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-channel.h"
34 #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 "migration/colo.h"
53 #include "hw/boards.h"
54 #include "hw/qdev-properties.h"
55 #include "hw/qdev-properties-system.h"
56 #include "monitor/monitor.h"
57 #include "net/announce.h"
58 #include "qemu/queue.h"
60 #include "qemu/yank.h"
63 #include "hw/vfio/vfio-common.h"
66 #define MAX_THROTTLE (128 << 20) /* Migration transfer speed throttling */
68 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
70 #define BUFFER_DELAY 100
71 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
73 /* Time in milliseconds we are allowed to stop the source,
74 * for sending the last part */
75 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
77 /* Maximum migrate downtime set to 2000 seconds */
78 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
79 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
81 /* Default compression thread count */
82 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
83 /* Default decompression thread count, usually decompression is at
84 * least 4 times as fast as compression.*/
85 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
86 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
87 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
88 /* Define default autoconverge cpu throttle migration parameters */
89 #define DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD 50
90 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
91 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
92 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
94 /* Migration XBZRLE default cache size */
95 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
97 /* The delay time (in ms) between two COLO checkpoints */
98 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
99 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
100 #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE
101 /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */
102 #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
103 /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
104 #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
106 /* Background transfer rate for postcopy, 0 means unlimited, note
107 * that page requests can still exceed this limit.
109 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
112 * Parameters for self_announce_delay giving a stream of RARP/ARP
113 * packets after migration.
115 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
116 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
117 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
118 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
120 static NotifierList migration_state_notifiers
=
121 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
123 /* Messages sent on the return path from destination to source */
124 enum mig_rp_message_type
{
125 MIG_RP_MSG_INVALID
= 0, /* Must be 0 */
126 MIG_RP_MSG_SHUT
, /* sibling will not send any more RP messages */
127 MIG_RP_MSG_PONG
, /* Response to a PING; data (seq: be32 ) */
129 MIG_RP_MSG_REQ_PAGES_ID
, /* data (start: be64, len: be32, id: string) */
130 MIG_RP_MSG_REQ_PAGES
, /* data (start: be64, len: be32) */
131 MIG_RP_MSG_RECV_BITMAP
, /* send recved_bitmap back to source */
132 MIG_RP_MSG_RESUME_ACK
, /* tell source that we are ready to resume */
137 /* When we add fault tolerance, we could have several
138 migrations at once. For now we don't need to add
139 dynamic creation of migration */
141 static MigrationState
*current_migration
;
142 static MigrationIncomingState
*current_incoming
;
144 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
145 static int migration_maybe_pause(MigrationState
*s
,
146 int *current_active_state
,
148 static void migrate_fd_cancel(MigrationState
*s
);
150 static gint
page_request_addr_cmp(gconstpointer ap
, gconstpointer bp
)
152 uintptr_t a
= (uintptr_t) ap
, b
= (uintptr_t) bp
;
154 return (a
> b
) - (a
< b
);
157 void migration_object_init(void)
161 /* This can only be called once. */
162 assert(!current_migration
);
163 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
166 * Init the migrate incoming object as well no matter whether
167 * we'll use it or not.
169 assert(!current_incoming
);
170 current_incoming
= g_new0(MigrationIncomingState
, 1);
171 current_incoming
->state
= MIGRATION_STATUS_NONE
;
172 current_incoming
->postcopy_remote_fds
=
173 g_array_new(FALSE
, TRUE
, sizeof(struct PostCopyFD
));
174 qemu_mutex_init(¤t_incoming
->rp_mutex
);
175 qemu_event_init(¤t_incoming
->main_thread_load_event
, false);
176 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_dst
, 0);
177 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fault
, 0);
178 qemu_mutex_init(¤t_incoming
->page_request_mutex
);
179 current_incoming
->page_requested
= g_tree_new(page_request_addr_cmp
);
181 if (!migration_object_check(current_migration
, &err
)) {
182 error_report_err(err
);
188 dirty_bitmap_mig_init();
191 void migration_shutdown(void)
194 * Cancel the current migration - that will (eventually)
195 * stop the migration using this structure
197 migrate_fd_cancel(current_migration
);
198 object_unref(OBJECT(current_migration
));
201 * Cancel outgoing migration of dirty bitmaps. It should
202 * at least unref used block nodes.
204 dirty_bitmap_mig_cancel_outgoing();
207 * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
208 * are non-critical data, and their loss never considered as
211 dirty_bitmap_mig_cancel_incoming();
215 MigrationState
*migrate_get_current(void)
217 /* This can only be called after the object created. */
218 assert(current_migration
);
219 return current_migration
;
222 MigrationIncomingState
*migration_incoming_get_current(void)
224 assert(current_incoming
);
225 return current_incoming
;
228 void migration_incoming_state_destroy(void)
230 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
232 if (mis
->to_src_file
) {
233 /* Tell source that we are done */
234 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
235 qemu_fclose(mis
->to_src_file
);
236 mis
->to_src_file
= NULL
;
239 if (mis
->from_src_file
) {
240 qemu_fclose(mis
->from_src_file
);
241 mis
->from_src_file
= NULL
;
243 if (mis
->postcopy_remote_fds
) {
244 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
245 mis
->postcopy_remote_fds
= NULL
;
248 qemu_event_reset(&mis
->main_thread_load_event
);
250 if (mis
->page_requested
) {
251 g_tree_destroy(mis
->page_requested
);
252 mis
->page_requested
= NULL
;
255 if (mis
->socket_address_list
) {
256 qapi_free_SocketAddressList(mis
->socket_address_list
);
257 mis
->socket_address_list
= NULL
;
260 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
263 static void migrate_generate_event(int new_state
)
265 if (migrate_use_events()) {
266 qapi_event_send_migration(new_state
);
270 static bool migrate_late_block_activate(void)
274 s
= migrate_get_current();
276 return s
->enabled_capabilities
[
277 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE
];
281 * Send a message on the return channel back to the source
284 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
285 enum mig_rp_message_type message_type
,
286 uint16_t len
, void *data
)
290 trace_migrate_send_rp_message((int)message_type
, len
);
291 qemu_mutex_lock(&mis
->rp_mutex
);
294 * It's possible that the file handle got lost due to network
297 if (!mis
->to_src_file
) {
302 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
303 qemu_put_be16(mis
->to_src_file
, len
);
304 qemu_put_buffer(mis
->to_src_file
, data
, len
);
305 qemu_fflush(mis
->to_src_file
);
307 /* It's possible that qemu file got error during sending */
308 ret
= qemu_file_get_error(mis
->to_src_file
);
311 qemu_mutex_unlock(&mis
->rp_mutex
);
315 /* Request one page from the source VM at the given start address.
316 * rb: the RAMBlock to request the page in
317 * Start: Address offset within the RB
318 * Len: Length in bytes required - must be a multiple of pagesize
320 int migrate_send_rp_message_req_pages(MigrationIncomingState
*mis
,
321 RAMBlock
*rb
, ram_addr_t start
)
323 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
324 size_t msglen
= 12; /* start + len */
325 size_t len
= qemu_ram_pagesize(rb
);
326 enum mig_rp_message_type msg_type
;
330 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
331 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
334 * We maintain the last ramblock that we requested for page. Note that we
335 * don't need locking because this function will only be called within the
336 * postcopy ram fault thread.
338 if (rb
!= mis
->last_rb
) {
341 rbname
= qemu_ram_get_idstr(rb
);
342 rbname_len
= strlen(rbname
);
344 assert(rbname_len
< 256);
346 bufc
[msglen
++] = rbname_len
;
347 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
348 msglen
+= rbname_len
;
349 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
351 msg_type
= MIG_RP_MSG_REQ_PAGES
;
354 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
357 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
,
358 RAMBlock
*rb
, ram_addr_t start
, uint64_t haddr
)
360 void *aligned
= (void *)(uintptr_t)(haddr
& (-qemu_ram_pagesize(rb
)));
361 bool received
= false;
363 WITH_QEMU_LOCK_GUARD(&mis
->page_request_mutex
) {
364 received
= ramblock_recv_bitmap_test_byte_offset(rb
, start
);
365 if (!received
&& !g_tree_lookup(mis
->page_requested
, aligned
)) {
367 * The page has not been received, and it's not yet in the page
368 * request list. Queue it. Set the value of element to 1, so that
369 * things like g_tree_lookup() will return TRUE (1) when found.
371 g_tree_insert(mis
->page_requested
, aligned
, (gpointer
)1);
372 mis
->page_requested_count
++;
373 trace_postcopy_page_req_add(aligned
, mis
->page_requested_count
);
378 * If the page is there, skip sending the message. We don't even need the
379 * lock because as long as the page arrived, it'll be there forever.
385 return migrate_send_rp_message_req_pages(mis
, rb
, start
);
388 static bool migration_colo_enabled
;
389 bool migration_incoming_colo_enabled(void)
391 return migration_colo_enabled
;
394 void migration_incoming_disable_colo(void)
396 ram_block_discard_disable(false);
397 migration_colo_enabled
= false;
400 int migration_incoming_enable_colo(void)
402 if (ram_block_discard_disable(true)) {
403 error_report("COLO: cannot disable RAM discard");
406 migration_colo_enabled
= true;
410 void migrate_add_address(SocketAddress
*address
)
412 MigrationIncomingState
*mis
= migration_incoming_get_current();
414 QAPI_LIST_PREPEND(mis
->socket_address_list
,
415 QAPI_CLONE(SocketAddress
, address
));
418 static void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
420 const char *p
= NULL
;
422 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
426 qapi_event_send_migration(MIGRATION_STATUS_SETUP
);
427 if (strstart(uri
, "tcp:", &p
) ||
428 strstart(uri
, "unix:", NULL
) ||
429 strstart(uri
, "vsock:", NULL
)) {
430 socket_start_incoming_migration(p
? p
: uri
, errp
);
432 } else if (strstart(uri
, "rdma:", &p
)) {
433 rdma_start_incoming_migration(p
, errp
);
435 } else if (strstart(uri
, "exec:", &p
)) {
436 exec_start_incoming_migration(p
, errp
);
437 } else if (strstart(uri
, "fd:", &p
)) {
438 fd_start_incoming_migration(p
, errp
);
440 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
441 error_setg(errp
, "unknown migration protocol: %s", uri
);
445 static void process_incoming_migration_bh(void *opaque
)
447 Error
*local_err
= NULL
;
448 MigrationIncomingState
*mis
= opaque
;
450 /* If capability late_block_activate is set:
451 * Only fire up the block code now if we're going to restart the
452 * VM, else 'cont' will do it.
453 * This causes file locking to happen; so we don't want it to happen
454 * unless we really are starting the VM.
456 if (!migrate_late_block_activate() ||
457 (autostart
&& (!global_state_received() ||
458 global_state_get_runstate() == RUN_STATE_RUNNING
))) {
459 /* Make sure all file formats flush their mutable metadata.
460 * If we get an error here, just don't restart the VM yet. */
461 bdrv_invalidate_cache_all(&local_err
);
463 error_report_err(local_err
);
470 * This must happen after all error conditions are dealt with and
471 * we're sure the VM is going to be running on this host.
