4 * Copyright IBM, Corp. 2008
7 * Anthony Liguori <aliguori@us.ibm.com>
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
16 #include "qemu/osdep.h"
17 #include "qemu/cutils.h"
18 #include "qemu/error-report.h"
19 #include "qemu/main-loop.h"
20 #include "migration/blocker.h"
24 #include "sysemu/runstate.h"
25 #include "sysemu/sysemu.h"
26 #include "sysemu/cpu-throttle.h"
29 #include "migration/global_state.h"
30 #include "migration/misc.h"
31 #include "migration.h"
33 #include "qemu-file.h"
35 #include "migration/vmstate.h"
36 #include "block/block.h"
37 #include "qapi/error.h"
38 #include "qapi/clone-visitor.h"
39 #include "qapi/qapi-visit-migration.h"
40 #include "qapi/qapi-visit-sockets.h"
41 #include "qapi/qapi-commands-migration.h"
42 #include "qapi/qapi-events-migration.h"
43 #include "qapi/qmp/qerror.h"
44 #include "qapi/qmp/qnull.h"
47 #include "postcopy-ram.h"
48 #include "qemu/thread.h"
50 #include "exec/target_page.h"
51 #include "io/channel-buffer.h"
52 #include "io/channel-tls.h"
53 #include "migration/colo.h"
54 #include "hw/boards.h"
55 #include "hw/qdev-properties.h"
56 #include "hw/qdev-properties-system.h"
57 #include "monitor/monitor.h"
58 #include "net/announce.h"
59 #include "qemu/queue.h"
61 #include "threadinfo.h"
62 #include "qemu/yank.h"
63 #include "sysemu/cpus.h"
64 #include "yank_functions.h"
65 #include "sysemu/qtest.h"
66 #include "ui/qemu-spice.h"
68 #define MAX_THROTTLE (128 << 20) /* Migration transfer speed throttling */
70 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
72 #define BUFFER_DELAY 100
73 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
75 /* Time in milliseconds we are allowed to stop the source,
76 * for sending the last part */
77 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
79 /* Maximum migrate downtime set to 2000 seconds */
80 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
81 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
83 /* Default compression thread count */
84 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
85 /* Default decompression thread count, usually decompression is at
86 * least 4 times as fast as compression.*/
87 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
88 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
89 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
90 /* Define default autoconverge cpu throttle migration parameters */
91 #define DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD 50
92 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
93 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
94 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
96 /* Migration XBZRLE default cache size */
97 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
99 /* The delay time (in ms) between two COLO checkpoints */
100 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
101 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
102 #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE
103 /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */
104 #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
105 /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
106 #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
108 /* Background transfer rate for postcopy, 0 means unlimited, note
109 * that page requests can still exceed this limit.
111 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
114 * Parameters for self_announce_delay giving a stream of RARP/ARP
115 * packets after migration.
117 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
118 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
119 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
120 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
122 static NotifierList migration_state_notifiers
=
123 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
125 /* Messages sent on the return path from destination to source */
126 enum mig_rp_message_type
{
127 MIG_RP_MSG_INVALID
= 0, /* Must be 0 */
128 MIG_RP_MSG_SHUT
, /* sibling will not send any more RP messages */
129 MIG_RP_MSG_PONG
, /* Response to a PING; data (seq: be32 ) */
131 MIG_RP_MSG_REQ_PAGES_ID
, /* data (start: be64, len: be32, id: string) */
132 MIG_RP_MSG_REQ_PAGES
, /* data (start: be64, len: be32) */
133 MIG_RP_MSG_RECV_BITMAP
, /* send recved_bitmap back to source */
134 MIG_RP_MSG_RESUME_ACK
, /* tell source that we are ready to resume */
139 /* Migration capabilities set */
140 struct MigrateCapsSet
{
141 int size
; /* Capability set size */
142 MigrationCapability caps
[]; /* Variadic array of capabilities */
144 typedef struct MigrateCapsSet MigrateCapsSet
;
146 /* Define and initialize MigrateCapsSet */
147 #define INITIALIZE_MIGRATE_CAPS_SET(_name, ...) \
148 MigrateCapsSet _name = { \
149 .size = sizeof((int []) { __VA_ARGS__ }) / sizeof(int), \
150 .caps = { __VA_ARGS__ } \
153 /* Background-snapshot compatibility check list */
155 INITIALIZE_MIGRATE_CAPS_SET(check_caps_background_snapshot
,
156 MIGRATION_CAPABILITY_POSTCOPY_RAM
,
157 MIGRATION_CAPABILITY_DIRTY_BITMAPS
,
158 MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
,
159 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE
,
160 MIGRATION_CAPABILITY_RETURN_PATH
,
161 MIGRATION_CAPABILITY_MULTIFD
,
162 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
,
163 MIGRATION_CAPABILITY_AUTO_CONVERGE
,
164 MIGRATION_CAPABILITY_RELEASE_RAM
,
165 MIGRATION_CAPABILITY_RDMA_PIN_ALL
,
166 MIGRATION_CAPABILITY_COMPRESS
,
167 MIGRATION_CAPABILITY_XBZRLE
,
168 MIGRATION_CAPABILITY_X_COLO
,
169 MIGRATION_CAPABILITY_VALIDATE_UUID
,
170 MIGRATION_CAPABILITY_ZERO_COPY_SEND
);
172 /* When we add fault tolerance, we could have several
173 migrations at once. For now we don't need to add
174 dynamic creation of migration */
176 static MigrationState
*current_migration
;
177 static MigrationIncomingState
*current_incoming
;
179 static GSList
*migration_blockers
;
181 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
182 static int migration_maybe_pause(MigrationState
*s
,
183 int *current_active_state
,
185 static void migrate_fd_cancel(MigrationState
*s
);
187 static bool migration_needs_multiple_sockets(void)
189 return migrate_use_multifd() || migrate_postcopy_preempt();
192 static bool uri_supports_multi_channels(const char *uri
)
194 return strstart(uri
, "tcp:", NULL
) || strstart(uri
, "unix:", NULL
) ||
195 strstart(uri
, "vsock:", NULL
);
199 migration_channels_and_uri_compatible(const char *uri
, Error
**errp
)
201 if (migration_needs_multiple_sockets() &&
202 !uri_supports_multi_channels(uri
)) {
203 error_setg(errp
, "Migration requires multi-channel URIs (e.g. tcp)");
210 static gint
page_request_addr_cmp(gconstpointer ap
, gconstpointer bp
)
212 uintptr_t a
= (uintptr_t) ap
, b
= (uintptr_t) bp
;
214 return (a
> b
) - (a
< b
);
217 void migration_object_init(void)
219 /* This can only be called once. */
220 assert(!current_migration
);
221 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
224 * Init the migrate incoming object as well no matter whether
225 * we'll use it or not.
227 assert(!current_incoming
);
228 current_incoming
= g_new0(MigrationIncomingState
, 1);
229 current_incoming
->state
= MIGRATION_STATUS_NONE
;
230 current_incoming
->postcopy_remote_fds
=
231 g_array_new(FALSE
, TRUE
, sizeof(struct PostCopyFD
));
232 qemu_mutex_init(¤t_incoming
->rp_mutex
);
233 qemu_mutex_init(¤t_incoming
->postcopy_prio_thread_mutex
);
234 qemu_event_init(¤t_incoming
->main_thread_load_event
, false);
235 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_dst
, 0);
236 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fault
, 0);
237 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fast_load
, 0);
238 qemu_sem_init(¤t_incoming
->postcopy_qemufile_dst_done
, 0);
240 qemu_mutex_init(¤t_incoming
->page_request_mutex
);
241 current_incoming
->page_requested
= g_tree_new(page_request_addr_cmp
);
243 migration_object_check(current_migration
, &error_fatal
);
247 dirty_bitmap_mig_init();
250 void migration_cancel(const Error
*error
)
253 migrate_set_error(current_migration
, error
);
255 migrate_fd_cancel(current_migration
);
258 void migration_shutdown(void)
261 * When the QEMU main thread exit, the COLO thread
262 * may wait a semaphore. So, we should wakeup the
263 * COLO thread before migration shutdown.
267 * Cancel the current migration - that will (eventually)
268 * stop the migration using this structure
270 migration_cancel(NULL
);
271 object_unref(OBJECT(current_migration
));
274 * Cancel outgoing migration of dirty bitmaps. It should
275 * at least unref used block nodes.
277 dirty_bitmap_mig_cancel_outgoing();
280 * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
281 * are non-critical data, and their loss never considered as
284 dirty_bitmap_mig_cancel_incoming();
288 MigrationState
*migrate_get_current(void)
290 /* This can only be called after the object created. */
291 assert(current_migration
);
292 return current_migration
;
295 MigrationIncomingState
*migration_incoming_get_current(void)
297 assert(current_incoming
);
298 return current_incoming
;
301 void migration_incoming_transport_cleanup(MigrationIncomingState
*mis
)
303 if (mis
->socket_address_list
) {
304 qapi_free_SocketAddressList(mis
->socket_address_list
);
305 mis
->socket_address_list
= NULL
;
308 if (mis
->transport_cleanup
) {
309 mis
->transport_cleanup(mis
->transport_data
);
310 mis
->transport_data
= mis
->transport_cleanup
= NULL
;
314 void migration_incoming_state_destroy(void)
316 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
318 multifd_load_cleanup();
320 if (mis
->to_src_file
) {
321 /* Tell source that we are done */
322 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
323 qemu_fclose(mis
->to_src_file
);
324 mis
->to_src_file
= NULL
;
327 if (mis
->from_src_file
) {
328 migration_ioc_unregister_yank_from_file(mis
->from_src_file
);
329 qemu_fclose(mis
->from_src_file
);
330 mis
->from_src_file
= NULL
;
332 if (mis
->postcopy_remote_fds
) {
333 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
334 mis
->postcopy_remote_fds
= NULL
;
337 migration_incoming_transport_cleanup(mis
);
338 qemu_event_reset(&mis
->main_thread_load_event
);
340 if (mis
->page_requested
) {
341 g_tree_destroy(mis
->page_requested
);
342 mis
->page_requested
= NULL
;
345 if (mis
->postcopy_qemufile_dst
) {
346 migration_ioc_unregister_yank_from_file(mis
->postcopy_qemufile_dst
);
347 qemu_fclose(mis
->postcopy_qemufile_dst
);
348 mis
->postcopy_qemufile_dst
= NULL
;
351 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
354 static void migrate_generate_event(int new_state
)
356 if (migrate_use_events()) {
357 qapi_event_send_migration(new_state
);
361 static bool migrate_late_block_activate(void)
365 s
= migrate_get_current();
367 return s
->enabled_capabilities
[
368 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE
];
372 * Send a message on the return channel back to the source
375 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
376 enum mig_rp_message_type message_type
,
377 uint16_t len
, void *data
)
381 trace_migrate_send_rp_message((int)message_type
, len
);
382 QEMU_LOCK_GUARD(&mis
->rp_mutex
);
385 * It's possible that the file handle got lost due to network
388 if (!mis
->to_src_file
) {
393 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
394 qemu_put_be16(mis
->to_src_file
, len
);
395 qemu_put_buffer(mis
->to_src_file
, data
, len
);
396 qemu_fflush(mis
->to_src_file
);
398 /* It's possible that qemu file got error during sending */
399 ret
= qemu_file_get_error(mis
->to_src_file
);
404 /* Request one page from the source VM at the given start address.
405 * rb: the RAMBlock to request the page in
406 * Start: Address offset within the RB
407 * Len: Length in bytes required - must be a multiple of pagesize
409 int migrate_send_rp_message_req_pages(MigrationIncomingState
*mis
,
410 RAMBlock
*rb
, ram_addr_t start
)
412 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
413 size_t msglen
= 12; /* start + len */
414 size_t len
= qemu_ram_pagesize(rb
);
415 enum mig_rp_message_type msg_type
;
419 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
420 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
423 * We maintain the last ramblock that we requested for page. Note that we
424 * don't need locking because this function will only be called within the
425 * postcopy ram fault thread.
427 if (rb
!= mis
->last_rb
) {
430 rbname
= qemu_ram_get_idstr(rb
);
431 rbname_len
= strlen(rbname
);
433 assert(rbname_len
< 256);
435 bufc
[msglen
++] = rbname_len
;
436 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
437 msglen
+= rbname_len
;
438 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
440 msg_type
= MIG_RP_MSG_REQ_PAGES
;
443 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
446 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
,
447 RAMBlock
*rb
, ram_addr_t start
, uint64_t haddr
)
449 void *aligned
= (void *)(uintptr_t)ROUND_DOWN(haddr
, qemu_ram_pagesize(rb
));
450 bool received
= false;
452 WITH_QEMU_LOCK_GUARD(&mis
->page_request_mutex
) {
453 received
= ramblock_recv_bitmap_test_byte_offset(rb
, start
);
454 if (!received
&& !g_tree_lookup(mis
->page_requested
, aligned
)) {
456 * The page has not been received, and it's not yet in the page
457 * request list. Queue it. Set the value of element to 1, so that
458 * things like g_tree_lookup() will return TRUE (1) when found.
460 g_tree_insert(mis
->page_requested
, aligned
, (gpointer
)1);
461 mis
->page_requested_count
++;
462 trace_postcopy_page_req_add(aligned
, mis
->page_requested_count
);
467 * If the page is there, skip sending the message. We don't even need the
468 * lock because as long as the page arrived, it'll be there forever.
474 return migrate_send_rp_message_req_pages(mis
, rb
, start
);
477 static bool migration_colo_enabled
;
478 bool migration_incoming_colo_enabled(void)
480 return migration_colo_enabled
;
483 void migration_incoming_disable_colo(void)
485 ram_block_discard_disable(false);
486 migration_colo_enabled
= false;
489 int migration_incoming_enable_colo(void)
491 if (ram_block_discard_disable(true)) {
492 error_report("COLO: cannot disable RAM discard");
495 migration_colo_enabled
= true;
499 void migrate_add_address(SocketAddress
*address
)
501 MigrationIncomingState
*mis
= migration_incoming_get_current();
503 QAPI_LIST_PREPEND(mis
->socket_address_list
,
504 QAPI_CLONE(SocketAddress
, address
));
507 static void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
509 const char *p
= NULL
;
511 /* URI is not suitable for migration? */
512 if (!migration_channels_and_uri_compatible(uri
, errp
)) {
516 qapi_event_send_migration(MIGRATION_STATUS_SETUP
);
517 if (strstart(uri
, "tcp:", &p
) ||
518 strstart(uri
, "unix:", NULL
) ||
519 strstart(uri
, "vsock:", NULL
)) {
520 socket_start_incoming_migration(p
? p
: uri
, errp
);
522 } else if (strstart(uri
, "rdma:", &p
)) {
523 rdma_start_incoming_migration(p
, errp
);
525 } else if (strstart(uri
, "exec:", &p
)) {
526 exec_start_incoming_migration(p
, errp
);
527 } else if (strstart(uri
, "fd:", &p
)) {
528 fd_start_incoming_migration(p
, errp
);
530 error_setg(errp
, "unknown migration protocol: %s", uri
);
534 static void process_incoming_migration_bh(void *opaque
)
536 Error
*local_err
= NULL
;
537 MigrationIncomingState
*mis
= opaque
;
539 /* If capability late_block_activate is set:
540 * Only fire up the block code now if we're going to restart the
541 * VM, else 'cont' will do it.
542 * This causes file locking to happen; so we don't want it to happen
543 * unless we really are starting the VM.
545 if (!migrate_late_block_activate() ||
546 (autostart
&& (!global_state_received() ||
547 global_state_get_runstate() == RUN_STATE_RUNNING
))) {
548 /* Make sure all file formats throw away their mutable metadata.
