Skip to content

crawlergo 动态爬虫 源码学习

字数
5684 字
阅读时间
28 分钟
更新日期
9/24/2021

crawlergo是一个使用chrome headless模式进行URL收集的浏览器爬虫。它对整个网页的关键位置与DOM渲染阶段进行HOOK,自动进行表单填充并提交,配合智能的JS事件触发,尽可能的收集网站暴露出的入口。内置URL去重模块,过滤掉了大量伪静态URL,对于大型网站仍保持较快的解析与抓取速度,最后得到高质量的请求结果集合。

crawlergo 目前支持以下特性:

  • 原生浏览器环境,协程池调度任务

  • 表单智能填充、自动化提交

  • 完整DOM事件收集,自动化触发

  • 智能URL去重,去掉大部分的重复请求

  • 全面分析收集,包括javascript文件内容、页面注释、robots.txt文件和常见路径Fuzz

  • 支持Host绑定,自动添加Referer

  • 支持请求代理,支持爬虫结果主动推送

Github: https://github.com/Qianlitp/crawlergo

作者开源了源码,我是很兴奋的,以前也有写一个的想法,但是开源的动态爬虫不多,看了其中几个。

调研

  1. https://github.com/fcavallarin/htcap
    • 递归dom搜索引擎
    • 发现ajax/fetch/jsonp/websocket请求
    • 支持cookie,代理,ua,http auth
    • 基于文本相似度的页面重复数据删除引擎
      • 根据文本长度 <256
        • simhash
      • else
        • ShinglePrint
    • 主要代码是python调用puppeteer,但是核心逻辑在js里
  2. https://github.com/go-rod/rod
    • 一个操作chrome headless的go库
    • 它比官方提供的chrome操作库更容易使用
    • 有效解决了chrome残留僵尸进程的问题
  3. https://github.com/lc/gau
    • 通过一些通用接口获取url信息
  4. https://github.com/jaeles-project/gospider
    • Web静态爬虫,也提供了一些方法获取更多URL
  5. https://github.com/chaitin/rad 1. rad虽然没有开源,但是它里面使用yaml进行的配置选项很多,通过配置选项可以大致知道它的一些特性。 2. 可以手动登陆 3. 启用图片 4. 显示对爬取url的一些限制
    1. 不允许的文件后缀
    2. 不允许的url关键字
    3. 不允许的域名
    4. 不允许的url
5. 设置下个页面最大点击和事件触发

Crawlergo

之前也想过写一个动态爬虫来对接扫描器,但是动态爬虫有很多细节都需要打磨,一直没时间做,现在有现成的源码参考能省下不少事。

主要看几个点

  • 对浏览器 JavaScript环境的hoook

  • dom的触发,表单填充

  • url如何去重

  • url的收集

目录结构

bash
├─cmd
  └─crawlergo # 程序主入口
├─examples
├─imgs
└─pkg
    ├─config  # 一些配置相关
    ├─engine  # chrome相关程序
    ├─filter  # 去重相关
    ├─js      # 一些注入的js
    ├─logger  # 日志
    ├─model   # url和请求相关的库
    └─tools   # 一些通用类库
        └─requests

根据源码的调用堆栈做了一个程序启动流程图

crawlergo

配置文件

pkg/config/config.go定义了一些默认的配置文件

go
const (
    DefaultUA               = "Mozilla/5.0 (Windows NT 6.1; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/79.0.3945.0 Safari/537.36"
    MaxTabsCount            = 10
    TabRunTimeout           = 20 * time.Second
    DefaultInputText        = "Crawlergo" // 默认输入的文字
    FormInputKeyword        = "Crawlergo" // form输入的文字,但是代码中没有引用这个变量的
    SuspectURLRegex         = `(?:"|')(((?:[a-zA-Z]{1,10}://|//)[^"'/]{1,}\.[a-zA-Z]{2,}[^"']{0,})|((?:/|\.\./|\./)[^"'><,;|*()(%%$^/\\\[\]][^"'><,;|()]{1,})|([a-zA-Z0-9_\-/]{1,}/[a-zA-Z0-9_\-/]{1,}\.(?:[a-zA-Z]{1,4}|action)(?:[\?|#][^"|']{0,}|))|([a-zA-Z0-9_\-/]{1,}/[a-zA-Z0-9_\-/]{3,}(?:[\?|#][^"|']{0,}|))|([a-zA-Z0-9_\-]{1,}\.(?:php|asp|aspx|jsp|json|action|html|js|txt|xml)(?:[\?|#][^"|']{0,}|)))(?:"|')` // url获取正则
    URLRegex                = `((https?|ftp|file):)?//[-A-Za-z0-9+&@#/%?=~_|!:,.;]+[-A-Za-z0-9+&@#/%=~_|]` // url获取正则
    AttrURLRegex            = ``
    DomContentLoadedTimeout = 5 * time.Second
    EventTriggerInterval    = 100 * time.Millisecond // 单位毫秒
    BeforeExitDelay         = 1 * time.Second
    DefaultEventTriggerMode = EventTriggerAsync
    MaxCrawlCount           = 200
)

// 请求的来源,记录了每个url的来源,可以根据这些关键词定位到相关的获取代码
const (
    FromTarget      = "Target"     //初始输入的目标
    FromNavigation  = "Navigation" //页面导航请求
    FromXHR         = "XHR"        //ajax异步请求
    FromDOM         = "DOM"        //dom解析出来的请求
    FromJSFile      = "JavaScript" //JS脚本中解析
    FromFuzz        = "PathFuzz"   //初始path fuzz
    FromRobots      = "robots.txt" //robots.txt
    FromComment     = "Comment"    //页面中的注释
    FromWebSocket   = "WebSocket"
    FromEventSource = "EventSource"
    FromFetch       = "Fetch"
    FromHistoryAPI  = "HistoryAPI"
    FromOpenWindow  = "OpenWindow"
    FromHashChange  = "HashChange"
    FromStaticRes   = "StaticResource"
    FromStaticRegex = "StaticRegex"
)

