代码 Python - ICP备案查询分析

Code Python
发布时间 :
字数:2.2k 阅读 :
  1. 请求分析
    1. __jsluid_s
    2. Uuid
    3. clientUid
    4. Token
  2. 验证码分析
    1. 点选
    2. 识别
  3. 转换
  4. 参考

请求分析

首先从最终的查询数据包来看

POST /icpproject_query/api/icpAbbreviateInfo/queryByCondition HTTP/1.1
Host: hlwicpfwc.miit.gov.cn
Cookie: __jsluid_s=ae00a1e14491f21530d47bd8091b587a
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:122.0) Gecko/20100101 Firefox/122.0
Accept: application/json, text/plain, */*
Accept-Language: zh-CN,zh;q=0.8,zh-TW;q=0.7,zh-HK;q=0.5,en-US;q=0.3,en;q=0.2
Accept-Encoding: gzip, deflate, br
Content-Type: application/json
Uuid: 42084b64747a4c96b90e7903812a4ec9
Sign: eyJ0eXBlIjozLCJleHREYXRhIjp7InZhZnljb2RlX2ltYWdlX2tleSI6IjQyMDg0YjY0NzQ3YTRjOTZiOTBlNzkwMzgxMmE0ZWM5In0sImUiOjE3MDY2MTAxMjYwNDF9.5phR5zdDcMcoVJN5y-LqTUgWvS11dEoJMpW1cuxNc0E
Rci: 77839f050ddc4596b9e22e3574c237f0
Token: eyJ0eXBlIjoxLCJ1IjoiMDk4ZjZiY2Q0NjIxZDM3M2NhZGU0ZTgzMjYyN2I0ZjYiLCJzIjoxNzA2NjA5NjE3NzMxLCJlIjoxNzA2NjEwMDk3NzMxfQ.pOCWH8ru6VkpgS3rbUk607mSGYlDqxpMrTI0ADrcc44
Content-Length: 96
Origin: https://beian.miit.gov.cn
Dnt: 1
Sec-Gpc: 1
Referer: https://beian.miit.gov.cn/
Sec-Fetch-Dest: empty
Sec-Fetch-Mode: cors
Sec-Fetch-Site: same-site
Te: trailers
Connection: close

{"pageNum":2,"pageSize":10,"unitName":"XXX有限公司","serviceType":1}

请求包中的意义不明的就只有这五段了

  1. Cookie: __jsluid_s=ae00a1e14491f21530d47bd8091b587a
  2. Uuid: 42084b64747a4c96b90e7903812a4ec9
  3. Sign: eyJ0eXBlIjozLCJleHREYXRhIjp7InZhZnljb2RlX2ltYWdlX2tleSI6IjQyMDg0YjY0NzQ3YTRjOTZiOTBlNzkwMzgxMmE0ZWM5In0sImUiOjE3MDY2MTAxMjYwNDF9.5phR5zdDcMcoVJN5y-LqTUgWvS11dEoJMpW1cuxNc0E
  4. Rci: 77839f050ddc4596b9e22e3574c237f0
  5. Token: eyJ0eXBlIjoxLCJ1IjoiMDk4ZjZiY2Q0NjIxZDM3M2NhZGU0ZTgzMjYyN2I0ZjYiLCJzIjoxNzA2NjA5NjE3NzMxLCJlIjoxNzA2NjEwMDk3NzMxfQ.pOCWH8ru6VkpgS3rbUk607mSGYlDqxpMrTI0ADrcc44

__jsluid_s

搜索ae00a1e14491f21530d47bd8091b587a，找到第一个数据包

HTTP/1.1 200 OK
Date: Tue, 30 Jan 2024 10:16:37 GMT
Content-Type: text/html;charset=UTF-8
Connection: close
Vary: Accept-Encoding
Vary: Accept-Encoding
Vary: Origin
Vary: Access-Control-Request-Method
Vary: Access-Control-Request-Headers
Access-Control-Allow-Origin: https://beian.miit.gov.cn
Access-Control-Expose-Headers: Access-Control-Allow-Origin, Access-Control-Allow-Credentials, rci
Access-Control-Allow-Credentials: true
Strict-Transport-Security: max-age=15724800; includeSubDomains
X-Via-JSL: ba1114a,-
Set-Cookie: __jsluid_s=ae00a1e14491f21530d47bd8091b587a; max-age=31536000; path=/; HttpOnly; SameSite=None; secure
X-Cache: bypass
Content-Length: 419

