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"
35 #include "monitor/monitor.h"
36 #include "sysemu/sysemu.h"
37 #include "qemu/timer.h"
38 #include "audio/audio.h"
39 #include "migration/migration.h"
40 #include "migration/postcopy-ram.h"
41 #include "qapi/qmp/qerror.h"
42 #include "qemu/error-report.h"
43 #include "qemu/sockets.h"
44 #include "qemu/queue.h"
45 #include "sysemu/cpus.h"
46 #include "exec/memory.h"
47 #include "qmp-commands.h"
49 #include "qemu/bitops.h"
51 #include "block/snapshot.h"
52 #include "block/qapi.h"
53 #include "qemu/cutils.h"
54 #include "io/channel-buffer.h"
55 #include "io/channel-file.h"
58 #define ETH_P_RARP 0x8035
60 #define ARP_HTYPE_ETH 0x0001
61 #define ARP_PTYPE_IP 0x0800
62 #define ARP_OP_REQUEST_REV 0x3
64 const unsigned int postcopy_ram_discard_version
= 0;
66 static bool skip_section_footers
;
68 static struct mig_cmd_args
{
69 ssize_t len
; /* -1 = variable */
72 [MIG_CMD_INVALID
] = { .len
= -1, .name
= "INVALID" },
73 [MIG_CMD_OPEN_RETURN_PATH
] = { .len
= 0, .name
= "OPEN_RETURN_PATH" },
74 [MIG_CMD_PING
] = { .len
= sizeof(uint32_t), .name
= "PING" },
75 [MIG_CMD_POSTCOPY_ADVISE
] = { .len
= 16, .name
= "POSTCOPY_ADVISE" },
76 [MIG_CMD_POSTCOPY_LISTEN
] = { .len
= 0, .name
= "POSTCOPY_LISTEN" },
77 [MIG_CMD_POSTCOPY_RUN
] = { .len
= 0, .name
= "POSTCOPY_RUN" },
78 [MIG_CMD_POSTCOPY_RAM_DISCARD
] = {
79 .len
= -1, .name
= "POSTCOPY_RAM_DISCARD" },
80 [MIG_CMD_PACKAGED
] = { .len
= 4, .name
= "PACKAGED" },
81 [MIG_CMD_MAX
] = { .len
= -1, .name
= "MAX" },
84 static int announce_self_create(uint8_t *buf
,
87 /* Ethernet header. */
88 memset(buf
, 0xff, 6); /* destination MAC addr */
89 memcpy(buf
+ 6, mac_addr
, 6); /* source MAC addr */
90 *(uint16_t *)(buf
+ 12) = htons(ETH_P_RARP
); /* ethertype */
93 *(uint16_t *)(buf
+ 14) = htons(ARP_HTYPE_ETH
); /* hardware addr space */
94 *(uint16_t *)(buf
+ 16) = htons(ARP_PTYPE_IP
); /* protocol addr space */
95 *(buf
+ 18) = 6; /* hardware addr length (ethernet) */
96 *(buf
+ 19) = 4; /* protocol addr length (IPv4) */
97 *(uint16_t *)(buf
+ 20) = htons(ARP_OP_REQUEST_REV
); /* opcode */
98 memcpy(buf
+ 22, mac_addr
, 6); /* source hw addr */
99 memset(buf
+ 28, 0x00, 4); /* source protocol addr */
100 memcpy(buf
+ 32, mac_addr
, 6); /* target hw addr */
101 memset(buf
+ 38, 0x00, 4); /* target protocol addr */
103 /* Padding to get up to 60 bytes (ethernet min packet size, minus FCS). */
104 memset(buf
+ 42, 0x00, 18);
106 return 60; /* len (FCS will be added by hardware) */
109 static void qemu_announce_self_iter(NICState
*nic
, void *opaque
)
114 trace_qemu_announce_self_iter(qemu_ether_ntoa(&nic
->conf
->macaddr
));
115 len
= announce_self_create(buf
, nic
->conf
->macaddr
.a
);
117 qemu_send_packet_raw(qemu_get_queue(nic
), buf
, len
);
121 static void qemu_announce_self_once(void *opaque
)
123 static int count
= SELF_ANNOUNCE_ROUNDS
;
124 QEMUTimer
*timer
= *(QEMUTimer
**)opaque
;
126 qemu_foreach_nic(qemu_announce_self_iter
, NULL
);
129 /* delay 50ms, 150ms, 250ms, ... */
130 timer_mod(timer
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) +
131 self_announce_delay(count
));
138 void qemu_announce_self(void)
140 static QEMUTimer
*timer
;
141 timer
= timer_new_ms(QEMU_CLOCK_REALTIME
, qemu_announce_self_once
, &timer
);
142 qemu_announce_self_once(&timer
);
145 /***********************************************************/
146 /* savevm/loadvm support */
148 static ssize_t
block_writev_buffer(void *opaque
, struct iovec
*iov
, int iovcnt
,
154 qemu_iovec_init_external(&qiov
, iov
, iovcnt
);
155 ret
= bdrv_writev_vmstate(opaque
, &qiov
, pos
);
163 static ssize_t
block_put_buffer(void *opaque
, const uint8_t *buf
,
164 int64_t pos
, size_t size
)
166 bdrv_save_vmstate(opaque
, buf
, pos
, size
);
170 static ssize_t
block_get_buffer(void *opaque
, uint8_t *buf
, int64_t pos
,
173 return bdrv_load_vmstate(opaque
, buf
, pos
, size
);
176 static int bdrv_fclose(void *opaque
)
178 return bdrv_flush(opaque
);
181 static const QEMUFileOps bdrv_read_ops
= {
182 .get_buffer
= block_get_buffer
,
186 static const QEMUFileOps bdrv_write_ops
= {
187 .put_buffer
= block_put_buffer
,
188 .writev_buffer
= block_writev_buffer
,
192 static QEMUFile
*qemu_fopen_bdrv(BlockDriverState
*bs
, int is_writable
)
195 return qemu_fopen_ops(bs
, &bdrv_write_ops
);
197 return qemu_fopen_ops(bs
, &bdrv_read_ops
);
201 /* QEMUFile timer support.
202 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
205 void timer_put(QEMUFile
*f
, QEMUTimer
*ts
)
207 uint64_t expire_time
;
209 expire_time
= timer_expire_time_ns(ts
);
210 qemu_put_be64(f
, expire_time
);
213 void timer_get(QEMUFile
*f
, QEMUTimer
*ts
)
215 uint64_t expire_time
;
217 expire_time
= qemu_get_be64(f
);
218 if (expire_time
!= -1) {
219 timer_mod_ns(ts
, expire_time
);
226 /* VMState timer support.
