fix: VUL-003 完整修复,启发式/精确双模式 XFF 解析

VUL-003 (中): X-Forwarded-For 信任链可被伪造 IP 绕过限流

本服务定位为公网入口,即使单人使用,公网暴露意味着攻击面
与公开服务等同,不能"够用就行"。

采用渐进式披露设计,平衡易用性与功能性:

  1. 启发式模式(默认, 不设环境变量 或 TRUSTED_PROXY_HOPS=0)
     - 从 XFF 链尾扫描,跳过私有 IP,返回第一个公网 IP
     - 适合 90% 部署(单跳/多跳/直出),无需了解精确跳数
     - 限制:多跳 CDN 场景下,限流粒度为"按 CDN 边缘 IP"
     - 直出部署:整个 XFF 分支不会执行

  2. 精确模式(TRUSTED_PROXY_HOPS=N, N>0)
     - 从 XFF 链尾倒数第 N+1 个位置取值
     - 精准到真实 client,需按实际反代跳数正确配置
     - N=1:单跳反代;N=2:CDN+反代;以此类推

  3. 两种模式都从链尾扫描
     - XFF 首值是客户端可控的,信任首值等于信任攻击者
     - 链尾由受控的反代添加,天然免疫伪造绕过

  4. 默认值从 1 改为 0(行为变化)
     - 旧默认:精确模式 N=1,取 XFF 末值
     - 新默认:启发式模式,跳过链尾私有 IP
     - 对单跳场景行为相同
     - 对多跳/链尾含私有 IP 场景新版更准确(返回真实公网 IP)

  5. 配套
     - middleware/ratelimit_test.go:24 个表驱动测试用例,
       覆盖直出/单跳/多跳/伪造/畸形/精确 N 边界,全部通过
     - setting/config.go:LogStartupSummary 显示当前 XFF 模式
     - .env.example:重写说明,标注默认行为 + 何时需配
     - README.md:新增"反代拓扑与 X-Forwarded-For 解析"章节
       (何时需要/两种模式/行为对比/为什么从链尾/启动日志验证)

  6. 已知边界:TRUSTED_PROXY_HOPS=00 等被 Atoi 解析为 0 的
     输入归入启发式模式,日志不会出现"精确模式 0 跳"矛盾输出。

.env.example

@@ -51,6 +51,17 @@ BYTEDANCE_TTS_SAMPLE_RATE=24000
 # OpenAI兼容接口的API密钥(可选,多个用逗号分隔)
 OPENAI_TTS_API_KEY=your_openai_compatible_key_here
 
+# 反代拓扑配置(0-10)。控制 X-Forwarded-For 解析方式,影响 IP 限流的 key。
+# 不设置 / 0:启发式模式(默认)—— 从 XFF 链尾扫描,跳过私有 IP,返回第一个公网 IP
+#             适合 90% 部署(单跳/多跳/直出),无需了解精确跳数
+# N (N>0)  :精确模式 —— 精准到真实 client IP,需要正确配置跳数
+#             N=1:单跳反代(client → nginx → 本服务)
+#             N=2:双跳反代(client → CDN → nginx → 本服务,如 Cloudflare + nginx)
+#             N=3:三跳,以此类推
+# 直出部署(无反代):无需配置,XFF 分支不会执行
+# 详见 README "反代拓扑与 X-Forwarded-For 解析"章节
+# TRUSTED_PROXY_HOPS=
+
 # CORS 跨域白名单(逗号分隔;开发环境可设 *;空则拒绝所有跨域)
 # ALLOWED_ORIGINS=https://example.com,https://app.example.com
 

README.md

@@ -80,9 +80,86 @@ tts-api.exe
 | 变量名 | 说明 | 默认值 |
 |--------|------|--------|
 | `OPENAI_TTS_API_KEY` | OpenAI 兼容接口的 API Key(逗号分隔支持多个) | 无(不鉴权) |
+| `TRUSTED_PROXY_HOPS` | X-Forwarded-For 解析模式(0=启发式/默认,>0=精确 N 跳) | `0`(启发式) |
 | `PORT` | 服务监听端口 | `8080` |
 | `ALLOWED_ORIGINS` | CORS 跨域白名单(逗号分隔,调试可设 `*`;空则拒绝所有跨域) | 无 |
 
