4 * Copyright Fujitsu, Corp. 2011, 2012
7 * Wen Congyang <wency@cn.fujitsu.com>
9 * This work is licensed under the terms of the GNU GPL, version 2 or later.
10 * See the COPYING file in the top-level directory.
14 #include "qemu/osdep.h"
15 #include "qemu/cutils.h"
17 #include "exec/hwaddr.h"
18 #include "monitor/monitor.h"
19 #include "sysemu/kvm.h"
20 #include "sysemu/dump.h"
21 #include "sysemu/memory_mapping.h"
22 #include "sysemu/runstate.h"
23 #include "sysemu/cpus.h"
24 #include "qapi/error.h"
25 #include "qapi/qapi-commands-dump.h"
26 #include "qapi/qapi-events-dump.h"
27 #include "qapi/qmp/qerror.h"
28 #include "qemu/error-report.h"
29 #include "qemu/main-loop.h"
30 #include "hw/misc/vmcoreinfo.h"
31 #include "migration/blocker.h"
39 #include <lzo/lzo1x.h>
44 #ifndef ELF_MACHINE_UNAME
45 #define ELF_MACHINE_UNAME "Unknown"
48 #define MAX_GUEST_NOTE_SIZE (1 << 20) /* 1MB should be enough */
50 static Error
*dump_migration_blocker
;
52 #define ELF_NOTE_SIZE(hdr_size, name_size, desc_size) \
53 ((DIV_ROUND_UP((hdr_size), 4) + \
54 DIV_ROUND_UP((name_size), 4) + \
55 DIV_ROUND_UP((desc_size), 4)) * 4)
57 uint16_t cpu_to_dump16(DumpState
*s
, uint16_t val
)
59 if (s
->dump_info
.d_endian
== ELFDATA2LSB
) {
60 val
= cpu_to_le16(val
);
62 val
= cpu_to_be16(val
);
68 uint32_t cpu_to_dump32(DumpState
*s
, uint32_t val
)
70 if (s
->dump_info
.d_endian
== ELFDATA2LSB
) {
71 val
= cpu_to_le32(val
);
73 val
= cpu_to_be32(val
);
79 uint64_t cpu_to_dump64(DumpState
*s
, uint64_t val
)
81 if (s
->dump_info
.d_endian
== ELFDATA2LSB
) {
82 val
= cpu_to_le64(val
);
84 val
= cpu_to_be64(val
);
90 static int dump_cleanup(DumpState
*s
)
92 guest_phys_blocks_free(&s
->guest_phys_blocks
);
93 memory_mapping_list_free(&s
->list
);
95 g_free(s
->guest_note
);
99 qemu_mutex_lock_iothread();
103 qemu_mutex_unlock_iothread();
106 migrate_del_blocker(dump_migration_blocker
);
111 static int fd_write_vmcore(const void *buf
, size_t size
, void *opaque
)
113 DumpState
*s
= opaque
;
116 written_size
= qemu_write_full(s
->fd
, buf
, size
);
117 if (written_size
!= size
) {
124 static void write_elf64_header(DumpState
*s
, Error
**errp
)
126 Elf64_Ehdr elf_header
;
129 memset(&elf_header
, 0, sizeof(Elf64_Ehdr
));
130 memcpy(&elf_header
, ELFMAG
, SELFMAG
);
131 elf_header
.e_ident
[EI_CLASS
] = ELFCLASS64
;
132 elf_header
.e_ident
[EI_DATA
] = s
->dump_info
.d_endian
;
133 elf_header
.e_ident
[EI_VERSION
] = EV_CURRENT
;
134 elf_header
.e_type
= cpu_to_dump16(s
, ET_CORE
);
135 elf_header
.e_machine
= cpu_to_dump16(s
, s
->dump_info
.d_machine
);
136 elf_header
.e_version
= cpu_to_dump32(s
, EV_CURRENT
);
137 elf_header
.e_ehsize
= cpu_to_dump16(s
, sizeof(elf_header
));
138 elf_header
.e_phoff
= cpu_to_dump64(s
, sizeof(Elf64_Ehdr
));
139 elf_header
.e_phentsize
= cpu_to_dump16(s
, sizeof(Elf64_Phdr
));
140 elf_header
.e_phnum
= cpu_to_dump16(s
, s
->phdr_num
);
141 if (s
->have_section
) {
142 uint64_t shoff
= sizeof(Elf64_Ehdr
) + sizeof(Elf64_Phdr
) * s
->sh_info
;
144 elf_header
.e_shoff
= cpu_to_dump64(s
, shoff
);
145 elf_header
.e_shentsize
= cpu_to_dump16(s
, sizeof(Elf64_Shdr
));
146 elf_header
.e_shnum
= cpu_to_dump16(s
, 1);
149 ret
= fd_write_vmcore(&elf_header
, sizeof(elf_header
), s
);
151 error_setg_errno(errp
, -ret
, "dump: failed to write elf header");
155 static void write_elf32_header(DumpState
*s
, Error
**errp
)
157 Elf32_Ehdr elf_header
;
160 memset(&elf_header
, 0, sizeof(Elf32_Ehdr
));
161 memcpy(&elf_header
, ELFMAG
, SELFMAG
);
162 elf_header
.e_ident
[EI_CLASS
] = ELFCLASS32
;
163 elf_header
.e_ident
[EI_DATA
] = s
->dump_info
.d_endian
;
164 elf_header
.e_ident
[EI_VERSION
] = EV_CURRENT
;
165 elf_header
.e_type
= cpu_to_dump16(s
, ET_CORE
);
166 elf_header
.e_machine
= cpu_to_dump16(s
, s
->dump_info
.d_machine
);
167 elf_header
.e_version
= cpu_to_dump32(s
, EV_CURRENT
);
168 elf_header
.e_ehsize
= cpu_to_dump16(s
, sizeof(elf_header
));
169 elf_header
.e_phoff
= cpu_to_dump32(s
, sizeof(Elf32_Ehdr
));
170 elf_header
.e_phentsize
= cpu_to_dump16(s
, sizeof(Elf32_Phdr
));
171 elf_header
.e_phnum
= cpu_to_dump16(s
, s
->phdr_num
);
172 if (s
->have_section
) {
173 uint32_t shoff
= sizeof(Elf32_Ehdr
) + sizeof(Elf32_Phdr
) * s
->sh_info
;
175 elf_header
.e_shoff
= cpu_to_dump32(s
, shoff
);
176 elf_header
.e_shentsize
= cpu_to_dump16(s
, sizeof(Elf32_Shdr
));
177 elf_header
.e_shnum
= cpu_to_dump16(s
, 1);
180 ret
= fd_write_vmcore(&elf_header
, sizeof(elf_header
), s
);
182 error_setg_errno(errp
, -ret
, "dump: failed to write elf header");
186 static void write_elf64_load(DumpState
*s
, MemoryMapping
*memory_mapping
,
187 int phdr_index
, hwaddr offset
,
188 hwaddr filesz
, Error
**errp
)
193 memset(&phdr
, 0, sizeof(Elf64_Phdr
));
194 phdr
.p_type
= cpu_to_dump32(s
, PT_LOAD
);
195 phdr
.p_offset
= cpu_to_dump64(s
, offset
);
196 phdr
.p_paddr
= cpu_to_dump64(s
, memory_mapping
->phys_addr
);
197 phdr
.p_filesz
= cpu_to_dump64(s
, filesz
);
198 phdr
.p_memsz
= cpu_to_dump64(s
, memory_mapping
->length
);
199 phdr
.p_vaddr
= cpu_to_dump64(s
, memory_mapping
->virt_addr
) ?: phdr
.p_paddr
;
201 assert(memory_mapping
->length
>= filesz
);
203 ret
= fd_write_vmcore(&phdr
, sizeof(Elf64_Phdr
), s
);
205 error_setg_errno(errp
, -ret
,
206 "dump: failed to write program header table");
210 static void write_elf32_load(DumpState
*s
, MemoryMapping