227 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
230 static int get_timer(QEMUFile
*f
, void *pv
, size_t size
)
237 static void put_timer(QEMUFile
*f
, void *pv
, size_t size
)
243 const VMStateInfo vmstate_info_timer
= {
250 typedef struct CompatEntry
{
255 typedef struct SaveStateEntry
{
256 QTAILQ_ENTRY(SaveStateEntry
) entry
;
263 const VMStateDescription
*vmsd
;
269 typedef struct SaveState
{
270 QTAILQ_HEAD(, SaveStateEntry
) handlers
;
271 int global_section_id
;
272 bool skip_configuration
;
277 static SaveState savevm_state
= {
278 .handlers
= QTAILQ_HEAD_INITIALIZER(savevm_state
.handlers
),
279 .global_section_id
= 0,
280 .skip_configuration
= false,
283 void savevm_skip_configuration(void)
285 savevm_state
.skip_configuration
= true;
289 static void configuration_pre_save(void *opaque
)
291 SaveState
*state
= opaque
;
292 const char *current_name
= MACHINE_GET_CLASS(current_machine
)->name
;
294 state
->len
= strlen(current_name
);
295 state
->name
= current_name
;
298 static int configuration_post_load(void *opaque
, int version_id
)
300 SaveState
*state
= opaque
;
301 const char *current_name
= MACHINE_GET_CLASS(current_machine
)->name
;
303 if (strncmp(state
->name
, current_name
, state
->len
) != 0) {
304 error_report("Machine type received is '%.*s' and local is '%s'",
305 (int) state
->len
, state
->name
, current_name
);
311 static const VMStateDescription vmstate_configuration
= {
312 .name
= "configuration",
314 .post_load
= configuration_post_load
,
315 .pre_save
= configuration_pre_save
,
316 .fields
= (VMStateField
[]) {
317 VMSTATE_UINT32(len
, SaveState
),
318 VMSTATE_VBUFFER_ALLOC_UINT32(name
, SaveState
, 0, NULL
, 0, len
),
319 VMSTATE_END_OF_LIST()
323 static void dump_vmstate_vmsd(FILE *out_file
,
324 const VMStateDescription
*vmsd
, int indent
,
327 static void dump_vmstate_vmsf(FILE *out_file
, const VMStateField
*field
,
330 fprintf(out_file
, "%*s{\n", indent
, "");
332 fprintf(out_file
, "%*s\"field\": \"%s\",\n", indent
, "", field
->name
);
333 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
335 fprintf(out_file
, "%*s\"field_exists\": %s,\n", indent
, "",
336 field
->field_exists
? "true" : "false");
337 fprintf(out_file
, "%*s\"size\": %zu", indent
, "", field
->size
);
338 if (field
->vmsd
!= NULL
) {
339 fprintf(out_file
, ",\n");
340 dump_vmstate_vmsd(out_file
, field
->vmsd
, indent
, false);
342 fprintf(out_file
, "\n%*s}", indent
- 2, "");
345 static void dump_vmstate_vmss(FILE *out_file
,
346 const VMStateDescription
**subsection
,
349 if (*subsection
!= NULL
) {
350 dump_vmstate_vmsd(out_file
, *subsection
, indent
, true);
354 static void dump_vmstate_vmsd(FILE *out_file
,
355 const VMStateDescription
*vmsd
, int indent
,
359 fprintf(out_file
, "%*s{\n", indent
, "");
361 fprintf(out_file
, "%*s\"%s\": {\n", indent
, "", "Description");
364 fprintf(out_file
, "%*s\"name\": \"%s\",\n", indent
, "", vmsd
->name
);
365 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
367 fprintf(out_file
, "%*s\"minimum_version_id\": %d", indent
, "",
368 vmsd
->minimum_version_id
);
369 if (vmsd
->fields
!= NULL
) {
370 const VMStateField
*field
= vmsd
->fields
;
373 fprintf(out_file
, ",\n%*s\"Fields\": [\n", indent
, "");
375 while (field
->name
!= NULL
) {
376 if (field
->flags
& VMS_MUST_EXIST
) {
377 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
382 fprintf(out_file
, ",\n");
384 dump_vmstate_vmsf(out_file
, field
, indent
+ 2);
388 fprintf(out_file
, "\n%*s]", indent
, "");
390 if (vmsd
->subsections
!= NULL
) {
391 const VMStateDescription
**subsection
= vmsd
->subsections
;
394 fprintf(out_file
, ",\n%*s\"Subsections\": [\n", indent
, "");
396 while (*subsection
!= NULL
) {
398 fprintf(out_file
, ",\n");
400 dump_vmstate_vmss(out_file
, subsection
, indent
+ 2);
404 fprintf(out_file
, "\n%*s]", indent
, "");
406 fprintf(out_file
, "\n%*s}", indent
- 2, "");
409 static void dump_machine_type(FILE *out_file
)
413 mc
= MACHINE_GET_CLASS(current_machine
);
415 fprintf(out_file
, " \"vmschkmachine\": {\n");
416 fprintf(out_file
, " \"Name\": \"%s\"\n", mc
->name
);
417 fprintf(out_file
, " },\n");
420 void dump_vmstate_json_to_file(FILE *out_file
)
425 fprintf(out_file
, "{\n");
426 dump_machine_type(out_file
);
429 list
= object_class_get_list(TYPE_DEVICE
, true);
430 for (elt
= list
; elt
; elt
= elt
->next
) {
431 DeviceClass
*dc
= OBJECT_CLASS_CHECK(DeviceClass
, elt
->data
,
441 fprintf(out_file
, ",\n");
443 name
= object_class_get_name(OBJECT_CLASS(dc
));
444 fprintf(out_file
, "%*s\"%s\": {\n", indent
, "", name
);
446 fprintf(out_file
, "%*s\"Name\": \"%s\",\n", indent
, "", name
);
447 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
448 dc
->vmsd
->version_id
);
449 fprintf(out_file
, "%*s\"minimum_version_id\": %d,\n", indent
, "",
450 dc
->vmsd
->minimum_version_id
);
452 dump_vmstate_vmsd(out_file
, dc
->vmsd
, indent
, false);
454 fprintf(out_file
, "\n%*s}", indent
- 2, "");
457 fprintf(out_file
, "\n}\n");
461 static int calculate_new_instance_id(const char *idstr
)
466 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
467 if (strcmp(idstr
, se
->idstr
) == 0
468 && instance_id
<= se
->instance_id
) {
469 instance_id
= se
->instance_id
+ 1;
475 static int calculate_compat_instance_id(const char *idstr
)
480 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
485 if (strcmp(idstr
, se
->compat
->idstr
) == 0
486 && instance_id
<= se
->compat
->instance_id
) {
487 instance_id
= se
->compat
->instance_id
+ 1;
493 /* TODO: Individual devices generally have very little idea about the rest
494 of the system, so instance_id should be removed/replaced.
495 Meanwhile pass -1 as instance_id if you do not already have a clearly
496 distinguishing id for all instances of your device class. */
497 int register_savevm_live(DeviceState
*dev
,
506 se
= g_new0(SaveStateEntry
, 1);
507 se
->version_id
= version_id
;
508 se
->section_id
= savevm_state
.global_section_id
++;
512 /* if this is a live_savem then set is_ram */
513 if (ops
->save_live_setup
!= NULL
) {
518 char *id
= qdev_get_dev_path(dev
);
520 pstrcpy(se
->idstr
, sizeof(se
->idstr
), id
);
521 pstrcat(se
->idstr
, sizeof(se
->idstr
), "/");
524 se
->compat
= g_new0(CompatEntry
, 1);
525 pstrcpy(se
->compat
->idstr
, sizeof(se
->compat
->idstr
), idstr
);
526 se
->compat
->instance_id
= instance_id
== -1 ?
