4 * Copyright IBM, Corp. 2008
7 * Anthony Liguori <aliguori@us.ibm.com>
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
12 * Contributions after 2012-01-13 are licensed under the terms of the
13 * GNU GPL, version 2 or (at your option) any later version.
16 #include "qemu/osdep.h"
17 #include "qemu/cutils.h"
18 #include "qemu/error-report.h"
19 #include "qemu/main-loop.h"
20 #include "migration/blocker.h"
24 #include "sysemu/runstate.h"
25 #include "sysemu/sysemu.h"
26 #include "sysemu/cpu-throttle.h"
29 #include "migration/global_state.h"
30 #include "migration/misc.h"
31 #include "migration.h"
33 #include "qemu-file-channel.h"
34 #include "qemu-file.h"
35 #include "migration/vmstate.h"
36 #include "block/block.h"
37 #include "qapi/error.h"
38 #include "qapi/clone-visitor.h"
39 #include "qapi/qapi-visit-migration.h"
40 #include "qapi/qapi-visit-sockets.h"
41 #include "qapi/qapi-commands-migration.h"
42 #include "qapi/qapi-events-migration.h"
43 #include "qapi/qmp/qerror.h"
44 #include "qapi/qmp/qnull.h"
47 #include "postcopy-ram.h"
48 #include "qemu/thread.h"
50 #include "exec/target_page.h"
51 #include "io/channel-buffer.h"
52 #include "migration/colo.h"
53 #include "hw/boards.h"
54 #include "hw/qdev-properties.h"
55 #include "hw/qdev-properties-system.h"
56 #include "monitor/monitor.h"
57 #include "net/announce.h"
58 #include "qemu/queue.h"
60 #include "qemu/yank.h"
61 #include "sysemu/cpus.h"
62 #include "yank_functions.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 /* Migration capabilities set */
136 struct MigrateCapsSet
{
137 int size
; /* Capability set size */
138 MigrationCapability caps
[]; /* Variadic array of capabilities */
140 typedef struct MigrateCapsSet MigrateCapsSet
;
142 /* Define and initialize MigrateCapsSet */
143 #define INITIALIZE_MIGRATE_CAPS_SET(_name, ...) \
144 MigrateCapsSet _name = { \
145 .size = sizeof((int []) { __VA_ARGS__ }) / sizeof(int), \
146 .caps = { __VA_ARGS__ } \
149 /* Background-snapshot compatibility check list */
151 INITIALIZE_MIGRATE_CAPS_SET(check_caps_background_snapshot
,
152 MIGRATION_CAPABILITY_POSTCOPY_RAM
,
153 MIGRATION_CAPABILITY_DIRTY_BITMAPS
,
154 MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
,
155 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE
,
156 MIGRATION_CAPABILITY_RETURN_PATH
,
157 MIGRATION_CAPABILITY_MULTIFD
,
158 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
,
159 MIGRATION_CAPABILITY_AUTO_CONVERGE
,
160 MIGRATION_CAPABILITY_RELEASE_RAM
,
161 MIGRATION_CAPABILITY_RDMA_PIN_ALL
,
162 MIGRATION_CAPABILITY_COMPRESS
,
163 MIGRATION_CAPABILITY_XBZRLE
,
164 MIGRATION_CAPABILITY_X_COLO
,
165 MIGRATION_CAPABILITY_VALIDATE_UUID
);
167 /* When we add fault tolerance, we could have several
168 migrations at once. For now we don't need to add
169 dynamic creation of migration */
171 static MigrationState
*current_migration
;
172 static MigrationIncomingState
*current_incoming
;
174 static GSList
*migration_blockers
;
176 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
177 static int migration_maybe_pause(MigrationState
*s
,
178 int *current_active_state
,
180 static void migrate_fd_cancel(MigrationState
*s
);
182 static gint
page_request_addr_cmp(gconstpointer ap
, gconstpointer bp
)
184 uintptr_t a
= (uintptr_t) ap
, b
= (uintptr_t) bp
;
186 return (a
> b
) - (a
< b
);
189 void migration_object_init(void)
191 /* This can only be called once. */
192 assert(!current_migration
);
193 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
196 * Init the migrate incoming object as well no matter whether
197 * we'll use it or not.
199 assert(!current_incoming
);
200 current_incoming
= g_new0(MigrationIncomingState
, 1);
201 current_incoming
->state
= MIGRATION_STATUS_NONE
;
202 current_incoming
->postcopy_remote_fds
=
203 g_array_new(FALSE
, TRUE
, sizeof(struct PostCopyFD
));
204 qemu_mutex_init(¤t_incoming
->rp_mutex
);
205 qemu_event_init(¤t_incoming
->main_thread_load_event
, false);
206 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_dst
, 0);
207 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fault
, 0);
208 qemu_mutex_init(¤t_incoming
->page_request_mutex
);
209 current_incoming
->page_requested
= g_tree_new(page_request_addr_cmp
);
211 migration_object_check(current_migration
, &error_fatal
);
215 dirty_bitmap_mig_init();
218 void migration_cancel(void)
220 migrate_fd_cancel(current_migration
);
223 void migration_shutdown(void)
226 * Cancel the current migration - that will (eventually)
227 * stop the migration using this structure
230 object_unref(OBJECT(current_migration
));
233 * Cancel outgoing migration of dirty bitmaps. It should
234 * at least unref used block nodes.
236 dirty_bitmap_mig_cancel_outgoing();
239 * Cancel incoming migration of dirty bitmaps. Dirty bitmaps
240 * are non-critical data, and their loss never considered as
243 dirty_bitmap_mig_cancel_incoming();
247 MigrationState
*migrate_get_current(void)
249 /* This can only be called after the object created. */
250 assert(current_migration
);
251 return current_migration
;
254 MigrationIncomingState
*migration_incoming_get_current(void)
256 assert(current_incoming
);
257 return current_incoming
;
260 void migration_incoming_state_destroy(void)
262 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
264 if (mis
->to_src_file
) {
265 /* Tell source that we are done */
266 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
267 qemu_fclose(mis
->to_src_file
);
268 mis
->to_src_file
= NULL
;
271 if (mis
->from_src_file
) {
272 migration_ioc_unregister_yank_from_file(mis
->from_src_file
);
273 qemu_fclose(mis
->from_src_file
);
274 mis
->from_src_file
= NULL
;
276 if (mis
->postcopy_remote_fds
) {
277 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
278 mis
->postcopy_remote_fds
= NULL
;
280 if (mis
->transport_cleanup
) {
281 mis
->transport_cleanup(mis
->transport_data
);
284 qemu_event_reset(&mis
->main_thread_load_event
);
286 if (mis
->page_requested
) {
287 g_tree_destroy(mis
->page_requested
);
288 mis
->page_requested
= NULL
;
291 if (mis
->socket_address_list
) {
292 qapi_free_SocketAddressList(mis
->socket_address_list
);
293 mis
->socket_address_list
= NULL
;
296 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
299 static void migrate_generate_event(int new_state
)
301 if (migrate_use_events()) {
302 qapi_event_send_migration(new_state
);
306 static bool migrate_late_block_activate(void)
310 s
= migrate_get_current();
312 return s
->enabled_capabilities
[
313 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE
];
317 * Send a message on the return channel back to the source
320 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
321 enum mig_rp_message_type message_type
,
322 uint16_t len
, void *data
)
326 trace_migrate_send_rp_message((int)message_type
, len
);
327 QEMU_LOCK_GUARD(&mis
->rp_mutex
);
330 * It's possible that the file handle got lost due to network
333 if (!mis
->to_src_file
) {
338 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
339 qemu_put_be16(mis
->to_src_file
, len
);
340 qemu_put_buffer(mis
->to_src_file
, data
, len
);
341 qemu_fflush(mis
->to_src_file
);
343 /* It's possible that qemu file got error during sending */
344 ret
= qemu_file_get_error(mis
->to_src_file
);
349 /* Request one page from the source VM at the given start address.
350 * rb: the RAMBlock to request the page in
351 * Start: Address offset within the RB
352 * Len: Length in bytes required - must be a multiple of pagesize
354 int migrate_send_rp_message_req_pages(MigrationIncomingState
*mis
,
355 RAMBlock
*rb
, ram_addr_t start
)
357 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
358 size_t msglen
= 12; /* start + len */
359 size_t len
= qemu_ram_pagesize(rb
);
360 enum mig_rp_message_type msg_type
;
364 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
365 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
368 * We maintain the last ramblock that we requested for page. Note that we
369 * don't need locking because this function will only be called within the
370 * postcopy ram fault thread.
372 if (rb
!= mis
->last_rb
) {
375 rbname
= qemu_ram_get_idstr(rb
);
376 rbname_len
= strlen(rbname
);
378 assert(rbname_len
< 256);
380 bufc
[msglen
++] = rbname_len
;
381 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
382 msglen
+= rbname_len
;
383 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
385 msg_type
= MIG_RP_MSG_REQ_PAGES
;
388 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
391 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
,
392 RAMBlock
*rb
, ram_addr_t start
, uint64_t haddr
)
394 void *aligned
= (void *)(uintptr_t)(haddr
& (-qemu_ram_pagesize(rb
)));
395 bool received
= false;
397 WITH_QEMU_LOCK_GUARD(&mis
->page_request_mutex
) {
398 received
= ramblock_recv_bitmap_test_byte_offset(rb
, start
);
399 if (!received
&& !g_tree_lookup(mis
->page_requested
, aligned
)) {
401 * The page has not been received, and it's not yet in the page
402 * request list. Queue it. Set the value of element to 1, so that
403 * things like g_tree_lookup() will return TRUE (1) when found.
405 g_tree_insert(mis
->page_requested
, aligned
, (gpointer
)1);
406 mis
->page_requested_count
++;
407 trace_postcopy_page_req_add(aligned
, mis
->page_requested_count
);
412 * If the page is there, skip sending the message. We don't even need the
413 * lock because as long as the page arrived, it'll be there forever.
419 return migrate_send_rp_message_req_pages(mis
, rb
, start
);
422 static bool migration_colo_enabled
;
423 bool migration_incoming_colo_enabled(void)
425 return migration_colo_enabled
;
428 void migration_incoming_disable_colo(void)
430 ram_block_discard_disable(false);
431 migration_colo_enabled
= false;
434 int migration_incoming_enable_colo(void)
436 if (ram_block_discard_disable(true)) {
437 error_report("COLO: cannot disable RAM discard");
440 migration_colo_enabled
= true;
444 void migrate_add_address(SocketAddress
*address
)
446 MigrationIncomingState
*mis
= migration_incoming_get_current();
448 QAPI_LIST_PREPEND(mis
->socket_address_list
,
449 QAPI_CLONE(SocketAddress
, address
));
452 static void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
454 const char *p
= NULL
;
456 migrate_protocol_allow_multifd(false); /* reset it anyway */
457 qapi_event_send_migration(MIGRATION_STATUS_SETUP
);
458 if (strstart(uri
, "tcp:", &p
) ||
459 strstart(uri
, "unix:", NULL
) ||
460 strstart(uri
, "vsock:", NULL
)) {
461 migrate_protocol_allow_multifd(true);
462 socket_start_incoming_migration(p
? p
: uri
, errp
);
464 } else if (strstart(uri
, "rdma:", &p
)) {
465 rdma_start_incoming_migration(p
, errp
);
467 } else if (strstart(uri
, "exec:", &p
)) {
468 exec_start_incoming_migration(p
, errp
);
469 } else if (strstart(uri
, "fd:", &p
)) {
470 fd_start_incoming_migration(p
, errp
);
472 error_setg(errp
, "unknown migration protocol: %s", uri
);
476 static void process_incoming_migration_bh(void *opaque
)
478 Error
*local_err
= NULL
;
479 MigrationIncomingState
*mis
= opaque
;
481 /* If capability late_block_activate is set:
482 * Only fire up the block code now if we're going to restart the
483 * VM, else 'cont' will do it.
484 * This causes file locking to happen; so we don't want it to happen
485 * unless we really are starting the VM.
487 if (!migrate_late_block_activate() ||
488 (autostart
&& (!global_state_received() ||
489 global_state_get_runstate() == RUN_STATE_RUNNING
))) {
490 /* Make sure all file formats flush their mutable metadata.
491 * If we get an error here, just don't restart the VM yet. */
492 bdrv_invalidate_cache_all(&local_err
);
494 error_report_err(local_err
);
501 * This must happen after all error conditions are dealt with and
502 * we're sure the VM is going to be running on this host.
504 qemu_announce_self(&mis
->announce_timer
, migrate_announce_params());
506 if (multifd_load_cleanup(&local_err
) != 0) {
507 error_report_err(local_err
);
510 /* If global state section was not received or we are in running
511 state, we need to obey autostart. Any other state is set with
514 dirty_bitmap_mig_before_vm_start();
516 if (!global_state_received() ||
517 global_state_get_runstate() == RUN_STATE_RUNNING
) {
521 runstate_set(RUN_STATE_PAUSED
);
523 } else if (migration_incoming_colo_enabled()) {
524 migration_incoming_disable_colo();
527 runstate_set(global_state_get_runstate());
530 * This must happen after any state changes since as soon as an external
531 * observer sees this event they might start to prod at the VM assuming
534 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
535 MIGRATION_STATUS_COMPLETED
);
536 qemu_bh_delete(mis
->bh
);
537 migration_incoming_state_destroy();
540 static void process_incoming_migration_co(void *opaque
)
542 MigrationIncomingState
*mis
= migration_incoming_get_current();
545 Error
*local_err
= NULL
;
547 assert(mis
->from_src_file
);
548 mis
->migration_incoming_co
= qemu_coroutine_self();
549 mis
->largest_page_size
= qemu_ram_pagesize_largest();
550 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
551 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
552 MIGRATION_STATUS_ACTIVE
);
553 ret
= qemu_loadvm_state(mis
->from_src_file
);
555 ps
= postcopy_state_get();
556 trace_process_incoming_migration_co_end(ret
, ps
);
557 if (ps
!= POSTCOPY_INCOMING_NONE
) {
558 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
560 * Where a migration had postcopy enabled (and thus went to advise)
561 * but managed to complete within the precopy period, we can use
564 postcopy_ram_incoming_cleanup(mis
);
565 } else if (ret
>= 0) {
567 * Postcopy was started, cleanup should happen at the end of the
570 trace_process_incoming_migration_co_postcopy_end_main();
573 /* Else if something went wrong then just fall out of the normal exit */
576 /* we get COLO info, and know if we are in COLO mode */
577 if (!ret
&& migration_incoming_colo_enabled()) {
578 /* Make sure all file formats flush their mutable metadata */
579 bdrv_invalidate_cache_all(&local_err
);
581 error_report_err(local_err
);
585 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
586 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
587 mis
->have_colo_incoming_thread
= true;
588 qemu_coroutine_yield();
590 /* Wait checkpoint incoming thread exit before free resource */
591 qemu_thread_join(&mis
->colo_incoming_thread
);
592 /* We hold the global iothread lock, so it is safe here */
593 colo_release_ram_cache();
597 error_report("load of migration failed: %s", strerror(-ret
));
600 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
601 qemu_bh_schedule(mis
->bh
);
602 mis
->migration_incoming_co
= NULL
;
606 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
607 MIGRATION_STATUS_FAILED
);
608 qemu_fclose(mis
->from_src_file
);
609 if (multifd_load_cleanup(&local_err
) != 0) {
610 error_report_err(local_err
);
616 * migration_incoming_setup: Setup incoming migration
617 * @f: file for main migration channel
618 * @errp: where to put errors
620 * Returns: %true on success, %false on error.
