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/error.h"
44 #include "qapi/qapi-commands-migration.h"
45 #include "qapi/qapi-commands-misc.h"
46 #include "qapi/qmp/qerror.h"
47 #include "qemu/error-report.h"
48 #include "sysemu/cpus.h"
49 #include "exec/memory.h"
50 #include "exec/target_page.h"
53 #include "block/snapshot.h"
54 #include "qemu/cutils.h"
55 #include "io/channel-buffer.h"
56 #include "io/channel-file.h"
57 #include "sysemu/replay.h"
59 #include "migration/colo.h"
62 #define ETH_P_RARP 0x8035
64 #define ARP_HTYPE_ETH 0x0001
65 #define ARP_PTYPE_IP 0x0800
66 #define ARP_OP_REQUEST_REV 0x3
68 const unsigned int postcopy_ram_discard_version
= 0;
70 /* Subcommands for QEMU_VM_COMMAND */
72 MIG_CMD_INVALID
= 0, /* Must be 0 */
73 MIG_CMD_OPEN_RETURN_PATH
, /* Tell the dest to open the Return path */
74 MIG_CMD_PING
, /* Request a PONG on the RP */
76 MIG_CMD_POSTCOPY_ADVISE
, /* Prior to any page transfers, just
77 warn we might want to do PC */
78 MIG_CMD_POSTCOPY_LISTEN
, /* Start listening for incoming
79 pages as it's running. */
80 MIG_CMD_POSTCOPY_RUN
, /* Start execution */
82 MIG_CMD_POSTCOPY_RAM_DISCARD
, /* A list of pages to discard that
83 were previously sent during
84 precopy but are dirty. */
85 MIG_CMD_PACKAGED
, /* Send a wrapped stream within this stream */
86 MIG_CMD_ENABLE_COLO
, /* Enable COLO */
87 MIG_CMD_POSTCOPY_RESUME
, /* resume postcopy on dest */
88 MIG_CMD_RECV_BITMAP
, /* Request for recved bitmap on dst */
92 #define MAX_VM_CMD_PACKAGED_SIZE UINT32_MAX
93 static struct mig_cmd_args
{
94 ssize_t len
; /* -1 = variable */
97 [MIG_CMD_INVALID
] = { .len
= -1, .name
= "INVALID" },
98 [MIG_CMD_OPEN_RETURN_PATH
] = { .len
= 0, .name
= "OPEN_RETURN_PATH" },
99 [MIG_CMD_PING
] = { .len
= sizeof(uint32_t), .name
= "PING" },
100 [MIG_CMD_POSTCOPY_ADVISE
] = { .len
= -1, .name
= "POSTCOPY_ADVISE" },
101 [MIG_CMD_POSTCOPY_LISTEN
] = { .len
= 0, .name
= "POSTCOPY_LISTEN" },
102 [MIG_CMD_POSTCOPY_RUN
] = { .len
= 0, .name
= "POSTCOPY_RUN" },
103 [MIG_CMD_POSTCOPY_RAM_DISCARD
] = {
104 .len
= -1, .name
= "POSTCOPY_RAM_DISCARD" },
105 [MIG_CMD_POSTCOPY_RESUME
] = { .len
= 0, .name
= "POSTCOPY_RESUME" },
106 [MIG_CMD_PACKAGED
] = { .len
= 4, .name
= "PACKAGED" },
107 [MIG_CMD_RECV_BITMAP
] = { .len
= -1, .name
= "RECV_BITMAP" },
108 [MIG_CMD_MAX
] = { .len
= -1, .name
= "MAX" },
111 /* Note for MIG_CMD_POSTCOPY_ADVISE:
112 * The format of arguments is depending on postcopy mode:
113 * - postcopy RAM only
114 * uint64_t host page size
115 * uint64_t taget page size
117 * - postcopy RAM and postcopy dirty bitmaps
118 * format is the same as for postcopy RAM only
120 * - postcopy dirty bitmaps only
121 * Nothing. Command length field is 0.
123 * Be careful: adding a new postcopy entity with some other parameters should
124 * not break format self-description ability. Good way is to introduce some
125 * generic extendable format with an exception for two old entities.
128 static int announce_self_create(uint8_t *buf
,
131 /* Ethernet header. */
132 memset(buf
, 0xff, 6); /* destination MAC addr */
133 memcpy(buf
+ 6, mac_addr
, 6); /* source MAC addr */
134 *(uint16_t *)(buf
+ 12) = htons(ETH_P_RARP
); /* ethertype */
137 *(uint16_t *)(buf
+ 14) = htons(ARP_HTYPE_ETH
); /* hardware addr space */
138 *(uint16_t *)(buf
+ 16) = htons(ARP_PTYPE_IP
); /* protocol addr space */
139 *(buf
+ 18) = 6; /* hardware addr length (ethernet) */
140 *(buf
+ 19) = 4; /* protocol addr length (IPv4) */
141 *(uint16_t *)(buf
+ 20) = htons(ARP_OP_REQUEST_REV
); /* opcode */
142 memcpy(buf
+ 22, mac_addr
, 6); /* source hw addr */
143 memset(buf
+ 28, 0x00, 4); /* source protocol addr */
144 memcpy(buf
+ 32, mac_addr
, 6); /* target hw addr */
145 memset(buf
+ 38, 0x00, 4); /* target protocol addr */
147 /* Padding to get up to 60 bytes (ethernet min packet size, minus FCS). */
148 memset(buf
+ 42, 0x00, 18);
150 return 60; /* len (FCS will be added by hardware) */
153 static void qemu_announce_self_iter(NICState
*nic
, void *opaque
)
158 trace_qemu_announce_self_iter(qemu_ether_ntoa(&nic
->conf
->macaddr
));
159 len
= announce_self_create(buf
, nic
->conf
->macaddr
.a
);
161 qemu_send_packet_raw(qemu_get_queue(nic
), buf
, len
);
165 static void qemu_announce_self_once(void *opaque
)
167 static int count
= SELF_ANNOUNCE_ROUNDS
;
168 QEMUTimer
*timer
= *(QEMUTimer
**)opaque
;
170 qemu_foreach_nic(qemu_announce_self_iter
, NULL
);
173 /* delay 50ms, 150ms, 250ms, ... */
174 timer_mod(timer
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) +
175 self_announce_delay(count
));
182 void qemu_announce_self(void)
184 static QEMUTimer
*timer
;
185 timer
= timer_new_ms(QEMU_CLOCK_REALTIME
, qemu_announce_self_once
, &timer
);
186 qemu_announce_self_once(&timer
);
189 /***********************************************************/
190 /* savevm/loadvm support */
192 static ssize_t
block_writev_buffer(void *opaque
, struct iovec
*iov
, int iovcnt
,
198 qemu_iovec_init_external(&qiov
, iov
, iovcnt
);
199 ret
= bdrv_writev_vmstate(opaque
, &qiov
, pos
);
207 static ssize_t
block_get_buffer(void *opaque
, uint8_t *buf
, int64_t pos
,
210 return bdrv_load_vmstate(opaque
, buf
, pos
, size
);
213 static int bdrv_fclose(void *opaque
)
215 return bdrv_flush(opaque
);
218 static const QEMUFileOps bdrv_read_ops
= {
219 .get_buffer
= block_get_buffer
,
223 static const QEMUFileOps bdrv_write_ops
= {
224 .writev_buffer
= block_writev_buffer
,
228 static QEMUFile
*qemu_fopen_bdrv(BlockDriverState
*bs
, int is_writable
)
231 return qemu_fopen_ops(bs
, &bdrv_write_ops
);
233 return qemu_fopen_ops(bs
, &bdrv_read_ops
);
237 /* QEMUFile timer support.
238 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
241 void timer_put(QEMUFile
*f
, QEMUTimer
*ts
)
243 uint64_t expire_time
;
245 expire_time
= timer_expire_time_ns(ts
);
246 qemu_put_be64(f
, expire_time
);
249 void timer_get(QEMUFile
*f
, QEMUTimer
*ts
)
251 uint64_t expire_time
;
253 expire_time
= qemu_get_be64(f
);
254 if (expire_time
!= -1) {
255 timer_mod_ns(ts
, expire_time
);
262 /* VMState timer support.
263 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
266 static int get_timer(QEMUFile
*f
, void *pv
, size_t size
, VMStateField
*field
)
273 static int put_timer(QEMUFile
*f
, void *pv
, size_t size
, VMStateField
*field
,
282 const VMStateInfo vmstate_info_timer
= {
289 typedef struct CompatEntry
{
294 typedef struct SaveStateEntry
{
295 QTAILQ_ENTRY(SaveStateEntry
) entry
;
300 /* version id read from the stream */
303 /* section id read from the stream */
306 const VMStateDescription
*vmsd
;
312 typedef struct SaveState
{
313 QTAILQ_HEAD(, SaveStateEntry
) handlers
;
314 int global_section_id
;
317 uint32_t target_page_bits
;
320 static SaveState savevm_state
= {
321 .handlers
= QTAILQ_HEAD_INITIALIZER(savevm_state
.handlers
),
322 .global_section_id
= 0,
325 static int configuration_pre_save(void *opaque
)
327 SaveState
*state
= opaque
;
328 const char *current_name
= MACHINE_GET_CLASS(current_machine
)->name
;
330 state
->len
= strlen(current_name
);
331 state
->name
= current_name
;
332 state
->target_page_bits
= qemu_target_page_bits();
337 static int configuration_pre_load(void *opaque
)
339 SaveState
*state
= opaque
;
341 /* If there is no target-page-bits subsection it means the source
342 * predates the variable-target-page-bits support and is using the
343 * minimum possible value for this CPU.