549 * If we get an error here, just don't restart the VM yet. */
550 bdrv_activate_all(&local_err
);
552 error_report_err(local_err
);
559 * This must happen after all error conditions are dealt with and
560 * we're sure the VM is going to be running on this host.
562 qemu_announce_self(&mis
->announce_timer
, migrate_announce_params());
564 multifd_load_shutdown();
566 dirty_bitmap_mig_before_vm_start();
568 if (!global_state_received() ||
569 global_state_get_runstate() == RUN_STATE_RUNNING
) {
573 runstate_set(RUN_STATE_PAUSED
);
575 } else if (migration_incoming_colo_enabled()) {
576 migration_incoming_disable_colo();
579 runstate_set(global_state_get_runstate());
582 * This must happen after any state changes since as soon as an external
583 * observer sees this event they might start to prod at the VM assuming
586 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
587 MIGRATION_STATUS_COMPLETED
);
588 qemu_bh_delete(mis
->bh
);
589 migration_incoming_state_destroy();
592 static void coroutine_fn
593 process_incoming_migration_co(void *opaque
)
595 MigrationIncomingState
*mis
= migration_incoming_get_current();
598 Error
*local_err
= NULL
;
600 assert(mis
->from_src_file
);
601 mis
->migration_incoming_co
= qemu_coroutine_self();
602 mis
->largest_page_size
= qemu_ram_pagesize_largest();
603 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
604 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
605 MIGRATION_STATUS_ACTIVE
);
606 ret
= qemu_loadvm_state(mis
->from_src_file
);
608 ps
= postcopy_state_get();
609 trace_process_incoming_migration_co_end(ret
, ps
);
610 if (ps
!= POSTCOPY_INCOMING_NONE
) {
611 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
613 * Where a migration had postcopy enabled (and thus went to advise)
614 * but managed to complete within the precopy period, we can use
617 postcopy_ram_incoming_cleanup(mis
);
618 } else if (ret
>= 0) {
620 * Postcopy was started, cleanup should happen at the end of the
623 trace_process_incoming_migration_co_postcopy_end_main();
626 /* Else if something went wrong then just fall out of the normal exit */
629 /* we get COLO info, and know if we are in COLO mode */
630 if (!ret
&& migration_incoming_colo_enabled()) {
631 /* Make sure all file formats throw away their mutable metadata */
632 bdrv_activate_all(&local_err
);
634 error_report_err(local_err
);
638 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
639 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
640 mis
->have_colo_incoming_thread
= true;
641 qemu_coroutine_yield();
643 qemu_mutex_unlock_iothread();
644 /* Wait checkpoint incoming thread exit before free resource */
645 qemu_thread_join(&mis
->colo_incoming_thread
);
646 qemu_mutex_lock_iothread();
647 /* We hold the global iothread lock, so it is safe here */
648 colo_release_ram_cache();
652 error_report("load of migration failed: %s", strerror(-ret
));
655 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
656 qemu_bh_schedule(mis
->bh
);
657 mis
->migration_incoming_co
= NULL
;
661 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
662 MIGRATION_STATUS_FAILED
);
663 qemu_fclose(mis
->from_src_file
);
665 multifd_load_cleanup();
671 * migration_incoming_setup: Setup incoming migration
672 * @f: file for main migration channel
673 * @errp: where to put errors
675 * Returns: %true on success, %false on error.
677 static bool migration_incoming_setup(QEMUFile
*f
, Error
**errp
)
679 MigrationIncomingState
*mis
= migration_incoming_get_current();
681 if (!mis
->from_src_file
) {
682 mis
->from_src_file
= f
;
684 qemu_file_set_blocking(f
, false);
688 void migration_incoming_process(void)
690 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
691 qemu_coroutine_enter(co
);
694 /* Returns true if recovered from a paused migration, otherwise false */
695 static bool postcopy_try_recover(void)
697 MigrationIncomingState
*mis
= migration_incoming_get_current();
699 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
700 /* Resumed from a paused postcopy migration */
702 /* This should be set already in migration_incoming_setup() */
703 assert(mis
->from_src_file
);
704 /* Postcopy has standalone thread to do vm load */
705 qemu_file_set_blocking(mis
->from_src_file
, true);
707 /* Re-configure the return path */
708 mis
->to_src_file
= qemu_file_get_return_path(mis
->from_src_file
);
710 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
711 MIGRATION_STATUS_POSTCOPY_RECOVER
);
714 * Here, we only wake up the main loading thread (while the
715 * rest threads will still be waiting), so that we can receive
716 * commands from source now, and answer it if needed. The
717 * rest threads will be woken up afterwards until we are sure
718 * that source is ready to reply to page requests.
720 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
727 void migration_fd_process_incoming(QEMUFile
*f
, Error
**errp
)
729 if (!migration_incoming_setup(f
, errp
)) {
732 if (postcopy_try_recover()) {
735 migration_incoming_process();
739 * Returns true when we want to start a new incoming migration process,
742 static bool migration_should_start_incoming(bool main_channel
)
744 /* Multifd doesn't start unless all channels are established */
745 if (migrate_use_multifd()) {
746 return migration_has_all_channels();
749 /* Preempt channel only starts when the main channel is created */
750 if (migrate_postcopy_preempt()) {
755 * For all the rest types of migration, we should only reach here when
756 * it's the main channel that's being created, and we should always
757 * proceed with this channel.
759 assert(main_channel
);
763 void migration_ioc_process_incoming(QIOChannel
*ioc
, Error
**errp
)
765 MigrationIncomingState
*mis
= migration_incoming_get_current();
766 Error
*local_err
= NULL
;
768 bool default_channel
= true;
769 uint32_t channel_magic
= 0;
772 if (migrate_use_multifd() && !migrate_postcopy_ram() &&
773 qio_channel_has_feature(ioc
, QIO_CHANNEL_FEATURE_READ_MSG_PEEK
)) {
775 * With multiple channels, it is possible that we receive channels
776 * out of order on destination side, causing incorrect mapping of
777 * source channels on destination side. Check channel MAGIC to
778 * decide type of channel. Please note this is best effort, postcopy
779 * preempt channel does not send any magic number so avoid it for
780 * postcopy live migration. Also tls live migration already does
781 * tls handshake while initializing main channel so with tls this
782 * issue is not possible.
784 ret
= migration_channel_read_peek(ioc
, (void *)&channel_magic
,
785 sizeof(channel_magic
), &local_err
);
788 error_propagate(errp
, local_err
);
792 default_channel
= (channel_magic
== cpu_to_be32(QEMU_VM_FILE_MAGIC
));
794 default_channel
= !mis
->from_src_file
;
797 if (multifd_load_setup(errp
) != 0) {
798 error_setg(errp
, "Failed to setup multifd channels");
802 if (default_channel
) {
803 f
= qemu_file_new_input(ioc
);
805 if (!migration_incoming_setup(f
, errp
)) {
809 /* Multiple connections */
810 assert(migration_needs_multiple_sockets());
811 if (migrate_use_multifd()) {
812 multifd_recv_new_channel(ioc
, &local_err
);
814 assert(migrate_postcopy_preempt());
815 f
= qemu_file_new_input(ioc
);
816 postcopy_preempt_new_channel(mis
, f
);
819 error_propagate(errp
, local_err
);
824 if (migration_should_start_incoming(default_channel
)) {
825 /* If it's a recovery, we're done */
826 if (postcopy_try_recover()) {
829 migration_incoming_process();
834 * @migration_has_all_channels: We have received all channels that we need
836 * Returns true when we have got connections to all the channels that
837 * we need for migration.
839 bool migration_has_all_channels(void)
841 MigrationIncomingState
*mis
= migration_incoming_get_current();
843 if (!mis
->from_src_file
) {
847 if (migrate_use_multifd()) {
848 return multifd_recv_all_channels_created();
851 if (migrate_postcopy_preempt()) {
852 return mis
->postcopy_qemufile_dst
!= NULL
;
859 * Send a 'SHUT' message on the return channel with the given value
860 * to indicate that we've finished with the RP. Non-0 value indicates
863 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
868 buf
= cpu_to_be32(value
);
869 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
873 * Send a 'PONG' message on the return channel with the given value
874 * (normally in response to a 'PING')
876 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
881 buf
= cpu_to_be32(value
);
882 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
885 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
893 * First, we send the header part. It contains only the len of
894 * idstr, and the idstr itself.
896 len
= strlen(block_name
);
898 memcpy(buf
+ 1, block_name
, len
);
900 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
901 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
906 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
909 * Next, we dump the received bitmap to the stream.
911 * TODO: currently we are safe since we are the only one that is
912 * using the to_src_file handle (fault thread is still paused),
913 * and it's ok even not taking the mutex. However the best way is
914 * to take the lock before sending the message header, and release
915 * the lock after sending the bitmap.
917 qemu_mutex_lock(&mis
->rp_mutex
);
918 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
919 qemu_mutex_unlock(&mis
->rp_mutex
);
921 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
924 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
928 buf
= cpu_to_be32(value
);
929 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
932 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
934 MigrationCapabilityStatusList
*head
= NULL
, **tail
= &head
;
935 MigrationCapabilityStatus
*caps
;
936 MigrationState
*s
= migrate_get_current();
939 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
940 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
941 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
945 caps
= g_malloc0(sizeof(*caps
));
946 caps
->capability
= i
;
947 caps
->state
= s
->enabled_capabilities
[i
];
948 QAPI_LIST_APPEND(tail
, caps
);
954 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
956 MigrationParameters
*params
;
957 MigrationState
*s
= migrate_get_current();
959 /* TODO use QAPI_CLONE() instead of duplicating it inline */
960 params
= g_malloc0(sizeof(*params
));
961 params
->has_compress_level
= true;
962 params
->compress_level
= s
->parameters
.compress_level
;
963 params
->has_compress_threads
= true;
964 params
->compress_threads
= s
->parameters
.compress_threads
;
965 params
->has_compress_wait_thread
= true;
966 params
->compress_wait_thread
= s
->parameters
.compress_wait_thread
;
967 params
->has_decompress_threads
= true;
968 params
->decompress_threads
= s
->parameters
.decompress_threads
;
969 params
->has_throttle_trigger_threshold
= true;
970 params
->throttle_trigger_threshold
= s
->parameters
.throttle_trigger_threshold
;
971 params
->has_cpu_throttle_initial
= true;
972 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
973 params
->has_cpu_throttle_increment
= true;
974 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
975 params
->has_cpu_throttle_tailslow
= true;
976 params
->cpu_throttle_tailslow
= s
->parameters
.cpu_throttle_tailslow
;
977 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
978 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
979 params
->tls_authz
= g_strdup(s
->parameters
.tls_authz
?
980 s
->parameters
.tls_authz
: "");
981 params
->has_max_bandwidth
= true;
982 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
983 params
->has_downtime_limit
= true;
984 params
->downtime_limit
= s
->parameters
.downtime_limit
;
985 params
->has_x_checkpoint_delay
= true;
986 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
987 params
->has_block_incremental
= true;
988 params
->block_incremental
= s
->parameters
.block_incremental
;
989 params
->has_multifd_channels
= true;
990 params
->multifd_channels
= s
->parameters
.multifd_channels
;
991 params
->has_multifd_compression
= true;
992 params
->multifd_compression
= s
->parameters
.multifd_compression
;
993 params
->has_multifd_zlib_level
= true;
994 params
->multifd_zlib_level
= s
->parameters
.multifd_zlib_level
;
995 params
->has_multifd_zstd_level
= true;
996 params
->multifd_zstd_level
= s
->parameters
.multifd_zstd_level
;
997 params
->has_xbzrle_cache_size
= true;
998 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
999 params
->has_max_postcopy_bandwidth
= true;
1000 params
->max_postcopy_bandwidth
= s
->parameters
.max_postcopy_bandwidth
;
1001 params
->has_max_cpu_throttle
= true;
1002 params
->max_cpu_throttle
= s
->parameters
.max_cpu_throttle
;
1003 params
->has_announce_initial
= true;
1004 params
->announce_initial
= s
->parameters
.announce_initial
;
1005 params
->has_announce_max
= true;
1006 params
->announce_max
= s
->parameters
.announce_max
;
1007 params
->has_announce_rounds
= true;
1008 params
->announce_rounds
= s
->parameters
.announce_rounds
;
1009 params
->has_announce_step
= true;
1010 params
->announce_step
= s
->parameters
.announce_step
;
1012 if (s
->parameters
.has_block_bitmap_mapping
) {
1013 params
->has_block_bitmap_mapping
= true;
1014 params
->block_bitmap_mapping
=
1015 QAPI_CLONE(BitmapMigrationNodeAliasList
,
1016 s
->parameters
.block_bitmap_mapping
);
1022 void qmp_client_migrate_info(const char *protocol
, const char *hostname
,
1023 bool has_port
, int64_t port
,
1024 bool has_tls_port
, int64_t tls_port
,
1025 const char *cert_subject
,
1028 if (strcmp(protocol
, "spice") == 0) {
1029 if (!qemu_using_spice(errp
)) {
1033 if (!has_port
&& !has_tls_port
) {
1034 error_setg(errp
, QERR_MISSING_PARAMETER
, "port/tls-port");
1038 if (qemu_spice
.migrate_info(hostname
,
1039 has_port
? port
: -1,
1040 has_tls_port
? tls_port
: -1,
1042 error_setg(errp
, "Could not set up display for migration");
1048 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "protocol", "'spice'");
1051 AnnounceParameters
*migrate_announce_params(void)
1053 static AnnounceParameters ap
;
1055 MigrationState
*s
= migrate_get_current();
1057 ap
.initial
= s
->parameters
.announce_initial
;
1058 ap
.max
= s
->parameters
.announce_max
;
1059 ap
.rounds
= s
->parameters
.announce_rounds
;
1060 ap
.step
= s
->parameters
.announce_step
;
1066 * Return true if we're already in the middle of a migration
1067 * (i.e. any of the active or setup states)
1069 bool migration_is_setup_or_active(int state
)
1072 case MIGRATION_STATUS_ACTIVE
:
1073 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1074 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1075 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1076 case MIGRATION_STATUS_SETUP
:
1077 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1078 case MIGRATION_STATUS_DEVICE
:
1079 case MIGRATION_STATUS_WAIT_UNPLUG
:
1080 case MIGRATION_STATUS_COLO
:
1089 bool migration_is_running(int state
)
1092 case MIGRATION_STATUS_ACTIVE
:
1093 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1094 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1095 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1096 case MIGRATION_STATUS_SETUP
:
1097 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1098 case MIGRATION_STATUS_DEVICE
:
1099 case MIGRATION_STATUS_WAIT_UNPLUG
:
1100 case MIGRATION_STATUS_CANCELLING
:
1109 static bool migrate_show_downtime(MigrationState
*s
)
1111 return (s
->state
== MIGRATION_STATUS_COMPLETED
) || migration_in_postcopy();
1114 static void populate_time_info(MigrationInfo
*info
, MigrationState
*s
)
1116 info
->has_status
= true;
1117 info
->has_setup_time
= true;
1118 info
->setup_time
= s
->setup_time
;
1120 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
1121 info
->has_total_time
= true;
1122 info
->total_time
= s
->total_time
;
1124 info
->has_total_time
= true;
1125 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) -
1129 if (migrate_show_downtime(s
)) {
1130 info
->has_downtime
= true;
1131 info
->downtime
= s
->downtime
;
1133 info
->has_expected_downtime
= true;
1134 info
->expected_downtime
= s
->expected_downtime
;
1138 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
1140 size_t page_size
= qemu_target_page_size();
1142 info
->ram
= g_malloc0(sizeof(*info
->ram
));
1143 info
->ram
->transferred
= stat64_get(&ram_atomic_counters
.transferred
);
1144 info
->ram
->total
= ram_bytes_total();
1145 info
->ram
->duplicate
= stat64_get(&ram_atomic_counters
.duplicate
);
1146 /* legacy value. It is not used anymore */
1147 info
->ram
->skipped
= 0;
1148 info
->ram
->normal
= stat64_get(&ram_atomic_counters
.normal
);
1149 info
->ram
->normal_bytes
= info
->ram
->normal
* page_size
;
1150 info
->ram
->mbps
= s
->mbps
;
1151 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
1152 info
->ram
->dirty_sync_missed_zero_copy
=
1153 ram_counters
.dirty_sync_missed_zero_copy
;
1154 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
1155 info
->ram
->page_size
= page_size
;
1156 info
->ram
->multifd_bytes
= ram_counters
.multifd_bytes
;
1157 info
->ram
->pages_per_second
= s
->pages_per_second
;
1158 info
->ram
->precopy_bytes
= ram_counters
.precopy_bytes
;
1159 info
->ram
->downtime_bytes
= ram_counters
.downtime_bytes
;
1160 info
->ram
->postcopy_bytes
= stat64_get(&ram_atomic_counters
.postcopy_bytes
);
1162 if (migrate_use_xbzrle()) {
1163 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
1164 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
1165 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
1166 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
1167 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
1168 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
1169 info
->xbzrle_cache
->encoding_rate
= xbzrle_counters
.encoding_rate
;
1170 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
1173 if (migrate_use_compression()) {
1174 info
->compression
= g_malloc0(sizeof(*info
->compression
));
1175 info
->compression
->pages
= compression_counters
.pages
;
1176 info
->compression
->busy
= compression_counters
.busy
;
1177 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
1178 info
->compression
->compressed_size
=
1179 compression_counters
.compressed_size
;
1180 info
->compression
->compression_rate
=
1181 compression_counters
.compression_rate
;
1184 if (cpu_throttle_active()) {
1185 info
->has_cpu_throttle_percentage
= true;
1186 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
1189 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
1190 info
->ram
->remaining
= ram_bytes_remaining();
1191 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
1195 static void populate_disk_info(MigrationInfo
*info
)
1197 if (blk_mig_active()) {
1198 info
->disk
= g_malloc0(sizeof(*info
->disk
));
1199 info
->disk
->transferred
= blk_mig_bytes_transferred();
1200 info
->disk
->remaining
= blk_mig_bytes_remaining();
1201 info
->disk
->total
= blk_mig_bytes_total();
1205 static void fill_source_migration_info(MigrationInfo
*info
)
1207 MigrationState
*s
= migrate_get_current();
1208 int state
= qatomic_read(&s
->state
);
1209 GSList
*cur_blocker
= migration_blockers
;
1211 info
->blocked_reasons
= NULL
;
1214 * There are two types of reasons a migration might be blocked;
1215 * a) devices marked in VMState as non-migratable, and
1216 * b) Explicit migration blockers
1217 * We need to add both of them here.