// 静态文件后缀,这些后缀的文件全部阻断读取
var StaticSuffix = []string{
    "png", "gif", "jpg", "mp4", "mp3", "mng", "pct", "bmp", "jpeg", "pst", "psp", "ttf",
    "tif", "tiff", "ai", "drw", "wma", "ogg", "wav", "ra", "aac", "mid", "au", "aiff",
    "dxf", "eps", "ps", "svg", "3gp", "asf", "asx", "avi", "mov", "mpg", "qt", "rm",
    "wmv", "m4a", "bin", "xls", "xlsx", "ppt", "pptx", "doc", "docx", "odt", "ods", "odg",
    "odp", "exe", "zip", "rar", "tar", "gz", "iso", "rss", "pdf", "txt", "dll", "ico",
    "gz2", "apk", "crt", "woff", "map", "woff2", "webp", "less", "dmg", "bz2", "otf", "swf",
    "flv", "mpeg", "dat", "xsl", "csv", "cab", "exif", "wps", "m4v", "rmvb",
}

// 动态文件的后缀,过滤器用于过滤伪静态用
var ScriptSuffix = []string{
    "php", "asp", "jsp", "asa",
}

// 默认不爬的url关键词
var DefaultIgnoreKeywords = []string{"logout", "quit", "exit"}

// 填充表单相关,这些是模糊匹配,匹配到了则使用对应的规则
var AllowedFormName = []string{"default", "mail", "code", "phone", "username", "password", "qq", "id_card", "url", "date", "number"}

var InputTextMap = map[string]map[string]interface{}{
    "mail": {
        "keyword": []string{"mail"},
        "value":   "crawlergo@gmail.com",
    },
    "code": {
        "keyword": []string{"yanzhengma", "code", "ver", "captcha"},
        "value":   "123a",
    },
    "phone": {
        "keyword": []string{"phone", "number", "tel", "shouji"},
        "value":   "18812345678",
    },
    "username": {
        "keyword": []string{"name", "user", "id", "login", "account"},
        "value":   "crawlergo@gmail.com",
    },
    "password": {
        "keyword": []string{"pass", "pwd"},
        "value":   "Crawlergo6.",
    },
    "qq": {
        "keyword": []string{"qq", "wechat", "tencent", "weixin"},
        "value":   "123456789",
    },
    "IDCard": {
        "keyword": []string{"card", "shenfen"},
        "value":   "511702197409284963",
    },
    "url": {
        "keyword": []string{"url", "site", "web", "blog", "link"},
        "value":   "https://crawlergo.nice.cn/",
    },
    "date": {
        "keyword": []string{"date", "time", "year", "now"},
        "value":   "2018-01-01",
    },
    "number": {
        "keyword": []string{"day", "age", "num", "count"},
        "value":   "10",
    },
}

URL过滤方式

crawlergo有两种过滤,氛围simplesmart

simple

simple过滤方式比较简单,就是将计算请求体的method、url、postdata 结合计算md5

image-20210924002811716

判断是否存在

image-20210924002839698

smart

smart过滤会对每个请求的参数name,参数value,path 进行标记,会标记成以下字段

go
const (
    CustomValueMark    = "{{Crawlergo}}"
    FixParamRepeatMark = "{{fix_param}}"
    FixPathMark        = "{{fix_path}}"
    TooLongMark        = "{{long}}"
    NumberMark         = "{{number}}"
    ChineseMark        = "{{chinese}}"
    UpperMark          = "{{upper}}"
    LowerMark          = "{{lower}}"
    UrlEncodeMark      = "{{urlencode}}"
    UnicodeMark        = "{{unicode}}"
    BoolMark           = "{{bool}}"
    ListMark           = "{{list}}"
    TimeMark           = "{{time}}"
    MixAlphaNumMark    = "{{mix_alpha_num}}"
    MixSymbolMark      = "{{mix_symbol}}"
    MixNumMark         = "{{mix_num}}"
    NoLowerAlphaMark   = "{{no_lower}}"
    MixStringMark      = "{{mix_str}}"
)

例如

?m=home&c=index&a=index
?type=202cb962ac59075b964b07152d234b70
?id=1
?msg=%E6%B6%88%E6%81%AF

处理结果即

?m=home&c=index&a=index
?type={hash}
?id={int}
?msg={urlencode}

会对超过阈值的结果,以及计算的唯一id进行比对,来判断是否过滤,总体来说是基于url的过滤方式。

使用的正则

go
var chineseRegex = regexp.MustCompile("[\u4e00-\u9fa5]+")
var urlencodeRegex = regexp.MustCompile("(?:%[A-Fa-f0-9]{2,6})+")
var unicodeRegex = regexp.MustCompile(`(?:\\u\w{4})+`)
var onlyAlphaRegex = regexp.MustCompile("^[a-zA-Z]+$")
var onlyAlphaUpperRegex = regexp.MustCompile("^[A-Z]+$")
var alphaUpperRegex = regexp.MustCompile("[A-Z]+")
var alphaLowerRegex = regexp.MustCompile("[a-z]+")
var replaceNumRegex = regexp.MustCompile(`[0-9]+\.[0-9]+|\d+`)
var onlyNumberRegex = regexp.MustCompile(`^[0-9]+$`)
var numberRegex = regexp.MustCompile(`[0-9]+`)
var OneNumberRegex = regexp.MustCompile(`[0-9]`)
var numSymbolRegex = regexp.MustCompile(`\.|_|-`)
var timeSymbolRegex = regexp.MustCompile(`-|:|\s`)
var onlyAlphaNumRegex = regexp.MustCompile(`^[0-9a-zA-Z]+$`)
var markedStringRegex = regexp.MustCompile(`^{{.+}}$`)
var htmlReplaceRegex = regexp.MustCompile(`\.shtml|\.html|\.htm`)

