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"
31 #include "hw/boards.h"
34 #include "hw/xen/xen.h"
36 #include "sysemu/sysemu.h"
37 #include "qemu/timer.h"
38 #include "migration/migration.h"
39 #include "postcopy-ram.h"
40 #include "qapi/qmp/qerror.h"
41 #include "qemu/error-report.h"
42 #include "qemu/queue.h"
43 #include "sysemu/cpus.h"
44 #include "exec/memory.h"
45 #include "qmp-commands.h"
47 #include "qemu/bitops.h"
49 #include "block/snapshot.h"
50 #include "qemu/cutils.h"
51 #include "io/channel-buffer.h"
52 #include "io/channel-file.h"
55 #define ETH_P_RARP 0x8035
57 #define ARP_HTYPE_ETH 0x0001
58 #define ARP_PTYPE_IP 0x0800
59 #define ARP_OP_REQUEST_REV 0x3
61 const unsigned int postcopy_ram_discard_version
= 0;
63 static bool skip_section_footers
;
65 static struct mig_cmd_args
{
66 ssize_t len
; /* -1 = variable */
69 [MIG_CMD_INVALID
] = { .len
= -1, .name
= "INVALID" },
70 [MIG_CMD_OPEN_RETURN_PATH
] = { .len
= 0, .name
= "OPEN_RETURN_PATH" },
71 [MIG_CMD_PING
] = { .len
= sizeof(uint32_t), .name
= "PING" },
72 [MIG_CMD_POSTCOPY_ADVISE
] = { .len
= 16, .name
= "POSTCOPY_ADVISE" },
73 [MIG_CMD_POSTCOPY_LISTEN
] = { .len
= 0, .name
= "POSTCOPY_LISTEN" },
74 [MIG_CMD_POSTCOPY_RUN
] = { .len
= 0, .name
= "POSTCOPY_RUN" },
75 [MIG_CMD_POSTCOPY_RAM_DISCARD
] = {
76 .len
= -1, .name
= "POSTCOPY_RAM_DISCARD" },
77 [MIG_CMD_PACKAGED
] = { .len
= 4, .name
= "PACKAGED" },
78 [MIG_CMD_MAX
] = { .len
= -1, .name
= "MAX" },
81 static int announce_self_create(uint8_t *buf
,
84 /* Ethernet header. */
85 memset(buf
, 0xff, 6); /* destination MAC addr */
86 memcpy(buf
+ 6, mac_addr
, 6); /* source MAC addr */
87 *(uint16_t *)(buf
+ 12) = htons(ETH_P_RARP
); /* ethertype */
90 *(uint16_t *)(buf
+ 14) = htons(ARP_HTYPE_ETH
); /* hardware addr space */
91 *(uint16_t *)(buf
+ 16) = htons(ARP_PTYPE_IP
); /* protocol addr space */
92 *(buf
+ 18) = 6; /* hardware addr length (ethernet) */
93 *(buf
+ 19) = 4; /* protocol addr length (IPv4) */
94 *(uint16_t *)(buf
+ 20) = htons(ARP_OP_REQUEST_REV
); /* opcode */
95 memcpy(buf
+ 22, mac_addr
, 6); /* source hw addr */
96 memset(buf
+ 28, 0x00, 4); /* source protocol addr */
97 memcpy(buf
+ 32, mac_addr
, 6); /* target hw addr */
98 memset(buf
+ 38, 0x00, 4); /* target protocol addr */
100 /* Padding to get up to 60 bytes (ethernet min packet size, minus FCS). */
101 memset(buf
+ 42, 0x00, 18);
103 return 60; /* len (FCS will be added by hardware) */
106 static void qemu_announce_self_iter(NICState
*nic
, void *opaque
)
111 trace_qemu_announce_self_iter(qemu_ether_ntoa(&nic
->conf
->macaddr
));
112 len
= announce_self_create(buf
, nic
->conf
->macaddr
.a
);
114 qemu_send_packet_raw(qemu_get_queue(nic
), buf
, len
);
118 static void qemu_announce_self_once(void *opaque
)
120 static int count
= SELF_ANNOUNCE_ROUNDS
;
121 QEMUTimer
*timer
= *(QEMUTimer
**)opaque
;
123 qemu_foreach_nic(qemu_announce_self_iter
, NULL
);
126 /* delay 50ms, 150ms, 250ms, ... */
127 timer_mod(timer
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) +
128 self_announce_delay(count
));
135 void qemu_announce_self(void)
137 static QEMUTimer
*timer
;
138 timer
= timer_new_ms(QEMU_CLOCK_REALTIME
, qemu_announce_self_once
, &timer
);
139 qemu_announce_self_once(&timer
);
142 /***********************************************************/
143 /* savevm/loadvm support */
145 static ssize_t
block_writev_buffer(void *opaque
, struct iovec
*iov
, int iovcnt
,
151 qemu_iovec_init_external(&qiov
, iov
, iovcnt
);
152 ret
= bdrv_writev_vmstate(opaque
, &qiov
, pos
);
160 static ssize_t
block_get_buffer(void *opaque
, uint8_t *buf
, int64_t pos
,
163 return bdrv_load_vmstate(opaque
, buf
, pos
, size
);
166 static int bdrv_fclose(void *opaque
)
168 return bdrv_flush(opaque
);
171 static const QEMUFileOps bdrv_read_ops
= {
172 .get_buffer
= block_get_buffer
,
176 static const QEMUFileOps bdrv_write_ops
= {
177 .writev_buffer
= block_writev_buffer
,
181 static QEMUFile
*qemu_fopen_bdrv(BlockDriverState
*bs
, int is_writable
)
184 return qemu_fopen_ops(bs
, &bdrv_write_ops
);
186 return qemu_fopen_ops(bs
, &bdrv_read_ops
);
190 /* QEMUFile timer support.
191 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
194 void timer_put(QEMUFile
*f
, QEMUTimer
*ts
)
196 uint64_t expire_time
;
198 expire_time
= timer_expire_time_ns(ts
);
199 qemu_put_be64(f
, expire_time
);
202 void timer_get(QEMUFile
*f
, QEMUTimer
*ts
)
204 uint64_t expire_time
;
206 expire_time
= qemu_get_be64(f
);
207 if (expire_time
!= -1) {
208 timer_mod_ns(ts
, expire_time
);
215 /* VMState timer support.
216 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
219 static int get_timer(QEMUFile
*f
, void *pv
, size_t size
, VMStateField
*field
)
226 static int put_timer(QEMUFile
*f
, void *pv
, size_t size
, VMStateField
*field
,
235 const VMStateInfo vmstate_info_timer
= {
242 typedef struct CompatEntry
{
247 typedef struct SaveStateEntry
{
248 QTAILQ_ENTRY(SaveStateEntry
) entry
;
255 const VMStateDescription
*vmsd
;
261 typedef struct SaveState
{
262 QTAILQ_HEAD(, SaveStateEntry
) handlers
;
263 int global_section_id
;
264 bool skip_configuration
;
267 uint32_t target_page_bits
;
270 static SaveState savevm_state
= {
271 .handlers
= QTAILQ_HEAD_INITIALIZER(savevm_state
.handlers
),
272 .global_section_id
= 0,
273 .skip_configuration
= false,
276 void savevm_skip_configuration(void)
278 savevm_state
.skip_configuration
= true;
282 static void configuration_pre_save(void *opaque
)
284 SaveState
*state
= opaque
;
285 const char *current_name
= MACHINE_GET_CLASS(current_machine
)->name
;
287 state
->len
= strlen(current_name
);
288 state
->name
= current_name
;
289 state
->target_page_bits
= TARGET_PAGE_BITS
;
292 static int configuration_pre_load(void *opaque
)
294 SaveState
*state
= opaque
;
296 /* If there is no target-page-bits subsection it means the source
297 * predates the variable-target-page-bits support and is using the
298 * minimum possible value for this CPU.
