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_tls_authz
= true;
761 params
->tls_authz
= g_strdup(s
->parameters
.tls_authz
);
762 params
->has_max_bandwidth
= true;
763 params
->max_bandwidth
= s
->parameters
.max_bandwidth
;
764 params
->has_downtime_limit
= true;
765 params
->downtime_limit
= s
->parameters
.downtime_limit
;
766 params
->has_x_checkpoint_delay
= true;
767 params
->x_checkpoint_delay
= s
->parameters
.x_checkpoint_delay
;
768 params
->has_block_incremental
= true;
769 params
->block_incremental
= s
->parameters
.block_incremental
;
770 params
->has_multifd_channels
= true;
771 params
->multifd_channels
= s
->parameters
.multifd_channels
;
772 params
->has_xbzrle_cache_size
= true;
773 params
->xbzrle_cache_size
= s
->parameters
.xbzrle_cache_size
;
774 params
->has_max_postcopy_bandwidth
= true;
775 params
->max_postcopy_bandwidth
= s
->parameters
.max_postcopy_bandwidth
;
776 params
->has_max_cpu_throttle
= true;
777 params
->max_cpu_throttle
= s
->parameters
.max_cpu_throttle
;
778 params
->has_announce_initial
= true;
779 params
->announce_initial
= s
->parameters
.announce_initial
;
780 params
->has_announce_max
= true;
781 params
->announce_max
= s
->parameters
.announce_max
;
782 params
->has_announce_rounds
= true;
783 params
->announce_rounds
= s
->parameters
.announce_rounds
;
784 params
->has_announce_step
= true;
785 params
->announce_step
= s
->parameters
.announce_step
;
790 AnnounceParameters
*migrate_announce_params(void)
792 static AnnounceParameters ap
;
794 MigrationState
*s
= migrate_get_current();
796 ap
.initial
= s
->parameters
.announce_initial
;
797 ap
.max
= s
->parameters
.announce_max
;
798 ap
.rounds
= s
->parameters
.announce_rounds
;
799 ap
.step
= s
->parameters
.announce_step
;
805 * Return true if we're already in the middle of a migration
806 * (i.e. any of the active or setup states)
808 bool migration_is_setup_or_active(int state
)
811 case MIGRATION_STATUS_ACTIVE
:
812 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
813 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
814 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
815 case MIGRATION_STATUS_SETUP
:
816 case MIGRATION_STATUS_PRE_SWITCHOVER
:
817 case MIGRATION_STATUS_DEVICE
:
826 static void populate_time_info(MigrationInfo
*info
, MigrationState
*s
)
828 info
->has_status
= true;
829 info
->has_setup_time
= true;
830 info
->setup_time
= s
->setup_time
;
831 if (s
->state
== MIGRATION_STATUS_COMPLETED
) {
832 info
->has_total_time
= true;
833 info
->total_time
= s
->total_time
;
834 info
->has_downtime
= true;
835 info
->downtime
= s
->downtime
;
837 info
->has_total_time
= true;
838 info
->total_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) -
840 info
->has_expected_downtime
= true;
841 info
->expected_downtime
= s
->expected_downtime
;
845 static void populate_ram_info(MigrationInfo
*info
, MigrationState
*s
)
847 info
->has_ram
= true;
848 info
->ram
= g_malloc0(sizeof(*info
->ram
));
849 info
->ram
->transferred
= ram_counters
.transferred
;
850 info
->ram
->total
= ram_bytes_total();
851 info
->ram
->duplicate
= ram_counters
.duplicate
;
852 /* legacy value. It is not used anymore */
853 info
->ram
->skipped
= 0;
854 info
->ram
->normal
= ram_counters
.normal
;
855 info
->ram
->normal_bytes
= ram_counters
.normal
*
856 qemu_target_page_size();
857 info
->ram
->mbps
= s
->mbps
;
858 info
->ram
->dirty_sync_count
= ram_counters
.dirty_sync_count
;
859 info
->ram
->postcopy_requests
= ram_counters
.postcopy_requests
;
860 info
->ram
->page_size
= qemu_target_page_size();
861 info
->ram
->multifd_bytes
= ram_counters
.multifd_bytes
;
862 info
->ram
->pages_per_second
= s
->pages_per_second
;
864 if (migrate_use_xbzrle()) {
865 info
->has_xbzrle_cache
= true;
866 info
->xbzrle_cache
= g_malloc0(sizeof(*info
->xbzrle_cache
));
867 info
->xbzrle_cache
->cache_size
= migrate_xbzrle_cache_size();
868 info
->xbzrle_cache
->bytes
= xbzrle_counters
.bytes
;
869 info
->xbzrle_cache
->pages
= xbzrle_counters
.pages
;
870 info
->xbzrle_cache
->cache_miss
= xbzrle_counters
.cache_miss
;
871 info
->xbzrle_cache
->cache_miss_rate
= xbzrle_counters
.cache_miss_rate
;
872 info
->xbzrle_cache
->overflow
= xbzrle_counters
.overflow
;
875 if (migrate_use_compression()) {
876 info
->has_compression
= true;
877 info
->compression
= g_malloc0(sizeof(*info
->compression
));
878 info
->compression
->pages
= compression_counters
.pages
;
879 info
->compression
->busy
= compression_counters
.busy
;
880 info
->compression
->busy_rate
= compression_counters
.busy_rate
;
881 info
->compression
->compressed_size
=
882 compression_counters
.compressed_size
;
883 info
->compression
->compression_rate
=
884 compression_counters
.compression_rate
;
887 if (cpu_throttle_active()) {
888 info
->has_cpu_throttle_percentage
= true;
889 info
->cpu_throttle_percentage
= cpu_throttle_get_percentage();
892 if (s
->state
!= MIGRATION_STATUS_COMPLETED
) {
893 info
->ram
->remaining
= ram_bytes_remaining();
894 info
->ram
->dirty_pages_rate
= ram_counters
.dirty_pages_rate
;
898 static void populate_disk_info(MigrationInfo
*info
)
900 if (blk_mig_active()) {
901 info
->has_disk
= true;
902 info
->disk
= g_malloc0(sizeof(*info
->disk
));
903 info
->disk
->transferred
= blk_mig_bytes_transferred();
904 info
->disk
->remaining
= blk_mig_bytes_remaining();
905 info
->disk
->total
= blk_mig_bytes_total();
909 static void fill_source_migration_info(MigrationInfo
*info
)
911 MigrationState
*s
= migrate_get_current();
914 case MIGRATION_STATUS_NONE
:
915 /* no migration has happened ever */
916 /* do not overwrite destination migration status */
919 case MIGRATION_STATUS_SETUP
:
920 info
->has_status
= true;
921 info
->has_total_time
= false;
923 case MIGRATION_STATUS_ACTIVE
:
924 case MIGRATION_STATUS_CANCELLING
:
925 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
926 case MIGRATION_STATUS_PRE_SWITCHOVER
:
927 case MIGRATION_STATUS_DEVICE
:
928 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
929 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
930 /* TODO add some postcopy stats */
931 populate_time_info(info
, s
);
932 populate_ram_info(info
, s
);
933 populate_disk_info(info
);
935 case MIGRATION_STATUS_COLO
:
936 info
->has_status
= true;
937 /* TODO: display COLO specific information (checkpoint info etc.) */
939 case MIGRATION_STATUS_COMPLETED
:
940 populate_time_info(info
, s
);
941 populate_ram_info(info
, s
);
943 case MIGRATION_STATUS_FAILED
:
944 info
->has_status
= true;
946 info
->has_error_desc
= true;
947 info
->error_desc
= g_strdup(error_get_pretty(s
->error
));
950 case MIGRATION_STATUS_CANCELLED
:
951 info
->has_status
= true;
954 info
->status
= s
->state
;
958 * @migration_caps_check - check capability validity
960 * @cap_list: old capability list, array of bool
961 * @params: new capabilities to be applied soon
962 * @errp: set *errp if the check failed, with reason
964 * Returns true if check passed, otherwise false.
966 static bool migrate_caps_check(bool *cap_list
,
967 MigrationCapabilityStatusList
*params
,
970 MigrationCapabilityStatusList
*cap
;
971 bool old_postcopy_cap
;
972 MigrationIncomingState
*mis
= migration_incoming_get_current();
974 old_postcopy_cap
= cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
976 for (cap
= params
; cap
; cap
= cap
->next
) {
977 cap_list
[cap
->value
->capability
] = cap
->value
->state
;
980 #ifndef CONFIG_LIVE_BLOCK_MIGRATION
981 if (cap_list
[MIGRATION_CAPABILITY_BLOCK
]) {
982 error_setg(errp
, "QEMU compiled without old-style (blk/-b, inc/-i) "
984 error_append_hint(errp
, "Use drive_mirror+NBD instead.\n");
989 #ifndef CONFIG_REPLICATION
990 if (cap_list
[MIGRATION_CAPABILITY_X_COLO
]) {
991 error_setg(errp
, "QEMU compiled without replication module"
992 " can't enable COLO");
993 error_append_hint(errp
, "Please enable replication before COLO.\n");
998 if (cap_list
[MIGRATION_CAPABILITY_POSTCOPY_RAM
]) {
999 if (cap_list
[MIGRATION_CAPABILITY_COMPRESS
]) {
1000 /* The decompression threads asynchronously write into RAM
1001 * rather than use the atomic copies needed to avoid
1002 * userfaulting. It should be possible to fix the decompression
1003 * threads for compatibility in future.
1005 error_setg(errp
, "Postcopy is not currently compatible "
1006 "with compression");
1010 /* This check is reasonably expensive, so only when it's being
1011 * set the first time, also it's only the destination that needs
1014 if (!old_postcopy_cap
&& runstate_check(RUN_STATE_INMIGRATE
) &&
1015 !postcopy_ram_supported_by_host(mis
)) {
1016 /* postcopy_ram_supported_by_host will have emitted a more
1019 error_setg(errp
, "Postcopy is not supported");
1023 if (cap_list
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
]) {
1024 error_setg(errp
, "Postcopy is not compatible with ignore-shared");
1032 static void fill_destination_migration_info(MigrationInfo
*info
)
1034 MigrationIncomingState
*mis
= migration_incoming_get_current();
1036 if (mis
->socket_address_list
) {
1037 info
->has_socket_address
= true;
1038 info
->socket_address
=
1039 QAPI_CLONE(SocketAddressList
, mis
->socket_address_list
);
1042 switch (mis
->state
) {
1043 case MIGRATION_STATUS_NONE
:
1046 case MIGRATION_STATUS_SETUP
:
1047 case MIGRATION_STATUS_CANCELLING
:
1048 case MIGRATION_STATUS_CANCELLED
:
1049 case MIGRATION_STATUS_ACTIVE
:
1050 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1051 case MIGRATION_STATUS_POSTCOPY_PAUSED
:
1052 case MIGRATION_STATUS_POSTCOPY_RECOVER
:
1053 case MIGRATION_STATUS_FAILED
:
1054 case MIGRATION_STATUS_COLO
:
1055 info
->has_status
= true;
1057 case MIGRATION_STATUS_COMPLETED
:
1058 info
->has_status
= true;
1059 fill_destination_postcopy_migration_info(info
);
1062 info
->status
= mis
->state
;
1065 MigrationInfo
*qmp_query_migrate(Error
**errp
)
1067 MigrationInfo
*info
= g_malloc0(sizeof(*info
));
1069 fill_destination_migration_info(info
);
1070 fill_source_migration_info(info
);
1075 void qmp_migrate_set_capabilities(MigrationCapabilityStatusList
*params
,
1078 MigrationState
*s
= migrate_get_current();
1079 MigrationCapabilityStatusList
*cap
;
1080 bool cap_list
[MIGRATION_CAPABILITY__MAX
];
1082 if (migration_is_setup_or_active(s
->state
)) {
1083 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1087 memcpy(cap_list
, s
->enabled_capabilities
, sizeof(cap_list
));
1088 if (!migrate_caps_check(cap_list
, params
, errp
)) {
1092 for (cap
= params
; cap
; cap
= cap
->next
) {
1093 s
->enabled_capabilities
[cap
->value
->capability
] = cap
->value
->state
;
1098 * Check whether the parameters are valid. Error will be put into errp
1099 * (if provided). Return true if valid, otherwise false.