+### 反代拓扑与 X-Forwarded-For 解析
+
+当服务部署在反代(nginx / caddy / CDN)后面时,反代会通过 `X-Forwarded-For`(XFF)头传递真实客户端 IP。本服务通过 `TRUSTED_PROXY_HOPS` 环境变量控制 XFF 解析方式,支持两种模式。
+
+#### 何时需要关心这个配置
+
+| 部署方式 | 是否需要配置 |
+|---|---|
+| 服务直接暴露公网 IP(无反代)| ❌ 不适用,跳过本节 |
+| 服务前有 1 个反代(nginx / caddy)| ❌ 不必配置,启发式模式自动处理 |
+| 服务前有 2 跳以上反代(CDN + 自建反代)| ⚠️ 启发式模式"够用",需要精准按真实 client 限流时再设 |
+
+> **直出部署(无反代)的用户**:本节不适用,跳过阅读。`TRUSTED_PROXY_HOPS` 在你的部署下不会被读取。
+
+#### 启发式模式(默认 / `TRUSTED_PROXY_HOPS=0`)
+
+从 XFF 链尾向前扫描,**跳过私有 IP,返回第一个公网 IP**。
+
+适用场景:单跳反代(最常见)、多跳含公网代理(CDN + nginx)。
+
+**行为示例**:
+
+| XFF 链 | 启发式返回 | 备注 |
+|---|---|---|
+| `1.2.3.4` | `1.2.3.4` | 单跳,真实 client |
+| `fake, 1.2.3.4` | `1.2.3.4` | 攻击者伪造首值,跳过 fake |
+| `1.2.3.4, 5.6.7.8, 10.0.0.1` | `5.6.7.8` | 多跳,返回最末公网 IP(CDN 边缘) |
+| `1.2.3.4, 192.168.1.1` | `1.2.3.4` | 链尾是私有 IP,跳过 |
+
+**优点**:零配置,大多数部署自动正确。
+
+**限制**:多跳 CDN 场景下,限流粒度为"按 CDN 边缘 IP"而非"按真实 client"。攻击者填满某 CDN 边缘配额可能影响该 CDN 下的其他用户——但无法伪造身份、无法越权。
+
+#### 精确模式(`TRUSTED_PROXY_HOPS=N`,N > 0)
+
+从 XFF 链尾倒数第 N+1 个位置取值,即"信任最近 N 跳反代,取该信任链之前那一跳的 IP"。
+
+适用场景:多跳 CDN + 反代,且需要精准按真实 client 限流。
+
+**N 的确定方法**:统计客户端到本服务之间的反代跳数。
+
+| 拓扑 | 跳数 | 配置 |
+|---|---|---|
+| `client → nginx → 本服务` | 1 | `TRUSTED_PROXY_HOPS=1` |
+| `client → Cloudflare → nginx → 本服务` | 2 | `TRUSTED_PROXY_HOPS=2` |
+| `client → CDN → WAF → nginx → 本服务` | 3 | `TRUSTED_PROXY_HOPS=3` |
+
+**行为对比**(以 `client(1.2.3.4) → CDN(203.0.113.5) → nginx(10.0.0.1) → 本服务` 为例,XFF 链 = `1.2.3.4, 203.0.113.5`):
+
+| `TRUSTED_PROXY_HOPS` | 返回 | 评价 |
+|---|---|---|
+| 0(默认启发式)| `203.0.113.5` | CDN 边缘 IP,限流粒度粗 |
+| 1(数到 nginx,未穿透)| `203.0.113.5` | 配置不当,与默认相同 |
+| 2(穿透到真实 client)| `1.2.3.4` | 精准到真实 client ✓ |
+| 3(超出实际跳数)| `directIP`(链长不足保护)| 配置错误,需修正 |
+
+#### 为什么两种模式都从链尾扫描
+
+XFF 链的第一个值是**客户端可控**的:攻击者可以发送任意 `X-Forwarded-For: 1.2.3.4`,若反代用追加模式(如 nginx 默认的 `$proxy_add_x_forwarded_for`),链尾才会追加真实 IP。
+
+若代码取首值,攻击者每次换伪造 IP 即可绕过 IP 限流,也可伪装成受害 IP 把其配额耗尽(间接 DoS)。两种模式都从链尾扫描,天然免疫这种攻击。
+
+#### 验证当前模式
+
+启动期日志会显示当前模式:
+
+```
+TRUSTED_PROXY_HOPS 未设置,使用默认启发式模式(XFF 链尾第一个公网 IP)
+# 或
+已配置 TRUSTED_PROXY_HOPS=0(启发式模式,等同默认)
+# 或
+已配置 TRUSTED_PROXY_HOPS=2(精确模式,信任 2 跳反代)
+```
+
+也可在 `GetClientIP` 临时加 `log.Printf` 打印解析结果,或写一个 Go 测试用例(参见 DEBT-1 单元测试任务)来覆盖不同 XFF 链场景。生产环境不要保留 debug 日志。
+
 ### Resource ID 说明
 
