4 * Copyright (c) 2003-2008 Fabrice Bellard
5 * Copyright (c) 2009-2015 Red Hat Inc
8 * Juan Quintela <quintela@redhat.com>
10 * Permission is hereby granted, free of charge, to any person obtaining a copy
11 * of this software and associated documentation files (the "Software"), to deal
12 * in the Software without restriction, including without limitation the rights
13 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
14 * copies of the Software, and to permit persons to whom the Software is
15 * furnished to do so, subject to the following conditions:
17 * The above copyright notice and this permission notice shall be included in
18 * all copies or substantial portions of the Software.
20 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
21 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
22 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
23 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
24 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
25 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
29 #include "qemu/osdep.h"
30 #include "hw/boards.h"
31 #include "hw/xen/xen.h"
33 #include "migration.h"
34 #include "migration/snapshot.h"
35 #include "migration/misc.h"
36 #include "migration/register.h"
37 #include "migration/global_state.h"
39 #include "qemu-file-channel.h"
40 #include "qemu-file.h"
42 #include "postcopy-ram.h"
43 #include "qapi/qmp/qerror.h"
44 #include "qemu/error-report.h"
45 #include "sysemu/cpus.h"
46 #include "exec/memory.h"
47 #include "exec/target_page.h"
48 #include "qmp-commands.h"
51 #include "block/snapshot.h"
52 #include "qemu/cutils.h"
53 #include "io/channel-buffer.h"
54 #include "io/channel-file.h"
57 #define ETH_P_RARP 0x8035
59 #define ARP_HTYPE_ETH 0x0001
60 #define ARP_PTYPE_IP 0x0800
61 #define ARP_OP_REQUEST_REV 0x3
63 const unsigned int postcopy_ram_discard_version
= 0;
65 static bool skip_section_footers
;
67 /* Subcommands for QEMU_VM_COMMAND */
69 MIG_CMD_INVALID
= 0, /* Must be 0 */
70 MIG_CMD_OPEN_RETURN_PATH
, /* Tell the dest to open the Return path */
71 MIG_CMD_PING
, /* Request a PONG on the RP */
73 MIG_CMD_POSTCOPY_ADVISE
, /* Prior to any page transfers, just
74 warn we might want to do PC */
75 MIG_CMD_POSTCOPY_LISTEN
, /* Start listening for incoming
76 pages as it's running. */
77 MIG_CMD_POSTCOPY_RUN
, /* Start execution */
79 MIG_CMD_POSTCOPY_RAM_DISCARD
, /* A list of pages to discard that
80 were previously sent during
81 precopy but are dirty. */
82 MIG_CMD_PACKAGED
, /* Send a wrapped stream within this stream */
86 #define MAX_VM_CMD_PACKAGED_SIZE (1ul << 24)
87 static struct mig_cmd_args
{
88 ssize_t len
; /* -1 = variable */
91 [MIG_CMD_INVALID
] = { .len
= -1, .name
= "INVALID" },
92 [MIG_CMD_OPEN_RETURN_PATH
] = { .len
= 0, .name
= "OPEN_RETURN_PATH" },
93 [MIG_CMD_PING
] = { .len
= sizeof(uint32_t), .name
= "PING" },
94 [MIG_CMD_POSTCOPY_ADVISE
] = { .len
= 16, .name
= "POSTCOPY_ADVISE" },
95 [MIG_CMD_POSTCOPY_LISTEN
] = { .len
= 0, .name
= "POSTCOPY_LISTEN" },
96 [MIG_CMD_POSTCOPY_RUN
] = { .len
= 0, .name
= "POSTCOPY_RUN" },
97 [MIG_CMD_POSTCOPY_RAM_DISCARD
] = {
98 .len
= -1, .name
= "POSTCOPY_RAM_DISCARD" },
99 [MIG_CMD_PACKAGED
] = { .len
= 4, .name
= "PACKAGED" },
100 [MIG_CMD_MAX
] = { .len
= -1, .name
= "MAX" },
103 static int announce_self_create(uint8_t *buf
,
106 /* Ethernet header. */
107 memset(buf
, 0xff, 6); /* destination MAC addr */
108 memcpy(buf
+ 6, mac_addr
, 6); /* source MAC addr */
109 *(uint16_t *)(buf
+ 12) = htons(ETH_P_RARP
); /* ethertype */
112 *(uint16_t *)(buf
+ 14) = htons(ARP_HTYPE_ETH
); /* hardware addr space */
113 *(uint16_t *)(buf
+ 16) = htons(ARP_PTYPE_IP
); /* protocol addr space */
114 *(buf
+ 18) = 6; /* hardware addr length (ethernet) */
115 *(buf
+ 19) = 4; /* protocol addr length (IPv4) */
116 *(uint16_t *)(buf
+ 20) = htons(ARP_OP_REQUEST_REV
); /* opcode */
117 memcpy(buf
+ 22, mac_addr
, 6); /* source hw addr */
118 memset(buf
+ 28, 0x00, 4); /* source protocol addr */
119 memcpy(buf
+ 32, mac_addr
, 6); /* target hw addr */
120 memset(buf
+ 38, 0x00, 4); /* target protocol addr */
122 /* Padding to get up to 60 bytes (ethernet min packet size, minus FCS). */
123 memset(buf
+ 42, 0x00, 18);
125 return 60; /* len (FCS will be added by hardware) */
128 static void qemu_announce_self_iter(NICState
*nic
, void *opaque
)
133 trace_qemu_announce_self_iter(qemu_ether_ntoa(&nic
->conf
->macaddr
));
134 len
= announce_self_create(buf
, nic
->conf
->macaddr
.a
);
136 qemu_send_packet_raw(qemu_get_queue(nic
), buf
, len
);
140 static void qemu_announce_self_once(void *opaque
)
142 static int count
= SELF_ANNOUNCE_ROUNDS
;
143 QEMUTimer
*timer
= *(QEMUTimer
**)opaque
;
145 qemu_foreach_nic(qemu_announce_self_iter
, NULL
);
148 /* delay 50ms, 150ms, 250ms, ... */
149 timer_mod(timer
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) +
150 self_announce_delay(count
));
157 void qemu_announce_self(void)
159 static QEMUTimer
*timer
;
160 timer
= timer_new_ms(QEMU_CLOCK_REALTIME
, qemu_announce_self_once
, &timer
);
161 qemu_announce_self_once(&timer
);
164 /***********************************************************/
165 /* savevm/loadvm support */
167 static ssize_t
block_writev_buffer(void *opaque
, struct iovec
*iov
, int iovcnt
,
173 qemu_iovec_init_external(&qiov
, iov
, iovcnt
);
174 ret
= bdrv_writev_vmstate(opaque
, &qiov
, pos
);
182 static ssize_t
block_get_buffer(void *opaque
, uint8_t *buf
, int64_t pos
,
185 return bdrv_load_vmstate(opaque
, buf
, pos
, size
);
188 static int bdrv_fclose(void *opaque
)
190 return bdrv_flush(opaque
);
193 static const QEMUFileOps bdrv_read_ops
= {
194 .get_buffer
= block_get_buffer
,
198 static const QEMUFileOps bdrv_write_ops
= {
199 .writev_buffer
= block_writev_buffer
,
203 static QEMUFile
*qemu_fopen_bdrv(BlockDriverState
*bs
, int is_writable
)
206 return qemu_fopen_ops(bs
, &bdrv_write_ops
);
208 return qemu_fopen_ops(bs
, &bdrv_read_ops
);
212 /* QEMUFile timer support.
213 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
216 void timer_put(QEMUFile
*f
, QEMUTimer
*ts
)
218 uint64_t expire_time
;
220 expire_time
= timer_expire_time_ns(ts
);
221 qemu_put_be64(f
, expire_time
);
224 void timer_get(QEMUFile
*f
, QEMUTimer
*ts
)
226 uint64_t expire_time
;
228 expire_time
= qemu_get_be64(f
);
229 if (expire_time
!= -1) {
230 timer_mod_ns(ts
, expire_time
);
237 /* VMState timer support.
