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 "monitor/monitor.h"
56 #include "net/announce.h"
57 #include "qemu/queue.h"
61 #include "hw/vfio/vfio-common.h"
64 #define MAX_THROTTLE (128 << 20) /* Migration transfer speed throttling */
66 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
68 #define BUFFER_DELAY 100
69 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
71 /* Time in milliseconds we are allowed to stop the source,
72 * for sending the last part */
73 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
75 /* Maximum migrate downtime set to 2000 seconds */
76 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
77 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
79 /* Default compression thread count */
80 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
81 /* Default decompression thread count, usually decompression is at
82 * least 4 times as fast as compression.*/
83 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
84 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
85 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
86 /* Define default autoconverge cpu throttle migration parameters */
87 #define DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD 50
88 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
89 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
90 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
92 /* Migration XBZRLE default cache size */
93 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
95 /* The delay time (in ms) between two COLO checkpoints */
96 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
97 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
98 #define DEFAULT_MIGRATE_MULTIFD_COMPRESSION MULTIFD_COMPRESSION_NONE
99 /* 0: means nocompress, 1: best speed, ... 9: best compress ratio */
100 #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
101 /* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
102 #define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
104 /* Background transfer rate for postcopy, 0 means unlimited, note
105 * that page requests can still exceed this limit.
107 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
110 * Parameters for self_announce_delay giving a stream of RARP/ARP
111 * packets after migration.
113 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
114 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
115 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
116 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
118 static NotifierList migration_state_notifiers
=
119 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
121 /* Messages sent on the return path from destination to source */
122 enum mig_rp_message_type
{
123 MIG_RP_MSG_INVALID
= 0, /* Must be 0 */
124 MIG_RP_MSG_SHUT
, /* sibling will not send any more RP messages */
125 MIG_RP_MSG_PONG
, /* Response to a PING; data (seq: be32 ) */
127 MIG_RP_MSG_REQ_PAGES_ID
, /* data (start: be64, len: be32, id: string) */
128 MIG_RP_MSG_REQ_PAGES
, /* data (start: be64, len: be32) */
129 MIG_RP_MSG_RECV_BITMAP
, /* send recved_bitmap back to source */
130 MIG_RP_MSG_RESUME_ACK
, /* tell source that we are ready to resume */
135 /* When we add fault tolerance, we could have several
136 migrations at once. For now we don't need to add
137 dynamic creation of migration */
139 static MigrationState
*current_migration
;
140 static MigrationIncomingState
*current_incoming
;
142 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
143 static int migration_maybe_pause(MigrationState
*s
,
144 int *current_active_state
,
146 static void migrate_fd_cancel(MigrationState
*s
);
148 static gint
page_request_addr_cmp(gconstpointer ap
, gconstpointer bp
)
150 uintptr_t a
= (uintptr_t) ap
, b
= (uintptr_t) bp
;
152 return (a
> b
) - (a
< b
);
155 void migration_object_init(void)
159 /* This can only be called once. */
160 assert(!current_migration
);
161 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
164 * Init the migrate incoming object as well no matter whether
165 * we'll use it or not.
167 assert(!current_incoming
);
168 current_incoming
= g_new0(MigrationIncomingState
, 1);
169 current_incoming
->state
= MIGRATION_STATUS_NONE
;
170 current_incoming
->postcopy_remote_fds
=
171 g_array_new(FALSE
, TRUE
, sizeof(struct PostCopyFD
));
172 qemu_mutex_init(¤t_incoming
->rp_mutex
);
173 qemu_event_init(¤t_incoming
->main_thread_load_event
, false);
174 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_dst
, 0);
175 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fault
, 0);
176 qemu_mutex_init(¤t_incoming
->page_request_mutex
);
177 current_incoming
->page_requested
= g_tree_new(page_request_addr_cmp
);
179 if (!migration_object_check(current_migration
, &err
)) {
180 error_report_err(err
);
186 dirty_bitmap_mig_init();
189 void migration_shutdown(void)
192 * Cancel the current migration - that will (eventually)
193 * stop the migration using this structure
195 migrate_fd_cancel(current_migration
);
196 object_unref(OBJECT(current_migration
));
199 * Cancel outgoing migration of dirty bitmaps. It should
200 * at least unref used block nodes.
202 dirty_bitmap_mig_cancel_outgoing();
205 * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
206 * are non-critical data, and their loss never considered as
209 dirty_bitmap_mig_cancel_incoming();
213 MigrationState
*migrate_get_current(void)
215 /* This can only be called after the object created. */
216 assert(current_migration
);
217 return current_migration
;
220 MigrationIncomingState
*migration_incoming_get_current(void)
222 assert(current_incoming
);
223 return current_incoming
;
226 void migration_incoming_state_destroy(void)
228 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
230 if (mis
->to_src_file
) {
231 /* Tell source that we are done */
232 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
233 qemu_fclose(mis
->to_src_file
);
234 mis
->to_src_file
= NULL
;
237 if (mis
->from_src_file
) {
238 qemu_fclose(mis
->from_src_file
);
239 mis
->from_src_file
= NULL
;
241 if (mis
->postcopy_remote_fds
) {
242 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
243 mis
->postcopy_remote_fds
= NULL
;
246 qemu_event_reset(&mis
->main_thread_load_event
);
248 if (mis
->page_requested
) {
249 g_tree_destroy(mis
->page_requested
);
250 mis
->page_requested
= NULL
;
253 if (mis
->socket_address_list
) {
254 qapi_free_SocketAddressList(mis
->socket_address_list
);
255 mis
->socket_address_list
= NULL
;
259 static void migrate_generate_event(int new_state
)
261 if (migrate_use_events()) {
262 qapi_event_send_migration(new_state
);
266 static bool migrate_late_block_activate(void)
270 s
= migrate_get_current();
272 return s
->enabled_capabilities
[
273 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE
];
277 * Send a message on the return channel back to the source
280 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
281 enum mig_rp_message_type message_type
,
282 uint16_t len
, void *data
)
286 trace_migrate_send_rp_message((int)message_type
, len
);
287 qemu_mutex_lock(&mis
->rp_mutex
);
290 * It's possible that the file handle got lost due to network
293 if (!mis
->to_src_file
) {
298 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
299 qemu_put_be16(mis
->to_src_file
, len
);
300 qemu_put_buffer(mis
->to_src_file
, data
, len
);
301 qemu_fflush(mis
->to_src_file
);
303 /* It's possible that qemu file got error during sending */
304 ret
= qemu_file_get_error(mis
->to_src_file
);
307 qemu_mutex_unlock(&mis
->rp_mutex
);
311 /* Request one page from the source VM at the given start address.
312 * rb: the RAMBlock to request the page in
313 * Start: Address offset within the RB
314 * Len: Length in bytes required - must be a multiple of pagesize
316 int migrate_send_rp_message_req_pages(MigrationIncomingState
*mis
,
317 RAMBlock
*rb
, ram_addr_t start
)
319 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
320 size_t msglen
= 12; /* start + len */
321 size_t len
= qemu_ram_pagesize(rb
);
322 enum mig_rp_message_type msg_type
;
326 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
327 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
330 * We maintain the last ramblock that we requested for page. Note that we
331 * don't need locking because this function will only be called within the
332 * postcopy ram fault thread.
334 if (rb
!= mis
->last_rb
) {
337 rbname
= qemu_ram_get_idstr(rb
);
338 rbname_len
= strlen(rbname
);
340 assert(rbname_len
< 256);
342 bufc
[msglen
++] = rbname_len
;
343 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
344 msglen
+= rbname_len
;
345 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
347 msg_type
= MIG_RP_MSG_REQ_PAGES
;
350 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
353 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
,
354 RAMBlock
*rb
, ram_addr_t start
, uint64_t haddr
)
356 void *aligned
= (void *)(uintptr_t)(haddr
& (-qemu_ram_pagesize(rb
)));
357 bool received
= false;
359 WITH_QEMU_LOCK_GUARD(&mis
->page_request_mutex
) {
360 received
= ramblock_recv_bitmap_test_byte_offset(rb
, start
);
361 if (!received
&& !g_tree_lookup(mis
->page_requested
, aligned
)) {
363 * The page has not been received, and it's not yet in the page
364 * request list. Queue it. Set the value of element to 1, so that
365 * things like g_tree_lookup() will return TRUE (1) when found.
367 g_tree_insert(mis
->page_requested
, aligned
, (gpointer
)1);
368 mis
->page_requested_count
++;
369 trace_postcopy_page_req_add(aligned
, mis
->page_requested_count
);
374 * If the page is there, skip sending the message. We don't even need the
375 * lock because as long as the page arrived, it'll be there forever.
381 return migrate_send_rp_message_req_pages(mis
, rb
, start
);
384 static bool migration_colo_enabled
;
385 bool migration_incoming_colo_enabled(void)
387 return migration_colo_enabled
;
390 void migration_incoming_disable_colo(void)
392 ram_block_discard_disable(false);
393 migration_colo_enabled
= false;
396 int migration_incoming_enable_colo(void)
398 if (ram_block_discard_disable(true)) {
399 error_report("COLO: cannot disable RAM discard");
402 migration_colo_enabled
= true;
406 void migrate_add_address(SocketAddress
*address
)
408 MigrationIncomingState
*mis
= migration_incoming_get_current();
409 SocketAddressList
*addrs
;
411 addrs
= g_new0(SocketAddressList
, 1);
412 addrs
->next
= mis
->socket_address_list
;
413 mis
->socket_address_list
= addrs
;
414 addrs
->value
= QAPI_CLONE(SocketAddress
, address
);
417 static void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
419 const char *p
= NULL
;
421 qapi_event_send_migration(MIGRATION_STATUS_SETUP
);
422 if (strstart(uri
, "tcp:", &p
) ||
423 strstart(uri
, "unix:", NULL
) ||
424 strstart(uri
, "vsock:", NULL
)) {
425 socket_start_incoming_migration(p
? p
: uri
, errp
);
427 } else if (strstart(uri
, "rdma:", &p
)) {
428 rdma_start_incoming_migration(p
, errp
);
430 } else if (strstart(uri
, "exec:", &p
)) {
431 exec_start_incoming_migration(p
, errp
);
432 } else if (strstart(uri
, "fd:", &p
)) {
433 fd_start_incoming_migration(p
, errp
);
435 error_setg(errp
, "unknown migration protocol: %s", uri
);
439 static void process_incoming_migration_bh(void *opaque
)
441 Error
*local_err
= NULL
;
442 MigrationIncomingState
*mis
= opaque
;
444 /* If capability late_block_activate is set:
445 * Only fire up the block code now if we're going to restart the
446 * VM, else 'cont' will do it.
447 * This causes file locking to happen; so we don't want it to happen
448 * unless we really are starting the VM.
450 if (!migrate_late_block_activate() ||
451 (autostart
&& (!global_state_received() ||
452 global_state_get_runstate() == RUN_STATE_RUNNING
))) {
453 /* Make sure all file formats flush their mutable metadata.
454 * If we get an error here, just don't restart the VM yet. */
455 bdrv_invalidate_cache_all(&local_err
);
457 error_report_err(local_err
);
464 * This must happen after all error conditions are dealt with and
465 * we're sure the VM is going to be running on this host.