300 state
->target_page_bits
= TARGET_PAGE_BITS_MIN
;
304 static int configuration_post_load(void *opaque
, int version_id
)
306 SaveState
*state
= opaque
;
307 const char *current_name
= MACHINE_GET_CLASS(current_machine
)->name
;
309 if (strncmp(state
->name
, current_name
, state
->len
) != 0) {
310 error_report("Machine type received is '%.*s' and local is '%s'",
311 (int) state
->len
, state
->name
, current_name
);
315 if (state
->target_page_bits
!= TARGET_PAGE_BITS
) {
316 error_report("Received TARGET_PAGE_BITS is %d but local is %d",
317 state
->target_page_bits
, TARGET_PAGE_BITS
);
324 /* The target-page-bits subsection is present only if the
325 * target page size is not the same as the default (ie the
326 * minimum page size for a variable-page-size guest CPU).
327 * If it is present then it contains the actual target page
328 * bits for the machine, and migration will fail if the
329 * two ends don't agree about it.
331 static bool vmstate_target_page_bits_needed(void *opaque
)
333 return TARGET_PAGE_BITS
> TARGET_PAGE_BITS_MIN
;
336 static const VMStateDescription vmstate_target_page_bits
= {
337 .name
= "configuration/target-page-bits",
339 .minimum_version_id
= 1,
340 .needed
= vmstate_target_page_bits_needed
,
341 .fields
= (VMStateField
[]) {
342 VMSTATE_UINT32(target_page_bits
, SaveState
),
343 VMSTATE_END_OF_LIST()
347 static const VMStateDescription vmstate_configuration
= {
348 .name
= "configuration",
350 .pre_load
= configuration_pre_load
,
351 .post_load
= configuration_post_load
,
352 .pre_save
= configuration_pre_save
,
353 .fields
= (VMStateField
[]) {
354 VMSTATE_UINT32(len
, SaveState
),
355 VMSTATE_VBUFFER_ALLOC_UINT32(name
, SaveState
, 0, NULL
, len
),
356 VMSTATE_END_OF_LIST()
358 .subsections
= (const VMStateDescription
*[]) {
359 &vmstate_target_page_bits
,
364 static void dump_vmstate_vmsd(FILE *out_file
,
365 const VMStateDescription
*vmsd
, int indent
,
368 static void dump_vmstate_vmsf(FILE *out_file
, const VMStateField
*field
,
371 fprintf(out_file
, "%*s{\n", indent
, "");
373 fprintf(out_file
, "%*s\"field\": \"%s\",\n", indent
, "", field
->name
);
374 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
376 fprintf(out_file
, "%*s\"field_exists\": %s,\n", indent
, "",
377 field
->field_exists
? "true" : "false");
378 fprintf(out_file
, "%*s\"size\": %zu", indent
, "", field
->size
);
379 if (field
->vmsd
!= NULL
) {
380 fprintf(out_file
, ",\n");
381 dump_vmstate_vmsd(out_file
, field
->vmsd
, indent
, false);
383 fprintf(out_file
, "\n%*s}", indent
- 2, "");
386 static void dump_vmstate_vmss(FILE *out_file
,
387 const VMStateDescription
**subsection
,
390 if (*subsection
!= NULL
) {
391 dump_vmstate_vmsd(out_file
, *subsection
, indent
, true);
395 static void dump_vmstate_vmsd(FILE *out_file
,
396 const VMStateDescription
*vmsd
, int indent
,
400 fprintf(out_file
, "%*s{\n", indent
, "");
402 fprintf(out_file
, "%*s\"%s\": {\n", indent
, "", "Description");
405 fprintf(out_file
, "%*s\"name\": \"%s\",\n", indent
, "", vmsd
->name
);
406 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
408 fprintf(out_file
, "%*s\"minimum_version_id\": %d", indent
, "",
409 vmsd
->minimum_version_id
);
410 if (vmsd
->fields
!= NULL
) {
411 const VMStateField
*field
= vmsd
->fields
;
414 fprintf(out_file
, ",\n%*s\"Fields\": [\n", indent
, "");
416 while (field
->name
!= NULL
) {
417 if (field
->flags
& VMS_MUST_EXIST
) {
418 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
423 fprintf(out_file
, ",\n");
425 dump_vmstate_vmsf(out_file
, field
, indent
+ 2);
429 fprintf(out_file
, "\n%*s]", indent
, "");
431 if (vmsd
->subsections
!= NULL
) {
432 const VMStateDescription
**subsection
= vmsd
->subsections
;
435 fprintf(out_file
, ",\n%*s\"Subsections\": [\n", indent
, "");
437 while (*subsection
!= NULL
) {
439 fprintf(out_file
, ",\n");
441 dump_vmstate_vmss(out_file
, subsection
, indent
+ 2);
445 fprintf(out_file
, "\n%*s]", indent
, "");
447 fprintf(out_file
, "\n%*s}", indent
- 2, "");
450 static void dump_machine_type(FILE *out_file
)
454 mc
= MACHINE_GET_CLASS(current_machine
);
456 fprintf(out_file
, " \"vmschkmachine\": {\n");
457 fprintf(out_file
, " \"Name\": \"%s\"\n", mc
->name
);
458 fprintf(out_file
, " },\n");
461 void dump_vmstate_json_to_file(FILE *out_file
)
466 fprintf(out_file
, "{\n");
467 dump_machine_type(out_file
);
470 list
= object_class_get_list(TYPE_DEVICE
, true);
471 for (elt
= list
; elt
; elt
= elt
->next
) {
472 DeviceClass
*dc
= OBJECT_CLASS_CHECK(DeviceClass
, elt
->data
,
482 fprintf(out_file
, ",\n");
484 name
= object_class_get_name(OBJECT_CLASS(dc
));
485 fprintf(out_file
, "%*s\"%s\": {\n", indent
, "", name
);
487 fprintf(out_file
, "%*s\"Name\": \"%s\",\n", indent
, "", name
);
488 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
489 dc
->vmsd
->version_id
);
490 fprintf(out_file
, "%*s\"minimum_version_id\": %d,\n", indent
, "",
491 dc
->vmsd
->minimum_version_id
);
493 dump_vmstate_vmsd(out_file
, dc
->vmsd
, indent
, false);
495 fprintf(out_file
, "\n%*s}", indent
- 2, "");
498 fprintf(out_file
, "\n}\n");
502 static int calculate_new_instance_id(const char *idstr
)
507 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
508 if (strcmp(idstr
, se
->idstr
) == 0
509 && instance_id
<= se
->instance_id
) {
510 instance_id
= se
->instance_id
+ 1;
516 static int calculate_compat_instance_id(const char *idstr
)
521 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
526 if (strcmp(idstr
, se
->compat
->idstr
) == 0
527 && instance_id
<= se
->compat
->instance_id
) {
528 instance_id
= se
->compat
->instance_id
+ 1;
534 static inline MigrationPriority
save_state_priority(SaveStateEntry
*se
)
537 return se
->vmsd
->priority
;
539 return MIG_PRI_DEFAULT
;
542 static void savevm_state_handler_insert(SaveStateEntry
*nse
)
544 MigrationPriority priority
= save_state_priority(nse
);
547 assert(priority
<= MIG_PRI_MAX
);
549 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
550 if (save_state_priority(se
) < priority
) {
556 QTAILQ_INSERT_BEFORE(se
, nse
, entry
);
558 QTAILQ_INSERT_TAIL(&savevm_state
.handlers
, nse
, entry
);
562 /* TODO: Individual devices generally have very little idea about the rest
563 of the system, so instance_id should be removed/replaced.
564 Meanwhile pass -1 as instance_id if you do not already have a clearly
565 distinguishing id for all instances of your device class. */
566 int register_savevm_live(DeviceState
*dev
,
575 se
= g_new0(SaveStateEntry
, 1);
576 se
->version_id
= version_id
;
577 se
->section_id
= savevm_state
.global_section_id
++;
581 /* if this is a live_savem then set is_ram */
582 if (ops
->save_live_setup
!= NULL
) {
587 char *id
= qdev_get_dev_path(dev
);
589 if (snprintf(se
->idstr
, sizeof(se
->idstr
), "%s/", id
) >=
591 error_report("Path too long for VMState (%s)", id
);
599 se
->compat
= g_new0(CompatEntry
, 1);
600 pstrcpy(se
->compat
->idstr
, sizeof(se
->compat
->idstr
), idstr
);
601 se
->compat
->instance_id
= instance_id
== -1 ?