*memory_mapping
,
211 int phdr_index
, hwaddr offset
,
212 hwaddr filesz
, Error
**errp
)
217 memset(&phdr
, 0, sizeof(Elf32_Phdr
));
218 phdr
.p_type
= cpu_to_dump32(s
, PT_LOAD
);
219 phdr
.p_offset
= cpu_to_dump32(s
, offset
);
220 phdr
.p_paddr
= cpu_to_dump32(s
, memory_mapping
->phys_addr
);
221 phdr
.p_filesz
= cpu_to_dump32(s
, filesz
);
222 phdr
.p_memsz
= cpu_to_dump32(s
, memory_mapping
->length
);
224 cpu_to_dump32(s
, memory_mapping
->virt_addr
) ?: phdr
.p_paddr
;
226 assert(memory_mapping
->length
>= filesz
);
228 ret
= fd_write_vmcore(&phdr
, sizeof(Elf32_Phdr
), s
);
230 error_setg_errno(errp
, -ret
,
231 "dump: failed to write program header table");
235 static void write_elf64_note(DumpState
*s
, Error
**errp
)
238 hwaddr begin
= s
->memory_offset
- s
->note_size
;
241 memset(&phdr
, 0, sizeof(Elf64_Phdr
));
242 phdr
.p_type
= cpu_to_dump32(s
, PT_NOTE
);
243 phdr
.p_offset
= cpu_to_dump64(s
, begin
);
245 phdr
.p_filesz
= cpu_to_dump64(s
, s
->note_size
);
246 phdr
.p_memsz
= cpu_to_dump64(s
, s
->note_size
);
249 ret
= fd_write_vmcore(&phdr
, sizeof(Elf64_Phdr
), s
);
251 error_setg_errno(errp
, -ret
,
252 "dump: failed to write program header table");
256 static inline int cpu_index(CPUState
*cpu
)
258 return cpu
->cpu_index
+ 1;
261 static void write_guest_note(WriteCoreDumpFunction f
, DumpState
*s
,
267 ret
= f(s
->guest_note
, s
->guest_note_size
, s
);
269 error_setg(errp
, "dump: failed to write guest note");
274 static void write_elf64_notes(WriteCoreDumpFunction f
, DumpState
*s
,
283 ret
= cpu_write_elf64_note(f
, cpu
, id
, s
);
285 error_setg(errp
, "dump: failed to write elf notes");
291 ret
= cpu_write_elf64_qemunote(f
, cpu
, s
);
293 error_setg(errp
, "dump: failed to write CPU status");
298 write_guest_note(f
, s
, errp
);
301 static void write_elf32_note(DumpState
*s
, Error
**errp
)
303 hwaddr begin
= s
->memory_offset
- s
->note_size
;
307 memset(&phdr
, 0, sizeof(Elf32_Phdr
));
308 phdr
.p_type
= cpu_to_dump32(s
, PT_NOTE
);
309 phdr
.p_offset
= cpu_to_dump32(s
, begin
);
311 phdr
.p_filesz
= cpu_to_dump32(s
, s
->note_size
);
312 phdr
.p_memsz
= cpu_to_dump32(s
, s
->note_size
);
315 ret
= fd_write_vmcore(&phdr
, sizeof(Elf32_Phdr
), s
);
317 error_setg_errno(errp
, -ret
,
318 "dump: failed to write program header table");
322 static void write_elf32_notes(WriteCoreDumpFunction f
, DumpState
*s
,
331 ret
= cpu_write_elf32_note(f
, cpu
, id
, s
);
333 error_setg(errp
, "dump: failed to write elf notes");
339 ret
= cpu_write_elf32_qemunote(f
, cpu
, s
);
341 error_setg(errp
, "dump: failed to write CPU status");
346 write_guest_note(f
, s
, errp
);
349 static void write_elf_section(DumpState
*s
, int type
, Error
**errp
)
358 shdr_size
= sizeof(Elf32_Shdr
);
359 memset(&shdr32
, 0, shdr_size
);
360 shdr32
.sh_info
= cpu_to_dump32(s
, s
->sh_info
);
363 shdr_size
= sizeof(Elf64_Shdr
);
364 memset(&shdr64
, 0, shdr_size
);
365 shdr64
.sh_info
= cpu_to_dump32(s
, s
->sh_info
);
369 ret
= fd_write_vmcore(shdr
, shdr_size
, s
);
371 error_setg_errno(errp
, -ret
,
372 "dump: failed to write section header table");
376 static void write_data(DumpState
*s
, void *buf
, int length
, Error
**errp
)
380 ret
= fd_write_vmcore(buf
, length
, s
);
382 error_setg_errno(errp
, -ret
, "dump: failed to save memory");
384 s
->written_size
+= length
;
388 /* write the memory to vmcore. 1 page per I/O. */
389 static void write_memory(DumpState
*s
, GuestPhysBlock
*block
, ram_addr_t start
,
390 int64_t size
, Error
**errp
)
393 Error
*local_err
= NULL
;
395 for (i
= 0; i
< size
/ s
->dump_info
.page_size
; i
++) {
396 write_data(s
, block
->host_addr
+ start
+ i
* s
->dump_info
.page_size
,
397 s
->dump_info
.page_size
, &local_err
);
399 error_propagate(errp
, local_err
);
404 if ((size
% s
->dump_info
.page_size
) != 0) {
405 write_data(s
, block
->host_addr
+ start
+ i
* s
->dump_info
.page_size
,
406 size
% s
->dump_info
.page_size
, &local_err
);
408 error_propagate(errp
, local_err
);
414 /* get the memory's offset and size in the vmcore */
415 static void get_offset_range(hwaddr phys_addr
,
416 ram_addr_t mapping_length
,
421 GuestPhysBlock
*block
;
422 hwaddr offset
= s
->memory_offset
;
423 int64_t size_in_block
, start
;
425 /* When the memory is not stored into vmcore, offset will be -1 */
430 if (phys_addr
< s
->begin
|| phys_addr
>= s
->begin
+ s
->length
) {
435 QTAILQ_FOREACH(block
, &s
->guest_phys_blocks
.head
, next
) {
437 if (block
->target_start
>= s
->begin
+ s
->length
||
438 block
->target_end
<= s
->begin
) {
439 /* This block is out of the range */
443 if (s
->begin
<= block
->target_start
) {
444 start
= block
->target_start
;
449 size_in_block
= block
->target_end
- start
;
450 if (s
->begin
+ s
->length
< block
->target_end
) {
451 size_in_block
-= block
->target_end
- (s
->begin
+ s
->length
);
454 start
= block
->target_start
;
455 size_in_block
= block
->target_end
- block
->target_start
;
458 if (phys_addr
>= start
&& phys_addr
< start
+ size_in_block
) {
459 *p_offset
= phys_addr
- start
+ offset
;
461 /* The offset range mapped from the vmcore file must not spill over
462 * the GuestPhysBlock, clamp it. The rest of the mapping will be
463 * zero-filled in memory at load time; see
464 * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
466 *p_filesz
= phys_addr
+ mapping_length
<= start
+ size_in_block
?
