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-stats.h"
35 #include "migration/vmstate.h"
36 #include "migration/misc.h"
37 #include "migration/register.h"
38 #include "migration/global_state.h"
39 #include "migration/channel-block.h"
41 #include "qemu-file.h"
43 #include "postcopy-ram.h"
44 #include "qapi/error.h"
45 #include "qapi/qapi-commands-migration.h"
46 #include "qapi/clone-visitor.h"
47 #include "qapi/qapi-builtin-visit.h"
48 #include "qemu/error-report.h"
49 #include "sysemu/cpus.h"
50 #include "exec/memory.h"
51 #include "exec/target_page.h"
55 #include "qemu/main-loop.h"
56 #include "block/snapshot.h"
57 #include "qemu/cutils.h"
58 #include "io/channel-buffer.h"
59 #include "io/channel-file.h"
60 #include "sysemu/replay.h"
61 #include "sysemu/runstate.h"
62 #include "sysemu/sysemu.h"
63 #include "sysemu/xen.h"
64 #include "migration/colo.h"
65 #include "qemu/bitmap.h"
66 #include "net/announce.h"
67 #include "qemu/yank.h"
68 #include "yank_functions.h"
69 #include "sysemu/qtest.h"
72 const unsigned int postcopy_ram_discard_version
;
74 /* Subcommands for QEMU_VM_COMMAND */
76 MIG_CMD_INVALID
= 0, /* Must be 0 */
77 MIG_CMD_OPEN_RETURN_PATH
, /* Tell the dest to open the Return path */
78 MIG_CMD_PING
, /* Request a PONG on the RP */
80 MIG_CMD_POSTCOPY_ADVISE
, /* Prior to any page transfers, just
81 warn we might want to do PC */
82 MIG_CMD_POSTCOPY_LISTEN
, /* Start listening for incoming
83 pages as it's running. */
84 MIG_CMD_POSTCOPY_RUN
, /* Start execution */
86 MIG_CMD_POSTCOPY_RAM_DISCARD
, /* A list of pages to discard that
87 were previously sent during
88 precopy but are dirty. */
89 MIG_CMD_PACKAGED
, /* Send a wrapped stream within this stream */
90 MIG_CMD_ENABLE_COLO
, /* Enable COLO */
91 MIG_CMD_POSTCOPY_RESUME
, /* resume postcopy on dest */
92 MIG_CMD_RECV_BITMAP
, /* Request for recved bitmap on dst */
96 #define MAX_VM_CMD_PACKAGED_SIZE UINT32_MAX
97 static struct mig_cmd_args
{
98 ssize_t len
; /* -1 = variable */
101 [MIG_CMD_INVALID
] = { .len
= -1, .name
= "INVALID" },
102 [MIG_CMD_OPEN_RETURN_PATH
] = { .len
= 0, .name
= "OPEN_RETURN_PATH" },
103 [MIG_CMD_PING
] = { .len
= sizeof(uint32_t), .name
= "PING" },
104 [MIG_CMD_POSTCOPY_ADVISE
] = { .len
= -1, .name
= "POSTCOPY_ADVISE" },
105 [MIG_CMD_POSTCOPY_LISTEN
] = { .len
= 0, .name
= "POSTCOPY_LISTEN" },
106 [MIG_CMD_POSTCOPY_RUN
] = { .len
= 0, .name
= "POSTCOPY_RUN" },
107 [MIG_CMD_POSTCOPY_RAM_DISCARD
] = {
108 .len
= -1, .name
= "POSTCOPY_RAM_DISCARD" },
109 [MIG_CMD_POSTCOPY_RESUME
] = { .len
= 0, .name
= "POSTCOPY_RESUME" },
110 [MIG_CMD_PACKAGED
] = { .len
= 4, .name
= "PACKAGED" },
111 [MIG_CMD_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
112 [MIG_CMD_MAX
] = { .len
= -1, .name
= "MAX" },
115 /* Note for MIG_CMD_POSTCOPY_ADVISE:
116 * The format of arguments is depending on postcopy mode:
117 * - postcopy RAM only
118 * uint64_t host page size
119 * uint64_t target page size
121 * - postcopy RAM and postcopy dirty bitmaps
122 * format is the same as for postcopy RAM only
124 * - postcopy dirty bitmaps only
125 * Nothing. Command length field is 0.
127 * Be careful: adding a new postcopy entity with some other parameters should
128 * not break format self-description ability. Good way is to introduce some
129 * generic extendable format with an exception for two old entities.
132 /***********************************************************/
133 /* savevm/loadvm support */
135 static QEMUFile
*qemu_fopen_bdrv(BlockDriverState
*bs
, int is_writable
)
138 return qemu_file_new_output(QIO_CHANNEL(qio_channel_block_new(bs
)));
140 return qemu_file_new_input(QIO_CHANNEL(qio_channel_block_new(bs
)));
145 /* QEMUFile timer support.
146 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
149 void timer_put(QEMUFile
*f
, QEMUTimer
*ts
)
151 uint64_t expire_time
;
153 expire_time
= timer_expire_time_ns(ts
);
154 qemu_put_be64(f
, expire_time
);
157 void timer_get(QEMUFile
*f
, QEMUTimer
*ts
)
159 uint64_t expire_time
;
161 expire_time
= qemu_get_be64(f
);
162 if (expire_time
!= -1) {
163 timer_mod_ns(ts
, expire_time
);
170 /* VMState timer support.
171 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
174 static int get_timer(QEMUFile
*f
, void *pv
, size_t size
,
175 const VMStateField
*field
)
182 static int put_timer(QEMUFile
*f
, void *pv
, size_t size
,
183 const VMStateField
*field
, JSONWriter
*vmdesc
)
191 const VMStateInfo vmstate_info_timer
= {
198 typedef struct CompatEntry
{
203 typedef struct SaveStateEntry
{
204 QTAILQ_ENTRY(SaveStateEntry
) entry
;
206 uint32_t instance_id
;
209 /* version id read from the stream */
212 /* section id read from the stream */
214 const SaveVMHandlers
*ops
;
215 const VMStateDescription
*vmsd
;
221 typedef struct SaveState
{
222 QTAILQ_HEAD(, SaveStateEntry
) handlers
;
223 SaveStateEntry
*handler_pri_head
[MIG_PRI_MAX
+ 1];
224 int global_section_id
;
227 uint32_t target_page_bits
;
229 MigrationCapability
*capabilities
;
233 static SaveState savevm_state
= {
234 .handlers
= QTAILQ_HEAD_INITIALIZER(savevm_state
.handlers
),
235 .handler_pri_head
= { [MIG_PRI_DEFAULT
... MIG_PRI_MAX
] = NULL
},
236 .global_section_id
= 0,
239 static SaveStateEntry
*find_se(const char *idstr
, uint32_t instance_id
);
241 static bool should_validate_capability(int capability
)
243 assert(capability
>= 0 && capability
< MIGRATION_CAPABILITY__MAX
);
244 /* Validate only new capabilities to keep compatibility. */
245 switch (capability
) {
246 case MIGRATION_CAPABILITY_X_IGNORE_SHARED
:
247 case MIGRATION_CAPABILITY_MAPPED_RAM
:
254 static uint32_t get_validatable_capabilities_count(void)
256 MigrationState
*s
= migrate_get_current();
259 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
260 if (should_validate_capability(i
) && s
->capabilities
[i
]) {
267 static int configuration_pre_save(void *opaque
)
269 SaveState
*state
= opaque
;
270 const char *current_name
= MACHINE_GET_CLASS(current_machine
)->name
;
271 MigrationState
*s
= migrate_get_current();
274 state
->len
= strlen(current_name
);
275 state
->name
= current_name
;
276 state
->target_page_bits
= qemu_target_page_bits();
278 state
->caps_count
= get_validatable_capabilities_count();
279 state
->capabilities
= g_renew(MigrationCapability
, state
->capabilities
,
281 for (i
= j
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
282 if (should_validate_capability(i
) && s
->capabilities
[i
]) {
283 state
->capabilities
[j
++] = i
;
286 state
->uuid
= qemu_uuid
;
291 static int configuration_post_save(void *opaque
)
293 SaveState
*state
= opaque
;
295 g_free(state
->capabilities
);
296 state
->capabilities
= NULL
;
297 state
->caps_count
= 0;
301 static int configuration_pre_load(void *opaque
)
303 SaveState
*state
= opaque
;
305 /* If there is no target-page-bits subsection it means the source
306 * predates the variable-target-page-bits support and is using the
307 * minimum possible value for this CPU.
309 state
->target_page_bits
= qemu_target_page_bits_min();
313 static bool configuration_validate_capabilities(SaveState
*state
)
316 MigrationState
*s
= migrate_get_current();
317 unsigned long *source_caps_bm
;
320 source_caps_bm
= bitmap_new(MIGRATION_CAPABILITY__MAX
);
321 for (i
= 0; i
< state
->caps_count
; i
++) {
322 MigrationCapability capability
= state
->capabilities
[i
];
323 set_bit(capability
, source_caps_bm
);
326 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
327 bool source_state
, target_state
;
328 if (!should_validate_capability(i
)) {
331 source_state
= test_bit(i
, source_caps_bm
);
332 target_state
= s
->capabilities
[i
];
333 if (source_state
!= target_state
) {
334 error_report("Capability %s is %s, but received capability is %s",
335 MigrationCapability_str(i
),
336 target_state
? "on" : "off",
337 source_state
? "on" : "off");
339 /* Don't break here to report all failed capabilities */
343 g_free(source_caps_bm
);
347 static int configuration_post_load(void *opaque
, int version_id
)
349 SaveState
*state
= opaque
;
350 const char *current_name
= MACHINE_GET_CLASS(current_machine
)->name
;
353 if (strncmp(state
->name
, current_name
, state
->len
) != 0) {
354 error_report("Machine type received is '%.*s' and local is '%s'",
355 (int) state
->len
, state
->name
, current_name
);
360 if (state
->target_page_bits
!= qemu_target_page_bits()) {
361 error_report("Received TARGET_PAGE_BITS is %d but local is %d",
362 state
->target_page_bits
, qemu_target_page_bits());
367 if (!configuration_validate_capabilities(state
)) {
373 g_free((void *)state
->name
);
376 g_free(state
->capabilities
);
377 state
->capabilities
= NULL
;
378 state
->caps_count
= 0;
383 static int get_capability(QEMUFile
*f
, void *pv
, size_t size
,
384 const VMStateField
*field
)
386 MigrationCapability
*capability
= pv
;
387 char capability_str
[UINT8_MAX
+ 1];
391 len
= qemu_get_byte(f
);
392 qemu_get_buffer(f
, (uint8_t *)capability_str
, len
);
393 capability_str
[len
] = '\0';
394 for (i
= 0; i
< MIGRATION_CAPABILITY__MAX
; i
++) {
395 if (!strcmp(MigrationCapability_str(i
), capability_str
)) {
400 error_report("Received unknown capability %s", capability_str
);
404 static int put_capability(QEMUFile
*f
, void *pv
, size_t size
,
405 const VMStateField
*field
, JSONWriter
*vmdesc
)
407 MigrationCapability
*capability
= pv
;
408 const char *capability_str
= MigrationCapability_str(*capability
);
409 size_t len
= strlen(capability_str
);
410 assert(len
<= UINT8_MAX
);
412 qemu_put_byte(f
, len
);
413 qemu_put_buffer(f
, (uint8_t *)capability_str
, len
);
417 static const VMStateInfo vmstate_info_capability
= {
418 .name
= "capability",
419 .get
= get_capability
,
420 .put
= put_capability
,
423 /* The target-page-bits subsection is present only if the
424 * target page size is not the same as the default (ie the
425 * minimum page size for a variable-page-size guest CPU).
426 * If it is present then it contains the actual target page
427 * bits for the machine, and migration will fail if the
428 * two ends don't agree about it.
