Galgame汉化中的逆向(八)哈希算法分析以krkrz_hxv4为例

这几年ai迅速发展,使得逆向分析门槛大幅下降。若依赖ai反而会觉得缺少乐趣,亦或是和ai扯皮半天,ai还睁眼说瞎话,你纠正错误到面红耳赤,它反手给你甩个“you reached rate limit”强行结束。逆向作为茶余饭后之娱乐活动,就和我们喜欢手动挡一样,完全由自己掌控的心流令人欲罢不能。本文将不依赖ai, 依旧以传统的逆向方法和技巧来呈现。

时隔三年再次发帖,提前祝大家新年快乐~

by devseed, 本贴论坛和我的博客同时发布。

0x0 background

近些年,wamsoft魔改的krkrz引入了hxv4解密方案,最大的区别是封包只存储文件哈希值,不存储文件名。游戏脚本(通常为*.scn)内资源文件以原始文件名存储,引擎运行时计算得到哈希值,从而找到封包内对应文件。由于哈希函数不可逆,这使得要想得到文件名变得非常麻烦(要么你得跑一边游戏所有分支剧情dump,或者干脆不要文件名了)。

目前主要有两种方案,运行时dump(krkrdump)、扫描对应的脚本构建字符串碰撞从而得到哈希值和文件名的映射(KrkrExtractForCxdecV2+krkr_hxv4_dumphash)。本文将以dc5ph为例分析hxv4的哈希函数,以及如何还原对应的算法和数据结构。

0x1 krkrz、hxv4

直接分析hxv4则是非常困难的,可以通过原版krkrz了解引擎大概加载流程,再针对性的进行寻找,原版Stream如下:

1
tTVPPlugin -> TVPCreateIStream -> _TVPCreateStream ->  tTVPArchive::CreateStream -> TVPStorageMediaManager.Open -> tTVPXP3Archive::CreateStreamByIndex -> Read

关于hxv4,可以参考hxcryptHxv4entry内容是加密的,先解密这个entry,之后得到filter key在用旧版cxdec方法解密各个文件entry。如下:

1
2
3
4
5
// decrypt hxv4 index
Xp3Stream::TryOpen -> HxCrypt::ReadIndex -> HxChachaDecryptor::Decrypt

// decrypt file content
HxFilter::Decrypt -> HxFilterSpan::DecryptHeader

解密相关参数示例如下,获取方式可以用我写的firda脚本krkr_hxv4_dumpkeydc5ph_hxv4_dumpkey

1
2
3
4
5
6
7
8
9
10
11
12
13
control_block.bin // 4096 bytes
hxpoint at 0x5b18f0c3
cxpoint at 0x5b183c6d
* key : b338a06fc12ba33610e7e4428c8389ca0342b418ae6a77e5287e3607e41fe65b
* nonce : ec668fc7eff5f388612eb56f1e6d4d6f
* filterkey : 4eef61df5f2e1771
* mask : 0x273
* offset : 0x178
* randtype : 1
* order : 04 00 02 03 06 01 07 05 04 05 00 01 03 02 00 02 01
* PrologOrder (garbro) : 0, 2, 1
* OddBranchOrder (garbro) : 3, 4, 0, 1, 2, 5
* EvenBranchOrder (garbro) : 2, 6, 3, 1, 0, 4, 5, 7

0x2 program flow

分析的切入点结合krkrz源码是虚函数的RTTI,找到关键函数v2linktTVPXP3ArchiveStream。找到此函数,即可调用read函数将其文件动态dump出来。

1
2
3
4
5
6
7
8
9
10
11
.rdata:00728520 ; class tTVPXP3ArchiveStream: TJS::tTJSBinaryStream;   (#classinformer)
.rdata:00728520 dd offset ??_R4tTVPXP3ArchiveStream@@6B@ ; const tTVPXP3ArchiveStream::`RTTI Complete Object Locator'
.rdata:00728524 ; const tTVPXP3ArchiveStream::`vftable'
.rdata:00728524 ??_7tTVPXP3ArchiveStream@@6B@ dd offset tTVPXP3ArchiveStream__Seek_437230
.rdata:00728524 ; DATA XREF: sub_436D90+41↑o
.rdata:00728524 ; sub_436E90+2A↑o
.rdata:00728528 dd offset tTVPXP3ArchiveStream__Read_4372E0
.rdata:0072852C dd offset sub_402CD0
.rdata:00728530 dd offset sub_4768F0
.rdata:00728534 dd offset tTVPXP3ArchiveStream__GetSize_437480
.rdata:00728538 dd offset tTVPXP3ArchiveStream__deconstruct_436E60