345 state
->target_page_bits
= qemu_target_page_bits_min();
349 static int configuration_post_load(void *opaque
, int version_id
)
351 SaveState
*state
= opaque
;
352 const char *current_name
= MACHINE_GET_CLASS(current_machine
)->name
;
354 if (strncmp(state
->name
, current_name
, state
->len
) != 0) {
355 error_report("Machine type received is '%.*s' and local is '%s'",
356 (int) state
->len
, state
->name
, current_name
);
360 if (state
->target_page_bits
!= qemu_target_page_bits()) {
361 error_report("Received TARGET_PAGE_BITS is %d but local is %d",
362 state
->target_page_bits
, qemu_target_page_bits());
369 /* The target-page-bits subsection is present only if the
370 * target page size is not the same as the default (ie the
371 * minimum page size for a variable-page-size guest CPU).
372 * If it is present then it contains the actual target page
373 * bits for the machine, and migration will fail if the
374 * two ends don't agree about it.
376 static bool vmstate_target_page_bits_needed(void *opaque
)
378 return qemu_target_page_bits()
379 > qemu_target_page_bits_min();
382 static const VMStateDescription vmstate_target_page_bits
= {
383 .name
= "configuration/target-page-bits",
385 .minimum_version_id
= 1,
386 .needed
= vmstate_target_page_bits_needed
,
387 .fields
= (VMStateField
[]) {
388 VMSTATE_UINT32(target_page_bits
, SaveState
),
389 VMSTATE_END_OF_LIST()
393 static const VMStateDescription vmstate_configuration
= {
394 .name
= "configuration",
396 .pre_load
= configuration_pre_load
,
397 .post_load
= configuration_post_load
,
398 .pre_save
= configuration_pre_save
,
399 .fields
= (VMStateField
[]) {
400 VMSTATE_UINT32(len
, SaveState
),
401 VMSTATE_VBUFFER_ALLOC_UINT32(name
, SaveState
, 0, NULL
, len
),
402 VMSTATE_END_OF_LIST()
404 .subsections
= (const VMStateDescription
*[]) {
405 &vmstate_target_page_bits
,
410 static void dump_vmstate_vmsd(FILE *out_file
,
411 const VMStateDescription
*vmsd
, int indent
,
414 static void dump_vmstate_vmsf(FILE *out_file
, const VMStateField
*field
,
417 fprintf(out_file
, "%*s{\n", indent
, "");
419 fprintf(out_file
, "%*s\"field\": \"%s\",\n", indent
, "", field
->name
);
420 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
422 fprintf(out_file
, "%*s\"field_exists\": %s,\n", indent
, "",
423 field
->field_exists
? "true" : "false");
424 fprintf(out_file
, "%*s\"size\": %zu", indent
, "", field
->size
);
425 if (field
->vmsd
!= NULL
) {
426 fprintf(out_file
, ",\n");
427 dump_vmstate_vmsd(out_file
, field
->vmsd
, indent
, false);
429 fprintf(out_file
, "\n%*s}", indent
- 2, "");
432 static void dump_vmstate_vmss(FILE *out_file
,
433 const VMStateDescription
**subsection
,
436 if (*subsection
!= NULL
) {
437 dump_vmstate_vmsd(out_file
, *subsection
, indent
, true);
441 static void dump_vmstate_vmsd(FILE *out_file
,
442 const VMStateDescription
*vmsd
, int indent
,
446 fprintf(out_file
, "%*s{\n", indent
, "");
448 fprintf(out_file
, "%*s\"%s\": {\n", indent
, "", "Description");
451 fprintf(out_file
, "%*s\"name\": \"%s\",\n", indent
, "", vmsd
->name
);
452 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
454 fprintf(out_file
, "%*s\"minimum_version_id\": %d", indent
, "",
455 vmsd
->minimum_version_id
);
456 if (vmsd
->fields
!= NULL
) {
457 const VMStateField
*field
= vmsd
->fields
;
460 fprintf(out_file
, ",\n%*s\"Fields\": [\n", indent
, "");
462 while (field
->name
!= NULL
) {
463 if (field
->flags
& VMS_MUST_EXIST
) {
464 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
469 fprintf(out_file
, ",\n");
471 dump_vmstate_vmsf(out_file
, field
, indent
+ 2);
475 fprintf(out_file
, "\n%*s]", indent
, "");
477 if (vmsd
->subsections
!= NULL
) {
478 const VMStateDescription
**subsection
= vmsd
->subsections
;
481 fprintf(out_file
, ",\n%*s\"Subsections\": [\n", indent
, "");
483 while (*subsection
!= NULL
) {
485 fprintf(out_file
, ",\n");
487 dump_vmstate_vmss(out_file
, subsection
, indent
+ 2);
491 fprintf(out_file
, "\n%*s]", indent
, "");
493 fprintf(out_file
, "\n%*s}", indent
- 2, "");
496 static void dump_machine_type(FILE *out_file
)
500 mc
= MACHINE_GET_CLASS(current_machine
);
502 fprintf(out_file
, " \"vmschkmachine\": {\n");
503 fprintf(out_file
, " \"Name\": \"%s\"\n", mc
->name
);
504 fprintf(out_file
, " },\n");
507 void dump_vmstate_json_to_file(FILE *out_file
)
512 fprintf(out_file
, "{\n");
513 dump_machine_type(out_file
);
516 list
= object_class_get_list(TYPE_DEVICE
, true);
517 for (elt
= list
; elt
; elt
= elt
->next
) {
518 DeviceClass
*dc
= OBJECT_CLASS_CHECK(DeviceClass
, elt
->data
,
528 fprintf(out_file
, ",\n");
530 name
= object_class_get_name(OBJECT_CLASS(dc
));
531 fprintf(out_file
, "%*s\"%s\": {\n", indent
, "", name
);
533 fprintf(out_file
, "%*s\"Name\": \"%s\",\n", indent
, "", name
);
534 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
535 dc
->vmsd
->version_id
);
536 fprintf(out_file
, "%*s\"minimum_version_id\": %d,\n", indent
, "",
537 dc
->vmsd
->minimum_version_id
);
539 dump_vmstate_vmsd(out_file
, dc
->vmsd
, indent
, false);
541 fprintf(out_file
, "\n%*s}", indent
- 2, "");
544 fprintf(out_file
, "\n}\n");
548 static int calculate_new_instance_id(const char *idstr
)
553 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
554 if (strcmp(idstr
, se
->idstr
) == 0
555 && instance_id
<= se
->instance_id
) {
556 instance_id
= se
->instance_id
+ 1;
562 static int calculate_compat_instance_id(const char *idstr
)
567 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
572 if (strcmp(idstr
, se
->compat
->idstr
) == 0
573 && instance_id
<= se
->compat
->instance_id
) {
574 instance_id
= se
->compat
->instance_id
+ 1;
580 static inline MigrationPriority
save_state_priority(SaveStateEntry
*se
)
583 return se
->vmsd
->priority
;
585 return MIG_PRI_DEFAULT
;
588 static void savevm_state_handler_insert(SaveStateEntry
*nse
)
590 MigrationPriority priority
= save_state_priority(nse
);
593 assert(priority
<= MIG_PRI_MAX
);
595 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
596 if (save_state_priority(se
) < priority
) {
602 QTAILQ_INSERT_BEFORE(se
, nse
, entry
);
604 QTAILQ_INSERT_TAIL(&savevm_state
.handlers
, nse
, entry
);
608 /* TODO: Individual devices generally have very little idea about the rest
609 of the system, so instance_id should be removed/replaced.
610 Meanwhile pass -1 as instance_id if you do not already have a clearly
611 distinguishing id for all instances of your device class. */
612 int register_savevm_live(DeviceState
*dev
,
621 se
= g_new0(SaveStateEntry
, 1);
622 se
->version_id
= version_id
;
623 se
->section_id
= savevm_state
.global_section_id
++;
627 /* if this is a live_savem then set is_ram */
628 if (ops
->save_setup
!= NULL
) {
633 char *id
= qdev_get_dev_path(dev
);
635 if (snprintf(se
->idstr
, sizeof(se
->idstr
), "%s/", id
) >=
637 error_report("Path too long for VMState (%s)", id
);
645 se
->compat
= g_new0(CompatEntry
, 1);
646 pstrcpy(se
->compat
->idstr
, sizeof(se
->compat
->idstr
), idstr
);
647 se
->compat
->instance_id
= instance_id
== -1 ?
648 calculate_compat_instance_id(idstr
) : instance_id
;
652 pstrcat(se
->idstr
, sizeof(se
->idstr
), idstr
);
654 if (instance_id
== -1) {
655 se
->instance_id
= calculate_new_instance_id(se
->idstr
);
657 se
->instance_id
= instance_id
;
659 assert(!se
->compat
|| se
->instance_id
== 0);
660 savevm_state_handler_insert(se
);
664 void unregister_savevm(DeviceState
*dev
, const char *idstr
, void *opaque
)
666 SaveStateEntry
*se
, *new_se
;
670 char *path
= qdev_get_dev_path(dev
);
672 pstrcpy(id
, sizeof(id
), path
);
673 pstrcat(id
, sizeof(id
), "/");
677 pstrcat(id
, sizeof(id
), idstr
);
679 QTAILQ_FOREACH_SAFE(se
, &savevm_state
.handlers
, entry
, new_se
) {
680 if (strcmp(se
->idstr
, id
) == 0 && se
->opaque
== opaque
) {
681 QTAILQ_REMOVE(&savevm_state
.handlers
, se
, entry
);
688 int vmstate_register_with_alias_id(DeviceState
*dev
, int instance_id
,
689 const VMStateDescription
*vmsd
,
690 void *opaque
, int alias_id
,
691 int required_for_version
,
696 /* If this triggers, alias support can be dropped for the vmsd. */
697 assert(alias_id
== -1 || required_for_version
>= vmsd
->minimum_version_id
);
699 se
= g_new0(SaveStateEntry
, 1);
700 se
->version_id
= vmsd
->version_id
;
701 se
->section_id
= savevm_state
.global_section_id
++;
704 se
->alias_id
= alias_id
;
707 char *id
= qdev_get_dev_path(dev
);
709 if (snprintf(se
->idstr
, sizeof(se
->idstr
), "%s/", id
) >=
711 error_setg(errp
, "Path too long for VMState (%s)", id
);
719 se
->compat
= g_new0(CompatEntry
, 1);
720 pstrcpy(se
->compat
->idstr
, sizeof(se
->compat
->idstr
), vmsd
->name
);
721 se
->compat
->instance_id
= instance_id
== -1 ?
