4 * Copyright (c) 2003-2008 Fabrice Bellard
5 * Copyright (c) 2009-2015 Red Hat Inc
8 * Juan Quintela <quintela@redhat.com>
10 * Permission is hereby granted, free of charge, to any person obtaining a copy
11 * of this software and associated documentation files (the "Software"), to deal
12 * in the Software without restriction, including without limitation the rights
13 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
14 * copies of the Software, and to permit persons to whom the Software is
15 * furnished to do so, subject to the following conditions:
17 * The above copyright notice and this permission notice shall be included in
18 * all copies or substantial portions of the Software.
20 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
21 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
23 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
24 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
25 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
29 #include "qemu/osdep.h"
30 #include "hw/boards.h"
32 #include "migration.h"
33 #include "migration/snapshot.h"
34 #include "migration/vmstate.h"
35 #include "migration/misc.h"
36 #include "migration/register.h"
37 #include "migration/global_state.h"
39 #include "qemu-file-channel.h"
40 #include "qemu-file.h"
42 #include "postcopy-ram.h"
43 #include "qapi/error.h"
44 #include "qapi/qapi-commands-migration.h"
45 #include "qapi/qmp/json-writer.h"
46 #include "qapi/qmp/qerror.h"
47 #include "qemu/error-report.h"
48 #include "sysemu/cpus.h"
49 #include "exec/memory.h"
50 #include "exec/target_page.h"
53 #include "qemu/main-loop.h"
54 #include "block/snapshot.h"
55 #include "qemu/cutils.h"
56 #include "io/channel-buffer.h"
57 #include "io/channel-file.h"
58 #include "sysemu/replay.h"
59 #include "sysemu/runstate.h"
60 #include "sysemu/sysemu.h"
61 #include "sysemu/xen.h"
62 #include "migration/colo.h"
63 #include "qemu/bitmap.h"
64 #include "net/announce.h"
65 #include "qemu/yank.h"
67 const unsigned int postcopy_ram_discard_version
;
69 /* Subcommands for QEMU_VM_COMMAND */
71 MIG_CMD_INVALID
= 0, /* Must be 0 */
72 MIG_CMD_OPEN_RETURN_PATH
, /* Tell the dest to open the Return path */
73 MIG_CMD_PING
, /* Request a PONG on the RP */
75 MIG_CMD_POSTCOPY_ADVISE
, /* Prior to any page transfers, just
76 warn we might want to do PC */
77 MIG_CMD_POSTCOPY_LISTEN
, /* Start listening for incoming
78 pages as it's running. */
79 MIG_CMD_POSTCOPY_RUN
, /* Start execution */
81 MIG_CMD_POSTCOPY_RAM_DISCARD
, /* A list of pages to discard that
82 were previously sent during
83 precopy but are dirty. */
84 MIG_CMD_PACKAGED
, /* Send a wrapped stream within this stream */
85 MIG_CMD_ENABLE_COLO
, /* Enable COLO */
86 MIG_CMD_POSTCOPY_RESUME
, /* resume postcopy on dest */
87 MIG_CMD_RECV_BITMAP
, /* Request for recved bitmap on dst */
91 #define MAX_VM_CMD_PACKAGED_SIZE UINT32_MAX
92 static struct mig_cmd_args
{
93 ssize_t len
; /* -1 = variable */
96 [MIG_CMD_INVALID
] = { .len
= -1, .name
= "INVALID" },
97 [MIG_CMD_OPEN_RETURN_PATH
] = { .len
= 0, .name
= "OPEN_RETURN_PATH" },
98 [MIG_CMD_PING
] = { .len
= sizeof(uint32_t), .name
= "PING" },
99 [MIG_CMD_POSTCOPY_ADVISE
] = { .len
= -1, .name
= "POSTCOPY_ADVISE" },
100 [MIG_CMD_POSTCOPY_LISTEN
] = { .len
= 0, .name
= "POSTCOPY_LISTEN" },
101 [MIG_CMD_POSTCOPY_RUN
] = { .len
= 0, .name
= "POSTCOPY_RUN" },
102 [MIG_CMD_POSTCOPY_RAM_DISCARD
] = {
103 .len
= -1, .name
= "POSTCOPY_RAM_DISCARD" },
104 [MIG_CMD_POSTCOPY_RESUME
] = { .len
= 0, .name
= "POSTCOPY_RESUME" },
105 [MIG_CMD_PACKAGED
] = { .len
= 4, .name
= "PACKAGED" },
106 [MIG_CMD_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
107 [MIG_CMD_MAX
] = { .len
= -1, .name
= "MAX" },
110 /* Note for MIG_CMD_POSTCOPY_ADVISE:
111 * The format of arguments is depending on postcopy mode:
112 * - postcopy RAM only
113 * uint64_t host page size
114 * uint64_t taget page size
116 * - postcopy RAM and postcopy dirty bitmaps
117 * format is the same as for postcopy RAM only
119 * - postcopy dirty bitmaps only
120 * Nothing. Command length field is 0.
122 * Be careful: adding a new postcopy entity with some other parameters should
123 * not break format self-description ability. Good way is to introduce some
124 * generic extendable format with an exception for two old entities.
127 /***********************************************************/
128 /* savevm/loadvm support */
130 static ssize_t
block_writev_buffer(void *opaque
, struct iovec
*iov
, int iovcnt
,
131 int64_t pos
, Error
**errp
)
136 qemu_iovec_init_external(&qiov
, iov
, iovcnt
);
137 ret
= bdrv_writev_vmstate(opaque
, &qiov
, pos
);
145 static ssize_t
block_get_buffer(void *opaque
, uint8_t *buf
, int64_t pos
,
146 size_t size
, Error
**errp
)
148 return bdrv_load_vmstate(opaque
, buf
, pos
, size
);
151 static int bdrv_fclose(void *opaque
, Error
**errp
)
153 return bdrv_flush(opaque
);
156 static const QEMUFileOps bdrv_read_ops
= {
157 .get_buffer
= block_get_buffer
,
161 static const QEMUFileOps bdrv_write_ops
= {
162 .writev_buffer
= block_writev_buffer
,
166 static QEMUFile
*qemu_fopen_bdrv(BlockDriverState
*bs
, int is_writable
)
169 return qemu_fopen_ops(bs
, &bdrv_write_ops
);
171 return qemu_fopen_ops(bs
, &bdrv_read_ops
);
175 /* QEMUFile timer support.
176 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
179 void timer_put(QEMUFile
*f
, QEMUTimer
*ts
)
181 uint64_t expire_time
;
183 expire_time
= timer_expire_time_ns(ts
);
184 qemu_put_be64(f
, expire_time
);
187 void timer_get(QEMUFile
*f
, QEMUTimer
*ts
)
189 uint64_t expire_time
;
191 expire_time
= qemu_get_be64(f
);
192 if (expire_time
!= -1) {
193 timer_mod_ns(ts
, expire_time
);
200 /* VMState timer support.
201 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
204 static int get_timer(QEMUFile
*f
, void *pv
, size_t size
,
205 const VMStateField
*field
)
212 static int put_timer(QEMUFile
*f
, void *pv
, size_t size
,
213 const VMStateField
*field
, JSONWriter
*vmdesc
)
221 const VMStateInfo vmstate_info_timer
= {
228 typedef struct CompatEntry
{
233 typedef struct SaveStateEntry
{
234 QTAILQ_ENTRY(SaveStateEntry
) entry
;
236 uint32_t instance_id
;
239 /* version id read from the stream */
242 /* section id read from the stream */
244 const SaveVMHandlers
*ops
;
245 const VMStateDescription
*vmsd
;
251 typedef struct SaveState
{
252 QTAILQ_HEAD(, SaveStateEntry
) handlers
;
253 SaveStateEntry
*handler_pri_head
[MIG_PRI_MAX
+ 1];
254 int global_section_id
;
257 uint32_t target_page_bits
;
259 MigrationCapability
*capabilities
;
263 static SaveState savevm_state
= {
264 .handlers
= QTAILQ_HEAD_INITIALIZER(savevm_state
.handlers
),
265 .handler_pri_head
= { [MIG_PRI_DEFAULT
... MIG_PRI_MAX
] = NULL
},
266 .global_section_id
= 0,
269 static bool should_validate_capability(int capability
)
271 assert(capability
>= 0 && capability
< MIGRATION_CAPABILITY__MAX
);
272 /* Validate only new capabilities to keep compatibility. */
273 switch (capability
) {
274 case MIGRATION_CAPABILITY_X_IGNORE_SHARED
:
281 static uint32_t get_validatable_capabilities_count(void)
283 MigrationState
*s
= migrate_get_current();
286 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
287 if (should_validate_capability(i
) && s
->enabled_capabilities
[i
]) {
294 static int configuration_pre_save(void *opaque
)
296 SaveState
*state
= opaque
;
297 const char *current_name
= MACHINE_GET_CLASS(current_machine
)->name
;
298 MigrationState
*s
= migrate_get_current();
301 state
->len
= strlen(current_name
);
302 state
->name
= current_name
;
303 state
->target_page_bits
= qemu_target_page_bits();
305 state
->caps_count
= get_validatable_capabilities_count();
306 state
->capabilities
= g_renew(MigrationCapability
, state
->capabilities
,
308 for (i
= j
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
309 if (should_validate_capability(i
) && s
->enabled_capabilities
[i
]) {
310 state
->capabilities
[j
++] = i
;
313 state
->uuid
= qemu_uuid
;
318 static int configuration_post_save(void *opaque
)
320 SaveState
*state
= opaque
;
322 g_free(state
->capabilities
);
323 state
->capabilities
= NULL
;
324 state
->caps_count
= 0;
328 static int configuration_pre_load(void *opaque
)
330 SaveState
*state
= opaque
;
332 /* If there is no target-page-bits subsection it means the source
333 * predates the variable-target-page-bits support and is using the
334 * minimum possible value for this CPU.