如果没有RTTI,则可以通过函数特征码定位(一般来说编译器变化不大的情况下,生成对应函数的代码差不多,所以可以自己编译一下,看看对应函数代码什么样)。TVPCreateStream函数和对应的代码如下。这个函数找到后继续跟TVPStorageMediaManager::Open_40CFD0函数顺藤摸瓜找到hxv4相关函数。不同于传统的krkrz filter解密函数,hxv4通过StorageMediaManager对stream接管很早。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
.text:0040EDB0                   ; =============== S U B R O U T I N E =======================================
.text:0040EDB0
.text:0040EDB0 ; signature: 55 8b ec 6a ff 68 ? ? ? ? 64 a1 ? ? ? ? 50 83 ec 5c 53 56 57 a1 ? ? ? ? 33 c5 50 8d 45 f4 64 a3 ? ? ? ? 89 65 f0 89 4d ec c7 45 ? ? ? ? ? e8 ? ? ? ? 8b 4d f4 64 89 0d ? ? ? ? 59 5f 5e 5b 8b e5 5d c3
.text:0040EDB0 ; void *__fastcall TVPCreateStream_40EDB0(void *name, uint32_t flags)
.text:0040EDB0 _TVPCreateStream_40EDB0 proc near ; CODE XREF: TVPCreateStream_40F040+35↓p
.text:0040EDB0
.text:0040EDB0 ; __unwind { // SEH_40EDB0
.text:0040EDB0 55 push ebp
.text:0040EDB1 8B EC mov ebp, esp
.text:0040EDB3 6A FF push 0FFFFFFFFh
.text:0040EDB5 68 C8 9F 69 00 push offset SEH_40EDB0
.text:0040EDBA 64 A1 00 00 00 00 mov eax, large fs:0
.text:0040EDC0 50 push eax
.text:0040EDC1 83 EC 24 sub esp, 24h
.text:0040EDC4 53 push ebx
.text:0040EDC5 56 push esi
.text:0040EDC6 57 push edi
.text:0040EDC7 A1 50 F9 76 00 mov eax, ___security_cookie
.text:0040EDCC 33 C5 xor eax, ebp
.text:0040EDCE 50 push eax
.text:0040EDCF 8D 45 F4 lea eax, [ebp+var_C]
.text:0040EDD2 64 A3 00 00 00 00 mov large fs:0, eax
.text:0040EDD8 89 65 F0 mov [ebp+var_10], esp
.text:0040EDDB 8B DA mov ebx, edx
.text:0040EDDD 89 5D DC mov [ebp+flags_alter1], ebx
.text:0040EDE0 8B F9 mov edi, ecx
.text:0040EDE2 C7 45 D4 84 64 78 mov [ebp+var_2C], offset stru_786484
.text:0040EDE2 00
.text:0040EDE9 68 84 64 78 00 push offset stru_786484 ; lpCriticalSection
.text:0040EDEE FF 15 24 03 6C 00 call ds:EnterCriticalSection
.text:0040EDF4 ; try {
.text:0040EDF4 C7 45 FC 00 00 00 mov [ebp+var_4], 0
.text:0040EDF4 00
.text:0040EDFB C7 45 EC 00 00 00 mov [ebp+name_alter1], 0

