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 "migration/blocker.h"
25 #include "migration/global_state.h"
26 #include "migration/misc.h"
27 #include "migration.h"
29 #include "qemu-file-channel.h"
30 #include "qemu-file.h"
31 #include "migration/vmstate.h"
32 #include "block/block.h"
33 #include "qapi/error.h"
34 #include "qapi/clone-visitor.h"
35 #include "qapi/qapi-visit-sockets.h"
36 #include "qapi/qapi-commands-migration.h"
37 #include "qapi/qapi-events-migration.h"
38 #include "qapi/qmp/qerror.h"
39 #include "qapi/qmp/qnull.h"
42 #include "postcopy-ram.h"
43 #include "qemu/thread.h"
45 #include "exec/target_page.h"
46 #include "io/channel-buffer.h"
47 #include "migration/colo.h"
48 #include "hw/boards.h"
49 #include "monitor/monitor.h"
50 #include "net/announce.h"
52 #define MAX_THROTTLE (32 << 20) /* Migration transfer speed throttling */
54 /* Amount of time to allocate to each "chunk" of bandwidth-throttled
56 #define BUFFER_DELAY 100
57 #define XFER_LIMIT_RATIO (1000 / BUFFER_DELAY)
59 /* Time in milliseconds we are allowed to stop the source,
60 * for sending the last part */
61 #define DEFAULT_MIGRATE_SET_DOWNTIME 300
63 /* Maximum migrate downtime set to 2000 seconds */
64 #define MAX_MIGRATE_DOWNTIME_SECONDS 2000
65 #define MAX_MIGRATE_DOWNTIME (MAX_MIGRATE_DOWNTIME_SECONDS * 1000)
67 /* Default compression thread count */
68 #define DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT 8
69 /* Default decompression thread count, usually decompression is at
70 * least 4 times as fast as compression.*/
71 #define DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT 2
72 /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
73 #define DEFAULT_MIGRATE_COMPRESS_LEVEL 1
74 /* Define default autoconverge cpu throttle migration parameters */
75 #define DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL 20
76 #define DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT 10
77 #define DEFAULT_MIGRATE_MAX_CPU_THROTTLE 99
79 /* Migration XBZRLE default cache size */
80 #define DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE (64 * 1024 * 1024)
82 /* The delay time (in ms) between two COLO checkpoints */
83 #define DEFAULT_MIGRATE_X_CHECKPOINT_DELAY (200 * 100)
84 #define DEFAULT_MIGRATE_MULTIFD_CHANNELS 2
86 /* Background transfer rate for postcopy, 0 means unlimited, note
87 * that page requests can still exceed this limit.
89 #define DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH 0
92 * Parameters for self_announce_delay giving a stream of RARP/ARP
93 * packets after migration.
95 #define DEFAULT_MIGRATE_ANNOUNCE_INITIAL 50
96 #define DEFAULT_MIGRATE_ANNOUNCE_MAX 550
97 #define DEFAULT_MIGRATE_ANNOUNCE_ROUNDS 5
98 #define DEFAULT_MIGRATE_ANNOUNCE_STEP 100
100 static NotifierList migration_state_notifiers
=
101 NOTIFIER_LIST_INITIALIZER(migration_state_notifiers
);
103 static bool deferred_incoming
;
105 /* Messages sent on the return path from destination to source */
106 enum mig_rp_message_type
{
107 MIG_RP_MSG_INVALID
= 0, /* Must be 0 */
108 MIG_RP_MSG_SHUT
, /* sibling will not send any more RP messages */
109 MIG_RP_MSG_PONG
, /* Response to a PING; data (seq: be32 ) */
111 MIG_RP_MSG_REQ_PAGES_ID
, /* data (start: be64, len: be32, id: string) */
112 MIG_RP_MSG_REQ_PAGES
, /* data (start: be64, len: be32) */
113 MIG_RP_MSG_RECV_BITMAP
, /* send recved_bitmap back to source */
114 MIG_RP_MSG_RESUME_ACK
, /* tell source that we are ready to resume */
119 /* When we add fault tolerance, we could have several
120 migrations at once. For now we don't need to add
121 dynamic creation of migration */
123 static MigrationState
*current_migration
;
124 static MigrationIncomingState
*current_incoming
;
126 static bool migration_object_check(MigrationState
*ms
, Error
**errp
);
127 static int migration_maybe_pause(MigrationState
*s
,
128 int *current_active_state
,
130 static void migrate_fd_cancel(MigrationState
*s
);
132 void migration_object_init(void)
134 MachineState
*ms
= MACHINE(qdev_get_machine());
137 /* This can only be called once. */
138 assert(!current_migration
);
139 current_migration
= MIGRATION_OBJ(object_new(TYPE_MIGRATION
));
142 * Init the migrate incoming object as well no matter whether
143 * we'll use it or not.
145 assert(!current_incoming
);
146 current_incoming
= g_new0(MigrationIncomingState
, 1);
147 current_incoming
->state
= MIGRATION_STATUS_NONE
;
148 current_incoming
->postcopy_remote_fds
=
149 g_array_new(FALSE
, TRUE
, sizeof(struct PostCopyFD
));
150 qemu_mutex_init(¤t_incoming
->rp_mutex
);
151 qemu_event_init(¤t_incoming
->main_thread_load_event
, false);
152 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_dst
, 0);
153 qemu_sem_init(¤t_incoming
->postcopy_pause_sem_fault
, 0);
155 init_dirty_bitmap_incoming_migration();
157 if (!migration_object_check(current_migration
, &err
)) {
158 error_report_err(err
);
163 * We cannot really do this in migration_instance_init() since at
164 * that time global properties are not yet applied, then this
165 * value will be definitely replaced by something else.
167 if (ms
->enforce_config_section
) {
168 current_migration
->send_configuration
= true;
172 void migration_shutdown(void)
175 * Cancel the current migration - that will (eventually)
176 * stop the migration using this structure
178 migrate_fd_cancel(current_migration
);
179 object_unref(OBJECT(current_migration
));
183 MigrationState
*migrate_get_current(void)
185 /* This can only be called after the object created. */
186 assert(current_migration
);
187 return current_migration
;
190 MigrationIncomingState
*migration_incoming_get_current(void)
192 assert(current_incoming
);
193 return current_incoming
;
196 void migration_incoming_state_destroy(void)
198 struct MigrationIncomingState
*mis
= migration_incoming_get_current();
200 if (mis
->to_src_file
) {
201 /* Tell source that we are done */
202 migrate_send_rp_shut(mis
, qemu_file_get_error(mis
->from_src_file
) != 0);
203 qemu_fclose(mis
->to_src_file
);
204 mis
->to_src_file
= NULL
;
207 if (mis
->from_src_file
) {
208 qemu_fclose(mis
->from_src_file
);
209 mis
->from_src_file
= NULL
;
211 if (mis
->postcopy_remote_fds
) {
212 g_array_free(mis
->postcopy_remote_fds
, TRUE
);
213 mis
->postcopy_remote_fds
= NULL
;
216 qemu_event_reset(&mis
->main_thread_load_event
);
218 if (mis
->socket_address_list
) {
219 qapi_free_SocketAddressList(mis
->socket_address_list
);
220 mis
->socket_address_list
= NULL
;
224 static void migrate_generate_event(int new_state
)
226 if (migrate_use_events()) {
227 qapi_event_send_migration(new_state
);
231 static bool migrate_late_block_activate(void)
235 s
= migrate_get_current();
237 return s
->enabled_capabilities
[
238 MIGRATION_CAPABILITY_LATE_BLOCK_ACTIVATE
];
242 * Called on -incoming with a defer: uri.
243 * The migration can be started later after any parameters have been
246 static void deferred_incoming_migration(Error
**errp
)
248 if (deferred_incoming
) {
249 error_setg(errp
, "Incoming migration already deferred");
251 deferred_incoming
= true;
255 * Send a message on the return channel back to the source
258 static int migrate_send_rp_message(MigrationIncomingState
*mis
,
259 enum mig_rp_message_type message_type
,
260 uint16_t len
, void *data
)
264 trace_migrate_send_rp_message((int)message_type
, len
);
265 qemu_mutex_lock(&mis
->rp_mutex
);
268 * It's possible that the file handle got lost due to network
271 if (!mis
->to_src_file
) {
276 qemu_put_be16(mis
->to_src_file
, (unsigned int)message_type
);
277 qemu_put_be16(mis
->to_src_file
, len
);
278 qemu_put_buffer(mis
->to_src_file
, data
, len
);
279 qemu_fflush(mis
->to_src_file
);
281 /* It's possible that qemu file got error during sending */
282 ret
= qemu_file_get_error(mis
->to_src_file
);
285 qemu_mutex_unlock(&mis
->rp_mutex
);
289 /* Request a range of pages from the source VM at the given
291 * rbname: Name of the RAMBlock to request the page in, if NULL it's the same
292 * as the last request (a name must have been given previously)
293 * Start: Address offset within the RB
294 * Len: Length in bytes required - must be a multiple of pagesize
296 int migrate_send_rp_req_pages(MigrationIncomingState
*mis
, const char *rbname
,
297 ram_addr_t start
, size_t len
)
299 uint8_t bufc
[12 + 1 + 255]; /* start (8), len (4), rbname up to 256 */
300 size_t msglen
= 12; /* start + len */
301 enum mig_rp_message_type msg_type
;
303 *(uint64_t *)bufc
= cpu_to_be64((uint64_t)start
);
304 *(uint32_t *)(bufc
+ 8) = cpu_to_be32((uint32_t)len
);
307 int rbname_len
= strlen(rbname
);
308 assert(rbname_len
< 256);
310 bufc
[msglen
++] = rbname_len
;
311 memcpy(bufc
+ msglen
, rbname
, rbname_len
);
312 msglen
+= rbname_len
;
313 msg_type
= MIG_RP_MSG_REQ_PAGES_ID
;
315 msg_type
= MIG_RP_MSG_REQ_PAGES
;
318 return migrate_send_rp_message(mis
, msg_type
, msglen
, bufc
);
321 static bool migration_colo_enabled
;
322 bool migration_incoming_colo_enabled(void)
324 return migration_colo_enabled
;
327 void migration_incoming_disable_colo(void)
329 migration_colo_enabled
= false;
332 void migration_incoming_enable_colo(void)
334 migration_colo_enabled
= true;
337 void migrate_add_address(SocketAddress
*address
)
339 MigrationIncomingState
*mis
= migration_incoming_get_current();
340 SocketAddressList
*addrs
;
342 addrs
= g_new0(SocketAddressList
, 1);
343 addrs
->next
= mis
->socket_address_list
;
344 mis
->socket_address_list
= addrs
;
345 addrs
->value
= QAPI_CLONE(SocketAddress
, address
);
348 void qemu_start_incoming_migration(const char *uri
, Error
**errp
)
352 qapi_event_send_migration(MIGRATION_STATUS_SETUP
);
353 if (!strcmp(uri
, "defer")) {
354 deferred_incoming_migration(errp
);
355 } else if (strstart(uri
, "tcp:", &p
)) {
356 tcp_start_incoming_migration(p
, errp
);
358 } else if (strstart(uri
, "rdma:", &p
)) {
359 rdma_start_incoming_migration(p
, errp
);
361 } else if (strstart(uri
, "exec:", &p
)) {
362 exec_start_incoming_migration(p
, errp
);
363 } else if (strstart(uri
, "unix:", &p
)) {
364 unix_start_incoming_migration(p
, errp
);
365 } else if (strstart(uri
, "fd:", &p
)) {
366 fd_start_incoming_migration(p
, errp
);
368 error_setg(errp
, "unknown migration protocol: %s", uri
);
372 static void process_incoming_migration_bh(void *opaque
)
374 Error
*local_err
= NULL
;
375 MigrationIncomingState
*mis
= opaque
;
377 /* If capability late_block_activate is set:
378 * Only fire up the block code now if we're going to restart the
379 * VM, else 'cont' will do it.
