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"
59 #define ETH_P_RARP 0x8035
61 #define ARP_HTYPE_ETH 0x0001
62 #define ARP_PTYPE_IP 0x0800
63 #define ARP_OP_REQUEST_REV 0x3
65 const unsigned int postcopy_ram_discard_version
= 0;
67 /* 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 UINT32_MAX
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
= -1, .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 /* Note for MIG_CMD_POSTCOPY_ADVISE:
104 * The format of arguments is depending on postcopy mode:
105 * - postcopy RAM only
106 * uint64_t host page size
107 * uint64_t taget page size
109 * - postcopy RAM and postcopy dirty bitmaps
110 * format is the same as for postcopy RAM only
112 * - postcopy dirty bitmaps only
113 * Nothing. Command length field is 0.
115 * Be careful: adding a new postcopy entity with some other parameters should
116 * not break format self-description ability. Good way is to introduce some
117 * generic extendable format with an exception for two old entities.
120 static int announce_self_create(uint8_t *buf
,
123 /* Ethernet header. */
124 memset(buf
, 0xff, 6); /* destination MAC addr */
125 memcpy(buf
+ 6, mac_addr
, 6); /* source MAC addr */
126 *(uint16_t *)(buf
+ 12) = htons(ETH_P_RARP
); /* ethertype */
129 *(uint16_t *)(buf
+ 14) = htons(ARP_HTYPE_ETH
); /* hardware addr space */
130 *(uint16_t *)(buf
+ 16) = htons(ARP_PTYPE_IP
); /* protocol addr space */
131 *(buf
+ 18) = 6; /* hardware addr length (ethernet) */
132 *(buf
+ 19) = 4; /* protocol addr length (IPv4) */
133 *(uint16_t *)(buf
+ 20) = htons(ARP_OP_REQUEST_REV
); /* opcode */
134 memcpy(buf
+ 22, mac_addr
, 6); /* source hw addr */
135 memset(buf
+ 28, 0x00, 4); /* source protocol addr */
136 memcpy(buf
+ 32, mac_addr
, 6); /* target hw addr */
137 memset(buf
+ 38, 0x00, 4); /* target protocol addr */
139 /* Padding to get up to 60 bytes (ethernet min packet size, minus FCS). */
140 memset(buf
+ 42, 0x00, 18);
142 return 60; /* len (FCS will be added by hardware) */
145 static void qemu_announce_self_iter(NICState
*nic
, void *opaque
)
150 trace_qemu_announce_self_iter(qemu_ether_ntoa(&nic
->conf
->macaddr
));
151 len
= announce_self_create(buf
, nic
->conf
->macaddr
.a
);
153 qemu_send_packet_raw(qemu_get_queue(nic
), buf
, len
);
157 static void qemu_announce_self_once(void *opaque
)
159 static int count
= SELF_ANNOUNCE_ROUNDS
;
160 QEMUTimer
*timer
= *(QEMUTimer
**)opaque
;
162 qemu_foreach_nic(qemu_announce_self_iter
, NULL
);
165 /* delay 50ms, 150ms, 250ms, ... */
166 timer_mod(timer
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) +
167 self_announce_delay(count
));
174 void qemu_announce_self(void)
176 static QEMUTimer
*timer
;
177 timer
= timer_new_ms(QEMU_CLOCK_REALTIME
, qemu_announce_self_once
, &timer
);
178 qemu_announce_self_once(&timer
);
181 /***********************************************************/
182 /* savevm/loadvm support */
184 static ssize_t
block_writev_buffer(void *opaque
, struct iovec
*iov
, int iovcnt
,
190 qemu_iovec_init_external(&qiov
, iov
, iovcnt
);
191 ret
= bdrv_writev_vmstate(opaque
, &qiov
, pos
);
199 static ssize_t
block_get_buffer(void *opaque
, uint8_t *buf
, int64_t pos
,
202 return bdrv_load_vmstate(opaque
, buf
, pos
, size
);
205 static int bdrv_fclose(void *opaque
)
207 return bdrv_flush(opaque
);
210 static const QEMUFileOps bdrv_read_ops
= {
211 .get_buffer
= block_get_buffer
,
215 static const QEMUFileOps bdrv_write_ops
= {
216 .writev_buffer
= block_writev_buffer
,
220 static QEMUFile
*qemu_fopen_bdrv(BlockDriverState
*bs
, int is_writable
)
223 return qemu_fopen_ops(bs
, &bdrv_write_ops
);
225 return qemu_fopen_ops(bs
, &bdrv_read_ops
);
229 /* QEMUFile timer support.
230 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
233 void timer_put(QEMUFile
*f
, QEMUTimer
*ts
)
235 uint64_t expire_time
;
237 expire_time
= timer_expire_time_ns(ts
);
238 qemu_put_be64(f
, expire_time
);
241 void timer_get(QEMUFile
*f
, QEMUTimer
*ts
)
243 uint64_t expire_time
;
245 expire_time
= qemu_get_be64(f
);
246 if (expire_time
!= -1) {
247 timer_mod_ns(ts
, expire_time
);
254 /* VMState timer support.
255 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
258 static int get_timer(QEMUFile
*f
, void *pv
, size_t size
, VMStateField
*field
)
265 static int put_timer(QEMUFile
*f
, void *pv
, size_t size
, VMStateField
*field
,
274 const VMStateInfo vmstate_info_timer
= {
281 typedef struct CompatEntry
{
286 typedef struct SaveStateEntry
{
287 QTAILQ_ENTRY(SaveStateEntry
) entry
;
292 /* version id read from the stream */
295 /* section id read from the stream */
298 const VMStateDescription
*vmsd
;
304 typedef struct SaveState
{
305 QTAILQ_HEAD(, SaveStateEntry
) handlers
;
306 int global_section_id
;
309 uint32_t target_page_bits
;
312 static SaveState savevm_state
= {
313 .handlers
= QTAILQ_HEAD_INITIALIZER(savevm_state
.handlers
),
314 .global_section_id
= 0,
317 static int configuration_pre_save(void *opaque
)
319 SaveState
*state
= opaque
;
320 const char *current_name
= MACHINE_GET_CLASS(current_machine
)->name
;
322 state
->len
= strlen(current_name
);
323 state
->name
= current_name
;
324 state
->target_page_bits
= qemu_target_page_bits();
329 static int configuration_pre_load(void *opaque
)
331 SaveState
*state
= opaque
;
333 /* If there is no target-page-bits subsection it means the source
334 * predates the variable-target-page-bits support and is using the
335 * minimum possible value for this CPU.
337 state
->target_page_bits
= qemu_target_page_bits_min();
341 static int configuration_post_load(void *opaque
, int version_id
)
343 SaveState
*state
= opaque
;
344 const char *current_name
= MACHINE_GET_CLASS(current_machine
)->name
;
346 if (strncmp(state
->name
, current_name
, state
->len
) != 0) {
347 error_report("Machine type received is '%.*s' and local is '%s'",
348 (int) state
->len
, state
->name
, current_name
);
352 if (state
->target_page_bits
!= qemu_target_page_bits()) {
353 error_report("Received TARGET_PAGE_BITS is %d but local is %d",
354 state
->target_page_bits
, qemu_target_page_bits());
361 /* The target-page-bits subsection is present only if the
362 * target page size is not the same as the default (ie the
363 * minimum page size for a variable-page-size guest CPU).
364 * If it is present then it contains the actual target page
365 * bits for the machine, and migration will fail if the
366 * two ends don't agree about it.