336 state
->target_page_bits
= qemu_target_page_bits_min();
340 static bool configuration_validate_capabilities(SaveState
*state
)
343 MigrationState
*s
= migrate_get_current();
344 unsigned long *source_caps_bm
;
347 source_caps_bm
= bitmap_new(MIGRATION_CAPABILITY__MAX
);
348 for (i
= 0; i
< state
->caps_count
; i
++) {
349 MigrationCapability capability
= state
->capabilities
[i
];
350 set_bit(capability
, source_caps_bm
);
353 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
354 bool source_state
, target_state
;
355 if (!should_validate_capability(i
)) {
358 source_state
= test_bit(i
, source_caps_bm
);
359 target_state
= s
->enabled_capabilities
[i
];
360 if (source_state
!= target_state
) {
361 error_report("Capability %s is %s, but received capability is %s",
362 MigrationCapability_str(i
),
363 target_state
? "on" : "off",
364 source_state
? "on" : "off");
366 /* Don't break here to report all failed capabilities */
370 g_free(source_caps_bm
);
374 static int configuration_post_load(void *opaque
, int version_id
)
376 SaveState
*state
= opaque
;
377 const char *current_name
= MACHINE_GET_CLASS(current_machine
)->name
;
380 if (strncmp(state
->name
, current_name
, state
->len
) != 0) {
381 error_report("Machine type received is '%.*s' and local is '%s'",
382 (int) state
->len
, state
->name
, current_name
);
387 if (state
->target_page_bits
!= qemu_target_page_bits()) {
388 error_report("Received TARGET_PAGE_BITS is %d but local is %d",
389 state
->target_page_bits
, qemu_target_page_bits());
394 if (!configuration_validate_capabilities(state
)) {
400 g_free((void *)state
->name
);
403 g_free(state
->capabilities
);
404 state
->capabilities
= NULL
;
405 state
->caps_count
= 0;
410 static int get_capability(QEMUFile
*f
, void *pv
, size_t size
,
411 const VMStateField
*field
)
413 MigrationCapability
*capability
= pv
;
414 char capability_str
[UINT8_MAX
+ 1];
418 len
= qemu_get_byte(f
);
419 qemu_get_buffer(f
, (uint8_t *)capability_str
, len
);
420 capability_str
[len
] = '\0';
421 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
422 if (!strcmp(MigrationCapability_str(i
), capability_str
)) {
427 error_report("Received unknown capability %s", capability_str
);
431 static int put_capability(QEMUFile
*f
, void *pv
, size_t size
,
432 const VMStateField
*field
, JSONWriter
*vmdesc
)
434 MigrationCapability
*capability
= pv
;
435 const char *capability_str
= MigrationCapability_str(*capability
);
436 size_t len
= strlen(capability_str
);
437 assert(len
<= UINT8_MAX
);
439 qemu_put_byte(f
, len
);
440 qemu_put_buffer(f
, (uint8_t *)capability_str
, len
);
444 static const VMStateInfo vmstate_info_capability
= {
445 .name
= "capability",
446 .get
= get_capability
,
447 .put
= put_capability
,
450 /* The target-page-bits subsection is present only if the
451 * target page size is not the same as the default (ie the
452 * minimum page size for a variable-page-size guest CPU).
453 * If it is present then it contains the actual target page
454 * bits for the machine, and migration will fail if the
455 * two ends don't agree about it.
457 static bool vmstate_target_page_bits_needed(void *opaque
)
459 return qemu_target_page_bits()
460 > qemu_target_page_bits_min();
463 static const VMStateDescription vmstate_target_page_bits
= {
464 .name
= "configuration/target-page-bits",
466 .minimum_version_id
= 1,
467 .needed
= vmstate_target_page_bits_needed
,
468 .fields
= (VMStateField
[]) {
469 VMSTATE_UINT32(target_page_bits
, SaveState
),
470 VMSTATE_END_OF_LIST()
474 static bool vmstate_capabilites_needed(void *opaque
)
476 return get_validatable_capabilities_count() > 0;
479 static const VMStateDescription vmstate_capabilites
= {
480 .name
= "configuration/capabilities",
482 .minimum_version_id
= 1,
483 .needed
= vmstate_capabilites_needed
,
484 .fields
= (VMStateField
[]) {
485 VMSTATE_UINT32_V(caps_count
, SaveState
, 1),
486 VMSTATE_VARRAY_UINT32_ALLOC(capabilities
, SaveState
, caps_count
, 1,
487 vmstate_info_capability
,
488 MigrationCapability
),
489 VMSTATE_END_OF_LIST()
493 static bool vmstate_uuid_needed(void *opaque
)
495 return qemu_uuid_set
&& migrate_validate_uuid();
498 static int vmstate_uuid_post_load(void *opaque
, int version_id
)
500 SaveState
*state
= opaque
;
501 char uuid_src
[UUID_FMT_LEN
+ 1];
502 char uuid_dst
[UUID_FMT_LEN
+ 1];
504 if (!qemu_uuid_set
) {
506 * It's warning because user might not know UUID in some cases,
507 * e.g. load an old snapshot
509 qemu_uuid_unparse(&state
->uuid
, uuid_src
);
510 warn_report("UUID is received %s, but local uuid isn't set",
514 if (!qemu_uuid_is_equal(&state
->uuid
, &qemu_uuid
)) {
515 qemu_uuid_unparse(&state
->uuid
, uuid_src
);
516 qemu_uuid_unparse(&qemu_uuid
, uuid_dst
);
517 error_report("UUID received is %s and local is %s", uuid_src
, uuid_dst
);
523 static const VMStateDescription vmstate_uuid
= {
524 .name
= "configuration/uuid",
526 .minimum_version_id
= 1,
527 .needed
= vmstate_uuid_needed
,
528 .post_load
= vmstate_uuid_post_load
,
529 .fields
= (VMStateField
[]) {
530 VMSTATE_UINT8_ARRAY_V(uuid
.data
, SaveState
, sizeof(QemuUUID
), 1),
531 VMSTATE_END_OF_LIST()
535 static const VMStateDescription vmstate_configuration
= {
536 .name
= "configuration",
538 .pre_load
= configuration_pre_load
,
539 .post_load
= configuration_post_load
,
540 .pre_save
= configuration_pre_save
,
541 .post_save
= configuration_post_save
,
542 .fields
= (VMStateField
[]) {
543 VMSTATE_UINT32(len
, SaveState
),
544 VMSTATE_VBUFFER_ALLOC_UINT32(name
, SaveState
, 0, NULL
, len
),
545 VMSTATE_END_OF_LIST()
547 .subsections
= (const VMStateDescription
*[]) {
548 &vmstate_target_page_bits
,
549 &vmstate_capabilites
,
555 static void dump_vmstate_vmsd(FILE *out_file
,
556 const VMStateDescription
*vmsd
, int indent
,
559 static void dump_vmstate_vmsf(FILE *out_file
, const VMStateField
*field
,
562 fprintf(out_file
, "%*s{\n", indent
, "");
564 fprintf(out_file
, "%*s\"field\": \"%s\",\n", indent
, "", field
->name
);
565 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
567 fprintf(out_file
, "%*s\"field_exists\": %s,\n", indent
, "",
568 field
->field_exists
? "true" : "false");
569 fprintf(out_file
, "%*s\"size\": %zu", indent
, "", field
->size
);
570 if (field
->vmsd
!= NULL
) {
571 fprintf(out_file
, ",\n");
572 dump_vmstate_vmsd(out_file
, field
->vmsd
, indent
, false);
574 fprintf(out_file
, "\n%*s}", indent
- 2, "");
577 static void dump_vmstate_vmss(FILE *out_file
,
578 const VMStateDescription
**subsection
,
581 if (*subsection
!= NULL
) {
582 dump_vmstate_vmsd(out_file
, *subsection
, indent
, true);
586 static void dump_vmstate_vmsd(FILE *out_file
,
587 const VMStateDescription
*vmsd
, int indent
,
591 fprintf(out_file
, "%*s{\n", indent
, "");
593 fprintf(out_file
, "%*s\"%s\": {\n", indent
, "", "Description");
596 fprintf(out_file
, "%*s\"name\": \"%s\",\n", indent
, "", vmsd
->name
);
597 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
599 fprintf(out_file
, "%*s\"minimum_version_id\": %d", indent
, "",
600 vmsd
->minimum_version_id
);
601 if (vmsd
->fields
!= NULL
) {
602 const VMStateField
*field
= vmsd
->fields
;
605 fprintf(out_file
, ",\n%*s\"Fields\": [\n", indent
, "");
607 while (field
->name
!= NULL
) {
608 if (field
->flags
& VMS_MUST_EXIST
) {
609 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
614 fprintf(out_file
, ",\n");
616 dump_vmstate_vmsf(out_file
, field
, indent
+ 2);
620 fprintf(out_file
, "\n%*s]", indent
, "");
622 if (vmsd
->subsections
!= NULL
) {
623 const VMStateDescription
**subsection
= vmsd
->subsections
;
626 fprintf(out_file
, ",\n%*s\"Subsections\": [\n", indent
, "");
628 while (*subsection
!= NULL
) {
630 fprintf(out_file
, ",\n");
632 dump_vmstate_vmss(out_file
, subsection
, indent
+ 2);
636 fprintf(out_file
, "\n%*s]", indent
, "");
638 fprintf(out_file
, "\n%*s}", indent
- 2, "");
641 static void dump_machine_type(FILE *out_file
)
645 mc
= MACHINE_GET_CLASS(current_machine
);
647 fprintf(out_file
, " \"vmschkmachine\": {\n");
648 fprintf(out_file
, " \"Name\": \"%s\"\n", mc
->name
);
649 fprintf(out_file
, " },\n");
652 void dump_vmstate_json_to_file(FILE *out_file
)
657 fprintf(out_file
, "{\n");
658 dump_machine_type(out_file
);
661 list
= object_class_get_list(TYPE_DEVICE
, true);
662 for (elt
= list
; elt
; elt
= elt
->next
) {
663 DeviceClass
*dc
= OBJECT_CLASS_CHECK(DeviceClass
, elt
->data
,
673 fprintf(out_file
, ",\n");
675 name
= object_class_get_name(OBJECT_CLASS(dc
));
676 fprintf(out_file
, "%*s\"%s\": {\n", indent
, "", name
);
678 fprintf(out_file
, "%*s\"Name\": \"%s\",\n", indent
, "", name
);
679 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
680 dc
->vmsd
->version_id
);
681 fprintf(out_file
, "%*s\"minimum_version_id\": %d,\n", indent
, "",
682 dc
->vmsd
->minimum_version_id
);
684 dump_vmstate_vmsd(out_file
, dc
->vmsd
, indent
, false);
686 fprintf(out_file
, "\n%*s}", indent
- 2, "");
689 fprintf(out_file
, "\n}\n");
694 static uint32_t calculate_new_instance_id(const char *idstr
)
697 uint32_t instance_id
= 0;
699 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
700 if (strcmp(idstr
, se
->idstr
) == 0
701 && instance_id
<= se
->instance_id
) {
702 instance_id
= se
->instance_id
+ 1;
705 /* Make sure we never loop over without being noticed */
706 assert(instance_id
!= VMSTATE_INSTANCE_ID_ANY
);
710 static int calculate_compat_instance_id(const char *idstr
)
715 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
720 if (strcmp(idstr
, se
->compat
->idstr
) == 0
721 && instance_id
<= se
->compat
->instance_id
) {
722 instance_id
= se
->compat
->instance_id
+ 1;
728 static inline MigrationPriority
save_state_priority(SaveStateEntry
*se
)
731 return se
->vmsd
->priority
;
733 return MIG_PRI_DEFAULT
;
736 static void savevm_state_handler_insert(SaveStateEntry
*nse
)
738 MigrationPriority priority
= save_state_priority(nse
);
742 assert(priority
<= MIG_PRI_MAX
);
744 for (i
= priority
- 1; i
>= 0; i
--) {
745 se
= savevm_state
.handler_pri_head
[i
];
747 assert(save_state_priority(se
) < priority
);
753 QTAILQ_INSERT_BEFORE(se
, nse
, entry
);
755 QTAILQ_INSERT_TAIL(&savevm_state
.handlers
, nse
, entry
);
758 if (savevm_state
.handler_pri_head
[priority
] == NULL
) {
759 savevm_state
.handler_pri_head
[priority
] = nse
;
763 static void savevm_state_handler_remove(SaveStateEntry
*se
)
765 SaveStateEntry
*next
;
766 MigrationPriority priority
= save_state_priority(se
);
768 if (se
== savevm_state
.handler_pri_head
[priority
]) {
769 next
= QTAILQ_NEXT(se
, entry
);
770 if (next
!= NULL
&& save_state_priority(next
) == priority
) {
771 savevm_state
.handler_pri_head
[priority
] = next
;
773 savevm_state
.handler_pri_head
[priority
] = NULL
;
776 QTAILQ_REMOVE(&savevm_state
.handlers
, se
, entry
);
779 /* TODO: Individual devices generally have very little idea about the rest
780 of the system, so instance_id should be removed/replaced.