对路径的处理

go
/**
标记路径
*/
func (s *SmartFilter) MarkPath(path string) string {
    pathParts := strings.Split(path, "/")
    for index, part := range pathParts {
        if len(part) >= 32 {
            pathParts[index] = TooLongMark
        } else if onlyNumberRegex.MatchString(numSymbolRegex.ReplaceAllString(part, "")) {
            pathParts[index] = NumberMark
        } else if strings.HasSuffix(part, ".html") || strings.HasSuffix(part, ".htm") || strings.HasSuffix(part, ".shtml") {
            part = htmlReplaceRegex.ReplaceAllString(part, "")
            // 大写、小写、数字混合
            if numberRegex.MatchString(part) && alphaUpperRegex.MatchString(part) && alphaLowerRegex.MatchString(part) {
                pathParts[index] = MixAlphaNumMark
                // 纯数字
            } else if b := numSymbolRegex.ReplaceAllString(part, ""); onlyNumberRegex.MatchString(b) {
                pathParts[index] = NumberMark
            }
            // 含有特殊符号
        } else if s.hasSpecialSymbol(part) {
            pathParts[index] = MixSymbolMark
        } else if chineseRegex.MatchString(part) {
            pathParts[index] = ChineseMark
        } else if unicodeRegex.MatchString(part) {
            pathParts[index] = UnicodeMark
        } else if onlyAlphaUpperRegex.MatchString(part) {
            pathParts[index] = UpperMark
            // 均为数字和一些符号组成
        } else if b := numSymbolRegex.ReplaceAllString(part, ""); onlyNumberRegex.MatchString(b) {
            pathParts[index] = NumberMark
            // 数字出现的次数超过3,视为伪静态path
        } else if b := OneNumberRegex.ReplaceAllString(part, "0"); strings.Count(b, "0") > 3 {
            pathParts[index] = MixNumMark
        }
    }
    newPath := strings.Join(pathParts, "/")
    return newPath
}

基于网页结构去重

作者原帖中的基于网页结构去重写的非常精彩 https://www.anquanke.com/post/id/178339#h2-17

参考的论文下载: https://patents.google.com/patent/CN101694668B/zh

作者的将网页特征向量抽离出来,索引存储,用来判断大量网页的相似度,非常惊艳,未来用到资产收集系统或者网络空间引擎上也都是非常不错的选择。

URL收集

robots

在爬虫之前,会先请求robots.txt,解析出所有链接,加入到待爬取页面。

源码中使用了一个正则来匹配

go
var urlFindRegex = regexp.MustCompile(`(?:Disallow|Allow):.*?(/.+)`)

看了下robots规范:https://baike.baidu.com/item/robots协议/2483797 ,应该还可以再优化一下,来处理一些表达式。

image-20210924110010389

DIR FUZZ

在爬虫之前,如果没有指定dir字典的话,默认会使用内置的字典

python
['11', '123', '2017', '2018', 'message', 'mis', 'model', 'abstract', 'account', 'act', 'action', 'activity', 'ad', 'address', 'ajax', 'alarm', 'api', 'app', 'ar', 'attachment', 'auth', 'authority', 'award', 'back', 'backup', 'bak', 'base', 'bbs', 'bbs1', 'cms', 'bd', 'gallery', 'game', 'gift', 'gold', 'bg', 'bin', 'blacklist', 'blog', 'bootstrap', 'brand', 'build', 'cache', 'caches', 'caching', 'cacti', 'cake', 'captcha', 'category', 'cdn', 'ch', 'check', 'city', 'class', 'classes', 'classic', 'client', 'cluster', 'collection', 'comment', 'commit', 'common', 'commons', 'components', 'conf', 'config', 'mysite', 'confs', 'console', 'consumer', 'content', 'control', 'controllers', 'core', 'crontab', 'crud', 'css', 'daily', 'dashboard', 'data', 'database', 'db', 'default', 'demo', 'dev', 'doc', 'download', 'duty', 'es', 'eva', 'examples', 'excel', 'export', 'ext', 'fe', 'feature', 'file', 'files', 'finance', 'flashchart', 'follow', 'forum', 'frame', 'framework', 'ft', 'group', 'gss', 'hello', 'helper', 'helpers', 'history', 'home', 'hr', 'htdocs', 'html', 'hunter', 'image', 'img11', 'import', 'improve', 'inc', 'include', 'includes', 'index', 'info', 'install', 'interface', 'item', 'jobconsume', 'jobs', 'json', 'kindeditor', 'l', 'languages', 'lib', 'libraries', 'libs', 'link', 'lite', 'local', 'log', 'login', 'logs', 'mail', 'main', 'maintenance', 'manage', 'manager', 'manufacturer', 'menus', 'models', 'modules', 'monitor', 'movie', 'mysql', 'n', 'nav', 'network', 'news', 'notice', 'nw', 'oauth', 'other', 'page', 'pages', 'passport', 'pay', 'pcheck', 'people', 'person', 'php', 'phprpc', 'phptest', 'picture', 'pl', 'platform', 'pm', 'portal', 'post', 'product', 'project', 'protected', 'proxy', 'ps', 'public', 'qq', 'question', 'quote', 'redirect', 'redisclient', 'report', 'resource', 'resources', 's', 'save', 'schedule', 'schema', 'script', 'scripts', 'search', 'security', 'server', 'service', 'shell', 'show', 'simple', 'site', 'sites', 'skin', 'sms', 'soap', 'sola', 'sort', 'spider', 'sql', 'stat', 'static', 'statistics', 'stats', 'submit', 'subways', 'survey', 'sv', 'syslog', 'system', 'tag', 'task', 'tasks', 'tcpdf', 'template', 'templates', 'test', 'tests', 'ticket', 'tmp', 'token', 'tool', 'tools', 'top', 'tpl', 'txt', 'upload', 'uploadify', 'uploads', 'url', 'user', 'util', 'v1', 'v2', 'vendor', 'view', 'views', 'web', 'weixin', 'widgets', 'wm', 'wordpress', 'workspace', 'ws', 'www', 'www2', 'wwwroot', 'zone', 'admin', 'admin_bak', 'mobile', 'm', 'js']

