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 static inline bool dump_is_64bit(DumpState
*s
)
59 return s
->dump_info
.d_class
== ELFCLASS64
;
62 static inline bool dump_has_filter(DumpState
*s
)
64 return s
->filter_area_length
> 0;
67 uint16_t cpu_to_dump16(DumpState
*s
, uint16_t val
)
69 if (s
->dump_info
.d_endian
== ELFDATA2LSB
) {
70 val
= cpu_to_le16(val
);
72 val
= cpu_to_be16(val
);
78 uint32_t cpu_to_dump32(DumpState
*s
, uint32_t val
)
80 if (s
->dump_info
.d_endian
== ELFDATA2LSB
) {
81 val
= cpu_to_le32(val
);
83 val
= cpu_to_be32(val
);
89 uint64_t cpu_to_dump64(DumpState
*s
, uint64_t val
)
91 if (s
->dump_info
.d_endian
== ELFDATA2LSB
) {
92 val
= cpu_to_le64(val
);
94 val
= cpu_to_be64(val
);
100 static int dump_cleanup(DumpState
*s
)
102 guest_phys_blocks_free(&s
->guest_phys_blocks
);
103 memory_mapping_list_free(&s
->list
);
105 g_free(s
->guest_note
);
106 s
->guest_note
= NULL
;
109 qemu_mutex_lock_iothread();
113 qemu_mutex_unlock_iothread();
116 migrate_del_blocker(dump_migration_blocker
);
121 static int fd_write_vmcore(const void *buf
, size_t size
, void *opaque
)
123 DumpState
*s
= opaque
;
126 written_size
= qemu_write_full(s
->fd
, buf
, size
);
127 if (written_size
!= size
) {
134 static void prepare_elf64_header(DumpState
*s
, Elf64_Ehdr
*elf_header
)
137 * phnum in the elf header is 16 bit, if we have more segments we
138 * set phnum to PN_XNUM and write the real number of segments to a
141 uint16_t phnum
= MIN(s
->phdr_num
, PN_XNUM
);
143 memset(elf_header
, 0, sizeof(Elf64_Ehdr
));
144 memcpy(elf_header
, ELFMAG
, SELFMAG
);
145 elf_header
->e_ident
[EI_CLASS
] = ELFCLASS64
;
146 elf_header
->e_ident
[EI_DATA
] = s
->dump_info
.d_endian
;
147 elf_header
->e_ident
[EI_VERSION
] = EV_CURRENT
;
148 elf_header
->e_type
= cpu_to_dump16(s
, ET_CORE
);
149 elf_header
->e_machine
= cpu_to_dump16(s
, s
->dump_info
.d_machine
);
150 elf_header
->e_version
= cpu_to_dump32(s
, EV_CURRENT
);
151 elf_header
->e_ehsize
= cpu_to_dump16(s
, sizeof(elf_header
));
152 elf_header
->e_phoff
= cpu_to_dump64(s
, s
->phdr_offset
);
153 elf_header
->e_phentsize
= cpu_to_dump16(s
, sizeof(Elf64_Phdr
));
154 elf_header
->e_phnum
= cpu_to_dump16(s
, phnum
);
156 elf_header
->e_shoff
= cpu_to_dump64(s
, s
->shdr_offset
);
157 elf_header
->e_shentsize
= cpu_to_dump16(s
, sizeof(Elf64_Shdr
));
158 elf_header
->e_shnum
= cpu_to_dump16(s
, s
->shdr_num
);
162 static void prepare_elf32_header(DumpState
*s
, Elf32_Ehdr
*elf_header
)
165 * phnum in the elf header is 16 bit, if we have more segments we
166 * set phnum to PN_XNUM and write the real number of segments to a
169 uint16_t phnum
= MIN(s
->phdr_num
, PN_XNUM
);
171 memset(elf_header
, 0, sizeof(Elf32_Ehdr
));
172 memcpy(elf_header
, ELFMAG
, SELFMAG
);
173 elf_header
->e_ident
[EI_CLASS
] = ELFCLASS32
;
174 elf_header
->e_ident
[EI_DATA
] = s
->dump_info
.d_endian
;
175 elf_header
->e_ident
[EI_VERSION
] = EV_CURRENT
;
176 elf_header
->e_type
= cpu_to_dump16(s
, ET_CORE
);
177 elf_header
->e_machine
= cpu_to_dump16(s
, s
->dump_info
.d_machine
);
178 elf_header
->e_version
= cpu_to_dump32(s
, EV_CURRENT
);
179 elf_header
->e_ehsize
= cpu_to_dump16(s
, sizeof(elf_header
));
180 elf_header
->e_phoff
= cpu_to_dump32(s
, s
->phdr_offset
);
181 elf_header
->e_phentsize
= cpu_to_dump16(s
, sizeof(Elf32_Phdr
));
182 elf_header
->e_phnum
= cpu_to_dump16(s
, phnum
);
184 elf_header
->e_shoff
= cpu_to_dump32(s
, s
->shdr_offset
);
185 elf_header
->e_shentsize
= cpu_to_dump16(s
, sizeof(Elf32_Shdr
));
186 elf_header
->e_shnum
= cpu_to_dump16(s
, s
->shdr_num
);
190 static void write_elf_header(DumpState
*s
, Error
**errp
)
192 Elf32_Ehdr elf32_header
;
193 Elf64_Ehdr elf64_header
;
198 if (dump_is_64bit(s
)) {
199 prepare_elf64_header(s
, &elf64_header
);
200 header_size
= sizeof(elf64_header
);
201 header_ptr
= &elf64_header
;
203 prepare_elf32_header(s
, &elf32_header
);
204 header_size
= sizeof(elf32_header
);
205 header_ptr
= &elf32_header
;
208 ret
= fd_write_vmcore(header_ptr
, header_size
, s
);
210 error_setg_errno(errp
, -ret
, "dump: failed to write elf header");
214 static void write_elf64_load(DumpState
*s
, MemoryMapping
*memory_mapping
,
215 int phdr_index
, hwaddr offset
,
216 hwaddr filesz
, Error
**errp
)
221 memset(&phdr
, 0, sizeof(Elf64_Phdr
));
222 phdr
.p_type
= cpu_to_dump32(s
, PT_LOAD
);
223 phdr
.p_offset
= cpu_to_dump64(s
, offset
);
224 phdr
.p_paddr
= cpu_to_dump64(s
, memory_mapping
->phys_addr
);
225 phdr
.p_filesz
= cpu_to_dump64(s
, filesz
);
226 phdr
.p_memsz
= cpu_to_dump64(s
, memory_mapping
->length
);
227 phdr
.p_vaddr
= cpu_to_dump64(s
, memory_mapping
->virt_addr
) ?: phdr
.p_paddr
;
229 assert(memory_mapping
->length
>= filesz
);
231 ret
= fd_write_vmcore(&phdr
, sizeof(Elf64_Phdr
), s
);
233 error_setg_errno(errp
, -ret
,
234 "dump: failed to write program header table");
238 static void write_elf32_load(DumpState
*s
, MemoryMapping
*memory_mapping
,
239 int phdr_index
, hwaddr offset
,
240 hwaddr filesz
, Error
**errp
)
245 memset(&phdr
, 0, sizeof(Elf32_Phdr
));
246 phdr
.p_type
= cpu_to_dump32(s
, PT_LOAD
);
247 phdr
.p_offset
= cpu_to_dump32(s
, offset
);
248 phdr
.p_paddr
= cpu_to_dump32(s
, memory_mapping