781 Meanwhile pass -1 as instance_id if you do not already have a clearly
782 distinguishing id for all instances of your device class. */
783 int register_savevm_live(const char *idstr
,
784 uint32_t instance_id
,
786 const SaveVMHandlers
*ops
,
791 se
= g_new0(SaveStateEntry
, 1);
792 se
->version_id
= version_id
;
793 se
->section_id
= savevm_state
.global_section_id
++;
797 /* if this is a live_savem then set is_ram */
798 if (ops
->save_setup
!= NULL
) {
802 pstrcat(se
->idstr
, sizeof(se
->idstr
), idstr
);
804 if (instance_id
== VMSTATE_INSTANCE_ID_ANY
) {
805 se
->instance_id
= calculate_new_instance_id(se
->idstr
);
807 se
->instance_id
= instance_id
;
809 assert(!se
->compat
|| se
->instance_id
== 0);
810 savevm_state_handler_insert(se
);
814 void unregister_savevm(VMStateIf
*obj
, const char *idstr
, void *opaque
)
816 SaveStateEntry
*se
, *new_se
;
820 char *oid
= vmstate_if_get_id(obj
);
822 pstrcpy(id
, sizeof(id
), oid
);
823 pstrcat(id
, sizeof(id
), "/");
827 pstrcat(id
, sizeof(id
), idstr
);
829 QTAILQ_FOREACH_SAFE(se
, &savevm_state
.handlers
, entry
, new_se
) {
830 if (strcmp(se
->idstr
, id
) == 0 && se
->opaque
== opaque
) {
831 savevm_state_handler_remove(se
);
838 int vmstate_register_with_alias_id(VMStateIf
*obj
, uint32_t instance_id
,
839 const VMStateDescription
*vmsd
,
840 void *opaque
, int alias_id
,
841 int required_for_version
,
846 /* If this triggers, alias support can be dropped for the vmsd. */
847 assert(alias_id
== -1 || required_for_version
>= vmsd
->minimum_version_id
);
849 se
= g_new0(SaveStateEntry
, 1);
850 se
->version_id
= vmsd
->version_id
;
851 se
->section_id
= savevm_state
.global_section_id
++;
854 se
->alias_id
= alias_id
;
857 char *id
= vmstate_if_get_id(obj
);
859 if (snprintf(se
->idstr
, sizeof(se
->idstr
), "%s/", id
) >=
861 error_setg(errp
, "Path too long for VMState (%s)", id
);
869 se
->compat
= g_new0(CompatEntry
, 1);
870 pstrcpy(se
->compat
->idstr
, sizeof(se
->compat
->idstr
), vmsd
->name
);
871 se
->compat
->instance_id
= instance_id
== VMSTATE_INSTANCE_ID_ANY
?
872 calculate_compat_instance_id(vmsd
->name
) : instance_id
;
873 instance_id
= VMSTATE_INSTANCE_ID_ANY
;
876 pstrcat(se
->idstr
, sizeof(se
->idstr
), vmsd
->name
);
878 if (instance_id
== VMSTATE_INSTANCE_ID_ANY
) {
879 se
->instance_id
= calculate_new_instance_id(se
->idstr
);
881 se
->instance_id
= instance_id
;
883 assert(!se
->compat
|| se
->instance_id
== 0);
884 savevm_state_handler_insert(se
);
888 void vmstate_unregister(VMStateIf
*obj
, const VMStateDescription
*vmsd
,
891 SaveStateEntry
*se
, *new_se
;
893 QTAILQ_FOREACH_SAFE(se
, &savevm_state
.handlers
, entry
, new_se
) {
894 if (se
->vmsd
== vmsd
&& se
->opaque
== opaque
) {
895 savevm_state_handler_remove(se
);
902 static int vmstate_load(QEMUFile
*f
, SaveStateEntry
*se
)
904 trace_vmstate_load(se
->idstr
, se
->vmsd
? se
->vmsd
->name
: "(old)");
905 if (!se
->vmsd
) { /* Old style */
906 return se
->ops
->load_state(f
, se
->opaque
, se
->load_version_id
);
908 return vmstate_load_state(f
, se
->vmsd
, se
->opaque
, se
->load_version_id
);
911 static void vmstate_save_old_style(QEMUFile
*f
, SaveStateEntry
*se
,
914 int64_t old_offset
, size
;
916 old_offset
= qemu_ftell_fast(f
);
917 se
->ops
->save_state(f
, se
->opaque
);
918 size
= qemu_ftell_fast(f
) - old_offset
;
921 json_writer_int64(vmdesc
, "size", size
);
922 json_writer_start_array(vmdesc
, "fields");
923 json_writer_start_object(vmdesc
, NULL
);
924 json_writer_str(vmdesc
, "name", "data");
925 json_writer_int64(vmdesc
, "size", size
);
926 json_writer_str(vmdesc
, "type", "buffer");
927 json_writer_end_object(vmdesc
);
928 json_writer_end_array(vmdesc
);
932 static int vmstate_save(QEMUFile
*f
, SaveStateEntry
*se
,
935 trace_vmstate_save(se
->idstr
, se
->vmsd
? se
->vmsd
->name
: "(old)");
937 vmstate_save_old_style(f
, se
, vmdesc
);
940 return vmstate_save_state(f
, se
->vmsd
, se
->opaque
, vmdesc
);
944 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
946 static void save_section_header(QEMUFile
*f
, SaveStateEntry
*se
,
947 uint8_t section_type
)
949 qemu_put_byte(f
, section_type
);
950 qemu_put_be32(f
, se
->section_id
);
952 if (section_type
== QEMU_VM_SECTION_FULL
||
953 section_type
== QEMU_VM_SECTION_START
) {
955 size_t len
= strlen(se
->idstr
);
956 qemu_put_byte(f
, len
);
957 qemu_put_buffer(f
, (uint8_t *)se
->idstr
, len
);
959 qemu_put_be32(f
, se
->instance_id
);
960 qemu_put_be32(f
, se
->version_id
);
965 * Write a footer onto device sections that catches cases misformatted device
968 static void save_section_footer(QEMUFile
*f
, SaveStateEntry
*se
)
970 if (migrate_get_current()->send_section_footer
) {
971 qemu_put_byte(f
, QEMU_VM_SECTION_FOOTER
);
972 qemu_put_be32(f
, se
->section_id
);
977 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
978 * command and associated data.
980 * @f: File to send command on
981 * @command: Command type to send
982 * @len: Length of associated data
983 * @data: Data associated with command.
985 static void qemu_savevm_command_send(QEMUFile
*f
,
986 enum qemu_vm_cmd command
,
990 trace_savevm_command_send(command
, len
);
991 qemu_put_byte(f
, QEMU_VM_COMMAND
);
992 qemu_put_be16(f
, (uint16_t)command
);
993 qemu_put_be16(f
, len
);
994 qemu_put_buffer(f
, data
, len
);
998 void qemu_savevm_send_colo_enable(QEMUFile
*f
)
1000 trace_savevm_send_colo_enable();
1001 qemu_savevm_command_send(f
, MIG_CMD_ENABLE_COLO
, 0, NULL
);
1004 void qemu_savevm_send_ping(QEMUFile
*f
, uint32_t value
)
1008 trace_savevm_send_ping(value
);
1009 buf
= cpu_to_be32(value
);
1010 qemu_savevm_command_send(f
, MIG_CMD_PING
, sizeof(value
), (uint8_t *)&buf
);
1013 void qemu_savevm_send_open_return_path(QEMUFile
*f
)
1015 trace_savevm_send_open_return_path();
1016 qemu_savevm_command_send(f
, MIG_CMD_OPEN_RETURN_PATH
, 0, NULL
);
1019 /* We have a buffer of data to send; we don't want that all to be loaded
1020 * by the command itself, so the command contains just the length of the
1021 * extra buffer that we then send straight after it.
1022 * TODO: Must be a better way to organise that
1028 int qemu_savevm_send_packaged(QEMUFile
*f
, const uint8_t *buf
, size_t len
)
1032 if (len
> MAX_VM_CMD_PACKAGED_SIZE
) {
1033 error_report("%s: Unreasonably large packaged state: %zu",
1038 tmp
= cpu_to_be32(len
);
1040 trace_qemu_savevm_send_packaged();
1041 qemu_savevm_command_send(f
, MIG_CMD_PACKAGED
, 4, (uint8_t *)&tmp
);
1043 qemu_put_buffer(f
, buf
, len
);
1048 /* Send prior to any postcopy transfer */
1049 void qemu_savevm_send_postcopy_advise(QEMUFile
*f
)
1051 if (migrate_postcopy_ram()) {
1053 tmp
[0] = cpu_to_be64(ram_pagesize_summary());
1054 tmp
[1] = cpu_to_be64(qemu_target_page_size());
1056 trace_qemu_savevm_send_postcopy_advise();
1057 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_ADVISE
,
1058 16, (uint8_t *)tmp
);
1060 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_ADVISE
, 0, NULL
);
1064 /* Sent prior to starting the destination running in postcopy, discard pages
1065 * that have already been sent but redirtied on the source.
