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 "monitor/monitor.h"
37 #include "sysemu/sysemu.h"
38 #include "qemu/timer.h"
39 #include "audio/audio.h"
40 #include "migration/migration.h"
41 #include "migration/postcopy-ram.h"
42 #include "qapi/qmp/qerror.h"
43 #include "qemu/error-report.h"
44 #include "qemu/sockets.h"
45 #include "qemu/queue.h"
46 #include "sysemu/cpus.h"
47 #include "exec/memory.h"
48 #include "qmp-commands.h"
50 #include "qemu/bitops.h"
52 #include "block/snapshot.h"
53 #include "block/qapi.h"
54 #include "qemu/cutils.h"
55 #include "io/channel-buffer.h"
56 #include "io/channel-file.h"
59 #define ETH_P_RARP 0x8035
61 #define ARP_HTYPE_ETH 0x0001
62 #define ARP_PTYPE_IP 0x0800
63 #define ARP_OP_REQUEST_REV 0x3
65 const unsigned int postcopy_ram_discard_version
= 0;
67 static bool skip_section_footers
;
69 static struct mig_cmd_args
{
70 ssize_t len
; /* -1 = variable */
73 [MIG_CMD_INVALID
] = { .len
= -1, .name
= "INVALID" },
74 [MIG_CMD_OPEN_RETURN_PATH
] = { .len
= 0, .name
= "OPEN_RETURN_PATH" },
75 [MIG_CMD_PING
] = { .len
= sizeof(uint32_t), .name
= "PING" },
76 [MIG_CMD_POSTCOPY_ADVISE
] = { .len
= 16, .name
= "POSTCOPY_ADVISE" },
77 [MIG_CMD_POSTCOPY_LISTEN
] = { .len
= 0, .name
= "POSTCOPY_LISTEN" },
78 [MIG_CMD_POSTCOPY_RUN
] = { .len
= 0, .name
= "POSTCOPY_RUN" },
79 [MIG_CMD_POSTCOPY_RAM_DISCARD
] = {
80 .len
= -1, .name
= "POSTCOPY_RAM_DISCARD" },
81 [MIG_CMD_PACKAGED
] = { .len
= 4, .name
= "PACKAGED" },
82 [MIG_CMD_MAX
] = { .len
= -1, .name
= "MAX" },
85 static int announce_self_create(uint8_t *buf
,
88 /* Ethernet header. */
89 memset(buf
, 0xff, 6); /* destination MAC addr */
90 memcpy(buf
+ 6, mac_addr
, 6); /* source MAC addr */
91 *(uint16_t *)(buf
+ 12) = htons(ETH_P_RARP
); /* ethertype */
94 *(uint16_t *)(buf
+ 14) = htons(ARP_HTYPE_ETH
); /* hardware addr space */
95 *(uint16_t *)(buf
+ 16) = htons(ARP_PTYPE_IP
); /* protocol addr space */
96 *(buf
+ 18) = 6; /* hardware addr length (ethernet) */
97 *(buf
+ 19) = 4; /* protocol addr length (IPv4) */
98 *(uint16_t *)(buf
+ 20) = htons(ARP_OP_REQUEST_REV
); /* opcode */
99 memcpy(buf
+ 22, mac_addr
, 6); /* source hw addr */
100 memset(buf
+ 28, 0x00, 4); /* source protocol addr */
101 memcpy(buf
+ 32, mac_addr
, 6); /* target hw addr */
102 memset(buf
+ 38, 0x00, 4); /* target protocol addr */
104 /* Padding to get up to 60 bytes (ethernet min packet size, minus FCS). */
105 memset(buf
+ 42, 0x00, 18);
107 return 60; /* len (FCS will be added by hardware) */
110 static void qemu_announce_self_iter(NICState
*nic
, void *opaque
)
115 trace_qemu_announce_self_iter(qemu_ether_ntoa(&nic
->conf
->macaddr
));
116 len
= announce_self_create(buf
, nic
->conf
->macaddr
.a
);
118 qemu_send_packet_raw(qemu_get_queue(nic
), buf
, len
);
122 static void qemu_announce_self_once(void *opaque
)
124 static int count
= SELF_ANNOUNCE_ROUNDS
;
125 QEMUTimer
*timer
= *(QEMUTimer
**)opaque
;
127 qemu_foreach_nic(qemu_announce_self_iter
, NULL
);
130 /* delay 50ms, 150ms, 250ms, ... */
131 timer_mod(timer
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) +
132 self_announce_delay(count
));
139 void qemu_announce_self(void)
141 static QEMUTimer
*timer
;
142 timer
= timer_new_ms(QEMU_CLOCK_REALTIME
, qemu_announce_self_once
, &timer
);
143 qemu_announce_self_once(&timer
);
146 /***********************************************************/
147 /* savevm/loadvm support */
149 static ssize_t
block_writev_buffer(void *opaque
, struct iovec
*iov
, int iovcnt
,
155 qemu_iovec_init_external(&qiov
, iov
, iovcnt
);
156 ret
= bdrv_writev_vmstate(opaque
, &qiov
, pos
);
164 static ssize_t
block_get_buffer(void *opaque
, uint8_t *buf
, int64_t pos
,
167 return bdrv_load_vmstate(opaque
, buf
, pos
, size
);
170 static int bdrv_fclose(void *opaque
)
172 return bdrv_flush(opaque
);
175 static const QEMUFileOps bdrv_read_ops
= {
176 .get_buffer
= block_get_buffer
,
180 static const QEMUFileOps bdrv_write_ops
= {
181 .writev_buffer
= block_writev_buffer
,
185 static QEMUFile
*qemu_fopen_bdrv(BlockDriverState
*bs
, int is_writable
)
188 return qemu_fopen_ops(bs
, &bdrv_write_ops
);
190 return qemu_fopen_ops(bs
, &bdrv_read_ops
);
194 /* QEMUFile timer support.
195 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
198 void timer_put(QEMUFile
*f
, QEMUTimer
*ts
)
200 uint64_t expire_time
;
202 expire_time
= timer_expire_time_ns(ts
);
203 qemu_put_be64(f
, expire_time
);
206 void timer_get(QEMUFile
*f
, QEMUTimer
*ts
)
208 uint64_t expire_time
;
210 expire_time
= qemu_get_be64(f
);
211 if (expire_time
!= -1) {
212 timer_mod_ns(ts
, expire_time
);
219 /* VMState timer support.