1101 static bool migrate_params_check(MigrationParameters
*params
, Error
**errp
)
1103 if (params
->has_compress_level
&&
1104 (params
->compress_level
> 9)) {
1105 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "compress_level",
1106 "is invalid, it should be in the range of 0 to 9");
1110 if (params
->has_compress_threads
&& (params
->compress_threads
< 1)) {
1111 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1113 "is invalid, it should be in the range of 1 to 255");
1117 if (params
->has_decompress_threads
&& (params
->decompress_threads
< 1)) {
1118 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1119 "decompress_threads",
1120 "is invalid, it should be in the range of 1 to 255");
1124 if (params
->has_cpu_throttle_initial
&&
1125 (params
->cpu_throttle_initial
< 1 ||
1126 params
->cpu_throttle_initial
> 99)) {
1127 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1128 "cpu_throttle_initial",
1129 "an integer in the range of 1 to 99");
1133 if (params
->has_cpu_throttle_increment
&&
1134 (params
->cpu_throttle_increment
< 1 ||
1135 params
->cpu_throttle_increment
> 99)) {
1136 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1137 "cpu_throttle_increment",
1138 "an integer in the range of 1 to 99");
1142 if (params
->has_max_bandwidth
&& (params
->max_bandwidth
> SIZE_MAX
)) {
1143 error_setg(errp
, "Parameter 'max_bandwidth' expects an integer in the"
1144 " range of 0 to %zu bytes/second", SIZE_MAX
);
1148 if (params
->has_downtime_limit
&&
1149 (params
->downtime_limit
> MAX_MIGRATE_DOWNTIME
)) {
1150 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
1151 "the range of 0 to %d milliseconds",
1152 MAX_MIGRATE_DOWNTIME
);
1156 /* x_checkpoint_delay is now always positive */
1158 if (params
->has_multifd_channels
&& (params
->multifd_channels
< 1)) {
1159 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1161 "is invalid, it should be in the range of 1 to 255");
1165 if (params
->has_xbzrle_cache_size
&&
1166 (params
->xbzrle_cache_size
< qemu_target_page_size() ||
1167 !is_power_of_2(params
->xbzrle_cache_size
))) {
1168 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1169 "xbzrle_cache_size",
1170 "is invalid, it should be bigger than target page size"
1171 " and a power of two");
1175 if (params
->has_max_cpu_throttle
&&
1176 (params
->max_cpu_throttle
< params
->cpu_throttle_initial
||
1177 params
->max_cpu_throttle
> 99)) {
1178 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1180 "an integer in the range of cpu_throttle_initial to 99");
1184 if (params
->has_announce_initial
&&
1185 params
->announce_initial
> 100000) {
1186 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1188 "is invalid, it must be less than 100000 ms");
1191 if (params
->has_announce_max
&&
1192 params
->announce_max
> 100000) {
1193 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1195 "is invalid, it must be less than 100000 ms");
1198 if (params
->has_announce_rounds
&&
1199 params
->announce_rounds
> 1000) {
1200 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1202 "is invalid, it must be in the range of 0 to 1000");
1205 if (params
->has_announce_step
&&
1206 (params
->announce_step
< 1 ||
1207 params
->announce_step
> 10000)) {
1208 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
,
1210 "is invalid, it must be in the range of 1 to 10000 ms");
1216 static void migrate_params_test_apply(MigrateSetParameters
*params
,
1217 MigrationParameters
*dest
)
1219 *dest
= migrate_get_current()->parameters
;
1221 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1223 if (params
->has_compress_level
) {
1224 dest
->compress_level
= params
->compress_level
;
1227 if (params
->has_compress_threads
) {
1228 dest
->compress_threads
= params
->compress_threads
;
1231 if (params
->has_compress_wait_thread
) {
1232 dest
->compress_wait_thread
= params
->compress_wait_thread
;
1235 if (params
->has_decompress_threads
) {
1236 dest
->decompress_threads
= params
->decompress_threads
;
1239 if (params
->has_cpu_throttle_initial
) {
1240 dest
->cpu_throttle_initial
= params
->cpu_throttle_initial
;
1243 if (params
->has_cpu_throttle_increment
) {
1244 dest
->cpu_throttle_increment
= params
->cpu_throttle_increment
;
1247 if (params
->has_tls_creds
) {
1248 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1249 dest
->tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1252 if (params
->has_tls_hostname
) {
1253 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1254 dest
->tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1257 if (params
->has_max_bandwidth
) {
1258 dest
->max_bandwidth
= params
->max_bandwidth
;
1261 if (params
->has_downtime_limit
) {
1262 dest
->downtime_limit
= params
->downtime_limit
;
1265 if (params
->has_x_checkpoint_delay
) {
1266 dest
->x_checkpoint_delay
= params
->x_checkpoint_delay
;
1269 if (params
->has_block_incremental
) {
1270 dest
->block_incremental
= params
->block_incremental
;
1272 if (params
->has_multifd_channels
) {
1273 dest
->multifd_channels
= params
->multifd_channels
;
1275 if (params
->has_xbzrle_cache_size
) {
1276 dest
->xbzrle_cache_size
= params
->xbzrle_cache_size
;
1278 if (params
->has_max_postcopy_bandwidth
) {
1279 dest
->max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1281 if (params
->has_max_cpu_throttle
) {
1282 dest
->max_cpu_throttle
= params
->max_cpu_throttle
;
1284 if (params
->has_announce_initial
) {
1285 dest
->announce_initial
= params
->announce_initial
;
1287 if (params
->has_announce_max
) {
1288 dest
->announce_max
= params
->announce_max
;
1290 if (params
->has_announce_rounds
) {
1291 dest
->announce_rounds
= params
->announce_rounds
;
1293 if (params
->has_announce_step
) {
1294 dest
->announce_step
= params
->announce_step
;
1298 static void migrate_params_apply(MigrateSetParameters
*params
, Error
**errp
)
1300 MigrationState
*s
= migrate_get_current();
1302 /* TODO use QAPI_CLONE() instead of duplicating it inline */
1304 if (params
->has_compress_level
) {
1305 s
->parameters
.compress_level
= params
->compress_level
;
1308 if (params
->has_compress_threads
) {
1309 s
->parameters
.compress_threads
= params
->compress_threads
;
1312 if (params
->has_compress_wait_thread
) {
1313 s
->parameters
.compress_wait_thread
= params
->compress_wait_thread
;
1316 if (params
->has_decompress_threads
) {
1317 s
->parameters
.decompress_threads
= params
->decompress_threads
;
1320 if (params
->has_cpu_throttle_initial
) {
1321 s
->parameters
.cpu_throttle_initial
= params
->cpu_throttle_initial
;
1324 if (params
->has_cpu_throttle_increment
) {
1325 s
->parameters
.cpu_throttle_increment
= params
->cpu_throttle_increment
;
1328 if (params
->has_tls_creds
) {
1329 g_free(s
->parameters
.tls_creds
);
1330 assert(params
->tls_creds
->type
== QTYPE_QSTRING
);
1331 s
->parameters
.tls_creds
= g_strdup(params
->tls_creds
->u
.s
);
1334 if (params
->has_tls_hostname
) {
1335 g_free(s
->parameters
.tls_hostname
);
1336 assert(params
->tls_hostname
->type
== QTYPE_QSTRING
);
1337 s
->parameters
.tls_hostname
= g_strdup(params
->tls_hostname
->u
.s
);
1340 if (params
->has_tls_authz
) {
1341 g_free(s
->parameters
.tls_authz
);
1342 assert(params
->tls_authz
->type
== QTYPE_QSTRING
);
1343 s
->parameters
.tls_authz
= g_strdup(params
->tls_authz
->u
.s
);
1346 if (params
->has_max_bandwidth
) {
1347 s
->parameters
.max_bandwidth
= params
->max_bandwidth
;
1348 if (s
->to_dst_file
&& !migration_in_postcopy()) {
1349 qemu_file_set_rate_limit(s
->to_dst_file
,
1350 s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
);
1354 if (params
->has_downtime_limit
) {
1355 s
->parameters
.downtime_limit
= params
->downtime_limit
;
1358 if (params
->has_x_checkpoint_delay
) {
1359 s
->parameters
.x_checkpoint_delay
= params
->x_checkpoint_delay
;
1360 if (migration_in_colo_state()) {
1361 colo_checkpoint_notify(s
);
1365 if (params
->has_block_incremental
) {
1366 s
->parameters
.block_incremental
= params
->block_incremental
;
1368 if (params
->has_multifd_channels
) {
1369 s
->parameters
.multifd_channels
= params
->multifd_channels
;
1371 if (params
->has_xbzrle_cache_size
) {
1372 s
->parameters
.xbzrle_cache_size
= params
->xbzrle_cache_size
;
1373 xbzrle_cache_resize(params
->xbzrle_cache_size
, errp
);
1375 if (params
->has_max_postcopy_bandwidth
) {
1376 s
->parameters
.max_postcopy_bandwidth
= params
->max_postcopy_bandwidth
;
1377 if (s
->to_dst_file
&& migration_in_postcopy()) {
1378 qemu_file_set_rate_limit(s
->to_dst_file
,
1379 s
->parameters
.max_postcopy_bandwidth
/ XFER_LIMIT_RATIO
);
1382 if (params
->has_max_cpu_throttle
) {
1383 s
->parameters
.max_cpu_throttle
= params
->max_cpu_throttle
;
1385 if (params
->has_announce_initial
) {
1386 s
->parameters
.announce_initial
= params
->announce_initial
;
1388 if (params
->has_announce_max
) {
1389 s
->parameters
.announce_max
= params
->announce_max
;
1391 if (params
->has_announce_rounds
) {
1392 s
->parameters
.announce_rounds
= params
->announce_rounds
;
1394 if (params
->has_announce_step
) {
1395 s
->parameters
.announce_step
= params
->announce_step
;
1399 void qmp_migrate_set_parameters(MigrateSetParameters
*params
, Error
**errp
)
1401 MigrationParameters tmp
;
1403 /* TODO Rewrite "" to null instead */
1404 if (params
->has_tls_creds
1405 && params
->tls_creds
->type
== QTYPE_QNULL
) {
1406 qobject_unref(params
->tls_creds
->u
.n
);
1407 params
->tls_creds
->type
= QTYPE_QSTRING
;
1408 params
->tls_creds
->u
.s
= strdup("");
1410 /* TODO Rewrite "" to null instead */
1411 if (params
->has_tls_hostname
1412 && params
->tls_hostname
->type
== QTYPE_QNULL
) {
1413 qobject_unref(params
->tls_hostname
->u
.n
);
1414 params
->tls_hostname
->type
= QTYPE_QSTRING
;
1415 params
->tls_hostname
->u
.s
= strdup("");
1418 migrate_params_test_apply(params
, &tmp
);
1420 if (!migrate_params_check(&tmp
, errp
)) {
1421 /* Invalid parameter */
1425 migrate_params_apply(params
, errp
);
1429 void qmp_migrate_start_postcopy(Error
**errp
)
1431 MigrationState
*s
= migrate_get_current();
1433 if (!migrate_postcopy()) {
1434 error_setg(errp
, "Enable postcopy with migrate_set_capability before"
1435 " the start of migration");
1439 if (s
->state
== MIGRATION_STATUS_NONE
) {
1440 error_setg(errp
, "Postcopy must be started after migration has been"
1445 * we don't error if migration has finished since that would be racy
1446 * with issuing this command.