238 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
241 static int get_timer(QEMUFile
*f
, void *pv
, size_t size
, VMStateField
*field
)
248 static int put_timer(QEMUFile
*f
, void *pv
, size_t size
, VMStateField
*field
,
257 const VMStateInfo vmstate_info_timer
= {
264 typedef struct CompatEntry
{
269 typedef struct SaveStateEntry
{
270 QTAILQ_ENTRY(SaveStateEntry
) entry
;
275 /* version id read from the stream */
278 /* section id read from the stream */
281 const VMStateDescription
*vmsd
;
287 typedef struct SaveState
{
288 QTAILQ_HEAD(, SaveStateEntry
) handlers
;
289 int global_section_id
;
290 bool skip_configuration
;
293 uint32_t target_page_bits
;
296 static SaveState savevm_state
= {
297 .handlers
= QTAILQ_HEAD_INITIALIZER(savevm_state
.handlers
),
298 .global_section_id
= 0,
299 .skip_configuration
= false,
302 void savevm_skip_configuration(void)
304 savevm_state
.skip_configuration
= true;
308 static void configuration_pre_save(void *opaque
)
310 SaveState
*state
= opaque
;
311 const char *current_name
= MACHINE_GET_CLASS(current_machine
)->name
;
313 state
->len
= strlen(current_name
);
314 state
->name
= current_name
;
315 state
->target_page_bits
= qemu_target_page_bits();
318 static int configuration_pre_load(void *opaque
)
320 SaveState
*state
= opaque
;
322 /* If there is no target-page-bits subsection it means the source
323 * predates the variable-target-page-bits support and is using the
324 * minimum possible value for this CPU.
326 state
->target_page_bits
= qemu_target_page_bits_min();
330 static int configuration_post_load(void *opaque
, int version_id
)
332 SaveState
*state
= opaque
;
333 const char *current_name
= MACHINE_GET_CLASS(current_machine
)->name
;
335 if (strncmp(state
->name
, current_name
, state
->len
) != 0) {
336 error_report("Machine type received is '%.*s' and local is '%s'",
337 (int) state
->len
, state
->name
, current_name
);
341 if (state
->target_page_bits
!= qemu_target_page_bits()) {
342 error_report("Received TARGET_PAGE_BITS is %d but local is %d",
343 state
->target_page_bits
, qemu_target_page_bits());
350 /* The target-page-bits subsection is present only if the
351 * target page size is not the same as the default (ie the
352 * minimum page size for a variable-page-size guest CPU).
353 * If it is present then it contains the actual target page
354 * bits for the machine, and migration will fail if the
355 * two ends don't agree about it.
357 static bool vmstate_target_page_bits_needed(void *opaque
)
359 return qemu_target_page_bits()
360 > qemu_target_page_bits_min();
363 static const VMStateDescription vmstate_target_page_bits
= {
364 .name
= "configuration/target-page-bits",
366 .minimum_version_id
= 1,
367 .needed
= vmstate_target_page_bits_needed
,
368 .fields
= (VMStateField
[]) {
369 VMSTATE_UINT32(target_page_bits
, SaveState
),
370 VMSTATE_END_OF_LIST()
374 static const VMStateDescription vmstate_configuration
= {
375 .name
= "configuration",
377 .pre_load
= configuration_pre_load
,
378 .post_load
= configuration_post_load
,
379 .pre_save
= configuration_pre_save
,
380 .fields
= (VMStateField
[]) {
381 VMSTATE_UINT32(len
, SaveState
),
382 VMSTATE_VBUFFER_ALLOC_UINT32(name
, SaveState
, 0, NULL
, len
),
383 VMSTATE_END_OF_LIST()
385 .subsections
= (const VMStateDescription
*[]) {
386 &vmstate_target_page_bits
,
391 static void dump_vmstate_vmsd(FILE *out_file
,
392 const VMStateDescription
*vmsd
, int indent
,
395 static void dump_vmstate_vmsf(FILE *out_file
, const VMStateField
*field
,
398 fprintf(out_file
, "%*s{\n", indent
, "");
400 fprintf(out_file
, "%*s\"field\": \"%s\",\n", indent
, "", field
->name
);
401 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
403 fprintf(out_file
, "%*s\"field_exists\": %s,\n", indent
, "",
404 field
->field_exists
? "true" : "false");
405 fprintf(out_file
, "%*s\"size\": %zu", indent
, "", field
->size
);
406 if (field
->vmsd
!= NULL
) {
407 fprintf(out_file
, ",\n");
408 dump_vmstate_vmsd(out_file
, field
->vmsd
, indent
, false);
410 fprintf(out_file
, "\n%*s}", indent
- 2, "");
413 static void dump_vmstate_vmss(FILE *out_file
,
414 const VMStateDescription
**subsection
,
417 if (*subsection
!= NULL
) {
418 dump_vmstate_vmsd(out_file
, *subsection
, indent
, true);
422 static void dump_vmstate_vmsd(FILE *out_file
,
423 const VMStateDescription
*vmsd
, int indent
,
427 fprintf(out_file
, "%*s{\n", indent
, "");
429 fprintf(out_file
, "%*s\"%s\": {\n", indent
, "", "Description");
432 fprintf(out_file
, "%*s\"name\": \"%s\",\n", indent
, "", vmsd
->name
);
433 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
435 fprintf(out_file
, "%*s\"minimum_version_id\": %d", indent
, "",
436 vmsd
->minimum_version_id
);
437 if (vmsd
->fields
!= NULL
) {
438 const VMStateField
*field
= vmsd
->fields
;
441 fprintf(out_file
, ",\n%*s\"Fields\": [\n", indent
, "");
443 while (field
->name
!= NULL
) {
444 if (field
->flags
& VMS_MUST_EXIST
) {
445 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
450 fprintf(out_file
, ",\n");
452 dump_vmstate_vmsf(out_file
, field
, indent
+ 2);
456 fprintf(out_file
, "\n%*s]", indent
, "");
458 if (vmsd
->subsections
!= NULL
) {
459 const VMStateDescription
**subsection
= vmsd
->subsections
;
462 fprintf(out_file
, ",\n%*s\"Subsections\": [\n", indent
, "");
464 while (*subsection
!= NULL
) {
466 fprintf(out_file
, ",\n");
468 dump_vmstate_vmss(out_file
, subsection
, indent
+ 2);
472 fprintf(out_file
, "\n%*s]", indent
, "");
474 fprintf(out_file
, "\n%*s}", indent
- 2, "");
477 static void dump_machine_type(FILE *out_file
)
481 mc
= MACHINE_GET_CLASS(current_machine
);
483 fprintf(out_file
, " \"vmschkmachine\": {\n");
484 fprintf(out_file
, " \"Name\": \"%s\"\n", mc
->name
);
485 fprintf(out_file
, " },\n");
488 void dump_vmstate_json_to_file(FILE *out_file
)
493 fprintf(out_file
, "{\n");
494 dump_machine_type(out_file
);
497 list
= object_class_get_list(TYPE_DEVICE
, true);
498 for (elt
= list
; elt
; elt
= elt
->next
) {
499 DeviceClass
*dc
= OBJECT_CLASS_CHECK(DeviceClass
, elt
->data
,
509 fprintf(out_file
, ",\n");
511 name
= object_class_get_name(OBJECT_CLASS(dc
));
512 fprintf(out_file
, "%*s\"%s\": {\n", indent
, "", name
);
514 fprintf(out_file
, "%*s\"Name\": \"%s\",\n", indent
, "", name
);
515 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
516 dc
->vmsd
->version_id
);
517 fprintf(out_file
, "%*s\"minimum_version_id\": %d,\n", indent
, "",
518 dc
->vmsd
->minimum_version_id
);
520 dump_vmstate_vmsd(out_file
, dc
->vmsd
, indent
, false);
522 fprintf(out_file
, "\n%*s}", indent
- 2, "");
525 fprintf(out_file
, "\n}\n");
529 static int calculate_new_instance_id(const char *idstr
)
534 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
535 if (strcmp(idstr
, se
->idstr
) == 0
536 && instance_id
<= se
->instance_id
) {
537 instance_id
= se
->instance_id
+ 1;
543 static int calculate_compat_instance_id(const char *idstr
)
548 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
553 if (strcmp(idstr
, se
->compat
->idstr
) == 0
554 && instance_id
<= se
->compat
->instance_id
) {
555 instance_id
= se
->compat
->instance_id
+ 1;
561 static inline MigrationPriority
save_state_priority(SaveStateEntry
*se
)
564 return se
->vmsd
->priority
;
566 return MIG_PRI_DEFAULT
;
569 static void savevm_state_handler_insert(SaveStateEntry
*nse
)
571 MigrationPriority priority
= save_state_priority(nse
);
574 assert(priority
<= MIG_PRI_MAX
);
576 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
577 if (save_state_priority(se
) < priority
) {
583 QTAILQ_INSERT_BEFORE(se
, nse
, entry
);
585 QTAILQ_INSERT_TAIL(&savevm_state
.handlers
, nse
, entry
);
589 /* TODO: Individual devices generally have very little idea about the rest
590 of the system, so instance_id should be removed/replaced.