368 static bool vmstate_target_page_bits_needed(void *opaque
)
370 return qemu_target_page_bits()
371 > qemu_target_page_bits_min();
374 static const VMStateDescription vmstate_target_page_bits
= {
375 .name
= "configuration/target-page-bits",
377 .minimum_version_id
= 1,
378 .needed
= vmstate_target_page_bits_needed
,
379 .fields
= (VMStateField
[]) {
380 VMSTATE_UINT32(target_page_bits
, SaveState
),
381 VMSTATE_END_OF_LIST()
385 static const VMStateDescription vmstate_configuration
= {
386 .name
= "configuration",
388 .pre_load
= configuration_pre_load
,
389 .post_load
= configuration_post_load
,
390 .pre_save
= configuration_pre_save
,
391 .fields
= (VMStateField
[]) {
392 VMSTATE_UINT32(len
, SaveState
),
393 VMSTATE_VBUFFER_ALLOC_UINT32(name
, SaveState
, 0, NULL
, len
),
394 VMSTATE_END_OF_LIST()
396 .subsections
= (const VMStateDescription
*[]) {
397 &vmstate_target_page_bits
,
402 static void dump_vmstate_vmsd(FILE *out_file
,
403 const VMStateDescription
*vmsd
, int indent
,
406 static void dump_vmstate_vmsf(FILE *out_file
, const VMStateField
*field
,
409 fprintf(out_file
, "%*s{\n", indent
, "");
411 fprintf(out_file
, "%*s\"field\": \"%s\",\n", indent
, "", field
->name
);
412 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
414 fprintf(out_file
, "%*s\"field_exists\": %s,\n", indent
, "",
415 field
->field_exists
? "true" : "false");
416 fprintf(out_file
, "%*s\"size\": %zu", indent
, "", field
->size
);
417 if (field
->vmsd
!= NULL
) {
418 fprintf(out_file
, ",\n");
419 dump_vmstate_vmsd(out_file
, field
->vmsd
, indent
, false);
421 fprintf(out_file
, "\n%*s}", indent
- 2, "");
424 static void dump_vmstate_vmss(FILE *out_file
,
425 const VMStateDescription
**subsection
,
428 if (*subsection
!= NULL
) {
429 dump_vmstate_vmsd(out_file
, *subsection
, indent
, true);
433 static void dump_vmstate_vmsd(FILE *out_file
,
434 const VMStateDescription
*vmsd
, int indent
,
438 fprintf(out_file
, "%*s{\n", indent
, "");
440 fprintf(out_file
, "%*s\"%s\": {\n", indent
, "", "Description");
443 fprintf(out_file
, "%*s\"name\": \"%s\",\n", indent
, "", vmsd
->name
);
444 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
446 fprintf(out_file
, "%*s\"minimum_version_id\": %d", indent
, "",
447 vmsd
->minimum_version_id
);
448 if (vmsd
->fields
!= NULL
) {
449 const VMStateField
*field
= vmsd
->fields
;
452 fprintf(out_file
, ",\n%*s\"Fields\": [\n", indent
, "");
454 while (field
->name
!= NULL
) {
455 if (field
->flags
& VMS_MUST_EXIST
) {
456 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
461 fprintf(out_file
, ",\n");
463 dump_vmstate_vmsf(out_file
, field
, indent
+ 2);
467 fprintf(out_file
, "\n%*s]", indent
, "");
469 if (vmsd
->subsections
!= NULL
) {
470 const VMStateDescription
**subsection
= vmsd
->subsections
;
473 fprintf(out_file
, ",\n%*s\"Subsections\": [\n", indent
, "");
475 while (*subsection
!= NULL
) {
477 fprintf(out_file
, ",\n");
479 dump_vmstate_vmss(out_file
, subsection
, indent
+ 2);
483 fprintf(out_file
, "\n%*s]", indent
, "");
485 fprintf(out_file
, "\n%*s}", indent
- 2, "");
488 static void dump_machine_type(FILE *out_file
)
492 mc
= MACHINE_GET_CLASS(current_machine
);
494 fprintf(out_file
, " \"vmschkmachine\": {\n");
495 fprintf(out_file
, " \"Name\": \"%s\"\n", mc
->name
);
496 fprintf(out_file
, " },\n");
499 void dump_vmstate_json_to_file(FILE *out_file
)
504 fprintf(out_file
, "{\n");
505 dump_machine_type(out_file
);
508 list
= object_class_get_list(TYPE_DEVICE
, true);
509 for (elt
= list
; elt
; elt
= elt
->next
) {
510 DeviceClass
*dc
= OBJECT_CLASS_CHECK(DeviceClass
, elt
->data
,
520 fprintf(out_file
, ",\n");
522 name
= object_class_get_name(OBJECT_CLASS(dc
));
523 fprintf(out_file
, "%*s\"%s\": {\n", indent
, "", name
);
525 fprintf(out_file
, "%*s\"Name\": \"%s\",\n", indent
, "", name
);
526 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
527 dc
->vmsd
->version_id
);
528 fprintf(out_file
, "%*s\"minimum_version_id\": %d,\n", indent
, "",
529 dc
->vmsd
->minimum_version_id
);
531 dump_vmstate_vmsd(out_file
, dc
->vmsd
, indent
, false);
533 fprintf(out_file
, "\n%*s}", indent
- 2, "");
536 fprintf(out_file
, "\n}\n");
540 static int calculate_new_instance_id(const char *idstr
)
545 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
546 if (strcmp(idstr
, se
->idstr
) == 0
547 && instance_id
<= se
->instance_id
) {
548 instance_id
= se
->instance_id
+ 1;
554 static int calculate_compat_instance_id(const char *idstr
)
559 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
564 if (strcmp(idstr
, se
->compat
->idstr
) == 0
565 && instance_id
<= se
->compat
->instance_id
) {
566 instance_id
= se
->compat
->instance_id
+ 1;
572 static inline MigrationPriority
save_state_priority(SaveStateEntry
*se
)
575 return se
->vmsd
->priority
;
577 return MIG_PRI_DEFAULT
;
580 static void savevm_state_handler_insert(SaveStateEntry
*nse
)
582 MigrationPriority priority
= save_state_priority(nse
);
585 assert(priority
<= MIG_PRI_MAX
);
587 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
588 if (save_state_priority(se
) < priority
) {
594 QTAILQ_INSERT_BEFORE(se
, nse
, entry
);
596 QTAILQ_INSERT_TAIL(&savevm_state
.handlers
, nse
, entry
);
600 /* TODO: Individual devices generally have very little idea about the rest
601 of the system, so instance_id should be removed/replaced.
602 Meanwhile pass -1 as instance_id if you do not already have a clearly
603 distinguishing id for all instances of your device class. */
604 int register_savevm_live(DeviceState
*dev
,
613 se
= g_new0(SaveStateEntry
, 1);
614 se
->version_id
= version_id
;
615 se
->section_id
= savevm_state
.global_section_id
++;
619 /* if this is a live_savem then set is_ram */
620 if (ops
->save_setup
!= NULL
) {
625 char *id
= qdev_get_dev_path(dev
);
627 if (snprintf(se
->idstr
, sizeof(se
->idstr
), "%s/", id
) >=
629 error_report("Path too long for VMState (%s)", id
);
637 se
->compat
= g_new0(CompatEntry
, 1);
638 pstrcpy(se
->compat
->idstr
, sizeof(se
->compat
->idstr
), idstr
);
639 se
->compat
->instance_id
= instance_id
== -1 ?
640 calculate_compat_instance_id(idstr
) : instance_id
;
644 pstrcat(se
->idstr
, sizeof(se
->idstr
), idstr
);
646 if (instance_id
== -1) {
647 se
->instance_id
= calculate_new_instance_id(se
->idstr
);
649 se
->instance_id
= instance_id
;
651 assert(!se
->compat
|| se
->instance_id
== 0);
652 savevm_state_handler_insert(se
);
656 void unregister_savevm(DeviceState
*dev
, const char *idstr
, void *opaque
)
658 SaveStateEntry
*se
, *new_se
;
662 char *path
= qdev_get_dev_path(dev
);
664 pstrcpy(id
, sizeof(id
), path
);
665 pstrcat(id
, sizeof(id
), "/");
669 pstrcat(id
, sizeof(id
), idstr
);
671 QTAILQ_FOREACH_SAFE(se
, &savevm_state
.handlers
, entry
, new_se
) {
672 if (strcmp(se
->idstr
, id
) == 0 && se
->opaque
== opaque
) {
673 QTAILQ_REMOVE(&savevm_state
.handlers
, se
, entry
);
680 int vmstate_register_with_alias_id(DeviceState
*dev
, int instance_id
,
681 const VMStateDescription
*vmsd
,
682 void *opaque
, int alias_id
,
683 int required_for_version
,
688 /* If this triggers, alias support can be dropped for the vmsd. */
689 assert(alias_id
== -1 || required_for_version
>= vmsd
->minimum_version_id
);
691 se
= g_new0(SaveStateEntry
, 1);
692 se
->version_id
= vmsd
->version_id
;
693 se
->section_id
= savevm_state
.global_section_id
++;
696 se
->alias_id
= alias_id
;
699 char *id
= qdev_get_dev_path(dev
);
701 if (snprintf(se
->idstr
, sizeof(se
->idstr
), "%s/", id
) >=
703 error_setg(errp
, "Path too long for VMState (%s)", id
);
711 se
->compat
= g_new0(CompatEntry
, 1);
712 pstrcpy(se
->compat
->idstr
, sizeof(se
->compat
->idstr
), vmsd
->name
);
713 se
->compat
->instance_id
= instance_id
== -1 ?