722 calculate_compat_instance_id(vmsd
->name
) : instance_id
;
726 pstrcat(se
->idstr
, sizeof(se
->idstr
), vmsd
->name
);
728 if (instance_id
== -1) {
729 se
->instance_id
= calculate_new_instance_id(se
->idstr
);
731 se
->instance_id
= instance_id
;
733 assert(!se
->compat
|| se
->instance_id
== 0);
734 savevm_state_handler_insert(se
);
738 void vmstate_unregister(DeviceState
*dev
, const VMStateDescription
*vmsd
,
741 SaveStateEntry
*se
, *new_se
;
743 QTAILQ_FOREACH_SAFE(se
, &savevm_state
.handlers
, entry
, new_se
) {
744 if (se
->vmsd
== vmsd
&& se
->opaque
== opaque
) {
745 QTAILQ_REMOVE(&savevm_state
.handlers
, se
, entry
);
752 static int vmstate_load(QEMUFile
*f
, SaveStateEntry
*se
)
754 trace_vmstate_load(se
->idstr
, se
->vmsd
? se
->vmsd
->name
: "(old)");
755 if (!se
->vmsd
) { /* Old style */
756 return se
->ops
->load_state(f
, se
->opaque
, se
->load_version_id
);
758 return vmstate_load_state(f
, se
->vmsd
, se
->opaque
, se
->load_version_id
);
761 static void vmstate_save_old_style(QEMUFile
*f
, SaveStateEntry
*se
, QJSON
*vmdesc
)
763 int64_t old_offset
, size
;
765 old_offset
= qemu_ftell_fast(f
);
766 se
->ops
->save_state(f
, se
->opaque
);
767 size
= qemu_ftell_fast(f
) - old_offset
;
770 json_prop_int(vmdesc
, "size", size
);
771 json_start_array(vmdesc
, "fields");
772 json_start_object(vmdesc
, NULL
);
773 json_prop_str(vmdesc
, "name", "data");
774 json_prop_int(vmdesc
, "size", size
);
775 json_prop_str(vmdesc
, "type", "buffer");
776 json_end_object(vmdesc
);
777 json_end_array(vmdesc
);
781 static int vmstate_save(QEMUFile
*f
, SaveStateEntry
*se
, QJSON
*vmdesc
)
783 trace_vmstate_save(se
->idstr
, se
->vmsd
? se
->vmsd
->name
: "(old)");
785 vmstate_save_old_style(f
, se
, vmdesc
);
788 return vmstate_save_state(f
, se
->vmsd
, se
->opaque
, vmdesc
);
792 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
794 static void save_section_header(QEMUFile
*f
, SaveStateEntry
*se
,
795 uint8_t section_type
)
797 qemu_put_byte(f
, section_type
);
798 qemu_put_be32(f
, se
->section_id
);
800 if (section_type
== QEMU_VM_SECTION_FULL
||
801 section_type
== QEMU_VM_SECTION_START
) {
803 size_t len
= strlen(se
->idstr
);
804 qemu_put_byte(f
, len
);
805 qemu_put_buffer(f
, (uint8_t *)se
->idstr
, len
);
807 qemu_put_be32(f
, se
->instance_id
);
808 qemu_put_be32(f
, se
->version_id
);
813 * Write a footer onto device sections that catches cases misformatted device
816 static void save_section_footer(QEMUFile
*f
, SaveStateEntry
*se
)
818 if (migrate_get_current()->send_section_footer
) {
819 qemu_put_byte(f
, QEMU_VM_SECTION_FOOTER
);
820 qemu_put_be32(f
, se
->section_id
);
825 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
826 * command and associated data.
828 * @f: File to send command on
829 * @command: Command type to send
830 * @len: Length of associated data
831 * @data: Data associated with command.
833 static void qemu_savevm_command_send(QEMUFile
*f
,
834 enum qemu_vm_cmd command
,
838 trace_savevm_command_send(command
, len
);
839 qemu_put_byte(f
, QEMU_VM_COMMAND
);
840 qemu_put_be16(f
, (uint16_t)command
);
841 qemu_put_be16(f
, len
);
842 qemu_put_buffer(f
, data
, len
);
846 void qemu_savevm_send_colo_enable(QEMUFile
*f
)
848 trace_savevm_send_colo_enable();
849 qemu_savevm_command_send(f
, MIG_CMD_ENABLE_COLO
, 0, NULL
);
852 void qemu_savevm_send_ping(QEMUFile
*f
, uint32_t value
)
856 trace_savevm_send_ping(value
);
857 buf
= cpu_to_be32(value
);
858 qemu_savevm_command_send(f
, MIG_CMD_PING
, sizeof(value
), (uint8_t *)&buf
);
861 void qemu_savevm_send_open_return_path(QEMUFile
*f
)
863 trace_savevm_send_open_return_path();
864 qemu_savevm_command_send(f
, MIG_CMD_OPEN_RETURN_PATH
, 0, NULL
);
867 /* We have a buffer of data to send; we don't want that all to be loaded
868 * by the command itself, so the command contains just the length of the
869 * extra buffer that we then send straight after it.
870 * TODO: Must be a better way to organise that
876 int qemu_savevm_send_packaged(QEMUFile
*f
, const uint8_t *buf
, size_t len
)
880 if (len
> MAX_VM_CMD_PACKAGED_SIZE
) {
881 error_report("%s: Unreasonably large packaged state: %zu",
886 tmp
= cpu_to_be32(len
);
888 trace_qemu_savevm_send_packaged();
889 qemu_savevm_command_send(f
, MIG_CMD_PACKAGED
, 4, (uint8_t *)&tmp
);
891 qemu_put_buffer(f
, buf
, len
);
896 /* Send prior to any postcopy transfer */
897 void qemu_savevm_send_postcopy_advise(QEMUFile
*f
)
899 if (migrate_postcopy_ram()) {
901 tmp
[0] = cpu_to_be64(ram_pagesize_summary());
902 tmp
[1] = cpu_to_be64(qemu_target_page_size());
904 trace_qemu_savevm_send_postcopy_advise();
905 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_ADVISE
,
908 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_ADVISE
, 0, NULL
);
912 /* Sent prior to starting the destination running in postcopy, discard pages
913 * that have already been sent but redirtied on the source.
914 * CMD_POSTCOPY_RAM_DISCARD consist of:
916 * byte Length of name field (not including 0)
917 * n x byte RAM block name
918 * byte 0 terminator (just for safety)
919 * n x Byte ranges within the named RAMBlock
920 * be64 Start of the range
923 * name: RAMBlock name that these entries are part of
924 * len: Number of page entries
925 * start_list: 'len' addresses
926 * length_list: 'len' addresses
929 void qemu_savevm_send_postcopy_ram_discard(QEMUFile
*f
, const char *name
,
931 uint64_t *start_list
,
932 uint64_t *length_list
)
937 size_t name_len
= strlen(name
);
939 trace_qemu_savevm_send_postcopy_ram_discard(name
, len
);
940 assert(name_len
< 256);
941 buf
= g_malloc0(1 + 1 + name_len
+ 1 + (8 + 8) * len
);
942 buf
[0] = postcopy_ram_discard_version
;
944 memcpy(buf
+ 2, name
, name_len
);
945 tmplen
= 2 + name_len
;
946 buf
[tmplen
++] = '\0';
948 for (t
= 0; t
< len
; t
++) {
949 stq_be_p(buf
+ tmplen
, start_list
[t
]);
951 stq_be_p(buf
+ tmplen
, length_list
[t
]);
954 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RAM_DISCARD
, tmplen
, buf
);
958 /* Get the destination into a state where it can receive postcopy data. */
959 void qemu_savevm_send_postcopy_listen(QEMUFile
*f
)
961 trace_savevm_send_postcopy_listen();
962 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_LISTEN
, 0, NULL
);
965 /* Kick the destination into running */
966 void qemu_savevm_send_postcopy_run(QEMUFile
*f
)
968 trace_savevm_send_postcopy_run();
969 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RUN
, 0, NULL
);
972 void qemu_savevm_send_postcopy_resume(QEMUFile
*f
)
974 trace_savevm_send_postcopy_resume();
975 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RESUME
, 0, NULL
);
978 void qemu_savevm_send_recv_bitmap(QEMUFile
*f
, char *block_name
)
983 trace_savevm_send_recv_bitmap(block_name
);
985 buf
[0] = len
= strlen(block_name
);
986 memcpy(buf
+ 1, block_name
, len
);
988 qemu_savevm_command_send(f
, MIG_CMD_RECV_BITMAP
, len
+ 1, (uint8_t *)buf
);
991 bool qemu_savevm_state_blocked(Error
**errp
)
995 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
996 if (se
->vmsd
&& se
->vmsd
->unmigratable
) {
997 error_setg(errp
, "State blocked by non-migratable device '%s'",
1005 void qemu_savevm_state_header(QEMUFile
*f
)
1007 trace_savevm_state_header();
1008 qemu_put_be32(f
, QEMU_VM_FILE_MAGIC
);
1009 qemu_put_be32(f
, QEMU_VM_FILE_VERSION
);
1011 if (migrate_get_current()->send_configuration
) {
1012 qemu_put_byte(f
, QEMU_VM_CONFIGURATION
);
1013 vmstate_save_state(f
, &vmstate_configuration
, &savevm_state
, 0);
1017 void qemu_savevm_state_setup(QEMUFile
*f
)
1022 trace_savevm_state_setup();
1023 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1024 if (!se
->ops
|| !se
->ops
->save_setup
) {
1027 if (se
->ops
&& se
->ops
->is_active
) {
1028 if (!se
->ops
->is_active(se
->opaque
)) {
1032 save_section_header(f
, se
, QEMU_VM_SECTION_START
);
1034 ret
= se
->ops
->save_setup(f
, se
->opaque
);
1035 save_section_footer(f
, se
);
1037 qemu_file_set_error(f
, ret
);
1043 int qemu_savevm_state_resume_prepare(MigrationState
*s
)
1048 trace_savevm_state_resume_prepare();
1050 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1051 if (!se
->ops
|| !se
->ops
->resume_prepare
) {
1054 if (se
->ops
&& se
->ops
->is_active
) {
1055 if (!se
->ops
->is_active(se
->opaque
)) {
1059 ret
= se
->ops
->resume_prepare(s
, se
->opaque
);
1069 * this function has three return values:
1070 * negative: there was one error, and we have -errno.