468 size_in_block
- (phys_addr
- start
);
472 offset
+= size_in_block
;
476 static void write_elf_loads(DumpState
*s
, Error
**errp
)
478 hwaddr offset
, filesz
;
479 MemoryMapping
*memory_mapping
;
480 uint32_t phdr_index
= 1;
482 Error
*local_err
= NULL
;
484 if (s
->have_section
) {
485 max_index
= s
->sh_info
;
487 max_index
= s
->phdr_num
;
490 QTAILQ_FOREACH(memory_mapping
, &s
->list
.head
, next
) {
491 get_offset_range(memory_mapping
->phys_addr
,
492 memory_mapping
->length
,
493 s
, &offset
, &filesz
);
494 if (s
->dump_info
.d_class
== ELFCLASS64
) {
495 write_elf64_load(s
, memory_mapping
, phdr_index
++, offset
,
498 write_elf32_load(s
, memory_mapping
, phdr_index
++, offset
,
503 error_propagate(errp
, local_err
);
507 if (phdr_index
>= max_index
) {
513 /* write elf header, PT_NOTE and elf note to vmcore. */
514 static void dump_begin(DumpState
*s
, Error
**errp
)
516 Error
*local_err
= NULL
;
519 * the vmcore's format is:
538 * we only know where the memory is saved after we write elf note into
542 /* write elf header to vmcore */
543 if (s
->dump_info
.d_class
== ELFCLASS64
) {
544 write_elf64_header(s
, &local_err
);
546 write_elf32_header(s
, &local_err
);
549 error_propagate(errp
, local_err
);
553 if (s
->dump_info
.d_class
== ELFCLASS64
) {
554 /* write PT_NOTE to vmcore */
555 write_elf64_note(s
, &local_err
);
557 error_propagate(errp
, local_err
);
561 /* write all PT_LOAD to vmcore */
562 write_elf_loads(s
, &local_err
);
564 error_propagate(errp
, local_err
);
568 /* write section to vmcore */
569 if (s
->have_section
) {
570 write_elf_section(s
, 1, &local_err
);
572 error_propagate(errp
, local_err
);
577 /* write notes to vmcore */
578 write_elf64_notes(fd_write_vmcore
, s
, &local_err
);
580 error_propagate(errp
, local_err
);
584 /* write PT_NOTE to vmcore */
585 write_elf32_note(s
, &local_err
);
587 error_propagate(errp
, local_err
);
591 /* write all PT_LOAD to vmcore */
592 write_elf_loads(s
, &local_err
);
594 error_propagate(errp
, local_err
);
598 /* write section to vmcore */
599 if (s
->have_section
) {
600 write_elf_section(s
, 0, &local_err
);
602 error_propagate(errp
, local_err
);
607 /* write notes to vmcore */
608 write_elf32_notes(fd_write_vmcore
, s
, &local_err
);
610 error_propagate(errp
, local_err
);
616 static int get_next_block(DumpState
*s
, GuestPhysBlock
*block
)
619 block
= QTAILQ_NEXT(block
, next
);
626 s
->next_block
= block
;
628 if (block
->target_start
>= s
->begin
+ s
->length
||
629 block
->target_end
<= s
->begin
) {
630 /* This block is out of the range */
634 if (s
->begin
> block
->target_start
) {
635 s
->start
= s
->begin
- block
->target_start
;
643 /* write all memory to vmcore */
644 static void dump_iterate(DumpState
*s
, Error
**errp
)
646 GuestPhysBlock
*block
;
648 Error
*local_err
= NULL
;
651 block
= s
->next_block
;
653 size
= block
->target_end
- block
->target_start
;
656 if (s
->begin
+ s
->length
< block
->target_end
) {
657 size
-= block
->target_end
- (s
->begin
+ s
->length
);
660 write_memory(s
, block
, s
->start
, size
, &local_err
);
662 error_propagate(errp
, local_err
);
666 } while (!get_next_block(s
, block
));
669 static void create_vmcore(DumpState
*s
, Error
**errp
)
671 Error
*local_err
= NULL
;
673 dump_begin(s
, &local_err
);
675 error_propagate(errp
, local_err
);
679 dump_iterate(s
, errp
);
682 static int write_start_flat_header(int fd
)
684 MakedumpfileHeader
*mh
;
687 QEMU_BUILD_BUG_ON(sizeof *mh
> MAX_SIZE_MDF_HEADER
);
688 mh
= g_malloc0(MAX_SIZE_MDF_HEADER
);
690 memcpy(mh
->signature
, MAKEDUMPFILE_SIGNATURE
,
691 MIN(sizeof mh
->signature
, sizeof MAKEDUMPFILE_SIGNATURE
));
693 mh
->type
= cpu_to_be64(TYPE_FLAT_HEADER
);
694 mh
->version
= cpu_to_be64(VERSION_FLAT_HEADER
);
697 written_size
= qemu_write_full(fd
, mh
, MAX_SIZE_MDF_HEADER
);
698 if (written_size
!= MAX_SIZE_MDF_HEADER
) {
706 static int write_end_flat_header(int fd
)
708 MakedumpfileDataHeader mdh
;
710 mdh
.offset
= END_FLAG_FLAT_HEADER
;
711 mdh
.buf_size
= END_FLAG_FLAT_HEADER
;
714 written_size
= qemu_write_full(fd
, &mdh
, sizeof(mdh
));
715 if (written_size
!= sizeof(mdh
)) {
722 static int write_buffer(int fd
, off_t offset
, const void *buf
, size_t size
)
725 MakedumpfileDataHeader mdh
;
727 mdh
.offset
= cpu_to_be64(offset
);
728 mdh
.buf_size
= cpu_to_be64(size
);
730 written_size
= qemu_write_full(fd
, &mdh
, sizeof(mdh
));
731 if (written_size
!= sizeof(mdh
)) {
735 written_size
= qemu_write_full(fd
, buf
, size
);
736 if (written_size
!= size
) {
743 static int buf_write_note(const void *buf
, size_t size
, void *opaque
)
745 DumpState
*s
= opaque
;
747 /* note_buf is not enough */
748 if (s
->note_buf_offset
+ size
> s
->note_size
) {
752 memcpy(s
->note_buf
+ s
->note_buf_offset
, buf
, size
);
754 s
->note_buf_offset
+= size
;
760 * This function retrieves various sizes from an elf header.
762 * @note has to be a valid ELF note. The return sizes are unmodified
763 * (not padded or rounded up to be multiple of 4).
765 static void get_note_sizes(DumpState
*s
, const void *note
,
766 uint64_t *note_head_size
,
770 uint64_t note_head_sz
;
774 if (s
->dump_info
.d_class
== ELFCLASS64
) {
775 const Elf64_Nhdr
*hdr
= note
;
776 note_head_sz
= sizeof(Elf64_Nhdr
);
777 name_sz
= tswap64(hdr
->n_namesz
);
778 desc_sz
= tswap64(hdr
->n_descsz
);
780 const Elf32_Nhdr
*hdr
= note
;
781 note_head_sz
= sizeof(Elf32_Nhdr
);
782 name_sz
= tswap32(hdr
->n_namesz
);
783 desc_sz
= tswap32(hdr
->n_descsz
);
786 if (note_head_size
) {
787 *note_head_size
= note_head_sz
;
790 *name_size
= name_sz
;
793 *desc_size
= desc_sz
;
797 static bool note_name_equal(DumpState
*s
,
798 const uint8_t *note
, const char *name
)
800 int len
= strlen(name
) + 1;
801 uint64_t head_size
, name_size
;
803 get_note_sizes(s
, note
, &head_size
, &name_size
, NULL
);
804 head_size
= ROUND_UP(head_size
, 4);
806 return name_size
== len
&& memcmp(note
+ head_size
, name
, len
) == 0;
809 /* write common header, sub header and elf note to vmcore */
810 static void create_header32(DumpState
*s
, Error
**errp
)
812 DiskDumpHeader32
*dh
= NULL
;
813 KdumpSubHeader32
*kh
= NULL
;
816 uint32_t sub_hdr_size
;
817 uint32_t bitmap_blocks
;
819 uint64_t offset_note
;
820 Error
*local_err
= NULL
;
822 /* write common header, the version of kdump-compressed format is 6th */
823 size
= sizeof(DiskDumpHeader32
);
824 dh
= g_malloc0(size
);
826 memcpy(dh
->signature
, KDUMP_SIGNATURE
, SIG_LEN
);
827 dh
->header_version
= cpu_to_dump32(s
, 6);
828 block_size
= s
->dump_info
.page_size
;
829 dh
->block_size
= cpu_to_dump32(s
, block_size
);
830 sub_hdr_size
= sizeof(struct KdumpSubHeader32
) + s
->note_size
;
831 sub_hdr_size