714 calculate_compat_instance_id(vmsd
->name
) : instance_id
;
718 pstrcat(se
->idstr
, sizeof(se
->idstr
), vmsd
->name
);
720 if (instance_id
== -1) {
721 se
->instance_id
= calculate_new_instance_id(se
->idstr
);
723 se
->instance_id
= instance_id
;
725 assert(!se
->compat
|| se
->instance_id
== 0);
726 savevm_state_handler_insert(se
);
730 void vmstate_unregister(DeviceState
*dev
, const VMStateDescription
*vmsd
,
733 SaveStateEntry
*se
, *new_se
;
735 QTAILQ_FOREACH_SAFE(se
, &savevm_state
.handlers
, entry
, new_se
) {
736 if (se
->vmsd
== vmsd
&& se
->opaque
== opaque
) {
737 QTAILQ_REMOVE(&savevm_state
.handlers
, se
, entry
);
744 static int vmstate_load(QEMUFile
*f
, SaveStateEntry
*se
)
746 trace_vmstate_load(se
->idstr
, se
->vmsd
? se
->vmsd
->name
: "(old)");
747 if (!se
->vmsd
) { /* Old style */
748 return se
->ops
->load_state(f
, se
->opaque
, se
->load_version_id
);
750 return vmstate_load_state(f
, se
->vmsd
, se
->opaque
, se
->load_version_id
);
753 static void vmstate_save_old_style(QEMUFile
*f
, SaveStateEntry
*se
, QJSON
*vmdesc
)
755 int64_t old_offset
, size
;
757 old_offset
= qemu_ftell_fast(f
);
758 se
->ops
->save_state(f
, se
->opaque
);
759 size
= qemu_ftell_fast(f
) - old_offset
;
762 json_prop_int(vmdesc
, "size", size
);
763 json_start_array(vmdesc
, "fields");
764 json_start_object(vmdesc
, NULL
);
765 json_prop_str(vmdesc
, "name", "data");
766 json_prop_int(vmdesc
, "size", size
);
767 json_prop_str(vmdesc
, "type", "buffer");
768 json_end_object(vmdesc
);
769 json_end_array(vmdesc
);
773 static int vmstate_save(QEMUFile
*f
, SaveStateEntry
*se
, QJSON
*vmdesc
)
775 trace_vmstate_save(se
->idstr
, se
->vmsd
? se
->vmsd
->name
: "(old)");
777 vmstate_save_old_style(f
, se
, vmdesc
);
780 return vmstate_save_state(f
, se
->vmsd
, se
->opaque
, vmdesc
);
784 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
786 static void save_section_header(QEMUFile
*f
, SaveStateEntry
*se
,
787 uint8_t section_type
)
789 qemu_put_byte(f
, section_type
);
790 qemu_put_be32(f
, se
->section_id
);
792 if (section_type
== QEMU_VM_SECTION_FULL
||
793 section_type
== QEMU_VM_SECTION_START
) {
795 size_t len
= strlen(se
->idstr
);
796 qemu_put_byte(f
, len
);
797 qemu_put_buffer(f
, (uint8_t *)se
->idstr
, len
);
799 qemu_put_be32(f
, se
->instance_id
);
800 qemu_put_be32(f
, se
->version_id
);
805 * Write a footer onto device sections that catches cases misformatted device
808 static void save_section_footer(QEMUFile
*f
, SaveStateEntry
*se
)
810 if (migrate_get_current()->send_section_footer
) {
811 qemu_put_byte(f
, QEMU_VM_SECTION_FOOTER
);
812 qemu_put_be32(f
, se
->section_id
);
817 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
818 * command and associated data.
820 * @f: File to send command on
821 * @command: Command type to send
822 * @len: Length of associated data
823 * @data: Data associated with command.
825 static void qemu_savevm_command_send(QEMUFile
*f
,
826 enum qemu_vm_cmd command
,
830 trace_savevm_command_send(command
, len
);
831 qemu_put_byte(f
, QEMU_VM_COMMAND
);
832 qemu_put_be16(f
, (uint16_t)command
);
833 qemu_put_be16(f
, len
);
834 qemu_put_buffer(f
, data
, len
);
838 void qemu_savevm_send_ping(QEMUFile
*f
, uint32_t value
)
842 trace_savevm_send_ping(value
);
843 buf
= cpu_to_be32(value
);
844 qemu_savevm_command_send(f
, MIG_CMD_PING
, sizeof(value
), (uint8_t *)&buf
);
847 void qemu_savevm_send_open_return_path(QEMUFile
*f
)
849 trace_savevm_send_open_return_path();
850 qemu_savevm_command_send(f
, MIG_CMD_OPEN_RETURN_PATH
, 0, NULL
);
853 /* We have a buffer of data to send; we don't want that all to be loaded
854 * by the command itself, so the command contains just the length of the
855 * extra buffer that we then send straight after it.
856 * TODO: Must be a better way to organise that
862 int qemu_savevm_send_packaged(QEMUFile
*f
, const uint8_t *buf
, size_t len
)
866 if (len
> MAX_VM_CMD_PACKAGED_SIZE
) {
867 error_report("%s: Unreasonably large packaged state: %zu",
872 tmp
= cpu_to_be32(len
);
874 trace_qemu_savevm_send_packaged();
875 qemu_savevm_command_send(f
, MIG_CMD_PACKAGED
, 4, (uint8_t *)&tmp
);
877 qemu_put_buffer(f
, buf
, len
);
882 /* Send prior to any postcopy transfer */
883 void qemu_savevm_send_postcopy_advise(QEMUFile
*f
)
885 if (migrate_postcopy_ram()) {
887 tmp
[0] = cpu_to_be64(ram_pagesize_summary());
888 tmp
[1] = cpu_to_be64(qemu_target_page_size());
890 trace_qemu_savevm_send_postcopy_advise();
891 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_ADVISE
,
894 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_ADVISE
, 0, NULL
);
898 /* Sent prior to starting the destination running in postcopy, discard pages
899 * that have already been sent but redirtied on the source.
900 * CMD_POSTCOPY_RAM_DISCARD consist of:
902 * byte Length of name field (not including 0)
903 * n x byte RAM block name
904 * byte 0 terminator (just for safety)
905 * n x Byte ranges within the named RAMBlock
906 * be64 Start of the range
909 * name: RAMBlock name that these entries are part of
910 * len: Number of page entries
911 * start_list: 'len' addresses
912 * length_list: 'len' addresses
915 void qemu_savevm_send_postcopy_ram_discard(QEMUFile
*f
, const char *name
,
917 uint64_t *start_list
,
918 uint64_t *length_list
)
923 size_t name_len
= strlen(name
);
925 trace_qemu_savevm_send_postcopy_ram_discard(name
, len
);
926 assert(name_len
< 256);
927 buf
= g_malloc0(1 + 1 + name_len
+ 1 + (8 + 8) * len
);
928 buf
[0] = postcopy_ram_discard_version
;
930 memcpy(buf
+ 2, name
, name_len
);
931 tmplen
= 2 + name_len
;
932 buf
[tmplen
++] = '\0';
934 for (t
= 0; t
< len
; t
++) {
935 stq_be_p(buf
+ tmplen
, start_list
[t
]);
937 stq_be_p(buf
+ tmplen
, length_list
[t
]);
940 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RAM_DISCARD
, tmplen
, buf
);
944 /* Get the destination into a state where it can receive postcopy data. */
945 void qemu_savevm_send_postcopy_listen(QEMUFile
*f
)
947 trace_savevm_send_postcopy_listen();
948 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_LISTEN
, 0, NULL
);
951 /* Kick the destination into running */
952 void qemu_savevm_send_postcopy_run(QEMUFile
*f
)
954 trace_savevm_send_postcopy_run();
955 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RUN
, 0, NULL
);
958 bool qemu_savevm_state_blocked(Error
**errp
)
962 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
963 if (se
->vmsd
&& se
->vmsd
->unmigratable
) {
964 error_setg(errp
, "State blocked by non-migratable device '%s'",
972 void qemu_savevm_state_header(QEMUFile
*f
)
974 trace_savevm_state_header();
975 qemu_put_be32(f
, QEMU_VM_FILE_MAGIC
);
976 qemu_put_be32(f
, QEMU_VM_FILE_VERSION
);
978 if (migrate_get_current()->send_configuration
) {
979 qemu_put_byte(f
, QEMU_VM_CONFIGURATION
);
980 vmstate_save_state(f
, &vmstate_configuration
, &savevm_state
, 0);
984 void qemu_savevm_state_setup(QEMUFile
*f
)
989 trace_savevm_state_setup();
990 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
991 if (!se
->ops
|| !se
->ops
->save_setup
) {
994 if (se
->ops
&& se
->ops
->is_active
) {
995 if (!se
->ops
->is_active(se
->opaque
)) {
999 save_section_header(f
, se
, QEMU_VM_SECTION_START
);
1001 ret
= se
->ops
->save_setup(f
, se
->opaque
);
1002 save_section_footer(f
, se
);
1004 qemu_file_set_error(f
, ret
);
1011 * this function has three return values:
1012 * negative: there was one error, and we have -errno.