622 static bool migration_incoming_setup(QEMUFile
*f
, Error
**errp
)
624 MigrationIncomingState
*mis
= migration_incoming_get_current();
626 if (multifd_load_setup(errp
) != 0) {
630 if (!mis
->from_src_file
) {
631 mis
->from_src_file
= f
;
633 qemu_file_set_blocking(f
, false);
637 void migration_incoming_process(void)
639 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
640 qemu_coroutine_enter(co
);
643 /* Returns true if recovered from a paused migration, otherwise false */
644 static bool postcopy_try_recover(QEMUFile
*f
)
646 MigrationIncomingState
*mis
= migration_incoming_get_current();
648 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
649 /* Resumed from a paused postcopy migration */
651 mis
->from_src_file
= f
;
652 /* Postcopy has standalone thread to do vm load */
653 qemu_file_set_blocking(f
, true);
655 /* Re-configure the return path */
656 mis
->to_src_file
= qemu_file_get_return_path(f
);
658 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
659 MIGRATION_STATUS_POSTCOPY_RECOVER
);
662 * Here, we only wake up the main loading thread (while the
663 * fault thread will still be waiting), so that we can receive
664 * commands from source now, and answer it if needed. The
665 * fault thread will be woken up afterwards until we are sure
666 * that source is ready to reply to page requests.
668 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
675 void migration_fd_process_incoming(QEMUFile
*f
, Error
**errp
)
677 if (postcopy_try_recover(f
)) {
681 if (!migration_incoming_setup(f
, errp
)) {
684 migration_incoming_process();
687 void migration_ioc_process_incoming(QIOChannel
*ioc
, Error
**errp
)
689 MigrationIncomingState
*mis
= migration_incoming_get_current();
690 Error
*local_err
= NULL
;
691 bool start_migration
;
693 if (!mis
->from_src_file
) {
694 /* The first connection (multifd may have multiple) */
695 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
697 /* If it's a recovery, we're done */
698 if (postcopy_try_recover(f
)) {
702 if (!migration_incoming_setup(f
, errp
)) {
707 * Common migration only needs one channel, so we can start
708 * right now. Multifd needs more than one channel, we wait.
710 start_migration
= !migrate_use_multifd();
712 /* Multiple connections */
713 assert(migrate_use_multifd());
714 start_migration
= multifd_recv_new_channel(ioc
, &local_err
);
716 error_propagate(errp
, local_err
);
721 if (start_migration
) {
722 migration_incoming_process();
727 * @migration_has_all_channels: We have received all channels that we need
729 * Returns true when we have got connections to all the channels that
730 * we need for migration.
732 bool migration_has_all_channels(void)
734 MigrationIncomingState
*mis
= migration_incoming_get_current();
737 all_channels
= multifd_recv_all_channels_created();
739 return all_channels
&& mis
->from_src_file
!= NULL
;
743 * Send a 'SHUT' message on the return channel with the given value
744 * to indicate that we've finished with the RP. Non-0 value indicates
747 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
752 buf
= cpu_to_be32(value
);
753 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
757 * Send a 'PONG' message on the return channel with the given value
758 * (normally in response to a 'PING')
760 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
765 buf
= cpu_to_be32(value
);
766 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
769 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
777 * First, we send the header part. It contains only the len of
778 * idstr, and the idstr itself.
780 len
= strlen(block_name
);
782 memcpy(buf
+ 1, block_name
, len
);
784 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
785 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
790 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
793 * Next, we dump the received bitmap to the stream.
795 * TODO: currently we are safe since we are the only one that is
796 * using the to_src_file handle (fault thread is still paused),
797 * and it's ok even not taking the mutex. However the best way is
798 * to take the lock before sending the message header, and release
799 * the lock after sending the bitmap.
801 qemu_mutex_lock(&mis
->rp_mutex
);
802 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
803 qemu_mutex_unlock(&mis
->rp_mutex
);
805 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
808 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
812 buf
= cpu_to_be32(value
);
813 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
816 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
818 MigrationCapabilityStatusList
*head
= NULL
, **tail
= &head
;
819 MigrationCapabilityStatus
*caps
;
820 MigrationState
*s
= migrate_get_current();
823 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
824 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
825 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
829 caps
= g_malloc0(sizeof(*caps
));
830 caps
->capability
= i
;
831 caps
->state
= s
->enabled_capabilities
[i
];
832 QAPI_LIST_APPEND(tail
, caps
);
838 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
840 MigrationParameters
*params
;
841 MigrationState
*s
= migrate_get_current();
843 /* TODO use QAPI_CLONE() instead of duplicating it inline */
844 params
= g_malloc0(sizeof(*params
));
845 params
->has_compress_level
= true;
846 params
->compress_level
= s
->parameters
.compress_level
;
847 params
->has_compress_threads
= true;
848 params
->compress_threads
= s
->parameters
.compress_threads
;
849 params
->has_compress_wait_thread
= true;
850 params
->compress_wait_thread
= s
->parameters
.compress_wait_thread
;
851 params
->has_decompress_threads
= true;
852 params
->decompress_threads
= s
->parameters
.decompress_threads
;
853 params
->has_throttle_trigger_threshold
= true;
854 params
->throttle_trigger_threshold
= s
->parameters
.throttle_trigger_threshold
;
855 params
->has_cpu_throttle_initial
= true;
856 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
857 params
->has_cpu_throttle_increment
= true;
858 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
859 params
->has_cpu_throttle_tailslow
= true;
860 params
->cpu_throttle_tailslow
= s
->parameters
.cpu_throttle_tailslow
;
861 params
->has_tls_creds
= true;
862 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
863 params
->has_tls_hostname
= true;
864 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
865 params
->has_tls_authz
= true;
866 params
->tls_authz
= g_strdup(s
->parameters
.tls_authz
?
867 s
->parameters
.tls_authz
: "");
868 params
->has_max_bandwidth
= true;
869 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
870 params
->has_downtime_limit
= true;
871 params
->downtime_limit
= s
->parameters
.downtime_limit
;
872 params
->has_x_checkpoint_delay
= true;
873 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
874 params
->has_block_incremental
= true;
875 params
->block_incremental
= s
->parameters
.block_incremental
;
876 params
->has_multifd_channels
= true;
877 params
->multifd_channels
= s
->parameters
.multifd_channels
;
878 params
->has_multifd_compression
= true;
879 params
->multifd_compression
= s
->parameters
.multifd_compression
;
880 params
->has_multifd_zlib_level
= true;
881 params
->multifd_zlib_level
= s
->parameters
.multifd_zlib_level
;
882 params
->has_multifd_zstd_level
= true;
883 params
->multifd_zstd_level
= s
->parameters
.multifd_zstd_level
;
884 params
->has_xbzrle_cache_size
= true;
885 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
886 params
->has_max_postcopy_bandwidth
= true;
887 params
->max_postcopy_bandwidth
= s
->parameters
.max_postcopy_bandwidth
;
888 params
->has_max_cpu_throttle
= true;
889 params
->max_cpu_throttle
= s
->parameters
.max_cpu_throttle
;
890 params
->has_announce_initial
= true;
891 params
->announce_initial
= s
->parameters
.announce_initial
;
892 params
->has_announce_max
= true;
893 params
->announce_max
= s
->parameters
.announce_max
;
894 params
->has_announce_rounds
= true;
895 params
->announce_rounds
= s
->parameters
.announce_rounds
;
896 params
->has_announce_step
= true;
897 params
->announce_step
= s
->parameters
.announce_step
;
899 if (s
->parameters
.has_block_bitmap_mapping
) {
900 params
->has_block_bitmap_mapping
= true;
901 params
->block_bitmap_mapping
=
902 QAPI_CLONE(BitmapMigrationNodeAliasList
,
903 s
->parameters
.block_bitmap_mapping
);
909 AnnounceParameters
*migrate_announce_params(void)
911 static AnnounceParameters ap
;
913 MigrationState
*s
= migrate_get_current();
915 ap
.initial
= s
->parameters
.announce_initial
;
916 ap
.max
= s
->parameters
.announce_max
;
917 ap
.rounds
= s
->parameters
.announce_rounds
;
918 ap
.step
= s
->parameters
.announce_step
;
924 * Return true if we're already in the middle of a migration
925 * (i.e. any of the active or setup states)
927 bool migration_is_setup_or_active(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_COLO
:
947 bool migration_is_running(int state
)
950 case MIGRATION_STATUS_ACTIVE
:
951 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
952 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
953 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
954 case MIGRATION_STATUS_SETUP
:
955 case MIGRATION_STATUS_PRE_SWITCHOVER
:
956 case MIGRATION_STATUS_DEVICE
:
957 case MIGRATION_STATUS_WAIT_UNPLUG
:
958 case MIGRATION_STATUS_CANCELLING
:
967 static void populate_time_info(MigrationInfo
*info
, MigrationState
*s
)
969 info
->has_status
= true;
970 info
->has_setup_time
= true;
971 info
->setup_time
= s
->setup_time
;
972 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
973 info
->has_total_time
= true;
974 info
->total_time
= s
->total_time
;
975 info
->has_downtime
= true;
976 info
->downtime
= s
->downtime
;
978 info
->has_total_time
= true;
979 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) -
981 info
->has_expected_downtime
= true;
982 info
->expected_downtime
= s
->expected_downtime
;
986 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
988 info
->has_ram
= true;
989 info
->ram
= g_malloc0(sizeof(*info
->ram
));
990 info
->ram
->transferred
= ram_counters
.transferred
;
991 info
->ram
->total
= ram_bytes_total();
992 info
->ram
->duplicate
= ram_counters
.duplicate
;
993 /* legacy value. It is not used anymore */
994 info
->ram
->skipped
= 0;
995 info
->ram
->normal
= ram_counters
.normal
;
996 info
->ram
->normal_bytes
= ram_counters
.normal
*
997 qemu_target_page_size();
998 info
->ram
->mbps
= s
->mbps
;
999 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
1000 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
1001 info
->ram
->page_size
= qemu_target_page_size();
1002 info
->ram
->multifd_bytes
= ram_counters
.multifd_bytes
;
1003 info
->ram
->pages_per_second
= s
->pages_per_second
;
1005 if (migrate_use_xbzrle()) {
1006 info
->has_xbzrle_cache
= true;
1007 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
1008 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
1009 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
1010 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
1011 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
1012 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
1013 info
->xbzrle_cache
->encoding_rate
= xbzrle_counters
.encoding_rate
;
1014 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
1017 if (migrate_use_compression()) {
1018 info
->has_compression
= true;
1019 info
->compression
= g_malloc0(sizeof(*info
->compression
));
1020 info
->compression
->pages
= compression_counters
.pages
;
1021 info
->compression
->busy
= compression_counters
.busy
;
1022 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
1023 info
->compression
->compressed_size
=
1024 compression_counters
.compressed_size
;
1025 info
->compression
->compression_rate
=
1026 compression_counters
.compression_rate
;
1029 if (cpu_throttle_active()) {
1030 info
->has_cpu_throttle_percentage
= true;
1031 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
1034 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
1035 info
->ram
->remaining
= ram_bytes_remaining();
1036 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
1040 static void populate_disk_info(MigrationInfo
*info
)
1042 if (blk_mig_active()) {
1043 info
->has_disk
= true;
1044 info
->disk
= g_malloc0(sizeof(*info
->disk
));
1045 info
->disk
->transferred
= blk_mig_bytes_transferred();
1046 info
->disk
->remaining
= blk_mig_bytes_remaining();
1047 info
->disk
->total
= blk_mig_bytes_total();
1051 static void fill_source_migration_info(MigrationInfo
*info
)
1053 MigrationState
*s
= migrate_get_current();
1054 GSList
*cur_blocker
= migration_blockers
;
1056 info
->blocked_reasons
= NULL
;
1059 * There are two types of reasons a migration might be blocked;
1060 * a) devices marked in VMState as non-migratable, and
1061 * b) Explicit migration blockers
1062 * We need to add both of them here.