527 calculate_compat_instance_id(idstr
) : instance_id
;
531 pstrcat(se
->idstr
, sizeof(se
->idstr
), idstr
);
533 if (instance_id
== -1) {
534 se
->instance_id
= calculate_new_instance_id(se
->idstr
);
536 se
->instance_id
= instance_id
;
538 assert(!se
->compat
|| se
->instance_id
== 0);
539 /* add at the end of list */
540 QTAILQ_INSERT_TAIL(&savevm_state
.handlers
, se
, entry
);
544 int register_savevm(DeviceState
*dev
,
548 SaveStateHandler
*save_state
,
549 LoadStateHandler
*load_state
,
552 SaveVMHandlers
*ops
= g_new0(SaveVMHandlers
, 1);
553 ops
->save_state
= save_state
;
554 ops
->load_state
= load_state
;
555 return register_savevm_live(dev
, idstr
, instance_id
, version_id
,
559 void unregister_savevm(DeviceState
*dev
, const char *idstr
, void *opaque
)
561 SaveStateEntry
*se
, *new_se
;
565 char *path
= qdev_get_dev_path(dev
);
567 pstrcpy(id
, sizeof(id
), path
);
568 pstrcat(id
, sizeof(id
), "/");
572 pstrcat(id
, sizeof(id
), idstr
);
574 QTAILQ_FOREACH_SAFE(se
, &savevm_state
.handlers
, entry
, new_se
) {
575 if (strcmp(se
->idstr
, id
) == 0 && se
->opaque
== opaque
) {
576 QTAILQ_REMOVE(&savevm_state
.handlers
, se
, entry
);
584 int vmstate_register_with_alias_id(DeviceState
*dev
, int instance_id
,
585 const VMStateDescription
*vmsd
,
586 void *opaque
, int alias_id
,
587 int required_for_version
)
591 /* If this triggers, alias support can be dropped for the vmsd. */
592 assert(alias_id
== -1 || required_for_version
>= vmsd
->minimum_version_id
);
594 se
= g_new0(SaveStateEntry
, 1);
595 se
->version_id
= vmsd
->version_id
;
596 se
->section_id
= savevm_state
.global_section_id
++;
599 se
->alias_id
= alias_id
;
602 char *id
= qdev_get_dev_path(dev
);
604 pstrcpy(se
->idstr
, sizeof(se
->idstr
), id
);
605 pstrcat(se
->idstr
, sizeof(se
->idstr
), "/");
608 se
->compat
= g_new0(CompatEntry
, 1);
609 pstrcpy(se
->compat
->idstr
, sizeof(se
->compat
->idstr
), vmsd
->name
);
610 se
->compat
->instance_id
= instance_id
== -1 ?
611 calculate_compat_instance_id(vmsd
->name
) : instance_id
;
615 pstrcat(se
->idstr
, sizeof(se
->idstr
), vmsd
->name
);
617 if (instance_id
== -1) {
618 se
->instance_id
= calculate_new_instance_id(se
->idstr
);
620 se
->instance_id
= instance_id
;
622 assert(!se
->compat
|| se
->instance_id
== 0);
623 /* add at the end of list */
624 QTAILQ_INSERT_TAIL(&savevm_state
.handlers
, se
, entry
);
628 void vmstate_unregister(DeviceState
*dev
, const VMStateDescription
*vmsd
,
631 SaveStateEntry
*se
, *new_se
;
633 QTAILQ_FOREACH_SAFE(se
, &savevm_state
.handlers
, entry
, new_se
) {
634 if (se
->vmsd
== vmsd
&& se
->opaque
== opaque
) {
635 QTAILQ_REMOVE(&savevm_state
.handlers
, se
, entry
);
642 static int vmstate_load(QEMUFile
*f
, SaveStateEntry
*se
, int version_id
)
644 trace_vmstate_load(se
->idstr
, se
->vmsd
? se
->vmsd
->name
: "(old)");
645 if (!se
->vmsd
) { /* Old style */
646 return se
->ops
->load_state(f
, se
->opaque
, version_id
);
648 return vmstate_load_state(f
, se
->vmsd
, se
->opaque
, version_id
);
651 static void vmstate_save_old_style(QEMUFile
*f
, SaveStateEntry
*se
, QJSON
*vmdesc
)
653 int64_t old_offset
, size
;
655 old_offset
= qemu_ftell_fast(f
);
656 se
->ops
->save_state(f
, se
->opaque
);
657 size
= qemu_ftell_fast(f
) - old_offset
;
660 json_prop_int(vmdesc
, "size", size
);
661 json_start_array(vmdesc
, "fields");
662 json_start_object(vmdesc
, NULL
);
663 json_prop_str(vmdesc
, "name", "data");
664 json_prop_int(vmdesc
, "size", size
);
665 json_prop_str(vmdesc
, "type", "buffer");
666 json_end_object(vmdesc
);
667 json_end_array(vmdesc
);
671 static void vmstate_save(QEMUFile
*f
, SaveStateEntry
*se
, QJSON
*vmdesc
)
673 trace_vmstate_save(se
->idstr
, se
->vmsd
? se
->vmsd
->name
: "(old)");
675 vmstate_save_old_style(f
, se
, vmdesc
);
678 vmstate_save_state(f
, se
->vmsd
, se
->opaque
, vmdesc
);
681 void savevm_skip_section_footers(void)
683 skip_section_footers
= true;
687 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
689 static void save_section_header(QEMUFile
*f
, SaveStateEntry
*se
,
690 uint8_t section_type
)
692 qemu_put_byte(f
, section_type
);
693 qemu_put_be32(f
, se
->section_id
);
695 if (section_type
== QEMU_VM_SECTION_FULL
||
696 section_type
== QEMU_VM_SECTION_START
) {
698 size_t len
= strlen(se
->idstr
);
699 qemu_put_byte(f
, len
);
700 qemu_put_buffer(f
, (uint8_t *)se
->idstr
, len
);
702 qemu_put_be32(f
, se
->instance_id
);
703 qemu_put_be32(f
, se
->version_id
);
708 * Write a footer onto device sections that catches cases misformatted device
711 static void save_section_footer(QEMUFile
*f
, SaveStateEntry
*se
)
713 if (!skip_section_footers
) {
714 qemu_put_byte(f
, QEMU_VM_SECTION_FOOTER
);
715 qemu_put_be32(f
, se
->section_id
);
720 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
721 * command and associated data.
723 * @f: File to send command on
724 * @command: Command type to send
725 * @len: Length of associated data
726 * @data: Data associated with command.
728 void qemu_savevm_command_send(QEMUFile
*f
,
729 enum qemu_vm_cmd command
,
733 trace_savevm_command_send(command
, len
);
734 qemu_put_byte(f
, QEMU_VM_COMMAND
);
735 qemu_put_be16(f
, (uint16_t)command
);
736 qemu_put_be16(f
, len
);
737 qemu_put_buffer(f
, data
, len
);
741 void qemu_savevm_send_ping(QEMUFile
*f
, uint32_t value
)
745 trace_savevm_send_ping(value
);
746 buf
= cpu_to_be32(value
);
747 qemu_savevm_command_send(f
, MIG_CMD_PING
, sizeof(value
), (uint8_t *)&buf
);
750 void qemu_savevm_send_open_return_path(QEMUFile
*f
)
752 trace_savevm_send_open_return_path();
753 qemu_savevm_command_send(f
, MIG_CMD_OPEN_RETURN_PATH
, 0, NULL
);
756 /* We have a buffer of data to send; we don't want that all to be loaded
757 * by the command itself, so the command contains just the length of the
758 * extra buffer that we then send straight after it.
759 * TODO: Must be a better way to organise that
765 int qemu_savevm_send_packaged(QEMUFile
*f
, const uint8_t *buf
, size_t len
)
769 if (len
> MAX_VM_CMD_PACKAGED_SIZE
) {
770 error_report("%s: Unreasonably large packaged state: %zu",
775 tmp
= cpu_to_be32(len
);
777 trace_qemu_savevm_send_packaged();
778 qemu_savevm_command_send(f
, MIG_CMD_PACKAGED
, 4, (uint8_t *)&tmp
);
780 qemu_put_buffer(f
, buf
, len
);
785 /* Send prior to any postcopy transfer */
786 void qemu_savevm_send_postcopy_advise(QEMUFile
*f
)
789 tmp
[0] = cpu_to_be64(getpagesize());
790 tmp
[1] = cpu_to_be64(1ul << qemu_target_page_bits());
792 trace_qemu_savevm_send_postcopy_advise();
793 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_ADVISE
, 16, (uint8_t *)tmp
);
796 /* Sent prior to starting the destination running in postcopy, discard pages
797 * that have already been sent but redirtied on the source.