430 static bool vmstate_target_page_bits_needed(void *opaque
)
432 return qemu_target_page_bits()
433 > qemu_target_page_bits_min();
436 static const VMStateDescription vmstate_target_page_bits
= {
437 .name
= "configuration/target-page-bits",
439 .minimum_version_id
= 1,
440 .needed
= vmstate_target_page_bits_needed
,
441 .fields
= (const VMStateField
[]) {
442 VMSTATE_UINT32(target_page_bits
, SaveState
),
443 VMSTATE_END_OF_LIST()
447 static bool vmstate_capabilites_needed(void *opaque
)
449 return get_validatable_capabilities_count() > 0;
452 static const VMStateDescription vmstate_capabilites
= {
453 .name
= "configuration/capabilities",
455 .minimum_version_id
= 1,
456 .needed
= vmstate_capabilites_needed
,
457 .fields
= (const VMStateField
[]) {
458 VMSTATE_UINT32_V(caps_count
, SaveState
, 1),
459 VMSTATE_VARRAY_UINT32_ALLOC(capabilities
, SaveState
, caps_count
, 1,
460 vmstate_info_capability
,
461 MigrationCapability
),
462 VMSTATE_END_OF_LIST()
466 static bool vmstate_uuid_needed(void *opaque
)
468 return qemu_uuid_set
&& migrate_validate_uuid();
471 static int vmstate_uuid_post_load(void *opaque
, int version_id
)
473 SaveState
*state
= opaque
;
474 char uuid_src
[UUID_STR_LEN
];
475 char uuid_dst
[UUID_STR_LEN
];
477 if (!qemu_uuid_set
) {
479 * It's warning because user might not know UUID in some cases,
480 * e.g. load an old snapshot
482 qemu_uuid_unparse(&state
->uuid
, uuid_src
);
483 warn_report("UUID is received %s, but local uuid isn't set",
487 if (!qemu_uuid_is_equal(&state
->uuid
, &qemu_uuid
)) {
488 qemu_uuid_unparse(&state
->uuid
, uuid_src
);
489 qemu_uuid_unparse(&qemu_uuid
, uuid_dst
);
490 error_report("UUID received is %s and local is %s", uuid_src
, uuid_dst
);
496 static const VMStateDescription vmstate_uuid
= {
497 .name
= "configuration/uuid",
499 .minimum_version_id
= 1,
500 .needed
= vmstate_uuid_needed
,
501 .post_load
= vmstate_uuid_post_load
,
502 .fields
= (const VMStateField
[]) {
503 VMSTATE_UINT8_ARRAY_V(uuid
.data
, SaveState
, sizeof(QemuUUID
), 1),
504 VMSTATE_END_OF_LIST()
508 static const VMStateDescription vmstate_configuration
= {
509 .name
= "configuration",
511 .pre_load
= configuration_pre_load
,
512 .post_load
= configuration_post_load
,
513 .pre_save
= configuration_pre_save
,
514 .post_save
= configuration_post_save
,
515 .fields
= (const VMStateField
[]) {
516 VMSTATE_UINT32(len
, SaveState
),
517 VMSTATE_VBUFFER_ALLOC_UINT32(name
, SaveState
, 0, NULL
, len
),
518 VMSTATE_END_OF_LIST()
520 .subsections
= (const VMStateDescription
* const []) {
521 &vmstate_target_page_bits
,
522 &vmstate_capabilites
,
528 static void dump_vmstate_vmsd(FILE *out_file
,
529 const VMStateDescription
*vmsd
, int indent
,
532 static void dump_vmstate_vmsf(FILE *out_file
, const VMStateField
*field
,
535 fprintf(out_file
, "%*s{\n", indent
, "");
537 fprintf(out_file
, "%*s\"field\": \"%s\",\n", indent
, "", field
->name
);
538 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
540 fprintf(out_file
, "%*s\"field_exists\": %s,\n", indent
, "",
541 field
->field_exists
? "true" : "false");
542 if (field
->flags
& VMS_ARRAY
) {
543 fprintf(out_file
, "%*s\"num\": %d,\n", indent
, "", field
->num
);
545 fprintf(out_file
, "%*s\"size\": %zu", indent
, "", field
->size
);
546 if (field
->vmsd
!= NULL
) {
547 fprintf(out_file
, ",\n");
548 dump_vmstate_vmsd(out_file
, field
->vmsd
, indent
, false);
550 fprintf(out_file
, "\n%*s}", indent
- 2, "");
553 static void dump_vmstate_vmss(FILE *out_file
,
554 const VMStateDescription
*subsection
,
557 if (subsection
!= NULL
) {
558 dump_vmstate_vmsd(out_file
, subsection
, indent
, true);
562 static void dump_vmstate_vmsd(FILE *out_file
,
563 const VMStateDescription
*vmsd
, int indent
,
567 fprintf(out_file
, "%*s{\n", indent
, "");
569 fprintf(out_file
, "%*s\"%s\": {\n", indent
, "", "Description");
572 fprintf(out_file
, "%*s\"name\": \"%s\",\n", indent
, "", vmsd
->name
);
573 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
575 fprintf(out_file
, "%*s\"minimum_version_id\": %d", indent
, "",
576 vmsd
->minimum_version_id
);
577 if (vmsd
->fields
!= NULL
) {
578 const VMStateField
*field
= vmsd
->fields
;
581 fprintf(out_file
, ",\n%*s\"Fields\": [\n", indent
, "");
583 while (field
->name
!= NULL
) {
584 if (field
->flags
& VMS_MUST_EXIST
) {
585 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
590 fprintf(out_file
, ",\n");
592 dump_vmstate_vmsf(out_file
, field
, indent
+ 2);
596 assert(field
->flags
== VMS_END
);
597 fprintf(out_file
, "\n%*s]", indent
, "");
599 if (vmsd
->subsections
!= NULL
) {
600 const VMStateDescription
* const *subsection
= vmsd
->subsections
;
603 fprintf(out_file
, ",\n%*s\"Subsections\": [\n", indent
, "");
605 while (*subsection
!= NULL
) {
607 fprintf(out_file
, ",\n");
609 dump_vmstate_vmss(out_file
, *subsection
, indent
+ 2);
613 fprintf(out_file
, "\n%*s]", indent
, "");
615 fprintf(out_file
, "\n%*s}", indent
- 2, "");
618 static void dump_machine_type(FILE *out_file
)
622 mc
= MACHINE_GET_CLASS(current_machine
);
624 fprintf(out_file
, " \"vmschkmachine\": {\n");
625 fprintf(out_file
, " \"Name\": \"%s\"\n", mc
->name
);
626 fprintf(out_file
, " },\n");
629 void dump_vmstate_json_to_file(FILE *out_file
)
634 fprintf(out_file
, "{\n");
635 dump_machine_type(out_file
);
638 list
= object_class_get_list(TYPE_DEVICE
, true);
639 for (elt
= list
; elt
; elt
= elt
->next
) {
640 DeviceClass
*dc
= OBJECT_CLASS_CHECK(DeviceClass
, elt
->data
,
650 fprintf(out_file
, ",\n");
652 name
= object_class_get_name(OBJECT_CLASS(dc
));
653 fprintf(out_file
, "%*s\"%s\": {\n", indent
, "", name
);
655 fprintf(out_file
, "%*s\"Name\": \"%s\",\n", indent
, "", name
);
656 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
657 dc
->vmsd
->version_id
);
658 fprintf(out_file
, "%*s\"minimum_version_id\": %d,\n", indent
, "",
659 dc
->vmsd
->minimum_version_id
);
661 dump_vmstate_vmsd(out_file
, dc
->vmsd
, indent
, false);
663 fprintf(out_file
, "\n%*s}", indent
- 2, "");
666 fprintf(out_file
, "\n}\n");
671 static uint32_t calculate_new_instance_id(const char *idstr
)
674 uint32_t instance_id
= 0;
676 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
677 if (strcmp(idstr
, se
->idstr
) == 0
678 && instance_id
<= se
->instance_id
) {
679 instance_id
= se
->instance_id
+ 1;
682 /* Make sure we never loop over without being noticed */
683 assert(instance_id
!= VMSTATE_INSTANCE_ID_ANY
);
687 static int calculate_compat_instance_id(const char *idstr
)
692 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
697 if (strcmp(idstr
, se
->compat
->idstr
) == 0
698 && instance_id
<= se
->compat
->instance_id
) {
699 instance_id
= se
->compat
->instance_id
+ 1;
705 static inline MigrationPriority
save_state_priority(SaveStateEntry
*se
)
708 return se
->vmsd
->priority
;
710 return MIG_PRI_DEFAULT
;
713 static void savevm_state_handler_insert(SaveStateEntry
*nse
)
715 MigrationPriority priority
= save_state_priority(nse
);
719 assert(priority
<= MIG_PRI_MAX
);
722 * This should never happen otherwise migration will probably fail
723 * silently somewhere because we can be wrongly applying one
724 * object properties upon another one. Bail out ASAP.
726 if (find_se(nse
->idstr
, nse
->instance_id
)) {
727 error_report("%s: Detected duplicate SaveStateEntry: "
728 "id=%s, instance_id=0x%"PRIx32
, __func__
,
729 nse
->idstr
, nse
->instance_id
);
733 for (i
= priority
- 1; i
>= 0; i
--) {
734 se
= savevm_state
.handler_pri_head
[i
];
736 assert(save_state_priority(se
) < priority
);
742 QTAILQ_INSERT_BEFORE(se
, nse
, entry
);
744 QTAILQ_INSERT_TAIL(&savevm_state
.handlers
, nse
, entry
);
747 if (savevm_state
.handler_pri_head
[priority
] == NULL
) {
748 savevm_state
.handler_pri_head
[priority
] = nse
;
752 static void savevm_state_handler_remove(SaveStateEntry
*se
)
754 SaveStateEntry
*next
;
755 MigrationPriority priority
= save_state_priority(se
);
757 if (se
== savevm_state
.handler_pri_head
[priority
]) {
758 next
= QTAILQ_NEXT(se
, entry
);
759 if (next
!= NULL
&& save_state_priority(next
) == priority
) {
760 savevm_state
.handler_pri_head
[priority
] = next
;
762 savevm_state
.handler_pri_head
[priority
] = NULL
;
765 QTAILQ_REMOVE(&savevm_state
.handlers
, se
, entry
);
768 /* TODO: Individual devices generally have very little idea about the rest
769 of the system, so instance_id should be removed/replaced.
770 Meanwhile pass -1 as instance_id if you do not already have a clearly
771 distinguishing id for all instances of your device class. */
772 int register_savevm_live(const char *idstr
,
773 uint32_t instance_id
,
775 const SaveVMHandlers
*ops
,
780 se
= g_new0(SaveStateEntry
, 1);
781 se
->version_id
= version_id
;
782 se
->section_id
= savevm_state
.global_section_id
++;
786 /* if this is a live_savem then set is_ram */
787 if (ops
->save_setup
!= NULL
) {
791 pstrcat(se
->idstr
, sizeof(se
->idstr
), idstr
);
793 if (instance_id
== VMSTATE_INSTANCE_ID_ANY
) {
794 se
->instance_id
= calculate_new_instance_id(se
->idstr
);
796 se
->instance_id
= instance_id
;
798 assert(!se
->compat
|| se
->instance_id
== 0);
799 savevm_state_handler_insert(se
);
803 void unregister_savevm(VMStateIf
*obj
, const char *idstr
, void *opaque
)
805 SaveStateEntry
*se
, *new_se
;
809 char *oid
= vmstate_if_get_id(obj
);
811 pstrcpy(id
, sizeof(id
), oid
);
812 pstrcat(id
, sizeof(id
), "/");
816 pstrcat(id
, sizeof(id
), idstr
);
818 QTAILQ_FOREACH_SAFE(se
, &savevm_state
.handlers
, entry
, new_se
) {
819 if (strcmp(se
->idstr
, id
) == 0 && se
->opaque
== opaque
) {
820 savevm_state_handler_remove(se
);
828 * Perform some basic checks on vmsd's at registration
831 static void vmstate_check(const VMStateDescription
*vmsd
)
833 const VMStateField
*field
= vmsd
->fields
;
834 const VMStateDescription
* const *subsection
= vmsd
->subsections
;
837 while (field
->name
) {
838 if (field
->flags
& (VMS_STRUCT
| VMS_VSTRUCT
)) {
839 /* Recurse to sub structures */
840 vmstate_check(field
->vmsd
);
845 /* Check for the end of field list canary */
846 if (field
->flags
!= VMS_END
) {
847 error_report("VMSTATE not ending with VMS_END: %s", vmsd
->name
);
848 g_assert_not_reached();
852 while (subsection
&& *subsection
) {
854 * The name of a subsection should start with the name of the
857 assert(!strncmp(vmsd
->name
, (*subsection
)->name
, strlen(vmsd
->name
)));
858 vmstate_check(*subsection
);
864 * See comment in hw/intc/xics.c:icp_realize()
866 * This function can be removed when
867 * pre_2_10_vmstate_register_dummy_icp() is removed.