220 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
223 static int get_timer(QEMUFile
*f
, void *pv
, size_t size
)
230 static void put_timer(QEMUFile
*f
, void *pv
, size_t size
)
236 const VMStateInfo vmstate_info_timer
= {
243 typedef struct CompatEntry
{
248 typedef struct SaveStateEntry
{
249 QTAILQ_ENTRY(SaveStateEntry
) entry
;
256 const VMStateDescription
*vmsd
;
262 typedef struct SaveState
{
263 QTAILQ_HEAD(, SaveStateEntry
) handlers
;
264 int global_section_id
;
265 bool skip_configuration
;
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
;
291 static int configuration_post_load(void *opaque
, int version_id
)
293 SaveState
*state
= opaque
;
294 const char *current_name
= MACHINE_GET_CLASS(current_machine
)->name
;
296 if (strncmp(state
->name
, current_name
, state
->len
) != 0) {
297 error_report("Machine type received is '%.*s' and local is '%s'",
298 (int) state
->len
, state
->name
, current_name
);
304 static const VMStateDescription vmstate_configuration
= {
305 .name
= "configuration",
307 .post_load
= configuration_post_load
,
308 .pre_save
= configuration_pre_save
,
309 .fields
= (VMStateField
[]) {
310 VMSTATE_UINT32(len
, SaveState
),
311 VMSTATE_VBUFFER_ALLOC_UINT32(name
, SaveState
, 0, NULL
, 0, len
),
312 VMSTATE_END_OF_LIST()
316 static void dump_vmstate_vmsd(FILE *out_file
,
317 const VMStateDescription
*vmsd
, int indent
,
320 static void dump_vmstate_vmsf(FILE *out_file
, const VMStateField
*field
,
323 fprintf(out_file
, "%*s{\n", indent
, "");
325 fprintf(out_file
, "%*s\"field\": \"%s\",\n", indent
, "", field
->name
);
326 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
328 fprintf(out_file
, "%*s\"field_exists\": %s,\n", indent
, "",
329 field
->field_exists
? "true" : "false");
330 fprintf(out_file
, "%*s\"size\": %zu", indent
, "", field
->size
);
331 if (field
->vmsd
!= NULL
) {
332 fprintf(out_file
, ",\n");
333 dump_vmstate_vmsd(out_file
, field
->vmsd
, indent
, false);
335 fprintf(out_file
, "\n%*s}", indent
- 2, "");
338 static void dump_vmstate_vmss(FILE *out_file
,
339 const VMStateDescription
**subsection
,
342 if (*subsection
!= NULL
) {
343 dump_vmstate_vmsd(out_file
, *subsection
, indent
, true);
347 static void dump_vmstate_vmsd(FILE *out_file
,
348 const VMStateDescription
*vmsd
, int indent
,
352 fprintf(out_file
, "%*s{\n", indent
, "");
354 fprintf(out_file
, "%*s\"%s\": {\n", indent
, "", "Description");
357 fprintf(out_file
, "%*s\"name\": \"%s\",\n", indent
, "", vmsd
->name
);
358 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
360 fprintf(out_file
, "%*s\"minimum_version_id\": %d", indent
, "",
361 vmsd
->minimum_version_id
);
362 if (vmsd
->fields
!= NULL
) {
363 const VMStateField
*field
= vmsd
->fields
;
366 fprintf(out_file
, ",\n%*s\"Fields\": [\n", indent
, "");
368 while (field
->name
!= NULL
) {
369 if (field
->flags
& VMS_MUST_EXIST
) {
370 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
375 fprintf(out_file
, ",\n");
377 dump_vmstate_vmsf(out_file
, field
, indent
+ 2);
381 fprintf(out_file
, "\n%*s]", indent
, "");
383 if (vmsd
->subsections
!= NULL
) {
384 const VMStateDescription
**subsection
= vmsd
->subsections
;
387 fprintf(out_file
, ",\n%*s\"Subsections\": [\n", indent
, "");
389 while (*subsection
!= NULL
) {
391 fprintf(out_file
, ",\n");
393 dump_vmstate_vmss(out_file
, subsection
, indent
+ 2);
397 fprintf(out_file
, "\n%*s]", indent
, "");
399 fprintf(out_file
, "\n%*s}", indent
- 2, "");
402 static void dump_machine_type(FILE *out_file
)
406 mc
= MACHINE_GET_CLASS(current_machine
);
408 fprintf(out_file
, " \"vmschkmachine\": {\n");
409 fprintf(out_file
, " \"Name\": \"%s\"\n", mc
->name
);
410 fprintf(out_file
, " },\n");
413 void dump_vmstate_json_to_file(FILE *out_file
)
418 fprintf(out_file
, "{\n");
419 dump_machine_type(out_file
);
422 list
= object_class_get_list(TYPE_DEVICE
, true);
423 for (elt
= list
; elt
; elt
= elt
->next
) {
424 DeviceClass
*dc
= OBJECT_CLASS_CHECK(DeviceClass
, elt
->data
,
434 fprintf(out_file
, ",\n");
436 name
= object_class_get_name(OBJECT_CLASS(dc
));
437 fprintf(out_file
, "%*s\"%s\": {\n", indent
, "", name
);
439 fprintf(out_file
, "%*s\"Name\": \"%s\",\n", indent
, "", name
);
440 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
441 dc
->vmsd
->version_id
);
442 fprintf(out_file
, "%*s\"minimum_version_id\": %d,\n", indent
, "",
443 dc
->vmsd
->minimum_version_id
);
445 dump_vmstate_vmsd(out_file
, dc
->vmsd
, indent
, false);
447 fprintf(out_file
, "\n%*s}", indent
- 2, "");
450 fprintf(out_file
, "\n}\n");
454 static int calculate_new_instance_id(const char *idstr
)
459 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
460 if (strcmp(idstr
, se
->idstr
) == 0
461 && instance_id
<= se
->instance_id
) {
462 instance_id
= se
->instance_id
+ 1;
468 static int calculate_compat_instance_id(const char *idstr
)
473 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
478 if (strcmp(idstr
, se
->compat
->idstr
) == 0
479 && instance_id
<= se
->compat
->instance_id
) {
480 instance_id
= se
->compat
->instance_id
+ 1;
486 /* TODO: Individual devices generally have very little idea about the rest
487 of the system, so instance_id should be removed/replaced.
488 Meanwhile pass -1 as instance_id if you do not already have a clearly
489 distinguishing id for all instances of your device class. */
490 int register_savevm_live(DeviceState
*dev
,
499 se
= g_new0(SaveStateEntry
, 1);
500 se
->version_id
= version_id
;
501 se
->section_id
= savevm_state
.global_section_id
++;
505 /* if this is a live_savem then set is_ram */
506 if (ops
->save_live_setup
!= NULL
) {
511 char *id
= qdev_get_dev_path(dev
);
513 pstrcpy(se
->idstr
, sizeof(se
->idstr
), id
);
514 pstrcat(se
->idstr
, sizeof(se
->idstr
), "/");
517 se
->compat
= g_new0(CompatEntry
, 1);
518 pstrcpy(se
->compat
->idstr
, sizeof(se
->compat
->idstr
), idstr
);
519 se
->compat
->instance_id
= instance_id
== -1 ?
520 calculate_compat_instance_id(idstr
) : instance_id
;
524 pstrcat(se
->idstr
, sizeof(se
->idstr
), idstr
);
526 if (instance_id
== -1) {
527 se
->instance_id
= calculate_new_instance_id(se
->idstr
);
529 se
->instance_id
= instance_id
;
531 assert(!se
->compat
|| se
->instance_id
== 0);
532 /* add at the end of list */
533 QTAILQ_INSERT_TAIL(&savevm_state
.handlers
, se
, entry
);
537 int register_savevm(DeviceState
*dev
,
541 SaveStateHandler
*save_state
,
542 LoadStateHandler
*load_state
,
545 SaveVMHandlers
*ops
= g_new0(SaveVMHandlers
, 1);
546 ops
->save_state
= save_state
;
547 ops
->load_state
= load_state
;
548 return register_savevm_live(dev
, idstr
, instance_id
, version_id
,
552 void unregister_savevm(DeviceState
*dev
, const char *idstr
, void *opaque
)
554 SaveStateEntry
*se
, *new_se
;
558 char *path
= qdev_get_dev_path(dev
);
560 pstrcpy(id
, sizeof(id
), path
);
561 pstrcat(id
, sizeof(id
), "/");
565 pstrcat(id
, sizeof(id
), idstr
);
567 QTAILQ_FOREACH_SAFE(se
, &savevm_state
.handlers
, entry
, new_se
) {
568 if (strcmp(se
->idstr
, id
) == 0 && se
->opaque
== opaque
) {
569 QTAILQ_REMOVE(&savevm_state
.handlers
, se
, entry
);
577 int vmstate_register_with_alias_id(DeviceState
*dev
, int instance_id
,
578 const VMStateDescription
*vmsd
,
579 void *opaque
, int alias_id
,
580 int required_for_version
)
584 /* If this triggers, alias support can be dropped for the vmsd. */
585 assert(alias_id
== -1 || required_for_version
>= vmsd
->minimum_version_id
);
587 se
= g_new0(SaveStateEntry
, 1);
588 se
->version_id
= vmsd
->version_id
;
589 se
->section_id
= savevm_state
.global_section_id
++;
592 se
->alias_id
= alias_id
;
595 char *id
= qdev_get_dev_path(dev
);
597 pstrcpy(se
->idstr
, sizeof(se
->idstr
), id
);
598 pstrcat(se
->idstr
, sizeof(se
->idstr
), "/");
601 se
->compat
= g_new0(CompatEntry
, 1);
602 pstrcpy(se
->compat
->idstr
, sizeof(se
->compat
->idstr
), vmsd
->name
);
603 se
->compat
->instance_id
= instance_id
== -1 ?