380 * This causes file locking to happen; so we don't want it to happen
381 * unless we really are starting the VM.
383 if (!migrate_late_block_activate() ||
384 (autostart
&& (!global_state_received() ||
385 global_state_get_runstate() == RUN_STATE_RUNNING
))) {
386 /* Make sure all file formats flush their mutable metadata.
387 * If we get an error here, just don't restart the VM yet. */
388 bdrv_invalidate_cache_all(&local_err
);
390 error_report_err(local_err
);
397 * This must happen after all error conditions are dealt with and
398 * we're sure the VM is going to be running on this host.
400 qemu_announce_self(&mis
->announce_timer
, migrate_announce_params());
402 if (multifd_load_cleanup(&local_err
) != 0) {
403 error_report_err(local_err
);
406 /* If global state section was not received or we are in running
407 state, we need to obey autostart. Any other state is set with
410 dirty_bitmap_mig_before_vm_start();
412 if (!global_state_received() ||
413 global_state_get_runstate() == RUN_STATE_RUNNING
) {
417 runstate_set(RUN_STATE_PAUSED
);
419 } else if (migration_incoming_colo_enabled()) {
420 migration_incoming_disable_colo();
423 runstate_set(global_state_get_runstate());
426 * This must happen after any state changes since as soon as an external
427 * observer sees this event they might start to prod at the VM assuming
430 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
431 MIGRATION_STATUS_COMPLETED
);
432 qemu_bh_delete(mis
->bh
);
433 migration_incoming_state_destroy();
436 static void process_incoming_migration_co(void *opaque
)
438 MigrationIncomingState
*mis
= migration_incoming_get_current();
441 Error
*local_err
= NULL
;
443 assert(mis
->from_src_file
);
444 mis
->migration_incoming_co
= qemu_coroutine_self();
445 mis
->largest_page_size
= qemu_ram_pagesize_largest();
446 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
447 migrate_set_state(&mis
->state
, MIGRATION_STATUS_NONE
,
448 MIGRATION_STATUS_ACTIVE
);
449 ret
= qemu_loadvm_state(mis
->from_src_file
);
451 ps
= postcopy_state_get();
452 trace_process_incoming_migration_co_end(ret
, ps
);
453 if (ps
!= POSTCOPY_INCOMING_NONE
) {
454 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
456 * Where a migration had postcopy enabled (and thus went to advise)
457 * but managed to complete within the precopy period, we can use
460 postcopy_ram_incoming_cleanup(mis
);
461 } else if (ret
>= 0) {
463 * Postcopy was started, cleanup should happen at the end of the
466 trace_process_incoming_migration_co_postcopy_end_main();
469 /* Else if something went wrong then just fall out of the normal exit */
472 /* we get COLO info, and know if we are in COLO mode */
473 if (!ret
&& migration_incoming_colo_enabled()) {
474 /* Make sure all file formats flush their mutable metadata */
475 bdrv_invalidate_cache_all(&local_err
);
477 error_report_err(local_err
);
481 if (colo_init_ram_cache() < 0) {
482 error_report("Init ram cache failed");
486 qemu_thread_create(&mis
->colo_incoming_thread
, "COLO incoming",
487 colo_process_incoming_thread
, mis
, QEMU_THREAD_JOINABLE
);
488 mis
->have_colo_incoming_thread
= true;
489 qemu_coroutine_yield();
491 /* Wait checkpoint incoming thread exit before free resource */
492 qemu_thread_join(&mis
->colo_incoming_thread
);
493 /* We hold the global iothread lock, so it is safe here */
494 colo_release_ram_cache();
498 error_report("load of migration failed: %s", strerror(-ret
));
501 mis
->bh
= qemu_bh_new(process_incoming_migration_bh
, mis
);
502 qemu_bh_schedule(mis
->bh
);
503 mis
->migration_incoming_co
= NULL
;
507 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
508 MIGRATION_STATUS_FAILED
);
509 qemu_fclose(mis
->from_src_file
);
510 if (multifd_load_cleanup(&local_err
) != 0) {
511 error_report_err(local_err
);
516 static void migration_incoming_setup(QEMUFile
*f
)
518 MigrationIncomingState
*mis
= migration_incoming_get_current();
520 if (multifd_load_setup() != 0) {
521 /* We haven't been able to create multifd threads
522 nothing better to do */
526 if (!mis
->from_src_file
) {
527 mis
->from_src_file
= f
;
529 qemu_file_set_blocking(f
, false);
532 void migration_incoming_process(void)
534 Coroutine
*co
= qemu_coroutine_create(process_incoming_migration_co
, NULL
);
535 qemu_coroutine_enter(co
);
538 /* Returns true if recovered from a paused migration, otherwise false */
539 static bool postcopy_try_recover(QEMUFile
*f
)
541 MigrationIncomingState
*mis
= migration_incoming_get_current();
543 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
544 /* Resumed from a paused postcopy migration */
546 mis
->from_src_file
= f
;
547 /* Postcopy has standalone thread to do vm load */
548 qemu_file_set_blocking(f
, true);
550 /* Re-configure the return path */
551 mis
->to_src_file
= qemu_file_get_return_path(f
);
553 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
554 MIGRATION_STATUS_POSTCOPY_RECOVER
);
557 * Here, we only wake up the main loading thread (while the
558 * fault thread will still be waiting), so that we can receive
559 * commands from source now, and answer it if needed. The
560 * fault thread will be woken up afterwards until we are sure
561 * that source is ready to reply to page requests.
563 qemu_sem_post(&mis
->postcopy_pause_sem_dst
);
570 void migration_fd_process_incoming(QEMUFile
*f
)
572 if (postcopy_try_recover(f
)) {
576 migration_incoming_setup(f
);
577 migration_incoming_process();
580 void migration_ioc_process_incoming(QIOChannel
*ioc
, Error
**errp
)
582 MigrationIncomingState
*mis
= migration_incoming_get_current();
583 bool start_migration
;
585 if (!mis
->from_src_file
) {
586 /* The first connection (multifd may have multiple) */
587 QEMUFile
*f
= qemu_fopen_channel_input(ioc
);
589 /* If it's a recovery, we're done */
590 if (postcopy_try_recover(f
)) {
594 migration_incoming_setup(f
);
597 * Common migration only needs one channel, so we can start
598 * right now. Multifd needs more than one channel, we wait.
600 start_migration
= !migrate_use_multifd();
602 Error
*local_err
= NULL
;
603 /* Multiple connections */
604 assert(migrate_use_multifd());
605 start_migration
= multifd_recv_new_channel(ioc
, &local_err
);
607 error_propagate(errp
, local_err
);
612 if (start_migration
) {
613 migration_incoming_process();
618 * @migration_has_all_channels: We have received all channels that we need
620 * Returns true when we have got connections to all the channels that
621 * we need for migration.
623 bool migration_has_all_channels(void)
625 MigrationIncomingState
*mis
= migration_incoming_get_current();
628 all_channels
= multifd_recv_all_channels_created();
630 return all_channels
&& mis
->from_src_file
!= NULL
;
634 * Send a 'SHUT' message on the return channel with the given value
635 * to indicate that we've finished with the RP. Non-0 value indicates
638 void migrate_send_rp_shut(MigrationIncomingState
*mis
,
643 buf
= cpu_to_be32(value
);
644 migrate_send_rp_message(mis
, MIG_RP_MSG_SHUT
, sizeof(buf
), &buf
);
648 * Send a 'PONG' message on the return channel with the given value
649 * (normally in response to a 'PING')
651 void migrate_send_rp_pong(MigrationIncomingState
*mis
,
656 buf
= cpu_to_be32(value
);
657 migrate_send_rp_message(mis
, MIG_RP_MSG_PONG
, sizeof(buf
), &buf
);
660 void migrate_send_rp_recv_bitmap(MigrationIncomingState
*mis
,
668 * First, we send the header part. It contains only the len of
669 * idstr, and the idstr itself.
671 len
= strlen(block_name
);
673 memcpy(buf
+ 1, block_name
, len
);
675 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
676 error_report("%s: MSG_RP_RECV_BITMAP only used for recovery",
681 migrate_send_rp_message(mis
, MIG_RP_MSG_RECV_BITMAP
, len
+ 1, buf
);
684 * Next, we dump the received bitmap to the stream.
686 * TODO: currently we are safe since we are the only one that is
687 * using the to_src_file handle (fault thread is still paused),
688 * and it's ok even not taking the mutex. However the best way is
689 * to take the lock before sending the message header, and release
690 * the lock after sending the bitmap.