869 int vmstate_replace_hack_for_ppc(VMStateIf
*obj
, int instance_id
,
870 const VMStateDescription
*vmsd
,
873 SaveStateEntry
*se
= find_se(vmsd
->name
, instance_id
);
876 savevm_state_handler_remove(se
);
878 return vmstate_register(obj
, instance_id
, vmsd
, opaque
);
881 int vmstate_register_with_alias_id(VMStateIf
*obj
, uint32_t instance_id
,
882 const VMStateDescription
*vmsd
,
883 void *opaque
, int alias_id
,
884 int required_for_version
,
889 /* If this triggers, alias support can be dropped for the vmsd. */
890 assert(alias_id
== -1 || required_for_version
>= vmsd
->minimum_version_id
);
892 se
= g_new0(SaveStateEntry
, 1);
893 se
->version_id
= vmsd
->version_id
;
894 se
->section_id
= savevm_state
.global_section_id
++;
897 se
->alias_id
= alias_id
;
900 char *id
= vmstate_if_get_id(obj
);
902 if (snprintf(se
->idstr
, sizeof(se
->idstr
), "%s/", id
) >=
904 error_setg(errp
, "Path too long for VMState (%s)", id
);
912 se
->compat
= g_new0(CompatEntry
, 1);
913 pstrcpy(se
->compat
->idstr
, sizeof(se
->compat
->idstr
), vmsd
->name
);
914 se
->compat
->instance_id
= instance_id
== VMSTATE_INSTANCE_ID_ANY
?
915 calculate_compat_instance_id(vmsd
->name
) : instance_id
;
916 instance_id
= VMSTATE_INSTANCE_ID_ANY
;
919 pstrcat(se
->idstr
, sizeof(se
->idstr
), vmsd
->name
);
921 if (instance_id
== VMSTATE_INSTANCE_ID_ANY
) {
922 se
->instance_id
= calculate_new_instance_id(se
->idstr
);
924 se
->instance_id
= instance_id
;
927 /* Perform a recursive sanity check during the test runs */
928 if (qtest_enabled()) {
931 assert(!se
->compat
|| se
->instance_id
== 0);
932 savevm_state_handler_insert(se
);
936 void vmstate_unregister(VMStateIf
*obj
, const VMStateDescription
*vmsd
,
939 SaveStateEntry
*se
, *new_se
;
941 QTAILQ_FOREACH_SAFE(se
, &savevm_state
.handlers
, entry
, new_se
) {
942 if (se
->vmsd
== vmsd
&& se
->opaque
== opaque
) {
943 savevm_state_handler_remove(se
);
950 static int vmstate_load(QEMUFile
*f
, SaveStateEntry
*se
)
952 trace_vmstate_load(se
->idstr
, se
->vmsd
? se
->vmsd
->name
: "(old)");
953 if (!se
->vmsd
) { /* Old style */
954 return se
->ops
->load_state(f
, se
->opaque
, se
->load_version_id
);
956 return vmstate_load_state(f
, se
->vmsd
, se
->opaque
, se
->load_version_id
);
959 static void vmstate_save_old_style(QEMUFile
*f
, SaveStateEntry
*se
,
962 uint64_t old_offset
= qemu_file_transferred(f
);
963 se
->ops
->save_state(f
, se
->opaque
);
964 uint64_t size
= qemu_file_transferred(f
) - old_offset
;
967 json_writer_int64(vmdesc
, "size", size
);
968 json_writer_start_array(vmdesc
, "fields");
969 json_writer_start_object(vmdesc
, NULL
);
970 json_writer_str(vmdesc
, "name", "data");
971 json_writer_int64(vmdesc
, "size", size
);
972 json_writer_str(vmdesc
, "type", "buffer");
973 json_writer_end_object(vmdesc
);
974 json_writer_end_array(vmdesc
);
979 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
981 static void save_section_header(QEMUFile
*f
, SaveStateEntry
*se
,
982 uint8_t section_type
)
984 qemu_put_byte(f
, section_type
);
985 qemu_put_be32(f
, se
->section_id
);
987 if (section_type
== QEMU_VM_SECTION_FULL
||
988 section_type
== QEMU_VM_SECTION_START
) {
990 size_t len
= strlen(se
->idstr
);
991 qemu_put_byte(f
, len
);
992 qemu_put_buffer(f
, (uint8_t *)se
->idstr
, len
);
994 qemu_put_be32(f
, se
->instance_id
);
995 qemu_put_be32(f
, se
->version_id
);
1000 * Write a footer onto device sections that catches cases misformatted device
1003 static void save_section_footer(QEMUFile
*f
, SaveStateEntry
*se
)
1005 if (migrate_get_current()->send_section_footer
) {
1006 qemu_put_byte(f
, QEMU_VM_SECTION_FOOTER
);
1007 qemu_put_be32(f
, se
->section_id
);
1011 static int vmstate_save(QEMUFile
*f
, SaveStateEntry
*se
, JSONWriter
*vmdesc
,
1016 if ((!se
->ops
|| !se
->ops
->save_state
) && !se
->vmsd
) {
1019 if (se
->vmsd
&& !vmstate_section_needed(se
->vmsd
, se
->opaque
)) {
1020 trace_savevm_section_skip(se
->idstr
, se
->section_id
);
1024 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1025 save_section_header(f
, se
, QEMU_VM_SECTION_FULL
);
1027 json_writer_start_object(vmdesc
, NULL
);
1028 json_writer_str(vmdesc
, "name", se
->idstr
);
1029 json_writer_int64(vmdesc
, "instance_id", se
->instance_id
);
1032 trace_vmstate_save(se
->idstr
, se
->vmsd
? se
->vmsd
->name
: "(old)");
1034 vmstate_save_old_style(f
, se
, vmdesc
);
1036 ret
= vmstate_save_state_with_err(f
, se
->vmsd
, se
->opaque
, vmdesc
,
1043 trace_savevm_section_end(se
->idstr
, se
->section_id
, 0);
1044 save_section_footer(f
, se
);
1046 json_writer_end_object(vmdesc
);
1051 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
1052 * command and associated data.
1054 * @f: File to send command on
1055 * @command: Command type to send
1056 * @len: Length of associated data
1057 * @data: Data associated with command.
1059 static void qemu_savevm_command_send(QEMUFile
*f
,
1060 enum qemu_vm_cmd command
,
1064 trace_savevm_command_send(command
, len
);
1065 qemu_put_byte(f
, QEMU_VM_COMMAND
);
1066 qemu_put_be16(f
, (uint16_t)command
);
1067 qemu_put_be16(f
, len
);
1068 qemu_put_buffer(f
, data
, len
);
1072 void qemu_savevm_send_colo_enable(QEMUFile
*f
)
1074 trace_savevm_send_colo_enable();
1075 qemu_savevm_command_send(f
, MIG_CMD_ENABLE_COLO
, 0, NULL
);
1078 void qemu_savevm_send_ping(QEMUFile
*f
, uint32_t value
)
1082 trace_savevm_send_ping(value
);
1083 buf
= cpu_to_be32(value
);
1084 qemu_savevm_command_send(f
, MIG_CMD_PING
, sizeof(value
), (uint8_t *)&buf
);
1087 void qemu_savevm_send_open_return_path(QEMUFile
*f
)
1089 trace_savevm_send_open_return_path();
1090 qemu_savevm_command_send(f
, MIG_CMD_OPEN_RETURN_PATH
, 0, NULL
);
1093 /* We have a buffer of data to send; we don't want that all to be loaded
1094 * by the command itself, so the command contains just the length of the
1095 * extra buffer that we then send straight after it.
1096 * TODO: Must be a better way to organise that
1102 int qemu_savevm_send_packaged(QEMUFile
*f
, const uint8_t *buf
, size_t len
)
1105 MigrationState
*ms
= migrate_get_current();
1106 Error
*local_err
= NULL
;
1108 if (len
> MAX_VM_CMD_PACKAGED_SIZE
) {
1109 error_setg(&local_err
, "%s: Unreasonably large packaged state: %zu",
1111 migrate_set_error(ms
, local_err
);
1112 error_report_err(local_err
);
1116 tmp
= cpu_to_be32(len
);
1118 trace_qemu_savevm_send_packaged();
1119 qemu_savevm_command_send(f
, MIG_CMD_PACKAGED
, 4, (uint8_t *)&tmp
);
1121 qemu_put_buffer(f
, buf
, len
);
1126 /* Send prior to any postcopy transfer */
1127 void qemu_savevm_send_postcopy_advise(QEMUFile
*f
)
1129 if (migrate_postcopy_ram()) {
1131 tmp
[0] = cpu_to_be64(ram_pagesize_summary());
1132 tmp
[1] = cpu_to_be64(qemu_target_page_size());
1134 trace_qemu_savevm_send_postcopy_advise();
1135 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_ADVISE
,
1136 16, (uint8_t *)tmp
);
1138 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_ADVISE
, 0, NULL
);
1142 /* Sent prior to starting the destination running in postcopy, discard pages
1143 * that have already been sent but redirtied on the source.
1144 * CMD_POSTCOPY_RAM_DISCARD consist of:
1146 * byte Length of name field (not including 0)
1147 * n x byte RAM block name
1148 * byte 0 terminator (just for safety)
1149 * n x Byte ranges within the named RAMBlock
1150 * be64 Start of the range
1153 * name: RAMBlock name that these entries are part of
1154 * len: Number of page entries
1155 * start_list: 'len' addresses
1156 * length_list: 'len' addresses
1159 void qemu_savevm_send_postcopy_ram_discard(QEMUFile
*f
, const char *name
,
1161 uint64_t *start_list
,
1162 uint64_t *length_list
)
1167 size_t name_len
= strlen(name
);
1169 trace_qemu_savevm_send_postcopy_ram_discard(name
, len
);
1170 assert(name_len
< 256);
1171 buf
= g_malloc0(1 + 1 + name_len
+ 1 + (8 + 8) * len
);
1172 buf
[0] = postcopy_ram_discard_version
;
1174 memcpy(buf
+ 2, name
, name_len
);
1175 tmplen
= 2 + name_len
;
1176 buf
[tmplen
++] = '\0';
1178 for (t
= 0; t
< len
; t
++) {
1179 stq_be_p(buf
+ tmplen
, start_list
[t
]);
1181 stq_be_p(buf
+ tmplen
, length_list
[t
]);
1184 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RAM_DISCARD
, tmplen
, buf
);
1188 /* Get the destination into a state where it can receive postcopy data. */
1189 void qemu_savevm_send_postcopy_listen(QEMUFile
*f
)
1191 trace_savevm_send_postcopy_listen();
1192 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_LISTEN
, 0, NULL
);
1195 /* Kick the destination into running */
1196 void qemu_savevm_send_postcopy_run(QEMUFile
*f
)
1198 trace_savevm_send_postcopy_run();
1199 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RUN
, 0, NULL
);
1202 void qemu_savevm_send_postcopy_resume(QEMUFile
*f
)
1204 trace_savevm_send_postcopy_resume();
1205 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RESUME
, 0, NULL
);
1208 void qemu_savevm_send_recv_bitmap(QEMUFile
*f
, char *block_name
)
1213 trace_savevm_send_recv_bitmap(block_name
);
1215 buf
[0] = len
= strlen(block_name
);
1216 memcpy(buf
+ 1, block_name
, len
);
1218 qemu_savevm_command_send(f
, MIG_CMD_RECV_BITMAP
, len
+ 1, (uint8_t *)buf
);
1221 bool qemu_savevm_state_blocked(Error
**errp
)
1225 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1226 if (se
->vmsd
&& se
->vmsd
->unmigratable
) {
1227 error_setg(errp
, "State blocked by non-migratable device '%s'",
1235 void qemu_savevm_non_migratable_list(strList
**reasons
)
1239 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1240 if (se
->vmsd
&& se
->vmsd
->unmigratable
) {
1241 QAPI_LIST_PREPEND(*reasons
,
1242 g_strdup_printf("non-migratable device: %s",
1248 void qemu_savevm_state_header(QEMUFile
*f
)
1250 MigrationState
*s
= migrate_get_current();
1252 s
->vmdesc
= json_writer_new(false);
1254 trace_savevm_state_header();
1255 qemu_put_be32(f
, QEMU_VM_FILE_MAGIC
);
1256 qemu_put_be32(f
, QEMU_VM_FILE_VERSION
);
1258 if (s
->send_configuration
) {
1259 qemu_put_byte(f
, QEMU_VM_CONFIGURATION
);
1262 * This starts the main json object and is paired with the
1263 * json_writer_end_object in
1264 * qemu_savevm_state_complete_precopy_non_iterable
1266 json_writer_start_object(s
->vmdesc
, NULL
);
1268 json_writer_start_object(s
->vmdesc
, "configuration");
1269 vmstate_save_state(f
, &vmstate_configuration
, &savevm_state
, s
->vmdesc
);
1270 json_writer_end_object(s
->vmdesc
);
1274 bool qemu_savevm_state_guest_unplug_pending(void)
1278 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1279 if (se
->vmsd
&& se
->vmsd
->dev_unplug_pending
&&
1280 se
->vmsd
->dev_unplug_pending(se
->opaque
)) {
1288 int qemu_savevm_state_prepare(Error
**errp
)
1293 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1294 if (!se
->ops
|| !se
->ops
->save_prepare
) {
1297 if (se
->ops
->is_active
) {
1298 if (!se
->ops
->is_active(se
->opaque
)) {
1303 ret
= se
->ops
->save_prepare(se
->opaque
, errp
);
1312 int qemu_savevm_state_setup(QEMUFile
*f
, Error
**errp
)
1315 MigrationState
*ms
= migrate_get_current();
1319 json_writer_int64(ms
->vmdesc
, "page_size", qemu_target_page_size());
1320 json_writer_start_array(ms
->vmdesc
, "devices");
1322 trace_savevm_state_setup();
1323 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1324 if (se
->vmsd
&& se
->vmsd
->early_setup
) {
1325 ret
= vmstate_save(f
, se
, ms
->vmdesc
, errp
);
1327 migrate_set_error(ms
, *errp
);
1328 qemu_file_set_error(f
, ret
);
1334 if (!se
->ops
|| !se
->ops
->save_setup
) {
1337 if (se
->ops
->is_active
) {
1338 if (!se
->ops
->is_active(se
->opaque
)) {
1342 save_section_header(f
, se
, QEMU_VM_SECTION_START
);
1344 ret
= se
->ops
->save_setup(f
, se
->opaque
, errp
);
1345 save_section_footer(f
, se
);
1347 qemu_file_set_error(f
, ret
);
1356 /* TODO: Should we check that errp is set in case of failure ? */
1357 return precopy_notify(PRECOPY_NOTIFY_SETUP
, errp
);
1360 int qemu_savevm_state_resume_prepare(MigrationState
*s
)
1365 trace_savevm_state_resume_prepare();
1367 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1368 if (!se
->ops
|| !se
->ops
->resume_prepare
) {
1371 if (se
->ops
->is_active
) {
1372 if (!se
->ops
->is_active(se
->opaque
)) {
1376 ret
= se
->ops
->resume_prepare(s
, se
->opaque
);
1386 * this function has three return values:
1387 * negative: there was one error, and we have -errno.