604 calculate_compat_instance_id(vmsd
->name
) : instance_id
;
608 pstrcat(se
->idstr
, sizeof(se
->idstr
), vmsd
->name
);
610 if (instance_id
== -1) {
611 se
->instance_id
= calculate_new_instance_id(se
->idstr
);
613 se
->instance_id
= instance_id
;
615 assert(!se
->compat
|| se
->instance_id
== 0);
616 /* add at the end of list */
617 QTAILQ_INSERT_TAIL(&savevm_state
.handlers
, se
, entry
);
621 void vmstate_unregister(DeviceState
*dev
, const VMStateDescription
*vmsd
,
624 SaveStateEntry
*se
, *new_se
;
626 QTAILQ_FOREACH_SAFE(se
, &savevm_state
.handlers
, entry
, new_se
) {
627 if (se
->vmsd
== vmsd
&& se
->opaque
== opaque
) {
628 QTAILQ_REMOVE(&savevm_state
.handlers
, se
, entry
);
635 static int vmstate_load(QEMUFile
*f
, SaveStateEntry
*se
, int version_id
)
637 trace_vmstate_load(se
->idstr
, se
->vmsd
? se
->vmsd
->name
: "(old)");
638 if (!se
->vmsd
) { /* Old style */
639 return se
->ops
->load_state(f
, se
->opaque
, version_id
);
641 return vmstate_load_state(f
, se
->vmsd
, se
->opaque
, version_id
);
644 static void vmstate_save_old_style(QEMUFile
*f
, SaveStateEntry
*se
, QJSON
*vmdesc
)
646 int64_t old_offset
, size
;
648 old_offset
= qemu_ftell_fast(f
);
649 se
->ops
->save_state(f
, se
->opaque
);
650 size
= qemu_ftell_fast(f
) - old_offset
;
653 json_prop_int(vmdesc
, "size", size
);
654 json_start_array(vmdesc
, "fields");
655 json_start_object(vmdesc
, NULL
);
656 json_prop_str(vmdesc
, "name", "data");
657 json_prop_int(vmdesc
, "size", size
);
658 json_prop_str(vmdesc
, "type", "buffer");
659 json_end_object(vmdesc
);
660 json_end_array(vmdesc
);
664 static void vmstate_save(QEMUFile
*f
, SaveStateEntry
*se
, QJSON
*vmdesc
)
666 trace_vmstate_save(se
->idstr
, se
->vmsd
? se
->vmsd
->name
: "(old)");
668 vmstate_save_old_style(f
, se
, vmdesc
);
671 vmstate_save_state(f
, se
->vmsd
, se
->opaque
, vmdesc
);
674 void savevm_skip_section_footers(void)
676 skip_section_footers
= true;
680 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
682 static void save_section_header(QEMUFile
*f
, SaveStateEntry
*se
,
683 uint8_t section_type
)
685 qemu_put_byte(f
, section_type
);
686 qemu_put_be32(f
, se
->section_id
);
688 if (section_type
== QEMU_VM_SECTION_FULL
||
689 section_type
== QEMU_VM_SECTION_START
) {
691 size_t len
= strlen(se
->idstr
);
692 qemu_put_byte(f
, len
);
693 qemu_put_buffer(f
, (uint8_t *)se
->idstr
, len
);
695 qemu_put_be32(f
, se
->instance_id
);
696 qemu_put_be32(f
, se
->version_id
);
701 * Write a footer onto device sections that catches cases misformatted device
704 static void save_section_footer(QEMUFile
*f
, SaveStateEntry
*se
)
706 if (!skip_section_footers
) {
707 qemu_put_byte(f
, QEMU_VM_SECTION_FOOTER
);
708 qemu_put_be32(f
, se
->section_id
);
713 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
714 * command and associated data.
716 * @f: File to send command on
717 * @command: Command type to send
718 * @len: Length of associated data
719 * @data: Data associated with command.
721 void qemu_savevm_command_send(QEMUFile
*f
,
722 enum qemu_vm_cmd command
,
726 trace_savevm_command_send(command
, len
);
727 qemu_put_byte(f
, QEMU_VM_COMMAND
);
728 qemu_put_be16(f
, (uint16_t)command
);
729 qemu_put_be16(f
, len
);
730 qemu_put_buffer(f
, data
, len
);
734 void qemu_savevm_send_ping(QEMUFile
*f
, uint32_t value
)
738 trace_savevm_send_ping(value
);
739 buf
= cpu_to_be32(value
);
740 qemu_savevm_command_send(f
, MIG_CMD_PING
, sizeof(value
), (uint8_t *)&buf
);
743 void qemu_savevm_send_open_return_path(QEMUFile
*f
)
745 trace_savevm_send_open_return_path();
746 qemu_savevm_command_send(f
, MIG_CMD_OPEN_RETURN_PATH
, 0, NULL
);
749 /* We have a buffer of data to send; we don't want that all to be loaded
750 * by the command itself, so the command contains just the length of the
751 * extra buffer that we then send straight after it.
752 * TODO: Must be a better way to organise that
758 int qemu_savevm_send_packaged(QEMUFile
*f
, const uint8_t *buf
, size_t len
)
762 if (len
> MAX_VM_CMD_PACKAGED_SIZE
) {
763 error_report("%s: Unreasonably large packaged state: %zu",
768 tmp
= cpu_to_be32(len
);
770 trace_qemu_savevm_send_packaged();
771 qemu_savevm_command_send(f
, MIG_CMD_PACKAGED
, 4, (uint8_t *)&tmp
);
773 qemu_put_buffer(f
, buf
, len
);
778 /* Send prior to any postcopy transfer */
779 void qemu_savevm_send_postcopy_advise(QEMUFile
*f
)
782 tmp
[0] = cpu_to_be64(getpagesize());
783 tmp
[1] = cpu_to_be64(1ul << qemu_target_page_bits());
785 trace_qemu_savevm_send_postcopy_advise();
786 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_ADVISE
, 16, (uint8_t *)tmp
);
789 /* Sent prior to starting the destination running in postcopy, discard pages
790 * that have already been sent but redirtied on the source.
791 * CMD_POSTCOPY_RAM_DISCARD consist of:
793 * byte Length of name field (not including 0)
794 * n x byte RAM block name
795 * byte 0 terminator (just for safety)
796 * n x Byte ranges within the named RAMBlock
797 * be64 Start of the range
800 * name: RAMBlock name that these entries are part of
801 * len: Number of page entries
802 * start_list: 'len' addresses
803 * length_list: 'len' addresses
806 void qemu_savevm_send_postcopy_ram_discard(QEMUFile
*f
, const char *name
,
808 uint64_t *start_list
,
809 uint64_t *length_list
)
814 size_t name_len
= strlen(name
);
816 trace_qemu_savevm_send_postcopy_ram_discard(name
, len
);
817 assert(name_len
< 256);
818 buf
= g_malloc0(1 + 1 + name_len
+ 1 + (8 + 8) * len
);
819 buf
[0] = postcopy_ram_discard_version
;
821 memcpy(buf
+ 2, name
, name_len
);
822 tmplen
= 2 + name_len
;
823 buf
[tmplen
++] = '\0';
825 for (t
= 0; t
< len
; t
++) {
826 stq_be_p(buf
+ tmplen
, start_list
[t
]);
828 stq_be_p(buf
+ tmplen
, length_list
[t
]);
831 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RAM_DISCARD
, tmplen
, buf
);
835 /* Get the destination into a state where it can receive postcopy data. */
836 void qemu_savevm_send_postcopy_listen(QEMUFile
*f
)
838 trace_savevm_send_postcopy_listen();
839 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_LISTEN
, 0, NULL
);
842 /* Kick the destination into running */
843 void qemu_savevm_send_postcopy_run(QEMUFile
*f
)
845 trace_savevm_send_postcopy_run();
846 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RUN
, 0, NULL
);
849 bool qemu_savevm_state_blocked(Error
**errp
)
853 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
854 if (se
->vmsd
&& se
->vmsd
->unmigratable
) {
855 error_setg(errp
, "State blocked by non-migratable device '%s'",
863 static bool enforce_config_section(void)
865 MachineState
*machine
= MACHINE(qdev_get_machine());
866 return machine
->enforce_config_section
;
869 void qemu_savevm_state_header(QEMUFile
*f
)
871 trace_savevm_state_header();
872 qemu_put_be32(f
, QEMU_VM_FILE_MAGIC
);
873 qemu_put_be32(f
, QEMU_VM_FILE_VERSION
);
875 if (!savevm_state
.skip_configuration
|| enforce_config_section()) {
876 qemu_put_byte(f
, QEMU_VM_CONFIGURATION
);
877 vmstate_save_state(f
, &vmstate_configuration
, &savevm_state
, 0);
882 void qemu_savevm_state_begin(QEMUFile
*f
,
883 const MigrationParams
*params
)
888 trace_savevm_state_begin();
889 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
890 if (!se
->ops
|| !se
->ops
->set_params
) {
893 se
->ops
->set_params(params
, se
->opaque
);
896 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
897 if (!se
->ops
|| !se
->ops
->save_live_setup
) {
900 if (se
->ops
&& se
->ops
->is_active
) {
901 if (!se
->ops
->is_active(se
->opaque
)) {
905 save_section_header(f
, se
, QEMU_VM_SECTION_START
);
907 ret
= se
->ops
->save_live_setup(f
, se
->opaque
);
908 save_section_footer(f
, se
);
910 qemu_file_set_error(f
, ret
);
917 * this function has three return values:
918 * negative: there was one error, and we have -errno.