1013 * 0 : We haven't finished, caller have to go again
1014 * 1 : We have finished, we can go to complete phase
1016 int qemu_savevm_state_iterate(QEMUFile
*f
, bool postcopy
)
1021 trace_savevm_state_iterate();
1022 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1023 if (!se
->ops
|| !se
->ops
->save_live_iterate
) {
1026 if (se
->ops
&& se
->ops
->is_active
) {
1027 if (!se
->ops
->is_active(se
->opaque
)) {
1032 * In the postcopy phase, any device that doesn't know how to
1033 * do postcopy should have saved it's state in the _complete
1034 * call that's already run, it might get confused if we call
1035 * iterate afterwards.
1038 !(se
->ops
->has_postcopy
&& se
->ops
->has_postcopy(se
->opaque
))) {
1041 if (qemu_file_rate_limit(f
)) {
1044 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1046 save_section_header(f
, se
, QEMU_VM_SECTION_PART
);
1048 ret
= se
->ops
->save_live_iterate(f
, se
->opaque
);
1049 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1050 save_section_footer(f
, se
);
1053 qemu_file_set_error(f
, ret
);
1056 /* Do not proceed to the next vmstate before this one reported
1057 completion of the current stage. This serializes the migration
1058 and reduces the probability that a faster changing state is
1059 synchronized over and over again. */
1066 static bool should_send_vmdesc(void)
1068 MachineState
*machine
= MACHINE(qdev_get_machine());
1069 bool in_postcopy
= migration_in_postcopy();
1070 return !machine
->suppress_vmdesc
&& !in_postcopy
;
1074 * Calls the save_live_complete_postcopy methods
1075 * causing the last few pages to be sent immediately and doing any associated
1077 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1078 * all the other devices, but that happens at the point we switch to postcopy.
1080 void qemu_savevm_state_complete_postcopy(QEMUFile
*f
)
1085 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1086 if (!se
->ops
|| !se
->ops
->save_live_complete_postcopy
) {
1089 if (se
->ops
&& se
->ops
->is_active
) {
1090 if (!se
->ops
->is_active(se
->opaque
)) {
1094 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1096 qemu_put_byte(f
, QEMU_VM_SECTION_END
);
1097 qemu_put_be32(f
, se
->section_id
);
1099 ret
= se
->ops
->save_live_complete_postcopy(f
, se
->opaque
);
1100 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1101 save_section_footer(f
, se
);
1103 qemu_file_set_error(f
, ret
);
1108 qemu_put_byte(f
, QEMU_VM_EOF
);
1112 int qemu_savevm_state_complete_precopy(QEMUFile
*f
, bool iterable_only
,
1113 bool inactivate_disks
)
1119 bool in_postcopy
= migration_in_postcopy();
1121 trace_savevm_state_complete_precopy();
1123 cpu_synchronize_all_states();
1125 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1127 (in_postcopy
&& se
->ops
->has_postcopy
&&
1128 se
->ops
->has_postcopy(se
->opaque
)) ||
1129 (in_postcopy
&& !iterable_only
) ||
1130 !se
->ops
->save_live_complete_precopy
) {
1134 if (se
->ops
&& se
->ops
->is_active
) {
1135 if (!se
->ops
->is_active(se
->opaque
)) {
1139 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1141 save_section_header(f
, se
, QEMU_VM_SECTION_END
);
1143 ret
= se
->ops
->save_live_complete_precopy(f
, se
->opaque
);
1144 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1145 save_section_footer(f
, se
);
1147 qemu_file_set_error(f
, ret
);
1152 if (iterable_only
) {
1156 vmdesc
= qjson_new();
1157 json_prop_int(vmdesc
, "page_size", qemu_target_page_size());
1158 json_start_array(vmdesc
, "devices");
1159 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1161 if ((!se
->ops
|| !se
->ops
->save_state
) && !se
->vmsd
) {
1164 if (se
->vmsd
&& !vmstate_save_needed(se
->vmsd
, se
->opaque
)) {
1165 trace_savevm_section_skip(se
->idstr
, se
->section_id
);
1169 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1171 json_start_object(vmdesc
, NULL
);
1172 json_prop_str(vmdesc
, "name", se
->idstr
);
1173 json_prop_int(vmdesc
, "instance_id", se
->instance_id
);
1175 save_section_header(f
, se
, QEMU_VM_SECTION_FULL
);
1176 ret
= vmstate_save(f
, se
, vmdesc
);
1178 qemu_file_set_error(f
, ret
);
1181 trace_savevm_section_end(se
->idstr
, se
->section_id
, 0);
1182 save_section_footer(f
, se
);
1184 json_end_object(vmdesc
);
1187 if (inactivate_disks
) {
1188 /* Inactivate before sending QEMU_VM_EOF so that the
1189 * bdrv_invalidate_cache_all() on the other end won't fail. */
1190 ret
= bdrv_inactivate_all();
1192 error_report("%s: bdrv_inactivate_all() failed (%d)",
1194 qemu_file_set_error(f
, ret
);
1199 /* Postcopy stream will still be going */
1200 qemu_put_byte(f
, QEMU_VM_EOF
);
1203 json_end_array(vmdesc
);
1204 qjson_finish(vmdesc
);
1205 vmdesc_len
= strlen(qjson_get_str(vmdesc
));
1207 if (should_send_vmdesc()) {
1208 qemu_put_byte(f
, QEMU_VM_VMDESCRIPTION
);
1209 qemu_put_be32(f
, vmdesc_len
);
1210 qemu_put_buffer(f
, (uint8_t *)qjson_get_str(vmdesc
), vmdesc_len
);
1212 qjson_destroy(vmdesc
);
1218 /* Give an estimate of the amount left to be transferred,
1219 * the result is split into the amount for units that can and
1220 * for units that can't do postcopy.