1388 * 0 : We haven't finished, caller have to go again
1389 * 1 : We have finished, we can go to complete phase
1391 int qemu_savevm_state_iterate(QEMUFile
*f
, bool postcopy
)
1394 bool all_finished
= true;
1397 trace_savevm_state_iterate();
1398 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1399 if (!se
->ops
|| !se
->ops
->save_live_iterate
) {
1402 if (se
->ops
->is_active
&&
1403 !se
->ops
->is_active(se
->opaque
)) {
1406 if (se
->ops
->is_active_iterate
&&
1407 !se
->ops
->is_active_iterate(se
->opaque
)) {
1411 * In the postcopy phase, any device that doesn't know how to
1412 * do postcopy should have saved it's state in the _complete
1413 * call that's already run, it might get confused if we call
1414 * iterate afterwards.
1417 !(se
->ops
->has_postcopy
&& se
->ops
->has_postcopy(se
->opaque
))) {
1420 if (migration_rate_exceeded(f
)) {
1423 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1425 save_section_header(f
, se
, QEMU_VM_SECTION_PART
);
1427 ret
= se
->ops
->save_live_iterate(f
, se
->opaque
);
1428 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1429 save_section_footer(f
, se
);
1432 error_report("failed to save SaveStateEntry with id(name): "
1434 se
->section_id
, se
->idstr
, ret
);
1435 qemu_file_set_error(f
, ret
);
1438 all_finished
= false;
1441 return all_finished
;
1444 static bool should_send_vmdesc(void)
1446 MachineState
*machine
= MACHINE(qdev_get_machine());
1447 bool in_postcopy
= migration_in_postcopy();
1448 return !machine
->suppress_vmdesc
&& !in_postcopy
;
1452 * Calls the save_live_complete_postcopy methods
1453 * causing the last few pages to be sent immediately and doing any associated
1455 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1456 * all the other devices, but that happens at the point we switch to postcopy.
1458 void qemu_savevm_state_complete_postcopy(QEMUFile
*f
)
1463 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1464 if (!se
->ops
|| !se
->ops
->save_live_complete_postcopy
) {
1467 if (se
->ops
->is_active
) {
1468 if (!se
->ops
->is_active(se
->opaque
)) {
1472 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1474 qemu_put_byte(f
, QEMU_VM_SECTION_END
);
1475 qemu_put_be32(f
, se
->section_id
);
1477 ret
= se
->ops
->save_live_complete_postcopy(f
, se
->opaque
);
1478 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1479 save_section_footer(f
, se
);
1481 qemu_file_set_error(f
, ret
);
1486 qemu_put_byte(f
, QEMU_VM_EOF
);
1491 int qemu_savevm_state_complete_precopy_iterable(QEMUFile
*f
, bool in_postcopy
)
1493 int64_t start_ts_each
, end_ts_each
;
1497 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1499 (in_postcopy
&& se
->ops
->has_postcopy
&&
1500 se
->ops
->has_postcopy(se
->opaque
)) ||
1501 !se
->ops
->save_live_complete_precopy
) {
1505 if (se
->ops
->is_active
) {
1506 if (!se
->ops
->is_active(se
->opaque
)) {
1511 start_ts_each
= qemu_clock_get_us(QEMU_CLOCK_REALTIME
);
1512 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1514 save_section_header(f
, se
, QEMU_VM_SECTION_END
);
1516 ret
= se
->ops
->save_live_complete_precopy(f
, se
->opaque
);
1517 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1518 save_section_footer(f
, se
);
1520 qemu_file_set_error(f
, ret
);
1523 end_ts_each
= qemu_clock_get_us(QEMU_CLOCK_REALTIME
);
1524 trace_vmstate_downtime_save("iterable", se
->idstr
, se
->instance_id
,
1525 end_ts_each
- start_ts_each
);
1528 trace_vmstate_downtime_checkpoint("src-iterable-saved");
1533 int qemu_savevm_state_complete_precopy_non_iterable(QEMUFile
*f
,
1535 bool inactivate_disks
)
1537 MigrationState
*ms
= migrate_get_current();
1538 int64_t start_ts_each
, end_ts_each
;
1539 JSONWriter
*vmdesc
= ms
->vmdesc
;
1542 Error
*local_err
= NULL
;
1545 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1546 if (se
->vmsd
&& se
->vmsd
->early_setup
) {
1547 /* Already saved during qemu_savevm_state_setup(). */
1551 start_ts_each
= qemu_clock_get_us(QEMU_CLOCK_REALTIME
);
1553 ret
= vmstate_save(f
, se
, vmdesc
, &local_err
);
1555 migrate_set_error(ms
, local_err
);
1556 error_report_err(local_err
);
1557 qemu_file_set_error(f
, ret
);
1561 end_ts_each
= qemu_clock_get_us(QEMU_CLOCK_REALTIME
);
1562 trace_vmstate_downtime_save("non-iterable", se
->idstr
, se
->instance_id
,
1563 end_ts_each
- start_ts_each
);
1566 if (inactivate_disks
) {
1567 /* Inactivate before sending QEMU_VM_EOF so that the
1568 * bdrv_activate_all() on the other end won't fail. */
1569 ret
= bdrv_inactivate_all();
1571 error_setg(&local_err
, "%s: bdrv_inactivate_all() failed (%d)",
1573 migrate_set_error(ms
, local_err
);
1574 error_report_err(local_err
);
1575 qemu_file_set_error(f
, ret
);
1580 /* Postcopy stream will still be going */
1581 qemu_put_byte(f
, QEMU_VM_EOF
);
1584 json_writer_end_array(vmdesc
);
1585 json_writer_end_object(vmdesc
);
1586 vmdesc_len
= strlen(json_writer_get(vmdesc
));
1588 if (should_send_vmdesc()) {
1589 qemu_put_byte(f
, QEMU_VM_VMDESCRIPTION
);
1590 qemu_put_be32(f
, vmdesc_len
);
1591 qemu_put_buffer(f
, (uint8_t *)json_writer_get(vmdesc
), vmdesc_len
);
1594 /* Free it now to detect any inconsistencies. */
1595 json_writer_free(vmdesc
);
1598 trace_vmstate_downtime_checkpoint("src-non-iterable-saved");
1603 int qemu_savevm_state_complete_precopy(QEMUFile
*f
, bool iterable_only
,
1604 bool inactivate_disks
)
1607 Error
*local_err
= NULL
;
1608 bool in_postcopy
= migration_in_postcopy();
1610 if (precopy_notify(PRECOPY_NOTIFY_COMPLETE
, &local_err
)) {
1611 error_report_err(local_err
);
1614 trace_savevm_state_complete_precopy();
1616 cpu_synchronize_all_states();
1618 if (!in_postcopy
|| iterable_only
) {
1619 ret
= qemu_savevm_state_complete_precopy_iterable(f
, in_postcopy
);
1625 if (iterable_only
) {
1629 ret
= qemu_savevm_state_complete_precopy_non_iterable(f
, in_postcopy
,
1636 return qemu_fflush(f
);
1639 /* Give an estimate of the amount left to be transferred,
1640 * the result is split into the amount for units that can and
1641 * for units that can't do postcopy.