692 qemu_mutex_lock(&mis
->rp_mutex
);
693 res
= ramblock_recv_bitmap_send(mis
->to_src_file
, block_name
);
694 qemu_mutex_unlock(&mis
->rp_mutex
);
696 trace_migrate_send_rp_recv_bitmap(block_name
, res
);
699 void migrate_send_rp_resume_ack(MigrationIncomingState
*mis
, uint32_t value
)
703 buf
= cpu_to_be32(value
);
704 migrate_send_rp_message(mis
, MIG_RP_MSG_RESUME_ACK
, sizeof(buf
), &buf
);
707 MigrationCapabilityStatusList
*qmp_query_migrate_capabilities(Error
**errp
)
709 MigrationCapabilityStatusList
*head
= NULL
;
710 MigrationCapabilityStatusList
*caps
;
711 MigrationState
*s
= migrate_get_current();
714 caps
= NULL
; /* silence compiler warning */
715 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
716 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
717 if (i
== MIGRATION_CAPABILITY_BLOCK
) {
722 head
= g_malloc0(sizeof(*caps
));
725 caps
->next
= g_malloc0(sizeof(*caps
));
729 g_malloc(sizeof(*caps
->value
));
730 caps
->value
->capability
= i
;
731 caps
->value
->state
= s
->enabled_capabilities
[i
];
737 MigrationParameters
*qmp_query_migrate_parameters(Error
**errp
)
739 MigrationParameters
*params
;
740 MigrationState
*s
= migrate_get_current();
742 /* TODO use QAPI_CLONE() instead of duplicating it inline */
743 params
= g_malloc0(sizeof(*params
));
744 params
->has_compress_level
= true;
745 params
->compress_level
= s
->parameters
.compress_level
;
746 params
->has_compress_threads
= true;
747 params
->compress_threads
= s
->parameters
.compress_threads
;
748 params
->has_compress_wait_thread
= true;
749 params
->compress_wait_thread
= s
->parameters
.compress_wait_thread
;
750 params
->has_decompress_threads
= true;
751 params
->decompress_threads
= s
->parameters
.decompress_threads
;
752 params
->has_cpu_throttle_initial
= true;
753 params
->cpu_throttle_initial
= s
->parameters
.cpu_throttle_initial
;
754 params
->has_cpu_throttle_increment
= true;
755 params
->cpu_throttle_increment
= s
->parameters
.cpu_throttle_increment
;
756 params
->has_tls_creds
= true;
757 params
->tls_creds
= g_strdup(s
->parameters
.tls_creds
);
758 params
->has_tls_hostname
= true;
759 params
->tls_hostname
= g_strdup(s
->parameters
.tls_hostname
);
760 params
->has_max_bandwidth
= true;
761 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
762 params
->has_downtime_limit
= true;
763 params
->downtime_limit
= s
->parameters
.downtime_limit
;
764 params
->has_x_checkpoint_delay
= true;
765 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
766 params
->has_block_incremental
= true;
767 params
->block_incremental
= s
->parameters
.block_incremental
;
768 params
->has_x_multifd_channels
= true;
769 params
->x_multifd_channels
= s
->parameters
.x_multifd_channels
;
770 params
->has_xbzrle_cache_size
= true;
771 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
772 params
->has_max_postcopy_bandwidth
= true;
773 params
->max_postcopy_bandwidth
= s
->parameters
.max_postcopy_bandwidth
;
774 params
->has_max_cpu_throttle
= true;
775 params
->max_cpu_throttle
= s
->parameters
.max_cpu_throttle
;
776 params
->has_announce_initial
= true;
777 params
->announce_initial
= s
->parameters
.announce_initial
;
778 params
->has_announce_max
= true;
779 params
->announce_max
= s
->parameters
.announce_max
;
780 params
->has_announce_rounds
= true;
781 params
->announce_rounds
= s
->parameters
.announce_rounds
;
782 params
->has_announce_step
= true;
783 params
->announce_step
= s
->parameters
.announce_step
;
788 AnnounceParameters
*migrate_announce_params(void)
790 static AnnounceParameters ap
;
792 MigrationState
*s
= migrate_get_current();
794 ap
.initial
= s
->parameters
.announce_initial
;
795 ap
.max
= s
->parameters
.announce_max
;
796 ap
.rounds
= s
->parameters
.announce_rounds
;
797 ap
.step
= s
->parameters
.announce_step
;
803 * Return true if we're already in the middle of a migration
804 * (i.e. any of the active or setup states)
806 bool migration_is_setup_or_active(int state
)
809 case MIGRATION_STATUS_ACTIVE
:
810 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
811 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
812 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
813 case MIGRATION_STATUS_SETUP
:
814 case MIGRATION_STATUS_PRE_SWITCHOVER
:
815 case MIGRATION_STATUS_DEVICE
:
824 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
826 info
->has_ram
= true;
827 info
->ram
= g_malloc0(sizeof(*info
->ram
));
828 info
->ram
->transferred
= ram_counters
.transferred
;
829 info
->ram
->total
= ram_bytes_total();
830 info
->ram
->duplicate
= ram_counters
.duplicate
;
831 /* legacy value. It is not used anymore */
832 info
->ram
->skipped
= 0;
833 info
->ram
->normal
= ram_counters
.normal
;
834 info
->ram
->normal_bytes
= ram_counters
.normal
*
835 qemu_target_page_size();
836 info
->ram
->mbps
= s
->mbps
;
837 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
838 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
839 info
->ram
->page_size
= qemu_target_page_size();
840 info
->ram
->multifd_bytes
= ram_counters
.multifd_bytes
;
841 info
->ram
->pages_per_second
= s
->pages_per_second
;
843 if (migrate_use_xbzrle()) {
844 info
->has_xbzrle_cache
= true;
845 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
846 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
847 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
848 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
849 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
850 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
851 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
854 if (migrate_use_compression()) {
855 info
->has_compression
= true;
856 info
->compression
= g_malloc0(sizeof(*info
->compression
));
857 info
->compression
->pages
= compression_counters
.pages
;
858 info
->compression
->busy
= compression_counters
.busy
;
859 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
860 info
->compression
->compressed_size
=
861 compression_counters
.compressed_size
;
862 info
->compression
->compression_rate
=
863 compression_counters
.compression_rate
;
866 if (cpu_throttle_active()) {
867 info
->has_cpu_throttle_percentage
= true;
868 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
871 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
872 info
->ram
->remaining
= ram_bytes_remaining();
873 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
877 static void populate_disk_info(MigrationInfo
*info
)
879 if (blk_mig_active()) {
880 info
->has_disk
= true;
881 info
->disk
= g_malloc0(sizeof(*info
->disk
));
882 info
->disk
->transferred
= blk_mig_bytes_transferred();
883 info
->disk
->remaining
= blk_mig_bytes_remaining();
884 info
->disk
->total
= blk_mig_bytes_total();
888 static void fill_source_migration_info(MigrationInfo
*info
)
890 MigrationState
*s
= migrate_get_current();
893 case MIGRATION_STATUS_NONE
:
894 /* no migration has happened ever */
895 /* do not overwrite destination migration status */
898 case MIGRATION_STATUS_SETUP
:
899 info
->has_status
= true;
900 info
->has_total_time
= false;
902 case MIGRATION_STATUS_ACTIVE
:
903 case MIGRATION_STATUS_CANCELLING
:
904 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
905 case MIGRATION_STATUS_PRE_SWITCHOVER
:
906 case MIGRATION_STATUS_DEVICE
:
907 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
908 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
909 /* TODO add some postcopy stats */
910 info
->has_status
= true;
911 info
->has_total_time
= true;
912 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
)
914 info
->has_expected_downtime
= true;
915 info
->expected_downtime
= s
->expected_downtime
;
916 info
->has_setup_time
= true;
917 info
->setup_time
= s
->setup_time
;
919 populate_ram_info(info
, s
);
920 populate_disk_info(info
);
922 case MIGRATION_STATUS_COLO
:
923 info
->has_status
= true;
924 /* TODO: display COLO specific information (checkpoint info etc.) */
926 case MIGRATION_STATUS_COMPLETED
:
927 info
->has_status
= true;
928 info
->has_total_time
= true;
929 info
->total_time
= s
->total_time
;
930 info
->has_downtime
= true;
931 info
->downtime
= s
->downtime
;
932 info
->has_setup_time
= true;
933 info
->setup_time
= s
->setup_time
;
935 populate_ram_info(info
, s
);
937 case MIGRATION_STATUS_FAILED
:
938 info
->has_status
= true;
940 info
->has_error_desc
= true;
941 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
944 case MIGRATION_STATUS_CANCELLED
:
945 info
->has_status
= true;
948 info
->status
= s
->state
;
952 * @migration_caps_check - check capability validity
954 * @cap_list: old capability list, array of bool
955 * @params: new capabilities to be applied soon
956 * @errp: set *errp if the check failed, with reason
958 * Returns true if check passed, otherwise false.
960 static bool migrate_caps_check(bool *cap_list
,
961 MigrationCapabilityStatusList
*params
,
964 MigrationCapabilityStatusList
*cap
;
965 bool old_postcopy_cap
;
966 MigrationIncomingState
*mis
= migration_incoming_get_current();
968 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
970 for (cap
= params
; cap
; cap
= cap
->next
) {
971 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
974 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
975 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
976 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
978 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
983 #ifndef CONFIG_REPLICATION
984 if (cap_list
[MIGRATION_CAPABILITY_X_COLO
]) {
985 error_setg(errp
, "QEMU compiled without replication module"
986 " can't enable COLO");
987 error_append_hint(errp
, "Please enable replication before COLO.\n");
992 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
993 if (cap_list
[MIGRATION_CAPABILITY_COMPRESS
]) {
994 /* The decompression threads asynchronously write into RAM
995 * rather than use the atomic copies needed to avoid
996 * userfaulting. It should be possible to fix the decompression
997 * threads for compatibility in future.
999 error_setg(errp
, "Postcopy is not currently compatible "
1000 "with compression");
1004 /* This check is reasonably expensive, so only when it's being
1005 * set the first time, also it's only the destination that needs
1008 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
1009 !postcopy_ram_supported_by_host(mis
)) {
1010 /* postcopy_ram_supported_by_host will have emitted a more
1013 error_setg(errp
, "Postcopy is not supported");
1017 if (cap_list
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
]) {
1018 error_setg(errp
, "Postcopy is not compatible with ignore-shared");
1026 static void fill_destination_migration_info(MigrationInfo
*info
)
1028 MigrationIncomingState
*mis
= migration_incoming_get_current();
1030 if (mis
->socket_address_list
) {
1031 info
->has_socket_address
= true;
1032 info
->socket_address
=
1033 QAPI_CLONE(SocketAddressList
, mis
->socket_address_list
);
1036 switch (mis
->state
) {
1037 case MIGRATION_STATUS_NONE
:
1040 case MIGRATION_STATUS_SETUP
:
1041 case MIGRATION_STATUS_CANCELLING
:
1042 case MIGRATION_STATUS_CANCELLED
:
1043 case MIGRATION_STATUS_ACTIVE
:
1044 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1045 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1046 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1047 case MIGRATION_STATUS_FAILED
:
1048 case MIGRATION_STATUS_COLO
:
1049 info
->has_status
= true;
1051 case MIGRATION_STATUS_COMPLETED
:
1052 info
->has_status
= true;
1053 fill_destination_postcopy_migration_info(info
);
1056 info
->status
= mis
->state
;
1059 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1061 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1063 fill_destination_migration_info(info
);
1064 fill_source_migration_info(info
);
1069 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
1072 MigrationState
*s
= migrate_get_current();
1073 MigrationCapabilityStatusList
*cap
;
1074 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
1076 if (migration_is_setup_or_active(s
->state
)) {
1077 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1081 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
1082 if (!migrate_caps_check(cap_list
, params
, errp
)) {
1086 for (cap
= params
; cap
; cap
= cap
->next
) {
1087 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
1092 * Check whether the parameters are valid. Error will be put into errp
1093 * (if provided). Return true if valid, otherwise false.