473 qemu_announce_self(&mis
->announce_timer
, migrate_announce_params());
475 if (multifd_load_cleanup(&local_err
) != 0) {
476 error_report_err(local_err
);
479 /* If global state section was not received or we are in running
480 state, we need to obey autostart. Any other state is set with
483 dirty_bitmap_mig_before_vm_start();
485 if (!global_state_received() ||
486 global_state_get_runstate() == RUN_STATE_RUNNING
) {
490 runstate_set(RUN_STATE_PAUSED
);
492 } else if (migration_incoming_colo_enabled()) {
493 migration_incoming_disable_colo();
496 runstate_set(global_state_get_runstate());
499 * This must happen after any state changes since as soon as an external
500 * observer sees this event they might start to prod at the VM assuming
503 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
504 MIGRATION_STATUS_COMPLETED
);
505 qemu_bh_delete(mis
->bh
);
506 migration_incoming_state_destroy();
509 static void process_incoming_migration_co(void *opaque
)
511 MigrationIncomingState
*mis
= migration_incoming_get_current();
514 Error
*local_err
= NULL
;
516 assert(mis
->from_src_file
);
517 mis
->migration_incoming_co
= qemu_coroutine_self();
518 mis
->largest_page_size
= qemu_ram_pagesize_largest();
519 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
520 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
521 MIGRATION_STATUS_ACTIVE
);
522 ret
= qemu_loadvm_state(mis
->from_src_file
);
524 ps
= postcopy_state_get();
525 trace_process_incoming_migration_co_end(ret
, ps
);
526 if (ps
!= POSTCOPY_INCOMING_NONE
) {
527 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
529 * Where a migration had postcopy enabled (and thus went to advise)
530 * but managed to complete within the precopy period, we can use
533 postcopy_ram_incoming_cleanup(mis
);
534 } else if (ret
>= 0) {
536 * Postcopy was started, cleanup should happen at the end of the
539 trace_process_incoming_migration_co_postcopy_end_main();
542 /* Else if something went wrong then just fall out of the normal exit */
545 /* we get COLO info, and know if we are in COLO mode */
546 if (!ret
&& migration_incoming_colo_enabled()) {
547 /* Make sure all file formats flush their mutable metadata */
548 bdrv_invalidate_cache_all(&local_err
);
550 error_report_err(local_err
);
554 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
555 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
556 mis
->have_colo_incoming_thread
= true;
557 qemu_coroutine_yield();
559 /* Wait checkpoint incoming thread exit before free resource */
560 qemu_thread_join(&mis
->colo_incoming_thread
);
561 /* We hold the global iothread lock, so it is safe here */
562 colo_release_ram_cache();
566 error_report("load of migration failed: %s", strerror(-ret
));
569 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
570 qemu_bh_schedule(mis
->bh
);
571 mis
->migration_incoming_co
= NULL
;
575 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
576 MIGRATION_STATUS_FAILED
);
577 qemu_fclose(mis
->from_src_file
);
578 if (multifd_load_cleanup(&local_err
) != 0) {
579 error_report_err(local_err
);
585 * @migration_incoming_setup: Setup incoming migration
587 * Returns 0 for no error or 1 for error
589 * @f: file for main migration channel
590 * @errp: where to put errors
592 static int migration_incoming_setup(QEMUFile
*f
, Error
**errp
)
594 MigrationIncomingState
*mis
= migration_incoming_get_current();
595 Error
*local_err
= NULL
;
597 if (multifd_load_setup(&local_err
) != 0) {
598 /* We haven't been able to create multifd threads
599 nothing better to do */
600 error_report_err(local_err
);
604 if (!mis
->from_src_file
) {
605 mis
->from_src_file
= f
;
607 qemu_file_set_blocking(f
, false);
611 void migration_incoming_process(void)
613 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
614 qemu_coroutine_enter(co
);
617 /* Returns true if recovered from a paused migration, otherwise false */
618 static bool postcopy_try_recover(QEMUFile
*f
)
620 MigrationIncomingState
*mis
= migration_incoming_get_current();
622 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
623 /* Resumed from a paused postcopy migration */
625 mis
->from_src_file
= f
;
626 /* Postcopy has standalone thread to do vm load */
627 qemu_file_set_blocking(f
, true);
629 /* Re-configure the return path */
630 mis
->to_src_file
= qemu_file_get_return_path(f
);
632 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
633 MIGRATION_STATUS_POSTCOPY_RECOVER
);
636 * Here, we only wake up the main loading thread (while the
637 * fault thread will still be waiting), so that we can receive
638 * commands from source now, and answer it if needed. The
639 * fault thread will be woken up afterwards until we are sure
640 * that source is ready to reply to page requests.
642 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
649 void migration_fd_process_incoming(QEMUFile
*f
, Error
**errp
)
651 Error
*local_err
= NULL
;
653 if (postcopy_try_recover(f
)) {
657 if (migration_incoming_setup(f
, &local_err
)) {
658 error_propagate(errp
, local_err
);
661 migration_incoming_process();
664 void migration_ioc_process_incoming(QIOChannel
*ioc
, Error
**errp
)
666 MigrationIncomingState
*mis
= migration_incoming_get_current();
667 Error
*local_err
= NULL
;
668 bool start_migration
;
670 if (!mis
->from_src_file
) {
671 /* The first connection (multifd may have multiple) */
672 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
674 /* If it's a recovery, we're done */
675 if (postcopy_try_recover(f
)) {
679 if (migration_incoming_setup(f
, &local_err
)) {
680 error_propagate(errp
, local_err
);
685 * Common migration only needs one channel, so we can start
686 * right now. Multifd needs more than one channel, we wait.
688 start_migration
= !migrate_use_multifd();
690 /* Multiple connections */
691 assert(migrate_use_multifd());
692 start_migration
= multifd_recv_new_channel(ioc
, &local_err
);
694 error_propagate(errp
, local_err
);
699 if (start_migration
) {
700 migration_incoming_process();
705 * @migration_has_all_channels: We have received all channels that we need
707 * Returns true when we have got connections to all the channels that
708 * we need for migration.
710 bool migration_has_all_channels(void)
712 MigrationIncomingState
*mis
= migration_incoming_get_current();
715 all_channels
= multifd_recv_all_channels_created();
717 return all_channels
&& mis
->from_src_file
!= NULL
;
721 * Send a 'SHUT' message on the return channel with the given value
722 * to indicate that we've finished with the RP. Non-0 value indicates
725 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
730 buf
= cpu_to_be32(value
);
731 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
735 * Send a 'PONG' message on the return channel with the given value
736 * (normally in response to a 'PING')
738 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
743 buf
= cpu_to_be32(value
);
744 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
747 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
755 * First, we send the header part. It contains only the len of
756 * idstr, and the idstr itself.
758 len
= strlen(block_name
);
760 memcpy(buf
+ 1, block_name
, len
);
762 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
763 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
768 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
771 * Next, we dump the received bitmap to the stream.
773 * TODO: currently we are safe since we are the only one that is
774 * using the to_src_file handle (fault thread is still paused),
775 * and it's ok even not taking the mutex. However the best way is
776 * to take the lock before sending the message header, and release
777 * the lock after sending the bitmap.
779 qemu_mutex_lock(&mis
->rp_mutex
);
780 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
781 qemu_mutex_unlock(&mis
->rp_mutex
);
783 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
786 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
790 buf
= cpu_to_be32(value
);
791 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
794 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
796 MigrationCapabilityStatusList
*head
= NULL
, **tail
= &head
;
797 MigrationCapabilityStatus
*caps
;
798 MigrationState
*s
= migrate_get_current();
801 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
802 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
803 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
807 caps
= g_malloc0(sizeof(*caps
));
808 caps
->capability
= i
;
809 caps
->state
= s
->enabled_capabilities
[i
];
810 QAPI_LIST_APPEND(tail
, caps
);
816 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
818 MigrationParameters
*params
;
819 MigrationState
*s
= migrate_get_current();
821 /* TODO use QAPI_CLONE() instead of duplicating it inline */
822 params
= g_malloc0(sizeof(*params
));
823 params
->has_compress_level
= true;
824 params
->compress_level
= s
->parameters
.compress_level
;
825 params
->has_compress_threads
= true;
826 params
->compress_threads
= s
->parameters
.compress_threads
;
827 params
->has_compress_wait_thread
= true;
828 params
->compress_wait_thread
= s
->parameters
.compress_wait_thread
;
829 params
->has_decompress_threads
= true;
830 params
->decompress_threads
= s
->parameters
.decompress_threads
;
831 params
->has_throttle_trigger_threshold
= true;
832 params
->throttle_trigger_threshold
= s
->parameters
.throttle_trigger_threshold
;
833 params
->has_cpu_throttle_initial
= true;
834 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
835 params
->has_cpu_throttle_increment
= true;
836 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
837 params
->has_cpu_throttle_tailslow
= true;
838 params
->cpu_throttle_tailslow
= s
->parameters
.cpu_throttle_tailslow
;
839 params
->has_tls_creds
= true;
840 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
841 params
->has_tls_hostname
= true;
842 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
843 params
->has_tls_authz
= true;
844 params
->tls_authz
= g_strdup(s
->parameters
.tls_authz
?
845 s
->parameters
.tls_authz
: "");
846 params
->has_max_bandwidth
= true;
847 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
848 params
->has_downtime_limit
= true;
849 params
->downtime_limit
= s
->parameters
.downtime_limit
;
850 params
->has_x_checkpoint_delay
= true;
851 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
852 params
->has_block_incremental
= true;
853 params
->block_incremental
= s
->parameters
.block_incremental
;
854 params
->has_multifd_channels
= true;
855 params
->multifd_channels
= s
->parameters
.multifd_channels
;
856 params
->has_multifd_compression
= true;
857 params
->multifd_compression
= s
->parameters
.multifd_compression
;
858 params
->has_multifd_zlib_level
= true;
859 params
->multifd_zlib_level
= s
->parameters
.multifd_zlib_level
;
860 params
->has_multifd_zstd_level
= true;
861 params
->multifd_zstd_level
= s
->parameters
.multifd_zstd_level
;
862 params
->has_xbzrle_cache_size
= true;
863 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
864 params
->has_max_postcopy_bandwidth
= true;
865 params
->max_postcopy_bandwidth
= s
->parameters
.max_postcopy_bandwidth
;
866 params
->has_max_cpu_throttle
= true;
867 params
->max_cpu_throttle
= s
->parameters
.max_cpu_throttle
;
868 params
->has_announce_initial
= true;
869 params
->announce_initial
= s
->parameters
.announce_initial
;
870 params
->has_announce_max
= true;
871 params
->announce_max
= s
->parameters
.announce_max
;
872 params
->has_announce_rounds
= true;
873 params
->announce_rounds
= s
->parameters
.announce_rounds
;
874 params
->has_announce_step
= true;
875 params
->announce_step
= s
->parameters
.announce_step
;
877 if (s
->parameters
.has_block_bitmap_mapping
) {
878 params
->has_block_bitmap_mapping
= true;
879 params
->block_bitmap_mapping
=
880 QAPI_CLONE(BitmapMigrationNodeAliasList
,
881 s
->parameters
.block_bitmap_mapping
);
887 AnnounceParameters
*migrate_announce_params(void)
889 static AnnounceParameters ap
;
891 MigrationState
*s
= migrate_get_current();
893 ap
.initial
= s
->parameters
.announce_initial
;
894 ap
.max
= s
->parameters
.announce_max
;
895 ap
.rounds
= s
->parameters
.announce_rounds
;
896 ap
.step
= s
->parameters
.announce_step
;
902 * Return true if we're already in the middle of a migration
903 * (i.e. any of the active or setup states)
905 bool migration_is_setup_or_active(int state
)
908 case MIGRATION_STATUS_ACTIVE
:
909 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
910 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
911 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
912 case MIGRATION_STATUS_SETUP
:
913 case MIGRATION_STATUS_PRE_SWITCHOVER
:
914 case MIGRATION_STATUS_DEVICE
:
915 case MIGRATION_STATUS_WAIT_UNPLUG
:
916 case MIGRATION_STATUS_COLO
:
925 bool migration_is_running(int state
)
928 case MIGRATION_STATUS_ACTIVE
:
929 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
930 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
931 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
932 case MIGRATION_STATUS_SETUP
:
933 case MIGRATION_STATUS_PRE_SWITCHOVER
:
934 case MIGRATION_STATUS_DEVICE
:
935 case MIGRATION_STATUS_WAIT_UNPLUG
:
936 case MIGRATION_STATUS_CANCELLING
:
945 static void populate_time_info(MigrationInfo
*info
, MigrationState
*s
)
947 info
->has_status
= true;
948 info
->has_setup_time
= true;
949 info
->setup_time
= s
->setup_time
;
950 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
951 info
->has_total_time
= true;
952 info
->total_time
= s
->total_time
;
953 info
->has_downtime
= true;
954 info
->downtime
= s
->downtime
;
956 info
->has_total_time
= true;
957 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) -
959 info
->has_expected_downtime
= true;
960 info
->expected_downtime
= s
->expected_downtime
;
964 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
966 info
->has_ram
= true;
967 info
->ram
= g_malloc0(sizeof(*info
->ram
));
968 info
->ram
->transferred
= ram_counters
.transferred
;
969 info
->ram
->total
= ram_bytes_total();
970 info
->ram
->duplicate
= ram_counters
.duplicate
;
971 /* legacy value. It is not used anymore */
972 info
->ram
->skipped