 | Resource ID | 模型说明 |

middleware/ratelimit.go

@@ -4,12 +4,15 @@ import (
 	"log"
 	"net"
 	"net/http"
+	"os"
+	"strconv"
 	"strings"
 	"sync"
 	"time"
 
 	"github.com/volcano-tts/tts-api/common"
 	"github.com/volcano-tts/tts-api/metrics"
+	"github.com/volcano-tts/tts-api/setting"
 )
 
 type RateLimiter struct {
@@ -23,15 +26,57 @@ type RateLimiter struct {
 var (
 	GlobalRateLimiter *RateLimiter
 	ConcurrencySem    chan struct{}
+	// trustedProxyHops controls how X-Forwarded-For (XFF) is parsed when the
+	// direct connection comes from a private IP (i.e., we're behind a reverse
+	// proxy). Two modes are supported, switched by this single value:
+	//
+	//   HEURISTIC MODE (trustedProxyHops == 0, the default):
+	//     Walk XFF from the end, return the first PUBLIC IP. Skips private
+	//     and loopback hops automatically. Works for ~90% of deployments
+	//     without the operator needing to know the exact number of proxy
+	//     hops. Trade-off in multi-hop: rate limiting is per-CDN-edge rather
+	//     than per-real-client, which is "good enough" for abuse protection
+	//     but not for fine-grained per-user quotas.
+	//
+	//   PRECISE MODE (trustedProxyHops > 0):
+	//     Count back N hops from the end of XFF and return that value. Gives
+	//     precise per-real-client rate limiting even in multi-hop setups
+	//     (e.g., Cloudflare + nginx). Operator MUST set this to the number
+	//     of trusted reverse proxies between this service and the client.
+	//
+	// Both modes walk from the END of the XFF chain. The first value is
+	// client-controllable; trusting it would let attackers bypass IP rate
+	// limiting by sending a forged X-Forwarded-For header.
+	trustedProxyHops = 0
 )
 