798 * CMD_POSTCOPY_RAM_DISCARD consist of:
800 * byte Length of name field (not including 0)
801 * n x byte RAM block name
802 * byte 0 terminator (just for safety)
803 * n x Byte ranges within the named RAMBlock
804 * be64 Start of the range
807 * name: RAMBlock name that these entries are part of
808 * len: Number of page entries
809 * start_list: 'len' addresses
810 * length_list: 'len' addresses
813 void qemu_savevm_send_postcopy_ram_discard(QEMUFile
*f
, const char *name
,
815 uint64_t *start_list
,
816 uint64_t *length_list
)
821 size_t name_len
= strlen(name
);
823 trace_qemu_savevm_send_postcopy_ram_discard(name
, len
);
824 assert(name_len
< 256);
825 buf
= g_malloc0(1 + 1 + name_len
+ 1 + (8 + 8) * len
);
826 buf
[0] = postcopy_ram_discard_version
;
828 memcpy(buf
+ 2, name
, name_len
);
829 tmplen
= 2 + name_len
;
830 buf
[tmplen
++] = '\0';
832 for (t
= 0; t
< len
; t
++) {
833 cpu_to_be64w((uint64_t *)(buf
+ tmplen
), start_list
[t
]);
835 cpu_to_be64w((uint64_t *)(buf
+ tmplen
), length_list
[t
]);
838 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RAM_DISCARD
, tmplen
, buf
);
842 /* Get the destination into a state where it can receive postcopy data. */
843 void qemu_savevm_send_postcopy_listen(QEMUFile
*f
)
845 trace_savevm_send_postcopy_listen();
846 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_LISTEN
, 0, NULL
);
849 /* Kick the destination into running */
850 void qemu_savevm_send_postcopy_run(QEMUFile
*f
)
852 trace_savevm_send_postcopy_run();
853 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RUN
, 0, NULL
);
856 bool qemu_savevm_state_blocked(Error
**errp
)
860 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
861 if (se
->vmsd
&& se
->vmsd
->unmigratable
) {
862 error_setg(errp
, "State blocked by non-migratable device '%s'",
870 static bool enforce_config_section(void)
872 MachineState
*machine
= MACHINE(qdev_get_machine());
873 return machine
->enforce_config_section
;
876 void qemu_savevm_state_header(QEMUFile
*f
)
878 trace_savevm_state_header();
879 qemu_put_be32(f
, QEMU_VM_FILE_MAGIC
);
880 qemu_put_be32(f
, QEMU_VM_FILE_VERSION
);
882 if (!savevm_state
.skip_configuration
|| enforce_config_section()) {
883 qemu_put_byte(f
, QEMU_VM_CONFIGURATION
);
884 vmstate_save_state(f
, &vmstate_configuration
, &savevm_state
, 0);
889 void qemu_savevm_state_begin(QEMUFile
*f
,
890 const MigrationParams
*params
)
895 trace_savevm_state_begin();
896 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
897 if (!se
->ops
|| !se
->ops
->set_params
) {
900 se
->ops
->set_params(params
, se
->opaque
);
903 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
904 if (!se
->ops
|| !se
->ops
->save_live_setup
) {
907 if (se
->ops
&& se
->ops
->is_active
) {
908 if (!se
->ops
->is_active(se
->opaque
)) {
912 save_section_header(f
, se
, QEMU_VM_SECTION_START
);
914 ret
= se
->ops
->save_live_setup(f
, se
->opaque
);
915 save_section_footer(f
, se
);
917 qemu_file_set_error(f
, ret
);
924 * this function has three return values:
925 * negative: there was one error, and we have -errno.
926 * 0 : We haven't finished, caller have to go again
927 * 1 : We have finished, we can go to complete phase
929 int qemu_savevm_state_iterate(QEMUFile
*f
, bool postcopy
)
934 trace_savevm_state_iterate();
935 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
936 if (!se
->ops
|| !se
->ops
->save_live_iterate
) {
939 if (se
->ops
&& se
->ops
->is_active
) {
940 if (!se
->ops
->is_active(se
->opaque
)) {
945 * In the postcopy phase, any device that doesn't know how to
946 * do postcopy should have saved it's state in the _complete
947 * call that's already run, it might get confused if we call
948 * iterate afterwards.
950 if (postcopy
&& !se
->ops
->save_live_complete_postcopy
) {
953 if (qemu_file_rate_limit(f
)) {
956 trace_savevm_section_start(se
->idstr
, se
->section_id
);
958 save_section_header(f
, se
, QEMU_VM_SECTION_PART
);
960 ret
= se
->ops
->save_live_iterate(f
, se
->opaque
);
961 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
962 save_section_footer(f
, se
);
965 qemu_file_set_error(f
, ret
);
968 /* Do not proceed to the next vmstate before this one reported
969 completion of the current stage. This serializes the migration
970 and reduces the probability that a faster changing state is
971 synchronized over and over again. */
978 static bool should_send_vmdesc(void)
980 MachineState
*machine
= MACHINE(qdev_get_machine());
981 bool in_postcopy
= migration_in_postcopy(migrate_get_current());
982 return !machine
->suppress_vmdesc
&& !in_postcopy
;
986 * Calls the save_live_complete_postcopy methods
987 * causing the last few pages to be sent immediately and doing any associated
989 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
990 * all the other devices, but that happens at the point we switch to postcopy.
992 void qemu_savevm_state_complete_postcopy(QEMUFile
*f
)
997 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
998 if (!se
->ops
|| !se
->ops
->save_live_complete_postcopy
) {
1001 if (se
->ops
&& se
->ops
->is_active
) {
1002 if (!se
->ops
->is_active(se
->opaque
)) {
1006 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1008 qemu_put_byte(f
, QEMU_VM_SECTION_END
);
1009 qemu_put_be32(f
, se
->section_id
);
1011 ret
= se
->ops
->save_live_complete_postcopy(f
, se
->opaque
);
1012 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1013 save_section_footer(f
, se
);
1015 qemu_file_set_error(f
, ret
);
1020 qemu_put_byte(f
, QEMU_VM_EOF
);
1024 void qemu_savevm_state_complete_precopy(QEMUFile
*f
, bool iterable_only
)
1030 bool in_postcopy
= migration_in_postcopy(migrate_get_current());
1032 trace_savevm_state_complete_precopy();
1034 cpu_synchronize_all_states();
1036 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1038 (in_postcopy
&& se
->ops
->save_live_complete_postcopy
) ||
1039 (in_postcopy
&& !iterable_only
) ||
1040 !se
->ops
->save_live_complete_precopy
) {
1044 if (se
->ops
&& se
->ops
->is_active
) {
1045 if (!se
->ops
->is_active(se
->opaque
)) {
1049 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1051 save_section_header(f
, se
, QEMU_VM_SECTION_END
);
1053 ret
= se
->ops
->save_live_complete_precopy(f
, se
->opaque
);
1054 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1055 save_section_footer(f
, se
);
1057 qemu_file_set_error(f
, ret
);
1062 if (iterable_only
) {
1066 vmdesc
= qjson_new();
1067 json_prop_int(vmdesc
, "page_size", TARGET_PAGE_SIZE
);
1068 json_start_array(vmdesc
, "devices");
1069 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1071 if ((!se
->ops
|| !se
->ops
->save_state
) && !se
->vmsd
) {
1074 if (se
->vmsd
&& !vmstate_save_needed(se
->vmsd
, se
->opaque
)) {
1075 trace_savevm_section_skip(se
->idstr
, se
->section_id
);
1079 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1081 json_start_object(vmdesc
, NULL
);
1082 json_prop_str(vmdesc
, "name", se
->idstr
);
1083 json_prop_int(vmdesc
, "instance_id", se
->instance_id
);
1085 save_section_header(f
, se
, QEMU_VM_SECTION_FULL
);
1086 vmstate_save(f
, se
, vmdesc
);
1087 trace_savevm_section_end(se
->idstr
, se
->section_id
, 0);
1088 save_section_footer(f
, se
);
1090 json_end_object(vmdesc
);
1094 /* Postcopy stream will still be going */
1095 qemu_put_byte(f
, QEMU_VM_EOF
);
1098 json_end_array(vmdesc
);
1099 qjson_finish(vmdesc
);
1100 vmdesc_len
= strlen(qjson_get_str(vmdesc
));
1102 if (should_send_vmdesc()) {
1103 qemu_put_byte(f
, QEMU_VM_VMDESCRIPTION
);
1104 qemu_put_be32(f
, vmdesc_len
);
1105 qemu_put_buffer(f
, (uint8_t *)qjson_get_str(vmdesc
), vmdesc_len
);
1107 qjson_destroy(vmdesc
);
1112 /* Give an estimate of the amount left to be transferred,
1113 * the result is split into the amount for units that can and
1114 * for units that can't do postcopy.