1219 qemu_savevm_non_migratable_list(&info
->blocked_reasons
);
1221 while (cur_blocker
) {
1222 QAPI_LIST_PREPEND(info
->blocked_reasons
,
1223 g_strdup(error_get_pretty(cur_blocker
->data
)));
1224 cur_blocker
= g_slist_next(cur_blocker
);
1226 info
->has_blocked_reasons
= info
->blocked_reasons
!= NULL
;
1229 case MIGRATION_STATUS_NONE
:
1230 /* no migration has happened ever */
1231 /* do not overwrite destination migration status */
1233 case MIGRATION_STATUS_SETUP
:
1234 info
->has_status
= true;
1235 info
->has_total_time
= false;
1237 case MIGRATION_STATUS_ACTIVE
:
1238 case MIGRATION_STATUS_CANCELLING
:
1239 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1240 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1241 case MIGRATION_STATUS_DEVICE
:
1242 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1243 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1244 /* TODO add some postcopy stats */
1245 populate_time_info(info
, s
);
1246 populate_ram_info(info
, s
);
1247 populate_disk_info(info
);
1248 populate_vfio_info(info
);
1250 case MIGRATION_STATUS_COLO
:
1251 info
->has_status
= true;
1252 /* TODO: display COLO specific information (checkpoint info etc.) */
1254 case MIGRATION_STATUS_COMPLETED
:
1255 populate_time_info(info
, s
);
1256 populate_ram_info(info
, s
);
1257 populate_vfio_info(info
);
1259 case MIGRATION_STATUS_FAILED
:
1260 info
->has_status
= true;
1262 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
1265 case MIGRATION_STATUS_CANCELLED
:
1266 info
->has_status
= true;
1268 case MIGRATION_STATUS_WAIT_UNPLUG
:
1269 info
->has_status
= true;
1272 info
->status
= state
;
1275 typedef enum WriteTrackingSupport
{
1276 WT_SUPPORT_UNKNOWN
= 0,
1278 WT_SUPPORT_AVAILABLE
,
1279 WT_SUPPORT_COMPATIBLE
1280 } WriteTrackingSupport
;
1283 WriteTrackingSupport
migrate_query_write_tracking(void)
1285 /* Check if kernel supports required UFFD features */
1286 if (!ram_write_tracking_available()) {
1287 return WT_SUPPORT_ABSENT
;
1290 * Check if current memory configuration is
1291 * compatible with required UFFD features.
1293 if (!ram_write_tracking_compatible()) {
1294 return WT_SUPPORT_AVAILABLE
;
1297 return WT_SUPPORT_COMPATIBLE
;
1301 * @migration_caps_check - check capability validity
1303 * @cap_list: old capability list, array of bool
1304 * @params: new capabilities to be applied soon
1305 * @errp: set *errp if the check failed, with reason
1307 * Returns true if check passed, otherwise false.
1309 static bool migrate_caps_check(bool *cap_list
,
1310 MigrationCapabilityStatusList
*params
,
1313 MigrationCapabilityStatusList
*cap
;
1314 bool old_postcopy_cap
;
1315 MigrationIncomingState
*mis
= migration_incoming_get_current();
1317 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1319 for (cap
= params
; cap
; cap
= cap
->next
) {
1320 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
1323 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1324 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
1325 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
1327 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
1332 #ifndef CONFIG_REPLICATION
1333 if (cap_list
[MIGRATION_CAPABILITY_X_COLO
]) {
1334 error_setg(errp
, "QEMU compiled without replication module"
1335 " can't enable COLO");
1336 error_append_hint(errp
, "Please enable replication before COLO.\n");
1341 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
1342 /* This check is reasonably expensive, so only when it's being
1343 * set the first time, also it's only the destination that needs
1346 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
1347 !postcopy_ram_supported_by_host(mis
)) {
1348 /* postcopy_ram_supported_by_host will have emitted a more
1351 error_setg(errp
, "Postcopy is not supported");
1355 if (cap_list
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
]) {
1356 error_setg(errp
, "Postcopy is not compatible with ignore-shared");
1361 if (cap_list
[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
]) {
1362 WriteTrackingSupport wt_support
;
1365 * Check if 'background-snapshot' capability is supported by
1366 * host kernel and compatible with guest memory configuration.
1368 wt_support
= migrate_query_write_tracking();
1369 if (wt_support
< WT_SUPPORT_AVAILABLE
) {
1370 error_setg(errp
, "Background-snapshot is not supported by host kernel");
1373 if (wt_support
< WT_SUPPORT_COMPATIBLE
) {
1374 error_setg(errp
, "Background-snapshot is not compatible "
1375 "with guest memory configuration");
1380 * Check if there are any migration capabilities
1381 * incompatible with 'background-snapshot'.
1383 for (idx
= 0; idx
< check_caps_background_snapshot
.size
; idx
++) {
1384 int incomp_cap
= check_caps_background_snapshot
.caps
[idx
];
1385 if (cap_list
[incomp_cap
]) {
1387 "Background-snapshot is not compatible with %s",
1388 MigrationCapability_str(incomp_cap
));
1395 if (cap_list
[MIGRATION_CAPABILITY_ZERO_COPY_SEND
] &&
1396 (!cap_list
[MIGRATION_CAPABILITY_MULTIFD
] ||
1397 cap_list
[MIGRATION_CAPABILITY_COMPRESS
] ||
1398 cap_list
[MIGRATION_CAPABILITY_XBZRLE
] ||
1399 migrate_multifd_compression() ||
1400 migrate_use_tls())) {
1402 "Zero copy only available for non-compressed non-TLS multifd migration");
1406 if (cap_list
[MIGRATION_CAPABILITY_ZERO_COPY_SEND
]) {
1408 "Zero copy currently only available on Linux");
1413 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_PREEMPT
]) {
1414 if (!cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
1415 error_setg(errp
, "Postcopy preempt requires postcopy-ram");
1420 * Preempt mode requires urgent pages to be sent in separate
1421 * channel, OTOH compression logic will disorder all pages into
1422 * different compression channels, which is not compatible with the
1423 * preempt assumptions on channel assignments.
1425 if (cap_list
[MIGRATION_CAPABILITY_COMPRESS
]) {
1426 error_setg(errp
, "Postcopy preempt not compatible with compress");
1431 if (cap_list
[MIGRATION_CAPABILITY_MULTIFD
]) {
1432 if (cap_list
[MIGRATION_CAPABILITY_COMPRESS
]) {
1433 error_setg(errp
, "Multifd is not compatible with compress");
1441 static void fill_destination_migration_info(MigrationInfo
*info
)
1443 MigrationIncomingState
*mis
= migration_incoming_get_current();
1445 if (mis
->socket_address_list
) {
1446 info
->has_socket_address
= true;
1447 info
->socket_address
=
1448 QAPI_CLONE(SocketAddressList
, mis
->socket_address_list
);
1451 switch (mis
->state
) {
1452 case MIGRATION_STATUS_NONE
:
1454 case MIGRATION_STATUS_SETUP
:
1455 case MIGRATION_STATUS_CANCELLING
:
1456 case MIGRATION_STATUS_CANCELLED
:
1457 case MIGRATION_STATUS_ACTIVE
:
1458 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1459 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1460 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1461 case MIGRATION_STATUS_FAILED
:
1462 case MIGRATION_STATUS_COLO
:
1463 info
->has_status
= true;
1465 case MIGRATION_STATUS_COMPLETED
:
1466 info
->has_status
= true;
1467 fill_destination_postcopy_migration_info(info
);
1470 info
->status
= mis
->state
;
1473 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1475 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1477 fill_destination_migration_info(info
);
1478 fill_source_migration_info(info
);
1483 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
1486 MigrationState
*s
= migrate_get_current();
1487 MigrationCapabilityStatusList
*cap
;
1488 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
1490 if (migration_is_running(s
->state
)) {
1491 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1495 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
1496 if (!migrate_caps_check(cap_list
, params
, errp
)) {
1500 for (cap
= params
; cap
; cap
= cap
->next
) {
1501 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
1506 * Check whether the parameters are valid. Error will be put into errp
1507 * (if provided). Return true if valid, otherwise false.