1071 * 0 : We haven't finished, caller have to go again
1072 * 1 : We have finished, we can go to complete phase
1074 int qemu_savevm_state_iterate(QEMUFile
*f
, bool postcopy
)
1079 trace_savevm_state_iterate();
1080 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1081 if (!se
->ops
|| !se
->ops
->save_live_iterate
) {
1084 if (se
->ops
&& se
->ops
->is_active
) {
1085 if (!se
->ops
->is_active(se
->opaque
)) {
1089 if (se
->ops
&& se
->ops
->is_active_iterate
) {
1090 if (!se
->ops
->is_active_iterate(se
->opaque
)) {
1095 * In the postcopy phase, any device that doesn't know how to
1096 * do postcopy should have saved it's state in the _complete
1097 * call that's already run, it might get confused if we call
1098 * iterate afterwards.
1101 !(se
->ops
->has_postcopy
&& se
->ops
->has_postcopy(se
->opaque
))) {
1104 if (qemu_file_rate_limit(f
)) {
1107 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1109 save_section_header(f
, se
, QEMU_VM_SECTION_PART
);
1111 ret
= se
->ops
->save_live_iterate(f
, se
->opaque
);
1112 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1113 save_section_footer(f
, se
);
1116 qemu_file_set_error(f
, ret
);
1119 /* Do not proceed to the next vmstate before this one reported
1120 completion of the current stage. This serializes the migration
1121 and reduces the probability that a faster changing state is
1122 synchronized over and over again. */
1129 static bool should_send_vmdesc(void)
1131 MachineState
*machine
= MACHINE(qdev_get_machine());
1132 bool in_postcopy
= migration_in_postcopy();
1133 return !machine
->suppress_vmdesc
&& !in_postcopy
;
1137 * Calls the save_live_complete_postcopy methods
1138 * causing the last few pages to be sent immediately and doing any associated
1140 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1141 * all the other devices, but that happens at the point we switch to postcopy.
1143 void qemu_savevm_state_complete_postcopy(QEMUFile
*f
)
1148 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1149 if (!se
->ops
|| !se
->ops
->save_live_complete_postcopy
) {
1152 if (se
->ops
&& se
->ops
->is_active
) {
1153 if (!se
->ops
->is_active(se
->opaque
)) {
1157 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1159 qemu_put_byte(f
, QEMU_VM_SECTION_END
);
1160 qemu_put_be32(f
, se
->section_id
);
1162 ret
= se
->ops
->save_live_complete_postcopy(f
, se
->opaque
);
1163 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1164 save_section_footer(f
, se
);
1166 qemu_file_set_error(f
, ret
);
1171 qemu_put_byte(f
, QEMU_VM_EOF
);
1175 int qemu_savevm_state_complete_precopy(QEMUFile
*f
, bool iterable_only
,
1176 bool inactivate_disks
)
1182 bool in_postcopy
= migration_in_postcopy();
1184 trace_savevm_state_complete_precopy();
1186 cpu_synchronize_all_states();
1188 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1190 (in_postcopy
&& se
->ops
->has_postcopy
&&
1191 se
->ops
->has_postcopy(se
->opaque
)) ||
1192 (in_postcopy
&& !iterable_only
) ||
1193 !se
->ops
->save_live_complete_precopy
) {
1197 if (se
->ops
&& se
->ops
->is_active
) {
1198 if (!se
->ops
->is_active(se
->opaque
)) {
1202 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1204 save_section_header(f
, se
, QEMU_VM_SECTION_END
);
1206 ret
= se
->ops
->save_live_complete_precopy(f
, se
->opaque
);
1207 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1208 save_section_footer(f
, se
);
1210 qemu_file_set_error(f
, ret
);
1215 if (iterable_only
) {
1219 vmdesc
= qjson_new();
1220 json_prop_int(vmdesc
, "page_size", qemu_target_page_size());
1221 json_start_array(vmdesc
, "devices");
1222 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1224 if ((!se
->ops
|| !se
->ops
->save_state
) && !se
->vmsd
) {
1227 if (se
->vmsd
&& !vmstate_save_needed(se
->vmsd
, se
->opaque
)) {
1228 trace_savevm_section_skip(se
->idstr
, se
->section_id
);
1232 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1234 json_start_object(vmdesc
, NULL
);
1235 json_prop_str(vmdesc
, "name", se
->idstr
);
1236 json_prop_int(vmdesc
, "instance_id", se
->instance_id
);
1238 save_section_header(f
, se
, QEMU_VM_SECTION_FULL
);
1239 ret
= vmstate_save(f
, se
, vmdesc
);
1241 qemu_file_set_error(f
, ret
);
1244 trace_savevm_section_end(se
->idstr
, se
->section_id
, 0);
1245 save_section_footer(f
, se
);
1247 json_end_object(vmdesc
);
1250 if (inactivate_disks
) {
1251 /* Inactivate before sending QEMU_VM_EOF so that the
1252 * bdrv_invalidate_cache_all() on the other end won't fail. */
1253 ret
= bdrv_inactivate_all();
1255 error_report("%s: bdrv_inactivate_all() failed (%d)",
1257 qemu_file_set_error(f
, ret
);
1262 /* Postcopy stream will still be going */
1263 qemu_put_byte(f
, QEMU_VM_EOF
);
1266 json_end_array(vmdesc
);
1267 qjson_finish(vmdesc
);
1268 vmdesc_len
= strlen(qjson_get_str(vmdesc
));
1270 if (should_send_vmdesc()) {
1271 qemu_put_byte(f
, QEMU_VM_VMDESCRIPTION
);
1272 qemu_put_be32(f
, vmdesc_len
);
1273 qemu_put_buffer(f
, (uint8_t *)qjson_get_str(vmdesc
), vmdesc_len
);
1275 qjson_destroy(vmdesc
);
1281 /* Give an estimate of the amount left to be transferred,
1282 * the result is split into the amount for units that can and
1283 * for units that can't do postcopy.