1643 void qemu_savevm_state_pending_estimate(uint64_t *must_precopy
,
1644 uint64_t *can_postcopy
)
1651 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1652 if (!se
->ops
|| !se
->ops
->state_pending_estimate
) {
1655 if (se
->ops
->is_active
) {
1656 if (!se
->ops
->is_active(se
->opaque
)) {
1660 se
->ops
->state_pending_estimate(se
->opaque
, must_precopy
, can_postcopy
);
1664 void qemu_savevm_state_pending_exact(uint64_t *must_precopy
,
1665 uint64_t *can_postcopy
)
1672 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1673 if (!se
->ops
|| !se
->ops
->state_pending_exact
) {
1676 if (se
->ops
->is_active
) {
1677 if (!se
->ops
->is_active(se
->opaque
)) {
1681 se
->ops
->state_pending_exact(se
->opaque
, must_precopy
, can_postcopy
);
1685 void qemu_savevm_state_cleanup(void)
1688 Error
*local_err
= NULL
;
1690 if (precopy_notify(PRECOPY_NOTIFY_CLEANUP
, &local_err
)) {
1691 error_report_err(local_err
);
1694 trace_savevm_state_cleanup();
1695 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1696 if (se
->ops
&& se
->ops
->save_cleanup
) {
1697 se
->ops
->save_cleanup(se
->opaque
);
1702 static int qemu_savevm_state(QEMUFile
*f
, Error
**errp
)
1705 MigrationState
*ms
= migrate_get_current();
1706 MigrationStatus status
;
1708 if (migration_is_running()) {
1709 error_setg(errp
, "There's a migration process in progress");
1713 ret
= migrate_init(ms
, errp
);
1717 ms
->to_dst_file
= f
;
1719 qemu_savevm_state_header(f
);
1720 ret
= qemu_savevm_state_setup(f
, errp
);
1725 while (qemu_file_get_error(f
) == 0) {
1726 if (qemu_savevm_state_iterate(f
, false) > 0) {
1731 ret
= qemu_file_get_error(f
);
1733 qemu_savevm_state_complete_precopy(f
, false, false);
1734 ret
= qemu_file_get_error(f
);
1737 error_setg_errno(errp
, -ret
, "Error while writing VM state");
1740 qemu_savevm_state_cleanup();
1743 status
= MIGRATION_STATUS_FAILED
;
1745 status
= MIGRATION_STATUS_COMPLETED
;
1747 migrate_set_state(&ms
->state
, MIGRATION_STATUS_SETUP
, status
);
1749 /* f is outer parameter, it should not stay in global migration state after
1750 * this function finished */
1751 ms
->to_dst_file
= NULL
;
1756 void qemu_savevm_live_state(QEMUFile
*f
)
1758 /* save QEMU_VM_SECTION_END section */
1759 qemu_savevm_state_complete_precopy(f
, true, false);
1760 qemu_put_byte(f
, QEMU_VM_EOF
);
1763 int qemu_save_device_state(QEMUFile
*f
)
1765 MigrationState
*ms
= migrate_get_current();
1766 Error
*local_err
= NULL
;
1769 if (!migration_in_colo_state()) {
1770 qemu_put_be32(f
, QEMU_VM_FILE_MAGIC
);
1771 qemu_put_be32(f
, QEMU_VM_FILE_VERSION
);
1773 cpu_synchronize_all_states();
1775 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1781 ret
= vmstate_save(f
, se
, NULL
, &local_err
);
1783 migrate_set_error(ms
, local_err
);
1784 error_report_err(local_err
);
1789 qemu_put_byte(f
, QEMU_VM_EOF
);
1791 return qemu_file_get_error(f
);
1794 static SaveStateEntry
*find_se(const char *idstr
, uint32_t instance_id
)
1798 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1799 if (!strcmp(se
->idstr
, idstr
) &&
1800 (instance_id
== se
->instance_id
||
1801 instance_id
== se
->alias_id
))
1803 /* Migrating from an older version? */
1804 if (strstr(se
->idstr
, idstr
) && se
->compat
) {
1805 if (!strcmp(se
->compat
->idstr
, idstr
) &&
1806 (instance_id
== se
->compat
->instance_id
||
1807 instance_id
== se
->alias_id
))
1814 enum LoadVMExitCodes
{
1815 /* Allow a command to quit all layers of nested loadvm loops */
1819 /* ------ incoming postcopy messages ------ */
1820 /* 'advise' arrives before any transfers just to tell us that a postcopy
1821 * *might* happen - it might be skipped if precopy transferred everything
1824 static int loadvm_postcopy_handle_advise(MigrationIncomingState
*mis
,
1827 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_ADVISE
);
1828 uint64_t remote_pagesize_summary
, local_pagesize_summary
, remote_tps
;
1829 size_t page_size
= qemu_target_page_size();
1830 Error
*local_err
= NULL
;
1832 trace_loadvm_postcopy_handle_advise();
1833 if (ps
!= POSTCOPY_INCOMING_NONE
) {
1834 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps
);
1840 if (migrate_postcopy_ram()) {
1841 error_report("RAM postcopy is enabled but have 0 byte advise");
1846 if (!migrate_postcopy_ram()) {
1847 error_report("RAM postcopy is disabled but have 16 byte advise");
1852 error_report("CMD_POSTCOPY_ADVISE invalid length (%d)", len
);
1856 if (!postcopy_ram_supported_by_host(mis
, &local_err
)) {
1857 error_report_err(local_err
);
1858 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
1862 remote_pagesize_summary
= qemu_get_be64(mis
->from_src_file
);
1863 local_pagesize_summary
= ram_pagesize_summary();
1865 if (remote_pagesize_summary
!= local_pagesize_summary
) {
1867 * This detects two potential causes of mismatch:
1868 * a) A mismatch in host page sizes
1869 * Some combinations of mismatch are probably possible but it gets
1870 * a bit more complicated. In particular we need to place whole
1871 * host pages on the dest at once, and we need to ensure that we
1872 * handle dirtying to make sure we never end up sending part of
1873 * a hostpage on it's own.
1874 * b) The use of different huge page sizes on source/destination
1875 * a more fine grain test is performed during RAM block migration
1876 * but this test here causes a nice early clear failure, and
1877 * also fails when passed to an older qemu that doesn't
1880 error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1882 remote_pagesize_summary
, local_pagesize_summary
);
1886 remote_tps
= qemu_get_be64(mis
->from_src_file
);
1887 if (remote_tps
!= page_size
) {
1889 * Again, some differences could be dealt with, but for now keep it
1892 error_report("Postcopy needs matching target page sizes (s=%d d=%zd)",
1893 (int)remote_tps
, page_size
);
1897 if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_ADVISE
, &local_err
)) {
1898 error_report_err(local_err
);
1902 if (ram_postcopy_incoming_init(mis
)) {
1909 /* After postcopy we will be told to throw some pages away since they're
1910 * dirty and will have to be demand fetched. Must happen before CPU is
1912 * There can be 0..many of these messages, each encoding multiple pages.
1914 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState
*mis
,
1919 PostcopyState ps
= postcopy_state_get();
1921 trace_loadvm_postcopy_ram_handle_discard();
1924 case POSTCOPY_INCOMING_ADVISE
:
1926 tmp
= postcopy_ram_prepare_discard(mis
);
1932 case POSTCOPY_INCOMING_DISCARD
:
1933 /* Expected state */
1937 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1941 /* We're expecting a
1943 * a RAM ID string (length byte, name, 0 term)
1944 * then at least 1 16 byte chunk
1946 if (len
< (1 + 1 + 1 + 1 + 2 * 8)) {
1947 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len
);
1951 tmp
= qemu_get_byte(mis
->from_src_file
);
1952 if (tmp
!= postcopy_ram_discard_version
) {
1953 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp
);
1957 if (!qemu_get_counted_string(mis
->from_src_file
, ramid
)) {
1958 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1961 tmp
= qemu_get_byte(mis
->from_src_file
);
1963 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp
);
1967 len
-= 3 + strlen(ramid
);
1969 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len
);
1972 trace_loadvm_postcopy_ram_handle_discard_header(ramid
, len
);
1974 uint64_t start_addr
, block_length
;
1975 start_addr
= qemu_get_be64(mis
->from_src_file
);
1976 block_length
= qemu_get_be64(mis
->from_src_file
);
1979 int ret
= ram_discard_range(ramid
, start_addr
, block_length
);
1984 trace_loadvm_postcopy_ram_handle_discard_end();
1990 * Triggered by a postcopy_listen command; this thread takes over reading
1991 * the input stream, leaving the main thread free to carry on loading the rest
1992 * of the device state (from RAM).
1993 * (TODO:This could do with being in a postcopy file - but there again it's
1994 * just another input loop, not that postcopy specific)
1996 static void *postcopy_ram_listen_thread(void *opaque
)
1998 MigrationIncomingState
*mis
= migration_incoming_get_current();
1999 QEMUFile
*f
= mis
->from_src_file
;
2001 MigrationState
*migr
= migrate_get_current();
2003 object_ref(OBJECT(migr
));
2005 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
2006 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2007 qemu_sem_post(&mis
->thread_sync_sem
);
2008 trace_postcopy_ram_listen_thread_start();
2010 rcu_register_thread();
2012 * Because we're a thread and not a coroutine we can't yield
2013 * in qemu_file, and thus we must be blocking now.
2015 qemu_file_set_blocking(f
, true);
2016 load_res
= qemu_loadvm_state_main(f
, mis
);
2019 * This is tricky, but, mis->from_src_file can change after it
2020 * returns, when postcopy recovery happened. In the future, we may
2021 * want a wrapper for the QEMUFile handle.
2023 f
= mis
->from_src_file
;
2025 /* And non-blocking again so we don't block in any cleanup */
2026 qemu_file_set_blocking(f
, false);
2028 trace_postcopy_ram_listen_thread_exit();
2030 qemu_file_set_error(f
, load_res
);
2031 dirty_bitmap_mig_cancel_incoming();
2032 if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING
&&
2033 !migrate_postcopy_ram() && migrate_dirty_bitmaps())
2035 error_report("%s: loadvm failed during postcopy: %d. All states "
2036 "are migrated except dirty bitmaps. Some dirty "
2037 "bitmaps may be lost, and present migrated dirty "
2038 "bitmaps are correctly migrated and valid.",
2039 __func__
, load_res
);
2040 load_res
= 0; /* prevent further exit() */
2042 error_report("%s: loadvm failed: %d", __func__
, load_res
);
2043 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2044 MIGRATION_STATUS_FAILED
);
2047 if (load_res
>= 0) {
2049 * This looks good, but it's possible that the device loading in the
2050 * main thread hasn't finished yet, and so we might not be in 'RUN'
2051 * state yet; wait for the end of the main thread.
2053 qemu_event_wait(&mis
->main_thread_load_event
);
2055 postcopy_ram_incoming_cleanup(mis
);
2059 * If something went wrong then we have a bad state so exit;
2060 * depending how far we got it might be possible at this point
2061 * to leave the guest running and fire MCEs for pages that never
2062 * arrived as a desperate recovery step.
2064 rcu_unregister_thread();
2068 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2069 MIGRATION_STATUS_COMPLETED
);
2071 * If everything has worked fine, then the main thread has waited
2072 * for us to start, and we're the last use of the mis.
2073 * (If something broke then qemu will have to exit anyway since it's
2074 * got a bad migration state).
2076 migration_incoming_state_destroy();
2077 qemu_loadvm_state_cleanup();
2079 rcu_unregister_thread();
2080 mis
->have_listen_thread
= false;
2081 postcopy_state_set(POSTCOPY_INCOMING_END
);
2083 object_unref(OBJECT(migr
));
2088 /* After this message we must be able to immediately receive postcopy data */
2089 static int loadvm_postcopy_handle_listen(MigrationIncomingState
*mis
)
2091 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_LISTENING
);
2092 Error
*local_err
= NULL
;
2094 trace_loadvm_postcopy_handle_listen("enter");
2096 if (ps
!= POSTCOPY_INCOMING_ADVISE
&& ps
!= POSTCOPY_INCOMING_DISCARD
) {
2097 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps
);
2100 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
2102 * A rare case, we entered listen without having to do any discards,
2103 * so do the setup that's normally done at the time of the 1st discard.
2105 if (migrate_postcopy_ram()) {
2106 postcopy_ram_prepare_discard(mis
);
2110 trace_loadvm_postcopy_handle_listen("after discard");
2113 * Sensitise RAM - can now generate requests for blocks that don't exist
2114 * However, at this point the CPU shouldn't be running, and the IO
2115 * shouldn't be doing anything yet so don't actually expect requests
2117 if (migrate_postcopy_ram()) {
2118 if (postcopy_ram_incoming_setup(mis
)) {
2119 postcopy_ram_incoming_cleanup(mis
);
2124 trace_loadvm_postcopy_handle_listen("after uffd");
2126 if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_LISTEN
, &local_err
)) {
2127 error_report_err(local_err
);
2131 mis
->have_listen_thread
= true;
2132 postcopy_thread_create(mis
, &mis
->listen_thread
, "postcopy/listen",
2133 postcopy_ram_listen_thread
, QEMU_THREAD_DETACHED
);
2134 trace_loadvm_postcopy_handle_listen("return");
2139 static void loadvm_postcopy_handle_run_bh(void *opaque
)
2141 Error
*local_err
= NULL
;
2142 MigrationIncomingState
*mis
= opaque
;
2144 trace_vmstate_downtime_checkpoint("dst-postcopy-bh-enter");
2146 /* TODO we should move all of this lot into postcopy_ram.c or a shared code
2149 cpu_synchronize_all_post_init();
2151 trace_vmstate_downtime_checkpoint("dst-postcopy-bh-cpu-synced");
2153 qemu_announce_self(&mis
->announce_timer
, migrate_announce_params());
2155 trace_vmstate_downtime_checkpoint("dst-postcopy-bh-announced");
2157 /* Make sure all file formats throw away their mutable metadata.
2158 * If we get an error here, just don't restart the VM yet. */
2159 bdrv_activate_all(&local_err
);
2161 error_report_err(local_err
);
2166 trace_vmstate_downtime_checkpoint("dst-postcopy-bh-cache-invalidated");
2168 dirty_bitmap_mig_before_vm_start();
2171 /* Hold onto your hats, starting the CPU */
2174 /* leave it paused and let management decide when to start the CPU */
2175 runstate_set(RUN_STATE_PAUSED
);
2178 trace_vmstate_downtime_checkpoint("dst-postcopy-bh-vm-started");
2181 /* After all discards we can start running and asking for pages */
2182 static int loadvm_postcopy_handle_run(MigrationIncomingState
*mis
)
2184 PostcopyState ps
= postcopy_state_get();
2186 trace_loadvm_postcopy_handle_run();
2187 if (ps
!= POSTCOPY_INCOMING_LISTENING
) {
2188 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps
);
2192 postcopy_state_set(POSTCOPY_INCOMING_RUNNING
);
2193 migration_bh_schedule(loadvm_postcopy_handle_run_bh
, mis
);
2195 /* We need to finish reading the stream from the package
2196 * and also stop reading anything more from the stream that loaded the
2197 * package (since it's now being read by the listener thread).
2198 * LOADVM_QUIT will quit all the layers of nested loadvm loops.