1064 qemu_savevm_non_migratable_list(&info
->blocked_reasons
);
1066 while (cur_blocker
) {
1067 QAPI_LIST_PREPEND(info
->blocked_reasons
,
1068 g_strdup(error_get_pretty(cur_blocker
->data
)));
1069 cur_blocker
= g_slist_next(cur_blocker
);
1071 info
->has_blocked_reasons
= info
->blocked_reasons
!= NULL
;
1074 case MIGRATION_STATUS_NONE
:
1075 /* no migration has happened ever */
1076 /* do not overwrite destination migration status */
1078 case MIGRATION_STATUS_SETUP
:
1079 info
->has_status
= true;
1080 info
->has_total_time
= false;
1082 case MIGRATION_STATUS_ACTIVE
:
1083 case MIGRATION_STATUS_CANCELLING
:
1084 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1085 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1086 case MIGRATION_STATUS_DEVICE
:
1087 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1088 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1089 /* TODO add some postcopy stats */
1090 populate_time_info(info
, s
);
1091 populate_ram_info(info
, s
);
1092 populate_disk_info(info
);
1093 populate_vfio_info(info
);
1095 case MIGRATION_STATUS_COLO
:
1096 info
->has_status
= true;
1097 /* TODO: display COLO specific information (checkpoint info etc.) */
1099 case MIGRATION_STATUS_COMPLETED
:
1100 populate_time_info(info
, s
);
1101 populate_ram_info(info
, s
);
1102 populate_vfio_info(info
);
1104 case MIGRATION_STATUS_FAILED
:
1105 info
->has_status
= true;
1107 info
->has_error_desc
= true;
1108 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
1111 case MIGRATION_STATUS_CANCELLED
:
1112 info
->has_status
= true;
1114 case MIGRATION_STATUS_WAIT_UNPLUG
:
1115 info
->has_status
= true;
1118 info
->status
= s
->state
;
1121 typedef enum WriteTrackingSupport
{
1122 WT_SUPPORT_UNKNOWN
= 0,
1124 WT_SUPPORT_AVAILABLE
,
1125 WT_SUPPORT_COMPATIBLE
1126 } WriteTrackingSupport
;
1129 WriteTrackingSupport
migrate_query_write_tracking(void)
1131 /* Check if kernel supports required UFFD features */
1132 if (!ram_write_tracking_available()) {
1133 return WT_SUPPORT_ABSENT
;
1136 * Check if current memory configuration is
1137 * compatible with required UFFD features.
1139 if (!ram_write_tracking_compatible()) {
1140 return WT_SUPPORT_AVAILABLE
;
1143 return WT_SUPPORT_COMPATIBLE
;
1147 * @migration_caps_check - check capability validity
1149 * @cap_list: old capability list, array of bool
1150 * @params: new capabilities to be applied soon
1151 * @errp: set *errp if the check failed, with reason
1153 * Returns true if check passed, otherwise false.
1155 static bool migrate_caps_check(bool *cap_list
,
1156 MigrationCapabilityStatusList
*params
,
1159 MigrationCapabilityStatusList
*cap
;
1160 bool old_postcopy_cap
;
1161 MigrationIncomingState
*mis
= migration_incoming_get_current();
1163 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
1165 for (cap
= params
; cap
; cap
= cap
->next
) {
1166 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
1169 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
1170 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
1171 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
1173 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
1178 #ifndef CONFIG_REPLICATION
1179 if (cap_list
[MIGRATION_CAPABILITY_X_COLO
]) {
1180 error_setg(errp
, "QEMU compiled without replication module"
1181 " can't enable COLO");
1182 error_append_hint(errp
, "Please enable replication before COLO.\n");
1187 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
1188 /* This check is reasonably expensive, so only when it's being
1189 * set the first time, also it's only the destination that needs
1192 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
1193 !postcopy_ram_supported_by_host(mis
)) {
1194 /* postcopy_ram_supported_by_host will have emitted a more
1197 error_setg(errp
, "Postcopy is not supported");
1201 if (cap_list
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
]) {
1202 error_setg(errp
, "Postcopy is not compatible with ignore-shared");
1207 if (cap_list
[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
]) {
1208 WriteTrackingSupport wt_support
;
1211 * Check if 'background-snapshot' capability is supported by
1212 * host kernel and compatible with guest memory configuration.
1214 wt_support
= migrate_query_write_tracking();
1215 if (wt_support
< WT_SUPPORT_AVAILABLE
) {
1216 error_setg(errp
, "Background-snapshot is not supported by host kernel");
1219 if (wt_support
< WT_SUPPORT_COMPATIBLE
) {
1220 error_setg(errp
, "Background-snapshot is not compatible "
1221 "with guest memory configuration");
1226 * Check if there are any migration capabilities
1227 * incompatible with 'background-snapshot'.
1229 for (idx
= 0; idx
< check_caps_background_snapshot
.size
; idx
++) {
1230 int incomp_cap
= check_caps_background_snapshot
.caps
[idx
];
1231 if (cap_list
[incomp_cap
]) {
1233 "Background-snapshot is not compatible with %s",
1234 MigrationCapability_str(incomp_cap
));
1240 /* incoming side only */
1241 if (runstate_check(RUN_STATE_INMIGRATE
) &&
1242 !migrate_multifd_is_allowed() &&
1243 cap_list
[MIGRATION_CAPABILITY_MULTIFD
]) {
1244 error_setg(errp
, "multifd is not supported by current protocol");
1251 static void fill_destination_migration_info(MigrationInfo
*info
)
1253 MigrationIncomingState
*mis
= migration_incoming_get_current();
1255 if (mis
->socket_address_list
) {
1256 info
->has_socket_address
= true;
1257 info
->socket_address
=
1258 QAPI_CLONE(SocketAddressList
, mis
->socket_address_list
);
1261 switch (mis
->state
) {
1262 case MIGRATION_STATUS_NONE
:
1264 case MIGRATION_STATUS_SETUP
:
1265 case MIGRATION_STATUS_CANCELLING
:
1266 case MIGRATION_STATUS_CANCELLED
:
1267 case MIGRATION_STATUS_ACTIVE
:
1268 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1269 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1270 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1271 case MIGRATION_STATUS_FAILED
:
1272 case MIGRATION_STATUS_COLO
:
1273 info
->has_status
= true;
1275 case MIGRATION_STATUS_COMPLETED
:
1276 info
->has_status
= true;
1277 fill_destination_postcopy_migration_info(info
);
1280 info
->status
= mis
->state
;
1283 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1285 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1287 fill_destination_migration_info(info
);
1288 fill_source_migration_info(info
);
1293 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
1296 MigrationState
*s
= migrate_get_current();
1297 MigrationCapabilityStatusList
*cap
;
1298 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
1300 if (migration_is_running(s
->state
)) {
1301 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1305 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
1306 if (!migrate_caps_check(cap_list
, params
, errp
)) {
1310 for (cap
= params
; cap
; cap
= cap
->next
) {
1311 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
1316 * Check whether the parameters are valid. Error will be put into errp
1317 * (if provided). Return true if valid, otherwise false.
1319 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
1321 if (params
->has_compress_level
&&
1322 (params
->compress_level
> 9)) {
1323 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
1324 "a value between 0 and 9");
1328 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
1329 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1331 "a value between 1 and 255");
1335 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
1336 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1337 "decompress_threads",
1338 "a value between 1 and 255");
1342 if (params
->has_throttle_trigger_threshold
&&
1343 (params
->throttle_trigger_threshold
< 1 ||
1344 params
->throttle_trigger_threshold
> 100)) {
1345 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1346 "throttle_trigger_threshold",
1347 "an integer in the range of 1 to 100");
1351 if (params
->has_cpu_throttle_initial
&&
1352 (params
->cpu_throttle_initial
< 1 ||
1353 params
->cpu_throttle_initial
> 99)) {
1354 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1355 "cpu_throttle_initial",
1356 "an integer in the range of 1 to 99");
1360 if (params
->has_cpu_throttle_increment
&&
1361 (params
->cpu_throttle_increment
< 1 ||
1362 params
->cpu_throttle_increment
> 99)) {
1363 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1364 "cpu_throttle_increment",
1365 "an integer in the range of 1 to 99");
1369 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
1370 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1372 "an integer in the range of 0 to "stringify(SIZE_MAX
)
1377 if (params
->has_downtime_limit
&&
1378 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
1379 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1381 "an integer in the range of 0 to "
1382 stringify(MAX_MIGRATE_DOWNTIME
)" ms");
1386 /* x_checkpoint_delay is now always positive */
1388 if (params
->has_multifd_channels
&& (params
->multifd_channels
< 1)) {
1389 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1391 "a value between 1 and 255");
1395 if (params
->has_multifd_zlib_level
&&
1396 (params
->multifd_zlib_level
> 9)) {
1397 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zlib_level",
1398 "a value between 0 and 9");
1402 if (params
->has_multifd_zstd_level
&&
1403 (params
->multifd_zstd_level
> 20)) {
1404 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "multifd_zstd_level",
1405 "a value between 0 and 20");
1409 if (params
->has_xbzrle_cache_size
&&
1410 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
1411 !is_power_of_2(params
->xbzrle_cache_size
))) {
1412 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1413 "xbzrle_cache_size",
1414 "a power of two no less than the target page size");
1418 if (params
->has_max_cpu_throttle
&&
1419 (params
->max_cpu_throttle
< params
->cpu_throttle_initial
||
1420 params
->max_cpu_throttle
> 99)) {
1421 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1423 "an integer in the range of cpu_throttle_initial to 99");
1427 if (params
->has_announce_initial
&&
1428 params
->announce_initial
> 100000) {
1429 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1431 "a value between 0 and 100000");
1434 if (params
->has_announce_max
&&
1435 params
->announce_max
> 100000) {
1436 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1438 "a value between 0 and 100000");
1441 if (params
->has_announce_rounds
&&
1442 params
->announce_rounds
> 1000) {
1443 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1445 "a value between 0 and 1000");
1448 if (params
->has_announce_step
&&
1449 (params
->announce_step
< 1 ||
1450 params
->announce_step
> 10000)) {
1451 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1453 "a value between 0 and 10000");
1457 if (params
->has_block_bitmap_mapping
&&
1458 !check_dirty_bitmap_mig_alias_map(params
->block_bitmap_mapping
, errp
)) {
1459 error_prepend(errp
, "Invalid mapping given for block-bitmap-mapping: ");
1466 static void migrate_params_test_apply(MigrateSetParameters
*params
,
1467 MigrationParameters
*dest
)
1469 *dest
= migrate_get_current()->parameters
;
1471 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1473 if (params
->has_compress_level
) {
1474 dest
->compress_level
= params
->compress_level
;
1477 if (params
->has_compress_threads
) {
1478 dest
->compress_threads
= params
->compress_threads
;
1481 if (params
->has_compress_wait_thread
) {
1482 dest
->compress_wait_thread
= params
->compress_wait_thread
;
1485 if (params
->has_decompress_threads
) {
1486 dest
->decompress_threads
= params
->decompress_threads
;
1489 if (params
->has_throttle_trigger_threshold
) {
1490 dest
->throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1493 if (params
->has_cpu_throttle_initial
) {
1494 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
1497 if (params
->has_cpu_throttle_increment
) {
1498 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
1501 if (params
->has_cpu_throttle_tailslow
) {
1502 dest
->cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1505 if (params
->has_tls_creds
) {
1506 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1507 dest
->tls_creds
= params
->tls_creds
->u
.s
;
1510 if (params
->has_tls_hostname
) {
1511 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1512 dest
->tls_hostname
= params
->tls_hostname
->u
.s
;
1515 if (params
->has_max_bandwidth
) {
1516 dest
->max_bandwidth
= params
->max_bandwidth
;
1519 if (params
->has_downtime_limit
) {
1520 dest
->downtime_limit
= params
->downtime_limit
;
1523 if (params
->has_x_checkpoint_delay
) {
1524 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
1527 if (params
->has_block_incremental
) {
1528 dest
->block_incremental
= params
->block_incremental
;
1530 if (params
->has_multifd_channels
) {
1531 dest
->multifd_channels
= params
->multifd_channels
;
1533 if (params
->has_multifd_compression
) {
1534 dest
->multifd_compression
= params
->multifd_compression
;
1536 if (params
->has_xbzrle_cache_size
) {
1537 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
1539 if (params
->has_max_postcopy_bandwidth
) {
1540 dest
->max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1542 if (params
->has_max_cpu_throttle
) {
1543 dest
->max_cpu_throttle
= params
->max_cpu_throttle
;
1545 if (params
->has_announce_initial
) {
1546 dest
->announce_initial
= params
->announce_initial
;
1548 if (params
->has_announce_max
) {
1549 dest
->announce_max
= params
->announce_max
;
1551 if (params
->has_announce_rounds
) {
1552 dest
->announce_rounds
= params
->announce_rounds
;
1554 if (params
->has_announce_step
) {
1555 dest
->announce_step
= params
->announce_step
;
1558 if (params
->has_block_bitmap_mapping
) {
1559 dest
->has_block_bitmap_mapping
= true;
1560 dest
->block_bitmap_mapping
= params
->block_bitmap_mapping
;
1564 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
1566 MigrationState
*s
= migrate_get_current();
1568 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1570 if (params
->has_compress_level
) {
1571 s
->parameters
.compress_level
= params
->compress_level
;
1574 if (params
->has_compress_threads
) {
1575 s
->parameters
.compress_threads
= params
->compress_threads
;
1578 if (params
->has_compress_wait_thread
) {
1579 s
->parameters
.compress_wait_thread
= params
->compress_wait_thread
;
1582 if (params
->has_decompress_threads
) {
1583 s
->parameters
.decompress_threads
= params
->decompress_threads
;
1586 if (params
->has_throttle_trigger_threshold
) {
1587 s
->parameters
.throttle_trigger_threshold
= params
->throttle_trigger_threshold
;
1590 if (params
->has_cpu_throttle_initial
) {
1591 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
1594 if (params
->has_cpu_throttle_increment
) {
1595 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
1598 if (params
->has_cpu_throttle_tailslow
) {
1599 s
->parameters
.cpu_throttle_tailslow
= params
->cpu_throttle_tailslow
;
1602 if (params
->has_tls_creds
) {
1603 g_free(s
->parameters
.tls_creds
);
1604 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1605 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1608 if (params
->has_tls_hostname
) {
1609 g_free(s
->parameters
.tls_hostname
);
1610 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1611 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1614 if (params
->has_tls_authz
) {
1615 g_free(s
->parameters
.tls_authz
);
1616 assert(params
->tls_authz
->type
== QTYPE_QSTRING
);
1617 s
->parameters
.tls_authz
= g_strdup(params
->tls_authz
->u
.s
);
1620 if (params
->has_max_bandwidth
) {
1621 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1622 if (s
->to_dst_file
&& !migration_in_postcopy()) {
1623 qemu_file_set_rate_limit(s
->to_dst_file
,
1624 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1628 if (params
->has_downtime_limit
) {
1629 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1632 if (params
->has_x_checkpoint_delay
) {
1633 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1634 if (migration_in_colo_state()) {
1635 colo_checkpoint_notify(s
);
1639 if (params
->has_block_incremental
) {
1640 s
->parameters
.block_incremental
= params
->block_incremental
;
1642 if (params
->has_multifd_channels
) {
1643 s
->parameters
.multifd_channels
= params
->multifd_channels
;
1645 if (params
->has_multifd_compression
) {
1646 s
->parameters
.multifd_compression
= params
->multifd_compression
;
1648 if (params
->has_xbzrle_cache_size
) {
1649 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1650 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1652 if (params
->has_max_postcopy_bandwidth
) {
1653 s
->parameters
.max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1654 if (s
->to_dst_file
&& migration_in_postcopy()) {
1655 qemu_file_set_rate_limit(s
->to_dst_file
,
1656 s
->parameters
.max_postcopy_bandwidth
/ XFER_LIMIT_RATIO
);
1659 if (params
->has_max_cpu_throttle
) {
1660 s
->parameters
.max_cpu_throttle
= params
->max_cpu_throttle
;
1662 if (params
->has_announce_initial
) {
1663 s
->parameters
.announce_initial
= params
->announce_initial
;
1665 if (params
->has_announce_max
) {
1666 s
->parameters
.announce_max
= params
->announce_max
;
1668 if (params
->has_announce_rounds
) {
1669 s
->parameters
.announce_rounds
= params
->announce_rounds
;
1671 if (params
->has_announce_step
) {
1672 s
->parameters
.announce_step
= params
->announce_step
;
1675 if (params
->has_block_bitmap_mapping
) {
1676 qapi_free_BitmapMigrationNodeAliasList(
1677 s
->parameters
.block_bitmap_mapping
);
1679 s
->parameters
.has_block_bitmap_mapping
= true;
1680 s
->parameters
.block_bitmap_mapping
=
1681 QAPI_CLONE(BitmapMigrationNodeAliasList
,
1682 params
->block_bitmap_mapping
);
1686 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1688 MigrationParameters tmp
;
1690 /* TODO Rewrite "" to null instead */
1691 if (params
->has_tls_creds
1692 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1693 qobject_unref(params
->tls_creds
->u
.n
);
1694 params
->tls_creds
->type
= QTYPE_QSTRING
;
1695 params
->tls_creds
->u
.s
= strdup("");
1697 /* TODO Rewrite "" to null instead */
1698 if (params
->has_tls_hostname
1699 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1700 qobject_unref(params
->tls_hostname
->u
.n
);
1701 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1702 params
->tls_hostname
->u
.s
= strdup("");
1705 migrate_params_test_apply(params
, &tmp
);
1707 if (!migrate_params_check(&tmp
, errp
)) {
1708 /* Invalid parameter */
1712 migrate_params_apply(params
, errp
);
1716 void qmp_migrate_start_postcopy(Error
**errp
)
1718 MigrationState
*s
= migrate_get_current();
1720 if (!migrate_postcopy()) {
1721 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1722 " the start of migration");
1726 if (s
->state
== MIGRATION_STATUS_NONE
) {
1727 error_setg(errp
, "Postcopy must be started after migration has been"
1732 * we don't error if migration has finished since that would be racy
1733 * with issuing this command.