1509 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
1511 if (params
->has_compress_level
&&
1512 (params
->compress_level
> 9)) {
1513 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
1514 "a value between 0 and 9");
1518 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
1519 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1521 "a value between 1 and 255");
1525 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
1526 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1527 "decompress_threads",
1528 "a value between 1 and 255");
1532 if (params
->has_throttle_trigger_threshold
&&
1533 (params
->throttle_trigger_threshold
< 1 ||
1534 params
->throttle_trigger_threshold
> 100)) {
1535 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1536 "throttle_trigger_threshold",
1537 "an integer in the range of 1 to 100");
1541 if (params
->has_cpu_throttle_initial
&&
1542 (params
->cpu_throttle_initial
< 1 ||
1543 params
->cpu_throttle_initial
> 99)) {
1544 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1545 "cpu_throttle_initial",
1546 "an integer in the range of 1 to 99");
1550 if (params
->has_cpu_throttle_increment
&&
1551 (params
->cpu_throttle_increment
< 1 ||
1552 params
->cpu_throttle_increment
> 99)) {
1553 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1554 "cpu_throttle_increment",
1555 "an integer in the range of 1 to 99");
1559 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
1560 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1562 "an integer in the range of 0 to "stringify(SIZE_MAX
)
1567 if (params
->has_downtime_limit
&&
1568 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
1569 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1571 "an integer in the range of 0 to "
1572 stringify(MAX_MIGRATE_DOWNTIME
)" ms");
1576 /* x_checkpoint_delay is now always positive */
1578 if (params
->has_multifd_channels
&& (params
->multifd_channels
< 1)) {
1579 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1581 "a value between 1 and 255");
1585 if (params
->has_multifd_zlib_level
&&
1586 (params
->multifd_zlib_level
> 9)) {
1587 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zlib_level",
1588 "a value between 0 and 9");
1592 if (params
->has_multifd_zstd_level
&&
1593 (params
->multifd_zstd_level
> 20)) {
1594 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zstd_level",
1595 "a value between 0 and 20");
1599 if (params
->has_xbzrle_cache_size
&&
1600 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
1601 !is_power_of_2(params
->xbzrle_cache_size
))) {
1602 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1603 "xbzrle_cache_size",
1604 "a power of two no less than the target page size");
1608 if (params
->has_max_cpu_throttle
&&
1609 (params
->max_cpu_throttle
< params
->cpu_throttle_initial
||
1610 params
->max_cpu_throttle
> 99)) {
1611 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1613 "an integer in the range of cpu_throttle_initial to 99");
1617 if (params
->has_announce_initial
&&
1618 params
->announce_initial
> 100000) {
1619 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1621 "a value between 0 and 100000");
1624 if (params
->has_announce_max
&&
1625 params
->announce_max
> 100000) {
1626 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1628 "a value between 0 and 100000");
1631 if (params
->has_announce_rounds
&&
1632 params
->announce_rounds
> 1000) {
1633 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1635 "a value between 0 and 1000");
1638 if (params
->has_announce_step
&&
1639 (params
->announce_step
< 1 ||
1640 params
->announce_step
> 10000)) {
1641 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1643 "a value between 0 and 10000");
1647 if (params
->has_block_bitmap_mapping
&&
1648 !check_dirty_bitmap_mig_alias_map(params
->block_bitmap_mapping
, errp
)) {
1649 error_prepend(errp
, "Invalid mapping given for block-bitmap-mapping: ");
1654 if (migrate_use_zero_copy_send() &&
1655 ((params
->has_multifd_compression
&& params
->multifd_compression
) ||
1656 (params
->tls_creds
&& *params
->tls_creds
))) {
1658 "Zero copy only available for non-compressed non-TLS multifd migration");
1666 static void migrate_params_test_apply(MigrateSetParameters
*params
,
1667 MigrationParameters
*dest
)
1669 *dest
= migrate_get_current()->parameters
;
1671 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1673 if (params
->has_compress_level
) {
1674 dest
->compress_level
= params
->compress_level
;
1677 if (params
->has_compress_threads
) {
1678 dest
->compress_threads
= params
->compress_threads
;
1681 if (params
->has_compress_wait_thread
) {
1682 dest
->compress_wait_thread
= params
->compress_wait_thread
;
1685 if (params
->has_decompress_threads
) {
1686 dest
->decompress_threads
= params
->decompress_threads
;
1689 if (params
->has_throttle_trigger_threshold
) {
1690 dest
->throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1693 if (params
->has_cpu_throttle_initial
) {
1694 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
1697 if (params
->has_cpu_throttle_increment
) {
1698 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
1701 if (params
->has_cpu_throttle_tailslow
) {
1702 dest
->cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1705 if (params
->tls_creds
) {
1706 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1707 dest
->tls_creds
= params
->tls_creds
->u
.s
;
1710 if (params
->tls_hostname
) {
1711 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1712 dest
->tls_hostname
= params
->tls_hostname
->u
.s
;
1715 if (params
->has_max_bandwidth
) {
1716 dest
->max_bandwidth
= params
->max_bandwidth
;
1719 if (params
->has_downtime_limit
) {
1720 dest
->downtime_limit
= params
->downtime_limit
;
1723 if (params
->has_x_checkpoint_delay
) {
1724 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
1727 if (params
->has_block_incremental
) {
1728 dest
->block_incremental
= params
->block_incremental
;
1730 if (params
->has_multifd_channels
) {
1731 dest
->multifd_channels
= params
->multifd_channels
;
1733 if (params
->has_multifd_compression
) {
1734 dest
->multifd_compression
= params
->multifd_compression
;
1736 if (params
->has_xbzrle_cache_size
) {
1737 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
1739 if (params
->has_max_postcopy_bandwidth
) {
1740 dest
->max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1742 if (params
->has_max_cpu_throttle
) {
1743 dest
->max_cpu_throttle
= params
->max_cpu_throttle
;
1745 if (params
->has_announce_initial
) {
1746 dest
->announce_initial
= params
->announce_initial
;
1748 if (params
->has_announce_max
) {
1749 dest
->announce_max
= params
->announce_max
;
1751 if (params
->has_announce_rounds
) {
1752 dest
->announce_rounds
= params
->announce_rounds
;
1754 if (params
->has_announce_step
) {
1755 dest
->announce_step
= params
->announce_step
;
1758 if (params
->has_block_bitmap_mapping
) {
1759 dest
->has_block_bitmap_mapping
= true;
1760 dest
->block_bitmap_mapping
= params
->block_bitmap_mapping
;
1764 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
1766 MigrationState
*s
= migrate_get_current();
1768 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1770 if (params
->has_compress_level
) {
1771 s
->parameters
.compress_level
= params
->compress_level
;
1774 if (params
->has_compress_threads
) {
1775 s
->parameters
.compress_threads
= params
->compress_threads
;
1778 if (params
->has_compress_wait_thread
) {
1779 s
->parameters
.compress_wait_thread
= params
->compress_wait_thread
;
1782 if (params
->has_decompress_threads
) {
1783 s
->parameters
.decompress_threads
= params
->decompress_threads
;
1786 if (params
->has_throttle_trigger_threshold
) {
1787 s
->parameters
.throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1790 if (params
->has_cpu_throttle_initial
) {
1791 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
1794 if (params
->has_cpu_throttle_increment
) {
1795 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
1798 if (params
->has_cpu_throttle_tailslow
) {
1799 s
->parameters
.cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1802 if (params
->tls_creds
) {
1803 g_free(s
->parameters
.tls_creds
);
1804 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1805 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1808 if (params
->tls_hostname
) {
1809 g_free(s
->parameters
.tls_hostname
);
1810 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1811 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1814 if (params
->tls_authz
) {
1815 g_free(s
->parameters
.tls_authz
);
1816 assert(params
->tls_authz
->type
== QTYPE_QSTRING
);
1817 s
->parameters
.tls_authz
= g_strdup(params
->tls_authz
->u
.s
);
1820 if (params
->has_max_bandwidth
) {
1821 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1822 if (s
->to_dst_file
&& !migration_in_postcopy()) {
1823 qemu_file_set_rate_limit(s
->to_dst_file
,
1824 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1828 if (params
->has_downtime_limit
) {
1829 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1832 if (params
->has_x_checkpoint_delay
) {
1833 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1834 if (migration_in_colo_state()) {
1835 colo_checkpoint_notify(s
);
1839 if (params
->has_block_incremental
) {
1840 s
->parameters
.block_incremental
= params
->block_incremental
;
1842 if (params
->has_multifd_channels
) {
1843 s
->parameters
.multifd_channels
= params
->multifd_channels
;
1845 if (params
->has_multifd_compression
) {
1846 s
->parameters
.multifd_compression
= params
->multifd_compression
;
1848 if (params
->has_xbzrle_cache_size
) {
1849 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1850 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1852 if (params
->has_max_postcopy_bandwidth
) {
1853 s
->parameters
.max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1854 if (s
->to_dst_file
&& migration_in_postcopy()) {
1855 qemu_file_set_rate_limit(s
->to_dst_file
,
1856 s
->parameters
.max_postcopy_bandwidth
/ XFER_LIMIT_RATIO
);
1859 if (params
->has_max_cpu_throttle
) {
1860 s
->parameters
.max_cpu_throttle
= params
->max_cpu_throttle
;
1862 if (params
->has_announce_initial
) {
1863 s
->parameters
.announce_initial
= params
->announce_initial
;
1865 if (params
->has_announce_max
) {
1866 s
->parameters
.announce_max
= params
->announce_max
;
1868 if (params
->has_announce_rounds
) {
1869 s
->parameters
.announce_rounds
= params
->announce_rounds
;
1871 if (params
->has_announce_step
) {
1872 s
->parameters
.announce_step
= params
->announce_step
;
1875 if (params
->has_block_bitmap_mapping
) {
1876 qapi_free_BitmapMigrationNodeAliasList(
1877 s
->parameters
.block_bitmap_mapping
);
1879 s
->parameters
.has_block_bitmap_mapping
= true;
1880 s
->parameters
.block_bitmap_mapping
=
1881 QAPI_CLONE(BitmapMigrationNodeAliasList
,
1882 params
->block_bitmap_mapping
);
1886 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1888 MigrationParameters tmp
;
1890 /* TODO Rewrite "" to null instead */
1891 if (params
->tls_creds
1892 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1893 qobject_unref(params
->tls_creds
->u
.n
);
1894 params
->tls_creds
->type
= QTYPE_QSTRING
;
1895 params
->tls_creds
->u
.s
= strdup("");
1897 /* TODO Rewrite "" to null instead */
1898 if (params
->tls_hostname
1899 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1900 qobject_unref(params
->tls_hostname
->u
.n
);
1901 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1902 params
->tls_hostname
->u
.s
= strdup("");
1905 migrate_params_test_apply(params
, &tmp
);
1907 if (!migrate_params_check(&tmp
, errp
)) {
1908 /* Invalid parameter */
1912 migrate_params_apply(params
, errp
);
1916 void qmp_migrate_start_postcopy(Error
**errp
)
1918 MigrationState
*s
= migrate_get_current();
1920 if (!migrate_postcopy()) {
1921 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1922 " the start of migration");
1926 if (s
->state
== MIGRATION_STATUS_NONE
) {
1927 error_setg(errp
, "Postcopy must be started after migration has been"
1932 * we don't error if migration has finished since that would be racy
1933 * with issuing this command.
1935 qatomic_set(&s
->start_postcopy
, true);
1938 /* shared migration helpers */
1940 void migrate_set_state(int *state
, int old_state
, int new_state
)
1942 assert(new_state
< MIGRATION_STATUS__MAX
);
1943 if (qatomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1944 trace_migrate_set_state(MigrationStatus_str(new_state
));
1945 migrate_generate_event(new_state
);
1949 static MigrationCapabilityStatus
*migrate_cap_add(MigrationCapability index
,
1952 MigrationCapabilityStatus
*cap
;
1954 cap
= g_new0(MigrationCapabilityStatus
, 1);
1955 cap
->capability
= index
;
1961 void migrate_set_block_enabled(bool value
, Error
**errp
)
1963 MigrationCapabilityStatusList
*cap
= NULL
;
1965 QAPI_LIST_PREPEND(cap
, migrate_cap_add(MIGRATION_CAPABILITY_BLOCK
, value
));
1966 qmp_migrate_set_capabilities(cap
, errp
);
1967 qapi_free_MigrationCapabilityStatusList(cap
);
1970 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1972 s
->parameters
.block_incremental
= value
;
1975 static void block_cleanup_parameters(MigrationState
*s
)
1977 if (s
->must_remove_block_options
) {
1978 /* setting to false can never fail */
1979 migrate_set_block_enabled(false, &error_abort
);
1980 migrate_set_block_incremental(s
, false);
1981 s
->must_remove_block_options
= false;
1985 static void migrate_fd_cleanup(MigrationState
*s
)
1987 qemu_bh_delete(s
->cleanup_bh
);
1988 s
->cleanup_bh
= NULL
;
1990 g_free(s
->hostname
);
1992 json_writer_free(s
->vmdesc
);
1995 qemu_savevm_state_cleanup();
1997 if (s
->to_dst_file
) {
2000 trace_migrate_fd_cleanup();
2001 qemu_mutex_unlock_iothread();
2002 if (s
->migration_thread_running
) {
2003 qemu_thread_join(&s
->thread
);
2004 s
->migration_thread_running
= false;
2006 qemu_mutex_lock_iothread();
2008 multifd_save_cleanup();
2009 qemu_mutex_lock(&s
->qemu_file_lock
);
2010 tmp
= s
->to_dst_file
;
2011 s
->to_dst_file
= NULL
;
2012 qemu_mutex_unlock(&s
->qemu_file_lock
);
2014 * Close the file handle without the lock to make sure the
2015 * critical section won't block for long.
2017 migration_ioc_unregister_yank_from_file(tmp
);
2021 if (s
->postcopy_qemufile_src
) {
2022 migration_ioc_unregister_yank_from_file(s
->postcopy_qemufile_src
);
2023 qemu_fclose(s
->postcopy_qemufile_src
);
2024 s
->postcopy_qemufile_src
= NULL
;
2027 assert(!migration_is_active(s
));
2029 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
2030 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
2031 MIGRATION_STATUS_CANCELLED
);
2035 /* It is used on info migrate. We can't free it */
2036 error_report_err(error_copy(s
->error
));
2038 notifier_list_notify(&migration_state_notifiers
, s
);
2039 block_cleanup_parameters(s
);
2040 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2043 static void migrate_fd_cleanup_schedule(MigrationState
*s
)
2046 * Ref the state for bh, because it may be called when
2047 * there're already no other refs
2049 object_ref(OBJECT(s
));
2050 qemu_bh_schedule(s
->cleanup_bh
);
2053 static void migrate_fd_cleanup_bh(void *opaque
)
2055 MigrationState
*s
= opaque
;
2056 migrate_fd_cleanup(s
);
2057 object_unref(OBJECT(s
));
2060 void migrate_set_error(MigrationState
*s
, const Error
*error
)
2062 QEMU_LOCK_GUARD(&s
->error_mutex
);
2064 s
->error
= error_copy(error
);
2068 static void migrate_error_free(MigrationState
*s
)
2070 QEMU_LOCK_GUARD(&s
->error_mutex
);
2072 error_free(s
->error
);
2077 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
2079 trace_migrate_fd_error(error_get_pretty(error
));
2080 assert(s
->to_dst_file
== NULL
);
2081 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2082 MIGRATION_STATUS_FAILED
);
2083 migrate_set_error(s
, error
);
2086 static void migrate_fd_cancel(MigrationState
*s
)
2089 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
2090 trace_migrate_fd_cancel();
2092 WITH_QEMU_LOCK_GUARD(&s
->qemu_file_lock
) {
2093 if (s
->rp_state
.from_dst_file
) {
2094 /* shutdown the rp socket, so causing the rp thread to shutdown */
2095 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
2100 old_state
= s
->state
;
2101 if (!migration_is_running(old_state
)) {
2104 /* If the migration is paused, kick it out of the pause */
2105 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
2106 qemu_sem_post(&s
->pause_sem
);
2108 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
2109 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
2112 * If we're unlucky the migration code might be stuck somewhere in a
2113 * send/write while the network has failed and is waiting to timeout;
2114 * if we've got shutdown(2) available then we can force it to quit.
2115 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
2116 * called in a bh, so there is no race against this cancel.
2118 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
2119 qemu_file_shutdown(f
);
2121 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
2122 Error
*local_err
= NULL
;
2124 bdrv_activate_all(&local_err
);
2126 error_report_err(local_err
);
2128 s
->block_inactive
= false;
2133 void add_migration_state_change_notifier(Notifier
*notify
)
2135 notifier_list_add(&migration_state_notifiers
, notify
);
2138 void remove_migration_state_change_notifier(Notifier
*notify
)
2140 notifier_remove(notify
);
2143 bool migration_in_setup(MigrationState
*s
)
2145 return s
->state
== MIGRATION_STATUS_SETUP
;
2148 bool migration_has_finished(MigrationState
*s
)
2150 return s
->state
== MIGRATION_STATUS_COMPLETED
;
2153 bool migration_has_failed(MigrationState
*s
)
2155 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
2156 s
->state
== MIGRATION_STATUS_FAILED
);
2159 bool migration_in_postcopy(void)
2161 MigrationState
*s
= migrate_get_current();
2164 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
2165 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
2166 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
2173 bool migration_in_postcopy_after_devices(MigrationState
*s
)
2175 return migration_in_postcopy() && s
->postcopy_after_devices
;
2178 bool migration_in_incoming_postcopy(void)
2180 PostcopyState ps
= postcopy_state_get();
2182 return ps
>= POSTCOPY_INCOMING_DISCARD
&& ps
< POSTCOPY_INCOMING_END
;
2185 bool migration_incoming_postcopy_advised(void)
2187 PostcopyState ps
= postcopy_state_get();
2189 return ps
>= POSTCOPY_INCOMING_ADVISE
&& ps
< POSTCOPY_INCOMING_END
;
2192 bool migration_in_bg_snapshot(void)
2194 MigrationState
*s
= migrate_get_current();
2196 return migrate_background_snapshot() &&
2197 migration_is_setup_or_active(s
->state
);
2200 bool migration_is_idle(void)
2202 MigrationState
*s
= current_migration
;
2209 case MIGRATION_STATUS_NONE
:
2210 case MIGRATION_STATUS_CANCELLED
:
2211 case MIGRATION_STATUS_COMPLETED
:
2212 case MIGRATION_STATUS_FAILED
:
2214 case MIGRATION_STATUS_SETUP
:
2215 case MIGRATION_STATUS_CANCELLING
:
2216 case MIGRATION_STATUS_ACTIVE
:
2217 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
2218 case MIGRATION_STATUS_COLO
:
2219 case MIGRATION_STATUS_PRE_SWITCHOVER
:
2220 case MIGRATION_STATUS_DEVICE
:
2221 case MIGRATION_STATUS_WAIT_UNPLUG
:
2223 case MIGRATION_STATUS__MAX
:
2224 g_assert_not_reached();
2230 bool migration_is_active(MigrationState
*s
)
2232 return (s
->state
== MIGRATION_STATUS_ACTIVE
||
2233 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2236 void migrate_init(MigrationState
*s
)
2239 * Reinitialise all migration state, except
2240 * parameters/capabilities that the user set, and
2245 s
->to_dst_file
= NULL
;
2246 s
->state
= MIGRATION_STATUS_NONE
;
2247 s
->rp_state
.from_dst_file
= NULL
;
2248 s
->rp_state
.error
= false;
2250 s
->pages_per_second
= 0.0;
2252 s
->expected_downtime
= 0;
2254 s
->start_postcopy
= false;
2255 s
->postcopy_after_devices
= false;
2256 s
->migration_thread_running
= false;
2257 error_free(s
->error
);
2261 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
2263 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2265 s
->vm_was_running
= false;
2266 s
->iteration_initial_bytes
= 0;
2267 s
->threshold_size
= 0;
2270 int migrate_add_blocker_internal(Error
*reason
, Error
**errp
)
2272 /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */
2273 if (runstate_check(RUN_STATE_SAVE_VM
) || !migration_is_idle()) {
2274 error_propagate_prepend(errp
, error_copy(reason
),
2275 "disallowing migration blocker "
2276 "(migration/snapshot in progress) for: ");
2280 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
2284 int migrate_add_blocker(Error
*reason
, Error
**errp
)
2286 if (only_migratable
) {
2287 error_propagate_prepend(errp
, error_copy(reason
),
2288 "disallowing migration blocker "
2289 "(--only-migratable) for: ");
2293 return migrate_add_blocker_internal(reason
, errp
);
2296 void migrate_del_blocker(Error
*reason
)
2298 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
2301 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
2303 Error
*local_err
= NULL
;
2304 static bool once
= true;
2307 error_setg(errp
, "The incoming migration has already been started");
2310 if (!runstate_check(RUN_STATE_INMIGRATE
)) {
2311 error_setg(errp
, "'-incoming' was not specified on the command line");
2315 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
2319 qemu_start_incoming_migration(uri
, &local_err
);
2322 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2323 error_propagate(errp
, local_err
);
2330 void qmp_migrate_recover(const char *uri
, Error
**errp
)
2332 MigrationIncomingState
*mis
= migration_incoming_get_current();
2335 * Don't even bother to use ERRP_GUARD() as it _must_ always be set by
2336 * callers (no one should ignore a recover failure); if there is, it's a
2337 * programming error.