= 0;
973 info
->ram
->normal
= ram_counters
.normal
;
974 info
->ram
->normal_bytes
= ram_counters
.normal
*
975 qemu_target_page_size();
976 info
->ram
->mbps
= s
->mbps
;
977 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
978 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
979 info
->ram
->page_size
= qemu_target_page_size();
980 info
->ram
->multifd_bytes
= ram_counters
.multifd_bytes
;
981 info
->ram
->pages_per_second
= s
->pages_per_second
;
983 if (migrate_use_xbzrle()) {
984 info
->has_xbzrle_cache
= true;
985 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
986 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
987 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
988 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
989 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
990 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
991 info
->xbzrle_cache
->encoding_rate
= xbzrle_counters
.encoding_rate
;
992 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
995 if (migrate_use_compression()) {
996 info
->has_compression
= true;
997 info
->compression
= g_malloc0(sizeof(*info
->compression
));
998 info
->compression
->pages
= compression_counters
.pages
;
999 info
->compression
->busy
= compression_counters
.busy
;
1000 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
1001 info
->compression
->compressed_size
=
1002 compression_counters
.compressed_size
;
1003 info
->compression
->compression_rate
=
1004 compression_counters
.compression_rate
;
1007 if (cpu_throttle_active()) {
1008 info
->has_cpu_throttle_percentage
= true;
1009 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
1012 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
1013 info
->ram
->remaining
= ram_bytes_remaining();
1014 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
1018 static void populate_disk_info(MigrationInfo
*info
)
1020 if (blk_mig_active()) {
1021 info
->has_disk
= true;
1022 info
->disk
= g_malloc0(sizeof(*info
->disk
));
1023 info
->disk
->transferred
= blk_mig_bytes_transferred();
1024 info
->disk
->remaining
= blk_mig_bytes_remaining();
1025 info
->disk
->total
= blk_mig_bytes_total();
1029 static void populate_vfio_info(MigrationInfo
*info
)
1032 if (vfio_mig_active()) {
1033 info
->has_vfio
= true;
1034 info
->vfio
= g_malloc0(sizeof(*info
->vfio
));
1035 info
->vfio
->transferred
= vfio_mig_bytes_transferred();
1040 static void fill_source_migration_info(MigrationInfo
*info
)
1042 MigrationState
*s
= migrate_get_current();
1045 case MIGRATION_STATUS_NONE
:
1046 /* no migration has happened ever */
1047 /* do not overwrite destination migration status */
1049 case MIGRATION_STATUS_SETUP
:
1050 info
->has_status
= true;
1051 info
->has_total_time
= false;
1053 case MIGRATION_STATUS_ACTIVE
:
1054 case MIGRATION_STATUS_CANCELLING
:
1055 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1056 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1057 case MIGRATION_STATUS_DEVICE
:
1058 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1059 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1060 /* TODO add some postcopy stats */
1061 populate_time_info(info
, s
);
1062 populate_ram_info(info
, s
);
1063 populate_disk_info(info
);
1064 populate_vfio_info(info
);
1066 case MIGRATION_STATUS_COLO
:
1067 info
->has_status
= true;
1068 /* TODO: display COLO specific information (checkpoint info etc.) */
1070 case MIGRATION_STATUS_COMPLETED
:
1071 populate_time_info(info
, s
);
1072 populate_ram_info(info
, s
);
1073 populate_vfio_info(info
);
1075 case MIGRATION_STATUS_FAILED
:
1076 info
->has_status
= true;
1078 info
->has_error_desc
= true;
1079 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
1082 case MIGRATION_STATUS_CANCELLED
:
1083 info
->has_status
= true;
1085 case MIGRATION_STATUS_WAIT_UNPLUG
:
1086 info
->has_status
= true;
1089 info
->status
= s
->state
;
1093 * @migration_caps_check - check capability validity
1095 * @cap_list: old capability list, array of bool
1096 * @params: new capabilities to be applied soon
1097 * @errp: set *errp if the check failed, with reason
1099 * Returns true if check passed, otherwise false.
1101 static bool migrate_caps_check(bool *cap_list
,
1102 MigrationCapabilityStatusList
*params
,
1105 MigrationCapabilityStatusList
*cap
;
1106 bool old_postcopy_cap
;
1107 MigrationIncomingState
*mis
= migration_incoming_get_current();
1109 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1111 for (cap
= params
; cap
; cap
= cap
->next
) {
1112 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
1115 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1116 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
1117 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
1119 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
1124 #ifndef CONFIG_REPLICATION
1125 if (cap_list
[MIGRATION_CAPABILITY_X_COLO
]) {
1126 error_setg(errp
, "QEMU compiled without replication module"
1127 " can't enable COLO");
1128 error_append_hint(errp
, "Please enable replication before COLO.\n");
1133 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
1134 /* This check is reasonably expensive, so only when it's being
1135 * set the first time, also it's only the destination that needs
1138 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
1139 !postcopy_ram_supported_by_host(mis
)) {
1140 /* postcopy_ram_supported_by_host will have emitted a more
1143 error_setg(errp
, "Postcopy is not supported");
1147 if (cap_list
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
]) {
1148 error_setg(errp
, "Postcopy is not compatible with ignore-shared");
1156 static void fill_destination_migration_info(MigrationInfo
*info
)
1158 MigrationIncomingState
*mis
= migration_incoming_get_current();
1160 if (mis
->socket_address_list
) {
1161 info
->has_socket_address
= true;
1162 info
->socket_address
=
1163 QAPI_CLONE(SocketAddressList
, mis
->socket_address_list
);
1166 switch (mis
->state
) {
1167 case MIGRATION_STATUS_NONE
:
1169 case MIGRATION_STATUS_SETUP
:
1170 case MIGRATION_STATUS_CANCELLING
:
1171 case MIGRATION_STATUS_CANCELLED
:
1172 case MIGRATION_STATUS_ACTIVE
:
1173 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1174 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1175 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1176 case MIGRATION_STATUS_FAILED
:
1177 case MIGRATION_STATUS_COLO
:
1178 info
->has_status
= true;
1180 case MIGRATION_STATUS_COMPLETED
:
1181 info
->has_status
= true;
1182 fill_destination_postcopy_migration_info(info
);
1185 info
->status
= mis
->state
;
1188 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1190 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1192 fill_destination_migration_info(info
);
1193 fill_source_migration_info(info
);
1198 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
1201 MigrationState
*s
= migrate_get_current();
1202 MigrationCapabilityStatusList
*cap
;
1203 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
1205 if (migration_is_running(s
->state
)) {
1206 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1210 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
1211 if (!migrate_caps_check(cap_list
, params
, errp
)) {
1215 for (cap
= params
; cap
; cap
= cap
->next
) {
1216 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
1221 * Check whether the parameters are valid. Error will be put into errp
1222 * (if provided). Return true if valid, otherwise false.
1224 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
1226 if (params
->has_compress_level
&&
1227 (params
->compress_level
> 9)) {
1228 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
1229 "is invalid, it should be in the range of 0 to 9");
1233 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
1234 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1236 "is invalid, it should be in the range of 1 to 255");
1240 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
1241 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1242 "decompress_threads",
1243 "is invalid, it should be in the range of 1 to 255");
1247 if (params
->has_throttle_trigger_threshold
&&
1248 (params
->throttle_trigger_threshold
< 1 ||
1249 params
->throttle_trigger_threshold
> 100)) {
1250 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1251 "throttle_trigger_threshold",
1252 "an integer in the range of 1 to 100");
1256 if (params
->has_cpu_throttle_initial
&&
1257 (params
->cpu_throttle_initial
< 1 ||
1258 params
->cpu_throttle_initial
> 99)) {
1259 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1260 "cpu_throttle_initial",
1261 "an integer in the range of 1 to 99");
1265 if (params
->has_cpu_throttle_increment
&&
1266 (params
->cpu_throttle_increment
< 1 ||
1267 params
->cpu_throttle_increment
> 99)) {
1268 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1269 "cpu_throttle_increment",
1270 "an integer in the range of 1 to 99");
1274 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
1275 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1277 "an integer in the range of 0 to "stringify(SIZE_MAX
)
1282 if (params
->has_downtime_limit
&&
1283 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
1284 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1286 "an integer in the range of 0 to "
1287 stringify(MAX_MIGRATE_DOWNTIME
)" ms");
1291 /* x_checkpoint_delay is now always positive */
1293 if (params
->has_multifd_channels
&& (params
->multifd_channels
< 1)) {
1294 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1296 "is invalid, it should be in the range of 1 to 255");
1300 if (params
->has_multifd_zlib_level
&&
1301 (params
->multifd_zlib_level
> 9)) {
1302 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zlib_level",
1303 "is invalid, it should be in the range of 0 to 9");
1307 if (params
->has_multifd_zstd_level
&&
1308 (params
->multifd_zstd_level
> 20)) {
1309 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zstd_level",
1310 "is invalid, it should be in the range of 0 to 20");
1314 if (params
->has_xbzrle_cache_size
&&
1315 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
1316 !is_power_of_2(params
->xbzrle_cache_size
))) {
1317 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1318 "xbzrle_cache_size",
1319 "is invalid, it should be bigger than target page size"
1320 " and a power of 2");
1324 if (params
->has_max_cpu_throttle
&&
1325 (params
->max_cpu_throttle
< params
->cpu_throttle_initial
||
1326 params
->max_cpu_throttle
> 99)) {
1327 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1329 "an integer in the range of cpu_throttle_initial to 99");
1333 if (params
->has_announce_initial
&&
1334 params
->announce_initial
> 100000) {
1335 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1337 "is invalid, it must be less than 100000 ms");
1340 if (params
->has_announce_max
&&
1341 params
->announce_max
> 100000) {
1342 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1344 "is invalid, it must be less than 100000 ms");
1347 if (params
->has_announce_rounds
&&
1348 params
->announce_rounds
> 1000) {
1349 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1351 "is invalid, it must be in the range of 0 to 1000");
1354 if (params
->has_announce_step
&&
1355 (params
->announce_step
< 1 ||
1356 params
->announce_step
> 10000)) {
1357 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1359 "is invalid, it must be in the range of 1 to 10000 ms");
1363 if (params
->has_block_bitmap_mapping
&&
1364 !check_dirty_bitmap_mig_alias_map(params
->block_bitmap_mapping
, errp
)) {
1365 error_prepend(errp
, "Invalid mapping given for block-bitmap-mapping: ");
1372 static void migrate_params_test_apply(MigrateSetParameters
*params
,
1373 MigrationParameters
*dest
)
1375 *dest
= migrate_get_current()->parameters
;
1377 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1379 if (params
->has_compress_level
) {
1380 dest
->compress_level
= params
->compress_level
;
1383 if (params
->has_compress_threads
) {
1384 dest
->compress_threads
= params
->compress_threads
;
1387 if (params
->has_compress_wait_thread
) {
1388 dest
->compress_wait_thread
= params
->compress_wait_thread
;
1391 if (params
->has_decompress_threads
) {
1392 dest
->decompress_threads
= params
->decompress_threads
;
1395 if (params
->has_throttle_trigger_threshold
) {
1396 dest
->throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1399 if (params
->has_cpu_throttle_initial
) {
1400 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
1403 if (params
->has_cpu_throttle_increment
) {
1404 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
1407 if (params
->has_cpu_throttle_tailslow
) {
1408 dest
->cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1411 if (params
->has_tls_creds
) {
1412 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1413 dest
->tls_creds
= params
->tls_creds
->u
.s
;
1416 if (params
->has_tls_hostname
) {
1417 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1418 dest
->tls_hostname
= params
->tls_hostname
->u
.s
;
1421 if (params
->has_max_bandwidth
) {
1422 dest
->max_bandwidth
= params
->max_bandwidth
;
1425 if (params
->has_downtime_limit
) {
1426 dest
->downtime_limit
= params
->downtime_limit
;
1429 if (params
->has_x_checkpoint_delay
) {
1430 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
1433 if (params
->has_block_incremental
) {
1434 dest
->block_incremental
= params
->block_incremental
;
1436 if (params
->has_multifd_channels
) {
1437 dest
->multifd_channels
= params
->multifd_channels
;
1439 if (params
->has_multifd_compression
) {
1440 dest
->multifd_compression
= params
->multifd_compression
;
1442 if (params
->has_xbzrle_cache_size
) {
1443 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
1445 if (params
->has_max_postcopy_bandwidth
) {
1446 dest
->max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1448 if (params
->has_max_cpu_throttle
) {
1449 dest
->max_cpu_throttle
= params
->max_cpu_throttle
;
1451 if (params
->has_announce_initial
) {
1452 dest
->announce_initial
= params
->announce_initial
;
1454 if (params
->has_announce_max
) {
1455 dest
->announce_max
= params
->announce_max
;
1457 if (params
->has_announce_rounds
) {
1458 dest
->announce_rounds
= params
