitmap_mapping
);
1830 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1832 MigrationParameters tmp
;
1834 /* TODO Rewrite "" to null instead */
1835 if (params
->tls_creds
1836 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1837 qobject_unref(params
->tls_creds
->u
.n
);
1838 params
->tls_creds
->type
= QTYPE_QSTRING
;
1839 params
->tls_creds
->u
.s
= strdup("");
1841 /* TODO Rewrite "" to null instead */
1842 if (params
->tls_hostname
1843 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1844 qobject_unref(params
->tls_hostname
->u
.n
);
1845 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1846 params
->tls_hostname
->u
.s
= strdup("");
1849 migrate_params_test_apply(params
, &tmp
);
1851 if (!migrate_params_check(&tmp
, errp
)) {
1852 /* Invalid parameter */
1856 migrate_params_apply(params
, errp
);
1860 void qmp_migrate_start_postcopy(Error
**errp
)
1862 MigrationState
*s
= migrate_get_current();
1864 if (!migrate_postcopy()) {
1865 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1866 " the start of migration");
1870 if (s
->state
== MIGRATION_STATUS_NONE
) {
1871 error_setg(errp
, "Postcopy must be started after migration has been"
1876 * we don't error if migration has finished since that would be racy
1877 * with issuing this command.
1879 qatomic_set(&s
->start_postcopy
, true);
1882 /* shared migration helpers */
1884 void migrate_set_state(int *state
, int old_state
, int new_state
)
1886 assert(new_state
< MIGRATION_STATUS__MAX
);
1887 if (qatomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1888 trace_migrate_set_state(MigrationStatus_str(new_state
));
1889 migrate_generate_event(new_state
);
1893 static MigrationCapabilityStatus
*migrate_cap_add(MigrationCapability index
,
1896 MigrationCapabilityStatus
*cap
;
1898 cap
= g_new0(MigrationCapabilityStatus
, 1);
1899 cap
->capability
= index
;
1905 void migrate_set_block_enabled(bool value
, Error
**errp
)
1907 MigrationCapabilityStatusList
*cap
= NULL
;
1909 QAPI_LIST_PREPEND(cap
, migrate_cap_add(MIGRATION_CAPABILITY_BLOCK
, value
));
1910 qmp_migrate_set_capabilities(cap
, errp
);
1911 qapi_free_MigrationCapabilityStatusList(cap
);
1914 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1916 s
->parameters
.block_incremental
= value
;
1919 static void block_cleanup_parameters(MigrationState
*s
)
1921 if (s
->must_remove_block_options
) {
1922 /* setting to false can never fail */
1923 migrate_set_block_enabled(false, &error_abort
);
1924 migrate_set_block_incremental(s
, false);
1925 s
->must_remove_block_options
= false;
1929 static void migrate_fd_cleanup(MigrationState
*s
)
1931 qemu_bh_delete(s
->cleanup_bh
);
1932 s
->cleanup_bh
= NULL
;
1934 g_free(s
->hostname
);
1937 qemu_savevm_state_cleanup();
1939 if (s
->to_dst_file
) {
1942 trace_migrate_fd_cleanup();
1943 qemu_mutex_unlock_iothread();
1944 if (s
->migration_thread_running
) {
1945 qemu_thread_join(&s
->thread
);
1946 s
->migration_thread_running
= false;
1948 qemu_mutex_lock_iothread();
1950 multifd_save_cleanup();
1951 qemu_mutex_lock(&s
->qemu_file_lock
);
1952 tmp
= s
->to_dst_file
;
1953 s
->to_dst_file
= NULL
;
1954 qemu_mutex_unlock(&s
->qemu_file_lock
);
1956 * Close the file handle without the lock to make sure the
1957 * critical section won't block for long.
1959 migration_ioc_unregister_yank_from_file(tmp
);
1963 if (s
->postcopy_qemufile_src
) {
1964 migration_ioc_unregister_yank_from_file(s
->postcopy_qemufile_src
);
1965 qemu_fclose(s
->postcopy_qemufile_src
);
1966 s
->postcopy_qemufile_src
= NULL
;
1969 assert(!migration_is_active(s
));
1971 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1972 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1973 MIGRATION_STATUS_CANCELLED
);
1977 /* It is used on info migrate. We can't free it */
1978 error_report_err(error_copy(s
->error
));
1980 notifier_list_notify(&migration_state_notifiers
, s
);
1981 block_cleanup_parameters(s
);
1982 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1985 static void migrate_fd_cleanup_schedule(MigrationState
*s
)
1988 * Ref the state for bh, because it may be called when
1989 * there're already no other refs
1991 object_ref(OBJECT(s
));
1992 qemu_bh_schedule(s
->cleanup_bh
);
1995 static void migrate_fd_cleanup_bh(void *opaque
)
1997 MigrationState
*s
= opaque
;
1998 migrate_fd_cleanup(s
);
1999 object_unref(OBJECT(s
));
2002 void migrate_set_error(MigrationState
*s
, const Error
*error
)
2004 QEMU_LOCK_GUARD(&s
->error_mutex
);
2006 s
->error
= error_copy(error
);
2010 static void migrate_error_free(MigrationState
*s
)
2012 QEMU_LOCK_GUARD(&s
->error_mutex
);
2014 error_free(s
->error
);
2019 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
2021 trace_migrate_fd_error(error_get_pretty(error
));
2022 assert(s
->to_dst_file
== NULL
);
2023 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2024 MIGRATION_STATUS_FAILED
);
2025 migrate_set_error(s
, error
);
2028 static void migrate_fd_cancel(MigrationState
*s
)
2031 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
2032 trace_migrate_fd_cancel();
2034 WITH_QEMU_LOCK_GUARD(&s
->qemu_file_lock
) {
2035 if (s
->rp_state
.from_dst_file
) {
2036 /* shutdown the rp socket, so causing the rp thread to shutdown */
2037 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
2042 old_state
= s
->state
;
2043 if (!migration_is_running(old_state
)) {
2046 /* If the migration is paused, kick it out of the pause */
2047 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
2048 qemu_sem_post(&s
->pause_sem
);
2050 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
2051 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
2054 * If we're unlucky the migration code might be stuck somewhere in a
2055 * send/write while the network has failed and is waiting to timeout;
2056 * if we've got shutdown(2) available then we can force it to quit.
2057 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
2058 * called in a bh, so there is no race against this cancel.
2060 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
2061 qemu_file_shutdown(f
);
2063 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
2064 Error
*local_err
= NULL
;
2066 bdrv_activate_all(&local_err
);
2068 error_report_err(local_err
);
2070 s
->block_inactive
= false;
2075 void add_migration_state_change_notifier(Notifier
*notify
)
2077 notifier_list_add(&migration_state_notifiers
, notify
);
2080 void remove_migration_state_change_notifier(Notifier
*notify
)
2082 notifier_remove(notify
);
2085 bool migration_in_setup(MigrationState
*s
)
2087 return s
->state
== MIGRATION_STATUS_SETUP
;
2090 bool migration_has_finished(MigrationState
*s
)
2092 return s
->state
== MIGRATION_STATUS_COMPLETED
;
2095 bool migration_has_failed(MigrationState
*s
)
2097 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
2098 s
->state
== MIGRATION_STATUS_FAILED
);
2101 bool migration_in_postcopy(void)
2103 MigrationState
*s
= migrate_get_current();
2106 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
2107 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
2108 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
2115 bool migration_in_postcopy_after_devices(MigrationState
*s
)
2117 return migration_in_postcopy() && s
->postcopy_after_devices
;
2120 bool migration_in_incoming_postcopy(void)
2122 PostcopyState ps
= postcopy_state_get();
2124 return ps
>= POSTCOPY_INCOMING_DISCARD
&& ps
< POSTCOPY_INCOMING_END
;
2127 bool migration_in_bg_snapshot(void)
2129 MigrationState
*s
= migrate_get_current();
2131 return migrate_background_snapshot() &&
2132 migration_is_setup_or_active(s
->state
);
2135 bool migration_is_idle(void)
2137 MigrationState
*s
= current_migration
;
2144 case MIGRATION_STATUS_NONE
:
2145 case MIGRATION_STATUS_CANCELLED
:
2146 case MIGRATION_STATUS_COMPLETED
:
2147 case MIGRATION_STATUS_FAILED
:
2149 case MIGRATION_STATUS_SETUP
:
2150 case MIGRATION_STATUS_CANCELLING
:
2151 case MIGRATION_STATUS_ACTIVE
:
2152 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
2153 case MIGRATION_STATUS_COLO
:
2154 case MIGRATION_STATUS_PRE_SWITCHOVER
:
2155 case MIGRATION_STATUS_DEVICE
:
2156 case MIGRATION_STATUS_WAIT_UNPLUG
:
2158 case MIGRATION_STATUS__MAX
:
2159 g_assert_not_reached();
2165 bool migration_is_active(MigrationState
*s
)
2167 return (s
->state
== MIGRATION_STATUS_ACTIVE
||
2168 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2171 void migrate_init(MigrationState
*s
)
2174 * Reinitialise all migration state, except
2175 * parameters/capabilities that the user set, and
2180 s
->to_dst_file
= NULL
;
2181 s
->state
= MIGRATION_STATUS_NONE
;
2182 s
->rp_state
.from_dst_file
= NULL
;
2183 s
->rp_state
.error
= false;
2185 s
->pages_per_second
= 0.0;
2187 s
->expected_downtime
= 0;
2189 s
->start_postcopy
= false;
2190 s
->postcopy_after_devices
= false;
2191 s
->migration_thread_running
= false;
2192 error_free(s
->error
);
2196 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
2198 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2200 s
->vm_was_running
= false;
2201 s
->iteration_initial_bytes
= 0;
2202 s
->threshold_size
= 0;
2205 int migrate_add_blocker_internal(Error
*reason
, Error
**errp
)
2207 /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */
2208 if (runstate_check(RUN_STATE_SAVE_VM
) || !migration_is_idle()) {
2209 error_propagate_prepend(errp
, error_copy(reason
),
2210 "disallowing migration blocker "
2211 "(migration/snapshot in progress) for: ");
2215 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
2219 int migrate_add_blocker(Error
*reason
, Error
**errp
)
2221 if (only_migratable
) {
2222 error_propagate_prepend(errp
, error_copy(reason
),
2223 "disallowing migration blocker "
2224 "(--only-migratable) for: ");
2228 return migrate_add_blocker_internal(reason
, errp
);
2231 void migrate_del_blocker(Error
*reason
)
2233 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
2236 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
2238 Error
*local_err
= NULL
;
2239 static bool once
= true;
2242 error_setg(errp
, "The incoming migration has already been started");
2245 if (!runstate_check(RUN_STATE_INMIGRATE
)) {
2246 error_setg(errp
, "'-incoming' was not specified on the command line");
2250 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
2254 qemu_start_incoming_migration(uri
, &local_err
);
2257 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2258 error_propagate(errp
, local_err
);
2265 void qmp_migrate_recover(const char *uri
, Error
**errp
)
2267 MigrationIncomingState
*mis
= migration_incoming_get_current();
2270 * Don't even bother to use ERRP_GUARD() as it _must_ always be set by
2271 * callers (no one should ignore a recover failure); if there is, it's a
2272 * programming error.
