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"
60 #define ETH_P_RARP 0x8035
62 #define ARP_HTYPE_ETH 0x0001
63 #define ARP_PTYPE_IP 0x0800
64 #define ARP_OP_REQUEST_REV 0x3
66 const unsigned int postcopy_ram_discard_version
= 0;
68 /* Subcommands for QEMU_VM_COMMAND */
70 MIG_CMD_INVALID
= 0, /* Must be 0 */
71 MIG_CMD_OPEN_RETURN_PATH
, /* Tell the dest to open the Return path */
72 MIG_CMD_PING
, /* Request a PONG on the RP */
74 MIG_CMD_POSTCOPY_ADVISE
, /* Prior to any page transfers, just
75 warn we might want to do PC */
76 MIG_CMD_POSTCOPY_LISTEN
, /* Start listening for incoming
77 pages as it's running. */
78 MIG_CMD_POSTCOPY_RUN
, /* Start execution */
80 MIG_CMD_POSTCOPY_RAM_DISCARD
, /* A list of pages to discard that
81 were previously sent during
82 precopy but are dirty. */
83 MIG_CMD_PACKAGED
, /* Send a wrapped stream within this stream */
87 #define MAX_VM_CMD_PACKAGED_SIZE UINT32_MAX
88 static struct mig_cmd_args
{
89 ssize_t len
; /* -1 = variable */
92 [MIG_CMD_INVALID
] = { .len
= -1, .name
= "INVALID" },
93 [MIG_CMD_OPEN_RETURN_PATH
] = { .len
= 0, .name
= "OPEN_RETURN_PATH" },
94 [MIG_CMD_PING
] = { .len
= sizeof(uint32_t), .name
= "PING" },
95 [MIG_CMD_POSTCOPY_ADVISE
] = { .len
= -1, .name
= "POSTCOPY_ADVISE" },
96 [MIG_CMD_POSTCOPY_LISTEN
] = { .len
= 0, .name
= "POSTCOPY_LISTEN" },
97 [MIG_CMD_POSTCOPY_RUN
] = { .len
= 0, .name
= "POSTCOPY_RUN" },
98 [MIG_CMD_POSTCOPY_RAM_DISCARD
] = {
99 .len
= -1, .name
= "POSTCOPY_RAM_DISCARD" },
100 [MIG_CMD_PACKAGED
] = { .len
= 4, .name
= "PACKAGED" },
101 [MIG_CMD_MAX
] = { .len
= -1, .name
= "MAX" },
104 /* Note for MIG_CMD_POSTCOPY_ADVISE:
105 * The format of arguments is depending on postcopy mode:
106 * - postcopy RAM only
107 * uint64_t host page size
108 * uint64_t taget page size
110 * - postcopy RAM and postcopy dirty bitmaps
111 * format is the same as for postcopy RAM only
113 * - postcopy dirty bitmaps only
114 * Nothing. Command length field is 0.
116 * Be careful: adding a new postcopy entity with some other parameters should
117 * not break format self-description ability. Good way is to introduce some
118 * generic extendable format with an exception for two old entities.
121 static int announce_self_create(uint8_t *buf
,
124 /* Ethernet header. */
125 memset(buf
, 0xff, 6); /* destination MAC addr */
126 memcpy(buf
+ 6, mac_addr
, 6); /* source MAC addr */
127 *(uint16_t *)(buf
+ 12) = htons(ETH_P_RARP
); /* ethertype */
130 *(uint16_t *)(buf
+ 14) = htons(ARP_HTYPE_ETH
); /* hardware addr space */
131 *(uint16_t *)(buf
+ 16) = htons(ARP_PTYPE_IP
); /* protocol addr space */
132 *(buf
+ 18) = 6; /* hardware addr length (ethernet) */
133 *(buf
+ 19) = 4; /* protocol addr length (IPv4) */
134 *(uint16_t *)(buf
+ 20) = htons(ARP_OP_REQUEST_REV
); /* opcode */
135 memcpy(buf
+ 22, mac_addr
, 6); /* source hw addr */
136 memset(buf
+ 28, 0x00, 4); /* source protocol addr */
137 memcpy(buf
+ 32, mac_addr
, 6); /* target hw addr */
138 memset(buf
+ 38, 0x00, 4); /* target protocol addr */
140 /* Padding to get up to 60 bytes (ethernet min packet size, minus FCS). */
141 memset(buf
+ 42, 0x00, 18);
143 return 60; /* len (FCS will be added by hardware) */
146 static void qemu_announce_self_iter(NICState
*nic
, void *opaque
)
151 trace_qemu_announce_self_iter(qemu_ether_ntoa(&nic
->conf
->macaddr
));
152 len
= announce_self_create(buf
, nic
->conf
->macaddr
.a
);
154 qemu_send_packet_raw(qemu_get_queue(nic
), buf
, len
);
158 static void qemu_announce_self_once(void *opaque
)
160 static int count
= SELF_ANNOUNCE_ROUNDS
;
161 QEMUTimer
*timer
= *(QEMUTimer
**)opaque
;
163 qemu_foreach_nic(qemu_announce_self_iter
, NULL
);
166 /* delay 50ms, 150ms, 250ms, ... */
167 timer_mod(timer
, qemu_clock_get_ms(QEMU_CLOCK_REALTIME
) +
168 self_announce_delay(count
));
175 void qemu_announce_self(void)
177 static QEMUTimer
*timer
;
178 timer
= timer_new_ms(QEMU_CLOCK_REALTIME
, qemu_announce_self_once
, &timer
);
179 qemu_announce_self_once(&timer
);
182 /***********************************************************/
183 /* savevm/loadvm support */
185 static ssize_t
block_writev_buffer(void *opaque
, struct iovec
*iov
, int iovcnt
,
191 qemu_iovec_init_external(&qiov
, iov
, iovcnt
);
192 ret
= bdrv_writev_vmstate(opaque
, &qiov
, pos
);
200 static ssize_t
block_get_buffer(void *opaque
, uint8_t *buf
, int64_t pos
,
203 return bdrv_load_vmstate(opaque
, buf
, pos
, size
);
206 static int bdrv_fclose(void *opaque
)
208 return bdrv_flush(opaque
);
211 static const QEMUFileOps bdrv_read_ops
= {
212 .get_buffer
= block_get_buffer
,
216 static const QEMUFileOps bdrv_write_ops
= {
217 .writev_buffer
= block_writev_buffer
,
221 static QEMUFile
*qemu_fopen_bdrv(BlockDriverState
*bs
, int is_writable
)
224 return qemu_fopen_ops(bs
, &bdrv_write_ops
);
226 return qemu_fopen_ops(bs
, &bdrv_read_ops
);
230 /* QEMUFile timer support.
231 * Not in qemu-file.c to not add qemu-timer.c as dependency to qemu-file.c
234 void timer_put(QEMUFile
*f
, QEMUTimer
*ts
)
236 uint64_t expire_time
;
238 expire_time
= timer_expire_time_ns(ts
);
239 qemu_put_be64(f
, expire_time
);
242 void timer_get(QEMUFile
*f
, QEMUTimer
*ts
)
244 uint64_t expire_time
;
246 expire_time
= qemu_get_be64(f
);
247 if (expire_time
!= -1) {
248 timer_mod_ns(ts
, expire_time
);
255 /* VMState timer support.
256 * Not in vmstate.c to not add qemu-timer.c as dependency to vmstate.c
259 static int get_timer(QEMUFile
*f
, void *pv
, size_t size
, VMStateField
*field
)
266 static int put_timer(QEMUFile
*f
, void *pv
, size_t size
, VMStateField
*field
,
275 const VMStateInfo vmstate_info_timer
= {
282 typedef struct CompatEntry
{
287 typedef struct SaveStateEntry
{
288 QTAILQ_ENTRY(SaveStateEntry
) entry
;
293 /* version id read from the stream */
296 /* section id read from the stream */
299 const VMStateDescription
*vmsd
;
305 typedef struct SaveState
{
306 QTAILQ_HEAD(, SaveStateEntry
) handlers
;
307 int global_section_id
;
310 uint32_t target_page_bits
;
313 static SaveState savevm_state
= {
314 .handlers
= QTAILQ_HEAD_INITIALIZER(savevm_state
.handlers
),
315 .global_section_id
= 0,
318 static int configuration_pre_save(void *opaque
)
320 SaveState
*state
= opaque
;
321 const char *current_name
= MACHINE_GET_CLASS(current_machine
)->name
;
323 state
->len
= strlen(current_name
);
324 state
->name
= current_name
;
325 state
->target_page_bits
= qemu_target_page_bits();
330 static int configuration_pre_load(void *opaque
)
332 SaveState
*state
= opaque
;
334 /* If there is no target-page-bits subsection it means the source
335 * predates the variable-target-page-bits support and is using the
336 * minimum possible value for this CPU.