2341 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2342 error_setg(errp
, "Migrate recover can only be run "
2343 "when postcopy is paused.");
2347 /* If there's an existing transport, release it */
2348 migration_incoming_transport_cleanup(mis
);
2351 * Note that this call will never start a real migration; it will
2352 * only re-setup the migration stream and poke existing migration
2353 * to continue using that newly established channel.
2355 qemu_start_incoming_migration(uri
, errp
);
2358 void qmp_migrate_pause(Error
**errp
)
2360 MigrationState
*ms
= migrate_get_current();
2361 MigrationIncomingState
*mis
= migration_incoming_get_current();
2364 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2365 /* Source side, during postcopy */
2366 qemu_mutex_lock(&ms
->qemu_file_lock
);
2367 ret
= qemu_file_shutdown(ms
->to_dst_file
);
2368 qemu_mutex_unlock(&ms
->qemu_file_lock
);
2370 error_setg(errp
, "Failed to pause source migration");
2375 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2376 ret
= qemu_file_shutdown(mis
->from_src_file
);
2378 error_setg(errp
, "Failed to pause destination migration");
2383 error_setg(errp
, "migrate-pause is currently only supported "
2384 "during postcopy-active state");
2387 bool migration_is_blocked(Error
**errp
)
2389 if (qemu_savevm_state_blocked(errp
)) {
2393 if (migration_blockers
) {
2394 error_propagate(errp
, error_copy(migration_blockers
->data
));
2401 /* Returns true if continue to migrate, or false if error detected */
2402 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
2403 bool resume
, Error
**errp
)
2405 Error
*local_err
= NULL
;
2408 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2409 error_setg(errp
, "Cannot resume if there is no "
2410 "paused migration");
2415 * Postcopy recovery won't work well with release-ram
2416 * capability since release-ram will drop the page buffer as
2417 * long as the page is put into the send buffer. So if there
2418 * is a network failure happened, any page buffers that have
2419 * not yet reached the destination VM but have already been
2420 * sent from the source VM will be lost forever. Let's refuse
2421 * the client from resuming such a postcopy migration.
2422 * Luckily release-ram was designed to only be used when src
2423 * and destination VMs are on the same host, so it should be
2426 if (migrate_release_ram()) {
2427 error_setg(errp
, "Postcopy recovery cannot work "
2428 "when release-ram capability is set");
2432 /* This is a resume, skip init status */
2436 if (migration_is_running(s
->state
)) {
2437 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
2441 if (runstate_check(RUN_STATE_INMIGRATE
)) {
2442 error_setg(errp
, "Guest is waiting for an incoming migration");
2446 if (runstate_check(RUN_STATE_POSTMIGRATE
)) {
2447 error_setg(errp
, "Can't migrate the vm that was paused due to "
2448 "previous migration");
2452 if (migration_is_blocked(errp
)) {
2456 if (blk
|| blk_inc
) {
2457 if (migrate_colo_enabled()) {
2458 error_setg(errp
, "No disk migration is required in COLO mode");
2461 if (migrate_use_block() || migrate_use_block_incremental()) {
2462 error_setg(errp
, "Command options are incompatible with "
2463 "current migration capabilities");
2466 migrate_set_block_enabled(true, &local_err
);
2468 error_propagate(errp
, local_err
);
2471 s
->must_remove_block_options
= true;
2475 migrate_set_block_incremental(s
, true);
2480 * set ram_counters compression_counters memory to zero for a
2483 memset(&ram_counters
, 0, sizeof(ram_counters
));
2484 memset(&compression_counters
, 0, sizeof(compression_counters
));
2489 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
2490 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
2491 bool has_resume
, bool resume
, Error
**errp
)
2493 Error
*local_err
= NULL
;
2494 MigrationState
*s
= migrate_get_current();
2495 const char *p
= NULL
;
2497 /* URI is not suitable for migration? */
2498 if (!migration_channels_and_uri_compatible(uri
, errp
)) {
2502 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
2503 has_resume
&& resume
, errp
)) {
2504 /* Error detected, put into errp */
2508 if (!(has_resume
&& resume
)) {
2509 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
2514 if (strstart(uri
, "tcp:", &p
) ||
2515 strstart(uri
, "unix:", NULL
) ||
2516 strstart(uri
, "vsock:", NULL
)) {
2517 socket_start_outgoing_migration(s
, p
? p
: uri
, &local_err
);
2519 } else if (strstart(uri
, "rdma:", &p
)) {
2520 rdma_start_outgoing_migration(s
, p
, &local_err
);
2522 } else if (strstart(uri
, "exec:", &p
)) {
2523 exec_start_outgoing_migration(s
, p
, &local_err
);
2524 } else if (strstart(uri
, "fd:", &p
)) {
2525 fd_start_outgoing_migration(s
, p
, &local_err
);
2527 if (!(has_resume
&& resume
)) {
2528 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2530 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
2531 "a valid migration protocol");
2532 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2533 MIGRATION_STATUS_FAILED
);
2534 block_cleanup_parameters(s
);
2539 if (!(has_resume
&& resume
)) {
2540 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2542 migrate_fd_error(s
, local_err
);
2543 error_propagate(errp
, local_err
);
2548 void qmp_migrate_cancel(Error
**errp
)
2550 migration_cancel(NULL
);
2553 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
2555 MigrationState
*s
= migrate_get_current();
2556 if (s
->state
!= state
) {
2557 error_setg(errp
, "Migration not in expected state: %s",
2558 MigrationStatus_str(s
->state
));
2561 qemu_sem_post(&s
->pause_sem
);
2564 bool migrate_release_ram(void)
2568 s
= migrate_get_current();
2570 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
2573 bool migrate_postcopy_ram(void)
2577 s
= migrate_get_current();
2579 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
2582 bool migrate_postcopy(void)
2584 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2587 bool migrate_auto_converge(void)
2591 s
= migrate_get_current();
2593 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
2596 bool migrate_zero_blocks(void)
2600 s
= migrate_get_current();
2602 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
2605 bool migrate_postcopy_blocktime(void)
2609 s
= migrate_get_current();
2611 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
2614 bool migrate_use_compression(void)
2618 s
= migrate_get_current();
2620 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
2623 int migrate_compress_level(void)
2627 s
= migrate_get_current();
2629 return s
->parameters
.compress_level
;
2632 int migrate_compress_threads(void)
2636 s
= migrate_get_current();
2638 return s
->parameters
.compress_threads
;
2641 int migrate_compress_wait_thread(void)
2645 s
= migrate_get_current();
2647 return s
->parameters
.compress_wait_thread
;
2650 int migrate_decompress_threads(void)
2654 s
= migrate_get_current();
2656 return s
->parameters
.decompress_threads
;
2659 bool migrate_dirty_bitmaps(void)
2663 s
= migrate_get_current();
2665 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
2668 bool migrate_ignore_shared(void)
2672 s
= migrate_get_current();
2674 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
];
2677 bool migrate_validate_uuid(void)
2681 s
= migrate_get_current();
2683 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_VALIDATE_UUID
];
2686 bool migrate_use_events(void)
2690 s
= migrate_get_current();
2692 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
2695 bool migrate_use_multifd(void)
2699 s
= migrate_get_current();
2701 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_MULTIFD
];
2704 bool migrate_pause_before_switchover(void)
2708 s
= migrate_get_current();
2710 return s
->enabled_capabilities
[
2711 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
2714 int migrate_multifd_channels(void)
2718 s
= migrate_get_current();
2720 return s
->parameters
.multifd_channels
;
2723 MultiFDCompression
migrate_multifd_compression(void)
2727 s
= migrate_get_current();
2729 assert(s
->parameters
.multifd_compression
< MULTIFD_COMPRESSION__MAX
);
2730 return s
->parameters
.multifd_compression
;
2733 int migrate_multifd_zlib_level(void)
2737 s
= migrate_get_current();
2739 return s
->parameters
.multifd_zlib_level
;
2742 int migrate_multifd_zstd_level(void)
2746 s
= migrate_get_current();
2748 return s
->parameters
.multifd_zstd_level
;
2752 bool migrate_use_zero_copy_send(void)
2756 s
= migrate_get_current();
2758 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_COPY_SEND
];
2762 int migrate_use_tls(void)
2766 s
= migrate_get_current();
2768 return s
->parameters
.tls_creds
&& *s
->parameters
.tls_creds
;
2771 int migrate_use_xbzrle(void)
2775 s
= migrate_get_current();
2777 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
2780 uint64_t migrate_xbzrle_cache_size(void)
2784 s
= migrate_get_current();
2786 return s
->parameters
.xbzrle_cache_size
;
2789 static int64_t migrate_max_postcopy_bandwidth(void)
2793 s
= migrate_get_current();
2795 return s
->parameters
.max_postcopy_bandwidth
;
2798 bool migrate_use_block(void)
2802 s
= migrate_get_current();
2804 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
2807 bool migrate_use_return_path(void)
2811 s
= migrate_get_current();
2813 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
2816 bool migrate_use_block_incremental(void)
2820 s
= migrate_get_current();
2822 return s
->parameters
.block_incremental
;
2825 bool migrate_background_snapshot(void)
2829 s
= migrate_get_current();
2831 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
];
2834 bool migrate_postcopy_preempt(void)
2838 s
= migrate_get_current();
2840 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_PREEMPT
];
2843 /* migration thread support */
2845 * Something bad happened to the RP stream, mark an error
2846 * The caller shall print or trace something to indicate why
2848 static void mark_source_rp_bad(MigrationState
*s
)
2850 s
->rp_state
.error
= true;
2853 static struct rp_cmd_args
{
2854 ssize_t len
; /* -1 = variable */
2857 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
2858 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
2859 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
2860 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
2861 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
2862 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
2863 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
2864 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
2868 * Process a request for pages received on the return path,
2869 * We're allowed to send more than requested (e.g. to round to our page size)
2870 * and we don't need to send pages that have already been sent.
2872 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
2873 ram_addr_t start
, size_t len
)
2875 long our_host_ps
= qemu_real_host_page_size();
2877 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
2880 * Since we currently insist on matching page sizes, just sanity check
2881 * we're being asked for whole host pages.
2883 if (!QEMU_IS_ALIGNED(start
, our_host_ps
) ||
2884 !QEMU_IS_ALIGNED(len
, our_host_ps
)) {
2885 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2886 " len: %zd", __func__
, start
, len
);
2887 mark_source_rp_bad(ms
);
2891 if (ram_save_queue_pages(rbname
, start
, len
)) {
2892 mark_source_rp_bad(ms
);
2896 /* Return true to retry, false to quit */
2897 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
2899 trace_postcopy_pause_return_path();
2901 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
2903 trace_postcopy_pause_return_path_continued();
2908 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
2910 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
2913 error_report("%s: invalid block name '%s'", __func__
, block_name
);
2917 /* Fetch the received bitmap and refresh the dirty bitmap */
2918 return ram_dirty_bitmap_reload(s
, block
);
2921 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
2923 trace_source_return_path_thread_resume_ack(value
);
2925 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
2926 error_report("%s: illegal resume_ack value %"PRIu32
,
2931 /* Now both sides are active. */
2932 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2933 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2935 /* Notify send thread that time to continue send pages */
2936 qemu_sem_post(&s
->rp_state
.rp_sem
);
2942 * Release ms->rp_state.from_dst_file (and postcopy_qemufile_src if
2943 * existed) in a safe way.
2945 static void migration_release_dst_files(MigrationState
*ms
)
2949 WITH_QEMU_LOCK_GUARD(&ms
->qemu_file_lock
) {
2951 * Reset the from_dst_file pointer first before releasing it, as we
2952 * can't block within lock section
2954 file
= ms
->rp_state
.from_dst_file
;
2955 ms
->rp_state
.from_dst_file
= NULL
;
2959 * Do the same to postcopy fast path socket too if there is. No
2960 * locking needed because this qemufile should only be managed by
2961 * return path thread.
2963 if (ms
->postcopy_qemufile_src
) {
2964 migration_ioc_unregister_yank_from_file(ms
->postcopy_qemufile_src
);
2965 qemu_file_shutdown(ms
->postcopy_qemufile_src
);
2966 qemu_fclose(ms
->postcopy_qemufile_src
);
2967 ms
->postcopy_qemufile_src
= NULL
;
2974 * Handles messages sent on the return path towards the source VM
2977 static void *source_return_path_thread(void *opaque
)
2979 MigrationState
*ms
= opaque
;
2980 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
2981 uint16_t header_len
, header_type
;
2983 uint32_t tmp32
, sibling_error
;
2984 ram_addr_t start
= 0; /* =0 to silence warning */
2985 size_t len
= 0, expected_len
;
2988 trace_source_return_path_thread_entry();
2989 rcu_register_thread();
2992 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
2993 migration_is_setup_or_active(ms
->state
)) {
2994 trace_source_return_path_thread_loop_top();
2995 header_type
= qemu_get_be16(rp
);
2996 header_len
= qemu_get_be16(rp
);
2998 if (qemu_file_get_error(rp
)) {
2999 mark_source_rp_bad(ms
);
3003 if (header_type
>= MIG_RP_MSG_MAX
||
3004 header_type
== MIG_RP_MSG_INVALID
) {
3005 error_report("RP: Received invalid message 0x%04x length 0x%04x",
3006 header_type
, header_len
);
3007 mark_source_rp_bad(ms
);
3011 if ((rp_cmd_args
[header_type
].len
!= -1 &&
3012 header_len
!= rp_cmd_args
[header_type
].len
) ||
3013 header_len
> sizeof(buf
)) {
3014 error_report("RP: Received '%s' message (0x%04x) with"
3015 "incorrect length %d expecting %zu",
3016 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
3017 (size_t)rp_cmd_args
[header_type
].len
);
3018 mark_source_rp_bad(ms
);
3022 /* We know we've got a valid header by this point */
3023 res
= qemu_get_buffer(rp
, buf
, header_len
);
3024 if (res
!= header_len
) {
3025 error_report("RP: Failed reading data for message 0x%04x"
3026 " read %d expected %d",
3027 header_type
, res
, header_len
);
3028 mark_source_rp_bad(ms
);
3032 /* OK, we have the message and the data */
3033 switch (header_type
) {
3034 case MIG_RP_MSG_SHUT
:
3035 sibling_error
= ldl_be_p(buf
);
3036 trace_source_return_path_thread_shut(sibling_error
);
3037 if (sibling_error
) {
3038 error_report("RP: Sibling indicated error %d", sibling_error
);
3039 mark_source_rp_bad(ms
);
3042 * We'll let the main thread deal with closing the RP
3043 * we could do a shutdown(2) on it, but we're the only user
3044 * anyway, so there's nothing gained.