->phys_addr
);
249 phdr
.p_filesz
= cpu_to_dump32(s
, filesz
);
250 phdr
.p_memsz
= cpu_to_dump32(s
, memory_mapping
->length
);
252 cpu_to_dump32(s
, memory_mapping
->virt_addr
) ?: phdr
.p_paddr
;
254 assert(memory_mapping
->length
>= filesz
);
256 ret
= fd_write_vmcore(&phdr
, sizeof(Elf32_Phdr
), s
);
258 error_setg_errno(errp
, -ret
,
259 "dump: failed to write program header table");
263 static void prepare_elf64_phdr_note(DumpState
*s
, Elf64_Phdr
*phdr
)
265 memset(phdr
, 0, sizeof(*phdr
));
266 phdr
->p_type
= cpu_to_dump32(s
, PT_NOTE
);
267 phdr
->p_offset
= cpu_to_dump64(s
, s
->note_offset
);
269 phdr
->p_filesz
= cpu_to_dump64(s
, s
->note_size
);
270 phdr
->p_memsz
= cpu_to_dump64(s
, s
->note_size
);
274 static inline int cpu_index(CPUState
*cpu
)
276 return cpu
->cpu_index
+ 1;
279 static void write_guest_note(WriteCoreDumpFunction f
, DumpState
*s
,
285 ret
= f(s
->guest_note
, s
->guest_note_size
, s
);
287 error_setg(errp
, "dump: failed to write guest note");
292 static void write_elf64_notes(WriteCoreDumpFunction f
, DumpState
*s
,
301 ret
= cpu_write_elf64_note(f
, cpu
, id
, s
);
303 error_setg(errp
, "dump: failed to write elf notes");
309 ret
= cpu_write_elf64_qemunote(f
, cpu
, s
);
311 error_setg(errp
, "dump: failed to write CPU status");
316 write_guest_note(f
, s
, errp
);
319 static void prepare_elf32_phdr_note(DumpState
*s
, Elf32_Phdr
*phdr
)
321 memset(phdr
, 0, sizeof(*phdr
));
322 phdr
->p_type
= cpu_to_dump32(s
, PT_NOTE
);
323 phdr
->p_offset
= cpu_to_dump32(s
, s
->note_offset
);
325 phdr
->p_filesz
= cpu_to_dump32(s
, s
->note_size
);
326 phdr
->p_memsz
= cpu_to_dump32(s
, s
->note_size
);
330 static void write_elf32_notes(WriteCoreDumpFunction f
, DumpState
*s
,
339 ret
= cpu_write_elf32_note(f
, cpu
, id
, s
);
341 error_setg(errp
, "dump: failed to write elf notes");
347 ret
= cpu_write_elf32_qemunote(f
, cpu
, s
);
349 error_setg(errp
, "dump: failed to write CPU status");
354 write_guest_note(f
, s
, errp
);
357 static void write_elf_phdr_note(DumpState
*s
, Error
**errp
)
366 if (dump_is_64bit(s
)) {
367 prepare_elf64_phdr_note(s
, &phdr64
);
368 size
= sizeof(phdr64
);
371 prepare_elf32_phdr_note(s
, &phdr32
);
372 size
= sizeof(phdr32
);
376 ret
= fd_write_vmcore(phdr
, size
, s
);
378 error_setg_errno(errp
, -ret
,
379 "dump: failed to write program header table");
383 static void write_elf_section(DumpState
*s
, int type
, Error
**errp
)
392 shdr_size
= sizeof(Elf32_Shdr
);
393 memset(&shdr32
, 0, shdr_size
);
394 shdr32
.sh_info
= cpu_to_dump32(s
, s
->phdr_num
);
397 shdr_size
= sizeof(Elf64_Shdr
);
398 memset(&shdr64
, 0, shdr_size
);
399 shdr64
.sh_info
= cpu_to_dump32(s
, s
->phdr_num
);
403 ret
= fd_write_vmcore(shdr
, shdr_size
, s
);
405 error_setg_errno(errp
, -ret
,
406 "dump: failed to write section header table");
410 static void write_data(DumpState
*s
, void *buf
, int length
, Error
**errp
)
414 ret
= fd_write_vmcore(buf
, length
, s
);
416 error_setg_errno(errp
, -ret
, "dump: failed to save memory");
418 s
->written_size
+= length
;
422 /* write the memory to vmcore. 1 page per I/O. */
423 static void write_memory(DumpState
*s
, GuestPhysBlock
*block
, ram_addr_t start
,
424 int64_t size
, Error
**errp
)
429 for (i
= 0; i
< size
/ s
->dump_info
.page_size
; i
++) {
430 write_data(s
, block
->host_addr
+ start
+ i
* s
->dump_info
.page_size
,
431 s
->dump_info
.page_size
, errp
);
437 if ((size
% s
->dump_info
.page_size
) != 0) {
438 write_data(s
, block
->host_addr
+ start
+ i
* s
->dump_info
.page_size
,
439 size
% s
->dump_info
.page_size
, errp
);
446 /* get the memory's offset and size in the vmcore */
447 static void get_offset_range(hwaddr phys_addr
,
448 ram_addr_t mapping_length
,
453 GuestPhysBlock
*block
;
454 hwaddr offset
= s
->memory_offset
;
455 int64_t size_in_block
, start
;
457 /* When the memory is not stored into vmcore, offset will be -1 */
461 if (dump_has_filter(s
)) {
462 if (phys_addr
< s
->filter_area_begin
||
463 phys_addr
>= s
->filter_area_begin
+ s
->filter_area_length
) {
468 QTAILQ_FOREACH(block
, &s
->guest_phys_blocks
.head
, next
) {
469 if (dump_has_filter(s
)) {
470 if (block
->target_start
>= s
->filter_area_begin
+ s
->filter_area_length
||
471 block
->target_end
<= s
->filter_area_begin
) {
472 /* This block is out of the range */
476 if (s
->filter_area_begin
<= block
->target_start
) {
477 start
= block
->target_start
;
479 start
= s
->filter_area_begin
;
482 size_in_block
= block
->target_end
- start
;
483 if (s
->filter_area_begin
+ s
->filter_area_length
< block
->target_end
) {
484 size_in_block
-= block
->target_end
- (s
->filter_area_begin
+ s
->filter_area_length
);
487 start
= block
->target_start
;
488 size_in_block
= block
->target_end
- block
->target_start
;
491 if (phys_addr
>= start
&& phys_addr
< start
+ size_in_block
) {
492 *p_offset
= phys_addr
- start
+ offset
;
494 /* The offset range mapped from the vmcore file must not spill over
495 * the GuestPhysBlock, clamp it. The rest of the mapping will be
496 * zero-filled in memory at load time; see
497 * <http://refspecs.linuxbase.org/elf/gabi4+/ch5.pheader.html>.
499 *p_filesz
= phys_addr
+ mapping_length
<= start
+ size_in_block
?