1735 qatomic_set(&s
->start_postcopy
, true);
1738 /* shared migration helpers */
1740 void migrate_set_state(int *state
, int old_state
, int new_state
)
1742 assert(new_state
< MIGRATION_STATUS__MAX
);
1743 if (qatomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1744 trace_migrate_set_state(MigrationStatus_str(new_state
));
1745 migrate_generate_event(new_state
);
1749 static MigrationCapabilityStatus
*migrate_cap_add(MigrationCapability index
,
1752 MigrationCapabilityStatus
*cap
;
1754 cap
= g_new0(MigrationCapabilityStatus
, 1);
1755 cap
->capability
= index
;
1761 void migrate_set_block_enabled(bool value
, Error
**errp
)
1763 MigrationCapabilityStatusList
*cap
= NULL
;
1765 QAPI_LIST_PREPEND(cap
, migrate_cap_add(MIGRATION_CAPABILITY_BLOCK
, value
));
1766 qmp_migrate_set_capabilities(cap
, errp
);
1767 qapi_free_MigrationCapabilityStatusList(cap
);
1770 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1772 s
->parameters
.block_incremental
= value
;
1775 static void block_cleanup_parameters(MigrationState
*s
)
1777 if (s
->must_remove_block_options
) {
1778 /* setting to false can never fail */
1779 migrate_set_block_enabled(false, &error_abort
);
1780 migrate_set_block_incremental(s
, false);
1781 s
->must_remove_block_options
= false;
1785 static void migrate_fd_cleanup(MigrationState
*s
)
1787 qemu_bh_delete(s
->cleanup_bh
);
1788 s
->cleanup_bh
= NULL
;
1790 qemu_savevm_state_cleanup();
1792 if (s
->to_dst_file
) {
1795 trace_migrate_fd_cleanup();
1796 qemu_mutex_unlock_iothread();
1797 if (s
->migration_thread_running
) {
1798 qemu_thread_join(&s
->thread
);
1799 s
->migration_thread_running
= false;
1801 qemu_mutex_lock_iothread();
1803 multifd_save_cleanup();
1804 qemu_mutex_lock(&s
->qemu_file_lock
);
1805 tmp
= s
->to_dst_file
;
1806 s
->to_dst_file
= NULL
;
1807 qemu_mutex_unlock(&s
->qemu_file_lock
);
1809 * Close the file handle without the lock to make sure the
1810 * critical section won't block for long.
1812 migration_ioc_unregister_yank_from_file(tmp
);
1816 assert(!migration_is_active(s
));
1818 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1819 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1820 MIGRATION_STATUS_CANCELLED
);
1824 /* It is used on info migrate. We can't free it */
1825 error_report_err(error_copy(s
->error
));
1827 notifier_list_notify(&migration_state_notifiers
, s
);
1828 block_cleanup_parameters(s
);
1829 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
1832 static void migrate_fd_cleanup_schedule(MigrationState
*s
)
1835 * Ref the state for bh, because it may be called when
1836 * there're already no other refs
1838 object_ref(OBJECT(s
));
1839 qemu_bh_schedule(s
->cleanup_bh
);
1842 static void migrate_fd_cleanup_bh(void *opaque
)
1844 MigrationState
*s
= opaque
;
1845 migrate_fd_cleanup(s
);
1846 object_unref(OBJECT(s
));
1849 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1851 QEMU_LOCK_GUARD(&s
->error_mutex
);
1853 s
->error
= error_copy(error
);
1857 static void migrate_error_free(MigrationState
*s
)
1859 QEMU_LOCK_GUARD(&s
->error_mutex
);
1861 error_free(s
->error
);
1866 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1868 trace_migrate_fd_error(error_get_pretty(error
));
1869 assert(s
->to_dst_file
== NULL
);
1870 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1871 MIGRATION_STATUS_FAILED
);
1872 migrate_set_error(s
, error
);
1875 static void migrate_fd_cancel(MigrationState
*s
)
1878 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1879 trace_migrate_fd_cancel();
1881 WITH_QEMU_LOCK_GUARD(&s
->qemu_file_lock
) {
1882 if (s
->rp_state
.from_dst_file
) {
1883 /* shutdown the rp socket, so causing the rp thread to shutdown */
1884 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1889 old_state
= s
->state
;
1890 if (!migration_is_running(old_state
)) {
1893 /* If the migration is paused, kick it out of the pause */
1894 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1895 qemu_sem_post(&s
->pause_sem
);
1897 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1898 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1901 * If we're unlucky the migration code might be stuck somewhere in a
1902 * send/write while the network has failed and is waiting to timeout;
1903 * if we've got shutdown(2) available then we can force it to quit.
1904 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1905 * called in a bh, so there is no race against this cancel.
1907 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1908 qemu_file_shutdown(f
);
1910 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1911 Error
*local_err
= NULL
;
1913 bdrv_invalidate_cache_all(&local_err
);
1915 error_report_err(local_err
);
1917 s
->block_inactive
= false;
1922 void add_migration_state_change_notifier(Notifier
*notify
)
1924 notifier_list_add(&migration_state_notifiers
, notify
);
1927 void remove_migration_state_change_notifier(Notifier
*notify
)
1929 notifier_remove(notify
);
1932 bool migration_in_setup(MigrationState
*s
)
1934 return s
->state
== MIGRATION_STATUS_SETUP
;
1937 bool migration_has_finished(MigrationState
*s
)
1939 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1942 bool migration_has_failed(MigrationState
*s
)
1944 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1945 s
->state
== MIGRATION_STATUS_FAILED
);
1948 bool migration_in_postcopy(void)
1950 MigrationState
*s
= migrate_get_current();
1953 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1954 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1955 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1962 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1964 return migration_in_postcopy() && s
->postcopy_after_devices
;
1967 bool migration_in_incoming_postcopy(void)
1969 PostcopyState ps
= postcopy_state_get();
1971 return ps
>= POSTCOPY_INCOMING_DISCARD
&& ps
< POSTCOPY_INCOMING_END
;
1974 bool migration_in_bg_snapshot(void)
1976 MigrationState
*s
= migrate_get_current();
1978 return migrate_background_snapshot() &&
1979 migration_is_setup_or_active(s
->state
);
1982 bool migration_is_idle(void)
1984 MigrationState
*s
= current_migration
;
1991 case MIGRATION_STATUS_NONE
:
1992 case MIGRATION_STATUS_CANCELLED
:
1993 case MIGRATION_STATUS_COMPLETED
:
1994 case MIGRATION_STATUS_FAILED
:
1996 case MIGRATION_STATUS_SETUP
:
1997 case MIGRATION_STATUS_CANCELLING
:
1998 case MIGRATION_STATUS_ACTIVE
:
1999 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
2000 case MIGRATION_STATUS_COLO
:
2001 case MIGRATION_STATUS_PRE_SWITCHOVER
:
2002 case MIGRATION_STATUS_DEVICE
:
2003 case MIGRATION_STATUS_WAIT_UNPLUG
:
2005 case MIGRATION_STATUS__MAX
:
2006 g_assert_not_reached();
2012 bool migration_is_active(MigrationState
*s
)
2014 return (s
->state
== MIGRATION_STATUS_ACTIVE
||
2015 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2018 void migrate_init(MigrationState
*s
)
2021 * Reinitialise all migration state, except
2022 * parameters/capabilities that the user set, and
2027 s
->to_dst_file
= NULL
;
2028 s
->state
= MIGRATION_STATUS_NONE
;
2029 s
->rp_state
.from_dst_file
= NULL
;
2030 s
->rp_state
.error
= false;
2032 s
->pages_per_second
= 0.0;
2034 s
->expected_downtime
= 0;
2036 s
->start_postcopy
= false;
2037 s
->postcopy_after_devices
= false;
2038 s
->migration_thread_running
= false;
2039 error_free(s
->error
);
2043 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
2045 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2047 s
->vm_was_running
= false;
2048 s
->iteration_initial_bytes
= 0;
2049 s
->threshold_size
= 0;
2052 int migrate_add_blocker_internal(Error
*reason
, Error
**errp
)
2054 /* Snapshots are similar to migrations, so check RUN_STATE_SAVE_VM too. */
2055 if (runstate_check(RUN_STATE_SAVE_VM
) || !migration_is_idle()) {
2056 error_propagate_prepend(errp
, error_copy(reason
),
2057 "disallowing migration blocker "
2058 "(migration/snapshot in progress) for: ");
2062 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
2066 int migrate_add_blocker(Error
*reason
, Error
**errp
)
2068 if (only_migratable
) {
2069 error_propagate_prepend(errp
, error_copy(reason
),
2070 "disallowing migration blocker "
2071 "(--only-migratable) for: ");
2075 return migrate_add_blocker_internal(reason
, errp
);
2078 void migrate_del_blocker(Error
*reason
)
2080 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
2083 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
2085 Error
*local_err
= NULL
;
2086 static bool once
= true;
2089 error_setg(errp
, "The incoming migration has already been started");
2092 if (!runstate_check(RUN_STATE_INMIGRATE
)) {
2093 error_setg(errp
, "'-incoming' was not specified on the command line");
2097 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
2101 qemu_start_incoming_migration(uri
, &local_err
);
2104 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2105 error_propagate(errp
, local_err
);
2112 void qmp_migrate_recover(const char *uri
, Error
**errp
)
2114 MigrationIncomingState
*mis
= migration_incoming_get_current();
2117 * Don't even bother to use ERRP_GUARD() as it _must_ always be set by
2118 * callers (no one should ignore a recover failure); if there is, it's a
2119 * programming error.
2123 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2124 error_setg(errp
, "Migrate recover can only be run "
2125 "when postcopy is paused.");
2129 if (qatomic_cmpxchg(&mis
->postcopy_recover_triggered
,
2130 false, true) == true) {
2131 error_setg(errp
, "Migrate recovery is triggered already");
2136 * Note that this call will never start a real migration; it will
2137 * only re-setup the migration stream and poke existing migration
2138 * to continue using that newly established channel.