1066 * CMD_POSTCOPY_RAM_DISCARD consist of:
1068 * byte Length of name field (not including 0)
1069 * n x byte RAM block name
1070 * byte 0 terminator (just for safety)
1071 * n x Byte ranges within the named RAMBlock
1072 * be64 Start of the range
1075 * name: RAMBlock name that these entries are part of
1076 * len: Number of page entries
1077 * start_list: 'len' addresses
1078 * length_list: 'len' addresses
1081 void qemu_savevm_send_postcopy_ram_discard(QEMUFile
*f
, const char *name
,
1083 uint64_t *start_list
,
1084 uint64_t *length_list
)
1089 size_t name_len
= strlen(name
);
1091 trace_qemu_savevm_send_postcopy_ram_discard(name
, len
);
1092 assert(name_len
< 256);
1093 buf
= g_malloc0(1 + 1 + name_len
+ 1 + (8 + 8) * len
);
1094 buf
[0] = postcopy_ram_discard_version
;
1096 memcpy(buf
+ 2, name
, name_len
);
1097 tmplen
= 2 + name_len
;
1098 buf
[tmplen
++] = '\0';
1100 for (t
= 0; t
< len
; t
++) {
1101 stq_be_p(buf
+ tmplen
, start_list
[t
]);
1103 stq_be_p(buf
+ tmplen
, length_list
[t
]);
1106 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RAM_DISCARD
, tmplen
, buf
);
1110 /* Get the destination into a state where it can receive postcopy data. */
1111 void qemu_savevm_send_postcopy_listen(QEMUFile
*f
)
1113 trace_savevm_send_postcopy_listen();
1114 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_LISTEN
, 0, NULL
);
1117 /* Kick the destination into running */
1118 void qemu_savevm_send_postcopy_run(QEMUFile
*f
)
1120 trace_savevm_send_postcopy_run();
1121 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RUN
, 0, NULL
);
1124 void qemu_savevm_send_postcopy_resume(QEMUFile
*f
)
1126 trace_savevm_send_postcopy_resume();
1127 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RESUME
, 0, NULL
);
1130 void qemu_savevm_send_recv_bitmap(QEMUFile
*f
, char *block_name
)
1135 trace_savevm_send_recv_bitmap(block_name
);
1137 buf
[0] = len
= strlen(block_name
);
1138 memcpy(buf
+ 1, block_name
, len
);
1140 qemu_savevm_command_send(f
, MIG_CMD_RECV_BITMAP
, len
+ 1, (uint8_t *)buf
);
1143 bool qemu_savevm_state_blocked(Error
**errp
)
1147 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1148 if (se
->vmsd
&& se
->vmsd
->unmigratable
) {
1149 error_setg(errp
, "State blocked by non-migratable device '%s'",
1157 void qemu_savevm_non_migratable_list(strList
**reasons
)
1161 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1162 if (se
->vmsd
&& se
->vmsd
->unmigratable
) {
1163 QAPI_LIST_PREPEND(*reasons
,
1164 g_strdup_printf("non-migratable device: %s",
1170 void qemu_savevm_state_header(QEMUFile
*f
)
1172 trace_savevm_state_header();
1173 qemu_put_be32(f
, QEMU_VM_FILE_MAGIC
);
1174 qemu_put_be32(f
, QEMU_VM_FILE_VERSION
);
1176 if (migrate_get_current()->send_configuration
) {
1177 qemu_put_byte(f
, QEMU_VM_CONFIGURATION
);
1178 vmstate_save_state(f
, &vmstate_configuration
, &savevm_state
, 0);
1182 bool qemu_savevm_state_guest_unplug_pending(void)
1186 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1187 if (se
->vmsd
&& se
->vmsd
->dev_unplug_pending
&&
1188 se
->vmsd
->dev_unplug_pending(se
->opaque
)) {
1196 void qemu_savevm_state_setup(QEMUFile
*f
)
1199 Error
*local_err
= NULL
;
1202 trace_savevm_state_setup();
1203 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1204 if (!se
->ops
|| !se
->ops
->save_setup
) {
1207 if (se
->ops
->is_active
) {
1208 if (!se
->ops
->is_active(se
->opaque
)) {
1212 save_section_header(f
, se
, QEMU_VM_SECTION_START
);
1214 ret
= se
->ops
->save_setup(f
, se
->opaque
);
1215 save_section_footer(f
, se
);
1217 qemu_file_set_error(f
, ret
);
1222 if (precopy_notify(PRECOPY_NOTIFY_SETUP
, &local_err
)) {
1223 error_report_err(local_err
);
1227 int qemu_savevm_state_resume_prepare(MigrationState
*s
)
1232 trace_savevm_state_resume_prepare();
1234 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1235 if (!se
->ops
|| !se
->ops
->resume_prepare
) {
1238 if (se
->ops
->is_active
) {
1239 if (!se
->ops
->is_active(se
->opaque
)) {
1243 ret
= se
->ops
->resume_prepare(s
, se
->opaque
);
1253 * this function has three return values:
1254 * negative: there was one error, and we have -errno.
1255 * 0 : We haven't finished, caller have to go again
1256 * 1 : We have finished, we can go to complete phase
1258 int qemu_savevm_state_iterate(QEMUFile
*f
, bool postcopy
)
1263 trace_savevm_state_iterate();
1264 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1265 if (!se
->ops
|| !se
->ops
->save_live_iterate
) {
1268 if (se
->ops
->is_active
&&
1269 !se
->ops
->is_active(se
->opaque
)) {
1272 if (se
->ops
->is_active_iterate
&&
1273 !se
->ops
->is_active_iterate(se
->opaque
)) {
1277 * In the postcopy phase, any device that doesn't know how to
1278 * do postcopy should have saved it's state in the _complete
1279 * call that's already run, it might get confused if we call
1280 * iterate afterwards.
1283 !(se
->ops
->has_postcopy
&& se
->ops
->has_postcopy(se
->opaque
))) {
1286 if (qemu_file_rate_limit(f
)) {
1289 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1291 save_section_header(f
, se
, QEMU_VM_SECTION_PART
);
1293 ret
= se
->ops
->save_live_iterate(f
, se
->opaque
);
1294 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1295 save_section_footer(f
, se
);
1298 error_report("failed to save SaveStateEntry with id(name): %d(%s)",
1299 se
->section_id
, se
->idstr
);
1300 qemu_file_set_error(f
, ret
);
1303 /* Do not proceed to the next vmstate before this one reported
1304 completion of the current stage. This serializes the migration
1305 and reduces the probability that a faster changing state is
1306 synchronized over and over again. */
1313 static bool should_send_vmdesc(void)
1315 MachineState
*machine
= MACHINE(qdev_get_machine());
1316 bool in_postcopy
= migration_in_postcopy();
1317 return !machine
->suppress_vmdesc
&& !in_postcopy
;
1321 * Calls the save_live_complete_postcopy methods
1322 * causing the last few pages to be sent immediately and doing any associated
1324 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1325 * all the other devices, but that happens at the point we switch to postcopy.
1327 void qemu_savevm_state_complete_postcopy(QEMUFile
*f
)
1332 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1333 if (!se
->ops
|| !se
->ops
->save_live_complete_postcopy
) {
1336 if (se
->ops
->is_active
) {
1337 if (!se
->ops
->is_active(se
->opaque
)) {
1341 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1343 qemu_put_byte(f
, QEMU_VM_SECTION_END
);
1344 qemu_put_be32(f
, se
->section_id
);
1346 ret
= se
->ops
->save_live_complete_postcopy(f
, se
->opaque
);
1347 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1348 save_section_footer(f
, se
);
1350 qemu_file_set_error(f
, ret
);
1355 qemu_put_byte(f
, QEMU_VM_EOF
);
1360 int qemu_savevm_state_complete_precopy_iterable(QEMUFile
*f
, bool in_postcopy
)
1365 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1367 (in_postcopy
&& se
->ops
->has_postcopy
&&
1368 se
->ops
->has_postcopy(se
->opaque
)) ||
1369 !se
->ops
->save_live_complete_precopy
) {
1373 if (se
->ops
->is_active
) {
1374 if (!se
->ops
->is_active(se
->opaque
)) {
1378 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1380 save_section_header(f
, se
, QEMU_VM_SECTION_END
);
1382 ret
= se
->ops
->save_live_complete_precopy(f
, se
->opaque
);
1383 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1384 save_section_footer(f
, se
);
1386 qemu_file_set_error(f
, ret
);
1394 int qemu_savevm_state_complete_precopy_non_iterable(QEMUFile
*f
,
1396 bool inactivate_disks
)
1398 g_autoptr(JSONWriter
) vmdesc
= NULL
;
1403 vmdesc
= json_writer_new(false);
1404 json_writer_start_object(vmdesc
, NULL
);
1405 json_writer_int64(vmdesc
, "page_size", qemu_target_page_size());
1406 json_writer_start_array(vmdesc
, "devices");
1407 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1409 if ((!se
->ops
|| !se
->ops
->save_state
) && !se
->vmsd
) {
1412 if (se
->vmsd
&& !vmstate_save_needed(se
->vmsd
, se
->opaque
)) {
1413 trace_savevm_section_skip(se
->idstr
, se
->section_id
);
1417 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1419 json_writer_start_object(vmdesc
, NULL
);
1420 json_writer_str(vmdesc
, "name", se
->idstr
);
1421 json_writer_int64(vmdesc
, "instance_id", se
->instance_id
);
1423 save_section_header(f
, se
, QEMU_VM_SECTION_FULL
);
1424 ret
= vmstate_save(f
, se
, vmdesc
);
1426 qemu_file_set_error(f
, ret
);
1429 trace_savevm_section_end(se
->idstr
, se
->section_id
, 0);
1430 save_section_footer(f
, se
);
1432 json_writer_end_object(vmdesc
);
1435 if (inactivate_disks
) {
1436 /* Inactivate before sending QEMU_VM_EOF so that the
1437 * bdrv_invalidate_cache_all() on the other end won't fail. */
1438 ret
= bdrv_inactivate_all();
1440 error_report("%s: bdrv_inactivate_all() failed (%d)",
1442 qemu_file_set_error(f
, ret
);
1447 /* Postcopy stream will still be going */
1448 qemu_put_byte(f
, QEMU_VM_EOF
);
1451 json_writer_end_array(vmdesc
);
1452 json_writer_end_object(vmdesc
);
1453 vmdesc_len
= strlen(json_writer_get(vmdesc
));
1455 if (should_send_vmdesc()) {
1456 qemu_put_byte(f
, QEMU_VM_VMDESCRIPTION
);
1457 qemu_put_be32(f
, vmdesc_len
);
1458 qemu_put_buffer(f
, (uint8_t *)json_writer_get(vmdesc
), vmdesc_len
);
1464 int qemu_savevm_state_complete_precopy(QEMUFile
*f
, bool iterable_only
,
1465 bool inactivate_disks
)
1468 Error
*local_err
= NULL
;
1469 bool in_postcopy
= migration_in_postcopy();
1471 if (precopy_notify(PRECOPY_NOTIFY_COMPLETE
, &local_err
)) {
1472 error_report_err(local_err
);
1475 trace_savevm_state_complete_precopy();
1477 cpu_synchronize_all_states();
1479 if (!in_postcopy
|| iterable_only
) {
1480 ret
= qemu_savevm_state_complete_precopy_iterable(f
, in_postcopy
);
1486 if (iterable_only
) {
1490 ret
= qemu_savevm_state_complete_precopy_non_iterable(f
, in_postcopy
,
1501 /* Give an estimate of the amount left to be transferred,
1502 * the result is split into the amount for units that can and
1503 * for units that can't do postcopy.