501 size_in_block
- (phys_addr
- start
);
505 offset
+= size_in_block
;
509 static void write_elf_phdr_loads(DumpState
*s
, Error
**errp
)
512 hwaddr offset
, filesz
;
513 MemoryMapping
*memory_mapping
;
514 uint32_t phdr_index
= 1;
516 QTAILQ_FOREACH(memory_mapping
, &s
->list
.head
, next
) {
517 get_offset_range(memory_mapping
->phys_addr
,
518 memory_mapping
->length
,
519 s
, &offset
, &filesz
);
520 if (dump_is_64bit(s
)) {
521 write_elf64_load(s
, memory_mapping
, phdr_index
++, offset
,
524 write_elf32_load(s
, memory_mapping
, phdr_index
++, offset
,
532 if (phdr_index
>= s
->phdr_num
) {
538 static void write_elf_notes(DumpState
*s
, Error
**errp
)
540 if (dump_is_64bit(s
)) {
541 write_elf64_notes(fd_write_vmcore
, s
, errp
);
543 write_elf32_notes(fd_write_vmcore
, s
, errp
);
547 /* write elf header, PT_NOTE and elf note to vmcore. */
548 static void dump_begin(DumpState
*s
, Error
**errp
)
553 * the vmcore's format is:
572 * we only know where the memory is saved after we write elf note into
576 /* write elf header to vmcore */
577 write_elf_header(s
, errp
);
582 /* write PT_NOTE to vmcore */
583 write_elf_phdr_note(s
, errp
);
588 /* write all PT_LOADs to vmcore */
589 write_elf_phdr_loads(s
, errp
);
594 /* write section to vmcore */
596 write_elf_section(s
, 1, errp
);
602 /* write notes to vmcore */
603 write_elf_notes(s
, errp
);
606 static int64_t dump_filtered_memblock_size(GuestPhysBlock
*block
,
607 int64_t filter_area_start
,
608 int64_t filter_area_length
)
610 int64_t size
, left
, right
;
612 /* No filter, return full size */
613 if (!filter_area_length
) {
614 return block
->target_end
- block
->target_start
;
617 /* calculate the overlapped region. */
618 left
= MAX(filter_area_start
, block
->target_start
);
619 right
= MIN(filter_area_start
+ filter_area_length
, block
->target_end
);
621 size
= size
> 0 ? size
: 0;
626 static int64_t dump_filtered_memblock_start(GuestPhysBlock
*block
,
627 int64_t filter_area_start
,
628 int64_t filter_area_length
)
630 if (filter_area_length
) {
631 /* return -1 if the block is not within filter area */
632 if (block
->target_start
>= filter_area_start
+ filter_area_length
||
633 block
->target_end
<= filter_area_start
) {
637 if (filter_area_start
> block
->target_start
) {
638 return filter_area_start
- block
->target_start
;
645 /* write all memory to vmcore */
646 static void dump_iterate(DumpState
*s
, Error
**errp
)
649 GuestPhysBlock
*block
;
650 int64_t memblock_size
, memblock_start
;
652 QTAILQ_FOREACH(block
, &s
->guest_phys_blocks
.head
, next
) {
653 memblock_start
= dump_filtered_memblock_start(block
, s
->filter_area_begin
, s
->filter_area_length
);
654 if (memblock_start
== -1) {
658 memblock_size
= dump_filtered_memblock_size(block
, s
->filter_area_begin
, s
->filter_area_length
);
660 /* Write the memory to file */
661 write_memory(s
, block
, memblock_start
, memblock_size
, errp
);
668 static void create_vmcore(DumpState
*s
, Error
**errp
)
677 dump_iterate(s
, errp
);
680 static int write_start_flat_header(int fd
)
682 MakedumpfileHeader
*mh
;
685 QEMU_BUILD_BUG_ON(sizeof *mh
> MAX_SIZE_MDF_HEADER
);
686 mh
= g_malloc0(MAX_SIZE_MDF_HEADER
);
688 memcpy(mh
->signature
, MAKEDUMPFILE_SIGNATURE
,
689 MIN(sizeof mh
->signature
, sizeof MAKEDUMPFILE_SIGNATURE
));
691 mh
->type
= cpu_to_be64(TYPE_FLAT_HEADER
);
692 mh
->version
= cpu_to_be64(VERSION_FLAT_HEADER
);
695 written_size
= qemu_write_full(fd
, mh
, MAX_SIZE_MDF_HEADER
);
696 if (written_size
!= MAX_SIZE_MDF_HEADER
) {
704 static int write_end_flat_header(int fd
)
706 MakedumpfileDataHeader mdh
;
708 mdh
.offset
= END_FLAG_FLAT_HEADER
;
709 mdh
.buf_size
= END_FLAG_FLAT_HEADER
;
712 written_size
= qemu_write_full(fd
, &mdh
, sizeof(mdh
));
713 if (written_size
!= sizeof(mdh
)) {
720 static int write_buffer(int fd
, off_t offset
, const void *buf
, size_t size
)
723 MakedumpfileDataHeader mdh
;
725 mdh
.offset
= cpu_to_be64(offset
);
726 mdh
.buf_size
= cpu_to_be64(size
);
728 written_size
= qemu_write_full(fd
, &mdh
, sizeof(mdh
));
729 if (written_size
!= sizeof(mdh
)) {
733 written_size
= qemu_write_full(fd
, buf
, size
);
734 if (written_size
!= size
) {
741 static int buf_write_note(const void *buf
, size_t size
, void *opaque
)
743 DumpState
*s
= opaque
;
745 /* note_buf is not enough */
746 if (s
->note_buf_offset
+ size
> s
->note_size
) {
750 memcpy(s
->note_buf
+ s
->note_buf_offset
, buf
, size
);
752 s
->note_buf_offset
+= size
;
758 * This function retrieves various sizes from an elf header.
760 * @note has to be a valid ELF note. The return sizes are unmodified
761 * (not padded or rounded up to be multiple of 4).
763 static void get_note_sizes(DumpState
*s
, const void *note
,
764 uint64_t *note_head_size
,
768 uint64_t note_head_sz
;
772 if (dump_is_64bit(s
)) {
773 const Elf64_Nhdr
*hdr
= note
;
774 note_head_sz
= sizeof(Elf64_Nhdr
);
775 name_sz
= tswap64(hdr
->n_namesz
);
776 desc_sz
= tswap64(hdr
->n_descsz
);
778 const Elf32_Nhdr
*hdr
= note
;
779 note_head_sz
= sizeof(Elf32_Nhdr
);
780 name_sz
= tswap32(hdr
->n_namesz
);
781 desc_sz
= tswap32(hdr
->n_descsz
);
784 if (note_head_size
) {
785 *note_head_size
= note_head_sz
;
788 *name_size
= name_sz
;
791 *desc_size
= desc_sz
;
795 static bool note_name_equal(DumpState
*s
,
796 const uint8_t *note
, const char *name
)
798 int len
= strlen(name
) + 1;
799 uint64_t head_size
, name_size
;
801 get_note_sizes(s
, note
, &head_size
, &name_size
, NULL
);
802 head_size
= ROUND_UP(head_size
, 4);
804 return name_size
== len
&& memcmp(note
+ head_size
, name
, len
) == 0;
807 /* write common header, sub header and elf note to vmcore */
808 static void create_header32(DumpState
*s
, Error
**errp
)
811 DiskDumpHeader32
*dh
= NULL
;
812 KdumpSubHeader32
*kh
= NULL
;
815 uint32_t sub_hdr_size
;
816 uint32_t bitmap_blocks
;
818 uint64_t offset_note
;
820 /* write common header, the version of kdump-compressed format is 6th */
821 size
= sizeof(DiskDumpHeader32
);
822 dh
= g_malloc0(size
);
824 memcpy(dh
->signature
, KDUMP_SIGNATURE
, SIG_LEN
);
825 dh
->header_version
= cpu_to_dump32(s
, 6);
826 block_size
= s
->dump_info
.page_size
;
827 dh
->block_size
= cpu_to_dump32(s
, block_size
);
828 sub_hdr_size
= sizeof(struct KdumpSubHeader32
) + s
->note_size
;
829 sub_hdr_size
= DIV_ROUND_UP(sub_hdr_size
, block_size
);
830 dh
->sub_hdr_size