1285 void qemu_savevm_state_pending(QEMUFile
*f
, uint64_t threshold_size
,
1286 uint64_t *res_precopy_only
,
1287 uint64_t *res_compatible
,
1288 uint64_t *res_postcopy_only
)
1292 *res_precopy_only
= 0;
1293 *res_compatible
= 0;
1294 *res_postcopy_only
= 0;
1297 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1298 if (!se
->ops
|| !se
->ops
->save_live_pending
) {
1301 if (se
->ops
&& se
->ops
->is_active
) {
1302 if (!se
->ops
->is_active(se
->opaque
)) {
1306 se
->ops
->save_live_pending(f
, se
->opaque
, threshold_size
,
1307 res_precopy_only
, res_compatible
,
1312 void qemu_savevm_state_cleanup(void)
1316 trace_savevm_state_cleanup();
1317 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1318 if (se
->ops
&& se
->ops
->save_cleanup
) {
1319 se
->ops
->save_cleanup(se
->opaque
);
1324 static int qemu_savevm_state(QEMUFile
*f
, Error
**errp
)
1327 MigrationState
*ms
= migrate_get_current();
1328 MigrationStatus status
;
1330 if (migration_is_setup_or_active(ms
->state
) ||
1331 ms
->state
== MIGRATION_STATUS_CANCELLING
||
1332 ms
->state
== MIGRATION_STATUS_COLO
) {
1333 error_setg(errp
, QERR_MIGRATION_ACTIVE
);
1337 if (migration_is_blocked(errp
)) {
1341 if (migrate_use_block()) {
1342 error_setg(errp
, "Block migration and snapshots are incompatible");
1347 ms
->to_dst_file
= f
;
1349 qemu_mutex_unlock_iothread();
1350 qemu_savevm_state_header(f
);
1351 qemu_savevm_state_setup(f
);
1352 qemu_mutex_lock_iothread();
1354 while (qemu_file_get_error(f
) == 0) {
1355 if (qemu_savevm_state_iterate(f
, false) > 0) {
1360 ret
= qemu_file_get_error(f
);
1362 qemu_savevm_state_complete_precopy(f
, false, false);
1363 ret
= qemu_file_get_error(f
);
1365 qemu_savevm_state_cleanup();
1367 error_setg_errno(errp
, -ret
, "Error while writing VM state");
1371 status
= MIGRATION_STATUS_FAILED
;
1373 status
= MIGRATION_STATUS_COMPLETED
;
1375 migrate_set_state(&ms
->state
, MIGRATION_STATUS_SETUP
, status
);
1377 /* f is outer parameter, it should not stay in global migration state after
1378 * this function finished */
1379 ms
->to_dst_file
= NULL
;
1384 void qemu_savevm_live_state(QEMUFile
*f
)
1386 /* save QEMU_VM_SECTION_END section */
1387 qemu_savevm_state_complete_precopy(f
, true, false);
1388 qemu_put_byte(f
, QEMU_VM_EOF
);
1391 int qemu_save_device_state(QEMUFile
*f
)
1395 if (!migration_in_colo_state()) {
1396 qemu_put_be32(f
, QEMU_VM_FILE_MAGIC
);
1397 qemu_put_be32(f
, QEMU_VM_FILE_VERSION
);
1399 cpu_synchronize_all_states();
1401 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1407 if ((!se
->ops
|| !se
->ops
->save_state
) && !se
->vmsd
) {
1410 if (se
->vmsd
&& !vmstate_save_needed(se
->vmsd
, se
->opaque
)) {
1414 save_section_header(f
, se
, QEMU_VM_SECTION_FULL
);
1416 ret
= vmstate_save(f
, se
, NULL
);
1421 save_section_footer(f
, se
);
1424 qemu_put_byte(f
, QEMU_VM_EOF
);
1426 return qemu_file_get_error(f
);
1429 static SaveStateEntry
*find_se(const char *idstr
, int instance_id
)
1433 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1434 if (!strcmp(se
->idstr
, idstr
) &&
1435 (instance_id
== se
->instance_id
||
1436 instance_id
== se
->alias_id
))
1438 /* Migrating from an older version? */
1439 if (strstr(se
->idstr
, idstr
) && se
->compat
) {
1440 if (!strcmp(se
->compat
->idstr
, idstr
) &&
1441 (instance_id
== se
->compat
->instance_id
||
1442 instance_id
== se
->alias_id
))
1449 enum LoadVMExitCodes
{
1450 /* Allow a command to quit all layers of nested loadvm loops */
1454 /* ------ incoming postcopy messages ------ */
1455 /* 'advise' arrives before any transfers just to tell us that a postcopy
1456 * *might* happen - it might be skipped if precopy transferred everything
1459 static int loadvm_postcopy_handle_advise(MigrationIncomingState
*mis
,
1462 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_ADVISE
);
1463 uint64_t remote_pagesize_summary
, local_pagesize_summary
, remote_tps
;
1464 Error
*local_err
= NULL
;
1466 trace_loadvm_postcopy_handle_advise();
1467 if (ps
!= POSTCOPY_INCOMING_NONE
) {
1468 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps
);
1474 if (migrate_postcopy_ram()) {
1475 error_report("RAM postcopy is enabled but have 0 byte advise");
1480 if (!migrate_postcopy_ram()) {
1481 error_report("RAM postcopy is disabled but have 16 byte advise");
1486 error_report("CMD_POSTCOPY_ADVISE invalid length (%d)", len
);
1490 if (!postcopy_ram_supported_by_host(mis
)) {
1491 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
1495 remote_pagesize_summary
= qemu_get_be64(mis
->from_src_file
);
1496 local_pagesize_summary
= ram_pagesize_summary();
1498 if (remote_pagesize_summary
!= local_pagesize_summary
) {
1500 * This detects two potential causes of mismatch:
1501 * a) A mismatch in host page sizes
1502 * Some combinations of mismatch are probably possible but it gets
1503 * a bit more complicated. In particular we need to place whole
1504 * host pages on the dest at once, and we need to ensure that we
1505 * handle dirtying to make sure we never end up sending part of
1506 * a hostpage on it's own.
1507 * b) The use of different huge page sizes on source/destination
1508 * a more fine grain test is performed during RAM block migration
1509 * but this test here causes a nice early clear failure, and
1510 * also fails when passed to an older qemu that doesn't
1513 error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1515 remote_pagesize_summary
, local_pagesize_summary
);
1519 remote_tps
= qemu_get_be64(mis
->from_src_file
);
1520 if (remote_tps
!= qemu_target_page_size()) {
1522 * Again, some differences could be dealt with, but for now keep it
1525 error_report("Postcopy needs matching target page sizes (s=%d d=%zd)",
1526 (int)remote_tps
, qemu_target_page_size());
1530 if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_ADVISE
, &local_err
)) {
1531 error_report_err(local_err
);
1535 if (ram_postcopy_incoming_init(mis
)) {
1539 postcopy_state_set(POSTCOPY_INCOMING_ADVISE
);
1544 /* After postcopy we will be told to throw some pages away since they're
1545 * dirty and will have to be demand fetched. Must happen before CPU is
1547 * There can be 0..many of these messages, each encoding multiple pages.
1549 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState
*mis
,
1554 PostcopyState ps
= postcopy_state_get();
1556 trace_loadvm_postcopy_ram_handle_discard();
1559 case POSTCOPY_INCOMING_ADVISE
:
1561 tmp
= postcopy_ram_prepare_discard(mis
);
1567 case POSTCOPY_INCOMING_DISCARD
:
1568 /* Expected state */
1572 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1576 /* We're expecting a
1578 * a RAM ID string (length byte, name, 0 term)
1579 * then at least 1 16 byte chunk
1581 if (len
< (1 + 1 + 1 + 1 + 2 * 8)) {
1582 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len
);
1586 tmp
= qemu_get_byte(mis
->from_src_file
);
1587 if (tmp
!= postcopy_ram_discard_version
) {
1588 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp
);
1592 if (!qemu_get_counted_string(mis
->from_src_file
, ramid
)) {
1593 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1596 tmp
= qemu_get_byte(mis
->from_src_file
);
1598 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp
);
1602 len
-= 3 + strlen(ramid
);
1604 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len
);
1607 trace_loadvm_postcopy_ram_handle_discard_header(ramid
, len
);
1609 uint64_t start_addr
, block_length
;
1610 start_addr
= qemu_get_be64(mis
->from_src_file
);
1611 block_length
= qemu_get_be64(mis
->from_src_file
);
1614 int ret
= ram_discard_range(ramid
, start_addr
, block_length
);
1619 trace_loadvm_postcopy_ram_handle_discard_end();
1625 * Triggered by a postcopy_listen command; this thread takes over reading
1626 * the input stream, leaving the main thread free to carry on loading the rest
1627 * of the device state (from RAM).
1628 * (TODO:This could do with being in a postcopy file - but there again it's
1629 * just another input loop, not that postcopy specific)
1631 static void *postcopy_ram_listen_thread(void *opaque
)
1633 MigrationIncomingState
*mis
= migration_incoming_get_current();
1634 QEMUFile
*f
= mis
->from_src_file
;
1637 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
1638 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1639 qemu_sem_post(&mis
->listen_thread_sem
);
1640 trace_postcopy_ram_listen_thread_start();
1642 rcu_register_thread();
1644 * Because we're a thread and not a coroutine we can't yield
1645 * in qemu_file, and thus we must be blocking now.
1647 qemu_file_set_blocking(f
, true);
1648 load_res
= qemu_loadvm_state_main(f
, mis
);
1651 * This is tricky, but, mis->from_src_file can change after it
1652 * returns, when postcopy recovery happened. In the future, we may
1653 * want a wrapper for the QEMUFile handle.
1655 f
= mis
->from_src_file
;
1657 /* And non-blocking again so we don't block in any cleanup */
1658 qemu_file_set_blocking(f
, false);
1660 trace_postcopy_ram_listen_thread_exit();
1662 error_report("%s: loadvm failed: %d", __func__
, load_res
);
1663 qemu_file_set_error(f
, load_res
);
1664 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1665 MIGRATION_STATUS_FAILED
);
1668 * This looks good, but it's possible that the device loading in the
1669 * main thread hasn't finished yet, and so we might not be in 'RUN'
1670 * state yet; wait for the end of the main thread.
1672 qemu_event_wait(&mis
->main_thread_load_event
);
1674 postcopy_ram_incoming_cleanup(mis
);
1678 * If something went wrong then we have a bad state so exit;
1679 * depending how far we got it might be possible at this point
1680 * to leave the guest running and fire MCEs for pages that never
1681 * arrived as a desperate recovery step.
1683 rcu_unregister_thread();
1687 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1688 MIGRATION_STATUS_COMPLETED
);
1690 * If everything has worked fine, then the main thread has waited
1691 * for us to start, and we're the last use of the mis.
1692 * (If something broke then qemu will have to exit anyway since it's
1693 * got a bad migration state).
1695 migration_incoming_state_destroy();
1696 qemu_loadvm_state_cleanup();
1698 rcu_unregister_thread();
1699 mis
->have_listen_thread
= false;
1703 /* After this message we must be able to immediately receive postcopy data */
1704 static int loadvm_postcopy_handle_listen(MigrationIncomingState
*mis
)
1706 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_LISTENING
);
1707 trace_loadvm_postcopy_handle_listen();
1708 Error
*local_err
= NULL
;
1710 if (ps
!= POSTCOPY_INCOMING_ADVISE
&& ps
!= POSTCOPY_INCOMING_DISCARD
) {
1711 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps
);
1714 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
1716 * A rare case, we entered listen without having to do any discards,
1717 * so do the setup that's normally done at the time of the 1st discard.