根据状态码判断

image-20210924110712189

爬虫时的url收集

  • 解析流量中的url

    • 获取xmr类型的请求

    • 正则解析js、html、json中url

    • 收集当前页面上的url信息

    "src", "href", "data-url", "data-href"
    object[data]
    注释中的url
    • image-20210924115821009

对浏览器环境的hook

在chrome的tab初始化时,会执行一段js代码,hook部分函数和事件来控制js的运行环境。

同时定义了一个js全局函数addLinkTest,通过这个函数可以与go进行交互。

image-20210924143623439

调用go函数

js初始化时对这个函数重新包装

js
const binding = window["addLink"];
window["addLink"] = async(...args) => {
    const me = window["addLink"];
    let callbacks = me['callbacks'];
    if (!callbacks) {
        callbacks = new Map();
        me['callbacks'] = callbacks;
    }
    const seq = (me['lastSeq'] || 0) + 1;
    me['lastSeq'] = seq;
    const promise = new Promise(fulfill => callbacks.set(seq, fulfill));
    binding(JSON.stringify({name: "addLink", seq, args}));
    return promise;
};

const bindingTest = window["Test"];
window["Test"] = async(...args) => {
    const me = window["Test"];
    let callbacks = me['callbacks'];
    if (!callbacks) {
        callbacks = new Map();
        me['callbacks'] = callbacks;
    }
    const seq = (me['lastSeq'] || 0) + 1;
    me['lastSeq'] = seq;
    const promise = new Promise(fulfill => callbacks.set(seq, fulfill));
    binding(JSON.stringify({name: "Test", seq, args}));
    return promise;
};

go处理逻辑

image-20210924144038030

执行完go函数后会再执行一段js

js
const DeliverResultJS = `
(function deliverResult(name, seq, result) {
    window[name]['callbacks'].get(seq)(result);
    window[name]['callbacks'].delete(seq);
})("%s", %v, "%s")

但是没看懂使用promise后回调调用的意义是什么。。

Bypass headless detect

js
// Pass the Webdriver Test.
Object.defineProperty(navigator, 'webdriver', {
    get: () => false,
});

// Pass the Plugins Length Test.
// Overwrite the plugins property to use a custom getter.
Object.defineProperty(navigator, 'plugins', {
    // This just needs to have length > 0 for the current test,
    // but we could mock the plugins too if necessary.
    get: () => [1, 2, 3, 4, 5],
});

// Pass the Chrome Test.
// We can mock this in as much depth as we need for the test.
window.chrome = {
    runtime: {},
};

// Pass the Permissions Test.
const originalQuery = window.navigator.permissions.query;
window.navigator.permissions.query = (parameters) => (
    parameters.name === 'notifications' ?
    Promise.resolve({ state: Notification.permission }) :
    originalQuery(parameters)
);

//Pass the Permissions Test. navigator.userAgent
Object.defineProperty(navigator, 'userAgent', {
    get: () => "Mozilla/5.0 (Windows NT 6.1; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/79.0.3945.0 Safari/537.36",
});

// 修改浏览器对象的属性
Object.defineProperty(navigator, 'platform', {
    get: function () { return 'win32'; }
});

Object.defineProperty(navigator, 'language', {
    get: function () { return 'zh-CN'; }
});

Object.defineProperty(navigator, 'languages', {
    get: function () { return ["zh-CN", "zh"]; }
});

记录url以及对前端框架的适配

js
// history api hook 许多前端框架都采用此API进行页面路由,记录url并取消操作
window.history.pushState = function(a, b, c) { 
    window.addLink(c, "HistoryAPI");
}
window.history.replaceState = function(a, b, c) { 
    window.addLink(c, "HistoryAPI");
}
Object.defineProperty(window.history,"pushState",{"writable": false, "configurable": false});
Object.defineProperty(window.history,"replaceState",{"writable": false, "configurable": false});
// 监听hash改变 Vue等框架默认使用hash部分进行前端页面路由
window.addEventListener("hashchange", function() {
    window.addLink(document.location.href, "HashChange");
});

// 监听窗口的打开和关闭,记录新窗口打开的url,并取消实际操作
// hook window.open 
window.open = function (url) {
    console.log("trying to open window.");
    window.addLink(url, "OpenWindow");
}
Object.defineProperty(window,"open",{"writable": false, "configurable": false});

// hook window close
window.close = function() {console.log("trying to close page.");};
Object.defineProperty(window,"close",{"writable": false, "configurable": false});

// hook window.WebSocket 、window.EventSource 、 window.fetch 等函数
var oldWebSocket = window.WebSocket;
window.WebSocket = function(url, arg) {
    window.addLink(url, "WebSocket");
    return new oldWebSocket(url, arg);
}

var oldEventSource = window.EventSource;
window.EventSource = function(url) {
    window.addLink(url, "EventSource");
    return new oldEventSource(url);
}

var oldFetch = window.fetch;
window.fetch = function(url) {
    window.addLink(url, "Fetch");
    return oldFetch(url);
}

hook setTimeout/SetInterval

js
// hook setTimeout
//window.__originalSetTimeout = window.setTimeout;
//window.setTimeout = function() {
//    arguments[1] = 0;
//    return window.__originalSetTimeout.apply(this, arguments);
//};
//Object.defineProperty(window,"setTimeout",{"writable": false, "configurable": false});