1448 atomic_set(&s
->start_postcopy
, true);
1451 /* shared migration helpers */
1453 void migrate_set_state(int *state
, int old_state
, int new_state
)
1455 assert(new_state
< MIGRATION_STATUS__MAX
);
1456 if (atomic_cmpxchg(state
, old_state
, new_state
) == old_state
) {
1457 trace_migrate_set_state(MigrationStatus_str(new_state
));
1458 migrate_generate_event(new_state
);
1462 static MigrationCapabilityStatusList
*migrate_cap_add(
1463 MigrationCapabilityStatusList
*list
,
1464 MigrationCapability index
,
1467 MigrationCapabilityStatusList
*cap
;
1469 cap
= g_new0(MigrationCapabilityStatusList
, 1);
1470 cap
->value
= g_new0(MigrationCapabilityStatus
, 1);
1471 cap
->value
->capability
= index
;
1472 cap
->value
->state
= state
;
1478 void migrate_set_block_enabled(bool value
, Error
**errp
)
1480 MigrationCapabilityStatusList
*cap
;
1482 cap
= migrate_cap_add(NULL
, MIGRATION_CAPABILITY_BLOCK
, value
);
1483 qmp_migrate_set_capabilities(cap
, errp
);
1484 qapi_free_MigrationCapabilityStatusList(cap
);
1487 static void migrate_set_block_incremental(MigrationState
*s
, bool value
)
1489 s
->parameters
.block_incremental
= value
;
1492 static void block_cleanup_parameters(MigrationState
*s
)
1494 if (s
->must_remove_block_options
) {
1495 /* setting to false can never fail */
1496 migrate_set_block_enabled(false, &error_abort
);
1497 migrate_set_block_incremental(s
, false);
1498 s
->must_remove_block_options
= false;
1502 static void migrate_fd_cleanup(MigrationState
*s
)
1504 qemu_bh_delete(s
->cleanup_bh
);
1505 s
->cleanup_bh
= NULL
;
1507 qemu_savevm_state_cleanup();
1509 if (s
->to_dst_file
) {
1512 trace_migrate_fd_cleanup();
1513 qemu_mutex_unlock_iothread();
1514 if (s
->migration_thread_running
) {
1515 qemu_thread_join(&s
->thread
);
1516 s
->migration_thread_running
= false;
1518 qemu_mutex_lock_iothread();
1520 multifd_save_cleanup();
1521 qemu_mutex_lock(&s
->qemu_file_lock
);
1522 tmp
= s
->to_dst_file
;
1523 s
->to_dst_file
= NULL
;
1524 qemu_mutex_unlock(&s
->qemu_file_lock
);
1526 * Close the file handle without the lock to make sure the
1527 * critical section won't block for long.
1532 assert((s
->state
!= MIGRATION_STATUS_ACTIVE
) &&
1533 (s
->state
!= MIGRATION_STATUS_POSTCOPY_ACTIVE
));
1535 if (s
->state
== MIGRATION_STATUS_CANCELLING
) {
1536 migrate_set_state(&s
->state
, MIGRATION_STATUS_CANCELLING
,
1537 MIGRATION_STATUS_CANCELLED
);
1541 /* It is used on info migrate. We can't free it */
1542 error_report_err(error_copy(s
->error
));
1544 notifier_list_notify(&migration_state_notifiers
, s
);
1545 block_cleanup_parameters(s
);
1548 static void migrate_fd_cleanup_schedule(MigrationState
*s
)
1551 * Ref the state for bh, because it may be called when
1552 * there're already no other refs
1554 object_ref(OBJECT(s
));
1555 qemu_bh_schedule(s
->cleanup_bh
);
1558 static void migrate_fd_cleanup_bh(void *opaque
)
1560 MigrationState
*s
= opaque
;
1561 migrate_fd_cleanup(s
);
1562 object_unref(OBJECT(s
));
1565 void migrate_set_error(MigrationState
*s
, const Error
*error
)
1567 qemu_mutex_lock(&s
->error_mutex
);
1569 s
->error
= error_copy(error
);
1571 qemu_mutex_unlock(&s
->error_mutex
);
1574 void migrate_fd_error(MigrationState
*s
, const Error
*error
)
1576 trace_migrate_fd_error(error_get_pretty(error
));
1577 assert(s
->to_dst_file
== NULL
);
1578 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1579 MIGRATION_STATUS_FAILED
);
1580 migrate_set_error(s
, error
);
1583 static void migrate_fd_cancel(MigrationState
*s
)
1586 QEMUFile
*f
= migrate_get_current()->to_dst_file
;
1587 trace_migrate_fd_cancel();
1589 if (s
->rp_state
.from_dst_file
) {
1590 /* shutdown the rp socket, so causing the rp thread to shutdown */
1591 qemu_file_shutdown(s
->rp_state
.from_dst_file
);
1595 old_state
= s
->state
;
1596 if (!migration_is_setup_or_active(old_state
)) {
1599 /* If the migration is paused, kick it out of the pause */
1600 if (old_state
== MIGRATION_STATUS_PRE_SWITCHOVER
) {
1601 qemu_sem_post(&s
->pause_sem
);
1603 migrate_set_state(&s
->state
, old_state
, MIGRATION_STATUS_CANCELLING
);
1604 } while (s
->state
!= MIGRATION_STATUS_CANCELLING
);
1607 * If we're unlucky the migration code might be stuck somewhere in a
1608 * send/write while the network has failed and is waiting to timeout;
1609 * if we've got shutdown(2) available then we can force it to quit.
1610 * The outgoing qemu file gets closed in migrate_fd_cleanup that is
1611 * called in a bh, so there is no race against this cancel.
1613 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& f
) {
1614 qemu_file_shutdown(f
);
1616 if (s
->state
== MIGRATION_STATUS_CANCELLING
&& s
->block_inactive
) {
1617 Error
*local_err
= NULL
;
1619 bdrv_invalidate_cache_all(&local_err
);
1621 error_report_err(local_err
);
1623 s
->block_inactive
= false;
1628 void add_migration_state_change_notifier(Notifier
*notify
)
1630 notifier_list_add(&migration_state_notifiers
, notify
);
1633 void remove_migration_state_change_notifier(Notifier
*notify
)
1635 notifier_remove(notify
);
1638 bool migration_in_setup(MigrationState
*s
)
1640 return s
->state
== MIGRATION_STATUS_SETUP
;
1643 bool migration_has_finished(MigrationState
*s
)
1645 return s
->state
== MIGRATION_STATUS_COMPLETED
;
1648 bool migration_has_failed(MigrationState
*s
)
1650 return (s
->state
== MIGRATION_STATUS_CANCELLED
||
1651 s
->state
== MIGRATION_STATUS_FAILED
);
1654 bool migration_in_postcopy(void)
1656 MigrationState
*s
= migrate_get_current();
1658 return (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1661 bool migration_in_postcopy_after_devices(MigrationState
*s
)
1663 return migration_in_postcopy() && s
->postcopy_after_devices
;
1666 bool migration_is_idle(void)
1668 MigrationState
*s
= current_migration
;
1675 case MIGRATION_STATUS_NONE
:
1676 case MIGRATION_STATUS_CANCELLED
:
1677 case MIGRATION_STATUS_COMPLETED
:
1678 case MIGRATION_STATUS_FAILED
:
1680 case MIGRATION_STATUS_SETUP
:
1681 case MIGRATION_STATUS_CANCELLING
:
1682 case MIGRATION_STATUS_ACTIVE
:
1683 case MIGRATION_STATUS_POSTCOPY_ACTIVE
:
1684 case MIGRATION_STATUS_COLO
:
1685 case MIGRATION_STATUS_PRE_SWITCHOVER
:
1686 case MIGRATION_STATUS_DEVICE
:
1688 case MIGRATION_STATUS__MAX
:
1689 g_assert_not_reached();
1695 void migrate_init(MigrationState
*s
)
1698 * Reinitialise all migration state, except
1699 * parameters/capabilities that the user set, and
1703 s
->to_dst_file
= NULL
;
1704 s
->state
= MIGRATION_STATUS_NONE
;
1705 s
->rp_state
.from_dst_file
= NULL
;
1706 s
->rp_state
.error
= false;
1708 s
->pages_per_second
= 0.0;
1710 s
->expected_downtime
= 0;
1712 s
->start_postcopy
= false;
1713 s
->postcopy_after_devices
= false;
1714 s
->migration_thread_running
= false;
1715 error_free(s
->error
);
1718 migrate_set_state(&s
->state
, MIGRATION_STATUS_NONE
, MIGRATION_STATUS_SETUP
);
1720 s
->start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
1722 s
->vm_was_running
= false;
1723 s
->iteration_initial_bytes
= 0;
1724 s
->threshold_size
= 0;
1727 static GSList
*migration_blockers
;
1729 int migrate_add_blocker(Error
*reason
, Error
**errp
)
1731 if (only_migratable
) {
1732 error_propagate_prepend(errp
, error_copy(reason
),
1733 "disallowing migration blocker "
1734 "(--only-migratable) for: ");
1738 if (migration_is_idle()) {
1739 migration_blockers
= g_slist_prepend(migration_blockers
, reason
);
1743 error_propagate_prepend(errp
, error_copy(reason
),
1744 "disallowing migration blocker "
1745 "(migration in progress) for: ");
1749 void migrate_del_blocker(Error
*reason
)
1751 migration_blockers
= g_slist_remove(migration_blockers
, reason
);
1754 void qmp_migrate_incoming(const char *uri
, Error
**errp
)
1756 Error
*local_err
= NULL
;
1757 static bool once
= true;
1759 if (!deferred_incoming
) {
1760 error_setg(errp
, "For use with '-incoming defer'");
1764 error_setg(errp
, "The incoming migration has already been started");
1767 qemu_start_incoming_migration(uri
, &local_err
);
1770 error_propagate(errp
, local_err
);
1777 void qmp_migrate_recover(const char *uri
, Error
**errp
)
1779 MigrationIncomingState
*mis
= migration_incoming_get_current();
1781 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1782 error_setg(errp
, "Migrate recover can only be run "
1783 "when postcopy is paused.");
1787 if (atomic_cmpxchg(&mis
->postcopy_recover_triggered
,
1788 false, true) == true) {
1789 error_setg(errp
, "Migrate recovery is triggered already");
1794 * Note that this call will never start a real migration; it will
1795 * only re-setup the migration stream and poke existing migration
1796 * to continue using that newly established channel.