1222 void qemu_savevm_state_pending(QEMUFile
*f
, uint64_t threshold_size
,
1223 uint64_t *res_non_postcopiable
,
1224 uint64_t *res_postcopiable
)
1228 *res_non_postcopiable
= 0;
1229 *res_postcopiable
= 0;
1232 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1233 if (!se
->ops
|| !se
->ops
->save_live_pending
) {
1236 if (se
->ops
&& se
->ops
->is_active
) {
1237 if (!se
->ops
->is_active(se
->opaque
)) {
1241 se
->ops
->save_live_pending(f
, se
->opaque
, threshold_size
,
1242 res_non_postcopiable
, res_postcopiable
);
1246 void qemu_savevm_state_cleanup(void)
1250 trace_savevm_state_cleanup();
1251 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1252 if (se
->ops
&& se
->ops
->save_cleanup
) {
1253 se
->ops
->save_cleanup(se
->opaque
);
1258 static int qemu_savevm_state(QEMUFile
*f
, Error
**errp
)
1261 MigrationState
*ms
= migrate_get_current();
1262 MigrationStatus status
;
1266 ms
->to_dst_file
= f
;
1268 if (migration_is_blocked(errp
)) {
1273 if (migrate_use_block()) {
1274 error_setg(errp
, "Block migration and snapshots are incompatible");
1279 qemu_mutex_unlock_iothread();
1280 qemu_savevm_state_header(f
);
1281 qemu_savevm_state_setup(f
);
1282 qemu_mutex_lock_iothread();
1284 while (qemu_file_get_error(f
) == 0) {
1285 if (qemu_savevm_state_iterate(f
, false) > 0) {
1290 ret
= qemu_file_get_error(f
);
1292 qemu_savevm_state_complete_precopy(f
, false, false);
1293 ret
= qemu_file_get_error(f
);
1295 qemu_savevm_state_cleanup();
1297 error_setg_errno(errp
, -ret
, "Error while writing VM state");
1302 status
= MIGRATION_STATUS_FAILED
;
1304 status
= MIGRATION_STATUS_COMPLETED
;
1306 migrate_set_state(&ms
->state
, MIGRATION_STATUS_SETUP
, status
);
1308 /* f is outer parameter, it should not stay in global migration state after
1309 * this function finished */
1310 ms
->to_dst_file
= NULL
;
1315 static int qemu_save_device_state(QEMUFile
*f
)
1319 qemu_put_be32(f
, QEMU_VM_FILE_MAGIC
);
1320 qemu_put_be32(f
, QEMU_VM_FILE_VERSION
);
1322 cpu_synchronize_all_states();
1324 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1330 if ((!se
->ops
|| !se
->ops
->save_state
) && !se
->vmsd
) {
1333 if (se
->vmsd
&& !vmstate_save_needed(se
->vmsd
, se
->opaque
)) {
1337 save_section_header(f
, se
, QEMU_VM_SECTION_FULL
);
1339 ret
= vmstate_save(f
, se
, NULL
);
1344 save_section_footer(f
, se
);
1347 qemu_put_byte(f
, QEMU_VM_EOF
);
1349 return qemu_file_get_error(f
);
1352 static SaveStateEntry
*find_se(const char *idstr
, int instance_id
)
1356 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1357 if (!strcmp(se
->idstr
, idstr
) &&
1358 (instance_id
== se
->instance_id
||
1359 instance_id
== se
->alias_id
))
1361 /* Migrating from an older version? */
1362 if (strstr(se
->idstr
, idstr
) && se
->compat
) {
1363 if (!strcmp(se
->compat
->idstr
, idstr
) &&
1364 (instance_id
== se
->compat
->instance_id
||
1365 instance_id
== se
->alias_id
))
1372 enum LoadVMExitCodes
{
1373 /* Allow a command to quit all layers of nested loadvm loops */
1377 static int qemu_loadvm_state_main(QEMUFile
*f
, MigrationIncomingState
*mis
);
1379 /* ------ incoming postcopy messages ------ */
1380 /* 'advise' arrives before any transfers just to tell us that a postcopy
1381 * *might* happen - it might be skipped if precopy transferred everything
1384 static int loadvm_postcopy_handle_advise(MigrationIncomingState
*mis
,
1387 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_ADVISE
);
1388 uint64_t remote_pagesize_summary
, local_pagesize_summary
, remote_tps
;
1390 trace_loadvm_postcopy_handle_advise();
1391 if (ps
!= POSTCOPY_INCOMING_NONE
) {
1392 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps
);
1398 if (migrate_postcopy_ram()) {
1399 error_report("RAM postcopy is enabled but have 0 byte advise");
1404 if (!migrate_postcopy_ram()) {
1405 error_report("RAM postcopy is disabled but have 16 byte advise");
1410 error_report("CMD_POSTCOPY_ADVISE invalid length (%d)", len
);
1414 if (!postcopy_ram_supported_by_host(mis
)) {
1415 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
1419 remote_pagesize_summary
= qemu_get_be64(mis
->from_src_file
);
1420 local_pagesize_summary
= ram_pagesize_summary();
1422 if (remote_pagesize_summary
!= local_pagesize_summary
) {
1424 * This detects two potential causes of mismatch:
1425 * a) A mismatch in host page sizes
1426 * Some combinations of mismatch are probably possible but it gets
1427 * a bit more complicated. In particular we need to place whole
1428 * host pages on the dest at once, and we need to ensure that we
1429 * handle dirtying to make sure we never end up sending part of
1430 * a hostpage on it's own.
1431 * b) The use of different huge page sizes on source/destination
1432 * a more fine grain test is performed during RAM block migration
1433 * but this test here causes a nice early clear failure, and
1434 * also fails when passed to an older qemu that doesn't
1437 error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1439 remote_pagesize_summary
, local_pagesize_summary
);
1443 remote_tps
= qemu_get_be64(mis
->from_src_file
);
1444 if (remote_tps
!= qemu_target_page_size()) {
1446 * Again, some differences could be dealt with, but for now keep it
1449 error_report("Postcopy needs matching target page sizes (s=%d d=%zd)",
1450 (int)remote_tps
, qemu_target_page_size());
1454 if (ram_postcopy_incoming_init(mis
)) {
1458 postcopy_state_set(POSTCOPY_INCOMING_ADVISE
);
1463 /* After postcopy we will be told to throw some pages away since they're
1464 * dirty and will have to be demand fetched. Must happen before CPU is
1466 * There can be 0..many of these messages, each encoding multiple pages.
1468 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState
*mis
,
1473 PostcopyState ps
= postcopy_state_get();
1475 trace_loadvm_postcopy_ram_handle_discard();
1478 case POSTCOPY_INCOMING_ADVISE
:
1480 tmp
= postcopy_ram_prepare_discard(mis
);
1486 case POSTCOPY_INCOMING_DISCARD
:
1487 /* Expected state */
1491 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1495 /* We're expecting a
1497 * a RAM ID string (length byte, name, 0 term)
1498 * then at least 1 16 byte chunk
1500 if (len
< (1 + 1 + 1 + 1 + 2 * 8)) {
1501 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len
);
1505 tmp
= qemu_get_byte(mis
->from_src_file
);
1506 if (tmp
!= postcopy_ram_discard_version
) {
1507 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp
);
1511 if (!qemu_get_counted_string(mis
->from_src_file
, ramid
)) {
1512 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1515 tmp
= qemu_get_byte(mis
->from_src_file
);
1517 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp
);
1521 len
-= 3 + strlen(ramid
);
1523 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len
);
1526 trace_loadvm_postcopy_ram_handle_discard_header(ramid
, len
);
1528 uint64_t start_addr
, block_length
;
1529 start_addr
= qemu_get_be64(mis
->from_src_file
);
1530 block_length
= qemu_get_be64(mis
->from_src_file
);
1533 int ret
= ram_discard_range(ramid
, start_addr
, block_length
);
1538 trace_loadvm_postcopy_ram_handle_discard_end();
1544 * Triggered by a postcopy_listen command; this thread takes over reading
1545 * the input stream, leaving the main thread free to carry on loading the rest
1546 * of the device state (from RAM).
1547 * (TODO:This could do with being in a postcopy file - but there again it's
1548 * just another input loop, not that postcopy specific)
1550 static void *postcopy_ram_listen_thread(void *opaque
)
1552 QEMUFile
*f
= opaque
;
1553 MigrationIncomingState
*mis
= migration_incoming_get_current();
1556 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
1557 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1558 qemu_sem_post(&mis
->listen_thread_sem
);
1559 trace_postcopy_ram_listen_thread_start();
1562 * Because we're a thread and not a coroutine we can't yield
1563 * in qemu_file, and thus we must be blocking now.
1565 qemu_file_set_blocking(f
, true);
1566 load_res
= qemu_loadvm_state_main(f
, mis
);
1567 /* And non-blocking again so we don't block in any cleanup */
1568 qemu_file_set_blocking(f
, false);
1570 trace_postcopy_ram_listen_thread_exit();
1572 error_report("%s: loadvm failed: %d", __func__
, load_res
);
1573 qemu_file_set_error(f
, load_res
);
1574 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1575 MIGRATION_STATUS_FAILED
);
1578 * This looks good, but it's possible that the device loading in the
1579 * main thread hasn't finished yet, and so we might not be in 'RUN'
1580 * state yet; wait for the end of the main thread.