// hook setInterval 时间设置为60秒 目的是减轻chrome的压力
window.__originalSetInterval = window.setInterval;
window.setInterval = function() {
    arguments[1] = 60000;
    return window.__originalSetInterval.apply(this, arguments);
};
Object.defineProperty(window,"setInterval",{"writable": false, "configurable": false});

这个hook操作没有明白,将setInterval强制设为了60s,我想应该来个判断,大于60s时再统一设置,60s,这样爬虫效率就变得太低了。

setTimeout取消了hook,毕竟只执行一次,可能不是很重要。

hook ajax

hook 原生ajax并限制最大请求数,可能是怕自动点击造成ajax爆炸

js
// 劫持原生ajax,并对每个请求设置最大请求次数
window.ajax_req_count_sec_auto = {};
XMLHttpRequest.prototype.__originalOpen = XMLHttpRequest.prototype.open;
XMLHttpRequest.prototype.open = function(method, url, async, user, password) {
    // hook code
    this.url = url;
    this.method = method;
    let name = method + url;
    if (!window.ajax_req_count_sec_auto.hasOwnProperty(name)) {
        window.ajax_req_count_sec_auto[name] = 1
    } else {
        window.ajax_req_count_sec_auto[name] += 1
    }

    if (window.ajax_req_count_sec_auto[name] <= 10) {
        return this.__originalOpen(method, url, true, user, password);
    }
}
Object.defineProperty(XMLHttpRequest.prototype,"open",{"writable": false, "configurable": false});

XMLHttpRequest.prototype.__originalSend = XMLHttpRequest.prototype.send;
XMLHttpRequest.prototype.send = function(data) {
    // hook code
    let name = this.method + this.url;
    if (window.ajax_req_count_sec_auto[name] <= 10) {
        return this.__originalSend(data);
    }
}
Object.defineProperty(XMLHttpRequest.prototype,"send",{"writable": false, "configurable": false});

XMLHttpRequest.prototype.__originalAbort = XMLHttpRequest.prototype.abort;
XMLHttpRequest.prototype.abort = function() {
    // hook code
}
Object.defineProperty(XMLHttpRequest.prototype,"abort",{"writable": false, "configurable": false});

锁定表单重置

爬虫在处理网页时,会先填充表单,接着触发事件去提交表单,但有时会意外点击到表单的重置按钮,造成内容清空,表单提交失败。所以为了防止这种情况的发生,需要Hook表单的重置并锁定不能修改。

js
// 锁定表单重置
HTMLFormElement.prototype.reset = function() {console.log("cancel reset form")};
Object.defineProperty(HTMLFormElement.prototype,"reset",{"writable": false, "configurable": false});

事件Hook

对dom0级和dom2级事件分别hook,用一个数组设定最大触发次数,所有js调用的事件,会对每个标签设置一个sec_auto_dom2_event_flag属性,方便后面寻找并自动触发。

javascript
// hook dom2 级事件监听
window.add_even_listener_count_sec_auto = {};
// record event func , hook addEventListener
let old_event_handle = Element.prototype.addEventListener;
Element.prototype.addEventListener = function(event_name, event_func, useCapture) {
    let name = "<" + this.tagName + "> " + this.id + this.name + this.getAttribute("class") + "|" + event_name;
    // console.log(name)
    // 对每个事件设定最大的添加次数,防止无限触发,最大次数为5
    if (!window.add_even_listener_count_sec_auto.hasOwnProperty(name)) {
        window.add_even_listener_count_sec_auto[name] = 1;
    } else if (window.add_even_listener_count_sec_auto[name] == 5) {
        return ;
    } else {
        window.add_even_listener_count_sec_auto[name] += 1;
    }
    if (this.hasAttribute("sec_auto_dom2_event_flag")) {
        let sec_auto_dom2_event_flag = this.getAttribute("sec_auto_dom2_event_flag");
        this.setAttribute("sec_auto_dom2_event_flag", sec_auto_dom2_event_flag + "|" + event_name);
    } else {
        this.setAttribute("sec_auto_dom2_event_flag", event_name);
    }
    old_event_handle.apply(this, arguments);
};

function dom0_listener_hook(that, event_name) {
    let name = "<" + that.tagName + "> " + that.id + that.name + that.getAttribute("class") + "|" + event_name;
    // console.log(name);
    // 对每个事件设定最大的添加次数,防止无限触发,最大次数为5
    if (!window.add_even_listener_count_sec_auto.hasOwnProperty(name)) {
        window.add_even_listener_count_sec_auto[name] = 1;
    } else if (window.add_even_listener_count_sec_auto[name] == 5) {
        return ;
    } else {
        window.add_even_listener_count_sec_auto[name] += 1;
    }
    if (that.hasAttribute("sec_auto_dom2_event_flag")) {
        let sec_auto_dom2_event_flag = that.getAttribute("sec_auto_dom2_event_flag");
        that.setAttribute("sec_auto_dom2_event_flag", sec_auto_dom2_event_flag + "|" + event_name);
    } else {
        that.setAttribute("sec_auto_dom2_event_flag", event_name);
    }
}