2276 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2277 error_setg(errp
, "Migrate recover can only be run "
2278 "when postcopy is paused.");
2282 /* If there's an existing transport, release it */
2283 migration_incoming_transport_cleanup(mis
);
2286 * Note that this call will never start a real migration; it will
2287 * only re-setup the migration stream and poke existing migration
2288 * to continue using that newly established channel.
2290 qemu_start_incoming_migration(uri
, errp
);
2293 void qmp_migrate_pause(Error
**errp
)
2295 MigrationState
*ms
= migrate_get_current();
2296 MigrationIncomingState
*mis
= migration_incoming_get_current();
2299 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2300 /* Source side, during postcopy */
2301 qemu_mutex_lock(&ms
->qemu_file_lock
);
2302 ret
= qemu_file_shutdown(ms
->to_dst_file
);
2303 qemu_mutex_unlock(&ms
->qemu_file_lock
);
2305 error_setg(errp
, "Failed to pause source migration");
2310 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2311 ret
= qemu_file_shutdown(mis
->from_src_file
);
2313 error_setg(errp
, "Failed to pause destination migration");
2318 error_setg(errp
, "migrate-pause is currently only supported "
2319 "during postcopy-active state");
2322 bool migration_is_blocked(Error
**errp
)
2324 if (qemu_savevm_state_blocked(errp
)) {
2328 if (migration_blockers
) {
2329 error_propagate(errp
, error_copy(migration_blockers
->data
));
2336 /* Returns true if continue to migrate, or false if error detected */
2337 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
2338 bool resume
, Error
**errp
)
2340 Error
*local_err
= NULL
;
2343 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2344 error_setg(errp
, "Cannot resume if there is no "
2345 "paused migration");
2350 * Postcopy recovery won't work well with release-ram
2351 * capability since release-ram will drop the page buffer as
2352 * long as the page is put into the send buffer. So if there
2353 * is a network failure happened, any page buffers that have
2354 * not yet reached the destination VM but have already been
2355 * sent from the source VM will be lost forever. Let's refuse
2356 * the client from resuming such a postcopy migration.
2357 * Luckily release-ram was designed to only be used when src
2358 * and destination VMs are on the same host, so it should be
2361 if (migrate_release_ram()) {
2362 error_setg(errp
, "Postcopy recovery cannot work "
2363 "when release-ram capability is set");
2367 /* This is a resume, skip init status */
2371 if (migration_is_running(s
->state
)) {
2372 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
2376 if (runstate_check(RUN_STATE_INMIGRATE
)) {
2377 error_setg(errp
, "Guest is waiting for an incoming migration");
2381 if (runstate_check(RUN_STATE_POSTMIGRATE
)) {
2382 error_setg(errp
, "Can't migrate the vm that was paused due to "
2383 "previous migration");
2387 if (migration_is_blocked(errp
)) {
2391 if (blk
|| blk_inc
) {
2392 if (migrate_colo_enabled()) {
2393 error_setg(errp
, "No disk migration is required in COLO mode");
2396 if (migrate_use_block() || migrate_use_block_incremental()) {
2397 error_setg(errp
, "Command options are incompatible with "
2398 "current migration capabilities");
2401 migrate_set_block_enabled(true, &local_err
);
2403 error_propagate(errp
, local_err
);
2406 s
->must_remove_block_options
= true;
2410 migrate_set_block_incremental(s
, true);
2415 * set ram_counters compression_counters memory to zero for a
2418 memset(&ram_counters
, 0, sizeof(ram_counters
));
2419 memset(&compression_counters
, 0, sizeof(compression_counters
));
2424 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
2425 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
2426 bool has_resume
, bool resume
, Error
**errp
)
2428 Error
*local_err
= NULL
;
2429 MigrationState
*s
= migrate_get_current();
2430 const char *p
= NULL
;
2432 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
2433 has_resume
&& resume
, errp
)) {
2434 /* Error detected, put into errp */
2438 if (!(has_resume
&& resume
)) {
2439 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
2444 migrate_protocol_allow_multi_channels(false);
2445 if (strstart(uri
, "tcp:", &p
) ||
2446 strstart(uri
, "unix:", NULL
) ||
2447 strstart(uri
, "vsock:", NULL
)) {
2448 migrate_protocol_allow_multi_channels(true);
2449 socket_start_outgoing_migration(s
, p
? p
: uri
, &local_err
);
2451 } else if (strstart(uri
, "rdma:", &p
)) {
2452 rdma_start_outgoing_migration(s
, p
, &local_err
);
2454 } else if (strstart(uri
, "exec:", &p
)) {
2455 exec_start_outgoing_migration(s
, p
, &local_err
);
2456 } else if (strstart(uri
, "fd:", &p
)) {
2457 fd_start_outgoing_migration(s
, p
, &local_err
);
2459 if (!(has_resume
&& resume
)) {
2460 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2462 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
2463 "a valid migration protocol");
2464 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2465 MIGRATION_STATUS_FAILED
);
2466 block_cleanup_parameters(s
);
2471 if (!(has_resume
&& resume
)) {
2472 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2474 migrate_fd_error(s
, local_err
);
2475 error_propagate(errp
, local_err
);
2480 void qmp_migrate_cancel(Error
**errp
)
2482 migration_cancel(NULL
);
2485 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
2487 MigrationState
*s
= migrate_get_current();
2488 if (s
->state
!= state
) {
2489 error_setg(errp
, "Migration not in expected state: %s",
2490 MigrationStatus_str(s
->state
));
2493 qemu_sem_post(&s
->pause_sem
);
2496 bool migrate_release_ram(void)
2500 s
= migrate_get_current();
2502 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
2505 bool migrate_postcopy_ram(void)
2509 s
= migrate_get_current();
2511 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
2514 bool migrate_postcopy(void)
2516 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2519 bool migrate_auto_converge(void)
2523 s
= migrate_get_current();
2525 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
2528 bool migrate_zero_blocks(void)
2532 s
= migrate_get_current();
2534 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
2537 bool migrate_postcopy_blocktime(void)
2541 s
= migrate_get_current();
2543 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
2546 bool migrate_use_compression(void)
2550 s
= migrate_get_current();
2552 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
2555 int migrate_compress_level(void)
2559 s
= migrate_get_current();
2561 return s
->parameters
.compress_level
;
2564 int migrate_compress_threads(void)
2568 s
= migrate_get_current();
2570 return s
->parameters
.compress_threads
;
2573 int migrate_compress_wait_thread(void)
2577 s
= migrate_get_current();
2579 return s
->parameters
.compress_wait_thread
;
2582 int migrate_decompress_threads(void)
2586 s
= migrate_get_current();
2588 return s
->parameters
.decompress_threads
;
2591 bool migrate_dirty_bitmaps(void)
2595 s
= migrate_get_current();
2597 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
2600 bool migrate_ignore_shared(void)
2604 s
= migrate_get_current();
2606 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
];
2609 bool migrate_validate_uuid(void)
2613 s
= migrate_get_current();
2615 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_VALIDATE_UUID
];
2618 bool migrate_use_events(void)
2622 s
= migrate_get_current();
2624 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
2627 bool migrate_use_multifd(void)
2631 s
= migrate_get_current();
2633 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_MULTIFD
];
2636 bool migrate_pause_before_switchover(void)
2640 s
= migrate_get_current();
2642 return s