338 state
->target_page_bits
= qemu_target_page_bits_min();
342 static int configuration_post_load(void *opaque
, int version_id
)
344 SaveState
*state
= opaque
;
345 const char *current_name
= MACHINE_GET_CLASS(current_machine
)->name
;
347 if (strncmp(state
->name
, current_name
, state
->len
) != 0) {
348 error_report("Machine type received is '%.*s' and local is '%s'",
349 (int) state
->len
, state
->name
, current_name
);
353 if (state
->target_page_bits
!= qemu_target_page_bits()) {
354 error_report("Received TARGET_PAGE_BITS is %d but local is %d",
355 state
->target_page_bits
, qemu_target_page_bits());
362 /* The target-page-bits subsection is present only if the
363 * target page size is not the same as the default (ie the
364 * minimum page size for a variable-page-size guest CPU).
365 * If it is present then it contains the actual target page
366 * bits for the machine, and migration will fail if the
367 * two ends don't agree about it.
369 static bool vmstate_target_page_bits_needed(void *opaque
)
371 return qemu_target_page_bits()
372 > qemu_target_page_bits_min();
375 static const VMStateDescription vmstate_target_page_bits
= {
376 .name
= "configuration/target-page-bits",
378 .minimum_version_id
= 1,
379 .needed
= vmstate_target_page_bits_needed
,
380 .fields
= (VMStateField
[]) {
381 VMSTATE_UINT32(target_page_bits
, SaveState
),
382 VMSTATE_END_OF_LIST()
386 static const VMStateDescription vmstate_configuration
= {
387 .name
= "configuration",
389 .pre_load
= configuration_pre_load
,
390 .post_load
= configuration_post_load
,
391 .pre_save
= configuration_pre_save
,
392 .fields
= (VMStateField
[]) {
393 VMSTATE_UINT32(len
, SaveState
),
394 VMSTATE_VBUFFER_ALLOC_UINT32(name
, SaveState
, 0, NULL
, len
),
395 VMSTATE_END_OF_LIST()
397 .subsections
= (const VMStateDescription
*[]) {
398 &vmstate_target_page_bits
,
403 static void dump_vmstate_vmsd(FILE *out_file
,
404 const VMStateDescription
*vmsd
, int indent
,
407 static void dump_vmstate_vmsf(FILE *out_file
, const VMStateField
*field
,
410 fprintf(out_file
, "%*s{\n", indent
, "");
412 fprintf(out_file
, "%*s\"field\": \"%s\",\n", indent
, "", field
->name
);
413 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
415 fprintf(out_file
, "%*s\"field_exists\": %s,\n", indent
, "",
416 field
->field_exists
? "true" : "false");
417 fprintf(out_file
, "%*s\"size\": %zu", indent
, "", field
->size
);
418 if (field
->vmsd
!= NULL
) {
419 fprintf(out_file
, ",\n");
420 dump_vmstate_vmsd(out_file
, field
->vmsd
, indent
, false);
422 fprintf(out_file
, "\n%*s}", indent
- 2, "");
425 static void dump_vmstate_vmss(FILE *out_file
,
426 const VMStateDescription
**subsection
,
429 if (*subsection
!= NULL
) {
430 dump_vmstate_vmsd(out_file
, *subsection
, indent
, true);
434 static void dump_vmstate_vmsd(FILE *out_file
,
435 const VMStateDescription
*vmsd
, int indent
,
439 fprintf(out_file
, "%*s{\n", indent
, "");
441 fprintf(out_file
, "%*s\"%s\": {\n", indent
, "", "Description");
444 fprintf(out_file
, "%*s\"name\": \"%s\",\n", indent
, "", vmsd
->name
);
445 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
447 fprintf(out_file
, "%*s\"minimum_version_id\": %d", indent
, "",
448 vmsd
->minimum_version_id
);
449 if (vmsd
->fields
!= NULL
) {
450 const VMStateField
*field
= vmsd
->fields
;
453 fprintf(out_file
, ",\n%*s\"Fields\": [\n", indent
, "");
455 while (field
->name
!= NULL
) {
456 if (field
->flags
& VMS_MUST_EXIST
) {
457 /* Ignore VMSTATE_VALIDATE bits; these don't get migrated */
462 fprintf(out_file
, ",\n");
464 dump_vmstate_vmsf(out_file
, field
, indent
+ 2);
468 fprintf(out_file
, "\n%*s]", indent
, "");
470 if (vmsd
->subsections
!= NULL
) {
471 const VMStateDescription
**subsection
= vmsd
->subsections
;
474 fprintf(out_file
, ",\n%*s\"Subsections\": [\n", indent
, "");
476 while (*subsection
!= NULL
) {
478 fprintf(out_file
, ",\n");
480 dump_vmstate_vmss(out_file
, subsection
, indent
+ 2);
484 fprintf(out_file
, "\n%*s]", indent
, "");
486 fprintf(out_file
, "\n%*s}", indent
- 2, "");
489 static void dump_machine_type(FILE *out_file
)
493 mc
= MACHINE_GET_CLASS(current_machine
);
495 fprintf(out_file
, " \"vmschkmachine\": {\n");
496 fprintf(out_file
, " \"Name\": \"%s\"\n", mc
->name
);
497 fprintf(out_file
, " },\n");
500 void dump_vmstate_json_to_file(FILE *out_file
)
505 fprintf(out_file
, "{\n");
506 dump_machine_type(out_file
);
509 list
= object_class_get_list(TYPE_DEVICE
, true);
510 for (elt
= list
; elt
; elt
= elt
->next
) {
511 DeviceClass
*dc
= OBJECT_CLASS_CHECK(DeviceClass
, elt
->data
,
521 fprintf(out_file
, ",\n");
523 name
= object_class_get_name(OBJECT_CLASS(dc
));
524 fprintf(out_file
, "%*s\"%s\": {\n", indent
, "", name
);
526 fprintf(out_file
, "%*s\"Name\": \"%s\",\n", indent
, "", name
);
527 fprintf(out_file
, "%*s\"version_id\": %d,\n", indent
, "",
528 dc
->vmsd
->version_id
);
529 fprintf(out_file
, "%*s\"minimum_version_id\": %d,\n", indent
, "",
530 dc
->vmsd
->minimum_version_id
);
532 dump_vmstate_vmsd(out_file
, dc
->vmsd
, indent
, false);
534 fprintf(out_file
, "\n%*s}", indent
- 2, "");
537 fprintf(out_file
, "\n}\n");
541 static int calculate_new_instance_id(const char *idstr
)
546 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
547 if (strcmp(idstr
, se
->idstr
) == 0
548 && instance_id
<= se
->instance_id
) {
549 instance_id
= se
->instance_id
+ 1;
555 static int calculate_compat_instance_id(const char *idstr
)
560 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
565 if (strcmp(idstr
, se
->compat
->idstr
) == 0
566 && instance_id
<= se
->compat
->instance_id
) {
567 instance_id
= se
->compat
->instance_id
+ 1;
573 static inline MigrationPriority
save_state_priority(SaveStateEntry
*se
)
576 return se
->vmsd
->priority
;
578 return MIG_PRI_DEFAULT
;
581 static void savevm_state_handler_insert(SaveStateEntry
*nse
)
583 MigrationPriority priority
= save_state_priority(nse
);
586 assert(priority
<= MIG_PRI_MAX
);
588 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
589 if (save_state_priority(se
) < priority
) {
595 QTAILQ_INSERT_BEFORE(se
, nse
, entry
);
597 QTAILQ_INSERT_TAIL(&savevm_state
.handlers
, nse
, entry
);
601 /* TODO: Individual devices generally have very little idea about the rest
602 of the system, so instance_id should be removed/replaced.
603 Meanwhile pass -1 as instance_id if you do not already have a clearly
604 distinguishing id for all instances of your device class. */
605 int register_savevm_live(DeviceState
*dev
,
614 se
= g_new0(SaveStateEntry
, 1);
615 se
->version_id
= version_id
;
616 se
->section_id
= savevm_state
.global_section_id
++;
620 /* if this is a live_savem then set is_ram */
621 if (ops
->save_setup
!= NULL
) {
626 char *id
= qdev_get_dev_path(dev
);
628 if (snprintf(se
->idstr
, sizeof(se
->idstr
), "%s/", id
) >=
630 error_report("Path too long for VMState (%s)", id
);
638 se
->compat
= g_new0(CompatEntry
, 1);
639 pstrcpy(se
->compat
->idstr
, sizeof(se
->compat
->idstr
), idstr
);
640 se
->compat
->instance_id
= instance_id
== -1 ?