1582 qemu_event_wait(&mis
->main_thread_load_event
);
1584 postcopy_ram_incoming_cleanup(mis
);
1588 * If something went wrong then we have a bad state so exit;
1589 * depending how far we got it might be possible at this point
1590 * to leave the guest running and fire MCEs for pages that never
1591 * arrived as a desperate recovery step.
1596 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1597 MIGRATION_STATUS_COMPLETED
);
1599 * If everything has worked fine, then the main thread has waited
1600 * for us to start, and we're the last use of the mis.
1601 * (If something broke then qemu will have to exit anyway since it's
1602 * got a bad migration state).
1604 migration_incoming_state_destroy();
1605 qemu_loadvm_state_cleanup();
1610 /* After this message we must be able to immediately receive postcopy data */
1611 static int loadvm_postcopy_handle_listen(MigrationIncomingState
*mis
)
1613 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_LISTENING
);
1614 trace_loadvm_postcopy_handle_listen();
1615 if (ps
!= POSTCOPY_INCOMING_ADVISE
&& ps
!= POSTCOPY_INCOMING_DISCARD
) {
1616 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps
);
1619 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
1621 * A rare case, we entered listen without having to do any discards,
1622 * so do the setup that's normally done at the time of the 1st discard.
1624 if (migrate_postcopy_ram()) {
1625 postcopy_ram_prepare_discard(mis
);
1630 * Sensitise RAM - can now generate requests for blocks that don't exist
1631 * However, at this point the CPU shouldn't be running, and the IO
1632 * shouldn't be doing anything yet so don't actually expect requests
1634 if (migrate_postcopy_ram()) {
1635 if (postcopy_ram_enable_notify(mis
)) {
1640 if (mis
->have_listen_thread
) {
1641 error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread");
1645 mis
->have_listen_thread
= true;
1646 /* Start up the listening thread and wait for it to signal ready */
1647 qemu_sem_init(&mis
->listen_thread_sem
, 0);
1648 qemu_thread_create(&mis
->listen_thread
, "postcopy/listen",
1649 postcopy_ram_listen_thread
, mis
->from_src_file
,
1650 QEMU_THREAD_DETACHED
);
1651 qemu_sem_wait(&mis
->listen_thread_sem
);
1652 qemu_sem_destroy(&mis
->listen_thread_sem
);
1662 static void loadvm_postcopy_handle_run_bh(void *opaque
)
1664 Error
*local_err
= NULL
;
1665 HandleRunBhData
*data
= opaque
;
1667 /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1670 cpu_synchronize_all_post_init();
1672 qemu_announce_self();
1674 /* Make sure all file formats flush their mutable metadata.
1675 * If we get an error here, just don't restart the VM yet. */
1676 bdrv_invalidate_cache_all(&local_err
);
1678 error_report_err(local_err
);
1683 trace_loadvm_postcopy_handle_run_cpu_sync();
1684 cpu_synchronize_all_post_init();
1686 trace_loadvm_postcopy_handle_run_vmstart();
1689 /* Hold onto your hats, starting the CPU */
1692 /* leave it paused and let management decide when to start the CPU */
1693 runstate_set(RUN_STATE_PAUSED
);
1696 qemu_bh_delete(data
->bh
);
1700 /* After all discards we can start running and asking for pages */
1701 static int loadvm_postcopy_handle_run(MigrationIncomingState
*mis
)
1703 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_RUNNING
);
1704 HandleRunBhData
*data
;
1706 trace_loadvm_postcopy_handle_run();
1707 if (ps
!= POSTCOPY_INCOMING_LISTENING
) {
1708 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps
);
1712 data
= g_new(HandleRunBhData
, 1);
1713 data
->bh
= qemu_bh_new(loadvm_postcopy_handle_run_bh
, data
);
1714 qemu_bh_schedule(data
->bh
);
1716 /* We need to finish reading the stream from the package
1717 * and also stop reading anything more from the stream that loaded the
1718 * package (since it's now being read by the listener thread).
1719 * LOADVM_QUIT will quit all the layers of nested loadvm loops.
1725 * Immediately following this command is a blob of data containing an embedded
1726 * chunk of migration stream; read it and load it.
1728 * @mis: Incoming state
1729 * @length: Length of packaged data to read
1731 * Returns: Negative values on error
1734 static int loadvm_handle_cmd_packaged(MigrationIncomingState
*mis
)
1738 QIOChannelBuffer
*bioc
;
1740 length
= qemu_get_be32(mis
->from_src_file
);
1741 trace_loadvm_handle_cmd_packaged(length
);
1743 if (length
> MAX_VM_CMD_PACKAGED_SIZE
) {
1744 error_report("Unreasonably large packaged state: %zu", length
);
1748 bioc
= qio_channel_buffer_new(length
);
1749 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-loadvm-buffer");
1750 ret
= qemu_get_buffer(mis
->from_src_file
,
1753 if (ret
!= length
) {
1754 object_unref(OBJECT(bioc
));
1755 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
1757 return (ret
< 0) ? ret
: -EAGAIN
;
1759 bioc
->usage
+= length
;
1760 trace_loadvm_handle_cmd_packaged_received(ret
);
1762 QEMUFile
*packf
= qemu_fopen_channel_input(QIO_CHANNEL(bioc
));
1764 ret
= qemu_loadvm_state_main(packf
, mis
);
1765 trace_loadvm_handle_cmd_packaged_main(ret
);
1767 object_unref(OBJECT(bioc
));
1773 * Process an incoming 'QEMU_VM_COMMAND'
1774 * 0 just a normal return
1775 * LOADVM_QUIT All good, but exit the loop
1778 static int loadvm_process_command(QEMUFile
*f
)
1780 MigrationIncomingState
*mis
= migration_incoming_get_current();
1785 cmd
= qemu_get_be16(f
);
1786 len
= qemu_get_be16(f
);
1788 /* Check validity before continue processing of cmds */
1789 if (qemu_file_get_error(f
)) {
1790 return qemu_file_get_error(f
);
1793 trace_loadvm_process_command(cmd
, len
);
1794 if (cmd
>= MIG_CMD_MAX
|| cmd
== MIG_CMD_INVALID
) {
1795 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd
, len
);
1799 if (mig_cmd_args
[cmd
].len
!= -1 && mig_cmd_args
[cmd
].len
!= len
) {
1800 error_report("%s received with bad length - expecting %zu, got %d",
1801 mig_cmd_args
[cmd
].name
,
1802 (size_t)mig_cmd_args
[cmd
].len
, len
);
1807 case MIG_CMD_OPEN_RETURN_PATH
:
1808 if (mis
->to_src_file
) {
1809 error_report("CMD_OPEN_RETURN_PATH called when RP already open");
1810 /* Not really a problem, so don't give up */
1813 mis
->to_src_file
= qemu_file_get_return_path(f
);
1814 if (!mis
->to_src_file
) {
1815 error_report("CMD_OPEN_RETURN_PATH failed");
1821 tmp32
= qemu_get_be32(f
);
1822 trace_loadvm_process_command_ping(tmp32
);
1823 if (!mis
->to_src_file
) {
1824 error_report("CMD_PING (0x%x) received with no return path",
1828 migrate_send_rp_pong(mis
, tmp32
);