1798 qemu_start_incoming_migration(uri
, errp
);
1801 void qmp_migrate_pause(Error
**errp
)
1803 MigrationState
*ms
= migrate_get_current();
1804 MigrationIncomingState
*mis
= migration_incoming_get_current();
1807 if (ms
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1808 /* Source side, during postcopy */
1809 qemu_mutex_lock(&ms
->qemu_file_lock
);
1810 ret
= qemu_file_shutdown(ms
->to_dst_file
);
1811 qemu_mutex_unlock(&ms
->qemu_file_lock
);
1813 error_setg(errp
, "Failed to pause source migration");
1818 if (mis
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
1819 ret
= qemu_file_shutdown(mis
->from_src_file
);
1821 error_setg(errp
, "Failed to pause destination migration");
1826 error_setg(errp
, "migrate-pause is currently only supported "
1827 "during postcopy-active state");
1830 bool migration_is_blocked(Error
**errp
)
1832 if (qemu_savevm_state_blocked(errp
)) {
1836 if (migration_blockers
) {
1837 error_propagate(errp
, error_copy(migration_blockers
->data
));
1844 /* Returns true if continue to migrate, or false if error detected */
1845 static bool migrate_prepare(MigrationState
*s
, bool blk
, bool blk_inc
,
1846 bool resume
, Error
**errp
)
1848 Error
*local_err
= NULL
;
1851 if (s
->state
!= MIGRATION_STATUS_POSTCOPY_PAUSED
) {
1852 error_setg(errp
, "Cannot resume if there is no "
1853 "paused migration");
1858 * Postcopy recovery won't work well with release-ram
1859 * capability since release-ram will drop the page buffer as
1860 * long as the page is put into the send buffer. So if there
1861 * is a network failure happened, any page buffers that have
1862 * not yet reached the destination VM but have already been
1863 * sent from the source VM will be lost forever. Let's refuse
1864 * the client from resuming such a postcopy migration.
1865 * Luckily release-ram was designed to only be used when src
1866 * and destination VMs are on the same host, so it should be
1869 if (migrate_release_ram()) {
1870 error_setg(errp
, "Postcopy recovery cannot work "
1871 "when release-ram capability is set");
1875 /* This is a resume, skip init status */
1879 if (migration_is_setup_or_active(s
->state
) ||
1880 s
->state
== MIGRATION_STATUS_CANCELLING
||
1881 s
->state
== MIGRATION_STATUS_COLO
) {
1882 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1886 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1887 error_setg(errp
, "Guest is waiting for an incoming migration");
1891 if (migration_is_blocked(errp
)) {
1895 if (blk
|| blk_inc
) {
1896 if (migrate_use_block() || migrate_use_block_incremental()) {
1897 error_setg(errp
, "Command options are incompatible with "
1898 "current migration capabilities");
1901 migrate_set_block_enabled(true, &local_err
);
1903 error_propagate(errp
, local_err
);
1906 s
->must_remove_block_options
= true;
1910 migrate_set_block_incremental(s
, true);
1915 * set ram_counters memory to zero for a
1918 memset(&ram_counters
, 0, sizeof(ram_counters
));
1923 void qmp_migrate(const char *uri
, bool has_blk
, bool blk
,
1924 bool has_inc
, bool inc
, bool has_detach
, bool detach
,
1925 bool has_resume
, bool resume
, Error
**errp
)
1927 Error
*local_err
= NULL
;
1928 MigrationState
*s
= migrate_get_current();
1931 if (!migrate_prepare(s
, has_blk
&& blk
, has_inc
&& inc
,
1932 has_resume
&& resume
, errp
)) {
1933 /* Error detected, put into errp */
1937 if (strstart(uri
, "tcp:", &p
)) {
1938 tcp_start_outgoing_migration(s
, p
, &local_err
);
1940 } else if (strstart(uri
, "rdma:", &p
)) {
1941 rdma_start_outgoing_migration(s
, p
, &local_err
);
1943 } else if (strstart(uri
, "exec:", &p
)) {
1944 exec_start_outgoing_migration(s
, p
, &local_err
);
1945 } else if (strstart(uri
, "unix:", &p
)) {
1946 unix_start_outgoing_migration(s
, p
, &local_err
);
1947 } else if (strstart(uri
, "fd:", &p
)) {
1948 fd_start_outgoing_migration(s
, p
, &local_err
);
1950 error_setg(errp
, QERR_INVALID_PARAMETER_VALUE
, "uri",
1951 "a valid migration protocol");
1952 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
1953 MIGRATION_STATUS_FAILED
);
1954 block_cleanup_parameters(s
);
1959 migrate_fd_error(s
, local_err
);
1960 error_propagate(errp
, local_err
);
1965 void qmp_migrate_cancel(Error
**errp
)
1967 migrate_fd_cancel(migrate_get_current());
1970 void qmp_migrate_continue(MigrationStatus state
, Error
**errp
)
1972 MigrationState
*s
= migrate_get_current();
1973 if (s
->state
!= state
) {
1974 error_setg(errp
, "Migration not in expected state: %s",
1975 MigrationStatus_str(s
->state
));
1978 qemu_sem_post(&s
->pause_sem
);
1981 void qmp_migrate_set_cache_size(int64_t value
, Error
**errp
)
1983 MigrateSetParameters p
= {
1984 .has_xbzrle_cache_size
= true,
1985 .xbzrle_cache_size
= value
,
1988 qmp_migrate_set_parameters(&p
, errp
);
1991 int64_t qmp_query_migrate_cache_size(Error
**errp
)
1993 return migrate_xbzrle_cache_size();
1996 void qmp_migrate_set_speed(int64_t value
, Error
**errp
)
1998 MigrateSetParameters p
= {
1999 .has_max_bandwidth
= true,
2000 .max_bandwidth
= value
,
2003 qmp_migrate_set_parameters(&p
, errp
);
2006 void qmp_migrate_set_downtime(double value
, Error
**errp
)
2008 if (value
< 0 || value
> MAX_MIGRATE_DOWNTIME_SECONDS
) {
2009 error_setg(errp
, "Parameter 'downtime_limit' expects an integer in "
2010 "the range of 0 to %d seconds",
2011 MAX_MIGRATE_DOWNTIME_SECONDS
);
2015 value
*= 1000; /* Convert to milliseconds */
2016 value
= MAX(0, MIN(INT64_MAX
, value
));
2018 MigrateSetParameters p
= {
2019 .has_downtime_limit
= true,
2020 .downtime_limit
= value
,
2023 qmp_migrate_set_parameters(&p
, errp
);
2026 bool migrate_release_ram(void)
2030 s
= migrate_get_current();
2032 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RELEASE_RAM
];
2035 bool migrate_postcopy_ram(void)
2039 s
= migrate_get_current();
2041 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_RAM
];
2044 bool migrate_postcopy(void)
2046 return migrate_postcopy_ram() || migrate_dirty_bitmaps();
2049 bool migrate_auto_converge(void)
2053 s
= migrate_get_current();
2055 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_AUTO_CONVERGE
];
2058 bool migrate_zero_blocks(void)
2062 s
= migrate_get_current();
2064 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_ZERO_BLOCKS
];
2067 bool migrate_postcopy_blocktime(void)
2071 s
= migrate_get_current();
2073 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_POSTCOPY_BLOCKTIME
];
2076 bool migrate_use_compression(void)
2080 s
= migrate_get_current();
2082 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_COMPRESS
];
2085 int migrate_compress_level(void)
2089 s
= migrate_get_current();
2091 return s
->parameters
.compress_level
;
2094 int migrate_compress_threads(void)
2098 s
= migrate_get_current();
2100 return s
->parameters
.compress_threads
;
2103 int migrate_compress_wait_thread(void)
2107 s
= migrate_get_current();
2109 return s
->parameters
.compress_wait_thread
;
2112 int migrate_decompress_threads(void)
2116 s
= migrate_get_current();
2118 return s
->parameters
.decompress_threads
;
2121 bool migrate_dirty_bitmaps(void)
2125 s
= migrate_get_current();
2127 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_DIRTY_BITMAPS
];
2130 bool migrate_ignore_shared(void)
2134 s
= migrate_get_current();
2136 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_IGNORE_SHARED
];
2139 bool migrate_use_events(void)
2143 s
= migrate_get_current();
2145 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_EVENTS
];
2148 bool migrate_use_multifd(void)
2152 s
= migrate_get_current();
2154 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_MULTIFD
];
2157 bool migrate_pause_before_switchover(void)
2161 s
= migrate_get_current();
2163 return s
->enabled_capabilities
[
2164 MIGRATION_CAPABILITY_PAUSE_BEFORE_SWITCHOVER
];
2167 int migrate_multifd_channels(void)
2171 s
= migrate_get_current();
2173 return s
->parameters
.multifd_channels
;
2176 int migrate_use_xbzrle(void)
2180 s
= migrate_get_current();
2182 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_XBZRLE
];
2185 int64_t migrate_xbzrle_cache_size(void)
2189 s
= migrate_get_current();
2191 return s
->parameters
.xbzrle_cache_size
;
2194 static int64_t migrate_max_postcopy_bandwidth(void)
2198 s
= migrate_get_current();
2200 return s
->parameters
.max_postcopy_bandwidth
;
2203 bool migrate_use_block(void)
2207 s
= migrate_get_current();
2209 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_BLOCK
];
2212 bool migrate_use_return_path(void)
2216 s
= migrate_get_current();
2218 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_RETURN_PATH
];
2221 bool migrate_use_block_incremental(void)
2225 s
= migrate_get_current();
2227 return s
->parameters
.block_incremental
;
2230 /* migration thread support */
2232 * Something bad happened to the RP stream, mark an error
2233 * The caller shall print or trace something to indicate why
2235 static void mark_source_rp_bad(MigrationState
*s
)
2237 s
->rp_state
.error
= true;
2240 static struct rp_cmd_args
{
2241 ssize_t len
; /* -1 = variable */
2244 [MIG_RP_MSG_INVALID
] = { .len
= -1, .name
= "INVALID" },
2245 [MIG_RP_MSG_SHUT
] = { .len
= 4, .name
= "SHUT" },
2246 [MIG_RP_MSG_PONG
] = { .len
= 4, .name
= "PONG" },
2247 [MIG_RP_MSG_REQ_PAGES
] = { .len
= 12, .name
= "REQ_PAGES" },
2248 [MIG_RP_MSG_REQ_PAGES_ID
] = { .len
= -1, .name
= "REQ_PAGES_ID" },
2249 [MIG_RP_MSG_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
2250 [MIG_RP_MSG_RESUME_ACK
] = { .len
= 4, .name
= "RESUME_ACK" },
2251 [MIG_RP_MSG_MAX
] = { .len
= -1, .name
= "MAX" },
2255 * Process a request for pages received on the return path,
2256 * We're allowed to send more than requested (e.g. to round to our page size)
2257 * and we don't need to send pages that have already been sent.
2259 static void migrate_handle_rp_req_pages(MigrationState
*ms
, const char* rbname
,
2260 ram_addr_t start
, size_t len
)
2262 long our_host_ps
= getpagesize();
2264 trace_migrate_handle_rp_req_pages(rbname
, start
, len
);
2267 * Since we currently insist on matching page sizes, just sanity check
2268 * we're being asked for whole host pages.