602 calculate_compat_instance_id(idstr
) : instance_id
;
606 pstrcat(se
->idstr
, sizeof(se
->idstr
), idstr
);
608 if (instance_id
== -1) {
609 se
->instance_id
= calculate_new_instance_id(se
->idstr
);
611 se
->instance_id
= instance_id
;
613 assert(!se
->compat
|| se
->instance_id
== 0);
614 savevm_state_handler_insert(se
);
618 int register_savevm(DeviceState
*dev
,
622 SaveStateHandler
*save_state
,
623 LoadStateHandler
*load_state
,
626 SaveVMHandlers
*ops
= g_new0(SaveVMHandlers
, 1);
627 ops
->save_state
= save_state
;
628 ops
->load_state
= load_state
;
629 return register_savevm_live(dev
, idstr
, instance_id
, version_id
,
633 void unregister_savevm(DeviceState
*dev
, const char *idstr
, void *opaque
)
635 SaveStateEntry
*se
, *new_se
;
639 char *path
= qdev_get_dev_path(dev
);
641 pstrcpy(id
, sizeof(id
), path
);
642 pstrcat(id
, sizeof(id
), "/");
646 pstrcat(id
, sizeof(id
), idstr
);
648 QTAILQ_FOREACH_SAFE(se
, &savevm_state
.handlers
, entry
, new_se
) {
649 if (strcmp(se
->idstr
, id
) == 0 && se
->opaque
== opaque
) {
650 QTAILQ_REMOVE(&savevm_state
.handlers
, se
, entry
);
658 int vmstate_register_with_alias_id(DeviceState
*dev
, int instance_id
,
659 const VMStateDescription
*vmsd
,
660 void *opaque
, int alias_id
,
661 int required_for_version
,
666 /* If this triggers, alias support can be dropped for the vmsd. */
667 assert(alias_id
== -1 || required_for_version
>= vmsd
->minimum_version_id
);
669 se
= g_new0(SaveStateEntry
, 1);
670 se
->version_id
= vmsd
->version_id
;
671 se
->section_id
= savevm_state
.global_section_id
++;
674 se
->alias_id
= alias_id
;
677 char *id
= qdev_get_dev_path(dev
);
679 if (snprintf(se
->idstr
, sizeof(se
->idstr
), "%s/", id
) >=
681 error_setg(errp
, "Path too long for VMState (%s)", id
);
689 se
->compat
= g_new0(CompatEntry
, 1);
690 pstrcpy(se
->compat
->idstr
, sizeof(se
->compat
->idstr
), vmsd
->name
);
691 se
->compat
->instance_id
= instance_id
== -1 ?
692 calculate_compat_instance_id(vmsd
->name
) : instance_id
;
696 pstrcat(se
->idstr
, sizeof(se
->idstr
), vmsd
->name
);
698 if (instance_id
== -1) {
699 se
->instance_id
= calculate_new_instance_id(se
->idstr
);
701 se
->instance_id
= instance_id
;
703 assert(!se
->compat
|| se
->instance_id
== 0);
704 savevm_state_handler_insert(se
);
708 void vmstate_unregister(DeviceState
*dev
, const VMStateDescription
*vmsd
,
711 SaveStateEntry
*se
, *new_se
;
713 QTAILQ_FOREACH_SAFE(se
, &savevm_state
.handlers
, entry
, new_se
) {
714 if (se
->vmsd
== vmsd
&& se
->opaque
== opaque
) {
715 QTAILQ_REMOVE(&savevm_state
.handlers
, se
, entry
);
722 static int vmstate_load(QEMUFile
*f
, SaveStateEntry
*se
, int version_id
)
724 trace_vmstate_load(se
->idstr
, se
->vmsd
? se
->vmsd
->name
: "(old)");
725 if (!se
->vmsd
) { /* Old style */
726 return se
->ops
->load_state(f
, se
->opaque
, version_id
);
728 return vmstate_load_state(f
, se
->vmsd
, se
->opaque
, version_id
);
731 static void vmstate_save_old_style(QEMUFile
*f
, SaveStateEntry
*se
, QJSON
*vmdesc
)
733 int64_t old_offset
, size
;
735 old_offset
= qemu_ftell_fast(f
);
736 se
->ops
->save_state(f
, se
->opaque
);
737 size
= qemu_ftell_fast(f
) - old_offset
;
740 json_prop_int(vmdesc
, "size", size
);
741 json_start_array(vmdesc
, "fields");
742 json_start_object(vmdesc
, NULL
);
743 json_prop_str(vmdesc
, "name", "data");
744 json_prop_int(vmdesc
, "size", size
);
745 json_prop_str(vmdesc
, "type", "buffer");
746 json_end_object(vmdesc
);
747 json_end_array(vmdesc
);
751 static void vmstate_save(QEMUFile
*f
, SaveStateEntry
*se
, QJSON
*vmdesc
)
753 trace_vmstate_save(se
->idstr
, se
->vmsd
? se
->vmsd
->name
: "(old)");
755 vmstate_save_old_style(f
, se
, vmdesc
);
758 vmstate_save_state(f
, se
->vmsd
, se
->opaque
, vmdesc
);
761 void savevm_skip_section_footers(void)
763 skip_section_footers
= true;
767 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
769 static void save_section_header(QEMUFile
*f
, SaveStateEntry
*se
,
770 uint8_t section_type
)
772 qemu_put_byte(f
, section_type
);
773 qemu_put_be32(f
, se
->section_id
);
775 if (section_type
== QEMU_VM_SECTION_FULL
||
776 section_type
== QEMU_VM_SECTION_START
) {
778 size_t len
= strlen(se
->idstr
);
779 qemu_put_byte(f
, len
);
780 qemu_put_buffer(f
, (uint8_t *)se
->idstr
, len
);
782 qemu_put_be32(f
, se
->instance_id
);
783 qemu_put_be32(f
, se
->version_id
);
788 * Write a footer onto device sections that catches cases misformatted device
791 static void save_section_footer(QEMUFile
*f
, SaveStateEntry
*se
)
793 if (!skip_section_footers
) {
794 qemu_put_byte(f
, QEMU_VM_SECTION_FOOTER
);
795 qemu_put_be32(f
, se
->section_id
);
800 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
801 * command and associated data.
803 * @f: File to send command on
804 * @command: Command type to send
805 * @len: Length of associated data
806 * @data: Data associated with command.
808 void qemu_savevm_command_send(QEMUFile
*f
,
809 enum qemu_vm_cmd command
,
813 trace_savevm_command_send(command
, len
);
814 qemu_put_byte(f
, QEMU_VM_COMMAND
);
815 qemu_put_be16(f
, (uint16_t)command
);
816 qemu_put_be16(f
, len
);
817 qemu_put_buffer(f
, data
, len
);
821 void qemu_savevm_send_ping(QEMUFile
*f
, uint32_t value
)
825 trace_savevm_send_ping(value
);
826 buf
= cpu_to_be32(value
);
827 qemu_savevm_command_send(f
, MIG_CMD_PING
, sizeof(value
), (uint8_t *)&buf
);
830 void qemu_savevm_send_open_return_path(QEMUFile
*f
)
832 trace_savevm_send_open_return_path();
833 qemu_savevm_command_send(f
, MIG_CMD_OPEN_RETURN_PATH
, 0, NULL
);
836 /* We have a buffer of data to send; we don't want that all to be loaded
837 * by the command itself, so the command contains just the length of the
838 * extra buffer that we then send straight after it.