->announce_rounds
;
1460 if (params
->has_announce_step
) {
1461 dest
->announce_step
= params
->announce_step
;
1464 if (params
->has_block_bitmap_mapping
) {
1465 dest
->has_block_bitmap_mapping
= true;
1466 dest
->block_bitmap_mapping
= params
->block_bitmap_mapping
;
1470 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
1472 MigrationState
*s
= migrate_get_current();
1474 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1476 if (params
->has_compress_level
) {
1477 s
->parameters
.compress_level
= params
->compress_level
;
1480 if (params
->has_compress_threads
) {
1481 s
->parameters
.compress_threads
= params
->compress_threads
;
1484 if (params
->has_compress_wait_thread
) {
1485 s
->parameters
.compress_wait_thread
= params
->compress_wait_thread
;
1488 if (params
->has_decompress_threads
) {
1489 s
->parameters
.decompress_threads
= params
->decompress_threads
;
1492 if (params
->has_throttle_trigger_threshold
) {
1493 s
->parameters
.throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1496 if (params
->has_cpu_throttle_initial
) {
1497 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
1500 if (params
->has_cpu_throttle_increment
) {
1501 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
1504 if (params
->has_cpu_throttle_tailslow
) {
1505 s
->parameters
.cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1508 if (params
->has_tls_creds
) {
1509 g_free(s
->parameters
.tls_creds
);
1510 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1511 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1514 if (params
->has_tls_hostname
) {
1515 g_free(s
->parameters
.tls_hostname
);
1516 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1517 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1520 if (params
->has_tls_authz
) {
1521 g_free(s
->parameters
.tls_authz
);
1522 assert(params
->tls_authz
->type
== QTYPE_QSTRING
);
1523 s
->parameters
.tls_authz
= g_strdup(params
->tls_authz
->u
.s
);
1526 if (params
->has_max_bandwidth
) {
1527 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1528 if (s
->to_dst_file
&& !migration_in_postcopy()) {
1529 qemu_file_set_rate_limit(s
->to_dst_file
,
1530 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1534 if (params
->has_downtime_limit
) {
1535 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1538 if (params
->has_x_checkpoint_delay
) {
1539 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1540 if (migration_in_colo_state()) {
1541 colo_checkpoint_notify(s
);
1545 if (params
->has_block_incremental
) {
1546 s
->parameters
.block_incremental
= params
->block_incremental
;
1548 if (params
->has_multifd_channels
) {
1549 s
->parameters
.multifd_channels
= params
->multifd_channels
;
1551 if (params
->has_multifd_compression
) {
1552 s
->parameters
.multifd_compression
= params
->multifd_compression
;
1554 if (params
->has_xbzrle_cache_size
) {
1555 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1556 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1558 if (params
->has_max_postcopy_bandwidth
) {
1559 s
->parameters
.max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1560 if (s
->to_dst_file
&& migration_in_postcopy()) {
1561 qemu_file_set_rate_limit(s
->to_dst_file
,
1562 s
->parameters
.max_postcopy_bandwidth
/ XFER_LIMIT_RATIO
);
1565 if (params
->has_max_cpu_throttle
) {
1566 s
->parameters
.max_cpu_throttle
= params
->max_cpu_throttle
;
1568 if (params
->has_announce_initial
) {
1569 s
->parameters
.announce_initial
= params
->announce_initial
;
1571 if (params
->has_announce_max
) {
1572 s
->parameters
.announce_max
= params
->announce_max
;
1574 if (params
->has_announce_rounds
) {
1575 s
->parameters
.announce_rounds
= params
->announce_rounds
;
1577 if (params
->has_announce_step
) {
1578 s
->parameters
.announce_step
= params
->announce_step
;
1581 if (params
->has_block_bitmap_mapping
) {
1582 qapi_free_BitmapMigrationNodeAliasList(
1583 s
->parameters
.block_bitmap_mapping
);
1585 s
->parameters
.has_block_bitmap_mapping
= true;
1586 s
->parameters
.block_bitmap_mapping
=
1587 QAPI_CLONE(BitmapMigrationNodeAliasList
,
1588 params
->block_bitmap_mapping
);
1592 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1594 MigrationParameters tmp
;
1596 /* TODO Rewrite "" to null instead */
1597 if (params
->has_tls_creds
1598 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1599 qobject_unref(params
->tls_creds
->u
.n
);
1600 params
->tls_creds
->type
= QTYPE_QSTRING
;
1601 params
->tls_creds
->u
.s
= strdup("");
1603 /* TODO Rewrite "" to null instead */
1604 if (params
->has_tls_hostname
1605 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1606 qobject_unref(params
->tls_hostname
->u
.n
);
1607 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1608 params
->tls_hostname
->u
.s
= strdup("");
1611 migrate_params_test_apply(params
, &tmp
);
1613 if (!migrate_params_check(&tmp
, errp
)) {
1614 /* Invalid parameter */
1618 migrate_params_apply(params
, errp
);
1622 void qmp_migrate_start_postcopy(Error
**errp
)
1624 MigrationState
*s
= migrate_get_current();
1626 if (!migrate_postcopy()) {
1627 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1628 " the start of migration");
1632 if (s
->state
== MIGRATION_STATUS_NONE
) {
1633 error_setg(errp
, "Postcopy must be started after migration has been"
1638 * we don't error if migration has finished since that would be racy
1639 * with issuing this command.
1641 qatomic_set(&s
->start_postcopy
, true);
1644 /* shared migration helpers */
1646 void migrate_set_state(int *state
, int old_state
, int new_state
)
1648 assert(new_state
< MIGRATION_STATUS__MAX
);
1649 if (qatomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1650 trace_migrate_set_state(MigrationStatus_str(new_state
));
1651 migrate_generate_event(new_state
);
1655 static MigrationCapabilityStatus
*migrate_cap_add(MigrationCapability index
,
1658 MigrationCapabilityStatus
*cap
;
1660 cap
= g_new0(MigrationCapabilityStatus
, 1);
1661 cap
->capability
= index
;
1667 void migrate_set_block_enabled(bool value
, Error
**errp
)
1669 MigrationCapabilityStatusList
*cap
= NULL
;
1671 QAPI_LIST_PREPEND(cap
, migrate_cap_add(MIGRATION_CAPABILITY_BLOCK
, value
));
1672 qmp_migrate_set_capabilities(cap
, errp
);
1673 qapi_free_MigrationCapabilityStatusList(cap
);
1676 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1678 s
->parameters
.block_incremental
= value
;
1681 static void block_cleanup_parameters(MigrationState
*s
)
1683 if (s
->must_remove_block_options
) {
1684 /* setting to false can never fail */
1685 migrate_set_block_enabled(false, &error_abort
);
1686 migrate_set_block_incremental(s
, false);
1687 s
->must_remove_block_options
= false;
1691 static void migrate_fd_cleanup(MigrationState
*s
)
1693 qemu_bh_delete(s
->cleanup_bh
);
1694 s
->cleanup_bh
= NULL
;
1696 qemu_savevm_state_cleanup();
1698 if (s
->to_dst_file
) {
1701 trace_migrate_fd_cleanup();
1702 qemu_mutex_unlock_iothread();
1703 if (s
->migration_thread_running
) {
1704 qemu_thread_join(&s
->thread
);
1705 s
->migration_thread_running
= false;
1707 qemu_mutex_lock_iothread();
1709 multifd_save_cleanup();
1710 qemu_mutex_lock(&s
->qemu_file_lock
);
1711 tmp
= s
->to_dst_file
;
1712 s
->to_dst_file
= NULL
;
1713 qemu_mutex_unlock(&s
->qemu_file_lock
);
1715 * Close the file handle without the lock to make sure the
1716 * critical section won't block for long.
1721 assert(!migration_is_active(s
));
1723 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1724 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1725 MIGRATION_STATUS_CANCELLED
);
1729 /* It is used on info migrate. We can't free it */
1730 error_report_err(error_copy(s
->error
));
1732 notifier_list_notify(&migration_state_notifiers
, s
);
1733 block_cleanup_parameters(s
);
1734 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1737 static void migrate_fd_cleanup_schedule(MigrationState
*s
)
1740 * Ref the state for bh, because it may be called when
1741 * there're already no other refs
1743 object_ref(OBJECT(s
));
1744 qemu_bh_schedule(s
->cleanup_bh
);
1747 static void migrate_fd_cleanup_bh(void *opaque
)
1749 MigrationState
*s
= opaque
;
1750 migrate_fd_cleanup(s
);
1751 object_unref(OBJECT(s
));
1754 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1756 QEMU_LOCK_GUARD(&s
->error_mutex
);
1758 s
->error
= error_copy(error
);
1762 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1764 trace_migrate_fd_error(error_get_pretty(error
));
1765 assert(s
->to_dst_file
== NULL
);
1766 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1767 MIGRATION_STATUS_FAILED
);
1768 migrate_set_error(s
, error
);
1771 static void migrate_fd_cancel(MigrationState
*s
)
1774 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1775 trace_migrate_fd_cancel();
1777 if (s
->rp_state
.from_dst_file
) {
1778 /* shutdown the rp socket, so causing the rp thread to shutdown */
1779 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1783 old_state
= s
->state
;
1784 if (!migration_is_running(old_state
)) {
1787 /* If the migration is paused, kick it out of the pause */
1788 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1789 qemu_sem_post(&s
->pause_sem
);
1791 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1792 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1795 * If we're unlucky the migration code might be stuck somewhere in a
1796 * send/write while the network has failed and is waiting to timeout;
1797 * if we've got shutdown(2) available then we can force it to quit.
1798 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1799 * called in a bh, so there is no race against this cancel.
1801 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1802 qemu_file_shutdown(f
);
1804 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1805 Error
*local_err
= NULL
;
1807 bdrv_invalidate_cache_all(&local_err
);
1809 error_report_err(local_err
);
1811 s
->block_inactive
= false;
1816 void add_migration_state_change_notifier(Notifier
*notify
)
1818 notifier_list_add(&migration_state_notifiers
, notify
);
1821 void remove_migration_state_change_notifier(Notifier
*notify
)
1823 notifier_remove(notify
);
1826 bool migration_in_setup(MigrationState
*s
)
1828 return s
->state
== MIGRATION_STATUS_SETUP
;
1831 bool migration_has_finished(MigrationState
*s
)
1833 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1836 bool migration_has_failed(MigrationState
*s
)
1838 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1839 s
->state
== MIGRATION_STATUS_FAILED
);
1842 bool migration_in_postcopy(void)
1844 MigrationState
*s
= migrate_get_current();
1847 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1848 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1849 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1856 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1858 return migration_in_postcopy() && s
->postcopy_after_devices
;
1861 bool migration_in_incoming_postcopy(void)
1863 PostcopyState ps
= postcopy_state_get();
1865 return ps
>= POSTCOPY_INCOMING_DISCARD
&& ps
< POSTCOPY_INCOMING_END
;
1868 bool migration_is_idle(void)
1870 MigrationState
*s
= current_migration
;
1877 case MIGRATION_STATUS_NONE
:
1878 case MIGRATION_STATUS_CANCELLED
:
1879 case MIGRATION_STATUS_COMPLETED
:
1880 case MIGRATION_STATUS_FAILED
:
1882 case MIGRATION_STATUS_SETUP
:
1883 case MIGRATION_STATUS_CANCELLING
:
1884 case MIGRATION_STATUS_ACTIVE
:
1885 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1886 case MIGRATION_STATUS_COLO
:
1887 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1888 case MIGRATION_STATUS_DEVICE
:
1889 case MIGRATION_STATUS_WAIT_UNPLUG
:
1891 case MIGRATION_STATUS__MAX
:
1892 g_assert_not_reached();
1898 bool migration_is_active(MigrationState
*s
)
1900 return (s
->state
== MIGRATION_STATUS_ACTIVE
||
1901 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1904 void migrate_init(MigrationState
*s
)
1907 * Reinitialise all migration state, except
1908 * parameters/capabilities that the user set, and
1912 s
->to_dst_file
= NULL
;
1913 s
->state
= MIGRATION_STATUS_NONE
;
1914 s
->rp_state
.from_dst_file
= NULL
;
1915 s
->rp_state
.error
= false;
1917 s
->pages_per_second
= 0.0;
1919 s
->expected_downtime
= 0;
1921 s
->start_postcopy
= false;
1922 s
->postcopy_after_devices
= false;
1923 s
->migration_thread_running
= false;
1924 error_free(s
->error
);
1928 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1930 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1932 s
->vm_was_running
= false;
1933 s
->iteration_initial_bytes
= 0;
1934 s
->threshold_size
= 0;
1937 static GSList
*migration_blockers
;
1939 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1941 if (only_migratable
) {
1942 error_propagate_prepend(errp
, error_copy(reason
),
1943 "disallowing migration blocker "
1944 "(--only-migratable) for: ");
1948 if (migration_is_idle()) {
1949 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1953 error_propagate_prepend(errp
, error_copy(reason
),
1954 "disallowing migration blocker "
1955 "(migration in progress) for: ");
1959 void migrate_del_blocker(Error
*reason
)
1961 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1964 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1966 Error
*local_err
= NULL
;
1967 static bool once
= true;
1970 error_setg(errp
, "The incoming migration has already been started");
1973 if (!runstate_check(RUN_STATE_INMIGRATE
)) {
1974 error_setg(errp
, "'-incoming' was not specified on the command line");
1978 qemu_start_incoming_migration(uri
, &local_err
);
1981 error_propagate(errp
, local_err
);
1988 void qmp_migrate_recover(const char *uri
, Error
**errp
)
1990 MigrationIncomingState
*mis
= migration_incoming_get_current();
1992 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1993 error_setg(errp
, "Migrate recover can only be run "
1994 "when postcopy is paused.");
1998 if (qatomic_cmpxchg(&mis
->postcopy_recover_triggered
,
1999 false, true) == true) {
2000 error_setg(errp
, "Migrate recovery is triggered already");
2005 * Note that this call will never start a real migration; it will
2006 * only re-setup the migration stream and poke existing migration
2007 * to continue using that newly established channel.