= DIV_ROUND_UP(sub_hdr_size
, block_size
);
832 dh
->sub_hdr_size
= cpu_to_dump32(s
, sub_hdr_size
);
833 /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
834 dh
->max_mapnr
= cpu_to_dump32(s
, MIN(s
->max_mapnr
, UINT_MAX
));
835 dh
->nr_cpus
= cpu_to_dump32(s
, s
->nr_cpus
);
836 bitmap_blocks
= DIV_ROUND_UP(s
->len_dump_bitmap
, block_size
) * 2;
837 dh
->bitmap_blocks
= cpu_to_dump32(s
, bitmap_blocks
);
838 strncpy(dh
->utsname
.machine
, ELF_MACHINE_UNAME
, sizeof(dh
->utsname
.machine
));
840 if (s
->flag_compress
& DUMP_DH_COMPRESSED_ZLIB
) {
841 status
|= DUMP_DH_COMPRESSED_ZLIB
;
844 if (s
->flag_compress
& DUMP_DH_COMPRESSED_LZO
) {
845 status
|= DUMP_DH_COMPRESSED_LZO
;
849 if (s
->flag_compress
& DUMP_DH_COMPRESSED_SNAPPY
) {
850 status
|= DUMP_DH_COMPRESSED_SNAPPY
;
853 dh
->status
= cpu_to_dump32(s
, status
);
855 if (write_buffer(s
->fd
, 0, dh
, size
) < 0) {
856 error_setg(errp
, "dump: failed to write disk dump header");
860 /* write sub header */
861 size
= sizeof(KdumpSubHeader32
);
862 kh
= g_malloc0(size
);
864 /* 64bit max_mapnr_64 */
865 kh
->max_mapnr_64
= cpu_to_dump64(s
, s
->max_mapnr
);
866 kh
->phys_base
= cpu_to_dump32(s
, s
->dump_info
.phys_base
);
867 kh
->dump_level
= cpu_to_dump32(s
, DUMP_LEVEL
);
869 offset_note
= DISKDUMP_HEADER_BLOCKS
* block_size
+ size
;
871 note_name_equal(s
, s
->guest_note
, "VMCOREINFO")) {
872 uint64_t hsize
, name_size
, size_vmcoreinfo_desc
, offset_vmcoreinfo
;
874 get_note_sizes(s
, s
->guest_note
,
875 &hsize
, &name_size
, &size_vmcoreinfo_desc
);
876 offset_vmcoreinfo
= offset_note
+ s
->note_size
- s
->guest_note_size
+
877 (DIV_ROUND_UP(hsize
, 4) + DIV_ROUND_UP(name_size
, 4)) * 4;
878 kh
->offset_vmcoreinfo
= cpu_to_dump64(s
, offset_vmcoreinfo
);
879 kh
->size_vmcoreinfo
= cpu_to_dump32(s
, size_vmcoreinfo_desc
);
882 kh
->offset_note
= cpu_to_dump64(s
, offset_note
);
883 kh
->note_size
= cpu_to_dump32(s
, s
->note_size
);
885 if (write_buffer(s
->fd
, DISKDUMP_HEADER_BLOCKS
*
886 block_size
, kh
, size
) < 0) {
887 error_setg(errp
, "dump: failed to write kdump sub header");
892 s
->note_buf
= g_malloc0(s
->note_size
);
893 s
->note_buf_offset
= 0;
895 /* use s->note_buf to store notes temporarily */
896 write_elf32_notes(buf_write_note
, s
, &local_err
);
898 error_propagate(errp
, local_err
);
901 if (write_buffer(s
->fd
, offset_note
, s
->note_buf
,
903 error_setg(errp
, "dump: failed to write notes");
907 /* get offset of dump_bitmap */
908 s
->offset_dump_bitmap
= (DISKDUMP_HEADER_BLOCKS
+ sub_hdr_size
) *
911 /* get offset of page */
912 s
->offset_page
= (DISKDUMP_HEADER_BLOCKS
+ sub_hdr_size
+ bitmap_blocks
) *
921 /* write common header, sub header and elf note to vmcore */
922 static void create_header64(DumpState
*s
, Error
**errp
)
924 DiskDumpHeader64
*dh
= NULL
;
925 KdumpSubHeader64
*kh
= NULL
;
928 uint32_t sub_hdr_size
;
929 uint32_t bitmap_blocks
;
931 uint64_t offset_note
;
932 Error
*local_err
= NULL
;
934 /* write common header, the version of kdump-compressed format is 6th */
935 size
= sizeof(DiskDumpHeader64
);
936 dh
= g_malloc0(size
);
938 memcpy(dh
->signature
, KDUMP_SIGNATURE
, SIG_LEN
);
939 dh
->header_version
= cpu_to_dump32(s
, 6);
940 block_size
= s
->dump_info
.page_size
;
941 dh
->block_size
= cpu_to_dump32(s
, block_size
);
942 sub_hdr_size
= sizeof(struct KdumpSubHeader64
) + s
->note_size
;
943 sub_hdr_size
= DIV_ROUND_UP(sub_hdr_size
, block_size
);
944 dh
->sub_hdr_size
= cpu_to_dump32(s
, sub_hdr_size
);
945 /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
946 dh
->max_mapnr
= cpu_to_dump32(s
, MIN(s
->max_mapnr
, UINT_MAX
));
947 dh
->nr_cpus
= cpu_to_dump32(s
, s
->nr_cpus
);
948 bitmap_blocks
= DIV_ROUND_UP(s
->len_dump_bitmap
, block_size
) * 2;
949 dh
->bitmap_blocks
= cpu_to_dump32(s
, bitmap_blocks
);
950 strncpy(dh
->utsname
.machine
, ELF_MACHINE_UNAME
, sizeof(dh
->utsname
.machine
));
952 if (s
->flag_compress
& DUMP_DH_COMPRESSED_ZLIB
) {
953 status
|= DUMP_DH_COMPRESSED_ZLIB
;
956 if (s
->flag_compress
& DUMP_DH_COMPRESSED_LZO
) {
957 status
|= DUMP_DH_COMPRESSED_LZO
;
961 if (s
->flag_compress
& DUMP_DH_COMPRESSED_SNAPPY
) {
962 status
|= DUMP_DH_COMPRESSED_SNAPPY
;
965 dh
->status
= cpu_to_dump32(s
, status
);
967 if (write_buffer(s
->fd
, 0, dh
, size
) < 0) {
968 error_setg(errp
, "dump: failed to write disk dump header");
972 /* write sub header */
973 size
= sizeof(KdumpSubHeader64
);
974 kh
= g_malloc0(size
);
976 /* 64bit max_mapnr_64 */
977 kh
->max_mapnr_64
= cpu_to_dump64(s
, s
->max_mapnr
);
978 kh
->phys_base
= cpu_to_dump64(s
, s
->dump_info
.phys_base
);
979 kh
->dump_level
= cpu_to_dump32(s
, DUMP_LEVEL
);
981 offset_note
= DISKDUMP_HEADER_BLOCKS
* block_size
+ size
;
983 note_name_equal(s
, s
->guest_note
, "VMCOREINFO")) {
984 uint64_t hsize
, name_size
, size_vmcoreinfo_desc
, offset_vmcoreinfo
;
986 get_note_sizes(s
, s
->guest_note
,
987 &hsize
, &name_size
, &size_vmcoreinfo_desc
);
988 offset_vmcoreinfo
= offset_note
+ s
->note_size
- s
->guest_note_size
+
989 (DIV_ROUND_UP(hsize
, 4) + DIV_ROUND_UP(name_size
, 4)) * 4;
990 kh
->offset_vmcoreinfo
= cpu_to_dump64(s
, offset_vmcoreinfo
);
991 kh
->size_vmcoreinfo
= cpu_to_dump64(s
, size_vmcoreinfo_desc
);
994 kh
->offset_note
= cpu_to_dump64(s
, offset_note
);
995 kh
->note_size
= cpu_to_dump64(s
, s
->note_size
);
997 if (write_buffer(s
->fd
, DISKDUMP_HEADER_BLOCKS
*
998 block_size
, kh
, size
) < 0) {
999 error_setg(errp
, "dump: failed to write kdump sub header");
1004 s
->note_buf
= g_malloc0(s
->note_size
);
1005 s
->note_buf_offset
= 0;
1007 /* use s->note_buf to store notes temporarily */
1008 write_elf64_notes(buf_write_note
, s
, &local_err
);
1010 error_propagate(errp
, local_err
);
1014 if (write_buffer(s
->fd
, offset_note
, s
->note_buf
,
1015 s
->note_size
) < 0) {
1016 error_setg(errp
, "dump: failed to write notes");
1020 /* get offset of dump_bitmap */
1021 s
->offset_dump_bitmap
= (DISKDUMP_HEADER_BLOCKS
+ sub_hdr_size
) *
1024 /* get offset of page */
1025 s
->offset_page
= (DISKDUMP_HEADER_BLOCKS
+ sub_hdr_size
+ bitmap_blocks
) *
1031 g_free(s
->note_buf
);
1034 static void write_dump_header(DumpState
*s
, Error
**errp
)
1036 if (s
->dump_info
.d_class
== ELFCLASS32
) {
1037 create_header32(s
, errp
);
1039 create_header64(s
, errp
);
1043 static size_t dump_bitmap_get_bufsize(DumpState
*s
)
1045 return s
->dump_info
.page_size
;
1049 * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
1050 * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
1051 * set_dump_bitmap will always leave the recently set bit un-sync. And setting
1052 * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
1053 * vmcore, ie. synchronizing un-sync bit into vmcore.