839 * TODO: Must be a better way to organise that
845 int qemu_savevm_send_packaged(QEMUFile
*f
, const uint8_t *buf
, size_t len
)
849 if (len
> MAX_VM_CMD_PACKAGED_SIZE
) {
850 error_report("%s: Unreasonably large packaged state: %zu",
855 tmp
= cpu_to_be32(len
);
857 trace_qemu_savevm_send_packaged();
858 qemu_savevm_command_send(f
, MIG_CMD_PACKAGED
, 4, (uint8_t *)&tmp
);
860 qemu_put_buffer(f
, buf
, len
);
865 /* Send prior to any postcopy transfer */
866 void qemu_savevm_send_postcopy_advise(QEMUFile
*f
)
869 tmp
[0] = cpu_to_be64(ram_pagesize_summary());
870 tmp
[1] = cpu_to_be64(qemu_target_page_size());
872 trace_qemu_savevm_send_postcopy_advise();
873 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_ADVISE
, 16, (uint8_t *)tmp
);
876 /* Sent prior to starting the destination running in postcopy, discard pages
877 * that have already been sent but redirtied on the source.
878 * CMD_POSTCOPY_RAM_DISCARD consist of:
880 * byte Length of name field (not including 0)
881 * n x byte RAM block name
882 * byte 0 terminator (just for safety)
883 * n x Byte ranges within the named RAMBlock
884 * be64 Start of the range
887 * name: RAMBlock name that these entries are part of
888 * len: Number of page entries
889 * start_list: 'len' addresses
890 * length_list: 'len' addresses
893 void qemu_savevm_send_postcopy_ram_discard(QEMUFile
*f
, const char *name
,
895 uint64_t *start_list
,
896 uint64_t *length_list
)
901 size_t name_len
= strlen(name
);
903 trace_qemu_savevm_send_postcopy_ram_discard(name
, len
);
904 assert(name_len
< 256);
905 buf
= g_malloc0(1 + 1 + name_len
+ 1 + (8 + 8) * len
);
906 buf
[0] = postcopy_ram_discard_version
;
908 memcpy(buf
+ 2, name
, name_len
);
909 tmplen
= 2 + name_len
;
910 buf
[tmplen
++] = '\0';
912 for (t
= 0; t
< len
; t
++) {
913 stq_be_p(buf
+ tmplen
, start_list
[t
]);
915 stq_be_p(buf
+ tmplen
, length_list
[t
]);
918 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RAM_DISCARD
, tmplen
, buf
);
922 /* Get the destination into a state where it can receive postcopy data. */
923 void qemu_savevm_send_postcopy_listen(QEMUFile
*f
)
925 trace_savevm_send_postcopy_listen();
926 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_LISTEN
, 0, NULL
);
929 /* Kick the destination into running */
930 void qemu_savevm_send_postcopy_run(QEMUFile
*f
)
932 trace_savevm_send_postcopy_run();
933 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RUN
, 0, NULL
);
936 bool qemu_savevm_state_blocked(Error
**errp
)
940 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
941 if (se
->vmsd
&& se
->vmsd
->unmigratable
) {
942 error_setg(errp
, "State blocked by non-migratable device '%s'",
950 static bool enforce_config_section(void)
952 MachineState
*machine
= MACHINE(qdev_get_machine());
953 return machine
->enforce_config_section
;
956 void qemu_savevm_state_header(QEMUFile
*f
)
958 trace_savevm_state_header();
959 qemu_put_be32(f
, QEMU_VM_FILE_MAGIC
);
960 qemu_put_be32(f
, QEMU_VM_FILE_VERSION
);
962 if (!savevm_state
.skip_configuration
|| enforce_config_section()) {
963 qemu_put_byte(f
, QEMU_VM_CONFIGURATION
);
964 vmstate_save_state(f
, &vmstate_configuration
, &savevm_state
, 0);
969 void qemu_savevm_state_begin(QEMUFile
*f
,
970 const MigrationParams
*params
)
975 trace_savevm_state_begin();
976 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
977 if (!se
->ops
|| !se
->ops
->set_params
) {
980 se
->ops
->set_params(params
, se
->opaque
);
983 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
984 if (!se
->ops
|| !se
->ops
->save_live_setup
) {
987 if (se
->ops
&& se
->ops
->is_active
) {
988 if (!se
->ops
->is_active(se
->opaque
)) {
992 save_section_header(f
, se
, QEMU_VM_SECTION_START
);
994 ret
= se
->ops
->save_live_setup(f
, se
->opaque
);
995 save_section_footer(f
, se
);
997 qemu_file_set_error(f
, ret
);
1004 * this function has three return values:
1005 * negative: there was one error, and we have -errno.
1006 * 0 : We haven't finished, caller have to go again
1007 * 1 : We have finished, we can go to complete phase
1009 int qemu_savevm_state_iterate(QEMUFile
*f
, bool postcopy
)
1014 trace_savevm_state_iterate();
1015 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1016 if (!se
->ops
|| !se
->ops
->save_live_iterate
) {
1019 if (se
->ops
&& se
->ops
->is_active
) {
1020 if (!se
->ops
->is_active(se
->opaque
)) {
1025 * In the postcopy phase, any device that doesn't know how to
1026 * do postcopy should have saved it's state in the _complete
1027 * call that's already run, it might get confused if we call
1028 * iterate afterwards.
1030 if (postcopy
&& !se
->ops
->save_live_complete_postcopy
) {
1033 if (qemu_file_rate_limit(f
)) {
1036 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1038 save_section_header(f
, se
, QEMU_VM_SECTION_PART
);
1040 ret
= se
->ops
->save_live_iterate(f
, se
->opaque
);
1041 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1042 save_section_footer(f
, se
);
1045 qemu_file_set_error(f
, ret
);
1048 /* Do not proceed to the next vmstate before this one reported
1049 completion of the current stage. This serializes the migration
1050 and reduces the probability that a faster changing state is
1051 synchronized over and over again. */
1058 static bool should_send_vmdesc(void)
1060 MachineState
*machine
= MACHINE(qdev_get_machine());
1061 bool in_postcopy
= migration_in_postcopy();
1062 return !machine
->suppress_vmdesc
&& !in_postcopy
;
1066 * Calls the save_live_complete_postcopy methods
1067 * causing the last few pages to be sent immediately and doing any associated
1069 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1070 * all the other devices, but that happens at the point we switch to postcopy.
1072 void qemu_savevm_state_complete_postcopy(QEMUFile
*f
)
1077 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1078 if (!se
->ops
|| !se
->ops
->save_live_complete_postcopy
) {
1081 if (se
->ops
&& se
->ops
->is_active
) {
1082 if (!se
->ops
->is_active(se
->opaque
)) {
1086 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1088 qemu_put_byte(f
, QEMU_VM_SECTION_END
);
1089 qemu_put_be32(f
, se
->section_id
);
1091 ret
= se
->ops
->save_live_complete_postcopy(f
, se
->opaque
);
1092 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1093 save_section_footer(f
, se
);
1095 qemu_file_set_error(f
, ret
);
1100 qemu_put_byte(f
, QEMU_VM_EOF
);
1104 void qemu_savevm_state_complete_precopy(QEMUFile
*f
, bool iterable_only
)
1110 bool in_postcopy
= migration_in_postcopy();
1112 trace_savevm_state_complete_precopy();
1114 cpu_synchronize_all_states();
1116 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1118 (in_postcopy
&& se
->ops
->save_live_complete_postcopy
) ||
1119 (in_postcopy
&& !iterable_only
) ||
1120 !se
->ops
->save_live_complete_precopy
) {
1124 if (se
->ops
&& se
->ops
->is_active
) {
1125 if (!se
->ops
->is_active(se
->opaque
)) {
1129 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1131 save_section_header(f
, se
, QEMU_VM_SECTION_END
);
1133 ret
= se
->ops
->save_live_complete_precopy(f
, se
->opaque
);
1134 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1135 save_section_footer(f
, se
);
1137 qemu_file_set_error(f
, ret
);
1142 if (iterable_only
) {
1146 vmdesc
= qjson_new();
1147 json_prop_int(vmdesc
, "page_size", TARGET_PAGE_SIZE
);
1148 json_start_array(vmdesc
, "devices");
1149 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1151 if ((!se
->ops
|| !se
->ops
->save_state
) && !se
->vmsd
) {
1154 if (se
->vmsd
&& !vmstate_save_needed(se
->vmsd
, se
->opaque
)) {
1155 trace_savevm_section_skip(se
->idstr
, se
->section_id
);
1159 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1161 json_start_object(vmdesc
, NULL
);
1162 json_prop_str(vmdesc
, "name", se
->idstr
);
1163 json_prop_int(vmdesc
, "instance_id", se
->instance_id
);
1165 save_section_header(f
, se
, QEMU_VM_SECTION_FULL
);
1166 vmstate_save(f
, se
, vmdesc
);
1167 trace_savevm_section_end(se
->idstr
, se
->section_id
, 0);
1168 save_section_footer(f
, se
);
1170 json_end_object(vmdesc
);
1174 /* Postcopy stream will still be going */
1175 qemu_put_byte(f
, QEMU_VM_EOF
);
1178 json_end_array(vmdesc
);
1179 qjson_finish(vmdesc
);
1180 vmdesc_len
= strlen(qjson_get_str(vmdesc
));
1182 if (should_send_vmdesc()) {
1183 qemu_put_byte(f
, QEMU_VM_VMDESCRIPTION
);
1184 qemu_put_be32(f
, vmdesc_len
);
1185 qemu_put_buffer(f
, (uint8_t *)qjson_get_str(vmdesc
), vmdesc_len
);
1187 qjson_destroy(vmdesc
);
1192 /* Give an estimate of the amount left to be transferred,
1193 * the result is split into the amount for units that can and
1194 * for units that can't do postcopy.