// hook dom0 级事件监听
Object.defineProperties(HTMLElement.prototype, {
    onclick: {set: function(newValue){onclick = newValue;dom0_listener_hook(this, "click");}},
    onchange: {set: function(newValue){onchange = newValue;dom0_listener_hook(this, "change");}},
    onblur: {set: function(newValue){onblur = newValue;dom0_listener_hook(this, "blur");}},
    ondblclick: {set: function(newValue){ondblclick = newValue;dom0_listener_hook(this, "dbclick");}},
    onfocus: {set: function(newValue){onfocus = newValue;dom0_listener_hook(this, "focus");}},
    onkeydown: {set: function(newValue){onkeydown = newValue;dom0_listener_hook(this, "keydown");}},
    onkeypress: {set: function(newValue){onkeypress = newValue;dom0_listener_hook(this, "keypress");}},
    onkeyup: {set: function(newValue){onkeyup = newValue;dom0_listener_hook(this, "keyup");}},
    onload: {set: function(newValue){onload = newValue;dom0_listener_hook(this, "load");}},
    onmousedown: {set: function(newValue){onmousedown = newValue;dom0_listener_hook(this, "mousedown");}},
    onmousemove: {set: function(newValue){onmousemove = newValue;dom0_listener_hook(this, "mousemove");}},
    onmouseout: {set: function(newValue){onmouseout = newValue;dom0_listener_hook(this, "mouseout");}},
    onmouseover: {set: function(newValue){onmouseover = newValue;dom0_listener_hook(this, "mouseover");}},
    onmouseup: {set: function(newValue){onmouseup = newValue;dom0_listener_hook(this, "mouseup");}},
    onreset: {set: function(newValue){onreset = newValue;dom0_listener_hook(this, "reset");}},
    onresize: {set: function(newValue){onresize = newValue;dom0_listener_hook(this, "resize");}},
    onselect: {set: function(newValue){onselect = newValue;dom0_listener_hook(this, "select");}},
    onsubmit: {set: function(newValue){onsubmit = newValue;dom0_listener_hook(this, "submit");}},
    onunload: {set: function(newValue){onunload = newValue;dom0_listener_hook(this, "unload");}},
    onabort: {set: function(newValue){onabort = newValue;dom0_listener_hook(this, "abort");}},
    onerror: {set: function(newValue){onerror = newValue;dom0_listener_hook(this, "error");}},
})

表单填充,事件触发

表单填充

image-20210924160620582

input处理

go
// 找出 type 为空 或者 type=text
for _, node := range nodes {
    // 兜底超时
    tCtxN, cancelN := context.WithTimeout(ctx, time.Second*5)
    attrType := node.AttributeValue("type")
    if attrType == "text" || attrType == "" {
        inputName := node.AttributeValue("id") + node.AttributeValue("class") + node.AttributeValue("name")
        value := f.GetMatchInputText(inputName)
        // 寻找匹配类型的值
        var nodeIds = []cdp.NodeID{node.NodeID}
        // 先使用模拟输入
        _ = chromedp.SendKeys(nodeIds, value, chromedp.ByNodeID).Do(tCtxN)
        // 再直接赋值JS属性
        _ = chromedp.SetAttributeValue(nodeIds, "value", value, chromedp.ByNodeID).Do(tCtxN)
    } else if attrType == "email" || attrType == "password" || attrType == "tel" {
        value := f.GetMatchInputText(attrType)
        // 寻找匹配类型的值
        var nodeIds = []cdp.NodeID{node.NodeID}
        // 先使用模拟输入
        _ = chromedp.SendKeys(nodeIds, value, chromedp.ByNodeID).Do(tCtxN)
        // 再直接赋值JS属性
        _ = chromedp.SetAttributeValue(nodeIds, "value", value, chromedp.ByNodeID).Do(tCtxN)
    } else if attrType == "radio" || attrType == "checkbox" {
        var nodeIds = []cdp.NodeID{node.NodeID}
        _ = chromedp.SetAttributeValue(nodeIds, "checked", "true", chromedp.ByNodeID).Do(tCtxN)
    } else if attrType == "file" || attrType == "image" {
        var nodeIds = []cdp.NodeID{node.NodeID}
        wd, _ := os.Getwd()
        filePath := wd + "/upload/image.png"
        _ = chromedp.RemoveAttribute(nodeIds, "accept", chromedp.ByNodeID).Do(tCtxN)
        _ = chromedp.RemoveAttribute(nodeIds, "required", chromedp.ByNodeID).Do(tCtxN)
        // 对于一些简单的限制,可以去掉,比如找到文件上传的dom节点并删除 accept 和 required 属性:
        _ = chromedp.SendKeys(nodeIds, filePath, chromedp.ByNodeID).Do(tCtxN)
    }
    cancelN()
}

multiSelect

css语法获取select第一个元素,设置属性即可

go
optionNodes, optionErr := f.tab.GetNodeIDs(`select option:first-child`)
    if optionErr != nil || len(optionNodes) == 0 {
        logger.Logger.Debug("fillMultiSelect: get select option element err")
        if optionErr != nil {
            logger.Logger.Debug(optionErr)
        }
        return
    }
    _ = chromedp.SetAttributeValue(optionNodes, "selected", "true", chromedp.ByNodeID).Do(tCtx)
    _ = chromedp.SetJavascriptAttribute(optionNodes, "selected", "true", chromedp.ByNodeID).Do(tCtx)

TextArea

找到后填充即可

go
textareaNodes, textareaErr := f.tab.GetNodeIDs(`textarea`)
if textareaErr != nil || len(textareaNodes) == 0 {
    logger.Logger.Debug("fillTextarea: get textarea element err")
    if textareaErr != nil {
        logger.Logger.Debug(textareaErr)
    }
    return
}

_ = chromedp.SendKeys(textareaNodes, value, chromedp.ByNodeID).Do(tCtx)