919 * 0 : We haven't finished, caller have to go again
920 * 1 : We have finished, we can go to complete phase
922 int qemu_savevm_state_iterate(QEMUFile
*f
, bool postcopy
)
927 trace_savevm_state_iterate();
928 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
929 if (!se
->ops
|| !se
->ops
->save_live_iterate
) {
932 if (se
->ops
&& se
->ops
->is_active
) {
933 if (!se
->ops
->is_active(se
->opaque
)) {
938 * In the postcopy phase, any device that doesn't know how to
939 * do postcopy should have saved it's state in the _complete
940 * call that's already run, it might get confused if we call
941 * iterate afterwards.
943 if (postcopy
&& !se
->ops
->save_live_complete_postcopy
) {
946 if (qemu_file_rate_limit(f
)) {
949 trace_savevm_section_start(se
->idstr
, se
->section_id
);
951 save_section_header(f
, se
, QEMU_VM_SECTION_PART
);
953 ret
= se
->ops
->save_live_iterate(f
, se
->opaque
);
954 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
955 save_section_footer(f
, se
);
958 qemu_file_set_error(f
, ret
);
961 /* Do not proceed to the next vmstate before this one reported
962 completion of the current stage. This serializes the migration
963 and reduces the probability that a faster changing state is
964 synchronized over and over again. */
971 static bool should_send_vmdesc(void)
973 MachineState
*machine
= MACHINE(qdev_get_machine());
974 bool in_postcopy
= migration_in_postcopy(migrate_get_current());
975 return !machine
->suppress_vmdesc
&& !in_postcopy
;
979 * Calls the save_live_complete_postcopy methods
980 * causing the last few pages to be sent immediately and doing any associated
982 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
983 * all the other devices, but that happens at the point we switch to postcopy.
985 void qemu_savevm_state_complete_postcopy(QEMUFile
*f
)
990 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
991 if (!se
->ops
|| !se
->ops
->save_live_complete_postcopy
) {
994 if (se
->ops
&& se
->ops
->is_active
) {
995 if (!se
->ops
->is_active(se
->opaque
)) {
999 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1001 qemu_put_byte(f
, QEMU_VM_SECTION_END
);
1002 qemu_put_be32(f
, se
->section_id
);
1004 ret
= se
->ops
->save_live_complete_postcopy(f
, se
->opaque
);
1005 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1006 save_section_footer(f
, se
);
1008 qemu_file_set_error(f
, ret
);
1013 qemu_put_byte(f
, QEMU_VM_EOF
);
1017 void qemu_savevm_state_complete_precopy(QEMUFile
*f
, bool iterable_only
)
1023 bool in_postcopy
= migration_in_postcopy(migrate_get_current());
1025 trace_savevm_state_complete_precopy();
1027 cpu_synchronize_all_states();
1029 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1031 (in_postcopy
&& se
->ops
->save_live_complete_postcopy
) ||
1032 (in_postcopy
&& !iterable_only
) ||
1033 !se
->ops
->save_live_complete_precopy
) {
1037 if (se
->ops
&& se
->ops
->is_active
) {
1038 if (!se
->ops
->is_active(se
->opaque
)) {
1042 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1044 save_section_header(f
, se
, QEMU_VM_SECTION_END
);
1046 ret
= se
->ops
->save_live_complete_precopy(f
, se
->opaque
);
1047 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1048 save_section_footer(f
, se
);
1050 qemu_file_set_error(f
, ret
);
1055 if (iterable_only
) {
1059 vmdesc
= qjson_new();
1060 json_prop_int(vmdesc
, "page_size", TARGET_PAGE_SIZE
);
1061 json_start_array(vmdesc
, "devices");
1062 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1064 if ((!se
->ops
|| !se
->ops
->save_state
) && !se
->vmsd
) {
1067 if (se
->vmsd
&& !vmstate_save_needed(se
->vmsd
, se
->opaque
)) {
1068 trace_savevm_section_skip(se
->idstr
, se
->section_id
);
1072 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1074 json_start_object(vmdesc
, NULL
);
1075 json_prop_str(vmdesc
, "name", se
->idstr
);
1076 json_prop_int(vmdesc
, "instance_id", se
->instance_id
);
1078 save_section_header(f
, se
, QEMU_VM_SECTION_FULL
);
1079 vmstate_save(f
, se
, vmdesc
);
1080 trace_savevm_section_end(se
->idstr
, se
->section_id
, 0);
1081 save_section_footer(f
, se
);
1083 json_end_object(vmdesc
);
1087 /* Postcopy stream will still be going */
1088 qemu_put_byte(f
, QEMU_VM_EOF
);
1091 json_end_array(vmdesc
);
1092 qjson_finish(vmdesc
);
1093 vmdesc_len
= strlen(qjson_get_str(vmdesc
));
1095 if (should_send_vmdesc()) {
1096 qemu_put_byte(f
, QEMU_VM_VMDESCRIPTION
);
1097 qemu_put_be32(f
, vmdesc_len
);
1098 qemu_put_buffer(f
, (uint8_t *)qjson_get_str(vmdesc
), vmdesc_len
);
1100 qjson_destroy(vmdesc
);
1105 /* Give an estimate of the amount left to be transferred,
1106 * the result is split into the amount for units that can and
1107 * for units that can't do postcopy.
1109 void qemu_savevm_state_pending(QEMUFile
*f
, uint64_t max_size
,
1110 uint64_t *res_non_postcopiable
,
1111 uint64_t *res_postcopiable
)
1115 *res_non_postcopiable
= 0;
1116 *res_postcopiable
= 0;
1119 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1120 if (!se
->ops
|| !se
->ops
->save_live_pending
) {
1123 if (se
->ops
&& se
->ops
->is_active
) {
1124 if (!se
->ops
->is_active(se
->opaque
)) {
1128 se
->ops
->save_live_pending(f
, se
->opaque
, max_size
,
1129 res_non_postcopiable
, res_postcopiable
);
1133 void qemu_savevm_state_cleanup(void)
1137 trace_savevm_state_cleanup();
1138 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1139 if (se
->ops
&& se
->ops
->cleanup
) {
1140 se
->ops
->cleanup(se
->opaque
);
1145 static int qemu_savevm_state(QEMUFile
*f
, Error
**errp
)
1148 MigrationParams params
= {
1152 MigrationState
*ms
= migrate_init(¶ms
);
1153 MigrationStatus status
;
1154 ms
->to_dst_file
= f
;
1156 if (migration_is_blocked(errp
)) {
1161 qemu_mutex_unlock_iothread();
1162 qemu_savevm_state_header(f
);
1163 qemu_savevm_state_begin(f
, ¶ms
);
1164 qemu_mutex_lock_iothread();
1166 while (qemu_file_get_error(f
) == 0) {
1167 if (qemu_savevm_state_iterate(f
, false) > 0) {
1172 ret
= qemu_file_get_error(f
);
1174 qemu_savevm_state_complete_precopy(f
, false);
1175 ret
= qemu_file_get_error(f
);
1177 qemu_savevm_state_cleanup();
1179 error_setg_errno(errp
, -ret
, "Error while writing VM state");
1184 status
= MIGRATION_STATUS_FAILED
;
1186 status
= MIGRATION_STATUS_COMPLETED
;
1188 migrate_set_state(&ms
->state
, MIGRATION_STATUS_SETUP
, status
);
1192 static int qemu_save_device_state(QEMUFile
*f
)
1196 qemu_put_be32(f
, QEMU_VM_FILE_MAGIC
);
1197 qemu_put_be32(f
, QEMU_VM_FILE_VERSION
);
1199 cpu_synchronize_all_states();
1201 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1205 if ((!se
->ops
|| !se
->ops
->save_state
) && !se
->vmsd
) {
1208 if (se
->vmsd
&& !vmstate_save_needed(se
->vmsd
, se
->opaque
)) {
1212 save_section_header(f
, se
, QEMU_VM_SECTION_FULL
);
1214 vmstate_save(f
, se
, NULL
);
1216 save_section_footer(f
, se
);
1219 qemu_put_byte(f
, QEMU_VM_EOF
);
1221 return qemu_file_get_error(f
);
1224 static SaveStateEntry
*find_se(const char *idstr
, int instance_id
)
1228 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1229 if (!strcmp(se
->idstr
, idstr
) &&
1230 (instance_id
== se
->instance_id
||
1231 instance_id
== se
->alias_id
))
1233 /* Migrating from an older version? */
1234 if (strstr(se
->idstr
, idstr
) && se
->compat
) {
1235 if (!strcmp(se
->compat
->idstr
, idstr
) &&
1236 (instance_id
== se
->compat
->instance_id
||
1237 instance_id
== se
->alias_id
))
1244 enum LoadVMExitCodes
{
1245 /* Allow a command to quit all layers of nested loadvm loops */
1249 static int qemu_loadvm_state_main(QEMUFile
*f
, MigrationIncomingState
*mis
);
1251 /* ------ incoming postcopy messages ------ */
1252 /* 'advise' arrives before any transfers just to tell us that a postcopy
1253 * *might* happen - it might be skipped if precopy transferred everything
1256 static int loadvm_postcopy_handle_advise(MigrationIncomingState
*mis
)
1258 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_ADVISE
);
1259 uint64_t remote_hps
, remote_tps
;
1261 trace_loadvm_postcopy_handle_advise();
1262 if (ps
!= POSTCOPY_INCOMING_NONE
) {
1263 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps
);
1267 if (!postcopy_ram_supported_by_host()) {
1271 remote_hps
= qemu_get_be64(mis
->from_src_file
);
1272 if (remote_hps
!= getpagesize()) {
1274 * Some combinations of mismatch are probably possible but it gets
1275 * a bit more complicated. In particular we need to place whole
1276 * host pages on the dest at once, and we need to ensure that we
1277 * handle dirtying to make sure we never end up sending part of
1278 * a hostpage on it's own.