= cpu_to_dump32(s
, sub_hdr_size
);
831 /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
832 dh
->max_mapnr
= cpu_to_dump32(s
, MIN(s
->max_mapnr
, UINT_MAX
));
833 dh
->nr_cpus
= cpu_to_dump32(s
, s
->nr_cpus
);
834 bitmap_blocks
= DIV_ROUND_UP(s
->len_dump_bitmap
, block_size
) * 2;
835 dh
->bitmap_blocks
= cpu_to_dump32(s
, bitmap_blocks
);
836 strncpy(dh
->utsname
.machine
, ELF_MACHINE_UNAME
, sizeof(dh
->utsname
.machine
));
838 if (s
->flag_compress
& DUMP_DH_COMPRESSED_ZLIB
) {
839 status
|= DUMP_DH_COMPRESSED_ZLIB
;
842 if (s
->flag_compress
& DUMP_DH_COMPRESSED_LZO
) {
843 status
|= DUMP_DH_COMPRESSED_LZO
;
847 if (s
->flag_compress
& DUMP_DH_COMPRESSED_SNAPPY
) {
848 status
|= DUMP_DH_COMPRESSED_SNAPPY
;
851 dh
->status
= cpu_to_dump32(s
, status
);
853 if (write_buffer(s
->fd
, 0, dh
, size
) < 0) {
854 error_setg(errp
, "dump: failed to write disk dump header");
858 /* write sub header */
859 size
= sizeof(KdumpSubHeader32
);
860 kh
= g_malloc0(size
);
862 /* 64bit max_mapnr_64 */
863 kh
->max_mapnr_64
= cpu_to_dump64(s
, s
->max_mapnr
);
864 kh
->phys_base
= cpu_to_dump32(s
, s
->dump_info
.phys_base
);
865 kh
->dump_level
= cpu_to_dump32(s
, DUMP_LEVEL
);
867 offset_note
= DISKDUMP_HEADER_BLOCKS
* block_size
+ size
;
869 note_name_equal(s
, s
->guest_note
, "VMCOREINFO")) {
870 uint64_t hsize
, name_size
, size_vmcoreinfo_desc
, offset_vmcoreinfo
;
872 get_note_sizes(s
, s
->guest_note
,
873 &hsize
, &name_size
, &size_vmcoreinfo_desc
);
874 offset_vmcoreinfo
= offset_note
+ s
->note_size
- s
->guest_note_size
+
875 (DIV_ROUND_UP(hsize
, 4) + DIV_ROUND_UP(name_size
, 4)) * 4;
876 kh
->offset_vmcoreinfo
= cpu_to_dump64(s
, offset_vmcoreinfo
);
877 kh
->size_vmcoreinfo
= cpu_to_dump32(s
, size_vmcoreinfo_desc
);
880 kh
->offset_note
= cpu_to_dump64(s
, offset_note
);
881 kh
->note_size
= cpu_to_dump32(s
, s
->note_size
);
883 if (write_buffer(s
->fd
, DISKDUMP_HEADER_BLOCKS
*
884 block_size
, kh
, size
) < 0) {
885 error_setg(errp
, "dump: failed to write kdump sub header");
890 s
->note_buf
= g_malloc0(s
->note_size
);
891 s
->note_buf_offset
= 0;
893 /* use s->note_buf to store notes temporarily */
894 write_elf32_notes(buf_write_note
, s
, errp
);
898 if (write_buffer(s
->fd
, offset_note
, s
->note_buf
,
900 error_setg(errp
, "dump: failed to write notes");
904 /* get offset of dump_bitmap */
905 s
->offset_dump_bitmap
= (DISKDUMP_HEADER_BLOCKS
+ sub_hdr_size
) *
908 /* get offset of page */
909 s
->offset_page
= (DISKDUMP_HEADER_BLOCKS
+ sub_hdr_size
+ bitmap_blocks
) *
918 /* write common header, sub header and elf note to vmcore */
919 static void create_header64(DumpState
*s
, Error
**errp
)
922 DiskDumpHeader64
*dh
= NULL
;
923 KdumpSubHeader64
*kh
= NULL
;
926 uint32_t sub_hdr_size
;
927 uint32_t bitmap_blocks
;
929 uint64_t offset_note
;
931 /* write common header, the version of kdump-compressed format is 6th */
932 size
= sizeof(DiskDumpHeader64
);
933 dh
= g_malloc0(size
);
935 memcpy(dh
->signature
, KDUMP_SIGNATURE
, SIG_LEN
);
936 dh
->header_version
= cpu_to_dump32(s
, 6);
937 block_size
= s
->dump_info
.page_size
;
938 dh
->block_size
= cpu_to_dump32(s
, block_size
);
939 sub_hdr_size
= sizeof(struct KdumpSubHeader64
) + s
->note_size
;
940 sub_hdr_size
= DIV_ROUND_UP(sub_hdr_size
, block_size
);
941 dh
->sub_hdr_size
= cpu_to_dump32(s
, sub_hdr_size
);
942 /* dh->max_mapnr may be truncated, full 64bit is in kh.max_mapnr_64 */
943 dh
->max_mapnr
= cpu_to_dump32(s
, MIN(s
->max_mapnr
, UINT_MAX
));
944 dh
->nr_cpus
= cpu_to_dump32(s
, s
->nr_cpus
);
945 bitmap_blocks
= DIV_ROUND_UP(s
->len_dump_bitmap
, block_size
) * 2;
946 dh
->bitmap_blocks
= cpu_to_dump32(s
, bitmap_blocks
);
947 strncpy(dh
->utsname
.machine
, ELF_MACHINE_UNAME
, sizeof(dh
->utsname
.machine
));
949 if (s
->flag_compress
& DUMP_DH_COMPRESSED_ZLIB
) {
950 status
|= DUMP_DH_COMPRESSED_ZLIB
;
953 if (s
->flag_compress
& DUMP_DH_COMPRESSED_LZO
) {
954 status
|= DUMP_DH_COMPRESSED_LZO
;
958 if (s
->flag_compress
& DUMP_DH_COMPRESSED_SNAPPY
) {
959 status
|= DUMP_DH_COMPRESSED_SNAPPY
;
962 dh
->status
= cpu_to_dump32(s
, status
);
964 if (write_buffer(s
->fd
, 0, dh
, size
) < 0) {
965 error_setg(errp
, "dump: failed to write disk dump header");
969 /* write sub header */
970 size
= sizeof(KdumpSubHeader64
);
971 kh
= g_malloc0(size
);
973 /* 64bit max_mapnr_64 */
974 kh
->max_mapnr_64
= cpu_to_dump64(s
, s
->max_mapnr
);
975 kh
->phys_base
= cpu_to_dump64(s
, s
->dump_info
.phys_base
);
976 kh
->dump_level
= cpu_to_dump32(s
, DUMP_LEVEL
);
978 offset_note
= DISKDUMP_HEADER_BLOCKS
* block_size
+ size
;
980 note_name_equal(s
, s
->guest_note
, "VMCOREINFO")) {
981 uint64_t hsize
, name_size
, size_vmcoreinfo_desc
, offset_vmcoreinfo
;
983 get_note_sizes(s
, s
->guest_note
,
984 &hsize
, &name_size
, &size_vmcoreinfo_desc
);
985 offset_vmcoreinfo
= offset_note
+ s
->note_size
- s
->guest_note_size
+
986 (DIV_ROUND_UP(hsize
, 4) + DIV_ROUND_UP(name_size
, 4)) * 4;
987 kh
->offset_vmcoreinfo
= cpu_to_dump64(s
, offset_vmcoreinfo
);
988 kh
->size_vmcoreinfo
= cpu_to_dump64(s
, size_vmcoreinfo_desc
);
991 kh
->offset_note
= cpu_to_dump64(s
, offset_note
);
992 kh
->note_size
= cpu_to_dump64(s
, s
->note_size
);
994 if (write_buffer(s
->fd
, DISKDUMP_HEADER_BLOCKS
*
995 block_size
, kh
, size
) < 0) {
996 error_setg(errp
, "dump: failed to write kdump sub header");
1001 s
->note_buf
= g_malloc0(s
->note_size
);
1002 s
->note_buf_offset
= 0;
1004 /* use s->note_buf to store notes temporarily */
1005 write_elf64_notes(buf_write_note
, s
, errp
);
1010 if (write_buffer(s
->fd
, offset_note
, s
->note_buf
,
1011 s
->note_size
) < 0) {
1012 error_setg(errp
, "dump: failed to write notes");
1016 /* get offset of dump_bitmap */
1017 s
->offset_dump_bitmap
= (DISKDUMP_HEADER_BLOCKS
+ sub_hdr_size
) *
1020 /* get offset of page */
1021 s
->offset_page
= (DISKDUMP_HEADER_BLOCKS
+ sub_hdr_size
+ bitmap_blocks
) *
1027 g_free(s
->note_buf
);
1030 static void write_dump_header(DumpState
*s
, Error
**errp
)
1032 if (dump_is_64bit(s
)) {
1033 create_header64(s
, errp
);
1035 create_header32(s
, errp
);
1039 static size_t dump_bitmap_get_bufsize(DumpState
*s
)
1041 return s
->dump_info
.page_size
;
1045 * set dump_bitmap sequencely. the bit before last_pfn is not allowed to be
1046 * rewritten, so if need to set the first bit, set last_pfn and pfn to 0.