641 calculate_compat_instance_id(idstr
) : instance_id
;
645 pstrcat(se
->idstr
, sizeof(se
->idstr
), idstr
);
647 if (instance_id
== -1) {
648 se
->instance_id
= calculate_new_instance_id(se
->idstr
);
650 se
->instance_id
= instance_id
;
652 assert(!se
->compat
|| se
->instance_id
== 0);
653 savevm_state_handler_insert(se
);
657 void unregister_savevm(DeviceState
*dev
, const char *idstr
, void *opaque
)
659 SaveStateEntry
*se
, *new_se
;
663 char *path
= qdev_get_dev_path(dev
);
665 pstrcpy(id
, sizeof(id
), path
);
666 pstrcat(id
, sizeof(id
), "/");
670 pstrcat(id
, sizeof(id
), idstr
);
672 QTAILQ_FOREACH_SAFE(se
, &savevm_state
.handlers
, entry
, new_se
) {
673 if (strcmp(se
->idstr
, id
) == 0 && se
->opaque
== opaque
) {
674 QTAILQ_REMOVE(&savevm_state
.handlers
, se
, entry
);
681 int vmstate_register_with_alias_id(DeviceState
*dev
, int instance_id
,
682 const VMStateDescription
*vmsd
,
683 void *opaque
, int alias_id
,
684 int required_for_version
,
689 /* If this triggers, alias support can be dropped for the vmsd. */
690 assert(alias_id
== -1 || required_for_version
>= vmsd
->minimum_version_id
);
692 se
= g_new0(SaveStateEntry
, 1);
693 se
->version_id
= vmsd
->version_id
;
694 se
->section_id
= savevm_state
.global_section_id
++;
697 se
->alias_id
= alias_id
;
700 char *id
= qdev_get_dev_path(dev
);
702 if (snprintf(se
->idstr
, sizeof(se
->idstr
), "%s/", id
) >=
704 error_setg(errp
, "Path too long for VMState (%s)", id
);
712 se
->compat
= g_new0(CompatEntry
, 1);
713 pstrcpy(se
->compat
->idstr
, sizeof(se
->compat
->idstr
), vmsd
->name
);
714 se
->compat
->instance_id
= instance_id
== -1 ?
715 calculate_compat_instance_id(vmsd
->name
) : instance_id
;
719 pstrcat(se
->idstr
, sizeof(se
->idstr
), vmsd
->name
);
721 if (instance_id
== -1) {
722 se
->instance_id
= calculate_new_instance_id(se
->idstr
);
724 se
->instance_id
= instance_id
;
726 assert(!se
->compat
|| se
->instance_id
== 0);
727 savevm_state_handler_insert(se
);
731 void vmstate_unregister(DeviceState
*dev
, const VMStateDescription
*vmsd
,
734 SaveStateEntry
*se
, *new_se
;
736 QTAILQ_FOREACH_SAFE(se
, &savevm_state
.handlers
, entry
, new_se
) {
737 if (se
->vmsd
== vmsd
&& se
->opaque
== opaque
) {
738 QTAILQ_REMOVE(&savevm_state
.handlers
, se
, entry
);
745 static int vmstate_load(QEMUFile
*f
, SaveStateEntry
*se
)
747 trace_vmstate_load(se
->idstr
, se
->vmsd
? se
->vmsd
->name
: "(old)");
748 if (!se
->vmsd
) { /* Old style */
749 return se
->ops
->load_state(f
, se
->opaque
, se
->load_version_id
);
751 return vmstate_load_state(f
, se
->vmsd
, se
->opaque
, se
->load_version_id
);
754 static void vmstate_save_old_style(QEMUFile
*f
, SaveStateEntry
*se
, QJSON
*vmdesc
)
756 int64_t old_offset
, size
;
758 old_offset
= qemu_ftell_fast(f
);
759 se
->ops
->save_state(f
, se
->opaque
);
760 size
= qemu_ftell_fast(f
) - old_offset
;
763 json_prop_int(vmdesc
, "size", size
);
764 json_start_array(vmdesc
, "fields");
765 json_start_object(vmdesc
, NULL
);
766 json_prop_str(vmdesc
, "name", "data");
767 json_prop_int(vmdesc
, "size", size
);
768 json_prop_str(vmdesc
, "type", "buffer");
769 json_end_object(vmdesc
);
770 json_end_array(vmdesc
);
774 static int vmstate_save(QEMUFile
*f
, SaveStateEntry
*se
, QJSON
*vmdesc
)
776 trace_vmstate_save(se
->idstr
, se
->vmsd
? se
->vmsd
->name
: "(old)");
778 vmstate_save_old_style(f
, se
, vmdesc
);
781 return vmstate_save_state(f
, se
->vmsd
, se
->opaque
, vmdesc
);
785 * Write the header for device section (QEMU_VM_SECTION START/END/PART/FULL)
787 static void save_section_header(QEMUFile
*f
, SaveStateEntry
*se
,
788 uint8_t section_type
)
790 qemu_put_byte(f
, section_type
);
791 qemu_put_be32(f
, se
->section_id
);
793 if (section_type
== QEMU_VM_SECTION_FULL
||
794 section_type
== QEMU_VM_SECTION_START
) {
796 size_t len
= strlen(se
->idstr
);
797 qemu_put_byte(f
, len
);
798 qemu_put_buffer(f
, (uint8_t *)se
->idstr
, len
);
800 qemu_put_be32(f
, se
->instance_id
);
801 qemu_put_be32(f
, se
->version_id
);
806 * Write a footer onto device sections that catches cases misformatted device
809 static void save_section_footer(QEMUFile
*f
, SaveStateEntry
*se
)
811 if (migrate_get_current()->send_section_footer
) {
812 qemu_put_byte(f
, QEMU_VM_SECTION_FOOTER
);
813 qemu_put_be32(f
, se
->section_id
);
818 * qemu_savevm_command_send: Send a 'QEMU_VM_COMMAND' type element with the
819 * command and associated data.
821 * @f: File to send command on
822 * @command: Command type to send
823 * @len: Length of associated data
824 * @data: Data associated with command.
826 static void qemu_savevm_command_send(QEMUFile
*f
,
827 enum qemu_vm_cmd command
,
831 trace_savevm_command_send(command
, len
);
832 qemu_put_byte(f
, QEMU_VM_COMMAND
);
833 qemu_put_be16(f
, (uint16_t)command
);
834 qemu_put_be16(f
, len
);
835 qemu_put_buffer(f
, data
, len
);
839 void qemu_savevm_send_ping(QEMUFile
*f
, uint32_t value
)
843 trace_savevm_send_ping(value
);
844 buf
= cpu_to_be32(value
);
845 qemu_savevm_command_send(f
, MIG_CMD_PING
, sizeof(value
), (uint8_t *)&buf
);
848 void qemu_savevm_send_open_return_path(QEMUFile
*f
)
850 trace_savevm_send_open_return_path();
851 qemu_savevm_command_send(f
, MIG_CMD_OPEN_RETURN_PATH
, 0, NULL
);
854 /* We have a buffer of data to send; we don't want that all to be loaded
855 * by the command itself, so the command contains just the length of the
856 * extra buffer that we then send straight after it.
857 * TODO: Must be a better way to organise that
863 int qemu_savevm_send_packaged(QEMUFile
*f
, const uint8_t *buf
, size_t len
)
867 if (len
> MAX_VM_CMD_PACKAGED_SIZE
) {
868 error_report("%s: Unreasonably large packaged state: %zu",
873 tmp
= cpu_to_be32(len
);
875 trace_qemu_savevm_send_packaged();
876 qemu_savevm_command_send(f
, MIG_CMD_PACKAGED
, 4, (uint8_t *)&tmp
);
878 qemu_put_buffer(f
, buf
, len
);
883 /* Send prior to any postcopy transfer */
884 void qemu_savevm_send_postcopy_advise(QEMUFile
*f
)
886 if (migrate_postcopy_ram()) {
888 tmp
[0] = cpu_to_be64(ram_pagesize_summary());
889 tmp
[1] = cpu_to_be64(qemu_target_page_size());
891 trace_qemu_savevm_send_postcopy_advise();
892 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_ADVISE
,
895 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_ADVISE
, 0, NULL
);
899 /* Sent prior to starting the destination running in postcopy, discard pages
900 * that have already been sent but redirtied on the source.
