@

v0.129.0: CAMPAIGN-4 fixes — rate-limiter key (F-B) + volume-blind estimate (F-A) + no-op claim status (F-C)

F-B (MED, security): shared clientIP(r) helper (XFF first-hop, else SplitHostPort
host, else raw) replaces requestIP + the duplicated inline derivation in handleLogin,
so login AND escrow re-auth key on the port-stripped host IP — distinct direct
connections no longer evade the failed-attempt counter. XFF-trust out of scope (commented).

F-A (MED, honesty): volumeSizer seam reads volume size from a container view
(docker run --rm -v vol:/vol:ro alpine du -sb /vol), replacing the host-path du that
returned 0 inside the containerized controller. Failed read -> size_unknown +
fits_on_dest forced false (never "fits"). Export pre-flight hard-aborts only on a
KNOWN doesn.t-fit. HDD branch unchanged.

F-C (LOW-MED): escrowClaimAPIHandler relays agent 404 -> clean 404 and 409 -> 409;
410 and genuine-unreachable 502 unchanged (was: 404 fell through to 502).

Tests + red-proofs: ratelimit_ip_test.go (F-B x6), estimate_volsize_test.go (F-A x3),
TestEscrowClaim_ProxySemantics +3 (F-C). Alpine busybox du -sb verified prod-valid.

Claude-Session: https://claude.ai/code/session_01LbMm4T7Ayzs1unB9pN6Uqd
@

controller/internal/web/auth.go

@@ -148,12 +148,10 @@ func (s *Server) handleLogin(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 
-	// Rate limit: check failed attempts from this IP
-	ip := r.RemoteAddr
-	if fwd := r.Header.Get("X-Forwarded-For"); fwd != "" {
-		ip = strings.Split(fwd, ",")[0]
-	}
-	ip = strings.TrimSpace(ip)
+	// Rate limit: check failed attempts from this host. clientIP strips the ephemeral port
+	// (CAMPAIGN-4 F-B) so distinct direct connections from one host share a key and the counter
+	// actually accrues; XFF first-hop still wins for proxied clients.
+	ip := clientIP(r)
 
 	s.loginAttemptMu.Lock()
 	attempt := s.loginAttempts[ip]

controller/internal/web/claim.go

@@ -9,6 +9,7 @@ import (
 	"encoding/json"
 	"fmt"
 	"io"
+	"net"
 	"net/http"
 	"os"
 	"strings"
@@ -193,12 +194,26 @@ func (s *Server) claimNow() time.Time {
 	return time.Now()
 }
 
-func requestIP(r *http.Request) string {
-	ip := r.RemoteAddr
+// clientIP returns the client IP used as the rate-limiter key. Order: the X-Forwarded-For first
+// hop (set by the traefik/Cloudflare proxy) wins; otherwise the HOST portion of RemoteAddr with the
+// ephemeral PORT stripped (net.SplitHostPort). This is the CAMPAIGN-4 F-B fix: keying on the raw
+// RemoteAddr (IP:PORT) meant every fresh direct connection from one host got a distinct ephemeral
+// port → a distinct key → the failed-attempt counter never accrued, so a direct-to-controller
+// (LAN/guest, non-proxied) path had NO brute-force protection. A RemoteAddr with no port
+// (tests/edge) or an IPv6 form is handled by SplitHostPort, falling back to the raw value.
+//
+// Accepted limitation (out of scope here): X-Forwarded-For is attacker-controlled on a direct path,
+// so a client rotating the first hop still evades the per-IP counter. This fix only closes the
+// port-in-key bug so the proxied / stable-source-IP case — the real deployment — works; it does NOT
+// attempt to establish XFF trust.
+func clientIP(r *http.Request) string {
 	if fwd := r.Header.Get("X-Forwarded-For"); fwd != "" {
-		ip = strings.Split(fwd, ",")[0]
+		return strings.TrimSpace(strings.Split(fwd, ",")[0])
 	}
-	return strings.TrimSpace(ip)
+	if host, _, err := net.SplitHostPort(r.RemoteAddr); err == nil {
+		return host
+	}
+	return strings.TrimSpace(r.RemoteAddr)
 }
 
 // ── the pages ────────────────────────────────────────────────────────────────────────────────
@@ -260,7 +275,7 @@ func (s *Server) handleClaimSubmit(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	wasReset := s.authEnabled() // a password already set → this is a reset, not a first-claim
-	ip := requestIP(r)
+	ip := clientIP(r)
 
 	if locked, _ := s.claimRateLocked(); locked {
 		s.handleClaimPage(w, r, "Túl sok próbálkozás — próbáld újra 15 perc múlva.", "")