自动化提交表单

提交表单也有一些需要注意的问题,直接点击form表单的提交按钮会导致页面重载,我们并不希望当前页面刷新,所以除了Hook住前端导航请求之外,我们还可以为form节点设置target属性,指向一个隐藏的iframe。具体操作的话就是新建隐藏iframe然后将form表单的target指向它即可

go
/**
设置form的target指向一个frame
*/
const NewFrameTemplate = `
(function sec_auto_new_iframe () {
    let frame = document.createElement("iframe");
    frame.setAttribute("name", "%s");
    frame.setAttribute("id", "%s");
    frame.setAttribute("style", "display: none");
    document.body.appendChild(frame);
})()
`
func (tab *Tab) setFormToFrame() {
    // 首先新建 frame
    nameStr := tools.RandSeq(8)
    tab.Evaluate(fmt.Sprintf(js.NewFrameTemplate, nameStr, nameStr))

    // 接下来将所有的 form 节点target都指向它
    ctx := tab.GetExecutor()
    formNodes, formErr := tab.GetNodeIDs(`form`)
    if formErr != nil || len(formNodes) == 0 {
        logger.Logger.Debug("setFormToFrame: get form element err")
        if formErr != nil {
            logger.Logger.Debug(formErr)
        }
        return
    }
    tCtx, cancel := context.WithTimeout(ctx, time.Second*2)
    defer cancel()
    _ = chromedp.SetAttributeValue(formNodes, "target", nameStr, chromedp.ByNodeID).Do(tCtx)
}

要成功的提交表单,就得正确触发表单的submit操作。不是所有的前端内容都有规范的表单格式,或许有一些form连个button都没有,所以这里有三种思路可供尝试,保险起见建议全部都运行一次:

  • 在form节点的子节点内寻找type=submit的节点,执行elementHandle.click()方法。
  • 直接对form节点执行JS语句:form.submit(),注意,如果form内有包含属性值name=submit的节点,将会抛出异常,所以注意捕获异常。
  • 在form节点的子节点内寻找所有button节点,全部执行一次elementHandle.click()方法。因为我们之前已经重定义并锁定了表单重置函数,所以不用担心会清空表单。

这样,绝大部分表单我们都能触发了。

go
/**
点击按钮 type=submit
*/
func (tab *Tab) clickSubmit() {
    defer tab.formSubmitWG.Done()

    // 首先点击按钮 type=submit
    ctx := tab.GetExecutor()

    // 获取所有的form节点 直接执行submit
    formNodes, formErr := tab.GetNodeIDs(`form`)
    if formErr != nil || len(formNodes) == 0 {
        logger.Logger.Debug("clickSubmit: get form element err")
        if formErr != nil {
            logger.Logger.Debug(formErr)
        }
        return
    }
    tCtx1, cancel1 := context.WithTimeout(ctx, time.Second*2)
    defer cancel1()
    _ = chromedp.Submit(formNodes, chromedp.ByNodeID).Do(tCtx1)

    // 获取所有的input标签
    inputNodes, inputErr := tab.GetNodeIDs(`form input[type=submit]`)
    if inputErr != nil || len(inputNodes) == 0 {
        logger.Logger.Debug("clickSubmit: get form input element err")
        if inputErr != nil {
            logger.Logger.Debug(inputErr)
        }
        return
    }
    tCtx2, cancel2 := context.WithTimeout(ctx, time.Second*2)
    defer cancel2()
    _ = chromedp.Click(inputNodes, chromedp.ByNodeID).Do(tCtx2)
}

/**
click all button
*/
func (tab *Tab) clickAllButton() {
    defer tab.formSubmitWG.Done()

    // 获取所有的form中的button节点
    ctx := tab.GetExecutor()
    // 获取所有的button标签
    btnNodeIDs, bErr := tab.GetNodeIDs(`form button`)
    if bErr != nil || len(btnNodeIDs) == 0 {
        logger.Logger.Debug("clickAllButton: get form button element err")
        if bErr != nil {
            logger.Logger.Debug(bErr)
        }
        return
    }
    tCtx, cancel1 := context.WithTimeout(ctx, time.Second*2)
    defer cancel1()
    _ = chromedp.Click(btnNodeIDs, chromedp.ByNodeID).Do(tCtx)

    // 使用JS的click方法进行点击
    var btnNodes []*cdp.Node
    tCtx2, cancel2 := context.WithTimeout(ctx, time.Second*2)
    defer cancel2()
    err := chromedp.Nodes(btnNodeIDs, &btnNodes, chromedp.ByNodeID).Do(tCtx2)
    if err != nil {
        return
    }
    for _, node := range btnNodes {
        _ = tab.EvaluateWithNode(js.FormNodeClickJS, node)
    }
}