1095 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
1097 if (params
->has_compress_level
&&
1098 (params
->compress_level
> 9)) {
1099 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
1100 "is invalid, it should be in the range of 0 to 9");
1104 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
1105 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1107 "is invalid, it should be in the range of 1 to 255");
1111 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
1112 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1113 "decompress_threads",
1114 "is invalid, it should be in the range of 1 to 255");
1118 if (params
->has_cpu_throttle_initial
&&
1119 (params
->cpu_throttle_initial
< 1 ||
1120 params
->cpu_throttle_initial
> 99)) {
1121 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1122 "cpu_throttle_initial",
1123 "an integer in the range of 1 to 99");
1127 if (params
->has_cpu_throttle_increment
&&
1128 (params
->cpu_throttle_increment
< 1 ||
1129 params
->cpu_throttle_increment
> 99)) {
1130 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1131 "cpu_throttle_increment",
1132 "an integer in the range of 1 to 99");
1136 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
1137 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
1138 " range of 0 to %zu bytes/second", SIZE_MAX
);
1142 if (params
->has_downtime_limit
&&
1143 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
1144 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1145 "the range of 0 to %d milliseconds",
1146 MAX_MIGRATE_DOWNTIME
);
1150 /* x_checkpoint_delay is now always positive */
1152 if (params
->has_x_multifd_channels
&& (params
->x_multifd_channels
< 1)) {
1153 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1155 "is invalid, it should be in the range of 1 to 255");
1159 if (params
->has_xbzrle_cache_size
&&
1160 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
1161 !is_power_of_2(params
->xbzrle_cache_size
))) {
1162 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1163 "xbzrle_cache_size",
1164 "is invalid, it should be bigger than target page size"
1165 " and a power of two");
1169 if (params
->has_max_cpu_throttle
&&
1170 (params
->max_cpu_throttle
< params
->cpu_throttle_initial
||
1171 params
->max_cpu_throttle
> 99)) {
1172 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1174 "an integer in the range of cpu_throttle_initial to 99");
1178 if (params
->has_announce_initial
&&
1179 params
->announce_initial
> 100000) {
1180 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1182 "is invalid, it must be less than 100000 ms");
1185 if (params
->has_announce_max
&&
1186 params
->announce_max
> 100000) {
1187 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1189 "is invalid, it must be less than 100000 ms");
1192 if (params
->has_announce_rounds
&&
1193 params
->announce_rounds
> 1000) {
1194 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1196 "is invalid, it must be in the range of 0 to 1000");
1199 if (params
->has_announce_step
&&
1200 (params
->announce_step
< 1 ||
1201 params
->announce_step
> 10000)) {
1202 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1204 "is invalid, it must be in the range of 1 to 10000 ms");
1210 static void migrate_params_test_apply(MigrateSetParameters
*params
,
1211 MigrationParameters
*dest
)
1213 *dest
= migrate_get_current()->parameters
;
1215 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1217 if (params
->has_compress_level
) {
1218 dest
->compress_level
= params
->compress_level
;
1221 if (params
->has_compress_threads
) {
1222 dest
->compress_threads
= params
->compress_threads
;
1225 if (params
->has_compress_wait_thread
) {
1226 dest
->compress_wait_thread
= params
->compress_wait_thread
;
1229 if (params
->has_decompress_threads
) {
1230 dest
->decompress_threads
= params
->decompress_threads
;
1233 if (params
->has_cpu_throttle_initial
) {
1234 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
1237 if (params
->has_cpu_throttle_increment
) {
1238 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
1241 if (params
->has_tls_creds
) {
1242 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1243 dest
->tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1246 if (params
->has_tls_hostname
) {
1247 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1248 dest
->tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1251 if (params
->has_max_bandwidth
) {
1252 dest
->max_bandwidth
= params
->max_bandwidth
;
1255 if (params
->has_downtime_limit
) {
1256 dest
->downtime_limit
= params
->downtime_limit
;
1259 if (params
->has_x_checkpoint_delay
) {
1260 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
1263 if (params
->has_block_incremental
) {
1264 dest
->block_incremental
= params
->block_incremental
;
1266 if (params
->has_x_multifd_channels
) {
1267 dest
->x_multifd_channels
= params
->x_multifd_channels
;
1269 if (params
->has_xbzrle_cache_size
) {
1270 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
1272 if (params
->has_max_postcopy_bandwidth
) {
1273 dest
->max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1275 if (params
->has_max_cpu_throttle
) {
1276 dest
->max_cpu_throttle
= params
->max_cpu_throttle
;
1278 if (params
->has_announce_initial
) {
1279 dest
->announce_initial
= params
->announce_initial
;
1281 if (params
->has_announce_max
) {
1282 dest
->announce_max
= params
->announce_max
;
1284 if (params
->has_announce_rounds
) {
1285 dest
->announce_rounds
= params
->announce_rounds
;
1287 if (params
->has_announce_step
) {
1288 dest
->announce_step
= params
->announce_step
;
1292 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
1294 MigrationState
*s
= migrate_get_current();
1296 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1298 if (params
->has_compress_level
) {
1299 s
->parameters
.compress_level
= params
->compress_level
;
1302 if (params
->has_compress_threads
) {
1303 s
->parameters
.compress_threads
= params
->compress_threads
;
1306 if (params
->has_compress_wait_thread
) {
1307 s
->parameters
.compress_wait_thread
= params
->compress_wait_thread
;
1310 if (params
->has_decompress_threads
) {
1311 s
->parameters
.decompress_threads
= params
->decompress_threads
;
1314 if (params
->has_cpu_throttle_initial
) {
1315 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
1318 if (params
->has_cpu_throttle_increment
) {
1319 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
1322 if (params
->has_tls_creds
) {
1323 g_free(s
->parameters
.tls_creds
);
1324 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1325 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1328 if (params
->has_tls_hostname
) {
1329 g_free(s
->parameters
.tls_hostname
);
1330 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1331 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1334 if (params
->has_max_bandwidth
) {
1335 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1336 if (s
->to_dst_file
) {
1337 qemu_file_set_rate_limit(s
->to_dst_file
,
1338 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1342 if (params
->has_downtime_limit
) {
1343 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1346 if (params
->has_x_checkpoint_delay
) {
1347 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1348 if (migration_in_colo_state()) {
1349 colo_checkpoint_notify(s
);
1353 if (params
->has_block_incremental
) {
1354 s
->parameters
.block_incremental
= params
->block_incremental
;
1356 if (params
->has_x_multifd_channels
) {
1357 s
->parameters
.x_multifd_channels
= params
->x_multifd_channels
;
1359 if (params
->has_xbzrle_cache_size
) {
1360 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1361 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1363 if (params
->has_max_postcopy_bandwidth
) {
1364 s
->parameters
.max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1366 if (params
->has_max_cpu_throttle
) {
1367 s
->parameters
.max_cpu_throttle
= params
->max_cpu_throttle
;
1369 if (params
->has_announce_initial
) {
1370 s
->parameters
.announce_initial
= params
->announce_initial
;
1372 if (params
->has_announce_max
) {
1373 s
->parameters
.announce_max
= params
->announce_max
;
1375 if (params
->has_announce_rounds
) {
1376 s
->parameters
.announce_rounds
= params
->announce_rounds
;
1378 if (params
->has_announce_step
) {
1379 s
->parameters
.announce_step
= params
->announce_step
;
1383 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1385 MigrationParameters tmp
;
1387 /* TODO Rewrite "" to null instead */
1388 if (params
->has_tls_creds
1389 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1390 qobject_unref(params
->tls_creds
->u
.n
);
1391 params
->tls_creds
->type
= QTYPE_QSTRING
;
1392 params
->tls_creds
->u
.s
= strdup("");
1394 /* TODO Rewrite "" to null instead */
1395 if (params
->has_tls_hostname
1396 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1397 qobject_unref(params
->tls_hostname
->u
.n
);
1398 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1399 params
->tls_hostname
->u
.s
= strdup("");
1402 migrate_params_test_apply(params
, &tmp
);
1404 if (!migrate_params_check(&tmp
, errp
)) {
1405 /* Invalid parameter */
1409 migrate_params_apply(params
, errp
);
1413 void qmp_migrate_start_postcopy(Error
**errp
)
1415 MigrationState
*s
= migrate_get_current();
1417 if (!migrate_postcopy()) {
1418 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1419 " the start of migration");
1423 if (s
->state
== MIGRATION_STATUS_NONE
) {
1424 error_setg(errp
, "Postcopy must be started after migration has been"
1429 * we don't error if migration has finished since that would be racy
1430 * with issuing this command.
1432 atomic_set(&s
->start_postcopy
, true);
1435 /* shared migration helpers */
1437 void migrate_set_state(int *state
, int old_state
, int new_state
)
1439 assert(new_state
< MIGRATION_STATUS__MAX
);
1440 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1441 trace_migrate_set_state(MigrationStatus_str(new_state
));
1442 migrate_generate_event(new_state
);
1446 static MigrationCapabilityStatusList
*migrate_cap_add(
1447 MigrationCapabilityStatusList
*list
,
1448 MigrationCapability index
,
1451 MigrationCapabilityStatusList
*cap
;
1453 cap
= g_new0(MigrationCapabilityStatusList
, 1);
1454 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
1455 cap
->value
->capability
= index
;
1456 cap
->value
->state
= state
;
1462 void migrate_set_block_enabled(bool value
, Error
**errp
)
1464 MigrationCapabilityStatusList
*cap
;
1466 cap
= migrate_cap_add(NULL
, MIGRATION_CAPABILITY_BLOCK
, value
);
1467 qmp_migrate_set_capabilities(cap
, errp
);
1468 qapi_free_MigrationCapabilityStatusList(cap
);
1471 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1473 s
->parameters
.block_incremental
= value
;
1476 static void block_cleanup_parameters(MigrationState
*s
)
1478 if (s
->must_remove_block_options
) {
1479 /* setting to false can never fail */
1480 migrate_set_block_enabled(false, &error_abort
);
1481 migrate_set_block_incremental(s
, false);
1482 s
->must_remove_block_options
= false;
1486 static void migrate_fd_cleanup(void *opaque
)
1488 MigrationState
*s
= opaque
;
1490 qemu_bh_delete(s
->cleanup_bh
);
1491 s
->cleanup_bh
= NULL
;
1493 qemu_savevm_state_cleanup();
1495 if (s
->to_dst_file
) {
1498 trace_migrate_fd_cleanup();
1499 qemu_mutex_unlock_iothread();
1500 if (s
->migration_thread_running
) {
1501 qemu_thread_join(&s
->thread
);
1502 s
->migration_thread_running
= false;
1504 qemu_mutex_lock_iothread();
1506 multifd_save_cleanup();
1507 qemu_mutex_lock(&s
->qemu_file_lock
);
1508 tmp
= s
->to_dst_file
;
1509 s
->to_dst_file
= NULL
;
1510 qemu_mutex_unlock(&s
->qemu_file_lock
);
1512 * Close the file handle without the lock to make sure the
1513 * critical section won't block for long.
1518 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
1519 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
1521 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1522 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1523 MIGRATION_STATUS_CANCELLED
);
1527 /* It is used on info migrate. We can't free it */
1528 error_report_err(error_copy(s
->error
));
1530 notifier_list_notify(&migration_state_notifiers
, s
);
1531 block_cleanup_parameters(s
);
1534 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1536 qemu_mutex_lock(&s
->error_mutex
);
1538 s
->error
= error_copy(error
);
1540 qemu_mutex_unlock(&s
->error_mutex
);
1543 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1545 trace_migrate_fd_error(error_get_pretty(error
));
1546 assert(s
->to_dst_file
== NULL
);
1547 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1548 MIGRATION_STATUS_FAILED
);
1549 migrate_set_error(s
, error
);
1552 static void migrate_fd_cancel(MigrationState
*s
)
1555 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1556 trace_migrate_fd_cancel();
1558 if (s
->rp_state
.from_dst_file
) {
1559 /* shutdown the rp socket, so causing the rp thread to shutdown */
1560 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1564 old_state
= s
->state
;
1565 if (!migration_is_setup_or_active(old_state
)) {
1568 /* If the migration is paused, kick it out of the pause */
1569 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1570 qemu_sem_post(&s
->pause_sem
);
1572 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1573 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1576 * If we're unlucky the migration code might be stuck somewhere in a
1577 * send/write while the network has failed and is waiting to timeout;
1578 * if we've got shutdown(2) available then we can force it to quit.
1579 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1580 * called in a bh, so there is no race against this cancel.