1505 void qemu_savevm_state_pending(QEMUFile
*f
, uint64_t threshold_size
,
1506 uint64_t *res_precopy_only
,
1507 uint64_t *res_compatible
,
1508 uint64_t *res_postcopy_only
)
1512 *res_precopy_only
= 0;
1513 *res_compatible
= 0;
1514 *res_postcopy_only
= 0;
1517 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1518 if (!se
->ops
|| !se
->ops
->save_live_pending
) {
1521 if (se
->ops
->is_active
) {
1522 if (!se
->ops
->is_active(se
->opaque
)) {
1526 se
->ops
->save_live_pending(f
, se
->opaque
, threshold_size
,
1527 res_precopy_only
, res_compatible
,
1532 void qemu_savevm_state_cleanup(void)
1535 Error
*local_err
= NULL
;
1537 if (precopy_notify(PRECOPY_NOTIFY_CLEANUP
, &local_err
)) {
1538 error_report_err(local_err
);
1541 trace_savevm_state_cleanup();
1542 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1543 if (se
->ops
&& se
->ops
->save_cleanup
) {
1544 se
->ops
->save_cleanup(se
->opaque
);
1549 static int qemu_savevm_state(QEMUFile
*f
, Error
**errp
)
1552 MigrationState
*ms
= migrate_get_current();
1553 MigrationStatus status
;
1555 if (migration_is_running(ms
->state
)) {
1556 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1560 if (migrate_use_block()) {
1561 error_setg(errp
, "Block migration and snapshots are incompatible");
1566 memset(&ram_counters
, 0, sizeof(ram_counters
));
1567 ms
->to_dst_file
= f
;
1569 qemu_mutex_unlock_iothread();
1570 qemu_savevm_state_header(f
);
1571 qemu_savevm_state_setup(f
);
1572 qemu_mutex_lock_iothread();
1574 while (qemu_file_get_error(f
) == 0) {
1575 if (qemu_savevm_state_iterate(f
, false) > 0) {
1580 ret
= qemu_file_get_error(f
);
1582 qemu_savevm_state_complete_precopy(f
, false, false);
1583 ret
= qemu_file_get_error(f
);
1585 qemu_savevm_state_cleanup();
1587 error_setg_errno(errp
, -ret
, "Error while writing VM state");
1591 status
= MIGRATION_STATUS_FAILED
;
1593 status
= MIGRATION_STATUS_COMPLETED
;
1595 migrate_set_state(&ms
->state
, MIGRATION_STATUS_SETUP
, status
);
1597 /* f is outer parameter, it should not stay in global migration state after
1598 * this function finished */
1599 ms
->to_dst_file
= NULL
;
1604 void qemu_savevm_live_state(QEMUFile
*f
)
1606 /* save QEMU_VM_SECTION_END section */
1607 qemu_savevm_state_complete_precopy(f
, true, false);
1608 qemu_put_byte(f
, QEMU_VM_EOF
);
1611 int qemu_save_device_state(QEMUFile
*f
)
1615 if (!migration_in_colo_state()) {
1616 qemu_put_be32(f
, QEMU_VM_FILE_MAGIC
);
1617 qemu_put_be32(f
, QEMU_VM_FILE_VERSION
);
1619 cpu_synchronize_all_states();
1621 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1627 if ((!se
->ops
|| !se
->ops
->save_state
) && !se
->vmsd
) {
1630 if (se
->vmsd
&& !vmstate_save_needed(se
->vmsd
, se
->opaque
)) {
1634 save_section_header(f
, se
, QEMU_VM_SECTION_FULL
);
1636 ret
= vmstate_save(f
, se
, NULL
);
1641 save_section_footer(f
, se
);
1644 qemu_put_byte(f
, QEMU_VM_EOF
);
1646 return qemu_file_get_error(f
);
1649 static SaveStateEntry
*find_se(const char *idstr
, uint32_t instance_id
)
1653 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1654 if (!strcmp(se
->idstr
, idstr
) &&
1655 (instance_id
== se
->instance_id
||
1656 instance_id
== se
->alias_id
))
1658 /* Migrating from an older version? */
1659 if (strstr(se
->idstr
, idstr
) && se
->compat
) {
1660 if (!strcmp(se
->compat
->idstr
, idstr
) &&
1661 (instance_id
== se
->compat
->instance_id
||
1662 instance_id
== se
->alias_id
))
1669 enum LoadVMExitCodes
{
1670 /* Allow a command to quit all layers of nested loadvm loops */
1674 /* ------ incoming postcopy messages ------ */
1675 /* 'advise' arrives before any transfers just to tell us that a postcopy
1676 * *might* happen - it might be skipped if precopy transferred everything
1679 static int loadvm_postcopy_handle_advise(MigrationIncomingState
*mis
,
1682 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_ADVISE
);
1683 uint64_t remote_pagesize_summary
, local_pagesize_summary
, remote_tps
;
1684 Error
*local_err
= NULL
;
1686 trace_loadvm_postcopy_handle_advise();
1687 if (ps
!= POSTCOPY_INCOMING_NONE
) {
1688 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps
);
1694 if (migrate_postcopy_ram()) {
1695 error_report("RAM postcopy is enabled but have 0 byte advise");
1700 if (!migrate_postcopy_ram()) {
1701 error_report("RAM postcopy is disabled but have 16 byte advise");
1706 error_report("CMD_POSTCOPY_ADVISE invalid length (%d)", len
);
1710 if (!postcopy_ram_supported_by_host(mis
)) {
1711 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
1715 remote_pagesize_summary
= qemu_get_be64(mis
->from_src_file
);
1716 local_pagesize_summary
= ram_pagesize_summary();
1718 if (remote_pagesize_summary
!= local_pagesize_summary
) {
1720 * This detects two potential causes of mismatch:
1721 * a) A mismatch in host page sizes
1722 * Some combinations of mismatch are probably possible but it gets
1723 * a bit more complicated. In particular we need to place whole
1724 * host pages on the dest at once, and we need to ensure that we
1725 * handle dirtying to make sure we never end up sending part of
1726 * a hostpage on it's own.
1727 * b) The use of different huge page sizes on source/destination
1728 * a more fine grain test is performed during RAM block migration
1729 * but this test here causes a nice early clear failure, and
1730 * also fails when passed to an older qemu that doesn't
1733 error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1735 remote_pagesize_summary
, local_pagesize_summary
);
1739 remote_tps
= qemu_get_be64(mis
->from_src_file
);
1740 if (remote_tps
!= qemu_target_page_size()) {
1742 * Again, some differences could be dealt with, but for now keep it
1745 error_report("Postcopy needs matching target page sizes (s=%d d=%zd)",
1746 (int)remote_tps
, qemu_target_page_size());
1750 if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_ADVISE
, &local_err
)) {
1751 error_report_err(local_err
);
1755 if (ram_postcopy_incoming_init(mis
)) {
1762 /* After postcopy we will be told to throw some pages away since they're
1763 * dirty and will have to be demand fetched. Must happen before CPU is
1765 * There can be 0..many of these messages, each encoding multiple pages.
1767 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState
*mis
,
1772 PostcopyState ps
= postcopy_state_get();
1774 trace_loadvm_postcopy_ram_handle_discard();
1777 case POSTCOPY_INCOMING_ADVISE
:
1779 tmp
= postcopy_ram_prepare_discard(mis
);
1785 case POSTCOPY_INCOMING_DISCARD
:
1786 /* Expected state */
1790 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1794 /* We're expecting a
1796 * a RAM ID string (length byte, name, 0 term)
1797 * then at least 1 16 byte chunk
1799 if (len
< (1 + 1 + 1 + 1 + 2 * 8)) {
1800 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len
);
1804 tmp
= qemu_get_byte(mis
->from_src_file
);
1805 if (tmp
!= postcopy_ram_discard_version
) {
1806 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp
);
1810 if (!qemu_get_counted_string(mis
->from_src_file
, ramid
)) {
1811 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1814 tmp
= qemu_get_byte(mis
->from_src_file
);
1816 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp
);
1820 len
-= 3 + strlen(ramid
);
1822 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len
);
1825 trace_loadvm_postcopy_ram_handle_discard_header(ramid
, len
);
1827 uint64_t start_addr
, block_length
;
1828 start_addr
= qemu_get_be64(mis
->from_src_file
);
1829 block_length
= qemu_get_be64(mis
->from_src_file
);
1832 int ret
= ram_discard_range(ramid
, start_addr
, block_length
);
1837 trace_loadvm_postcopy_ram_handle_discard_end();
1843 * Triggered by a postcopy_listen command; this thread takes over reading
1844 * the input stream, leaving the main thread free to carry on loading the rest
1845 * of the device state (from RAM).
1846 * (TODO:This could do with being in a postcopy file - but there again it's
1847 * just another input loop, not that postcopy specific)
1849 static void *postcopy_ram_listen_thread(void *opaque
)
1851 MigrationIncomingState
*mis
= migration_incoming_get_current();
1852 QEMUFile
*f
= mis
->from_src_file
;
1854 MigrationState
*migr
= migrate_get_current();
1856 object_ref(OBJECT(migr
));
1858 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
1859 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1860 qemu_sem_post(&mis
->listen_thread_sem
);
1861 trace_postcopy_ram_listen_thread_start();
1863 rcu_register_thread();
1865 * Because we're a thread and not a coroutine we can't yield
1866 * in qemu_file, and thus we must be blocking now.
1868 qemu_file_set_blocking(f
, true);
1869 load_res
= qemu_loadvm_state_main(f
, mis
);
1872 * This is tricky, but, mis->from_src_file can change after it
1873 * returns, when postcopy recovery happened. In the future, we may
1874 * want a wrapper for the QEMUFile handle.
