首页
社区
课程
招聘
Galgame汉化中的逆向(七):动态汉化分析2_以AZsystem引擎为例
2023-1-21 10:56 19220

Galgame汉化中的逆向(七):动态汉化分析2_以AZsystem引擎为例

2023-1-21 10:56
19220

Galgame汉化中的逆向(七):动态汉化分析2_以AZsystem引擎为例

好久没发帖了,不知不觉又到了除夕,祝大家新年快乐~

 

by devseed, 本贴论坛和我的博客同时发布
本贴代码开源详见我的github: GalgameReverse, ReverseUtil

 

上篇链接:Galgame汉化中的逆向(六):动态汉化分析_以MAJIROv3引擎为例

0x0 前言

上节 Galgame汉化中的逆向(六):动态汉化分析_以MAJIROv3引擎为例,我们介绍了动态汉化。动态汉化不用分析封包结构,不用分析opcode,看上去很方便,但是动态汉化解决同步问题会很麻烦,比如说改完文本后backlog文本仍是日文、返回主界面再载入文本没有变动等问题。动态汉化也有可能出现莫名其妙的崩溃bug,且这些bug不容易被调试。

 

针对动态汉化的上述缺点,本节我们将介绍一种这种半动态汉化的方案。与上节的方法不同,本节不进行文本级替换,而是文件级别的替换。即去hook相关函数,动态将解密后的缓冲区替换为我们汉化后的文件。适合于那种封包与加密特别麻烦或复杂的游戏。

 

本文将以azsystem为例,来分析:

  • 引擎如何加载游戏脚本,如何定位关键点提取脚本
  • 引擎如何加载图片,如何解压各通道数据,如何将图片数据送入帧缓存渲染
  • 汉化如何用inline hook对加载后的内容进行替换

lamune_chstest2

0x1 脚本文件分析与提取

(1) asb文件的分析

和上节相同,第一步先分析文件,无论静态分析算法还是动态dump缓冲区,先把文件提取出来。

 

由于方法差不多,这里不再详细展开了。

 

这个游戏封包为.arc文件,用文件长度哈希值来作为加密密钥,里面有若干个.asb脚本文件。IDA里面直接搜.asb字符串就能找到相关函数了,读取脚本文件函数如下:

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
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
int __thiscall sub_43112A(_DWORD *this, char *script_name)
{
  char *raw_data; // edi
  int v4; // eax
  unsigned int v5; // ecx
  _DWORD *v7[4]; // [esp+8h] [ebp-34h] BYREF
  int v8; // [esp+18h] [ebp-24h] BYREF
  unsigned int compressed_size; // [esp+1Ch] [ebp-20h]
  unsigned int raw_size; // [esp+20h] [ebp-1Ch]
  int v11; // [esp+24h] [ebp-18h]
  int (__thiscall **v12)(void *, char); // [esp+28h] [ebp-14h]
  char *compressed_data; // [esp+2Ch] [ebp-10h]
  int v14; // [esp+38h] [ebp-4h]
 
  v7[0] = off_460A6C;
  sub_40BD95(v7);
  v14 = 1;
  v12 = &off_462CDC;
  v11 = 0;
  sub_430FC9((int)this);
  if ( fopen_40C102(v7, script_name, 0x80000000) != 1 )
  {
    logprintf_407C41("CScript::Create", byte_4679CC, script_name);
    goto LABEL_13;
  }
  readfile_40C03E(v7, (char *)&v8, 0xC);
  if ( v8 == 0x1A425341 ) // asb\x1a
  {
    compressed_data = (char *)operator new(compressed_size);
    raw_data = (char *)operator new(raw_size);
    readfile_40C03E(v7, compressed_data, compressed_size);
    if ( sub_430F6A(compressed_data, compressed_size, raw_size) )
    {
      v4 = decompress_40AB65(compressed_data, compressed_size, raw_data, raw_size);// decompress
      v5 = raw_size;
      if ( v4 == raw_size )
      {
        this[4] = 0;
        this[1] = raw_data;
        this[2] = v5;
        this[3] = raw_data;
        this[5] = raw_data;
        v11 = 1;
LABEL_10:
        if ( compressed_data )
          j__free(compressed_data);
        goto LABEL_13;
      }
      logprintf_407C41("CScript::Create", byte_467A38, script_name);
    }
    else
    {
      logprintf_407C41("CScript::Create", byte_467A0C, script_name);// error
    }
    if ( raw_data )
      j__free(raw_data);
    goto LABEL_10;
  }
LABEL_13:
  v14 = -1;
  v12 = &off_462CDC;
  v7[0] = off_460A6C;
  sub_40BFDD(v7);
  return v11;
}

lamune_loadscript

 

简单分析后,我们可以得到asb的文件头结构、校验文本函数、解压函数以下结论,具体如下:

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
28
29
30
typedef struct {
        s8 magic[4];                        /* "ASB" */
        u32 comprlen;
        u32 uncomprlen;
        u32 unknown;
} asb_header_t;
 
typedef struct {
        s8 magic[4];                        /* "ASB\x1a" 通过此magic来定位*/
        u32 comprlen;
        u32 uncomprlen;
} asb1a_header_t;
 
// CScript.constructor, 这里不再自己构造了,在游戏调用的时候记录下this指针
void *__thiscall sub_43277F(_DWORD *this)
 