1582 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1583 qemu_file_shutdown(f
);
1585 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1586 Error
*local_err
= NULL
;
1588 bdrv_invalidate_cache_all(&local_err
);
1590 error_report_err(local_err
);
1592 s
->block_inactive
= false;
1597 void add_migration_state_change_notifier(Notifier
*notify
)
1599 notifier_list_add(&migration_state_notifiers
, notify
);
1602 void remove_migration_state_change_notifier(Notifier
*notify
)
1604 notifier_remove(notify
);
1607 bool migration_in_setup(MigrationState
*s
)
1609 return s
->state
== MIGRATION_STATUS_SETUP
;
1612 bool migration_has_finished(MigrationState
*s
)
1614 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1617 bool migration_has_failed(MigrationState
*s
)
1619 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1620 s
->state
== MIGRATION_STATUS_FAILED
);
1623 bool migration_in_postcopy(void)
1625 MigrationState
*s
= migrate_get_current();
1627 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1630 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1632 return migration_in_postcopy() && s
->postcopy_after_devices
;
1635 bool migration_is_idle(void)
1637 MigrationState
*s
= migrate_get_current();
1640 case MIGRATION_STATUS_NONE
:
1641 case MIGRATION_STATUS_CANCELLED
:
1642 case MIGRATION_STATUS_COMPLETED
:
1643 case MIGRATION_STATUS_FAILED
:
1645 case MIGRATION_STATUS_SETUP
:
1646 case MIGRATION_STATUS_CANCELLING
:
1647 case MIGRATION_STATUS_ACTIVE
:
1648 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1649 case MIGRATION_STATUS_COLO
:
1650 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1651 case MIGRATION_STATUS_DEVICE
:
1653 case MIGRATION_STATUS__MAX
:
1654 g_assert_not_reached();
1660 void migrate_init(MigrationState
*s
)
1663 * Reinitialise all migration state, except
1664 * parameters/capabilities that the user set, and
1670 s
->to_dst_file
= NULL
;
1671 s
->state
= MIGRATION_STATUS_NONE
;
1672 s
->rp_state
.from_dst_file
= NULL
;
1673 s
->rp_state
.error
= false;
1675 s
->pages_per_second
= 0.0;
1677 s
->expected_downtime
= 0;
1679 s
->start_postcopy
= false;
1680 s
->postcopy_after_devices
= false;
1681 s
->migration_thread_running
= false;
1682 error_free(s
->error
);
1685 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1687 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1689 s
->vm_was_running
= false;
1690 s
->iteration_initial_bytes
= 0;
1691 s
->threshold_size
= 0;
1694 static GSList
*migration_blockers
;
1696 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1698 if (migrate_get_current()->only_migratable
) {
1699 error_propagate_prepend(errp
, error_copy(reason
),
1700 "disallowing migration blocker "
1701 "(--only_migratable) for: ");
1705 if (migration_is_idle()) {
1706 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1710 error_propagate_prepend(errp
, error_copy(reason
),
1711 "disallowing migration blocker "
1712 "(migration in progress) for: ");
1716 void migrate_del_blocker(Error
*reason
)
1718 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1721 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1723 Error
*local_err
= NULL
;
1724 static bool once
= true;
1726 if (!deferred_incoming
) {
1727 error_setg(errp
, "For use with '-incoming defer'");
1731 error_setg(errp
, "The incoming migration has already been started");
1734 qemu_start_incoming_migration(uri
, &local_err
);
1737 error_propagate(errp
, local_err
);
1744 void qmp_migrate_recover(const char *uri
, Error
**errp
)
1746 MigrationIncomingState
*mis
= migration_incoming_get_current();
1748 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1749 error_setg(errp
, "Migrate recover can only be run "
1750 "when postcopy is paused.");
1754 if (atomic_cmpxchg(&mis
->postcopy_recover_triggered
,
1755 false, true) == true) {
1756 error_setg(errp
, "Migrate recovery is triggered already");
1761 * Note that this call will never start a real migration; it will
1762 * only re-setup the migration stream and poke existing migration
1763 * to continue using that newly established channel.
1765 qemu_start_incoming_migration(uri
, errp
);
1768 void qmp_migrate_pause(Error
**errp
)
1770 MigrationState
*ms
= migrate_get_current();
1771 MigrationIncomingState
*mis
= migration_incoming_get_current();
1774 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1775 /* Source side, during postcopy */
1776 qemu_mutex_lock(&ms
->qemu_file_lock
);
1777 ret
= qemu_file_shutdown(ms
->to_dst_file
);
1778 qemu_mutex_unlock(&ms
->qemu_file_lock
);
1780 error_setg(errp
, "Failed to pause source migration");
1785 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1786 ret
= qemu_file_shutdown(mis
->from_src_file
);
1788 error_setg(errp
, "Failed to pause destination migration");
1793 error_setg(errp
, "migrate-pause is currently only supported "
1794 "during postcopy-active state");
1797 bool migration_is_blocked(Error
**errp
)
1799 if (qemu_savevm_state_blocked(errp
)) {
1803 if (migration_blockers
) {
1804 error_propagate(errp
, error_copy(migration_blockers
->data
));
1811 /* Returns true if continue to migrate, or false if error detected */
1812 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
1813 bool resume
, Error
**errp
)
1815 Error
*local_err
= NULL
;
1818 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1819 error_setg(errp
, "Cannot resume if there is no "
1820 "paused migration");
1825 * Postcopy recovery won't work well with release-ram
1826 * capability since release-ram will drop the page buffer as
1827 * long as the page is put into the send buffer. So if there
1828 * is a network failure happened, any page buffers that have
1829 * not yet reached the destination VM but have already been
1830 * sent from the source VM will be lost forever. Let's refuse
1831 * the client from resuming such a postcopy migration.
1832 * Luckily release-ram was designed to only be used when src
1833 * and destination VMs are on the same host, so it should be
1836 if (migrate_release_ram()) {
1837 error_setg(errp
, "Postcopy recovery cannot work "
1838 "when release-ram capability is set");
1842 /* This is a resume, skip init status */
1846 if (migration_is_setup_or_active(s
->state
) ||
1847 s
->state
== MIGRATION_STATUS_CANCELLING
||
1848 s
->state
== MIGRATION_STATUS_COLO
) {
1849 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1853 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1854 error_setg(errp
, "Guest is waiting for an incoming migration");
1858 if (migration_is_blocked(errp
)) {
1862 if (blk
|| blk_inc
) {
1863 if (migrate_use_block() || migrate_use_block_incremental()) {
1864 error_setg(errp
, "Command options are incompatible with "
1865 "current migration capabilities");
1868 migrate_set_block_enabled(true, &local_err
);
1870 error_propagate(errp
, local_err
);
1873 s
->must_remove_block_options
= true;
1877 migrate_set_block_incremental(s
, true);
1885 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1886 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1887 bool has_resume
, bool resume
, Error
**errp
)
1889 Error
*local_err
= NULL
;
1890 MigrationState
*s
= migrate_get_current();
1893 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
1894 has_resume
&& resume
, errp
)) {
1895 /* Error detected, put into errp */
1899 if (strstart(uri
, "tcp:", &p
)) {
1900 tcp_start_outgoing_migration(s
, p
, &local_err
);
1902 } else if (strstart(uri
, "rdma:", &p
)) {
1903 rdma_start_outgoing_migration(s
, p
, &local_err
);
1905 } else if (strstart(uri
, "exec:", &p
)) {
1906 exec_start_outgoing_migration(s
, p
, &local_err
);
1907 } else if (strstart(uri
, "unix:", &p
)) {
1908 unix_start_outgoing_migration(s
, p
, &local_err
);
1909 } else if (strstart(uri
, "fd:", &p
)) {
1910 fd_start_outgoing_migration(s
, p
, &local_err
);
1912 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1913 "a valid migration protocol");
1914 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1915 MIGRATION_STATUS_FAILED
);
1916 block_cleanup_parameters(s
);
1921 migrate_fd_error(s
, local_err
);
1922 error_propagate(errp
, local_err
);
1927 void qmp_migrate_cancel(Error
**errp
)
1929 migrate_fd_cancel(migrate_get_current());
1932 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
1934 MigrationState
*s
= migrate_get_current();
1935 if (s
->state
!= state
) {
1936 error_setg(errp
, "Migration not in expected state: %s",
1937 MigrationStatus_str(s
->state
));
1940 qemu_sem_post(&s
->pause_sem
);
1943 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1945 MigrateSetParameters p
= {
1946 .has_xbzrle_cache_size
= true,
1947 .xbzrle_cache_size
= value
,
1950 qmp_migrate_set_parameters(&p
, errp
);
1953 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1955 return migrate_xbzrle_cache_size();
1958 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1960 MigrateSetParameters p
= {
1961 .has_max_bandwidth
= true,
1962 .max_bandwidth
= value
,
1965 qmp_migrate_set_parameters(&p
, errp
);
1968 void qmp_migrate_set_downtime(double value
, Error
**errp
)
1970 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
1971 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1972 "the range of 0 to %d seconds",
1973 MAX_MIGRATE_DOWNTIME_SECONDS
);
1977 value
*= 1000; /* Convert to milliseconds */
1978 value
= MAX(0, MIN(INT64_MAX
, value
));
1980 MigrateSetParameters p
= {
1981 .has_downtime_limit
= true,
1982 .downtime_limit
= value
,
1985 qmp_migrate_set_parameters(&p
, errp
);
1988 bool migrate_release_ram(void)
1992 s
= migrate_get_current();
1994 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
1997 bool migrate_postcopy_ram(void)
2001 s
= migrate_get_current();
2003 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
2006 bool migrate_postcopy(void)
2008 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2011 bool migrate_auto_converge(void)
2015 s
= migrate_get_current();
2017 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
2020 bool migrate_zero_blocks(void)
2024 s
= migrate_get_current();
2026 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
2029 bool migrate_postcopy_blocktime(void)
2033 s
= migrate_get_current();
2035 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
2038 bool migrate_use_compression(void)
2042 s
= migrate_get_current();
2044 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
2047 int migrate_compress_level(void)
2051 s
= migrate_get_current();
2053 return s
->parameters
.compress_level
;
2056 int migrate_compress_threads(void)
2060 s
= migrate_get_current();
2062 return s
->parameters
.compress_threads
;
2065 int migrate_compress_wait_thread(void)
2069 s
= migrate_get_current();
2071 return s
->parameters
.compress_wait_thread
;
2074 int migrate_decompress_threads(void)
2078 s
= migrate_get_current();
2080 return s
->parameters
.decompress_threads
;
2083 bool migrate_dirty_bitmaps(void)
2087 s
= migrate_get_current();
2089 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
2092 bool migrate_ignore_shared(void)
2096 s
= migrate_get_current();
2098 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
];
2101 bool migrate_use_events(void)
2105 s
= migrate_get_current();
2107 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
2110 bool migrate_use_multifd(void)
2114 s
= migrate_get_current();
2116 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_MULTIFD
];
2119 bool migrate_pause_before_switchover(void)
2123 s
= migrate_get_current();
2125 return s
->enabled_capabilities
[
2126 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
2129 int migrate_multifd_channels(void)
2133 s
= migrate_get_current();
2135 return s
->parameters
.x_multifd_channels
;
2138 int migrate_use_xbzrle(void)
2142 s
= migrate_get_current();
2144 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
2147 int64_t migrate_xbzrle_cache_size(void)
2151 s
= migrate_get_current();
2153 return s
->parameters
.xbzrle_cache_size
;
2156 static int64_t migrate_max_postcopy_bandwidth(void)
2160 s
= migrate_get_current();
2162 return s
->parameters
.max_postcopy_bandwidth
;
2165 bool migrate_use_block(void)
2169 s
= migrate_get_current();
2171 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
2174 bool migrate_use_return_path(void)
2178 s
= migrate_get_current();
2180 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
2183 bool migrate_use_block_incremental(void)
2187 s
= migrate_get_current();
2189 return s
->parameters
.block_incremental
;
2192 /* migration thread support */
2194 * Something bad happened to the RP stream, mark an error
2195 * The caller shall print or trace something to indicate why
2197 static void mark_source_rp_bad(MigrationState
*s
)
2199 s
->rp_state
.error
= true;
2202 static struct rp_cmd_args
{
2203 ssize_t len
; /* -1 = variable */
2206 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
2207 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
2208 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
2209 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
2210 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
2211 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
2212 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
2213 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
2217 * Process a request for pages received on the return path,
2218 * We're allowed to send more than requested (e.g. to round to our page size)
2219 * and we don't need to send pages that have already been sent.
2221 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
2222 ram_addr_t start
, size_t len
)
2224 long our_host_ps
= getpagesize();
2226 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
2229 * Since we currently insist on matching page sizes, just sanity check
2230 * we're being asked for whole host pages.