901 * CMD_POSTCOPY_RAM_DISCARD consist of:
903 * byte Length of name field (not including 0)
904 * n x byte RAM block name
905 * byte 0 terminator (just for safety)
906 * n x Byte ranges within the named RAMBlock
907 * be64 Start of the range
910 * name: RAMBlock name that these entries are part of
911 * len: Number of page entries
912 * start_list: 'len' addresses
913 * length_list: 'len' addresses
916 void qemu_savevm_send_postcopy_ram_discard(QEMUFile
*f
, const char *name
,
918 uint64_t *start_list
,
919 uint64_t *length_list
)
924 size_t name_len
= strlen(name
);
926 trace_qemu_savevm_send_postcopy_ram_discard(name
, len
);
927 assert(name_len
< 256);
928 buf
= g_malloc0(1 + 1 + name_len
+ 1 + (8 + 8) * len
);
929 buf
[0] = postcopy_ram_discard_version
;
931 memcpy(buf
+ 2, name
, name_len
);
932 tmplen
= 2 + name_len
;
933 buf
[tmplen
++] = '\0';
935 for (t
= 0; t
< len
; t
++) {
936 stq_be_p(buf
+ tmplen
, start_list
[t
]);
938 stq_be_p(buf
+ tmplen
, length_list
[t
]);
941 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RAM_DISCARD
, tmplen
, buf
);
945 /* Get the destination into a state where it can receive postcopy data. */
946 void qemu_savevm_send_postcopy_listen(QEMUFile
*f
)
948 trace_savevm_send_postcopy_listen();
949 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_LISTEN
, 0, NULL
);
952 /* Kick the destination into running */
953 void qemu_savevm_send_postcopy_run(QEMUFile
*f
)
955 trace_savevm_send_postcopy_run();
956 qemu_savevm_command_send(f
, MIG_CMD_POSTCOPY_RUN
, 0, NULL
);
959 bool qemu_savevm_state_blocked(Error
**errp
)
963 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
964 if (se
->vmsd
&& se
->vmsd
->unmigratable
) {
965 error_setg(errp
, "State blocked by non-migratable device '%s'",
973 void qemu_savevm_state_header(QEMUFile
*f
)
975 trace_savevm_state_header();
976 qemu_put_be32(f
, QEMU_VM_FILE_MAGIC
);
977 qemu_put_be32(f
, QEMU_VM_FILE_VERSION
);
979 if (migrate_get_current()->send_configuration
) {
980 qemu_put_byte(f
, QEMU_VM_CONFIGURATION
);
981 vmstate_save_state(f
, &vmstate_configuration
, &savevm_state
, 0);
985 void qemu_savevm_state_setup(QEMUFile
*f
)
990 trace_savevm_state_setup();
991 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
992 if (!se
->ops
|| !se
->ops
->save_setup
) {
995 if (se
->ops
&& se
->ops
->is_active
) {
996 if (!se
->ops
->is_active(se
->opaque
)) {
1000 save_section_header(f
, se
, QEMU_VM_SECTION_START
);
1002 ret
= se
->ops
->save_setup(f
, se
->opaque
);
1003 save_section_footer(f
, se
);
1005 qemu_file_set_error(f
, ret
);
1012 * this function has three return values:
1013 * negative: there was one error, and we have -errno.
1014 * 0 : We haven't finished, caller have to go again
1015 * 1 : We have finished, we can go to complete phase
1017 int qemu_savevm_state_iterate(QEMUFile
*f
, bool postcopy
)
1022 trace_savevm_state_iterate();
1023 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1024 if (!se
->ops
|| !se
->ops
->save_live_iterate
) {
1027 if (se
->ops
&& se
->ops
->is_active
) {
1028 if (!se
->ops
->is_active(se
->opaque
)) {
1032 if (se
->ops
&& se
->ops
->is_active_iterate
) {
1033 if (!se
->ops
->is_active_iterate(se
->opaque
)) {
1038 * In the postcopy phase, any device that doesn't know how to
1039 * do postcopy should have saved it's state in the _complete
1040 * call that's already run, it might get confused if we call
1041 * iterate afterwards.
1044 !(se
->ops
->has_postcopy
&& se
->ops
->has_postcopy(se
->opaque
))) {
1047 if (qemu_file_rate_limit(f
)) {
1050 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1052 save_section_header(f
, se
, QEMU_VM_SECTION_PART
);
1054 ret
= se
->ops
->save_live_iterate(f
, se
->opaque
);
1055 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1056 save_section_footer(f
, se
);
1059 qemu_file_set_error(f
, ret
);
1062 /* Do not proceed to the next vmstate before this one reported
1063 completion of the current stage. This serializes the migration
1064 and reduces the probability that a faster changing state is
1065 synchronized over and over again. */
1072 static bool should_send_vmdesc(void)
1074 MachineState
*machine
= MACHINE(qdev_get_machine());
1075 bool in_postcopy
= migration_in_postcopy();
1076 return !machine
->suppress_vmdesc
&& !in_postcopy
;
1080 * Calls the save_live_complete_postcopy methods
1081 * causing the last few pages to be sent immediately and doing any associated
1083 * Note postcopy also calls qemu_savevm_state_complete_precopy to complete
1084 * all the other devices, but that happens at the point we switch to postcopy.
1086 void qemu_savevm_state_complete_postcopy(QEMUFile
*f
)
1091 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1092 if (!se
->ops
|| !se
->ops
->save_live_complete_postcopy
) {
1095 if (se
->ops
&& se
->ops
->is_active
) {
1096 if (!se
->ops
->is_active(se
->opaque
)) {
1100 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1102 qemu_put_byte(f
, QEMU_VM_SECTION_END
);
1103 qemu_put_be32(f
, se
->section_id
);
1105 ret
= se
->ops
->save_live_complete_postcopy(f
, se
->opaque
);
1106 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1107 save_section_footer(f
, se
);
1109 qemu_file_set_error(f
, ret
);
1114 qemu_put_byte(f
, QEMU_VM_EOF
);
1118 int qemu_savevm_state_complete_precopy(QEMUFile
*f
, bool iterable_only
,
1119 bool inactivate_disks
)
1125 bool in_postcopy
= migration_in_postcopy();
1127 trace_savevm_state_complete_precopy();
1129 cpu_synchronize_all_states();
1131 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1133 (in_postcopy
&& se
->ops
->has_postcopy
&&
1134 se
->ops
->has_postcopy(se
->opaque
)) ||
1135 (in_postcopy
&& !iterable_only
) ||
1136 !se
->ops
->save_live_complete_precopy
) {
1140 if (se
->ops
&& se
->ops
->is_active
) {
1141 if (!se
->ops
->is_active(se
->opaque
)) {
1145 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1147 save_section_header(f
, se
, QEMU_VM_SECTION_END
);
1149 ret
= se
->ops
->save_live_complete_precopy(f
, se
->opaque
);
1150 trace_savevm_section_end(se
->idstr
, se
->section_id
, ret
);
1151 save_section_footer(f
, se
);
1153 qemu_file_set_error(f
, ret
);
1158 if (iterable_only
) {
1162 vmdesc
= qjson_new();
1163 json_prop_int(vmdesc
, "page_size", qemu_target_page_size());
1164 json_start_array(vmdesc
, "devices");
1165 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1167 if ((!se
->ops
|| !se
->ops
->save_state
) && !se
->vmsd
) {
1170 if (se
->vmsd
&& !vmstate_save_needed(se
->vmsd
, se
->opaque
)) {
1171 trace_savevm_section_skip(se
->idstr
, se
->section_id
);
1175 trace_savevm_section_start(se
->idstr
, se
->section_id
);
1177 json_start_object(vmdesc
, NULL
);
1178 json_prop_str(vmdesc
, "name", se
->idstr
);
1179 json_prop_int(vmdesc
, "instance_id", se
->instance_id
);
1181 save_section_header(f
, se
, QEMU_VM_SECTION_FULL
);
1182 ret
= vmstate_save(f
, se
, vmdesc
);
1184 qemu_file_set_error(f
, ret
);
1187 trace_savevm_section_end(se
->idstr
, se
->section_id
, 0);
1188 save_section_footer(f
, se
);
1190 json_end_object(vmdesc
);
1193 if (inactivate_disks
) {
1194 /* Inactivate before sending QEMU_VM_EOF so that the
1195 * bdrv_invalidate_cache_all() on the other end won't fail. */
1196 ret
= bdrv_inactivate_all();
1198 error_report("%s: bdrv_inactivate_all() failed (%d)",
1200 qemu_file_set_error(f
, ret
);
1205 /* Postcopy stream will still be going */
1206 qemu_put_byte(f
, QEMU_VM_EOF
);
1209 json_end_array(vmdesc
);
1210 qjson_finish(vmdesc
);
1211 vmdesc_len
= strlen(qjson_get_str(vmdesc
));
1213 if (should_send_vmdesc()) {
1214 qemu_put_byte(f
, QEMU_VM_VMDESCRIPTION
);
1215 qemu_put_be32(f
, vmdesc_len
);
1216 qemu_put_buffer(f
, (uint8_t *)qjson_get_str(vmdesc
), vmdesc_len
);
1218 qjson_destroy(vmdesc
);
1224 /* Give an estimate of the amount left to be transferred,
1225 * the result is split into the amount for units that can and
1226 * for units that can't do postcopy.