2009 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2010 qemu_start_incoming_migration(uri
, errp
);
2013 void qmp_migrate_pause(Error
**errp
)
2015 MigrationState
*ms
= migrate_get_current();
2016 MigrationIncomingState
*mis
= migration_incoming_get_current();
2019 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2020 /* Source side, during postcopy */
2021 qemu_mutex_lock(&ms
->qemu_file_lock
);
2022 ret
= qemu_file_shutdown(ms
->to_dst_file
);
2023 qemu_mutex_unlock(&ms
->qemu_file_lock
);
2025 error_setg(errp
, "Failed to pause source migration");
2030 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2031 ret
= qemu_file_shutdown(mis
->from_src_file
);
2033 error_setg(errp
, "Failed to pause destination migration");
2038 error_setg(errp
, "migrate-pause is currently only supported "
2039 "during postcopy-active state");
2042 bool migration_is_blocked(Error
**errp
)
2044 if (qemu_savevm_state_blocked(errp
)) {
2048 if (migration_blockers
) {
2049 error_propagate(errp
, error_copy(migration_blockers
->data
));
2056 /* Returns true if continue to migrate, or false if error detected */
2057 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
2058 bool resume
, Error
**errp
)
2060 Error
*local_err
= NULL
;
2063 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2064 error_setg(errp
, "Cannot resume if there is no "
2065 "paused migration");
2070 * Postcopy recovery won't work well with release-ram
2071 * capability since release-ram will drop the page buffer as
2072 * long as the page is put into the send buffer. So if there
2073 * is a network failure happened, any page buffers that have
2074 * not yet reached the destination VM but have already been
2075 * sent from the source VM will be lost forever. Let's refuse
2076 * the client from resuming such a postcopy migration.
2077 * Luckily release-ram was designed to only be used when src
2078 * and destination VMs are on the same host, so it should be
2081 if (migrate_release_ram()) {
2082 error_setg(errp
, "Postcopy recovery cannot work "
2083 "when release-ram capability is set");
2087 /* This is a resume, skip init status */
2091 if (migration_is_running(s
->state
)) {
2092 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
2096 if (runstate_check(RUN_STATE_INMIGRATE
)) {
2097 error_setg(errp
, "Guest is waiting for an incoming migration");
2101 if (runstate_check(RUN_STATE_POSTMIGRATE
)) {
2102 error_setg(errp
, "Can't migrate the vm that was paused due to "
2103 "previous migration");
2107 if (migration_is_blocked(errp
)) {
2111 if (blk
|| blk_inc
) {
2112 if (migrate_use_block() || migrate_use_block_incremental()) {
2113 error_setg(errp
, "Command options are incompatible with "
2114 "current migration capabilities");
2117 migrate_set_block_enabled(true, &local_err
);
2119 error_propagate(errp
, local_err
);
2122 s
->must_remove_block_options
= true;
2126 migrate_set_block_incremental(s
, true);
2131 * set ram_counters memory to zero for a
2134 memset(&ram_counters
, 0, sizeof(ram_counters
));
2139 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
2140 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
2141 bool has_resume
, bool resume
, Error
**errp
)
2143 Error
*local_err
= NULL
;
2144 MigrationState
*s
= migrate_get_current();
2145 const char *p
= NULL
;
2147 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
2148 has_resume
&& resume
, errp
)) {
2149 /* Error detected, put into errp */
2153 if (!(has_resume
&& resume
)) {
2154 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
2159 if (strstart(uri
, "tcp:", &p
) ||
2160 strstart(uri
, "unix:", NULL
) ||
2161 strstart(uri
, "vsock:", NULL
)) {
2162 socket_start_outgoing_migration(s
, p
? p
: uri
, &local_err
);
2164 } else if (strstart(uri
, "rdma:", &p
)) {
2165 rdma_start_outgoing_migration(s
, p
, &local_err
);
2167 } else if (strstart(uri
, "exec:", &p
)) {
2168 exec_start_outgoing_migration(s
, p
, &local_err
);
2169 } else if (strstart(uri
, "fd:", &p
)) {
2170 fd_start_outgoing_migration(s
, p
, &local_err
);
2172 if (!(has_resume
&& resume
)) {
2173 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2175 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
2176 "a valid migration protocol");
2177 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2178 MIGRATION_STATUS_FAILED
);
2179 block_cleanup_parameters(s
);
2184 if (!(has_resume
&& resume
)) {
2185 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2187 migrate_fd_error(s
, local_err
);
2188 error_propagate(errp
, local_err
);
2193 void qmp_migrate_cancel(Error
**errp
)
2195 migrate_fd_cancel(migrate_get_current());
2198 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
2200 MigrationState
*s
= migrate_get_current();
2201 if (s
->state
!= state
) {
2202 error_setg(errp
, "Migration not in expected state: %s",
2203 MigrationStatus_str(s
->state
));
2206 qemu_sem_post(&s
->pause_sem
);
2209 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
2211 MigrateSetParameters p
= {
2212 .has_xbzrle_cache_size
= true,
2213 .xbzrle_cache_size
= value
,
2216 qmp_migrate_set_parameters(&p
, errp
);
2219 int64_t qmp_query_migrate_cache_size(Error
**errp
)
2221 return migrate_xbzrle_cache_size();
2224 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
2226 MigrateSetParameters p
= {
2227 .has_max_bandwidth
= true,
2228 .max_bandwidth
= value
,
2231 qmp_migrate_set_parameters(&p
, errp
);
2234 void qmp_migrate_set_downtime(double value
, Error
**errp
)
2236 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
2237 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
2239 "an integer in the range of 0 to "
2240 stringify(MAX_MIGRATE_DOWNTIME_SECONDS
)" seconds");
2244 value
*= 1000; /* Convert to milliseconds */
2246 MigrateSetParameters p
= {
2247 .has_downtime_limit
= true,
2248 .downtime_limit
= (int64_t)value
,
2251 qmp_migrate_set_parameters(&p
, errp
);
2254 bool migrate_release_ram(void)
2258 s
= migrate_get_current();
2260 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
2263 bool migrate_postcopy_ram(void)
2267 s
= migrate_get_current();
2269 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
2272 bool migrate_postcopy(void)
2274 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2277 bool migrate_auto_converge(void)
2281 s
= migrate_get_current();
2283 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
2286 bool migrate_zero_blocks(void)
2290 s
= migrate_get_current();
2292 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
2295 bool migrate_postcopy_blocktime(void)
2299 s
= migrate_get_current();
2301 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
2304 bool migrate_use_compression(void)
2308 s
= migrate_get_current();
2310 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
2313 int migrate_compress_level(void)
2317 s
= migrate_get_current();
2319 return s
->parameters
.compress_level
;
2322 int migrate_compress_threads(void)
2326 s
= migrate_get_current();
2328 return s
->parameters
.compress_threads
;
2331 int migrate_compress_wait_thread(void)
2335 s
= migrate_get_current();
2337 return s
->parameters
.compress_wait_thread
;
2340 int migrate_decompress_threads(void)
2344 s
= migrate_get_current();
2346 return s
->parameters
.decompress_threads
;
2349 bool migrate_dirty_bitmaps(void)
2353 s
= migrate_get_current();
2355 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
2358 bool migrate_ignore_shared(void)
2362 s
= migrate_get_current();
2364 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
];
2367 bool migrate_validate_uuid(void)
2371 s
= migrate_get_current();
2373 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_VALIDATE_UUID
];
2376 bool migrate_use_events(void)
2380 s
= migrate_get_current();
2382 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
2385 bool migrate_use_multifd(void)
2389 s
= migrate_get_current();
2391 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_MULTIFD
];
2394 bool migrate_pause_before_switchover(void)
2398 s
= migrate_get_current();
2400 return s
->enabled_capabilities
[
2401 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
2404 int migrate_multifd_channels(void)
2408 s
= migrate_get_current();
2410 return s
->parameters
.multifd_channels
;
2413 MultiFDCompression
migrate_multifd_compression(void)
2417 s
= migrate_get_current();
2419 return s
->parameters
.multifd_compression
;
2422 int migrate_multifd_zlib_level(void)
2426 s
= migrate_get_current();
2428 return s
->parameters
.multifd_zlib_level
;
2431 int migrate_multifd_zstd_level(void)
2435 s
= migrate_get_current();
2437 return s
->parameters
.multifd_zstd_level
;
2440 int migrate_use_xbzrle(void)
2444 s
= migrate_get_current();
2446 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
2449 int64_t migrate_xbzrle_cache_size(void)
2453 s
= migrate_get_current();
2455 return s
->parameters
.xbzrle_cache_size
;
2458 static int64_t migrate_max_postcopy_bandwidth(void)
2462 s
= migrate_get_current();
2464 return s
->parameters
.max_postcopy_bandwidth
;
2467 bool migrate_use_block(void)
2471 s
= migrate_get_current();
2473 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
2476 bool migrate_use_return_path(void)
2480 s
= migrate_get_current();
2482 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
2485 bool migrate_use_block_incremental(void)
2489 s
= migrate_get_current();
2491 return s
->parameters
.block_incremental
;
2494 /* migration thread support */
2496 * Something bad happened to the RP stream, mark an error
2497 * The caller shall print or trace something to indicate why
2499 static void mark_source_rp_bad(MigrationState
*s
)
2501 s
->rp_state
.error
= true;
2504 static struct rp_cmd_args
{
2505 ssize_t len
; /* -1 = variable */
2508 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
2509 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
2510 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
2511 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
2512 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
2513 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
2514 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
2515 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
2519 * Process a request for pages received on the return path,
2520 * We're allowed to send more than requested (e.g. to round to our page size)
2521 * and we don't need to send pages that have already been sent.
2523 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
2524 ram_addr_t start
, size_t len
)
2526 long our_host_ps
= qemu_real_host_page_size
;
2528 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
2531 * Since we currently insist on matching page sizes, just sanity check
2532 * we're being asked for whole host pages.