{"code":200,"msg":"操作成功","params":{"bussiness":"eyJ0eXBlIjoxLCJ1IjoiMDk4ZjZiY2Q0NjIxZDM3M2NhZGU0ZTgzMjYyN2I0ZjYiLCJzIjoxNzA2NjA5NjE3NzMxLCJlIjoxNzA2NjEwMDk3NzMxfQ.pOCWH8ru6VkpgS3rbUk607mSGYlDqxpMrTI0ADrcc44","expire":300000,"refresh":"eyJ0eXBlIjoyLCJ1IjoiMDk4ZjZiY2Q0NjIxZDM3M2NhZGU0ZTgzMjYyN2I0ZjYiLCJzIjoxNzA2NjA5NjE3NzMxLCJlIjoxNzA2NjEwMzk3NzMxfQ.xnbwyV593KAWwv9bTIQKgqDPdzRssNAyteKbbjkBsRw"},"success":true}

对应的请求包中有个不知从何而来的0d5b07c64a82f07dc6e9ec472e39458f

POST /icpproject_query/api/auth HTTP/1.1
Host: hlwicpfwc.miit.gov.cn
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:122.0) Gecko/20100101 Firefox/122.0
Accept: */*
Accept-Language: zh-CN,zh;q=0.8,zh-TW;q=0.7,zh-HK;q=0.5,en-US;q=0.3,en;q=0.2
Accept-Encoding: gzip, deflate, br
Content-Type: application/x-www-form-urlencoded; charset=UTF-8
Content-Length: 64
Origin: https://beian.miit.gov.cn
Dnt: 1
Sec-Gpc: 1
Referer: https://beian.miit.gov.cn/
Sec-Fetch-Dest: empty
Sec-Fetch-Mode: cors
Sec-Fetch-Site: same-site
Te: trailers
Connection: close

authKey=0d5b07c64a82f07dc6e9ec472e39458f&timeStamp=1706609797163

其中702f32ff914c3cfcf62507d34a2fdda0这个没有搜到其他数据包，于是搜索authKey

从下往上看，最后一个authKey是

接着往上找n

n是w.authKey(g,A,t)，往上可以看到t的值是时间戳(new Date).getTime()（13位），而g和A则是由函数auth传入

authKey就是把g,A,t三个值拼接后计算出md5

继续查找函数auth看到g和A均为字符串test

打个断点看看

与实验结果一致

将计算值替换到请求包中重放能够获得正确的响应包

不正确的话如下

Uuid

搜索42084b64747a4c96b90e7903812a4ec9，找到第一个数据包

Uuid的值来自响应数据包的json格式body中的params[‘uuid’]

HTTP/1.1 200 OK
Date: Tue, 30 Jan 2024 10:16:55 GMT
Content-Type: application/json
Connection: close
Vary: Accept-Encoding
Vary: Accept-Encoding
Vary: Origin
Vary: Access-Control-Request-Method
Vary: Access-Control-Request-Headers
Access-Control-Allow-Origin: https://beian.miit.gov.cn
Access-Control-Expose-Headers: Access-Control-Allow-Origin, Access-Control-Allow-Credentials, rci
Access-Control-Allow-Credentials: true
Strict-Transport-Security: max-age=15724800; includeSubDomains
X-Via-JSL: 559ed21,-
X-Cache: bypass
Content-Length: 269568

{"code":200,"msg":"操作成功","params":{"bigImage":"iVBORw0KGgo……BJRU5ErkJggg==","secretKey":"QctFZYMwYcKSku7Y","smallImage":"iVBORw0KGgo……ElFTkSuQmCC","uuid":"42084b64747a4c96b90e7903812a4ec9","wordCount":4},"success":true}

对应的请求包为

POST /icpproject_query/api/image/getCheckImagePoint HTTP/1.1
Host: hlwicpfwc.miit.gov.cn
Cookie: __jsluid_s=ae00a1e14491f21530d47bd8091b587a
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:122.0) Gecko/20100101 Firefox/122.0
Accept: application/json, text/plain, */*
Accept-Language: zh-CN,zh;q=0.8,zh-TW;q=0.7,zh-HK;q=0.5,en-US;q=0.3,en;q=0.2
Accept-Encoding: gzip, deflate, br
Content-Type: application/json
Token: eyJ0eXBlIjoxLCJ1IjoiMDk4ZjZiY2Q0NjIxZDM3M2NhZGU0ZTgzMjYyN2I0ZjYiLCJzIjoxNzA2NjA5NjE3NzMxLCJlIjoxNzA2NjEwMDk3NzMxfQ.pOCWH8ru6VkpgS3rbUk607mSGYlDqxpMrTI0ADrcc44
Content-Length: 58
Origin: https://beian.miit.gov.cn
Dnt: 1
Sec-Gpc: 1
Referer: https://beian.miit.gov.cn/
Sec-Fetch-Dest: empty
Sec-Fetch-Mode: cors
Sec-Fetch-Site: same-site
Te: trailers
Connection: close