1831 case MIG_CMD_PACKAGED
:
1832 return loadvm_handle_cmd_packaged(mis
);
1834 case MIG_CMD_POSTCOPY_ADVISE
:
1835 return loadvm_postcopy_handle_advise(mis
, len
);
1837 case MIG_CMD_POSTCOPY_LISTEN
:
1838 return loadvm_postcopy_handle_listen(mis
);
1840 case MIG_CMD_POSTCOPY_RUN
:
1841 return loadvm_postcopy_handle_run(mis
);
1843 case MIG_CMD_POSTCOPY_RAM_DISCARD
:
1844 return loadvm_postcopy_ram_handle_discard(mis
, len
);
1851 * Read a footer off the wire and check that it matches the expected section
1853 * Returns: true if the footer was good
1854 * false if there is a problem (and calls error_report to say why)
1856 static bool check_section_footer(QEMUFile
*f
, SaveStateEntry
*se
)
1860 uint32_t read_section_id
;
1862 if (!migrate_get_current()->send_section_footer
) {
1863 /* No footer to check */
1867 read_mark
= qemu_get_byte(f
);
1869 ret
= qemu_file_get_error(f
);
1871 error_report("%s: Read section footer failed: %d",
1876 if (read_mark
!= QEMU_VM_SECTION_FOOTER
) {
1877 error_report("Missing section footer for %s", se
->idstr
);
1881 read_section_id
= qemu_get_be32(f
);
1882 if (read_section_id
!= se
->load_section_id
) {
1883 error_report("Mismatched section id in footer for %s -"
1884 " read 0x%x expected 0x%x",
1885 se
->idstr
, read_section_id
, se
->load_section_id
);
1894 qemu_loadvm_section_start_full(QEMUFile
*f
, MigrationIncomingState
*mis
)
1896 uint32_t instance_id
, version_id
, section_id
;
1901 /* Read section start */
1902 section_id
= qemu_get_be32(f
);
1903 if (!qemu_get_counted_string(f
, idstr
)) {
1904 error_report("Unable to read ID string for section %u",
1908 instance_id
= qemu_get_be32(f
);
1909 version_id
= qemu_get_be32(f
);
1911 ret
= qemu_file_get_error(f
);
1913 error_report("%s: Failed to read instance/version ID: %d",
1918 trace_qemu_loadvm_state_section_startfull(section_id
, idstr
,
1919 instance_id
, version_id
);
1920 /* Find savevm section */
1921 se
= find_se(idstr
, instance_id
);
1923 error_report("Unknown savevm section or instance '%s' %d",
1924 idstr
, instance_id
);
1928 /* Validate version */
1929 if (version_id
> se
->version_id
) {
1930 error_report("savevm: unsupported version %d for '%s' v%d",
1931 version_id
, idstr
, se
->version_id
);
1934 se
->load_version_id
= version_id
;
1935 se
->load_section_id
= section_id
;
1937 /* Validate if it is a device's state */
1938 if (xen_enabled() && se
->is_ram
) {
1939 error_report("loadvm: %s RAM loading not allowed on Xen", idstr
);
1943 ret
= vmstate_load(f
, se
);
1945 error_report("error while loading state for instance 0x%x of"
1946 " device '%s'", instance_id
, idstr
);
1949 if (!check_section_footer(f
, se
)) {
1957 qemu_loadvm_section_part_end(QEMUFile
*f
, MigrationIncomingState
*mis
)
1959 uint32_t section_id
;
1963 section_id
= qemu_get_be32(f
);
1965 ret
= qemu_file_get_error(f
);
1967 error_report("%s: Failed to read section ID: %d",
1972 trace_qemu_loadvm_state_section_partend(section_id
);
1973 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1974 if (se
->load_section_id
== section_id
) {
1979 error_report("Unknown savevm section %d", section_id
);
1983 ret
= vmstate_load(f
, se
);
1985 error_report("error while loading state section id %d(%s)",
1986 section_id
, se
->idstr
);
1989 if (!check_section_footer(f
, se
)) {
1996 static int qemu_loadvm_state_setup(QEMUFile
*f
)
2001 trace_loadvm_state_setup();
2002 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
2003 if (!se
->ops
|| !se
->ops
->load_setup
) {
2006 if (se
->ops
&& se
->ops
->is_active
) {
2007 if (!se
->ops
->is_active(se
->opaque
)) {
2012 ret
= se
->ops
->load_setup(f
, se
->opaque
);
2014 qemu_file_set_error(f
, ret
);
2015 error_report("Load state of device %s failed", se
->idstr
);
2022 void qemu_loadvm_state_cleanup(void)
2026 trace_loadvm_state_cleanup();
2027 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
2028 if (se
->ops
&& se
->ops
->load_cleanup
) {
2029 se
->ops
->load_cleanup(se
->opaque
);
2034 static int qemu_loadvm_state_main(QEMUFile
*f
, MigrationIncomingState
*mis
)
2036 uint8_t section_type
;
2040 section_type
= qemu_get_byte(f
);
2042 if (qemu_file_get_error(f
)) {
2043 ret
= qemu_file_get_error(f
);
2047 trace_qemu_loadvm_state_section(section_type
);
2048 switch (section_type
) {
2049 case QEMU_VM_SECTION_START
:
2050 case QEMU_VM_SECTION_FULL
:
2051 ret
= qemu_loadvm_section_start_full(f
, mis
);
2056 case QEMU_VM_SECTION_PART
:
2057 case QEMU_VM_SECTION_END
:
2058 ret
= qemu_loadvm_section_part_end(f
, mis
);
2063 case QEMU_VM_COMMAND
:
2064 ret
= loadvm_process_command(f
);
2065 trace_qemu_loadvm_state_section_command(ret
);
2066 if ((ret
< 0) || (ret
& LOADVM_QUIT
)) {
2071 /* This is the end of migration */
2074 error_report("Unknown savevm section type %d", section_type
);
2082 qemu_file_set_error(f
, ret
);
2087 int qemu_loadvm_state(QEMUFile
*f
)
2089 MigrationIncomingState
*mis
= migration_incoming_get_current();
2090 Error
*local_err
= NULL
;
2094 if (qemu_savevm_state_blocked(&local_err
)) {
2095 error_report_err(local_err
);
2099 v
= qemu_get_be32(f
);
2100 if (v
!= QEMU_VM_FILE_MAGIC
) {
2101 error_report("Not a migration stream");
2105 v
= qemu_get_be32(f
);
2106 if (v
== QEMU_VM_FILE_VERSION_COMPAT
) {
2107 error_report("SaveVM v2 format is obsolete and don't work anymore");
2110 if (v
!= QEMU_VM_FILE_VERSION
) {
2111 error_report("Unsupported migration stream version");
2115 if (qemu_loadvm_state_setup(f
) != 0) {
2119 if (migrate_get_current()->send_configuration
) {
2120 if (qemu_get_byte(f
) != QEMU_VM_CONFIGURATION
) {
2121 error_report("Configuration section missing");
2124 ret
= vmstate_load_state(f
, &vmstate_configuration
, &savevm_state
, 0);
2131 cpu_synchronize_all_pre_loadvm();
2133 ret
= qemu_loadvm_state_main(f
, mis
);
2134 qemu_event_set(&mis
->main_thread_load_event
);
2136 trace_qemu_loadvm_state_post_main(ret
);
2138 if (mis
->have_listen_thread
) {
2139 /* Listen thread still going, can't clean up yet */
2144 ret
= qemu_file_get_error(f
);
2148 * Try to read in the VMDESC section as well, so that dumping tools that
2149 * intercept our migration stream have the chance to see it.
2152 /* We've got to be careful; if we don't read the data and just shut the fd
2153 * then the sender can error if we close while it's still sending.
2154 * We also mustn't read data that isn't there; some transports (RDMA)
2155 * will stall waiting for that data when the source has already closed.