->enabled_capabilities
[
2643 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
2646 int migrate_multifd_channels(void)
2650 s
= migrate_get_current();
2652 return s
->parameters
.multifd_channels
;
2655 MultiFDCompression
migrate_multifd_compression(void)
2659 s
= migrate_get_current();
2661 assert(s
->parameters
.multifd_compression
< MULTIFD_COMPRESSION__MAX
);
2662 return s
->parameters
.multifd_compression
;
2665 int migrate_multifd_zlib_level(void)
2669 s
= migrate_get_current();
2671 return s
->parameters
.multifd_zlib_level
;
2674 int migrate_multifd_zstd_level(void)
2678 s
= migrate_get_current();
2680 return s
->parameters
.multifd_zstd_level
;
2684 bool migrate_use_zero_copy_send(void)
2688 s
= migrate_get_current();
2690 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_COPY_SEND
];
2694 int migrate_use_tls(void)
2698 s
= migrate_get_current();
2700 return s
->parameters
.tls_creds
&& *s
->parameters
.tls_creds
;
2703 int migrate_use_xbzrle(void)
2707 s
= migrate_get_current();
2709 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
2712 uint64_t migrate_xbzrle_cache_size(void)
2716 s
= migrate_get_current();
2718 return s
->parameters
.xbzrle_cache_size
;
2721 static int64_t migrate_max_postcopy_bandwidth(void)
2725 s
= migrate_get_current();
2727 return s
->parameters
.max_postcopy_bandwidth
;
2730 bool migrate_use_block(void)
2734 s
= migrate_get_current();
2736 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
2739 bool migrate_use_return_path(void)
2743 s
= migrate_get_current();
2745 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
2748 bool migrate_use_block_incremental(void)
2752 s
= migrate_get_current();
2754 return s
->parameters
.block_incremental
;
2757 bool migrate_background_snapshot(void)
2761 s
= migrate_get_current();
2763 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
];
2766 bool migrate_postcopy_preempt(void)
2770 s
= migrate_get_current();
2772 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_PREEMPT
];
2775 /* migration thread support */
2777 * Something bad happened to the RP stream, mark an error
2778 * The caller shall print or trace something to indicate why
2780 static void mark_source_rp_bad(MigrationState
*s
)
2782 s
->rp_state
.error
= true;
2785 static struct rp_cmd_args
{
2786 ssize_t len
; /* -1 = variable */
2789 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
2790 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
2791 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
2792 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
2793 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
2794 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
2795 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
2796 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
2800 * Process a request for pages received on the return path,
2801 * We're allowed to send more than requested (e.g. to round to our page size)
2802 * and we don't need to send pages that have already been sent.
2804 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
2805 ram_addr_t start
, size_t len
)
2807 long our_host_ps
= qemu_real_host_page_size();
2809 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
2812 * Since we currently insist on matching page sizes, just sanity check
2813 * we're being asked for whole host pages.
2815 if (!QEMU_IS_ALIGNED(start
, our_host_ps
) ||
2816 !QEMU_IS_ALIGNED(len
, our_host_ps
)) {
2817 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2818 " len: %zd", __func__
, start
, len
);
2819 mark_source_rp_bad(ms
);
2823 if (ram_save_queue_pages(rbname
, start
, len
)) {
2824 mark_source_rp_bad(ms
);
2828 /* Return true to retry, false to quit */
2829 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
2831 trace_postcopy_pause_return_path();
2833 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
2835 trace_postcopy_pause_return_path_continued();
2840 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
2842 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
2845 error_report("%s: invalid block name '%s'", __func__
, block_name
);
2849 /* Fetch the received bitmap and refresh the dirty bitmap */
2850 return ram_dirty_bitmap_reload(s
, block
);
2853 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
2855 trace_source_return_path_thread_resume_ack(value
);
2857 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
2858 error_report("%s: illegal resume_ack value %"PRIu32
,
2863 /* Now both sides are active. */
2864 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2865 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2867 /* Notify send thread that time to continue send pages */
2868 qemu_sem_post(&s
->rp_state
.rp_sem
);
2874 * Release ms->rp_state.from_dst_file (and postcopy_qemufile_src if
2875 * existed) in a safe way.
2877 static void migration_release_dst_files(MigrationState
*ms
)
2881 WITH_QEMU_LOCK_GUARD(&ms
->qemu_file_lock
) {
2883 * Reset the from_dst_file pointer first before releasing it, as we
2884 * can't block within lock section
2886 file
= ms
->rp_state
.from_dst_file
;
2887 ms
->rp_state
.from_dst_file
= NULL
;
2891 * Do the same to postcopy fast path socket too if there is. No
2892 * locking needed because this qemufile should only be managed by
2893 * return path thread.
2895 if (ms
->postcopy_qemufile_src
) {
2896 migration_ioc_unregister_yank_from_file(ms
->postcopy_qemufile_src
);
2897 qemu_file_shutdown(ms
->postcopy_qemufile_src
);
2898 qemu_fclose(ms
->postcopy_qemufile_src
);
2899 ms
->postcopy_qemufile_src
= NULL
;
2906 * Handles messages sent on the return path towards the source VM
2909 static void *source_return_path_thread(void *opaque
)
2911 MigrationState
*ms
= opaque
;
2912 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
2913 uint16_t header_len
, header_type
;
2915 uint32_t tmp32
, sibling_error
;
2916 ram_addr_t start
= 0; /* =0 to silence warning */
2917 size_t len
= 0, expected_len
;
2920 trace_source_return_path_thread_entry();
2921 rcu_register_thread();
2924 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
2925 migration_is_setup_or_active(ms
->state
)) {
2926 trace_source_return_path_thread_loop_top();
2927 header_type
= qemu_get_be16(rp
);
2928 header_len
= qemu_get_be16(rp
);
2930 if (qemu_file_get_error(rp
)) {
2931 mark_source_rp_bad(ms
);
2935 if (header_type
>= MIG_RP_MSG_MAX
||
2936 header_type
== MIG_RP_MSG_INVALID
) {
2937 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2938 header_type
, header_len
);
2939 mark_source_rp_bad(ms
);
2943 if ((rp_cmd_args
[header_type
].len
!= -1 &&
2944 header_len
!= rp_cmd_args
[header_type
].len
) ||
2945 header_len
> sizeof(buf
)) {
2946 error_report("RP: Received '%s' message (0x%04x) with"
2947 "incorrect length %d expecting %zu",
2948 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
2949 (size_t)rp_cmd_args
[header_type
].len
);
2950 mark_source_rp_bad(ms
);
2954 /* We know we've got a valid header by this point */
2955 res
= qemu_get_buffer(rp
, buf
, header_len
);
2956 if (res
!= header_len
) {
2957 error_report("RP: Failed reading data for message 0x%04x"
2958 " read %d expected %d",
2959 header_type
, res
, header_len
);
2960 mark_source_rp_bad(ms
);
2964 /* OK, we have the message and the data */
2965 switch (header_type
) {
2966 case MIG_RP_MSG_SHUT
:
2967 sibling_error
= ldl_be_p(buf
);
2968 trace_source_return_path_thread_shut(sibling_error
);
2969 if (sibling_error
) {
2970 error_report("RP: Sibling indicated error %d", sibling_error
);
2971 mark_source_rp_bad(ms
);
2974 * We'll let the main thread deal with closing the RP
2975 * we could do a shutdown(2) on it, but we're the only user
2976 * anyway, so there's nothing gained.