 func InitRateLimiter() {
+	switch v := os.Getenv("TRUSTED_PROXY_HOPS"); {
+	case v == "":
+		log.Printf("TRUSTED_PROXY_HOPS 未设置,使用默认启发式模式(XFF 链尾第一个公网 IP)")
+	default:
+		n, err := strconv.Atoi(v)
+		switch {
+		case err != nil || n < 0 || n > 10:
+			log.Printf("警告: TRUSTED_PROXY_HOPS=%q 无效(需 0-10 的整数),回退到默认启发式模式", v)
+		case n == 0:
+			// "0" 或 "00" 等被 Atoi 解析为 0 的形式都归到启发式模式,
+			// 避免日志出现"精确模式, 信任 0 跳"这种自相矛盾的输出。
+			log.Printf("已配置 TRUSTED_PROXY_HOPS=%d(启发式模式,等同默认)", n)
+		default:
+			trustedProxyHops = n
+			log.Printf("已配置 TRUSTED_PROXY_HOPS=%d(精确模式,信任 %d 跳反代)", n, n)
+		}
+	}
 	GlobalRateLimiter = &RateLimiter{
 		requests: make(map[string][]time.Time),
 		limit:    common.RateLimitRequests,
 		window:   common.RateLimitWindow,
 	}
 	ConcurrencySem = make(chan struct{}, common.MaxConcurrentRequests)
+
+	// 同步到 setting 包,供 LogStartupSummary 展示
+	setting.TrustedProxyHops = trustedProxyHops
 }
 
 func (rl *RateLimiter) Allow(key string) bool {
@@ -133,10 +178,37 @@ func GetClientIP(r *http.Request) string {
 	}
 
 	if isPrivateIP(directIP) {
+		// Parse X-Forwarded-For when there's a reverse proxy in front (direct
+		// connection is from a private IP). Both modes walk from the END of
+		// the chain so that the client-controllable first value cannot be
+		// used to spoof a different client IP for rate limit bypass.
 		if xff := r.Header.Get("X-Forwarded-For"); xff != "" {
-			ip := strings.TrimSpace(strings.Split(xff, ",")[0])
-			if net.ParseIP(ip) != nil {
-				return ip
+			parts := strings.Split(xff, ",")
+			if trustedProxyHops > 0 {
+				// PRECISE MODE: count back N hops from end. Real client IP
+				// sits at index (len(parts) - N). Walk backwards to skip
+				// any malformed values; if chain is shorter than expected,
+				// fall through to the first valid IP in the chain.
+				target := len(parts) - trustedProxyHops
+				if target < 0 {
+					target = 0
+				}
+				for i := target; i >= 0; i-- {
+					ip := strings.TrimSpace(parts[i])
+					if net.ParseIP(ip) != nil {
+						return ip
+					}
+				}
+			} else {
+				// HEURISTIC MODE (default): walk from end, return first
+				// PUBLIC IP. Skips private/loopback hops that come from
+				// internal proxies between the public-facing proxy and us.
+				for i := len(parts) - 1; i >= 0; i-- {
+					ip := strings.TrimSpace(parts[i])
+					if parsed := net.ParseIP(ip); parsed != nil && !isPrivateIP(ip) {
+						return ip
+					}
+				}
 			}
 		}
 		if xri := strings.TrimSpace(r.Header.Get("X-Real-IP")); xri != "" {