事件触发

  1. 对JavaScript协议的内联事件触发,执行以下js
js
(async function click_all_a_tag_javascript(){
    let nodeListHref = document.querySelectorAll("[href]");
    nodeListHref = window.randArr(nodeListHref);
    for (let node of nodeListHref) {
        let attrValue = node.getAttribute("href");
        if (attrValue.toLocaleLowerCase().startsWith("javascript:")) {
            await window.sleep(%f);
            try {
                eval(attrValue.substring(11));
            }
            catch {}
        }
    }
    let nodeListSrc = document.querySelectorAll("[src]");
    nodeListSrc = window.randArr(nodeListSrc);
    for (let node of nodeListSrc) {
        let attrValue = node.getAttribute("src");
        if (attrValue.toLocaleLowerCase().startsWith("javascript:")) {
            await window.sleep(%f);
            try {
                eval(attrValue.substring(11));
            }
            catch {}
        }
    }
})()
  1. 对常见的内联事件触发
js
(async function trigger_all_inline_event(){
    let eventNames = ["onabort", "onblur", "onchange", "onclick", "ondblclick", "onerror", "onfocus", "onkeydown", "onkeypress", "onkeyup", "onload", "onmousedown", "onmousemove", "onmouseout", "onmouseover", "onmouseup", "onreset", "onresize", "onselect", "onsubmit", "onunload"];
    for (let eventName of eventNames) {
        let event = eventName.replace("on", "");
        let nodeList = document.querySelectorAll("[" + eventName + "]");
        if (nodeList.length > 100) {
            nodeList = nodeList.slice(0, 100);
        }
        nodeList = window.randArr(nodeList);
        for (let node of nodeList) {
            await window.sleep(%f);
            let evt = document.createEvent('CustomEvent');
            evt.initCustomEvent(event, false, true, null);
            try {
                node.dispatchEvent(evt);
            }
            catch {}
        }
    }
})()
  1. 对之前hook的事件触发,对于某些节点,可能会存在子节点也响应的事件,为了性能考虑,可以将层数控制到三层,且对兄弟节点随机选择一个触发。简单画图说明:
1. ![img](assert/37uKa_t01b9e672b3cf4a890e.png)
js
(async function trigger_all_dom2_custom_event() {
    function transmit_child(node, event, loop) {
        let _loop = loop + 1
        if (_loop > 4) {
            return;
        }
        if (node.nodeType === 1) {
            if (node.hasChildNodes) {
                let index = parseInt(Math.random()*node.children.length,10);
                try {
                    node.children[index].dispatchEvent(event);
                } catch(e) {}
                let max = node.children.length>5?5:node.children.length;
                for (let count=0;count<max;count++) {
                    let index = parseInt(Math.random()*node.children.length,10);
                    transmit_child(node.children[index], event, _loop);
                }
            }
        }
    }
    let nodes = document.querySelectorAll("[sec_auto_dom2_event_flag]");
    if (nodes.length > 200) {
        nodes = nodes.slice(0, 200);
    }
    nodes = window.randArr(nodes);
    for (let node of nodes) {
        let loop = 0;
        await window.sleep(%f);
        let event_name_list = node.getAttribute("sec_auto_dom2_event_flag").split("|");
        let event_name_set = new Set(event_name_list);
        event_name_list = [...event_name_set];
        for (let event_name of event_name_list) {
            let evt = document.createEvent('CustomEvent');
            evt.initCustomEvent(event_name, true, true, null);

            if (event_name == "click" || event_name == "focus" || event_name == "mouseover" || event_name == "select") {
                transmit_child(node, evt, loop);
            }
            if ( (node.className && node.className.includes("close")) || (node.id && node.id.includes("close"))) {
                continue;
            }

            try {
                node.dispatchEvent(evt);
            } catch(e) {}
        }
    }
})()
  1. 监控插入的节点,如果新增节点的href src含有JavaScript协议,则手动触发。这似乎会漏一些内联事件的触发。
js
(function init_observer_sec_auto_b() {
 window.dom_listener_func_sec_auto = function (e) {
     let node = e.target;
     let nodeListSrc = node.querySelectorAll("[src]");
     for (let each of nodeListSrc) {
         if (each.src) {
             window.addLink(each.src, "DOM");
             let attrValue = each.getAttribute("src");
             if (attrValue.toLocaleLowerCase().startsWith("javascript:")) {
                 try {
                     eval(attrValue.substring(11));
                 }
                 catch {}
             }
         }
     }

     let nodeListHref = node.querySelectorAll("[href]");
     nodeListHref = window.randArr(nodeListHref);
     for (let each of nodeListHref) {
         if (each.href) {
             window.addLink(each.href, "DOM");
             let attrValue = each.getAttribute("href");
             if (attrValue.toLocaleLowerCase().startsWith("javascript:")) {
                 try {
                     eval(attrValue.substring(11));
                 }
                 catch {}
             }
         }
     }
 };
 document.addEventListener('DOMNodeInserted', window.dom_listener_func_sec_auto, true);
 document.addEventListener('DOMSubtreeModified', window.dom_listener_func_sec_auto, true);
 document.addEventListener('DOMNodeInsertedIntoDocument', window.dom_listener_func_sec_auto, true);
 document.addEventListener('DOMAttrModified', window.dom_listener_func_sec_auto, true);
})()

窗口阻塞处理

crawler处理了 alert()/prompt() 基础认证等等的阻塞。

go
chromedp.ListenTarget(*tab.Ctx, func(v interface{}) {
        switch v := v.(type) {
        //case *network.EventLoadingFailed:
        //    logger.Logger.Error("EventLoadingFailed ", v.ErrorText)
        // 401 407 要求认证 此时会阻塞当前页面 需要处理解决
        case *fetch.EventAuthRequired:
            tab.WG.Add(1)
            go tab.HandleAuthRequired(v)

        // close Dialog
        case *page.EventJavascriptDialogOpening:
            tab.WG.Add(1)
            go tab.dismissDialog()
        }
})

但是还有 打印 和 文件上传窗口可能阻塞窗口

打印事件可以hook函数,文件上传窗口可以用Page.setInterceptFileChooserDialog过滤。

End

一些还可以优化的部分,表单填充可以识别参数长度max-lengthmin-length

从Crawlergo的设计和源码中能提取出很多东西来,

  • 基于网页结构的大量网页快速相似匹配,如果能集成到那些网络空间引擎中应该会很好玩,但似乎还没有一家做过。

有了原生的动态爬虫支持,对自动化漏扫也有了更多的想法,例如通过hook一些触发函数,污点检测来检测dom xss,爬虫的原始请求包可以直接推到w13scan中。有了自动化爬虫,后续所有流量都可以存储一份,直接用搜索语法来找到相同参数的页面进行poc测试等等。。

作者的代码风格太不go了,想重写一份了。

参考

撰写