1055 static int set_dump_bitmap(uint64_t last_pfn
, uint64_t pfn
, bool value
,
1056 uint8_t *buf
, DumpState
*s
)
1058 off_t old_offset
, new_offset
;
1059 off_t offset_bitmap1
, offset_bitmap2
;
1061 size_t bitmap_bufsize
= dump_bitmap_get_bufsize(s
);
1062 size_t bits_per_buf
= bitmap_bufsize
* CHAR_BIT
;
1064 /* should not set the previous place */
1065 assert(last_pfn
<= pfn
);
1068 * if the bit needed to be set is not cached in buf, flush the data in buf
1069 * to vmcore firstly.
1070 * making new_offset be bigger than old_offset can also sync remained data
1073 old_offset
= bitmap_bufsize
* (last_pfn
/ bits_per_buf
);
1074 new_offset
= bitmap_bufsize
* (pfn
/ bits_per_buf
);
1076 while (old_offset
< new_offset
) {
1077 /* calculate the offset and write dump_bitmap */
1078 offset_bitmap1
= s
->offset_dump_bitmap
+ old_offset
;
1079 if (write_buffer(s
->fd
, offset_bitmap1
, buf
,
1080 bitmap_bufsize
) < 0) {
1084 /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
1085 offset_bitmap2
= s
->offset_dump_bitmap
+ s
->len_dump_bitmap
+
1087 if (write_buffer(s
->fd
, offset_bitmap2
, buf
,
1088 bitmap_bufsize
) < 0) {
1092 memset(buf
, 0, bitmap_bufsize
);
1093 old_offset
+= bitmap_bufsize
;
1096 /* get the exact place of the bit in the buf, and set it */
1097 byte
= (pfn
% bits_per_buf
) / CHAR_BIT
;
1098 bit
= (pfn
% bits_per_buf
) % CHAR_BIT
;
1100 buf
[byte
] |= 1u << bit
;
1102 buf
[byte
] &= ~(1u << bit
);
1108 static uint64_t dump_paddr_to_pfn(DumpState
*s
, uint64_t addr
)
1110 int target_page_shift
= ctz32(s
->dump_info
.page_size
);
1112 return (addr
>> target_page_shift
) - ARCH_PFN_OFFSET
;
1115 static uint64_t dump_pfn_to_paddr(DumpState
*s
, uint64_t pfn
)
1117 int target_page_shift
= ctz32(s
->dump_info
.page_size
);
1119 return (pfn
+ ARCH_PFN_OFFSET
) << target_page_shift
;
1123 * exam every page and return the page frame number and the address of the page.
1124 * bufptr can be NULL. note: the blocks here is supposed to reflect guest-phys
1125 * blocks, so block->target_start and block->target_end should be interal
1126 * multiples of the target page size.
1128 static bool get_next_page(GuestPhysBlock
**blockptr
, uint64_t *pfnptr
,
1129 uint8_t **bufptr
, DumpState
*s
)
1131 GuestPhysBlock
*block
= *blockptr
;
1132 hwaddr addr
, target_page_mask
= ~((hwaddr
)s
->dump_info
.page_size
- 1);
1135 /* block == NULL means the start of the iteration */
1137 block
= QTAILQ_FIRST(&s
->guest_phys_blocks
.head
);
1139 assert((block
->target_start
& ~target_page_mask
) == 0);
1140 assert((block
->target_end
& ~target_page_mask
) == 0);
1141 *pfnptr
= dump_paddr_to_pfn(s
, block
->target_start
);
1143 *bufptr
= block
->host_addr
;
1148 *pfnptr
= *pfnptr
+ 1;
1149 addr
= dump_pfn_to_paddr(s
, *pfnptr
);
1151 if ((addr
>= block
->target_start
) &&
1152 (addr
+ s
->dump_info
.page_size
<= block
->target_end
)) {
1153 buf
= block
->host_addr
+ (addr
- block
->target_start
);
1155 /* the next page is in the next block */
1156 block
= QTAILQ_NEXT(block
, next
);
1161 assert((block
->target_start
& ~target_page_mask
) == 0);
1162 assert((block
->target_end
& ~target_page_mask
) == 0);
1163 *pfnptr
= dump_paddr_to_pfn(s
, block
->target_start
);
1164 buf
= block
->host_addr
;
1174 static void write_dump_bitmap(DumpState
*s
, Error
**errp
)
1177 uint64_t last_pfn
, pfn
;
1178 void *dump_bitmap_buf
;
1179 size_t num_dumpable
;
1180 GuestPhysBlock
*block_iter
= NULL
;
1181 size_t bitmap_bufsize
= dump_bitmap_get_bufsize(s
);
1182 size_t bits_per_buf
= bitmap_bufsize
* CHAR_BIT
;
1184 /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1185 dump_bitmap_buf
= g_malloc0(bitmap_bufsize
);
1191 * exam memory page by page, and set the bit in dump_bitmap corresponded
1192 * to the existing page.
1194 while (get_next_page(&block_iter
, &pfn
, NULL
, s
)) {
1195 ret
= set_dump_bitmap(last_pfn
, pfn
, true, dump_bitmap_buf
, s
);
1197 error_setg(errp
, "dump: failed to set dump_bitmap");
1206 * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1207 * set the remaining bits from last_pfn to the end of the bitmap buffer to
1208 * 0. With those set, the un-sync bit will be synchronized into the vmcore.
1210 if (num_dumpable
> 0) {
1211 ret
= set_dump_bitmap(last_pfn
, last_pfn
+ bits_per_buf
, false,
1212 dump_bitmap_buf
, s
);
1214 error_setg(errp
, "dump: failed to sync dump_bitmap");
1219 /* number of dumpable pages that will be dumped later */
1220 s
->num_dumpable
= num_dumpable
;
1223 g_free(dump_bitmap_buf
);
1226 static void prepare_data_cache(DataCache
*data_cache
, DumpState
*s
,
1229 data_cache
->fd
= s
->fd
;
1230 data_cache
->data_size
= 0;
1231 data_cache
->buf_size
= 4 * dump_bitmap_get_bufsize(s
);
1232 data_cache
->buf
= g_malloc0(data_cache
->buf_size
);
1233 data_cache
->offset
= offset
;
1236 static int write_cache(DataCache
*dc
, const void *buf
, size_t size
,
1240 * dc->buf_size should not be less than size, otherwise dc will never be
1243 assert(size
<= dc
->buf_size
);
1246 * if flag_sync is set, synchronize data in dc->buf into vmcore.