1876 f
= mis
->from_src_file
;
1878 /* And non-blocking again so we don't block in any cleanup */
1879 qemu_file_set_blocking(f
, false);
1881 trace_postcopy_ram_listen_thread_exit();
1883 qemu_file_set_error(f
, load_res
);
1884 dirty_bitmap_mig_cancel_incoming();
1885 if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING
&&
1886 !migrate_postcopy_ram() && migrate_dirty_bitmaps())
1888 error_report("%s: loadvm failed during postcopy: %d. All states "
1889 "are migrated except dirty bitmaps. Some dirty "
1890 "bitmaps may be lost, and present migrated dirty "
1891 "bitmaps are correctly migrated and valid.",
1892 __func__
, load_res
);
1893 load_res
= 0; /* prevent further exit() */
1895 error_report("%s: loadvm failed: %d", __func__
, load_res
);
1896 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1897 MIGRATION_STATUS_FAILED
);
1900 if (load_res
>= 0) {
1902 * This looks good, but it's possible that the device loading in the
1903 * main thread hasn't finished yet, and so we might not be in 'RUN'
1904 * state yet; wait for the end of the main thread.
1906 qemu_event_wait(&mis
->main_thread_load_event
);
1908 postcopy_ram_incoming_cleanup(mis
);
1912 * If something went wrong then we have a bad state so exit;
1913 * depending how far we got it might be possible at this point
1914 * to leave the guest running and fire MCEs for pages that never
1915 * arrived as a desperate recovery step.
1917 rcu_unregister_thread();
1921 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1922 MIGRATION_STATUS_COMPLETED
);
1924 * If everything has worked fine, then the main thread has waited
1925 * for us to start, and we're the last use of the mis.
1926 * (If something broke then qemu will have to exit anyway since it's
1927 * got a bad migration state).
1929 migration_incoming_state_destroy();
1930 qemu_loadvm_state_cleanup();
1932 rcu_unregister_thread();
1933 mis
->have_listen_thread
= false;
1934 postcopy_state_set(POSTCOPY_INCOMING_END
);
1936 object_unref(OBJECT(migr
));
1941 /* After this message we must be able to immediately receive postcopy data */
1942 static int loadvm_postcopy_handle_listen(MigrationIncomingState
*mis
)
1944 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_LISTENING
);
1945 trace_loadvm_postcopy_handle_listen();
1946 Error
*local_err
= NULL
;
1948 if (ps
!= POSTCOPY_INCOMING_ADVISE
&& ps
!= POSTCOPY_INCOMING_DISCARD
) {
1949 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps
);
1952 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
1954 * A rare case, we entered listen without having to do any discards,
1955 * so do the setup that's normally done at the time of the 1st discard.
1957 if (migrate_postcopy_ram()) {
1958 postcopy_ram_prepare_discard(mis
);
1963 * Sensitise RAM - can now generate requests for blocks that don't exist
1964 * However, at this point the CPU shouldn't be running, and the IO
1965 * shouldn't be doing anything yet so don't actually expect requests
1967 if (migrate_postcopy_ram()) {
1968 if (postcopy_ram_incoming_setup(mis
)) {
1969 postcopy_ram_incoming_cleanup(mis
);
1974 if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_LISTEN
, &local_err
)) {
1975 error_report_err(local_err
);
1979 mis
->have_listen_thread
= true;
1980 /* Start up the listening thread and wait for it to signal ready */
1981 qemu_sem_init(&mis
->listen_thread_sem
, 0);
1982 qemu_thread_create(&mis
->listen_thread
, "postcopy/listen",
1983 postcopy_ram_listen_thread
, NULL
,
1984 QEMU_THREAD_DETACHED
);
1985 qemu_sem_wait(&mis
->listen_thread_sem
);
1986 qemu_sem_destroy(&mis
->listen_thread_sem
);
1991 static void loadvm_postcopy_handle_run_bh(void *opaque
)
1993 Error
*local_err
= NULL
;
1994 MigrationIncomingState
*mis
= opaque
;
1996 /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1999 cpu_synchronize_all_post_init();
2001 qemu_announce_self(&mis
->announce_timer
, migrate_announce_params());
2003 /* Make sure all file formats flush their mutable metadata.
2004 * If we get an error here, just don't restart the VM yet. */
2005 bdrv_invalidate_cache_all(&local_err
);
2007 error_report_err(local_err
);
2012 trace_loadvm_postcopy_handle_run_cpu_sync();
2014 trace_loadvm_postcopy_handle_run_vmstart();
2016 dirty_bitmap_mig_before_vm_start();
2019 /* Hold onto your hats, starting the CPU */
2022 /* leave it paused and let management decide when to start the CPU */
2023 runstate_set(RUN_STATE_PAUSED
);
2026 qemu_bh_delete(mis
->bh
);
2029 /* After all discards we can start running and asking for pages */
2030 static int loadvm_postcopy_handle_run(MigrationIncomingState
*mis
)
2032 PostcopyState ps
= postcopy_state_get();
2034 trace_loadvm_postcopy_handle_run();
2035 if (ps
!= POSTCOPY_INCOMING_LISTENING
) {
2036 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps
);
2040 postcopy_state_set(POSTCOPY_INCOMING_RUNNING
);
2041 mis
->bh
= qemu_bh_new(loadvm_postcopy_handle_run_bh
, mis
);
2042 qemu_bh_schedule(mis
->bh
);
2044 /* We need to finish reading the stream from the package
2045 * and also stop reading anything more from the stream that loaded the
2046 * package (since it's now being read by the listener thread).
2047 * LOADVM_QUIT will quit all the layers of nested loadvm loops.
2052 /* We must be with page_request_mutex held */
2053 static gboolean
postcopy_sync_page_req(gpointer key
, gpointer value
,
2056 MigrationIncomingState
*mis
= data
;
2057 void *host_addr
= (void *) key
;
2058 ram_addr_t rb_offset
;
2062 rb
= qemu_ram_block_from_host(host_addr
, true, &rb_offset
);
2065 * This should _never_ happen. However be nice for a migrating VM to
2066 * not crash/assert. Post an error (note: intended to not use *_once
2067 * because we do want to see all the illegal addresses; and this can
2068 * never be triggered by the guest so we're safe) and move on next.
2070 error_report("%s: illegal host addr %p", __func__
, host_addr
);
2071 /* Try the next entry */
2075 ret
= migrate_send_rp_message_req_pages(mis
, rb
, rb_offset
);
2077 /* Please refer to above comment. */
2078 error_report("%s: send rp message failed for addr %p",
2079 __func__
, host_addr
);
2083 trace_postcopy_page_req_sync(host_addr
);
2088 static void migrate_send_rp_req_pages_pending(MigrationIncomingState
*mis
)
2090 WITH_QEMU_LOCK_GUARD(&mis
->page_request_mutex
) {
2091 g_tree_foreach(mis
->page_requested
, postcopy_sync_page_req
, mis
);
2095 static int loadvm_postcopy_handle_resume(MigrationIncomingState
*mis
)
2097 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2098 error_report("%s: illegal resume received", __func__
);
2099 /* Don't fail the load, only for this. */
2104 * Reset the last_rb before we resend any page req to source again, since
2105 * the source should have it reset already.
2107 mis
->last_rb
= NULL
;
2110 * This means source VM is ready to resume the postcopy migration.
2112 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2113 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2115 trace_loadvm_postcopy_handle_resume();
2117 /* Tell source that "we are ready" */
2118 migrate_send_rp_resume_ack(mis
, MIGRATION_RESUME_ACK_VALUE
);
2121 * After a postcopy recovery, the source should have lost the postcopy
2122 * queue, or potentially the requested pages could have been lost during
2123 * the network down phase. Let's re-sync with the source VM by re-sending
2124 * all the pending pages that we eagerly need, so these threads won't get
2125 * blocked too long due to the recovery.
2127 * Without this procedure, the faulted destination VM threads (waiting for
2128 * page requests right before the postcopy is interrupted) can keep hanging
2129 * until the pages are sent by the source during the background copying of
2130 * pages, or another thread faulted on the same address accidentally.
2132 migrate_send_rp_req_pages_pending(mis
);
2135 * It's time to switch state and release the fault thread to continue
2136 * service page faults. Note that this should be explicitly after the
2137 * above call to migrate_send_rp_req_pages_pending(). In short:
2138 * migrate_send_rp_message_req_pages() is not thread safe, yet.
2140 qemu_sem_post(&mis
->postcopy_pause_sem_fault
);
2146 * Immediately following this command is a blob of data containing an embedded
2147 * chunk of migration stream; read it and load it.