2270 if (start
& (our_host_ps
-1) ||
2271 (len
& (our_host_ps
-1))) {
2272 error_report("%s: Misaligned page request, start: " RAM_ADDR_FMT
2273 " len: %zd", __func__
, start
, len
);
2274 mark_source_rp_bad(ms
);
2278 if (ram_save_queue_pages(rbname
, start
, len
)) {
2279 mark_source_rp_bad(ms
);
2283 /* Return true to retry, false to quit */
2284 static bool postcopy_pause_return_path_thread(MigrationState
*s
)
2286 trace_postcopy_pause_return_path();
2288 qemu_sem_wait(&s
->postcopy_pause_rp_sem
);
2290 trace_postcopy_pause_return_path_continued();
2295 static int migrate_handle_rp_recv_bitmap(MigrationState
*s
, char *block_name
)
2297 RAMBlock
*block
= qemu_ram_block_by_name(block_name
);
2300 error_report("%s: invalid block name '%s'", __func__
, block_name
);
2304 /* Fetch the received bitmap and refresh the dirty bitmap */
2305 return ram_dirty_bitmap_reload(s
, block
);
2308 static int migrate_handle_rp_resume_ack(MigrationState
*s
, uint32_t value
)
2310 trace_source_return_path_thread_resume_ack(value
);
2312 if (value
!= MIGRATION_RESUME_ACK_VALUE
) {
2313 error_report("%s: illegal resume_ack value %"PRIu32
,
2318 /* Now both sides are active. */
2319 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2320 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2322 /* Notify send thread that time to continue send pages */
2323 qemu_sem_post(&s
->rp_state
.rp_sem
);
2329 * Handles messages sent on the return path towards the source VM
2332 static void *source_return_path_thread(void *opaque
)
2334 MigrationState
*ms
= opaque
;
2335 QEMUFile
*rp
= ms
->rp_state
.from_dst_file
;
2336 uint16_t header_len
, header_type
;
2338 uint32_t tmp32
, sibling_error
;
2339 ram_addr_t start
= 0; /* =0 to silence warning */
2340 size_t len
= 0, expected_len
;
2343 trace_source_return_path_thread_entry();
2344 rcu_register_thread();
2347 while (!ms
->rp_state
.error
&& !qemu_file_get_error(rp
) &&
2348 migration_is_setup_or_active(ms
->state
)) {
2349 trace_source_return_path_thread_loop_top();
2350 header_type
= qemu_get_be16(rp
);
2351 header_len
= qemu_get_be16(rp
);
2353 if (qemu_file_get_error(rp
)) {
2354 mark_source_rp_bad(ms
);
2358 if (header_type
>= MIG_RP_MSG_MAX
||
2359 header_type
== MIG_RP_MSG_INVALID
) {
2360 error_report("RP: Received invalid message 0x%04x length 0x%04x",
2361 header_type
, header_len
);
2362 mark_source_rp_bad(ms
);
2366 if ((rp_cmd_args
[header_type
].len
!= -1 &&
2367 header_len
!= rp_cmd_args
[header_type
].len
) ||
2368 header_len
> sizeof(buf
)) {
2369 error_report("RP: Received '%s' message (0x%04x) with"
2370 "incorrect length %d expecting %zu",
2371 rp_cmd_args
[header_type
].name
, header_type
, header_len
,
2372 (size_t)rp_cmd_args
[header_type
].len
);
2373 mark_source_rp_bad(ms
);
2377 /* We know we've got a valid header by this point */
2378 res
= qemu_get_buffer(rp
, buf
, header_len
);
2379 if (res
!= header_len
) {
2380 error_report("RP: Failed reading data for message 0x%04x"
2381 " read %d expected %d",
2382 header_type
, res
, header_len
);
2383 mark_source_rp_bad(ms
);
2387 /* OK, we have the message and the data */
2388 switch (header_type
) {
2389 case MIG_RP_MSG_SHUT
:
2390 sibling_error
= ldl_be_p(buf
);
2391 trace_source_return_path_thread_shut(sibling_error
);
2392 if (sibling_error
) {
2393 error_report("RP: Sibling indicated error %d", sibling_error
);
2394 mark_source_rp_bad(ms
);
2397 * We'll let the main thread deal with closing the RP
2398 * we could do a shutdown(2) on it, but we're the only user
2399 * anyway, so there's nothing gained.
2403 case MIG_RP_MSG_PONG
:
2404 tmp32
= ldl_be_p(buf
);
2405 trace_source_return_path_thread_pong(tmp32
);
2408 case MIG_RP_MSG_REQ_PAGES
:
2409 start
= ldq_be_p(buf
);
2410 len
= ldl_be_p(buf
+ 8);
2411 migrate_handle_rp_req_pages(ms
, NULL
, start
, len
);
2414 case MIG_RP_MSG_REQ_PAGES_ID
:
2415 expected_len
= 12 + 1; /* header + termination */
2417 if (header_len
>= expected_len
) {
2418 start
= ldq_be_p(buf
);
2419 len
= ldl_be_p(buf
+ 8);
2420 /* Now we expect an idstr */
2421 tmp32
= buf
[12]; /* Length of the following idstr */
2422 buf
[13 + tmp32
] = '\0';
2423 expected_len
+= tmp32
;
2425 if (header_len
!= expected_len
) {
2426 error_report("RP: Req_Page_id with length %d expecting %zd",
2427 header_len
, expected_len
);
2428 mark_source_rp_bad(ms
);
2431 migrate_handle_rp_req_pages(ms
, (char *)&buf
[13], start
, len
);
2434 case MIG_RP_MSG_RECV_BITMAP
:
2435 if (header_len
< 1) {
2436 error_report("%s: missing block name", __func__
);
2437 mark_source_rp_bad(ms
);
2440 /* Format: len (1B) + idstr (<255B). This ends the idstr. */
2441 buf
[buf
[0] + 1] = '\0';
2442 if (migrate_handle_rp_recv_bitmap(ms
, (char *)(buf
+ 1))) {
2443 mark_source_rp_bad(ms
);
2448 case MIG_RP_MSG_RESUME_ACK
:
2449 tmp32
= ldl_be_p(buf
);
2450 if (migrate_handle_rp_resume_ack(ms
, tmp32
)) {
2451 mark_source_rp_bad(ms
);
2462 res
= qemu_file_get_error(rp
);
2466 * Maybe there is something we can do: it looks like a
2467 * network down issue, and we pause for a recovery.
2469 if (postcopy_pause_return_path_thread(ms
)) {
2470 /* Reload rp, reset the rest */
2471 if (rp
!= ms
->rp_state
.from_dst_file
) {
2473 rp
= ms
->rp_state
.from_dst_file
;
2475 ms
->rp_state
.error
= false;
2480 trace_source_return_path_thread_bad_end();
2481 mark_source_rp_bad(ms
);
2484 trace_source_return_path_thread_end();
2485 ms
->rp_state
.from_dst_file
= NULL
;
2487 rcu_unregister_thread();
2491 static int open_return_path_on_source(MigrationState
*ms
,
2495 ms
->rp_state
.from_dst_file
= qemu_file_get_return_path(ms
->to_dst_file
);
2496 if (!ms
->rp_state
.from_dst_file
) {
2500 trace_open_return_path_on_source();
2502 if (!create_thread
) {
2507 qemu_thread_create(&ms
->rp_state
.rp_thread
, "return path",
2508 source_return_path_thread
, ms
, QEMU_THREAD_JOINABLE
);
2510 trace_open_return_path_on_source_continue();
2515 /* Returns 0 if the RP was ok, otherwise there was an error on the RP */
2516 static int await_return_path_close_on_source(MigrationState
*ms
)
2519 * If this is a normal exit then the destination will send a SHUT and the
2520 * rp_thread will exit, however if there's an error we need to cause
2523 if (qemu_file_get_error(ms
->to_dst_file
) && ms
->rp_state
.from_dst_file
) {
2525 * shutdown(2), if we have it, will cause it to unblock if it's stuck
2526 * waiting for the destination.
2528 qemu_file_shutdown(ms
->rp_state
.from_dst_file
);
2529 mark_source_rp_bad(ms
);
2531 trace_await_return_path_close_on_source_joining();
2532 qemu_thread_join(&ms
->rp_state
.rp_thread
);
2533 trace_await_return_path_close_on_source_close();
2534 return ms
->rp_state
.error
;
2538 * Switch from normal iteration to postcopy
2539 * Returns non-0 on error
2541 static int postcopy_start(MigrationState
*ms
)
2544 QIOChannelBuffer
*bioc
;
2546 int64_t time_at_stop
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2547 int64_t bandwidth
= migrate_max_postcopy_bandwidth();
2548 bool restart_block
= false;
2549 int cur_state
= MIGRATION_STATUS_ACTIVE
;
2550 if (!migrate_pause_before_switchover()) {
2551 migrate_set_state(&ms
->state
, MIGRATION_STATUS_ACTIVE
,
2552 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2555 trace_postcopy_start();
2556 qemu_mutex_lock_iothread();
2557 trace_postcopy_start_set_run();
2559 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2560 global_state_store();
2561 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2566 ret
= migration_maybe_pause(ms
, &cur_state
,
2567 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2572 ret
= bdrv_inactivate_all();
2576 restart_block
= true;
2579 * Cause any non-postcopiable, but iterative devices to
2580 * send out their final data.
2582 qemu_savevm_state_complete_precopy(ms
->to_dst_file
, true, false);
2585 * in Finish migrate and with the io-lock held everything should
2586 * be quiet, but we've potentially still got dirty pages and we
2587 * need to tell the destination to throw any pages it's already received
2590 if (migrate_postcopy_ram()) {
2591 if (ram_postcopy_send_discard_bitmap(ms
)) {
2592 error_report("postcopy send discard bitmap failed");
2598 * send rest of state - note things that are doing postcopy
2599 * will notice we're in POSTCOPY_ACTIVE and not actually
2600 * wrap their state up here
2602 /* 0 max-postcopy-bandwidth means unlimited */
2604 qemu_file_set_rate_limit(ms
->to_dst_file
, INT64_MAX
);
2606 qemu_file_set_rate_limit(ms
->to_dst_file
, bandwidth
/ XFER_LIMIT_RATIO
);
2608 if (migrate_postcopy_ram()) {
2609 /* Ping just for debugging, helps line traces up */
2610 qemu_savevm_send_ping(ms
->to_dst_file
, 2);
2614 * While loading the device state we may trigger page transfer
2615 * requests and the fd must be free to process those, and thus
2616 * the destination must read the whole device state off the fd before
2617 * it starts processing it. Unfortunately the ad-hoc migration format
2618 * doesn't allow the destination to know the size to read without fully
2619 * parsing it through each devices load-state code (especially the open
2620 * coded devices that use get/put).
2621 * So we wrap the device state up in a package with a length at the start;
2622 * to do this we use a qemu_buf to hold the whole of the device state.
2624 bioc
= qio_channel_buffer_new(4096);
2625 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-postcopy-buffer");
2626 fb
= qemu_fopen_channel_output(QIO_CHANNEL(bioc
));
2627 object_unref(OBJECT(bioc
));
2630 * Make sure the receiver can get incoming pages before we send the rest
2633 qemu_savevm_send_postcopy_listen(fb
);
2635 qemu_savevm_state_complete_precopy(fb
, false, false);
2636 if (migrate_postcopy_ram()) {
2637 qemu_savevm_send_ping(fb
, 3);
2640 qemu_savevm_send_postcopy_run(fb
);
2642 /* <><> end of stuff going into the package */
2644 /* Last point of recovery; as soon as we send the package the destination
2645 * can open devices and potentially start running.