1247 * otherwise check if the space is enough for caching data in buf, if not,
1248 * write the data in dc->buf to dc->fd and reset dc->buf
1250 if ((!flag_sync
&& dc
->data_size
+ size
> dc
->buf_size
) ||
1251 (flag_sync
&& dc
->data_size
> 0)) {
1252 if (write_buffer(dc
->fd
, dc
->offset
, dc
->buf
, dc
->data_size
) < 0) {
1256 dc
->offset
+= dc
->data_size
;
1261 memcpy(dc
->buf
+ dc
->data_size
, buf
, size
);
1262 dc
->data_size
+= size
;
1268 static void free_data_cache(DataCache
*data_cache
)
1270 g_free(data_cache
->buf
);
1273 static size_t get_len_buf_out(size_t page_size
, uint32_t flag_compress
)
1275 switch (flag_compress
) {
1276 case DUMP_DH_COMPRESSED_ZLIB
:
1277 return compressBound(page_size
);
1279 case DUMP_DH_COMPRESSED_LZO
:
1281 * LZO will expand incompressible data by a little amount. Please check
1282 * the following URL to see the expansion calculation:
1283 * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1285 return page_size
+ page_size
/ 16 + 64 + 3;
1287 #ifdef CONFIG_SNAPPY
1288 case DUMP_DH_COMPRESSED_SNAPPY
:
1289 return snappy_max_compressed_length(page_size
);
1295 static void write_dump_pages(DumpState
*s
, Error
**errp
)
1298 DataCache page_desc
, page_data
;
1299 size_t len_buf_out
, size_out
;
1301 lzo_bytep wrkmem
= NULL
;
1303 uint8_t *buf_out
= NULL
;
1304 off_t offset_desc
, offset_data
;
1305 PageDescriptor pd
, pd_zero
;
1307 GuestPhysBlock
*block_iter
= NULL
;
1310 /* get offset of page_desc and page_data in dump file */
1311 offset_desc
= s
->offset_page
;
1312 offset_data
= offset_desc
+ sizeof(PageDescriptor
) * s
->num_dumpable
;
1314 prepare_data_cache(&page_desc
, s
, offset_desc
);
1315 prepare_data_cache(&page_data
, s
, offset_data
);
1317 /* prepare buffer to store compressed data */
1318 len_buf_out
= get_len_buf_out(s
->dump_info
.page_size
, s
->flag_compress
);
1319 assert(len_buf_out
!= 0);
1322 wrkmem
= g_malloc(LZO1X_1_MEM_COMPRESS
);
1325 buf_out
= g_malloc(len_buf_out
);
1328 * init zero page's page_desc and page_data, because every zero page
1329 * uses the same page_data
1331 pd_zero
.size
= cpu_to_dump32(s
, s
->dump_info
.page_size
);
1332 pd_zero
.flags
= cpu_to_dump32(s
, 0);
1333 pd_zero
.offset
= cpu_to_dump64(s
, offset_data
);
1334 pd_zero
.page_flags
= cpu_to_dump64(s
, 0);
1335 buf
= g_malloc0(s
->dump_info
.page_size
);
1336 ret
= write_cache(&page_data
, buf
, s
->dump_info
.page_size
, false);
1339 error_setg(errp
, "dump: failed to write page data (zero page)");
1343 offset_data
+= s
->dump_info
.page_size
;
1346 * dump memory to vmcore page by page. zero page will all be resided in the
1347 * first page of page section
1349 while (get_next_page(&block_iter
, &pfn_iter
, &buf
, s
)) {
1350 /* check zero page */
1351 if (buffer_is_zero(buf
, s
->dump_info
.page_size
)) {
1352 ret
= write_cache(&page_desc
, &pd_zero
, sizeof(PageDescriptor
),
1355 error_setg(errp
, "dump: failed to write page desc");
1360 * not zero page, then:
1361 * 1. compress the page
1362 * 2. write the compressed page into the cache of page_data
1363 * 3. get page desc of the compressed page and write it into the
1364 * cache of page_desc
1366 * only one compression format will be used here, for
1367 * s->flag_compress is set. But when compression fails to work,
1368 * we fall back to save in plaintext.
1370 size_out
= len_buf_out
;
1371 if ((s
->flag_compress
& DUMP_DH_COMPRESSED_ZLIB
) &&
1372 (compress2(buf_out
, (uLongf
*)&size_out
, buf
,
1373 s
->dump_info
.page_size
, Z_BEST_SPEED
) == Z_OK
) &&
1374 (size_out
< s
->dump_info
.page_size
)) {
1375 pd
.flags
= cpu_to_dump32(s
, DUMP_DH_COMPRESSED_ZLIB
);
1376 pd
.size
= cpu_to_dump32(s
, size_out
);
1378 ret
= write_cache(&page_data
, buf_out
, size_out
, false);
1380 error_setg(errp
, "dump: failed to write page data");
1384 } else if ((s
->flag_compress
& DUMP_DH_COMPRESSED_LZO
) &&
1385 (lzo1x_1_compress(buf
, s
->dump_info
.page_size
, buf_out
,
1386 (lzo_uint
*)&size_out
, wrkmem
) == LZO_E_OK
) &&
1387 (size_out
< s
->dump_info
.page_size
)) {
1388 pd
.flags
= cpu_to_dump32(s
, DUMP_DH_COMPRESSED_LZO
);
1389 pd
.size
= cpu_to_dump32(s
, size_out
);
1391 ret
= write_cache(&page_data
, buf_out
, size_out
, false);
1393 error_setg(errp
, "dump: failed to write page data");
1397 #ifdef CONFIG_SNAPPY
1398 } else if ((s
->flag_compress
& DUMP_DH_COMPRESSED_SNAPPY
) &&
1399 (snappy_compress((char *)buf
, s
->dump_info
.page_size
,
1400 (char *)buf_out
, &size_out
) == SNAPPY_OK
) &&
1401 (size_out
< s
->dump_info
.page_size
)) {
1402 pd
.flags
= cpu_to_dump32(s
, DUMP_DH_COMPRESSED_SNAPPY
);
1403 pd
.size
= cpu_to_dump32(s
, size_out
);
1405 ret
= write_cache(&page_data
, buf_out
, size_out
, false);
1407 error_setg(errp
, "dump: failed to write page data");
1413 * fall back to save in plaintext, size_out should be
1414 * assigned the target's page size
1416 pd
.flags
= cpu_to_dump32(s
, 0);
1417 size_out
= s
->dump_info
.page_size
;
1418 pd
.size
= cpu_to_dump32(s
, size_out
);
1420 ret
= write_cache(&page_data
, buf
,
1421 s
->dump_info
.page_size
, false);
1423 error_setg(errp
, "dump: failed to write page data");
1428 /* get and write page desc here */
1429 pd
.page_flags
= cpu_to_dump64(s
, 0);
1430 pd
.offset
= cpu_to_dump64(s
, offset_data
);
1431 offset_data
+= size_out
;
1433 ret
= write_cache(&page_desc
, &pd
, sizeof(PageDescriptor
), false);
1435 error_setg(errp
, "dump: failed to write page desc");
1439 s
->written_size
+= s
->dump_info
.page_size
;
1442 ret
= write_cache(&page_desc
, NULL
, 0, true);
1444 error_setg(errp
, "dump: failed to sync cache for page_desc");
1447 ret
= write_cache(&page_data
, NULL
, 0, true);
1449 error_setg(errp
, "dump: failed to sync cache for page_data");
1454 free_data_cache(&page_desc
);
1455 free_data_cache(&page_data
);
1464 static void create_kdump_vmcore(DumpState
*s
, Error
**errp
)
1467 Error
*local_err
= NULL
;
1470 * the kdump-compressed format is:
1472 * +------------------------------------------+ 0x0
1473 * | main header (struct disk_dump_header) |
1474 * |------------------------------------------+ block 1
1475 * | sub header (struct kdump_sub_header) |
1476 * |------------------------------------------+ block 2
1477 * | 1st-dump_bitmap |
1478 * |------------------------------------------+ block 2 + X blocks
1479 * | 2nd-dump_bitmap | (aligned by block)
1480 * |------------------------------------------+ block 2 + 2 * X blocks
1481 * | page desc for pfn 0 (struct page_desc) | (aligned by block)
1482 * | page desc for pfn 1 (struct page_desc) |
1484 * |------------------------------------------| (not aligned by block)
1485 * | page data (pfn 0) |
1486 * | page data (pfn 1) |
1488 * +------------------------------------------+
1491 ret
= write_start_flat_header(s
->fd
);
1493 error_setg(errp
, "dump: failed to write start flat header");