1047 * set_dump_bitmap will always leave the recently set bit un-sync. And setting
1048 * (last bit + sizeof(buf) * 8) to 0 will do flushing the content in buf into
1049 * vmcore, ie. synchronizing un-sync bit into vmcore.
1051 static int set_dump_bitmap(uint64_t last_pfn
, uint64_t pfn
, bool value
,
1052 uint8_t *buf
, DumpState
*s
)
1054 off_t old_offset
, new_offset
;
1055 off_t offset_bitmap1
, offset_bitmap2
;
1057 size_t bitmap_bufsize
= dump_bitmap_get_bufsize(s
);
1058 size_t bits_per_buf
= bitmap_bufsize
* CHAR_BIT
;
1060 /* should not set the previous place */
1061 assert(last_pfn
<= pfn
);
1064 * if the bit needed to be set is not cached in buf, flush the data in buf
1065 * to vmcore firstly.
1066 * making new_offset be bigger than old_offset can also sync remained data
1069 old_offset
= bitmap_bufsize
* (last_pfn
/ bits_per_buf
);
1070 new_offset
= bitmap_bufsize
* (pfn
/ bits_per_buf
);
1072 while (old_offset
< new_offset
) {
1073 /* calculate the offset and write dump_bitmap */
1074 offset_bitmap1
= s
->offset_dump_bitmap
+ old_offset
;
1075 if (write_buffer(s
->fd
, offset_bitmap1
, buf
,
1076 bitmap_bufsize
) < 0) {
1080 /* dump level 1 is chosen, so 1st and 2nd bitmap are same */
1081 offset_bitmap2
= s
->offset_dump_bitmap
+ s
->len_dump_bitmap
+
1083 if (write_buffer(s
->fd
, offset_bitmap2
, buf
,
1084 bitmap_bufsize
) < 0) {
1088 memset(buf
, 0, bitmap_bufsize
);
1089 old_offset
+= bitmap_bufsize
;
1092 /* get the exact place of the bit in the buf, and set it */
1093 byte
= (pfn
% bits_per_buf
) / CHAR_BIT
;
1094 bit
= (pfn
% bits_per_buf
) % CHAR_BIT
;
1096 buf
[byte
] |= 1u << bit
;
1098 buf
[byte
] &= ~(1u << bit
);
1104 static uint64_t dump_paddr_to_pfn(DumpState
*s
, uint64_t addr
)
1106 int target_page_shift
= ctz32(s
->dump_info
.page_size
);
1108 return (addr
>> target_page_shift
) - ARCH_PFN_OFFSET
;
1111 static uint64_t dump_pfn_to_paddr(DumpState
*s
, uint64_t pfn
)
1113 int target_page_shift
= ctz32(s
->dump_info
.page_size
);
1115 return (pfn
+ ARCH_PFN_OFFSET
) << target_page_shift
;
1119 * Return the page frame number and the page content in *bufptr. bufptr can be
1120 * NULL. If not NULL, *bufptr must contains a target page size of pre-allocated
1121 * memory. This is not necessarily the memory returned.
1123 static bool get_next_page(GuestPhysBlock
**blockptr
, uint64_t *pfnptr
,
1124 uint8_t **bufptr
, DumpState
*s
)
1126 GuestPhysBlock
*block
= *blockptr
;
1127 uint32_t page_size
= s
->dump_info
.page_size
;
1128 uint8_t *buf
= NULL
, *hbuf
;
1131 /* block == NULL means the start of the iteration */
1133 block
= QTAILQ_FIRST(&s
->guest_phys_blocks
.head
);
1135 addr
= block
->target_start
;
1136 *pfnptr
= dump_paddr_to_pfn(s
, addr
);
1139 addr
= dump_pfn_to_paddr(s
, *pfnptr
);
1141 assert(block
!= NULL
);
1144 if (addr
>= block
->target_start
&& addr
< block
->target_end
) {
1145 size_t n
= MIN(block
->target_end
- addr
, page_size
- addr
% page_size
);
1146 hbuf
= block
->host_addr
+ (addr
- block
->target_start
);
1148 if (n
== page_size
) {
1149 /* this is a whole target page, go for it */
1150 assert(addr
% page_size
== 0);
1153 } else if (bufptr
) {
1156 memset(buf
, 0, page_size
);
1162 memcpy(buf
+ addr
% page_size
, hbuf
, n
);
1164 if (addr
% page_size
== 0) {
1165 /* we filled up the page */
1169 /* the next page is in the next block */
1170 *blockptr
= block
= QTAILQ_NEXT(block
, next
);
1175 addr
= block
->target_start
;
1176 /* are we still in the same page? */
1177 if (dump_paddr_to_pfn(s
, addr
) != *pfnptr
) {
1179 /* no, but we already filled something earlier, return it */
1182 /* else continue from there */
1183 *pfnptr
= dump_paddr_to_pfn(s
, addr
);
1196 static void write_dump_bitmap(DumpState
*s
, Error
**errp
)
1199 uint64_t last_pfn
, pfn
;
1200 void *dump_bitmap_buf
;
1201 size_t num_dumpable
;
1202 GuestPhysBlock
*block_iter
= NULL
;
1203 size_t bitmap_bufsize
= dump_bitmap_get_bufsize(s
);
1204 size_t bits_per_buf
= bitmap_bufsize
* CHAR_BIT
;
1206 /* dump_bitmap_buf is used to store dump_bitmap temporarily */
1207 dump_bitmap_buf
= g_malloc0(bitmap_bufsize
);
1213 * exam memory page by page, and set the bit in dump_bitmap corresponded
1214 * to the existing page.
1216 while (get_next_page(&block_iter
, &pfn
, NULL
, s
)) {
1217 ret
= set_dump_bitmap(last_pfn
, pfn
, true, dump_bitmap_buf
, s
);
1219 error_setg(errp
, "dump: failed to set dump_bitmap");
1228 * set_dump_bitmap will always leave the recently set bit un-sync. Here we
1229 * set the remaining bits from last_pfn to the end of the bitmap buffer to
1230 * 0. With those set, the un-sync bit will be synchronized into the vmcore.