591 Meanwhile pass -1 as instance_id if you do not already have a clearly
592 distinguishing id for all instances of your device class. */
593 int register_savevm_live(DeviceState
*dev
,
602 se
= g_new0(SaveStateEntry
, 1);
603 se
->version_id
= version_id
;
604 se
->section_id
= savevm_state
.global_section_id
++;
608 /* if this is a live_savem then set is_ram */
609 if (ops
->save_live_setup
!= NULL
) {
614 char *id
= qdev_get_dev_path(dev
);
616 if (snprintf(se
->idstr
, sizeof(se
->idstr
), "%s/", id
) >=
618 error_report("Path too long for VMState (%s)", id
);
626 se
->compat
= g_new0(CompatEntry
, 1);
627 pstrcpy(se
->compat
->idstr
, sizeof(se
->compat
->idstr
), idstr
);
628 se
->compat
->instance_id
= instance_id
== -1 ?
629 calculate_compat_instance_id(idstr
) : instance_id
;
633 pstrcat(se
->idstr
, sizeof(se
->idstr
), idstr
);
635 if (instance_id
== -1) {
636 se
->instance_id
= calculate_new_instance_id(se
->idstr
);
638 se
->instance_id
= instance_id
;
640 assert(!se
->compat
|| se
->instance_id
== 0);
641 savevm_state_handler_insert(se
);
645 void unregister_savevm(DeviceState
*dev
, const char *idstr
, void *opaque
)
647 SaveStateEntry
*se
, *new_se
;
651 char *path
= qdev_get_dev_path(dev
);
653 pstrcpy(id
, sizeof(id
), path
);
654 pstrcat(id
, sizeof(id
), "/");
658 pstrcat(id
, sizeof(id
), idstr
);
660 QTAILQ_FOREACH_SAFE(se
, &savevm_state
.handlers
, entry
, new_se
) {
661 if (strcmp(se
->idstr
, id
) == 0 && se
->opaque
== opaque
) {
662 QTAILQ_REMOVE(&savevm_state
.handlers
, se
, entry
);
669 int vmstate_register_with_alias_id(DeviceState
*dev
, int instance_id
,
670 const VMStateDescription
*vmsd
,
671 void *opaque
, int alias_id
,
672 int required_for_version
,
677 /* If this triggers, alias support can be dropped for the vmsd. */
678 assert(alias_id
== -1 || required_for_version
>= vmsd
->minimum_version_id
);
680 se
= g_new0(SaveStateEntry
, 1);
681 se
->version_id
= vmsd
->version_id
;
682 se
->section_id
= savevm_state
.global_section_id
++;
685 se
->alias_id
= alias_id
;
688 char *id
= qdev_get_dev_path(dev
);
690 if (snprintf(se
->idstr
, sizeof(se
->idstr
), "%s/", id
) >=
692 error_setg(errp
, "Path too long for VMState (%s)", id
);
700 se
->compat
= g_new0(CompatEntry
, 1);
701 pstrcpy(se
->compat
->idstr
, sizeof(se
->compat
->idstr
), vmsd
->name
);
702 se
->compat
->instance_id
= instance_id
== -1 ?
703 calculate_compat_instance_id(vmsd
->name
) : instance_id
;
707 pstrcat(se
->idstr
, sizeof(se
->idstr
), vmsd
->name
);
709 if (instance_id
== -1) {
710 se
->instance_id
= calculate_new_instance_id(se
->idstr
);
712 se
->instance_id
= instance_id
;
714 assert(!se
->compat
|| se
->instance_id
== 0);
715 savevm_state_handler_insert(se
);
719 void vmstate_unregister(DeviceState
*dev
, const VMStateDescription
*vmsd
,
722 SaveStateEntry
*se
, *new_se
;
724 QTAILQ_FOREACH_SAFE(se
, &savevm_state
.handlers
, entry
, new_se
) {
725 if (se
->vmsd
== vmsd
&& se
->opaque
== opaque
) {
726 QTAILQ_REMOVE(&savevm_state
.handlers
, se
, entry
);
733 static int vmstate_load(QEMUFile
*f
, SaveStateEntry
*se
)
735 trace_vmstate_load(se
->idstr
, se
->vmsd
? se
->vmsd
->name
: "(old)");
736 if (!se
->vmsd
) { /* Old style */
737 return se
->ops
->load_state(f
, se
->opaque
, se
->load_version_id
);
739 return vmstate_load_state(f
, se
->vmsd
, se
->opaque
, se
->load_version_id
);
742 static void vmstate_save_old_style(QEMUFile
*f
, SaveStateEntry
*se
, QJSON
*vmdesc
)
744 int64_t old_offset
, size
;
746 old_offset
= qemu_ftell_fast(f
);
747 se
->ops
->save_state(f
, se
->opaque
);
748 size
= qemu_ftell_fast(f
) - old_offset
;
751 json_prop_int(vmdesc
, "size", size
);
752 json_start_array(vmdesc
, "fields");
753 json_start_object(vmdesc
, NULL
);
754 json_prop_str(vmdesc
, "name", "data");
755 json_prop_int(vmdesc
, "size", size
);
756 json_prop_str(vmdesc
, "type", "buffer");
757 json_end_object(vmdesc
);
758 json_end_array(vmdesc
);
762 static void vmstate_save(QEMUFile
*f
, SaveStateEntry
*se
, QJSON
*vmdesc
)
764 trace_vmstate_save(se
->idstr
, se
->vmsd
? se
->vmsd
->name
: "(old)");
766 vmstate_save_old_style(f
, se
, vmdesc
);
769 vmstate_save_state(f
, se
->vmsd
, se
->opaque
, vmdesc
);
772 void savevm_skip_section_footers(void)
774 skip_section_footers
= true;
778 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
780 static void save_section_header(QEMUFile
*f
, SaveStateEntry
*se
,
781 uint8_t section_type
)
783 qemu_put_byte(f
, section_type
);
784 qemu_put_be32(f
, se
->section_id
);
786 if (section_type
== QEMU_VM_SECTION_FULL
||
787 section_type
== QEMU_VM_SECTION_START
) {
789 size_t len
= strlen(se
->idstr
);
790 qemu_put_byte(f
, len
);
791 qemu_put_buffer(f
, (uint8_t *)se
->idstr
, len
);
793 qemu_put_be32(f
, se
->instance_id
);
794 qemu_put_be32(f
, se
->version_id
);
799 * Write a footer onto device sections that catches cases misformatted device
802 static void save_section_footer(QEMUFile
*f
, SaveStateEntry
*se
)
804 if (!skip_section_footers
) {
805 qemu_put_byte(f
, QEMU_VM_SECTION_FOOTER
);
806 qemu_put_be32(f
, se
->section_id
);
811 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
812 * command and associated data.
814 * @f: File to send command on
815 * @command: Command type to send
816 * @len: Length of associated data
817 * @data: Data associated with command.
819 static void qemu_savevm_command_send(QEMUFile
*f
,
820 enum qemu_vm_cmd command
,
824 trace_savevm_command_send(command
, len
);
825 qemu_put_byte(f
, QEMU_VM_COMMAND
);
826 qemu_put_be16(f
, (uint16_t)command
);
827 qemu_put_be16(f
, len
);
828 qemu_put_buffer(f
, data
, len
);
832 void qemu_savevm_send_ping(QEMUFile
*f
, uint32_t value
)
836 trace_savevm_send_ping(value
);
837 buf
= cpu_to_be32(value
);
838 qemu_savevm_command_send(f
, MIG_CMD_PING
, sizeof(value
), (uint8_t *)&buf
);
841 void qemu_savevm_send_open_return_path(QEMUFile
*f
)
843 trace_savevm_send_open_return_path();
844 qemu_savevm_command_send(f
, MIG_CMD_OPEN_RETURN_PATH
, 0, NULL
);
847 /* We have a buffer of data to send; we don't want that all to be loaded
848 * by the command itself, so the command contains just the length of the
849 * extra buffer that we then send straight after it.