// check_valid
BOOL __stdcall sub_430F6A(char *compressed_data, int compressed_size, int raw_size)
 
// decompress
sub_40AB65(char *compressed_data, int compressed_len, char *raw_data, int raw_len)
 
0043112A    | B8 9EE54500  | mov eax,lamune.45E59E |load_script(char* name)
 
004311D4    | FF75 E4  | push dword ptr ss:[ebp-1C]  | raw_len
004311D7    | 8D4D EC  | lea ecx,dword ptr ss:[ebp-14]
004311DA    | 57         | push edi  | raw_data
004311DB    | FF75 E0    | push dword ptr ss:[ebp-20] | compressed_len
004311DE    | FF75 F0    | push dword ptr ss:[ebp-10] | compressed_data
004311E1    | E8 7F99FDFF| call lamune.40AB65| decompress

(2) asb文件的解密与提取

提取只需要hooksub_40AB65,frida代码如下:

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
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
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
/*
        for lamune.exe v1.0
        open the game to title, then
        frida -l lamune_hook.js -n lamune.exe
        next go to the prologue to dump all asbs
*/
function install_decompress_hook(outdir='./dump')
{
       // hook decompress function to dump
       const addr_decompress = ptr(0x40AB65);
       var raw_asbname = "";
       var raw_asbdata = ptr(0);
       var raw_asbsize = 0;
       Interceptor.attach(addr_decompress, {
           onEnter: function(args)
           {
               raw_asbdata = ptr(args[2]);
               raw_asbsize = args[3].toUInt32();
               raw_asbname = ptr(this.context.ebp).add(8).
                               readPointer().readAnsiString();
           },
           onLeave: function(retval)
           {
               //var asbname = asbname_buf.readAnsiString();
               var asbname = raw_asbname;
               console.log(asbname,
                   ", raw_asbdata addr at", raw_asbdata,
                   ", raw_asbsize ", raw_asbsize)
               try{
                   var fp = new File(outdir+"/"+asbname, 'wb');
                   fp.write(raw_asbdata.readByteArray(raw_asbsize));
                   fp.close();
               }
               catch(e)
               {
                   console.log("file error!", e);
               }
 
           }
       })
}
 
function dump_asbs(names, outdir="./dump")
{
    const addr_loadscript = ptr(0x43112A);
    const load_script = new NativeFunction(addr_loadscript, 
        'void', ['pointer', "pointer"], 'thiscall');
    console.log("load_script at:", load_script)
 
    // use this to store c++ context
    var pthis = ptr(0)
    Interceptor.attach(addr_loadscript, {
        onEnter: function(args)
        {
            pthis = ptr(this.context.ecx)
        }
    })
    install_decompress_hook(outdir)
 
    // wait for c++ context
    while(!pthis.toInt32())
    {
        Thread.sleep(0.2);
    }
 
    // dump all scripts
    var name_buf = Memory.alloc(0x100);
    for(var i=0;i<names.length;i++)
    {
        console.log("try to dump", names[i], ", this=",pthis);
        name_buf.writeAnsiString(names[i]);
        load_script(pthis, name_buf);
    }
    console.log("dump asbs finished!\n");
}
 
function dump_scenario()
{
    var names_v103 = ["00suzuk.asb"]
    dump_asbs(names_v103)
}

用其他工具如arc unpack可以得到arc封包的文件名,把文件名录入frida脚本,即可dump出全部asb脚本。

 

lamune_dumpasb

0x2 动态替换脚本文件

(1) 替换解密的asb缓冲区

结合上面文件分析,我们可以在004311E1| E8 7F99FDFF| call lamune.40AB65| decompress进行inlinehook,在此直接加载我们已经解密并汉化的asb文件。解密的缓冲区是前面new出来的,我们还需要修改缓冲区大小。另外还要nop掉缓冲区crc校验的函数。

 

上节我们用了detours,这期我们来手动inlinehook,步骤如下:

  1. 在需要hook的位置用5字节call(E9)jmp(E8) 进行相对跳转到我们的函数上,

    机器码为E8 XXXXXXXX, E9 XXXXXXXX

    XXXXXXXX为相对于下一条指令的偏移,即targetva - (va + 5)

  2. 执行完后hook的函数后,结尾手动修复一下被我们修改5字节破坏的代码,跳转到下个指令处。

动态替换解密后的缓冲区脚本代码如下:

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
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
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
/* for hook new decompressed buffer
0043119A   | FF75 E0   | push dword ptr ss:[ebp-20]
0043119D   | E8 A1510000  | call lamune.436343 | new
004311A2   | FF75 E4    | push dword ptr ss:[ebp-1C]  | [ebp-1c] raw_size
004311A5   | 8945 F0  | mov dword ptr ss:[ebp-10],eax
004311A8   | E8 96510000         | call lamune.436343  | new raw_buf
*/
const DWORD g_newrawbufi_4311A2 = 0x4311A2;
const DWORD g_newrawbufo_4311A8 = 0x4311A8;
 