1228 void qemu_savevm_state_pending(QEMUFile
*f
, uint64_t threshold_size
,
1229 uint64_t *res_precopy_only
,
1230 uint64_t *res_compatible
,
1231 uint64_t *res_postcopy_only
)
1235 *res_precopy_only
= 0;
1236 *res_compatible
= 0;
1237 *res_postcopy_only
= 0;
1240 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1241 if (!se
->ops
|| !se
->ops
->save_live_pending
) {
1244 if (se
->ops
&& se
->ops
->is_active
) {
1245 if (!se
->ops
->is_active(se
->opaque
)) {
1249 se
->ops
->save_live_pending(f
, se
->opaque
, threshold_size
,
1250 res_precopy_only
, res_compatible
,
1255 void qemu_savevm_state_cleanup(void)
1259 trace_savevm_state_cleanup();
1260 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1261 if (se
->ops
&& se
->ops
->save_cleanup
) {
1262 se
->ops
->save_cleanup(se
->opaque
);
1267 static int qemu_savevm_state(QEMUFile
*f
, Error
**errp
)
1270 MigrationState
*ms
= migrate_get_current();
1271 MigrationStatus status
;
1275 ms
->to_dst_file
= f
;
1277 if (migration_is_blocked(errp
)) {
1282 if (migrate_use_block()) {
1283 error_setg(errp
, "Block migration and snapshots are incompatible");
1288 qemu_mutex_unlock_iothread();
1289 qemu_savevm_state_header(f
);
1290 qemu_savevm_state_setup(f
);
1291 qemu_mutex_lock_iothread();
1293 while (qemu_file_get_error(f
) == 0) {
1294 if (qemu_savevm_state_iterate(f
, false) > 0) {
1299 ret
= qemu_file_get_error(f
);
1301 qemu_savevm_state_complete_precopy(f
, false, false);
1302 ret
= qemu_file_get_error(f
);
1304 qemu_savevm_state_cleanup();
1306 error_setg_errno(errp
, -ret
, "Error while writing VM state");
1311 status
= MIGRATION_STATUS_FAILED
;
1313 status
= MIGRATION_STATUS_COMPLETED
;
1315 migrate_set_state(&ms
->state
, MIGRATION_STATUS_SETUP
, status
);
1317 /* f is outer parameter, it should not stay in global migration state after
1318 * this function finished */
1319 ms
->to_dst_file
= NULL
;
1324 static int qemu_save_device_state(QEMUFile
*f
)
1328 qemu_put_be32(f
, QEMU_VM_FILE_MAGIC
);
1329 qemu_put_be32(f
, QEMU_VM_FILE_VERSION
);
1331 cpu_synchronize_all_states();
1333 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1339 if ((!se
->ops
|| !se
->ops
->save_state
) && !se
->vmsd
) {
1342 if (se
->vmsd
&& !vmstate_save_needed(se
->vmsd
, se
->opaque
)) {
1346 save_section_header(f
, se
, QEMU_VM_SECTION_FULL
);
1348 ret
= vmstate_save(f
, se
, NULL
);
1353 save_section_footer(f
, se
);
1356 qemu_put_byte(f
, QEMU_VM_EOF
);
1358 return qemu_file_get_error(f
);
1361 static SaveStateEntry
*find_se(const char *idstr
, int instance_id
)
1365 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1366 if (!strcmp(se
->idstr
, idstr
) &&
1367 (instance_id
== se
->instance_id
||
1368 instance_id
== se
->alias_id
))
1370 /* Migrating from an older version? */
1371 if (strstr(se
->idstr
, idstr
) && se
->compat
) {
1372 if (!strcmp(se
->compat
->idstr
, idstr
) &&
1373 (instance_id
== se
->compat
->instance_id
||
1374 instance_id
== se
->alias_id
))
1381 enum LoadVMExitCodes
{
1382 /* Allow a command to quit all layers of nested loadvm loops */
1386 static int qemu_loadvm_state_main(QEMUFile
*f
, MigrationIncomingState
*mis
);
1388 /* ------ incoming postcopy messages ------ */
1389 /* 'advise' arrives before any transfers just to tell us that a postcopy
1390 * *might* happen - it might be skipped if precopy transferred everything
1393 static int loadvm_postcopy_handle_advise(MigrationIncomingState
*mis
,
1396 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_ADVISE
);
1397 uint64_t remote_pagesize_summary
, local_pagesize_summary
, remote_tps
;
1399 trace_loadvm_postcopy_handle_advise();
1400 if (ps
!= POSTCOPY_INCOMING_NONE
) {
1401 error_report("CMD_POSTCOPY_ADVISE in wrong postcopy state (%d)", ps
);
1407 if (migrate_postcopy_ram()) {
1408 error_report("RAM postcopy is enabled but have 0 byte advise");
1413 if (!migrate_postcopy_ram()) {
1414 error_report("RAM postcopy is disabled but have 16 byte advise");
1419 error_report("CMD_POSTCOPY_ADVISE invalid length (%d)", len
);
1423 if (!postcopy_ram_supported_by_host(mis
)) {
1424 postcopy_state_set(POSTCOPY_INCOMING_NONE
);
1428 remote_pagesize_summary
= qemu_get_be64(mis
->from_src_file
);
1429 local_pagesize_summary
= ram_pagesize_summary();
1431 if (remote_pagesize_summary
!= local_pagesize_summary
) {
1433 * This detects two potential causes of mismatch:
1434 * a) A mismatch in host page sizes
1435 * Some combinations of mismatch are probably possible but it gets
1436 * a bit more complicated. In particular we need to place whole
1437 * host pages on the dest at once, and we need to ensure that we
1438 * handle dirtying to make sure we never end up sending part of
1439 * a hostpage on it's own.
1440 * b) The use of different huge page sizes on source/destination
1441 * a more fine grain test is performed during RAM block migration
1442 * but this test here causes a nice early clear failure, and
1443 * also fails when passed to an older qemu that doesn't
1446 error_report("Postcopy needs matching RAM page sizes (s=%" PRIx64
1448 remote_pagesize_summary
, local_pagesize_summary
);
1452 remote_tps
= qemu_get_be64(mis
->from_src_file
);
1453 if (remote_tps
!= qemu_target_page_size()) {
1455 * Again, some differences could be dealt with, but for now keep it
1458 error_report("Postcopy needs matching target page sizes (s=%d d=%zd)",
1459 (int)remote_tps
, qemu_target_page_size());
1463 if (ram_postcopy_incoming_init(mis
)) {
1467 postcopy_state_set(POSTCOPY_INCOMING_ADVISE
);
1472 /* After postcopy we will be told to throw some pages away since they're
1473 * dirty and will have to be demand fetched. Must happen before CPU is
1475 * There can be 0..many of these messages, each encoding multiple pages.
1477 static int loadvm_postcopy_ram_handle_discard(MigrationIncomingState
*mis
,
1482 PostcopyState ps
= postcopy_state_get();
1484 trace_loadvm_postcopy_ram_handle_discard();
1487 case POSTCOPY_INCOMING_ADVISE
:
1489 tmp
= postcopy_ram_prepare_discard(mis
);
1495 case POSTCOPY_INCOMING_DISCARD
:
1496 /* Expected state */
1500 error_report("CMD_POSTCOPY_RAM_DISCARD in wrong postcopy state (%d)",
1504 /* We're expecting a
1506 * a RAM ID string (length byte, name, 0 term)
1507 * then at least 1 16 byte chunk
1509 if (len
< (1 + 1 + 1 + 1 + 2 * 8)) {
1510 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len
);
1514 tmp
= qemu_get_byte(mis
->from_src_file
);
1515 if (tmp
!= postcopy_ram_discard_version
) {
1516 error_report("CMD_POSTCOPY_RAM_DISCARD invalid version (%d)", tmp
);
1520 if (!qemu_get_counted_string(mis
->from_src_file
, ramid
)) {
1521 error_report("CMD_POSTCOPY_RAM_DISCARD Failed to read RAMBlock ID");
1524 tmp
= qemu_get_byte(mis
->from_src_file
);
1526 error_report("CMD_POSTCOPY_RAM_DISCARD missing nil (%d)", tmp
);
1530 len
-= 3 + strlen(ramid
);
1532 error_report("CMD_POSTCOPY_RAM_DISCARD invalid length (%d)", len
);
1535 trace_loadvm_postcopy_ram_handle_discard_header(ramid
, len
);
1537 uint64_t start_addr
, block_length
;
1538 start_addr
= qemu_get_be64(mis
->from_src_file
);
1539 block_length
= qemu_get_be64(mis
->from_src_file
);
1542 int ret
= ram_discard_range(ramid
, start_addr
, block_length
);
1547 trace_loadvm_postcopy_ram_handle_discard_end();
1553 * Triggered by a postcopy_listen command; this thread takes over reading
1554 * the input stream, leaving the main thread free to carry on loading the rest
1555 * of the device state (from RAM).
1556 * (TODO:This could do with being in a postcopy file - but there again it's
1557 * just another input loop, not that postcopy specific)
1559 static void *postcopy_ram_listen_thread(void *opaque
)
1561 QEMUFile
*f
= opaque
;
1562 MigrationIncomingState
*mis
= migration_incoming_get_current();
1565 migrate_set_state(&mis
->state
, MIGRATION_STATUS_ACTIVE
,
1566 MIGRATION_STATUS_POSTCOPY_ACTIVE
);
1567 qemu_sem_post(&mis
->listen_thread_sem
);
1568 trace_postcopy_ram_listen_thread_start();
1571 * Because we're a thread and not a coroutine we can't yield
1572 * in qemu_file, and thus we must be blocking now.