3048 case MIG_RP_MSG_PONG
:
3049 tmp32
= ldl_be_p(buf
);
3050 trace_source_return_path_thread_pong(tmp32
);
3051 qemu_sem_post(&ms
->rp_state
.rp_pong_acks
);
3054 case MIG_RP_MSG_REQ_PAGES
:
3055 start
= ldq_be_p(buf
);
3056 len
= ldl_be_p(buf
+ 8);
3057 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
3060 case MIG_RP_MSG_REQ_PAGES_ID
:
3061 expected_len
= 12 + 1; /* header + termination */
3063 if (header_len
>= expected_len
) {
3064 start
= ldq_be_p(buf
);
3065 len
= ldl_be_p(buf
+ 8);
3066 /* Now we expect an idstr */
3067 tmp32
= buf
[12]; /* Length of the following idstr */
3068 buf
[13 + tmp32
] = '\0';
3069 expected_len
+= tmp32
;
3071 if (header_len
!= expected_len
) {
3072 error_report("RP: Req_Page_id with length %d expecting %zd",
3073 header_len
, expected_len
);
3074 mark_source_rp_bad(ms
);
3077 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
3080 case MIG_RP_MSG_RECV_BITMAP
:
3081 if (header_len
< 1) {
3082 error_report("%s: missing block name", __func__
);
3083 mark_source_rp_bad(ms
);
3086 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
3087 buf
[buf
[0] + 1] = '\0';
3088 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
3089 mark_source_rp_bad(ms
);
3094 case MIG_RP_MSG_RESUME_ACK
:
3095 tmp32
= ldl_be_p(buf
);
3096 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
3097 mark_source_rp_bad(ms
);
3108 res
= qemu_file_get_error(rp
);
3110 if (res
&& migration_in_postcopy()) {
3112 * Maybe there is something we can do: it looks like a
3113 * network down issue, and we pause for a recovery.
3115 migration_release_dst_files(ms
);
3117 if (postcopy_pause_return_path_thread(ms
)) {
3119 * Reload rp, reset the rest. Referencing it is safe since
3120 * it's reset only by us above, or when migration completes
3122 rp
= ms
->rp_state
.from_dst_file
;
3123 ms
->rp_state
.error
= false;
3128 trace_source_return_path_thread_bad_end();
3129 mark_source_rp_bad(ms
);
3132 trace_source_return_path_thread_end();
3133 migration_release_dst_files(ms
);
3134 rcu_unregister_thread();
3138 static int open_return_path_on_source(MigrationState
*ms
,
3141 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
3142 if (!ms
->rp_state
.from_dst_file
) {
3146 trace_open_return_path_on_source();
3148 if (!create_thread
) {
3153 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
3154 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
3155 ms
->rp_state
.rp_thread_created
= true;
3157 trace_open_return_path_on_source_continue();
3162 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
3163 static int await_return_path_close_on_source(MigrationState
*ms
)
3166 * If this is a normal exit then the destination will send a SHUT and the
3167 * rp_thread will exit, however if there's an error we need to cause
3170 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
3172 * shutdown(2), if we have it, will cause it to unblock if it's stuck
3173 * waiting for the destination.
3175 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
3176 mark_source_rp_bad(ms
);
3178 trace_await_return_path_close_on_source_joining();
3179 qemu_thread_join(&ms
->rp_state
.rp_thread
);
3180 ms
->rp_state
.rp_thread_created
= false;
3181 trace_await_return_path_close_on_source_close();
3182 return ms
->rp_state
.error
;
3186 migration_wait_main_channel(MigrationState
*ms
)
3188 /* Wait until one PONG message received */
3189 qemu_sem_wait(&ms
->rp_state
.rp_pong_acks
);
3193 * Switch from normal iteration to postcopy
3194 * Returns non-0 on error
3196 static int postcopy_start(MigrationState
*ms
)
3199 QIOChannelBuffer
*bioc
;
3201 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3202 int64_t bandwidth
= migrate_max_postcopy_bandwidth();
3203 bool restart_block
= false;
3204 int cur_state
= MIGRATION_STATUS_ACTIVE
;
3206 if (migrate_postcopy_preempt()) {
3207 migration_wait_main_channel(ms
);
3208 if (postcopy_preempt_establish_channel(ms
)) {
3209 migrate_set_state(&ms
->state
, ms
->state
, MIGRATION_STATUS_FAILED
);
3214 if (!migrate_pause_before_switchover()) {
3215 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
3216 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3219 trace_postcopy_start();
3220 qemu_mutex_lock_iothread();
3221 trace_postcopy_start_set_run();
3223 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3224 global_state_store();
3225 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
3230 ret
= migration_maybe_pause(ms
, &cur_state
,
3231 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3236 ret
= bdrv_inactivate_all();
3240 restart_block
= true;
3243 * Cause any non-postcopiable, but iterative devices to
3244 * send out their final data.
3246 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
3249 * in Finish migrate and with the io-lock held everything should
3250 * be quiet, but we've potentially still got dirty pages and we
3251 * need to tell the destination to throw any pages it's already received
3254 if (migrate_postcopy_ram()) {
3255 ram_postcopy_send_discard_bitmap(ms
);
3259 * send rest of state - note things that are doing postcopy
3260 * will notice we're in POSTCOPY_ACTIVE and not actually
3261 * wrap their state up here
3263 /* 0 max-postcopy-bandwidth means unlimited */
3265 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
3267 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
3269 if (migrate_postcopy_ram()) {
3270 /* Ping just for debugging, helps line traces up */
3271 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
3275 * While loading the device state we may trigger page transfer
3276 * requests and the fd must be free to process those, and thus
3277 * the destination must read the whole device state off the fd before
3278 * it starts processing it. Unfortunately the ad-hoc migration format
3279 * doesn't allow the destination to know the size to read without fully
3280 * parsing it through each devices load-state code (especially the open
3281 * coded devices that use get/put).
3282 * So we wrap the device state up in a package with a length at the start;
3283 * to do this we use a qemu_buf to hold the whole of the device state.
3285 bioc
= qio_channel_buffer_new(4096);
3286 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
3287 fb
= qemu_file_new_output(QIO_CHANNEL(bioc
));
3288 object_unref(OBJECT(bioc
));
3291 * Make sure the receiver can get incoming pages before we send the rest
3294 qemu_savevm_send_postcopy_listen(fb
);
3296 qemu_savevm_state_complete_precopy(fb
, false, false);
3297 if (migrate_postcopy_ram()) {
3298 qemu_savevm_send_ping(fb
, 3);
3301 qemu_savevm_send_postcopy_run(fb
);
3303 /* <><> end of stuff going into the package */
3305 /* Last point of recovery; as soon as we send the package the destination
3306 * can open devices and potentially start running.
3307 * Lets just check again we've not got any errors.
3309 ret
= qemu_file_get_error(ms
->to_dst_file
);
3311 error_report("postcopy_start: Migration stream errored (pre package)");
3315 restart_block
= false;
3317 /* Now send that blob */
3318 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
3323 /* Send a notify to give a chance for anything that needs to happen
3324 * at the transition to postcopy and after the device state; in particular
3325 * spice needs to trigger a transition now
3327 ms
->postcopy_after_devices
= true;
3328 notifier_list_notify(&migration_state_notifiers
, ms
);
3330 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
3332 qemu_mutex_unlock_iothread();
3334 if (migrate_postcopy_ram()) {
3336 * Although this ping is just for debug, it could potentially be
3337 * used for getting a better measurement of downtime at the source.
3339 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
3342 if (migrate_release_ram()) {
3343 ram_postcopy_migrated_memory_release(ms
);
3346 ret
= qemu_file_get_error(ms
->to_dst_file
);
3348 error_report("postcopy_start: Migration stream errored");
3349 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
3350 MIGRATION_STATUS_FAILED
);
3353 trace_postcopy_preempt_enabled(migrate_postcopy_preempt());
3360 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
3361 MIGRATION_STATUS_FAILED
);
3362 if (restart_block
) {
3363 /* A failure happened early enough that we know the destination hasn't
3364 * accessed block devices, so we're safe to recover.
3366 Error
*local_err
= NULL
;
3368 bdrv_activate_all(&local_err
);
3370 error_report_err(local_err
);
3373 qemu_mutex_unlock_iothread();
3378 * migration_maybe_pause: Pause if required to by
3379 * migrate_pause_before_switchover called with the iothread locked
3380 * Returns: 0 on success
3382 static int migration_maybe_pause(MigrationState
*s
,
3383 int *current_active_state
,
3386 if (!migrate_pause_before_switchover()) {
3390 /* Since leaving this state is not atomic with posting the semaphore
3391 * it's possible that someone could have issued multiple migrate_continue
3392 * and the semaphore is incorrectly positive at this point;
3393 * the docs say it's undefined to reinit a semaphore that's already
3394 * init'd, so use timedwait to eat up any existing posts.
3396 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
3397 /* This block intentionally left blank */
3401 * If the migration is cancelled when it is in the completion phase,
3402 * the migration state is set to MIGRATION_STATUS_CANCELLING.
3403 * So we don't need to wait a semaphore, otherwise we would always
3404 * wait for the 'pause_sem' semaphore.
3406 if (s
->state
!= MIGRATION_STATUS_CANCELLING
) {
3407 qemu_mutex_unlock_iothread();
3408 migrate_set_state(&s
->state
, *current_active_state
,
3409 MIGRATION_STATUS_PRE_SWITCHOVER
);
3410 qemu_sem_wait(&s
->pause_sem
);
3411 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
3413 *current_active_state
= new_state
;
3414 qemu_mutex_lock_iothread();
3417 return s
->state
== new_state
? 0 : -EINVAL
;
3421 * migration_completion: Used by migration_thread when there's not much left.
3422 * The caller 'breaks' the loop when this returns.
3424 * @s: Current migration state
3426 static void migration_completion(MigrationState
*s
)
3429 int current_active_state
= s
->state
;
3431 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
3432 qemu_mutex_lock_iothread();
3433 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3434 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3435 s
->vm_was_running
= runstate_is_running();
3436 ret
= global_state_store();
3439 bool inactivate
= !migrate_colo_enabled();
3440 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
3441 trace_migration_completion_vm_stop(ret
);
3443 ret
= migration_maybe_pause(s
, ¤t_active_state
,
3444 MIGRATION_STATUS_DEVICE
);
3447 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
3448 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
3451 if (inactivate
&& ret
>= 0) {
3452 s
->block_inactive
= true;
3455 qemu_mutex_unlock_iothread();
3460 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3461 trace_migration_completion_postcopy_end();
3463 qemu_mutex_lock_iothread();
3464 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
3465 qemu_mutex_unlock_iothread();
3468 * Shutdown the postcopy fast path thread. This is only needed
3469 * when dest QEMU binary is old (7.1/7.2). QEMU 8.0+ doesn't need
3472 if (migrate_postcopy_preempt() && s
->preempt_pre_7_2
) {
3473 postcopy_preempt_shutdown_file(s
);
3476 trace_migration_completion_postcopy_end_after_complete();
3482 * If rp was opened we must clean up the thread before
3483 * cleaning everything else up (since if there are no failures
3484 * it will wait for the destination to send it's status in
3487 if (s
->rp_state
.rp_thread_created
) {
3489 trace_migration_return_path_end_before();
3490 rp_error
= await_return_path_close_on_source(s
);
3491 trace_migration_return_path_end_after(rp_error
);
3493 goto fail_invalidate
;
3497 if (qemu_file_get_error(s
->to_dst_file
)) {
3498 trace_migration_completion_file_err();
3499 goto fail_invalidate
;
3502 if (migrate_colo_enabled() && s
->state
== MIGRATION_STATUS_ACTIVE
) {
3503 /* COLO does not support postcopy */
3504 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
3505 MIGRATION_STATUS_COLO
);
3507 migrate_set_state(&s
->state
, current_active_state
,
3508 MIGRATION_STATUS_COMPLETED
);
3514 /* If not doing postcopy, vm_start() will be called: let's regain
3515 * control on images.
3517 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
3518 s
->state
== MIGRATION_STATUS_DEVICE
) {
3519 Error
*local_err
= NULL
;
3521 qemu_mutex_lock_iothread();
3522 bdrv_activate_all(&local_err
);
3524 error_report_err(local_err
);
3526 s
->block_inactive
= false;
3528 qemu_mutex_unlock_iothread();
3532 migrate_set_state(&s
->state
, current_active_state
,
3533 MIGRATION_STATUS_FAILED
);
3537 * bg_migration_completion: Used by bg_migration_thread when after all the
3538 * RAM has been saved. The caller 'breaks' the loop when this returns.
3540 * @s: Current migration state
3542 static void bg_migration_completion(MigrationState
*s
)
3544 int current_active_state
= s
->state
;
3547 * Stop tracking RAM writes - un-protect memory, un-register UFFD
3548 * memory ranges, flush kernel wait queues and wake up threads
3549 * waiting for write fault to be resolved.
3551 ram_write_tracking_stop();
3553 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
3555 * By this moment we have RAM content saved into the migration stream.
3556 * The next step is to flush the non-RAM content (device state)
3557 * right after the ram content. The device state has been stored into
3558 * the temporary buffer before RAM saving started.
3560 qemu_put_buffer(s
->to_dst_file
, s
->bioc
->data
, s
->bioc
->usage
);
3561 qemu_fflush(s
->to_dst_file
);
3562 } else if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
3566 if (qemu_file_get_error(s
->to_dst_file
)) {
3567 trace_migration_completion_file_err();
3571 migrate_set_state(&s
->state
, current_active_state
,
3572 MIGRATION_STATUS_COMPLETED
);
3576 migrate_set_state(&s
->state
, current_active_state
,
3577 MIGRATION_STATUS_FAILED
);
3580 bool migrate_colo_enabled(void)
3582 MigrationState
*s
= migrate_get_current();
3583 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
3586 typedef enum MigThrError
{
3587 /* No error detected */
3588 MIG_THR_ERR_NONE
= 0,
3589 /* Detected error, but resumed successfully */
3590 MIG_THR_ERR_RECOVERED
= 1,
3591 /* Detected fatal error, need to exit */
3592 MIG_THR_ERR_FATAL
= 2,
3595 static int postcopy_resume_handshake(MigrationState
*s
)
3597 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
3599 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3600 qemu_sem_wait(&s
->rp_state
.rp_sem
);
3603 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3610 /* Return zero if success, or <0 for error */
3611 static int postcopy_do_resume(MigrationState
*s
)
3616 * Call all the resume_prepare() hooks, so that modules can be
3617 * ready for the migration resume.
3619 ret
= qemu_savevm_state_resume_prepare(s
);
3621 error_report("%s: resume_prepare() failure detected: %d",
3627 * If preempt is enabled, re-establish the preempt channel. Note that
3628 * we do it after resume prepare to make sure the main channel will be
3629 * created before the preempt channel. E.g. with weak network, the
3630 * dest QEMU may get messed up with the preempt and main channels on
3631 * the order of connection setup. This guarantees the correct order.
3633 ret
= postcopy_preempt_establish_channel(s
);
3635 error_report("%s: postcopy_preempt_establish_channel(): %d",
3641 * Last handshake with destination on the resume (destination will
3642 * switch to postcopy-active afterwards)
3644 ret
= postcopy_resume_handshake(s
);
3646 error_report("%s: handshake failed: %d", __func__
, ret
);
3654 * We don't return until we are in a safe state to continue current
3655 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
3656 * MIG_THR_ERR_FATAL if unrecovery failure happened.