1196 void qemu_savevm_state_pending(QEMUFile
*f
, uint64_t threshold_size
,
1197 uint64_t *res_non_postcopiable
,
1198 uint64_t *res_postcopiable
)
1202 *res_non_postcopiable
= 0;
1203 *res_postcopiable
= 0;
1206 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1207 if (!se
->ops
|| !se
->ops
->save_live_pending
) {
1210 if (se
->ops
&& se
->ops
->is_active
) {
1211 if (!se
->ops
->is_active(se
->opaque
)) {
1215 se
->ops
->save_live_pending(f
, se
->opaque
, threshold_size
,
1216 res_non_postcopiable
, res_postcopiable
);
1220 void qemu_savevm_state_cleanup(void)
1224 trace_savevm_state_cleanup();
1225 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1226 if (se
->ops
&& se
->ops
->cleanup
) {
1227 se
->ops
->cleanup(se
->opaque
);
1232 static int qemu_savevm_state(QEMUFile
*f
, Error
**errp
)
1235 MigrationParams params
= {
1239 MigrationState
*ms
= migrate_init(¶ms
);
1240 MigrationStatus status
;
1241 ms
->to_dst_file
= f
;
1243 if (migration_is_blocked(errp
)) {
1248 qemu_mutex_unlock_iothread();
1249 qemu_savevm_state_header(f
);
1250 qemu_savevm_state_begin(f
, ¶ms
);
1251 qemu_mutex_lock_iothread();
1253 while (qemu_file_get_error(f
) == 0) {
1254 if (qemu_savevm_state_iterate(f
, false) > 0) {
1259 ret
= qemu_file_get_error(f
);
1261 qemu_savevm_state_complete_precopy(f
, false);
1262 ret
= qemu_file_get_error(f
);
1264 qemu_savevm_state_cleanup();
1266 error_setg_errno(errp
, -ret
, "Error while writing VM state");
1271 status
= MIGRATION_STATUS_FAILED
;
1273 status
= MIGRATION_STATUS_COMPLETED
;
1275 migrate_set_state(&ms
->state
, MIGRATION_STATUS_SETUP
, status
);
1277 /* f is outer parameter, it should not stay in global migration state after
1278 * this function finished */
1279 ms
->to_dst_file
= NULL
;
1284 static int qemu_save_device_state(QEMUFile
*f
)
1288 qemu_put_be32(f
, QEMU_VM_FILE_MAGIC
);
1289 qemu_put_be32(f
, QEMU_VM_FILE_VERSION
);
1291 cpu_synchronize_all_states();
1293 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1297 if ((!se
->ops
|| !se
->ops
->save_state
) && !se
->vmsd
) {
1300 if (se
->vmsd
&& !vmstate_save_needed(se
->vmsd
, se
->opaque
)) {
1304 save_section_header(f
, se
, QEMU_VM_SECTION_FULL
);
1306 vmstate_save(f
, se
, NULL
);
1308 save_section_footer(f
, se
);
1311 qemu_put_byte(f
, QEMU_VM_EOF
);
1313 return qemu_file_get_error(f
);
1316 static SaveStateEntry
*find_se(const char *idstr
, int instance_id
)
1320 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1321 if (!strcmp(se
->idstr
, idstr
) &&
1322 (instance_id
== se
->instance_id
||
1323 instance_id
== se
->alias_id
))
1325 /* Migrating from an older version? */
1326 if (strstr(se
->idstr
, idstr
) && se
->compat
) {
1327 if (!strcmp(se
->compat
->idstr
, idstr
) &&
1328 (instance_id
== se
->compat
->instance_id
||
1329 instance_id
== se
->alias_id
))
1336 enum LoadVMExitCodes
{
1337 /* Allow a command to quit all layers of nested loadvm loops */
1341 static int qemu_loadvm_state_main(QEMUFile
*f
, MigrationIncomingState
*mis
);
1343 /* ------ incoming postcopy messages ------ */
1344 /* 'advise' arrives before any transfers just to tell us that a postcopy
1345 * *might* happen - it might be skipped if precopy transferred everything
1348 static int loadvm_postcopy_handle_advise(MigrationIncomingState
*mis
)
1350 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_ADVISE
);
1351 uint64_t remote_pagesize_summary
, local_pagesize_summary
, remote_tps
;
1353 trace_loadvm_postcopy_handle_advise();
1354 if (ps
!= POSTCOPY_INCOMING_NONE
) {
1355 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps
);
1359 if (!postcopy_ram_supported_by_host()) {
1360 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
1364 remote_pagesize_summary
= qemu_get_be64(mis
->from_src_file
);
1365 local_pagesize_summary
= ram_pagesize_summary();
1367 if (remote_pagesize_summary
!= local_pagesize_summary
) {
1369 * This detects two potential causes of mismatch:
1370 * a) A mismatch in host page sizes
1371 * Some combinations of mismatch are probably possible but it gets
1372 * a bit more complicated. In particular we need to place whole
1373 * host pages on the dest at once, and we need to ensure that we
1374 * handle dirtying to make sure we never end up sending part of
1375 * a hostpage on it's own.
1376 * b) The use of different huge page sizes on source/destination
1377 * a more fine grain test is performed during RAM block migration
1378 * but this test here causes a nice early clear failure, and
1379 * also fails when passed to an older qemu that doesn't
1382 error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1384 remote_pagesize_summary
, local_pagesize_summary
);
1388 remote_tps
= qemu_get_be64(mis
->from_src_file
);
1389 if (remote_tps
!= qemu_target_page_size()) {
1391 * Again, some differences could be dealt with, but for now keep it
1394 error_report("Postcopy needs matching target page sizes (s=%d d=%zd)",
1395 (int)remote_tps
, qemu_target_page_size());
1399 if (ram_postcopy_incoming_init(mis
)) {
1403 postcopy_state_set(POSTCOPY_INCOMING_ADVISE
);
1408 /* After postcopy we will be told to throw some pages away since they're
1409 * dirty and will have to be demand fetched. Must happen before CPU is
1411 * There can be 0..many of these messages, each encoding multiple pages.