2980 case MIG_RP_MSG_PONG
:
2981 tmp32
= ldl_be_p(buf
);
2982 trace_source_return_path_thread_pong(tmp32
);
2985 case MIG_RP_MSG_REQ_PAGES
:
2986 start
= ldq_be_p(buf
);
2987 len
= ldl_be_p(buf
+ 8);
2988 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
2991 case MIG_RP_MSG_REQ_PAGES_ID
:
2992 expected_len
= 12 + 1; /* header + termination */
2994 if (header_len
>= expected_len
) {
2995 start
= ldq_be_p(buf
);
2996 len
= ldl_be_p(buf
+ 8);
2997 /* Now we expect an idstr */
2998 tmp32
= buf
[12]; /* Length of the following idstr */
2999 buf
[13 + tmp32
] = '\0';
3000 expected_len
+= tmp32
;
3002 if (header_len
!= expected_len
) {
3003 error_report("RP: Req_Page_id with length %d expecting %zd",
3004 header_len
, expected_len
);
3005 mark_source_rp_bad(ms
);
3008 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
3011 case MIG_RP_MSG_RECV_BITMAP
:
3012 if (header_len
< 1) {
3013 error_report("%s: missing block name", __func__
);
3014 mark_source_rp_bad(ms
);
3017 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
3018 buf
[buf
[0] + 1] = '\0';
3019 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
3020 mark_source_rp_bad(ms
);
3025 case MIG_RP_MSG_RESUME_ACK
:
3026 tmp32
= ldl_be_p(buf
);
3027 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
3028 mark_source_rp_bad(ms
);
3039 res
= qemu_file_get_error(rp
);
3041 if (res
&& migration_in_postcopy()) {
3043 * Maybe there is something we can do: it looks like a
3044 * network down issue, and we pause for a recovery.
3046 migration_release_dst_files(ms
);
3048 if (postcopy_pause_return_path_thread(ms
)) {
3050 * Reload rp, reset the rest. Referencing it is safe since
3051 * it's reset only by us above, or when migration completes
3053 rp
= ms
->rp_state
.from_dst_file
;
3054 ms
->rp_state
.error
= false;
3059 trace_source_return_path_thread_bad_end();
3060 mark_source_rp_bad(ms
);
3063 trace_source_return_path_thread_end();
3064 migration_release_dst_files(ms
);
3065 rcu_unregister_thread();
3069 static int open_return_path_on_source(MigrationState
*ms
,
3072 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
3073 if (!ms
->rp_state
.from_dst_file
) {
3077 trace_open_return_path_on_source();
3079 if (!create_thread
) {
3084 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
3085 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
3086 ms
->rp_state
.rp_thread_created
= true;
3088 trace_open_return_path_on_source_continue();
3093 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
3094 static int await_return_path_close_on_source(MigrationState
*ms
)
3097 * If this is a normal exit then the destination will send a SHUT and the
3098 * rp_thread will exit, however if there's an error we need to cause
3101 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
3103 * shutdown(2), if we have it, will cause it to unblock if it's stuck
3104 * waiting for the destination.
3106 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
3107 mark_source_rp_bad(ms
);
3109 trace_await_return_path_close_on_source_joining();
3110 qemu_thread_join(&ms
->rp_state
.rp_thread
);
3111 ms
->rp_state
.rp_thread_created
= false;
3112 trace_await_return_path_close_on_source_close();
3113 return ms
->rp_state
.error
;
3117 * Switch from normal iteration to postcopy
3118 * Returns non-0 on error
3120 static int postcopy_start(MigrationState
*ms
)
3123 QIOChannelBuffer
*bioc
;
3125 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3126 int64_t bandwidth
= migrate_max_postcopy_bandwidth();
3127 bool restart_block
= false;
3128 int cur_state
= MIGRATION_STATUS_ACTIVE
;
3130 if (postcopy_preempt_wait_channel(ms
)) {
3131 migrate_set_state(&ms
->state
, ms
->state
, MIGRATION_STATUS_FAILED
);
3135 if (!migrate_pause_before_switchover()) {
3136 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
3137 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3140 trace_postcopy_start();
3141 qemu_mutex_lock_iothread();
3142 trace_postcopy_start_set_run();
3144 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3145 global_state_store();
3146 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
3151 ret
= migration_maybe_pause(ms
, &cur_state
,
3152 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3157 ret
= bdrv_inactivate_all();
3161 restart_block
= true;
3164 * Cause any non-postcopiable, but iterative devices to
3165 * send out their final data.
3167 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
3170 * in Finish migrate and with the io-lock held everything should
3171 * be quiet, but we've potentially still got dirty pages and we
3172 * need to tell the destination to throw any pages it's already received
3175 if (migrate_postcopy_ram()) {
3176 ram_postcopy_send_discard_bitmap(ms
);
3180 * send rest of state - note things that are doing postcopy
3181 * will notice we're in POSTCOPY_ACTIVE and not actually
3182 * wrap their state up here
3184 /* 0 max-postcopy-bandwidth means unlimited */
3186 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
3188 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
3190 if (migrate_postcopy_ram()) {
3191 /* Ping just for debugging, helps line traces up */
3192 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
3196 * While loading the device state we may trigger page transfer
3197 * requests and the fd must be free to process those, and thus
3198 * the destination must read the whole device state off the fd before
3199 * it starts processing it. Unfortunately the ad-hoc migration format
3200 * doesn't allow the destination to know the size to read without fully
3201 * parsing it through each devices load-state code (especially the open
3202 * coded devices that use get/put).
3203 * So we wrap the device state up in a package with a length at the start;
3204 * to do this we use a qemu_buf to hold the whole of the device state.
3206 bioc
= qio_channel_buffer_new(4096);
3207 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
3208 fb
= qemu_file_new_output(QIO_CHANNEL(bioc
));
3209 object_unref(OBJECT(bioc
));
3212 * Make sure the receiver can get incoming pages before we send the rest
3215 qemu_savevm_send_postcopy_listen(fb
);
3217 qemu_savevm_state_complete_precopy(fb
, false, false);
3218 if (migrate_postcopy_ram()) {
3219 qemu_savevm_send_ping(fb
, 3);
3222 qemu_savevm_send_postcopy_run(fb
);
3224 /* <><> end of stuff going into the package */
3226 /* Last point of recovery; as soon as we send the package the destination
3227 * can open devices and potentially start running.
3228 * Lets just check again we've not got any errors.
3230 ret
= qemu_file_get_error(ms
->to_dst_file
);
3232 error_report("postcopy_start: Migration stream errored (pre package)");
3236 restart_block
= false;
3238 /* Now send that blob */
3239 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
3244 /* Send a notify to give a chance for anything that needs to happen
3245 * at the transition to postcopy and after the device state; in particular
3246 * spice needs to trigger a transition now
3248 ms
->postcopy_after_devices
= true;
3249 notifier_list_notify(&migration_state_notifiers
, ms
);
3251 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
3253 qemu_mutex_unlock_iothread();
3255 if (migrate_postcopy_ram()) {
3257 * Although this ping is just for debug, it could potentially be
3258 * used for getting a better measurement of downtime at the source.
3260 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
3263 if (migrate_release_ram()) {
3264 ram_postcopy_migrated_memory_release(ms
);
3267 ret
= qemu_file_get_error(ms
->to_dst_file
);
3269 error_report("postcopy_start: Migration stream errored");
3270 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
3271 MIGRATION_STATUS_FAILED
);
3274 trace_postcopy_preempt_enabled(migrate_postcopy_preempt());
3281 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
3282 MIGRATION_STATUS_FAILED
);
3283 if (restart_block
) {
3284 /* A failure happened early enough that we know the destination hasn't
3285 * accessed block devices, so we're safe to recover.
3287 Error
*local_err
= NULL
;
3289 bdrv_activate_all(&local_err
);
3291 error_report_err(local_err
);
3294 qemu_mutex_unlock_iothread();
3299 * migration_maybe_pause: Pause if required to by
3300 * migrate_pause_before_switchover called with the iothread locked
3301 * Returns: 0 on success
3303 static int migration_maybe_pause(MigrationState
*s
,
3304 int *current_active_state
,
3307 if (!migrate_pause_before_switchover()) {
3311 /* Since leaving this state is not atomic with posting the semaphore
3312 * it's possible that someone could have issued multiple migrate_continue
3313 * and the semaphore is incorrectly positive at this point;
3314 * the docs say it's undefined to reinit a semaphore that's already
3315 * init'd, so use timedwait to eat up any existing posts.
3317 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
3318 /* This block intentionally left blank */
3322 * If the migration is cancelled when it is in the completion phase,
3323 * the migration state is set to MIGRATION_STATUS_CANCELLING.
3324 * So we don't need to wait a semaphore, otherwise we would always
3325 * wait for the 'pause_sem' semaphore.
3327 if (s
->state
!= MIGRATION_STATUS_CANCELLING
) {
3328 qemu_mutex_unlock_iothread();
3329 migrate_set_state(&s
->state
, *current_active_state
,
3330 MIGRATION_STATUS_PRE_SWITCHOVER
);
3331 qemu_sem_wait(&s
->pause_sem
);
3332 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
3334 *current_active_state
= new_state
;
3335 qemu_mutex_lock_iothread();
3338 return s
->state
== new_state
? 0 : -EINVAL
;
3342 * migration_completion: Used by migration_thread when there's not much left.
3343 * The caller 'breaks' the loop when this returns.