3658 static MigThrError
postcopy_pause(MigrationState
*s
)
3660 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3666 * Current channel is possibly broken. Release it. Note that this is
3667 * guaranteed even without lock because to_dst_file should only be
3668 * modified by the migration thread. That also guarantees that the
3669 * unregister of yank is safe too without the lock. It should be safe
3670 * even to be within the qemu_file_lock, but we didn't do that to avoid
3671 * taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make
3672 * the qemu_file_lock critical section as small as possible.
3674 assert(s
->to_dst_file
);
3675 migration_ioc_unregister_yank_from_file(s
->to_dst_file
);
3676 qemu_mutex_lock(&s
->qemu_file_lock
);
3677 file
= s
->to_dst_file
;
3678 s
->to_dst_file
= NULL
;
3679 qemu_mutex_unlock(&s
->qemu_file_lock
);
3681 qemu_file_shutdown(file
);
3684 migrate_set_state(&s
->state
, s
->state
,
3685 MIGRATION_STATUS_POSTCOPY_PAUSED
);
3687 error_report("Detected IO failure for postcopy. "
3688 "Migration paused.");
3691 * We wait until things fixed up. Then someone will setup the
3692 * status back for us.
3694 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
3695 qemu_sem_wait(&s
->postcopy_pause_sem
);
3698 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3699 /* Woken up by a recover procedure. Give it a shot */
3702 * Firstly, let's wake up the return path now, with a new
3703 * return path channel.
3705 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
3707 /* Do the resume logic */
3708 if (postcopy_do_resume(s
) == 0) {
3709 /* Let's continue! */
3710 trace_postcopy_pause_continued();
3711 return MIG_THR_ERR_RECOVERED
;
3714 * Something wrong happened during the recovery, let's
3715 * pause again. Pause is always better than throwing
3721 /* This is not right... Time to quit. */
3722 return MIG_THR_ERR_FATAL
;
3727 static MigThrError
migration_detect_error(MigrationState
*s
)
3730 int state
= s
->state
;
3731 Error
*local_error
= NULL
;
3733 if (state
== MIGRATION_STATUS_CANCELLING
||
3734 state
== MIGRATION_STATUS_CANCELLED
) {
3735 /* End the migration, but don't set the state to failed */
3736 return MIG_THR_ERR_FATAL
;
3740 * Try to detect any file errors. Note that postcopy_qemufile_src will
3741 * be NULL when postcopy preempt is not enabled.
3743 ret
= qemu_file_get_error_obj_any(s
->to_dst_file
,
3744 s
->postcopy_qemufile_src
,
3747 /* Everything is fine */
3748 assert(!local_error
);
3749 return MIG_THR_ERR_NONE
;
3753 migrate_set_error(s
, local_error
);
3754 error_free(local_error
);
3757 if (state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
) {
3759 * For postcopy, we allow the network to be down for a
3760 * while. After that, it can be continued by a
3763 return postcopy_pause(s
);
3766 * For precopy (or postcopy with error outside IO), we fail
3769 migrate_set_state(&s
->state
, state
, MIGRATION_STATUS_FAILED
);
3770 trace_migration_thread_file_err();
3772 /* Time to stop the migration, now. */
3773 return MIG_THR_ERR_FATAL
;
3777 /* How many bytes have we transferred since the beginning of the migration */
3778 static uint64_t migration_total_bytes(MigrationState
*s
)
3780 return qemu_file_total_transferred(s
->to_dst_file
) +
3781 ram_counters
.multifd_bytes
;
3784 static void migration_calculate_complete(MigrationState
*s
)
3786 uint64_t bytes
= migration_total_bytes(s
);
3787 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3788 int64_t transfer_time
;
3790 s
->total_time
= end_time
- s
->start_time
;
3793 * It's still not set, so we are precopy migration. For
3794 * postcopy, downtime is calculated during postcopy_start().
3796 s
->downtime
= end_time
- s
->downtime_start
;
3799 transfer_time
= s
->total_time
- s
->setup_time
;
3800 if (transfer_time
) {
3801 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
3805 static void update_iteration_initial_status(MigrationState
*s
)
3808 * Update these three fields at the same time to avoid mismatch info lead
3809 * wrong speed calculation.
3811 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3812 s
->iteration_initial_bytes
= migration_total_bytes(s
);
3813 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
3816 static void migration_update_counters(MigrationState
*s
,
3817 int64_t current_time
)
3819 uint64_t transferred
, transferred_pages
, time_spent
;
3820 uint64_t current_bytes
; /* bytes transferred since the beginning */
3823 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
3827 current_bytes
= migration_total_bytes(s
);
3828 transferred
= current_bytes
- s
->iteration_initial_bytes
;
3829 time_spent
= current_time
- s
->iteration_start_time
;
3830 bandwidth
= (double)transferred
/ time_spent
;
3831 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
3833 s
->mbps
= (((double) transferred
* 8.0) /
3834 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
3836 transferred_pages
= ram_get_total_transferred_pages() -
3837 s
->iteration_initial_pages
;
3838 s
->pages_per_second
= (double) transferred_pages
/
3839 (((double) time_spent
/ 1000.0));
3842 * if we haven't sent anything, we don't want to
3843 * recalculate. 10000 is a small enough number for our purposes
3845 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
3846 s
->expected_downtime
= ram_counters
.remaining
/ bandwidth
;
3849 qemu_file_reset_rate_limit(s
->to_dst_file
);
3851 update_iteration_initial_status(s
);
3853 trace_migrate_transferred(transferred
, time_spent
,
3854 bandwidth
, s
->threshold_size
);
3857 /* Migration thread iteration status */
3859 MIG_ITERATE_RESUME
, /* Resume current iteration */
3860 MIG_ITERATE_SKIP
, /* Skip current iteration */
3861 MIG_ITERATE_BREAK
, /* Break the loop */
3865 * Return true if continue to the next iteration directly, false
3868 static MigIterateState
migration_iteration_run(MigrationState
*s
)
3870 uint64_t must_precopy
, can_postcopy
;
3871 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
3873 qemu_savevm_state_pending_estimate(&must_precopy
, &can_postcopy
);
3874 uint64_t pending_size
= must_precopy
+ can_postcopy
;
3876 trace_migrate_pending_estimate(pending_size
, must_precopy
, can_postcopy
);
3878 if (must_precopy
<= s
->threshold_size
) {
3879 qemu_savevm_state_pending_exact(&must_precopy
, &can_postcopy
);
3880 pending_size
= must_precopy
+ can_postcopy
;
3881 trace_migrate_pending_exact(pending_size
, must_precopy
, can_postcopy
);
3884 if (!pending_size
|| pending_size
< s
->threshold_size
) {
3885 trace_migration_thread_low_pending(pending_size
);
3886 migration_completion(s
);
3887 return MIG_ITERATE_BREAK
;
3890 /* Still a significant amount to transfer */
3891 if (!in_postcopy
&& must_precopy
<= s
->threshold_size
&&
3892 qatomic_read(&s
->start_postcopy
)) {
3893 if (postcopy_start(s
)) {
3894 error_report("%s: postcopy failed to start", __func__
);
3896 return MIG_ITERATE_SKIP
;
3899 /* Just another iteration step */
3900 qemu_savevm_state_iterate(s
->to_dst_file
, in_postcopy
);
3901 return MIG_ITERATE_RESUME
;
3904 static void migration_iteration_finish(MigrationState
*s
)
3906 /* If we enabled cpu throttling for auto-converge, turn it off. */
3907 cpu_throttle_stop();
3909 qemu_mutex_lock_iothread();
3911 case MIGRATION_STATUS_COMPLETED
:
3912 migration_calculate_complete(s
);
3913 runstate_set(RUN_STATE_POSTMIGRATE
);
3915 case MIGRATION_STATUS_COLO
:
3916 if (!migrate_colo_enabled()) {
3917 error_report("%s: critical error: calling COLO code without "
3918 "COLO enabled", __func__
);
3920 migrate_start_colo_process(s
);
3921 s
->vm_was_running
= true;
3923 case MIGRATION_STATUS_FAILED
:
3924 case MIGRATION_STATUS_CANCELLED
:
3925 case MIGRATION_STATUS_CANCELLING
:
3926 if (s
->vm_was_running
) {
3927 if (!runstate_check(RUN_STATE_SHUTDOWN
)) {
3931 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
3932 runstate_set(RUN_STATE_POSTMIGRATE
);
3938 /* Should not reach here, but if so, forgive the VM. */
3939 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3942 migrate_fd_cleanup_schedule(s
);
3943 qemu_mutex_unlock_iothread();
3946 static void bg_migration_iteration_finish(MigrationState
*s
)
3948 qemu_mutex_lock_iothread();
3950 case MIGRATION_STATUS_COMPLETED
:
3951 migration_calculate_complete(s
);
3954 case MIGRATION_STATUS_ACTIVE
:
3955 case MIGRATION_STATUS_FAILED
:
3956 case MIGRATION_STATUS_CANCELLED
:
3957 case MIGRATION_STATUS_CANCELLING
:
3961 /* Should not reach here, but if so, forgive the VM. */
3962 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3966 migrate_fd_cleanup_schedule(s
);
3967 qemu_mutex_unlock_iothread();
3971 * Return true if continue to the next iteration directly, false
3974 static MigIterateState
bg_migration_iteration_run(MigrationState
*s
)
3978 res
= qemu_savevm_state_iterate(s
->to_dst_file
, false);
3980 bg_migration_completion(s
);
3981 return MIG_ITERATE_BREAK
;
3984 return MIG_ITERATE_RESUME
;
3987 void migration_make_urgent_request(void)
3989 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
3992 void migration_consume_urgent_request(void)
3994 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
3997 /* Returns true if the rate limiting was broken by an urgent request */
3998 bool migration_rate_limit(void)
4000 int64_t now
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
4001 MigrationState
*s
= migrate_get_current();
4003 bool urgent
= false;
4004 migration_update_counters(s
, now
);
4005 if (qemu_file_rate_limit(s
->to_dst_file
)) {
4007 if (qemu_file_get_error(s
->to_dst_file
)) {
4011 * Wait for a delay to do rate limiting OR
4012 * something urgent to post the semaphore.
4014 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- now
;
4015 trace_migration_rate_limit_pre(ms
);
4016 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
4018 * We were woken by one or more urgent things but
4019 * the timedwait will have consumed one of them.
4020 * The service routine for the urgent wake will dec
4021 * the semaphore itself for each item it consumes,
4022 * so add this one we just eat back.
4024 qemu_sem_post(&s
->rate_limit_sem
);
4027 trace_migration_rate_limit_post(urgent
);
4033 * if failover devices are present, wait they are completely
4037 static void qemu_savevm_wait_unplug(MigrationState
*s
, int old_state
,
4040 if (qemu_savevm_state_guest_unplug_pending()) {
4041 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_WAIT_UNPLUG
);
4043 while (s
->state
== MIGRATION_STATUS_WAIT_UNPLUG
&&
4044 qemu_savevm_state_guest_unplug_pending()) {
4045 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
4047 if (s
->state
!= MIGRATION_STATUS_WAIT_UNPLUG
) {
4048 int timeout
= 120; /* 30 seconds */
4050 * migration has been canceled
4051 * but as we have started an unplug we must wait the end
4052 * to be able to plug back the card
4054 while (timeout
-- && qemu_savevm_state_guest_unplug_pending()) {
4055 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
4057 if (qemu_savevm_state_guest_unplug_pending() &&
4059 warn_report("migration: partially unplugged device on "
4064 migrate_set_state(&s
->state
, MIGRATION_STATUS_WAIT_UNPLUG
, new_state
);
4066 migrate_set_state(&s
->state
, old_state
, new_state
);
4071 * Master migration thread on the source VM.
4072 * It drives the migration and pumps the data down the outgoing channel.
4074 static void *migration_thread(void *opaque
)
4076 MigrationState
*s
= opaque
;
4077 MigrationThread
*thread
= NULL
;
4078 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
4079 MigThrError thr_error
;
4080 bool urgent
= false;
4082 thread
= MigrationThreadAdd("live_migration", qemu_get_thread_id());
4084 rcu_register_thread();
4086 object_ref(OBJECT(s
));
4087 update_iteration_initial_status(s
);
4089 qemu_savevm_state_header(s
->to_dst_file
);
4092 * If we opened the return path, we need to make sure dst has it
4095 if (s
->rp_state
.rp_thread_created
) {
4096 /* Now tell the dest that it should open its end so it can reply */
4097 qemu_savevm_send_open_return_path(s
->to_dst_file
);
4099 /* And do a ping that will make stuff easier to debug */
4100 qemu_savevm_send_ping(s
->to_dst_file
, 1);
4103 if (migrate_postcopy()) {
4105 * Tell the destination that we *might* want to do postcopy later;
4106 * if the other end can't do postcopy it should fail now, nice and
4109 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
4112 if (migrate_colo_enabled()) {
4113 /* Notify migration destination that we enable COLO */
4114 qemu_savevm_send_colo_enable(s
->to_dst_file
);
4117 qemu_savevm_state_setup(s
->to_dst_file
);
4119 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
4120 MIGRATION_STATUS_ACTIVE
);
4122 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
4124 trace_migration_thread_setup_complete();
4126 while (migration_is_active(s
)) {
4127 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
4128 MigIterateState iter_state
= migration_iteration_run(s
);
4129 if (iter_state
== MIG_ITERATE_SKIP
) {
4131 } else if (iter_state
== MIG_ITERATE_BREAK
) {
4137 * Try to detect any kind of failures, and see whether we
4138 * should stop the migration now.
4140 thr_error
= migration_detect_error(s
);
4141 if (thr_error
== MIG_THR_ERR_FATAL
) {
4142 /* Stop migration */
4144 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
4146 * Just recovered from a e.g. network failure, reset all
4147 * the local variables. This is important to avoid
4148 * breaking transferred_bytes and bandwidth calculation
4150 update_iteration_initial_status(s
);
4153 urgent
= migration_rate_limit();
4156 trace_migration_thread_after_loop();
4157 migration_iteration_finish(s
);
4158 object_unref(OBJECT(s
));
4159 rcu_unregister_thread();
4160 MigrationThreadDel(thread
);
4164 static void bg_migration_vm_start_bh(void *opaque
)
4166 MigrationState
*s
= opaque
;
4168 qemu_bh_delete(s
->vm_start_bh
);
4169 s
->vm_start_bh
= NULL
;
4172 s
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - s
->downtime_start
;
4176 * Background snapshot thread, based on live migration code.
4177 * This is an alternative implementation of live migration mechanism
4178 * introduced specifically to support background snapshots.
4180 * It takes advantage of userfault_fd write protection mechanism introduced
4181 * in v5.7 kernel. Compared to existing dirty page logging migration much
4182 * lesser stream traffic is produced resulting in smaller snapshot images,
4183 * simply cause of no page duplicates can get into the stream.
4185 * Another key point is that generated vmstate stream reflects machine state
4186 * 'frozen' at the beginning of snapshot creation compared to dirty page logging
4187 * mechanism, which effectively results in that saved snapshot is the state of VM
4188 * at the end of the process.
4190 static void *bg_migration_thread(void *opaque
)
4192 MigrationState
*s
= opaque
;
4193 int64_t setup_start
;
4194 MigThrError thr_error
;
4196 bool early_fail
= true;
4198 rcu_register_thread();
4199 object_ref(OBJECT(s
));
4201 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
4203 setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
4205 * We want to save vmstate for the moment when migration has been
4206 * initiated but also we want to save RAM content while VM is running.