1413 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState
*mis
,
1418 PostcopyState ps
= postcopy_state_get();
1420 trace_loadvm_postcopy_ram_handle_discard();
1423 case POSTCOPY_INCOMING_ADVISE
:
1425 tmp
= postcopy_ram_prepare_discard(mis
);
1431 case POSTCOPY_INCOMING_DISCARD
:
1432 /* Expected state */
1436 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1440 /* We're expecting a
1442 * a RAM ID string (length byte, name, 0 term)
1443 * then at least 1 16 byte chunk
1445 if (len
< (1 + 1 + 1 + 1 + 2 * 8)) {
1446 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len
);
1450 tmp
= qemu_get_byte(mis
->from_src_file
);
1451 if (tmp
!= postcopy_ram_discard_version
) {
1452 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp
);
1456 if (!qemu_get_counted_string(mis
->from_src_file
, ramid
)) {
1457 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1460 tmp
= qemu_get_byte(mis
->from_src_file
);
1462 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp
);
1466 len
-= 3 + strlen(ramid
);
1468 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len
);
1471 trace_loadvm_postcopy_ram_handle_discard_header(ramid
, len
);
1473 uint64_t start_addr
, block_length
;
1474 start_addr
= qemu_get_be64(mis
->from_src_file
);
1475 block_length
= qemu_get_be64(mis
->from_src_file
);
1478 int ret
= ram_discard_range(ramid
, start_addr
, block_length
);
1483 trace_loadvm_postcopy_ram_handle_discard_end();
1489 * Triggered by a postcopy_listen command; this thread takes over reading
1490 * the input stream, leaving the main thread free to carry on loading the rest
1491 * of the device state (from RAM).
1492 * (TODO:This could do with being in a postcopy file - but there again it's
1493 * just another input loop, not that postcopy specific)
1495 static void *postcopy_ram_listen_thread(void *opaque
)
1497 QEMUFile
*f
= opaque
;
1498 MigrationIncomingState
*mis
= migration_incoming_get_current();
1501 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
1502 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1503 qemu_sem_post(&mis
->listen_thread_sem
);
1504 trace_postcopy_ram_listen_thread_start();
1507 * Because we're a thread and not a coroutine we can't yield
1508 * in qemu_file, and thus we must be blocking now.
1510 qemu_file_set_blocking(f
, true);
1511 load_res
= qemu_loadvm_state_main(f
, mis
);
1512 /* And non-blocking again so we don't block in any cleanup */
1513 qemu_file_set_blocking(f
, false);
1515 trace_postcopy_ram_listen_thread_exit();
1517 error_report("%s: loadvm failed: %d", __func__
, load_res
);
1518 qemu_file_set_error(f
, load_res
);
1519 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1520 MIGRATION_STATUS_FAILED
);
1523 * This looks good, but it's possible that the device loading in the
1524 * main thread hasn't finished yet, and so we might not be in 'RUN'
1525 * state yet; wait for the end of the main thread.
1527 qemu_event_wait(&mis
->main_thread_load_event
);
1529 postcopy_ram_incoming_cleanup(mis
);
1533 * If something went wrong then we have a bad state so exit;
1534 * depending how far we got it might be possible at this point
1535 * to leave the guest running and fire MCEs for pages that never
1536 * arrived as a desperate recovery step.
1541 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1542 MIGRATION_STATUS_COMPLETED
);
1544 * If everything has worked fine, then the main thread has waited
1545 * for us to start, and we're the last use of the mis.
1546 * (If something broke then qemu will have to exit anyway since it's
1547 * got a bad migration state).
1549 migration_incoming_state_destroy();
1555 /* After this message we must be able to immediately receive postcopy data */
1556 static int loadvm_postcopy_handle_listen(MigrationIncomingState
*mis
)
1558 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_LISTENING
);
1559 trace_loadvm_postcopy_handle_listen();
1560 if (ps
!= POSTCOPY_INCOMING_ADVISE
&& ps
!= POSTCOPY_INCOMING_DISCARD
) {
1561 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps
);
1564 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
1566 * A rare case, we entered listen without having to do any discards,
1567 * so do the setup that's normally done at the time of the 1st discard.
1569 postcopy_ram_prepare_discard(mis
);
1573 * Sensitise RAM - can now generate requests for blocks that don't exist
1574 * However, at this point the CPU shouldn't be running, and the IO
1575 * shouldn't be doing anything yet so don't actually expect requests
1577 if (postcopy_ram_enable_notify(mis
)) {
1581 if (mis
->have_listen_thread
) {
1582 error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread");
1586 mis
->have_listen_thread
= true;
1587 /* Start up the listening thread and wait for it to signal ready */
1588 qemu_sem_init(&mis
->listen_thread_sem
, 0);
1589 qemu_thread_create(&mis
->listen_thread
, "postcopy/listen",
1590 postcopy_ram_listen_thread
, mis
->from_src_file
,
1591 QEMU_THREAD_DETACHED
);
1592 qemu_sem_wait(&mis
->listen_thread_sem
);
1593 qemu_sem_destroy(&mis
->listen_thread_sem
);
1603 static void loadvm_postcopy_handle_run_bh(void *opaque
)
1605 Error
*local_err
= NULL
;
1606 HandleRunBhData
*data
= opaque
;
1608 /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1611 cpu_synchronize_all_post_init();
1613 qemu_announce_self();
1615 /* Make sure all file formats flush their mutable metadata.
1616 * If we get an error here, just don't restart the VM yet. */
1617 bdrv_invalidate_cache_all(&local_err
);
1619 error_report_err(local_err
);
1624 trace_loadvm_postcopy_handle_run_cpu_sync();
1625 cpu_synchronize_all_post_init();
1627 trace_loadvm_postcopy_handle_run_vmstart();
1630 /* Hold onto your hats, starting the CPU */
1633 /* leave it paused and let management decide when to start the CPU */
1634 runstate_set(RUN_STATE_PAUSED
);
1637 qemu_bh_delete(data
->bh
);
1641 /* After all discards we can start running and asking for pages */
1642 static int loadvm_postcopy_handle_run(MigrationIncomingState
*mis
)
1644 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_RUNNING
);
1645 HandleRunBhData
*data
;
1647 trace_loadvm_postcopy_handle_run();
1648 if (ps
!= POSTCOPY_INCOMING_LISTENING
) {
1649 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps
);
1653 data
= g_new(HandleRunBhData
, 1);
1654 data
->bh
= qemu_bh_new(loadvm_postcopy_handle_run_bh
, data
);
1655 qemu_bh_schedule(data
->bh
);
1657 /* We need to finish reading the stream from the package
1658 * and also stop reading anything more from the stream that loaded the
1659 * package (since it's now being read by the listener thread).
1660 * LOADVM_QUIT will quit all the layers of nested loadvm loops.
1666 * Immediately following this command is a blob of data containing an embedded
1667 * chunk of migration stream; read it and load it.