1116 void qemu_savevm_state_pending(QEMUFile
*f
, uint64_t max_size
,
1117 uint64_t *res_non_postcopiable
,
1118 uint64_t *res_postcopiable
)
1122 *res_non_postcopiable
= 0;
1123 *res_postcopiable
= 0;
1126 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1127 if (!se
->ops
|| !se
->ops
->save_live_pending
) {
1130 if (se
->ops
&& se
->ops
->is_active
) {
1131 if (!se
->ops
->is_active(se
->opaque
)) {
1135 se
->ops
->save_live_pending(f
, se
->opaque
, max_size
,
1136 res_non_postcopiable
, res_postcopiable
);
1140 void qemu_savevm_state_cleanup(void)
1144 trace_savevm_state_cleanup();
1145 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1146 if (se
->ops
&& se
->ops
->cleanup
) {
1147 se
->ops
->cleanup(se
->opaque
);
1152 static int qemu_savevm_state(QEMUFile
*f
, Error
**errp
)
1155 MigrationParams params
= {
1159 MigrationState
*ms
= migrate_init(¶ms
);
1160 ms
->to_dst_file
= f
;
1162 if (migration_is_blocked(errp
)) {
1166 qemu_mutex_unlock_iothread();
1167 qemu_savevm_state_header(f
);
1168 qemu_savevm_state_begin(f
, ¶ms
);
1169 qemu_mutex_lock_iothread();
1171 while (qemu_file_get_error(f
) == 0) {
1172 if (qemu_savevm_state_iterate(f
, false) > 0) {
1177 ret
= qemu_file_get_error(f
);
1179 qemu_savevm_state_complete_precopy(f
, false);
1180 ret
= qemu_file_get_error(f
);
1182 qemu_savevm_state_cleanup();
1184 error_setg_errno(errp
, -ret
, "Error while writing VM state");
1189 static int qemu_save_device_state(QEMUFile
*f
)
1193 qemu_put_be32(f
, QEMU_VM_FILE_MAGIC
);
1194 qemu_put_be32(f
, QEMU_VM_FILE_VERSION
);
1196 cpu_synchronize_all_states();
1198 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1202 if ((!se
->ops
|| !se
->ops
->save_state
) && !se
->vmsd
) {
1205 if (se
->vmsd
&& !vmstate_save_needed(se
->vmsd
, se
->opaque
)) {
1209 save_section_header(f
, se
, QEMU_VM_SECTION_FULL
);
1211 vmstate_save(f
, se
, NULL
);
1213 save_section_footer(f
, se
);
1216 qemu_put_byte(f
, QEMU_VM_EOF
);
1218 return qemu_file_get_error(f
);
1221 static SaveStateEntry
*find_se(const char *idstr
, int instance_id
)
1225 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1226 if (!strcmp(se
->idstr
, idstr
) &&
1227 (instance_id
== se
->instance_id
||
1228 instance_id
== se
->alias_id
))
1230 /* Migrating from an older version? */
1231 if (strstr(se
->idstr
, idstr
) && se
->compat
) {
1232 if (!strcmp(se
->compat
->idstr
, idstr
) &&
1233 (instance_id
== se
->compat
->instance_id
||
1234 instance_id
== se
->alias_id
))
1241 enum LoadVMExitCodes
{
1242 /* Allow a command to quit all layers of nested loadvm loops */
1246 static int qemu_loadvm_state_main(QEMUFile
*f
, MigrationIncomingState
*mis
);
1248 /* ------ incoming postcopy messages ------ */
1249 /* 'advise' arrives before any transfers just to tell us that a postcopy
1250 * *might* happen - it might be skipped if precopy transferred everything
1253 static int loadvm_postcopy_handle_advise(MigrationIncomingState
*mis
)
1255 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_ADVISE
);
1256 uint64_t remote_hps
, remote_tps
;
1258 trace_loadvm_postcopy_handle_advise();
1259 if (ps
!= POSTCOPY_INCOMING_NONE
) {
1260 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps
);
1264 if (!postcopy_ram_supported_by_host()) {
1268 remote_hps
= qemu_get_be64(mis
->from_src_file
);
1269 if (remote_hps
!= getpagesize()) {
1271 * Some combinations of mismatch are probably possible but it gets
1272 * a bit more complicated. In particular we need to place whole
1273 * host pages on the dest at once, and we need to ensure that we
1274 * handle dirtying to make sure we never end up sending part of
1275 * a hostpage on it's own.
1277 error_report("Postcopy needs matching host page sizes (s=%d d=%d)",
1278 (int)remote_hps
, getpagesize());
1282 remote_tps
= qemu_get_be64(mis
->from_src_file
);
1283 if (remote_tps
!= (1ul << qemu_target_page_bits())) {
1285 * Again, some differences could be dealt with, but for now keep it
1288 error_report("Postcopy needs matching target page sizes (s=%d d=%d)",
1289 (int)remote_tps
, 1 << qemu_target_page_bits());
1293 if (ram_postcopy_incoming_init(mis
)) {
1297 postcopy_state_set(POSTCOPY_INCOMING_ADVISE
);
1302 /* After postcopy we will be told to throw some pages away since they're
1303 * dirty and will have to be demand fetched. Must happen before CPU is
1305 * There can be 0..many of these messages, each encoding multiple pages.
1307 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState
*mis
,
1312 PostcopyState ps
= postcopy_state_get();
1314 trace_loadvm_postcopy_ram_handle_discard();
1317 case POSTCOPY_INCOMING_ADVISE
:
1319 tmp
= postcopy_ram_prepare_discard(mis
);
1325 case POSTCOPY_INCOMING_DISCARD
:
1326 /* Expected state */
1330 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1334 /* We're expecting a
1336 * a RAM ID string (length byte, name, 0 term)
1337 * then at least 1 16 byte chunk
1339 if (len
< (1 + 1 + 1 + 1 + 2 * 8)) {
1340 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len
);
1344 tmp
= qemu_get_byte(mis
->from_src_file
);
1345 if (tmp
!= postcopy_ram_discard_version
) {
1346 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp
);
1350 if (!qemu_get_counted_string(mis
->from_src_file
, ramid
)) {
1351 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1354 tmp
= qemu_get_byte(mis
->from_src_file
);
1356 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp
);
1360 len
-= 3 + strlen(ramid
);
1362 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len
);
1365 trace_loadvm_postcopy_ram_handle_discard_header(ramid
, len
);
1367 uint64_t start_addr
, block_length
;
1368 start_addr
= qemu_get_be64(mis
->from_src_file
);
1369 block_length
= qemu_get_be64(mis
->from_src_file
);
1372 int ret
= ram_discard_range(mis
, ramid
, start_addr
,
1378 trace_loadvm_postcopy_ram_handle_discard_end();
1384 * Triggered by a postcopy_listen command; this thread takes over reading
1385 * the input stream, leaving the main thread free to carry on loading the rest
1386 * of the device state (from RAM).
1387 * (TODO:This could do with being in a postcopy file - but there again it's
1388 * just another input loop, not that postcopy specific)
1390 static void *postcopy_ram_listen_thread(void *opaque
)
1392 QEMUFile
*f
= opaque
;
1393 MigrationIncomingState
*mis
= migration_incoming_get_current();
1396 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
1397 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1398 qemu_sem_post(&mis
->listen_thread_sem
);
1399 trace_postcopy_ram_listen_thread_start();
1402 * Because we're a thread and not a coroutine we can't yield
1403 * in qemu_file, and thus we must be blocking now.