2232 if (start
& (our_host_ps
-1) ||
2233 (len
& (our_host_ps
-1))) {
2234 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2235 " len: %zd", __func__
, start
, len
);
2236 mark_source_rp_bad(ms
);
2240 if (ram_save_queue_pages(rbname
, start
, len
)) {
2241 mark_source_rp_bad(ms
);
2245 /* Return true to retry, false to quit */
2246 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
2248 trace_postcopy_pause_return_path();
2250 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
2252 trace_postcopy_pause_return_path_continued();
2257 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
2259 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
2262 error_report("%s: invalid block name '%s'", __func__
, block_name
);
2266 /* Fetch the received bitmap and refresh the dirty bitmap */
2267 return ram_dirty_bitmap_reload(s
, block
);
2270 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
2272 trace_source_return_path_thread_resume_ack(value
);
2274 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
2275 error_report("%s: illegal resume_ack value %"PRIu32
,
2280 /* Now both sides are active. */
2281 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2282 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2284 /* Notify send thread that time to continue send pages */
2285 qemu_sem_post(&s
->rp_state
.rp_sem
);
2291 * Handles messages sent on the return path towards the source VM
2294 static void *source_return_path_thread(void *opaque
)
2296 MigrationState
*ms
= opaque
;
2297 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
2298 uint16_t header_len
, header_type
;
2300 uint32_t tmp32
, sibling_error
;
2301 ram_addr_t start
= 0; /* =0 to silence warning */
2302 size_t len
= 0, expected_len
;
2305 trace_source_return_path_thread_entry();
2306 rcu_register_thread();
2309 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
2310 migration_is_setup_or_active(ms
->state
)) {
2311 trace_source_return_path_thread_loop_top();
2312 header_type
= qemu_get_be16(rp
);
2313 header_len
= qemu_get_be16(rp
);
2315 if (qemu_file_get_error(rp
)) {
2316 mark_source_rp_bad(ms
);
2320 if (header_type
>= MIG_RP_MSG_MAX
||
2321 header_type
== MIG_RP_MSG_INVALID
) {
2322 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2323 header_type
, header_len
);
2324 mark_source_rp_bad(ms
);
2328 if ((rp_cmd_args
[header_type
].len
!= -1 &&
2329 header_len
!= rp_cmd_args
[header_type
].len
) ||
2330 header_len
> sizeof(buf
)) {
2331 error_report("RP: Received '%s' message (0x%04x) with"
2332 "incorrect length %d expecting %zu",
2333 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
2334 (size_t)rp_cmd_args
[header_type
].len
);
2335 mark_source_rp_bad(ms
);
2339 /* We know we've got a valid header by this point */
2340 res
= qemu_get_buffer(rp
, buf
, header_len
);
2341 if (res
!= header_len
) {
2342 error_report("RP: Failed reading data for message 0x%04x"
2343 " read %d expected %d",
2344 header_type
, res
, header_len
);
2345 mark_source_rp_bad(ms
);
2349 /* OK, we have the message and the data */
2350 switch (header_type
) {
2351 case MIG_RP_MSG_SHUT
:
2352 sibling_error
= ldl_be_p(buf
);
2353 trace_source_return_path_thread_shut(sibling_error
);
2354 if (sibling_error
) {
2355 error_report("RP: Sibling indicated error %d", sibling_error
);
2356 mark_source_rp_bad(ms
);
2359 * We'll let the main thread deal with closing the RP
2360 * we could do a shutdown(2) on it, but we're the only user
2361 * anyway, so there's nothing gained.
2365 case MIG_RP_MSG_PONG
:
2366 tmp32
= ldl_be_p(buf
);
2367 trace_source_return_path_thread_pong(tmp32
);
2370 case MIG_RP_MSG_REQ_PAGES
:
2371 start
= ldq_be_p(buf
);
2372 len
= ldl_be_p(buf
+ 8);
2373 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
2376 case MIG_RP_MSG_REQ_PAGES_ID
:
2377 expected_len
= 12 + 1; /* header + termination */
2379 if (header_len
>= expected_len
) {
2380 start
= ldq_be_p(buf
);
2381 len
= ldl_be_p(buf
+ 8);
2382 /* Now we expect an idstr */
2383 tmp32
= buf
[12]; /* Length of the following idstr */
2384 buf
[13 + tmp32
] = '\0';
2385 expected_len
+= tmp32
;
2387 if (header_len
!= expected_len
) {
2388 error_report("RP: Req_Page_id with length %d expecting %zd",
2389 header_len
, expected_len
);
2390 mark_source_rp_bad(ms
);
2393 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
2396 case MIG_RP_MSG_RECV_BITMAP
:
2397 if (header_len
< 1) {
2398 error_report("%s: missing block name", __func__
);
2399 mark_source_rp_bad(ms
);
2402 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2403 buf
[buf
[0] + 1] = '\0';
2404 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
2405 mark_source_rp_bad(ms
);
2410 case MIG_RP_MSG_RESUME_ACK
:
2411 tmp32
= ldl_be_p(buf
);
2412 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
2413 mark_source_rp_bad(ms
);
2424 res
= qemu_file_get_error(rp
);
2428 * Maybe there is something we can do: it looks like a
2429 * network down issue, and we pause for a recovery.
2431 if (postcopy_pause_return_path_thread(ms
)) {
2432 /* Reload rp, reset the rest */
2433 if (rp
!= ms
->rp_state
.from_dst_file
) {
2435 rp
= ms
->rp_state
.from_dst_file
;
2437 ms
->rp_state
.error
= false;
2442 trace_source_return_path_thread_bad_end();
2443 mark_source_rp_bad(ms
);
2446 trace_source_return_path_thread_end();
2447 ms
->rp_state
.from_dst_file
= NULL
;
2449 rcu_unregister_thread();
2453 static int open_return_path_on_source(MigrationState
*ms
,
2457 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
2458 if (!ms
->rp_state
.from_dst_file
) {
2462 trace_open_return_path_on_source();
2464 if (!create_thread
) {
2469 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
2470 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
2472 trace_open_return_path_on_source_continue();
2477 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2478 static int await_return_path_close_on_source(MigrationState
*ms
)
2481 * If this is a normal exit then the destination will send a SHUT and the
2482 * rp_thread will exit, however if there's an error we need to cause
2485 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
2487 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2488 * waiting for the destination.
2490 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
2491 mark_source_rp_bad(ms
);
2493 trace_await_return_path_close_on_source_joining();
2494 qemu_thread_join(&ms
->rp_state
.rp_thread
);
2495 trace_await_return_path_close_on_source_close();
2496 return ms
->rp_state
.error
;
2500 * Switch from normal iteration to postcopy
2501 * Returns non-0 on error
2503 static int postcopy_start(MigrationState
*ms
)
2506 QIOChannelBuffer
*bioc
;
2508 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2509 int64_t bandwidth
= migrate_max_postcopy_bandwidth();
2510 bool restart_block
= false;
2511 int cur_state
= MIGRATION_STATUS_ACTIVE
;
2512 if (!migrate_pause_before_switchover()) {
2513 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
2514 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2517 trace_postcopy_start();
2518 qemu_mutex_lock_iothread();
2519 trace_postcopy_start_set_run();
2521 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2522 global_state_store();
2523 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2528 ret
= migration_maybe_pause(ms
, &cur_state
,
2529 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2534 ret
= bdrv_inactivate_all();
2538 restart_block
= true;
2541 * Cause any non-postcopiable, but iterative devices to
2542 * send out their final data.
2544 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2547 * in Finish migrate and with the io-lock held everything should
2548 * be quiet, but we've potentially still got dirty pages and we
2549 * need to tell the destination to throw any pages it's already received
2552 if (migrate_postcopy_ram()) {
2553 if (ram_postcopy_send_discard_bitmap(ms
)) {
2554 error_report("postcopy send discard bitmap failed");
2560 * send rest of state - note things that are doing postcopy
2561 * will notice we're in POSTCOPY_ACTIVE and not actually
2562 * wrap their state up here
2564 /* 0 max-postcopy-bandwidth means unlimited */
2566 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
2568 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
2570 if (migrate_postcopy_ram()) {
2571 /* Ping just for debugging, helps line traces up */
2572 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
2576 * While loading the device state we may trigger page transfer
2577 * requests and the fd must be free to process those, and thus
2578 * the destination must read the whole device state off the fd before
2579 * it starts processing it. Unfortunately the ad-hoc migration format
2580 * doesn't allow the destination to know the size to read without fully
2581 * parsing it through each devices load-state code (especially the open
2582 * coded devices that use get/put).
2583 * So we wrap the device state up in a package with a length at the start;
2584 * to do this we use a qemu_buf to hold the whole of the device state.
2586 bioc
= qio_channel_buffer_new(4096);
2587 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
2588 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
2589 object_unref(OBJECT(bioc
));
2592 * Make sure the receiver can get incoming pages before we send the rest
2595 qemu_savevm_send_postcopy_listen(fb
);
2597 qemu_savevm_state_complete_precopy(fb
, false, false);
2598 if (migrate_postcopy_ram()) {
2599 qemu_savevm_send_ping(fb
, 3);
2602 qemu_savevm_send_postcopy_run(fb
);
2604 /* <><> end of stuff going into the package */
2606 /* Last point of recovery; as soon as we send the package the destination
2607 * can open devices and potentially start running.
2608 * Lets just check again we've not got any errors.
2610 ret
= qemu_file_get_error(ms
->to_dst_file
);
2612 error_report("postcopy_start: Migration stream errored (pre package)");
2616 restart_block
= false;
2618 /* Now send that blob */
2619 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2624 /* Send a notify to give a chance for anything that needs to happen
2625 * at the transition to postcopy and after the device state; in particular
2626 * spice needs to trigger a transition now
2628 ms
->postcopy_after_devices
= true;
2629 notifier_list_notify(&migration_state_notifiers
, ms
);
2631 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2633 qemu_mutex_unlock_iothread();
2635 if (migrate_postcopy_ram()) {
2637 * Although this ping is just for debug, it could potentially be
2638 * used for getting a better measurement of downtime at the source.
2640 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2643 if (migrate_release_ram()) {
2644 ram_postcopy_migrated_memory_release(ms
);
2647 ret
= qemu_file_get_error(ms
->to_dst_file
);
2649 error_report("postcopy_start: Migration stream errored");
2650 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2651 MIGRATION_STATUS_FAILED
);
2659 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2660 MIGRATION_STATUS_FAILED
);
2661 if (restart_block
) {
2662 /* A failure happened early enough that we know the destination hasn't
2663 * accessed block devices, so we're safe to recover.
2665 Error
*local_err
= NULL
;
2667 bdrv_invalidate_cache_all(&local_err
);
2669 error_report_err(local_err
);
2672 qemu_mutex_unlock_iothread();
2677 * migration_maybe_pause: Pause if required to by
2678 * migrate_pause_before_switchover called with the iothread locked
2679 * Returns: 0 on success
2681 static int migration_maybe_pause(MigrationState
*s
,
2682 int *current_active_state
,
2685 if (!migrate_pause_before_switchover()) {
2689 /* Since leaving this state is not atomic with posting the semaphore
2690 * it's possible that someone could have issued multiple migrate_continue
2691 * and the semaphore is incorrectly positive at this point;
2692 * the docs say it's undefined to reinit a semaphore that's already
2693 * init'd, so use timedwait to eat up any existing posts.