{"clientUid":"point-a3f7374f-986a-4b6a-87bf-ba035ae77c9b"}

发现此处已经有Token和clientUid两个值，那下面继续分析这两个哪来的

clientUid

搜索clientUid找到发送请求包对应的函数，发现是读取localStorage的值

既然是getItem(“point”)那就搜setItem(“point”

貌似是随机得到的

随便生成一个uuid的值，确实也可以得到正确的响应

Token

拿着eyJ0eXB……NzMxfQ搜索发现第一个包就是之前获取Cookie值__jsluid_s的响应包😅

看来可以在拿Cookie时顺便搞定他

Sign

搜索eyJ0eXB……MjYwNDF9，发现的第一个数据包如下

对应的请求包

POST /icpproject_query/api/image/checkImage HTTP/1.1
Host: hlwicpfwc.miit.gov.cn
Cookie: __jsluid_s=ae00a1e14491f21530d47bd8091b587a
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:122.0) Gecko/20100101 Firefox/122.0
Accept: application/json, text/plain, */*
Accept-Language: zh-CN,zh;q=0.8,zh-TW;q=0.7,zh-HK;q=0.5,en-US;q=0.3,en;q=0.2
Accept-Encoding: gzip, deflate, br
Content-Type: application/json
Token: eyJ0eXBlIjoxLCJ1IjoiMDk4ZjZiY2Q0NjIxZDM3M2NhZGU0ZTgzMjYyN2I0ZjYiLCJzIjoxNzA2NjA5NjE3NzMxLCJlIjoxNzA2NjEwMDk3NzMxfQ.pOCWH8ru6VkpgS3rbUk607mSGYlDqxpMrTI0ADrcc44
Content-Length: 255
Origin: https://beian.miit.gov.cn
Dnt: 1
Sec-Gpc: 1
Referer: https://beian.miit.gov.cn/
Sec-Fetch-Dest: empty
Sec-Fetch-Mode: cors
Sec-Fetch-Site: same-site
Te: trailers
Connection: close

{"token":"42084b64747a4c96b90e7903812a4ec9","secretKey":"QctFZYMwYcKSku7Y","clientUid":"point-a3f7374f-986a-4b6a-87bf-ba035ae77c9b","pointJson":"0CLJn2Mxxj1rsH3COFz0roWmZDHzgW+vlUt6vsQljB6o9V6j6ncEFMkNT68eqbwIXmIuFNhe8u7cfGFhFOAY2Fny68afmFITf+KFccUur6U="}

然后发现token、secretKey和clientUid都来自刚才获取Uuid的数据包中，看来pointJson就是验证码的值了

搜索pointJson，看到pointJson是函数h的结果

往上找传入参数g.checkPointArr发现是一个数组，g.secretKey就是请求包中secretKey的值

函数h是m的实例化，m看着就是个aes加密函数

A传入为空的话就赋值XwKsGlMcdPMEhR1B，偏移量无疑了

尝试打断点，与判断结果一致

剩下的就是看怎么识别验证码了

验证码分析

点选

试了下ddddocr对bigImage.png的文字的识别率，基本是没法看

但是点选位置的识别率还是很不错的

在某个项目下面看到有介绍用孪生神经网络来实现的，然后看了一下b站很多点选验证码也是用这个思路实现

smallImage.png的文字位置和间距都固定，每个字体大概是24x24

每个字的起始点（以左上为准）在(167, 13)，(200, 13)，(233, 13)，(266, 13)