2140 qemu_start_incoming_migration(uri
, errp
);
2142 /* Safe to dereference with the assert above */
2144 /* Reset the flag so user could still retry */
2145 qatomic_set(&mis
->postcopy_recover_triggered
, false);
2149 void qmp_migrate_pause(Error
**errp
)
2151 MigrationState
*ms
= migrate_get_current();
2152 MigrationIncomingState
*mis
= migration_incoming_get_current();
2155 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2156 /* Source side, during postcopy */
2157 qemu_mutex_lock(&ms
->qemu_file_lock
);
2158 ret
= qemu_file_shutdown(ms
->to_dst_file
);
2159 qemu_mutex_unlock(&ms
->qemu_file_lock
);
2161 error_setg(errp
, "Failed to pause source migration");
2166 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2167 ret
= qemu_file_shutdown(mis
->from_src_file
);
2169 error_setg(errp
, "Failed to pause destination migration");
2174 error_setg(errp
, "migrate-pause is currently only supported "
2175 "during postcopy-active state");
2178 bool migration_is_blocked(Error
**errp
)
2180 if (qemu_savevm_state_blocked(errp
)) {
2184 if (migration_blockers
) {
2185 error_propagate(errp
, error_copy(migration_blockers
->data
));
2192 /* Returns true if continue to migrate, or false if error detected */
2193 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
2194 bool resume
, Error
**errp
)
2196 Error
*local_err
= NULL
;
2199 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2200 error_setg(errp
, "Cannot resume if there is no "
2201 "paused migration");
2206 * Postcopy recovery won't work well with release-ram
2207 * capability since release-ram will drop the page buffer as
2208 * long as the page is put into the send buffer. So if there
2209 * is a network failure happened, any page buffers that have
2210 * not yet reached the destination VM but have already been
2211 * sent from the source VM will be lost forever. Let's refuse
2212 * the client from resuming such a postcopy migration.
2213 * Luckily release-ram was designed to only be used when src
2214 * and destination VMs are on the same host, so it should be
2217 if (migrate_release_ram()) {
2218 error_setg(errp
, "Postcopy recovery cannot work "
2219 "when release-ram capability is set");
2223 /* This is a resume, skip init status */
2227 if (migration_is_running(s
->state
)) {
2228 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
2232 if (runstate_check(RUN_STATE_INMIGRATE
)) {
2233 error_setg(errp
, "Guest is waiting for an incoming migration");
2237 if (runstate_check(RUN_STATE_POSTMIGRATE
)) {
2238 error_setg(errp
, "Can't migrate the vm that was paused due to "
2239 "previous migration");
2243 if (migration_is_blocked(errp
)) {
2247 if (blk
|| blk_inc
) {
2248 if (migrate_colo_enabled()) {
2249 error_setg(errp
, "No disk migration is required in COLO mode");
2252 if (migrate_use_block() || migrate_use_block_incremental()) {
2253 error_setg(errp
, "Command options are incompatible with "
2254 "current migration capabilities");
2257 migrate_set_block_enabled(true, &local_err
);
2259 error_propagate(errp
, local_err
);
2262 s
->must_remove_block_options
= true;
2266 migrate_set_block_incremental(s
, true);
2271 * set ram_counters memory to zero for a
2274 memset(&ram_counters
, 0, sizeof(ram_counters
));
2279 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
2280 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
2281 bool has_resume
, bool resume
, Error
**errp
)
2283 Error
*local_err
= NULL
;
2284 MigrationState
*s
= migrate_get_current();
2285 const char *p
= NULL
;
2287 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
2288 has_resume
&& resume
, errp
)) {
2289 /* Error detected, put into errp */
2293 if (!(has_resume
&& resume
)) {
2294 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
2299 migrate_protocol_allow_multifd(false);
2300 if (strstart(uri
, "tcp:", &p
) ||
2301 strstart(uri
, "unix:", NULL
) ||
2302 strstart(uri
, "vsock:", NULL
)) {
2303 migrate_protocol_allow_multifd(true);
2304 socket_start_outgoing_migration(s
, p
? p
: uri
, &local_err
);
2306 } else if (strstart(uri
, "rdma:", &p
)) {
2307 rdma_start_outgoing_migration(s
, p
, &local_err
);
2309 } else if (strstart(uri
, "exec:", &p
)) {
2310 exec_start_outgoing_migration(s
, p
, &local_err
);
2311 } else if (strstart(uri
, "fd:", &p
)) {
2312 fd_start_outgoing_migration(s
, p
, &local_err
);
2314 if (!(has_resume
&& resume
)) {
2315 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2317 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
2318 "a valid migration protocol");
2319 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
2320 MIGRATION_STATUS_FAILED
);
2321 block_cleanup_parameters(s
);
2326 if (!(has_resume
&& resume
)) {
2327 yank_unregister_instance(MIGRATION_YANK_INSTANCE
);
2329 migrate_fd_error(s
, local_err
);
2330 error_propagate(errp
, local_err
);
2335 void qmp_migrate_cancel(Error
**errp
)
2340 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
2342 MigrationState
*s
= migrate_get_current();
2343 if (s
->state
!= state
) {
2344 error_setg(errp
, "Migration not in expected state: %s",
2345 MigrationStatus_str(s
->state
));
2348 qemu_sem_post(&s
->pause_sem
);
2351 bool migrate_release_ram(void)
2355 s
= migrate_get_current();
2357 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
2360 bool migrate_postcopy_ram(void)
2364 s
= migrate_get_current();
2366 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
2369 bool migrate_postcopy(void)
2371 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2374 bool migrate_auto_converge(void)
2378 s
= migrate_get_current();
2380 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
2383 bool migrate_zero_blocks(void)
2387 s
= migrate_get_current();
2389 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
2392 bool migrate_postcopy_blocktime(void)
2396 s
= migrate_get_current();
2398 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
2401 bool migrate_use_compression(void)
2405 s
= migrate_get_current();
2407 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
2410 int migrate_compress_level(void)
2414 s
= migrate_get_current();
2416 return s
->parameters
.compress_level
;
2419 int migrate_compress_threads(void)
2423 s
= migrate_get_current();
2425 return s
->parameters
.compress_threads
;
2428 int migrate_compress_wait_thread(void)
2432 s
= migrate_get_current();
2434 return s
->parameters
.compress_wait_thread
;
2437 int migrate_decompress_threads(void)
2441 s
= migrate_get_current();
2443 return s
->parameters
.decompress_threads
;
2446 bool migrate_dirty_bitmaps(void)
2450 s
= migrate_get_current();
2452 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
2455 bool migrate_ignore_shared(void)
2459 s
= migrate_get_current();
2461 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
];
2464 bool migrate_validate_uuid(void)
2468 s
= migrate_get_current();
2470 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_VALIDATE_UUID
];
2473 bool migrate_use_events(void)
2477 s
= migrate_get_current();
2479 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
2482 bool migrate_use_multifd(void)
2486 s
= migrate_get_current();
2488 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_MULTIFD
];
2491 bool migrate_pause_before_switchover(void)
2495 s
= migrate_get_current();
2497 return s
->enabled_capabilities
[
2498 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
2501 int migrate_multifd_channels(void)
2505 s
= migrate_get_current();
2507 return s
->parameters
.multifd_channels
;
2510 MultiFDCompression
migrate_multifd_compression(void)
2514 s
= migrate_get_current();
2516 return s
->parameters
.multifd_compression
;
2519 int migrate_multifd_zlib_level(void)
2523 s
= migrate_get_current();
2525 return s
->parameters
.multifd_zlib_level
;
2528 int migrate_multifd_zstd_level(void)
2532 s
= migrate_get_current();
2534 return s
->parameters
.multifd_zstd_level
;
2537 int migrate_use_xbzrle(void)
2541 s
= migrate_get_current();
2543 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
2546 uint64_t migrate_xbzrle_cache_size(void)
2550 s
= migrate_get_current();
2552 return s
->parameters
.xbzrle_cache_size
;
2555 static int64_t migrate_max_postcopy_bandwidth(void)
2559 s
= migrate_get_current();
2561 return s
->parameters
.max_postcopy_bandwidth
;
2564 bool migrate_use_block(void)
2568 s
= migrate_get_current();
2570 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
2573 bool migrate_use_return_path(void)
2577 s
= migrate_get_current();
2579 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
2582 bool migrate_use_block_incremental(void)
2586 s
= migrate_get_current();
2588 return s
->parameters
.block_incremental
;
2591 bool migrate_background_snapshot(void)
2595 s
= migrate_get_current();
2597 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
];
2600 /* migration thread support */
2602 * Something bad happened to the RP stream, mark an error
2603 * The caller shall print or trace something to indicate why
2605 static void mark_source_rp_bad(MigrationState
*s
)
2607 s
->rp_state
.error
= true;
2610 static struct rp_cmd_args
{
2611 ssize_t len
; /* -1 = variable */
2614 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
2615 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
2616 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
2617 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
2618 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
2619 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
2620 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
2621 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
2625 * Process a request for pages received on the return path,
2626 * We're allowed to send more than requested (e.g. to round to our page size)
2627 * and we don't need to send pages that have already been sent.
2629 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
2630 ram_addr_t start
, size_t len
)
2632 long our_host_ps
= qemu_real_host_page_size
;
2634 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
2637 * Since we currently insist on matching page sizes, just sanity check
2638 * we're being asked for whole host pages.
2640 if (start
& (our_host_ps
- 1) ||
2641 (len
& (our_host_ps
- 1))) {
2642 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2643 " len: %zd", __func__
, start
, len
);
2644 mark_source_rp_bad(ms
);
2648 if (ram_save_queue_pages(rbname
, start
, len
)) {
2649 mark_source_rp_bad(ms
);
2653 /* Return true to retry, false to quit */
2654 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
2656 trace_postcopy_pause_return_path();
2658 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
2660 trace_postcopy_pause_return_path_continued();
2665 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
2667 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
2670 error_report("%s: invalid block name '%s'", __func__
, block_name
);
2674 /* Fetch the received bitmap and refresh the dirty bitmap */
2675 return ram_dirty_bitmap_reload(s
, block
);
2678 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
2680 trace_source_return_path_thread_resume_ack(value
);
2682 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
2683 error_report("%s: illegal resume_ack value %"PRIu32
,
2688 /* Now both sides are active. */
2689 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2690 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2692 /* Notify send thread that time to continue send pages */
2693 qemu_sem_post(&s
->rp_state
.rp_sem
);
2698 /* Release ms->rp_state.from_dst_file in a safe way */
2699 static void migration_release_from_dst_file(MigrationState
*ms
)
2703 WITH_QEMU_LOCK_GUARD(&ms
->qemu_file_lock
) {
2705 * Reset the from_dst_file pointer first before releasing it, as we
2706 * can't block within lock section
2708 file
= ms
->rp_state
.from_dst_file
;
2709 ms
->rp_state
.from_dst_file
= NULL
;
2716 * Handles messages sent on the return path towards the source VM
2719 static void *source_return_path_thread(void *opaque
)
2721 MigrationState
*ms
= opaque
;
2722 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
2723 uint16_t header_len
, header_type
;
2725 uint32_t tmp32
, sibling_error
;
2726 ram_addr_t start
= 0; /* =0 to silence warning */
2727 size_t len
= 0, expected_len
;
2730 trace_source_return_path_thread_entry();
2731 rcu_register_thread();
2734 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
2735 migration_is_setup_or_active(ms
->state
)) {
2736 trace_source_return_path_thread_loop_top();
2737 header_type
= qemu_get_be16(rp
);
2738 header_len
= qemu_get_be16(rp
);
2740 if (qemu_file_get_error(rp
)) {
2741 mark_source_rp_bad(ms
);
2745 if (header_type
>= MIG_RP_MSG_MAX
||
2746 header_type
== MIG_RP_MSG_INVALID
) {
2747 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2748 header_type
, header_len
);
2749 mark_source_rp_bad(ms
);
2753 if ((rp_cmd_args
[header_type
].len
!= -1 &&
2754 header_len
!= rp_cmd_args
[header_type
].len
) ||
2755 header_len
> sizeof(buf
)) {
2756 error_report("RP: Received '%s' message (0x%04x) with"
2757 "incorrect length %d expecting %zu",
2758 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
2759 (size_t)rp_cmd_args
[header_type
].len
);
2760 mark_source_rp_bad(ms
);
2764 /* We know we've got a valid header by this point */
2765 res
= qemu_get_buffer(rp
, buf
, header_len
);
2766 if (res
!= header_len
) {
2767 error_report("RP: Failed reading data for message 0x%04x"
2768 " read %d expected %d",
2769 header_type
, res
, header_len
);
2770 mark_source_rp_bad(ms
);
2774 /* OK, we have the message and the data */
2775 switch (header_type
) {
2776 case MIG_RP_MSG_SHUT
:
2777 sibling_error
= ldl_be_p(buf
);
2778 trace_source_return_path_thread_shut(sibling_error
);
2779 if (sibling_error
) {
2780 error_report("RP: Sibling indicated error %d", sibling_error
);
2781 mark_source_rp_bad(ms
);
2784 * We'll let the main thread deal with closing the RP
2785 * we could do a shutdown(2) on it, but we're the only user
2786 * anyway, so there's nothing gained.
2790 case MIG_RP_MSG_PONG
:
2791 tmp32
= ldl_be_p(buf
);
2792 trace_source_return_path_thread_pong(tmp32
);
2795 case MIG_RP_MSG_REQ_PAGES
:
2796 start
= ldq_be_p(buf
);
2797 len
= ldl_be_p(buf
+ 8);
2798 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
2801 case MIG_RP_MSG_REQ_PAGES_ID
:
2802 expected_len
= 12 + 1; /* header + termination */
2804 if (header_len
>= expected_len
) {
2805 start
= ldq_be_p(buf
);
2806 len
= ldl_be_p(buf
+ 8);
2807 /* Now we expect an idstr */
2808 tmp32
= buf
[12]; /* Length of the following idstr */
2809 buf
[13 + tmp32
] = '\0';
2810 expected_len
+= tmp32
;
2812 if (header_len
!= expected_len
) {
2813 error_report("RP: Req_Page_id with length %d expecting %zd",
2814 header_len
, expected_len
);
2815 mark_source_rp_bad(ms
);
2818 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
2821 case MIG_RP_MSG_RECV_BITMAP
:
2822 if (header_len
< 1) {
2823 error_report("%s: missing block name", __func__
);
2824 mark_source_rp_bad(ms
);
2827 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2828 buf
[buf
[0] + 1] = '\0';
2829 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
2830 mark_source_rp_bad(ms
);
2835 case MIG_RP_MSG_RESUME_ACK
:
2836 tmp32
= ldl_be_p(buf
);
2837 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
2838 mark_source_rp_bad(ms
);
2849 res
= qemu_file_get_error(rp
);
2851 if (res
== -EIO
&& migration_in_postcopy()) {
2853 * Maybe there is something we can do: it looks like a
2854 * network down issue, and we pause for a recovery.