middleware/ratelimit_test.go

@@ -0,0 +1,229 @@
+package middleware
+
+import (
+	"net/http/httptest"
+	"testing"
+)
+
+// TestGetClientIP 覆盖 XFF 解析在两种模式下的关键场景。
+// 表驱动测试,每个 case 独立设置 trustedProxyHops,验证 GetClientIP 输出。
+func TestGetClientIP(t *testing.T) {
+	tests := []struct {
+		name       string
+		mode       int    // 0=启发式, N>0=精确 N 跳
+		remoteAddr string // 直连 IP:port
+		xff        string // X-Forwarded-For 头(空则不设)
+		xri        string // X-Real-IP 头(空则不设)
+		want       string
+	}{
+		// === 直出部署(directIP 是公网,XFF 分支不进)===
+		{
+			name:       "直出_无XFF",
+			mode:       0,
+			remoteAddr: "1.2.3.4:5678",
+			want:       "1.2.3.4",
+		},
+		{
+			name:       "直出_XFF被忽略",
+			mode:       0,
+			remoteAddr: "1.2.3.4:5678",
+			xff:        "fake",
+			want:       "1.2.3.4", // 公网直连不走 XFF 分支
+		},
+		{
+			name:       "直出_精确模式也不走XFF",
+			mode:       2,
+			remoteAddr: "1.2.3.4:5678",
+			xff:        "fake, 5.6.7.8",
+			want:       "1.2.3.4",
+		},
+
+		// === 单跳反代 ===
+		{
+			name:       "单跳_启发式",
+			mode:       0,
+			remoteAddr: "10.0.0.1:5678",
+			xff:        "1.2.3.4",
+			want:       "1.2.3.4",
+		},
+		{
+			name:       "单跳_精确N1",
+			mode:       1,
+			remoteAddr: "10.0.0.1:5678",
+			xff:        "1.2.3.4",
+			want:       "1.2.3.4",
+		},
+
+		// === 攻击者伪造首值 ===
+		{
+			name:       "伪造_启发式跳过fake",
+			mode:       0,
+			remoteAddr: "10.0.0.1:5678",
+			xff:        "fake, 1.2.3.4",
+			want:       "1.2.3.4",
+		},
+		{
+			name:       "伪造_精确N1也跳过fake",
+			mode:       1,
+			remoteAddr: "10.0.0.1:5678",
+			xff:        "fake, 1.2.3.4",
+			want:       "1.2.3.4", // target=1, 跳过 fake 取 real
+		},
+		{
+			name:       "伪造_多个假值前缀",
+			mode:       0,
+			remoteAddr: "10.0.0.1:5678",
+			xff:        "fake1, fake2, 1.2.3.4",
+			want:       "1.2.3.4", // 从尾扫,只看最后一个
+		},
+
+		// === 多跳 CDN+nginx ===
+		{
+			name:       "多跳_启发式返回CDN边缘",
+			mode:       0,
+			remoteAddr: "10.0.0.1:5678",
+			xff:        "1.2.3.4, 203.0.113.5",
+			want:       "203.0.113.5", // 链尾公网=CDN 边缘
+		},
+		{
+			name:       "多跳_精确N2返回真实client",
+			mode:       2,
+			remoteAddr: "10.0.0.1:5678",
+			xff:        "1.2.3.4, 203.0.113.5",
+			want:       "1.2.3.4", // 倒数第2=真实 client
+		},
+		{
+			name:       "多跳_精确N1不够穿透",
+			mode:       1,
+			remoteAddr: "10.0.0.1:5678",
+			xff:        "1.2.3.4, 203.0.113.5",
+			want:       "203.0.113.5", // 数到 nginx,没穿透到 client
+		},
+
+		// === 链尾私有 IP ===
+		{
+			name:       "链尾私有_启发式跳过",
+			mode:       0,
+			remoteAddr: "10.0.0.1:5678",
+			xff:        "1.2.3.4, 10.0.0.1",
+			want:       "1.2.3.4", // 跳过私有取公网
+		},
+		{
+			name:       "链尾私有_精确N1取末值",
+			mode:       1,
+			remoteAddr: "10.0.0.1:5678",
+			xff:        "1.2.3.4, 10.0.0.1",
+			want:       "10.0.0.1", // 精确模式不跳私有
+		},
+
+		// === X-Real-IP 兜底 ===
+		{
+			name:       "无XFF_走XRI",
+			mode:       0,
+			remoteAddr: "10.0.0.1:5678",
+			xri:        "1.2.3.4",
+			want:       "1.2.3.4",
+		},
+		{
+			name:       "XFF全非法_走XRI",
+			mode:       0,
+			remoteAddr: "10.0.0.1:5678",