3345 * @s: Current migration state
3347 static void migration_completion(MigrationState
*s
)
3350 int current_active_state
= s
->state
;
3352 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
3353 qemu_mutex_lock_iothread();
3354 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3355 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3356 s
->vm_was_running
= runstate_is_running();
3357 ret
= global_state_store();
3360 bool inactivate
= !migrate_colo_enabled();
3361 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
3362 trace_migration_completion_vm_stop(ret
);
3364 ret
= migration_maybe_pause(s
, ¤t_active_state
,
3365 MIGRATION_STATUS_DEVICE
);
3368 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
3369 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
3372 if (inactivate
&& ret
>= 0) {
3373 s
->block_inactive
= true;
3376 qemu_mutex_unlock_iothread();
3381 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3382 trace_migration_completion_postcopy_end();
3384 qemu_mutex_lock_iothread();
3385 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
3386 qemu_mutex_unlock_iothread();
3388 /* Shutdown the postcopy fast path thread */
3389 if (migrate_postcopy_preempt()) {
3390 postcopy_preempt_shutdown_file(s
);
3393 trace_migration_completion_postcopy_end_after_complete();
3399 * If rp was opened we must clean up the thread before
3400 * cleaning everything else up (since if there are no failures
3401 * it will wait for the destination to send it's status in
3404 if (s
->rp_state
.rp_thread_created
) {
3406 trace_migration_return_path_end_before();
3407 rp_error
= await_return_path_close_on_source(s
);
3408 trace_migration_return_path_end_after(rp_error
);
3410 goto fail_invalidate
;
3414 if (qemu_file_get_error(s
->to_dst_file
)) {
3415 trace_migration_completion_file_err();
3416 goto fail_invalidate
;
3419 if (migrate_colo_enabled() && s
->state
== MIGRATION_STATUS_ACTIVE
) {
3420 /* COLO does not support postcopy */
3421 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
3422 MIGRATION_STATUS_COLO
);
3424 migrate_set_state(&s
->state
, current_active_state
,
3425 MIGRATION_STATUS_COMPLETED
);
3431 /* If not doing postcopy, vm_start() will be called: let's regain
3432 * control on images.
3434 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
3435 s
->state
== MIGRATION_STATUS_DEVICE
) {
3436 Error
*local_err
= NULL
;
3438 qemu_mutex_lock_iothread();
3439 bdrv_activate_all(&local_err
);
3441 error_report_err(local_err
);
3443 s
->block_inactive
= false;
3445 qemu_mutex_unlock_iothread();
3449 migrate_set_state(&s
->state
, current_active_state
,
3450 MIGRATION_STATUS_FAILED
);
3454 * bg_migration_completion: Used by bg_migration_thread when after all the
3455 * RAM has been saved. The caller 'breaks' the loop when this returns.
3457 * @s: Current migration state
3459 static void bg_migration_completion(MigrationState
*s
)
3461 int current_active_state
= s
->state
;
3464 * Stop tracking RAM writes - un-protect memory, un-register UFFD
3465 * memory ranges, flush kernel wait queues and wake up threads
3466 * waiting for write fault to be resolved.
3468 ram_write_tracking_stop();
3470 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
3472 * By this moment we have RAM content saved into the migration stream.
3473 * The next step is to flush the non-RAM content (device state)
3474 * right after the ram content. The device state has been stored into
3475 * the temporary buffer before RAM saving started.
3477 qemu_put_buffer(s
->to_dst_file
, s
->bioc
->data
, s
->bioc
->usage
);
3478 qemu_fflush(s
->to_dst_file
);
3479 } else if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
3483 if (qemu_file_get_error(s
->to_dst_file
)) {
3484 trace_migration_completion_file_err();
3488 migrate_set_state(&s
->state
, current_active_state
,
3489 MIGRATION_STATUS_COMPLETED
);
3493 migrate_set_state(&s
->state
, current_active_state
,
3494 MIGRATION_STATUS_FAILED
);
3497 bool migrate_colo_enabled(void)
3499 MigrationState
*s
= migrate_get_current();
3500 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
3503 typedef enum MigThrError
{
3504 /* No error detected */
3505 MIG_THR_ERR_NONE
= 0,
3506 /* Detected error, but resumed successfully */
3507 MIG_THR_ERR_RECOVERED
= 1,
3508 /* Detected fatal error, need to exit */
3509 MIG_THR_ERR_FATAL
= 2,
3512 static int postcopy_resume_handshake(MigrationState
*s
)
3514 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
3516 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3517 qemu_sem_wait(&s
->rp_state
.rp_sem
);
3520 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3527 /* Return zero if success, or <0 for error */
3528 static int postcopy_do_resume(MigrationState
*s
)
3533 * Call all the resume_prepare() hooks, so that modules can be
3534 * ready for the migration resume.
3536 ret
= qemu_savevm_state_resume_prepare(s
);
3538 error_report("%s: resume_prepare() failure detected: %d",
3544 * Last handshake with destination on the resume (destination will
3545 * switch to postcopy-active afterwards)
3547 ret
= postcopy_resume_handshake(s
);
3549 error_report("%s: handshake failed: %d", __func__
, ret
);
3557 * We don't return until we are in a safe state to continue current
3558 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
3559 * MIG_THR_ERR_FATAL if unrecovery failure happened.
3561 static MigThrError
postcopy_pause(MigrationState
*s
)
3563 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3569 * Current channel is possibly broken. Release it. Note that this is
3570 * guaranteed even without lock because to_dst_file should only be
3571 * modified by the migration thread. That also guarantees that the
3572 * unregister of yank is safe too without the lock. It should be safe
3573 * even to be within the qemu_file_lock, but we didn't do that to avoid
3574 * taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make
3575 * the qemu_file_lock critical section as small as possible.
3577 assert(s
->to_dst_file
);
3578 migration_ioc_unregister_yank_from_file(s
->to_dst_file
);
3579 qemu_mutex_lock(&s
->qemu_file_lock
);
3580 file
= s
->to_dst_file
;
3581 s
->to_dst_file
= NULL
;
3582 qemu_mutex_unlock(&s
->qemu_file_lock
);
3584 qemu_file_shutdown(file
);
3587 migrate_set_state(&s
->state
, s
->state
,
3588 MIGRATION_STATUS_POSTCOPY_PAUSED
);
3590 error_report("Detected IO failure for postcopy. "
3591 "Migration paused.");
3594 * We wait until things fixed up. Then someone will setup the
3595 * status back for us.
3597 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
3598 qemu_sem_wait(&s
->postcopy_pause_sem
);
3601 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3602 /* Woken up by a recover procedure. Give it a shot */
3604 if (postcopy_preempt_wait_channel(s
)) {
3606 * Preempt enabled, and new channel create failed; loop
3607 * back to wait for another recovery.
3613 * Firstly, let's wake up the return path now, with a new
3614 * return path channel.
3616 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
3618 /* Do the resume logic */
3619 if (postcopy_do_resume(s
) == 0) {
3620 /* Let's continue! */
3621 trace_postcopy_pause_continued();
3622 return MIG_THR_ERR_RECOVERED
;
3625 * Something wrong happened during the recovery, let's
3626 * pause again. Pause is always better than throwing
3632 /* This is not right... Time to quit. */
3633 return MIG_THR_ERR_FATAL
;
3638 static MigThrError
migration_detect_error(MigrationState
*s
)
3641 int state
= s
->state
;
3642 Error
*local_error
= NULL
;
3644 if (state
== MIGRATION_STATUS_CANCELLING
||
3645 state
== MIGRATION_STATUS_CANCELLED
) {
3646 /* End the migration, but don't set the state to failed */
3647 return MIG_THR_ERR_FATAL
;
3651 * Try to detect any file errors. Note that postcopy_qemufile_src will
3652 * be NULL when postcopy preempt is not enabled.
3654 ret
= qemu_file_get_error_obj_any(s
->to_dst_file
,
3655 s
->postcopy_qemufile_src
,
3658 /* Everything is fine */
3659 assert(!local_error
);
3660 return MIG_THR_ERR_NONE
;
3664 migrate_set_error(s
, local_error
);
3665 error_free(local_error
);
3668 if (state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
) {
3670 * For postcopy, we allow the network to be down for a
3671 * while. After that, it can be continued by a
3674 return postcopy_pause(s
);
3677 * For precopy (or postcopy with error outside IO), we fail
3680 migrate_set_state(&s
->state
, state
, MIGRATION_STATUS_FAILED
);
3681 trace_migration_thread_file_err();
3683 /* Time to stop the migration, now. */
3684 return MIG_THR_ERR_FATAL
;
3688 /* How many bytes have we transferred since the beginning of the migration */
3689 static uint64_t migration_total_bytes(MigrationState
*s
)
3691 return qemu_file_total_transferred(s
->to_dst_file
) +
3692 ram_counters
.multifd_bytes
;
3695 static void migration_calculate_complete(MigrationState
*s
)
3697 uint64_t bytes
= migration_total_bytes(s
);
3698 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3699 int64_t transfer_time
;
3701 s
->total_time
= end_time
- s
->start_time
;
3704 * It's still not set, so we are precopy migration. For
3705 * postcopy, downtime is calculated during postcopy_start().