467 qemu_announce_self(&mis
->announce_timer
, migrate_announce_params());
469 if (multifd_load_cleanup(&local_err
) != 0) {
470 error_report_err(local_err
);
473 /* If global state section was not received or we are in running
474 state, we need to obey autostart. Any other state is set with
477 dirty_bitmap_mig_before_vm_start();
479 if (!global_state_received() ||
480 global_state_get_runstate() == RUN_STATE_RUNNING
) {
484 runstate_set(RUN_STATE_PAUSED
);
486 } else if (migration_incoming_colo_enabled()) {
487 migration_incoming_disable_colo();
490 runstate_set(global_state_get_runstate());
493 * This must happen after any state changes since as soon as an external
494 * observer sees this event they might start to prod at the VM assuming
497 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
498 MIGRATION_STATUS_COMPLETED
);
499 qemu_bh_delete(mis
->bh
);
500 migration_incoming_state_destroy();
503 static void process_incoming_migration_co(void *opaque
)
505 MigrationIncomingState
*mis
= migration_incoming_get_current();
508 Error
*local_err
= NULL
;
510 assert(mis
->from_src_file
);
511 mis
->migration_incoming_co
= qemu_coroutine_self();
512 mis
->largest_page_size
= qemu_ram_pagesize_largest();
513 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
514 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
515 MIGRATION_STATUS_ACTIVE
);
516 ret
= qemu_loadvm_state(mis
->from_src_file
);
518 ps
= postcopy_state_get();
519 trace_process_incoming_migration_co_end(ret
, ps
);
520 if (ps
!= POSTCOPY_INCOMING_NONE
) {
521 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
523 * Where a migration had postcopy enabled (and thus went to advise)
524 * but managed to complete within the precopy period, we can use
527 postcopy_ram_incoming_cleanup(mis
);
528 } else if (ret
>= 0) {
530 * Postcopy was started, cleanup should happen at the end of the
533 trace_process_incoming_migration_co_postcopy_end_main();
536 /* Else if something went wrong then just fall out of the normal exit */
539 /* we get COLO info, and know if we are in COLO mode */
540 if (!ret
&& migration_incoming_colo_enabled()) {
541 /* Make sure all file formats flush their mutable metadata */
542 bdrv_invalidate_cache_all(&local_err
);
544 error_report_err(local_err
);
548 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
549 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
550 mis
->have_colo_incoming_thread
= true;
551 qemu_coroutine_yield();
553 /* Wait checkpoint incoming thread exit before free resource */
554 qemu_thread_join(&mis
->colo_incoming_thread
);
555 /* We hold the global iothread lock, so it is safe here */
556 colo_release_ram_cache();
560 error_report("load of migration failed: %s", strerror(-ret
));
563 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
564 qemu_bh_schedule(mis
->bh
);
565 mis
->migration_incoming_co
= NULL
;
569 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
570 MIGRATION_STATUS_FAILED
);
571 qemu_fclose(mis
->from_src_file
);
572 if (multifd_load_cleanup(&local_err
) != 0) {
573 error_report_err(local_err
);
579 * @migration_incoming_setup: Setup incoming migration
581 * Returns 0 for no error or 1 for error
583 * @f: file for main migration channel
584 * @errp: where to put errors
586 static int migration_incoming_setup(QEMUFile
*f
, Error
**errp
)
588 MigrationIncomingState
*mis
= migration_incoming_get_current();
589 Error
*local_err
= NULL
;
591 if (multifd_load_setup(&local_err
) != 0) {
592 /* We haven't been able to create multifd threads
593 nothing better to do */
594 error_report_err(local_err
);
598 if (!mis
->from_src_file
) {
599 mis
->from_src_file
= f
;
601 qemu_file_set_blocking(f
, false);
605 void migration_incoming_process(void)
607 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
608 qemu_coroutine_enter(co
);
611 /* Returns true if recovered from a paused migration, otherwise false */
612 static bool postcopy_try_recover(QEMUFile
*f
)
614 MigrationIncomingState
*mis
= migration_incoming_get_current();
616 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
617 /* Resumed from a paused postcopy migration */
619 mis
->from_src_file
= f
;
620 /* Postcopy has standalone thread to do vm load */
621 qemu_file_set_blocking(f
, true);
623 /* Re-configure the return path */
624 mis
->to_src_file
= qemu_file_get_return_path(f
);
626 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
627 MIGRATION_STATUS_POSTCOPY_RECOVER
);
630 * Here, we only wake up the main loading thread (while the
631 * fault thread will still be waiting), so that we can receive
632 * commands from source now, and answer it if needed. The
633 * fault thread will be woken up afterwards until we are sure
634 * that source is ready to reply to page requests.
636 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
643 void migration_fd_process_incoming(QEMUFile
*f
, Error
**errp
)
645 Error
*local_err
= NULL
;
647 if (postcopy_try_recover(f
)) {
651 if (migration_incoming_setup(f
, &local_err
)) {
652 error_propagate(errp
, local_err
);
655 migration_incoming_process();
658 void migration_ioc_process_incoming(QIOChannel
*ioc
, Error
**errp
)
660 MigrationIncomingState
*mis
= migration_incoming_get_current();
661 Error
*local_err
= NULL
;
662 bool start_migration
;
664 if (!mis
->from_src_file
) {
665 /* The first connection (multifd may have multiple) */
666 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
668 /* If it's a recovery, we're done */
669 if (postcopy_try_recover(f
)) {
673 if (migration_incoming_setup(f
, &local_err
)) {
674 error_propagate(errp
, local_err
);
679 * Common migration only needs one channel, so we can start
680 * right now. Multifd needs more than one channel, we wait.
682 start_migration
= !migrate_use_multifd();
684 /* Multiple connections */
685 assert(migrate_use_multifd());
686 start_migration
= multifd_recv_new_channel(ioc
, &local_err
);
688 error_propagate(errp
, local_err
);
693 if (start_migration
) {
694 migration_incoming_process();
699 * @migration_has_all_channels: We have received all channels that we need
701 * Returns true when we have got connections to all the channels that
702 * we need for migration.
704 bool migration_has_all_channels(void)
706 MigrationIncomingState
*mis
= migration_incoming_get_current();
709 all_channels
= multifd_recv_all_channels_created();
711 return all_channels
&& mis
->from_src_file
!= NULL
;
715 * Send a 'SHUT' message on the return channel with the given value
716 * to indicate that we've finished with the RP. Non-0 value indicates
719 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
724 buf
= cpu_to_be32(value
);
725 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
729 * Send a 'PONG' message on the return channel with the given value
730 * (normally in response to a 'PING')
732 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
737 buf
= cpu_to_be32(value
);
738 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
741 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
749 * First, we send the header part. It contains only the len of
750 * idstr, and the idstr itself.
752 len
= strlen(block_name
);
754 memcpy(buf
+ 1, block_name
, len
);
756 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
757 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
762 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
765 * Next, we dump the received bitmap to the stream.
767 * TODO: currently we are safe since we are the only one that is
768 * using the to_src_file handle (fault thread is still paused),
769 * and it's ok even not taking the mutex. However the best way is
770 * to take the lock before sending the message header, and release
771 * the lock after sending the bitmap.
773 qemu_mutex_lock(&mis
->rp_mutex
);
774 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
775 qemu_mutex_unlock(&mis
->rp_mutex
);
777 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
780 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
784 buf
= cpu_to_be32(value
);
785 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
788 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
790 MigrationCapabilityStatusList
*head
= NULL
;
791 MigrationCapabilityStatusList
*caps
;
792 MigrationState
*s
= migrate_get_current();
795 caps
= NULL
; /* silence compiler warning */
796 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
797 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
798 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
803 head
= g_malloc0(sizeof(*caps
));
806 caps
->next
= g_malloc0(sizeof(*caps
));
810 g_malloc(sizeof(*caps
->value
));
811 caps
->value
->capability
= i
;
812 caps
->value
->state
= s
->enabled_capabilities
[i
];
818 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
820 MigrationParameters
*params
;
821 MigrationState
*s
= migrate_get_current();
823 /* TODO use QAPI_CLONE() instead of duplicating it inline */
824 params
= g_malloc0(sizeof(*params
));
825 params
->has_compress_level
= true;
826 params
->compress_level
= s
->parameters
.compress_level
;
827 params
->has_compress_threads
= true;
828 params
->compress_threads
= s
->parameters
.compress_threads
;
829 params
->has_compress_wait_thread
= true;
830 params
->compress_wait_thread
= s
->parameters
.compress_wait_thread
;
831 params
->has_decompress_threads
= true;
832 params
->decompress_threads
= s
->parameters
.decompress_threads
;
833 params
->has_throttle_trigger_threshold
= true;
834 params
->throttle_trigger_threshold
= s
->parameters
.throttle_trigger_threshold
;
835 params
->has_cpu_throttle_initial
= true;
836 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
837 params
->has_cpu_throttle_increment
= true;
838 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
839 params
->has_cpu_throttle_tailslow
= true;
840 params
->cpu_throttle_tailslow
= s
->parameters
.cpu_throttle_tailslow
;
841 params
->has_tls_creds
= true;
842 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
843 params
->has_tls_hostname
= true;
844 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
845 params
->has_tls_authz
= true;
846 params
->tls_authz
= g_strdup(s
->parameters
.tls_authz
?
847 s
->parameters
.tls_authz
: "");
848 params
->has_max_bandwidth
= true;
849 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
850 params
->has_downtime_limit
= true;
851 params
->downtime_limit
= s
->parameters
.downtime_limit
;
852 params
->has_x_checkpoint_delay
= true;
853 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
854 params
->has_block_incremental
= true;
855 params
->block_incremental
= s
->parameters
.block_incremental
;
856 params
->has_multifd_channels
= true;
857 params
->multifd_channels
= s
->parameters
.multifd_channels
;
858 params
->has_multifd_compression
= true;
859 params
->multifd_compression
= s
->parameters
.multifd_compression
;
860 params
->has_multifd_zlib_level
= true;
861 params
->multifd_zlib_level
= s
->parameters
.multifd_zlib_level
;
862 params
->has_multifd_zstd_level
= true;
863 params
->multifd_zstd_level
= s
->parameters
.multifd_zstd_level
;
864 params
->has_xbzrle_cache_size
= true;
865 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
866 params
->has_max_postcopy_bandwidth
= true;
867 params
->max_postcopy_bandwidth
= s
->parameters
.max_postcopy_bandwidth
;
868 params
->has_max_cpu_throttle
= true;
869 params
->max_cpu_throttle
= s
->parameters
.max_cpu_throttle
;
870 params
->has_announce_initial
= true;
871 params
->announce_initial
= s
->parameters
.announce_initial
;
872 params
->has_announce_max
= true;
873 params
->announce_max
= s
->parameters
.announce_max
;
874 params
->has_announce_rounds
= true;
875 params
->announce_rounds
= s
->parameters
.announce_rounds
;
876 params
->has_announce_step
= true;
877 params
->announce_step
= s
->parameters
.announce_step
;
879 if (s
->parameters
.has_block_bitmap_mapping
) {
880 params
->has_block_bitmap_mapping
= true;
881 params
->block_bitmap_mapping
=
882 QAPI_CLONE(BitmapMigrationNodeAliasList
,
883 s
->parameters
.block_bitmap_mapping
);
889 AnnounceParameters
*migrate_announce_params(void)
891 static AnnounceParameters ap
;
893 MigrationState
*s
= migrate_get_current();
895 ap
.initial
= s
->parameters
.announce_initial
;
896 ap
.max
= s
->parameters
.announce_max
;
897 ap
.rounds
= s
->parameters
.announce_rounds
;
898 ap
.step
= s
->parameters
.announce_step
;
904 * Return true if we're already in the middle of a migration
905 * (i.e. any of the active or setup states)
907 bool migration_is_setup_or_active(int state
)
910 case MIGRATION_STATUS_ACTIVE
:
911 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
912 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
913 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