1405 qemu_file_set_blocking(f
, true);
1406 load_res
= qemu_loadvm_state_main(f
, mis
);
1407 /* And non-blocking again so we don't block in any cleanup */
1408 qemu_file_set_blocking(f
, false);
1410 trace_postcopy_ram_listen_thread_exit();
1412 error_report("%s: loadvm failed: %d", __func__
, load_res
);
1413 qemu_file_set_error(f
, load_res
);
1414 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1415 MIGRATION_STATUS_FAILED
);
1418 * This looks good, but it's possible that the device loading in the
1419 * main thread hasn't finished yet, and so we might not be in 'RUN'
1420 * state yet; wait for the end of the main thread.
1422 qemu_event_wait(&mis
->main_thread_load_event
);
1424 postcopy_ram_incoming_cleanup(mis
);
1428 * If something went wrong then we have a bad state so exit;
1429 * depending how far we got it might be possible at this point
1430 * to leave the guest running and fire MCEs for pages that never
1431 * arrived as a desperate recovery step.
1436 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1437 MIGRATION_STATUS_COMPLETED
);
1439 * If everything has worked fine, then the main thread has waited
1440 * for us to start, and we're the last use of the mis.
1441 * (If something broke then qemu will have to exit anyway since it's
1442 * got a bad migration state).
1444 migration_incoming_state_destroy();
1450 /* After this message we must be able to immediately receive postcopy data */
1451 static int loadvm_postcopy_handle_listen(MigrationIncomingState
*mis
)
1453 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_LISTENING
);
1454 trace_loadvm_postcopy_handle_listen();
1455 if (ps
!= POSTCOPY_INCOMING_ADVISE
&& ps
!= POSTCOPY_INCOMING_DISCARD
) {
1456 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps
);
1459 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
1461 * A rare case, we entered listen without having to do any discards,
1462 * so do the setup that's normally done at the time of the 1st discard.
1464 postcopy_ram_prepare_discard(mis
);
1468 * Sensitise RAM - can now generate requests for blocks that don't exist
1469 * However, at this point the CPU shouldn't be running, and the IO
1470 * shouldn't be doing anything yet so don't actually expect requests
1472 if (postcopy_ram_enable_notify(mis
)) {
1476 if (mis
->have_listen_thread
) {
1477 error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread");
1481 mis
->have_listen_thread
= true;
1482 /* Start up the listening thread and wait for it to signal ready */
1483 qemu_sem_init(&mis
->listen_thread_sem
, 0);
1484 qemu_thread_create(&mis
->listen_thread
, "postcopy/listen",
1485 postcopy_ram_listen_thread
, mis
->from_src_file
,
1486 QEMU_THREAD_DETACHED
);
1487 qemu_sem_wait(&mis
->listen_thread_sem
);
1488 qemu_sem_destroy(&mis
->listen_thread_sem
);
1498 static void loadvm_postcopy_handle_run_bh(void *opaque
)
1500 Error
*local_err
= NULL
;
1501 HandleRunBhData
*data
= opaque
;
1503 /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1506 cpu_synchronize_all_post_init();
1508 qemu_announce_self();
1510 /* Make sure all file formats flush their mutable metadata */
1511 bdrv_invalidate_cache_all(&local_err
);
1513 error_report_err(local_err
);
1516 trace_loadvm_postcopy_handle_run_cpu_sync();
1517 cpu_synchronize_all_post_init();
1519 trace_loadvm_postcopy_handle_run_vmstart();
1522 /* Hold onto your hats, starting the CPU */
1525 /* leave it paused and let management decide when to start the CPU */
1526 runstate_set(RUN_STATE_PAUSED
);
1529 qemu_bh_delete(data
->bh
);
1533 /* After all discards we can start running and asking for pages */
1534 static int loadvm_postcopy_handle_run(MigrationIncomingState
*mis
)
1536 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_RUNNING
);
1537 HandleRunBhData
*data
;
1539 trace_loadvm_postcopy_handle_run();
1540 if (ps
!= POSTCOPY_INCOMING_LISTENING
) {
1541 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps
);
1545 data
= g_new(HandleRunBhData
, 1);
1546 data
->bh
= qemu_bh_new(loadvm_postcopy_handle_run_bh
, data
);
1547 qemu_bh_schedule(data
->bh
);
1549 /* We need to finish reading the stream from the package
1550 * and also stop reading anything more from the stream that loaded the
1551 * package (since it's now being read by the listener thread).
1552 * LOADVM_QUIT will quit all the layers of nested loadvm loops.
1558 * Immediately following this command is a blob of data containing an embedded
1559 * chunk of migration stream; read it and load it.
1561 * @mis: Incoming state
1562 * @length: Length of packaged data to read
1564 * Returns: Negative values on error
1567 static int loadvm_handle_cmd_packaged(MigrationIncomingState
*mis
)
1571 QIOChannelBuffer
*bioc
;
1573 length
= qemu_get_be32(mis
->from_src_file
);
1574 trace_loadvm_handle_cmd_packaged(length
);
1576 if (length
> MAX_VM_CMD_PACKAGED_SIZE
) {
1577 error_report("Unreasonably large packaged state: %zu", length
);
1581 bioc
= qio_channel_buffer_new(length
);
1582 ret
= qemu_get_buffer(mis
->from_src_file
,
1585 if (ret
!= length
) {
1586 object_unref(OBJECT(bioc
));
1587 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
1589 return (ret
< 0) ? ret
: -EAGAIN
;
1591 bioc
->usage
+= length
;
1592 trace_loadvm_handle_cmd_packaged_received(ret
);
1594 QEMUFile
*packf
= qemu_fopen_channel_input(QIO_CHANNEL(bioc
));
1596 ret
= qemu_loadvm_state_main(packf
, mis
);
1597 trace_loadvm_handle_cmd_packaged_main(ret
);
1599 object_unref(OBJECT(bioc
));
1605 * Process an incoming 'QEMU_VM_COMMAND'
1606 * 0 just a normal return
1607 * LOADVM_QUIT All good, but exit the loop
1610 static int loadvm_process_command(QEMUFile
*f
)
1612 MigrationIncomingState
*mis
= migration_incoming_get_current();
1617 cmd
= qemu_get_be16(f
);
1618 len
= qemu_get_be16(f
);
1620 trace_loadvm_process_command(cmd
, len
);
1621 if (cmd
>= MIG_CMD_MAX
|| cmd
== MIG_CMD_INVALID
) {
1622 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd
, len
);
1626 if (mig_cmd_args
[cmd
].len
!= -1 && mig_cmd_args
[cmd
].len
!= len
) {
1627 error_report("%s received with bad length - expecting %zu, got %d",
1628 mig_cmd_args
[cmd
].name
,
