Files
Felis/internal/api/middleware.go
flyemoji a8c4077202 fix(api): log the panic value and stack behind the opaque 500
withRecover turned a panicking handler into a 500 envelope and threw the panic
away. The client is meant to get an opaque "internal error" — that part is
right — but nothing was written server-side, so a recovered panic was an
untraceable 500: an operator holding "internal error" has no message, no
stack and no request to grep for, and diagnosis degrades into guessing against
a live install. That is what it cost during the email-OTP report.

The panic value, a stack and the method+path are now logged first, keyed by
the same request_id writeError already stamps on unmapped errors, so the
client envelope and the server log can be joined. The test pins all three
markers plus the unchanged 500/"panic" response, because a silent recover
looks exactly like a working one from the outside.
2026-07-22 14:40:42 +09:00

157 lines
6.1 KiB
Go

package api
import (
"context"
"crypto/rand"
"encoding/hex"
"log"
"net/http"
"runtime/debug"
)
// withRequestID assigns a request id (honoring a WELL-FORMED inbound X-Request-Id)
// and echoes it on the response and into the context for the error envelope.
//
// A caller-supplied id is honored for cross-service tracing, but only after
// validation: the id is echoed to the client, embedded in the error envelope, AND
// persisted verbatim into audit_logs.request_id, so an unvalidated one is an
// audit-integrity vector — an arbitrarily long value bloats the audit row and a
// stray control byte could smuggle a forged line into a log sink. A rejected id is
// replaced with a fresh server-minted one: that one request loses its inbound trace
// link, which is strictly better than storing attacker-controlled text.
func withRequestID(next http.Handler) http.Handler {
return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
id := r.Header.Get("X-Request-Id")
if !validRequestID(id) {
id = newRequestID()
}
w.Header().Set("X-Request-Id", id)
ctx := context.WithValue(r.Context(), ctxKeyRequestID, id)
next.ServeHTTP(w, r.WithContext(ctx))
})
}
// maxRequestIDLen caps an inbound X-Request-Id we are willing to echo and persist.
// 64 characters comfortably fits a UUID or a typical distributed-trace id while
// bounding what reaches audit_logs.request_id.
const maxRequestIDLen = 64
// validRequestID reports whether an inbound X-Request-Id is safe to echo and store:
// non-empty, within maxRequestIDLen, and restricted to an unambiguous, log-safe
// charset (ASCII alphanumerics plus '-', '_', '.'). The byte-length check bounds it
// regardless of encoding, and the charset excludes whitespace, CR/LF, and every
// other control or multibyte rune, so nothing that survives can pollute a log line
// or the audit row.
func validRequestID(id string) bool {
if id == "" || len(id) > maxRequestIDLen {
return false
}
for _, c := range id {
switch {
case c >= 'a' && c <= 'z', c >= 'A' && c <= 'Z', c >= '0' && c <= '9':
case c == '-', c == '_', c == '.':
default:
return false
}
}
return true
}
// withRecover turns a panicking handler into a 500 envelope instead of a
// dropped connection.
//
// The client gets an opaque "internal error", but the panic value and stack are
// logged FIRST, keyed by the same request_id — the same contract writeError keeps
// for unmapped errors (see errors.go). Without it a recovered panic is an
// untraceable 500: an operator holding "internal error" has nothing to grep for,
// and diagnosis degrades into guessing against a live install.
func withRecover(next http.Handler) http.Handler {
return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
defer func() {
if rec := recover(); rec != nil {
log.Printf("api: %s %s: panic (request_id=%s): %v\n%s",
r.Method, r.URL.Path, requestIDFromContext(r.Context()), rec, debug.Stack())
writeError(w, r, newError(http.StatusInternalServerError, "panic", "internal error"))
}
}()
next.ServeHTTP(w, r)
})
}
// requireInternal enforces service-token auth for the internal face. It never
// applies Zero Trust (spec §14 red line).
func (a *API) requireInternal(next http.Handler) http.Handler {
return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
if err := a.Internal.Authenticate(r); err != nil {
writeError(w, r, errUnauthorized)
return
}
next.ServeHTTP(w, r)
})
}
// requireExternal enforces Access-JWT auth for the external face and stashes the
// resolved Principal in the request context.
func (a *API) requireExternal(next http.Handler) http.Handler {
return http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
p, err := a.External.Authenticate(r)
if err != nil || p == nil {
writeError(w, r, errUnauthorized)
return
}
// op.console door gate: the operator console is staff-only, so a request that
// arrives on the admin host from a non-admin principal is refused HERE, before
// any handler. Authentication alone (a player's passkey/email/bind session) is
// not access — internal permission is verified on top of it, so possessing a
// valid credential never "lets you in" to op.console. On the player console
// (console.<root_domain>) hostIsAdminConsole is false, so this is inert; in
// production Cloudflare Access already blocks non-staff at the edge and this is
// the defense-in-depth backstop for the passwordless (no-Zero-Trust) face.
if hostIsAdminConsole(r, a.RootDomain, a.AdminHostname) && !p.IsAdmin() {
writeError(w, r, errForbidden)
return
}
ctx := context.WithValue(r.Context(), ctxKeyPrincipal, p)
next.ServeHTTP(w, r.WithContext(ctx))
})
}
// adminOnly gates an external-face handler on the admin Zero-Trust path. The
// Access middleware has already authenticated; this enforces that admin-tier
// operations both carry role=admin AND arrived via admin.* (spec §14).
func (a *API) adminOnly(next http.HandlerFunc) http.HandlerFunc {
return func(w http.ResponseWriter, r *http.Request) {
if p := principalFromContext(r.Context()); !p.IsAdmin() {
writeError(w, r, errForbidden)
return
}
next(w, r)
}
}
// ownerOnly gates a handler on the owner role — the single platform-level
// identity above admin. It is stricter than adminOnly: a plain admin with
// role=admin and valid admin Access path is still refused here. The owner
// arrives through the same admin Zero-Trust path, so adminOnly is not a
// prerequisite (the two guards are orthogonal).
func (a *API) ownerOnly(next http.HandlerFunc) http.HandlerFunc {
return func(w http.ResponseWriter, r *http.Request) {
if p := principalFromContext(r.Context()); !p.IsOwner() {
writeError(w, r, errForbidden)
return
}
next(w, r)
}
}
// newRequestID returns a short random hex id. crypto/rand never fails on the
// platforms we target; on the impossible error path we fall back to a constant
// so a request still gets a (non-unique) id rather than crashing.
func newRequestID() string {
var b [8]byte
if _, err := rand.Read(b[:]); err != nil {
return "req-unknown"
}
return hex.EncodeToString(b[:])
}