914 case MIGRATION_STATUS_SETUP
:
915 case MIGRATION_STATUS_PRE_SWITCHOVER
:
916 case MIGRATION_STATUS_DEVICE
:
917 case MIGRATION_STATUS_WAIT_UNPLUG
:
918 case MIGRATION_STATUS_COLO
:
927 bool migration_is_running(int state
)
930 case MIGRATION_STATUS_ACTIVE
:
931 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
932 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
933 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
934 case MIGRATION_STATUS_SETUP
:
935 case MIGRATION_STATUS_PRE_SWITCHOVER
:
936 case MIGRATION_STATUS_DEVICE
:
937 case MIGRATION_STATUS_WAIT_UNPLUG
:
938 case MIGRATION_STATUS_CANCELLING
:
947 static void populate_time_info(MigrationInfo
*info
, MigrationState
*s
)
949 info
->has_status
= true;
950 info
->has_setup_time
= true;
951 info
->setup_time
= s
->setup_time
;
952 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
953 info
->has_total_time
= true;
954 info
->total_time
= s
->total_time
;
955 info
->has_downtime
= true;
956 info
->downtime
= s
->downtime
;
958 info
->has_total_time
= true;
959 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) -
961 info
->has_expected_downtime
= true;
962 info
->expected_downtime
= s
->expected_downtime
;
966 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
968 info
->has_ram
= true;
969 info
->ram
= g_malloc0(sizeof(*info
->ram
));
970 info
->ram
->transferred
= ram_counters
.transferred
;
971 info
->ram
->total
= ram_bytes_total();
972 info
->ram
->duplicate
= ram_counters
.duplicate
;
973 /* legacy value. It is not used anymore */
974 info
->ram
->skipped
= 0;
975 info
->ram
->normal
= ram_counters
.normal
;
976 info
->ram
->normal_bytes
= ram_counters
.normal
*
977 qemu_target_page_size();
978 info
->ram
->mbps
= s
->mbps
;
979 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
980 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
981 info
->ram
->page_size
= qemu_target_page_size();
982 info
->ram
->multifd_bytes
= ram_counters
.multifd_bytes
;
983 info
->ram
->pages_per_second
= s
->pages_per_second
;
985 if (migrate_use_xbzrle()) {
986 info
->has_xbzrle_cache
= true;
987 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
988 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
989 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
990 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
991 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
992 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
993 info
->xbzrle_cache
->encoding_rate
= xbzrle_counters
.encoding_rate
;
994 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
997 if (migrate_use_compression()) {
998 info
->has_compression
= true;
999 info
->compression
= g_malloc0(sizeof(*info
->compression
));
1000 info
->compression
->pages
= compression_counters
.pages
;
1001 info
->compression
->busy
= compression_counters
.busy
;
1002 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
1003 info
->compression
->compressed_size
=
1004 compression_counters
.compressed_size
;
1005 info
->compression
->compression_rate
=
1006 compression_counters
.compression_rate
;
1009 if (cpu_throttle_active()) {
1010 info
->has_cpu_throttle_percentage
= true;
1011 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
1014 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
1015 info
->ram
->remaining
= ram_bytes_remaining();
1016 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
1020 static void populate_disk_info(MigrationInfo
*info
)
1022 if (blk_mig_active()) {
1023 info
->has_disk
= true;
1024 info
->disk
= g_malloc0(sizeof(*info
->disk
));
1025 info
->disk
->transferred
= blk_mig_bytes_transferred();
1026 info
->disk
->remaining
= blk_mig_bytes_remaining();
1027 info
->disk
->total
= blk_mig_bytes_total();
1031 static void populate_vfio_info(MigrationInfo
*info
)
1034 if (vfio_mig_active()) {
1035 info
->has_vfio
= true;
1036 info
->vfio
= g_malloc0(sizeof(*info
->vfio
));
1037 info
->vfio
->transferred
= vfio_mig_bytes_transferred();
1042 static void fill_source_migration_info(MigrationInfo
*info
)
1044 MigrationState
*s
= migrate_get_current();
1047 case MIGRATION_STATUS_NONE
:
1048 /* no migration has happened ever */
1049 /* do not overwrite destination migration status */
1051 case MIGRATION_STATUS_SETUP
:
1052 info
->has_status
= true;
1053 info
->has_total_time
= false;
1055 case MIGRATION_STATUS_ACTIVE
:
1056 case MIGRATION_STATUS_CANCELLING
:
1057 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1058 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1059 case MIGRATION_STATUS_DEVICE
:
1060 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1061 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1062 /* TODO add some postcopy stats */
1063 populate_time_info(info
, s
);
1064 populate_ram_info(info
, s
);
1065 populate_disk_info(info
);
1066 populate_vfio_info(info
);
1068 case MIGRATION_STATUS_COLO
:
1069 info
->has_status
= true;
1070 /* TODO: display COLO specific information (checkpoint info etc.) */
1072 case MIGRATION_STATUS_COMPLETED
:
1073 populate_time_info(info
, s
);
1074 populate_ram_info(info
, s
);
1075 populate_vfio_info(info
);
1077 case MIGRATION_STATUS_FAILED
:
1078 info
->has_status
= true;
1080 info
->has_error_desc
= true;
1081 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
1084 case MIGRATION_STATUS_CANCELLED
:
1085 info
->has_status
= true;
1087 case MIGRATION_STATUS_WAIT_UNPLUG
:
1088 info
->has_status
= true;
1091 info
->status
= s
->state
;
1095 * @migration_caps_check - check capability validity
1097 * @cap_list: old capability list, array of bool
1098 * @params: new capabilities to be applied soon
1099 * @errp: set *errp if the check failed, with reason
1101 * Returns true if check passed, otherwise false.
1103 static bool migrate_caps_check(bool *cap_list
,
1104 MigrationCapabilityStatusList
*params
,
1107 MigrationCapabilityStatusList
*cap
;
1108 bool old_postcopy_cap
;
1109 MigrationIncomingState
*mis
= migration_incoming_get_current();
1111 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1113 for (cap
= params
; cap
; cap
= cap
->next
) {
1114 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
1117 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1118 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
1119 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
1121 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
1126 #ifndef CONFIG_REPLICATION
1127 if (cap_list
[MIGRATION_CAPABILITY_X_COLO
]) {
1128 error_setg(errp
, "QEMU compiled without replication module"
1129 " can't enable COLO");
1130 error_append_hint(errp
, "Please enable replication before COLO.\n");
1135 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
1136 /* This check is reasonably expensive, so only when it's being
1137 * set the first time, also it's only the destination that needs
1140 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
1141 !postcopy_ram_supported_by_host(mis
)) {
1142 /* postcopy_ram_supported_by_host will have emitted a more
1145 error_setg(errp
, "Postcopy is not supported");
1149 if (cap_list
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
]) {
1150 error_setg(errp
, "Postcopy is not compatible with ignore-shared");
1158 static void fill_destination_migration_info(MigrationInfo
*info
)
1160 MigrationIncomingState
*mis
= migration_incoming_get_current();
1162 if (mis
->socket_address_list
) {
1163 info
->has_socket_address
= true;
1164 info
->socket_address
=
1165 QAPI_CLONE(SocketAddressList
, mis
->socket_address_list
);
1168 switch (mis
->state
) {
1169 case MIGRATION_STATUS_NONE
:
1171 case MIGRATION_STATUS_SETUP
:
1172 case MIGRATION_STATUS_CANCELLING
:
1173 case MIGRATION_STATUS_CANCELLED
:
1174 case MIGRATION_STATUS_ACTIVE
:
1175 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1176 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1177 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1178 case MIGRATION_STATUS_FAILED
:
1179 case MIGRATION_STATUS_COLO
:
1180 info
->has_status
= true;
1182 case MIGRATION_STATUS_COMPLETED
:
1183 info
->has_status
= true;
1184 fill_destination_postcopy_migration_info(info
);
1187 info
->status
= mis
->state
;
1190 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1192 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1194 fill_destination_migration_info(info
);
1195 fill_source_migration_info(info
);
1200 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
1203 MigrationState
*s
= migrate_get_current();
1204 MigrationCapabilityStatusList
*cap
;
1205 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
1207 if (migration_is_running(s
->state
)) {
1208 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1212 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
1213 if (!migrate_caps_check(cap_list
, params
, errp
)) {
1217 for (cap
= params
; cap
; cap
= cap
->next
) {
1218 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
1223 * Check whether the parameters are valid. Error will be put into errp
1224 * (if provided). Return true if valid, otherwise false.
1226 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
1228 if (params
->has_compress_level
&&
1229 (params
->compress_level
> 9)) {
1230 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
1231 "is invalid, it should be in the range of 0 to 9");
1235 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
1236 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1238 "is invalid, it should be in the range of 1 to 255");
1242 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
1243 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1244 "decompress_threads",
1245 "is invalid, it should be in the range of 1 to 255");
1249 if (params
->has_throttle_trigger_threshold
&&
1250 (params
->throttle_trigger_threshold
< 1 ||
1251 params
->throttle_trigger_threshold
> 100)) {
1252 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1253 "throttle_trigger_threshold",
1254 "an integer in the range of 1 to 100");
1258 if (params
->has_cpu_throttle_initial
&&
1259 (params
->cpu_throttle_initial
< 1 ||
1260 params
->cpu_throttle_initial
> 99)) {
1261 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1262 "cpu_throttle_initial",
1263 "an integer in the range of 1 to 99");
1267 if (params
->has_cpu_throttle_increment
&&
1268 (params
->cpu_throttle_increment
< 1 ||
1269 params
->cpu_throttle_increment
> 99)) {
1270 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1271 "cpu_throttle_increment",
1272 "an integer in the range of 1 to 99");
1276 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
1277 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1279 "an integer in the range of 0 to "stringify(SIZE_MAX
)
1284 if (params
->has_downtime_limit
&&
1285 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
1286 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1288 "an integer in the range of 0 to "
1289 stringify(MAX_MIGRATE_DOWNTIME
)" ms");
1293 /* x_checkpoint_delay is now always positive */
1295 if (params
->has_multifd_channels
&& (params
->multifd_channels
< 1)) {
1296 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1298 "is invalid, it should be in the range of 1 to 255");
1302 if (params
->has_multifd_zlib_level
&&
1303 (params
->multifd_zlib_level
> 9)) {
1304 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zlib_level",
1305 "is invalid, it should be in the range of 0 to 9");
1309 if (params
->has_multifd_zstd_level
&&
1310 (params
->multifd_zstd_level
> 20)) {
1311 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zstd_level",
1312 "is invalid, it should be in the range of 0 to 20");
1316 if (params
->has_xbzrle_cache_size
&&
1317 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
1318 !is_power_of_2(params
->xbzrle_cache_size
))) {
1319 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1320 "xbzrle_cache_size",
1321 "is invalid, it should be bigger than target page size"
1322 " and a power of 2");
1326 if (params
->has_max_cpu_throttle
&&
1327 (params
->max_cpu_throttle
< params
->cpu_throttle_initial
||
1328 params
->max_cpu_throttle
> 99)) {
1329 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1331 "an integer in the range of cpu_throttle_initial to 99");
1335 if (params
->has_announce_initial
&&
1336 params
->announce_initial
> 100000) {
1337 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1339 "is invalid, it must be less than 100000 ms");
1342 if (params
->has_announce_max
&&
1343 params
->announce_max
> 100000) {
1344 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1346 "is invalid, it must be less than 100000 ms");
1349 if (params
->has_announce_rounds
&&
1350 params
->announce_rounds
> 1000) {
1351 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1353 "is invalid, it must be in the range of 0 to 1000");