1629 (size_t)mig_cmd_args
[cmd
].len
, len
);
1634 case MIG_CMD_OPEN_RETURN_PATH
:
1635 if (mis
->to_src_file
) {
1636 error_report("CMD_OPEN_RETURN_PATH called when RP already open");
1637 /* Not really a problem, so don't give up */
1640 mis
->to_src_file
= qemu_file_get_return_path(f
);
1641 if (!mis
->to_src_file
) {
1642 error_report("CMD_OPEN_RETURN_PATH failed");
1648 tmp32
= qemu_get_be32(f
);
1649 trace_loadvm_process_command_ping(tmp32
);
1650 if (!mis
->to_src_file
) {
1651 error_report("CMD_PING (0x%x) received with no return path",
1655 migrate_send_rp_pong(mis
, tmp32
);
1658 case MIG_CMD_PACKAGED
:
1659 return loadvm_handle_cmd_packaged(mis
);
1661 case MIG_CMD_POSTCOPY_ADVISE
:
1662 return loadvm_postcopy_handle_advise(mis
);
1664 case MIG_CMD_POSTCOPY_LISTEN
:
1665 return loadvm_postcopy_handle_listen(mis
);
1667 case MIG_CMD_POSTCOPY_RUN
:
1668 return loadvm_postcopy_handle_run(mis
);
1670 case MIG_CMD_POSTCOPY_RAM_DISCARD
:
1671 return loadvm_postcopy_ram_handle_discard(mis
, len
);
1677 struct LoadStateEntry
{
1678 QLIST_ENTRY(LoadStateEntry
) entry
;
1685 * Read a footer off the wire and check that it matches the expected section
1687 * Returns: true if the footer was good
1688 * false if there is a problem (and calls error_report to say why)
1690 static bool check_section_footer(QEMUFile
*f
, LoadStateEntry
*le
)
1693 uint32_t read_section_id
;
1695 if (skip_section_footers
) {
1696 /* No footer to check */
1700 read_mark
= qemu_get_byte(f
);
1702 if (read_mark
!= QEMU_VM_SECTION_FOOTER
) {
1703 error_report("Missing section footer for %s", le
->se
->idstr
);
1707 read_section_id
= qemu_get_be32(f
);
1708 if (read_section_id
!= le
->section_id
) {
1709 error_report("Mismatched section id in footer for %s -"
1710 " read 0x%x expected 0x%x",
1711 le
->se
->idstr
, read_section_id
, le
->section_id
);
1719 void loadvm_free_handlers(MigrationIncomingState
*mis
)
1721 LoadStateEntry
*le
, *new_le
;
1723 QLIST_FOREACH_SAFE(le
, &mis
->loadvm_handlers
, entry
, new_le
) {
1724 QLIST_REMOVE(le
, entry
);
1730 qemu_loadvm_section_start_full(QEMUFile
*f
, MigrationIncomingState
*mis
)
1732 uint32_t instance_id
, version_id
, section_id
;
1738 /* Read section start */
1739 section_id
= qemu_get_be32(f
);
1740 if (!qemu_get_counted_string(f
, idstr
)) {
1741 error_report("Unable to read ID string for section %u",
1745 instance_id
= qemu_get_be32(f
);
1746 version_id
= qemu_get_be32(f
);
1748 trace_qemu_loadvm_state_section_startfull(section_id
, idstr
,
1749 instance_id
, version_id
);
1750 /* Find savevm section */
1751 se
= find_se(idstr
, instance_id
);
1753 error_report("Unknown savevm section or instance '%s' %d",
1754 idstr
, instance_id
);
1758 /* Validate version */
1759 if (version_id
> se
->version_id
) {
1760 error_report("savevm: unsupported version %d for '%s' v%d",
1761 version_id
, idstr
, se
->version_id
);
1766 le
= g_malloc0(sizeof(*le
));
1769 le
->section_id
= section_id
;
1770 le
->version_id
= version_id
;
1771 QLIST_INSERT_HEAD(&mis
->loadvm_handlers
, le
, entry
);
1773 ret
= vmstate_load(f
, le
->se
, le
->version_id
);
1775 error_report("error while loading state for instance 0x%x of"
1776 " device '%s'", instance_id
, idstr
);
1779 if (!check_section_footer(f
, le
)) {
1787 qemu_loadvm_section_part_end(QEMUFile
*f
, MigrationIncomingState
*mis
)
1789 uint32_t section_id
;
1793 section_id
= qemu_get_be32(f
);
1795 trace_qemu_loadvm_state_section_partend(section_id
);
1796 QLIST_FOREACH(le
, &mis
->loadvm_handlers
, entry
) {
1797 if (le
->section_id
== section_id
) {
1802 error_report("Unknown savevm section %d", section_id
);
1806 ret
= vmstate_load(f
, le
->se
, le
->version_id
);
1808 error_report("error while loading state section id %d(%s)",
1809 section_id
, le
->se
->idstr
);
1812 if (!check_section_footer(f
, le
)) {
1819 static int qemu_loadvm_state_main(QEMUFile
*f
, MigrationIncomingState
*mis
)
1821 uint8_t section_type
;
1824 while ((section_type
= qemu_get_byte(f
)) != QEMU_VM_EOF
) {
1826 trace_qemu_loadvm_state_section(section_type
);
1827 switch (section_type
) {
1828 case QEMU_VM_SECTION_START
:
1829 case QEMU_VM_SECTION_FULL
:
1830 ret
= qemu_loadvm_section_start_full(f
, mis
);
1835 case QEMU_VM_SECTION_PART
:
1836 case QEMU_VM_SECTION_END
:
1837 ret
= qemu_loadvm_section_part_end(f
, mis
);
1842 case QEMU_VM_COMMAND
:
1843 ret
= loadvm_process_command(f
);
1844 trace_qemu_loadvm_state_section_command(ret
);
1845 if ((ret
< 0) || (ret
& LOADVM_QUIT
)) {
1850 error_report("Unknown savevm section type %d", section_type
);
1858 int qemu_loadvm_state(QEMUFile
*f
)
1860 MigrationIncomingState
*mis
= migration_incoming_get_current();
1861 Error
*local_err
= NULL
;
1865 if (qemu_savevm_state_blocked(&local_err
)) {
1866 error_report_err(local_err
);
1870 v
= qemu_get_be32(f
);
1871 if (v
!= QEMU_VM_FILE_MAGIC
) {
1872 error_report("Not a migration stream");
1876 v
= qemu_get_be32(f
);
1877 if (v
== QEMU_VM_FILE_VERSION_COMPAT
) {
1878 error_report("SaveVM v2 format is obsolete and don't work anymore");
1881 if (v
!= QEMU_VM_FILE_VERSION
) {
1882 error_report("Unsupported migration stream version");
1886 if (!savevm_state
.skip_configuration
|| enforce_config_section()) {
1887 if (qemu_get_byte(f
) != QEMU_VM_CONFIGURATION
) {
1888 error_report("Configuration section missing");
1891 ret
= vmstate_load_state(f
, &vmstate_configuration
, &savevm_state
, 0);
1898 ret
= qemu_loadvm_state_main(f
, mis
);
1899 qemu_event_set(&mis
->main_thread_load_event
);
1901 trace_qemu_loadvm_state_post_main(ret
);
1903 if (mis
->have_listen_thread
) {
1904 /* Listen thread still going, can't clean up yet */
1909 ret
= qemu_file_get_error(f
);
1913 * Try to read in the VMDESC section as well, so that dumping tools that
1914 * intercept our migration stream have the chance to see it.
1917 /* We've got to be careful; if we don't read the data and just shut the fd
1918 * then the sender can error if we close while it's still sending.
1919 * We also mustn't read data that isn't there; some transports (RDMA)
1920 * will stall waiting for that data when the source has already closed.