/* for hook decompress asb
.text:004311D4 FF 75 E4          push    [ebp+raw_size]  ; raw_len
.text:004311D7 8D 4D EC          lea     ecx, [ebp+var_14]
.text:004311DA 57                push    edi             ; raw_data
.text:004311DB FF 75 E0    push [ebp+compressed_size] ; compressed_len
.text:004311DE FF 75 F0    push [ebp+compressed_data] ; compressed_data
.text:004311E1 E8 7F 99 FD FF    call    decompress_40AB65
*/
const DWORD g_decompressasbi_4311E1 = 0x4311E1;
const DWORD g_decompressasbo_40AB65 = 0x40AB65;
 
// inlinehook stubs
void __declspec(naked) newrawbuf_hook_4311A2()
{
    __asm{
        pushad;
        xor eax, eax;
        // size_t __stdcall load_rawasb(char *name, PBYTE buf)
        push eax;
        push [ebp+8];
        call load_rawasb;
        test eax, eax;
        je newrawbuf_hook_end;
        mov [ebp-0x1c], eax; // change raw buf size
        newrawbuf_hook_end:
        popad;
 
        // fix origin code
        push dword ptr [ebp-0x1c];
        mov dword ptr [ebp-0x10], eax;
        jmp dword ptr ds:[g_newrawbufo_4311A8];
    }
}
 
void __declspec(naked) decompressasb_hook_4311E1()
{
    //sub_40AB65(char *compressed_data, int compressed_len, char *raw_data, int raw_len)
    __asm {
        push [esp+0xc]; // after push ret addr, above, raw_buf
        push [ebp+0x8];  // asbname
        call load_rawasb;
        test eax, eax;
        je decompress_origin;
        ret 0x10;
        decompress_origin:
        mov eax, 0x99E15CB4; // this is the original corrent crc value
        mov dword ptr ds:[0x0047E718], eax; // this is not worked...
        jmp dword ptr ds:[g_decompressasbo_40AB65];
    }
}
 
// hook install functions
void install_asbhook()
{
    /* inlinehook check_valid
    .text:0040AB8A 6A 00             push    0
    .text:0040AB8C 8D 43 FC          lea     eax, [ebx-4]
    .text:0040AB8F 50                push    eax
    .text:0040AB90 8D 77 04          lea     esi, [edi+4]
    .text:0040AB93 56                push    esi
    .text:0040AB94 E8 27 D9 FF FF    call    makecrc_4084C0
    .text:0040AB99 83 C4 0C          add     esp, 0Ch
    .text:0040AB9C 39 07             cmp     [edi], eax
    .text:0040AB9E 75 64             jnz     short loc_40AC04
    */
    BYTE nop2[0x2]={0x90, 0x90};
    winhook_patchmemory((LPVOID)0x4311d2,
        nop2, sizeof(nop2));
    winhook_patchmemory((LPVOID)0x40AB9E,
        nop2, sizeof(nop2));
 
    // inlinehook newrawdata
    BYTE jmpE8buf[0x5]={0xE9}; // jmp relative
    *(DWORD*)(jmpE8buf+1) = (DWORD)newrawbuf_hook_4311A2- 
        ((DWORD)g_newrawbufi_4311A2 + sizeof(jmpE8buf));
    winhook_patchmemory((LPVOID)g_newrawbufi_4311A2,
        jmpE8buf, sizeof(jmpE8buf));
 
    // inlinehook decompress
    BYTE callE9buf[0x5]={0xE8}; // call relative
    *(DWORD*)(callE9buf+1) =(DWORD)decompressasb_hook_4311E1- 
        ((DWORD)g_decompressasbi_4311E1 + sizeof(jmpE8buf));
    winhook_patchmemory((LPVOID)g_decompressasbi_4311E1,
        callE9buf, sizeof(callE9buf));
}

上面代码中load_rawasb即为我们读取对应解密文件的代码,这里为了减少零碎文件,我采取了从zip文件中读取的方法。

 

此处不再赘述,详见我的github

(2) 修改sjis检测字节支持gbk编码

导入中文文本后,经测试发现一大堆半角乱码。

 

这是因为有sjis首字节字符编码范围检测,不在sjis范围内的字符将被解析为单字节文本。

 

lamune_sjis_error

 

与其他游戏不同,此游戏不是用cmp ax, 0x81等指令来检测sjis字符,而且位置过多过于分散,修改起来很麻烦。

 

这部分定位我们可以在TextOutA下断点,往上慢慢找,可以看到下图位置:

 

lamune_sjischeck

 

这里非常巧妙,用一条c^0x20 + 0x5f > 0x3B就可以判断是否为sjis首字符了,具体分析如下:

1
2
3
4
5
6
7
8
.text:004340F6 loc_4340F6:
.text:004340F6 mov     ecx, [ebp+74h+var_4]
.text:004340F9 mov     cl, [ecx]
.text:004340FB mov     dl, cl
.text:004340FD xor     dl, 20h
.text:00434100 add     dl, 5Fh ; '_'
.text:00434103 cmp     dl, 3Bh ; ';'
.text:00434106 ja      loc_434215

修改方法也很简单,把上面xoraddnoppatch,编码检测改为cmp dl, 0x80即可。

 

修改完后,虽然文本框正确了,但我们发现backlog中文本还有乱码。

 

这时候就要在搜索其他地方的检测字符函数了,可以试着搜cmp al|bl|cl|dl, 0x3b,逐个下断点,启动backlog看哪里断下。

 

lamune_backlog_error

 

lamune_checksjis_search

(3) asb opcode分析

0nana.asb为例,这个opcode是对齐的,很工整,如下图:

 

lamune_script_decrypt

 

总结起来就是optype 4, oplengh 4, payload n结构,超长文本只需要修正一下oplenghjmp相关的指令就行了,如下:

1
2
3
4
5
6
7
8
9
10
// from the file start, there are several opcodes entries
optype 4, oplengh 4, payload n
 
[26|27 00 00 00], oplengh 4, [00]*0x10, optext // 26 music, 27 text
[0d 00 00 00], oplengh 4, [00]*4, option_num 4, [00] * 8, text1, 00, text2 ... // option
[0a 00 00 00], [18 00 00 00] , addr 4, [00]*4, unknow1 4, unknow2 4 // jmp
[0b 00 00 00], [1c 00 00 00] , addr 4, [00]*4, unknow1 4, unknow2 4 // option jmp
 
00 00 00 00 FF FF FF FF FF FF FF FF 00 00 00 00
00 00 00 00 00 00 00 00 // end with that

将测试文本导入后,我们可以完成超长文本的汉化测试了~

 

lamune_chstest

0x3 图片文件的加载和渲染分析

(1) 定位图片显示缓冲区

这个游戏是通过Windows compatible DC进行绘图的,我们可以在CreateDIBinfo下断点,然后一层层往上跟,找到在缓冲区填充像素的函数,之后bitblt到帧缓冲位置。这里有个麻烦事,这游戏有很多虚函数通过虚表来寻址,如v3=(*(**v7+12))(*v7, v5, v10,a3这种。静态跟起来很费劲,可以尝试动态来看虚表。由于跟踪过于繁琐了,具体流程从略了,callback和具体调用流程如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
0019FD80  0040EF75            50        return to lamune.sub_40EE6F+106 from ???           User // CreateDIBinfo
0019FDD0  00401E77  0040EE6F  34        return to lamune.sub_401D0F+168 from lamune.sub_40 User
0019FE04  0040955D            24        return to lamune.sub_40951B+42 from ???            User
0019FE28  0040686C  0040951B  24        return to lamune.sub_406813+59 from lamune.sub_409 User
0019FE4C  004383EA  00406813  A4        return to lamune.EntryPoint+184 from lamune.sub_40 User
0019FEF0  0043827E  0043F210  84        return to lamune.EntryPoint+18 from lamune.sub_43F User
 
DWORD __thiscall sub_42A199(int *this) // neko_logo.cpb
| loadimg_419E03(off_473088, "neko_logo.cpb", this + 0x214);
  | readcpb_40C03E((_DWORD **)this + 1, cpb_header, 0x10);
  | (*(_DWORD *)*v7+4))(*v7, cpb_header) //check magic cpb\x1a
  | (*(v8 + 0x3C))(v10[4], v10[5])   // 0041D3FB, read full
  | v3=(*(**v7+12))(*v7, v5, v10,a3);// 0041D453, check depth
    |return (*(*v6 + 0x10))(v6, a2, a4);// 0041E36F, 0041ddb8,decompress
          |decompress2_40AA38(char *compressed_buf, size_t |compressed_size, char *raw_buf, size_t raw_len) // lzss?
| sub_40C9C1(DWORD *this, int a2, int a3, int *a4, DWORD *a5)
  |sub_4101EB(v9 + 2, a2, a3, a4, *a5, a5[1], a5[2], a5[3], 0);// bltalpha
  |(*(this[2] + 0x48))(this + 2, a2, a3, a4, *a5, a5[1], a5[2], a5[3], 0xCC0020);// 004123E1, to bitblt

(2) cpb图片加载

上面我们来讲了一下定位方法,和整体加载流程。在这节我们来分析一下cpb文件如何读取和加载渲染到屏幕上的。

.1 cpb结构

cpb中像素是分通道存储的,数据结构如下:

1
2
3
4
5
6
7
8
9
10
11
12
13
00000000 cpb1a_header_t  struc ; (sizeof=0x20, mappedto_128)
00000000 ; XREF: decompresscpb_41E36F/r
00000000 magic           db 4 dup(?) ; string(C)
00000004 unknow1         db ?
00000005 color_depth     db ?
00000006 unknow2         db ?
00000007 version         db ?
00000008 width           dw ? ; XREF: decompresscpb_41E36F+39/r
0000000A height          dw ? ; XREF: decompresscpb_41E36F+3E/r
0000000C max_comprlen    dd ? ; XREF: decompresscpb_41E36F+56/r
00000010 comprlen dd 4 dup(?); XREF: decompresscpb_41E36F+93/r
00000010  ; decompresscpb_41E36F+B7/r ...
00000020 cpb1a_header_t  ends