2149 * @mis: Incoming state
2150 * @length: Length of packaged data to read
2152 * Returns: Negative values on error
2155 static int loadvm_handle_cmd_packaged(MigrationIncomingState
*mis
)
2159 QIOChannelBuffer
*bioc
;
2161 length
= qemu_get_be32(mis
->from_src_file
);
2162 trace_loadvm_handle_cmd_packaged(length
);
2164 if (length
> MAX_VM_CMD_PACKAGED_SIZE
) {
2165 error_report("Unreasonably large packaged state: %zu", length
);
2169 bioc
= qio_channel_buffer_new(length
);
2170 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-loadvm-buffer");
2171 ret
= qemu_get_buffer(mis
->from_src_file
,
2174 if (ret
!= length
) {
2175 object_unref(OBJECT(bioc
));
2176 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
2178 return (ret
< 0) ? ret
: -EAGAIN
;
2180 bioc
->usage
+= length
;
2181 trace_loadvm_handle_cmd_packaged_received(ret
);
2183 QEMUFile
*packf
= qemu_fopen_channel_input(QIO_CHANNEL(bioc
));
2185 ret
= qemu_loadvm_state_main(packf
, mis
);
2186 trace_loadvm_handle_cmd_packaged_main(ret
);
2188 object_unref(OBJECT(bioc
));
2194 * Handle request that source requests for recved_bitmap on
2195 * destination. Payload format:
2197 * len (1 byte) + ramblock_name (<255 bytes)
2199 static int loadvm_handle_recv_bitmap(MigrationIncomingState
*mis
,
2202 QEMUFile
*file
= mis
->from_src_file
;
2204 char block_name
[256];
2207 cnt
= qemu_get_counted_string(file
, block_name
);
2209 error_report("%s: failed to read block name", __func__
);
2213 /* Validate before using the data */
2214 if (qemu_file_get_error(file
)) {
2215 return qemu_file_get_error(file
);
2218 if (len
!= cnt
+ 1) {
2219 error_report("%s: invalid payload length (%d)", __func__
, len
);
2223 rb
= qemu_ram_block_by_name(block_name
);
2225 error_report("%s: block '%s' not found", __func__
, block_name
);
2229 migrate_send_rp_recv_bitmap(mis
, block_name
);
2231 trace_loadvm_handle_recv_bitmap(block_name
);
2236 static int loadvm_process_enable_colo(MigrationIncomingState
*mis
)
2238 int ret
= migration_incoming_enable_colo();
2241 ret
= colo_init_ram_cache();
2243 migration_incoming_disable_colo();
2250 * Process an incoming 'QEMU_VM_COMMAND'
2251 * 0 just a normal return
2252 * LOADVM_QUIT All good, but exit the loop
2255 static int loadvm_process_command(QEMUFile
*f
)
2257 MigrationIncomingState
*mis
= migration_incoming_get_current();
2262 cmd
= qemu_get_be16(f
);
2263 len
= qemu_get_be16(f
);
2265 /* Check validity before continue processing of cmds */
2266 if (qemu_file_get_error(f
)) {
2267 return qemu_file_get_error(f
);
2270 trace_loadvm_process_command(cmd
, len
);
2271 if (cmd
>= MIG_CMD_MAX
|| cmd
== MIG_CMD_INVALID
) {
2272 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd
, len
);
2276 if (mig_cmd_args
[cmd
].len
!= -1 && mig_cmd_args
[cmd
].len
!= len
) {
2277 error_report("%s received with bad length - expecting %zu, got %d",
2278 mig_cmd_args
[cmd
].name
,
2279 (size_t)mig_cmd_args
[cmd
].len
, len
);
2284 case MIG_CMD_OPEN_RETURN_PATH
:
2285 if (mis
->to_src_file
) {
2286 error_report("CMD_OPEN_RETURN_PATH called when RP already open");
2287 /* Not really a problem, so don't give up */
2290 mis
->to_src_file
= qemu_file_get_return_path(f
);
2291 if (!mis
->to_src_file
) {
2292 error_report("CMD_OPEN_RETURN_PATH failed");
2298 tmp32
= qemu_get_be32(f
);
2299 trace_loadvm_process_command_ping(tmp32
);
2300 if (!mis
->to_src_file
) {
2301 error_report("CMD_PING (0x%x) received with no return path",
2305 migrate_send_rp_pong(mis
, tmp32
);
2308 case MIG_CMD_PACKAGED
:
2309 return loadvm_handle_cmd_packaged(mis
);
2311 case MIG_CMD_POSTCOPY_ADVISE
:
2312 return loadvm_postcopy_handle_advise(mis
, len
);
2314 case MIG_CMD_POSTCOPY_LISTEN
:
2315 return loadvm_postcopy_handle_listen(mis
);
2317 case MIG_CMD_POSTCOPY_RUN
:
2318 return loadvm_postcopy_handle_run(mis
);
2320 case MIG_CMD_POSTCOPY_RAM_DISCARD
:
2321 return loadvm_postcopy_ram_handle_discard(mis
, len
);
2323 case MIG_CMD_POSTCOPY_RESUME
:
2324 return loadvm_postcopy_handle_resume(mis
);
2326 case MIG_CMD_RECV_BITMAP
:
2327 return loadvm_handle_recv_bitmap(mis
, len
);
2329 case MIG_CMD_ENABLE_COLO
:
2330 return loadvm_process_enable_colo(mis
);
2337 * Read a footer off the wire and check that it matches the expected section
2339 * Returns: true if the footer was good
2340 * false if there is a problem (and calls error_report to say why)
2342 static bool check_section_footer(QEMUFile
*f
, SaveStateEntry
*se
)
2346 uint32_t read_section_id
;
2348 if (!migrate_get_current()->send_section_footer
) {
2349 /* No footer to check */
2353 read_mark
= qemu_get_byte(f
);
2355 ret
= qemu_file_get_error(f
);
2357 error_report("%s: Read section footer failed: %d",
2362 if (read_mark
!= QEMU_VM_SECTION_FOOTER
) {
2363 error_report("Missing section footer for %s", se
->idstr
);
2367 read_section_id
= qemu_get_be32(f
);
2368 if (read_section_id
!= se
->load_section_id
) {
2369 error_report("Mismatched section id in footer for %s -"
2370 " read 0x%x expected 0x%x",
2371 se
->idstr
, read_section_id
, se
->load_section_id
);
2380 qemu_loadvm_section_start_full(QEMUFile
*f
, MigrationIncomingState
*mis
)
2382 uint32_t instance_id
, version_id
, section_id
;
2387 /* Read section start */
2388 section_id
= qemu_get_be32(f
);
2389 if (!qemu_get_counted_string(f
, idstr
)) {
2390 error_report("Unable to read ID string for section %u",
2394 instance_id
= qemu_get_be32(f
);
2395 version_id
= qemu_get_be32(f
);
2397 ret
= qemu_file_get_error(f
);
2399 error_report("%s: Failed to read instance/version ID: %d",
2404 trace_qemu_loadvm_state_section_startfull(section_id
, idstr
,
2405 instance_id
, version_id
);
2406 /* Find savevm section */
2407 se
= find_se(idstr
, instance_id
);
2409 error_report("Unknown savevm section or instance '%s' %"PRIu32
". "
2410 "Make sure that your current VM setup matches your "
2411 "saved VM setup, including any hotplugged devices",
2412 idstr
, instance_id
);
2416 /* Validate version */
2417 if (version_id
> se
->version_id
) {
2418 error_report("savevm: unsupported version %d for '%s' v%d",
2419 version_id
, idstr
, se
->version_id
);
2422 se
->load_version_id
= version_id
;
2423 se
->load_section_id
= section_id
;
2425 /* Validate if it is a device's state */
2426 if (xen_enabled() && se
->is_ram
) {
2427 error_report("loadvm: %s RAM loading not allowed on Xen", idstr
);
2431 ret
= vmstate_load(f
, se
);
2433 error_report("error while loading state for instance 0x%"PRIx32
" of"
2434 " device '%s'", instance_id
, idstr
);
2437 if (!check_section_footer(f
, se
)) {
2445 qemu_loadvm_section_part_end(QEMUFile
*f
, MigrationIncomingState
*mis
)
2447 uint32_t section_id
;
2451 section_id
= qemu_get_be32(f
);
2453 ret
= qemu_file_get_error(f
);
2455 error_report("%s: Failed to read section ID: %d",
2460 trace_qemu_loadvm_state_section_partend(section_id
);
2461 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
2462 if (se
->load_section_id
== section_id
) {
2467 error_report("Unknown savevm section %d", section_id
);
2471 ret
= vmstate_load(f
, se
);
2473 error_report("error while loading state section id %d(%s)",
2474 section_id
, se
->idstr
);
2477 if (!check_section_footer(f
, se
)) {
2484 static int qemu_loadvm_state_header(QEMUFile
*f
)
2489 v
= qemu_get_be32(f
);
2490 if (v
!= QEMU_VM_FILE_MAGIC
) {
2491 error_report("Not a migration stream");
2495 v
= qemu_get_be32(f
);
2496 if (v
== QEMU_VM_FILE_VERSION_COMPAT
) {
2497 error_report("SaveVM v2 format is obsolete and don't work anymore");
2500 if (v
!= QEMU_VM_FILE_VERSION
) {
2501 error_report("Unsupported migration stream version");
2505 if (migrate_get_current()->send_configuration
) {
2506 if (qemu_get_byte(f
) != QEMU_VM_CONFIGURATION
) {
2507 error_report("Configuration section missing");
2508 qemu_loadvm_state_cleanup();
2511 ret
= vmstate_load_state(f
, &vmstate_configuration
, &savevm_state
, 0);
2514 qemu_loadvm_state_cleanup();
2521 static int qemu_loadvm_state_setup(QEMUFile
*f
)
2526 trace_loadvm_state_setup();
2527 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
2528 if (!se
->ops
|| !se
->ops
->load_setup
) {
2531 if (se
->ops
->is_active
) {
2532 if (!se
->ops
->is_active(se
->opaque
)) {
2537 ret
= se
->ops
->load_setup(f
, se
->opaque
);
2539 qemu_file_set_error(f
, ret
);
2540 error_report("Load state of device %s failed", se
->idstr
);
2547 void qemu_loadvm_state_cleanup(void)
2551 trace_loadvm_state_cleanup();
2552 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
2553 if (se
->ops
&& se
->ops
->load_cleanup
) {
2554 se
->ops
->load_cleanup(se
->opaque
);
2559 /* Return true if we should continue the migration, or false. */
2560 static bool postcopy_pause_incoming(MigrationIncomingState
*mis
)
2562 trace_postcopy_pause_incoming();
2564 assert(migrate_postcopy_ram());
2566 /* Clear the triggered bit to allow one recovery */
2567 mis
->postcopy_recover_triggered
= false;
2569 assert(mis
->from_src_file
);
2570 qemu_file_shutdown(mis
->from_src_file
);
2571 qemu_fclose(mis
->from_src_file
);
2572 mis
->from_src_file
= NULL
;
2574 assert(mis
->to_src_file
);
2575 qemu_file_shutdown(mis
->to_src_file
);
2576 qemu_mutex_lock(&mis
->rp_mutex
);
2577 qemu_fclose(mis
->to_src_file
);
2578 mis
->to_src_file
= NULL
;
2579 qemu_mutex_unlock(&mis
->rp_mutex
);
2581 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2582 MIGRATION_STATUS_POSTCOPY_PAUSED
);
2584 /* Notify the fault thread for the invalidated file handle */
2585 postcopy_fault_thread_notify(mis
);
2587 error_report("Detected IO failure for postcopy. "
2588 "Migration paused.");
2590 while (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2591 qemu_sem_wait(&mis
->postcopy_pause_sem_dst
);
2594 trace_postcopy_pause_incoming_continued();
2599 int qemu_loadvm_state_main(QEMUFile
*f
, MigrationIncomingState
*mis
)
2601 uint8_t section_type
;
2606 section_type
= qemu_get_byte(f
);
2608 if (qemu_file_get_error(f
)) {
2609 ret
= qemu_file_get_error(f
);
2613 trace_qemu_loadvm_state_section(section_type
);
2614 switch (section_type
) {
2615 case QEMU_VM_SECTION_START
:
2616 case QEMU_VM_SECTION_FULL
:
2617 ret
= qemu_loadvm_section_start_full(f
, mis
);
2622 case QEMU_VM_SECTION_PART
:
2623 case QEMU_VM_SECTION_END
:
2624 ret
= qemu_loadvm_section_part_end(f
, mis
);
2629 case QEMU_VM_COMMAND
:
2630 ret
= loadvm_process_command(f
);
2631 trace_qemu_loadvm_state_section_command(ret
);
2632 if ((ret
< 0) || (ret
== LOADVM_QUIT
)) {
2637 /* This is the end of migration */
2640 error_report("Unknown savevm section type %d", section_type
);
2648 qemu_file_set_error(f
, ret
);
2650 /* Cancel bitmaps incoming regardless of recovery */
2651 dirty_bitmap_mig_cancel_incoming();
2654 * If we are during an active postcopy, then we pause instead
2655 * of bail out to at least keep the VM's dirty data. Note
2656 * that POSTCOPY_INCOMING_LISTENING stage is still not enough,
2657 * during which we're still receiving device states and we
2658 * still haven't yet started the VM on destination.