调试后可知hxv4的dll藏在exe资源文件中,去hookLoadlibraryW后可知他会在C盘生成类似于krkr_xxx/yyy.dll,hxv4相关的文件解密还有哈希函数都在里面,切入点如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
.rdata:1008199C                   ; struct struct DefaultCompoundHasher<PathNameHashTrait>: struct CompoundStringHasher;   (#classinformer)
.rdata:1008199C F4 6B 09 10 dd offset ??_R4?$DefaultCompoundHasher@UPathNameHashTrait@@@@6B@ ; const DefaultCompoundHasher<PathNameHashTrait>::`RTTI Complete Object Locator'
.rdata:100819A0 ; const DefaultCompoundHasher<struct PathNameHashTrait>::`vftable'
.rdata:100819A0 C0 67 01 10 ??_7?$DefaultCompoundHasher@UPathNameHashTrait@@@@6B@ dd offset au_re_j__free_0_12
.rdata:100819A0 ; DATA XREF: sub_10016680+2B↑o
.rdata:100819A0 ; deconstructor
.rdata:100819A4 F0 69 01 10 dd offset DirHashCompute_100169F0
.rdata:100819A8
.rdata:100819A8 ; struct struct DefaultCompoundHasher<FileNameHashTrait>: struct CompoundStringHasher; (#classinformer)
.rdata:100819A8 40 6C 09 10 dd offset ??_R4?$DefaultCompoundHasher@UFileNameHashTrait@@@@6B@ ; const DefaultCompoundHasher<FileNameHashTrait>::`RTTI Complete Object Locator'
.rdata:100819AC ; const DefaultCompoundHasher<struct FileNameHashTrait>::`vftable'
.rdata:100819AC 80 67 01 10 ??_7?$DefaultCompoundHasher@UFileNameHashTrait@@@@6B@ dd offset au_re_j__free_0_11
.rdata:100819AC ; DATA XREF: sub_10016580+2B↑o
.rdata:100819B0 00 69 01 10 dd offset FileHashCompute_10016900

00000000 struct IStringHasher_VptrTable // sizeof=0x8
00000000 {
00000000 void *Destruct;
00000004 void *Calculate;
00000008 };

00000000 struct IStringHasher // sizeof=0xC
00000000 {
00000000 IStringHasher_VptrTable *VptrTable;
00000004 uint8_t *salt;
00000008 int saltsize;
0000000C };

从切入点顺藤摸瓜,可以归纳出下列数据结构和行为逻辑:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65

typedef tjs_int(__fastcall *FuncHxv4CalcHash)(Hxv4CompoundHasher* _this, void* _edx,
OUT tTJSVariant* hash, const tTJSString* str, const tTJSString* seed);

typedef struct Hxv4CompoundHasher
{
struct
{
void* destruct;
FuncHxv4CalcHash calc;
} *vftable; // offset 0
tjs_uint8* salt; // offset 0x4
tjs_int saltsize; // offset 0x8
} Hxv4CompoundHasher;

typedef struct Hxv4DirHasher
{
Hxv4CompoundHasher base;
tjs_uint8 saltdata[0x10];
} Hxv4DirHasher;

typedef struct Hxv4FileHasher
{
Hxv4CompoundHasher base;
tjs_uint8 saltdata[0x20];
} Hxv4FileHasher;

typedef struct Hxv4CompoundStorageMedia
{
void* vftable;
int nref;
uint32_t reserve1;
tTJSString prefix;
tTJSString seed; //offset 0x10
CRITICAL_SECTION critical_section;
uint8_t reserve2[0x20];
tTJSString* start;
tTJSString* pos;
tTJSString* end;
Hxv4DirHasher* dirhasher; // offset 0x58
Hxv4FileHasher* filehasher;
} Hxv4CompoundStorageMedia;

// hook from here
unsigned int __cdecl CreateCompoundStorageMedia_100059D0(
CompoundStorageMedia **retTVPStorageMedia,
int tjsVarPrefix,
int argc,
int *argv)
{
...
if ( argc > 1 ) CompoundStorageMedia::Init_1000A3D0(*retTVPStorageMedia, 0, *argv, argv[1]);
...
*retTVPStorageMedia = v10;
TVPRegisterStorageMedia_100068C0(v10); // in exe it will tTVPStorageMediaManager::Open
...
}

int __thiscall CompoundStorageMedia::Init_1000A3D0(CompoundStorageMedia *this, int a2, void *seed_variant, size_t Size)
{
...
this->PathNameHasher = (IStringHasher *)PathNameHasher::Init_10016890(Size);
this->FileNameHasher = (IStringHasher *)FileHasher::Init_10016820(Size);
...
}

至此我们已经找到了哈希函数了,可以动态附加到游戏程序里,直接调用Hxv4CompoundHasher::vftable->calc来计算任意字符串,详见krkr_hxv4_dumphash

dc5ph_hxv4_dumphash

0x3 hash function

动态dump hash后,更进一步,我们要怎么分析算法逻辑,并且能够静态复现呢?最笨的方法是直接把相关逻辑的C伪代码或者汇编代码搬出来,逐个模拟实现(汇编可用unicorn模拟)。但是这种方法费时费力,一个哈希函数动辄上千行,还特别容易出错。所以去年分析到动态调用这一步就没再继续,最近看了看发现可以从特征进行分析,从而得以继续。

