net_guard.rs
net_guard.rs
//! 站外抓取的 SSRF 防護(link-preview 與 image-proxy 共用)。
//!
//! 規則:
//! 1. 只允許 http/https
//! 2. 目的地是私網/迴環/link-local/CGNAT/IPv6-ULA → 拒絕
//! 3. 呼叫端在連線後還要用 `resp.remote_addr()` 對實際 peer IP 再驗一次
//! (DNS rebinding/域名指向私網/redirect 進內網,最終都落在這層)
use std::net::IpAddr;
/// 私網/迴環/link-local/CGNAT → 一律拒絕(SSRF 防護)
pub fn is_blocked_ip(ip: &IpAddr) -> bool {
match ip {
IpAddr::V4(v4) => {
v4.is_private()
|| v4.is_loopback()
|| v4.is_link_local()
|| v4.is_broadcast()
|| v4.is_unspecified()
|| v4.is_documentation()
// CGNAT 100.64.0.0/10
|| (v4.octets()[0] == 100 && (64..=127).contains(&v4.octets()[1]))
}
IpAddr::V6(v6) => {
v6.is_loopback()
|| v6.is_unspecified()
// fc00::/7 unique-local、fe80::/10 link-local
|| (v6.segments()[0] & 0xfe00) == 0xfc00
|| (v6.segments()[0] & 0xffc0) == 0xfe80
}
}
}
/// 驗證 URL 可安全抓取;回傳正規化後的 URL 與 host
pub fn validate_url(raw: &str) -> Option<(String, String)> {
let u = reqwest::Url::parse(raw).ok()?;
if !matches!(u.scheme(), "http" | "https") {
return None;
}
let host = u.host_str()?.to_string();
// 直接寫 IP 的先擋(走 DNS 的在 fetch 時由 resolve 再擋一次)。
// IPv6 字面量 host_str() 帶方括號("[::1]"),先剝掉才 parse 得出來——
// 不剝的話 parse 失敗會靜默跳過檢查,變成 v6 繞過(整合測試抓到的真 bug)。
let bare = host.trim_start_matches('[').trim_end_matches(']');
if let Ok(ip) = bare.parse::<IpAddr>()
&& is_blocked_ip(&ip)
{
return None;
}
if host.eq_ignore_ascii_case("localhost") {
return None;
}
Some((u.to_string(), host))
}
#[cfg(test)]
mod tests {
use super::*;
/// 每個網段各給一個斷言——這裡刻意「一段一筆」而不是包成迴圈跑幾個代表值。
///
/// 理由是 cargo-mutants 跑出來的結果:這支檔案原本 26 個變異只擋掉 9 個,漏掉的包括
/// 把 `||` 換成 `&&`(等於「所有條件同時成立才擋」→ 私有網段幾乎全部放行)、
/// 把 CIDR 遮罩的 `&` 換成 `|`/`^`、把 `==` 換成 `!=`。這些改動都不會讓程式崩潰,
/// 只會讓 SSRF 守衛安靜地失效——正是最需要測試盯住的那種錯。
/// 每段一筆斷言,任何一個條件被動到都會有具體的測試失敗指出是哪一段。
#[test]
fn blocks_every_private_range() {
let blocked = [
// IPv4
("10.0.0.1", "private 10/8"),
("172.16.0.1", "private 172.16/12 下界"),
("172.31.255.255", "private 172.16/12 上界"),
("192.168.1.1", "private 192.168/16"),
("127.0.0.1", "loopback"),
("169.254.169.254", "link-local(雲端 metadata 端點)"),
("255.255.255.255", "broadcast"),
("0.0.0.0", "unspecified"),
("192.0.2.1", "documentation TEST-NET-1"),
("198.51.100.1", "documentation TEST-NET-2"),
("203.0.113.1", "documentation TEST-NET-3"),
("100.64.0.1", "CGNAT 下界"),
("100.127.255.255", "CGNAT 上界"),
// IPv6
("::1", "v6 loopback"),
("::", "v6 unspecified"),
("fc00::1", "v6 ULA fc00::/7 下界"),
("fd00::1", "v6 ULA(實務上最常見的 fd00::/8)"),
("fdff:ffff::1", "v6 ULA fc00::/7 上界"),
("fe80::1", "v6 link-local fe80::/10 下界"),
("febf:ffff::1", "v6 link-local fe80::/10 上界"),
];
for (ip, why) in blocked {
assert!(is_blocked_ip(&ip.parse().unwrap()), "{ip} 應被擋({why})");
}
}
/// 邊界外的公開位址必須放行——只驗「有擋住」的話,把整個函式改成永遠回 true
/// 也會全綠(實際上 cargo-mutants 就是這樣穿過去的)。
/// 這裡的值都刻意貼著上面各網段的邊界,順便釘住遮罩與範圍比較的算法。
#[test]
fn allows_public_addresses_just_outside_the_ranges() {
let allowed = [
("8.8.8.8", "一般公開位址"),
("93.184.216.34", "一般公開位址"),
("172.15.255.255", "貼著 private 172.16/12 下界之外"),
("172.32.0.1", "貼著 private 172.16/12 上界之外"),
("100.63.255.255", "貼著 CGNAT 下界之外"),
("100.128.0.1", "貼著 CGNAT 上界之外"),
("101.64.0.1", "第一段不是 100,第二段落在 CGNAT 區間內"),
("2001:4860:4860::8888", "公開 v6"),
("fbff:ffff::1", "貼著 ULA fc00::/7 之外"),
("fe00::1", "在 fc00::/7 與 fe80::/10 的縫隙"),
("fec0::1", "site-local(已廢棄,不在 fe80::/10)"),
];
for (ip, why) in allowed {
assert!(!is_blocked_ip(&ip.parse().unwrap()), "{ip} 不該被擋({why})");
}
}
#[test]
fn validate_url_rejects_unsafe_targets() {
for (raw, why) in [
("http://127.0.0.1/x", "迴環字面量"),
("http://169.254.169.254/latest/meta-data/", "雲端 metadata"),
("http://10.1.2.3/x", "私網字面量"),
("http://[::1]/x", "v6 字面量帶方括號——剝不掉就會靜默跳過檢查"),
("http://[fd00::1]/x", "v6 ULA 字面量"),
("http://localhost/x", "localhost 主機名"),
("http://LOCALHOST/x", "localhost 大小寫不敏感"),
("ftp://example.com/x", "非 http(s) scheme"),
("file:///etc/passwd", "file scheme"),
("javascript:alert(1)", "javascript scheme"),
("не-url", "根本不是 URL"),
] {
assert!(validate_url(raw).is_none(), "{raw} 應被拒({why})");
}
}
#[test]
fn validate_url_accepts_normal_targets() {
let (url, host) = validate_url("https://example.com/a/b?c=d").expect("一般 https 應通過");
assert_eq!(host, "example.com");
assert!(url.starts_with("https://example.com/a/b"), "回傳正規化後的 URL:{url}");
let (_, host) = validate_url("http://8.8.8.8/x").expect("公開 IP 字面量應通過");
assert_eq!(host, "8.8.8.8");
}
}