3707 s
->downtime
= end_time
- s
->downtime_start
;
3710 transfer_time
= s
->total_time
- s
->setup_time
;
3711 if (transfer_time
) {
3712 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
3716 static void update_iteration_initial_status(MigrationState
*s
)
3719 * Update these three fields at the same time to avoid mismatch info lead
3720 * wrong speed calculation.
3722 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3723 s
->iteration_initial_bytes
= migration_total_bytes(s
);
3724 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
3727 static void migration_update_counters(MigrationState
*s
,
3728 int64_t current_time
)
3730 uint64_t transferred
, transferred_pages
, time_spent
;
3731 uint64_t current_bytes
; /* bytes transferred since the beginning */
3734 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
3738 current_bytes
= migration_total_bytes(s
);
3739 transferred
= current_bytes
- s
->iteration_initial_bytes
;
3740 time_spent
= current_time
- s
->iteration_start_time
;
3741 bandwidth
= (double)transferred
/ time_spent
;
3742 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
3744 s
->mbps
= (((double) transferred
* 8.0) /
3745 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
3747 transferred_pages
= ram_get_total_transferred_pages() -
3748 s
->iteration_initial_pages
;
3749 s
->pages_per_second
= (double) transferred_pages
/
3750 (((double) time_spent
/ 1000.0));
3753 * if we haven't sent anything, we don't want to
3754 * recalculate. 10000 is a small enough number for our purposes
3756 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
3757 s
->expected_downtime
= ram_counters
.remaining
/ bandwidth
;
3760 qemu_file_reset_rate_limit(s
->to_dst_file
);
3762 update_iteration_initial_status(s
);
3764 trace_migrate_transferred(transferred
, time_spent
,
3765 bandwidth
, s
->threshold_size
);
3768 /* Migration thread iteration status */
3770 MIG_ITERATE_RESUME
, /* Resume current iteration */
3771 MIG_ITERATE_SKIP
, /* Skip current iteration */
3772 MIG_ITERATE_BREAK
, /* Break the loop */
3776 * Return true if continue to the next iteration directly, false
3779 static MigIterateState
migration_iteration_run(MigrationState
*s
)
3781 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
3782 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
3784 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
3785 &pend_compat
, &pend_post
);
3786 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
3788 trace_migrate_pending(pending_size
, s
->threshold_size
,
3789 pend_pre
, pend_compat
, pend_post
);
3791 if (pending_size
&& pending_size
>= s
->threshold_size
) {
3792 /* Still a significant amount to transfer */
3793 if (!in_postcopy
&& pend_pre
<= s
->threshold_size
&&
3794 qatomic_read(&s
->start_postcopy
)) {
3795 if (postcopy_start(s
)) {
3796 error_report("%s: postcopy failed to start", __func__
);
3798 return MIG_ITERATE_SKIP
;
3800 /* Just another iteration step */
3801 qemu_savevm_state_iterate(s
->to_dst_file
, in_postcopy
);
3803 trace_migration_thread_low_pending(pending_size
);
3804 migration_completion(s
);
3805 return MIG_ITERATE_BREAK
;
3808 return MIG_ITERATE_RESUME
;
3811 static void migration_iteration_finish(MigrationState
*s
)
3813 /* If we enabled cpu throttling for auto-converge, turn it off. */
3814 cpu_throttle_stop();
3816 qemu_mutex_lock_iothread();
3818 case MIGRATION_STATUS_COMPLETED
:
3819 migration_calculate_complete(s
);
3820 runstate_set(RUN_STATE_POSTMIGRATE
);
3822 case MIGRATION_STATUS_COLO
:
3823 if (!migrate_colo_enabled()) {
3824 error_report("%s: critical error: calling COLO code without "
3825 "COLO enabled", __func__
);
3827 migrate_start_colo_process(s
);
3828 s
->vm_was_running
= true;
3830 case MIGRATION_STATUS_FAILED
:
3831 case MIGRATION_STATUS_CANCELLED
:
3832 case MIGRATION_STATUS_CANCELLING
:
3833 if (s
->vm_was_running
) {
3834 if (!runstate_check(RUN_STATE_SHUTDOWN
)) {
3838 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
3839 runstate_set(RUN_STATE_POSTMIGRATE
);
3845 /* Should not reach here, but if so, forgive the VM. */
3846 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3849 migrate_fd_cleanup_schedule(s
);
3850 qemu_mutex_unlock_iothread();
3853 static void bg_migration_iteration_finish(MigrationState
*s
)
3855 qemu_mutex_lock_iothread();
3857 case MIGRATION_STATUS_COMPLETED
:
3858 migration_calculate_complete(s
);
3861 case MIGRATION_STATUS_ACTIVE
:
3862 case MIGRATION_STATUS_FAILED
:
3863 case MIGRATION_STATUS_CANCELLED
:
3864 case MIGRATION_STATUS_CANCELLING
:
3868 /* Should not reach here, but if so, forgive the VM. */
3869 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3873 migrate_fd_cleanup_schedule(s
);
3874 qemu_mutex_unlock_iothread();
3878 * Return true if continue to the next iteration directly, false
3881 static MigIterateState
bg_migration_iteration_run(MigrationState
*s
)
3885 res
= qemu_savevm_state_iterate(s
->to_dst_file
, false);
3887 bg_migration_completion(s
);
3888 return MIG_ITERATE_BREAK
;
3891 return MIG_ITERATE_RESUME
;
3894 void migration_make_urgent_request(void)
3896 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
3899 void migration_consume_urgent_request(void)
3901 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
3904 /* Returns true if the rate limiting was broken by an urgent request */
3905 bool migration_rate_limit(void)
3907 int64_t now
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3908 MigrationState
*s
= migrate_get_current();
3910 bool urgent
= false;
3911 migration_update_counters(s
, now
);
3912 if (qemu_file_rate_limit(s
->to_dst_file
)) {
3914 if (qemu_file_get_error(s
->to_dst_file
)) {
3918 * Wait for a delay to do rate limiting OR
3919 * something urgent to post the semaphore.
3921 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- now
;
3922 trace_migration_rate_limit_pre(ms
);
3923 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
3925 * We were woken by one or more urgent things but
3926 * the timedwait will have consumed one of them.
3927 * The service routine for the urgent wake will dec
3928 * the semaphore itself for each item it consumes,
3929 * so add this one we just eat back.
3931 qemu_sem_post(&s
->rate_limit_sem
);
3934 trace_migration_rate_limit_post(urgent
);
3940 * if failover devices are present, wait they are completely
3944 static void qemu_savevm_wait_unplug(MigrationState
*s
, int old_state
,
3947 if (qemu_savevm_state_guest_unplug_pending()) {
3948 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_WAIT_UNPLUG
);
3950 while (s
->state
== MIGRATION_STATUS_WAIT_UNPLUG
&&
3951 qemu_savevm_state_guest_unplug_pending()) {
3952 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
3954 if (s
->state
!= MIGRATION_STATUS_WAIT_UNPLUG
) {
3955 int timeout
= 120; /* 30 seconds */
3957 * migration has been canceled
3958 * but as we have started an unplug we must wait the end
3959 * to be able to plug back the card
3961 while (timeout
-- && qemu_savevm_state_guest_unplug_pending()) {
3962 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
3964 if (qemu_savevm_state_guest_unplug_pending() &&
3966 warn_report("migration: partially unplugged device on "
3971 migrate_set_state(&s
->state
, MIGRATION_STATUS_WAIT_UNPLUG
, new_state
);
3973 migrate_set_state(&s
->state
, old_state
, new_state
);
3978 * Master migration thread on the source VM.
3979 * It drives the migration and pumps the data down the outgoing channel.
3981 static void *migration_thread(void *opaque
)
3983 MigrationState
*s
= opaque
;
3984 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3985 MigThrError thr_error
;
3986 bool urgent
= false;
3988 rcu_register_thread();
3990 object_ref(OBJECT(s
));
3991 update_iteration_initial_status(s
);
3993 qemu_savevm_state_header(s
->to_dst_file
);
3996 * If we opened the return path, we need to make sure dst has it
3999 if (s
->rp_state
.rp_thread_created
) {
4000 /* Now tell the dest that it should open its end so it can reply */
4001 qemu_savevm_send_open_return_path(s
->to_dst_file
);
4003 /* And do a ping that will make stuff easier to debug */
4004 qemu_savevm_send_ping(s
->to_dst_file
, 1);
4007 if (migrate_postcopy()) {
4009 * Tell the destination that we *might* want to do postcopy later;
4010 * if the other end can't do postcopy it should fail now, nice and
4013 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
4016 if (migrate_colo_enabled()) {
4017 /* Notify migration destination that we enable COLO */
4018 qemu_savevm_send_colo_enable(s
->to_dst_file
);
4021 qemu_savevm_state_setup(s
->to_dst_file
);
4023 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
4024 MIGRATION_STATUS_ACTIVE
);
4026 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
4028 trace_migration_thread_setup_complete();
4030 while (migration_is_active(s
)) {
4031 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
4032 MigIterateState iter_state
= migration_iteration_run(s
);
4033 if (iter_state
== MIG_ITERATE_SKIP
) {
4035 } else if (iter_state
== MIG_ITERATE_BREAK
) {
4041 * Try to detect any kind of failures, and see whether we
4042 * should stop the migration now.