1669 * @mis: Incoming state
1670 * @length: Length of packaged data to read
1672 * Returns: Negative values on error
1675 static int loadvm_handle_cmd_packaged(MigrationIncomingState
*mis
)
1679 QIOChannelBuffer
*bioc
;
1681 length
= qemu_get_be32(mis
->from_src_file
);
1682 trace_loadvm_handle_cmd_packaged(length
);
1684 if (length
> MAX_VM_CMD_PACKAGED_SIZE
) {
1685 error_report("Unreasonably large packaged state: %zu", length
);
1689 bioc
= qio_channel_buffer_new(length
);
1690 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-loadvm-buffer");
1691 ret
= qemu_get_buffer(mis
->from_src_file
,
1694 if (ret
!= length
) {
1695 object_unref(OBJECT(bioc
));
1696 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
1698 return (ret
< 0) ? ret
: -EAGAIN
;
1700 bioc
->usage
+= length
;
1701 trace_loadvm_handle_cmd_packaged_received(ret
);
1703 QEMUFile
*packf
= qemu_fopen_channel_input(QIO_CHANNEL(bioc
));
1705 ret
= qemu_loadvm_state_main(packf
, mis
);
1706 trace_loadvm_handle_cmd_packaged_main(ret
);
1708 object_unref(OBJECT(bioc
));
1714 * Process an incoming 'QEMU_VM_COMMAND'
1715 * 0 just a normal return
1716 * LOADVM_QUIT All good, but exit the loop
1719 static int loadvm_process_command(QEMUFile
*f
)
1721 MigrationIncomingState
*mis
= migration_incoming_get_current();
1726 cmd
= qemu_get_be16(f
);
1727 len
= qemu_get_be16(f
);
1729 trace_loadvm_process_command(cmd
, len
);
1730 if (cmd
>= MIG_CMD_MAX
|| cmd
== MIG_CMD_INVALID
) {
1731 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd
, len
);
1735 if (mig_cmd_args
[cmd
].len
!= -1 && mig_cmd_args
[cmd
].len
!= len
) {
1736 error_report("%s received with bad length - expecting %zu, got %d",
1737 mig_cmd_args
[cmd
].name
,
1738 (size_t)mig_cmd_args
[cmd
].len
, len
);
1743 case MIG_CMD_OPEN_RETURN_PATH
:
1744 if (mis
->to_src_file
) {
1745 error_report("CMD_OPEN_RETURN_PATH called when RP already open");
1746 /* Not really a problem, so don't give up */
1749 mis
->to_src_file
= qemu_file_get_return_path(f
);
1750 if (!mis
->to_src_file
) {
1751 error_report("CMD_OPEN_RETURN_PATH failed");
1757 tmp32
= qemu_get_be32(f
);
1758 trace_loadvm_process_command_ping(tmp32
);
1759 if (!mis
->to_src_file
) {
1760 error_report("CMD_PING (0x%x) received with no return path",
1764 migrate_send_rp_pong(mis
, tmp32
);
1767 case MIG_CMD_PACKAGED
:
1768 return loadvm_handle_cmd_packaged(mis
);
1770 case MIG_CMD_POSTCOPY_ADVISE
:
1771 return loadvm_postcopy_handle_advise(mis
);
1773 case MIG_CMD_POSTCOPY_LISTEN
:
1774 return loadvm_postcopy_handle_listen(mis
);
1776 case MIG_CMD_POSTCOPY_RUN
:
1777 return loadvm_postcopy_handle_run(mis
);
1779 case MIG_CMD_POSTCOPY_RAM_DISCARD
:
1780 return loadvm_postcopy_ram_handle_discard(mis
, len
);
1786 struct LoadStateEntry
{
1787 QLIST_ENTRY(LoadStateEntry
) entry
;
1794 * Read a footer off the wire and check that it matches the expected section
1796 * Returns: true if the footer was good
1797 * false if there is a problem (and calls error_report to say why)
1799 static bool check_section_footer(QEMUFile
*f
, LoadStateEntry
*le
)
1802 uint32_t read_section_id
;
1804 if (skip_section_footers
) {
1805 /* No footer to check */
1809 read_mark
= qemu_get_byte(f
);
1811 if (read_mark
!= QEMU_VM_SECTION_FOOTER
) {
1812 error_report("Missing section footer for %s", le
->se
->idstr
);
1816 read_section_id
= qemu_get_be32(f
);
1817 if (read_section_id
!= le
->section_id
) {
1818 error_report("Mismatched section id in footer for %s -"
1819 " read 0x%x expected 0x%x",
1820 le
->se
->idstr
, read_section_id
, le
->section_id
);
1828 void loadvm_free_handlers(MigrationIncomingState
*mis
)
1830 LoadStateEntry
*le
, *new_le
;
1832 QLIST_FOREACH_SAFE(le
, &mis
->loadvm_handlers
, entry
, new_le
) {
1833 QLIST_REMOVE(le
, entry
);
1839 qemu_loadvm_section_start_full(QEMUFile
*f
, MigrationIncomingState
*mis
)
1841 uint32_t instance_id
, version_id
, section_id
;
1847 /* Read section start */
1848 section_id
= qemu_get_be32(f
);
1849 if (!qemu_get_counted_string(f
, idstr
)) {
1850 error_report("Unable to read ID string for section %u",
1854 instance_id
= qemu_get_be32(f
);
1855 version_id
= qemu_get_be32(f
);
1857 trace_qemu_loadvm_state_section_startfull(section_id
, idstr
,
1858 instance_id
, version_id
);
1859 /* Find savevm section */
1860 se
= find_se(idstr
, instance_id
);
1862 error_report("Unknown savevm section or instance '%s' %d",
1863 idstr
, instance_id
);
1867 /* Validate version */
1868 if (version_id
> se
->version_id
) {
1869 error_report("savevm: unsupported version %d for '%s' v%d",
1870 version_id
, idstr
, se
->version_id
);
1874 /* Validate if it is a device's state */
1875 if (xen_enabled() && se
->is_ram
) {
1876 error_report("loadvm: %s RAM loading not allowed on Xen", idstr
);
1881 le
= g_malloc0(sizeof(*le
));
1884 le
->section_id
= section_id
;
1885 le
->version_id
= version_id
;
1886 QLIST_INSERT_HEAD(&mis
->loadvm_handlers
, le
, entry
);
1888 ret
= vmstate_load(f
, le
->se
, le
->version_id
);
1890 error_report("error while loading state for instance 0x%x of"
1891 " device '%s'", instance_id
, idstr
);
1894 if (!check_section_footer(f
, le
)) {
1902 qemu_loadvm_section_part_end(QEMUFile
*f
, MigrationIncomingState
*mis
)
1904 uint32_t section_id
;
1908 section_id
= qemu_get_be32(f
);
1910 trace_qemu_loadvm_state_section_partend(section_id
);
1911 QLIST_FOREACH(le
, &mis
->loadvm_handlers
, entry
) {
1912 if (le
->section_id
== section_id
) {
1917 error_report("Unknown savevm section %d", section_id
);
1921 ret
= vmstate_load(f
, le
->se
, le
->version_id
);
1923 error_report("error while loading state section id %d(%s)",
1924 section_id
, le
->se
->idstr
);
1927 if (!check_section_footer(f
, le
)) {
1934 static int qemu_loadvm_state_main(QEMUFile
*f
, MigrationIncomingState
*mis
)
1936 uint8_t section_type
;
1939 while ((section_type
= qemu_get_byte(f
)) != QEMU_VM_EOF
) {
1941 trace_qemu_loadvm_state_section(section_type
);
1942 switch (section_type
) {
1943 case QEMU_VM_SECTION_START
:
1944 case QEMU_VM_SECTION_FULL
:
1945 ret
= qemu_loadvm_section_start_full(f
, mis
);
1950 case QEMU_VM_SECTION_PART
:
1951 case QEMU_VM_SECTION_END
:
1952 ret
= qemu_loadvm_section_part_end(f
, mis
);
1957 case QEMU_VM_COMMAND
:
1958 ret
= loadvm_process_command(f
);
1959 trace_qemu_loadvm_state_section_command(ret
);
1960 if ((ret
< 0) || (ret
& LOADVM_QUIT
)) {
1965 error_report("Unknown savevm section type %d", section_type
);
1973 qemu_file_set_error(f
, ret
);
1978 int qemu_loadvm_state(QEMUFile
*f
)
1980 MigrationIncomingState
*mis
= migration_incoming_get_current();
1981 Error
*local_err
= NULL
;
1985 if (qemu_savevm_state_blocked(&local_err
)) {
1986 error_report_err(local_err
);
1990 v
= qemu_get_be32(f
);
1991 if (v
!= QEMU_VM_FILE_MAGIC
) {
1992 error_report("Not a migration stream");
1996 v
= qemu_get_be32(f
);
1997 if (v
== QEMU_VM_FILE_VERSION_COMPAT
) {
1998 error_report("SaveVM v2 format is obsolete and don't work anymore");
2001 if (v
!= QEMU_VM_FILE_VERSION
) {
2002 error_report("Unsupported migration stream version");
2006 if (!savevm_state
.skip_configuration
|| enforce_config_section()) {
2007 if (qemu_get_byte(f
) != QEMU_VM_CONFIGURATION
) {
2008 error_report("Configuration section missing");
2011 ret
= vmstate_load_state(f
, &vmstate_configuration
, &savevm_state
, 0);
2018 ret
= qemu_loadvm_state_main(f
, mis
);
2019 qemu_event_set(&mis
->main_thread_load_event
);
2021 trace_qemu_loadvm_state_post_main(ret
);
2023 if (mis
->have_listen_thread
) {
2024 /* Listen thread still going, can't clean up yet */
2029 ret
= qemu_file_get_error(f
);
2033 * Try to read in the VMDESC section as well, so that dumping tools that
2034 * intercept our migration stream have the chance to see it.