1232 if (num_dumpable
> 0) {
1233 ret
= set_dump_bitmap(last_pfn
, last_pfn
+ bits_per_buf
, false,
1234 dump_bitmap_buf
, s
);
1236 error_setg(errp
, "dump: failed to sync dump_bitmap");
1241 /* number of dumpable pages that will be dumped later */
1242 s
->num_dumpable
= num_dumpable
;
1245 g_free(dump_bitmap_buf
);
1248 static void prepare_data_cache(DataCache
*data_cache
, DumpState
*s
,
1251 data_cache
->fd
= s
->fd
;
1252 data_cache
->data_size
= 0;
1253 data_cache
->buf_size
= 4 * dump_bitmap_get_bufsize(s
);
1254 data_cache
->buf
= g_malloc0(data_cache
->buf_size
);
1255 data_cache
->offset
= offset
;
1258 static int write_cache(DataCache
*dc
, const void *buf
, size_t size
,
1262 * dc->buf_size should not be less than size, otherwise dc will never be
1265 assert(size
<= dc
->buf_size
);
1268 * if flag_sync is set, synchronize data in dc->buf into vmcore.
1269 * otherwise check if the space is enough for caching data in buf, if not,
1270 * write the data in dc->buf to dc->fd and reset dc->buf
1272 if ((!flag_sync
&& dc
->data_size
+ size
> dc
->buf_size
) ||
1273 (flag_sync
&& dc
->data_size
> 0)) {
1274 if (write_buffer(dc
->fd
, dc
->offset
, dc
->buf
, dc
->data_size
) < 0) {
1278 dc
->offset
+= dc
->data_size
;
1283 memcpy(dc
->buf
+ dc
->data_size
, buf
, size
);
1284 dc
->data_size
+= size
;
1290 static void free_data_cache(DataCache
*data_cache
)
1292 g_free(data_cache
->buf
);
1295 static size_t get_len_buf_out(size_t page_size
, uint32_t flag_compress
)
1297 switch (flag_compress
) {
1298 case DUMP_DH_COMPRESSED_ZLIB
:
1299 return compressBound(page_size
);
1301 case DUMP_DH_COMPRESSED_LZO
:
1303 * LZO will expand incompressible data by a little amount. Please check
1304 * the following URL to see the expansion calculation:
1305 * http://www.oberhumer.com/opensource/lzo/lzofaq.php
1307 return page_size
+ page_size
/ 16 + 64 + 3;
1309 #ifdef CONFIG_SNAPPY
1310 case DUMP_DH_COMPRESSED_SNAPPY
:
1311 return snappy_max_compressed_length(page_size
);
1317 static void write_dump_pages(DumpState
*s
, Error
**errp
)
1320 DataCache page_desc
, page_data
;
1321 size_t len_buf_out
, size_out
;
1323 lzo_bytep wrkmem
= NULL
;
1325 uint8_t *buf_out
= NULL
;
1326 off_t offset_desc
, offset_data
;
1327 PageDescriptor pd
, pd_zero
;
1329 GuestPhysBlock
*block_iter
= NULL
;
1331 g_autofree
uint8_t *page
= NULL
;
1333 /* get offset of page_desc and page_data in dump file */
1334 offset_desc
= s
->offset_page
;
1335 offset_data
= offset_desc
+ sizeof(PageDescriptor
) * s
->num_dumpable
;
1337 prepare_data_cache(&page_desc
, s
, offset_desc
);
1338 prepare_data_cache(&page_data
, s
, offset_data
);
1340 /* prepare buffer to store compressed data */
1341 len_buf_out
= get_len_buf_out(s
->dump_info
.page_size
, s
->flag_compress
);
1342 assert(len_buf_out
!= 0);
1345 wrkmem
= g_malloc(LZO1X_1_MEM_COMPRESS
);
1348 buf_out
= g_malloc(len_buf_out
);
1351 * init zero page's page_desc and page_data, because every zero page
1352 * uses the same page_data
1354 pd_zero
.size
= cpu_to_dump32(s
, s
->dump_info
.page_size
);
1355 pd_zero
.flags
= cpu_to_dump32(s
, 0);
1356 pd_zero
.offset
= cpu_to_dump64(s
, offset_data
);
1357 pd_zero
.page_flags
= cpu_to_dump64(s
, 0);
1358 buf
= g_malloc0(s
->dump_info
.page_size
);
1359 ret
= write_cache(&page_data
, buf
, s
->dump_info
.page_size
, false);
1362 error_setg(errp
, "dump: failed to write page data (zero page)");
1366 offset_data
+= s
->dump_info
.page_size
;
1367 page
= g_malloc(s
->dump_info
.page_size
);
1370 * dump memory to vmcore page by page. zero page will all be resided in the
1371 * first page of page section
1373 for (buf
= page
; get_next_page(&block_iter
, &pfn_iter
, &buf
, s
); buf
= page
) {
1374 /* check zero page */
1375 if (buffer_is_zero(buf
, s
->dump_info
.page_size
)) {
1376 ret
= write_cache(&page_desc
, &pd_zero
, sizeof(PageDescriptor
),
1379 error_setg(errp
, "dump: failed to write page desc");
1384 * not zero page, then:
1385 * 1. compress the page
1386 * 2. write the compressed page into the cache of page_data
1387 * 3. get page desc of the compressed page and write it into the
1388 * cache of page_desc
1390 * only one compression format will be used here, for
1391 * s->flag_compress is set. But when compression fails to work,
1392 * we fall back to save in plaintext.
1394 size_out
= len_buf_out
;
1395 if ((s
->flag_compress
& DUMP_DH_COMPRESSED_ZLIB
) &&
1396 (compress2(buf_out
, (uLongf
*)&size_out
, buf
,
1397 s
->dump_info
.page_size
, Z_BEST_SPEED
) == Z_OK
) &&
1398 (size_out
< s
->dump_info
.page_size
)) {
1399 pd
.flags
= cpu_to_dump32(s
, DUMP_DH_COMPRESSED_ZLIB
);
1400 pd
.size
= cpu_to_dump32(s
, size_out
);
1402 ret
= write_cache(&page_data
, buf_out
, size_out
, false);
1404 error_setg(errp
, "dump: failed to write page data");
1408 } else if ((s
->flag_compress
& DUMP_DH_COMPRESSED_LZO
) &&
1409 (lzo1x_1_compress(buf
, s
->dump_info
.page_size
, buf_out
,
1410 (lzo_uint
*)&size_out
, wrkmem
) == LZO_E_OK
) &&
1411 (size_out
< s
->dump_info
.page_size
)) {
1412 pd
.flags
= cpu_to_dump32(s
, DUMP_DH_COMPRESSED_LZO
);
1413 pd
.size
= cpu_to_dump32(s
, size_out
);
1415 ret
= write_cache(&page_data
, buf_out
, size_out
, false);
1417 error_setg(errp
, "dump: failed to write page data");
1421 #ifdef CONFIG_SNAPPY
1422 } else if ((s
->flag_compress
& DUMP_DH_COMPRESSED_SNAPPY
) &&
1423 (snappy_compress((char *)buf
, s
->dump_info
.page_size
,
1424 (char *)buf_out
, &size_out
) == SNAPPY_OK
) &&
1425 (size_out
< s
->dump_info
.page_size
)) {
1426 pd
.flags
= cpu_to_dump32(s
, DUMP_DH_COMPRESSED_SNAPPY
);
1427 pd
.size
= cpu_to_dump32(s
, size_out
);
1429 ret
= write_cache(&page_data
, buf_out
, size_out
, false);
1431 error_setg(errp
, "dump: failed to write page data");
1437 * fall back to save in plaintext, size_out should be
1438 * assigned the target's page size
1440 pd
.flags
= cpu_to_dump32(s
, 0);
1441 size_out
= s
->dump_info
.page_size
;
1442 pd
.size
= cpu_to_dump32(s
, size_out
);
1444 ret
= write_cache(&page_data
, buf
,
1445 s
->dump_info
.page_size
, false);
1447 error_setg(errp
, "dump: failed to write page data");
1452 /* get and write page desc here */
1453 pd
.page_flags