1356 if (params
->has_announce_step
&&
1357 (params
->announce_step
< 1 ||
1358 params
->announce_step
> 10000)) {
1359 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1361 "is invalid, it must be in the range of 1 to 10000 ms");
1365 if (params
->has_block_bitmap_mapping
&&
1366 !check_dirty_bitmap_mig_alias_map(params
->block_bitmap_mapping
, errp
)) {
1367 error_prepend(errp
, "Invalid mapping given for block-bitmap-mapping: ");
1374 static void migrate_params_test_apply(MigrateSetParameters
*params
,
1375 MigrationParameters
*dest
)
1377 *dest
= migrate_get_current()->parameters
;
1379 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1381 if (params
->has_compress_level
) {
1382 dest
->compress_level
= params
->compress_level
;
1385 if (params
->has_compress_threads
) {
1386 dest
->compress_threads
= params
->compress_threads
;
1389 if (params
->has_compress_wait_thread
) {
1390 dest
->compress_wait_thread
= params
->compress_wait_thread
;
1393 if (params
->has_decompress_threads
) {
1394 dest
->decompress_threads
= params
->decompress_threads
;
1397 if (params
->has_throttle_trigger_threshold
) {
1398 dest
->throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1401 if (params
->has_cpu_throttle_initial
) {
1402 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
1405 if (params
->has_cpu_throttle_increment
) {
1406 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
1409 if (params
->has_cpu_throttle_tailslow
) {
1410 dest
->cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1413 if (params
->has_tls_creds
) {
1414 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1415 dest
->tls_creds
= params
->tls_creds
->u
.s
;
1418 if (params
->has_tls_hostname
) {
1419 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1420 dest
->tls_hostname
= params
->tls_hostname
->u
.s
;
1423 if (params
->has_max_bandwidth
) {
1424 dest
->max_bandwidth
= params
->max_bandwidth
;
1427 if (params
->has_downtime_limit
) {
1428 dest
->downtime_limit
= params
->downtime_limit
;
1431 if (params
->has_x_checkpoint_delay
) {
1432 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
1435 if (params
->has_block_incremental
) {
1436 dest
->block_incremental
= params
->block_incremental
;
1438 if (params
->has_multifd_channels
) {
1439 dest
->multifd_channels
= params
->multifd_channels
;
1441 if (params
->has_multifd_compression
) {
1442 dest
->multifd_compression
= params
->multifd_compression
;
1444 if (params
->has_xbzrle_cache_size
) {
1445 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
1447 if (params
->has_max_postcopy_bandwidth
) {
1448 dest
->max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1450 if (params
->has_max_cpu_throttle
) {
1451 dest
->max_cpu_throttle
= params
->max_cpu_throttle
;
1453 if (params
->has_announce_initial
) {
1454 dest
->announce_initial
= params
->announce_initial
;
1456 if (params
->has_announce_max
) {
1457 dest
->announce_max
= params
->announce_max
;
1459 if (params
->has_announce_rounds
) {
1460 dest
->announce_rounds
= params
->announce_rounds
;
1462 if (params
->has_announce_step
) {
1463 dest
->announce_step
= params
->announce_step
;
1466 if (params
->has_block_bitmap_mapping
) {
1467 dest
->has_block_bitmap_mapping
= true;
1468 dest
->block_bitmap_mapping
= params
->block_bitmap_mapping
;
1472 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
1474 MigrationState
*s
= migrate_get_current();
1476 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1478 if (params
->has_compress_level
) {
1479 s
->parameters
.compress_level
= params
->compress_level
;
1482 if (params
->has_compress_threads
) {
1483 s
->parameters
.compress_threads
= params
->compress_threads
;
1486 if (params
->has_compress_wait_thread
) {
1487 s
->parameters
.compress_wait_thread
= params
->compress_wait_thread
;
1490 if (params
->has_decompress_threads
) {
1491 s
->parameters
.decompress_threads
= params
->decompress_threads
;
1494 if (params
->has_throttle_trigger_threshold
) {
1495 s
->parameters
.throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1498 if (params
->has_cpu_throttle_initial
) {
1499 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
1502 if (params
->has_cpu_throttle_increment
) {
1503 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
1506 if (params
->has_cpu_throttle_tailslow
) {
1507 s
->parameters
.cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1510 if (params
->has_tls_creds
) {
1511 g_free(s
->parameters
.tls_creds
);
1512 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1513 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1516 if (params
->has_tls_hostname
) {
1517 g_free(s
->parameters
.tls_hostname
);
1518 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1519 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1522 if (params
->has_tls_authz
) {
1523 g_free(s
->parameters
.tls_authz
);
1524 assert(params
->tls_authz
->type
== QTYPE_QSTRING
);
1525 s
->parameters
.tls_authz
= g_strdup(params
->tls_authz
->u
.s
);
1528 if (params
->has_max_bandwidth
) {
1529 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1530 if (s
->to_dst_file
&& !migration_in_postcopy()) {
1531 qemu_file_set_rate_limit(s
->to_dst_file
,
1532 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1536 if (params
->has_downtime_limit
) {
1537 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1540 if (params
->has_x_checkpoint_delay
) {
1541 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1542 if (migration_in_colo_state()) {
1543 colo_checkpoint_notify(s
);
1547 if (params
->has_block_incremental
) {
1548 s
->parameters
.block_incremental
= params
->block_incremental
;
1550 if (params
->has_multifd_channels
) {
1551 s
->parameters
.multifd_channels
= params
->multifd_channels
;
1553 if (params
->has_multifd_compression
) {
1554 s
->parameters
.multifd_compression
= params
->multifd_compression
;
1556 if (params
->has_xbzrle_cache_size
) {
1557 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1558 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1560 if (params
->has_max_postcopy_bandwidth
) {
1561 s
->parameters
.max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1562 if (s
->to_dst_file
&& migration_in_postcopy()) {
1563 qemu_file_set_rate_limit(s
->to_dst_file
,
1564 s
->parameters
.max_postcopy_bandwidth
/ XFER_LIMIT_RATIO
);
1567 if (params
->has_max_cpu_throttle
) {
1568 s
->parameters
.max_cpu_throttle
= params
->max_cpu_throttle
;
1570 if (params
->has_announce_initial
) {
1571 s
->parameters
.announce_initial
= params
->announce_initial
;
1573 if (params
->has_announce_max
) {
1574 s
->parameters
.announce_max
= params
->announce_max
;
1576 if (params
->has_announce_rounds
) {
1577 s
->parameters
.announce_rounds
= params
->announce_rounds
;
1579 if (params
->has_announce_step
) {
1580 s
->parameters
.announce_step
= params
->announce_step
;
1583 if (params
->has_block_bitmap_mapping
) {
1584 qapi_free_BitmapMigrationNodeAliasList(
1585 s
->parameters
.block_bitmap_mapping
);
1587 s
->parameters
.has_block_bitmap_mapping
= true;
1588 s
->parameters
.block_bitmap_mapping
=
1589 QAPI_CLONE(BitmapMigrationNodeAliasList
,
1590 params
->block_bitmap_mapping
);
1594 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1596 MigrationParameters tmp
;
1598 /* TODO Rewrite "" to null instead */
1599 if (params
->has_tls_creds
1600 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1601 qobject_unref(params
->tls_creds
->u
.n
);
1602 params
->tls_creds
->type
= QTYPE_QSTRING
;
1603 params
->tls_creds
->u
.s
= strdup("");
1605 /* TODO Rewrite "" to null instead */
1606 if (params
->has_tls_hostname
1607 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1608 qobject_unref(params
->tls_hostname
->u
.n
);
1609 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1610 params
->tls_hostname
->u
.s
= strdup("");
1613 migrate_params_test_apply(params
, &tmp
);
1615 if (!migrate_params_check(&tmp
, errp
)) {
1616 /* Invalid parameter */
1620 migrate_params_apply(params
, errp
);
1624 void qmp_migrate_start_postcopy(Error
**errp
)
1626 MigrationState
*s
= migrate_get_current();
1628 if (!migrate_postcopy()) {
1629 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1630 " the start of migration");
1634 if (s
->state
== MIGRATION_STATUS_NONE
) {
1635 error_setg(errp
, "Postcopy must be started after migration has been"
1640 * we don't error if migration has finished since that would be racy
1641 * with issuing this command.
1643 qatomic_set(&s
->start_postcopy
, true);
1646 /* shared migration helpers */
1648 void migrate_set_state(int *state
, int old_state
, int new_state
)
1650 assert(new_state
< MIGRATION_STATUS__MAX
);
1651 if (qatomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1652 trace_migrate_set_state(MigrationStatus_str(new_state
));
1653 migrate_generate_event(new_state
);
1657 static MigrationCapabilityStatusList
*migrate_cap_add(
1658 MigrationCapabilityStatusList
*list
,
1659 MigrationCapability index
,
1662 MigrationCapabilityStatusList
*cap
;
1664 cap
= g_new0(MigrationCapabilityStatusList
, 1);
1665 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
1666 cap
->value
->capability
= index
;
1667 cap
->value
->state
= state
;
1673 void migrate_set_block_enabled(bool value
, Error
**errp
)
1675 MigrationCapabilityStatusList
*cap
;
1677 cap
= migrate_cap_add(NULL
, MIGRATION_CAPABILITY_BLOCK
, value
);
1678 qmp_migrate_set_capabilities(cap
, errp
);
1679 qapi_free_MigrationCapabilityStatusList(cap
);
1682 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1684 s
->parameters
.block_incremental
= value
;
1687 static void block_cleanup_parameters(MigrationState
*s
)
1689 if (s
->must_remove_block_options
) {
1690 /* setting to false can never fail */
1691 migrate_set_block_enabled(false, &error_abort
);
1692 migrate_set_block_incremental(s
, false);
1693 s
->must_remove_block_options
= false;
1697 static void migrate_fd_cleanup(MigrationState
*s
)
1699 qemu_bh_delete(s
->cleanup_bh
);
1700 s
->cleanup_bh
= NULL
;
1702 qemu_savevm_state_cleanup();
1704 if (s
->to_dst_file
) {
1707 trace_migrate_fd_cleanup();
1708 qemu_mutex_unlock_iothread();
1709 if (s
->migration_thread_running
) {
1710 qemu_thread_join(&s
->thread
);
1711 s
->migration_thread_running
= false;
1713 qemu_mutex_lock_iothread();
1715 multifd_save_cleanup();
1716 qemu_mutex_lock(&s
->qemu_file_lock
);
1717 tmp
= s
->to_dst_file
;
1718 s
->to_dst_file
= NULL
;
1719 qemu_mutex_unlock(&s
->qemu_file_lock
);
1721 * Close the file handle without the lock to make sure the
1722 * critical section won't block for long.
1727 assert(!migration_is_active(s
));
1729 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1730 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1731 MIGRATION_STATUS_CANCELLED
);
1735 /* It is used on info migrate. We can't free it */
1736 error_report_err(error_copy(s
->error
));
1738 notifier_list_notify(&migration_state_notifiers
, s
);
1739 block_cleanup_parameters(s
);
1742 static void migrate_fd_cleanup_schedule(MigrationState
*s
)
1745 * Ref the state for bh, because it may be called when
1746 * there're already no other refs
1748 object_ref(OBJECT(s
));
1749 qemu_bh_schedule(s
->cleanup_bh
);
1752 static void migrate_fd_cleanup_bh(void *opaque
)
1754 MigrationState
*s
= opaque
;
1755 migrate_fd_cleanup(s
);
1756 object_unref(OBJECT(s
));
1759 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1761 QEMU_LOCK_GUARD(&s
->error_mutex
);
1763 s
->error
= error_copy(error
);
1767 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1769 trace_migrate_fd_error(error_get_pretty(error
));
1770 assert(s
->to_dst_file
== NULL
);
1771 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1772 MIGRATION_STATUS_FAILED
);
1773 migrate_set_error(s
, error
);
1776 static void migrate_fd_cancel(MigrationState
*s
)
1779 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1780 trace_migrate_fd_cancel();
1782 if (s
->rp_state
.from_dst_file
) {
1783 /* shutdown the rp socket, so causing the rp thread to shutdown */
1784 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1788 old_state
= s
->state
;
1789 if (!migration_is_running(old_state
)) {
1792 /* If the migration is paused, kick it out of the pause */
1793 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1794 qemu_sem_post(&s
->pause_sem
);
1796 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1797 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1800 * If we're unlucky the migration code might be stuck somewhere in a
1801 * send/write while the network has failed and is waiting to timeout;
1802 * if we've got shutdown(2) available then we can force it to quit.
1803 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1804 * called in a bh, so there is no race against this cancel.