站在开发者的角度想,大部分游戏不会自己研制一套全新的哈希算法,大多数是用现有的方法,或者在现有的方法上改改参数或流程。因此我们还原算法的主要目标,是寻找当前算法是哪个原有算法的改版。那么哈希算法如何进行呢?通常是下面几个步骤:

init(key, salt) -> update(buf, lastvalue) -> final(outsize)

还原哈希算法,函数的输入输出还有函数内相关常数需要重点关注。本游戏有两种哈希算法,计算文件名的hash、计算文件夹的hash。

file hash

计算文件哈希算法如下,tTJSString使得此函数变得很乱, 移除掉后可以很清晰的看到计算哈希的流程。输入为文本unicode编码,输出为32字节。之前动态调用计算得到一组值为!scnlist.txt,C1F625E3A4BB508E082A52A8B032F4B3D2F34FF7FB3A30502574717DE6579126

filehash_init_1000E070 -> filehash_update_100159F0 -> filehash_final32_10016B00

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
int __userpurge FileHashCompute_10016900@<eax>(
IStringHasher *this@<ecx>,
void *hashValueRet,
tTJSString_S *rawstr,
tTJSString_S *seed)
{
size_t (__stdcall *v4)(tTJSString_S *); // eax
int (__stdcall *v5)(tTJSString_S *); // eax
uint8_t *raw_cstr; // eax
int (__stdcall *v7)(tTJSString_S *); // eax
int seed_len; // edi
int (__stdcall *v9)(tTJSString_S *); // eax
uint8_t *seed_cstr; // eax
size_t v12; // [esp-8h] [ebp-94h]
filehash_ctx ctx; // [esp+Ch] [ebp-80h] BYREF

filehash_init_1000E070(&ctx, 0x20u, this->salt, this->saltsize);
v4 = (size_t (__stdcall *)(tTJSString_S *))tTJSString::length_100AD158;
if ( !tTJSString::length_100AD158 )
{
v4 = (size_t (__stdcall *)(tTJSString_S *))findfunc_10016420((void (__stdcall *)(_DWORD))aTjsIntTtjsstri);
tTJSString::length_100AD158 = (int)v4;
}
v4(rawstr);
v5 = (int (__stdcall *)(tTJSString_S *))TJSString::c_str_100AD0F4;
if ( !TJSString::c_str_100AD0F4 )
{
v5 = (int (__stdcall *)(tTJSString_S *))findfunc_10016420((void (__stdcall *)(_DWORD))aConstTjsCharTt);
TJSString::c_str_100AD0F4 = (int)v5;
}
raw_cstr = (uint8_t *)v5(rawstr);
filehash_update_100159F0(&ctx, raw_cstr, v12); // v12=2*rawstr_len
if ( seed )
{
v7 = (int (__stdcall *)(tTJSString_S *))tTJSString::length_100AD158;
if ( !tTJSString::length_100AD158 )
{
v7 = (int (__stdcall *)(tTJSString_S *))findfunc_10016420((void (__stdcall *)(_DWORD))aTjsIntTtjsstri);
tTJSString::length_100AD158 = (int)v7;
}
seed_len = v7(seed);
v9 = (int (__stdcall *)(tTJSString_S *))TJSString::c_str_100AD0F4;
if ( !TJSString::c_str_100AD0F4 )
{
v9 = (int (__stdcall *)(tTJSString_S *))findfunc_10016420((void (__stdcall *)(_DWORD))aConstTjsCharTt);
TJSString::c_str_100AD0F4 = (int)v9;
}
seed_cstr = (uint8_t *)v9(seed);
filehash_update_100159F0(&ctx, seed_cstr, 2 * seed_len);
}
return filehash_final32_10016B00((int)&ctx, hashValueRet);// return 32, fill hashvalueret
}

