Core principle of WAF bypass: the WAF and the backend application parse the same HTTP request differently; exploit this difference so the WAF "sees" a legitimate request while the backend "sees" the malicious payload.
In-depth reference
- Encoding bypass payload (double URL/Unicode/HTML/mixed encoding) → references/encoding-bypass-payloads.md
- HTTP protocol layer bypass (chunked transfer/Content-Type/method switching/HTTP2/smuggling) → references/http-protocol-bypass.md
- Parameter layer bypass (HPP/array syntax/Multipart) → references/parameter-bypass.md
- Payload mutation (generic encoding/splitting/format switching) → references/payload-mutation.md
Phase 0: WAF identification
0.1 Detect whether a WAF is present
# 发送明显恶意请求，观察响应
curl -s "http://TARGET/?id=1' OR '1'='1" -D-
curl -s "http://TARGET/?id=<script>alert(1)</script>" -D-
curl -s "http://TARGET/?cmd=;id" -D-
# 对比正常请求和恶意请求的响应差异
# WAF 拦截特征：403/406 状态码、特定拦截页面、不同的 Server 头
0.2 WAF fingerprinting
| Feature | WAF product |
|---|---|
Server: cloudflare / cf-ray header |
Cloudflare |
X-Sucuri-ID header |
Sucuri |
Response contains ModSecurity |
ModSecurity |
Response contains 安全狗 / safedog |
SafeDog |
Response contains 宝塔 / bt.cn |
BT Panel WAF |
X-Powered-By-Anquanbao |
Anbai WAF |
Response contains yunsuo |
Yunsuo |
| Alibaba Cloud 403 page | Alibaba Cloud Shield |
| Tencent Cloud specific 403 | Tencent Cloud WAF |
General bypass check flow
Payload 被拦截 → 403/拦截页
├── 1. 编码绕过
│ ├── 双重 URL 编码
│ ├── Unicode 编码
│ └── 混合大小写 + NULL 字节
├── 2. HTTP 层
│ ├── 分块传输
│ ├── Content-Type 切换
│ ├── HTTP 方法切换
│ └── HTTP/2
├── 3. 参数层
│ ├── 参数污染 (HPP)
│ ├── 数组/JSON 嵌套
│ └── Multipart 包裹
├── 4. Payload 变形
│ ├── 空格替代（注释/Tab/换行）
│ ├── 函数名替代
│ ├── 拼接/编码函数
│ └── 通配符/变量
└── 5. 逻辑层
├── 分多次请求发送（先探测再利用）
└── 利用白名单路径（/api/health + 路径穿越）
See the corresponding references file for detailed payloads for each branch
Phase 5: Domestic/commercial WAF fingerprints and bypasses
| WAF | Fingerprint feature | Bypass approach |
|---|---|---|
| F5 BIG-IP ASM | X-Content-Type-Options, ASM interception page |
Encoding obfuscation, chunking, method switching |
| ChinaNetCenter | X-Cache: NS, ChinaNetCenter interception page |
Parameter pollution, case variation, Unicode encoding |
| UEWAF | Specific interception page/JS challenge | Direct request to backend IP/domain, bypass JS validation |
| Cloudflare | cf-ray、server: cloudflare |
Origin IP direct connection, encoding bypass |
| SafeDog/BT Panel | Page feature | Comment characters, equivalent functions, encoding |
Phase 6: Captcha unlocking (captcha bypass)
When a WAF/risk control uses captchas to block high-frequency attacks:
- Logic bypass: reuse captcha, empty code, delete parameter, universal captcha
- Recognition: OCR/captcha-solving platforms recognize image captchas; for click-based captchas, simulate clicks by coordinates
- Trigger point switching: switch endpoint/parameter to bypass the captcha trigger logic
- See captcha-bypass-methodology for detailed methods
Phase 7: Proxy pool/free proxy IP rotation
When a WAF rate-limits/bans by IP:
- Use free proxies/proxy pools to switch the egress IP and evade IP bans
- Note proxy availability and speed; prefer high-anonymity proxies
- Combine with low-rate (slow) requests to reduce the probability of triggering risk control
- Free proxy sources change dynamically; verify proxy availability before testing
Phase 8: SQL injection bypass (F5/ChinaNetCenter scenarios)
- Comment character replacement:
--→#→/*!*/→;%00 - Keyword obfuscation: case, inline comments
/*!union*/, equivalent functions - Encoding: URL double encoding, Unicode, hex
- Combined use of chunked transfer + parameter pollution