1806 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1807 qemu_file_shutdown(f
);
1809 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1810 Error
*local_err
= NULL
;
1812 bdrv_invalidate_cache_all(&local_err
);
1814 error_report_err(local_err
);
1816 s
->block_inactive
= false;
1821 void add_migration_state_change_notifier(Notifier
*notify
)
1823 notifier_list_add(&migration_state_notifiers
, notify
);
1826 void remove_migration_state_change_notifier(Notifier
*notify
)
1828 notifier_remove(notify
);
1831 bool migration_in_setup(MigrationState
*s
)
1833 return s
->state
== MIGRATION_STATUS_SETUP
;
1836 bool migration_has_finished(MigrationState
*s
)
1838 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1841 bool migration_has_failed(MigrationState
*s
)
1843 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1844 s
->state
== MIGRATION_STATUS_FAILED
);
1847 bool migration_in_postcopy(void)
1849 MigrationState
*s
= migrate_get_current();
1852 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1853 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1854 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1861 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1863 return migration_in_postcopy() && s
->postcopy_after_devices
;
1866 bool migration_in_incoming_postcopy(void)
1868 PostcopyState ps
= postcopy_state_get();
1870 return ps
>= POSTCOPY_INCOMING_DISCARD
&& ps
< POSTCOPY_INCOMING_END
;
1873 bool migration_is_idle(void)
1875 MigrationState
*s
= current_migration
;
1882 case MIGRATION_STATUS_NONE
:
1883 case MIGRATION_STATUS_CANCELLED
:
1884 case MIGRATION_STATUS_COMPLETED
:
1885 case MIGRATION_STATUS_FAILED
:
1887 case MIGRATION_STATUS_SETUP
:
1888 case MIGRATION_STATUS_CANCELLING
:
1889 case MIGRATION_STATUS_ACTIVE
:
1890 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1891 case MIGRATION_STATUS_COLO
:
1892 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1893 case MIGRATION_STATUS_DEVICE
:
1894 case MIGRATION_STATUS_WAIT_UNPLUG
:
1896 case MIGRATION_STATUS__MAX
:
1897 g_assert_not_reached();
1903 bool migration_is_active(MigrationState
*s
)
1905 return (s
->state
== MIGRATION_STATUS_ACTIVE
||
1906 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1909 void migrate_init(MigrationState
*s
)
1912 * Reinitialise all migration state, except
1913 * parameters/capabilities that the user set, and
1917 s
->to_dst_file
= NULL
;
1918 s
->state
= MIGRATION_STATUS_NONE
;
1919 s
->rp_state
.from_dst_file
= NULL
;
1920 s
->rp_state
.error
= false;
1922 s
->pages_per_second
= 0.0;
1924 s
->expected_downtime
= 0;
1926 s
->start_postcopy
= false;
1927 s
->postcopy_after_devices
= false;
1928 s
->migration_thread_running
= false;
1929 error_free(s
->error
);
1933 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1935 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1937 s
->vm_was_running
= false;
1938 s
->iteration_initial_bytes
= 0;
1939 s
->threshold_size
= 0;
1942 static GSList
*migration_blockers
;
1944 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1946 if (only_migratable
) {
1947 error_propagate_prepend(errp
, error_copy(reason
),
1948 "disallowing migration blocker "
1949 "(--only-migratable) for: ");
1953 if (migration_is_idle()) {
1954 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1958 error_propagate_prepend(errp
, error_copy(reason
),
1959 "disallowing migration blocker "
1960 "(migration in progress) for: ");
1964 void migrate_del_blocker(Error
*reason
)
1966 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1969 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1971 Error
*local_err
= NULL
;
1972 static bool once
= true;
1975 error_setg(errp
, "The incoming migration has already been started");
1978 if (!runstate_check(RUN_STATE_INMIGRATE
)) {
1979 error_setg(errp
, "'-incoming' was not specified on the command line");
1983 qemu_start_incoming_migration(uri
, &local_err
);
1986 error_propagate(errp
, local_err
);
1993 void qmp_migrate_recover(const char *uri
, Error
**errp
)
1995 MigrationIncomingState
*mis
= migration_incoming_get_current();
1997 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1998 error_setg(errp
, "Migrate recover can only be run "
1999 "when postcopy is paused.");
2003 if (qatomic_cmpxchg(&mis
->postcopy_recover_triggered
,
2004 false, true) == true) {
2005 error_setg(errp
, "Migrate recovery is triggered already");
2010 * Note that this call will never start a real migration; it will
2011 * only re-setup the migration stream and poke existing migration
2012 * to continue using that newly established channel.
2014 qemu_start_incoming_migration(uri
, errp
);
2017 void qmp_migrate_pause(Error
**errp
)
2019 MigrationState
*ms
= migrate_get_current();
2020 MigrationIncomingState
*mis
= migration_incoming_get_current();
2023 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2024 /* Source side, during postcopy */
2025 qemu_mutex_lock(&ms
->qemu_file_lock
);
2026 ret
= qemu_file_shutdown(ms
->to_dst_file
);
2027 qemu_mutex_unlock(&ms
->qemu_file_lock
);
2029 error_setg(errp
, "Failed to pause source migration");
2034 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2035 ret
= qemu_file_shutdown(mis
->from_src_file
);
2037 error_setg(errp
, "Failed to pause destination migration");
2042 error_setg(errp
, "migrate-pause is currently only supported "
2043 "during postcopy-active state");
2046 bool migration_is_blocked(Error
**errp
)
2048 if (qemu_savevm_state_blocked(errp
)) {
2052 if (migration_blockers
) {
2053 error_propagate(errp
, error_copy(migration_blockers
->data
));
2060 /* Returns true if continue to migrate, or false if error detected */
2061 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
2062 bool resume
, Error
**errp
)
2064 Error
*local_err
= NULL
;
2067 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2068 error_setg(errp
, "Cannot resume if there is no "
2069 "paused migration");
2074 * Postcopy recovery won't work well with release-ram
2075 * capability since release-ram will drop the page buffer as
2076 * long as the page is put into the send buffer. So if there
2077 * is a network failure happened, any page buffers that have
2078 * not yet reached the destination VM but have already been
2079 * sent from the source VM will be lost forever. Let's refuse
2080 * the client from resuming such a postcopy migration.
2081 * Luckily release-ram was designed to only be used when src
2082 * and destination VMs are on the same host, so it should be
2085 if (migrate_release_ram()) {
2086 error_setg(errp
, "Postcopy recovery cannot work "
2087 "when release-ram capability is set");
2091 /* This is a resume, skip init status */
2095 if (migration_is_running(s
->state
)) {
2096 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
2100 if (runstate_check(RUN_STATE_INMIGRATE
)) {
2101 error_setg(errp
, "Guest is waiting for an incoming migration");
2105 if (migration_is_blocked(errp
)) {
2109 if (blk
|| blk_inc
) {
2110 if (migrate_use_block() || migrate_use_block_incremental()) {
2111 error_setg(errp
, "Command options are incompatible with "
2112 "current migration capabilities");
2115 migrate_set_block_enabled(true, &local_err
);
2117 error_propagate(errp
, local_err
);
2120 s
->must_remove_block_options
= true;
2124 migrate_set_block_incremental(s
, true);
2129 * set ram_counters memory to zero for a
2132 memset(&ram_counters
, 0, sizeof(ram_counters
));
2137 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
2138 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
2139 bool has_resume
, bool resume
, Error
**errp
)
2141 Error
*local_err
= NULL
;
2142 MigrationState
*s
= migrate_get_current();
2143 const char *p
= NULL
;
2145 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
2146 has_resume
&& resume
, errp
)) {
2147 /* Error detected, put into errp */
2151 if (strstart(uri
, "tcp:", &p
) ||
2152 strstart(uri
, "unix:", NULL
) ||
2153 strstart(uri
, "vsock:", NULL
)) {
2154 socket_start_outgoing_migration(s
, p
? p
: uri
, &local_err
);
2156 } else if (strstart(uri
, "rdma:", &p
)) {
2157 rdma_start_outgoing_migration(s
, p
, &local_err
);
2159 } else if (strstart(uri
, "exec:", &p
)) {
2160 exec_start_outgoing_migration(s
, p
, &local_err
);
2161 } else if (strstart(uri
, "fd:", &p
)) {
2162 fd_start_outgoing_migration(s
, p
, &local_err
);
2164 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
2165 "a valid migration protocol");
2166 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2167 MIGRATION_STATUS_FAILED
);
2168 block_cleanup_parameters(s
);
2173 migrate_fd_error(s
, local_err
);
2174 error_propagate(errp
, local_err
);
2179 void qmp_migrate_cancel(Error
**errp
)
2181 migrate_fd_cancel(migrate_get_current());
2184 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
2186 MigrationState
*s
= migrate_get_current();
2187 if (s
->state
!= state
) {
2188 error_setg(errp
, "Migration not in expected state: %s",
2189 MigrationStatus_str(s
->state
));
2192 qemu_sem_post(&s
->pause_sem
);
2195 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
2197 MigrateSetParameters p
= {
2198 .has_xbzrle_cache_size
= true,
2199 .xbzrle_cache_size
= value
,
2202 qmp_migrate_set_parameters(&p
, errp
);
2205 int64_t qmp_query_migrate_cache_size(Error
**errp
)
2207 return migrate_xbzrle_cache_size();
2210 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
2212 MigrateSetParameters p
= {
2213 .has_max_bandwidth
= true,
2214 .max_bandwidth
= value
,
2217 qmp_migrate_set_parameters(&p
, errp
);
2220 void qmp_migrate_set_downtime(double value
, Error
**errp
)
2222 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
2223 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
2225 "an integer in the range of 0 to "
2226 stringify(MAX_MIGRATE_DOWNTIME_SECONDS
)" seconds");
2230 value
*= 1000; /* Convert to milliseconds */
2232 MigrateSetParameters p
= {
2233 .has_downtime_limit
= true,
2234 .downtime_limit
= (int64_t)value
,
2237 qmp_migrate_set_parameters(&p
, errp
);
2240 bool migrate_release_ram(void)
2244 s
= migrate_get_current();
2246 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
2249 bool migrate_postcopy_ram(void)
2253 s
= migrate_get_current();
2255 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
2258 bool migrate_postcopy(void)
2260 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2263 bool migrate_auto_converge(void)
2267 s
= migrate_get_current();
2269 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
2272 bool migrate_zero_blocks(void)
2276 s
= migrate_get_current();
2278 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
2281 bool migrate_postcopy_blocktime(void)
2285 s
= migrate_get_current();
2287 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
2290 bool migrate_use_compression(void)
2294 s
= migrate_get_current();
2296 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
2299 int migrate_compress_level(void)
2303 s
= migrate_get_current();
2305 return s
->parameters
.compress_level
;
2308 int migrate_compress_threads(void)
2312 s
= migrate_get_current();
2314 return s
->parameters
.compress_threads
;
2317 int migrate_compress_wait_thread(void)
2321 s
= migrate_get_current();
2323 return s
->parameters
.compress_wait_thread
;
2326 int migrate_decompress_threads(void)
2330 s
= migrate_get_current();
2332 return s
->parameters
.decompress_threads
;
2335 bool migrate_dirty_bitmaps(void)
2339 s
= migrate_get_current();
2341 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
2344 bool migrate_ignore_shared(void)
2348 s
= migrate_get_current();
2350 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
];
2353 bool migrate_validate_uuid(void)
2357 s
= migrate_get_current();
2359 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_VALIDATE_UUID
];
2362 bool migrate_use_events(void)
2366 s
= migrate_get_current();
2368 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
2371 bool migrate_use_multifd(void)
2375 s
= migrate_get_current();
2377 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_MULTIFD
];
2380 bool migrate_pause_before_switchover(void)
2384 s
= migrate_get_current();
2386 return s
->enabled_capabilities
[
2387 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
2390 int migrate_multifd_channels(void)
2394 s
= migrate_get_current();
2396 return s
->parameters
.multifd_channels
;
2399 MultiFDCompression
migrate_multifd_compression(void)
2403 s
= migrate_get_current();
2405 return s
->parameters
.multifd_compression
;
2408 int migrate_multifd_zlib_level(void)
2412 s
= migrate_get_current();
2414 return s
->parameters
.multifd_zlib_level
;
2417 int migrate_multifd_zstd_level(void)
2421 s
= migrate_get_current();
2423 return s
->parameters
.multifd_zstd_level
;
2426 int migrate_use_xbzrle(void)
2430 s
= migrate_get_current();
2432 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
2435 int64_t migrate_xbzrle_cache_size(void)
2439 s
= migrate_get_current();
2441 return s
->parameters
.xbzrle_cache_size
;
2444 static int64_t migrate_max_postcopy_bandwidth(void)
2448 s
= migrate_get_current();
2450 return s
->parameters
.max_postcopy_bandwidth
;
2453 bool migrate_use_block(void)
2457 s
= migrate_get_current();
2459 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
2462 bool migrate_use_return_path(void)
2466 s
= migrate_get_current();
2468 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
2471 bool migrate_use_block_incremental(void)
2475 s
= migrate_get_current();
2477 return s
->parameters
.block_incremental
;
2480 /* migration thread support */
2482 * Something bad happened to the RP stream, mark an error
2483 * The caller shall print or trace something to indicate why
2485 static void mark_source_rp_bad(MigrationState
*s
)
2487 s
->rp_state
.error
= true;
2490 static struct rp_cmd_args
{
2491 ssize_t len
; /* -1 = variable */
2494 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
2495 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
2496 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
2497 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
2498 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
2499 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
2500 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
2501 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
2505 * Process a request for pages received on the return path,
2506 * We're allowed to send more than requested (e.g. to round to our page size)
2507 * and we don't need to send pages that have already been sent.
2509 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
2510 ram_addr_t start
, size_t len
)
2512 long our_host_ps
= qemu_real_host_page_size
;
2514 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
2517 * Since we currently insist on matching page sizes, just sanity check
2518 * we're being asked for whole host pages.