controller/internal/web/escrow_handlers.go

@@ -173,7 +173,7 @@ func (s *Server) escrowStartAPIHandler(w http.ResponseWriter, r *http.Request) {
 		escrowJSON(w, http.StatusForbidden, nil, "A vezérlőpult jelszava nincs beállítva — előbb állítson be jelszót.")
 		return
 	}
-	ip := requestIP(r)
+	ip := clientIP(r)
 	if s.escrowRateLimited(ip) {
 		s.logger.Printf("[WARN] [web] escrow start rate limited for %s", ip)
 		escrowJSON(w, http.StatusTooManyRequests, nil, "Túl sok sikertelen próbálkozás, próbálja újra 1 perc múlva.")
@@ -254,12 +254,20 @@ func (s *Server) escrowClaimAPIHandler(w http.ResponseWriter, r *http.Request) {
 	}
 	code, status, err := agent.EscrowCeremonyClaim(r.Context())
 	if err != nil {
-		if status == http.StatusGone {
+		switch status {
+		case http.StatusGone: // 410 — the code was minted but never shown; permanently void.
 			escrowJSON(w, http.StatusGone, nil, "A kód létrejött, de nem lett megjelenítve — biztonsági okból újra nem kérhető le. Indítsa újra a folyamatot: az új kód a régit érvényteleníti.")
-			return
+		case http.StatusNotFound: // 404 — no ceremony has run (e.g. phase:none post-reboot). F-C:
+			// this used to fall through to a 502; a bad-gateway class code for "nothing to claim"
+			// is wrong. Relay a clean, honest 4xx.
+			escrowJSON(w, http.StatusNotFound, nil, "Nincs aktív helyreállítási folyamat — előbb indítsa el a kódkészítést.")
+		case http.StatusConflict: // 409 — the ceremony state doesn't allow a claim right now. F-C.
+			escrowJSON(w, http.StatusConflict, nil, "A folyamat jelenlegi állapotában a kód nem kérhető le.")
+		default: // status 0 (agent unreachable / transport error) or a genuine agent 5xx — a real
+			// bad gateway; keep 502.
+			s.logger.Printf("[WARN] [web] escrow claim failed (status %d)", status) // reason text may echo agent detail; the code itself is never in errors
+			escrowJSON(w, http.StatusBadGateway, nil, "A kód lekérése nem sikerült.")
 		}
-		s.logger.Printf("[WARN] [web] escrow claim failed (status %d)", status) // reason text may echo agent detail; the code itself is never in errors
-		escrowJSON(w, http.StatusBadGateway, nil, "A kód lekérése nem sikerült.")
 		return
 	}
 	s.logger.Printf("[INFO] [web] escrow recovery code claimed (one-shot; not logged)")

controller/internal/web/escrow_wizard_test.go

@@ -286,6 +286,43 @@ func TestEscrowClaim_ProxySemantics(t *testing.T) {
 			t.Fatalf("gone: got %d %s", w.Code, w.Body.String())
 		}
 	})
+	// Scenario I (F-C fix): agent 404 "no ceremony has run" → a clean 404, NOT 502.
+	t.Run("no_active_ceremony_404_not_502", func(t *testing.T) {
+		h := newEscrowWizardHarness(t)
+		h.agent.claimStatus = http.StatusNotFound
+		h.agent.claimErr = fmt.Errorf("no ceremony has run")
+		w := httptest.NewRecorder()
+		h.s.escrowClaimAPIHandler(w, httptest.NewRequest("POST", "/api/escrow/claim", nil))
+		if w.Code != http.StatusNotFound {
+			t.Fatalf("no-ceremony claim must be 404, got %d %s", w.Code, w.Body.String())
+		}
+		if !strings.Contains(w.Body.String(), "Nincs aktív helyreállítási folyamat") {
+			t.Fatalf("expected honest Hungarian no-ceremony message, got %s", w.Body.String())
+		}
+	})
+	// Scenario J (regression): agent 409 → passed through as 409, not 502.
+	t.Run("conflict_409", func(t *testing.T) {
+		h := newEscrowWizardHarness(t)
+		h.agent.claimStatus = http.StatusConflict
+		h.agent.claimErr = fmt.Errorf("conflict")
+		w := httptest.NewRecorder()
+		h.s.escrowClaimAPIHandler(w, httptest.NewRequest("POST", "/api/escrow/claim", nil))
+		if w.Code != http.StatusConflict {
+			t.Fatalf("409 must pass through, got %d", w.Code)
+		}
+	})
+	// Scenario K (regression): a genuinely-unreachable agent (status 0) stays 502 — that IS a real
+	// bad gateway; the fix must NOT over-correct it to a 4xx.
+	t.Run("unreachable_stays_502", func(t *testing.T) {
+		h := newEscrowWizardHarness(t)
+		h.agent.claimStatus = 0
+		h.agent.claimErr = fmt.Errorf("dial tcp: connection refused")
+		w := httptest.NewRecorder()
+		h.s.escrowClaimAPIHandler(w, httptest.NewRequest("POST", "/api/escrow/claim", nil))
+		if w.Code != http.StatusBadGateway {
+			t.Fatalf("unreachable agent must stay 502, got %d", w.Code)
+		}
+	})
 }
 