1497 write_dump_header(s
, &local_err
);
1499 error_propagate(errp
, local_err
);
1503 write_dump_bitmap(s
, &local_err
);
1505 error_propagate(errp
, local_err
);
1509 write_dump_pages(s
, &local_err
);
1511 error_propagate(errp
, local_err
);
1515 ret
= write_end_flat_header(s
->fd
);
1517 error_setg(errp
, "dump: failed to write end flat header");
1522 static ram_addr_t
get_start_block(DumpState
*s
)
1524 GuestPhysBlock
*block
;
1526 if (!s
->has_filter
) {
1527 s
->next_block
= QTAILQ_FIRST(&s
->guest_phys_blocks
.head
);
1531 QTAILQ_FOREACH(block
, &s
->guest_phys_blocks
.head
, next
) {
1532 if (block
->target_start
>= s
->begin
+ s
->length
||
1533 block
->target_end
<= s
->begin
) {
1534 /* This block is out of the range */
1538 s
->next_block
= block
;
1539 if (s
->begin
> block
->target_start
) {
1540 s
->start
= s
->begin
- block
->target_start
;
1550 static void get_max_mapnr(DumpState
*s
)
1552 GuestPhysBlock
*last_block
;
1554 last_block
= QTAILQ_LAST(&s
->guest_phys_blocks
.head
);
1555 s
->max_mapnr
= dump_paddr_to_pfn(s
, last_block
->target_end
);
1558 static DumpState dump_state_global
= { .status
= DUMP_STATUS_NONE
};
1560 static void dump_state_prepare(DumpState
*s
)
1562 /* zero the struct, setting status to active */
1563 *s
= (DumpState
) { .status
= DUMP_STATUS_ACTIVE
};
1566 bool qemu_system_dump_in_progress(void)
1568 DumpState
*state
= &dump_state_global
;
1569 return (qatomic_read(&state
->status
) == DUMP_STATUS_ACTIVE
);
1572 /* calculate total size of memory to be dumped (taking filter into
1574 static int64_t dump_calculate_size(DumpState
*s
)
1576 GuestPhysBlock
*block
;
1577 int64_t size
= 0, total
= 0, left
= 0, right
= 0;
1579 QTAILQ_FOREACH(block
, &s
->guest_phys_blocks
.head
, next
) {
1580 if (s
->has_filter
) {
1581 /* calculate the overlapped region. */
1582 left
= MAX(s
->begin
, block
->target_start
);
1583 right
= MIN(s
->begin
+ s
->length
, block
->target_end
);
1584 size
= right
- left
;
1585 size
= size
> 0 ? size
: 0;
1587 /* count the whole region in */
1588 size
= (block
->target_end
- block
->target_start
);
1596 static void vmcoreinfo_update_phys_base(DumpState
*s
)
1598 uint64_t size
, note_head_size
, name_size
, phys_base
;
1603 if (!note_name_equal(s
, s
->guest_note
, "VMCOREINFO")) {
1607 get_note_sizes(s
, s
->guest_note
, ¬e_head_size
, &name_size
, &size
);
1608 note_head_size
= ROUND_UP(note_head_size
, 4);
1610 vmci
= s
->guest_note
+ note_head_size
+ ROUND_UP(name_size
, 4);
1611 *(vmci
+ size
) = '\0';
1613 lines
= g_strsplit((char *)vmci
, "\n", -1);
1614 for (i
= 0; lines
[i
]; i
++) {
1615 const char *prefix
= NULL
;
1617 if (s
->dump_info
.d_machine
== EM_X86_64
) {
1618 prefix
= "NUMBER(phys_base)=";
1619 } else if (s
->dump_info
.d_machine
== EM_AARCH64
) {
1620 prefix
= "NUMBER(PHYS_OFFSET)=";
1623 if (prefix
&& g_str_has_prefix(lines
[i
], prefix
)) {
1624 if (qemu_strtou64(lines
[i
] + strlen(prefix
), NULL
, 16,
1626 warn_report("Failed to read %s", prefix
);
1628 s
->dump_info
.phys_base
= phys_base
;
1637 static void dump_init(DumpState
*s
, int fd
, bool has_format
,
1638 DumpGuestMemoryFormat format
, bool paging
, bool has_filter
,
1639 int64_t begin
, int64_t length
, Error
**errp
)
1641 VMCoreInfoState
*vmci
= vmcoreinfo_find();
1647 s
->has_format
= has_format
;
1649 s
->written_size
= 0;
1651 /* kdump-compressed is conflict with paging and filter */
1652 if (has_format
&& format
!= DUMP_GUEST_MEMORY_FORMAT_ELF
) {
1653 assert(!paging
&& !has_filter
);
1656 if (runstate_is_running()) {
1657 vm_stop(RUN_STATE_SAVE_VM
);
1663 /* If we use KVM, we should synchronize the registers before we get dump
1664 * info or physmap info.
1666 cpu_synchronize_all_states();
1673 s
->has_filter
= has_filter
;
1677 memory_mapping_list_init(&s
->list
);
1679 guest_phys_blocks_init(&s
->guest_phys_blocks
);
1680 guest_phys_blocks_append(&s
->guest_phys_blocks
);
1681 s
->total_size
= dump_calculate_size(s
);
1682 #ifdef DEBUG_DUMP_GUEST_MEMORY
1683 fprintf(stderr
, "DUMP: total memory to dump: %lu\n", s
->total_size
);
1686 /* it does not make sense to dump non-existent memory */
1687 if (!s
->total_size
) {
1688 error_setg(errp
, "dump: no guest memory to dump");
1692 s
->start
= get_start_block(s
);
1693 if (s
->start
== -1) {
1694 error_setg(errp
, QERR_INVALID_PARAMETER
, "begin");
1698 /* get dump info: endian, class and architecture.
1699 * If the target architecture is not supported, cpu_get_dump_info() will
1702 ret
= cpu_get_dump_info(&s
->dump_info
, &s
->guest_phys_blocks
);
1704 error_setg(errp
, QERR_UNSUPPORTED
);
1708 if (!s
->dump_info
.page_size
) {
1709 s
->dump_info
.page_size
= TARGET_PAGE_SIZE
;
1712 s
->note_size
= cpu_get_note_size(s
->dump_info
.d_class
,
1713 s
->dump_info
.d_machine
, nr_cpus
);
1714 if (s
->note_size
< 0) {
1715 error_setg(errp
, QERR_UNSUPPORTED
);
1720 * The goal of this block is to (a) update the previously guessed
1721 * phys_base, (b) copy the guest note out of the guest.
1722 * Failure to do so is not fatal for dumping.
1725 uint64_t addr
, note_head_size
, name_size
, desc_size
;
1729 note_head_size
= s
->dump_info
.d_class
== ELFCLASS32
?
1730 sizeof(Elf32_Nhdr
) : sizeof(Elf64_Nhdr
);
1732 format
= le16_to_cpu(vmci
->vmcoreinfo
.guest_format
);
1733 size
= le32_to_cpu(vmci
->vmcoreinfo
.size
);
1734 addr
= le64_to_cpu(vmci
->vmcoreinfo
.paddr
);
1735 if (!vmci
->has_vmcoreinfo
) {
1736 warn_report("guest note is not present");
1737 } else if (size
< note_head_size
|| size
> MAX_GUEST_NOTE_SIZE
) {
1738 warn_report("guest note size is invalid: %" PRIu32
, size
);
1739 } else if (format
!= FW_CFG_VMCOREINFO_FORMAT_ELF
) {
1740 warn_report("guest note format is unsupported: %" PRIu16
, format
);
1742 s
->guest_note
= g_malloc(size
+ 1); /* +1 for adding \0 */
1743 cpu_physical_memory_read(addr
, s
->guest_note
, size
);
1745 get_note_sizes(s
, s
->guest_note
, NULL
, &name_size
, &desc_size
);
1746 s
->guest_note_size
= ELF_NOTE_SIZE(note_head_size
, name_size
,
1748 if (name_size
> MAX_GUEST_NOTE_SIZE
||
1749 desc_size
> MAX_GUEST_NOTE_SIZE
||
1750 s
->guest_note_size
> size
) {
1751 warn_report("Invalid guest note header");
1752 g_free(s
->guest_note
);
1753 s
->guest_note
= NULL
;
1755 vmcoreinfo_update_phys_base(s
);
1756 s
->note_size
+= s
->guest_note_size
;
1761 /* get memory mapping */
1763 qemu_get_guest_memory_mapping(&s
->list
, &s
->guest_phys_blocks
, &err
);
1765 error_propagate(errp
, err
);
1769 qemu_get_guest_simple_memory_mapping(&s
->list
, &s
->guest_phys_blocks
);
1772 s
->nr_cpus
= nr_cpus
;
1777 tmp
= DIV_ROUND_UP(DIV_ROUND_UP(s
->max_mapnr
, CHAR_BIT
),
1778 s
->dump_info
.page_size
);
1779 s
->len_dump_bitmap
= tmp
* s
->dump_info
.page_size
;
1781 /* init for kdump-compressed format */
1782 if (has_format
&& format
!= DUMP_GUEST_MEMORY_FORMAT_ELF