4044 thr_error
= migration_detect_error(s
);
4045 if (thr_error
== MIG_THR_ERR_FATAL
) {
4046 /* Stop migration */
4048 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
4050 * Just recovered from a e.g. network failure, reset all
4051 * the local variables. This is important to avoid
4052 * breaking transferred_bytes and bandwidth calculation
4054 update_iteration_initial_status(s
);
4057 urgent
= migration_rate_limit();
4060 trace_migration_thread_after_loop();
4061 migration_iteration_finish(s
);
4062 object_unref(OBJECT(s
));
4063 rcu_unregister_thread();
4067 static void bg_migration_vm_start_bh(void *opaque
)
4069 MigrationState
*s
= opaque
;
4071 qemu_bh_delete(s
->vm_start_bh
);
4072 s
->vm_start_bh
= NULL
;
4075 s
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - s
->downtime_start
;
4079 * Background snapshot thread, based on live migration code.
4080 * This is an alternative implementation of live migration mechanism
4081 * introduced specifically to support background snapshots.
4083 * It takes advantage of userfault_fd write protection mechanism introduced
4084 * in v5.7 kernel. Compared to existing dirty page logging migration much
4085 * lesser stream traffic is produced resulting in smaller snapshot images,
4086 * simply cause of no page duplicates can get into the stream.
4088 * Another key point is that generated vmstate stream reflects machine state
4089 * 'frozen' at the beginning of snapshot creation compared to dirty page logging
4090 * mechanism, which effectively results in that saved snapshot is the state of VM
4091 * at the end of the process.
4093 static void *bg_migration_thread(void *opaque
)
4095 MigrationState
*s
= opaque
;
4096 int64_t setup_start
;
4097 MigThrError thr_error
;
4099 bool early_fail
= true;
4101 rcu_register_thread();
4102 object_ref(OBJECT(s
));
4104 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
4106 setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
4108 * We want to save vmstate for the moment when migration has been
4109 * initiated but also we want to save RAM content while VM is running.
4110 * The RAM content should appear first in the vmstate. So, we first
4111 * stash the non-RAM part of the vmstate to the temporary buffer,
4112 * then write RAM part of the vmstate to the migration stream
4113 * with vCPUs running and, finally, write stashed non-RAM part of
4114 * the vmstate from the buffer to the migration stream.
4116 s
->bioc
= qio_channel_buffer_new(512 * 1024);
4117 qio_channel_set_name(QIO_CHANNEL(s
->bioc
), "vmstate-buffer");
4118 fb
= qemu_file_new_output(QIO_CHANNEL(s
->bioc
));
4119 object_unref(OBJECT(s
->bioc
));
4121 update_iteration_initial_status(s
);
4124 * Prepare for tracking memory writes with UFFD-WP - populate
4125 * RAM pages before protecting.
4128 ram_write_tracking_prepare();
4131 qemu_savevm_state_header(s
->to_dst_file
);
4132 qemu_savevm_state_setup(s
->to_dst_file
);
4134 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
4135 MIGRATION_STATUS_ACTIVE
);
4137 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
4139 trace_migration_thread_setup_complete();
4140 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
4142 qemu_mutex_lock_iothread();
4145 * If VM is currently in suspended state, then, to make a valid runstate
4146 * transition in vm_stop_force_state() we need to wakeup it up.
4148 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
4149 s
->vm_was_running
= runstate_is_running();
4151 if (global_state_store()) {
4154 /* Forcibly stop VM before saving state of vCPUs and devices */
4155 if (vm_stop_force_state(RUN_STATE_PAUSED
)) {
4159 * Put vCPUs in sync with shadow context structures, then
4160 * save their state to channel-buffer along with devices.
4162 cpu_synchronize_all_states();
4163 if (qemu_savevm_state_complete_precopy_non_iterable(fb
, false, false)) {
4167 * Since we are going to get non-iterable state data directly
4168 * from s->bioc->data, explicit flush is needed here.
4172 /* Now initialize UFFD context and start tracking RAM writes */
4173 if (ram_write_tracking_start()) {
4179 * Start VM from BH handler to avoid write-fault lock here.
4180 * UFFD-WP protection for the whole RAM is already enabled so
4181 * calling VM state change notifiers from vm_start() would initiate
4182 * writes to virtio VQs memory which is in write-protected region.
4184 s
->vm_start_bh
= qemu_bh_new(bg_migration_vm_start_bh
, s
);
4185 qemu_bh_schedule(s
->vm_start_bh
);
4187 qemu_mutex_unlock_iothread();
4189 while (migration_is_active(s
)) {
4190 MigIterateState iter_state
= bg_migration_iteration_run(s
);
4191 if (iter_state
== MIG_ITERATE_SKIP
) {
4193 } else if (iter_state
== MIG_ITERATE_BREAK
) {
4198 * Try to detect any kind of failures, and see whether we
4199 * should stop the migration now.
4201 thr_error
= migration_detect_error(s
);
4202 if (thr_error
== MIG_THR_ERR_FATAL
) {
4203 /* Stop migration */
4207 migration_update_counters(s
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
));
4210 trace_migration_thread_after_loop();
4214 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
4215 MIGRATION_STATUS_FAILED
);
4216 qemu_mutex_unlock_iothread();
4219 bg_migration_iteration_finish(s
);
4222 object_unref(OBJECT(s
));
4223 rcu_unregister_thread();
4228 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
4230 Error
*local_err
= NULL
;
4232 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
4235 * If there's a previous error, free it and prepare for another one.
4236 * Meanwhile if migration completes successfully, there won't have an error
4237 * dumped when calling migrate_fd_cleanup().
4239 migrate_error_free(s
);
4241 s
->expected_downtime
= s
->parameters
.downtime_limit
;
4243 assert(s
->cleanup_bh
);
4245 assert(!s
->cleanup_bh
);
4246 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup_bh
, s
);
4249 migrate_fd_error(s
, error_in
);
4252 * Don't do cleanup for resume if channel is invalid, but only dump
4253 * the error. We wait for another channel connect from the user.
4254 * The error_report still gives HMP user a hint on what failed.
4255 * It's normally done in migrate_fd_cleanup(), but call it here
4258 error_report_err(error_copy(s
->error
));
4260 migrate_fd_cleanup(s
);
4266 /* This is a resumed migration */
4267 rate_limit
= s
->parameters
.max_postcopy_bandwidth
/
4270 /* This is a fresh new migration */
4271 rate_limit
= s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
;
4273 /* Notify before starting migration thread */
4274 notifier_list_notify(&migration_state_notifiers
, s
);
4277 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
4278 qemu_file_set_blocking(s
->to_dst_file
, true);
4281 * Open the return path. For postcopy, it is used exclusively. For
4282 * precopy, only if user specified "return-path" capability would
4283 * QEMU uses the return path.