2534 if (start
& (our_host_ps
- 1) ||
2535 (len
& (our_host_ps
- 1))) {
2536 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2537 " len: %zd", __func__
, start
, len
);
2538 mark_source_rp_bad(ms
);
2542 if (ram_save_queue_pages(rbname
, start
, len
)) {
2543 mark_source_rp_bad(ms
);
2547 /* Return true to retry, false to quit */
2548 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
2550 trace_postcopy_pause_return_path();
2552 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
2554 trace_postcopy_pause_return_path_continued();
2559 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
2561 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
2564 error_report("%s: invalid block name '%s'", __func__
, block_name
);
2568 /* Fetch the received bitmap and refresh the dirty bitmap */
2569 return ram_dirty_bitmap_reload(s
, block
);
2572 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
2574 trace_source_return_path_thread_resume_ack(value
);
2576 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
2577 error_report("%s: illegal resume_ack value %"PRIu32
,
2582 /* Now both sides are active. */
2583 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2584 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2586 /* Notify send thread that time to continue send pages */
2587 qemu_sem_post(&s
->rp_state
.rp_sem
);
2593 * Handles messages sent on the return path towards the source VM
2596 static void *source_return_path_thread(void *opaque
)
2598 MigrationState
*ms
= opaque
;
2599 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
2600 uint16_t header_len
, header_type
;
2602 uint32_t tmp32
, sibling_error
;
2603 ram_addr_t start
= 0; /* =0 to silence warning */
2604 size_t len
= 0, expected_len
;
2607 trace_source_return_path_thread_entry();
2608 rcu_register_thread();
2611 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
2612 migration_is_setup_or_active(ms
->state
)) {
2613 trace_source_return_path_thread_loop_top();
2614 header_type
= qemu_get_be16(rp
);
2615 header_len
= qemu_get_be16(rp
);
2617 if (qemu_file_get_error(rp
)) {
2618 mark_source_rp_bad(ms
);
2622 if (header_type
>= MIG_RP_MSG_MAX
||
2623 header_type
== MIG_RP_MSG_INVALID
) {
2624 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2625 header_type
, header_len
);
2626 mark_source_rp_bad(ms
);
2630 if ((rp_cmd_args
[header_type
].len
!= -1 &&
2631 header_len
!= rp_cmd_args
[header_type
].len
) ||
2632 header_len
> sizeof(buf
)) {
2633 error_report("RP: Received '%s' message (0x%04x) with"
2634 "incorrect length %d expecting %zu",
2635 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
2636 (size_t)rp_cmd_args
[header_type
].len
);
2637 mark_source_rp_bad(ms
);
2641 /* We know we've got a valid header by this point */
2642 res
= qemu_get_buffer(rp
, buf
, header_len
);
2643 if (res
!= header_len
) {
2644 error_report("RP: Failed reading data for message 0x%04x"
2645 " read %d expected %d",
2646 header_type
, res
, header_len
);
2647 mark_source_rp_bad(ms
);
2651 /* OK, we have the message and the data */
2652 switch (header_type
) {
2653 case MIG_RP_MSG_SHUT
:
2654 sibling_error
= ldl_be_p(buf
);
2655 trace_source_return_path_thread_shut(sibling_error
);
2656 if (sibling_error
) {
2657 error_report("RP: Sibling indicated error %d", sibling_error
);
2658 mark_source_rp_bad(ms
);
2661 * We'll let the main thread deal with closing the RP
2662 * we could do a shutdown(2) on it, but we're the only user
2663 * anyway, so there's nothing gained.
2667 case MIG_RP_MSG_PONG
:
2668 tmp32
= ldl_be_p(buf
);
2669 trace_source_return_path_thread_pong(tmp32
);
2672 case MIG_RP_MSG_REQ_PAGES
:
2673 start
= ldq_be_p(buf
);
2674 len
= ldl_be_p(buf
+ 8);
2675 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
2678 case MIG_RP_MSG_REQ_PAGES_ID
:
2679 expected_len
= 12 + 1; /* header + termination */
2681 if (header_len
>= expected_len
) {
2682 start
= ldq_be_p(buf
);
2683 len
= ldl_be_p(buf
+ 8);
2684 /* Now we expect an idstr */
2685 tmp32
= buf
[12]; /* Length of the following idstr */
2686 buf
[13 + tmp32
] = '\0';
2687 expected_len
+= tmp32
;
2689 if (header_len
!= expected_len
) {
2690 error_report("RP: Req_Page_id with length %d expecting %zd",
2691 header_len
, expected_len
);
2692 mark_source_rp_bad(ms
);
2695 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
2698 case MIG_RP_MSG_RECV_BITMAP
:
2699 if (header_len
< 1) {
2700 error_report("%s: missing block name", __func__
);
2701 mark_source_rp_bad(ms
);
2704 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2705 buf
[buf
[0] + 1] = '\0';
2706 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
2707 mark_source_rp_bad(ms
);
2712 case MIG_RP_MSG_RESUME_ACK
:
2713 tmp32
= ldl_be_p(buf
);
2714 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
2715 mark_source_rp_bad(ms
);
2726 res
= qemu_file_get_error(rp
);
2728 if (res
== -EIO
&& migration_in_postcopy()) {
2730 * Maybe there is something we can do: it looks like a
2731 * network down issue, and we pause for a recovery.
2733 if (postcopy_pause_return_path_thread(ms
)) {
2734 /* Reload rp, reset the rest */
2735 if (rp
!= ms
->rp_state
.from_dst_file
) {
2737 rp
= ms
->rp_state
.from_dst_file
;
2739 ms
->rp_state
.error
= false;
2744 trace_source_return_path_thread_bad_end();
2745 mark_source_rp_bad(ms
);
2748 trace_source_return_path_thread_end();
2749 ms
->rp_state
.from_dst_file
= NULL
;
2751 rcu_unregister_thread();
2755 static int open_return_path_on_source(MigrationState
*ms
,
2759 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
2760 if (!ms
->rp_state
.from_dst_file
) {
2764 trace_open_return_path_on_source();
2766 if (!create_thread
) {
2771 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
2772 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
2774 trace_open_return_path_on_source_continue();
2779 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2780 static int await_return_path_close_on_source(MigrationState
*ms
)
2783 * If this is a normal exit then the destination will send a SHUT and the
2784 * rp_thread will exit, however if there's an error we need to cause
2787 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
2789 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2790 * waiting for the destination.
2792 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
2793 mark_source_rp_bad(ms
);
2795 trace_await_return_path_close_on_source_joining();
2796 qemu_thread_join(&ms
->rp_state
.rp_thread
);
2797 trace_await_return_path_close_on_source_close();
2798 return ms
->rp_state
.error
;
2802 * Switch from normal iteration to postcopy
2803 * Returns non-0 on error
2805 static int postcopy_start(MigrationState
*ms
)
2808 QIOChannelBuffer
*bioc
;
2810 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2811 int64_t bandwidth
= migrate_max_postcopy_bandwidth();
2812 bool restart_block
= false;
2813 int cur_state
= MIGRATION_STATUS_ACTIVE
;
2814 if (!migrate_pause_before_switchover()) {
2815 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
2816 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2819 trace_postcopy_start();
2820 qemu_mutex_lock_iothread();
2821 trace_postcopy_start_set_run();
2823 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2824 global_state_store();
2825 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2830 ret
= migration_maybe_pause(ms
, &cur_state
,
2831 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2836 ret
= bdrv_inactivate_all();
2840 restart_block
= true;
2843 * Cause any non-postcopiable, but iterative devices to
2844 * send out their final data.
2846 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2849 * in Finish migrate and with the io-lock held everything should
2850 * be quiet, but we've potentially still got dirty pages and we
2851 * need to tell the destination to throw any pages it's already received
2854 if (migrate_postcopy_ram()) {
2855 if (ram_postcopy_send_discard_bitmap(ms
)) {
2856 error_report("postcopy send discard bitmap failed");
2862 * send rest of state - note things that are doing postcopy
2863 * will notice we're in POSTCOPY_ACTIVE and not actually
2864 * wrap their state up here
2866 /* 0 max-postcopy-bandwidth means unlimited */
2868 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
2870 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
2872 if (migrate_postcopy_ram()) {
2873 /* Ping just for debugging, helps line traces up */
2874 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
2878 * While loading the device state we may trigger page transfer
2879 * requests and the fd must be free to process those, and thus
2880 * the destination must read the whole device state off the fd before
2881 * it starts processing it. Unfortunately the ad-hoc migration format
2882 * doesn't allow the destination to know the size to read without fully
2883 * parsing it through each devices load-state code (especially the open
2884 * coded devices that use get/put).
2885 * So we wrap the device state up in a package with a length at the start;
2886 * to do this we use a qemu_buf to hold the whole of the device state.
2888 bioc
= qio_channel_buffer_new(4096);
2889 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
2890 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
2891 object_unref(OBJECT(bioc
));
2894 * Make sure the receiver can get incoming pages before we send the rest
2897 qemu_savevm_send_postcopy_listen(fb
);
2899 qemu_savevm_state_complete_precopy(fb
, false, false);
2900 if (migrate_postcopy_ram()) {
2901 qemu_savevm_send_ping(fb
, 3);
2904 qemu_savevm_send_postcopy_run(fb
);
2906 /* <><> end of stuff going into the package */
2908 /* Last point of recovery; as soon as we send the package the destination
2909 * can open devices and potentially start running.
2910 * Lets just check again we've not got any errors.
2912 ret
= qemu_file_get_error(ms
->to_dst_file
);
2914 error_report("postcopy_start: Migration stream errored (pre package)");
2918 restart_block
= false;
2920 /* Now send that blob */
2921 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2926 /* Send a notify to give a chance for anything that needs to happen
2927 * at the transition to postcopy and after the device state; in particular
2928 * spice needs to trigger a transition now
2930 ms
->postcopy_after_devices
= true;
2931 notifier_list_notify(&migration_state_notifiers
, ms
);
2933 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2935 qemu_mutex_unlock_iothread();
2937 if (migrate_postcopy_ram()) {
2939 * Although this ping is just for debug, it could potentially be
2940 * used for getting a better measurement of downtime at the source.
2942 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2945 if (migrate_release_ram()) {
2946 ram_postcopy_migrated_memory_release(ms
);
2949 ret
= qemu_file_get_error(ms
->to_dst_file
);
2951 error_report("postcopy_start: Migration stream errored");
2952 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2953 MIGRATION_STATUS_FAILED
);
2961 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2962 MIGRATION_STATUS_FAILED
);
2963 if (restart_block
) {
2964 /* A failure happened early enough that we know the destination hasn't
2965 * accessed block devices, so we're safe to recover.
2967 Error
*local_err
= NULL
;
2969 bdrv_invalidate_cache_all(&local_err
);
2971 error_report_err(local_err
);
2974 qemu_mutex_unlock_iothread();
2979 * migration_maybe_pause: Pause if required to by
2980 * migrate_pause_before_switchover called with the iothread locked
2981 * Returns: 0 on success
2983 static int migration_maybe_pause(MigrationState
*s
,
2984 int *current_active_state
,
2987 if (!migrate_pause_before_switchover()) {
2991 /* Since leaving this state is not atomic with posting the semaphore
2992 * it's possible that someone could have issued multiple migrate_continue
2993 * and the semaphore is incorrectly positive at this point;
2994 * the docs say it's undefined to reinit a semaphore that's already
2995 * init'd, so use timedwait to eat up any existing posts.
2997 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2998 /* This block intentionally left blank */
3002 * If the migration is cancelled when it is in the completion phase,
3003 * the migration state is set to MIGRATION_STATUS_CANCELLING.
3004 * So we don't need to wait a semaphore, otherwise we would always
3005 * wait for the 'pause_sem' semaphore.
3007 if (s
->state
!= MIGRATION_STATUS_CANCELLING
) {
3008 qemu_mutex_unlock_iothread();
3009 migrate_set_state(&s
->state
, *current_active_state
,
3010 MIGRATION_STATUS_PRE_SWITCHOVER
);
3011 qemu_sem_wait(&s
->pause_sem
);
3012 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
3014 *current_active_state
= new_state
;
3015 qemu_mutex_lock_iothread();
3018 return s
->state
== new_state
? 0 : -EINVAL
;
3022 * migration_completion: Used by migration_thread when there's not much left.
3023 * The caller 'breaks' the loop when this returns.