850 * TODO: Must be a better way to organise that
856 int qemu_savevm_send_packaged(QEMUFile
*f
, const uint8_t *buf
, size_t len
)
860 if (len
> MAX_VM_CMD_PACKAGED_SIZE
) {
861 error_report("%s: Unreasonably large packaged state: %zu",
866 tmp
= cpu_to_be32(len
);
868 trace_qemu_savevm_send_packaged();
869 qemu_savevm_command_send(f
, MIG_CMD_PACKAGED
, 4, (uint8_t *)&tmp
);
871 qemu_put_buffer(f
, buf
, len
);
876 /* Send prior to any postcopy transfer */
877 void qemu_savevm_send_postcopy_advise(QEMUFile
*f
)
880 tmp
[0] = cpu_to_be64(ram_pagesize_summary());
881 tmp
[1] = cpu_to_be64(qemu_target_page_size());
883 trace_qemu_savevm_send_postcopy_advise();
884 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_ADVISE
, 16, (uint8_t *)tmp
);
887 /* Sent prior to starting the destination running in postcopy, discard pages
888 * that have already been sent but redirtied on the source.
889 * CMD_POSTCOPY_RAM_DISCARD consist of:
891 * byte Length of name field (not including 0)
892 * n x byte RAM block name
893 * byte 0 terminator (just for safety)
894 * n x Byte ranges within the named RAMBlock
895 * be64 Start of the range
898 * name: RAMBlock name that these entries are part of
899 * len: Number of page entries
900 * start_list: 'len' addresses
901 * length_list: 'len' addresses
904 void qemu_savevm_send_postcopy_ram_discard(QEMUFile
*f
, const char *name
,
906 uint64_t *start_list
,
907 uint64_t *length_list
)
912 size_t name_len
= strlen(name
);
914 trace_qemu_savevm_send_postcopy_ram_discard(name
, len
);
915 assert(name_len
< 256);
916 buf
= g_malloc0(1 + 1 + name_len
+ 1 + (8 + 8) * len
);
917 buf
[0] = postcopy_ram_discard_version
;
919 memcpy(buf
+ 2, name
, name_len
);
920 tmplen
= 2 + name_len
;
921 buf
[tmplen
++] = '\0';
923 for (t
= 0; t
< len
; t
++) {
924 stq_be_p(buf
+ tmplen
, start_list
[t
]);
926 stq_be_p(buf
+ tmplen
, length_list
[t
]);
929 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RAM_DISCARD
, tmplen
, buf
);
933 /* Get the destination into a state where it can receive postcopy data. */
934 void qemu_savevm_send_postcopy_listen(QEMUFile
*f
)
936 trace_savevm_send_postcopy_listen();
937 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_LISTEN
, 0, NULL
);
940 /* Kick the destination into running */
941 void qemu_savevm_send_postcopy_run(QEMUFile
*f
)
943 trace_savevm_send_postcopy_run();
944 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RUN
, 0, NULL
);
947 bool qemu_savevm_state_blocked(Error
**errp
)
951 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
952 if (se
->vmsd
&& se
->vmsd
->unmigratable
) {
953 error_setg(errp
, "State blocked by non-migratable device '%s'",
961 static bool enforce_config_section(void)
963 MachineState
*machine
= MACHINE(qdev_get_machine());
964 return machine
->enforce_config_section
;
967 void qemu_savevm_state_header(QEMUFile
*f
)
969 trace_savevm_state_header();
970 qemu_put_be32(f
, QEMU_VM_FILE_MAGIC
);
971 qemu_put_be32(f
, QEMU_VM_FILE_VERSION
);
973 if (!savevm_state
.skip_configuration
|| enforce_config_section()) {
974 qemu_put_byte(f
, QEMU_VM_CONFIGURATION
);
975 vmstate_save_state(f
, &vmstate_configuration
, &savevm_state
, 0);
980 void qemu_savevm_state_begin(QEMUFile
*f
)
985 trace_savevm_state_begin();
986 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
987 if (!se
->ops
|| !se
->ops
->save_live_setup
) {
990 if (se
->ops
&& se
->ops
->is_active
) {
991 if (!se
->ops
->is_active(se
->opaque
)) {
995 save_section_header(f
, se
, QEMU_VM_SECTION_START
);
997 ret
= se
->ops
->save_live_setup(f
, se
->opaque
);
998 save_section_footer(f
, se
);
1000 qemu_file_set_error(f
, ret
);
1007 * this function has three return values:
1008 * negative: there was one error, and we have -errno.
1009 * 0 : We haven't finished, caller have to go again
1010 * 1 : We have finished, we can go to complete phase
1012 int qemu_savevm_state_iterate(QEMUFile
*f
, bool postcopy
)
1017 trace_savevm_state_iterate();
1018 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1019 if (!se
->ops
|| !se
->ops
->save_live_iterate
) {
1022 if (se
->ops
&& se
->ops
->is_active
) {
1023 if (!se
->ops
->is_active(se
->opaque
)) {
1028 * In the postcopy phase, any device that doesn't know how to
1029 * do postcopy should have saved it's state in the _complete
1030 * call that's already run, it might get confused if we call
1031 * iterate afterwards.
1033 if (postcopy
&& !se
->ops
->save_live_complete_postcopy
) {
1036 if (qemu_file_rate_limit(f
)) {
1039 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1041 save_section_header(f
, se
, QEMU_VM_SECTION_PART
);
1043 ret
= se
->ops
->save_live_iterate(f
, se
->opaque
);
1044 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1045 save_section_footer(f
, se
);
1048 qemu_file_set_error(f
, ret
);
1051 /* Do not proceed to the next vmstate before this one reported
1052 completion of the current stage. This serializes the migration
1053 and reduces the probability that a faster changing state is
1054 synchronized over and over again. */
1061 static bool should_send_vmdesc(void)
1063 MachineState
*machine
= MACHINE(qdev_get_machine());
1064 bool in_postcopy
= migration_in_postcopy();
1065 return !machine
->suppress_vmdesc
&& !in_postcopy
;
1069 * Calls the save_live_complete_postcopy methods
1070 * causing the last few pages to be sent immediately and doing any associated
1072 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1073 * all the other devices, but that happens at the point we switch to postcopy.