.2 prepare DC

在渲染图片之前,游戏引擎先进行DC的初始化。

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
void *__thiscall sub_40FDC2(void **this, LONG width, int height)
{
  void *result; // eax
  HBITMAP v5; // eax
  HDC dc; // eax
  int (__thiscall **v7)(void **, _DWORD); // eax
  BITMAPINFO pbmi; // [esp+8h] [ebp-2Ch] BYREF
 
  if ( (void *)width == this[41] && (void *)height == this[42] )
    return (void *)(*((int (__thiscall **)(void **, _DWORD))*this + 26))(this, 0);
  (*((void (__thiscall **)(void **))*this + 13))(this);
  if ( width > 0 && height > 0 )
  {
    memset(&pbmi, 0, sizeof(pbmi));
    pbmi.bmiHeader.biHeight = -height;
    pbmi.bmiHeader.biSize = 0x28;// struct size
    pbmi.bmiHeader.biWidth = width;
    pbmi.bmiHeader.biPlanes = 1// must be 1
    pbmi.bmiHeader.biBitCount = 32;    // rgba
    pbmi.bmiHeader.biCompression = 0;
    v5 = CreateDIBSection(0, &pbmi, 0, this + 0x28, 0, 0); // this+0x28
    this[37] = v5;
    if ( v5 )
    {
      dc = CreateCompatibleDC(0);
      this[0x27] = dc;
      if ( dc )
      {
        this[0x26] = SelectObject(dc, this[0x25]);
        this[0x29] = (void *)width;
        this[0x2A] = (void *)height;
        this[0x2E] = (void *)(height - 1);
        v7 = (int (__thiscall **)(void **, _DWORD))*this;
        this[0x2B] = 0;
        this[0x2C] = 0;
        this[0x2D] = (void *)(width - 1);
        result = (void *)v7[0x1A](this, 0); // 0041295A, FillRect
        this[0x52] = result;
        return result;
      }
    }
    (*((void (__thiscall **)(void **))*this + 0xD))(this);
  }
  return 0;
}

.3 load cpb

这部分是读取cpb到内存里,并检验文件头等信息

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
int __thiscall loadimg_419E03(_DWORD *this, char *filename, int *a3)
{
  int v3; // ebx
  _DWORD **v5; // edi
  _DWORD *v7; // esi
  int v8; // eax
  int v9; // eax
  int v10[7]; // [esp+Ch] [ebp-2Ch] BYREF
  char cpb_header[16]; // [esp+28h] [ebp-10h] BYREF
  int i; // [esp+40h] [ebp+8h]
 
  v3 = 0;
  if ( !filename || !a3 )
    return 0;
  v5 = (this + 1);
  if ( fopen_40C102(this + 1, filename, 0x80000000) != 1 )
  {
    logprintf_407C41("CGraphicLoader::GDILoad", "指定されたファイルが見つかりません [%s]", filename);
    return 0;
  }
  readcpb_40C03E(this + 1, cpb_header, 0x10);   // this+1 fp
  i = 0;
  v7 = this + 5// for test magic?
  do
  {
    if ( *v7 )
    {
      if ( (*(**v7 + 4))(*v7, cpb_header) == 1 )// 0041D0E8, 0041D3E9
     // check magic cpb\x1a,
      {
        sub_40C0A0(v5, 0, 0);
        memset(v10, 0, sizeof(v10));
        v3 = (*(**v7 + 8))(*v7, v5, v10); // 0041D3FB, read full header
        if ( v3 == 1 )
        {
          v8 = *a3;   // 0041D3FB, read full header
          v9 = v10[3] == 1 ? (*(v8 + 0x3C))(v10[4], v10[5]) : (*(v8 + 0x38))(v10[4], v10[5]);
          v3 = v9;
          if ( v9 == 1 )
          {
            sub_40C0A0(v5, 0, 0);
            v3 = (*(**v7 + 12))(*v7, v5, v10, a3);// 0041D453, check depth and decompress
            if ( v3 == 1 )
              break;
          }
        }
      }
    }
    ++i;
    ++v7;
  }
  while ( i < 4 );
  sub_40BFDD(v5);
  return v3;
}

加载后,会根据通道数不同调用不同的解压缩函数。

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
int __thiscall sub_41D453(_DWORD *this, int a2, int a3, int a4)
{
  int v6; // esi
  int v8; // ebx
  int v9; // [esp+8h] [ebp-4h]
 
  v9 = 0;
  if ( (*(*a4 + 0x2C))(a4) != 1 )    // 00401291, mov
    return 0;
  v6 = this[*(a3 + 0x18) + 1];
  if ( !v6 )
    return 0;
  if ( (*(*a4 + 0x1C))(a4) == 8 )// 00401278, mov
  {
    if ( *(a3 + 4) == 8 )   // 8bit with color panel
      return (*(*v6 + 4))(v6, a2, a4);
  }
  else if ( (*(*a4 + 0x1C))(a4) == 32 ) // 32bit rgba
  {
    v8 = *(a3 + 4);
    if ( v8 == 8 )
      return (*(*v6 + 8))(v6, a2, a4);
    if ( v8 == 24 )
      return (*(*v6 + 0xC))(v6, a2, a4);  // 0041e1c8 decompresscpb24
    if ( v8 != 32 || (*(*a4 + 0x30))(a4) != 1 ) // 00401298, mov
      return v9;
    return (*(*v6 + 0x10))(v6, a2, a4); // 0041E36F,decompresscpb32
  }
  return v9;
}

.4 decompress cpb

这个游戏有多个cpb解压函数,对应着不同通道数的文件,这里以32位图为例分析。

 

注意这里vv1 = (*(*obja + 0xC))(obja)中的vv1值为prepare dc中的v5 = CreateDIBSection(0, &pbmi, 0, this + 0x28, 0, 0) 此句的DIB缓冲区。

 

