Skip to content

某在线翻译js破解

字数
2100 字
阅读时间
13 分钟
更新日期
4/17/2021

想给信息流( i.hacking 8.com )换一个翻译引擎,于是看到这个。

它的翻译效果感觉挺不错的,所以就来分析一下怎么调用。

经过调试,它会发两个包,第一个包用于生成游客的jwt token

POST /v1/user/jwt/generate HTTP/1.1
Host: api.interpreter.caiyunai.com
Connection: close
Content-Length: 49
Pragma: no-cache
Cache-Control: no-cache
sec-ch-ua: "Google Chrome";v="89", "Chromium";v="89", ";Not A Brand";v="99"
os-version: 
sec-ch-ua-mobile: ?0
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 11_2_3) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/89.0.4389.114 Safari/537.36
app-name: xy
Content-Type: application/json;charset=UTF-8
Accept: application/json, text/plain, */*
device-id: 
os-type: web
X-Authorization: token:qgemv4jr1y38jyq6vhvi
version: 1.8.0
Origin: https://fanyi.caiyunapp.com
Sec-Fetch-Site: cross-site
Sec-Fetch-Mode: cors
Sec-Fetch-Dest: empty
Referer: https://fanyi.caiyunapp.com/
Accept-Encoding: gzip, deflate
Accept-Language: zh-CN,zh;q=0.9,en;q=0.8

{"browser_id":"6d3a8a9d1e39302beb6fca2dd1310954"}

返回

HTTP/1.1 200 OK
Server: nginx/1.16.1
Date: Sat, 17 Apr 2021 04:10:16 GMT
Content-Type: application/json
Connection: close
Access-Control-Allow-Origin: https://fanyi.caiyunapp.com
Access-Control-Allow-Credentials: true
Access-Control-Allow-Methods: GET, POST, OPTIONS
Access-Control-Allow-Headers: X-Requested-With,Content-Type,X-Authorization,app-name,version,os-type,os-version,device-id
Content-Length: 286

{"jwt":"eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJicm93c2VyX2lkIjoiNmQzYThhOWQxZTM5MzAyYmViNmZjYTJkZDEzMTA5NTQiLCJ2ZXJzaW9uIjoxLCJpcF9hZGRyZXNzIjoiMTIzLjExNy4xNzMuMTc1IiwidG9rZW4iOiJxZ2VtdjRqcjF5MzhqeXE2dmh2aSIsImV4cCI6MTYxODYzMzUxNn0.0vkqXovbGTHE8JW_a31CsAKA4yUWBmkvgfxyC2Axfqk","rc":0}

第二个包,发出翻译请求

POST /v1/translator HTTP/1.1
Host: api.interpreter.caiyunai.com
Connection: close
Content-Length: 201
sec-ch-ua: "Google Chrome";v="89", "Chromium";v="89", ";Not A Brand";v="99"
os-version: 
sec-ch-ua-mobile: ?0
X-Authorization: token:qgemv4jr1y38jyq6vhvi
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 11_2_3) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/89.0.4389.114 Safari/537.36
app-name: xy
Content-Type: application/json;charset=UTF-8
Accept: application/json, text/plain, */*
device-id: 
os-type: web
T-Authorization: eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJicm93c2VyX2lkIjoiNmQzYThhOWQxZTM5MzAyYmViNmZjYTJkZDEzMTA5NTQiLCJ2ZXJzaW9uIjoxLCJpcF9hZGRyZXNzIjoiMTIzLjExNy4xNzMuMTc1IiwidG9rZW4iOiJxZ2VtdjRqcjF5MzhqeXE2dmh2aSIsImV4cCI6MTYxODYzMzUxNn0.0vkqXovbGTHE8JW_a31CsAKA4yUWBmkvgfxyC2Axfqk
version: 1.8.0
Origin: https://fanyi.caiyunapp.com
Sec-Fetch-Site: cross-site
Sec-Fetch-Mode: cors
Sec-Fetch-Dest: empty
Referer: https://fanyi.caiyunapp.com/
Accept-Encoding: gzip, deflate
Accept-Language: zh-CN,zh;q=0.9,en;q=0.8