1075 void qemu_savevm_state_complete_postcopy(QEMUFile
*f
)
1080 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1081 if (!se
->ops
|| !se
->ops
->save_live_complete_postcopy
) {
1084 if (se
->ops
&& se
->ops
->is_active
) {
1085 if (!se
->ops
->is_active(se
->opaque
)) {
1089 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1091 qemu_put_byte(f
, QEMU_VM_SECTION_END
);
1092 qemu_put_be32(f
, se
->section_id
);
1094 ret
= se
->ops
->save_live_complete_postcopy(f
, se
->opaque
);
1095 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1096 save_section_footer(f
, se
);
1098 qemu_file_set_error(f
, ret
);
1103 qemu_put_byte(f
, QEMU_VM_EOF
);
1107 int qemu_savevm_state_complete_precopy(QEMUFile
*f
, bool iterable_only
,
1108 bool inactivate_disks
)
1114 bool in_postcopy
= migration_in_postcopy();
1116 trace_savevm_state_complete_precopy();
1118 cpu_synchronize_all_states();
1120 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1122 (in_postcopy
&& se
->ops
->save_live_complete_postcopy
) ||
1123 (in_postcopy
&& !iterable_only
) ||
1124 !se
->ops
->save_live_complete_precopy
) {
1128 if (se
->ops
&& se
->ops
->is_active
) {
1129 if (!se
->ops
->is_active(se
->opaque
)) {
1133 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1135 save_section_header(f
, se
, QEMU_VM_SECTION_END
);
1137 ret
= se
->ops
->save_live_complete_precopy(f
, se
->opaque
);
1138 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1139 save_section_footer(f
, se
);
1141 qemu_file_set_error(f
, ret
);
1146 if (iterable_only
) {
1150 vmdesc
= qjson_new();
1151 json_prop_int(vmdesc
, "page_size", qemu_target_page_size());
1152 json_start_array(vmdesc
, "devices");
1153 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1155 if ((!se
->ops
|| !se
->ops
->save_state
) && !se
->vmsd
) {
1158 if (se
->vmsd
&& !vmstate_save_needed(se
->vmsd
, se
->opaque
)) {
1159 trace_savevm_section_skip(se
->idstr
, se
->section_id
);
1163 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1165 json_start_object(vmdesc
, NULL
);
1166 json_prop_str(vmdesc
, "name", se
->idstr
);
1167 json_prop_int(vmdesc
, "instance_id", se
->instance_id
);
1169 save_section_header(f
, se
, QEMU_VM_SECTION_FULL
);
1170 vmstate_save(f
, se
, vmdesc
);
1171 trace_savevm_section_end(se
->idstr
, se
->section_id
, 0);
1172 save_section_footer(f
, se
);
1174 json_end_object(vmdesc
);
1177 if (inactivate_disks
) {
1178 /* Inactivate before sending QEMU_VM_EOF so that the
1179 * bdrv_invalidate_cache_all() on the other end won't fail. */
1180 ret
= bdrv_inactivate_all();
1182 qemu_file_set_error(f
, ret
);
1187 /* Postcopy stream will still be going */
1188 qemu_put_byte(f
, QEMU_VM_EOF
);
1191 json_end_array(vmdesc
);
1192 qjson_finish(vmdesc
);
1193 vmdesc_len
= strlen(qjson_get_str(vmdesc
));
1195 if (should_send_vmdesc()) {
1196 qemu_put_byte(f
, QEMU_VM_VMDESCRIPTION
);
1197 qemu_put_be32(f
, vmdesc_len
);
1198 qemu_put_buffer(f
, (uint8_t *)qjson_get_str(vmdesc
), vmdesc_len
);
1200 qjson_destroy(vmdesc
);
1206 /* Give an estimate of the amount left to be transferred,
1207 * the result is split into the amount for units that can and
1208 * for units that can't do postcopy.
1210 void qemu_savevm_state_pending(QEMUFile
*f
, uint64_t threshold_size
,
1211 uint64_t *res_non_postcopiable
,
1212 uint64_t *res_postcopiable
)
1216 *res_non_postcopiable
= 0;
1217 *res_postcopiable
= 0;
1220 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1221 if (!se
->ops
|| !se
->ops
->save_live_pending
) {
1224 if (se
->ops
&& se
->ops
->is_active
) {
1225 if (!se
->ops
->is_active(se
->opaque
)) {
1229 se
->ops
->save_live_pending(f
, se
->opaque
, threshold_size
,
1230 res_non_postcopiable
, res_postcopiable
);
1234 void qemu_savevm_state_cleanup(void)
1238 trace_savevm_state_cleanup();
1239 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1240 if (se
->ops
&& se
->ops
->cleanup
) {
1241 se
->ops
->cleanup(se
->opaque
);
1246 static int qemu_savevm_state(QEMUFile
*f
, Error
**errp
)
1249 MigrationState
*ms
= migrate_init();
1250 MigrationStatus status
;
1251 ms
->to_dst_file
= f
;
1253 if (migration_is_blocked(errp
)) {
1258 if (migrate_use_block()) {
1259 error_setg(errp
, "Block migration and snapshots are incompatible");
1264 qemu_mutex_unlock_iothread();
1265 qemu_savevm_state_header(f
);
1266 qemu_savevm_state_begin(f
);
1267 qemu_mutex_lock_iothread();
1269 while (qemu_file_get_error(f
) == 0) {
1270 if (qemu_savevm_state_iterate(f
, false) > 0) {
1275 ret
= qemu_file_get_error(f
);
1277 qemu_savevm_state_complete_precopy(f
, false, false);
1278 ret
= qemu_file_get_error(f
);
1280 qemu_savevm_state_cleanup();
1282 error_setg_errno(errp
, -ret
, "Error while writing VM state");
1287 status
= MIGRATION_STATUS_FAILED
;
1289 status
= MIGRATION_STATUS_COMPLETED
;
1291 migrate_set_state(&ms
->state
, MIGRATION_STATUS_SETUP
, status
);
1293 /* f is outer parameter, it should not stay in global migration state after
1294 * this function finished */
1295 ms
->to_dst_file
= NULL
;
1300 static int qemu_save_device_state(QEMUFile
*f
)
1304 qemu_put_be32(f
, QEMU_VM_FILE_MAGIC
);
1305 qemu_put_be32(f
, QEMU_VM_FILE_VERSION
);
1307 cpu_synchronize_all_states();
1309 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1313 if ((!se
->ops
|| !se
->ops
->save_state
) && !se
->vmsd
) {
1316 if (se
->vmsd
&& !vmstate_save_needed(se
->vmsd
, se
->opaque
)) {
1320 save_section_header(f
, se
, QEMU_VM_SECTION_FULL
);
1322 vmstate_save(f
, se
, NULL
);
1324 save_section_footer(f
, se
);
1327 qemu_put_byte(f
, QEMU_VM_EOF
);
1329 return qemu_file_get_error(f
);
1332 static SaveStateEntry
*find_se(const char *idstr
, int instance_id
)
1336 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1337 if (!strcmp(se
->idstr
, idstr
) &&
1338 (instance_id
== se
->instance_id
||
1339 instance_id
== se
->alias_id
))
1341 /* Migrating from an older version? */
1342 if (strstr(se
->idstr
, idstr
) && se
->compat
) {
1343 if (!strcmp(se
->compat
->idstr
, idstr
) &&
1344 (instance_id
== se
->compat
->instance_id
||
1345 instance_id
== se
->alias_id
))
1352 enum LoadVMExitCodes
{
1353 /* Allow a command to quit all layers of nested loadvm loops */
1357 static int qemu_loadvm_state_main(QEMUFile
*f
, MigrationIncomingState
*mis
);
1359 /* ------ incoming postcopy messages ------ */
1360 /* 'advise' arrives before any transfers just to tell us that a postcopy
1361 * *might* happen - it might be skipped if precopy transferred everything
1364 static int loadvm_postcopy_handle_advise(MigrationIncomingState
*mis
)
1366 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_ADVISE
);
1367 uint64_t remote_pagesize_summary
, local_pagesize_summary
, remote_tps
;
1369 trace_loadvm_postcopy_handle_advise();
1370 if (ps
!= POSTCOPY_INCOMING_NONE
) {
1371 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps
);
1375 if (!postcopy_ram_supported_by_host()) {
1376 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
1380 remote_pagesize_summary
= qemu_get_be64(mis
->from_src_file
);
1381 local_pagesize_summary
= ram_pagesize_summary();
1383 if (remote_pagesize_summary
!= local_pagesize_summary
) {
1385 * This detects two potential causes of mismatch:
1386 * a) A mismatch in host page sizes
1387 * Some combinations of mismatch are probably possible but it gets
1388 * a bit more complicated. In particular we need to place whole
1389 * host pages on the dest at once, and we need to ensure that we
1390 * handle dirtying to make sure we never end up sending part of
1391 * a hostpage on it's own.
1392 * b) The use of different huge page sizes on source/destination
1393 * a more fine grain test is performed during RAM block migration
1394 * but this test here causes a nice early clear failure, and
1395 * also fails when passed to an older qemu that doesn't
1398 error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1400 remote_pagesize_summary
, local_pagesize_summary
);
1404 remote_tps
= qemu_get_be64(mis
->from_src_file
);
1405 if (remote_tps
!= qemu_target_page_size()) {
1407 * Again, some differences could be dealt with, but for now keep it
1410 error_report("Postcopy needs matching target page sizes (s=%d d=%zd)",
1411 (int)remote_tps
, qemu_target_page_size());
1415 if (ram_postcopy_incoming_init(mis
)) {
1419 postcopy_state_set(POSTCOPY_INCOMING_ADVISE
);
1424 /* After postcopy we will be told to throw some pages away since they're
1425 * dirty and will have to be demand fetched. Must happen before CPU is
1427 * There can be 0..many of these messages, each encoding multiple pages.