1922 if (ret
== 0 && should_send_vmdesc()) {
1925 uint8_t section_type
= qemu_get_byte(f
);
1927 if (section_type
!= QEMU_VM_VMDESCRIPTION
) {
1928 error_report("Expected vmdescription section, but got %d",
1931 * It doesn't seem worth failing at this point since
1932 * we apparently have an otherwise valid VM state
1935 buf
= g_malloc(0x1000);
1936 size
= qemu_get_be32(f
);
1939 uint32_t read_chunk
= MIN(size
, 0x1000);
1940 qemu_get_buffer(f
, buf
, read_chunk
);
1947 cpu_synchronize_all_post_init();
1952 void hmp_savevm(Monitor
*mon
, const QDict
*qdict
)
1954 BlockDriverState
*bs
, *bs1
;
1955 QEMUSnapshotInfo sn1
, *sn
= &sn1
, old_sn1
, *old_sn
= &old_sn1
;
1958 int saved_vm_running
;
1959 uint64_t vm_state_size
;
1962 const char *name
= qdict_get_try_str(qdict
, "name");
1963 Error
*local_err
= NULL
;
1964 AioContext
*aio_context
;
1966 if (!bdrv_all_can_snapshot(&bs
)) {
1967 monitor_printf(mon
, "Device '%s' is writable but does not "
1968 "support snapshots.\n", bdrv_get_device_name(bs
));
1972 /* Delete old snapshots of the same name */
1973 if (name
&& bdrv_all_delete_snapshot(name
, &bs1
, &local_err
) < 0) {
1974 error_reportf_err(local_err
,
1975 "Error while deleting snapshot on device '%s': ",
1976 bdrv_get_device_name(bs1
));
1980 bs
= bdrv_all_find_vmstate_bs();
1982 monitor_printf(mon
, "No block device can accept snapshots\n");
1985 aio_context
= bdrv_get_aio_context(bs
);
1987 saved_vm_running
= runstate_is_running();
1989 ret
= global_state_store();
1991 monitor_printf(mon
, "Error saving global state\n");
1994 vm_stop(RUN_STATE_SAVE_VM
);
1996 aio_context_acquire(aio_context
);
1998 memset(sn
, 0, sizeof(*sn
));
2000 /* fill auxiliary fields */
2001 qemu_gettimeofday(&tv
);
2002 sn
->date_sec
= tv
.tv_sec
;
2003 sn
->date_nsec
= tv
.tv_usec
* 1000;
2004 sn
->vm_clock_nsec
= qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
);
2007 ret
= bdrv_snapshot_find(bs
, old_sn
, name
);
2009 pstrcpy(sn
->name
, sizeof(sn
->name
), old_sn
->name
);
2010 pstrcpy(sn
->id_str
, sizeof(sn
->id_str
), old_sn
->id_str
);
2012 pstrcpy(sn
->name
, sizeof(sn
->name
), name
);
2015 /* cast below needed for OpenBSD where tv_sec is still 'long' */
2016 localtime_r((const time_t *)&tv
.tv_sec
, &tm
);
2017 strftime(sn
->name
, sizeof(sn
->name
), "vm-%Y%m%d%H%M%S", &tm
);
2020 /* save the VM state */
2021 f
= qemu_fopen_bdrv(bs
, 1);
2023 monitor_printf(mon
, "Could not open VM state file\n");
2026 ret
= qemu_savevm_state(f
, &local_err
);
2027 vm_state_size
= qemu_ftell(f
);
2030 error_report_err(local_err
);
2034 ret
= bdrv_all_create_snapshot(sn
, bs
, vm_state_size
, &bs
);
2036 monitor_printf(mon
, "Error while creating snapshot on '%s'\n",
2037 bdrv_get_device_name(bs
));
2041 aio_context_release(aio_context
);
2042 if (saved_vm_running
) {
2047 void qmp_xen_save_devices_state(const char *filename
, Error
**errp
)
2050 QIOChannelFile
*ioc
;
2051 int saved_vm_running
;
2054 saved_vm_running
= runstate_is_running();
2055 vm_stop(RUN_STATE_SAVE_VM
);
2056 global_state_store_running();
2058 ioc
= qio_channel_file_new_path(filename
, O_WRONLY
| O_CREAT
, 0660, errp
);
2062 f
= qemu_fopen_channel_output(QIO_CHANNEL(ioc
));
2063 ret
= qemu_save_device_state(f
);
2066 error_setg(errp
, QERR_IO_ERROR
);
2070 if (saved_vm_running
) {
2075 int load_vmstate(const char *name
)
2077 BlockDriverState
*bs
, *bs_vm_state
;
2078 QEMUSnapshotInfo sn
;
2081 AioContext
*aio_context
;
2083 if (!bdrv_all_can_snapshot(&bs
)) {
2084 error_report("Device '%s' is writable but does not support snapshots.",
2085 bdrv_get_device_name(bs
));
2088 ret
= bdrv_all_find_snapshot(name
, &bs
);
2090 error_report("Device '%s' does not have the requested snapshot '%s'",
2091 bdrv_get_device_name(bs
), name
);
2095 bs_vm_state
= bdrv_all_find_vmstate_bs();
2097 error_report("No block device supports snapshots");
2100 aio_context
= bdrv_get_aio_context(bs_vm_state
);
2102 /* Don't even try to load empty VM states */
2103 aio_context_acquire(aio_context
);
2104 ret
= bdrv_snapshot_find(bs_vm_state
, &sn
, name
);
2105 aio_context_release(aio_context
);
2108 } else if (sn
.vm_state_size
== 0) {
2109 error_report("This is a disk-only snapshot. Revert to it offline "
2114 /* Flush all IO requests so they don't interfere with the new state. */
2117 ret
= bdrv_all_goto_snapshot(name
, &bs
);
2119 error_report("Error %d while activating snapshot '%s' on '%s'",
2120 ret
, name
, bdrv_get_device_name(bs
));
2124 /* restore the VM state */
2125 f
= qemu_fopen_bdrv(bs_vm_state
, 0);
2127 error_report("Could not open VM state file");
2131 qemu_system_reset(VMRESET_SILENT
);
2132 migration_incoming_state_new(f
);
2134 aio_context_acquire(aio_context
);
2135 ret
= qemu_loadvm_state(f
);
2137 aio_context_release(aio_context
);
2139 migration_incoming_state_destroy();
2141 error_report("Error %d while loading VM state", ret
);
2148 void hmp_delvm(Monitor
*mon
, const QDict
*qdict
)
2150 BlockDriverState
*bs
;
2152 const char *name
= qdict_get_str(qdict
, "name");
2154 if (bdrv_all_delete_snapshot(name
, &bs
, &err
) < 0) {
2155 error_reportf_err(err
,
2156 "Error while deleting snapshot on device '%s': ",
2157 bdrv_get_device_name(bs
));
2161 void hmp_info_snapshots(Monitor
*mon
, const QDict
*qdict
)
2163 BlockDriverState
*bs
, *bs1
;
2164 QEMUSnapshotInfo
*sn_tab
, *sn
;
2167 int *available_snapshots
;
2168 AioContext
*aio_context
;
2170 bs
= bdrv_all_find_vmstate_bs();
2172 monitor_printf(mon
, "No available block device supports snapshots\n");
2175 aio_context
= bdrv_get_aio_context(bs
);
2177 aio_context_acquire(aio_context
);
2178 nb_sns
= bdrv_snapshot_list(bs
, &sn_tab
);
2179 aio_context_release(aio_context
);
2182 monitor_printf(mon
, "bdrv_snapshot_list: error %d\n", nb_sns
);
2187 monitor_printf(mon
, "There is no snapshot available.\n");
2191 available_snapshots
= g_new0(int, nb_sns
);
2193 for (i
= 0; i
< nb_sns
; i
++) {
2194 if (bdrv_all_find_snapshot(sn_tab
[i
].id_str
, &bs1
) == 0) {
2195 available_snapshots
[total
] = i
;
2201 bdrv_snapshot_dump((fprintf_function
)monitor_printf
, mon
, NULL
);
2202 monitor_printf(mon
, "\n");
2203 for (i
= 0; i
< total
; i
++) {
2204 sn
= &sn_tab
[available_snapshots
[i
]];
2205 bdrv_snapshot_dump((fprintf_function
)monitor_printf
, mon
, sn
);
2206 monitor_printf(mon
, "\n");
2209 monitor_printf(mon
, "There is no suitable snapshot available\n");
2213 g_free(available_snapshots
);
2217 void vmstate_register_ram(MemoryRegion
*mr
, DeviceState
*dev
)
2219 qemu_ram_set_idstr(mr
->ram_block
,
2220 memory_region_name(mr
), dev
);
2223 void vmstate_unregister_ram(MemoryRegion
*mr
, DeviceState
*dev
)
2225 qemu_ram_unset_idstr(mr
->ram_block
);
2228 void vmstate_register_ram_global(MemoryRegion
*mr
)
2230 vmstate_register_ram(mr
, NULL
);