4207 * The RAM content should appear first in the vmstate. So, we first
4208 * stash the non-RAM part of the vmstate to the temporary buffer,
4209 * then write RAM part of the vmstate to the migration stream
4210 * with vCPUs running and, finally, write stashed non-RAM part of
4211 * the vmstate from the buffer to the migration stream.
4213 s
->bioc
= qio_channel_buffer_new(512 * 1024);
4214 qio_channel_set_name(QIO_CHANNEL(s
->bioc
), "vmstate-buffer");
4215 fb
= qemu_file_new_output(QIO_CHANNEL(s
->bioc
));
4216 object_unref(OBJECT(s
->bioc
));
4218 update_iteration_initial_status(s
);
4221 * Prepare for tracking memory writes with UFFD-WP - populate
4222 * RAM pages before protecting.
4225 ram_write_tracking_prepare();
4228 qemu_savevm_state_header(s
->to_dst_file
);
4229 qemu_savevm_state_setup(s
->to_dst_file
);
4231 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
4232 MIGRATION_STATUS_ACTIVE
);
4234 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
4236 trace_migration_thread_setup_complete();
4237 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
4239 qemu_mutex_lock_iothread();
4242 * If VM is currently in suspended state, then, to make a valid runstate
4243 * transition in vm_stop_force_state() we need to wakeup it up.
4245 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
4246 s
->vm_was_running
= runstate_is_running();
4248 if (global_state_store()) {
4251 /* Forcibly stop VM before saving state of vCPUs and devices */
4252 if (vm_stop_force_state(RUN_STATE_PAUSED
)) {
4256 * Put vCPUs in sync with shadow context structures, then
4257 * save their state to channel-buffer along with devices.
4259 cpu_synchronize_all_states();
4260 if (qemu_savevm_state_complete_precopy_non_iterable(fb
, false, false)) {
4264 * Since we are going to get non-iterable state data directly
4265 * from s->bioc->data, explicit flush is needed here.
4269 /* Now initialize UFFD context and start tracking RAM writes */
4270 if (ram_write_tracking_start()) {
4276 * Start VM from BH handler to avoid write-fault lock here.
4277 * UFFD-WP protection for the whole RAM is already enabled so
4278 * calling VM state change notifiers from vm_start() would initiate
4279 * writes to virtio VQs memory which is in write-protected region.
4281 s
->vm_start_bh
= qemu_bh_new(bg_migration_vm_start_bh
, s
);
4282 qemu_bh_schedule(s
->vm_start_bh
);
4284 qemu_mutex_unlock_iothread();
4286 while (migration_is_active(s
)) {
4287 MigIterateState iter_state
= bg_migration_iteration_run(s
);
4288 if (iter_state
== MIG_ITERATE_SKIP
) {
4290 } else if (iter_state
== MIG_ITERATE_BREAK
) {
4295 * Try to detect any kind of failures, and see whether we
4296 * should stop the migration now.
4298 thr_error
= migration_detect_error(s
);
4299 if (thr_error
== MIG_THR_ERR_FATAL
) {
4300 /* Stop migration */
4304 migration_update_counters(s
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
));
4307 trace_migration_thread_after_loop();
4311 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
4312 MIGRATION_STATUS_FAILED
);
4313 qemu_mutex_unlock_iothread();
4316 bg_migration_iteration_finish(s
);
4319 object_unref(OBJECT(s
));
4320 rcu_unregister_thread();
4325 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
4327 Error
*local_err
= NULL
;
4329 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
4332 * If there's a previous error, free it and prepare for another one.
4333 * Meanwhile if migration completes successfully, there won't have an error
4334 * dumped when calling migrate_fd_cleanup().
4336 migrate_error_free(s
);
4338 s
->expected_downtime
= s
->parameters
.downtime_limit
;
4340 assert(s
->cleanup_bh
);
4342 assert(!s
->cleanup_bh
);
4343 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup_bh
, s
);
4346 migrate_fd_error(s
, error_in
);
4349 * Don't do cleanup for resume if channel is invalid, but only dump
4350 * the error. We wait for another channel connect from the user.
4351 * The error_report still gives HMP user a hint on what failed.
4352 * It's normally done in migrate_fd_cleanup(), but call it here
4355 error_report_err(error_copy(s
->error
));
4357 migrate_fd_cleanup(s
);
4363 /* This is a resumed migration */
4364 rate_limit
= s
->parameters
.max_postcopy_bandwidth
/
4367 /* This is a fresh new migration */
4368 rate_limit
= s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
;
4370 /* Notify before starting migration thread */
4371 notifier_list_notify(&migration_state_notifiers
, s
);
4374 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
4375 qemu_file_set_blocking(s
->to_dst_file
, true);
4378 * Open the return path. For postcopy, it is used exclusively. For
4379 * precopy, only if user specified "return-path" capability would
4380 * QEMU uses the return path.
4382 if (migrate_postcopy_ram() || migrate_use_return_path()) {
4383 if (open_return_path_on_source(s
, !resume
)) {
4384 error_report("Unable to open return-path for postcopy");
4385 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
4386 migrate_fd_cleanup(s
);
4392 /* Wakeup the main migration thread to do the recovery */
4393 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
4394 MIGRATION_STATUS_POSTCOPY_RECOVER
);
4395 qemu_sem_post(&s
->postcopy_pause_sem
);
4399 if (multifd_save_setup(&local_err
) != 0) {
4400 error_report_err(local_err
);
4401 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
4402 MIGRATION_STATUS_FAILED
);
4403 migrate_fd_cleanup(s
);
4407 if (migrate_background_snapshot()) {
4408 qemu_thread_create(&s
->thread
, "bg_snapshot",
4409 bg_migration_thread
, s
, QEMU_THREAD_JOINABLE
);
4411 qemu_thread_create(&s
->thread
, "live_migration",
4412 migration_thread
, s
, QEMU_THREAD_JOINABLE
);
4414 s
->migration_thread_running
= true;
4417 void migration_global_dump(Monitor
*mon
)
4419 MigrationState
*ms
= migrate_get_current();
4421 monitor_printf(mon
, "globals:\n");
4422 monitor_printf(mon
, "store-global-state: %s\n",
4423 ms
->store_global_state
? "on" : "off");
4424 monitor_printf(mon
, "only-migratable: %s\n",
4425 only_migratable
? "on" : "off");
4426 monitor_printf(mon
, "send-configuration: %s\n",
4427 ms
->send_configuration
? "on" : "off");
4428 monitor_printf(mon
, "send-section-footer: %s\n",
4429 ms
->send_section_footer
? "on" : "off");
4430 monitor_printf(mon
, "decompress-error-check: %s\n",
4431 ms
->decompress_error_check
? "on" : "off");
4432 monitor_printf(mon
, "clear-bitmap-shift: %u\n",
4433 ms
->clear_bitmap_shift
);
4436 #define DEFINE_PROP_MIG_CAP(name, x) \
4437 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
4439 static Property migration_properties
[] = {
4440 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
4441 store_global_state
, true),
4442 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
4443 send_configuration
, true),
4444 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
4445 send_section_footer
, true),
4446 DEFINE_PROP_BOOL("decompress-error-check", MigrationState
,
4447 decompress_error_check
, true),
4448 DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState
,
4449 clear_bitmap_shift
, CLEAR_BITMAP_SHIFT_DEFAULT
),
4450 DEFINE_PROP_BOOL("x-preempt-pre-7-2", MigrationState
,
4451 preempt_pre_7_2
, false),
4453 /* Migration parameters */
4454 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
4455 parameters
.compress_level
,
4456 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
4457 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
4458 parameters
.compress_threads
,
4459 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
4460 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState
,
4461 parameters
.compress_wait_thread
, true),
4462 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
4463 parameters
.decompress_threads
,
4464 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
4465 DEFINE_PROP_UINT8("x-throttle-trigger-threshold", MigrationState
,
4466 parameters
.throttle_trigger_threshold
,
4467 DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD
),
4468 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
4469 parameters
.cpu_throttle_initial
,
4470 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
4471 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
4472 parameters
.cpu_throttle_increment
,
4473 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
4474 DEFINE_PROP_BOOL("x-cpu-throttle-tailslow", MigrationState
,
4475 parameters
.cpu_throttle_tailslow
, false),
4476 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
4477 parameters
.max_bandwidth
, MAX_THROTTLE
),
4478 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
4479 parameters
.downtime_limit
,
4480 DEFAULT_MIGRATE_SET_DOWNTIME
),
4481 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
4482 parameters
.x_checkpoint_delay
,
4483 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
4484 DEFINE_PROP_UINT8("multifd-channels", MigrationState
,
4485 parameters
.multifd_channels
,
4486 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
4487 DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState
,
4488 parameters
.multifd_compression
,
4489 DEFAULT_MIGRATE_MULTIFD_COMPRESSION
),
4490 DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState
,
4491 parameters
.multifd_zlib_level
,
4492 DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL
),
4493 DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState
,
4494 parameters
.multifd_zstd_level
,
4495 DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL
),
4496 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
4497 parameters
.xbzrle_cache_size
,
4498 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
4499 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState
,
4500 parameters
.max_postcopy_bandwidth
,
4501 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH
),
4502 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState
,
4503 parameters
.max_cpu_throttle
,
4504 DEFAULT_MIGRATE_MAX_CPU_THROTTLE
),
4505 DEFINE_PROP_SIZE("announce-initial", MigrationState
,
4506 parameters
.announce_initial
,
4507 DEFAULT_MIGRATE_ANNOUNCE_INITIAL
),
4508 DEFINE_PROP_SIZE("announce-max", MigrationState
,
4509 parameters
.announce_max
,
4510 DEFAULT_MIGRATE_ANNOUNCE_MAX
),
4511 DEFINE_PROP_SIZE("announce-rounds", MigrationState
,
4512 parameters
.announce_rounds
,
4513 DEFAULT_MIGRATE_ANNOUNCE_ROUNDS
),
4514 DEFINE_PROP_SIZE("announce-step", MigrationState
,
4515 parameters
.announce_step
,
4516 DEFAULT_MIGRATE_ANNOUNCE_STEP
),
4517 DEFINE_PROP_STRING("tls-creds", MigrationState
, parameters
.tls_creds
),
4518 DEFINE_PROP_STRING("tls-hostname", MigrationState
, parameters
.tls_hostname
),
4519 DEFINE_PROP_STRING("tls-authz", MigrationState
, parameters
.tls_authz
),
4521 /* Migration capabilities */
4522 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
4523 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
4524 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
4525 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
4526 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
4527 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
4528 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
4529 DEFINE_PROP_MIG_CAP("x-postcopy-preempt",
4530 MIGRATION_CAPABILITY_POSTCOPY_PREEMPT
),
4531 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
4532 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
4533 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
4534 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
4535 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD
),
4536 DEFINE_PROP_MIG_CAP("x-background-snapshot",
4537 MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
),
4539 DEFINE_PROP_MIG_CAP("x-zero-copy-send",
4540 MIGRATION_CAPABILITY_ZERO_COPY_SEND
),
4543 DEFINE_PROP_END_OF_LIST(),
4546 static void migration_class_init(ObjectClass
*klass
, void *data
)
4548 DeviceClass
*dc
= DEVICE_CLASS(klass
);
4550 dc
->user_creatable
= false;
4551 device_class_set_props(dc
, migration_properties
);
4554 static void migration_instance_finalize(Object
*obj
)
4556 MigrationState
*ms
= MIGRATION_OBJ(obj
);
4558 qemu_mutex_destroy(&ms
->error_mutex
);
4559 qemu_mutex_destroy(&ms
->qemu_file_lock
);
4560 qemu_sem_destroy(&ms
->wait_unplug_sem
);
4561 qemu_sem_destroy(&ms
->rate_limit_sem
);
4562 qemu_sem_destroy(&ms
->pause_sem
);
4563 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
4564 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
4565 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
4566 qemu_sem_destroy(&ms
->rp_state
.rp_pong_acks
);
4567 qemu_sem_destroy(&ms
->postcopy_qemufile_src_sem
);
4568 error_free(ms
->error
);
4571 static void migration_instance_init(Object
*obj
)
4573 MigrationState
*ms
= MIGRATION_OBJ(obj
);
4574 MigrationParameters
*params
= &ms
->parameters
;
4576 ms
->state
= MIGRATION_STATUS_NONE
;
4578 ms
->pages_per_second
= -1;
4579 qemu_sem_init(&ms
->pause_sem
, 0);
4580 qemu_mutex_init(&ms
->error_mutex
);
4582 params
->tls_hostname
= g_strdup("");
4583 params
->tls_creds
= g_strdup("");
4585 /* Set has_* up only for parameter checks */
4586 params
->has_compress_level
= true;
4587 params
->has_compress_threads
= true;
4588 params
->has_compress_wait_thread
= true;
4589 params
->has_decompress_threads
= true;
4590 params
->has_throttle_trigger_threshold
= true;
4591 params
->has_cpu_throttle_initial
= true;
4592 params
->has_cpu_throttle_increment
= true;
4593 params
->has_cpu_throttle_tailslow
= true;
4594 params
->has_max_bandwidth
= true;
4595 params
->has_downtime_limit
= true;
4596 params
->has_x_checkpoint_delay
= true;
4597 params
->has_block_incremental
= true;
4598 params
->has_multifd_channels
= true;
4599 params
->has_multifd_compression
= true;
4600 params
->has_multifd_zlib_level
= true;
4601 params
->has_multifd_zstd_level
= true;
4602 params
->has_xbzrle_cache_size
= true;
4603 params
->has_max_postcopy_bandwidth
= true;
4604 params
->has_max_cpu_throttle
= true;
4605 params
->has_announce_initial
= true;
4606 params
->has_announce_max
= true;
4607 params
->has_announce_rounds
= true;
4608 params
->has_announce_step
= true;
4610 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
4611 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
4612 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
4613 qemu_sem_init(&ms
->rp_state
.rp_pong_acks
, 0);
4614 qemu_sem_init(&ms
->rate_limit_sem
, 0);
4615 qemu_sem_init(&ms
->wait_unplug_sem
, 0);
4616 qemu_sem_init(&ms
->postcopy_qemufile_src_sem
, 0);
4617 qemu_mutex_init(&ms
->qemu_file_lock
);
4621 * Return true if check pass, false otherwise. Error will be put
4622 * inside errp if provided.
4624 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
4626 MigrationCapabilityStatusList
*head
= NULL
;
4627 /* Assuming all off */
4628 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
4631 if (!migrate_params_check(&ms
->parameters
, errp
)) {
4635 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
4636 if (ms
->enabled_capabilities
[i
]) {
4637 QAPI_LIST_PREPEND(head
, migrate_cap_add(i
, true));
4641 ret
= migrate_caps_check(cap_list
, head
, errp
);
4643 /* It works with head == NULL */
4644 qapi_free_MigrationCapabilityStatusList(head
);
4649 static const TypeInfo migration_type
= {
4650 .name
= TYPE_MIGRATION
,
4652 * NOTE: TYPE_MIGRATION is not really a device, as the object is
4653 * not created using qdev_new(), it is not attached to the qdev
4654 * device tree, and it is never realized.
4656 * TODO: Make this TYPE_OBJECT once QOM provides something like
4657 * TYPE_DEVICE's "-global" properties.
4659 .parent
= TYPE_DEVICE
,
4660 .class_init
= migration_class_init
,
4661 .class_size
= sizeof(MigrationClass
),
4662 .instance_size
= sizeof(MigrationState
),
4663 .instance_init
= migration_instance_init
,
4664 .instance_finalize
= migration_instance_finalize
,
4667 static void register_migration_types(void)
4669 type_register_static(&migration_type
);
4672 type_init(register_migration_types
);