1280 error_report("Postcopy needs matching host page sizes (s=%d d=%d)",
1281 (int)remote_hps
, getpagesize());
1285 remote_tps
= qemu_get_be64(mis
->from_src_file
);
1286 if (remote_tps
!= (1ul << qemu_target_page_bits())) {
1288 * Again, some differences could be dealt with, but for now keep it
1291 error_report("Postcopy needs matching target page sizes (s=%d d=%d)",
1292 (int)remote_tps
, 1 << qemu_target_page_bits());
1296 if (ram_postcopy_incoming_init(mis
)) {
1300 postcopy_state_set(POSTCOPY_INCOMING_ADVISE
);
1305 /* After postcopy we will be told to throw some pages away since they're
1306 * dirty and will have to be demand fetched. Must happen before CPU is
1308 * There can be 0..many of these messages, each encoding multiple pages.
1310 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState
*mis
,
1315 PostcopyState ps
= postcopy_state_get();
1317 trace_loadvm_postcopy_ram_handle_discard();
1320 case POSTCOPY_INCOMING_ADVISE
:
1322 tmp
= postcopy_ram_prepare_discard(mis
);
1328 case POSTCOPY_INCOMING_DISCARD
:
1329 /* Expected state */
1333 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1337 /* We're expecting a
1339 * a RAM ID string (length byte, name, 0 term)
1340 * then at least 1 16 byte chunk
1342 if (len
< (1 + 1 + 1 + 1 + 2 * 8)) {
1343 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len
);
1347 tmp
= qemu_get_byte(mis
->from_src_file
);
1348 if (tmp
!= postcopy_ram_discard_version
) {
1349 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp
);
1353 if (!qemu_get_counted_string(mis
->from_src_file
, ramid
)) {
1354 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1357 tmp
= qemu_get_byte(mis
->from_src_file
);
1359 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp
);
1363 len
-= 3 + strlen(ramid
);
1365 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len
);
1368 trace_loadvm_postcopy_ram_handle_discard_header(ramid
, len
);
1370 uint64_t start_addr
, block_length
;
1371 start_addr
= qemu_get_be64(mis
->from_src_file
);
1372 block_length
= qemu_get_be64(mis
->from_src_file
);
1375 int ret
= ram_discard_range(mis
, ramid
, start_addr
,
1381 trace_loadvm_postcopy_ram_handle_discard_end();
1387 * Triggered by a postcopy_listen command; this thread takes over reading
1388 * the input stream, leaving the main thread free to carry on loading the rest
1389 * of the device state (from RAM).
1390 * (TODO:This could do with being in a postcopy file - but there again it's
1391 * just another input loop, not that postcopy specific)
1393 static void *postcopy_ram_listen_thread(void *opaque
)
1395 QEMUFile
*f
= opaque
;
1396 MigrationIncomingState
*mis
= migration_incoming_get_current();
1399 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
1400 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1401 qemu_sem_post(&mis
->listen_thread_sem
);
1402 trace_postcopy_ram_listen_thread_start();
1405 * Because we're a thread and not a coroutine we can't yield
1406 * in qemu_file, and thus we must be blocking now.
1408 qemu_file_set_blocking(f
, true);
1409 load_res
= qemu_loadvm_state_main(f
, mis
);
1410 /* And non-blocking again so we don't block in any cleanup */
1411 qemu_file_set_blocking(f
, false);
1413 trace_postcopy_ram_listen_thread_exit();
1415 error_report("%s: loadvm failed: %d", __func__
, load_res
);
1416 qemu_file_set_error(f
, load_res
);
1417 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1418 MIGRATION_STATUS_FAILED
);
1421 * This looks good, but it's possible that the device loading in the
1422 * main thread hasn't finished yet, and so we might not be in 'RUN'
1423 * state yet; wait for the end of the main thread.
1425 qemu_event_wait(&mis
->main_thread_load_event
);
1427 postcopy_ram_incoming_cleanup(mis
);
1431 * If something went wrong then we have a bad state so exit;
1432 * depending how far we got it might be possible at this point
1433 * to leave the guest running and fire MCEs for pages that never
1434 * arrived as a desperate recovery step.
1439 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1440 MIGRATION_STATUS_COMPLETED
);
1442 * If everything has worked fine, then the main thread has waited
1443 * for us to start, and we're the last use of the mis.
1444 * (If something broke then qemu will have to exit anyway since it's
1445 * got a bad migration state).
1447 migration_incoming_state_destroy();
1453 /* After this message we must be able to immediately receive postcopy data */
1454 static int loadvm_postcopy_handle_listen(MigrationIncomingState
*mis
)
1456 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_LISTENING
);
1457 trace_loadvm_postcopy_handle_listen();
1458 if (ps
!= POSTCOPY_INCOMING_ADVISE
&& ps
!= POSTCOPY_INCOMING_DISCARD
) {
1459 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps
);
1462 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
1464 * A rare case, we entered listen without having to do any discards,
1465 * so do the setup that's normally done at the time of the 1st discard.