3025 * @s: Current migration state
3027 static void migration_completion(MigrationState
*s
)
3030 int current_active_state
= s
->state
;
3032 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
3033 qemu_mutex_lock_iothread();
3034 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3035 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3036 s
->vm_was_running
= runstate_is_running();
3037 ret
= global_state_store();
3040 bool inactivate
= !migrate_colo_enabled();
3041 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
3043 ret
= migration_maybe_pause(s
, ¤t_active_state
,
3044 MIGRATION_STATUS_DEVICE
);
3047 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
3048 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
3051 if (inactivate
&& ret
>= 0) {
3052 s
->block_inactive
= true;
3055 qemu_mutex_unlock_iothread();
3060 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3061 trace_migration_completion_postcopy_end();
3063 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
3064 trace_migration_completion_postcopy_end_after_complete();
3065 } else if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
3070 * If rp was opened we must clean up the thread before
3071 * cleaning everything else up (since if there are no failures
3072 * it will wait for the destination to send it's status in
3075 if (s
->rp_state
.from_dst_file
) {
3077 trace_migration_return_path_end_before();
3078 rp_error
= await_return_path_close_on_source(s
);
3079 trace_migration_return_path_end_after(rp_error
);
3081 goto fail_invalidate
;
3085 if (qemu_file_get_error(s
->to_dst_file
)) {
3086 trace_migration_completion_file_err();
3087 goto fail_invalidate
;
3090 if (!migrate_colo_enabled()) {
3091 migrate_set_state(&s
->state
, current_active_state
,
3092 MIGRATION_STATUS_COMPLETED
);
3098 /* If not doing postcopy, vm_start() will be called: let's regain
3099 * control on images.
3101 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
3102 s
->state
== MIGRATION_STATUS_DEVICE
) {
3103 Error
*local_err
= NULL
;
3105 qemu_mutex_lock_iothread();
3106 bdrv_invalidate_cache_all(&local_err
);
3108 error_report_err(local_err
);
3110 s
->block_inactive
= false;
3112 qemu_mutex_unlock_iothread();
3116 migrate_set_state(&s
->state
, current_active_state
,
3117 MIGRATION_STATUS_FAILED
);
3120 bool migrate_colo_enabled(void)
3122 MigrationState
*s
= migrate_get_current();
3123 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
3126 typedef enum MigThrError
{
3127 /* No error detected */
3128 MIG_THR_ERR_NONE
= 0,
3129 /* Detected error, but resumed successfully */
3130 MIG_THR_ERR_RECOVERED
= 1,
3131 /* Detected fatal error, need to exit */
3132 MIG_THR_ERR_FATAL
= 2,
3135 static int postcopy_resume_handshake(MigrationState
*s
)
3137 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
3139 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3140 qemu_sem_wait(&s
->rp_state
.rp_sem
);
3143 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3150 /* Return zero if success, or <0 for error */
3151 static int postcopy_do_resume(MigrationState
*s
)
3156 * Call all the resume_prepare() hooks, so that modules can be
3157 * ready for the migration resume.
3159 ret
= qemu_savevm_state_resume_prepare(s
);
3161 error_report("%s: resume_prepare() failure detected: %d",
3167 * Last handshake with destination on the resume (destination will
3168 * switch to postcopy-active afterwards)
3170 ret
= postcopy_resume_handshake(s
);
3172 error_report("%s: handshake failed: %d", __func__
, ret
);
3180 * We don't return until we are in a safe state to continue current
3181 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
3182 * MIG_THR_ERR_FATAL if unrecovery failure happened.
3184 static MigThrError
postcopy_pause(MigrationState
*s
)
3186 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3191 /* Current channel is possibly broken. Release it. */
3192 assert(s
->to_dst_file
);
3193 qemu_mutex_lock(&s
->qemu_file_lock
);
3194 file
= s
->to_dst_file
;
3195 s
->to_dst_file
= NULL
;
3196 qemu_mutex_unlock(&s
->qemu_file_lock
);
3198 qemu_file_shutdown(file
);
3201 migrate_set_state(&s
->state
, s
->state
,
3202 MIGRATION_STATUS_POSTCOPY_PAUSED
);
3204 error_report("Detected IO failure for postcopy. "
3205 "Migration paused.");
3208 * We wait until things fixed up. Then someone will setup the
3209 * status back for us.
3211 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
3212 qemu_sem_wait(&s
->postcopy_pause_sem
);
3215 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3216 /* Woken up by a recover procedure. Give it a shot */
3219 * Firstly, let's wake up the return path now, with a new
3220 * return path channel.
3222 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
3224 /* Do the resume logic */
3225 if (postcopy_do_resume(s
) == 0) {
3226 /* Let's continue! */
3227 trace_postcopy_pause_continued();
3228 return MIG_THR_ERR_RECOVERED
;
3231 * Something wrong happened during the recovery, let's
3232 * pause again. Pause is always better than throwing
3238 /* This is not right... Time to quit. */
3239 return MIG_THR_ERR_FATAL
;
3244 static MigThrError
migration_detect_error(MigrationState
*s
)
3247 int state
= s
->state
;
3248 Error
*local_error
= NULL
;
3250 if (state
== MIGRATION_STATUS_CANCELLING
||
3251 state
== MIGRATION_STATUS_CANCELLED
) {
3252 /* End the migration, but don't set the state to failed */
3253 return MIG_THR_ERR_FATAL
;
3256 /* Try to detect any file errors */
3257 ret
= qemu_file_get_error_obj(s
->to_dst_file
, &local_error
);
3259 /* Everything is fine */
3260 assert(!local_error
);
3261 return MIG_THR_ERR_NONE
;
3265 migrate_set_error(s
, local_error
);
3266 error_free(local_error
);
3269 if (state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
== -EIO
) {
3271 * For postcopy, we allow the network to be down for a
3272 * while. After that, it can be continued by a
3275 return postcopy_pause(s
);
3278 * For precopy (or postcopy with error outside IO), we fail
3281 migrate_set_state(&s
->state
, state
, MIGRATION_STATUS_FAILED
);
3282 trace_migration_thread_file_err();
3284 /* Time to stop the migration, now. */
3285 return MIG_THR_ERR_FATAL
;
3289 /* How many bytes have we transferred since the beginning of the migration */
3290 static uint64_t migration_total_bytes(MigrationState
*s
)
3292 return qemu_ftell(s
->to_dst_file
) + ram_counters
.multifd_bytes
;
3295 static void migration_calculate_complete(MigrationState
*s
)
3297 uint64_t bytes
= migration_total_bytes(s
);
3298 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3299 int64_t transfer_time
;
3301 s
->total_time
= end_time
- s
->start_time
;
3304 * It's still not set, so we are precopy migration. For
3305 * postcopy, downtime is calculated during postcopy_start().
3307 s
->downtime
= end_time
- s
->downtime_start
;
3310 transfer_time
= s
->total_time
- s
->setup_time
;
3311 if (transfer_time
) {
3312 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
3316 static void update_iteration_initial_status(MigrationState
*s
)
3319 * Update these three fields at the same time to avoid mismatch info lead
3320 * wrong speed calculation.
3322 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3323 s
->iteration_initial_bytes
= migration_total_bytes(s
);
3324 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
3327 static void migration_update_counters(MigrationState
*s
,
3328 int64_t current_time
)
3330 uint64_t transferred
, transferred_pages
, time_spent
;
3331 uint64_t current_bytes
; /* bytes transferred since the beginning */
3334 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
3338 current_bytes
= migration_total_bytes(s
);
3339 transferred
= current_bytes
- s
->iteration_initial_bytes
;
3340 time_spent
= current_time
- s
->iteration_start_time
;
3341 bandwidth
= (double)transferred
/ time_spent
;
3342 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
3344 s
->mbps
= (((double) transferred
* 8.0) /
3345 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
3347 transferred_pages
= ram_get_total_transferred_pages() -
3348 s
->iteration_initial_pages
;
3349 s
->pages_per_second
= (double) transferred_pages
/
3350 (((double) time_spent
/ 1000.0));
3353 * if we haven't sent anything, we don't want to
3354 * recalculate. 10000 is a small enough number for our purposes
3356 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
3357 s
->expected_downtime
= ram_counters
.remaining
/ bandwidth
;
3360 qemu_file_reset_rate_limit(s
->to_dst_file
);
3362 update_iteration_initial_status(s
);
3364 trace_migrate_transferred(transferred
, time_spent
,
3365 bandwidth
, s
->threshold_size
);
3368 /* Migration thread iteration status */
3370 MIG_ITERATE_RESUME
, /* Resume current iteration */
3371 MIG_ITERATE_SKIP
, /* Skip current iteration */
3372 MIG_ITERATE_BREAK
, /* Break the loop */
3376 * Return true if continue to the next iteration directly, false
3379 static MigIterateState
migration_iteration_run(MigrationState
*s
)
3381 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
3382 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
3384 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
3385 &pend_compat
, &pend_post
);
3386 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
3388 trace_migrate_pending(pending_size
, s
->threshold_size
,
3389 pend_pre
, pend_compat
, pend_post
);
3391 if (pending_size
&& pending_size
>= s
->threshold_size
) {
3392 /* Still a significant amount to transfer */
3393 if (!in_postcopy
&& pend_pre
<= s
->threshold_size
&&
3394 qatomic_read(&s
->start_postcopy
)) {
3395 if (postcopy_start(s
)) {
3396 error_report("%s: postcopy failed to start", __func__
);
3398 return MIG_ITERATE_SKIP
;
3400 /* Just another iteration step */
3401 qemu_savevm_state_iterate(s
->to_dst_file
, in_postcopy
);
3403 trace_migration_thread_low_pending(pending_size
);
3404 migration_completion(s
);
3405 return MIG_ITERATE_BREAK
;
3408 return MIG_ITERATE_RESUME
;
3411 static void migration_iteration_finish(MigrationState
*s
)
3413 /* If we enabled cpu throttling for auto-converge, turn it off. */
3414 cpu_throttle_stop();
3416 qemu_mutex_lock_iothread();
3418 case MIGRATION_STATUS_COMPLETED
:
3419 migration_calculate_complete(s
);
3420 runstate_set(RUN_STATE_POSTMIGRATE
);
3423 case MIGRATION_STATUS_ACTIVE
:
3425 * We should really assert here, but since it's during
3426 * migration, let's try to reduce the usage of assertions.
3428 if (!migrate_colo_enabled()) {
3429 error_report("%s: critical error: calling COLO code without "
3430 "COLO enabled", __func__
);
3432 migrate_start_colo_process(s
);
3434 * Fixme: we will run VM in COLO no matter its old running state.
3435 * After exited COLO, we will keep running.
3437 s
->vm_was_running
= true;
3439 case MIGRATION_STATUS_FAILED
:
3440 case MIGRATION_STATUS_CANCELLED
:
3441 case MIGRATION_STATUS_CANCELLING
:
3442 if (s
->vm_was_running
) {
3445 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
3446 runstate_set(RUN_STATE_POSTMIGRATE
);
3452 /* Should not reach here, but if so, forgive the VM. */
3453 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3456 migrate_fd_cleanup_schedule(s
);
3457 qemu_mutex_unlock_iothread();
3460 void migration_make_urgent_request(void)
3462 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
3465 void migration_consume_urgent_request(void)
3467 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
3470 /* Returns true if the rate limiting was broken by an urgent request */
3471 bool migration_rate_limit(void)
3473 int64_t now
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3474 MigrationState
*s
= migrate_get_current();
3476 bool urgent
= false;
3477 migration_update_counters(s
, now
);
3478 if (qemu_file_rate_limit(s
->to_dst_file
)) {
3480 if (qemu_file_get_error(s
->to_dst_file
)) {
3484 * Wait for a delay to do rate limiting OR
3485 * something urgent to post the semaphore.
3487 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- now
;
3488 trace_migration_rate_limit_pre(ms
);
3489 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
3491 * We were woken by one or more urgent things but
3492 * the timedwait will have consumed one of them.
3493 * The service routine for the urgent wake will dec
3494 * the semaphore itself for each item it consumes,
3495 * so add this one we just eat back.
3497 qemu_sem_post(&s
->rate_limit_sem
);
3500 trace_migration_rate_limit_post(urgent
);
3506 * Master migration thread on the source VM.
3507 * It drives the migration and pumps the data down the outgoing channel.