= cpu_to_dump64(s
, 0);
1454 pd
.offset
= cpu_to_dump64(s
, offset_data
);
1455 offset_data
+= size_out
;
1457 ret
= write_cache(&page_desc
, &pd
, sizeof(PageDescriptor
), false);
1459 error_setg(errp
, "dump: failed to write page desc");
1463 s
->written_size
+= s
->dump_info
.page_size
;
1466 ret
= write_cache(&page_desc
, NULL
, 0, true);
1468 error_setg(errp
, "dump: failed to sync cache for page_desc");
1471 ret
= write_cache(&page_data
, NULL
, 0, true);
1473 error_setg(errp
, "dump: failed to sync cache for page_data");
1478 free_data_cache(&page_desc
);
1479 free_data_cache(&page_data
);
1488 static void create_kdump_vmcore(DumpState
*s
, Error
**errp
)
1494 * the kdump-compressed format is:
1496 * +------------------------------------------+ 0x0
1497 * | main header (struct disk_dump_header) |
1498 * |------------------------------------------+ block 1
1499 * | sub header (struct kdump_sub_header) |
1500 * |------------------------------------------+ block 2
1501 * | 1st-dump_bitmap |
1502 * |------------------------------------------+ block 2 + X blocks
1503 * | 2nd-dump_bitmap | (aligned by block)
1504 * |------------------------------------------+ block 2 + 2 * X blocks
1505 * | page desc for pfn 0 (struct page_desc) | (aligned by block)
1506 * | page desc for pfn 1 (struct page_desc) |
1508 * |------------------------------------------| (not aligned by block)
1509 * | page data (pfn 0) |
1510 * | page data (pfn 1) |
1512 * +------------------------------------------+
1515 ret
= write_start_flat_header(s
->fd
);
1517 error_setg(errp
, "dump: failed to write start flat header");
1521 write_dump_header(s
, errp
);
1526 write_dump_bitmap(s
, errp
);
1531 write_dump_pages(s
, errp
);
1536 ret
= write_end_flat_header(s
->fd
);
1538 error_setg(errp
, "dump: failed to write end flat header");
1543 static int validate_start_block(DumpState
*s
)
1545 GuestPhysBlock
*block
;
1547 if (!dump_has_filter(s
)) {
1551 QTAILQ_FOREACH(block
, &s
->guest_phys_blocks
.head
, next
) {
1552 /* This block is out of the range */
1553 if (block
->target_start
>= s
->filter_area_begin
+ s
->filter_area_length
||
1554 block
->target_end
<= s
->filter_area_begin
) {
1563 static void get_max_mapnr(DumpState
*s
)
1565 GuestPhysBlock
*last_block
;
1567 last_block
= QTAILQ_LAST(&s
->guest_phys_blocks
.head
);
1568 s
->max_mapnr
= dump_paddr_to_pfn(s
, last_block
->target_end
);
1571 static DumpState dump_state_global
= { .status
= DUMP_STATUS_NONE
};
1573 static void dump_state_prepare(DumpState
*s
)
1575 /* zero the struct, setting status to active */
1576 *s
= (DumpState
) { .status
= DUMP_STATUS_ACTIVE
};
1579 bool qemu_system_dump_in_progress(void)
1581 DumpState
*state
= &dump_state_global
;
1582 return (qatomic_read(&state
->status
) == DUMP_STATUS_ACTIVE
);
1586 * calculate total size of memory to be dumped (taking filter into
1589 static int64_t dump_calculate_size(DumpState
*s
)
1591 GuestPhysBlock
*block
;
1594 QTAILQ_FOREACH(block
, &s
->guest_phys_blocks
.head
, next
) {
1595 total
+= dump_filtered_memblock_size(block
,
1596 s
->filter_area_begin
,
1597 s
->filter_area_length
);
1603 static void vmcoreinfo_update_phys_base(DumpState
*s
)
1605 uint64_t size
, note_head_size
, name_size
, phys_base
;
1610 if (!note_name_equal(s
, s
->guest_note
, "VMCOREINFO")) {
1614 get_note_sizes(s
, s
->guest_note
, ¬e_head_size
, &name_size
, &size
);
1615 note_head_size
= ROUND_UP(note_head_size
, 4);
1617 vmci
= s
->guest_note
+ note_head_size
+ ROUND_UP(name_size
, 4);
1618 *(vmci
+ size
) = '\0';
1620 lines
= g_strsplit((char *)vmci
, "\n", -1);
1621 for (i
= 0; lines
[i
]; i
++) {
1622 const char *prefix
= NULL
;
1624 if (s
->dump_info
.d_machine
== EM_X86_64
) {
1625 prefix
= "NUMBER(phys_base)=";
1626 } else if (s
->dump_info
.d_machine
== EM_AARCH64
) {
1627 prefix
= "NUMBER(PHYS_OFFSET)=";
1630 if (prefix
&& g_str_has_prefix(lines
[i
], prefix
)) {
1631 if (qemu_strtou64(lines
[i
] + strlen(prefix
), NULL
, 16,
1633 warn_report("Failed to read %s", prefix
);
1635 s
->dump_info
.phys_base
= phys_base
;
1644 static void dump_init(DumpState
*s
, int fd
, bool has_format
,
1645 DumpGuestMemoryFormat format
, bool paging
, bool has_filter
,
1646 int64_t begin
, int64_t length
, Error
**errp
)
1649 VMCoreInfoState
*vmci
= vmcoreinfo_find();
1654 s
->has_format
= has_format
;
1656 s
->written_size
= 0;
1658 /* kdump-compressed is conflict with paging and filter */
1659 if (has_format
&& format
!= DUMP_GUEST_MEMORY_FORMAT_ELF
) {
1660 assert(!paging
&& !has_filter
);
1663 if (runstate_is_running()) {
1664 vm_stop(RUN_STATE_SAVE_VM
);
1670 /* If we use KVM, we should synchronize the registers before we get dump
1671 * info or physmap info.
1673 cpu_synchronize_all_states();
1680 if (has_filter
&& !length
) {
1681 error_setg(errp
, QERR_INVALID_PARAMETER
, "length");
1684 s
->filter_area_begin
= begin
;
1685 s
->filter_area_length
= length
;
1687 memory_mapping_list_init(&s
->list
);
1689 guest_phys_blocks_init(&s
->guest_phys_blocks
);
1690 guest_phys_blocks_append(&s
->guest_phys_blocks
);
1691 s
->total_size
= dump_calculate_size(s
);
1692 #ifdef DEBUG_DUMP_GUEST_MEMORY
1693 fprintf(stderr
, "DUMP: total memory to dump: %lu\n", s
->total_size
);
1696 /* it does not make sense to dump non-existent memory */
1697 if (!s
->total_size
) {
1698 error_setg(errp
, "dump: no guest memory to dump");
1702 /* Is the filter filtering everything? */
1703 if (validate_start_block(s
) == -1) {
1704 error_setg(errp
, QERR_INVALID_PARAMETER
, "begin");
1708 /* get dump info: endian, class and architecture.
1709 * If the target architecture is not supported, cpu_get_dump_info() will
1712 ret
= cpu_get_dump_info(&s
->dump_info
, &s
->guest_phys_blocks
);
1714 error_setg(errp
, QERR_UNSUPPORTED
);
1718 if (!s
->dump_info
.page_size
) {
1719 s
->dump_info
.page_size
= TARGET_PAGE_SIZE
;
1722 s
->note_size
= cpu_get_note_size(s
->dump_info
.d_class
,
1723 s
->dump_info
.d_machine
, nr_cpus
);
1724 if (s
->note_size
< 0) {
1725 error_setg(errp
, QERR_UNSUPPORTED
);
1730 * The goal of this block is to (a) update the previously guessed
1731 * phys_base, (b) copy the guest note out of the guest.