2520 if (start
& (our_host_ps
- 1) ||
2521 (len
& (our_host_ps
- 1))) {
2522 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2523 " len: %zd", __func__
, start
, len
);
2524 mark_source_rp_bad(ms
);
2528 if (ram_save_queue_pages(rbname
, start
, len
)) {
2529 mark_source_rp_bad(ms
);
2533 /* Return true to retry, false to quit */
2534 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
2536 trace_postcopy_pause_return_path();
2538 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
2540 trace_postcopy_pause_return_path_continued();
2545 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
2547 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
2550 error_report("%s: invalid block name '%s'", __func__
, block_name
);
2554 /* Fetch the received bitmap and refresh the dirty bitmap */
2555 return ram_dirty_bitmap_reload(s
, block
);
2558 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
2560 trace_source_return_path_thread_resume_ack(value
);
2562 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
2563 error_report("%s: illegal resume_ack value %"PRIu32
,
2568 /* Now both sides are active. */
2569 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2570 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2572 /* Notify send thread that time to continue send pages */
2573 qemu_sem_post(&s
->rp_state
.rp_sem
);
2579 * Handles messages sent on the return path towards the source VM
2582 static void *source_return_path_thread(void *opaque
)
2584 MigrationState
*ms
= opaque
;
2585 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
2586 uint16_t header_len
, header_type
;
2588 uint32_t tmp32
, sibling_error
;
2589 ram_addr_t start
= 0; /* =0 to silence warning */
2590 size_t len
= 0, expected_len
;
2593 trace_source_return_path_thread_entry();
2594 rcu_register_thread();
2597 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
2598 migration_is_setup_or_active(ms
->state
)) {
2599 trace_source_return_path_thread_loop_top();
2600 header_type
= qemu_get_be16(rp
);
2601 header_len
= qemu_get_be16(rp
);
2603 if (qemu_file_get_error(rp
)) {
2604 mark_source_rp_bad(ms
);
2608 if (header_type
>= MIG_RP_MSG_MAX
||
2609 header_type
== MIG_RP_MSG_INVALID
) {
2610 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2611 header_type
, header_len
);
2612 mark_source_rp_bad(ms
);
2616 if ((rp_cmd_args
[header_type
].len
!= -1 &&
2617 header_len
!= rp_cmd_args
[header_type
].len
) ||
2618 header_len
> sizeof(buf
)) {
2619 error_report("RP: Received '%s' message (0x%04x) with"
2620 "incorrect length %d expecting %zu",
2621 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
2622 (size_t)rp_cmd_args
[header_type
].len
);
2623 mark_source_rp_bad(ms
);
2627 /* We know we've got a valid header by this point */
2628 res
= qemu_get_buffer(rp
, buf
, header_len
);
2629 if (res
!= header_len
) {
2630 error_report("RP: Failed reading data for message 0x%04x"
2631 " read %d expected %d",
2632 header_type
, res
, header_len
);
2633 mark_source_rp_bad(ms
);
2637 /* OK, we have the message and the data */
2638 switch (header_type
) {
2639 case MIG_RP_MSG_SHUT
:
2640 sibling_error
= ldl_be_p(buf
);
2641 trace_source_return_path_thread_shut(sibling_error
);
2642 if (sibling_error
) {
2643 error_report("RP: Sibling indicated error %d", sibling_error
);
2644 mark_source_rp_bad(ms
);
2647 * We'll let the main thread deal with closing the RP
2648 * we could do a shutdown(2) on it, but we're the only user
2649 * anyway, so there's nothing gained.
2653 case MIG_RP_MSG_PONG
:
2654 tmp32
= ldl_be_p(buf
);
2655 trace_source_return_path_thread_pong(tmp32
);
2658 case MIG_RP_MSG_REQ_PAGES
:
2659 start
= ldq_be_p(buf
);
2660 len
= ldl_be_p(buf
+ 8);
2661 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
2664 case MIG_RP_MSG_REQ_PAGES_ID
:
2665 expected_len
= 12 + 1; /* header + termination */
2667 if (header_len
>= expected_len
) {
2668 start
= ldq_be_p(buf
);
2669 len
= ldl_be_p(buf
+ 8);
2670 /* Now we expect an idstr */
2671 tmp32
= buf
[12]; /* Length of the following idstr */
2672 buf
[13 + tmp32
] = '\0';
2673 expected_len
+= tmp32
;
2675 if (header_len
!= expected_len
) {
2676 error_report("RP: Req_Page_id with length %d expecting %zd",
2677 header_len
, expected_len
);
2678 mark_source_rp_bad(ms
);
2681 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
2684 case MIG_RP_MSG_RECV_BITMAP
:
2685 if (header_len
< 1) {
2686 error_report("%s: missing block name", __func__
);
2687 mark_source_rp_bad(ms
);
2690 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2691 buf
[buf
[0] + 1] = '\0';
2692 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
2693 mark_source_rp_bad(ms
);
2698 case MIG_RP_MSG_RESUME_ACK
:
2699 tmp32
= ldl_be_p(buf
);
2700 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
2701 mark_source_rp_bad(ms
);
2712 res
= qemu_file_get_error(rp
);
2714 if (res
== -EIO
&& migration_in_postcopy()) {
2716 * Maybe there is something we can do: it looks like a
2717 * network down issue, and we pause for a recovery.
2719 if (postcopy_pause_return_path_thread(ms
)) {
2720 /* Reload rp, reset the rest */
2721 if (rp
!= ms
->rp_state
.from_dst_file
) {
2723 rp
= ms
->rp_state
.from_dst_file
;
2725 ms
->rp_state
.error
= false;
2730 trace_source_return_path_thread_bad_end();
2731 mark_source_rp_bad(ms
);
2734 trace_source_return_path_thread_end();
2735 ms
->rp_state
.from_dst_file
= NULL
;
2737 rcu_unregister_thread();
2741 static int open_return_path_on_source(MigrationState
*ms
,
2745 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
2746 if (!ms
->rp_state
.from_dst_file
) {
2750 trace_open_return_path_on_source();
2752 if (!create_thread
) {
2757 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
2758 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
2760 trace_open_return_path_on_source_continue();
2765 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2766 static int await_return_path_close_on_source(MigrationState
*ms
)
2769 * If this is a normal exit then the destination will send a SHUT and the
2770 * rp_thread will exit, however if there's an error we need to cause
2773 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
2775 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2776 * waiting for the destination.
2778 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
2779 mark_source_rp_bad(ms
);
2781 trace_await_return_path_close_on_source_joining();
2782 qemu_thread_join(&ms
->rp_state
.rp_thread
);
2783 trace_await_return_path_close_on_source_close();
2784 return ms
->rp_state
.error
;
2788 * Switch from normal iteration to postcopy
2789 * Returns non-0 on error
2791 static int postcopy_start(MigrationState
*ms
)
2794 QIOChannelBuffer
*bioc
;
2796 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2797 int64_t bandwidth
= migrate_max_postcopy_bandwidth();
2798 bool restart_block
= false;
2799 int cur_state
= MIGRATION_STATUS_ACTIVE
;
2800 if (!migrate_pause_before_switchover()) {
2801 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
2802 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2805 trace_postcopy_start();
2806 qemu_mutex_lock_iothread();
2807 trace_postcopy_start_set_run();
2809 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2810 global_state_store();
2811 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2816 ret
= migration_maybe_pause(ms
, &cur_state
,
2817 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2822 ret
= bdrv_inactivate_all();
2826 restart_block
= true;
2829 * Cause any non-postcopiable, but iterative devices to
2830 * send out their final data.
2832 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2835 * in Finish migrate and with the io-lock held everything should
2836 * be quiet, but we've potentially still got dirty pages and we
2837 * need to tell the destination to throw any pages it's already received
2840 if (migrate_postcopy_ram()) {
2841 if (ram_postcopy_send_discard_bitmap(ms
)) {
2842 error_report("postcopy send discard bitmap failed");
2848 * send rest of state - note things that are doing postcopy
2849 * will notice we're in POSTCOPY_ACTIVE and not actually
2850 * wrap their state up here
2852 /* 0 max-postcopy-bandwidth means unlimited */
2854 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
2856 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
2858 if (migrate_postcopy_ram()) {
2859 /* Ping just for debugging, helps line traces up */
2860 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
2864 * While loading the device state we may trigger page transfer
2865 * requests and the fd must be free to process those, and thus
2866 * the destination must read the whole device state off the fd before
2867 * it starts processing it. Unfortunately the ad-hoc migration format
2868 * doesn't allow the destination to know the size to read without fully
2869 * parsing it through each devices load-state code (especially the open
2870 * coded devices that use get/put).
2871 * So we wrap the device state up in a package with a length at the start;
2872 * to do this we use a qemu_buf to hold the whole of the device state.
2874 bioc
= qio_channel_buffer_new(4096);
2875 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
2876 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
2877 object_unref(OBJECT(bioc
));
2880 * Make sure the receiver can get incoming pages before we send the rest
2883 qemu_savevm_send_postcopy_listen(fb
);
2885 qemu_savevm_state_complete_precopy(fb
, false, false);
2886 if (migrate_postcopy_ram()) {
2887 qemu_savevm_send_ping(fb
, 3);
2890 qemu_savevm_send_postcopy_run(fb
);
2892 /* <><> end of stuff going into the package */
2894 /* Last point of recovery; as soon as we send the package the destination
2895 * can open devices and potentially start running.
2896 * Lets just check again we've not got any errors.
2898 ret
= qemu_file_get_error(ms
->to_dst_file
);
2900 error_report("postcopy_start: Migration stream errored (pre package)");
2904 restart_block
= false;
2906 /* Now send that blob */
2907 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2912 /* Send a notify to give a chance for anything that needs to happen
2913 * at the transition to postcopy and after the device state; in particular
2914 * spice needs to trigger a transition now
2916 ms
->postcopy_after_devices
= true;
2917 notifier_list_notify(&migration_state_notifiers
, ms
);
2919 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2921 qemu_mutex_unlock_iothread();
2923 if (migrate_postcopy_ram()) {
2925 * Although this ping is just for debug, it could potentially be
2926 * used for getting a better measurement of downtime at the source.
2928 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2931 if (migrate_release_ram()) {
2932 ram_postcopy_migrated_memory_release(ms
);
2935 ret
= qemu_file_get_error(ms
->to_dst_file
);
2937 error_report("postcopy_start: Migration stream errored");
2938 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2939 MIGRATION_STATUS_FAILED
);
2947 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2948 MIGRATION_STATUS_FAILED
);
2949 if (restart_block
) {
2950 /* A failure happened early enough that we know the destination hasn't
2951 * accessed block devices, so we're safe to recover.
2953 Error
*local_err
= NULL
;
2955 bdrv_invalidate_cache_all(&local_err
);
2957 error_report_err(local_err
);
2960 qemu_mutex_unlock_iothread();
2965 * migration_maybe_pause: Pause if required to by
2966 * migrate_pause_before_switchover called with the iothread locked
2967 * Returns: 0 on success
2969 static int migration_maybe_pause(MigrationState
*s
,
2970 int *current_active_state
,
2973 if (!migrate_pause_before_switchover()) {
2977 /* Since leaving this state is not atomic with posting the semaphore
2978 * it's possible that someone could have issued multiple migrate_continue
2979 * and the semaphore is incorrectly positive at this point;
2980 * the docs say it's undefined to reinit a semaphore that's already
2981 * init'd, so use timedwait to eat up any existing posts.
2983 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2984 /* This block intentionally left blank */
2988 * If the migration is cancelled when it is in the completion phase,
2989 * the migration state is set to MIGRATION_STATUS_CANCELLING.
2990 * So we don't need to wait a semaphore, otherwise we would always
2991 * wait for the 'pause_sem' semaphore.
2993 if (s
->state
!= MIGRATION_STATUS_CANCELLING
) {
2994 qemu_mutex_unlock_iothread();
2995 migrate_set_state(&s
->state
, *current_active_state
,
2996 MIGRATION_STATUS_PRE_SWITCHOVER
);
2997 qemu_sem_wait(&s
->pause_sem
);
2998 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
3000 *current_active_state
= new_state
;
3001 qemu_mutex_lock_iothread();
3004 return s
->state
== new_state
? 0 : -EINVAL
;
3008 * migration_completion: Used by migration_thread when there's not much left.
3009 * The caller 'breaks' the loop when this returns.