2203 /* We must be with page_request_mutex held */
2204 static gboolean
postcopy_sync_page_req(gpointer key
, gpointer value
,
2207 MigrationIncomingState
*mis
= data
;
2208 void *host_addr
= (void *) key
;
2209 ram_addr_t rb_offset
;
2213 rb
= qemu_ram_block_from_host(host_addr
, true, &rb_offset
);
2216 * This should _never_ happen. However be nice for a migrating VM to
2217 * not crash/assert. Post an error (note: intended to not use *_once
2218 * because we do want to see all the illegal addresses; and this can
2219 * never be triggered by the guest so we're safe) and move on next.
2221 error_report("%s: illegal host addr %p", __func__
, host_addr
);
2222 /* Try the next entry */
2226 ret
= migrate_send_rp_message_req_pages(mis
, rb
, rb_offset
);
2228 /* Please refer to above comment. */
2229 error_report("%s: send rp message failed for addr %p",
2230 __func__
, host_addr
);
2234 trace_postcopy_page_req_sync(host_addr
);
2239 static void migrate_send_rp_req_pages_pending(MigrationIncomingState
*mis
)
2241 WITH_QEMU_LOCK_GUARD(&mis
->page_request_mutex
) {
2242 g_tree_foreach(mis
->page_requested
, postcopy_sync_page_req
, mis
);
2246 static int loadvm_postcopy_handle_resume(MigrationIncomingState
*mis
)
2248 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
2249 error_report("%s: illegal resume received", __func__
);
2250 /* Don't fail the load, only for this. */
2255 * Reset the last_rb before we resend any page req to source again, since
2256 * the source should have it reset already.
2258 mis
->last_rb
= NULL
;
2261 * This means source VM is ready to resume the postcopy migration.
2263 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
2264 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
2266 trace_loadvm_postcopy_handle_resume();
2268 /* Tell source that "we are ready" */
2269 migrate_send_rp_resume_ack(mis
, MIGRATION_RESUME_ACK_VALUE
);
2272 * After a postcopy recovery, the source should have lost the postcopy
2273 * queue, or potentially the requested pages could have been lost during
2274 * the network down phase. Let's re-sync with the source VM by re-sending
2275 * all the pending pages that we eagerly need, so these threads won't get
2276 * blocked too long due to the recovery.
2278 * Without this procedure, the faulted destination VM threads (waiting for
2279 * page requests right before the postcopy is interrupted) can keep hanging
2280 * until the pages are sent by the source during the background copying of
2281 * pages, or another thread faulted on the same address accidentally.
2283 migrate_send_rp_req_pages_pending(mis
);
2286 * It's time to switch state and release the fault thread to continue
2287 * service page faults. Note that this should be explicitly after the
2288 * above call to migrate_send_rp_req_pages_pending(). In short:
2289 * migrate_send_rp_message_req_pages() is not thread safe, yet.
2291 qemu_sem_post(&mis
->postcopy_pause_sem_fault
);
2293 if (migrate_postcopy_preempt()) {
2295 * The preempt channel will be created in async manner, now let's
2296 * wait for it and make sure it's created.
2298 qemu_sem_wait(&mis
->postcopy_qemufile_dst_done
);
2299 assert(mis
->postcopy_qemufile_dst
);
2300 /* Kick the fast ram load thread too */
2301 qemu_sem_post(&mis
->postcopy_pause_sem_fast_load
);
2308 * Immediately following this command is a blob of data containing an embedded
2309 * chunk of migration stream; read it and load it.
2311 * @mis: Incoming state
2312 * @length: Length of packaged data to read
2314 * Returns: Negative values on error
2317 static int loadvm_handle_cmd_packaged(MigrationIncomingState
*mis
)
2321 QIOChannelBuffer
*bioc
;
2323 length
= qemu_get_be32(mis
->from_src_file
);
2324 trace_loadvm_handle_cmd_packaged(length
);
2326 if (length
> MAX_VM_CMD_PACKAGED_SIZE
) {
2327 error_report("Unreasonably large packaged state: %zu", length
);
2331 bioc
= qio_channel_buffer_new(length
);
2332 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-loadvm-buffer");
2333 ret
= qemu_get_buffer(mis
->from_src_file
,
2336 if (ret
!= length
) {
2337 object_unref(OBJECT(bioc
));
2338 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
2340 return (ret
< 0) ? ret
: -EAGAIN
;
2342 bioc
->usage
+= length
;
2343 trace_loadvm_handle_cmd_packaged_received(ret
);
2345 QEMUFile
*packf
= qemu_file_new_input(QIO_CHANNEL(bioc
));
2348 * Before loading the guest states, ensure that the preempt channel has
2349 * been ready to use, as some of the states (e.g. via virtio_load) might
2350 * trigger page faults that will be handled through the preempt channel.
2351 * So yield to the main thread in the case that the channel create event
2352 * hasn't been dispatched.
2354 * TODO: if we can move migration loadvm out of main thread, then we
2355 * won't block main thread from polling the accept() fds. We can drop
2356 * this as a whole when that is done.
2359 if (!migrate_postcopy_preempt() || !qemu_in_coroutine() ||
2360 mis
->postcopy_qemufile_dst
) {
2364 aio_co_schedule(qemu_get_current_aio_context(), qemu_coroutine_self());
2365 qemu_coroutine_yield();
2368 ret
= qemu_loadvm_state_main(packf
, mis
);
2369 trace_loadvm_handle_cmd_packaged_main(ret
);
2371 object_unref(OBJECT(bioc
));
2377 * Handle request that source requests for recved_bitmap on
2378 * destination. Payload format:
2380 * len (1 byte) + ramblock_name (<255 bytes)
2382 static int loadvm_handle_recv_bitmap(MigrationIncomingState
*mis
,
2385 QEMUFile
*file
= mis
->from_src_file
;
2387 char block_name
[256];
2390 cnt
= qemu_get_counted_string(file
, block_name
);
2392 error_report("%s: failed to read block name", __func__
);
2396 /* Validate before using the data */
2397 if (qemu_file_get_error(file
)) {
2398 return qemu_file_get_error(file
);
2401 if (len
!= cnt
+ 1) {
2402 error_report("%s: invalid payload length (%d)", __func__
, len
);
2406 rb
= qemu_ram_block_by_name(block_name
);
2408 error_report("%s: block '%s' not found", __func__
, block_name
);
2412 migrate_send_rp_recv_bitmap(mis
, block_name
);
2414 trace_loadvm_handle_recv_bitmap(block_name
);
2419 static int loadvm_process_enable_colo(MigrationIncomingState
*mis
)
2421 int ret
= migration_incoming_enable_colo();
2424 ret
= colo_init_ram_cache();
2426 migration_incoming_disable_colo();
2433 * Process an incoming 'QEMU_VM_COMMAND'
2434 * 0 just a normal return
2435 * LOADVM_QUIT All good, but exit the loop
2438 static int loadvm_process_command(QEMUFile
*f
)
2440 MigrationIncomingState
*mis
= migration_incoming_get_current();
2445 cmd
= qemu_get_be16(f
);
2446 len
= qemu_get_be16(f
);
2448 /* Check validity before continue processing of cmds */
2449 if (qemu_file_get_error(f
)) {
2450 return qemu_file_get_error(f
);
2453 if (cmd
>= MIG_CMD_MAX
|| cmd
== MIG_CMD_INVALID
) {
2454 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd
, len
);
2458 trace_loadvm_process_command(mig_cmd_args
[cmd
].name
, len
);
2460 if (mig_cmd_args
[cmd
].len
!= -1 && mig_cmd_args
[cmd
].len
!= len
) {
2461 error_report("%s received with bad length - expecting %zu, got %d",
2462 mig_cmd_args
[cmd
].name
,
2463 (size_t)mig_cmd_args
[cmd
].len
, len
);
2468 case MIG_CMD_OPEN_RETURN_PATH
:
2469 if (mis
->to_src_file
) {
2470 error_report("CMD_OPEN_RETURN_PATH called when RP already open");
2471 /* Not really a problem, so don't give up */
2474 mis
->to_src_file
= qemu_file_get_return_path(f
);
2475 if (!mis
->to_src_file
) {
2476 error_report("CMD_OPEN_RETURN_PATH failed");
2481 * Switchover ack is enabled but no device uses it, so send an ACK to
2482 * source that it's OK to switchover. Do it here, after return path has
2485 if (migrate_switchover_ack() && !mis
->switchover_ack_pending_num
) {
2486 int ret
= migrate_send_rp_switchover_ack(mis
);
2489 "Could not send switchover ack RP MSG, err %d (%s)", ret
,
2497 tmp32
= qemu_get_be32(f
);
2498 trace_loadvm_process_command_ping(tmp32
);
2499 if (!mis
->to_src_file
) {
2500 error_report("CMD_PING (0x%x) received with no return path",
2504 migrate_send_rp_pong(mis
, tmp32
);
2507 case MIG_CMD_PACKAGED
:
2508 return loadvm_handle_cmd_packaged(mis
);
2510 case MIG_CMD_POSTCOPY_ADVISE
:
2511 return loadvm_postcopy_handle_advise(mis
, len
);
2513 case MIG_CMD_POSTCOPY_LISTEN
:
2514 return loadvm_postcopy_handle_listen(mis
);
2516 case MIG_CMD_POSTCOPY_RUN
:
2517 return loadvm_postcopy_handle_run(mis
);
2519 case MIG_CMD_POSTCOPY_RAM_DISCARD
:
2520 return loadvm_postcopy_ram_handle_discard(mis
, len
);
2522 case MIG_CMD_POSTCOPY_RESUME
:
2523 return loadvm_postcopy_handle_resume(mis
);
2525 case MIG_CMD_RECV_BITMAP
:
2526 return loadvm_handle_recv_bitmap(mis
, len
);
2528 case MIG_CMD_ENABLE_COLO
:
2529 return loadvm_process_enable_colo(mis
);
2536 * Read a footer off the wire and check that it matches the expected section
2538 * Returns: true if the footer was good
2539 * false if there is a problem (and calls error_report to say why)
2541 static bool check_section_footer(QEMUFile
*f
, SaveStateEntry
*se
)
2545 uint32_t read_section_id
;
2547 if (!migrate_get_current()->send_section_footer
) {
2548 /* No footer to check */
2552 read_mark
= qemu_get_byte(f
);
2554 ret
= qemu_file_get_error(f
);
2556 error_report("%s: Read section footer failed: %d",
2561 if (read_mark
!= QEMU_VM_SECTION_FOOTER
) {
2562 error_report("Missing section footer for %s", se
->idstr
);
2566 read_section_id
= qemu_get_be32(f
);
2567 if (read_section_id
!= se
->load_section_id
) {
2568 error_report("Mismatched section id in footer for %s -"
2569 " read 0x%x expected 0x%x",
2570 se
->idstr
, read_section_id
, se
->load_section_id
);
2579 qemu_loadvm_section_start_full(QEMUFile
*f
, MigrationIncomingState
*mis
,
2582 bool trace_downtime
= (type