2646 * Lets just check again we've not got any errors.
2648 ret
= qemu_file_get_error(ms
->to_dst_file
);
2650 error_report("postcopy_start: Migration stream errored (pre package)");
2654 restart_block
= false;
2656 /* Now send that blob */
2657 if (qemu_savevm_send_packaged(ms
->to_dst_file
, bioc
->data
, bioc
->usage
)) {
2662 /* Send a notify to give a chance for anything that needs to happen
2663 * at the transition to postcopy and after the device state; in particular
2664 * spice needs to trigger a transition now
2666 ms
->postcopy_after_devices
= true;
2667 notifier_list_notify(&migration_state_notifiers
, ms
);
2669 ms
->downtime
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) - time_at_stop
;
2671 qemu_mutex_unlock_iothread();
2673 if (migrate_postcopy_ram()) {
2675 * Although this ping is just for debug, it could potentially be
2676 * used for getting a better measurement of downtime at the source.
2678 qemu_savevm_send_ping(ms
->to_dst_file
, 4);
2681 if (migrate_release_ram()) {
2682 ram_postcopy_migrated_memory_release(ms
);
2685 ret
= qemu_file_get_error(ms
->to_dst_file
);
2687 error_report("postcopy_start: Migration stream errored");
2688 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2689 MIGRATION_STATUS_FAILED
);
2697 migrate_set_state(&ms
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2698 MIGRATION_STATUS_FAILED
);
2699 if (restart_block
) {
2700 /* A failure happened early enough that we know the destination hasn't
2701 * accessed block devices, so we're safe to recover.
2703 Error
*local_err
= NULL
;
2705 bdrv_invalidate_cache_all(&local_err
);
2707 error_report_err(local_err
);
2710 qemu_mutex_unlock_iothread();
2715 * migration_maybe_pause: Pause if required to by
2716 * migrate_pause_before_switchover called with the iothread locked
2717 * Returns: 0 on success
2719 static int migration_maybe_pause(MigrationState
*s
,
2720 int *current_active_state
,
2723 if (!migrate_pause_before_switchover()) {
2727 /* Since leaving this state is not atomic with posting the semaphore
2728 * it's possible that someone could have issued multiple migrate_continue
2729 * and the semaphore is incorrectly positive at this point;
2730 * the docs say it's undefined to reinit a semaphore that's already
2731 * init'd, so use timedwait to eat up any existing posts.
2733 while (qemu_sem_timedwait(&s
->pause_sem
, 1) == 0) {
2734 /* This block intentionally left blank */
2737 qemu_mutex_unlock_iothread();
2738 migrate_set_state(&s
->state
, *current_active_state
,
2739 MIGRATION_STATUS_PRE_SWITCHOVER
);
2740 qemu_sem_wait(&s
->pause_sem
);
2741 migrate_set_state(&s
->state
, MIGRATION_STATUS_PRE_SWITCHOVER
,
2743 *current_active_state
= new_state
;
2744 qemu_mutex_lock_iothread();
2746 return s
->state
== new_state
? 0 : -EINVAL
;
2750 * migration_completion: Used by migration_thread when there's not much left.
2751 * The caller 'breaks' the loop when this returns.
2753 * @s: Current migration state
2755 static void migration_completion(MigrationState
*s
)
2758 int current_active_state
= s
->state
;
2760 if (s
->state
== MIGRATION_STATUS_ACTIVE
) {
2761 qemu_mutex_lock_iothread();
2762 s
->downtime_start
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
2763 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, NULL
);
2764 s
->vm_was_running
= runstate_is_running();
2765 ret
= global_state_store();
2768 bool inactivate
= !migrate_colo_enabled();
2769 ret
= vm_stop_force_state(RUN_STATE_FINISH_MIGRATE
);
2771 ret
= migration_maybe_pause(s
, ¤t_active_state
,
2772 MIGRATION_STATUS_DEVICE
);
2775 qemu_file_set_rate_limit(s
->to_dst_file
, INT64_MAX
);
2776 ret
= qemu_savevm_state_complete_precopy(s
->to_dst_file
, false,
2779 if (inactivate
&& ret
>= 0) {
2780 s
->block_inactive
= true;
2783 qemu_mutex_unlock_iothread();
2788 } else if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2789 trace_migration_completion_postcopy_end();
2791 qemu_savevm_state_complete_postcopy(s
->to_dst_file
);
2792 trace_migration_completion_postcopy_end_after_complete();
2796 * If rp was opened we must clean up the thread before
2797 * cleaning everything else up (since if there are no failures
2798 * it will wait for the destination to send it's status in
2801 if (s
->rp_state
.from_dst_file
) {
2803 trace_migration_return_path_end_before();
2804 rp_error
= await_return_path_close_on_source(s
);
2805 trace_migration_return_path_end_after(rp_error
);
2807 goto fail_invalidate
;
2811 if (qemu_file_get_error(s
->to_dst_file
)) {
2812 trace_migration_completion_file_err();
2813 goto fail_invalidate
;
2816 if (!migrate_colo_enabled()) {
2817 migrate_set_state(&s
->state
, current_active_state
,
2818 MIGRATION_STATUS_COMPLETED
);
2824 /* If not doing postcopy, vm_start() will be called: let's regain
2825 * control on images.
2827 if (s
->state
== MIGRATION_STATUS_ACTIVE
||
2828 s
->state
== MIGRATION_STATUS_DEVICE
) {
2829 Error
*local_err
= NULL
;
2831 qemu_mutex_lock_iothread();
2832 bdrv_invalidate_cache_all(&local_err
);
2834 error_report_err(local_err
);
2836 s
->block_inactive
= false;
2838 qemu_mutex_unlock_iothread();
2842 migrate_set_state(&s
->state
, current_active_state
,
2843 MIGRATION_STATUS_FAILED
);
2846 bool migrate_colo_enabled(void)
2848 MigrationState
*s
= migrate_get_current();
2849 return s
->enabled_capabilities
[MIGRATION_CAPABILITY_X_COLO
];
2852 typedef enum MigThrError
{
2853 /* No error detected */
2854 MIG_THR_ERR_NONE
= 0,
2855 /* Detected error, but resumed successfully */
2856 MIG_THR_ERR_RECOVERED
= 1,
2857 /* Detected fatal error, need to exit */
2858 MIG_THR_ERR_FATAL
= 2,
2861 static int postcopy_resume_handshake(MigrationState
*s
)
2863 qemu_savevm_send_postcopy_resume(s
->to_dst_file
);
2865 while (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2866 qemu_sem_wait(&s
->rp_state
.rp_sem
);
2869 if (s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
2876 /* Return zero if success, or <0 for error */
2877 static int postcopy_do_resume(MigrationState
*s
)
2882 * Call all the resume_prepare() hooks, so that modules can be
2883 * ready for the migration resume.
2885 ret
= qemu_savevm_state_resume_prepare(s
);
2887 error_report("%s: resume_prepare() failure detected: %d",
2893 * Last handshake with destination on the resume (destination will
2894 * switch to postcopy-active afterwards)
2896 ret
= postcopy_resume_handshake(s
);
2898 error_report("%s: handshake failed: %d", __func__
, ret
);
2906 * We don't return until we are in a safe state to continue current
2907 * postcopy migration. Returns MIG_THR_ERR_RECOVERED if recovered, or
2908 * MIG_THR_ERR_FATAL if unrecovery failure happened.
2910 static MigThrError
postcopy_pause(MigrationState
*s
)
2912 assert(s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2917 migrate_set_state(&s
->state
, s
->state
,
2918 MIGRATION_STATUS_POSTCOPY_PAUSED
);
2920 /* Current channel is possibly broken. Release it. */
2921 assert(s
->to_dst_file
);
2922 qemu_mutex_lock(&s
->qemu_file_lock
);
2923 file
= s
->to_dst_file
;
2924 s
->to_dst_file
= NULL
;
2925 qemu_mutex_unlock(&s
->qemu_file_lock
);
2927 qemu_file_shutdown(file
);
2930 error_report("Detected IO failure for postcopy. "
2931 "Migration paused.");
2934 * We wait until things fixed up. Then someone will setup the
2935 * status back for us.
2937 while (s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2938 qemu_sem_wait(&s
->postcopy_pause_sem
);
2941 if (s
->state
== MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2942 /* Woken up by a recover procedure. Give it a shot */
2945 * Firstly, let's wake up the return path now, with a new
2946 * return path channel.
2948 qemu_sem_post(&s
->postcopy_pause_rp_sem
);
2950 /* Do the resume logic */
2951 if (postcopy_do_resume(s
) == 0) {
2952 /* Let's continue! */
2953 trace_postcopy_pause_continued();
2954 return MIG_THR_ERR_RECOVERED
;
2957 * Something wrong happened during the recovery, let's
2958 * pause again. Pause is always better than throwing
2964 /* This is not right... Time to quit. */
2965 return MIG_THR_ERR_FATAL
;
2970 static MigThrError
migration_detect_error(MigrationState
*s
)
2973 int state
= s
->state
;
2974 Error
*local_error
= NULL
;
2976 if (state
== MIGRATION_STATUS_CANCELLING
||
2977 state
== MIGRATION_STATUS_CANCELLED
) {
2978 /* End the migration, but don't set the state to failed */
2979 return MIG_THR_ERR_FATAL
;
2982 /* Try to detect any file errors */
2983 ret
= qemu_file_get_error_obj(s
->to_dst_file
, &local_error
);
2985 /* Everything is fine */
2986 assert(!local_error
);
2987 return MIG_THR_ERR_NONE
;
2991 migrate_set_error(s
, local_error
);
2992 error_free(local_error
);
2995 if (state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
&& ret
== -EIO
) {
2997 * For postcopy, we allow the network to be down for a
2998 * while. After that, it can be continued by a
3001 return postcopy_pause(s
);
3004 * For precopy (or postcopy with error outside IO), we fail
3007 migrate_set_state(&s
->state
, state
, MIGRATION_STATUS_FAILED
);
3008 trace_migration_thread_file_err();
3010 /* Time to stop the migration, now. */
3011 return MIG_THR_ERR_FATAL
;
3015 /* How many bytes have we transferred since the beggining of the migration */
3016 static uint64_t migration_total_bytes(MigrationState
*s
)
3018 return qemu_ftell(s
->to_dst_file
) + ram_counters
.multifd_bytes
;
3021 static void migration_calculate_complete(MigrationState
*s
)
3023 uint64_t bytes
= migration_total_bytes(s
);
3024 int64_t end_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3025 int64_t transfer_time
;
3027 s
->total_time
= end_time
- s
->start_time
;
3030 * It's still not set, so we are precopy migration. For
3031 * postcopy, downtime is calculated during postcopy_start().
3033 s
->downtime
= end_time
- s
->downtime_start
;
3036 transfer_time
= s
->total_time
- s
->setup_time
;
3037 if (transfer_time
) {
3038 s
->mbps
= ((double) bytes
* 8.0) / transfer_time
/ 1000;
3042 static void update_iteration_initial_status(MigrationState
*s
)
3045 * Update these three fields at the same time to avoid mismatch info lead
3046 * wrong speed calculation.