2856 migration_release_from_dst_file(ms
);
2858 if (postcopy_pause_return_path_thread(ms
)) {
2860 * Reload rp, reset the rest. Referencing it is safe since
2861 * it's reset only by us above, or when migration completes
2863 rp
= ms
->rp_state
.from_dst_file
;
2864 ms
->rp_state
.error
= false;
2869 trace_source_return_path_thread_bad_end();
2870 mark_source_rp_bad(ms
);
2873 trace_source_return_path_thread_end();
2874 migration_release_from_dst_file(ms
);
2875 rcu_unregister_thread();
2879 static int open_return_path_on_source(MigrationState
*ms
,
2882 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
2883 if (!ms
->rp_state
.from_dst_file
) {
2887 trace_open_return_path_on_source();
2889 if (!create_thread
) {
2894 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
2895 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
2896 ms
->rp_state
.rp_thread_created
= true;
2898 trace_open_return_path_on_source_continue();
2903 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2904 static int await_return_path_close_on_source(MigrationState
*ms
)
2907 * If this is a normal exit then the destination will send a SHUT and the
2908 * rp_thread will exit, however if there's an error we need to cause
2911 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
2913 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2914 * waiting for the destination.
2916 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
2917 mark_source_rp_bad(ms
);
2919 trace_await_return_path_close_on_source_joining();
2920 qemu_thread_join(&ms
->rp_state
.rp_thread
);
2921 ms
->rp_state
.rp_thread_created
= false;
2922 trace_await_return_path_close_on_source_close();
2923 return ms
->rp_state
.error
;
2927 * Switch from normal iteration to postcopy
2928 * Returns non-0 on error
2930 static int postcopy_start(MigrationState
*ms
)
2933 QIOChannelBuffer
*bioc
;
2935 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2936 int64_t bandwidth
= migrate_max_postcopy_bandwidth();
2937 bool restart_block
= false;
2938 int cur_state
= MIGRATION_STATUS_ACTIVE
;
2939 if (!migrate_pause_before_switchover()) {
2940 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
2941 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2944 trace_postcopy_start();
2945 qemu_mutex_lock_iothread();
2946 trace_postcopy_start_set_run();
2948 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2949 global_state_store();
2950 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2955 ret
= migration_maybe_pause(ms
, &cur_state
,
2956 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2961 ret
= bdrv_inactivate_all();
2965 restart_block
= true;
2968 * Cause any non-postcopiable, but iterative devices to
2969 * send out their final data.
2971 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2974 * in Finish migrate and with the io-lock held everything should
2975 * be quiet, but we've potentially still got dirty pages and we
2976 * need to tell the destination to throw any pages it's already received
2979 if (migrate_postcopy_ram()) {
2980 if (ram_postcopy_send_discard_bitmap(ms
)) {
2981 error_report("postcopy send discard bitmap failed");
2987 * send rest of state - note things that are doing postcopy
2988 * will notice we're in POSTCOPY_ACTIVE and not actually
2989 * wrap their state up here
2991 /* 0 max-postcopy-bandwidth means unlimited */
2993 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
2995 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
2997 if (migrate_postcopy_ram()) {
2998 /* Ping just for debugging, helps line traces up */
2999 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
3003 * While loading the device state we may trigger page transfer
3004 * requests and the fd must be free to process those, and thus
3005 * the destination must read the whole device state off the fd before
3006 * it starts processing it. Unfortunately the ad-hoc migration format
3007 * doesn't allow the destination to know the size to read without fully
3008 * parsing it through each devices load-state code (especially the open
3009 * coded devices that use get/put).
3010 * So we wrap the device state up in a package with a length at the start;
3011 * to do this we use a qemu_buf to hold the whole of the device state.
3013 bioc
= qio_channel_buffer_new(4096);
3014 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
3015 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
3016 object_unref(OBJECT(bioc
));
3019 * Make sure the receiver can get incoming pages before we send the rest
3022 qemu_savevm_send_postcopy_listen(fb
);
3024 qemu_savevm_state_complete_precopy(fb
, false, false);
3025 if (migrate_postcopy_ram()) {
3026 qemu_savevm_send_ping(fb
, 3);
3029 qemu_savevm_send_postcopy_run(fb
);
3031 /* <><> end of stuff going into the package */
3033 /* Last point of recovery; as soon as we send the package the destination
3034 * can open devices and potentially start running.
3035 * Lets just check again we've not got any errors.
3037 ret
= qemu_file_get_error(ms
->to_dst_file
);
3039 error_report("postcopy_start: Migration stream errored (pre package)");
3043 restart_block
= false;
3045 /* Now send that blob */
3046 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
3051 /* Send a notify to give a chance for anything that needs to happen
3052 * at the transition to postcopy and after the device state; in particular
3053 * spice needs to trigger a transition now
3055 ms
->postcopy_after_devices
= true;
3056 notifier_list_notify(&migration_state_notifiers
, ms
);
3058 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
3060 qemu_mutex_unlock_iothread();
3062 if (migrate_postcopy_ram()) {
3064 * Although this ping is just for debug, it could potentially be
3065 * used for getting a better measurement of downtime at the source.
3067 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
3070 if (migrate_release_ram()) {
3071 ram_postcopy_migrated_memory_release(ms
);
3074 ret
= qemu_file_get_error(ms
->to_dst_file
);
3076 error_report("postcopy_start: Migration stream errored");
3077 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
3078 MIGRATION_STATUS_FAILED
);
3086 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
3087 MIGRATION_STATUS_FAILED
);
3088 if (restart_block
) {
3089 /* A failure happened early enough that we know the destination hasn't
3090 * accessed block devices, so we're safe to recover.
3092 Error
*local_err
= NULL
;
3094 bdrv_invalidate_cache_all(&local_err
);
3096 error_report_err(local_err
);
3099 qemu_mutex_unlock_iothread();
3104 * migration_maybe_pause: Pause if required to by
3105 * migrate_pause_before_switchover called with the iothread locked
3106 * Returns: 0 on success
3108 static int migration_maybe_pause(MigrationState
*s
,
3109 int *current_active_state
,
3112 if (!migrate_pause_before_switchover()) {
3116 /* Since leaving this state is not atomic with posting the semaphore
3117 * it's possible that someone could have issued multiple migrate_continue
3118 * and the semaphore is incorrectly positive at this point;
3119 * the docs say it's undefined to reinit a semaphore that's already
3120 * init'd, so use timedwait to eat up any existing posts.
3122 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
3123 /* This block intentionally left blank */
3127 * If the migration is cancelled when it is in the completion phase,
3128 * the migration state is set to MIGRATION_STATUS_CANCELLING.
3129 * So we don't need to wait a semaphore, otherwise we would always
3130 * wait for the 'pause_sem' semaphore.
3132 if (s
->state
!= MIGRATION_STATUS_CANCELLING
) {
3133 qemu_mutex_unlock_iothread();
3134 migrate_set_state(&s
->state
, *current_active_state
,
3135 MIGRATION_STATUS_PRE_SWITCHOVER
);
3136 qemu_sem_wait(&s
->pause_sem
);
3137 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
3139 *current_active_state
= new_state
;
3140 qemu_mutex_lock_iothread();
3143 return s
->state
== new_state
? 0 : -EINVAL
;
3147 * migration_completion: Used by migration_thread when there's not much left.
3148 * The caller 'breaks' the loop when this returns.
3150 * @s: Current migration state
3152 static void migration_completion(MigrationState
*s
)
3155 int current_active_state
= s
->state
;
3157 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
3158 qemu_mutex_lock_iothread();
3159 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3160 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3161 s
->vm_was_running
= runstate_is_running();
3162 ret
= global_state_store();
3165 bool inactivate
= !migrate_colo_enabled();
3166 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
3167 trace_migration_completion_vm_stop(ret
);
3169 ret
= migration_maybe_pause(s
, ¤t_active_state
,
3170 MIGRATION_STATUS_DEVICE
);
3173 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
3174 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
3177 if (inactivate
&& ret
>= 0) {
3178 s
->block_inactive
= true;
3181 qemu_mutex_unlock_iothread();
3186 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3187 trace_migration_completion_postcopy_end();
3189 qemu_mutex_lock_iothread();
3190 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
3191 qemu_mutex_unlock_iothread();
3193 trace_migration_completion_postcopy_end_after_complete();
3194 } else if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
3199 * If rp was opened we must clean up the thread before
3200 * cleaning everything else up (since if there are no failures
3201 * it will wait for the destination to send it's status in
3204 if (s
->rp_state
.rp_thread_created
) {
3206 trace_migration_return_path_end_before();
3207 rp_error
= await_return_path_close_on_source(s
);
3208 trace_migration_return_path_end_after(rp_error
);
3210 goto fail_invalidate
;
3214 if (qemu_file_get_error(s
->to_dst_file
)) {
3215 trace_migration_completion_file_err();
3216 goto fail_invalidate
;
3219 if (!migrate_colo_enabled()) {
3220 migrate_set_state(&s
->state
, current_active_state
,
3221 MIGRATION_STATUS_COMPLETED
);
3227 /* If not doing postcopy, vm_start() will be called: let's regain
3228 * control on images.
3230 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
3231 s
->state
== MIGRATION_STATUS_DEVICE
) {
3232 Error
*local_err
= NULL
;
3234 qemu_mutex_lock_iothread();
3235 bdrv_invalidate_cache_all(&local_err
);
3237 error_report_err(local_err
);
3239 s
->block_inactive
= false;
3241 qemu_mutex_unlock_iothread();
3245 migrate_set_state(&s
->state
, current_active_state
,
3246 MIGRATION_STATUS_FAILED
);
3250 * bg_migration_completion: Used by bg_migration_thread when after all the
3251 * RAM has been saved. The caller 'breaks' the loop when this returns.
3253 * @s: Current migration state
3255 static void bg_migration_completion(MigrationState
*s
)
3257 int current_active_state
= s
->state
;
3260 * Stop tracking RAM writes - un-protect memory, un-register UFFD
3261 * memory ranges, flush kernel wait queues and wake up threads
3262 * waiting for write fault to be resolved.
3264 ram_write_tracking_stop();
3266 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
3268 * By this moment we have RAM content saved into the migration stream.
3269 * The next step is to flush the non-RAM content (device state)
3270 * right after the ram content. The device state has been stored into
3271 * the temporary buffer before RAM saving started.
3273 qemu_put_buffer(s
->to_dst_file
, s
->bioc
->data
, s
->bioc
->usage
);
3274 qemu_fflush(s
->to_dst_file
);
3275 } else if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
3279 if (qemu_file_get_error(s
->to_dst_file
)) {
3280 trace_migration_completion_file_err();
3284 migrate_set_state(&s
->state
, current_active_state
,
3285 MIGRATION_STATUS_COMPLETED
);
3289 migrate_set_state(&s
->state
, current_active_state
,
3290 MIGRATION_STATUS_FAILED
);
3293 bool migrate_colo_enabled(void)
3295 MigrationState
*s
= migrate_get_current();
3296 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
3299 typedef enum MigThrError
{
3300 /* No error detected */
3301 MIG_THR_ERR_NONE
= 0,
3302 /* Detected error, but resumed successfully */
3303 MIG_THR_ERR_RECOVERED
= 1,
3304 /* Detected fatal error, need to exit */
3305 MIG_THR_ERR_FATAL
= 2,
3308 static int postcopy_resume_handshake(MigrationState
*s
)
3310 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
3312 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3313 qemu_sem_wait(&s
->rp_state
.rp_sem
);
3316 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3323 /* Return zero if success, or <0 for error */
3324 static int postcopy_do_resume(MigrationState
*s
)
3329 * Call all the resume_prepare() hooks, so that modules can be
3330 * ready for the migration resume.
3332 ret
= qemu_savevm_state_resume_prepare(s
);
3334 error_report("%s: resume_prepare() failure detected: %d",
3340 * Last handshake with destination on the resume (destination will
3341 * switch to postcopy-active afterwards)
3343 ret
= postcopy_resume_handshake(s
);
3345 error_report("%s: handshake failed: %d", __func__
, ret
);
3353 * We don't return until we are in a safe state to continue current
3354 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
3355 * MIG_THR_ERR_FATAL if unrecovery failure happened.
3357 static MigThrError
postcopy_pause(MigrationState
*s
)
3359 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3365 * Current channel is possibly broken. Release it. Note that this is
3366 * guaranteed even without lock because to_dst_file should only be
3367 * modified by the migration thread. That also guarantees that the
3368 * unregister of yank is safe too without the lock. It should be safe
3369 * even to be within the qemu_file_lock, but we didn't do that to avoid
3370 * taking more mutex (yank_lock) within qemu_file_lock. TL;DR: we make
3371 * the qemu_file_lock critical section as small as possible.
3373 assert(s
->to_dst_file
);
3374 migration_ioc_unregister_yank_from_file(s
->to_dst_file
);
3375 qemu_mutex_lock(&s
->qemu_file_lock
);
3376 file
= s
->to_dst_file
;
3377 s
->to_dst_file
= NULL
;
3378 qemu_mutex_unlock(&s
->qemu_file_lock
);
3380 qemu_file_shutdown(file
);
3383 migrate_set_state(&s
->state
, s
->state
,
3384 MIGRATION_STATUS_POSTCOPY_PAUSED
);
3386 error_report("Detected IO failure for postcopy. "
3387 "Migration paused.");
3390 * We wait until things fixed up. Then someone will setup the
3391 * status back for us.
3393 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
3394 qemu_sem_wait(&s
->postcopy_pause_sem
);
3397 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
3398 /* Woken up by a recover procedure. Give it a shot */
3401 * Firstly, let's wake up the return path now, with a new
3402 * return path channel.