我们可以替换decompress_channel_40AA38后的缓冲区为汉化后的图片,然后让游戏引擎帮我们复制到DIB缓冲区内。

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
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
int __thiscall decompresscpb32_41E36F(void *this, int *obj)
{
  int v3; // eax
  size_t pixels; // esi
  char *raw_buf; // eax MAPDST
  char *pchanel1; // ebx
  int vv2; // eax
  char *pchannel0; // edi
  _BYTE *v11; // esi
  _BYTE *v12; // eax
  int v13; // edx
  cpb1a_header_t cpb_header; // [esp+408h] [ebp-58h] BYREF
  int v16; // [esp+428h] [ebp-38h]
  int v17; // [esp+42Ch] [ebp-34h]
  char *v18; // [esp+430h] [ebp-30h]
  int width; // [esp+434h] [ebp-2Ch] MAPDST
  char *compressed_buf; // [esp+438h] [ebp-28h] MAPDST
  int pcurvv2; // [esp+43Ch] [ebp-24h]
  int i; // [esp+444h] [ebp-1Ch]
  char *pchanel3; // [esp+448h] [ebp-18h]
  _BYTE *vv1; // [esp+44Ch] [ebp-14h]
  int j; // [esp+450h] [ebp-10h]
  int v27; // [esp+45Ch] [ebp-4h]
  char *obja; // [esp+46Ch] [ebp+Ch] MAPDST
  char *pchanel2; // [esp+46Ch] [ebp+Ch]
 
  v27 = 0;
  j = 0;
  if ( readcpb_40C03E(obj, cpb_header.magic, 0x20) )
  {
    v3 = *obja;
    width = cpb_header.width;
    i = cpb_header.height;
    pixels = cpb_header.width * cpb_header.height;
    v17 = (*(v3 + 0x24))(obja);                 // 00401283, mov
    compressed_buf = operator new(cpb_header.max_comprlen);
    raw_buf = operator new(4 * pixels);
    pchanel1 = &raw_buf[pixels];
    pchanel2 = &raw_buf[pixels + pixels];
    pchanel3 = &pchanel2[pixels];
    vv1 = (*(*obja + 0xC))(obja); // 0040125C, mov this+0x28, get hdc buffer
        //  CreateDIBSection(0, &pbmi, 0, this + 0x28, 0, 0);
    vv2 = (*(*obja + 0x20))(obja);
    pcurvv2 = vv2;
    if ( readcpb_40C03E(obj, compressed_buf, cpb_header.comprlen[0])
      && decompress_channel_40AA38(compressed_buf, cpb_header.comprlen[0], raw_buf, pixels) != -1
      && readcpb_40C03E(obj, compressed_buf, cpb_header.comprlen[1])
      && decompress_channel_40AA38(compressed_buf, cpb_header.comprlen[1], pchanel1, pixels) != -1
      && readcpb_40C03E(obj, compressed_buf, cpb_header.comprlen[2])
      && decompress_channel_40AA38(compressed_buf, cpb_header.comprlen[2], pchanel2, pixels) != -1
      && readcpb_40C03E(obj, compressed_buf, cpb_header.comprlen[3])
      && decompress_channel_40AA38(compressed_buf, cpb_header.comprlen[3], pchanel3, pixels) != -1 )
    {
      if ( i > 0 )
      {
        pchannel0 = &pchanel1[-pixels];
        ++vv1;
        j = i;
        do    // copy data to dc buf
        {
          if ( width > 0 )
          {
            v11 = vv1;
            v16 = pchanel2 - pchanel1;
            v18 = (pchanel3 - pchanel1);
            v12 = pchanel1;
            v13 = pcurvv2 - pchanel1;
            i = width;
            do
            {
              v11[1] = v12[pchannel0 - pchanel1];
              *v11 = *v12;
              *(v11 - 1) = v12[v16];
              v12[v13] = v12[v18];
              ++v12;
              v11 += 4;
              --i;
            }
            while ( i );
          }
          pchanel2 += width;
          pchanel3 += width;
          vv1 += 4 * width;
          pcurvv2 += v17;
          pchanel1 += width;
          pchannel0 += width;
          --j;
        }
        while ( j );
      }
      j = 1;
    }
    if ( raw_buf )
      j__free(raw_buf);
    if ( compressed_buf )
      j__free(compressed_buf);
  }
  return j;
}

解压各通道算法,看起来有点像lzss改版?

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 __stdcall decompress_channel_40AA38(char *compressed_buf, size_t compressed_size, char *raw_buf, size_t raw_len)
{
  char *v5; // ebx
  char *v6; // edx
  char *v7; // esi
  char *v8; // edi
  unsigned int v9; // ecx
  signed int v10; // eax
  unsigned int v11; // ecx
  char *v12; // esi
  char v13; // cf
  bool v14; // cc
  unsigned int v15; // [esp+Ch] [ebp-4h]
  signed int dstsizea; // [esp+24h] [ebp+14h]
 
  if ( *(compressed_buf + 4) > raw_len )
    return -1;
  v5 = compressed_buf + 20;
  v6 = &compressed_buf[*(compressed_buf + 1) + 20];
  v7 = &v6[*(compressed_buf + 2)];
  dstsizea = *(compressed_buf + 4);
  v8 = raw_buf;
  v15 = 0x80808080;
  do
  {
    if ( (v15 & *v5) != 0 )
    {
      v9 = *v6;
      v6 += 2;
      v10 = (v9 >> 13) + 3;
      qmemcpy(v8, &v8[-(v9 & 0x1FFF) - 1], v10);
      v8 += v10;
    }
    else
    {
      v11 = *v7 + 1;
      v12 = v7 + 1;
      v10 = v11;
      qmemcpy(v8, v12, v11);
      v7 = &v12[v11];
      v8 += v11;
    }
    v13 = v15 & 1;
    v15 = __ROR4__(v15, 1);
    if ( v13 )
      ++v5;
    v14 = dstsizea <= v10;
    dstsizea -= v10;
  }
  while ( !v14 );
  return v8 - raw_buf;
}