2037 /* We've got to be careful; if we don't read the data and just shut the fd
2038 * then the sender can error if we close while it's still sending.
2039 * We also mustn't read data that isn't there; some transports (RDMA)
2040 * will stall waiting for that data when the source has already closed.
2042 if (ret
== 0 && should_send_vmdesc()) {
2045 uint8_t section_type
= qemu_get_byte(f
);
2047 if (section_type
!= QEMU_VM_VMDESCRIPTION
) {
2048 error_report("Expected vmdescription section, but got %d",
2051 * It doesn't seem worth failing at this point since
2052 * we apparently have an otherwise valid VM state
2055 buf
= g_malloc(0x1000);
2056 size
= qemu_get_be32(f
);
2059 uint32_t read_chunk
= MIN(size
, 0x1000);
2060 qemu_get_buffer(f
, buf
, read_chunk
);
2067 cpu_synchronize_all_post_init();
2072 int save_vmstate(const char *name
)
2074 BlockDriverState
*bs
, *bs1
;
2075 QEMUSnapshotInfo sn1
, *sn
= &sn1
, old_sn1
, *old_sn
= &old_sn1
;
2078 int saved_vm_running
;
2079 uint64_t vm_state_size
;
2082 Error
*local_err
= NULL
;
2083 AioContext
*aio_context
;
2085 if (!bdrv_all_can_snapshot(&bs
)) {
2086 error_report("Device '%s' is writable but does not support snapshots",
2087 bdrv_get_device_name(bs
));
2091 /* Delete old snapshots of the same name */
2093 ret
= bdrv_all_delete_snapshot(name
, &bs1
, &local_err
);
2095 error_reportf_err(local_err
,
2096 "Error while deleting snapshot on device '%s': ",
2097 bdrv_get_device_name(bs1
));
2102 bs
= bdrv_all_find_vmstate_bs();
2104 error_report("No block device can accept snapshots");
2107 aio_context
= bdrv_get_aio_context(bs
);
2109 saved_vm_running
= runstate_is_running();
2111 ret
= global_state_store();
2113 error_report("Error saving global state");
2116 vm_stop(RUN_STATE_SAVE_VM
);
2118 aio_context_acquire(aio_context
);
2120 memset(sn
, 0, sizeof(*sn
));
2122 /* fill auxiliary fields */
2123 qemu_gettimeofday(&tv
);
2124 sn
->date_sec
= tv
.tv_sec
;
2125 sn
->date_nsec
= tv
.tv_usec
* 1000;
2126 sn
->vm_clock_nsec
= qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
);
2129 ret
= bdrv_snapshot_find(bs
, old_sn
, name
);
2131 pstrcpy(sn
->name
, sizeof(sn
->name
), old_sn
->name
);
2132 pstrcpy(sn
->id_str
, sizeof(sn
->id_str
), old_sn
->id_str
);
2134 pstrcpy(sn
->name
, sizeof(sn
->name
), name
);
2137 /* cast below needed for OpenBSD where tv_sec is still 'long' */
2138 localtime_r((const time_t *)&tv
.tv_sec
, &tm
);
2139 strftime(sn
->name
, sizeof(sn
->name
), "vm-%Y%m%d%H%M%S", &tm
);
2142 /* save the VM state */
2143 f
= qemu_fopen_bdrv(bs
, 1);
2145 error_report("Could not open VM state file");
2148 ret
= qemu_savevm_state(f
, &local_err
);
2149 vm_state_size
= qemu_ftell(f
);
2152 error_report_err(local_err
);
2156 ret
= bdrv_all_create_snapshot(sn
, bs
, vm_state_size
, &bs
);
2158 error_report("Error while creating snapshot on '%s'",
2159 bdrv_get_device_name(bs
));
2166 aio_context_release(aio_context
);
2167 if (saved_vm_running
) {
2173 void qmp_xen_save_devices_state(const char *filename
, Error
**errp
)
2176 QIOChannelFile
*ioc
;
2177 int saved_vm_running
;
2180 saved_vm_running
= runstate_is_running();
2181 vm_stop(RUN_STATE_SAVE_VM
);
2182 global_state_store_running();
2184 ioc
= qio_channel_file_new_path(filename
, O_WRONLY
| O_CREAT
, 0660, errp
);
2188 qio_channel_set_name(QIO_CHANNEL(ioc
), "migration-xen-save-state");
2189 f
= qemu_fopen_channel_output(QIO_CHANNEL(ioc
));
2190 ret
= qemu_save_device_state(f
);
2193 error_setg(errp
, QERR_IO_ERROR
);
2197 if (saved_vm_running
) {
2202 void qmp_xen_load_devices_state(const char *filename
, Error
**errp
)
2205 QIOChannelFile
*ioc
;
2208 /* Guest must be paused before loading the device state; the RAM state
2209 * will already have been loaded by xc
2211 if (runstate_is_running()) {
2212 error_setg(errp
, "Cannot update device state while vm is running");
2215 vm_stop(RUN_STATE_RESTORE_VM
);
2217 ioc
= qio_channel_file_new_path(filename
, O_RDONLY
| O_BINARY
, 0, errp
);
2221 qio_channel_set_name(QIO_CHANNEL(ioc
), "migration-xen-load-state");
2222 f
= qemu_fopen_channel_input(QIO_CHANNEL(ioc
));
2224 ret
= qemu_loadvm_state(f
);
2227 error_setg(errp
, QERR_IO_ERROR
);
2229 migration_incoming_state_destroy();
2232 int load_vmstate(const char *name
)
2234 BlockDriverState
*bs
, *bs_vm_state
;
2235 QEMUSnapshotInfo sn
;
2238 AioContext
*aio_context
;
2239 MigrationIncomingState
*mis
= migration_incoming_get_current();
2241 if (!bdrv_all_can_snapshot(&bs
)) {
2242 error_report("Device '%s' is writable but does not support snapshots",
2243 bdrv_get_device_name(bs
));
2246 ret
= bdrv_all_find_snapshot(name
, &bs
);
2248 error_report("Device '%s' does not have the requested snapshot '%s'",
2249 bdrv_get_device_name(bs
), name
);
2253 bs_vm_state
= bdrv_all_find_vmstate_bs();
2255 error_report("No block device supports snapshots");
2258 aio_context
= bdrv_get_aio_context(bs_vm_state
);
2260 /* Don't even try to load empty VM states */
2261 aio_context_acquire(aio_context
);
2262 ret
= bdrv_snapshot_find(bs_vm_state
, &sn
, name
);
2263 aio_context_release(aio_context
);
2266 } else if (sn
.vm_state_size
== 0) {
2267 error_report("This is a disk-only snapshot. Revert to it offline "
2272 /* Flush all IO requests so they don't interfere with the new state. */
2275 ret
= bdrv_all_goto_snapshot(name
, &bs
);
2277 error_report("Error %d while activating snapshot '%s' on '%s'",
2278 ret
, name
, bdrv_get_device_name(bs
));
2282 /* restore the VM state */
2283 f
= qemu_fopen_bdrv(bs_vm_state
, 0);
2285 error_report("Could not open VM state file");
2289 qemu_system_reset(VMRESET_SILENT
);
2290 mis
->from_src_file
= f
;
2292 aio_context_acquire(aio_context
);
2293 ret
= qemu_loadvm_state(f
);
2295 aio_context_release(aio_context
);
2297 migration_incoming_state_destroy();
2299 error_report("Error %d while loading VM state", ret
);
2306 void vmstate_register_ram(MemoryRegion
*mr
, DeviceState
*dev
)
2308 qemu_ram_set_idstr(mr
->ram_block
,
2309 memory_region_name(mr
), dev
);
2312 void vmstate_unregister_ram(MemoryRegion
*mr
, DeviceState
*dev
)
2314 qemu_ram_unset_idstr(mr
->ram_block
);
2317 void vmstate_register_ram_global(MemoryRegion
*mr
)
2319 vmstate_register_ram(mr
, NULL
);