3404 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
3406 /* Do the resume logic */
3407 if (postcopy_do_resume(s
) == 0) {
3408 /* Let's continue! */
3409 trace_postcopy_pause_continued();
3410 return MIG_THR_ERR_RECOVERED
;
3413 * Something wrong happened during the recovery, let's
3414 * pause again. Pause is always better than throwing
3420 /* This is not right... Time to quit. */
3421 return MIG_THR_ERR_FATAL
;
3426 static MigThrError
migration_detect_error(MigrationState
*s
)
3429 int state
= s
->state
;
3430 Error
*local_error
= NULL
;
3432 if (state
== MIGRATION_STATUS_CANCELLING
||
3433 state
== MIGRATION_STATUS_CANCELLED
) {
3434 /* End the migration, but don't set the state to failed */
3435 return MIG_THR_ERR_FATAL
;
3438 /* Try to detect any file errors */
3439 ret
= qemu_file_get_error_obj(s
->to_dst_file
, &local_error
);
3441 /* Everything is fine */
3442 assert(!local_error
);
3443 return MIG_THR_ERR_NONE
;
3447 migrate_set_error(s
, local_error
);
3448 error_free(local_error
);
3451 if (state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
== -EIO
) {
3453 * For postcopy, we allow the network to be down for a
3454 * while. After that, it can be continued by a
3457 return postcopy_pause(s
);
3460 * For precopy (or postcopy with error outside IO), we fail
3463 migrate_set_state(&s
->state
, state
, MIGRATION_STATUS_FAILED
);
3464 trace_migration_thread_file_err();
3466 /* Time to stop the migration, now. */
3467 return MIG_THR_ERR_FATAL
;
3471 /* How many bytes have we transferred since the beginning of the migration */
3472 static uint64_t migration_total_bytes(MigrationState
*s
)
3474 return qemu_ftell(s
->to_dst_file
) + ram_counters
.multifd_bytes
;
3477 static void migration_calculate_complete(MigrationState
*s
)
3479 uint64_t bytes
= migration_total_bytes(s
);
3480 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3481 int64_t transfer_time
;
3483 s
->total_time
= end_time
- s
->start_time
;
3486 * It's still not set, so we are precopy migration. For
3487 * postcopy, downtime is calculated during postcopy_start().
3489 s
->downtime
= end_time
- s
->downtime_start
;
3492 transfer_time
= s
->total_time
- s
->setup_time
;
3493 if (transfer_time
) {
3494 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
3498 static void update_iteration_initial_status(MigrationState
*s
)
3501 * Update these three fields at the same time to avoid mismatch info lead
3502 * wrong speed calculation.
3504 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3505 s
->iteration_initial_bytes
= migration_total_bytes(s
);
3506 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
3509 static void migration_update_counters(MigrationState
*s
,
3510 int64_t current_time
)
3512 uint64_t transferred
, transferred_pages
, time_spent
;
3513 uint64_t current_bytes
; /* bytes transferred since the beginning */
3516 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
3520 current_bytes
= migration_total_bytes(s
);
3521 transferred
= current_bytes
- s
->iteration_initial_bytes
;
3522 time_spent
= current_time
- s
->iteration_start_time
;
3523 bandwidth
= (double)transferred
/ time_spent
;
3524 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
3526 s
->mbps
= (((double) transferred
* 8.0) /
3527 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
3529 transferred_pages
= ram_get_total_transferred_pages() -
3530 s
->iteration_initial_pages
;
3531 s
->pages_per_second
= (double) transferred_pages
/
3532 (((double) time_spent
/ 1000.0));
3535 * if we haven't sent anything, we don't want to
3536 * recalculate. 10000 is a small enough number for our purposes
3538 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
3539 s
->expected_downtime
= ram_counters
.remaining
/ bandwidth
;
3542 qemu_file_reset_rate_limit(s
->to_dst_file
);
3544 update_iteration_initial_status(s
);
3546 trace_migrate_transferred(transferred
, time_spent
,
3547 bandwidth
, s
->threshold_size
);
3550 /* Migration thread iteration status */
3552 MIG_ITERATE_RESUME
, /* Resume current iteration */
3553 MIG_ITERATE_SKIP
, /* Skip current iteration */
3554 MIG_ITERATE_BREAK
, /* Break the loop */
3558 * Return true if continue to the next iteration directly, false
3561 static MigIterateState
migration_iteration_run(MigrationState
*s
)
3563 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
3564 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
3566 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
3567 &pend_compat
, &pend_post
);
3568 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
3570 trace_migrate_pending(pending_size
, s
->threshold_size
,
3571 pend_pre
, pend_compat
, pend_post
);
3573 if (pending_size
&& pending_size
>= s
->threshold_size
) {
3574 /* Still a significant amount to transfer */
3575 if (!in_postcopy
&& pend_pre
<= s
->threshold_size
&&
3576 qatomic_read(&s
->start_postcopy
)) {
3577 if (postcopy_start(s
)) {
3578 error_report("%s: postcopy failed to start", __func__
);
3580 return MIG_ITERATE_SKIP
;
3582 /* Just another iteration step */
3583 qemu_savevm_state_iterate(s
->to_dst_file
, in_postcopy
);
3585 trace_migration_thread_low_pending(pending_size
);
3586 migration_completion(s
);
3587 return MIG_ITERATE_BREAK
;
3590 return MIG_ITERATE_RESUME
;
3593 static void migration_iteration_finish(MigrationState
*s
)
3595 /* If we enabled cpu throttling for auto-converge, turn it off. */
3596 cpu_throttle_stop();
3598 qemu_mutex_lock_iothread();
3600 case MIGRATION_STATUS_COMPLETED
:
3601 migration_calculate_complete(s
);
3602 runstate_set(RUN_STATE_POSTMIGRATE
);
3605 case MIGRATION_STATUS_ACTIVE
:
3607 * We should really assert here, but since it's during
3608 * migration, let's try to reduce the usage of assertions.
3610 if (!migrate_colo_enabled()) {
3611 error_report("%s: critical error: calling COLO code without "
3612 "COLO enabled", __func__
);
3614 migrate_start_colo_process(s
);
3616 * Fixme: we will run VM in COLO no matter its old running state.
3617 * After exited COLO, we will keep running.
3619 s
->vm_was_running
= true;
3621 case MIGRATION_STATUS_FAILED
:
3622 case MIGRATION_STATUS_CANCELLED
:
3623 case MIGRATION_STATUS_CANCELLING
:
3624 if (s
->vm_was_running
) {
3627 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
3628 runstate_set(RUN_STATE_POSTMIGRATE
);
3634 /* Should not reach here, but if so, forgive the VM. */
3635 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3638 migrate_fd_cleanup_schedule(s
);
3639 qemu_mutex_unlock_iothread();
3642 static void bg_migration_iteration_finish(MigrationState
*s
)
3644 qemu_mutex_lock_iothread();
3646 case MIGRATION_STATUS_COMPLETED
:
3647 migration_calculate_complete(s
);
3650 case MIGRATION_STATUS_ACTIVE
:
3651 case MIGRATION_STATUS_FAILED
:
3652 case MIGRATION_STATUS_CANCELLED
:
3653 case MIGRATION_STATUS_CANCELLING
:
3657 /* Should not reach here, but if so, forgive the VM. */
3658 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3662 migrate_fd_cleanup_schedule(s
);
3663 qemu_mutex_unlock_iothread();
3667 * Return true if continue to the next iteration directly, false
3670 static MigIterateState
bg_migration_iteration_run(MigrationState
*s
)
3674 res
= qemu_savevm_state_iterate(s
->to_dst_file
, false);
3676 bg_migration_completion(s
);
3677 return MIG_ITERATE_BREAK
;
3680 return MIG_ITERATE_RESUME
;
3683 void migration_make_urgent_request(void)
3685 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
3688 void migration_consume_urgent_request(void)
3690 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
3693 /* Returns true if the rate limiting was broken by an urgent request */
3694 bool migration_rate_limit(void)
3696 int64_t now
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3697 MigrationState
*s
= migrate_get_current();
3699 bool urgent
= false;
3700 migration_update_counters(s
, now
);
3701 if (qemu_file_rate_limit(s
->to_dst_file
)) {
3703 if (qemu_file_get_error(s
->to_dst_file
)) {
3707 * Wait for a delay to do rate limiting OR
3708 * something urgent to post the semaphore.
3710 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- now
;
3711 trace_migration_rate_limit_pre(ms
);
3712 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
3714 * We were woken by one or more urgent things but
3715 * the timedwait will have consumed one of them.
3716 * The service routine for the urgent wake will dec
3717 * the semaphore itself for each item it consumes,
3718 * so add this one we just eat back.
3720 qemu_sem_post(&s
->rate_limit_sem
);
3723 trace_migration_rate_limit_post(urgent
);
3729 * if failover devices are present, wait they are completely
3733 static void qemu_savevm_wait_unplug(MigrationState
*s
, int old_state
,
3736 if (qemu_savevm_state_guest_unplug_pending()) {
3737 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_WAIT_UNPLUG
);
3739 while (s
->state
== MIGRATION_STATUS_WAIT_UNPLUG
&&
3740 qemu_savevm_state_guest_unplug_pending()) {
3741 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
3743 if (s
->state
!= MIGRATION_STATUS_WAIT_UNPLUG
) {
3744 int timeout
= 120; /* 30 seconds */
3746 * migration has been canceled
3747 * but as we have started an unplug we must wait the end
3748 * to be able to plug back the card
3750 while (timeout
-- && qemu_savevm_state_guest_unplug_pending()) {
3751 qemu_sem_timedwait(&s
->wait_unplug_sem
, 250);
3753 if (qemu_savevm_state_guest_unplug_pending()) {
3754 warn_report("migration: partially unplugged device on "
3759 migrate_set_state(&s
->state
, MIGRATION_STATUS_WAIT_UNPLUG
, new_state
);
3761 migrate_set_state(&s
->state
, old_state
, new_state
);
3766 * Master migration thread on the source VM.
3767 * It drives the migration and pumps the data down the outgoing channel.
3769 static void *migration_thread(void *opaque
)
3771 MigrationState
*s
= opaque
;
3772 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3773 MigThrError thr_error
;
3774 bool urgent
= false;
3776 rcu_register_thread();
3778 object_ref(OBJECT(s
));
3779 update_iteration_initial_status(s
);
3781 qemu_savevm_state_header(s
->to_dst_file
);
3784 * If we opened the return path, we need to make sure dst has it
3787 if (s
->rp_state
.rp_thread_created
) {
3788 /* Now tell the dest that it should open its end so it can reply */
3789 qemu_savevm_send_open_return_path(s
->to_dst_file
);
3791 /* And do a ping that will make stuff easier to debug */
3792 qemu_savevm_send_ping(s
->to_dst_file
, 1);
3795 if (migrate_postcopy()) {
3797 * Tell the destination that we *might* want to do postcopy later;
3798 * if the other end can't do postcopy it should fail now, nice and
3801 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
3804 if (migrate_colo_enabled()) {
3805 /* Notify migration destination that we enable COLO */
3806 qemu_savevm_send_colo_enable(s
->to_dst_file
);
3809 qemu_savevm_state_setup(s
->to_dst_file
);
3811 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
3812 MIGRATION_STATUS_ACTIVE
);
3814 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3816 trace_migration_thread_setup_complete();
3818 while (migration_is_active(s
)) {
3819 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
3820 MigIterateState iter_state
= migration_iteration_run(s
);
3821 if (iter_state
== MIG_ITERATE_SKIP
) {
3823 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3829 * Try to detect any kind of failures, and see whether we
3830 * should stop the migration now.
3832 thr_error
= migration_detect_error(s
);
3833 if (thr_error
== MIG_THR_ERR_FATAL
) {
3834 /* Stop migration */
3836 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
3838 * Just recovered from a e.g. network failure, reset all
3839 * the local variables. This is important to avoid
3840 * breaking transferred_bytes and bandwidth calculation
3842 update_iteration_initial_status(s
);
3845 urgent
= migration_rate_limit();
3848 trace_migration_thread_after_loop();
3849 migration_iteration_finish(s
);
3850 object_unref(OBJECT(s
));
3851 rcu_unregister_thread();
3855 static void bg_migration_vm_start_bh(void *opaque
)
3857 MigrationState
*s
= opaque
;
3859 qemu_bh_delete(s
->vm_start_bh
);
3860 s
->vm_start_bh
= NULL
;
3863 s
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - s
->downtime_start
;
3867 * Background snapshot thread, based on live migration code.
3868 * This is an alternative implementation of live migration mechanism
3869 * introduced specifically to support background snapshots.
3871 * It takes advantage of userfault_fd write protection mechanism introduced
3872 * in v5.7 kernel. Compared to existing dirty page logging migration much
3873 * lesser stream traffic is produced resulting in smaller snapshot images,
3874 * simply cause of no page duplicates can get into the stream.
3876 * Another key point is that generated vmstate stream reflects machine state
3877 * 'frozen' at the beginning of snapshot creation compared to dirty page logging
3878 * mechanism, which effectively results in that saved snapshot is the state of VM
3879 * at the end of the process.
3881 static void *bg_migration_thread(void *opaque
)
3883 MigrationState
*s
= opaque
;
3884 int64_t setup_start
;
3885 MigThrError thr_error
;
3887 bool early_fail
= true;
3889 rcu_register_thread();
3890 object_ref(OBJECT(s
));
3892 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
3894 setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3896 * We want to save vmstate for the moment when migration has been
3897 * initiated but also we want to save RAM content while VM is running.
3898 * The RAM content should appear first in the vmstate. So, we first
3899 * stash the non-RAM part of the vmstate to the temporary buffer,
3900 * then write RAM part of the vmstate to the migration stream
3901 * with vCPUs running and, finally, write stashed non-RAM part of
3902 * the vmstate from the buffer to the migration stream.
3904 s
->bioc
= qio_channel_buffer_new(512 * 1024);
3905 qio_channel_set_name(QIO_CHANNEL(s
->bioc
), "vmstate-buffer");
3906 fb
= qemu_fopen_channel_output(QIO_CHANNEL(s
->bioc
));
3907 object_unref(OBJECT(s
->bioc
));
3909 update_iteration_initial_status(s
);
3912 * Prepare for tracking memory writes with UFFD-WP - populate
3913 * RAM pages before protecting.