1719 if (migrate_postcopy_ram()) {
1720 postcopy_ram_prepare_discard(mis
);
1725 * Sensitise RAM - can now generate requests for blocks that don't exist
1726 * However, at this point the CPU shouldn't be running, and the IO
1727 * shouldn't be doing anything yet so don't actually expect requests
1729 if (migrate_postcopy_ram()) {
1730 if (postcopy_ram_enable_notify(mis
)) {
1735 if (postcopy_notify(POSTCOPY_NOTIFY_INBOUND_LISTEN
, &local_err
)) {
1736 error_report_err(local_err
);
1740 if (mis
->have_listen_thread
) {
1741 error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread");
1745 mis
->have_listen_thread
= true;
1746 /* Start up the listening thread and wait for it to signal ready */
1747 qemu_sem_init(&mis
->listen_thread_sem
, 0);
1748 qemu_thread_create(&mis
->listen_thread
, "postcopy/listen",
1749 postcopy_ram_listen_thread
, NULL
,
1750 QEMU_THREAD_DETACHED
);
1751 qemu_sem_wait(&mis
->listen_thread_sem
);
1752 qemu_sem_destroy(&mis
->listen_thread_sem
);
1762 static void loadvm_postcopy_handle_run_bh(void *opaque
)
1764 Error
*local_err
= NULL
;
1765 HandleRunBhData
*data
= opaque
;
1767 /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1770 cpu_synchronize_all_post_init();
1772 qemu_announce_self();
1774 /* Make sure all file formats flush their mutable metadata.
1775 * If we get an error here, just don't restart the VM yet. */
1776 bdrv_invalidate_cache_all(&local_err
);
1778 error_report_err(local_err
);
1783 trace_loadvm_postcopy_handle_run_cpu_sync();
1784 cpu_synchronize_all_post_init();
1786 trace_loadvm_postcopy_handle_run_vmstart();
1788 dirty_bitmap_mig_before_vm_start();
1791 /* Hold onto your hats, starting the CPU */
1794 /* leave it paused and let management decide when to start the CPU */
1795 runstate_set(RUN_STATE_PAUSED
);
1798 qemu_bh_delete(data
->bh
);
1802 /* After all discards we can start running and asking for pages */
1803 static int loadvm_postcopy_handle_run(MigrationIncomingState
*mis
)
1805 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_RUNNING
);
1806 HandleRunBhData
*data
;
1808 trace_loadvm_postcopy_handle_run();
1809 if (ps
!= POSTCOPY_INCOMING_LISTENING
) {
1810 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps
);
1814 data
= g_new(HandleRunBhData
, 1);
1815 data
->bh
= qemu_bh_new(loadvm_postcopy_handle_run_bh
, data
);
1816 qemu_bh_schedule(data
->bh
);
1818 /* We need to finish reading the stream from the package
1819 * and also stop reading anything more from the stream that loaded the
1820 * package (since it's now being read by the listener thread).
1821 * LOADVM_QUIT will quit all the layers of nested loadvm loops.
1826 static int loadvm_postcopy_handle_resume(MigrationIncomingState
*mis
)
1828 if (mis
->state
!= MIGRATION_STATUS_POSTCOPY_RECOVER
) {
1829 error_report("%s: illegal resume received", __func__
);
1830 /* Don't fail the load, only for this. */
1835 * This means source VM is ready to resume the postcopy migration.
1836 * It's time to switch state and release the fault thread to
1837 * continue service page faults.
1839 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_RECOVER
,
1840 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1841 qemu_sem_post(&mis
->postcopy_pause_sem_fault
);
1843 trace_loadvm_postcopy_handle_resume();
1845 /* Tell source that "we are ready" */
1846 migrate_send_rp_resume_ack(mis
, MIGRATION_RESUME_ACK_VALUE
);
1852 * Immediately following this command is a blob of data containing an embedded
1853 * chunk of migration stream; read it and load it.
1855 * @mis: Incoming state
1856 * @length: Length of packaged data to read
1858 * Returns: Negative values on error
1861 static int loadvm_handle_cmd_packaged(MigrationIncomingState
*mis
)
1865 QIOChannelBuffer
*bioc
;
1867 length
= qemu_get_be32(mis
->from_src_file
);
1868 trace_loadvm_handle_cmd_packaged(length
);
1870 if (length
> MAX_VM_CMD_PACKAGED_SIZE
) {
1871 error_report("Unreasonably large packaged state: %zu", length
);
1875 bioc
= qio_channel_buffer_new(length
);
1876 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-loadvm-buffer");
1877 ret
= qemu_get_buffer(mis
->from_src_file
,
1880 if (ret
!= length
) {
1881 object_unref(OBJECT(bioc
));
1882 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
1884 return (ret
< 0) ? ret
: -EAGAIN
;
1886 bioc
->usage
+= length
;
1887 trace_loadvm_handle_cmd_packaged_received(ret
);
1889 QEMUFile
*packf
= qemu_fopen_channel_input(QIO_CHANNEL(bioc
));
1891 ret
= qemu_loadvm_state_main(packf
, mis
);
1892 trace_loadvm_handle_cmd_packaged_main(ret
);
1894 object_unref(OBJECT(bioc
));
1900 * Handle request that source requests for recved_bitmap on
1901 * destination. Payload format:
1903 * len (1 byte) + ramblock_name (<255 bytes)
1905 static int loadvm_handle_recv_bitmap(MigrationIncomingState
*mis
,
1908 QEMUFile
*file
= mis
->from_src_file
;
1910 char block_name
[256];
1913 cnt
= qemu_get_counted_string(file
, block_name
);
1915 error_report("%s: failed to read block name", __func__
);
1919 /* Validate before using the data */
1920 if (qemu_file_get_error(file
)) {
1921 return qemu_file_get_error(file
);
1924 if (len
!= cnt
+ 1) {
1925 error_report("%s: invalid payload length (%d)", __func__
, len
);
1929 rb
= qemu_ram_block_by_name(block_name
);
1931 error_report("%s: block '%s' not found", __func__
, block_name
);
1935 migrate_send_rp_recv_bitmap(mis
, block_name
);
1937 trace_loadvm_handle_recv_bitmap(block_name
);
1942 static int loadvm_process_enable_colo(MigrationIncomingState
*mis
)
1944 migration_incoming_enable_colo();
1945 return colo_init_ram_cache();
1949 * Process an incoming 'QEMU_VM_COMMAND'
1950 * 0 just a normal return
1951 * LOADVM_QUIT All good, but exit the loop
1954 static int loadvm_process_command(QEMUFile
*f
)
1956 MigrationIncomingState
*mis
= migration_incoming_get_current();
1961 cmd
= qemu_get_be16(f
);
1962 len
= qemu_get_be16(f
);
1964 /* Check validity before continue processing of cmds */
1965 if (qemu_file_get_error(f
)) {
1966 return qemu_file_get_error(f
);
1969 trace_loadvm_process_command(cmd
, len
);
1970 if (cmd
>= MIG_CMD_MAX
|| cmd
== MIG_CMD_INVALID
) {
1971 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd
, len
);
1975 if (mig_cmd_args
[cmd
].len
!= -1 && mig_cmd_args
[cmd
].len
!= len
) {
1976 error_report("%s received with bad length - expecting %zu, got %d",
1977 mig_cmd_args
[cmd
].name
,
1978 (size_t)mig_cmd_args
[cmd
].len
, len
);
1983 case MIG_CMD_OPEN_RETURN_PATH
:
1984 if (mis
->to_src_file
) {
1985 error_report("CMD_OPEN_RETURN_PATH called when RP already open");
1986 /* Not really a problem, so don't give up */
1989 mis
->to_src_file
= qemu_file_get_return_path(f
);
1990 if (!mis
->to_src_file
) {
1991 error_report("CMD_OPEN_RETURN_PATH failed");
1997 tmp32
= qemu_get_be32(f
);
1998 trace_loadvm_process_command_ping(tmp32
);
1999 if (!mis
->to_src_file
) {
2000 error_report("CMD_PING (0x%x) received with no return path",
2004 migrate_send_rp_pong(mis
, tmp32
);
2007 case MIG_CMD_PACKAGED
:
2008 return loadvm_handle_cmd_packaged(mis
);
2010 case MIG_CMD_POSTCOPY_ADVISE
:
2011 return loadvm_postcopy_handle_advise(mis
, len
);
2013 case MIG_CMD_POSTCOPY_LISTEN
:
2014 return loadvm_postcopy_handle_listen(mis
);
2016 case MIG_CMD_POSTCOPY_RUN
:
2017 return loadvm_postcopy_handle_run(mis
);
2019 case MIG_CMD_POSTCOPY_RAM_DISCARD
:
2020 return loadvm_postcopy_ram_handle_discard(mis
, len
);
2022 case MIG_CMD_POSTCOPY_RESUME
:
2023 return loadvm_postcopy_handle_resume(mis
);
2025 case MIG_CMD_RECV_BITMAP
:
2026 return loadvm_handle_recv_bitmap(mis
, len
);
2028 case MIG_CMD_ENABLE_COLO
:
2029 return loadvm_process_enable_colo(mis
);
2036 * Read a footer off the wire and check that it matches the expected section