1429 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState
*mis
,
1434 PostcopyState ps
= postcopy_state_get();
1436 trace_loadvm_postcopy_ram_handle_discard();
1439 case POSTCOPY_INCOMING_ADVISE
:
1441 tmp
= postcopy_ram_prepare_discard(mis
);
1447 case POSTCOPY_INCOMING_DISCARD
:
1448 /* Expected state */
1452 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1456 /* We're expecting a
1458 * a RAM ID string (length byte, name, 0 term)
1459 * then at least 1 16 byte chunk
1461 if (len
< (1 + 1 + 1 + 1 + 2 * 8)) {
1462 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len
);
1466 tmp
= qemu_get_byte(mis
->from_src_file
);
1467 if (tmp
!= postcopy_ram_discard_version
) {
1468 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp
);
1472 if (!qemu_get_counted_string(mis
->from_src_file
, ramid
)) {
1473 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1476 tmp
= qemu_get_byte(mis
->from_src_file
);
1478 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp
);
1482 len
-= 3 + strlen(ramid
);
1484 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len
);
1487 trace_loadvm_postcopy_ram_handle_discard_header(ramid
, len
);
1489 uint64_t start_addr
, block_length
;
1490 start_addr
= qemu_get_be64(mis
->from_src_file
);
1491 block_length
= qemu_get_be64(mis
->from_src_file
);
1494 int ret
= ram_discard_range(ramid
, start_addr
, block_length
);
1499 trace_loadvm_postcopy_ram_handle_discard_end();
1505 * Triggered by a postcopy_listen command; this thread takes over reading
1506 * the input stream, leaving the main thread free to carry on loading the rest
1507 * of the device state (from RAM).
1508 * (TODO:This could do with being in a postcopy file - but there again it's
1509 * just another input loop, not that postcopy specific)
1511 static void *postcopy_ram_listen_thread(void *opaque
)
1513 QEMUFile
*f
= opaque
;
1514 MigrationIncomingState
*mis
= migration_incoming_get_current();
1517 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
1518 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1519 qemu_sem_post(&mis
->listen_thread_sem
);
1520 trace_postcopy_ram_listen_thread_start();
1523 * Because we're a thread and not a coroutine we can't yield
1524 * in qemu_file, and thus we must be blocking now.
1526 qemu_file_set_blocking(f
, true);
1527 load_res
= qemu_loadvm_state_main(f
, mis
);
1528 /* And non-blocking again so we don't block in any cleanup */
1529 qemu_file_set_blocking(f
, false);
1531 trace_postcopy_ram_listen_thread_exit();
1533 error_report("%s: loadvm failed: %d", __func__
, load_res
);
1534 qemu_file_set_error(f
, load_res
);
1535 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1536 MIGRATION_STATUS_FAILED
);
1539 * This looks good, but it's possible that the device loading in the
1540 * main thread hasn't finished yet, and so we might not be in 'RUN'
1541 * state yet; wait for the end of the main thread.
1543 qemu_event_wait(&mis
->main_thread_load_event
);
1545 postcopy_ram_incoming_cleanup(mis
);
1549 * If something went wrong then we have a bad state so exit;
1550 * depending how far we got it might be possible at this point
1551 * to leave the guest running and fire MCEs for pages that never
1552 * arrived as a desperate recovery step.
1557 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1558 MIGRATION_STATUS_COMPLETED
);
1560 * If everything has worked fine, then the main thread has waited
1561 * for us to start, and we're the last use of the mis.
1562 * (If something broke then qemu will have to exit anyway since it's
1563 * got a bad migration state).
1565 migration_incoming_state_destroy();
1571 /* After this message we must be able to immediately receive postcopy data */
1572 static int loadvm_postcopy_handle_listen(MigrationIncomingState
*mis
)
1574 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_LISTENING
);
1575 trace_loadvm_postcopy_handle_listen();
1576 if (ps
!= POSTCOPY_INCOMING_ADVISE
&& ps
!= POSTCOPY_INCOMING_DISCARD
) {
1577 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps
);
1580 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
1582 * A rare case, we entered listen without having to do any discards,
1583 * so do the setup that's normally done at the time of the 1st discard.
1585 postcopy_ram_prepare_discard(mis
);
1589 * Sensitise RAM - can now generate requests for blocks that don't exist
1590 * However, at this point the CPU shouldn't be running, and the IO
1591 * shouldn't be doing anything yet so don't actually expect requests
1593 if (postcopy_ram_enable_notify(mis
)) {
1597 if (mis
->have_listen_thread
) {
1598 error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread");
1602 mis
->have_listen_thread
= true;
1603 /* Start up the listening thread and wait for it to signal ready */
1604 qemu_sem_init(&mis
->listen_thread_sem
, 0);
1605 qemu_thread_create(&mis
->listen_thread
, "postcopy/listen",
1606 postcopy_ram_listen_thread
, mis
->from_src_file
,
1607 QEMU_THREAD_DETACHED
);
1608 qemu_sem_wait(&mis
->listen_thread_sem
);
1609 qemu_sem_destroy(&mis
->listen_thread_sem
);
1619 static void loadvm_postcopy_handle_run_bh(void *opaque
)
1621 Error
*local_err
= NULL
;
1622 HandleRunBhData
*data
= opaque
;
1624 /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1627 cpu_synchronize_all_post_init();
1629 qemu_announce_self();
1631 /* Make sure all file formats flush their mutable metadata.
1632 * If we get an error here, just don't restart the VM yet. */
1633 bdrv_invalidate_cache_all(&local_err
);
1635 error_report_err(local_err
);
1640 trace_loadvm_postcopy_handle_run_cpu_sync();
1641 cpu_synchronize_all_post_init();
1643 trace_loadvm_postcopy_handle_run_vmstart();
1646 /* Hold onto your hats, starting the CPU */
1649 /* leave it paused and let management decide when to start the CPU */
1650 runstate_set(RUN_STATE_PAUSED
);
1653 qemu_bh_delete(data
->bh
);
1657 /* After all discards we can start running and asking for pages */
1658 static int loadvm_postcopy_handle_run(MigrationIncomingState
*mis
)
1660 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_RUNNING
);
1661 HandleRunBhData
*data
;
1663 trace_loadvm_postcopy_handle_run();
1664 if (ps
!= POSTCOPY_INCOMING_LISTENING
) {
1665 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps
);
1669 data
= g_new(HandleRunBhData
, 1);
1670 data
->bh
= qemu_bh_new(loadvm_postcopy_handle_run_bh
, data
);
1671 qemu_bh_schedule(data
->bh
);
1673 /* We need to finish reading the stream from the package
1674 * and also stop reading anything more from the stream that loaded the
1675 * package (since it's now being read by the listener thread).
1676 * LOADVM_QUIT will quit all the layers of nested loadvm loops.
1682 * Immediately following this command is a blob of data containing an embedded
1683 * chunk of migration stream; read it and load it.