4285 if (migrate_postcopy_ram() || migrate_use_return_path()) {
4286 if (open_return_path_on_source(s
, !resume
)) {
4287 error_report("Unable to open return-path for postcopy");
4288 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
4289 migrate_fd_cleanup(s
);
4294 /* This needs to be done before resuming a postcopy */
4295 if (postcopy_preempt_setup(s
, &local_err
)) {
4296 error_report_err(local_err
);
4297 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
4298 MIGRATION_STATUS_FAILED
);
4299 migrate_fd_cleanup(s
);
4304 /* Wakeup the main migration thread to do the recovery */
4305 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
4306 MIGRATION_STATUS_POSTCOPY_RECOVER
);
4307 qemu_sem_post(&s
->postcopy_pause_sem
);
4311 if (multifd_save_setup(&local_err
) != 0) {
4312 error_report_err(local_err
);
4313 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
4314 MIGRATION_STATUS_FAILED
);
4315 migrate_fd_cleanup(s
);
4319 if (migrate_background_snapshot()) {
4320 qemu_thread_create(&s
->thread
, "bg_snapshot",
4321 bg_migration_thread
, s
, QEMU_THREAD_JOINABLE
);
4323 qemu_thread_create(&s
->thread
, "live_migration",
4324 migration_thread
, s
, QEMU_THREAD_JOINABLE
);
4326 s
->migration_thread_running
= true;
4329 void migration_global_dump(Monitor
*mon
)
4331 MigrationState
*ms
= migrate_get_current();
4333 monitor_printf(mon
, "globals:\n");
4334 monitor_printf(mon
, "store-global-state: %s\n",
4335 ms
->store_global_state
? "on" : "off");
4336 monitor_printf(mon
, "only-migratable: %s\n",
4337 only_migratable
? "on" : "off");
4338 monitor_printf(mon
, "send-configuration: %s\n",
4339 ms
->send_configuration
? "on" : "off");
4340 monitor_printf(mon
, "send-section-footer: %s\n",
4341 ms
->send_section_footer
? "on" : "off");
4342 monitor_printf(mon
, "decompress-error-check: %s\n",
4343 ms
->decompress_error_check
? "on" : "off");
4344 monitor_printf(mon
, "clear-bitmap-shift: %u\n",
4345 ms
->clear_bitmap_shift
);
4348 #define DEFINE_PROP_MIG_CAP(name, x) \
4349 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
4351 static Property migration_properties
[] = {
4352 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
4353 store_global_state
, true),
4354 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
4355 send_configuration
, true),
4356 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
4357 send_section_footer
, true),
4358 DEFINE_PROP_BOOL("decompress-error-check", MigrationState
,
4359 decompress_error_check
, true),
4360 DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState
,
4361 clear_bitmap_shift
, CLEAR_BITMAP_SHIFT_DEFAULT
),
4363 /* Migration parameters */
4364 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
4365 parameters
.compress_level
,
4366 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
4367 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
4368 parameters
.compress_threads
,
4369 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
4370 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState
,
4371 parameters
.compress_wait_thread
, true),
4372 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
4373 parameters
.decompress_threads
,
4374 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
4375 DEFINE_PROP_UINT8("x-throttle-trigger-threshold", MigrationState
,
4376 parameters
.throttle_trigger_threshold
,
4377 DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD
),
4378 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
4379 parameters
.cpu_throttle_initial
,
4380 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
4381 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
4382 parameters
.cpu_throttle_increment
,
4383 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
4384 DEFINE_PROP_BOOL("x-cpu-throttle-tailslow", MigrationState
,
4385 parameters
.cpu_throttle_tailslow
, false),
4386 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
4387 parameters
.max_bandwidth
, MAX_THROTTLE
),
4388 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
4389 parameters
.downtime_limit
,
4390 DEFAULT_MIGRATE_SET_DOWNTIME
),
4391 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
4392 parameters
.x_checkpoint_delay
,
4393 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
4394 DEFINE_PROP_UINT8("multifd-channels", MigrationState
,
4395 parameters
.multifd_channels
,
4396 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
4397 DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState
,
4398 parameters
.multifd_compression
,
4399 DEFAULT_MIGRATE_MULTIFD_COMPRESSION
),
4400 DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState
,
4401 parameters
.multifd_zlib_level
,
4402 DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL
),
4403 DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState
,
4404 parameters
.multifd_zstd_level
,
4405 DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL
),
4406 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
4407 parameters
.xbzrle_cache_size
,
4408 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
4409 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState
,
4410 parameters
.max_postcopy_bandwidth
,
4411 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH
),
4412 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState
,
4413 parameters
.max_cpu_throttle
,
4414 DEFAULT_MIGRATE_MAX_CPU_THROTTLE
),
4415 DEFINE_PROP_SIZE("announce-initial", MigrationState
,
4416 parameters
.announce_initial
,
4417 DEFAULT_MIGRATE_ANNOUNCE_INITIAL
),
4418 DEFINE_PROP_SIZE("announce-max", MigrationState
,
4419 parameters
.announce_max
,
4420 DEFAULT_MIGRATE_ANNOUNCE_MAX
),
4421 DEFINE_PROP_SIZE("announce-rounds", MigrationState
,
4422 parameters
.announce_rounds
,
4423 DEFAULT_MIGRATE_ANNOUNCE_ROUNDS
),
4424 DEFINE_PROP_SIZE("announce-step", MigrationState
,
4425 parameters
.announce_step
,
4426 DEFAULT_MIGRATE_ANNOUNCE_STEP
),
4427 DEFINE_PROP_STRING("tls-creds", MigrationState
, parameters
.tls_creds
),
4428 DEFINE_PROP_STRING("tls-hostname", MigrationState
, parameters
.tls_hostname
),
4429 DEFINE_PROP_STRING("tls-authz", MigrationState
, parameters
.tls_authz
),
4431 /* Migration capabilities */
4432 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
4433 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
4434 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
4435 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
4436 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
4437 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
4438 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
4439 DEFINE_PROP_MIG_CAP("x-postcopy-preempt",
4440 MIGRATION_CAPABILITY_POSTCOPY_PREEMPT
),
4441 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
4442 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
4443 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
4444 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
4445 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD
),
4446 DEFINE_PROP_MIG_CAP("x-background-snapshot",
4447 MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
),
4449 DEFINE_PROP_MIG_CAP("x-zero-copy-send",
4450 MIGRATION_CAPABILITY_ZERO_COPY_SEND
),
4453 DEFINE_PROP_END_OF_LIST(),
4456 static void migration_class_init(ObjectClass
*klass
, void *data
)
4458 DeviceClass
*dc
= DEVICE_CLASS(klass
);
4460 dc
->user_creatable
= false;
4461 device_class_set_props(dc
, migration_properties
);
4464 static void migration_instance_finalize(Object
*obj
)
4466 MigrationState
*ms
= MIGRATION_OBJ(obj
);
4468 qemu_mutex_destroy(&ms
->error_mutex
);
4469 qemu_mutex_destroy(&ms
->qemu_file_lock
);
4470 qemu_sem_destroy(&ms
->wait_unplug_sem
);
4471 qemu_sem_destroy(&ms
->rate_limit_sem
);
4472 qemu_sem_destroy(&ms
->pause_sem
);
4473 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
4474 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
4475 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
4476 qemu_sem_destroy(&ms
->postcopy_qemufile_src_sem
);
4477 error_free(ms
->error
);
4480 static void migration_instance_init(Object
*obj
)
4482 MigrationState
*ms
= MIGRATION_OBJ(obj
);
4483 MigrationParameters
*params
= &ms
->parameters
;
4485 ms
->state
= MIGRATION_STATUS_NONE
;
4487 ms
->pages_per_second
= -1;
4488 qemu_sem_init(&ms
->pause_sem
, 0);
4489 qemu_mutex_init(&ms
->error_mutex
);
4491 params
->tls_hostname
= g_strdup("");
4492 params
->tls_creds
= g_strdup("");
4494 /* Set has_* up only for parameter checks */
4495 params
->has_compress_level
= true;
4496 params
->has_compress_threads
= true;
4497 params
->has_compress_wait_thread
= true;
4498 params
->has_decompress_threads
= true;
4499 params
->has_throttle_trigger_threshold
= true;
4500 params
->has_cpu_throttle_initial
= true;
4501 params
->has_cpu_throttle_increment
= true;
4502 params
->has_cpu_throttle_tailslow
= true;
4503 params
->has_max_bandwidth
= true;
4504 params
->has_downtime_limit
= true;
4505 params
->has_x_checkpoint_delay
= true;
4506 params
->has_block_incremental
= true;
4507 params
->has_multifd_channels
= true;
4508 params
->has_multifd_compression
= true;
4509 params
->has_multifd_zlib_level
= true;
4510 params
->has_multifd_zstd_level
= true;
4511 params
->has_xbzrle_cache_size
= true;
4512 params
->has_max_postcopy_bandwidth
= true;
4513 params
->has_max_cpu_throttle
= true;
4514 params
->has_announce_initial
= true;
4515 params
->has_announce_max
= true;
4516 params
->has_announce_rounds
= true;
4517 params
->has_announce_step
= true;
4519 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
4520 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
4521 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
4522 qemu_sem_init(&ms
->rate_limit_sem
, 0);
4523 qemu_sem_init(&ms
->wait_unplug_sem
, 0);
4524 qemu_sem_init(&ms
->postcopy_qemufile_src_sem
, 0);
4525 qemu_mutex_init(&ms
->qemu_file_lock
);
4529 * Return true if check pass, false otherwise. Error will be put
4530 * inside errp if provided.
4532 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
4534 MigrationCapabilityStatusList
*head
= NULL
;
4535 /* Assuming all off */
4536 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
4539 if (!migrate_params_check(&ms
->parameters
, errp
)) {
4543 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
4544 if (ms
->enabled_capabilities
[i
]) {
4545 QAPI_LIST_PREPEND(head
, migrate_cap_add(i
, true));
4549 ret
= migrate_caps_check(cap_list
, head
, errp
);
4551 /* It works with head == NULL */
4552 qapi_free_MigrationCapabilityStatusList(head
);
4557 static const TypeInfo migration_type
= {
4558 .name
= TYPE_MIGRATION
,
4560 * NOTE: TYPE_MIGRATION is not really a device, as the object is
4561 * not created using qdev_new(), it is not attached to the qdev
4562 * device tree, and it is never realized.
4564 * TODO: Make this TYPE_OBJECT once QOM provides something like
4565 * TYPE_DEVICE's "-global" properties.
4567 .parent
= TYPE_DEVICE
,
4568 .class_init
= migration_class_init
,
4569 .class_size
= sizeof(MigrationClass
),
4570 .instance_size
= sizeof(MigrationState
),
4571 .instance_init
= migration_instance_init
,
4572 .instance_finalize
= migration_instance_finalize
,
4575 static void register_migration_types(void)
4577 type_register_static(&migration_type
);
4580 type_init(register_migration_types
);