2038 * Returns: true if the footer was good
2039 * false if there is a problem (and calls error_report to say why)
2041 static bool check_section_footer(QEMUFile
*f
, SaveStateEntry
*se
)
2045 uint32_t read_section_id
;
2047 if (!migrate_get_current()->send_section_footer
) {
2048 /* No footer to check */
2052 read_mark
= qemu_get_byte(f
);
2054 ret
= qemu_file_get_error(f
);
2056 error_report("%s: Read section footer failed: %d",
2061 if (read_mark
!= QEMU_VM_SECTION_FOOTER
) {
2062 error_report("Missing section footer for %s", se
->idstr
);
2066 read_section_id
= qemu_get_be32(f
);
2067 if (read_section_id
!= se
->load_section_id
) {
2068 error_report("Mismatched section id in footer for %s -"
2069 " read 0x%x expected 0x%x",
2070 se
->idstr
, read_section_id
, se
->load_section_id
);
2079 qemu_loadvm_section_start_full(QEMUFile
*f
, MigrationIncomingState
*mis
)
2081 uint32_t instance_id
, version_id
, section_id
;
2086 /* Read section start */
2087 section_id
= qemu_get_be32(f
);
2088 if (!qemu_get_counted_string(f
, idstr
)) {
2089 error_report("Unable to read ID string for section %u",
2093 instance_id
= qemu_get_be32(f
);
2094 version_id
= qemu_get_be32(f
);
2096 ret
= qemu_file_get_error(f
);
2098 error_report("%s: Failed to read instance/version ID: %d",
2103 trace_qemu_loadvm_state_section_startfull(section_id
, idstr
,
2104 instance_id
, version_id
);
2105 /* Find savevm section */
2106 se
= find_se(idstr
, instance_id
);
2108 error_report("Unknown savevm section or instance '%s' %d. "
2109 "Make sure that your current VM setup matches your "
2110 "saved VM setup, including any hotplugged devices",
2111 idstr
, instance_id
);
2115 /* Validate version */
2116 if (version_id
> se
->version_id
) {
2117 error_report("savevm: unsupported version %d for '%s' v%d",
2118 version_id
, idstr
, se
->version_id
);
2121 se
->load_version_id
= version_id
;
2122 se
->load_section_id
= section_id
;
2124 /* Validate if it is a device's state */
2125 if (xen_enabled() && se
->is_ram
) {
2126 error_report("loadvm: %s RAM loading not allowed on Xen", idstr
);
2130 ret
= vmstate_load(f
, se
);
2132 error_report("error while loading state for instance 0x%x of"
2133 " device '%s'", instance_id
, idstr
);
2136 if (!check_section_footer(f
, se
)) {
2144 qemu_loadvm_section_part_end(QEMUFile
*f
, MigrationIncomingState
*mis
)
2146 uint32_t section_id
;
2150 section_id
= qemu_get_be32(f
);
2152 ret
= qemu_file_get_error(f
);
2154 error_report("%s: Failed to read section ID: %d",
2159 trace_qemu_loadvm_state_section_partend(section_id
);
2160 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
2161 if (se
->load_section_id
== section_id
) {
2166 error_report("Unknown savevm section %d", section_id
);
2170 ret
= vmstate_load(f
, se
);
2172 error_report("error while loading state section id %d(%s)",
2173 section_id
, se
->idstr
);
2176 if (!check_section_footer(f
, se
)) {
2183 static int qemu_loadvm_state_setup(QEMUFile
*f
)
2188 trace_loadvm_state_setup();
2189 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
2190 if (!se
->ops
|| !se
->ops
->load_setup
) {
2193 if (se
->ops
&& se
->ops
->is_active
) {
2194 if (!se
->ops
->is_active(se
->opaque
)) {
2199 ret
= se
->ops
->load_setup(f
, se
->opaque
);
2201 qemu_file_set_error(f
, ret
);
2202 error_report("Load state of device %s failed", se
->idstr
);
2209 void qemu_loadvm_state_cleanup(void)
2213 trace_loadvm_state_cleanup();
2214 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
2215 if (se
->ops
&& se
->ops
->load_cleanup
) {
2216 se
->ops
->load_cleanup(se
->opaque
);
2221 /* Return true if we should continue the migration, or false. */
2222 static bool postcopy_pause_incoming(MigrationIncomingState
*mis
)
2224 trace_postcopy_pause_incoming();
2226 /* Clear the triggered bit to allow one recovery */
2227 mis
->postcopy_recover_triggered
= false;
2229 assert(mis
->from_src_file
);
2230 qemu_file_shutdown(mis
->from_src_file
);
2231 qemu_fclose(mis
->from_src_file
);
2232 mis
->from_src_file
= NULL
;
2234 assert(mis
->to_src_file
);
2235 qemu_file_shutdown(mis
->to_src_file
);
2236 qemu_mutex_lock(&mis
->rp_mutex
);
2237 qemu_fclose(mis
->to_src_file
);
2238 mis
->to_src_file
= NULL
;
2239 qemu_mutex_unlock(&mis
->rp_mutex
);
2241 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
2242 MIGRATION_STATUS_POSTCOPY_PAUSED
);
2244 /* Notify the fault thread for the invalidated file handle */
2245 postcopy_fault_thread_notify(mis
);
2247 error_report("Detected IO failure for postcopy. "
2248 "Migration paused.");
2250 while (mis
->state
== MIGRATION_STATUS_POSTCOPY_PAUSED
) {
2251 qemu_sem_wait(&mis
->postcopy_pause_sem_dst
);
2254 trace_postcopy_pause_incoming_continued();
2259 int qemu_loadvm_state_main(QEMUFile
*f
, MigrationIncomingState
*mis
)
2261 uint8_t section_type
;
2266 section_type
= qemu_get_byte(f
);
2268 if (qemu_file_get_error(f
)) {
2269 ret
= qemu_file_get_error(f
);
2273 trace_qemu_loadvm_state_section(section_type
);
2274 switch (section_type
) {
2275 case QEMU_VM_SECTION_START
:
2276 case QEMU_VM_SECTION_FULL
:
2277 ret
= qemu_loadvm_section_start_full(f
, mis
);
2282 case QEMU_VM_SECTION_PART
:
2283 case QEMU_VM_SECTION_END
:
2284 ret
= qemu_loadvm_section_part_end(f
, mis
);
2289 case QEMU_VM_COMMAND
:
2290 ret
= loadvm_process_command(f
);
2291 trace_qemu_loadvm_state_section_command(ret
);
2292 if ((ret
< 0) || (ret
& LOADVM_QUIT
)) {
2297 /* This is the end of migration */
2300 error_report("Unknown savevm section type %d", section_type
);
2308 qemu_file_set_error(f
, ret
);
2311 * If we are during an active postcopy, then we pause instead
2312 * of bail out to at least keep the VM's dirty data. Note
2313 * that POSTCOPY_INCOMING_LISTENING stage is still not enough,
2314 * during which we're still receiving device states and we
2315 * still haven't yet started the VM on destination.
2317 if (postcopy_state_get() == POSTCOPY_INCOMING_RUNNING
&&
2318 postcopy_pause_incoming(mis
)) {
2319 /* Reset f to point to the newly created channel */
2320 f
= mis
->from_src_file
;
2327 int qemu_loadvm_state(QEMUFile
*f
)
2329 MigrationIncomingState
*mis
= migration_incoming_get_current();
2330 Error
*local_err
= NULL
;
2334 if (qemu_savevm_state_blocked(&local_err
)) {
2335 error_report_err(local_err
);
2339 v
= qemu_get_be32(f
);
2340 if (v
!= QEMU_VM_FILE_MAGIC
) {
2341 error_report("Not a migration stream");
2345 v
= qemu_get_be32(f
);
2346 if (v
== QEMU_VM_FILE_VERSION_COMPAT
) {
2347 error_report("SaveVM v2 format is obsolete and don't work anymore");
2350 if (v
!= QEMU_VM_FILE_VERSION
) {
2351 error_report("Unsupported migration stream version");
2355 if (qemu_loadvm_state_setup(f
) != 0) {
2359 if (migrate_get_current()->send_configuration
) {
2360 if (qemu_get_byte(f
) != QEMU_VM_CONFIGURATION
) {
2361 error_report("Configuration section missing");
2362 qemu_loadvm_state_cleanup();
2365 ret
= vmstate_load_state(f
, &vmstate_configuration
, &savevm_state
, 0);
2368 qemu_loadvm_state_cleanup();
2373 cpu_synchronize_all_pre_loadvm();
2375 ret
= qemu_loadvm_state_main(f
, mis
);
2376 qemu_event_set(&mis
->main_thread_load_event
);
2378 trace_qemu_loadvm_state_post_main(ret
);
2380 if (mis
->have_listen_thread
) {
2381 /* Listen thread still going, can't clean up yet */
2386 ret
= qemu_file_get_error(f
);
2390 * Try to read in the VMDESC section as well, so that dumping tools that
2391 * intercept our migration stream have the chance to see it.
2394 /* We've got to be careful; if we don't read the data and just shut the fd
2395 * then the sender can error if we close while it's still sending.
2396 * We also mustn't read data that isn't there; some transports (RDMA)
2397 * will stall waiting for that data when the source has already closed.