2157 if (ret
== 0 && should_send_vmdesc()) {
2160 uint8_t section_type
= qemu_get_byte(f
);
2162 if (section_type
!= QEMU_VM_VMDESCRIPTION
) {
2163 error_report("Expected vmdescription section, but got %d",
2166 * It doesn't seem worth failing at this point since
2167 * we apparently have an otherwise valid VM state
2170 buf
= g_malloc(0x1000);
2171 size
= qemu_get_be32(f
);
2174 uint32_t read_chunk
= MIN(size
, 0x1000);
2175 qemu_get_buffer(f
, buf
, read_chunk
);
2182 qemu_loadvm_state_cleanup();
2183 cpu_synchronize_all_post_init();
2188 int save_snapshot(const char *name
, Error
**errp
)
2190 BlockDriverState
*bs
, *bs1
;
2191 QEMUSnapshotInfo sn1
, *sn
= &sn1
, old_sn1
, *old_sn
= &old_sn1
;
2194 int saved_vm_running
;
2195 uint64_t vm_state_size
;
2198 AioContext
*aio_context
;
2200 if (!bdrv_all_can_snapshot(&bs
)) {
2201 error_setg(errp
, "Device '%s' is writable but does not support "
2202 "snapshots", bdrv_get_device_name(bs
));
2206 /* Delete old snapshots of the same name */
2208 ret
= bdrv_all_delete_snapshot(name
, &bs1
, errp
);
2210 error_prepend(errp
, "Error while deleting snapshot on device "
2211 "'%s': ", bdrv_get_device_name(bs1
));
2216 bs
= bdrv_all_find_vmstate_bs();
2218 error_setg(errp
, "No block device can accept snapshots");
2221 aio_context
= bdrv_get_aio_context(bs
);
2223 saved_vm_running
= runstate_is_running();
2225 ret
= global_state_store();
2227 error_setg(errp
, "Error saving global state");
2230 vm_stop(RUN_STATE_SAVE_VM
);
2232 bdrv_drain_all_begin();
2234 aio_context_acquire(aio_context
);
2236 memset(sn
, 0, sizeof(*sn
));
2238 /* fill auxiliary fields */
2239 qemu_gettimeofday(&tv
);
2240 sn
->date_sec
= tv
.tv_sec
;
2241 sn
->date_nsec
= tv
.tv_usec
* 1000;
2242 sn
->vm_clock_nsec
= qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
);
2245 ret
= bdrv_snapshot_find(bs
, old_sn
, name
);
2247 pstrcpy(sn
->name
, sizeof(sn
->name
), old_sn
->name
);
2248 pstrcpy(sn
->id_str
, sizeof(sn
->id_str
), old_sn
->id_str
);
2250 pstrcpy(sn
->name
, sizeof(sn
->name
), name
);
2253 /* cast below needed for OpenBSD where tv_sec is still 'long' */
2254 localtime_r((const time_t *)&tv
.tv_sec
, &tm
);
2255 strftime(sn
->name
, sizeof(sn
->name
), "vm-%Y%m%d%H%M%S", &tm
);
2258 /* save the VM state */
2259 f
= qemu_fopen_bdrv(bs
, 1);
2261 error_setg(errp
, "Could not open VM state file");
2264 ret
= qemu_savevm_state(f
, errp
);
2265 vm_state_size
= qemu_ftell(f
);
2271 /* The bdrv_all_create_snapshot() call that follows acquires the AioContext
2272 * for itself. BDRV_POLL_WHILE() does not support nested locking because
2273 * it only releases the lock once. Therefore synchronous I/O will deadlock
2274 * unless we release the AioContext before bdrv_all_create_snapshot().
2276 aio_context_release(aio_context
);
2279 ret
= bdrv_all_create_snapshot(sn
, bs
, vm_state_size
, &bs
);
2281 error_setg(errp
, "Error while creating snapshot on '%s'",
2282 bdrv_get_device_name(bs
));
2290 aio_context_release(aio_context
);
2293 bdrv_drain_all_end();
2295 if (saved_vm_running
) {
2301 void qmp_xen_save_devices_state(const char *filename
, bool has_live
, bool live
,
2305 QIOChannelFile
*ioc
;
2306 int saved_vm_running
;
2310 /* live default to true so old version of Xen tool stack can have a
2311 * successfull live migration */
2315 saved_vm_running
= runstate_is_running();
2316 vm_stop(RUN_STATE_SAVE_VM
);
2317 global_state_store_running();
2319 ioc
= qio_channel_file_new_path(filename
, O_WRONLY
| O_CREAT
, 0660, errp
);
2323 qio_channel_set_name(QIO_CHANNEL(ioc
), "migration-xen-save-state");
2324 f
= qemu_fopen_channel_output(QIO_CHANNEL(ioc
));
2325 object_unref(OBJECT(ioc
));
2326 ret
= qemu_save_device_state(f
);
2327 if (ret
< 0 || qemu_fclose(f
) < 0) {
2328 error_setg(errp
, QERR_IO_ERROR
);
2330 /* libxl calls the QMP command "stop" before calling
2331 * "xen-save-devices-state" and in case of migration failure, libxl
2332 * would call "cont".
2333 * So call bdrv_inactivate_all (release locks) here to let the other
2334 * side of the migration take controle of the images.
2336 if (live
&& !saved_vm_running
) {
2337 ret
= bdrv_inactivate_all();
2339 error_setg(errp
, "%s: bdrv_inactivate_all() failed (%d)",
2346 if (saved_vm_running
) {
2351 void qmp_xen_load_devices_state(const char *filename
, Error
**errp
)
2354 QIOChannelFile
*ioc
;
2357 /* Guest must be paused before loading the device state; the RAM state
2358 * will already have been loaded by xc
2360 if (runstate_is_running()) {
2361 error_setg(errp
, "Cannot update device state while vm is running");
2364 vm_stop(RUN_STATE_RESTORE_VM
);
2366 ioc
= qio_channel_file_new_path(filename
, O_RDONLY
| O_BINARY
, 0, errp
);
2370 qio_channel_set_name(QIO_CHANNEL(ioc
), "migration-xen-load-state");
2371 f
= qemu_fopen_channel_input(QIO_CHANNEL(ioc
));
2372 object_unref(OBJECT(ioc
));
2374 ret
= qemu_loadvm_state(f
);
2377 error_setg(errp
, QERR_IO_ERROR
);
2379 migration_incoming_state_destroy();
2382 int load_snapshot(const char *name
, Error
**errp
)
2384 BlockDriverState
*bs
, *bs_vm_state
;
2385 QEMUSnapshotInfo sn
;
2388 AioContext
*aio_context
;
2389 MigrationIncomingState
*mis
= migration_incoming_get_current();
2391 if (!bdrv_all_can_snapshot(&bs
)) {
2393 "Device '%s' is writable but does not support snapshots",
2394 bdrv_get_device_name(bs
));
2397 ret
= bdrv_all_find_snapshot(name
, &bs
);
2400 "Device '%s' does not have the requested snapshot '%s'",
2401 bdrv_get_device_name(bs
), name
);
2405 bs_vm_state
= bdrv_all_find_vmstate_bs();
2407 error_setg(errp
, "No block device supports snapshots");
2410 aio_context
= bdrv_get_aio_context(bs_vm_state
);
2412 /* Don't even try to load empty VM states */
2413 aio_context_acquire(aio_context
);
2414 ret
= bdrv_snapshot_find(bs_vm_state
, &sn
, name
);
2415 aio_context_release(aio_context
);
2418 } else if (sn
.vm_state_size
== 0) {
2419 error_setg(errp
, "This is a disk-only snapshot. Revert to it "
2420 " offline using qemu-img");
2424 /* Flush all IO requests so they don't interfere with the new state. */
2425 bdrv_drain_all_begin();
2427 ret
= bdrv_all_goto_snapshot(name
, &bs
, errp
);
2429 error_prepend(errp
, "Could not load snapshot '%s' on '%s': ",
2430 name
, bdrv_get_device_name(bs
));
2434 /* restore the VM state */
2435 f
= qemu_fopen_bdrv(bs_vm_state
, 0);
2437 error_setg(errp
, "Could not open VM state file");
2442 qemu_system_reset(SHUTDOWN_CAUSE_NONE
);
2443 mis
->from_src_file
= f
;
2445 aio_context_acquire(aio_context
);
2446 ret
= qemu_loadvm_state(f
);
2447 migration_incoming_state_destroy();
2448 aio_context_release(aio_context
);
2450 bdrv_drain_all_end();
2453 error_setg(errp
, "Error %d while loading VM state", ret
);
2460 bdrv_drain_all_end();
2464 void vmstate_register_ram(MemoryRegion
*mr
, DeviceState
*dev
)
2466 qemu_ram_set_idstr(mr
->ram_block
,
2467 memory_region_name(mr
), dev
);
2470 void vmstate_unregister_ram(MemoryRegion
*mr
, DeviceState
*dev
)
2472 qemu_ram_unset_idstr(mr
->ram_block
);
2475 void vmstate_register_ram_global(MemoryRegion
*mr
)
2477 vmstate_register_ram(mr
, NULL
);
2480 bool vmstate_check_only_migratable(const VMStateDescription
*vmsd
)
2482 /* check needed if --only-migratable is specified */
2483 if (!migrate_get_current()->only_migratable
) {
2487 return !(vmsd
&& vmsd
->unmigratable
);