1574 qemu_file_set_blocking(f
, true);
1575 load_res
= qemu_loadvm_state_main(f
, mis
);
1576 /* And non-blocking again so we don't block in any cleanup */
1577 qemu_file_set_blocking(f
, false);
1579 trace_postcopy_ram_listen_thread_exit();
1581 error_report("%s: loadvm failed: %d", __func__
, load_res
);
1582 qemu_file_set_error(f
, load_res
);
1583 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1584 MIGRATION_STATUS_FAILED
);
1587 * This looks good, but it's possible that the device loading in the
1588 * main thread hasn't finished yet, and so we might not be in 'RUN'
1589 * state yet; wait for the end of the main thread.
1591 qemu_event_wait(&mis
->main_thread_load_event
);
1593 postcopy_ram_incoming_cleanup(mis
);
1597 * If something went wrong then we have a bad state so exit;
1598 * depending how far we got it might be possible at this point
1599 * to leave the guest running and fire MCEs for pages that never
1600 * arrived as a desperate recovery step.
1605 migrate_set_state(&mis
->state
, MIGRATION_STATUS_POSTCOPY_ACTIVE
,
1606 MIGRATION_STATUS_COMPLETED
);
1608 * If everything has worked fine, then the main thread has waited
1609 * for us to start, and we're the last use of the mis.
1610 * (If something broke then qemu will have to exit anyway since it's
1611 * got a bad migration state).
1613 migration_incoming_state_destroy();
1614 qemu_loadvm_state_cleanup();
1619 /* After this message we must be able to immediately receive postcopy data */
1620 static int loadvm_postcopy_handle_listen(MigrationIncomingState
*mis
)
1622 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_LISTENING
);
1623 trace_loadvm_postcopy_handle_listen();
1624 if (ps
!= POSTCOPY_INCOMING_ADVISE
&& ps
!= POSTCOPY_INCOMING_DISCARD
) {
1625 error_report("CMD_POSTCOPY_LISTEN in wrong postcopy state (%d)", ps
);
1628 if (ps
== POSTCOPY_INCOMING_ADVISE
) {
1630 * A rare case, we entered listen without having to do any discards,
1631 * so do the setup that's normally done at the time of the 1st discard.
1633 if (migrate_postcopy_ram()) {
1634 postcopy_ram_prepare_discard(mis
);
1639 * Sensitise RAM - can now generate requests for blocks that don't exist
1640 * However, at this point the CPU shouldn't be running, and the IO
1641 * shouldn't be doing anything yet so don't actually expect requests
1643 if (migrate_postcopy_ram()) {
1644 if (postcopy_ram_enable_notify(mis
)) {
1649 if (mis
->have_listen_thread
) {
1650 error_report("CMD_POSTCOPY_RAM_LISTEN already has a listen thread");
1654 mis
->have_listen_thread
= true;
1655 /* Start up the listening thread and wait for it to signal ready */
1656 qemu_sem_init(&mis
->listen_thread_sem
, 0);
1657 qemu_thread_create(&mis
->listen_thread
, "postcopy/listen",
1658 postcopy_ram_listen_thread
, mis
->from_src_file
,
1659 QEMU_THREAD_DETACHED
);
1660 qemu_sem_wait(&mis
->listen_thread_sem
);
1661 qemu_sem_destroy(&mis
->listen_thread_sem
);
1671 static void loadvm_postcopy_handle_run_bh(void *opaque
)
1673 Error
*local_err
= NULL
;
1674 HandleRunBhData
*data
= opaque
;
1676 /* TODO we should move all of this lot into postcopy_ram.c or a shared code
1679 cpu_synchronize_all_post_init();
1681 qemu_announce_self();
1683 /* Make sure all file formats flush their mutable metadata.
1684 * If we get an error here, just don't restart the VM yet. */
1685 bdrv_invalidate_cache_all(&local_err
);
1687 error_report_err(local_err
);
1692 trace_loadvm_postcopy_handle_run_cpu_sync();
1693 cpu_synchronize_all_post_init();
1695 trace_loadvm_postcopy_handle_run_vmstart();
1697 dirty_bitmap_mig_before_vm_start();
1700 /* Hold onto your hats, starting the CPU */
1703 /* leave it paused and let management decide when to start the CPU */
1704 runstate_set(RUN_STATE_PAUSED
);
1707 qemu_bh_delete(data
->bh
);
1711 /* After all discards we can start running and asking for pages */
1712 static int loadvm_postcopy_handle_run(MigrationIncomingState
*mis
)
1714 PostcopyState ps
= postcopy_state_set(POSTCOPY_INCOMING_RUNNING
);
1715 HandleRunBhData
*data
;
1717 trace_loadvm_postcopy_handle_run();
1718 if (ps
!= POSTCOPY_INCOMING_LISTENING
) {
1719 error_report("CMD_POSTCOPY_RUN in wrong postcopy state (%d)", ps
);
1723 data
= g_new(HandleRunBhData
, 1);
1724 data
->bh
= qemu_bh_new(loadvm_postcopy_handle_run_bh
, data
);
1725 qemu_bh_schedule(data
->bh
);
1727 /* We need to finish reading the stream from the package
1728 * and also stop reading anything more from the stream that loaded the
1729 * package (since it's now being read by the listener thread).
1730 * LOADVM_QUIT will quit all the layers of nested loadvm loops.
1736 * Immediately following this command is a blob of data containing an embedded
1737 * chunk of migration stream; read it and load it.
1739 * @mis: Incoming state
1740 * @length: Length of packaged data to read
1742 * Returns: Negative values on error
1745 static int loadvm_handle_cmd_packaged(MigrationIncomingState
*mis
)
1749 QIOChannelBuffer
*bioc
;
1751 length
= qemu_get_be32(mis
->from_src_file
);
1752 trace_loadvm_handle_cmd_packaged(length
);
1754 if (length
> MAX_VM_CMD_PACKAGED_SIZE
) {
1755 error_report("Unreasonably large packaged state: %zu", length
);
1759 bioc
= qio_channel_buffer_new(length
);
1760 qio_channel_set_name(QIO_CHANNEL(bioc
), "migration-loadvm-buffer");
1761 ret
= qemu_get_buffer(mis
->from_src_file
,
1764 if (ret
!= length
) {
1765 object_unref(OBJECT(bioc
));
1766 error_report("CMD_PACKAGED: Buffer receive fail ret=%d length=%zu",
1768 return (ret
< 0) ? ret
: -EAGAIN
;
1770 bioc
->usage
+= length
;
1771 trace_loadvm_handle_cmd_packaged_received(ret
);
1773 QEMUFile
*packf
= qemu_fopen_channel_input(QIO_CHANNEL(bioc
));
1775 ret
= qemu_loadvm_state_main(packf
, mis
);
1776 trace_loadvm_handle_cmd_packaged_main(ret
);
1778 object_unref(OBJECT(bioc
));
1784 * Process an incoming 'QEMU_VM_COMMAND'
1785 * 0 just a normal return
1786 * LOADVM_QUIT All good, but exit the loop
1789 static int loadvm_process_command(QEMUFile
*f
)
1791 MigrationIncomingState
*mis
= migration_incoming_get_current();
1796 cmd
= qemu_get_be16(f
);
1797 len
= qemu_get_be16(f
);
1799 /* Check validity before continue processing of cmds */
1800 if (qemu_file_get_error(f
)) {
1801 return qemu_file_get_error(f
);
1804 trace_loadvm_process_command(cmd
, len
);
1805 if (cmd
>= MIG_CMD_MAX
|| cmd
== MIG_CMD_INVALID
) {
1806 error_report("MIG_CMD 0x%x unknown (len 0x%x)", cmd
, len
);
1810 if (mig_cmd_args
[cmd
].len
!= -1 && mig_cmd_args
[cmd
].len
!= len
) {
1811 error_report("%s received with bad length - expecting %zu, got %d",
1812 mig_cmd_args
[cmd
].name
,
1813 (size_t)mig_cmd_args
[cmd
].len
, len
);
1818 case MIG_CMD_OPEN_RETURN_PATH
:
1819 if (mis
->to_src_file
) {
1820 error_report("CMD_OPEN_RETURN_PATH called when RP already open");
1821 /* Not really a problem, so don't give up */