1685 * @mis: Incoming state
1686 * @length: Length of packaged data to read
1688 * Returns: Negative values on error
1691 static int loadvm_handle_cmd_packaged(MigrationIncomingState
*mis
)
1695 QIOChannelBuffer
*bioc
;
1697 length
= qemu_get_be32(mis
->from_src_file
);
1698 trace_loadvm_handle_cmd_packaged(length
);
1700 if (length
> MAX_VM_CMD_PACKAGED_SIZE
) {
1701 error_report("Unreasonably large packaged state: %zu", length
);
1705 bioc
= qio_channel_buffer_new(length
);
1706 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-loadvm-buffer");
1707 ret
= qemu_get_buffer(mis
->from_src_file
,
1710 if (ret
!= length
) {
1711 object_unref(OBJECT(bioc
));
1712 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
1714 return (ret
< 0) ? ret
: -EAGAIN
;
1716 bioc
->usage
+= length
;
1717 trace_loadvm_handle_cmd_packaged_received(ret
);
1719 QEMUFile
*packf
= qemu_fopen_channel_input(QIO_CHANNEL(bioc
));
1721 ret
= qemu_loadvm_state_main(packf
, mis
);
1722 trace_loadvm_handle_cmd_packaged_main(ret
);
1724 object_unref(OBJECT(bioc
));
1730 * Process an incoming 'QEMU_VM_COMMAND'
1731 * 0 just a normal return
1732 * LOADVM_QUIT All good, but exit the loop
1735 static int loadvm_process_command(QEMUFile
*f
)
1737 MigrationIncomingState
*mis
= migration_incoming_get_current();
1742 cmd
= qemu_get_be16(f
);
1743 len
= qemu_get_be16(f
);
1745 trace_loadvm_process_command(cmd
, len
);
1746 if (cmd
>= MIG_CMD_MAX
|| cmd
== MIG_CMD_INVALID
) {
1747 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd
, len
);
1751 if (mig_cmd_args
[cmd
].len
!= -1 && mig_cmd_args
[cmd
].len
!= len
) {
1752 error_report("%s received with bad length - expecting %zu, got %d",
1753 mig_cmd_args
[cmd
].name
,
1754 (size_t)mig_cmd_args
[cmd
].len
, len
);
1759 case MIG_CMD_OPEN_RETURN_PATH
:
1760 if (mis
->to_src_file
) {
1761 error_report("CMD_OPEN_RETURN_PATH called when RP already open");
1762 /* Not really a problem, so don't give up */
1765 mis
->to_src_file
= qemu_file_get_return_path(f
);
1766 if (!mis
->to_src_file
) {
1767 error_report("CMD_OPEN_RETURN_PATH failed");
1773 tmp32
= qemu_get_be32(f
);
1774 trace_loadvm_process_command_ping(tmp32
);
1775 if (!mis
->to_src_file
) {
1776 error_report("CMD_PING (0x%x) received with no return path",
1780 migrate_send_rp_pong(mis
, tmp32
);
1783 case MIG_CMD_PACKAGED
:
1784 return loadvm_handle_cmd_packaged(mis
);
1786 case MIG_CMD_POSTCOPY_ADVISE
:
1787 return loadvm_postcopy_handle_advise(mis
);
1789 case MIG_CMD_POSTCOPY_LISTEN
:
1790 return loadvm_postcopy_handle_listen(mis
);
1792 case MIG_CMD_POSTCOPY_RUN
:
1793 return loadvm_postcopy_handle_run(mis
);
1795 case MIG_CMD_POSTCOPY_RAM_DISCARD
:
1796 return loadvm_postcopy_ram_handle_discard(mis
, len
);
1803 * Read a footer off the wire and check that it matches the expected section
1805 * Returns: true if the footer was good
1806 * false if there is a problem (and calls error_report to say why)
1808 static bool check_section_footer(QEMUFile
*f
, SaveStateEntry
*se
)
1811 uint32_t read_section_id
;
1813 if (skip_section_footers
) {
1814 /* No footer to check */
1818 read_mark
= qemu_get_byte(f
);
1820 if (read_mark
!= QEMU_VM_SECTION_FOOTER
) {
1821 error_report("Missing section footer for %s", se
->idstr
);
1825 read_section_id
= qemu_get_be32(f
);
1826 if (read_section_id
!= se
->load_section_id
) {
1827 error_report("Mismatched section id in footer for %s -"
1828 " read 0x%x expected 0x%x",
1829 se
->idstr
, read_section_id
, se
->load_section_id
);
1838 qemu_loadvm_section_start_full(QEMUFile
*f
, MigrationIncomingState
*mis
)
1840 uint32_t instance_id
, version_id
, section_id
;
1845 /* Read section start */
1846 section_id
= qemu_get_be32(f
);
1847 if (!qemu_get_counted_string(f
, idstr
)) {
1848 error_report("Unable to read ID string for section %u",
1852 instance_id
= qemu_get_be32(f
);
1853 version_id
= qemu_get_be32(f
);
1855 trace_qemu_loadvm_state_section_startfull(section_id
, idstr
,
1856 instance_id
, version_id
);
1857 /* Find savevm section */
1858 se
= find_se(idstr
, instance_id
);
1860 error_report("Unknown savevm section or instance '%s' %d",
1861 idstr
, instance_id
);
1865 /* Validate version */
1866 if (version_id
> se
->version_id
) {
1867 error_report("savevm: unsupported version %d for '%s' v%d",
1868 version_id
, idstr
, se
->version_id
);
1871 se
->load_version_id
= version_id
;
1872 se
->load_section_id
= section_id
;
1874 /* Validate if it is a device's state */
1875 if (xen_enabled() && se
->is_ram
) {
1876 error_report("loadvm: %s RAM loading not allowed on Xen", idstr
);
1880 ret
= vmstate_load(f
, se
);
1882 error_report("error while loading state for instance 0x%x of"
1883 " device '%s'", instance_id
, idstr
);
1886 if (!check_section_footer(f
, se
)) {
1894 qemu_loadvm_section_part_end(QEMUFile
*f
, MigrationIncomingState
*mis
)
1896 uint32_t section_id
;
1900 section_id
= qemu_get_be32(f
);
1902 trace_qemu_loadvm_state_section_partend(section_id
);
1903 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1904 if (se
->load_section_id
== section_id
) {
1909 error_report("Unknown savevm section %d", section_id
);
1913 ret
= vmstate_load(f
, se
);
1915 error_report("error while loading state section id %d(%s)",
1916 section_id
, se
->idstr
);
1919 if (!check_section_footer(f
, se
)) {
1926 static int qemu_loadvm_state_main(QEMUFile
*f
, MigrationIncomingState
*mis
)
1928 uint8_t section_type
;
1931 while ((section_type
= qemu_get_byte(f
)) != QEMU_VM_EOF
) {
1933 trace_qemu_loadvm_state_section(section_type
);
1934 switch (section_type
) {
1935 case QEMU_VM_SECTION_START
:
1936 case QEMU_VM_SECTION_FULL
:
1937 ret
= qemu_loadvm_section_start_full(f
, mis
);
1942 case QEMU_VM_SECTION_PART
:
1943 case QEMU_VM_SECTION_END
:
1944 ret
= qemu_loadvm_section_part_end(f
, mis
);
1949 case QEMU_VM_COMMAND
:
1950 ret
= loadvm_process_command(f
);
1951 trace_qemu_loadvm_state_section_command(ret
);
1952 if ((ret
< 0) || (ret
& LOADVM_QUIT
)) {
1957 error_report("Unknown savevm section type %d", section_type
);
1965 qemu_file_set_error(f
, ret
);
1970 int qemu_loadvm_state(QEMUFile
*f
)
1972 MigrationIncomingState
*mis
= migration_incoming_get_current();
1973 Error
*local_err
= NULL
;
1977 if (qemu_savevm_state_blocked(&local_err
)) {
1978 error_report_err(local_err
);
1982 v
= qemu_get_be32(f
);
1983 if (v
!= QEMU_VM_FILE_MAGIC
) {
1984 error_report("Not a migration stream");
1988 v
= qemu_get_be32(f
);
1989 if (v
== QEMU_VM_FILE_VERSION_COMPAT
) {
1990 error_report("SaveVM v2 format is obsolete and don't work anymore");
1993 if (v
!= QEMU_VM_FILE_VERSION
) {
1994 error_report("Unsupported migration stream version");
1998 if (!savevm_state
.skip_configuration
|| enforce_config_section()) {
1999 if (qemu_get_byte(f
) != QEMU_VM_CONFIGURATION
) {
2000 error_report("Configuration section missing");
2003 ret
= vmstate_load_state(f
, &vmstate_configuration
, &savevm_state
, 0);
2010 cpu_synchronize_all_pre_loadvm();
2012 ret
= qemu_loadvm_state_main(f
, mis
);
2013 qemu_event_set(&mis
->main_thread_load_event
);
2015 trace_qemu_loadvm_state_post_main(ret
);
2017 if (mis
->have_listen_thread
) {
2018 /* Listen thread still going, can't clean up yet */
2023 ret
= qemu_file_get_error(f
);
2027 * Try to read in the VMDESC section as well, so that dumping tools that
2028 * intercept our migration stream have the chance to see it.
2031 /* We've got to be careful; if we don't read the data and just shut the fd
2032 * then the sender can error if we close while it's still sending.
2033 * We also mustn't read data that isn't there; some transports (RDMA)
2034 * will stall waiting for that data when the source has already closed.