2660 * Only RAM postcopy supports recovery. Still, if RAM postcopy is
2661 * enabled, canceled bitmaps postcopy will not affect RAM postcopy
2664 if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING
&&
2665 migrate_postcopy_ram() && postcopy_pause_incoming(mis
)) {
2666 /* Reset f to point to the newly created channel */
2667 f
= mis
->from_src_file
;
2674 int qemu_loadvm_state(QEMUFile
*f
)
2676 MigrationIncomingState
*mis
= migration_incoming_get_current();
2677 Error
*local_err
= NULL
;
2680 if (qemu_savevm_state_blocked(&local_err
)) {
2681 error_report_err(local_err
);
2685 ret
= qemu_loadvm_state_header(f
);
2690 if (qemu_loadvm_state_setup(f
) != 0) {
2694 cpu_synchronize_all_pre_loadvm();
2696 ret
= qemu_loadvm_state_main(f
, mis
);
2697 qemu_event_set(&mis
->main_thread_load_event
);
2699 trace_qemu_loadvm_state_post_main(ret
);
2701 if (mis
->have_listen_thread
) {
2702 /* Listen thread still going, can't clean up yet */
2707 ret
= qemu_file_get_error(f
);
2711 * Try to read in the VMDESC section as well, so that dumping tools that
2712 * intercept our migration stream have the chance to see it.
2715 /* We've got to be careful; if we don't read the data and just shut the fd
2716 * then the sender can error if we close while it's still sending.
2717 * We also mustn't read data that isn't there; some transports (RDMA)
2718 * will stall waiting for that data when the source has already closed.
2720 if (ret
== 0 && should_send_vmdesc()) {
2723 uint8_t section_type
= qemu_get_byte(f
);
2725 if (section_type
!= QEMU_VM_VMDESCRIPTION
) {
2726 error_report("Expected vmdescription section, but got %d",
2729 * It doesn't seem worth failing at this point since
2730 * we apparently have an otherwise valid VM state
2733 buf
= g_malloc(0x1000);
2734 size
= qemu_get_be32(f
);
2737 uint32_t read_chunk
= MIN(size
, 0x1000);
2738 qemu_get_buffer(f
, buf
, read_chunk
);
2745 qemu_loadvm_state_cleanup();
2746 cpu_synchronize_all_post_init();
2751 int qemu_load_device_state(QEMUFile
*f
)
2753 MigrationIncomingState
*mis
= migration_incoming_get_current();
2756 /* Load QEMU_VM_SECTION_FULL section */
2757 ret
= qemu_loadvm_state_main(f
, mis
);
2759 error_report("Failed to load device state: %d", ret
);
2763 cpu_synchronize_all_post_init();
2767 int save_snapshot(const char *name
, Error
**errp
)
2769 BlockDriverState
*bs
;
2770 QEMUSnapshotInfo sn1
, *sn
= &sn1
;
2773 int saved_vm_running
;
2774 uint64_t vm_state_size
;
2777 AioContext
*aio_context
;
2779 if (migration_is_blocked(errp
)) {
2783 if (!replay_can_snapshot()) {
2784 error_setg(errp
, "Record/replay does not allow making snapshot "
2785 "right now. Try once more later.");
2789 if (!bdrv_all_can_snapshot(errp
)) {
2793 /* Delete old snapshots of the same name */
2795 if (bdrv_all_delete_snapshot(name
, errp
) < 0) {
2800 bs
= bdrv_all_find_vmstate_bs(errp
);
2804 aio_context
= bdrv_get_aio_context(bs
);
2806 saved_vm_running
= runstate_is_running();
2808 ret
= global_state_store();
2810 error_setg(errp
, "Error saving global state");
2813 vm_stop(RUN_STATE_SAVE_VM
);
2815 bdrv_drain_all_begin();
2817 aio_context_acquire(aio_context
);
2819 memset(sn
, 0, sizeof(*sn
));
2821 /* fill auxiliary fields */
2822 qemu_gettimeofday(&tv
);
2823 sn
->date_sec
= tv
.tv_sec
;
2824 sn
->date_nsec
= tv
.tv_usec
* 1000;
2825 sn
->vm_clock_nsec
= qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
);
2826 if (replay_mode
!= REPLAY_MODE_NONE
) {
2827 sn
->icount
= replay_get_current_icount();
2833 pstrcpy(sn
->name
, sizeof(sn
->name
), name
);
2835 /* cast below needed for OpenBSD where tv_sec is still 'long' */
2836 localtime_r((const time_t *)&tv
.tv_sec
, &tm
);
2837 strftime(sn
->name
, sizeof(sn
->name
), "vm-%Y%m%d%H%M%S", &tm
);
2840 /* save the VM state */
2841 f
= qemu_fopen_bdrv(bs
, 1);
2843 error_setg(errp
, "Could not open VM state file");
2846 ret
= qemu_savevm_state(f
, errp
);
2847 vm_state_size
= qemu_ftell(f
);
2848 ret2
= qemu_fclose(f
);
2857 /* The bdrv_all_create_snapshot() call that follows acquires the AioContext
2858 * for itself. BDRV_POLL_WHILE() does not support nested locking because
2859 * it only releases the lock once. Therefore synchronous I/O will deadlock
2860 * unless we release the AioContext before bdrv_all_create_snapshot().
2862 aio_context_release(aio_context
);
2865 ret
= bdrv_all_create_snapshot(sn
, bs
, vm_state_size
, errp
);
2867 bdrv_all_delete_snapshot(sn
->name
, NULL
);
2875 aio_context_release(aio_context
);
2878 bdrv_drain_all_end();
2880 if (saved_vm_running
) {
2886 void qmp_xen_save_devices_state(const char *filename
, bool has_live
, bool live
,
2890 QIOChannelFile
*ioc
;
2891 int saved_vm_running
;
2895 /* live default to true so old version of Xen tool stack can have a
2896 * successful live migration */
2900 saved_vm_running
= runstate_is_running();
2901 vm_stop(RUN_STATE_SAVE_VM
);
2902 global_state_store_running();
2904 ioc
= qio_channel_file_new_path(filename
, O_WRONLY
| O_CREAT
| O_TRUNC
,
2909 qio_channel_set_name(QIO_CHANNEL(ioc
), "migration-xen-save-state");
2910 f
= qemu_fopen_channel_output(QIO_CHANNEL(ioc
));
2911 object_unref(OBJECT(ioc
));
2912 ret
= qemu_save_device_state(f
);
2913 if (ret
< 0 || qemu_fclose(f
) < 0) {
2914 error_setg(errp
, QERR_IO_ERROR
);
2916 /* libxl calls the QMP command "stop" before calling
2917 * "xen-save-devices-state" and in case of migration failure, libxl
2918 * would call "cont".
2919 * So call bdrv_inactivate_all (release locks) here to let the other
2920 * side of the migration take control of the images.
2922 if (live
&& !saved_vm_running
) {
2923 ret
= bdrv_inactivate_all();
2925 error_setg(errp
, "%s: bdrv_inactivate_all() failed (%d)",
2932 if (saved_vm_running
) {
2937 void qmp_xen_load_devices_state(const char *filename
, Error
**errp
)
2940 QIOChannelFile
*ioc
;
2943 /* Guest must be paused before loading the device state; the RAM state
2944 * will already have been loaded by xc
2946 if (runstate_is_running()) {
2947 error_setg(errp
, "Cannot update device state while vm is running");
2950 vm_stop(RUN_STATE_RESTORE_VM
);
2952 ioc
= qio_channel_file_new_path(filename
, O_RDONLY
| O_BINARY
, 0, errp
);
2956 qio_channel_set_name(QIO_CHANNEL(ioc
), "migration-xen-load-state");
2957 f
= qemu_fopen_channel_input(QIO_CHANNEL(ioc
));
2958 object_unref(OBJECT(ioc
));
2960 ret
= qemu_loadvm_state(f
);
2963 error_setg(errp
, QERR_IO_ERROR
);
2965 migration_incoming_state_destroy();
2968 int load_snapshot(const char *name
, Error
**errp
)
2970 BlockDriverState
*bs_vm_state
;
2971 QEMUSnapshotInfo sn
;
2974 AioContext
*aio_context
;
2975 MigrationIncomingState
*mis
= migration_incoming_get_current();
2977 if (!bdrv_all_can_snapshot(errp
)) {
2980 ret
= bdrv_all_find_snapshot(name
, errp
);
2985 bs_vm_state
= bdrv_all_find_vmstate_bs(errp
);
2989 aio_context
= bdrv_get_aio_context(bs_vm_state
);
2991 /* Don't even try to load empty VM states */
2992 aio_context_acquire(aio_context
);
2993 ret
= bdrv_snapshot_find(bs_vm_state
, &sn
, name
);
2994 aio_context_release(aio_context
);
2997 } else if (sn
.vm_state_size
== 0) {
2998 error_setg(errp
, "This is a disk-only snapshot. Revert to it "
2999 " offline using qemu-img");
3004 * Flush the record/replay queue. Now the VM state is going
3005 * to change. Therefore we don't need to preserve its consistency
3007 replay_flush_events();
3009 /* Flush all IO requests so they don't interfere with the new state. */
3010 bdrv_drain_all_begin();
3012 ret
= bdrv_all_goto_snapshot(name
, errp
);
3017 /* restore the VM state */
3018 f
= qemu_fopen_bdrv(bs_vm_state
, 0);
3020 error_setg(errp
, "Could not open VM state file");
3025 qemu_system_reset(SHUTDOWN_CAUSE_NONE
);
3026 mis
->from_src_file
= f
;
3028 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
3032 aio_context_acquire(aio_context
);
3033 ret
= qemu_loadvm_state(f
);
3034 migration_incoming_state_destroy();
3035 aio_context_release(aio_context
);
3037 bdrv_drain_all_end();
3040 error_setg(errp
, "Error %d while loading VM state", ret
);
3047 bdrv_drain_all_end();
3051 void vmstate_register_ram(MemoryRegion
*mr
, DeviceState
*dev
)
3053 qemu_ram_set_idstr(mr
->ram_block
,
3054 memory_region_name(mr
), dev
);
3055 qemu_ram_set_migratable(mr
->ram_block
);
3058 void vmstate_unregister_ram(MemoryRegion
*mr
, DeviceState
*dev
)
3060 qemu_ram_unset_idstr(mr
->ram_block
);
3061 qemu_ram_unset_migratable(mr
->ram_block
);
3064 void vmstate_register_ram_global(MemoryRegion
*mr
)
3066 vmstate_register_ram(mr
, NULL
);
3069 bool vmstate_check_only_migratable(const VMStateDescription
*vmsd
)
3071 /* check needed if --only-migratable is specified */
3072 if (!only_migratable
) {
3076 return !(vmsd
&& vmsd
->unmigratable
);