== QEMU_VM_SECTION_FULL
);
2583 uint32_t instance_id
, version_id
, section_id
;
2584 int64_t start_ts
, end_ts
;
2589 /* Read section start */
2590 section_id
= qemu_get_be32(f
);
2591 if (!qemu_get_counted_string(f
, idstr
)) {
2592 error_report("Unable to read ID string for section %u",
2596 instance_id
= qemu_get_be32(f
);
2597 version_id
= qemu_get_be32(f
);
2599 ret
= qemu_file_get_error(f
);
2601 error_report("%s: Failed to read instance/version ID: %d",
2606 trace_qemu_loadvm_state_section_startfull(section_id
, idstr
,
2607 instance_id
, version_id
);
2608 /* Find savevm section */
2609 se
= find_se(idstr
, instance_id
);
2611 error_report("Unknown savevm section or instance '%s' %"PRIu32
". "
2612 "Make sure that your current VM setup matches your "
2613 "saved VM setup, including any hotplugged devices",
2614 idstr
, instance_id
);
2618 /* Validate version */
2619 if (version_id
> se
->version_id
) {
2620 error_report("savevm: unsupported version %d for '%s' v%d",
2621 version_id
, idstr
, se
->version_id
);
2624 se
->load_version_id
= version_id
;
2625 se
->load_section_id
= section_id
;
2627 /* Validate if it is a device's state */
2628 if (xen_enabled() && se
->is_ram
) {
2629 error_report("loadvm: %s RAM loading not allowed on Xen", idstr
);
2633 if (trace_downtime
) {
2634 start_ts
= qemu_clock_get_us(QEMU_CLOCK_REALTIME
);
2637 ret
= vmstate_load(f
, se
);
2639 error_report("error while loading state for instance 0x%"PRIx32
" of"
2640 " device '%s'", instance_id
, idstr
);
2644 if (trace_downtime
) {
2645 end_ts
= qemu_clock_get_us(QEMU_CLOCK_REALTIME
);
2646 trace_vmstate_downtime_load("non-iterable", se
->idstr
,
2647 se
->instance_id
, end_ts
- start_ts
);
2650 if (!check_section_footer(f
, se
)) {
2658 qemu_loadvm_section_part_end(QEMUFile
*f
, MigrationIncomingState
*mis
,
2661 bool trace_downtime
= (type
== QEMU_VM_SECTION_END
);
2662 int64_t start_ts
, end_ts
;
2663 uint32_t section_id
;
2667 section_id
= qemu_get_be32(f
);
2669 ret
= qemu_file_get_error(f
);
2671 error_report("%s: Failed to read section ID: %d",
2676 trace_qemu_loadvm_state_section_partend(section_id
);
2677 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
2678 if (se
->load_section_id
== section_id
) {
2683 error_report("Unknown savevm section %d", section_id
);
2687 if (trace_downtime
) {
2688 start_ts
= qemu_clock_get_us(QEMU_CLOCK_REALTIME
);
2691 ret
= vmstate_load(f
, se
);
2693 error_report("error while loading state section id %d(%s)",
2694 section_id
, se
->idstr
);
2698 if (trace_downtime
) {
2699 end_ts
= qemu_clock_get_us(QEMU_CLOCK_REALTIME
);
2700 trace_vmstate_downtime_load("iterable", se
->idstr
,
2701 se
->instance_id
, end_ts
- start_ts
);
2704 if (!check_section_footer(f
, se
)) {
2711 static int qemu_loadvm_state_header(QEMUFile
*f
)
2716 v
= qemu_get_be32(f
);
2717 if (v
!= QEMU_VM_FILE_MAGIC
) {
2718 error_report("Not a migration stream");
2722 v
= qemu_get_be32(f
);
2723 if (v
== QEMU_VM_FILE_VERSION_COMPAT
) {
2724 error_report("SaveVM v2 format is obsolete and don't work anymore");
2727 if (v
!= QEMU_VM_FILE_VERSION
) {
2728 error_report("Unsupported migration stream version");
2732 if (migrate_get_current()->send_configuration
) {
2733 if (qemu_get_byte(f
) != QEMU_VM_CONFIGURATION
) {
2734 error_report("Configuration section missing");
2735 qemu_loadvm_state_cleanup();
2738 ret
= vmstate_load_state(f
, &vmstate_configuration
, &savevm_state
, 0);
2741 qemu_loadvm_state_cleanup();
2748 static void qemu_loadvm_state_switchover_ack_needed(MigrationIncomingState
*mis
)
2752 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
2753 if (!se
->ops
|| !se
->ops
->switchover_ack_needed
) {
2757 if (se
->ops
->switchover_ack_needed(se
->opaque
)) {
2758 mis
->switchover_ack_pending_num
++;
2762 trace_loadvm_state_switchover_ack_needed(mis
->switchover_ack_pending_num
);
2765 static int qemu_loadvm_state_setup(QEMUFile
*f
, Error
**errp
)
2771 trace_loadvm_state_setup();
2772 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
2773 if (!se
->ops
|| !se
->ops
->load_setup
) {
2776 if (se
->ops
->is_active
) {
2777 if (!se
->ops
->is_active(se
->opaque
)) {
2782 ret
= se
->ops
->load_setup(f
, se
->opaque
, errp
);
2784 error_prepend(errp
, "Load state of device %s failed: ",
2786 qemu_file_set_error(f
, ret
);
2793 void qemu_loadvm_state_cleanup(void)
2797 trace_loadvm_state_cleanup();
2798 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
2799 if (se
->ops
&& se
->ops
->load_cleanup
) {
2800 se
->ops
->load_cleanup(se
->opaque
);
2805 /* Return true if we should continue the migration, or false. */
2806 static bool postcopy_pause_incoming(MigrationIncomingState
*mis
)
2810 trace_postcopy_pause_incoming();
2812 assert(migrate_postcopy_ram());
2815 * Unregister yank with either from/to src would work, since ioc behind it
2818 migration_ioc_unregister_yank_from_file(mis
->from_src_file
);
2820 assert(mis
->from_src_file
);
2821 qemu_file_shutdown(mis
->from_src_file
);
2822 qemu_fclose(mis
->from_src_file
);
2823 mis
->from_src_file
= NULL
;
2825 assert(mis
->to_src_file
);
2826 qemu_file_shutdown(mis
->to_src_file
);
2827 qemu_mutex_lock(&mis
->rp_mutex
);
2828 qemu_fclose(mis
->to_src_file
);
2829 mis
->to_src_file
= NULL
;
2830 qemu_mutex_unlock(&mis
->rp_mutex
);
2833 * NOTE: this must happen before reset the PostcopyTmpPages below,
2834 * otherwise it's racy to reset those fields when the fast load thread
2835 * can be accessing it in parallel.
2837 if (mis
->postcopy_qemufile_dst
) {
2838 qemu_file_shutdown(mis
->postcopy_qemufile_dst
);
2839 /* Take the mutex to make sure the fast ram load thread halted */
2840 qemu_mutex_lock(&mis
->postcopy_prio_thread_mutex
);
2841 migration_ioc_unregister_yank_from_file(mis
->postcopy_qemufile_dst
);
2842 qemu_fclose(mis
->postcopy_qemufile_dst
);
2843 mis
->postcopy_qemufile_dst
= NULL
;
2844 qemu_mutex_unlock(&mis
->postcopy_prio_thread_mutex
);
2847 /* Current state can be either ACTIVE or RECOVER */
2848 migrate_set_state(&mis
->state
, mis
->state
,
2849 MIGRATION_STATUS_POSTCOPY_PAUSED
);
2851 /* Notify the fault thread for the invalidated file handle */
2852 postcopy_fault_thread_notify(mis
);
2855 * If network is interrupted, any temp page we received will be useless
2856 * because we didn't mark them as "received" in receivedmap. After a
2857 * proper recovery later (which will sync src dirty bitmap with receivedmap
2858 * on dest) these cached small pages will be resent again.
2860 for (i
= 0; i
< mis
->postcopy_channels
; i
++) {
2861 postcopy_temp_page_reset(&mis
->postcopy_tmp_pages
[i
]);
2864 error_report("Detected IO failure for postcopy. "
2865 "Migration paused.");
2867 while (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2868 qemu_sem_wait(&mis
->postcopy_pause_sem_dst
);
2871 trace_postcopy_pause_incoming_continued();
2876 int qemu_loadvm_state_main(QEMUFile
*f
, MigrationIncomingState
*mis
)
2878 uint8_t section_type
;
2883 section_type
= qemu_get_byte(f
);
2885 ret
= qemu_file_get_error_obj_any(f
, mis
->postcopy_qemufile_dst
, NULL
);
2890 trace_qemu_loadvm_state_section(section_type
);
2891 switch (section_type
) {
2892 case QEMU_VM_SECTION_START
:
2893 case QEMU_VM_SECTION_FULL
:
2894 ret
= qemu_loadvm_section_start_full(f
, mis
, section_type
);
2899 case QEMU_VM_SECTION_PART
:
2900 case QEMU_VM_SECTION_END
:
2901 ret
= qemu_loadvm_section_part_end(f
, mis
, section_type
);
2906 case QEMU_VM_COMMAND
:
2907 ret
= loadvm_process_command(f
);
2908 trace_qemu_loadvm_state_section_command(ret
);
2909 if ((ret
< 0) || (ret
== LOADVM_QUIT
)) {
2914 /* This is the end of migration */
2917 error_report("Unknown savevm section type %d", section_type
);
2925 qemu_file_set_error(f
, ret
);
2927 /* Cancel bitmaps incoming regardless of recovery */
2928 dirty_bitmap_mig_cancel_incoming();
2931 * If we are during an active postcopy, then we pause instead
2932 * of bail out to at least keep the VM's dirty data. Note
2933 * that POSTCOPY_INCOMING_LISTENING stage is still not enough,
2934 * during which we're still receiving device states and we
2935 * still haven't yet started the VM on destination.
2937 * Only RAM postcopy supports recovery. Still, if RAM postcopy is
2938 * enabled, canceled bitmaps postcopy will not affect RAM postcopy
2941 if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING
&&
2942 migrate_postcopy_ram() && postcopy_pause_incoming(mis
)) {
2943 /* Reset f to point to the newly created channel */
2944 f
= mis
->from_src_file
;
2951 int qemu_loadvm_state(QEMUFile
*f
)
2953 MigrationIncomingState
*mis
= migration_incoming_get_current();
2954 Error
*local_err
= NULL
;
2957 if (qemu_savevm_state_blocked(&local_err
)) {
2958 error_report_err(local_err
);
2962 ret
= qemu_loadvm_state_header(f
);
2967 if (qemu_loadvm_state_setup(f
, &local_err
) != 0) {
2968 error_report_err(local_err
);
2972 if (migrate_switchover_ack()) {
2973 qemu_loadvm_state_switchover_ack_needed(mis
);
2976 cpu_synchronize_all_pre_loadvm();
2978 ret
= qemu_loadvm_state_main(f
, mis
);
2979 qemu_event_set(&mis
->main_thread_load_event
);
2981 trace_qemu_loadvm_state_post_main(ret
);
2983 if (mis
->have_listen_thread
) {
2984 /* Listen thread still going, can't clean up yet */
2989 ret
= qemu_file_get_error(f
);
2993 * Try to read in the VMDESC section as well, so that dumping tools that
2994 * intercept our migration stream have the chance to see it.
2997 /* We've got to be careful; if we don't read the data and just shut the fd
2998 * then the sender can error if we close while it's still sending.
2999 * We also mustn't read data that isn't there; some transports (RDMA)
3000 * will stall waiting for that data when the source has already closed.