.5 bitblt screen dc

最后再通过bitblt到屏幕帧缓存中,至此整个游戏图片渲染分析完毕。

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
// for bitblt
BOOL __thiscall sub_4123E1(void *this, int x, int y, int a4, int x1, int a6, int a7, int a8, DWORD rop)
{
  int v10; // edi
  int v11; // ebx
  int v12; // eax
  int y1; // edi
  int v14; // eax
  int x_c; // ebx
  HDC hdc; // eax
  HDC srchdc; // [esp-10h] [ebp-20h]
 
  v10 = a7 - x1 + 1;
  v11 = a8 - a6 + 1;
  if ( x >= 0 )
  {
    if ( x + v10 > (*(*this + 16))(this) )// 00401263, mov
    {
      if ( (*(*this + 16))(this) - x <= 0 )
        return 0;
      a7 = (*(*this + 16))(this) + x1 - x - 1;
    }
  }
  else
  {
    if ( v10 + x <= 0 )
      return 0;
    x1 -= x;
    v12 = (*(*this + 16))(this) + x1 - 1;
    if ( v12 < a7 )
      a7 = v12;
    x = 0;
  }
  if ( y >= 0 )
  {
    if ( y + v11 > (*(*this + 20))(this) )// 0040126A, mov
    {
      if ( (*(*this + 20))(this) - y <= 0 )
        return 0;
      a8 = (*(*this + 20))(this) + a6 - y - 1;
    }
    y1 = a6;
  }
  else
  {
    if ( v11 + y <= 0 )
      return 0;
    y1 = a6 - y;
    v14 = (*(*this + 20))(this) + a6 - 1 - y;
    if ( v14 < a8 )
      a8 = v14;
    y = 0;
  }
  x_c = a7 - x1 + 1;
  if ( x_c > 0 && a8 - y1 + 1 > 0 )
  {
    srchdc = (*(*a4 + 4))(a4);// 0040124E, mov
    hdc = (*(*this + 4))(this);// 0040124E, mov
    return BitBlt(hdc, x, y, x_c, a8 - y1 + 1, srchdc, x1, y1, rop);
  }
  return 0;

0x4 动态替换图片文件

为了搞明白这个游戏游戏引擎图像如何渲染的,我把很多的虚函数都跟了一遍。

 

其实汉化图片只需要逆向到如何解压cpb文件那里就足够了。这个游戏麻烦地方在于不同通道对应的不同处理函数,要依次来hook替换缓冲区。另外在读取文件适合要记录一些文件名,用于缓冲区动态替换我们汉化的图片。

 

以24位图片代码替换为例,代码如下:

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
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
/* for hook decompressed cpb24 buffer
0041E2DB   | 8B55 0C  | mov edx,dword ptr ss:[ebp+C]
0041E2DE   | 8BC7   | mov eax,edi
0041E2E0   | 2BC6   | sub eax,esi
0041E2E2   | 42   | inc edx
0041E2E3   | 8955 0C  | mov dword ptr ss:[ebp+C],edx
0041E2E6   | 894D EC  | mov dword ptr ss:[ebp-14],ecx
0041E2E9   | 85DB | test ebx,ebx
0041E2EB   | 7E 35 | jle lamune_chs.41E322
*/
const char* g_curcpbname = NULL;
const DWORD g_copycpb24i_41E2DB = 0x41E2DB;
const DWORD g_copycpb24o_41E2E0 = 0x41E2E0;
 
void __declspec(naked) loadcpb_hook_419E03()
{
    __asm {
        push eax;
        mov eax, dword ptr [esp+8]; // after push eax
        mov g_curcpbname, eax;
        pop eax;
 
        // fix origin code
        push ebp;
        mov ebp, esp;
        sub esp, 0x2c;
        jmp dword ptr ds:[g_loadcpbo_419E09];
    }
}
 
void __declspec(naked) copycpb24_hook_41E2DB()
{
    __asm {
        pushad;
        push [ebp-0x20];
        push g_curcpbname;
        // size_t __stdcall load_rawcpb(char *name, PBYTE buf)
        call load_rawcpb;
        popad;
        // fix origin code
        mov edx,dword ptr [ebp+0xC];
        mov eax,edi;
        jmp dword ptr ds:[g_copycpb24o_41E2E0];
    }
}
 
void install_cpbhook()
{
    // inlinehook loadcpb
    BYTE jmpE8buf[0x5]={0xE9}; // jmp relative
    *(DWORD*)(jmpE8buf+1) = (DWORD)loadcpb_hook_419E03- 
        ((DWORD)g_loadcpbi_419E03 + sizeof(jmpE8buf));
    winhook_patchmemory((LPVOID)g_loadcpbi_419E03,
        jmpE8buf, sizeof(jmpE8buf));
 