3048 s
->iteration_start_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3049 s
->iteration_initial_bytes
= migration_total_bytes(s
);
3050 s
->iteration_initial_pages
= ram_get_total_transferred_pages();
3053 static void migration_update_counters(MigrationState
*s
,
3054 int64_t current_time
)
3056 uint64_t transferred
, transferred_pages
, time_spent
;
3057 uint64_t current_bytes
; /* bytes transferred since the beginning */
3060 if (current_time
< s
->iteration_start_time
+ BUFFER_DELAY
) {
3064 current_bytes
= migration_total_bytes(s
);
3065 transferred
= current_bytes
- s
->iteration_initial_bytes
;
3066 time_spent
= current_time
- s
->iteration_start_time
;
3067 bandwidth
= (double)transferred
/ time_spent
;
3068 s
->threshold_size
= bandwidth
* s
->parameters
.downtime_limit
;
3070 s
->mbps
= (((double) transferred
* 8.0) /
3071 ((double) time_spent
/ 1000.0)) / 1000.0 / 1000.0;
3073 transferred_pages
= ram_get_total_transferred_pages() -
3074 s
->iteration_initial_pages
;
3075 s
->pages_per_second
= (double) transferred_pages
/
3076 (((double) time_spent
/ 1000.0));
3079 * if we haven't sent anything, we don't want to
3080 * recalculate. 10000 is a small enough number for our purposes
3082 if (ram_counters
.dirty_pages_rate
&& transferred
> 10000) {
3083 s
->expected_downtime
= ram_counters
.remaining
/ bandwidth
;
3086 qemu_file_reset_rate_limit(s
->to_dst_file
);
3088 update_iteration_initial_status(s
);
3090 trace_migrate_transferred(transferred
, time_spent
,
3091 bandwidth
, s
->threshold_size
);
3094 /* Migration thread iteration status */
3096 MIG_ITERATE_RESUME
, /* Resume current iteration */
3097 MIG_ITERATE_SKIP
, /* Skip current iteration */
3098 MIG_ITERATE_BREAK
, /* Break the loop */
3102 * Return true if continue to the next iteration directly, false
3105 static MigIterateState
migration_iteration_run(MigrationState
*s
)
3107 uint64_t pending_size
, pend_pre
, pend_compat
, pend_post
;
3108 bool in_postcopy
= s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
;
3110 qemu_savevm_state_pending(s
->to_dst_file
, s
->threshold_size
, &pend_pre
,
3111 &pend_compat
, &pend_post
);
3112 pending_size
= pend_pre
+ pend_compat
+ pend_post
;
3114 trace_migrate_pending(pending_size
, s
->threshold_size
,
3115 pend_pre
, pend_compat
, pend_post
);
3117 if (pending_size
&& pending_size
>= s
->threshold_size
) {
3118 /* Still a significant amount to transfer */
3119 if (!in_postcopy
&& pend_pre
<= s
->threshold_size
&&
3120 atomic_read(&s
->start_postcopy
)) {
3121 if (postcopy_start(s
)) {
3122 error_report("%s: postcopy failed to start", __func__
);
3124 return MIG_ITERATE_SKIP
;
3126 /* Just another iteration step */
3127 qemu_savevm_state_iterate(s
->to_dst_file
,
3128 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
);
3130 trace_migration_thread_low_pending(pending_size
);
3131 migration_completion(s
);
3132 return MIG_ITERATE_BREAK
;
3135 return MIG_ITERATE_RESUME
;
3138 static void migration_iteration_finish(MigrationState
*s
)
3140 /* If we enabled cpu throttling for auto-converge, turn it off. */
3141 cpu_throttle_stop();
3143 qemu_mutex_lock_iothread();
3145 case MIGRATION_STATUS_COMPLETED
:
3146 migration_calculate_complete(s
);
3147 runstate_set(RUN_STATE_POSTMIGRATE
);
3150 case MIGRATION_STATUS_ACTIVE
:
3152 * We should really assert here, but since it's during
3153 * migration, let's try to reduce the usage of assertions.
3155 if (!migrate_colo_enabled()) {
3156 error_report("%s: critical error: calling COLO code without "
3157 "COLO enabled", __func__
);
3159 migrate_start_colo_process(s
);
3161 * Fixme: we will run VM in COLO no matter its old running state.
3162 * After exited COLO, we will keep running.
3164 s
->vm_was_running
= true;
3166 case MIGRATION_STATUS_FAILED
:
3167 case MIGRATION_STATUS_CANCELLED
:
3168 case MIGRATION_STATUS_CANCELLING
:
3169 if (s
->vm_was_running
) {
3172 if (runstate_check(RUN_STATE_FINISH_MIGRATE
)) {
3173 runstate_set(RUN_STATE_POSTMIGRATE
);
3179 /* Should not reach here, but if so, forgive the VM. */
3180 error_report("%s: Unknown ending state %d", __func__
, s
->state
);
3183 migrate_fd_cleanup_schedule(s
);
3184 qemu_mutex_unlock_iothread();
3187 void migration_make_urgent_request(void)
3189 qemu_sem_post(&migrate_get_current()->rate_limit_sem
);
3192 void migration_consume_urgent_request(void)
3194 qemu_sem_wait(&migrate_get_current()->rate_limit_sem
);
3198 * Master migration thread on the source VM.
3199 * It drives the migration and pumps the data down the outgoing channel.
3201 static void *migration_thread(void *opaque
)
3203 MigrationState
*s
= opaque
;
3204 int64_t setup_start
= qemu_clock_get_ms(QEMU_CLOCK_HOST
);
3205 MigThrError thr_error
;
3206 bool urgent
= false;
3208 rcu_register_thread();
3210 object_ref(OBJECT(s
));
3211 update_iteration_initial_status(s
);
3213 qemu_savevm_state_header(s
->to_dst_file
);
3216 * If we opened the return path, we need to make sure dst has it
3219 if (s
->rp_state
.from_dst_file
) {
3220 /* Now tell the dest that it should open its end so it can reply */
3221 qemu_savevm_send_open_return_path(s
->to_dst_file
);
3223 /* And do a ping that will make stuff easier to debug */
3224 qemu_savevm_send_ping(s
->to_dst_file
, 1);
3227 if (migrate_postcopy()) {
3229 * Tell the destination that we *might* want to do postcopy later;
3230 * if the other end can't do postcopy it should fail now, nice and
3233 qemu_savevm_send_postcopy_advise(s
->to_dst_file
);
3236 if (migrate_colo_enabled()) {
3237 /* Notify migration destination that we enable COLO */
3238 qemu_savevm_send_colo_enable(s
->to_dst_file
);
3241 qemu_savevm_state_setup(s
->to_dst_file
);
3243 s
->setup_time
= qemu_clock_get_ms(QEMU_CLOCK_HOST
) - setup_start
;
3244 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3245 MIGRATION_STATUS_ACTIVE
);
3247 trace_migration_thread_setup_complete();
3249 while (s
->state
== MIGRATION_STATUS_ACTIVE
||
3250 s
->state
== MIGRATION_STATUS_POSTCOPY_ACTIVE
) {
3251 int64_t current_time
;
3253 if (urgent
|| !qemu_file_rate_limit(s
->to_dst_file
)) {
3254 MigIterateState iter_state
= migration_iteration_run(s
);
3255 if (iter_state
== MIG_ITERATE_SKIP
) {
3257 } else if (iter_state
== MIG_ITERATE_BREAK
) {
3263 * Try to detect any kind of failures, and see whether we
3264 * should stop the migration now.
3266 thr_error
= migration_detect_error(s
);
3267 if (thr_error
== MIG_THR_ERR_FATAL
) {
3268 /* Stop migration */
3270 } else if (thr_error
== MIG_THR_ERR_RECOVERED
) {
3272 * Just recovered from a e.g. network failure, reset all
3273 * the local variables. This is important to avoid
3274 * breaking transferred_bytes and bandwidth calculation
3276 update_iteration_initial_status(s
);
3279 current_time
= qemu_clock_get_ms(QEMU_CLOCK_REALTIME
);
3281 migration_update_counters(s
, current_time
);
3284 if (qemu_file_rate_limit(s
->to_dst_file
)) {
3285 /* Wait for a delay to do rate limiting OR
3286 * something urgent to post the semaphore.
3288 int ms
= s
->iteration_start_time
+ BUFFER_DELAY
- current_time
;
3289 trace_migration_thread_ratelimit_pre(ms
);
3290 if (qemu_sem_timedwait(&s
->rate_limit_sem
, ms
) == 0) {
3291 /* We were worken by one or more urgent things but
3292 * the timedwait will have consumed one of them.
3293 * The service routine for the urgent wake will dec
3294 * the semaphore itself for each item it consumes,
3295 * so add this one we just eat back.
3297 qemu_sem_post(&s
->rate_limit_sem
);
3300 trace_migration_thread_ratelimit_post(urgent
);
3304 trace_migration_thread_after_loop();
3305 migration_iteration_finish(s
);
3306 object_unref(OBJECT(s
));
3307 rcu_unregister_thread();
3311 void migrate_fd_connect(MigrationState
*s
, Error
*error_in
)
3314 bool resume
= s
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
;
3316 s
->expected_downtime
= s
->parameters
.downtime_limit
;
3317 s
->cleanup_bh
= qemu_bh_new(migrate_fd_cleanup_bh
, s
);
3319 migrate_fd_error(s
, error_in
);
3320 migrate_fd_cleanup(s
);
3325 /* This is a resumed migration */
3326 rate_limit
= INT64_MAX
;
3328 /* This is a fresh new migration */
3329 rate_limit
= s
->parameters
.max_bandwidth
/ XFER_LIMIT_RATIO
;
3331 /* Notify before starting migration thread */
3332 notifier_list_notify(&migration_state_notifiers
, s
);
3335 qemu_file_set_rate_limit(s
->to_dst_file
, rate_limit
);
3336 qemu_file_set_blocking(s
->to_dst_file
, true);
3339 * Open the return path. For postcopy, it is used exclusively. For
3340 * precopy, only if user specified "return-path" capability would
3341 * QEMU uses the return path.