1824 mis
->to_src_file
= qemu_file_get_return_path(f
);
1825 if (!mis
->to_src_file
) {
1826 error_report("CMD_OPEN_RETURN_PATH failed");
1832 tmp32
= qemu_get_be32(f
);
1833 trace_loadvm_process_command_ping(tmp32
);
1834 if (!mis
->to_src_file
) {
1835 error_report("CMD_PING (0x%x) received with no return path",
1839 migrate_send_rp_pong(mis
, tmp32
);
1842 case MIG_CMD_PACKAGED
:
1843 return loadvm_handle_cmd_packaged(mis
);
1845 case MIG_CMD_POSTCOPY_ADVISE
:
1846 return loadvm_postcopy_handle_advise(mis
, len
);
1848 case MIG_CMD_POSTCOPY_LISTEN
:
1849 return loadvm_postcopy_handle_listen(mis
);
1851 case MIG_CMD_POSTCOPY_RUN
:
1852 return loadvm_postcopy_handle_run(mis
);
1854 case MIG_CMD_POSTCOPY_RAM_DISCARD
:
1855 return loadvm_postcopy_ram_handle_discard(mis
, len
);
1862 * Read a footer off the wire and check that it matches the expected section
1864 * Returns: true if the footer was good
1865 * false if there is a problem (and calls error_report to say why)
1867 static bool check_section_footer(QEMUFile
*f
, SaveStateEntry
*se
)
1871 uint32_t read_section_id
;
1873 if (!migrate_get_current()->send_section_footer
) {
1874 /* No footer to check */
1878 read_mark
= qemu_get_byte(f
);
1880 ret
= qemu_file_get_error(f
);
1882 error_report("%s: Read section footer failed: %d",
1887 if (read_mark
!= QEMU_VM_SECTION_FOOTER
) {
1888 error_report("Missing section footer for %s", se
->idstr
);
1892 read_section_id
= qemu_get_be32(f
);
1893 if (read_section_id
!= se
->load_section_id
) {
1894 error_report("Mismatched section id in footer for %s -"
1895 " read 0x%x expected 0x%x",
1896 se
->idstr
, read_section_id
, se
->load_section_id
);
1905 qemu_loadvm_section_start_full(QEMUFile
*f
, MigrationIncomingState
*mis
)
1907 uint32_t instance_id
, version_id
, section_id
;
1912 /* Read section start */
1913 section_id
= qemu_get_be32(f
);
1914 if (!qemu_get_counted_string(f
, idstr
)) {
1915 error_report("Unable to read ID string for section %u",
1919 instance_id
= qemu_get_be32(f
);
1920 version_id
= qemu_get_be32(f
);
1922 ret
= qemu_file_get_error(f
);
1924 error_report("%s: Failed to read instance/version ID: %d",
1929 trace_qemu_loadvm_state_section_startfull(section_id
, idstr
,
1930 instance_id
, version_id
);
1931 /* Find savevm section */
1932 se
= find_se(idstr
, instance_id
);
1934 error_report("Unknown savevm section or instance '%s' %d",
1935 idstr
, instance_id
);
1939 /* Validate version */
1940 if (version_id
> se
->version_id
) {
1941 error_report("savevm: unsupported version %d for '%s' v%d",
1942 version_id
, idstr
, se
->version_id
);
1945 se
->load_version_id
= version_id
;
1946 se
->load_section_id
= section_id
;
1948 /* Validate if it is a device's state */
1949 if (xen_enabled() && se
->is_ram
) {
1950 error_report("loadvm: %s RAM loading not allowed on Xen", idstr
);
1954 ret
= vmstate_load(f
, se
);
1956 error_report("error while loading state for instance 0x%x of"
1957 " device '%s'", instance_id
, idstr
);
1960 if (!check_section_footer(f
, se
)) {
1968 qemu_loadvm_section_part_end(QEMUFile
*f
, MigrationIncomingState
*mis
)
1970 uint32_t section_id
;
1974 section_id
= qemu_get_be32(f
);
1976 ret
= qemu_file_get_error(f
);
1978 error_report("%s: Failed to read section ID: %d",
1983 trace_qemu_loadvm_state_section_partend(section_id
);
1984 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
1985 if (se
->load_section_id
== section_id
) {
1990 error_report("Unknown savevm section %d", section_id
);
1994 ret
= vmstate_load(f
, se
);
1996 error_report("error while loading state section id %d(%s)",
1997 section_id
, se
->idstr
);
2000 if (!check_section_footer(f
, se
)) {
2007 static int qemu_loadvm_state_setup(QEMUFile
*f
)
2012 trace_loadvm_state_setup();
2013 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
2014 if (!se
->ops
|| !se
->ops
->load_setup
) {
2017 if (se
->ops
&& se
->ops
->is_active
) {
2018 if (!se
->ops
->is_active(se
->opaque
)) {
2023 ret
= se
->ops
->load_setup(f
, se
->opaque
);
2025 qemu_file_set_error(f
, ret
);
2026 error_report("Load state of device %s failed", se
->idstr
);
2033 void qemu_loadvm_state_cleanup(void)
2037 trace_loadvm_state_cleanup();
2038 QTAILQ_FOREACH(se
, &savevm_state
.handlers
, entry
) {
2039 if (se
->ops
&& se
->ops
->load_cleanup
) {
2040 se
->ops
->load_cleanup(se
->opaque
);
2045 static int qemu_loadvm_state_main(QEMUFile
*f
, MigrationIncomingState
*mis
)
2047 uint8_t section_type
;
2051 section_type
= qemu_get_byte(f
);
2053 if (qemu_file_get_error(f
)) {
2054 ret
= qemu_file_get_error(f
);
2058 trace_qemu_loadvm_state_section(section_type
);
2059 switch (section_type
) {
2060 case QEMU_VM_SECTION_START
:
2061 case QEMU_VM_SECTION_FULL
:
2062 ret
= qemu_loadvm_section_start_full(f
, mis
);
2067 case QEMU_VM_SECTION_PART
:
2068 case QEMU_VM_SECTION_END
:
2069 ret
= qemu_loadvm_section_part_end(f
, mis
);
2074 case QEMU_VM_COMMAND
:
2075 ret
= loadvm_process_command(f
);
2076 trace_qemu_loadvm_state_section_command(ret
);
2077 if ((ret
< 0) || (ret
& LOADVM_QUIT
)) {
2082 /* This is the end of migration */
2085 error_report("Unknown savevm section type %d", section_type
);
2093 qemu_file_set_error(f
, ret
);
2098 int qemu_loadvm_state(QEMUFile
*f
)
2100 MigrationIncomingState
*mis
= migration_incoming_get_current();
2101 Error
*local_err
= NULL
;
2105 if (qemu_savevm_state_blocked(&local_err
)) {
2106 error_report_err(local_err
);
2110 v
= qemu_get_be32(f
);
2111 if (v
!= QEMU_VM_FILE_MAGIC
) {
2112 error_report("Not a migration stream");
2116 v
= qemu_get_be32(f
);
2117 if (v
== QEMU_VM_FILE_VERSION_COMPAT
) {
2118 error_report("SaveVM v2 format is obsolete and don't work anymore");
2121 if (v
!= QEMU_VM_FILE_VERSION
) {
2122 error_report("Unsupported migration stream version");
2126 if (qemu_loadvm_state_setup(f
) != 0) {
2130 if (migrate_get_current()->send_configuration
) {
2131 if (qemu_get_byte(f
) != QEMU_VM_CONFIGURATION
) {
2132 error_report("Configuration section missing");
2135 ret
= vmstate_load_state(f
, &vmstate_configuration
, &savevm_state
, 0);
2142 cpu_synchronize_all_pre_loadvm();
2144 ret
= qemu_loadvm_state_main(f
, mis
);
2145 qemu_event_set(&mis
->main_thread_load_event
);
2147 trace_qemu_loadvm_state_post_main(ret
);
2149 if (mis
->have_listen_thread
) {
2150 /* Listen thread still going, can't clean up yet */
2155 ret
= qemu_file_get_error(f
);
2159 * Try to read in the VMDESC section as well, so that dumping tools that
2160 * intercept our migration stream have the chance to see it.
2163 /* We've got to be careful; if we don't read the data and just shut the fd
2164 * then the sender can error if we close while it's still sending.
2165 * We also mustn't read data that isn't there; some transports (RDMA)
2166 * will stall waiting for that data when the source has already closed.