1732 * Failure to do so is not fatal for dumping.
1735 uint64_t addr
, note_head_size
, name_size
, desc_size
;
1739 note_head_size
= dump_is_64bit(s
) ?
1740 sizeof(Elf64_Nhdr
) : sizeof(Elf32_Nhdr
);
1742 format
= le16_to_cpu(vmci
->vmcoreinfo
.guest_format
);
1743 size
= le32_to_cpu(vmci
->vmcoreinfo
.size
);
1744 addr
= le64_to_cpu(vmci
->vmcoreinfo
.paddr
);
1745 if (!vmci
->has_vmcoreinfo
) {
1746 warn_report("guest note is not present");
1747 } else if (size
< note_head_size
|| size
> MAX_GUEST_NOTE_SIZE
) {
1748 warn_report("guest note size is invalid: %" PRIu32
, size
);
1749 } else if (format
!= FW_CFG_VMCOREINFO_FORMAT_ELF
) {
1750 warn_report("guest note format is unsupported: %" PRIu16
, format
);
1752 s
->guest_note
= g_malloc(size
+ 1); /* +1 for adding \0 */
1753 cpu_physical_memory_read(addr
, s
->guest_note
, size
);
1755 get_note_sizes(s
, s
->guest_note
, NULL
, &name_size
, &desc_size
);
1756 s
->guest_note_size
= ELF_NOTE_SIZE(note_head_size
, name_size
,
1758 if (name_size
> MAX_GUEST_NOTE_SIZE
||
1759 desc_size
> MAX_GUEST_NOTE_SIZE
||
1760 s
->guest_note_size
> size
) {
1761 warn_report("Invalid guest note header");
1762 g_free(s
->guest_note
);
1763 s
->guest_note
= NULL
;
1765 vmcoreinfo_update_phys_base(s
);
1766 s
->note_size
+= s
->guest_note_size
;
1771 /* get memory mapping */
1773 qemu_get_guest_memory_mapping(&s
->list
, &s
->guest_phys_blocks
, errp
);
1778 qemu_get_guest_simple_memory_mapping(&s
->list
, &s
->guest_phys_blocks
);
1781 s
->nr_cpus
= nr_cpus
;
1786 tmp
= DIV_ROUND_UP(DIV_ROUND_UP(s
->max_mapnr
, CHAR_BIT
),
1787 s
->dump_info
.page_size
);
1788 s
->len_dump_bitmap
= tmp
* s
->dump_info
.page_size
;
1790 /* init for kdump-compressed format */
1791 if (has_format
&& format
!= DUMP_GUEST_MEMORY_FORMAT_ELF
) {
1793 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB
:
1794 s
->flag_compress
= DUMP_DH_COMPRESSED_ZLIB
;
1797 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO
:
1799 if (lzo_init() != LZO_E_OK
) {
1800 error_setg(errp
, "failed to initialize the LZO library");
1804 s
->flag_compress
= DUMP_DH_COMPRESSED_LZO
;
1807 case DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY
:
1808 s
->flag_compress
= DUMP_DH_COMPRESSED_SNAPPY
;
1812 s
->flag_compress
= 0;
1818 if (dump_has_filter(s
)) {
1819 memory_mapping_filter(&s
->list
, s
->filter_area_begin
, s
->filter_area_length
);
1823 * calculate phdr_num
1825 * the type of ehdr->e_phnum is uint16_t, so we should avoid overflow
1827 s
->phdr_num
= 1; /* PT_NOTE */
1828 if (s
->list
.num
< UINT16_MAX
- 2) {
1830 s
->phdr_num
+= s
->list
.num
;
1832 /* sh_info of section 0 holds the real number of phdrs */
1835 /* the type of shdr->sh_info is uint32_t, so we should avoid overflow */
1836 if (s
->list
.num
<= UINT32_MAX
- 1) {
1837 s
->phdr_num
+= s
->list
.num
;
1839 s
->phdr_num
= UINT32_MAX
;
1843 if (dump_is_64bit(s
)) {
1844 s
->phdr_offset
= sizeof(Elf64_Ehdr
);
1845 s
->shdr_offset
= s
->phdr_offset
+ sizeof(Elf64_Phdr
) * s
->phdr_num
;
1846 s
->note_offset
= s
->shdr_offset
+ sizeof(Elf64_Shdr
) * s
->shdr_num
;
1847 s
->memory_offset
= s
->note_offset
+ s
->note_size
;
1850 s
->phdr_offset
= sizeof(Elf32_Ehdr
);
1851 s
->shdr_offset
= s
->phdr_offset
+ sizeof(Elf32_Phdr
) * s
->phdr_num
;
1852 s
->note_offset
= s
->shdr_offset
+ sizeof(Elf32_Shdr
) * s
->shdr_num
;
1853 s
->memory_offset
= s
->note_offset
+ s
->note_size
;
1862 /* this operation might be time consuming. */
1863 static void dump_process(DumpState
*s
, Error
**errp
)
1866 DumpQueryResult
*result
= NULL
;
1868 if (s
->has_format
&& s
->format
== DUMP_GUEST_MEMORY_FORMAT_WIN_DMP
) {
1869 #ifdef TARGET_X86_64
1870 create_win_dump(s
, errp
);
1872 } else if (s
->has_format
&& s
->format
!= DUMP_GUEST_MEMORY_FORMAT_ELF
) {
1873 create_kdump_vmcore(s
, errp
);
1875 create_vmcore(s
, errp
);
1878 /* make sure status is written after written_size updates */
1880 qatomic_set(&s
->status
,
1881 (*errp
? DUMP_STATUS_FAILED
: DUMP_STATUS_COMPLETED
));
1883 /* send DUMP_COMPLETED message (unconditionally) */
1884 result
= qmp_query_dump(NULL
);
1885 /* should never fail */
1887 qapi_event_send_dump_completed(result
, !!*errp
, (*errp
?
1888 error_get_pretty(*errp
) : NULL
));
1889 qapi_free_DumpQueryResult(result
);
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
)
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
, errp
);
2024 qatomic_set(&s
->status
, DUMP_STATUS_FAILED
);
2031 qemu_thread_create(&s
->dump_thread
, "dump_thread", dump_thread
,
2032 s
, QEMU_THREAD_DETACHED
);
2035 dump_process(s
, errp
);
2039 DumpGuestMemoryCapability
*qmp_query_dump_guest_memory_capability(Error
**errp
)
2041 DumpGuestMemoryCapability
*cap
=
2042 g_new0(DumpGuestMemoryCapability
, 1);
2043 DumpGuestMemoryFormatList
**tail
= &cap
->formats
;
2045 /* elf is always available */
2046 QAPI_LIST_APPEND(tail
, DUMP_GUEST_MEMORY_FORMAT_ELF
);
2048 /* kdump-zlib is always available */
2049 QAPI_LIST_APPEND(tail
, DUMP_GUEST_MEMORY_FORMAT_KDUMP_ZLIB
);
2051 /* add new item if kdump-lzo is available */
2053 QAPI_LIST_APPEND(tail
, DUMP_GUEST_MEMORY_FORMAT_KDUMP_LZO
);
2056 /* add new item if kdump-snappy is available */
2057 #ifdef CONFIG_SNAPPY
2058 QAPI_LIST_APPEND(tail
, DUMP_GUEST_MEMORY_FORMAT_KDUMP_SNAPPY
);
2061 /* Windows dump is available only if target is x86_64 */
2062 #ifdef TARGET_X86_64
2063 QAPI_LIST_APPEND(tail
, DUMP_GUEST_MEMORY_FORMAT_WIN_DMP
);