1467 postcopy_ram_prepare_discard(mis
);
1471 * Sensitise RAM - can now generate requests for blocks that don't exist
1472 * However, at this point the CPU shouldn't be running, and the IO
1473 * shouldn't be doing anything yet so don't actually expect requests
1475 if (postcopy_ram_enable_notify(mis
)) {
1479 if (mis
->have_listen_thread
) {
1480 error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread");
1484 mis
->have_listen_thread
= true;
1485 /* Start up the listening thread and wait for it to signal ready */
1486 qemu_sem_init(&mis
->listen_thread_sem
, 0);
1487 qemu_thread_create(&mis
->listen_thread
, "postcopy/listen",
1488 postcopy_ram_listen_thread
, mis
->from_src_file
,
1489 QEMU_THREAD_DETACHED
);
1490 qemu_sem_wait(&mis
->listen_thread_sem
);
1491 qemu_sem_destroy(&mis
->listen_thread_sem
);
1501 static void loadvm_postcopy_handle_run_bh(void *opaque
)
1503 Error
*local_err
= NULL
;
1504 HandleRunBhData
*data
= opaque
;
1506 /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1509 cpu_synchronize_all_post_init();
1511 qemu_announce_self();
1513 /* Make sure all file formats flush their mutable metadata */
1514 bdrv_invalidate_cache_all(&local_err
);
1516 error_report_err(local_err
);
1519 trace_loadvm_postcopy_handle_run_cpu_sync();
1520 cpu_synchronize_all_post_init();
1522 trace_loadvm_postcopy_handle_run_vmstart();
1525 /* Hold onto your hats, starting the CPU */
1528 /* leave it paused and let management decide when to start the CPU */
1529 runstate_set(RUN_STATE_PAUSED
);
1532 qemu_bh_delete(data
->bh
);
1536 /* After all discards we can start running and asking for pages */
1537 static int loadvm_postcopy_handle_run(MigrationIncomingState
*mis
)
1539 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_RUNNING
);
1540 HandleRunBhData
*data
;
1542 trace_loadvm_postcopy_handle_run();
1543 if (ps
!= POSTCOPY_INCOMING_LISTENING
) {
1544 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps
);
1548 data
= g_new(HandleRunBhData
, 1);
1549 data
->bh
= qemu_bh_new(loadvm_postcopy_handle_run_bh
, data
);
1550 qemu_bh_schedule(data
->bh
);
1552 /* We need to finish reading the stream from the package
1553 * and also stop reading anything more from the stream that loaded the
1554 * package (since it's now being read by the listener thread).
1555 * LOADVM_QUIT will quit all the layers of nested loadvm loops.
1561 * Immediately following this command is a blob of data containing an embedded
1562 * chunk of migration stream; read it and load it.
1564 * @mis: Incoming state
1565 * @length: Length of packaged data to read
1567 * Returns: Negative values on error
1570 static int loadvm_handle_cmd_packaged(MigrationIncomingState
*mis
)
1574 QIOChannelBuffer
*bioc
;
1576 length
= qemu_get_be32(mis
->from_src_file
);
1577 trace_loadvm_handle_cmd_packaged(length
);
1579 if (length
> MAX_VM_CMD_PACKAGED_SIZE
) {
1580 error_report("Unreasonably large packaged state: %zu", length
);
1584 bioc
= qio_channel_buffer_new(length
);
1585 ret
= qemu_get_buffer(mis
->from_src_file
,
1588 if (ret
!= length
) {
1589 object_unref(OBJECT(bioc
));
1590 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
1592 return (ret
< 0) ? ret
: -EAGAIN
;
1594 bioc
->usage
+= length
;
1595 trace_loadvm_handle_cmd_packaged_received(ret
);
1597 QEMUFile
*packf
= qemu_fopen_channel_input(QIO_CHANNEL(bioc
));
1599 ret
= qemu_loadvm_state_main(packf
, mis
);
1600 trace_loadvm_handle_cmd_packaged_main(ret
);
1602 object_unref(OBJECT(bioc
));
1608 * Process an incoming 'QEMU_VM_COMMAND'
1609 * 0 just a normal return
1610 * LOADVM_QUIT All good, but exit the loop
1613 static int loadvm_process_command(QEMUFile
*f
)
1615 MigrationIncomingState
*mis
= migration_incoming_get_current();
1620 cmd
= qemu_get_be16(f
);
1621 len
= qemu_get_be16(f
);
1623 trace_loadvm_process_command(cmd
, len
);
1624 if (cmd
>= MIG_CMD_MAX
|| cmd
== MIG_CMD_INVALID
) {
1625 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd
, len
);
1629 if (mig_cmd_args
[cmd
].len
!= -1 && mig_cmd_args
[cmd
].len
!= len
) {
1630 error_report("%s received with bad length - expecting %zu, got %d",
1631 mig_cmd_args
[cmd
].name
,
1632 (size_t)mig_cmd_args
[cmd
].len
, len
);
1637 case MIG_CMD_OPEN_RETURN_PATH
:
1638 if (mis
->to_src_file
) {
1639 error_report("CMD_OPEN_RETURN_PATH called when RP already open");
1640 /* Not really a problem, so don't give up */
1643 mis
->to_src_file
= qemu_file_get_return_path(f
);
1644 if (!mis
->to_src_file
) {
1645 error_report("CMD_OPEN_RETURN_PATH failed");
1651 tmp32
= qemu_get_be32(f
);
1652 trace_loadvm_process_command_ping(tmp32
);
1653 if (!mis
->to_src_file
) {
1654 error_report("CMD_PING (0x%x) received with no return path",
1658 migrate_send_rp_pong(mis
, tmp32
);
1661 case MIG_CMD_PACKAGED
:
1662 return loadvm_handle_cmd_packaged(mis
);
1664 case MIG_CMD_POSTCOPY_ADVISE
:
1665 return loadvm_postcopy_handle_advise(mis
);
1667 case MIG_CMD_POSTCOPY_LISTEN
:
1668 return loadvm_postcopy_handle_listen(mis
);
1670 case MIG_CMD_POSTCOPY_RUN
:
1671 return loadvm_postcopy_handle_run(mis
);
1673 case MIG_CMD_POSTCOPY_RAM_DISCARD
:
1674 return loadvm_postcopy_ram_handle_discard(mis
, len
);
1680 struct LoadStateEntry
{
1681 QLIST_ENTRY(LoadStateEntry
) entry
;
1688 * Read a footer off the wire and check that it matches the expected section
1690 * Returns: true if the footer was good
1691 * false if there is a problem (and calls error_report to say why)
1693 static bool check_section_footer(QEMUFile
*f
, LoadStateEntry
*le
)
1696 uint32_t read_section_id
;
1698 if (skip_section_footers
) {
1699 /* No footer to check */
1703 read_mark
= qemu_get_byte(f
);
1705 if (read_mark
!= QEMU_VM_SECTION_FOOTER
) {
1706 error_report("Missing section footer for %s", le
->se
->idstr
);
1710 read_section_id
= qemu_get_be32(f
);
1711 if (read_section_id
!= le
->section_id
) {
1712 error_report("Mismatched section id in footer for %s -"
1713 " read 0x%x expected 0x%x",
1714 le
->se
->idstr
, read_section_id
, le
->section_id
);
1722 void loadvm_free_handlers(MigrationIncomingState
*mis
)
1724 LoadStateEntry
*le
, *new_le
;
1726 QLIST_FOREACH_SAFE(le
, &mis
->loadvm_handlers
, entry
, new_le
) {
1727 QLIST_REMOVE(le
, entry
);
1733 qemu_loadvm_section_start_full(QEMUFile
*f
, MigrationIncomingState
*mis
)
1735 uint32_t instance_id
, version_id
, section_id
;
1741 /* Read section start */
1742 section_id
= qemu_get_be32(f
);
1743 if (!qemu_get_counted_string(f
, idstr
)) {
1744 error_report("Unable to read ID string for section %u",
1748 instance_id
= qemu_get_be32(f
);
1749 version_id
= qemu_get_be32(f
);
1751 trace_qemu_loadvm_state_section_startfull(section_id
, idstr
,
1752 instance_id
, version_id
);
1753 /* Find savevm section */
1754 se