3509 static void *migration_thread(void *opaque
)
3511 MigrationState
*s
= opaque
;
3512 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3513 MigThrError thr_error
;
3514 bool urgent
= false;
3516 rcu_register_thread();
3518 object_ref(OBJECT(s
));
3519 update_iteration_initial_status(s
);
3521 qemu_savevm_state_header(s
->to_dst_file
);
3524 * If we opened the return path, we need to make sure dst has it
3527 if (s
->rp_state
.from_dst_file
) {
3528 /* Now tell the dest that it should open its end so it can reply */
3529 qemu_savevm_send_open_return_path(s
->to_dst_file
);
3531 /* And do a ping that will make stuff easier to debug */
3532 qemu_savevm_send_ping(s
->to_dst_file
, 1);
3535 if (migrate_postcopy()) {
3537 * Tell the destination that we *might* want to do postcopy later;
3538 * if the other end can't do postcopy it should fail now, nice and
3541 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
3544 if (migrate_colo_enabled()) {
3545 /* Notify migration destination that we enable COLO */
3546 qemu_savevm_send_colo_enable(s
->to_dst_file
);
3549 qemu_savevm_state_setup(s
->to_dst_file
);
3551 if (qemu_savevm_state_guest_unplug_pending()) {
3552 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3553 MIGRATION_STATUS_WAIT_UNPLUG
);
3555 while (s
->state
== MIGRATION_STATUS_WAIT_UNPLUG
&&
3556 qemu_savevm_state_guest_unplug_pending()) {
3557 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
3560 migrate_set_state(&s
->state
, MIGRATION_STATUS_WAIT_UNPLUG
,
3561 MIGRATION_STATUS_ACTIVE
);
3564 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3565 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3566 MIGRATION_STATUS_ACTIVE
);
3568 trace_migration_thread_setup_complete();
3570 while (migration_is_active(s
)) {
3571 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
3572 MigIterateState iter_state
= migration_iteration_run(s
);
3573 if (iter_state
== MIG_ITERATE_SKIP
) {
3575 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3581 * Try to detect any kind of failures, and see whether we
3582 * should stop the migration now.
3584 thr_error
= migration_detect_error(s
);
3585 if (thr_error
== MIG_THR_ERR_FATAL
) {
3586 /* Stop migration */
3588 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
3590 * Just recovered from a e.g. network failure, reset all
3591 * the local variables. This is important to avoid
3592 * breaking transferred_bytes and bandwidth calculation
3594 update_iteration_initial_status(s
);
3597 urgent
= migration_rate_limit();
3600 trace_migration_thread_after_loop();
3601 migration_iteration_finish(s
);
3602 object_unref(OBJECT(s
));
3603 rcu_unregister_thread();
3607 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
3609 Error
*local_err
= NULL
;
3611 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
3613 s
->expected_downtime
= s
->parameters
.downtime_limit
;
3615 assert(s
->cleanup_bh
);
3617 assert(!s
->cleanup_bh
);
3618 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup_bh
, s
);
3621 migrate_fd_error(s
, error_in
);
3622 migrate_fd_cleanup(s
);
3627 /* This is a resumed migration */
3628 rate_limit
= s
->parameters
.max_postcopy_bandwidth
/
3631 /* This is a fresh new migration */
3632 rate_limit
= s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
;
3634 /* Notify before starting migration thread */
3635 notifier_list_notify(&migration_state_notifiers
, s
);
3638 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
3639 qemu_file_set_blocking(s
->to_dst_file
, true);
3642 * Open the return path. For postcopy, it is used exclusively. For
3643 * precopy, only if user specified "return-path" capability would
3644 * QEMU uses the return path.
3646 if (migrate_postcopy_ram() || migrate_use_return_path()) {
3647 if (open_return_path_on_source(s
, !resume
)) {
3648 error_report("Unable to open return-path for postcopy");
3649 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
3650 migrate_fd_cleanup(s
);
3656 /* Wakeup the main migration thread to do the recovery */
3657 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
3658 MIGRATION_STATUS_POSTCOPY_RECOVER
);
3659 qemu_sem_post(&s
->postcopy_pause_sem
);
3663 if (multifd_save_setup(&local_err
) != 0) {
3664 error_report_err(local_err
);
3665 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3666 MIGRATION_STATUS_FAILED
);
3667 migrate_fd_cleanup(s
);
3670 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
3671 QEMU_THREAD_JOINABLE
);
3672 s
->migration_thread_running
= true;
3675 void migration_global_dump(Monitor
*mon
)
3677 MigrationState
*ms
= migrate_get_current();
3679 monitor_printf(mon
, "globals:\n");
3680 monitor_printf(mon
, "store-global-state: %s\n",
3681 ms
->store_global_state
? "on" : "off");
3682 monitor_printf(mon
, "only-migratable: %s\n",
3683 only_migratable
? "on" : "off");
3684 monitor_printf(mon
, "send-configuration: %s\n",
3685 ms
->send_configuration
? "on" : "off");
3686 monitor_printf(mon
, "send-section-footer: %s\n",
3687 ms
->send_section_footer
? "on" : "off");
3688 monitor_printf(mon
, "decompress-error-check: %s\n",
3689 ms
->decompress_error_check
? "on" : "off");
3690 monitor_printf(mon
, "clear-bitmap-shift: %u\n",
3691 ms
->clear_bitmap_shift
);
3694 #define DEFINE_PROP_MIG_CAP(name, x) \
3695 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3697 static Property migration_properties
[] = {
3698 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
3699 store_global_state
, true),
3700 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
3701 send_configuration
, true),
3702 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
3703 send_section_footer
, true),
3704 DEFINE_PROP_BOOL("decompress-error-check", MigrationState
,
3705 decompress_error_check
, true),
3706 DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState
,
3707 clear_bitmap_shift
, CLEAR_BITMAP_SHIFT_DEFAULT
),
3709 /* Migration parameters */
3710 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
3711 parameters
.compress_level
,
3712 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
3713 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
3714 parameters
.compress_threads
,
3715 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
3716 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState
,
3717 parameters
.compress_wait_thread
, true),
3718 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
3719 parameters
.decompress_threads
,
3720 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
3721 DEFINE_PROP_UINT8("x-throttle-trigger-threshold", MigrationState
,
3722 parameters
.throttle_trigger_threshold
,
3723 DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD
),
3724 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
3725 parameters
.cpu_throttle_initial
,
3726 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
3727 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
3728 parameters
.cpu_throttle_increment
,
3729 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
3730 DEFINE_PROP_BOOL("x-cpu-throttle-tailslow", MigrationState
,
3731 parameters
.cpu_throttle_tailslow
, false),
3732 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
3733 parameters
.max_bandwidth
, MAX_THROTTLE
),
3734 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
3735 parameters
.downtime_limit
,
3736 DEFAULT_MIGRATE_SET_DOWNTIME
),
3737 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
3738 parameters
.x_checkpoint_delay
,
3739 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
3740 DEFINE_PROP_UINT8("multifd-channels", MigrationState
,
3741 parameters
.multifd_channels
,
3742 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
3743 DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState
,
3744 parameters
.multifd_compression
,
3745 DEFAULT_MIGRATE_MULTIFD_COMPRESSION
),
3746 DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState
,
3747 parameters
.multifd_zlib_level
,
3748 DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL
),
3749 DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState
,
3750 parameters
.multifd_zstd_level
,
3751 DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL
),
3752 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
3753 parameters
.xbzrle_cache_size
,
3754 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
3755 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState
,
3756 parameters
.max_postcopy_bandwidth
,
3757 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH
),
3758 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState
,
3759 parameters
.max_cpu_throttle
,
3760 DEFAULT_MIGRATE_MAX_CPU_THROTTLE
),
3761 DEFINE_PROP_SIZE("announce-initial", MigrationState
,
3762 parameters
.announce_initial
,
3763 DEFAULT_MIGRATE_ANNOUNCE_INITIAL
),
3764 DEFINE_PROP_SIZE("announce-max", MigrationState
,
3765 parameters
.announce_max
,
3766 DEFAULT_MIGRATE_ANNOUNCE_MAX
),
3767 DEFINE_PROP_SIZE("announce-rounds", MigrationState
,
3768 parameters
.announce_rounds
,
3769 DEFAULT_MIGRATE_ANNOUNCE_ROUNDS
),
3770 DEFINE_PROP_SIZE("announce-step", MigrationState
,
3771 parameters
.announce_step
,
3772 DEFAULT_MIGRATE_ANNOUNCE_STEP
),
3774 /* Migration capabilities */
3775 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
3776 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
3777 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
3778 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
3779 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
3780 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
3781 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
3782 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
3783 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
3784 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
3785 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
3786 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD
),
3788 DEFINE_PROP_END_OF_LIST(),
3791 static void migration_class_init(ObjectClass
*klass
, void *data
)
3793 DeviceClass
*dc
= DEVICE_CLASS(klass
);
3795 dc
->user_creatable
= false;
3796 device_class_set_props(dc
, migration_properties
);
3799 static void migration_instance_finalize(Object
*obj
)
3801 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3802 MigrationParameters
*params
= &ms
->parameters
;
3804 qemu_mutex_destroy(&ms
->error_mutex
);
3805 qemu_mutex_destroy(&ms
->qemu_file_lock
);
3806 g_free(params
->tls_hostname
);
3807 g_free(params
->tls_creds
);
3808 qemu_sem_destroy(&ms
->wait_unplug_sem
);
3809 qemu_sem_destroy(&ms
->rate_limit_sem
);
3810 qemu_sem_destroy(&ms
->pause_sem
);
3811 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
3812 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
3813 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
3814 error_free(ms
->error
);
3817 static void migration_instance_init(Object
*obj
)
3819 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3820 MigrationParameters
*params
= &ms
->parameters
;
3822 ms
->state
= MIGRATION_STATUS_NONE
;
3824 ms
->pages_per_second
= -1;
3825 qemu_sem_init(&ms
->pause_sem
, 0);
3826 qemu_mutex_init(&ms
->error_mutex
);
3828 params
->tls_hostname
= g_strdup("");
3829 params
->tls_creds
= g_strdup("");
3831 /* Set has_* up only for parameter checks */
3832 params
->has_compress_level
= true;
3833 params
->has_compress_threads
= true;
3834 params
->has_decompress_threads
= true;
3835 params
->has_throttle_trigger_threshold
= true;
3836 params
->has_cpu_throttle_initial
= true;
3837 params
->has_cpu_throttle_increment
= true;
3838 params
->has_cpu_throttle_tailslow
= true;
3839 params
->has_max_bandwidth
= true;
3840 params
->has_downtime_limit
= true;
3841 params
->has_x_checkpoint_delay
= true;
3842 params
->has_block_incremental
= true;
3843 params
->has_multifd_channels
= true;
3844 params
->has_multifd_compression
= true;
3845 params
->has_multifd_zlib_level
= true;
3846 params
->has_multifd_zstd_level
= true;
3847 params
->has_xbzrle_cache_size
= true;
3848 params
->has_max_postcopy_bandwidth
= true;
3849 params
->has_max_cpu_throttle
= true;
3850 params
->has_announce_initial
= true;
3851 params
->has_announce_max
= true;
3852 params
->has_announce_rounds
= true;
3853 params
->has_announce_step
= true;
3855 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
3856 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
3857 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
3858 qemu_sem_init(&ms
->rate_limit_sem
, 0);
3859 qemu_sem_init(&ms
->wait_unplug_sem
, 0);
3860 qemu_mutex_init(&ms
->qemu_file_lock
);
3864 * Return true if check pass, false otherwise. Error will be put
3865 * inside errp if provided.
3867 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
3869 MigrationCapabilityStatusList
*head
= NULL
;
3870 /* Assuming all off */
3871 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
3874 if (!migrate_params_check(&ms
->parameters
, errp
)) {
3878 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
3879 if (ms
->enabled_capabilities
[i
]) {
3880 QAPI_LIST_PREPEND(head
, migrate_cap_add(i
, true));
3884 ret
= migrate_caps_check(cap_list
, head
, errp
);
3886 /* It works with head == NULL */
3887 qapi_free_MigrationCapabilityStatusList(head
);
3892 static const TypeInfo migration_type
= {
3893 .name
= TYPE_MIGRATION
,
3895 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3896 * not created using qdev_new(), it is not attached to the qdev
3897 * device tree, and it is never realized.
3899 * TODO: Make this TYPE_OBJECT once QOM provides something like
3900 * TYPE_DEVICE's "-global" properties.
3902 .parent
= TYPE_DEVICE
,
3903 .class_init
= migration_class_init
,
3904 .class_size
= sizeof(MigrationClass
),
3905 .instance_size
= sizeof(MigrationState
),
3906 .instance_init
= migration_instance_init
,
3907 .instance_finalize
= migration_instance_finalize
,
3910 static void register_migration_types(void)
3912 type_register_static(&migration_type
);
3915 type_init(register_migration_types
);