2399 if (ret
== 0 && should_send_vmdesc()) {
2402 uint8_t section_type
= qemu_get_byte(f
);
2404 if (section_type
!= QEMU_VM_VMDESCRIPTION
) {
2405 error_report("Expected vmdescription section, but got %d",
2408 * It doesn't seem worth failing at this point since
2409 * we apparently have an otherwise valid VM state
2412 buf
= g_malloc(0x1000);
2413 size
= qemu_get_be32(f
);
2416 uint32_t read_chunk
= MIN(size
, 0x1000);
2417 qemu_get_buffer(f
, buf
, read_chunk
);
2424 qemu_loadvm_state_cleanup();
2425 cpu_synchronize_all_post_init();
2430 int qemu_load_device_state(QEMUFile
*f
)
2432 MigrationIncomingState
*mis
= migration_incoming_get_current();
2435 /* Load QEMU_VM_SECTION_FULL section */
2436 ret
= qemu_loadvm_state_main(f
, mis
);
2438 error_report("Failed to load device state: %d", ret
);
2442 cpu_synchronize_all_post_init();
2446 int save_snapshot(const char *name
, Error
**errp
)
2448 BlockDriverState
*bs
, *bs1
;
2449 QEMUSnapshotInfo sn1
, *sn
= &sn1
, old_sn1
, *old_sn
= &old_sn1
;
2452 int saved_vm_running
;
2453 uint64_t vm_state_size
;
2456 AioContext
*aio_context
;
2458 if (!replay_can_snapshot()) {
2459 error_setg(errp
, "Record/replay does not allow making snapshot "
2460 "right now. Try once more later.");
2464 if (!bdrv_all_can_snapshot(&bs
)) {
2465 error_setg(errp
, "Device '%s' is writable but does not support "
2466 "snapshots", bdrv_get_device_name(bs
));
2470 /* Delete old snapshots of the same name */
2472 ret
= bdrv_all_delete_snapshot(name
, &bs1
, errp
);
2474 error_prepend(errp
, "Error while deleting snapshot on device "
2475 "'%s': ", bdrv_get_device_name(bs1
));
2480 bs
= bdrv_all_find_vmstate_bs();
2482 error_setg(errp
, "No block device can accept snapshots");
2485 aio_context
= bdrv_get_aio_context(bs
);
2487 saved_vm_running
= runstate_is_running();
2489 ret
= global_state_store();
2491 error_setg(errp
, "Error saving global state");
2494 vm_stop(RUN_STATE_SAVE_VM
);
2496 bdrv_drain_all_begin();
2498 aio_context_acquire(aio_context
);
2500 memset(sn
, 0, sizeof(*sn
));
2502 /* fill auxiliary fields */
2503 qemu_gettimeofday(&tv
);
2504 sn
->date_sec
= tv
.tv_sec
;
2505 sn
->date_nsec
= tv
.tv_usec
* 1000;
2506 sn
->vm_clock_nsec
= qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
);
2509 ret
= bdrv_snapshot_find(bs
, old_sn
, name
);
2511 pstrcpy(sn
->name
, sizeof(sn
->name
), old_sn
->name
);
2512 pstrcpy(sn
->id_str
, sizeof(sn
->id_str
), old_sn
->id_str
);
2514 pstrcpy(sn
->name
, sizeof(sn
->name
), name
);
2517 /* cast below needed for OpenBSD where tv_sec is still 'long' */
2518 localtime_r((const time_t *)&tv
.tv_sec
, &tm
);
2519 strftime(sn
->name
, sizeof(sn
->name
), "vm-%Y%m%d%H%M%S", &tm
);
2522 /* save the VM state */
2523 f
= qemu_fopen_bdrv(bs
, 1);
2525 error_setg(errp
, "Could not open VM state file");
2528 ret
= qemu_savevm_state(f
, errp
);
2529 vm_state_size
= qemu_ftell(f
);
2535 /* The bdrv_all_create_snapshot() call that follows acquires the AioContext
2536 * for itself. BDRV_POLL_WHILE() does not support nested locking because
2537 * it only releases the lock once. Therefore synchronous I/O will deadlock
2538 * unless we release the AioContext before bdrv_all_create_snapshot().
2540 aio_context_release(aio_context
);
2543 ret
= bdrv_all_create_snapshot(sn
, bs
, vm_state_size
, &bs
);
2545 error_setg(errp
, "Error while creating snapshot on '%s'",
2546 bdrv_get_device_name(bs
));
2554 aio_context_release(aio_context
);
2557 bdrv_drain_all_end();
2559 if (saved_vm_running
) {
2565 void qmp_xen_save_devices_state(const char *filename
, bool has_live
, bool live
,
2569 QIOChannelFile
*ioc
;
2570 int saved_vm_running
;
2574 /* live default to true so old version of Xen tool stack can have a
2575 * successfull live migration */
2579 saved_vm_running
= runstate_is_running();
2580 vm_stop(RUN_STATE_SAVE_VM
);
2581 global_state_store_running();
2583 ioc
= qio_channel_file_new_path(filename
, O_WRONLY
| O_CREAT
, 0660, errp
);
2587 qio_channel_set_name(QIO_CHANNEL(ioc
), "migration-xen-save-state");
2588 f
= qemu_fopen_channel_output(QIO_CHANNEL(ioc
));
2589 object_unref(OBJECT(ioc
));
2590 ret
= qemu_save_device_state(f
);
2591 if (ret
< 0 || qemu_fclose(f
) < 0) {
2592 error_setg(errp
, QERR_IO_ERROR
);
2594 /* libxl calls the QMP command "stop" before calling
2595 * "xen-save-devices-state" and in case of migration failure, libxl
2596 * would call "cont".
2597 * So call bdrv_inactivate_all (release locks) here to let the other
2598 * side of the migration take controle of the images.
2600 if (live
&& !saved_vm_running
) {
2601 ret
= bdrv_inactivate_all();
2603 error_setg(errp
, "%s: bdrv_inactivate_all() failed (%d)",
2610 if (saved_vm_running
) {
2615 void qmp_xen_load_devices_state(const char *filename
, Error
**errp
)
2618 QIOChannelFile
*ioc
;
2621 /* Guest must be paused before loading the device state; the RAM state
2622 * will already have been loaded by xc
2624 if (runstate_is_running()) {
2625 error_setg(errp
, "Cannot update device state while vm is running");
2628 vm_stop(RUN_STATE_RESTORE_VM
);
2630 ioc
= qio_channel_file_new_path(filename
, O_RDONLY
| O_BINARY
, 0, errp
);
2634 qio_channel_set_name(QIO_CHANNEL(ioc
), "migration-xen-load-state");
2635 f
= qemu_fopen_channel_input(QIO_CHANNEL(ioc
));
2636 object_unref(OBJECT(ioc
));
2638 ret
= qemu_loadvm_state(f
);
2641 error_setg(errp
, QERR_IO_ERROR
);
2643 migration_incoming_state_destroy();
2646 int load_snapshot(const char *name
, Error
**errp
)
2648 BlockDriverState
*bs
, *bs_vm_state
;
2649 QEMUSnapshotInfo sn
;
2652 AioContext
*aio_context
;
2653 MigrationIncomingState
*mis
= migration_incoming_get_current();
2655 if (!replay_can_snapshot()) {
2656 error_setg(errp
, "Record/replay does not allow loading snapshot "
2657 "right now. Try once more later.");
2661 if (!bdrv_all_can_snapshot(&bs
)) {
2663 "Device '%s' is writable but does not support snapshots",
2664 bdrv_get_device_name(bs
));
2667 ret
= bdrv_all_find_snapshot(name
, &bs
);
2670 "Device '%s' does not have the requested snapshot '%s'",
2671 bdrv_get_device_name(bs
), name
);
2675 bs_vm_state
= bdrv_all_find_vmstate_bs();
2677 error_setg(errp
, "No block device supports snapshots");
2680 aio_context
= bdrv_get_aio_context(bs_vm_state
);
2682 /* Don't even try to load empty VM states */
2683 aio_context_acquire(aio_context
);
2684 ret
= bdrv_snapshot_find(bs_vm_state
, &sn
, name
);
2685 aio_context_release(aio_context
);
2688 } else if (sn
.vm_state_size
== 0) {
2689 error_setg(errp
, "This is a disk-only snapshot. Revert to it "
2690 " offline using qemu-img");
2694 /* Flush all IO requests so they don't interfere with the new state. */
2695 bdrv_drain_all_begin();
2697 ret
= bdrv_all_goto_snapshot(name
, &bs
, errp
);
2699 error_prepend(errp
, "Could not load snapshot '%s' on '%s': ",
2700 name
, bdrv_get_device_name(bs
));
2704 /* restore the VM state */
2705 f
= qemu_fopen_bdrv(bs_vm_state
, 0);
2707 error_setg(errp
, "Could not open VM state file");
2712 qemu_system_reset(SHUTDOWN_CAUSE_NONE
);
2713 mis
->from_src_file
= f
;
2715 aio_context_acquire(aio_context
);
2716 ret
= qemu_loadvm_state(f
);
2717 migration_incoming_state_destroy();
2718 aio_context_release(aio_context
);
2720 bdrv_drain_all_end();
2723 error_setg(errp
, "Error %d while loading VM state", ret
);
2730 bdrv_drain_all_end();
2734 void vmstate_register_ram(MemoryRegion
*mr
, DeviceState
*dev
)
2736 qemu_ram_set_idstr(mr
->ram_block
,
2737 memory_region_name(mr
), dev
);
2738 qemu_ram_set_migratable(mr
->ram_block
);
2741 void vmstate_unregister_ram(MemoryRegion
*mr
, DeviceState
*dev
)
2743 qemu_ram_unset_idstr(mr
->ram_block
);
2744 qemu_ram_unset_migratable(mr
->ram_block
);
2747 void vmstate_register_ram_global(MemoryRegion
*mr
)
2749 vmstate_register_ram(mr
, NULL
);
2752 bool vmstate_check_only_migratable(const VMStateDescription
*vmsd
)
2754 /* check needed if --only-migratable is specified */
2755 if (!migrate_get_current()->only_migratable
) {
2759 return !(vmsd
&& vmsd
->unmigratable
);