2695 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2696 /* This block intentionally left blank */
2699 qemu_mutex_unlock_iothread();
2700 migrate_set_state(&s
->state
, *current_active_state
,
2701 MIGRATION_STATUS_PRE_SWITCHOVER
);
2702 qemu_sem_wait(&s
->pause_sem
);
2703 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2705 *current_active_state
= new_state
;
2706 qemu_mutex_lock_iothread();
2708 return s
->state
== new_state
? 0 : -EINVAL
;
2712 * migration_completion: Used by migration_thread when there's not much left.
2713 * The caller 'breaks' the loop when this returns.
2715 * @s: Current migration state
2717 static void migration_completion(MigrationState
*s
)
2720 int current_active_state
= s
->state
;
2722 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2723 qemu_mutex_lock_iothread();
2724 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2725 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2726 s
->vm_was_running
= runstate_is_running();
2727 ret
= global_state_store();
2730 bool inactivate
= !migrate_colo_enabled();
2731 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2733 ret
= migration_maybe_pause(s
, ¤t_active_state
,
2734 MIGRATION_STATUS_DEVICE
);
2737 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
2738 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
2741 if (inactivate
&& ret
>= 0) {
2742 s
->block_inactive
= true;
2745 qemu_mutex_unlock_iothread();
2750 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2751 trace_migration_completion_postcopy_end();
2753 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
2754 trace_migration_completion_postcopy_end_after_complete();
2758 * If rp was opened we must clean up the thread before
2759 * cleaning everything else up (since if there are no failures
2760 * it will wait for the destination to send it's status in
2763 if (s
->rp_state
.from_dst_file
) {
2765 trace_migration_return_path_end_before();
2766 rp_error
= await_return_path_close_on_source(s
);
2767 trace_migration_return_path_end_after(rp_error
);
2769 goto fail_invalidate
;
2773 if (qemu_file_get_error(s
->to_dst_file
)) {
2774 trace_migration_completion_file_err();
2775 goto fail_invalidate
;
2778 if (!migrate_colo_enabled()) {
2779 migrate_set_state(&s
->state
, current_active_state
,
2780 MIGRATION_STATUS_COMPLETED
);
2786 /* If not doing postcopy, vm_start() will be called: let's regain
2787 * control on images.
2789 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
2790 s
->state
== MIGRATION_STATUS_DEVICE
) {
2791 Error
*local_err
= NULL
;
2793 qemu_mutex_lock_iothread();
2794 bdrv_invalidate_cache_all(&local_err
);
2796 error_report_err(local_err
);
2798 s
->block_inactive
= false;
2800 qemu_mutex_unlock_iothread();
2804 migrate_set_state(&s
->state
, current_active_state
,
2805 MIGRATION_STATUS_FAILED
);
2808 bool migrate_colo_enabled(void)
2810 MigrationState
*s
= migrate_get_current();
2811 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
2814 typedef enum MigThrError
{
2815 /* No error detected */
2816 MIG_THR_ERR_NONE
= 0,
2817 /* Detected error, but resumed successfully */
2818 MIG_THR_ERR_RECOVERED
= 1,
2819 /* Detected fatal error, need to exit */
2820 MIG_THR_ERR_FATAL
= 2,
2823 static int postcopy_resume_handshake(MigrationState
*s
)
2825 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
2827 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2828 qemu_sem_wait(&s
->rp_state
.rp_sem
);
2831 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2838 /* Return zero if success, or <0 for error */
2839 static int postcopy_do_resume(MigrationState
*s
)
2844 * Call all the resume_prepare() hooks, so that modules can be
2845 * ready for the migration resume.
2847 ret
= qemu_savevm_state_resume_prepare(s
);
2849 error_report("%s: resume_prepare() failure detected: %d",
2855 * Last handshake with destination on the resume (destination will
2856 * switch to postcopy-active afterwards)
2858 ret
= postcopy_resume_handshake(s
);
2860 error_report("%s: handshake failed: %d", __func__
, ret
);
2868 * We don't return until we are in a safe state to continue current
2869 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2870 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2872 static MigThrError
postcopy_pause(MigrationState
*s
)
2874 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2879 migrate_set_state(&s
->state
, s
->state
,
2880 MIGRATION_STATUS_POSTCOPY_PAUSED
);
2882 /* Current channel is possibly broken. Release it. */
2883 assert(s
->to_dst_file
);
2884 qemu_mutex_lock(&s
->qemu_file_lock
);
2885 file
= s
->to_dst_file
;
2886 s
->to_dst_file
= NULL
;
2887 qemu_mutex_unlock(&s
->qemu_file_lock
);
2889 qemu_file_shutdown(file
);
2892 error_report("Detected IO failure for postcopy. "
2893 "Migration paused.");
2896 * We wait until things fixed up. Then someone will setup the
2897 * status back for us.
2899 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2900 qemu_sem_wait(&s
->postcopy_pause_sem
);
2903 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2904 /* Woken up by a recover procedure. Give it a shot */
2907 * Firstly, let's wake up the return path now, with a new
2908 * return path channel.
2910 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
2912 /* Do the resume logic */
2913 if (postcopy_do_resume(s
) == 0) {
2914 /* Let's continue! */
2915 trace_postcopy_pause_continued();
2916 return MIG_THR_ERR_RECOVERED
;
2919 * Something wrong happened during the recovery, let's
2920 * pause again. Pause is always better than throwing
2926 /* This is not right... Time to quit. */
2927 return MIG_THR_ERR_FATAL
;
2932 static MigThrError
migration_detect_error(MigrationState
*s
)
2935 int state
= s
->state
;
2937 if (state
== MIGRATION_STATUS_CANCELLING
||
2938 state
== MIGRATION_STATUS_CANCELLED
) {
2939 /* End the migration, but don't set the state to failed */
2940 return MIG_THR_ERR_FATAL
;
2943 /* Try to detect any file errors */
2944 ret
= qemu_file_get_error(s
->to_dst_file
);
2947 /* Everything is fine */
2948 return MIG_THR_ERR_NONE
;
2951 if (state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
== -EIO
) {
2953 * For postcopy, we allow the network to be down for a
2954 * while. After that, it can be continued by a
2957 return postcopy_pause(s
);
2960 * For precopy (or postcopy with error outside IO), we fail
2963 migrate_set_state(&s
->state
, state
, MIGRATION_STATUS_FAILED
);
2964 trace_migration_thread_file_err();
2966 /* Time to stop the migration, now. */
2967 return MIG_THR_ERR_FATAL
;
2971 /* How many bytes have we transferred since the beggining of the migration */
2972 static uint64_t migration_total_bytes(MigrationState
*s
)
2974 return qemu_ftell(s
->to_dst_file
) + ram_counters
.multifd_bytes
;
2977 static void migration_calculate_complete(MigrationState
*s
)
2979 uint64_t bytes
= migration_total_bytes(s
);
2980 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2981 int64_t transfer_time
;
2983 s
->total_time
= end_time
- s
->start_time
;
2986 * It's still not set, so we are precopy migration. For
2987 * postcopy, downtime is calculated during postcopy_start().
2989 s
->downtime
= end_time
- s
->downtime_start
;
2992 transfer_time
= s
->total_time
- s
->setup_time
;
2993 if (transfer_time
) {
2994 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
2998 static void migration_update_counters(MigrationState
*s
,
2999 int64_t current_time
)
3001 uint64_t transferred
, transferred_pages
, time_spent
;
3002 uint64_t current_bytes
; /* bytes transferred since the beginning */
3005 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
3009 current_bytes
= migration_total_bytes(s
);
3010 transferred
= current_bytes
- s
->iteration_initial_bytes
;
3011 time_spent
= current_time
- s
->iteration_start_time
;
3012 bandwidth
= (double)transferred
/ time_spent
;
3013 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
3015 s
->mbps
= (((double) transferred
* 8.0) /
3016 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
3018 transferred_pages
= ram_get_total_transferred_pages() -
3019 s
->iteration_initial_pages
;
3020 s
->pages_per_second
= (double) transferred_pages
/
3021 (((double) time_spent
/ 1000.0));
3024 * if we haven't sent anything, we don't want to
3025 * recalculate. 10000 is a small enough number for our purposes
3027 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
3028 s
->expected_downtime
= ram_counters
.remaining
/ bandwidth
;
3031 qemu_file_reset_rate_limit(s
->to_dst_file
);
3033 s
->iteration_start_time
= current_time
;
3034 s
->iteration_initial_bytes
= current_bytes
;
3035 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
3037 trace_migrate_transferred(transferred
, time_spent
,
3038 bandwidth
, s
->threshold_size
);
3041 /* Migration thread iteration status */
3043 MIG_ITERATE_RESUME
, /* Resume current iteration */
3044 MIG_ITERATE_SKIP
, /* Skip current iteration */
3045 MIG_ITERATE_BREAK
, /* Break the loop */
3049 * Return true if continue to the next iteration directly, false
3052 static MigIterateState
migration_iteration_run(MigrationState
*s
)
3054 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
3055 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
3057 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
3058 &pend_compat
, &pend_post
);
3059 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
3061 trace_migrate_pending(pending_size
, s
->threshold_size
,
3062 pend_pre
, pend_compat
, pend_post
);
3064 if (pending_size
&& pending_size
>= s
->threshold_size
) {
3065 /* Still a significant amount to transfer */
3066 if (migrate_postcopy() && !in_postcopy
&&
3067 pend_pre
<= s
->threshold_size
&&
3068 atomic_read(&s
->start_postcopy
)) {
3069 if (postcopy_start(s
)) {
3070 error_report("%s: postcopy failed to start", __func__
);
3072 return MIG_ITERATE_SKIP
;
3074 /* Just another iteration step */
3075 qemu_savevm_state_iterate(s
->to_dst_file
,
3076 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3078 trace_migration_thread_low_pending(pending_size
);
3079 migration_completion(s
);
3080 return MIG_ITERATE_BREAK
;
3083 return MIG_ITERATE_RESUME
;
3086 static void migration_iteration_finish(MigrationState
*s
)
3088 /* If we enabled cpu throttling for auto-converge, turn it off. */
3089 cpu_throttle_stop();
3091 qemu_mutex_lock_iothread();
3093 case MIGRATION_STATUS_COMPLETED
:
3094 migration_calculate_complete(s
);
3095 runstate_set(RUN_STATE_POSTMIGRATE
);
3098 case MIGRATION_STATUS_ACTIVE
:
3100 * We should really assert here, but since it's during
3101 * migration, let's try to reduce the usage of assertions.
3103 if (!migrate_colo_enabled()) {
3104 error_report("%s: critical error: calling COLO code without "
3105 "COLO enabled", __func__
);
3107 migrate_start_colo_process(s
);
3109 * Fixme: we will run VM in COLO no matter its old running state.
3110 * After exited COLO, we will keep running.
3112 s
->vm_was_running
= true;
3114 case MIGRATION_STATUS_FAILED
:
3115 case MIGRATION_STATUS_CANCELLED
:
3116 case MIGRATION_STATUS_CANCELLING
:
3117 if (s
->vm_was_running
) {
3120 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
3121 runstate_set(RUN_STATE_POSTMIGRATE
);
3127 /* Should not reach here, but if so, forgive the VM. */
3128 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3131 qemu_bh_schedule(s
->cleanup_bh
);
3132 qemu_mutex_unlock_iothread();
3135 void migration_make_urgent_request(void)
3137 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
3140 void migration_consume_urgent_request(void)
3142 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
3146 * Master migration thread on the source VM.
3147 * It drives the migration and pumps the data down the outgoing channel.
