逐跳头与端到端头

概念概述

逐跳头与端到端头（Hop-by-Hop Versus End-to-End Headers）指的是：Connection、Transfer-Encoding 等逐跳头仅作用于单条链路，中间代理必须消费它们而不能原样转发，否则会引发框架歧义与请求走私。

从更严格的定义看，HTTP headers are classified as end-to-end, meaningful across every hop, or hop-by-hop, meaningful only for one connection and consumed by each intermediary that processes the message. 该概念在工程实践中的价值，在于它能帮助团队用统一的语言描述一个具体的质量属性或设计约束，从而在评审与协作中减少歧义。

为什么需要它

The distinction governs what a proxy may copy when it relays a message and which headers a server must strip before forwarding. 它与 HTTP Trailer Fields、CAA Records Constraining Certificates、Equal-Cost Multipath Routing 等概念经常一起出现，用于解决相近的工程问题。理解它的适用场景，有助于在多种备选方案之间做出有依据的取舍，而不是凭直觉选择技术方案。

常见误解

• all headers can be forwarded verbatim by any proxy
• the Connection header only signals connection close behaviour

需要避免的反模式

• blindly copying every request header through an application proxy
• treating Transfer-Encoding and Content-Length as interchangeable

实践要点

1. The Connection header lists field names that are themselves hop-by-hop, so a proxy must remove both Connection and each name it lists before passing the message along.
2. Standard hop-by-hop fields include Keep-Alive, Transfer-Encoding, TE, Trailer, Upgrade, and the PROXY-related fields, none of which should survive past the hop that interpreted them.
3. Failing to strip them can smuggle headers past security checks or create framing ambiguity between a front proxy and its backend, which is a known path to request smuggling.

参考来源

https://datatracker.ietf.org/doc/html/rfc9110

代码示例

export function parseContentDisposition(header) {
  if (typeof header !== "string" || header.trim() === "") {
    throw new TypeError("header must be a non-empty string");
  }
  const segments = splitParameters(header);
  const type = segments.shift().trim().toLowerCase();
  if (type !== "inline" && type !== "attachment") {
    throw new Error(`unsupported disposition type "${type}"`);
  }
  const parameters = new Map();
  for (const segment of segments) {
    const equals = segment.indexOf("=");
    if (equals === -1) continue;
    const name = segment.slice(0, equals).trim().toLowerCase();
    if (name === "") continue;
    parameters.set(name, unquote(segment.slice(equals + 1).trim()));
  }
  return { type, filename: resolveFilename(parameters), parameters };
}

function splitParameters(header) {
  const segments = [];
  let current = "";
  let quoted = false;
  for (const character of header) {
    if (character === '"') quoted = !quoted;
    if (character === ";" && !quoted) {
      segments.push(current);
      current = "";
      continue;
    }
    current += character;
  }
  segments.push(current);
  return segments;
}

function unquote(value) {
  if (value.length >= 2 && value.startsWith('"') && value.endsWith('"')) {
    return value.slice(1, -1).replace(/\\(.)/g, "$1");
  }
  return value;
}

function resolveFilename(parameters) {
  const extended = parameters.get("filename*");
  if (extended !== undefined) {
    const match = /^([\w-]+)'([\w-]*)'(.+)$/.exec(extended);
    if (match !== null) {
      try {
        return sanitize(decodeURIComponent(match[3]));
      } catch {
        return sanitize(extended);
      }
    }
  }
  const plain = parameters.get("filename");
  return plain === undefined ? null : sanitize(plain);
}

function sanitize(filename) {
  return filename.split(/[\\/]/).pop() || null;
}

console.log(parseContentDisposition('attachment; filename="report.pdf"'));
console.log(parseContentDisposition("attachment; filename*=UTF-8''%E4%B8%AD%E6%96%87.pdf"));
console.log(parseContentDisposition('inline; filename="../../etc/passwd"'));

import re
from pathlib import PurePosixPath
from urllib.parse import unquote

NAME_STAR = re.compile(r"filename\*\s*=\s*([^;]+)", re.IGNORECASE)
NAME = re.compile(r'filename\s*=\s*(?:"([^"]*)"|([^;]+))', re.IGNORECASE)


def safe_basename(value: str) -> str | None:
    candidate = PurePosixPath(value.replace("\\", "/")).name.strip()
    if candidate in {"", ".", ".."} or "\x00" in candidate:
        return None
    return candidate


def attachment_filename(header: str) -> str | None:
    match = NAME_STAR.search(header)
    if match:
        value = match.group(1).strip().strip('"')
        charset, _, remainder = value.partition("'")
        _, _, encoded = remainder.partition("'")
        try:
            decoded = unquote(
                encoded or remainder,
                encoding=charset or "utf-8",
                errors="strict",
            )
        except (LookupError, UnicodeDecodeError):
            decoded = unquote(encoded or remainder)
        return safe_basename(decoded)
    match = NAME.search(header)
    if not match:
        return None
    value = match.group(1) if match.group(1) is not None else match.group(2)
    return safe_basename(unquote(value.strip()))


def main() -> None:
    headers = [
        'attachment; filename="report final.pdf"',
        "attachment; filename=../../etc/passwd",
        "attachment; filename*=UTF-8''%E5%B9%B4%E6%8A%A5.pdf",
        'attachment; filename=".."',
    ]
    for header in headers:
        print(attachment_filename(header))


if __name__ == "__main__":
    main()