= find_se(idstr
, instance_id
);
1756 error_report("Unknown savevm section or instance '%s' %d",
1757 idstr
, instance_id
);
1761 /* Validate version */
1762 if (version_id
> se
->version_id
) {
1763 error_report("savevm: unsupported version %d for '%s' v%d",
1764 version_id
, idstr
, se
->version_id
);
1768 /* Validate if it is a device's state */
1769 if (xen_enabled() && se
->is_ram
) {
1770 error_report("loadvm: %s RAM loading not allowed on Xen", idstr
);
1775 le
= g_malloc0(sizeof(*le
));
1778 le
->section_id
= section_id
;
1779 le
->version_id
= version_id
;
1780 QLIST_INSERT_HEAD(&mis
->loadvm_handlers
, le
, entry
);
1782 ret
= vmstate_load(f
, le
->se
, le
->version_id
);
1784 error_report("error while loading state for instance 0x%x of"
1785 " device '%s'", instance_id
, idstr
);
1788 if (!check_section_footer(f
, le
)) {
1796 qemu_loadvm_section_part_end(QEMUFile
*f
, MigrationIncomingState
*mis
)
1798 uint32_t section_id
;
1802 section_id
= qemu_get_be32(f
);
1804 trace_qemu_loadvm_state_section_partend(section_id
);
1805 QLIST_FOREACH(le
, &mis
->loadvm_handlers
, entry
) {
1806 if (le
->section_id
== section_id
) {
1811 error_report("Unknown savevm section %d", section_id
);
1815 ret
= vmstate_load(f
, le
->se
, le
->version_id
);
1817 error_report("error while loading state section id %d(%s)",
1818 section_id
, le
->se
->idstr
);
1821 if (!check_section_footer(f
, le
)) {
1828 static int qemu_loadvm_state_main(QEMUFile
*f
, MigrationIncomingState
*mis
)
1830 uint8_t section_type
;
1833 while ((section_type
= qemu_get_byte(f
)) != QEMU_VM_EOF
) {
1835 trace_qemu_loadvm_state_section(section_type
);
1836 switch (section_type
) {
1837 case QEMU_VM_SECTION_START
:
1838 case QEMU_VM_SECTION_FULL
:
1839 ret
= qemu_loadvm_section_start_full(f
, mis
);
1844 case QEMU_VM_SECTION_PART
:
1845 case QEMU_VM_SECTION_END
:
1846 ret
= qemu_loadvm_section_part_end(f
, mis
);
1851 case QEMU_VM_COMMAND
:
1852 ret
= loadvm_process_command(f
);
1853 trace_qemu_loadvm_state_section_command(ret
);
1854 if ((ret
< 0) || (ret
& LOADVM_QUIT
)) {
1859 error_report("Unknown savevm section type %d", section_type
);
1867 qemu_file_set_error(f
, ret
);
1872 int qemu_loadvm_state(QEMUFile
*f
)
1874 MigrationIncomingState
*mis
= migration_incoming_get_current();
1875 Error
*local_err
= NULL
;
1879 if (qemu_savevm_state_blocked(&local_err
)) {
1880 error_report_err(local_err
);
1884 v
= qemu_get_be32(f
);
1885 if (v
!= QEMU_VM_FILE_MAGIC
) {
1886 error_report("Not a migration stream");
1890 v
= qemu_get_be32(f
);
1891 if (v
== QEMU_VM_FILE_VERSION_COMPAT
) {
1892 error_report("SaveVM v2 format is obsolete and don't work anymore");
1895 if (v
!= QEMU_VM_FILE_VERSION
) {
1896 error_report("Unsupported migration stream version");
1900 if (!savevm_state
.skip_configuration
|| enforce_config_section()) {
1901 if (qemu_get_byte(f
) != QEMU_VM_CONFIGURATION
) {
1902 error_report("Configuration section missing");
1905 ret
= vmstate_load_state(f
, &vmstate_configuration
, &savevm_state
, 0);
1912 ret
= qemu_loadvm_state_main(f
, mis
);
1913 qemu_event_set(&mis
->main_thread_load_event
);
1915 trace_qemu_loadvm_state_post_main(ret
);
1917 if (mis
->have_listen_thread
) {
1918 /* Listen thread still going, can't clean up yet */
1923 ret
= qemu_file_get_error(f
);
1927 * Try to read in the VMDESC section as well, so that dumping tools that
1928 * intercept our migration stream have the chance to see it.
1931 /* We've got to be careful; if we don't read the data and just shut the fd
1932 * then the sender can error if we close while it's still sending.
1933 * We also mustn't read data that isn't there; some transports (RDMA)
1934 * will stall waiting for that data when the source has already closed.
1936 if (ret
== 0 && should_send_vmdesc()) {
1939 uint8_t section_type
= qemu_get_byte(f
);
1941 if (section_type
!= QEMU_VM_VMDESCRIPTION
) {
1942 error_report("Expected vmdescription section, but got %d",
1945 * It doesn't seem worth failing at this point since
1946 * we apparently have an otherwise valid VM state
1949 buf
= g_malloc(0x1000);
1950 size
= qemu_get_be32(f
);
1953 uint32_t read_chunk
= MIN(size
, 0x1000);
1954 qemu_get_buffer(f
, buf
, read_chunk
);
1961 cpu_synchronize_all_post_init();
1966 void hmp_savevm(Monitor
*mon
, const QDict
*qdict
)
1968 BlockDriverState
*bs
, *bs1
;
1969 QEMUSnapshotInfo sn1
, *sn
= &sn1
, old_sn1
, *old_sn
= &old_sn1
;
1972 int saved_vm_running
;
1973 uint64_t vm_state_size
;
1976 const char *name
= qdict_get_try_str(qdict
, "name");
1977 Error
*local_err
= NULL
;
1978 AioContext
*aio_context
;
1980 if (!bdrv_all_can_snapshot(&bs
)) {
1981 monitor_printf(mon
, "Device '%s' is writable but does not "
1982 "support snapshots.\n", bdrv_get_device_name(bs
));
1986 /* Delete old snapshots of the same name */
1987 if (name
&& bdrv_all_delete_snapshot(name
, &bs1
, &local_err
) < 0) {
1988 error_reportf_err(local_err
,
1989 "Error while deleting snapshot on device '%s': ",
1990 bdrv_get_device_name(bs1
));
1994 bs
= bdrv_all_find_vmstate_bs();
1996 monitor_printf(mon
, "No block device can accept snapshots\n");
1999 aio_context
= bdrv_get_aio_context(bs
);
2001 saved_vm_running
= runstate_is_running();
2003 ret
= global_state_store();
2005 monitor_printf(mon
, "Error saving global state\n");
2008 vm_stop(RUN_STATE_SAVE_VM
);
2010 aio_context_acquire(aio_context
);
2012 memset(sn
, 0, sizeof(*sn
));
2014 /* fill auxiliary fields */
2015 qemu_gettimeofday(&tv
);
2016 sn
->date_sec
= tv
.tv_sec
;
2017 sn
->date_nsec
= tv
.tv_usec
* 1000;
2018 sn
->vm_clock_nsec
= qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
);
2021 ret
= bdrv_snapshot_find(bs
, old_sn
, name
);
2023 pstrcpy(sn
->name
, sizeof(sn
->name
), old_sn
->name
);
2024 pstrcpy(sn
->id_str
, sizeof(sn
->id_str
), old_sn
->id_str
);
2026 pstrcpy(sn
->name
, sizeof(sn
->name
), name
);
2029 /* cast below needed for OpenBSD where tv_sec is still 'long' */
2030 localtime_r((const time_t *)&tv
.tv_sec
, &tm
);
2031 strftime(sn
->name
, sizeof(sn
->name
), "vm-%Y%m%d%H%M%S", &tm
);
2034 /* save the VM state */
2035 f
= qemu_fopen_bdrv(bs
, 1);
2037 monitor_printf(mon
, "Could not open VM state file\n");
2040 ret
= qemu_savevm_state(f
, &local_err
);
2041 vm_state_size
= qemu_ftell(f
);
2044 error_report_err(local_err
);
2048 ret
= bdrv_all_create_snapshot(sn
, bs
, vm_state_size
, &bs
);
2050 monitor_printf(mon
, "Error while creating snapshot on '%s'\n",
2051 bdrv_get_device_name(bs
));
2055 aio_context_release(aio_context
);
2056 if (saved_vm_running
) {
2061 void qmp_xen_save_devices_state(const char *filename
, Error
**errp
)
2064 QIOChannelFile
*ioc
;
2065 int saved_vm_running
;
2068 saved_vm_running
= runstate_is_running();
2069 vm_stop(RUN_STATE_SAVE_VM