这里我们重点关注filehash_init_1000E070其中的init_filehash_ctx_10014140, 搜索立即数6A09E667h,可知这是sha256的table,其中blake2s算法也共用这个table。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
filehash_ctx *__thiscall filehash_init_1000E070(filehash_ctx *pctx, size_t outlen, uint8_t *key, size_t keylen)
{
size_t v5; // eax
uint8_t tmp[32]; // [esp+Ch] [ebp-64h] BYREF
uint8_t tmp2[64]; // [esp+2Ch] [ebp-44h] BYREF

if ( outlen && outlen <= 0x20 && (key || !keylen) )
{
tmp[0] = outlen;
memset(&tmp[9], 0, 23);
*(_QWORD *)&tmp[1] = (unsigned __int8)keylen;
*(_WORD *)&tmp[2] = 0x101;
sub_10014260((uint8_t *)pctx, tmp);
if ( key && keylen )
{
memset(tmp2, 0, sizeof(tmp2));
v5 = 64;
if ( keylen < 64 )
v5 = keylen;
memmove_0(tmp2, key, v5);
filehash_update_100159F0(pctx, tmp2, 0x40u);
memset(tmp2, 0, sizeof(tmp2));
}
}
else
{
init_filehash_ctx_10014140(pctx); // init filehash iv
}
return pctx;
}

.text:10014140 ; void *__thiscall init_filehash_ctx_10014140(filehash_ctx *pctx)
.text:10014140 ; sub_10010410+F5↑p ...
.text:10014140 push esi
.text:10014141 mov esi, ecx
.text:10014143 push 40h ; '@' ; Size
.text:10014145 mov dword ptr [esi], 6A09E667h
.text:1001414B mov dword ptr [esi+4], 0BB67AE85h
.text:10014152 mov dword ptr [esi+8], 3C6EF372h
.text:10014159 mov dword ptr [esi+0Ch], 0A54FF53Ah
.text:10014160 mov dword ptr [esi+10h], 510E527Fh
.text:10014167 mov dword ptr [esi+14h], 9B05688Ch
.text:1001416E mov dword ptr [esi+18h], 1F83D9ABh
.text:10014175 mov dword ptr [esi+1Ch], 5BE0CD19h
.text:1001417C lea eax, [esi+30h]
.text:1001417F push 0 ; Val
.text:10014181 mov dword ptr [esi+20h], 0
.text:10014188 mov dword ptr [esi+24h], 0
.text:1001418F mov dword ptr [esi+28h], 0
.text:10014196 mov dword ptr [esi+2Ch], 0
.text:1001419D push eax ; void *
.text:1001419E mov eax, ds:off_10080BD4
.text:100141A3 call eax ; _memset
.text:100141A5 add esp, 0Ch
.text:100141A8 mov dword ptr [esi+70h], 0
.text:100141AF mov dword ptr [esi+74h], 0
.text:100141B6 mov byte ptr [esi+78h], 0
.text:100141BA pop esi
.text:100141BB retn

这时候就大胆假设小心求证了, 把blake2s的结构放进去看看能不能成立。经测试,这个和原版的blake2s结构体布局略有区别(input缓存和pos跑到下面了,并且多了个pos),如下结构体是能对上的。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63

struct filehash_ctx
{
uint32_t h[8]; // chained state
uint32_t t[2]; // total number of bytes
size_t c; // pointer for b[]
size_t outlen; // digest size
uint8_t b[64]; // input buf
uint32_t pos;
};

void *__thiscall filehash_update_100159F0(filehash_ctx *pctx, uint8_t *src)
{
uint8_t *v2; // ebx
size_t v4; // edi
uint32_t pos; // ecx
void *result; // eax
uint32_t Srca; // [esp+10h] [ebp+8h]
uint8_t *Srcb; // [esp+10h] [ebp+8h]
size_t srcsize; // [esp+14h] [ebp+Ch]

v2 = src;
if ( src )
{
v4 = srcsize;
if ( srcsize )
{
pos = pctx->pos;
result = (void *)(64 - pos);
Srca = 64 - pos;
if ( srcsize > 64 - pos )
{
memmove_0(&pctx->b[pos], v2, 64 - pos);
pctx->t[0] += 64;
pctx->t[1] += pctx->t[0] < 0x40;
result = (void *)filehash_compress_10012500(pctx, (int)pctx->b);
v4 = srcsize - Srca;
v2 += Srca;
if ( srcsize - Srca > 0x40 )
{
Srcb = (uint8_t *)(((v4 - 65) >> 6) + 1);
do
{
pctx->t[0] += 64;
pctx->t[1] += pctx->t[0] < 0x40;
result = (void *)filehash_compress_10012500(pctx, (int)v2);
v2 += 64;
v4 -= 64;
--Srcb;
}
while ( Srcb );
}
pctx->pos = 0;
}
if ( v4 )
{
result = memmove_0(&pctx->b[pctx->pos], v2, v4);
pctx->pos += v4;
}
}
}
return result;
}