因为每个汉字都会有些许差别，所以用逗号的间隔9来，刚好加上汉字的24是33

识别

搜索了一下这类验证码的识别，基本都是使用孪生网络的方案

首先写个代码把验证码中bigImage的点选文字和smallImage的待点选文字识别并截取出来做样本

Pytorch 搭建自己的孪生神经网络比较图片相似性平台（Bubbliiiing 深度学习 教程）（环境要求Python3.7，高了依赖装不上😅）

  1. 克隆仓库
git clone https://github.com/bubbliiiing/Siamese-pytorch.git

  1. pyenv安装Python3.7（最高版本Windows：3.7.9，Linux：3.7.17）

    pyenv install 3.7.17

  2. 创建虚拟

    cd Siamese-pytorch && python -m venv ./venv

  3. 安装依赖

    pip install torch==1.2.0 torchvision==0.4.0 -f https://download.pytorch.org/whl/torch_stable.html

    pip install -r requirements.txt

下载vgg16-397923af.pth（VGG预训练模型）保存至model_data目录下

train.py

diff --git a/train.py b/train.py  
index 9288dc3..ceb6f3f 100644  
--- a/train.py  
+++ b/train.py  
@@ -20,7 +20,7 @@ if __name__ == "__main__":  
    #   是否使用Cuda  
    #   没有GPU可以设置成False  
    #----------------------------------------------------#  
-    Cuda            = True  
+    Cuda            = False  
    #---------------------------------------------------------------------#  
    #   distributed     用于指定是否使用单机多卡分布式运行-  
    #                   终端指令仅支持Ubuntu。CUDA_VISIBLE_DEVICES用于在Ubuntu下指定显卡。-  
@@ -41,7 +41,7 @@ if __name__ == "__main__":  
    #   fp16        是否使用混合精度训练_  
    #               可减少约一半的显存、需要pytorch1.7.1以上D  
    #---------------------------------------------------------------------#  
-    fp16            = False  
+    fp16            = True  
    #----------------------------------------------------#  
    #   数据集存放的路径-  
    #----------------------------------------------------#  
@@ -84,7 +84,7 @@ if __name__ == "__main__":  
    #   网络一般不从0开始训练，至少会使用主干部分的权值，有些论文提到可以不用预训练，主要原因是他们 数据集较大 且 调参能力优秀。-  
    #   如果一定要训练网络的主干部分，可以了解imagenet数据集，首先训练分类模型，分类模型的 主干部分 和该模型通用，基于此进行训练。#  
    #----------------------------------------------------------------------------------------------------------------------------#  
-    model_path      = ""  
+    model_path      = "model_data/vgg16-397923af.pth"  
   
    #----------------------------------------------------------------------------------------------------------------------------#  
    #   显存不足与数据集大小无关，提示显存不足请调小batch_size。-

因为没N卡就把Cuda关了，依照视频教程训练会在logs目录下生成best_epoch_weights.pth

转换

参考微软文档将pth格式的PyTorch专用训练模型转换为onnx的通用模型文件

import torch.onnx 

#Function to Convert to ONNX 
def Convert_ONNX(): 
    
    # 导出模型之前必须调用 model.eval() 或 model.train(False)
    # 将模型设置为推理模式
    model.eval() 

    # 创建一个虚拟输入张量
    dummy_input = torch.randn(1, input_size, requires_grad=True)  

    # 导出模型 
    torch.onnx.export(model,  # model being run 
         dummy_input,  # 模型输入
         "ImageClassifier.onnx",  # 保存模型的位置
         export_params=True,  # 在模型文件中存储训练好的参数权重
         opset_version=10,  # 导出模型的 ONNX 版本
         do_constant_folding=True,  # 是否执行常量折叠优化
         input_names = ['modelInput'],  # 模型的输入名称
         output_names = ['modelOutput'],  # 模型的输出名称
         dynamic_axes={'modelInput' : {0 : 'batch_size'}, 'modelOutput' : {0 : 'batch_size'}})  # 变长轴
    print('Model has been converted to ONNX')

if __name__ == "__main__": 

    model = Network() 
    path = "myFirstModel.pth" 
    model.load_state_dict(torch.load(path)) 
 
    # Conversion to ONNX 
    Convert_ONNX()

参考

某查询网站点选逆向分析
将 PyTorch 训练模型转换为 ONNX | Microsoft Learn

转载请注明来源，欢迎对文章中的引用来源进行考证，欢迎指出任何有错误或不够清晰的表达。可以在下面评论区评论，也可以邮件至 cnlnnn@qq.com

文章标题:代码 Python - ICP备案查询分析

字数:2.2k

本文作者:cnlnn

发布时间:2024-02-02, 16:15:00

最后更新:2024-10-06, 14:21:44

原始链接:https://cnlnn.pages.dev/posts/icp_query_analyze/

版权声明: "署名-非商用-相同方式共享 4.0" 转载请保留原文链接及作者。

©2024-2024 cnlnn

Built with Hexo and 3-hexo theme