{"source":"heelo","trans_type":"auto2zh","request_id":"web_fanyi","media":"text","os_type":"web","dict":true,"cached":true,"replaced":true,"detect":true,"browser_id":"6d3a8a9d1e39302beb6fca2dd1310954"}

收到

HTTP/1.1 200 OK
Server: nginx/1.16.1
Date: Sat, 17 Apr 2021 04:17:45 GMT
Content-Type: application/json
Connection: close
Access-Control-Allow-Origin: https://fanyi.caiyunapp.com
Access-Control-Allow-Credentials: true
Access-Control-Allow-Methods: GET, POST, OPTIONS
Access-Control-Allow-Headers: X-Requested-With,Content-Type,X-Authorization,app-name,version,os-type,os-version,device-id,T-Authorization
Content-Length: 66

{"confidence":0.8642611644,"target":"FTIyoT8=","isdict":0,"rc":0}

target解密

可以看到"target":"FTIyoT8="是一个密文,但是前台是明文显示的,所以翻翻js找解密的算法。

通关监视变量,找到这个解密函数

js
function Le(t) {
            var e;
            return function(t) {
                return ke.decode(t)
            }((e = function(t) {
                return "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz".indexOf(t)
            }
            ,
            t.split("").map(function(t) {
                return e(t) > -1 ? "NOPQRSTUVWXYZABCDEFGHIJKLMnopqrstuvwxyzabcdefghijklm"[e(t)] : t
            }).join("")))
        }