filehash_compress_10012500函数超级长, 不过大概流程和blake2s也能对上。G操作里面一大堆ROTR32

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
int __thiscall filehash_compress_10012500(filehash_ctx *ctx, int last)
{
...
v3 = 0;
v4 = (unsigned __int8 *)(last + 2);
do
{
*(&v663 + v3++) = *(v4 - 2) | (unsigned __int16)(*(v4 - 1) << 8) | ((*v4 | (unsigned __int16)(v4[1] << 8)) << 16);
v4 += 4;
}
while ( v3 < 16 );
v584 = *(_DWORD *)&ctx->h[12];
v5 = *(_DWORD *)&ctx->h[16];
v461 = __ROL4__((v5 + v663 + *(_DWORD *)ctx->h) ^ *(_DWORD *)&ctx->h[32] ^ 0x510E527F, 16);
v6 = *(_DWORD *)&ctx->h[20];
v624 = v5 + v663 + *(_DWORD *)ctx->h;
v7 = v461 + 1779033703;
v8 = __ROR4__(v5 ^ (v461 + 1779033703), 12);
v462 = __ROR4__((v8 + v664 + v624) ^ v461, 8);
v414 = v462 + v7;
v343 = __ROR4__(v8 ^ (v462 + v7), 7);
v529 = v6 + v665 + *(_DWORD *)&ctx->h[4];
v625 = v8 + v664 + v624;
v9 = __ROL4__(v529 ^ *(_DWORD *)&ctx->h[36] ^ 0x9B05688C, 16);
...
}

至此我们已经确信, file hash大概率是基于blake2s的算法了,先不去详细分析上面那个巨长函数,先尝试一下是不是标准blake2s。很幸运,这个游戏并没有大改,salt为空,原版函数再加上xp3hnp的seed(动态调试得到的)即可搞定。

1
2
3
4
5
from hashlib import blake2s
h = blake2s(digest_size=32)
h.update("!scnlist.txt".encode("utf-16le"))
h.update("xp3hnp".encode("utf-16le"))
print(h.hexdigest()) // c1f625e3a4bb508e082a52a8b032f4b3d2f34ff7fb3a30502574717de6579126

dir hash

分析文件夹哈希与文件哈希方法类似。输出8字节,动态计算的一组值为ED,FEF68C92D344F4F6

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
int __userpurge DirHashCompute_100169F0@<eax>(
IStringHasher *this@<ecx>,
void *hashValueRet,
tTJSString_S *rawstr,
tTJSString_S *seed) // seed=xp3hnp
{
int v4; // edi
int v5; // esi
int (__stdcall *v6)(tTJSString_S *); // eax
int rawstr_len; // edi
int (__stdcall *v8)(tTJSString_S *); // eax
uint8_t *rawstr_cstr; // eax
int (__stdcall *v10)(tTJSString_S *); // eax
int seed_len; // edi
int (__stdcall *v12)(tTJSString_S *); // eax
uint8_t *seed_cstr; // eax
unsigned int saltsize; // [esp-4h] [ebp-54h]
uint8_t buf[80]; // [esp+0h] [ebp-50h] BYREF