3002 if (ret
== 0 && should_send_vmdesc()) {
3005 uint8_t section_type
= qemu_get_byte(f
);
3007 if (section_type
!= QEMU_VM_VMDESCRIPTION
) {
3008 error_report("Expected vmdescription section, but got %d",
3011 * It doesn't seem worth failing at this point since
3012 * we apparently have an otherwise valid VM state
3015 buf
= g_malloc(0x1000);
3016 size
= qemu_get_be32(f
);
3019 uint32_t read_chunk
= MIN(size
, 0x1000);
3020 qemu_get_buffer(f
, buf
, read_chunk
);
3027 qemu_loadvm_state_cleanup();
3028 cpu_synchronize_all_post_init();
3033 int qemu_load_device_state(QEMUFile
*f
)
3035 MigrationIncomingState
*mis
= migration_incoming_get_current();
3038 /* Load QEMU_VM_SECTION_FULL section */
3039 ret
= qemu_loadvm_state_main(f
, mis
);
3041 error_report("Failed to load device state: %d", ret
);
3045 cpu_synchronize_all_post_init();
3049 int qemu_loadvm_approve_switchover(void)
3051 MigrationIncomingState
*mis
= migration_incoming_get_current();
3053 if (!mis
->switchover_ack_pending_num
) {
3057 mis
->switchover_ack_pending_num
--;
3058 trace_loadvm_approve_switchover(mis
->switchover_ack_pending_num
);
3060 if (mis
->switchover_ack_pending_num
) {
3064 return migrate_send_rp_switchover_ack(mis
);
3067 bool save_snapshot(const char *name
, bool overwrite
, const char *vmstate
,
3068 bool has_devices
, strList
*devices
, Error
**errp
)
3070 BlockDriverState
*bs
;
3071 QEMUSnapshotInfo sn1
, *sn
= &sn1
;
3074 RunState saved_state
= runstate_get();
3075 uint64_t vm_state_size
;
3076 g_autoptr(GDateTime
) now
= g_date_time_new_now_local();
3078 GLOBAL_STATE_CODE();
3080 if (migration_is_blocked(errp
)) {
3084 if (!replay_can_snapshot()) {
3085 error_setg(errp
, "Record/replay does not allow making snapshot "
3086 "right now. Try once more later.");
3090 if (!bdrv_all_can_snapshot(has_devices
, devices
, errp
)) {
3094 /* Delete old snapshots of the same name */
3097 if (bdrv_all_delete_snapshot(name
, has_devices
,
3098 devices
, errp
) < 0) {
3102 ret2
= bdrv_all_has_snapshot(name
, has_devices
, devices
, errp
);
3108 "Snapshot '%s' already exists in one or more devices",
3115 bs
= bdrv_all_find_vmstate_bs(vmstate
, has_devices
, devices
, errp
);
3120 global_state_store();
3121 vm_stop(RUN_STATE_SAVE_VM
);
3123 bdrv_drain_all_begin();
3125 memset(sn
, 0, sizeof(*sn
));
3127 /* fill auxiliary fields */
3128 sn
->date_sec
= g_date_time_to_unix(now
);
3129 sn
->date_nsec
= g_date_time_get_microsecond(now
) * 1000;
3130 sn
->vm_clock_nsec
= qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
);
3131 if (replay_mode
!= REPLAY_MODE_NONE
) {
3132 sn
->icount
= replay_get_current_icount();
3138 pstrcpy(sn
->name
, sizeof(sn
->name
), name
);
3140 g_autofree
char *autoname
= g_date_time_format(now
, "vm-%Y%m%d%H%M%S");
3141 pstrcpy(sn
->name
, sizeof(sn
->name
), autoname
);
3144 /* save the VM state */
3145 f
= qemu_fopen_bdrv(bs
, 1);
3147 error_setg(errp
, "Could not open VM state file");
3150 ret
= qemu_savevm_state(f
, errp
);
3151 vm_state_size
= qemu_file_transferred(f
);
3152 ret2
= qemu_fclose(f
);
3161 ret
= bdrv_all_create_snapshot(sn
, bs
, vm_state_size
,
3162 has_devices
, devices
, errp
);
3164 bdrv_all_delete_snapshot(sn
->name
, has_devices
, devices
, NULL
);
3171 bdrv_drain_all_end();
3173 vm_resume(saved_state
);
3177 void qmp_xen_save_devices_state(const char *filename
, bool has_live
, bool live
,
3181 QIOChannelFile
*ioc
;
3182 int saved_vm_running
;
3186 /* live default to true so old version of Xen tool stack can have a
3187 * successful live migration */
3191 saved_vm_running
= runstate_is_running();
3192 vm_stop(RUN_STATE_SAVE_VM
);
3193 global_state_store_running();
3195 ioc
= qio_channel_file_new_path(filename
, O_WRONLY
| O_CREAT
| O_TRUNC
,
3200 qio_channel_set_name(QIO_CHANNEL(ioc
), "migration-xen-save-state");
3201 f
= qemu_file_new_output(QIO_CHANNEL(ioc
));
3202 object_unref(OBJECT(ioc
));
3203 ret
= qemu_save_device_state(f
);
3204 if (ret
< 0 || qemu_fclose(f
) < 0) {
3205 error_setg(errp
, "saving Xen device state failed");
3207 /* libxl calls the QMP command "stop" before calling
3208 * "xen-save-devices-state" and in case of migration failure, libxl
3209 * would call "cont".
3210 * So call bdrv_inactivate_all (release locks) here to let the other
3211 * side of the migration take control of the images.
3213 if (live
&& !saved_vm_running
) {
3214 ret
= bdrv_inactivate_all();
3216 error_setg(errp
, "%s: bdrv_inactivate_all() failed (%d)",
3223 if (saved_vm_running
) {
3228 void qmp_xen_load_devices_state(const char *filename
, Error
**errp
)
3231 QIOChannelFile
*ioc
;
3234 /* Guest must be paused before loading the device state; the RAM state
3235 * will already have been loaded by xc
3237 if (runstate_is_running()) {
3238 error_setg(errp
, "Cannot update device state while vm is running");
3241 vm_stop(RUN_STATE_RESTORE_VM
);
3243 ioc
= qio_channel_file_new_path(filename
, O_RDONLY
| O_BINARY
, 0, errp
);
3247 qio_channel_set_name(QIO_CHANNEL(ioc
), "migration-xen-load-state");
3248 f
= qemu_file_new_input(QIO_CHANNEL(ioc
));
3249 object_unref(OBJECT(ioc
));
3251 ret
= qemu_loadvm_state(f
);
3254 error_setg(errp
, "loading Xen device state failed");
3256 migration_incoming_state_destroy();
3259 bool load_snapshot(const char *name
, const char *vmstate
,
3260 bool has_devices
, strList
*devices
, Error
**errp
)
3262 BlockDriverState
*bs_vm_state
;
3263 QEMUSnapshotInfo sn
;
3266 MigrationIncomingState
*mis
= migration_incoming_get_current();
3268 if (!bdrv_all_can_snapshot(has_devices
, devices
, errp
)) {
3271 ret
= bdrv_all_has_snapshot(name
, has_devices
, devices
, errp
);
3276 error_setg(errp
, "Snapshot '%s' does not exist in one or more devices",
3281 bs_vm_state
= bdrv_all_find_vmstate_bs(vmstate
, has_devices
, devices
, errp
);
3286 /* Don't even try to load empty VM states */
3287 ret
= bdrv_snapshot_find(bs_vm_state
, &sn
, name
);
3290 } else if (sn
.vm_state_size
== 0) {
3291 error_setg(errp
, "This is a disk-only snapshot. Revert to it "
3292 " offline using qemu-img");
3297 * Flush the record/replay queue. Now the VM state is going
3298 * to change. Therefore we don't need to preserve its consistency
3300 replay_flush_events();
3302 /* Flush all IO requests so they don't interfere with the new state. */
3303 bdrv_drain_all_begin();
3305 ret
= bdrv_all_goto_snapshot(name
, has_devices
, devices
, errp
);
3310 /* restore the VM state */
3311 f
= qemu_fopen_bdrv(bs_vm_state
, 0);
3313 error_setg(errp
, "Could not open VM state file");
3317 qemu_system_reset(SHUTDOWN_CAUSE_SNAPSHOT_LOAD
);
3318 mis
->from_src_file
= f
;
3320 if (!yank_register_instance(MIGRATION_YANK_INSTANCE
, errp
)) {
3324 ret
= qemu_loadvm_state(f
);
3325 migration_incoming_state_destroy();
3327 bdrv_drain_all_end();
3330 error_setg(errp
, "Error %d while loading VM state", ret
);
3337 bdrv_drain_all_end();
3341 void load_snapshot_resume(RunState state
)
3344 if (state
== RUN_STATE_RUNNING
&& runstate_get() == RUN_STATE_SUSPENDED
) {
3345 qemu_system_wakeup_request(QEMU_WAKEUP_REASON_OTHER
, &error_abort
);
3349 bool delete_snapshot(const char *name
, bool has_devices
,
3350 strList
*devices
, Error
**errp
)
3352 if (!bdrv_all_can_snapshot(has_devices
, devices
, errp
)) {
3356 if (bdrv_all_delete_snapshot(name
, has_devices
, devices
, errp
) < 0) {
3363 void vmstate_register_ram(MemoryRegion
*mr
, DeviceState
*dev
)
3365 qemu_ram_set_idstr(mr
->ram_block
,
3366 memory_region_name(mr
), dev
);
3367 qemu_ram_set_migratable(mr
->ram_block
);
3370 void vmstate_unregister_ram(MemoryRegion
*mr
, DeviceState
*dev
)
3372 qemu_ram_unset_idstr(mr
->ram_block
);
3373 qemu_ram_unset_migratable(mr
->ram_block
);
3376 void vmstate_register_ram_global(MemoryRegion
*mr
)
3378 vmstate_register_ram(mr
, NULL
);
3381 bool vmstate_check_only_migratable(const VMStateDescription
*vmsd
)
3383 /* check needed if --only-migratable is specified */
3384 if (!only_migratable
) {
3388 return !(vmsd
&& vmsd
->unmigratable
);
3391 typedef struct SnapshotJob
{
3401 static void qmp_snapshot_job_free(SnapshotJob
*s
)
3405 qapi_free_strList(s
->devices
);
3409 static void snapshot_load_job_bh(void *opaque
)
3412 SnapshotJob
*s
= container_of(job
, SnapshotJob
, common
);
3413 RunState orig_state
= runstate_get();
3415 job_progress_set_remaining(&s
->common
, 1);
3417 vm_stop(RUN_STATE_RESTORE_VM
);
3419 s
->ret
= load_snapshot(s
->tag
, s
->vmstate
, true, s
->devices
, s
->errp
);
3421 load_snapshot_resume(orig_state
);
3424 job_progress_update(&s
->common
, 1);
3426 qmp_snapshot_job_free(s
);
3430 static void snapshot_save_job_bh(void *opaque
)
3433 SnapshotJob
*s
= container_of(job
, SnapshotJob
, common
);
3435 job_progress_set_remaining(&s
->common
, 1);
3436 s
->ret
= save_snapshot(s
->tag
, false, s
->vmstate
,
3437 true, s
->devices
, s
->errp
);
3438 job_progress_update(&s
->common
, 1);
3440 qmp_snapshot_job_free(s
);
3444 static void snapshot_delete_job_bh(void *opaque
)
3447 SnapshotJob
*s
= container_of(job
, SnapshotJob
, common
);
3449 job_progress_set_remaining(&s
->common
, 1);
3450 s
->ret
= delete_snapshot(s
->tag
, true, s
->devices
, s
->errp
);
3451 job_progress_update(&s
->common
, 1);
3453 qmp_snapshot_job_free(s
);
3457 static int coroutine_fn
snapshot_save_job_run(Job
*job
, Error
**errp
)
3459 SnapshotJob
*s
= container_of(job
, SnapshotJob
, common
);
3461 s
->co
= qemu_coroutine_self();
3462 aio_bh_schedule_oneshot(qemu_get_aio_context(),
3463 snapshot_save_job_bh
, job
);
3464 qemu_coroutine_yield();
3465 return s
->ret
? 0 : -1;
3468 static int coroutine_fn
snapshot_load_job_run(Job
*job
, Error
**errp
)
3470 SnapshotJob
*s
= container_of(job
, SnapshotJob
, common
);
3472 s
->co
= qemu_coroutine_self();
3473 aio_bh_schedule_oneshot(qemu_get_aio_context(),
3474 snapshot_load_job_bh
, job
);
3475 qemu_coroutine_yield();
3476 return s
->ret
? 0 : -1;
3479 static int coroutine_fn
snapshot_delete_job_run(Job
*job
, Error
**errp
)
3481 SnapshotJob
*s
= container_of(job
, SnapshotJob
, common
);
3483 s
->co
= qemu_coroutine_self();
3484 aio_bh_schedule_oneshot(qemu_get_aio_context(),
3485 snapshot_delete_job_bh
, job
);
3486 qemu_coroutine_yield();
3487 return s
->ret
? 0 : -1;
3491 static const JobDriver snapshot_load_job_driver
= {
3492 .instance_size
= sizeof(SnapshotJob
),
3493 .job_type
= JOB_TYPE_SNAPSHOT_LOAD
,
3494 .run
= snapshot_load_job_run
,
3497 static const JobDriver snapshot_save_job_driver
= {
3498 .instance_size
= sizeof(SnapshotJob
),
3499 .job_type
= JOB_TYPE_SNAPSHOT_SAVE
,
3500 .run
= snapshot_save_job_run
,
3503 static const JobDriver snapshot_delete_job_driver
= {
3504 .instance_size
= sizeof(SnapshotJob
),
3505 .job_type
= JOB_TYPE_SNAPSHOT_DELETE
,
3506 .run
= snapshot_delete_job_run
,
3510 void qmp_snapshot_save(const char *job_id
,
3512 const char *vmstate
,
3518 s
= job_create(job_id
, &snapshot_save_job_driver
, NULL
,
3519 qemu_get_aio_context(), JOB_MANUAL_DISMISS
,
3525 s
->tag
= g_strdup(tag
);
3526 s
->vmstate
= g_strdup(vmstate
);
3527 s
->devices
= QAPI_CLONE(strList
, devices
);
3529 job_start(&s
->common
);
3532 void qmp_snapshot_load(const char *job_id
,
3534 const char *vmstate
,
3540 s
= job_create(job_id
, &snapshot_load_job_driver
, NULL
,
3541 qemu_get_aio_context(), JOB_MANUAL_DISMISS
,
3547 s
->tag
= g_strdup(tag
);
3548 s
->vmstate
= g_strdup(vmstate
);
3549 s
->devices
= QAPI_CLONE(strList
, devices
);
3551 job_start(&s
->common
);
3554 void qmp_snapshot_delete(const char *job_id
,
3561 s
= job_create(job_id
, &snapshot_delete_job_driver
, NULL
,
3562 qemu_get_aio_context(), JOB_MANUAL_DISMISS
,
3568 s
->tag
= g_strdup(tag
);
3569 s
->devices
= QAPI_CLONE(strList
, devices
);
3571 job_start(&s
->common
);