3011 * @s: Current migration state
3013 static void migration_completion(MigrationState
*s
)
3016 int current_active_state
= s
->state
;
3018 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
3019 qemu_mutex_lock_iothread();
3020 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3021 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3022 s
->vm_was_running
= runstate_is_running();
3023 ret
= global_state_store();
3026 bool inactivate
= !migrate_colo_enabled();
3027 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
3029 ret
= migration_maybe_pause(s
, ¤t_active_state
,
3030 MIGRATION_STATUS_DEVICE
);
3033 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
3034 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
3037 if (inactivate
&& ret
>= 0) {
3038 s
->block_inactive
= true;
3041 qemu_mutex_unlock_iothread();
3046 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3047 trace_migration_completion_postcopy_end();
3049 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
3050 trace_migration_completion_postcopy_end_after_complete();
3051 } else if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
3056 * If rp was opened we must clean up the thread before
3057 * cleaning everything else up (since if there are no failures
3058 * it will wait for the destination to send it's status in
3061 if (s
->rp_state
.from_dst_file
) {
3063 trace_migration_return_path_end_before();
3064 rp_error
= await_return_path_close_on_source(s
);
3065 trace_migration_return_path_end_after(rp_error
);
3067 goto fail_invalidate
;
3071 if (qemu_file_get_error(s
->to_dst_file
)) {
3072 trace_migration_completion_file_err();
3073 goto fail_invalidate
;
3076 if (!migrate_colo_enabled()) {
3077 migrate_set_state(&s
->state
, current_active_state
,
3078 MIGRATION_STATUS_COMPLETED
);
3084 /* If not doing postcopy, vm_start() will be called: let's regain
3085 * control on images.
3087 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
3088 s
->state
== MIGRATION_STATUS_DEVICE
) {
3089 Error
*local_err
= NULL
;
3091 qemu_mutex_lock_iothread();
3092 bdrv_invalidate_cache_all(&local_err
);
3094 error_report_err(local_err
);
3096 s
->block_inactive
= false;
3098 qemu_mutex_unlock_iothread();
3102 migrate_set_state(&s
->state
, current_active_state
,
3103 MIGRATION_STATUS_FAILED
);
3106 bool migrate_colo_enabled(void)
3108 MigrationState
*s
= migrate_get_current();
3109 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
3112 typedef enum MigThrError
{
3113 /* No error detected */
3114 MIG_THR_ERR_NONE
= 0,
3115 /* Detected error, but resumed successfully */
3116 MIG_THR_ERR_RECOVERED
= 1,
3117 /* Detected fatal error, need to exit */
3118 MIG_THR_ERR_FATAL
= 2,
3121 static int postcopy_resume_handshake(MigrationState
*s
)
3123 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
3125 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3126 qemu_sem_wait(&s
->rp_state
.rp_sem
);
3129 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3136 /* Return zero if success, or <0 for error */
3137 static int postcopy_do_resume(MigrationState
*s
)
3142 * Call all the resume_prepare() hooks, so that modules can be
3143 * ready for the migration resume.
3145 ret
= qemu_savevm_state_resume_prepare(s
);
3147 error_report("%s: resume_prepare() failure detected: %d",
3153 * Last handshake with destination on the resume (destination will
3154 * switch to postcopy-active afterwards)
3156 ret
= postcopy_resume_handshake(s
);
3158 error_report("%s: handshake failed: %d", __func__
, ret
);
3166 * We don't return until we are in a safe state to continue current
3167 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
3168 * MIG_THR_ERR_FATAL if unrecovery failure happened.
3170 static MigThrError
postcopy_pause(MigrationState
*s
)
3172 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3177 /* Current channel is possibly broken. Release it. */
3178 assert(s
->to_dst_file
);
3179 qemu_mutex_lock(&s
->qemu_file_lock
);
3180 file
= s
->to_dst_file
;
3181 s
->to_dst_file
= NULL
;
3182 qemu_mutex_unlock(&s
->qemu_file_lock
);
3184 qemu_file_shutdown(file
);
3187 migrate_set_state(&s
->state
, s
->state
,
3188 MIGRATION_STATUS_POSTCOPY_PAUSED
);
3190 error_report("Detected IO failure for postcopy. "
3191 "Migration paused.");
3194 * We wait until things fixed up. Then someone will setup the
3195 * status back for us.
3197 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
3198 qemu_sem_wait(&s
->postcopy_pause_sem
);
3201 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3202 /* Woken up by a recover procedure. Give it a shot */
3205 * Firstly, let's wake up the return path now, with a new
3206 * return path channel.
3208 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
3210 /* Do the resume logic */
3211 if (postcopy_do_resume(s
) == 0) {
3212 /* Let's continue! */
3213 trace_postcopy_pause_continued();
3214 return MIG_THR_ERR_RECOVERED
;
3217 * Something wrong happened during the recovery, let's
3218 * pause again. Pause is always better than throwing
3224 /* This is not right... Time to quit. */
3225 return MIG_THR_ERR_FATAL
;
3230 static MigThrError
migration_detect_error(MigrationState
*s
)
3233 int state
= s
->state
;
3234 Error
*local_error
= NULL
;
3236 if (state
== MIGRATION_STATUS_CANCELLING
||
3237 state
== MIGRATION_STATUS_CANCELLED
) {
3238 /* End the migration, but don't set the state to failed */
3239 return MIG_THR_ERR_FATAL
;
3242 /* Try to detect any file errors */
3243 ret
= qemu_file_get_error_obj(s
->to_dst_file
, &local_error
);
3245 /* Everything is fine */
3246 assert(!local_error
);
3247 return MIG_THR_ERR_NONE
;
3251 migrate_set_error(s
, local_error
);
3252 error_free(local_error
);
3255 if (state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
== -EIO
) {
3257 * For postcopy, we allow the network to be down for a
3258 * while. After that, it can be continued by a
3261 return postcopy_pause(s
);
3264 * For precopy (or postcopy with error outside IO), we fail
3267 migrate_set_state(&s
->state
, state
, MIGRATION_STATUS_FAILED
);
3268 trace_migration_thread_file_err();
3270 /* Time to stop the migration, now. */
3271 return MIG_THR_ERR_FATAL
;
3275 /* How many bytes have we transferred since the beginning of the migration */
3276 static uint64_t migration_total_bytes(MigrationState
*s
)
3278 return qemu_ftell(s
->to_dst_file
) + ram_counters
.multifd_bytes
;
3281 static void migration_calculate_complete(MigrationState
*s
)
3283 uint64_t bytes
= migration_total_bytes(s
);
3284 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3285 int64_t transfer_time
;
3287 s
->total_time
= end_time
- s
->start_time
;
3290 * It's still not set, so we are precopy migration. For
3291 * postcopy, downtime is calculated during postcopy_start().
3293 s
->downtime
= end_time
- s
->downtime_start
;
3296 transfer_time
= s
->total_time
- s
->setup_time
;
3297 if (transfer_time
) {
3298 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
3302 static void update_iteration_initial_status(MigrationState
*s
)
3305 * Update these three fields at the same time to avoid mismatch info lead
3306 * wrong speed calculation.
3308 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3309 s
->iteration_initial_bytes
= migration_total_bytes(s
);
3310 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
3313 static void migration_update_counters(MigrationState
*s
,
3314 int64_t current_time
)
3316 uint64_t transferred
, transferred_pages
, time_spent
;
3317 uint64_t current_bytes
; /* bytes transferred since the beginning */
3320 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
3324 current_bytes
= migration_total_bytes(s
);
3325 transferred
= current_bytes
- s
->iteration_initial_bytes
;
3326 time_spent
= current_time
- s
->iteration_start_time
;
3327 bandwidth
= (double)transferred
/ time_spent
;
3328 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
3330 s
->mbps
= (((double) transferred
* 8.0) /
3331 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
3333 transferred_pages
= ram_get_total_transferred_pages() -
3334 s
->iteration_initial_pages
;
3335 s
->pages_per_second
= (double) transferred_pages
/
3336 (((double) time_spent
/ 1000.0));
3339 * if we haven't sent anything, we don't want to
3340 * recalculate. 10000 is a small enough number for our purposes
3342 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
3343 s
->expected_downtime
= ram_counters
.remaining
/ bandwidth
;
3346 qemu_file_reset_rate_limit(s
->to_dst_file
);
3348 update_iteration_initial_status(s
);
3350 trace_migrate_transferred(transferred
, time_spent
,
3351 bandwidth
, s
->threshold_size
);
3354 /* Migration thread iteration status */
3356 MIG_ITERATE_RESUME
, /* Resume current iteration */
3357 MIG_ITERATE_SKIP
, /* Skip current iteration */
3358 MIG_ITERATE_BREAK
, /* Break the loop */
3362 * Return true if continue to the next iteration directly, false
3365 static MigIterateState
migration_iteration_run(MigrationState
*s
)
3367 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
3368 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
3370 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
3371 &pend_compat
, &pend_post
);
3372 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
3374 trace_migrate_pending(pending_size
, s
->threshold_size
,
3375 pend_pre
, pend_compat
, pend_post
);
3377 if (pending_size
&& pending_size
>= s
->threshold_size
) {
3378 /* Still a significant amount to transfer */
3379 if (!in_postcopy
&& pend_pre
<= s
->threshold_size
&&
3380 qatomic_read(&s
->start_postcopy
)) {
3381 if (postcopy_start(s
)) {
3382 error_report("%s: postcopy failed to start", __func__
);
3384 return MIG_ITERATE_SKIP
;
3386 /* Just another iteration step */
3387 qemu_savevm_state_iterate(s
->to_dst_file
, in_postcopy
);
3389 trace_migration_thread_low_pending(pending_size
);
3390 migration_completion(s
);
3391 return MIG_ITERATE_BREAK
;
3394 return MIG_ITERATE_RESUME
;
3397 static void migration_iteration_finish(MigrationState
*s
)
3399 /* If we enabled cpu throttling for auto-converge, turn it off. */
3400 cpu_throttle_stop();
3402 qemu_mutex_lock_iothread();
3404 case MIGRATION_STATUS_COMPLETED
:
3405 migration_calculate_complete(s
);
3406 runstate_set(RUN_STATE_POSTMIGRATE
);
3409 case MIGRATION_STATUS_ACTIVE
:
3411 * We should really assert here, but since it's during
3412 * migration, let's try to reduce the usage of assertions.
3414 if (!migrate_colo_enabled()) {
3415 error_report("%s: critical error: calling COLO code without "
3416 "COLO enabled", __func__
);
3418 migrate_start_colo_process(s
);
3420 * Fixme: we will run VM in COLO no matter its old running state.
3421 * After exited COLO, we will keep running.
3423 s
->vm_was_running
= true;
3425 case MIGRATION_STATUS_FAILED
:
3426 case MIGRATION_STATUS_CANCELLED
:
3427 case MIGRATION_STATUS_CANCELLING
:
3428 if (s
->vm_was_running
) {
3431 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
3432 runstate_set(RUN_STATE_POSTMIGRATE
);
3438 /* Should not reach here, but if so, forgive the VM. */
3439 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3442 migrate_fd_cleanup_schedule(s
);
3443 qemu_mutex_unlock_iothread();
3446 void migration_make_urgent_request(void)
3448 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
3451 void migration_consume_urgent_request(void)
3453 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
3456 /* Returns true if the rate limiting was broken by an urgent request */
3457 bool migration_rate_limit(void)
3459 int64_t now
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3460 MigrationState
*s
= migrate_get_current();
3462 bool urgent
= false;
3463 migration_update_counters(s
, now
);
3464 if (qemu_file_rate_limit(s
->to_dst_file
)) {
3466 if (qemu_file_get_error(s
->to_dst_file
)) {
3470 * Wait for a delay to do rate limiting OR
3471 * something urgent to post the semaphore.
3473 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- now
;
3474 trace_migration_rate_limit_pre(ms
);
3475 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
3477 * We were woken by one or more urgent things but
3478 * the timedwait will have consumed one of them.
3479 * The service routine for the urgent wake will dec
3480 * the semaphore itself for each item it consumes,
3481 * so add this one we just eat back.
3483 qemu_sem_post(&s
->rate_limit_sem
);
3486 trace_migration_rate_limit_post(urgent
);
3492 * Master migration thread on the source VM.
3493 * It drives the migration and pumps the data down the outgoing channel.