百度了一下关键字符串,这就是一个rot13+base64解密算法

发包的参数

发包的请求头上有几个参数

X-AuthorizationT-Authorizationbrowser_id,知道它们怎么来的就能自己构造了。

T-Authorization就是生成的jwt token

X-Authorization在js中直接就硬编码了

image-20210417145454067

browser_id

追踪到生成jwt这里

image-20210417144642259

browser_id = e,e已经被赋值了,跟踪堆栈向上回溯

image-20210417144757249

这就是它的生成算法了,可以看到它是先合并数组i,然后对ix64hash128运算来着。i是啥,打印一下

image-20210417144942057

它就是从e中获取valuee是啥,看旁边的监视器

image-20210417145047587

js
[
    {
        "key": "userAgent",
        "value": "Mozilla/5.0 (Macintosh; Intel Mac OS X 11_2_3) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/89.0.4389.114 Safari/537.36"
    },
    {
        "key": "webdriver",
        "value": false
    },
    {
        "key": "language",
        "value": "zh-CN"
    },
    {
        "key": "colorDepth",
        "value": 30
    },
    {
        "key": "deviceMemory",
        "value": 8
    },
    {
        "key": "hardwareConcurrency",
        "value": 8
    },
    {
        "key": "screenResolution",
        "value": [
            900,
            1440
        ]
    },
    {
        "key": "availableScreenResolution",
        "value": [
            875,
            1385
        ]
    },
    {
        "key": "timezoneOffset",
        "value": -480
    },
    {
        "key": "timezone",
        "value": "Asia/Shanghai"
    },
    {
        "key": "sessionStorage",
        "value": true
    },
    {
        "key": "localStorage",
        "value": true
    },
    {
        "key": "indexedDb",
        "value": true
    },
    {
        "key": "addBehavior",
        "value": false
    },
    {
        "key": "openDatabase",
        "value": true
    },
    {
        "key": "cpuClass",
        "value": "not available"
    },
    {
        "key": "platform",
        "value": "MacIntel"
    },
    {
        "key": "plugins",
        "value": [
            [
                "Chrome PDF Plugin",
                "Portable Document Format",
                [
                    [
                        "application/x-google-chrome-pdf",
                        "pdf"
                    ]
                ]
            ],
            [
                "Chrome PDF Viewer",
                "",
                [
                    [
                        "application/pdf",
                        "pdf"
                    ]
                ]
            ],
            [
                "Native Client",
                "",
                [
                    [
                        "application/x-nacl",
                        ""
                    ],
                    [
                        "application/x-pnacl",
                        ""
                    ]
                ]
            ]
        ]
    },
    {
        "key": "canvas",
        "value": [
            "canvas winding:yes",
            "canvas fp:data:image/png;base64,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"
        ]
    },
    {
        "key": "webgl",
        "value": [
            "data:image/png;base64,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",
            "extensions:ANGLE_instanced_arrays;EXT_blend_minmax;EXT_color_buffer_half_float;EXT_disjoint_timer_query;EXT_float_blend;EXT_frag_depth;EXT_shader_texture_lod;EXT_texture_compression_rgtc;EXT_texture_filter_anisotropic;WEBKIT_EXT_texture_filter_anisotropic;EXT_sRGB;OES_element_index_uint;OES_fbo_render_mipmap;OES_standard_derivatives;OES_texture_float;OES_texture_float_linear;OES_texture_half_float;OES_texture_half_float_linear;OES_vertex_array_object;WEBGL_color_buffer_float;WEBGL_compressed_texture_s3tc;WEBKIT_WEBGL_compressed_texture_s3tc;WEBGL_compressed_texture_s3tc_srgb;WEBGL_debug_renderer_info;WEBGL_debug_shaders;WEBGL_depth_texture;WEBKIT_WEBGL_depth_texture;WEBGL_draw_buffers;WEBGL_lose_context;WEBKIT_WEBGL_lose_context;WEBGL_multi_draw",
            "webgl aliased line width range:[1, 1]",
            "webgl aliased point size range:[1, 255.875]",
            "webgl alpha bits:8",
            "webgl antialiasing:yes",
            "webgl blue bits:8",
            "webgl depth bits:24",
            "webgl green bits:8",
            "webgl max anisotropy:16",
            "webgl max combined texture image units:80",
            "webgl max cube map texture size:16384",
            "webgl max fragment uniform vectors:1024",
            "webgl max render buffer size:16384",
            "webgl max texture image units:16",
            "webgl max texture size:16384",
            "webgl max varying vectors:15",
            "webgl max vertex attribs:16",
            "webgl max vertex texture image units:16",
            "webgl max vertex uniform vectors:1024",
            "webgl max viewport dims:[16384, 16384]",
            "webgl red bits:8",
            "webgl renderer:WebKit WebGL",
            "webgl shading language version:WebGL GLSL ES 1.0 (OpenGL ES GLSL ES 1.0 Chromium)",
            "webgl stencil bits:0",
            "webgl vendor:WebKit",
            "webgl version:WebGL 1.0 (OpenGL ES 2.0 Chromium)",
            "webgl unmasked vendor:Intel Inc.",
            "webgl unmasked renderer:Intel(R) Iris(TM) Plus Graphics 655",
            "webgl vertex shader high float precision:23",
            "webgl vertex shader high float precision rangeMin:127",
            "webgl vertex shader high float precision rangeMax:127",
            "webgl vertex shader medium float precision:23",
            "webgl vertex shader medium float precision rangeMin:127",
            "webgl vertex shader medium float precision rangeMax:127",
            "webgl vertex shader low float precision:23",
            "webgl vertex shader low float precision rangeMin:127",
            "webgl vertex shader low float precision rangeMax:127",
            "webgl fragment shader high float precision:23",
            "webgl fragment shader high float precision rangeMin:127",
            "webgl fragment shader high float precision rangeMax:127",
            "webgl fragment shader medium float precision:23",
            "webgl fragment shader medium float precision rangeMin:127",
            "webgl fragment shader medium float precision rangeMax:127",
            "webgl fragment shader low float precision:23",
            "webgl fragment shader low float precision rangeMin:127",
            "webgl fragment shader low float precision rangeMax:127",
            "webgl vertex shader high int precision:0",
            "webgl vertex shader high int precision rangeMin:31",
            "webgl vertex shader high int precision rangeMax:30",
            "webgl vertex shader medium int precision:0",
            "webgl vertex shader medium int precision rangeMin:31",
            "webgl vertex shader medium int precision rangeMax:30",
            "webgl vertex shader low int precision:0",
            "webgl vertex shader low int precision rangeMin:31",
            "webgl vertex shader low int precision rangeMax:30",
            "webgl fragment shader high int precision:0",
            "webgl fragment shader high int precision rangeMin:31",
            "webgl fragment shader high int precision rangeMax:30",
            "webgl fragment shader medium int precision:0",
            "webgl fragment shader medium int precision rangeMin:31",
            "webgl fragment shader medium int precision rangeMax:30",
            "webgl fragment shader low int precision:0",
            "webgl fragment shader low int precision rangeMin:31",
            "webgl fragment shader low int precision rangeMax:30"
        ]
    },
    {
        "key": "webglVendorAndRenderer",
        "value": "Intel Inc.~Intel(R) Iris(TM) Plus Graphics 655"
    },
    {
        "key": "adBlock",
        "value": false
    },
    {
        "key": "hasLiedLanguages",
        "value": false
    },
    {
        "key": "hasLiedResolution",
        "value": false
    },
    {
        "key": "hasLiedOs",
        "value": false
    },
    {
        "key": "hasLiedBrowser",
        "value": false
    },
    {
        "key": "touchSupport",
        "value": [
            0,
            false,
            false
        ]
    },
    {
        "key": "fonts",
        "value": [
            "Andale Mono",
            "Arial",
            "Arial Black",
            "Arial Hebrew",
            "Arial Narrow",
            "Arial Rounded MT Bold",
            "Arial Unicode MS",
            "Comic Sans MS",
            "Courier",
            "Courier New",
            "Geneva",
            "Georgia",
            "Helvetica",
            "Helvetica Neue",
            "Impact",
            "LUCIDA GRANDE",
            "Microsoft Sans Serif",
            "Monaco",
            "Palatino",
            "Tahoma",
            "Times",
            "Times New Roman",
            "Trebuchet MS",
            "Verdana",
            "Wingdings",
            "Wingdings 2",
            "Wingdings 3"
        ]
    },
    {
        "key": "audio",
        "value": "124.0434806260746"
    }
]