3343 if (migrate_postcopy_ram() || migrate_use_return_path()) {
3344 if (open_return_path_on_source(s
, !resume
)) {
3345 error_report("Unable to open return-path for postcopy");
3346 migrate_set_state(&s
->state
, s
->state
, MIGRATION_STATUS_FAILED
);
3347 migrate_fd_cleanup(s
);
3353 /* Wakeup the main migration thread to do the recovery */
3354 migrate_set_state(&s
->state
, MIGRATION_STATUS_POSTCOPY_PAUSED
,
3355 MIGRATION_STATUS_POSTCOPY_RECOVER
);
3356 qemu_sem_post(&s
->postcopy_pause_sem
);
3360 if (multifd_save_setup() != 0) {
3361 migrate_set_state(&s
->state
, MIGRATION_STATUS_SETUP
,
3362 MIGRATION_STATUS_FAILED
);
3363 migrate_fd_cleanup(s
);
3366 qemu_thread_create(&s
->thread
, "live_migration", migration_thread
, s
,
3367 QEMU_THREAD_JOINABLE
);
3368 s
->migration_thread_running
= true;
3371 void migration_global_dump(Monitor
*mon
)
3373 MigrationState
*ms
= migrate_get_current();
3375 monitor_printf(mon
, "globals:\n");
3376 monitor_printf(mon
, "store-global-state: %s\n",
3377 ms
->store_global_state
? "on" : "off");
3378 monitor_printf(mon
, "only-migratable: %s\n",
3379 only_migratable
? "on" : "off");
3380 monitor_printf(mon
, "send-configuration: %s\n",
3381 ms
->send_configuration
? "on" : "off");
3382 monitor_printf(mon
, "send-section-footer: %s\n",
3383 ms
->send_section_footer
? "on" : "off");
3384 monitor_printf(mon
, "decompress-error-check: %s\n",
3385 ms
->decompress_error_check
? "on" : "off");
3386 monitor_printf(mon
, "clear-bitmap-shift: %u\n",
3387 ms
->clear_bitmap_shift
);
3390 #define DEFINE_PROP_MIG_CAP(name, x) \
3391 DEFINE_PROP_BOOL(name, MigrationState, enabled_capabilities[x], false)
3393 static Property migration_properties
[] = {
3394 DEFINE_PROP_BOOL("store-global-state", MigrationState
,
3395 store_global_state
, true),
3396 DEFINE_PROP_BOOL("send-configuration", MigrationState
,
3397 send_configuration
, true),
3398 DEFINE_PROP_BOOL("send-section-footer", MigrationState
,
3399 send_section_footer
, true),
3400 DEFINE_PROP_BOOL("decompress-error-check", MigrationState
,
3401 decompress_error_check
, true),
3402 DEFINE_PROP_UINT8("x-clear-bitmap-shift", MigrationState
,
3403 clear_bitmap_shift
, CLEAR_BITMAP_SHIFT_DEFAULT
),
3405 /* Migration parameters */
3406 DEFINE_PROP_UINT8("x-compress-level", MigrationState
,
3407 parameters
.compress_level
,
3408 DEFAULT_MIGRATE_COMPRESS_LEVEL
),
3409 DEFINE_PROP_UINT8("x-compress-threads", MigrationState
,
3410 parameters
.compress_threads
,
3411 DEFAULT_MIGRATE_COMPRESS_THREAD_COUNT
),
3412 DEFINE_PROP_BOOL("x-compress-wait-thread", MigrationState
,
3413 parameters
.compress_wait_thread
, true),
3414 DEFINE_PROP_UINT8("x-decompress-threads", MigrationState
,
3415 parameters
.decompress_threads
,
3416 DEFAULT_MIGRATE_DECOMPRESS_THREAD_COUNT
),
3417 DEFINE_PROP_UINT8("x-cpu-throttle-initial", MigrationState
,
3418 parameters
.cpu_throttle_initial
,
3419 DEFAULT_MIGRATE_CPU_THROTTLE_INITIAL
),
3420 DEFINE_PROP_UINT8("x-cpu-throttle-increment", MigrationState
,
3421 parameters
.cpu_throttle_increment
,
3422 DEFAULT_MIGRATE_CPU_THROTTLE_INCREMENT
),
3423 DEFINE_PROP_SIZE("x-max-bandwidth", MigrationState
,
3424 parameters
.max_bandwidth
, MAX_THROTTLE
),
3425 DEFINE_PROP_UINT64("x-downtime-limit", MigrationState
,
3426 parameters
.downtime_limit
,
3427 DEFAULT_MIGRATE_SET_DOWNTIME
),
3428 DEFINE_PROP_UINT32("x-checkpoint-delay", MigrationState
,
3429 parameters
.x_checkpoint_delay
,
3430 DEFAULT_MIGRATE_X_CHECKPOINT_DELAY
),
3431 DEFINE_PROP_UINT8("multifd-channels", MigrationState
,
3432 parameters
.multifd_channels
,
3433 DEFAULT_MIGRATE_MULTIFD_CHANNELS
),
3434 DEFINE_PROP_SIZE("xbzrle-cache-size", MigrationState
,
3435 parameters
.xbzrle_cache_size
,
3436 DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE
),
3437 DEFINE_PROP_SIZE("max-postcopy-bandwidth", MigrationState
,
3438 parameters
.max_postcopy_bandwidth
,
3439 DEFAULT_MIGRATE_MAX_POSTCOPY_BANDWIDTH
),
3440 DEFINE_PROP_UINT8("max-cpu-throttle", MigrationState
,
3441 parameters
.max_cpu_throttle
,
3442 DEFAULT_MIGRATE_MAX_CPU_THROTTLE
),
3443 DEFINE_PROP_SIZE("announce-initial", MigrationState
,
3444 parameters
.announce_initial
,
3445 DEFAULT_MIGRATE_ANNOUNCE_INITIAL
),
3446 DEFINE_PROP_SIZE("announce-max", MigrationState
,
3447 parameters
.announce_max
,
3448 DEFAULT_MIGRATE_ANNOUNCE_MAX
),
3449 DEFINE_PROP_SIZE("announce-rounds", MigrationState
,
3450 parameters
.announce_rounds
,
3451 DEFAULT_MIGRATE_ANNOUNCE_ROUNDS
),
3452 DEFINE_PROP_SIZE("announce-step", MigrationState
,
3453 parameters
.announce_step
,
3454 DEFAULT_MIGRATE_ANNOUNCE_STEP
),
3456 /* Migration capabilities */
3457 DEFINE_PROP_MIG_CAP("x-xbzrle", MIGRATION_CAPABILITY_XBZRLE
),
3458 DEFINE_PROP_MIG_CAP("x-rdma-pin-all", MIGRATION_CAPABILITY_RDMA_PIN_ALL
),
3459 DEFINE_PROP_MIG_CAP("x-auto-converge", MIGRATION_CAPABILITY_AUTO_CONVERGE
),
3460 DEFINE_PROP_MIG_CAP("x-zero-blocks", MIGRATION_CAPABILITY_ZERO_BLOCKS
),
3461 DEFINE_PROP_MIG_CAP("x-compress", MIGRATION_CAPABILITY_COMPRESS
),
3462 DEFINE_PROP_MIG_CAP("x-events", MIGRATION_CAPABILITY_EVENTS
),
3463 DEFINE_PROP_MIG_CAP("x-postcopy-ram", MIGRATION_CAPABILITY_POSTCOPY_RAM
),
3464 DEFINE_PROP_MIG_CAP("x-colo", MIGRATION_CAPABILITY_X_COLO
),
3465 DEFINE_PROP_MIG_CAP("x-release-ram", MIGRATION_CAPABILITY_RELEASE_RAM
),
3466 DEFINE_PROP_MIG_CAP("x-block", MIGRATION_CAPABILITY_BLOCK
),
3467 DEFINE_PROP_MIG_CAP("x-return-path", MIGRATION_CAPABILITY_RETURN_PATH
),
3468 DEFINE_PROP_MIG_CAP("x-multifd", MIGRATION_CAPABILITY_MULTIFD
),
3470 DEFINE_PROP_END_OF_LIST(),
3473 static void migration_class_init(ObjectClass
*klass
, void *data
)
3475 DeviceClass
*dc
= DEVICE_CLASS(klass
);
3477 dc
->user_creatable
= false;
3478 dc
->props
= migration_properties
;
3481 static void migration_instance_finalize(Object
*obj
)
3483 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3484 MigrationParameters
*params
= &ms
->parameters
;
3486 qemu_mutex_destroy(&ms
->error_mutex
);
3487 qemu_mutex_destroy(&ms
->qemu_file_lock
);
3488 g_free(params
->tls_hostname
);
3489 g_free(params
->tls_creds
);
3490 qemu_sem_destroy(&ms
->rate_limit_sem
);
3491 qemu_sem_destroy(&ms
->pause_sem
);
3492 qemu_sem_destroy(&ms
->postcopy_pause_sem
);
3493 qemu_sem_destroy(&ms
->postcopy_pause_rp_sem
);
3494 qemu_sem_destroy(&ms
->rp_state
.rp_sem
);
3495 error_free(ms
->error
);
3498 static void migration_instance_init(Object
*obj
)
3500 MigrationState
*ms
= MIGRATION_OBJ(obj
);
3501 MigrationParameters
*params
= &ms
->parameters
;
3503 ms
->state
= MIGRATION_STATUS_NONE
;
3505 ms
->pages_per_second
= -1;
3506 qemu_sem_init(&ms
->pause_sem
, 0);
3507 qemu_mutex_init(&ms
->error_mutex
);
3509 params
->tls_hostname
= g_strdup("");
3510 params
->tls_creds
= g_strdup("");
3512 /* Set has_* up only for parameter checks */
3513 params
->has_compress_level
= true;
3514 params
->has_compress_threads
= true;
3515 params
->has_decompress_threads
= true;
3516 params
->has_cpu_throttle_initial
= true;
3517 params
->has_cpu_throttle_increment
= true;
3518 params
->has_max_bandwidth
= true;
3519 params
->has_downtime_limit
= true;
3520 params
->has_x_checkpoint_delay
= true;
3521 params
->has_block_incremental
= true;
3522 params
->has_multifd_channels
= true;
3523 params
->has_xbzrle_cache_size
= true;
3524 params
->has_max_postcopy_bandwidth
= true;
3525 params
->has_max_cpu_throttle
= true;
3526 params
->has_announce_initial
= true;
3527 params
->has_announce_max
= true;
3528 params
->has_announce_rounds
= true;
3529 params
->has_announce_step
= true;
3531 qemu_sem_init(&ms
->postcopy_pause_sem
, 0);
3532 qemu_sem_init(&ms
->postcopy_pause_rp_sem
, 0);
3533 qemu_sem_init(&ms
->rp_state
.rp_sem
, 0);
3534 qemu_sem_init(&ms
->rate_limit_sem
, 0);
3535 qemu_mutex_init(&ms
->qemu_file_lock
);
3539 * Return true if check pass, false otherwise. Error will be put
3540 * inside errp if provided.
3542 static bool migration_object_check(MigrationState
*ms
, Error
**errp
)
3544 MigrationCapabilityStatusList
*head
= NULL
;
3545 /* Assuming all off */
3546 bool cap_list
[MIGRATION_CAPABILITY__MAX
] = { 0 }, ret
;
3549 if (!migrate_params_check(&ms
->parameters
, errp
)) {
3553 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
3554 if (ms
->enabled_capabilities
[i
]) {
3555 head
= migrate_cap_add(head
, i
, true);
3559 ret
= migrate_caps_check(cap_list
, head
, errp
);
3561 /* It works with head == NULL */
3562 qapi_free_MigrationCapabilityStatusList(head
);
3567 static const TypeInfo migration_type
= {
3568 .name
= TYPE_MIGRATION
,
3570 * NOTE: TYPE_MIGRATION is not really a device, as the object is
3571 * not created using qdev_create(), it is not attached to the qdev
3572 * device tree, and it is never realized.
3574 * TODO: Make this TYPE_OBJECT once QOM provides something like
3575 * TYPE_DEVICE's "-global" properties.
3577 .parent
= TYPE_DEVICE
,
3578 .class_init
= migration_class_init
,
3579 .class_size
= sizeof(MigrationClass
),
3580 .instance_size
= sizeof(MigrationState
),
3581 .instance_init
= migration_instance_init
,
3582 .instance_finalize
= migration_instance_finalize
,
3585 static void register_migration_types(void)
3587 type_register_static(&migration_type
);
3590 type_init(register_migration_types
);