) {
1784 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB
:
1785 s
->flag_compress
= DUMP_DH_COMPRESSED_ZLIB
;
1788 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO
:
1790 if (lzo_init() != LZO_E_OK
) {
1791 error_setg(errp
, "failed to initialize the LZO library");
1795 s
->flag_compress
= DUMP_DH_COMPRESSED_LZO
;
1798 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY
:
1799 s
->flag_compress
= DUMP_DH_COMPRESSED_SNAPPY
;
1803 s
->flag_compress
= 0;
1809 if (s
->has_filter
) {
1810 memory_mapping_filter(&s
->list
, s
->begin
, s
->length
);
1814 * calculate phdr_num
1816 * the type of ehdr->e_phnum is uint16_t, so we should avoid overflow
1818 s
->phdr_num
= 1; /* PT_NOTE */
1819 if (s
->list
.num
< UINT16_MAX
- 2) {
1820 s
->phdr_num
+= s
->list
.num
;
1821 s
->have_section
= false;
1823 s
->have_section
= true;
1824 s
->phdr_num
= PN_XNUM
;
1825 s
->sh_info
= 1; /* PT_NOTE */
1827 /* the type of shdr->sh_info is uint32_t, so we should avoid overflow */
1828 if (s
->list
.num
<= UINT32_MAX
- 1) {
1829 s
->sh_info
+= s
->list
.num
;
1831 s
->sh_info
= UINT32_MAX
;
1835 if (s
->dump_info
.d_class
== ELFCLASS64
) {
1836 if (s
->have_section
) {
1837 s
->memory_offset
= sizeof(Elf64_Ehdr
) +
1838 sizeof(Elf64_Phdr
) * s
->sh_info
+
1839 sizeof(Elf64_Shdr
) + s
->note_size
;
1841 s
->memory_offset
= sizeof(Elf64_Ehdr
) +
1842 sizeof(Elf64_Phdr
) * s
->phdr_num
+ s
->note_size
;
1845 if (s
->have_section
) {
1846 s
->memory_offset
= sizeof(Elf32_Ehdr
) +
1847 sizeof(Elf32_Phdr
) * s
->sh_info
+
1848 sizeof(Elf32_Shdr
) + s
->note_size
;
1850 s
->memory_offset
= sizeof(Elf32_Ehdr
) +
1851 sizeof(Elf32_Phdr
) * s
->phdr_num
+ s
->note_size
;
1861 /* this operation might be time consuming. */
1862 static void dump_process(DumpState
*s
, Error
**errp
)
1864 Error
*local_err
= NULL
;
1865 DumpQueryResult
*result
= NULL
;
1867 if (s
->has_format
&& s
->format
== DUMP_GUEST_MEMORY_FORMAT_WIN_DMP
) {
1868 #ifdef TARGET_X86_64
1869 create_win_dump(s
, &local_err
);
1871 } else if (s
->has_format
&& s
->format
!= DUMP_GUEST_MEMORY_FORMAT_ELF
) {
1872 create_kdump_vmcore(s
, &local_err
);
1874 create_vmcore(s
, &local_err
);
1877 /* make sure status is written after written_size updates */
1879 qatomic_set(&s
->status
,
1880 (local_err
? DUMP_STATUS_FAILED
: DUMP_STATUS_COMPLETED
));
1882 /* send DUMP_COMPLETED message (unconditionally) */
1883 result
= qmp_query_dump(NULL
);
1884 /* should never fail */
1886 qapi_event_send_dump_completed(result
, !!local_err
, (local_err
?
1887 error_get_pretty(local_err
) : NULL
));
1888 qapi_free_DumpQueryResult(result
);
1890 error_propagate(errp
, local_err
);
1894 static void *dump_thread(void *data
)
1896 DumpState
*s
= (DumpState
*)data
;
1897 dump_process(s
, NULL
);
1901 DumpQueryResult
*qmp_query_dump(Error
**errp
)
1903 DumpQueryResult
*result
= g_new(DumpQueryResult
, 1);
1904 DumpState
*state
= &dump_state_global
;
1905 result
->status
= qatomic_read(&state
->status
);
1906 /* make sure we are reading status and written_size in order */
1908 result
->completed
= state
->written_size
;
1909 result
->total
= state
->total_size
;
1913 void qmp_dump_guest_memory(bool paging
, const char *file
,
1914 bool has_detach
, bool detach
,
1915 bool has_begin
, int64_t begin
, bool has_length
,
1916 int64_t length
, bool has_format
,
1917 DumpGuestMemoryFormat format
, Error
**errp
)
1922 Error
*local_err
= NULL
;
1923 bool detach_p
= false;
1925 if (runstate_check(RUN_STATE_INMIGRATE
)) {
1926 error_setg(errp
, "Dump not allowed during incoming migration.");
1930 /* if there is a dump in background, we should wait until the dump
1932 if (qemu_system_dump_in_progress()) {
1933 error_setg(errp
, "There is a dump in process, please wait.");
1938 * kdump-compressed format need the whole memory dumped, so paging or
1939 * filter is not supported here.
1941 if ((has_format
&& format
!= DUMP_GUEST_MEMORY_FORMAT_ELF
) &&
1942 (paging
|| has_begin
|| has_length
)) {
1943 error_setg(errp
, "kdump-compressed format doesn't support paging or "
1947 if (has_begin
&& !has_length
) {
1948 error_setg(errp
, QERR_MISSING_PARAMETER
, "length");
1951 if (!has_begin
&& has_length
) {
1952 error_setg(errp
, QERR_MISSING_PARAMETER
, "begin");
1959 /* check whether lzo/snappy is supported */
1961 if (has_format
&& format
== DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO
) {
1962 error_setg(errp
, "kdump-lzo is not available now");
1967 #ifndef CONFIG_SNAPPY
1968 if (has_format
&& format
== DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY
) {
1969 error_setg(errp
, "kdump-snappy is not available now");
1974 #ifndef TARGET_X86_64
1975 if (has_format
&& format
== DUMP_GUEST_MEMORY_FORMAT_WIN_DMP
) {
1976 error_setg(errp
, "Windows dump is only available for x86-64");
1982 if (strstart(file
, "fd:", &p
)) {
1983 fd
= monitor_get_fd(monitor_cur(), p
, errp
);
1990 if (strstart(file
, "file:", &p
)) {
1991 fd
= qemu_open_old(p
, O_WRONLY
| O_CREAT
| O_TRUNC
| O_BINARY
, S_IRUSR
);
1993 error_setg_file_open(errp
, errno
, p
);
1999 error_setg(errp
, QERR_INVALID_PARAMETER
, "protocol");
2003 if (!dump_migration_blocker
) {
2004 error_setg(&dump_migration_blocker
,
2005 "Live migration disabled: dump-guest-memory in progress");
2009 * Allows even for -only-migratable, but forbid migration during the
2010 * process of dump guest memory.
2012 if (migrate_add_blocker_internal(dump_migration_blocker
, errp
)) {
2013 /* Remember to release the fd before passing it over to dump state */
2018 s
= &dump_state_global
;
2019 dump_state_prepare(s
);
2021 dump_init(s
, fd
, has_format
, format
, paging
, has_begin
,
2022 begin
, length
, &local_err
);
2024 error_propagate(errp
, local_err
);
2025 qatomic_set(&s
->status
, DUMP_STATUS_FAILED
);
2032 qemu_thread_create(&s
->dump_thread
, "dump_thread", dump_thread
,
2033 s
, QEMU_THREAD_DETACHED
);
2036 dump_process(s
, errp
);
2040 DumpGuestMemoryCapability
*qmp_query_dump_guest_memory_capability(Error
**errp
)
2042 DumpGuestMemoryCapability
*cap
=
2043 g_new0(DumpGuestMemoryCapability
, 1);
2044 DumpGuestMemoryFormatList
**tail
= &cap
->formats
;
2046 /* elf is always available */
2047 QAPI_LIST_APPEND(tail
, DUMP_GUEST_MEMORY_FORMAT_ELF
);
2049 /* kdump-zlib is always available */
2050 QAPI_LIST_APPEND(tail
, DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB
);
2052 /* add new item if kdump-lzo is available */
2054 QAPI_LIST_APPEND(tail
, DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO
);
2057 /* add new item if kdump-snappy is available */
2058 #ifdef CONFIG_SNAPPY
2059 QAPI_LIST_APPEND(tail
, DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY
);
2062 /* Windows dump is available only if target is x86_64 */
2063 #ifdef TARGET_X86_64
2064 QAPI_LIST_APPEND(tail
, DUMP_GUEST_MEMORY_FORMAT_WIN_DMP
);