2168 if (ret
== 0 && should_send_vmdesc()) {
2171 uint8_t section_type
= qemu_get_byte(f
);
2173 if (section_type
!= QEMU_VM_VMDESCRIPTION
) {
2174 error_report("Expected vmdescription section, but got %d",
2177 * It doesn't seem worth failing at this point since
2178 * we apparently have an otherwise valid VM state
2181 buf
= g_malloc(0x1000);
2182 size
= qemu_get_be32(f
);
2185 uint32_t read_chunk
= MIN(size
, 0x1000);
2186 qemu_get_buffer(f
, buf
, read_chunk
);
2193 qemu_loadvm_state_cleanup();
2194 cpu_synchronize_all_post_init();
2199 int save_snapshot(const char *name
, Error
**errp
)
2201 BlockDriverState
*bs
, *bs1
;
2202 QEMUSnapshotInfo sn1
, *sn
= &sn1
, old_sn1
, *old_sn
= &old_sn1
;
2205 int saved_vm_running
;
2206 uint64_t vm_state_size
;
2209 AioContext
*aio_context
;
2211 if (!replay_can_snapshot()) {
2212 error_report("Record/replay does not allow making snapshot "
2213 "right now. Try once more later.");
2217 if (!bdrv_all_can_snapshot(&bs
)) {
2218 error_setg(errp
, "Device '%s' is writable but does not support "
2219 "snapshots", bdrv_get_device_name(bs
));
2223 /* Delete old snapshots of the same name */
2225 ret
= bdrv_all_delete_snapshot(name
, &bs1
, errp
);
2227 error_prepend(errp
, "Error while deleting snapshot on device "
2228 "'%s': ", bdrv_get_device_name(bs1
));
2233 bs
= bdrv_all_find_vmstate_bs();
2235 error_setg(errp
, "No block device can accept snapshots");
2238 aio_context
= bdrv_get_aio_context(bs
);
2240 saved_vm_running
= runstate_is_running();
2242 ret
= global_state_store();
2244 error_setg(errp
, "Error saving global state");
2247 vm_stop(RUN_STATE_SAVE_VM
);
2249 bdrv_drain_all_begin();
2251 aio_context_acquire(aio_context
);
2253 memset(sn
, 0, sizeof(*sn
));
2255 /* fill auxiliary fields */
2256 qemu_gettimeofday(&tv
);
2257 sn
->date_sec
= tv
.tv_sec
;
2258 sn
->date_nsec
= tv
.tv_usec
* 1000;
2259 sn
->vm_clock_nsec
= qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL
);
2262 ret
= bdrv_snapshot_find(bs
, old_sn
, name
);
2264 pstrcpy(sn
->name
, sizeof(sn
->name
), old_sn
->name
);
2265 pstrcpy(sn
->id_str
, sizeof(sn
->id_str
), old_sn
->id_str
);
2267 pstrcpy(sn
->name
, sizeof(sn
->name
), name
);
2270 /* cast below needed for OpenBSD where tv_sec is still 'long' */
2271 localtime_r((const time_t *)&tv
.tv_sec
, &tm
);
2272 strftime(sn
->name
, sizeof(sn
->name
), "vm-%Y%m%d%H%M%S", &tm
);
2275 /* save the VM state */
2276 f
= qemu_fopen_bdrv(bs
, 1);
2278 error_setg(errp
, "Could not open VM state file");
2281 ret
= qemu_savevm_state(f
, errp
);
2282 vm_state_size
= qemu_ftell(f
);
2288 /* The bdrv_all_create_snapshot() call that follows acquires the AioContext
2289 * for itself. BDRV_POLL_WHILE() does not support nested locking because
2290 * it only releases the lock once. Therefore synchronous I/O will deadlock
2291 * unless we release the AioContext before bdrv_all_create_snapshot().
2293 aio_context_release(aio_context
);
2296 ret
= bdrv_all_create_snapshot(sn
, bs
, vm_state_size
, &bs
);
2298 error_setg(errp
, "Error while creating snapshot on '%s'",
2299 bdrv_get_device_name(bs
));
2307 aio_context_release(aio_context
);
2310 bdrv_drain_all_end();
2312 if (saved_vm_running
) {
2318 void qmp_xen_save_devices_state(const char *filename
, bool has_live
, bool live
,
2322 QIOChannelFile
*ioc
;
2323 int saved_vm_running
;
2327 /* live default to true so old version of Xen tool stack can have a
2328 * successfull live migration */
2332 saved_vm_running
= runstate_is_running();
2333 vm_stop(RUN_STATE_SAVE_VM
);
2334 global_state_store_running();
2336 ioc
= qio_channel_file_new_path(filename
, O_WRONLY
| O_CREAT
, 0660, errp
);
2340 qio_channel_set_name(QIO_CHANNEL(ioc
), "migration-xen-save-state");
2341 f
= qemu_fopen_channel_output(QIO_CHANNEL(ioc
));
2342 object_unref(OBJECT(ioc
));
2343 ret
= qemu_save_device_state(f
);
2344 if (ret
< 0 || qemu_fclose(f
) < 0) {
2345 error_setg(errp
, QERR_IO_ERROR
);
2347 /* libxl calls the QMP command "stop" before calling
2348 * "xen-save-devices-state" and in case of migration failure, libxl
2349 * would call "cont".
2350 * So call bdrv_inactivate_all (release locks) here to let the other
2351 * side of the migration take controle of the images.
2353 if (live
&& !saved_vm_running
) {
2354 ret
= bdrv_inactivate_all();
2356 error_setg(errp
, "%s: bdrv_inactivate_all() failed (%d)",
2363 if (saved_vm_running
) {
2368 void qmp_xen_load_devices_state(const char *filename
, Error
**errp
)
2371 QIOChannelFile
*ioc
;
2374 /* Guest must be paused before loading the device state; the RAM state
2375 * will already have been loaded by xc
2377 if (runstate_is_running()) {
2378 error_setg(errp
, "Cannot update device state while vm is running");
2381 vm_stop(RUN_STATE_RESTORE_VM
);
2383 ioc
= qio_channel_file_new_path(filename
, O_RDONLY
| O_BINARY
, 0, errp
);
2387 qio_channel_set_name(QIO_CHANNEL(ioc
), "migration-xen-load-state");
2388 f
= qemu_fopen_channel_input(QIO_CHANNEL(ioc
));
2389 object_unref(OBJECT(ioc
));
2391 ret
= qemu_loadvm_state(f
);
2394 error_setg(errp
, QERR_IO_ERROR
);
2396 migration_incoming_state_destroy();
2399 int load_snapshot(const char *name
, Error
**errp
)
2401 BlockDriverState
*bs
, *bs_vm_state
;
2402 QEMUSnapshotInfo sn
;
2405 AioContext
*aio_context
;
2406 MigrationIncomingState
*mis
= migration_incoming_get_current();
2408 if (!replay_can_snapshot()) {
2409 error_report("Record/replay does not allow loading snapshot "
2410 "right now. Try once more later.");
2414 if (!bdrv_all_can_snapshot(&bs
)) {
2416 "Device '%s' is writable but does not support snapshots",
2417 bdrv_get_device_name(bs
));
2420 ret
= bdrv_all_find_snapshot(name
, &bs
);
2423 "Device '%s' does not have the requested snapshot '%s'",
2424 bdrv_get_device_name(bs
), name
);
2428 bs_vm_state
= bdrv_all_find_vmstate_bs();
2430 error_setg(errp
, "No block device supports snapshots");
2433 aio_context
= bdrv_get_aio_context(bs_vm_state
);
2435 /* Don't even try to load empty VM states */
2436 aio_context_acquire(aio_context
);
2437 ret
= bdrv_snapshot_find(bs_vm_state
, &sn
, name
);
2438 aio_context_release(aio_context
);
2441 } else if (sn
.vm_state_size
== 0) {
2442 error_setg(errp
, "This is a disk-only snapshot. Revert to it "
2443 " offline using qemu-img");
2447 /* Flush all IO requests so they don't interfere with the new state. */
2448 bdrv_drain_all_begin();
2450 ret
= bdrv_all_goto_snapshot(name
, &bs
, errp
);
2452 error_prepend(errp
, "Could not load snapshot '%s' on '%s': ",
2453 name
, bdrv_get_device_name(bs
));
2457 /* restore the VM state */
2458 f
= qemu_fopen_bdrv(bs_vm_state
, 0);
2460 error_setg(errp
, "Could not open VM state file");
2465 qemu_system_reset(SHUTDOWN_CAUSE_NONE
);
2466 mis
->from_src_file
= f
;
2468 aio_context_acquire(aio_context
);
2469 ret
= qemu_loadvm_state(f
);
2470 migration_incoming_state_destroy();
2471 aio_context_release(aio_context
);
2473 bdrv_drain_all_end();
2476 error_setg(errp
, "Error %d while loading VM state", ret
);
2483 bdrv_drain_all_end();
2487 void vmstate_register_ram(MemoryRegion
*mr
, DeviceState
*dev
)
2489 qemu_ram_set_idstr(mr
->ram_block
,
2490 memory_region_name(mr
), dev
);
2493 void vmstate_unregister_ram(MemoryRegion
*mr
, DeviceState
*dev
)
2495 qemu_ram_unset_idstr(mr
->ram_block
);
2498 void vmstate_register_ram_global(MemoryRegion
*mr
)
2500 vmstate_register_ram(mr
, NULL
);
2503 bool vmstate_check_only_migratable(const VMStateDescription
*vmsd
)
2505 /* check needed if --only-migratable is specified */
2506 if (!migrate_get_current()->only_migratable
) {
2510 return !(vmsd
&& vmsd
->unmigratable
);