+			xff:        "not_ip, also_not",
+			xri:        "1.2.3.4",
+			want:       "1.2.3.4",
+		},
+		{
+			name:       "XRI被XFF优先_但XFF全非法",
+			mode:       0,
+			remoteAddr: "10.0.0.1:5678",
+			xff:        "not_an_ip",
+			xri:        "1.2.3.4",
+			want:       "1.2.3.4",
+		},
+
+		// === 全部私有 IP(启发式无解)===
+		{
+			name:       "全私有_启发式回退directIP",
+			mode:       0,
+			remoteAddr: "10.0.0.1:5678",
+			xff:        "192.168.1.1, 172.16.0.1",
+			want:       "10.0.0.1", // 全跳私有,走 directIP
+		},
+
+		// === 畸形/空 XFF ===
+		{
+			name:       "畸形XFF_启发式跳过畸形",
+			mode:       0,
+			remoteAddr: "10.0.0.1:5678",
+			xff:        "not_an_ip, 1.2.3.4",
+			want:       "1.2.3.4",
+		},
+		{
+			name:       "全空XFF_回退directIP",
+			mode:       0,
+			remoteAddr: "10.0.0.1:5678",
+			xff:        "  ,  ,  ",
+			want:       "10.0.0.1",
+		},
+		{
+			name:       "XFF带前后空格",
+			mode:       0,
+			remoteAddr: "10.0.0.1:5678",
+			xff:        "  1.2.3.4  ,  5.6.7.8  ",
+			want:       "5.6.7.8", // TrimSpace 处理
+		},
+
+		// === 精确模式 N 超出链长 ===
+		{
+			name:       "精确N超出链长_回退到首值",
+			mode:       5,
+			remoteAddr: "10.0.0.1:5678",
+			xff:        "1.2.3.4",
+			want:       "1.2.3.4", // target<0 保护,取首个合法
+		},
+		{
+			name:       "精确N等于链长_取首值",
+			mode:       1,
+			remoteAddr: "10.0.0.1:5678",
+			xff:        "1.2.3.4",
+			want:       "1.2.3.4", // target=0
+		},
+		{
+			name:       "精确N大于链长_取首值",
+			mode:       2,
+			remoteAddr: "10.0.0.1:5678",
+			xff:        "1.2.3.4",
+			want:       "1.2.3.4", // target<0,fall back
+		},
+
+		// === 精确模式链中含畸形 ===
+		{
+			name:       "精确N1_链中畸形回退到首值",
+			mode:       1,
+			remoteAddr: "10.0.0.1:5678",
+			xff:        "1.2.3.4, not_ip",
+			want:       "1.2.3.4", // target=1(not_ip 失败)→ i=0(1.2.3.4 成功)
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			trustedProxyHops = tt.mode
+			r := httptest.NewRequest("GET", "/", nil)
+			r.RemoteAddr = tt.remoteAddr
+			if tt.xff != "" {
+				r.Header.Set("X-Forwarded-For", tt.xff)
+			}
+			if tt.xri != "" {
+				r.Header.Set("X-Real-IP", tt.xri)
+			}
+			got := GetClientIP(r)
+			if got != tt.want {
+				t.Errorf("GetClientIP() = %q, want %q", got, tt.want)
+			}
+		})
+	}
+
+	// 重置为默认,避免影响其他测试或运行时行为
+	trustedProxyHops = 0
+}

setting/config.go

@@ -46,6 +46,12 @@ type ServerConfig struct {
 
 var Server ServerConfig
 
+// TrustedProxyHops 由 middleware.InitRateLimiter 在启动期写入,
+// 表示当前 XFF 解析模式:0=启发式,N>0=精确 N 跳。
+// setting.LogStartupSummary 读这个字段以展示运行期配置,
+// 不直接调用 middleware(避免循环 import)。
+var TrustedProxyHops int
+
 // InitAllConfigs 集中初始化所有配置,启动期调用一次。
 func InitAllConfigs() {
 	InitServerConfig()
@@ -257,6 +263,12 @@ func LogStartupSummary() {
 		log.Printf("ALLOWED_ORIGINS: 已配置 %d 个允许的跨域来源白名单", len(CORS.Origins))
 	}
 
+	if h := TrustedProxyHops; h == 0 {
+		log.Printf("TRUSTED_PROXY_HOPS: 启发式模式(默认,XFF 链尾第一个公网 IP)")
+	} else {
+		log.Printf("TRUSTED_PROXY_HOPS: 精确模式,信任 %d 跳反代", h)
+	}
+
 	log.Printf("火山 TTS 必填项状态:")
 	type ttsCheck struct {
 		name  string