3149 static void *migration_thread(void *opaque
)
3151 MigrationState
*s
= opaque
;
3152 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3153 MigThrError thr_error
;
3154 bool urgent
= false;
3156 rcu_register_thread();
3158 object_ref(OBJECT(s
));
3159 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3161 qemu_savevm_state_header(s
->to_dst_file
);
3164 * If we opened the return path, we need to make sure dst has it
3167 if (s
->rp_state
.from_dst_file
) {
3168 /* Now tell the dest that it should open its end so it can reply */
3169 qemu_savevm_send_open_return_path(s
->to_dst_file
);
3171 /* And do a ping that will make stuff easier to debug */
3172 qemu_savevm_send_ping(s
->to_dst_file
, 1);
3175 if (migrate_postcopy()) {
3177 * Tell the destination that we *might* want to do postcopy later;
3178 * if the other end can't do postcopy it should fail now, nice and
3181 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
3184 if (migrate_colo_enabled()) {
3185 /* Notify migration destination that we enable COLO */
3186 qemu_savevm_send_colo_enable(s
->to_dst_file
);
3189 qemu_savevm_state_setup(s
->to_dst_file
);
3191 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3192 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3193 MIGRATION_STATUS_ACTIVE
);
3195 trace_migration_thread_setup_complete();
3197 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
3198 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3199 int64_t current_time
;
3201 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
3202 MigIterateState iter_state
= migration_iteration_run(s
);
3203 if (iter_state
== MIG_ITERATE_SKIP
) {
3205 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3211 * Try to detect any kind of failures, and see whether we
3212 * should stop the migration now.
3214 thr_error
= migration_detect_error(s
);
3215 if (thr_error
== MIG_THR_ERR_FATAL
) {
3216 /* Stop migration */
3218 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
3220 * Just recovered from a e.g. network failure, reset all
3221 * the local variables. This is important to avoid
3222 * breaking transferred_bytes and bandwidth calculation
3224 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3225 s
->iteration_initial_bytes
= 0;
3228 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3230 migration_update_counters(s
, current_time
);
3233 if (qemu_file_rate_limit(s
->to_dst_file
)) {
3234 /* Wait for a delay to do rate limiting OR
3235 * something urgent to post the semaphore.
3237 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- current_time
;
3238 trace_migration_thread_ratelimit_pre(ms
);
3239 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
3240 /* We were worken by one or more urgent things but
3241 * the timedwait will have consumed one of them.
3242 * The service routine for the urgent wake will dec
3243 * the semaphore itself for each item it consumes,
3244 * so add this one we just eat back.
3246 qemu_sem_post(&s
->rate_limit_sem
);
3249 trace_migration_thread_ratelimit_post(urgent
);
3253 trace_migration_thread_after_loop();
3254 migration_iteration_finish(s
);
3255 object_unref(OBJECT(s
));
3256 rcu_unregister_thread();
3260 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
3263 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
3265 s
->expected_downtime
= s
->parameters
.downtime_limit
;
3266 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup
, s
);
3268 migrate_fd_error(s
, error_in
);
3269 migrate_fd_cleanup(s
);
3274 /* This is a resumed migration */
3275 rate_limit
= INT64_MAX
;
3277 /* This is a fresh new migration */
3278 rate_limit
= s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
;
3280 /* Notify before starting migration thread */
3281 notifier_list_notify(&migration_state_notifiers
, s
);
3284 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
3285 qemu_file_set_blocking(s
->to_dst_file
, true);
3288 * Open the return path. For postcopy, it is used exclusively. For
3289 * precopy, only if user specified "return-path" capability would
3290 * QEMU uses the return path.
3292 if (migrate_postcopy_ram() || migrate_use_return_path()) {
3293 if (open_return_path_on_source(s
, !resume
)) {
3294 error_report("Unable to open return-path for postcopy");
3295 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
3296 migrate_fd_cleanup(s
);
3302 /* Wakeup the main migration thread to do the recovery */
3303 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
3304 MIGRATION_STATUS_POSTCOPY_RECOVER
);
3305 qemu_sem_post(&s
->postcopy_pause_sem
);
3309 if (multifd_save_setup() != 0) {
3310 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3311 MIGRATION_STATUS_FAILED
);
3312 migrate_fd_cleanup(s
);
3315 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
3316 QEMU_THREAD_JOINABLE
);
3317 s
->migration_thread_running
= true;
3320 void migration_global_dump(Monitor
*mon
)
3322 MigrationState
*ms
= migrate_get_current();
3324 monitor_printf(mon
, "globals:\n");
3325 monitor_printf(mon
, "store-global-state: %s\n",
3326 ms
->store_global_state
? "on" : "off");
3327 monitor_printf(mon
, "only-migratable: %s\n",
3328 ms
->only_migratable
? "on" : "off");
3329 monitor_printf(mon
, "send-configuration: %s\n",
3330 ms
->send_configuration
? "on" : "off");
3331 monitor_printf(mon
, "send-section-footer: %s\n",
3332 ms
->send_section_footer
? "on" : "off");
3333 monitor_printf(mon
, "decompress-error-check: %s\n",
3334 ms
->decompress_error_check
? "on" : "off");
3337 #define DEFINE_PROP_MIG_CAP(name, x) \
3338 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3340 static Property migration_properties
[] = {
3341 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
3342 store_global_state
, true),
3343 DEFINE_PROP_BOOL("only-migratable", MigrationState
, only_migratable
, false),
3344 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
3345 send_configuration
, true),
3346 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
3347 send_section_footer
, true),
3348 DEFINE_PROP_BOOL("decompress-error-check", MigrationState
,
3349 decompress_error_check
, true),
3351 /* Migration parameters */
3352 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
3353 parameters
.compress_level
,
3354 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
3355 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
3356 parameters
.compress_threads
,
3357 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
3358 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState
,
3359 parameters
.compress_wait_thread
, true),
3360 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
3361 parameters
.decompress_threads
,
3362 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
3363 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
3364 parameters
.cpu_throttle_initial
,
3365 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
3366 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
3367 parameters
.cpu_throttle_increment
,
3368 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
3369 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
3370 parameters
.max_bandwidth
, MAX_THROTTLE
),
3371 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
3372 parameters
.downtime_limit
,
3373 DEFAULT_MIGRATE_SET_DOWNTIME
),
3374 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
3375 parameters
.x_checkpoint_delay
,
3376 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
3377 DEFINE_PROP_UINT8("x-multifd-channels", MigrationState
,
3378 parameters
.x_multifd_channels
,
3379 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
3380 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
3381 parameters
.xbzrle_cache_size
,
3382 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
3383 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState
,
3384 parameters
.max_postcopy_bandwidth
,
3385 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH
),
3386 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState
,
3387 parameters
.max_cpu_throttle
,
3388 DEFAULT_MIGRATE_MAX_CPU_THROTTLE
),
3389 DEFINE_PROP_SIZE("announce-initial", MigrationState
,
3390 parameters
.announce_initial
,
3391 DEFAULT_MIGRATE_ANNOUNCE_INITIAL
),
3392 DEFINE_PROP_SIZE("announce-max", MigrationState
,
3393 parameters
.announce_max
,
3394 DEFAULT_MIGRATE_ANNOUNCE_MAX
),
3395 DEFINE_PROP_SIZE("announce-rounds", MigrationState
,
3396 parameters
.announce_rounds
,
3397 DEFAULT_MIGRATE_ANNOUNCE_ROUNDS
),
3398 DEFINE_PROP_SIZE("announce-step", MigrationState
,
3399 parameters
.announce_step
,
3400 DEFAULT_MIGRATE_ANNOUNCE_STEP
),
3402 /* Migration capabilities */
3403 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
3404 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
3405 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
3406 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
3407 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
3408 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
3409 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
3410 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
3411 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
3412 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
3413 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
3414 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_X_MULTIFD
),
3416 DEFINE_PROP_END_OF_LIST(),
3419 static void migration_class_init(ObjectClass
*klass
, void *data
)
3421 DeviceClass
*dc
= DEVICE_CLASS(klass
);
3423 dc
->user_creatable
= false;
3424 dc
->props
= migration_properties
;
3427 static void migration_instance_finalize(Object
*obj
)
3429 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3430 MigrationParameters
*params
= &ms
->parameters
;
3432 qemu_mutex_destroy(&ms
->error_mutex
);
3433 qemu_mutex_destroy(&ms
->qemu_file_lock
);
3434 g_free(params
->tls_hostname
);
3435 g_free(params
->tls_creds
);
3436 qemu_sem_destroy(&ms
->rate_limit_sem
);
3437 qemu_sem_destroy(&ms
->pause_sem
);
3438 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
3439 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
3440 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
3441 error_free(ms
->error
);
3444 static void migration_instance_init(Object
*obj
)
3446 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3447 MigrationParameters
*params
= &ms
->parameters
;
3449 ms
->state
= MIGRATION_STATUS_NONE
;
3451 ms
->pages_per_second
= -1;
3452 qemu_sem_init(&ms
->pause_sem
, 0);
3453 qemu_mutex_init(&ms
->error_mutex
);
3455 params
->tls_hostname
= g_strdup("");
3456 params
->tls_creds
= g_strdup("");
3458 /* Set has_* up only for parameter checks */
3459 params
->has_compress_level
= true;
3460 params
->has_compress_threads
= true;
3461 params
->has_decompress_threads
= true;
3462 params
->has_cpu_throttle_initial
= true;
3463 params
->has_cpu_throttle_increment
= true;
3464 params
->has_max_bandwidth
= true;
3465 params
->has_downtime_limit
= true;
3466 params
->has_x_checkpoint_delay
= true;
3467 params
->has_block_incremental
= true;
3468 params
->has_x_multifd_channels
= true;
3469 params
->has_xbzrle_cache_size
= true;
3470 params
->has_max_postcopy_bandwidth
= true;
3471 params
->has_max_cpu_throttle
= true;
3472 params
->has_announce_initial
= true;
3473 params
->has_announce_max
= true;
3474 params
->has_announce_rounds
= true;
3475 params
->has_announce_step
= true;
3477 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
3478 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
3479 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
3480 qemu_sem_init(&ms
->rate_limit_sem
, 0);
3481 qemu_mutex_init(&ms
->qemu_file_lock
);
3485 * Return true if check pass, false otherwise. Error will be put
3486 * inside errp if provided.
3488 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
3490 MigrationCapabilityStatusList
*head
= NULL
;
3491 /* Assuming all off */
3492 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
3495 if (!migrate_params_check(&ms
->parameters
, errp
)) {
3499 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
3500 if (ms
->enabled_capabilities
[i
]) {
3501 head
= migrate_cap_add(head
, i
, true);
3505 ret
= migrate_caps_check(cap_list
, head
, errp
);
3507 /* It works with head == NULL */
3508 qapi_free_MigrationCapabilityStatusList(head
);
3513 static const TypeInfo migration_type
= {
3514 .name
= TYPE_MIGRATION
,
3516 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3517 * not created using qdev_create(), it is not attached to the qdev
3518 * device tree, and it is never realized.
3520 * TODO: Make this TYPE_OBJECT once QOM provides something like
3521 * TYPE_DEVICE's "-global" properties.
3523 .parent
= TYPE_DEVICE
,
3524 .class_init
= migration_class_init
,
3525 .class_size
= sizeof(MigrationClass
),
3526 .instance_size
= sizeof(MigrationState
),
3527 .instance_init
= migration_instance_init
,
3528 .instance_finalize
= migration_instance_finalize
,
3531 static void register_migration_types(void)
3533 type_register_static(&migration_type
);
3536 type_init(register_migration_types
);