3916 ram_write_tracking_prepare();
3919 qemu_savevm_state_header(s
->to_dst_file
);
3920 qemu_savevm_state_setup(s
->to_dst_file
);
3922 qemu_savevm_wait_unplug(s
, MIGRATION_STATUS_SETUP
,
3923 MIGRATION_STATUS_ACTIVE
);
3925 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3927 trace_migration_thread_setup_complete();
3928 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3930 qemu_mutex_lock_iothread();
3933 * If VM is currently in suspended state, then, to make a valid runstate
3934 * transition in vm_stop_force_state() we need to wakeup it up.
3936 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
3937 s
->vm_was_running
= runstate_is_running();
3939 if (global_state_store()) {
3942 /* Forcibly stop VM before saving state of vCPUs and devices */
3943 if (vm_stop_force_state(RUN_STATE_PAUSED
)) {
3947 * Put vCPUs in sync with shadow context structures, then
3948 * save their state to channel-buffer along with devices.
3950 cpu_synchronize_all_states();
3951 if (qemu_savevm_state_complete_precopy_non_iterable(fb
, false, false)) {
3955 * Since we are going to get non-iterable state data directly
3956 * from s->bioc->data, explicit flush is needed here.
3960 /* Now initialize UFFD context and start tracking RAM writes */
3961 if (ram_write_tracking_start()) {
3967 * Start VM from BH handler to avoid write-fault lock here.
3968 * UFFD-WP protection for the whole RAM is already enabled so
3969 * calling VM state change notifiers from vm_start() would initiate
3970 * writes to virtio VQs memory which is in write-protected region.
3972 s
->vm_start_bh
= qemu_bh_new(bg_migration_vm_start_bh
, s
);
3973 qemu_bh_schedule(s
->vm_start_bh
);
3975 qemu_mutex_unlock_iothread();
3977 while (migration_is_active(s
)) {
3978 MigIterateState iter_state
= bg_migration_iteration_run(s
);
3979 if (iter_state
== MIG_ITERATE_SKIP
) {
3981 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3986 * Try to detect any kind of failures, and see whether we
3987 * should stop the migration now.
3989 thr_error
= migration_detect_error(s
);
3990 if (thr_error
== MIG_THR_ERR_FATAL
) {
3991 /* Stop migration */
3995 migration_update_counters(s
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
));
3998 trace_migration_thread_after_loop();
4002 migrate_set_state(&s
->state
, MIGRATION_STATUS_ACTIVE
,
4003 MIGRATION_STATUS_FAILED
);
4004 qemu_mutex_unlock_iothread();
4007 bg_migration_iteration_finish(s
);
4010 object_unref(OBJECT(s
));
4011 rcu_unregister_thread();
4016 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
4018 Error
*local_err
= NULL
;
4020 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
4023 * If there's a previous error, free it and prepare for another one.
4024 * Meanwhile if migration completes successfully, there won't have an error
4025 * dumped when calling migrate_fd_cleanup().
4027 migrate_error_free(s
);
4029 s
->expected_downtime
= s
->parameters
.downtime_limit
;
4031 assert(s
->cleanup_bh
);
4033 assert(!s
->cleanup_bh
);
4034 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup_bh
, s
);
4037 migrate_fd_error(s
, error_in
);
4040 * Don't do cleanup for resume if channel is invalid, but only dump
4041 * the error. We wait for another channel connect from the user.
4042 * The error_report still gives HMP user a hint on what failed.
4043 * It's normally done in migrate_fd_cleanup(), but call it here
4046 error_report_err(error_copy(s
->error
));
4048 migrate_fd_cleanup(s
);
4054 /* This is a resumed migration */
4055 rate_limit
= s
->parameters
.max_postcopy_bandwidth
/
4058 /* This is a fresh new migration */
4059 rate_limit
= s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
;
4061 /* Notify before starting migration thread */
4062 notifier_list_notify(&migration_state_notifiers
, s
);
4065 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
4066 qemu_file_set_blocking(s
->to_dst_file
, true);
4069 * Open the return path. For postcopy, it is used exclusively. For
4070 * precopy, only if user specified "return-path" capability would
4071 * QEMU uses the return path.
4073 if (migrate_postcopy_ram() || migrate_use_return_path()) {
4074 if (open_return_path_on_source(s
, !resume
)) {
4075 error_report("Unable to open return-path for postcopy");
4076 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
4077 migrate_fd_cleanup(s
);
4083 /* Wakeup the main migration thread to do the recovery */
4084 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
4085 MIGRATION_STATUS_POSTCOPY_RECOVER
);
4086 qemu_sem_post(&s
->postcopy_pause_sem
);
4090 if (multifd_save_setup(&local_err
) != 0) {
4091 error_report_err(local_err
);
4092 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
4093 MIGRATION_STATUS_FAILED
);
4094 migrate_fd_cleanup(s
);
4098 if (migrate_background_snapshot()) {
4099 qemu_thread_create(&s
->thread
, "bg_snapshot",
4100 bg_migration_thread
, s
, QEMU_THREAD_JOINABLE
);
4102 qemu_thread_create(&s
->thread
, "live_migration",
4103 migration_thread
, s
, QEMU_THREAD_JOINABLE
);
4105 s
->migration_thread_running
= true;
4108 void migration_global_dump(Monitor
*mon
)
4110 MigrationState
*ms
= migrate_get_current();
4112 monitor_printf(mon
, "globals:\n");
4113 monitor_printf(mon
, "store-global-state: %s\n",
4114 ms
->store_global_state
? "on" : "off");
4115 monitor_printf(mon
, "only-migratable: %s\n",
4116 only_migratable
? "on" : "off");
4117 monitor_printf(mon
, "send-configuration: %s\n",
4118 ms
->send_configuration
? "on" : "off");
4119 monitor_printf(mon
, "send-section-footer: %s\n",
4120 ms
->send_section_footer
? "on" : "off");
4121 monitor_printf(mon
, "decompress-error-check: %s\n",
4122 ms
->decompress_error_check
? "on" : "off");
4123 monitor_printf(mon
, "clear-bitmap-shift: %u\n",
4124 ms
->clear_bitmap_shift
);
4127 #define DEFINE_PROP_MIG_CAP(name, x) \
4128 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
4130 static Property migration_properties
[] = {
4131 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
4132 store_global_state
, true),
4133 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
4134 send_configuration
, true),
4135 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
4136 send_section_footer
, true),
4137 DEFINE_PROP_BOOL("decompress-error-check", MigrationState
,
4138 decompress_error_check
, true),
4139 DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState
,
4140 clear_bitmap_shift
, CLEAR_BITMAP_SHIFT_DEFAULT
),
4142 /* Migration parameters */
4143 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
4144 parameters
.compress_level
,
4145 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
4146 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
4147 parameters
.compress_threads
,
4148 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
4149 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState
,
4150 parameters
.compress_wait_thread
, true),
4151 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
4152 parameters
.decompress_threads
,
4153 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
4154 DEFINE_PROP_UINT8("x-throttle-trigger-threshold", MigrationState
,
4155 parameters
.throttle_trigger_threshold
,
4156 DEFAULT_MIGRATE_THROTTLE_TRIGGER_THRESHOLD
),
4157 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
4158 parameters
.cpu_throttle_initial
,
4159 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
4160 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
4161 parameters
.cpu_throttle_increment
,
4162 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
4163 DEFINE_PROP_BOOL("x-cpu-throttle-tailslow", MigrationState
,
4164 parameters
.cpu_throttle_tailslow
, false),
4165 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
4166 parameters
.max_bandwidth
, MAX_THROTTLE
),
4167 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
4168 parameters
.downtime_limit
,
4169 DEFAULT_MIGRATE_SET_DOWNTIME
),
4170 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
4171 parameters
.x_checkpoint_delay
,
4172 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
4173 DEFINE_PROP_UINT8("multifd-channels", MigrationState
,
4174 parameters
.multifd_channels
,
4175 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
4176 DEFINE_PROP_MULTIFD_COMPRESSION("multifd-compression", MigrationState
,
4177 parameters
.multifd_compression
,
4178 DEFAULT_MIGRATE_MULTIFD_COMPRESSION
),
4179 DEFINE_PROP_UINT8("multifd-zlib-level", MigrationState
,
4180 parameters
.multifd_zlib_level
,
4181 DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL
),
4182 DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState
,
4183 parameters
.multifd_zstd_level
,
4184 DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL
),
4185 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
4186 parameters
.xbzrle_cache_size
,
4187 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
4188 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState
,
4189 parameters
.max_postcopy_bandwidth
,
4190 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH
),
4191 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState
,
4192 parameters
.max_cpu_throttle
,
4193 DEFAULT_MIGRATE_MAX_CPU_THROTTLE
),
4194 DEFINE_PROP_SIZE("announce-initial", MigrationState
,
4195 parameters
.announce_initial
,
4196 DEFAULT_MIGRATE_ANNOUNCE_INITIAL
),
4197 DEFINE_PROP_SIZE("announce-max", MigrationState
,
4198 parameters
.announce_max
,
4199 DEFAULT_MIGRATE_ANNOUNCE_MAX
),
4200 DEFINE_PROP_SIZE("announce-rounds", MigrationState
,
4201 parameters
.announce_rounds
,
4202 DEFAULT_MIGRATE_ANNOUNCE_ROUNDS
),
4203 DEFINE_PROP_SIZE("announce-step", MigrationState
,
4204 parameters
.announce_step
,
4205 DEFAULT_MIGRATE_ANNOUNCE_STEP
),
4207 /* Migration capabilities */
4208 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
4209 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
4210 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
4211 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
4212 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
4213 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
4214 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
4215 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
4216 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
4217 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
4218 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
4219 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD
),
4220 DEFINE_PROP_MIG_CAP("x-background-snapshot",
4221 MIGRATION_CAPABILITY_BACKGROUND_SNAPSHOT
),
4223 DEFINE_PROP_END_OF_LIST(),
4226 static void migration_class_init(ObjectClass
*klass
, void *data
)
4228 DeviceClass
*dc
= DEVICE_CLASS(klass
);
4230 dc
->user_creatable
= false;
4231 device_class_set_props(dc
, migration_properties
);
4234 static void migration_instance_finalize(Object
*obj
)
4236 MigrationState
*ms
= MIGRATION_OBJ(obj
);
4237 MigrationParameters
*params
= &ms
->parameters
;
4239 qemu_mutex_destroy(&ms
->error_mutex
);
4240 qemu_mutex_destroy(&ms
->qemu_file_lock
);
4241 g_free(params
->tls_hostname
);
4242 g_free(params
->tls_creds
);
4243 qemu_sem_destroy(&ms
->wait_unplug_sem
);
4244 qemu_sem_destroy(&ms
->rate_limit_sem
);
4245 qemu_sem_destroy(&ms
->pause_sem
);
4246 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
4247 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
4248 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
4249 error_free(ms
->error
);
4252 static void migration_instance_init(Object
*obj
)
4254 MigrationState
*ms
= MIGRATION_OBJ(obj
);
4255 MigrationParameters
*params
= &ms
->parameters
;
4257 ms
->state
= MIGRATION_STATUS_NONE
;
4259 ms
->pages_per_second
= -1;
4260 qemu_sem_init(&ms
->pause_sem
, 0);
4261 qemu_mutex_init(&ms
->error_mutex
);
4263 params
->tls_hostname
= g_strdup("");
4264 params
->tls_creds
= g_strdup("");
4266 /* Set has_* up only for parameter checks */
4267 params
->has_compress_level
= true;
4268 params
->has_compress_threads
= true;
4269 params
->has_decompress_threads
= true;
4270 params
->has_throttle_trigger_threshold
= true;
4271 params
->has_cpu_throttle_initial
= true;
4272 params
->has_cpu_throttle_increment
= true;
4273 params
->has_cpu_throttle_tailslow
= true;
4274 params
->has_max_bandwidth
= true;
4275 params
->has_downtime_limit
= true;
4276 params
->has_x_checkpoint_delay
= true;
4277 params
->has_block_incremental
= true;
4278 params
->has_multifd_channels
= true;
4279 params
->has_multifd_compression
= true;
4280 params
->has_multifd_zlib_level
= true;
4281 params
->has_multifd_zstd_level
= true;
4282 params
->has_xbzrle_cache_size
= true;
4283 params
->has_max_postcopy_bandwidth
= true;
4284 params
->has_max_cpu_throttle
= true;
4285 params
->has_announce_initial
= true;
4286 params
->has_announce_max
= true;
4287 params
->has_announce_rounds
= true;
4288 params
->has_announce_step
= true;
4290 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
4291 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
4292 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
4293 qemu_sem_init(&ms
->rate_limit_sem
, 0);
4294 qemu_sem_init(&ms
->wait_unplug_sem
, 0);
4295 qemu_mutex_init(&ms
->qemu_file_lock
);
4299 * Return true if check pass, false otherwise. Error will be put
4300 * inside errp if provided.
4302 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
4304 MigrationCapabilityStatusList
*head
= NULL
;
4305 /* Assuming all off */
4306 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
4309 if (!migrate_params_check(&ms
->parameters
, errp
)) {
4313 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
4314 if (ms
->enabled_capabilities
[i
]) {
4315 QAPI_LIST_PREPEND(head
, migrate_cap_add(i
, true));
4319 ret
= migrate_caps_check(cap_list
, head
, errp
);
4321 /* It works with head == NULL */
4322 qapi_free_MigrationCapabilityStatusList(head
);
4327 static const TypeInfo migration_type
= {
4328 .name
= TYPE_MIGRATION
,
4330 * NOTE: TYPE_MIGRATION is not really a device, as the object is
4331 * not created using qdev_new(), it is not attached to the qdev
4332 * device tree, and it is never realized.
4334 * TODO: Make this TYPE_OBJECT once QOM provides something like
4335 * TYPE_DEVICE's "-global" properties.
4337 .parent
= TYPE_DEVICE
,
4338 .class_init
= migration_class_init
,
4339 .class_size
= sizeof(MigrationClass
),
4340 .instance_size
= sizeof(MigrationState
),
4341 .instance_init
= migration_instance_init
,
4342 .instance_finalize
= migration_instance_finalize
,
4345 static void register_migration_types(void)
4347 type_register_static(&migration_type
);
4350 type_init(register_migration_types
);