3495 static void *migration_thread(void *opaque
)
3497 MigrationState
*s
= opaque
;
3498 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3499 MigThrError thr_error
;
3500 bool urgent
= false;
3502 rcu_register_thread();
3504 object_ref(OBJECT(s
));
3505 update_iteration_initial_status(s
);
3507 qemu_savevm_state_header(s
->to_dst_file
);
3510 * If we opened the return path, we need to make sure dst has it
3513 if (s
->rp_state
.from_dst_file
) {
3514 /* Now tell the dest that it should open its end so it can reply */
3515 qemu_savevm_send_open_return_path(s
->to_dst_file
);
3517 /* And do a ping that will make stuff easier to debug */
3518 qemu_savevm_send_ping(s
->to_dst_file
, 1);
3521 if (migrate_postcopy()) {
3523 * Tell the destination that we *might* want to do postcopy later;
3524 * if the other end can't do postcopy it should fail now, nice and
3527 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
3530 if (migrate_colo_enabled()) {
3531 /* Notify migration destination that we enable COLO */
3532 qemu_savevm_send_colo_enable(s
->to_dst_file
);
3535 qemu_savevm_state_setup(s
->to_dst_file
);
3537 if (qemu_savevm_state_guest_unplug_pending()) {
3538 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3539 MIGRATION_STATUS_WAIT_UNPLUG
);
3541 while (s
->state
== MIGRATION_STATUS_WAIT_UNPLUG
&&
3542 qemu_savevm_state_guest_unplug_pending()) {
3543 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
3546 migrate_set_state(&s
->state
, MIGRATION_STATUS_WAIT_UNPLUG
,
3547 MIGRATION_STATUS_ACTIVE
);
3550 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3551 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3552 MIGRATION_STATUS_ACTIVE
);
3554 trace_migration_thread_setup_complete();
3556 while (migration_is_active(s
)) {
3557 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
3558 MigIterateState iter_state
= migration_iteration_run(s
);
3559 if (iter_state
== MIG_ITERATE_SKIP
) {
3561 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3567 * Try to detect any kind of failures, and see whether we
3568 * should stop the migration now.
3570 thr_error
= migration_detect_error(s
);
3571 if (thr_error
== MIG_THR_ERR_FATAL
) {
3572 /* Stop migration */
3574 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
3576 * Just recovered from a e.g. network failure, reset all
3577 * the local variables. This is important to avoid
3578 * breaking transferred_bytes and bandwidth calculation
3580 update_iteration_initial_status(s
);
3583 urgent
= migration_rate_limit();
3586 trace_migration_thread_after_loop();
3587 migration_iteration_finish(s
);
3588 object_unref(OBJECT(s
));
3589 rcu_unregister_thread();
3593 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
3595 Error
*local_err
= NULL
;
3597 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
3599 s
->expected_downtime
= s
->parameters
.downtime_limit
;
3601 assert(s
->cleanup_bh
);
3603 assert(!s
->cleanup_bh
);
3604 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup_bh
, s
);
3607 migrate_fd_error(s
, error_in
);
3608 migrate_fd_cleanup(s
);
3613 /* This is a resumed migration */
3614 rate_limit
= s
->parameters
.max_postcopy_bandwidth
/
3617 /* This is a fresh new migration */
3618 rate_limit
= s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
;
3620 /* Notify before starting migration thread */
3621 notifier_list_notify(&migration_state_notifiers
, s
);
3624 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
3625 qemu_file_set_blocking(s
->to_dst_file
, true);
3628 * Open the return path. For postcopy, it is used exclusively. For
3629 * precopy, only if user specified "return-path" capability would
3630 * QEMU uses the return path.
3632 if (migrate_postcopy_ram() || migrate_use_return_path()) {
3633 if (open_return_path_on_source(s
, !resume
)) {
3634 error_report("Unable to open return-path for postcopy");
3635 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
3636 migrate_fd_cleanup(s
);
3642 /* Wakeup the main migration thread to do the recovery */
3643 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
3644 MIGRATION_STATUS_POSTCOPY_RECOVER
);
3645 qemu_sem_post(&s
->postcopy_pause_sem
);
3649 if (multifd_save_setup(&local_err
) != 0) {
3650 error_report_err(local_err
);
3651 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3652 MIGRATION_STATUS_FAILED
);
3653 migrate_fd_cleanup(s
);
3656 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
3657 QEMU_THREAD_JOINABLE
);
3658 s
->migration_thread_running
= true;
3661 void migration_global_dump(Monitor
*mon
)
3663 MigrationState
*ms
= migrate_get_current();
3665 monitor_printf(mon
, "globals:\n");
3666 monitor_printf(mon
, "store-global-state: %s\n",
3667 ms
->store_global_state
? "on" : "off");
3668 monitor_printf(mon
, "only-migratable: %s\n",
3669 only_migratable
? "on" : "off");
3670 monitor_printf(mon
, "send-configuration: %s\n",
3671 ms
->send_configuration
? "on" : "off");
3672 monitor_printf(mon
, "send-section-footer: %s\n",
3673 ms
->send_section_footer
? "on" : "off");
3674 monitor_printf(mon
, "decompress-error-check: %s\n",
3675 ms
->decompress_error_check
? "on" : "off");
3676 monitor_printf(mon
, "clear-bitmap-shift: %u\n",
3677 ms
->clear_bitmap_shift
);
3680 #define DEFINE_PROP_MIG_CAP(name, x) \
3681 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3683 static Property migration_properties
[] = {
3684 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
3685 store_global_state
, true),
3686 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
3687 send_configuration
, true),
3688 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
3689 send_section_footer
, true),
3690 DEFINE_PROP_BOOL("decompress-error-check", MigrationState
,
3691 decompress_error_check
, true),
3692 DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState
,
3693 clear_bitmap_shift
, CLEAR_BITMAP_SHIFT_DEFAULT
),
3695 /* Migration parameters */
3696 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
3697 parameters
.compress_level
,
3698 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
3699 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
3700 parameters
.compress_threads
,
3701 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
3702 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState
,
3703 parameters
.compress_wait_thread
, true),
3704 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
3705 parameters
.decompress_threads
,
3706 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
3707 DEFINE_PROP_UINT8("x-throttle-trigger-threshold", MigrationState
,
3708 parameters
.throttle_trigger_threshold
,
3709 DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD
),
3710 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
3711 parameters
.cpu_throttle_initial
,
3712 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
3713 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
3714 parameters
.cpu_throttle_increment
,
3715 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
3716 DEFINE_PROP_BOOL("x-cpu-throttle-tailslow", MigrationState
,
3717 parameters
.cpu_throttle_tailslow
, false),
3718 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
3719 parameters
.max_bandwidth
, MAX_THROTTLE
),
3720 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
3721 parameters
.downtime_limit
,
3722 DEFAULT_MIGRATE_SET_DOWNTIME
),
3723 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
3724 parameters
.x_checkpoint_delay
,
3725 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
3726 DEFINE_PROP_UINT8("multifd-channels", MigrationState
,
3727 parameters
.multifd_channels
,
3728 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
3729 DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState
,
3730 parameters
.multifd_compression
,
3731 DEFAULT_MIGRATE_MULTIFD_COMPRESSION
),
3732 DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState
,
3733 parameters
.multifd_zlib_level
,
3734 DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL
),
3735 DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState
,
3736 parameters
.multifd_zstd_level
,
3737 DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL
),
3738 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
3739 parameters
.xbzrle_cache_size
,
3740 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
3741 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState
,
3742 parameters
.max_postcopy_bandwidth
,
3743 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH
),
3744 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState
,
3745 parameters
.max_cpu_throttle
,
3746 DEFAULT_MIGRATE_MAX_CPU_THROTTLE
),
3747 DEFINE_PROP_SIZE("announce-initial", MigrationState
,
3748 parameters
.announce_initial
,
3749 DEFAULT_MIGRATE_ANNOUNCE_INITIAL
),
3750 DEFINE_PROP_SIZE("announce-max", MigrationState
,
3751 parameters
.announce_max
,
3752 DEFAULT_MIGRATE_ANNOUNCE_MAX
),
3753 DEFINE_PROP_SIZE("announce-rounds", MigrationState
,
3754 parameters
.announce_rounds
,
3755 DEFAULT_MIGRATE_ANNOUNCE_ROUNDS
),
3756 DEFINE_PROP_SIZE("announce-step", MigrationState
,
3757 parameters
.announce_step
,
3758 DEFAULT_MIGRATE_ANNOUNCE_STEP
),
3760 /* Migration capabilities */
3761 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
3762 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
3763 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
3764 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
3765 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
3766 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
3767 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
3768 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
3769 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
3770 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
3771 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
3772 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD
),
3774 DEFINE_PROP_END_OF_LIST(),
3777 static void migration_class_init(ObjectClass
*klass
, void *data
)
3779 DeviceClass
*dc
= DEVICE_CLASS(klass
);
3781 dc
->user_creatable
= false;
3782 device_class_set_props(dc
, migration_properties
);
3785 static void migration_instance_finalize(Object
*obj
)
3787 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3788 MigrationParameters
*params
= &ms
->parameters
;
3790 qemu_mutex_destroy(&ms
->error_mutex
);
3791 qemu_mutex_destroy(&ms
->qemu_file_lock
);
3792 g_free(params
->tls_hostname
);
3793 g_free(params
->tls_creds
);
3794 qemu_sem_destroy(&ms
->wait_unplug_sem
);
3795 qemu_sem_destroy(&ms
->rate_limit_sem
);
3796 qemu_sem_destroy(&ms
->pause_sem
);
3797 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
3798 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
3799 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
3800 error_free(ms
->error
);
3803 static void migration_instance_init(Object
*obj
)
3805 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3806 MigrationParameters
*params
= &ms
->parameters
;
3808 ms
->state
= MIGRATION_STATUS_NONE
;
3810 ms
->pages_per_second
= -1;
3811 qemu_sem_init(&ms
->pause_sem
, 0);
3812 qemu_mutex_init(&ms
->error_mutex
);
3814 params
->tls_hostname
= g_strdup("");
3815 params
->tls_creds
= g_strdup("");
3817 /* Set has_* up only for parameter checks */
3818 params
->has_compress_level
= true;
3819 params
->has_compress_threads
= true;
3820 params
->has_decompress_threads
= true;
3821 params
->has_throttle_trigger_threshold
= true;
3822 params
->has_cpu_throttle_initial
= true;
3823 params
->has_cpu_throttle_increment
= true;
3824 params
->has_cpu_throttle_tailslow
= true;
3825 params
->has_max_bandwidth
= true;
3826 params
->has_downtime_limit
= true;
3827 params
->has_x_checkpoint_delay
= true;
3828 params
->has_block_incremental
= true;
3829 params
->has_multifd_channels
= true;
3830 params
->has_multifd_compression
= true;
3831 params
->has_multifd_zlib_level
= true;
3832 params
->has_multifd_zstd_level
= true;
3833 params
->has_xbzrle_cache_size
= true;
3834 params
->has_max_postcopy_bandwidth
= true;
3835 params
->has_max_cpu_throttle
= true;
3836 params
->has_announce_initial
= true;
3837 params
->has_announce_max
= true;
3838 params
->has_announce_rounds
= true;
3839 params
->has_announce_step
= true;
3841 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
3842 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
3843 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
3844 qemu_sem_init(&ms
->rate_limit_sem
, 0);
3845 qemu_sem_init(&ms
->wait_unplug_sem
, 0);
3846 qemu_mutex_init(&ms
->qemu_file_lock
);
3850 * Return true if check pass, false otherwise. Error will be put
3851 * inside errp if provided.
3853 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
3855 MigrationCapabilityStatusList
*head
= NULL
;
3856 /* Assuming all off */
3857 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
3860 if (!migrate_params_check(&ms
->parameters
, errp
)) {
3864 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
3865 if (ms
->enabled_capabilities
[i
]) {
3866 head
= migrate_cap_add(head
, i
, true);
3870 ret
= migrate_caps_check(cap_list
, head
, errp
);
3872 /* It works with head == NULL */
3873 qapi_free_MigrationCapabilityStatusList(head
);
3878 static const TypeInfo migration_type
= {
3879 .name
= TYPE_MIGRATION
,
3881 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3882 * not created using qdev_new(), it is not attached to the qdev
3883 * device tree, and it is never realized.
3885 * TODO: Make this TYPE_OBJECT once QOM provides something like
3886 * TYPE_DEVICE's "-global" properties.
3888 .parent
= TYPE_DEVICE
,
3889 .class_init
= migration_class_init
,
3890 .class_size
= sizeof(MigrationClass
),
3891 .instance_size
= sizeof(MigrationState
),
3892 .instance_init
= migration_instance_init
,
3893 .instance_finalize
= migration_instance_finalize
,
3896 static void register_migration_types(void)
3898 type_register_static(&migration_type
);
3901 type_init(register_migration_types
);