saltsize = this->saltsize;
qmemcpy(buf, "uespemosmodnarodarenegylsetybdet", 32);// hash outsize is 32
dirhash_init_100172E0(buf, this->salt, saltsize);
v6 = (int (__stdcall *)(tTJSString_S *))tTJSString::length_100AD158;
if ( !tTJSString::length_100AD158 )
{
v6 = (int (__stdcall *)(tTJSString_S *))findfunc_10016420((void (__stdcall *)(_DWORD))aTjsIntTtjsstri);
tTJSString::length_100AD158 = (int)v6;
}
rawstr_len = ((int (__stdcall *)(tTJSString_S *, int, int))v6)(rawstr, v4, v5);
v8 = (int (__stdcall *)(tTJSString_S *))TJSString::c_str_100AD0F4;
if ( !TJSString::c_str_100AD0F4 )
{
v8 = (int (__stdcall *)(tTJSString_S *))findfunc_10016420((void (__stdcall *)(_DWORD))aConstTjsCharTt);
TJSString::c_str_100AD0F4 = (int)v8;
}
rawstr_cstr = (uint8_t *)v8(rawstr);
di rhash_update_10017480(buf, rawstr_cstr, 2 * rawstr_len);
if ( seed )
{
v10 = (int (__stdcall *)(tTJSString_S *))tTJSString::length_100AD158;
if ( !tTJSString::length_100AD158 )
{
v10 = (int (__stdcall *)(tTJSString_S *))findfunc_10016420((void (__stdcall *)(_DWORD))aTjsIntTtjsstri);
tTJSString::length_100AD158 = (int)v10;
}
seed_len = v10(seed);
v12 = (int (__stdcall *)(tTJSString_S *))TJSString::c_str_100AD0F4;
if ( !TJSString::c_str_100AD0F4 )
{
v12 = (int (__stdcall *)(tTJSString_S *))findfunc_10016420((void (__stdcall *)(_DWORD))aConstTjsCharTt);
TJSString::c_str_100AD0F4 = (int)v12;
}
seed_cstr = (uint8_t *)v12(seed);
dirhash_update_10017480(buf, seed_cstr, 2 * seed_len);
}
return dirhash_final8_10016BD0(buf, hashValueRet);// fill hashValueRet, return hash size
}

看到这行诡异的字符串uespemosmodnarodarenegylsetybdet,一开始以为是key,结果并不是,他是init的参数立即数内联过来了。搜索0x736F6D6570736575,第一条就是siphash

1
2
3
4
5
6
7
8
9
10
11
12
13
14
.text:100169F0                 push    ebp
.text:100169F1 mov ebp, esp
.text:100169F3 sub esp, 50h
.text:100169F6 push dword ptr [ecx+8]
.text:100169F9 mov dword ptr [ebp+buf], 70736575h
.text:10016A00 push dword ptr [ecx+4]
.text:10016A03 lea ecx, [ebp+buf]
.text:10016A06 mov dword ptr [ebp+buf+4], 736F6D65h
.text:10016A0D mov dword ptr [ebp+buf+8], 6E646F6Dh
.text:10016A14 mov dword ptr [ebp+buf+0Ch], 646F7261h
.text:10016A1B mov dword ptr [ebp+buf+10h], 6E657261h
.text:10016A22 mov dword ptr [ebp+buf+14h], 6C796765h
.text:10016A29 mov dword ptr [ebp+buf+18h], 79746573h
.text:10016A30 mov dword ptr [ebp+buf+1Ch], 74656462h

然后分析与上面类似,先写程序确定一下是不是常规的方法。经测试,文件夹哈希用了原版的siphash_2_4方案。

1
2
3
4
5
import siphash
h = siphash.SipHash_2_4(b"\x00" * 16)
h.update("ED".encode("utf-16le"))
h.update("xp3hnp".encode("utf-16le"))
print(h.hexdigest()) # FEF68C92D344F4F6

epilogue

好久没写逆向分析文章了,目前看除了我开源的krkr_hxv4_dumphash,没有公开的资料来具体分析这个臭名昭著的hxv4哈希函数,故写此文。写逆向游戏分析的文章不是想象中的那么容易,一写就是几个小时。因为分析游戏大多时间间隔很长,有时候卡住了往往要过几天才突然有灵感,有些关键地方可能突然想到了或者排查了半天刚好找到,整理并回顾这些过程也花了些时间。这些突破点往往不容易在文章里准确的表达,而且逆向本身也有很多很繁琐的流程,面面俱到都写进去反而使得文章冗长,整体流程不清晰。因此本文以分析哈希函数为主,其他部分仅写了关键流程和数据结构,略去了繁琐的调试过程,希望可以抛砖引玉,享受在逆向抽丝剥茧的乐趣中。

reference

krkrz
KrkrExtractForCxdecV2
krkrdump
hxcrypt
krkr_hxv4_dumpkey
krkr_hxv4_dumphash
blake2s
siphash