);
2070 global_state_store_running();
2072 ioc
= qio_channel_file_new_path(filename
, O_WRONLY
| O_CREAT
, 0660, errp
);
2076 f
= qemu_fopen_channel_output(QIO_CHANNEL(ioc
));
2077 ret
= qemu_save_device_state(f
);
2080 error_setg(errp
, QERR_IO_ERROR
);
2084 if (saved_vm_running
) {
2089 void qmp_xen_load_devices_state(const char *filename
, Error
**errp
)
2092 QIOChannelFile
*ioc
;
2095 /* Guest must be paused before loading the device state; the RAM state
2096 * will already have been loaded by xc
2098 if (runstate_is_running()) {
2099 error_setg(errp
, "Cannot update device state while vm is running");
2102 vm_stop(RUN_STATE_RESTORE_VM
);
2104 ioc
= qio_channel_file_new_path(filename
, O_RDONLY
| O_BINARY
, 0, errp
);
2108 f
= qemu_fopen_channel_input(QIO_CHANNEL(ioc
));
2110 migration_incoming_state_new(f
);
2111 ret
= qemu_loadvm_state(f
);
2114 error_setg(errp
, QERR_IO_ERROR
);
2116 migration_incoming_state_destroy();
2119 int load_vmstate(const char *name
)
2121 BlockDriverState
*bs
, *bs_vm_state
;
2122 QEMUSnapshotInfo sn
;
2125 AioContext
*aio_context
;
2127 if (!bdrv_all_can_snapshot(&bs
)) {
2128 error_report("Device '%s' is writable but does not support snapshots.",
2129 bdrv_get_device_name(bs
));
2132 ret
= bdrv_all_find_snapshot(name
, &bs
);
2134 error_report("Device '%s' does not have the requested snapshot '%s'",
2135 bdrv_get_device_name(bs
), name
);
2139 bs_vm_state
= bdrv_all_find_vmstate_bs();
2141 error_report("No block device supports snapshots");
2144 aio_context
= bdrv_get_aio_context(bs_vm_state
);
2146 /* Don't even try to load empty VM states */
2147 aio_context_acquire(aio_context
);
2148 ret
= bdrv_snapshot_find(bs_vm_state
, &sn
, name
);
2149 aio_context_release(aio_context
);
2152 } else if (sn
.vm_state_size
== 0) {
2153 error_report("This is a disk-only snapshot. Revert to it offline "
2158 /* Flush all IO requests so they don't interfere with the new state. */
2161 ret
= bdrv_all_goto_snapshot(name
, &bs
);
2163 error_report("Error %d while activating snapshot '%s' on '%s'",
2164 ret
, name
, bdrv_get_device_name(bs
));
2168 /* restore the VM state */
2169 f
= qemu_fopen_bdrv(bs_vm_state
, 0);
2171 error_report("Could not open VM state file");
2175 qemu_system_reset(VMRESET_SILENT
);
2176 migration_incoming_state_new(f
);
2178 aio_context_acquire(aio_context
);
2179 ret
= qemu_loadvm_state(f
);
2181 aio_context_release(aio_context
);
2183 migration_incoming_state_destroy();
2185 error_report("Error %d while loading VM state", ret
);
2192 void hmp_delvm(Monitor
*mon
, const QDict
*qdict
)
2194 BlockDriverState
*bs
;
2196 const char *name
= qdict_get_str(qdict
, "name");
2198 if (bdrv_all_delete_snapshot(name
, &bs
, &err
) < 0) {
2199 error_reportf_err(err
,
2200 "Error while deleting snapshot on device '%s': ",
2201 bdrv_get_device_name(bs
));
2205 void hmp_info_snapshots(Monitor
*mon
, const QDict
*qdict
)
2207 BlockDriverState
*bs
, *bs1
;
2208 BdrvNextIterator it1
;
2209 QEMUSnapshotInfo
*sn_tab
, *sn
;
2210 bool no_snapshot
= true;
2213 int *global_snapshots
;
2214 AioContext
*aio_context
;
2216 typedef struct SnapshotEntry
{
2217 QEMUSnapshotInfo sn
;
2218 QTAILQ_ENTRY(SnapshotEntry
) next
;
2221 typedef struct ImageEntry
{
2222 const char *imagename
;
2223 QTAILQ_ENTRY(ImageEntry
) next
;
2224 QTAILQ_HEAD(, SnapshotEntry
) snapshots
;
2227 QTAILQ_HEAD(, ImageEntry
) image_list
=
2228 QTAILQ_HEAD_INITIALIZER(image_list
);
2230 ImageEntry
*image_entry
, *next_ie
;
2231 SnapshotEntry
*snapshot_entry
;
2233 bs
= bdrv_all_find_vmstate_bs();
2235 monitor_printf(mon
, "No available block device supports snapshots\n");
2238 aio_context
= bdrv_get_aio_context(bs
);
2240 aio_context_acquire(aio_context
);
2241 nb_sns
= bdrv_snapshot_list(bs
, &sn_tab
);
2242 aio_context_release(aio_context
);
2245 monitor_printf(mon
, "bdrv_snapshot_list: error %d\n", nb_sns
);
2249 for (bs1
= bdrv_first(&it1
); bs1
; bs1
= bdrv_next(&it1
)) {
2253 AioContext
*ctx
= bdrv_get_aio_context(bs1
);
2255 aio_context_acquire(ctx
);
2256 if (bdrv_can_snapshot(bs1
)) {
2258 bs1_nb_sns
= bdrv_snapshot_list(bs1
, &sn
);
2259 if (bs1_nb_sns
> 0) {
2260 no_snapshot
= false;
2261 ie
= g_new0(ImageEntry
, 1);
2262 ie
->imagename
= bdrv_get_device_name(bs1
);
2263 QTAILQ_INIT(&ie
->snapshots
);
2264 QTAILQ_INSERT_TAIL(&image_list
, ie
, next
);
2265 for (i
= 0; i
< bs1_nb_sns
; i
++) {
2266 se
= g_new0(SnapshotEntry
, 1);
2268 QTAILQ_INSERT_TAIL(&ie
->snapshots
, se
, next
);
2273 aio_context_release(ctx
);
2277 monitor_printf(mon
, "There is no snapshot available.\n");
2281 global_snapshots
= g_new0(int, nb_sns
);
2283 for (i
= 0; i
< nb_sns
; i
++) {
2284 SnapshotEntry
*next_sn
;
2285 if (bdrv_all_find_snapshot(sn_tab
[i
].name
, &bs1
) == 0) {
2286 global_snapshots
[total
] = i
;
2288 QTAILQ_FOREACH(image_entry
, &image_list
, next
) {
2289 QTAILQ_FOREACH_SAFE(snapshot_entry
, &image_entry
->snapshots
,
2291 if (!strcmp(sn_tab
[i
].name
, snapshot_entry
->sn
.name
)) {
2292 QTAILQ_REMOVE(&image_entry
->snapshots
, snapshot_entry
,
2294 g_free(snapshot_entry
);
2301 monitor_printf(mon
, "List of snapshots present on all disks:\n");
2304 bdrv_snapshot_dump((fprintf_function
)monitor_printf
, mon
, NULL
);
2305 monitor_printf(mon
, "\n");
2306 for (i
= 0; i
< total
; i
++) {
2307 sn
= &sn_tab
[global_snapshots
[i
]];
2308 /* The ID is not guaranteed to be the same on all images, so
2311 pstrcpy(sn
->id_str
, sizeof(sn
->id_str
), "--");
2312 bdrv_snapshot_dump((fprintf_function
)monitor_printf
, mon
, sn
);
2313 monitor_printf(mon
, "\n");
2316 monitor_printf(mon
, "None\n");
2319 QTAILQ_FOREACH(image_entry
, &image_list
, next
) {
2320 if (QTAILQ_EMPTY(&image_entry
->snapshots
)) {
2324 "\nList of partial (non-loadable) snapshots on '%s':\n",
2325 image_entry
->imagename
);
2326 bdrv_snapshot_dump((fprintf_function
)monitor_printf
, mon
, NULL
);
2327 monitor_printf(mon
, "\n");
2328 QTAILQ_FOREACH(snapshot_entry
, &image_entry
->snapshots
, next
) {
2329 bdrv_snapshot_dump((fprintf_function
)monitor_printf
, mon
,
2330 &snapshot_entry
->sn
);
2331 monitor_printf(mon
, "\n");
2335 QTAILQ_FOREACH_SAFE(image_entry
, &image_list
, next
, next_ie
) {
2336 SnapshotEntry
*next_sn
;
2337 QTAILQ_FOREACH_SAFE(snapshot_entry
, &image_entry
->snapshots
, next
,
2339 g_free(snapshot_entry
);
2341 g_free(image_entry
);
2344 g_free(global_snapshots
);
2348 void vmstate_register_ram(MemoryRegion
*mr
, DeviceState
*dev
)
2350 qemu_ram_set_idstr(mr
->ram_block
,
2351 memory_region_name(mr
), dev
);
2354 void vmstate_unregister_ram(MemoryRegion
*mr
, DeviceState
*dev
)
2356 qemu_ram_unset_idstr(mr
->ram_block
);
2359 void vmstate_register_ram_global(MemoryRegion
*mr
)
2361 vmstate_register_ram(mr
, NULL
);