2036 if (ret
== 0 && should_send_vmdesc()) {
2039 uint8_t section_type
= qemu_get_byte(f
);
2041 if (section_type
!= QEMU_VM_VMDESCRIPTION
) {
2042 error_report("Expected vmdescription section, but got %d",
2045 * It doesn't seem worth failing at this point since
2046 * we apparently have an otherwise valid VM state
2049 buf
= g_malloc(0x1000);
2050 size
= qemu_get_be32(f
);
2053 uint32_t read_chunk
= MIN(size
, 0x1000);
2054 qemu_get_buffer(f
, buf
, read_chunk
);
2061 cpu_synchronize_all_post_init();
2066 int save_snapshot(const char *name
, Error
**errp
)
2068 BlockDriverState
*bs
, *bs1
;
2069 QEMUSnapshotInfo sn1
, *sn
= &sn1
, old_sn1
, *old_sn
= &old_sn1
;
2072 int saved_vm_running
;
2073 uint64_t vm_state_size
;
2076 AioContext
*aio_context
;
2078 if (!bdrv_all_can_snapshot(&bs
)) {
2079 error_setg(errp
, "Device '%s' is writable but does not support "
2080 "snapshots", bdrv_get_device_name(bs
));
2084 /* Delete old snapshots of the same name */
2086 ret
= bdrv_all_delete_snapshot(name
, &bs1
, errp
);
2088 error_prepend(errp
, "Error while deleting snapshot on device "
2089 "'%s': ", bdrv_get_device_name(bs1
));
2094 bs
= bdrv_all_find_vmstate_bs();
2096 error_setg(errp
, "No block device can accept snapshots");
2099 aio_context
= bdrv_get_aio_context(bs
);
2101 saved_vm_running
= runstate_is_running();
2103 ret
= global_state_store();
2105 error_setg(errp
, "Error saving global state");
2108 vm_stop(RUN_STATE_SAVE_VM
);
2110 bdrv_drain_all_begin();
2112 aio_context_acquire(aio_context
);
2114 memset(sn
, 0, sizeof(*sn
));
2116 /* fill auxiliary fields */
2117 qemu_gettimeofday(&tv
);
2118 sn
->date_sec
= tv
.tv_sec
;
2119 sn
->date_nsec
= tv
.tv_usec
* 1000;
2120 sn
->vm_clock_nsec
= qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
);
2123 ret
= bdrv_snapshot_find(bs
, old_sn
, name
);
2125 pstrcpy(sn
->name
, sizeof(sn
->name
), old_sn
->name
);
2126 pstrcpy(sn
->id_str
, sizeof(sn
->id_str
), old_sn
->id_str
);
2128 pstrcpy(sn
->name
, sizeof(sn
->name
), name
);
2131 /* cast below needed for OpenBSD where tv_sec is still 'long' */
2132 localtime_r((const time_t *)&tv
.tv_sec
, &tm
);
2133 strftime(sn
->name
, sizeof(sn
->name
), "vm-%Y%m%d%H%M%S", &tm
);
2136 /* save the VM state */
2137 f
= qemu_fopen_bdrv(bs
, 1);
2139 error_setg(errp
, "Could not open VM state file");
2142 ret
= qemu_savevm_state(f
, errp
);
2143 vm_state_size
= qemu_ftell(f
);
2149 /* The bdrv_all_create_snapshot() call that follows acquires the AioContext
2150 * for itself. BDRV_POLL_WHILE() does not support nested locking because
2151 * it only releases the lock once. Therefore synchronous I/O will deadlock
2152 * unless we release the AioContext before bdrv_all_create_snapshot().
2154 aio_context_release(aio_context
);
2157 ret
= bdrv_all_create_snapshot(sn
, bs
, vm_state_size
, &bs
);
2159 error_setg(errp
, "Error while creating snapshot on '%s'",
2160 bdrv_get_device_name(bs
));
2168 aio_context_release(aio_context
);
2171 bdrv_drain_all_end();
2173 if (saved_vm_running
) {
2179 void qmp_xen_save_devices_state(const char *filename
, Error
**errp
)
2182 QIOChannelFile
*ioc
;
2183 int saved_vm_running
;
2186 saved_vm_running
= runstate_is_running();
2187 vm_stop(RUN_STATE_SAVE_VM
);
2188 global_state_store_running();
2190 ioc
= qio_channel_file_new_path(filename
, O_WRONLY
| O_CREAT
, 0660, errp
);
2194 qio_channel_set_name(QIO_CHANNEL(ioc
), "migration-xen-save-state");
2195 f
= qemu_fopen_channel_output(QIO_CHANNEL(ioc
));
2196 ret
= qemu_save_device_state(f
);
2199 error_setg(errp
, QERR_IO_ERROR
);
2203 if (saved_vm_running
) {
2208 void qmp_xen_load_devices_state(const char *filename
, Error
**errp
)
2211 QIOChannelFile
*ioc
;
2214 /* Guest must be paused before loading the device state; the RAM state
2215 * will already have been loaded by xc
2217 if (runstate_is_running()) {
2218 error_setg(errp
, "Cannot update device state while vm is running");
2221 vm_stop(RUN_STATE_RESTORE_VM
);
2223 ioc
= qio_channel_file_new_path(filename
, O_RDONLY
| O_BINARY
, 0, errp
);
2227 qio_channel_set_name(QIO_CHANNEL(ioc
), "migration-xen-load-state");
2228 f
= qemu_fopen_channel_input(QIO_CHANNEL(ioc
));
2230 ret
= qemu_loadvm_state(f
);
2233 error_setg(errp
, QERR_IO_ERROR
);
2235 migration_incoming_state_destroy();
2238 int load_snapshot(const char *name
, Error
**errp
)
2240 BlockDriverState
*bs
, *bs_vm_state
;
2241 QEMUSnapshotInfo sn
;
2244 AioContext
*aio_context
;
2245 MigrationIncomingState
*mis
= migration_incoming_get_current();
2247 if (!bdrv_all_can_snapshot(&bs
)) {
2249 "Device '%s' is writable but does not support snapshots",
2250 bdrv_get_device_name(bs
));
2253 ret
= bdrv_all_find_snapshot(name
, &bs
);
2256 "Device '%s' does not have the requested snapshot '%s'",
2257 bdrv_get_device_name(bs
), name
);
2261 bs_vm_state
= bdrv_all_find_vmstate_bs();
2263 error_setg(errp
, "No block device supports snapshots");
2266 aio_context
= bdrv_get_aio_context(bs_vm_state
);
2268 /* Don't even try to load empty VM states */
2269 aio_context_acquire(aio_context
);
2270 ret
= bdrv_snapshot_find(bs_vm_state
, &sn
, name
);
2271 aio_context_release(aio_context
);
2274 } else if (sn
.vm_state_size
== 0) {
2275 error_setg(errp
, "This is a disk-only snapshot. Revert to it "
2276 " offline using qemu-img");
2280 /* Flush all IO requests so they don't interfere with the new state. */
2281 bdrv_drain_all_begin();
2283 ret
= bdrv_all_goto_snapshot(name
, &bs
);
2285 error_setg(errp
, "Error %d while activating snapshot '%s' on '%s'",
2286 ret
, name
, bdrv_get_device_name(bs
));
2290 /* restore the VM state */
2291 f
= qemu_fopen_bdrv(bs_vm_state
, 0);
2293 error_setg(errp
, "Could not open VM state file");
2298 qemu_system_reset(SHUTDOWN_CAUSE_NONE
);
2299 mis
->from_src_file
= f
;
2301 aio_context_acquire(aio_context
);
2302 ret
= qemu_loadvm_state(f
);
2303 migration_incoming_state_destroy();
2304 aio_context_release(aio_context
);
2306 bdrv_drain_all_end();
2309 error_setg(errp
, "Error %d while loading VM state", ret
);
2316 bdrv_drain_all_end();
2320 void vmstate_register_ram(MemoryRegion
*mr
, DeviceState
*dev
)
2322 qemu_ram_set_idstr(mr
->ram_block
,
2323 memory_region_name(mr
), dev
);
2326 void vmstate_unregister_ram(MemoryRegion
*mr
, DeviceState
*dev
)
2328 qemu_ram_unset_idstr(mr
->ram_block
);
2331 void vmstate_register_ram_global(MemoryRegion
*mr
)
2333 vmstate_register_ram(mr
, NULL
);
2336 bool vmstate_check_only_migratable(const VMStateDescription
*vmsd
)
2338 /* check needed if --only-migratable is specified */
2339 if (!only_migratable
) {
2343 return !(vmsd
&& vmsd
->unmigratable
);