将它转换为string类型,然后做x64sha128运算

image-20210417145230102

browser_id长度是32位,我们也可以完全伪造一个32位的,用个随机数+md5就可以了。

image-20210417145248157

Python调用代码

最后给出python调用的代码

python
import base64
import hashlib
import json

import requests
from django.utils.crypto import get_random_string


def rot13(params):
    t = "NOPQRSTUVWXYZABCDEFGHIJKLMnopqrstuvwxyzabcdefghijklm"
    o = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz"
    s = []
    for i in list(params):
        index = o.find(i)
        if index > -1:
            s.append(t[index])
        else:
            s.append(i)
    return ''.join(s)


def caiyun_decode(s):
    s = rot13(s)
    s2 = base64.b64decode(s).decode('utf-8')
    return s2


def generate_jwt():
    url = "https://api.interpreter.caiyunai.com/v1/user/jwt/generate"
    headers = {
        "X-Authorization": "token:qgemv4jr1y38jyq6vhvi",
        "Content-Type": "application/json;charset=UTF-8",
        "app-name": "xy"
    }
    encrypt = hashlib.md5()
    encrypt.update(get_random_string(10, "qweasdzxcrtyfghvbnuiopjklm1234567890").encode())
    result = encrypt.hexdigest()
    data = {"browser_id": result}
    r = requests.post(url, data=json.dumps(data), headers=headers)
    return r.json()["jwt"], result


def translater(s, jwt, brower_id):
    url = "https://api.interpreter.caiyunai.com/v1/translator"
    headers = {
        "X-Authorization": "token:qgemv4jr1y38jyq6vhvi",
        "Content-Type": "application/json;charset=UTF-8",
        "app-name": "xy",
        "T-Authorization": jwt,
    }
    data = {"source": s.splitlines(), "trans_type": "en2zh", "request_id": "web_fanyi", "media": "text",
            "os_type": "web",
            "dict": False, "cached": True, "replaced": True, "browser_id": brower_id}
    r = requests.post(url, data=json.dumps(data), headers=headers)
    return r.json()


if __name__ == '__main__':
    s = caiyun_decode("5bvE55Fb5YvN5Yvd5oPC5bdN5ora6Xrw5Lnm5YdT6Y+M5Yvd6Mrh6nXL77lZ5bvE5LnM5YdT6Y+M5Yvd57T7Bt==")
    print(s)
    s, s1 = generate_jwt()
    print(s)
    s = translater("hello", s, s1)
    print(s)
    for item in s["target"]:
        print(caiyun_decode(item))
    # "5oiR55So5LiA5Liq5bCP5oqA5ben6Kej5Yaz5LqG6L+Z5Liq6Zeu6aKY77yM5oiR5YaZ5LqG6L+Z5Liq57G7Og=="

输出结果

image-20210417145824095

撰写