    // inlinehook copycpb24
    *(DWORD*)(jmpE8buf+1) = (DWORD)copycpb24_hook_41E2DB- 
        ((DWORD)g_copycpb24i_41E2DB + sizeof(jmpE8buf));
    winhook_patchmemory((LPVOID)g_copycpb24i_41E2DB,
        jmpE8buf, sizeof(jmpE8buf));
}

这里采取的是png格式存储的汉化图片,为了方便用了stb进行加载。

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
size_t __stdcall load_rawcpb(char *name, PBYTE buf)
{
    char path[MAX_PATH] = {SYSGRAPH_DIR "/" "\0"};
    strcat(path, name);
    strcpy(path + strlen(path)-       
           strlen(SYSGRAPH_EXT),SYSGRAPH_EXT);
 
    int width, height, channel;
    printf("load_rawcpb(%s, %p)", path, buf);
    size_t entry_size = load_arc_entry(path, NULL);
    const BYTE *tmpbuf = (BYTE*)malloc(entry_size);
    load_arc_entry(path, (PBYTE)tmpbuf);
    char* img = (char*)stbi_load_from_memory(tmpbuf,
        entry_size, &width, &height, &channel, 0);
    free((void*)tmpbuf);
 
    if(!img)
    {
        printf(" not found!\n");
        return 0;
    }
    printf(" width=%d, heigth=%d, channel=%d\n",
        width, height, channel);
    for(int y=0;y<height;y++)
    {
        for(int x=0;x<width;x++)
        {
            char r = *(img + channel * (width*y + x) + 0);
            char g = *(img + channel * (width*y + x) + 1);
            char b = *(img + channel * (width*y + x) + 2);
 
 
            *(buf + 0*height*width + width*y+x) = r;
            *(buf + 1*height*width + width*y+x) = g;
            *(buf + 2*height*width + width*y+x) = b;
            if(channel==4)
            {
                char a = *(img + channel * (width*y + x) + 3);
                *(buf + 3*height*width + width*y+x) = a;
            }
        }
    }
 
    stbi_image_free(img);
    return width*height*channel;
}

加载后遇到渲染bug,我们把对应缓冲区dump出来放到ct2中进行查看,确定原因。

 

lamune_frame_error

 

lamune_texture

 

这里发现原来是stbi_load_from_memory函数对于tga格式有些问题,换成png格式最后参数为0,问题解决。

 

lamune_picture_chs

 

至此,图片汉化问题全部解决。

0x5 后记

这个游戏我逆向了一周多把引擎的加载方式搞明白了,之后又测试导入翻译断断续续修复bug一个月,基本上汉化完美了。这里有个坑,通关后没法打开gallary。这是官方的bug,下载了升级补丁可以修复。但是之前给我的文件是初版游戏,我说基于这个版本分析的。还得把旧版搬到新版上,非常麻烦。这个故事告诉我们,以后汉化要第一时间检查更新补丁。

 

整体来讲,这游戏有三大难点。难点之一在封包上,有加密和校验非常麻烦,因此我们采取了动态替换解密后的缓冲区;其二,图像缓冲区不好找,里面有大量虚函数,需要一点点跟;其三,sjis字符检测过于分散,需要手动一个个调整,而且也是用非主流方式判断的。因此,我认为此游戏比较适合半动态汉化。这种基于文件的替换方式可以免去复杂的封包,同时相比文本层面上的全动态汉化,可以更方便调试,少引发一些文本同步之类的问题。

 

另外我用stb加载图片,这里遇到了问题,xp上运行会崩溃。lamune_xp_error

 

调试定位在了mov eax, large fs:2Ch上,这是因为这个库用了__declspec(thread),在win xpLoadLibrary遇到tls就会崩,定义宏#define STBI_NO_THREAD_LOCALS即可解决。

 

然后进行了若干测试,我这个汉化兼容补丁性还不错~ win xp, win7, win8 , win10甚至连linux wineexagear都测试了,可以说是全平台兼容了~ 完结撒花~

 

lamune_exager

 

lamune_xp_iat

 

lamune_wine


[培训]《安卓高级研修班(网课)》月薪三万计划

收藏
点赞6
打赏
分享
最新回复 (4)
雪    币: 52
活跃值: (3274)
能力值: ( LV2,RANK:10 )
在线值:
发帖
回帖
粉丝
DirtyAngle 2023-1-21 15:56
2
0
什么? 这年头还有xp
雪    币: 6296
活跃值: (4927)
能力值: ( LV12,RANK:250 )
在线值:
发帖
回帖
粉丝
devseed 4 2023-1-25 17:48
3
0
DirtyAngle 什么? 这年头还有xp
大家都喜欢用xp来玩galgame呀~
雪    币: 3418
活跃值: (3627)
能力值: ( LV8,RANK:131 )
在线值:
发帖
回帖
粉丝
coneco 2 2023-1-28 10:20
4
0
感谢分享!
雪    币: 229
活跃值: (1647)
能力值: ( LV2,RANK:10 )
在线值:
发帖
回帖
粉丝
轻装前行 2023-3-16 11:55
5
0
还得是你们二次元啊
游客
登录 | 注册 方可回帖
返回