 // The status proxy relays the agent's non-secret job view verbatim.

controller/internal/web/ratelimit_ip_test.go

@@ -0,0 +1,151 @@
+package web
+
+import (
+	"fmt"
+	"io"
+	"log"
+	"net/http"
+	"net/http/httptest"
+	"net/url"
+	"strings"
+	"testing"
+
+	"gitea.dooplex.hu/admin/felhom-controller/internal/config"
+	"golang.org/x/crypto/bcrypt"
+)
+
+// rateLimitTestServer builds a minimal Server with a real password hash and an initialized
+// loginAttempts map, templates loaded (renderLogin needs s.tmpl). No agent/backup/settings — the
+// re-auth rate limiter fires before any of those.
+func rateLimitTestServer(t *testing.T) *Server {
+	t.Helper()
+	cfg := &config.Config{}
+	hash, _ := bcrypt.GenerateFromPassword([]byte("correct-pass"), bcrypt.MinCost)
+	cfg.Web.PasswordHash = string(hash)
+	s := &Server{cfg: cfg, logger: log.New(io.Discard, "", 0), version: "test",
+		loginAttempts: map[string]*loginAttempt{}, sessions: map[string]*session{}}
+	s.loadTemplates()
+	return s
+}
+
+func doLogin(s *Server, remoteAddr, xff, password string) *httptest.ResponseRecorder {
+	form := url.Values{"password": {password}}
+	r := httptest.NewRequest(http.MethodPost, "/login", strings.NewReader(form.Encode()))
+	r.Header.Set("Content-Type", "application/x-www-form-urlencoded")
+	r.RemoteAddr = remoteAddr
+	if xff != "" {
+		r.Header.Set("X-Forwarded-For", xff)
+	}
+	w := httptest.NewRecorder()
+	s.handleLogin(w, r)
+	return w
+}
+
+func ex(s string) string {
+	if len(s) > 200 {
+		return s[:200]
+	}
+	return s
+}
+
+// Scenario A (the F-B fix): 6 failed logins from ONE host over DISTINCT ephemeral ports and NO
+// X-Forwarded-For must share a rate-limit key and trip the limiter on attempt 6. Pre-fix (raw
+// RemoteAddr key) each port is its own key → never limited. This is the F-B red-proof anchor.
+func TestLoginRateLimit_DirectDistinctPorts_Limited(t *testing.T) {
+	s := rateLimitTestServer(t)
+	var last string
+	for i := 1; i <= 6; i++ {
+		last = doLogin(s, fmt.Sprintf("127.0.0.1:%d", 5000+i), "", "wrong").Body.String()
+		if i < 6 && !strings.Contains(last, "Hibás jelszó") {
+			t.Fatalf("attempt %d expected Hibás jelszó, got: %s", i, ex(last))
+		}
+	}
+	if !strings.Contains(last, "Túl sok sikertelen") {
+		t.Fatalf("attempt 6 (distinct ports, no XFF) MUST be rate-limited; got: %s", ex(last))
+	}
+}
+
+// Scenario B (regression): stable X-Forwarded-For across 6 attempts still limits on 6.
+func TestLoginRateLimit_StableXFF_Limited(t *testing.T) {
+	s := rateLimitTestServer(t)
+	var last string
+	for i := 1; i <= 6; i++ {
+		last = doLogin(s, fmt.Sprintf("10.9.9.9:%d", 5000+i), "203.0.113.9", "wrong").Body.String()
+	}
+	if !strings.Contains(last, "Túl sok sikertelen") {
+		t.Fatalf("attempt 6 with a stable XFF MUST be rate-limited; got: %s", ex(last))
+	}
+}
+
+// Scenario C (documented accepted limitation): rotating the X-Forwarded-For first hop evades the
+// per-IP counter. This is NOT what the fix targets (XFF is attacker-controlled on a direct path);
+// the test pins the known behavior so a future XFF-trust change is a conscious decision.
+func TestLoginRateLimit_RotatingXFF_NotLimited(t *testing.T) {
+	s := rateLimitTestServer(t)
+	var last string
+	for i := 1; i <= 6; i++ {
+		last = doLogin(s, "10.9.9.9:5000", fmt.Sprintf("203.0.113.%d", i), "wrong").Body.String()
+	}
+	if strings.Contains(last, "Túl sok sikertelen") {
+		t.Fatalf("rotating XFF is a known evasion (out of scope) — expected NOT limited")
+	}
+	if !strings.Contains(last, "Hibás jelszó") {
+		t.Fatalf("expected Hibás jelszó on rotating XFF; got: %s", ex(last))
+	}
+}
+
+// Scenario D: the escrow wizard re-auth path shares the SAME fixed key (proves the shared clientIP
+// helper reached escrow_handlers, not just handleLogin). 6 wrong wizard passwords over distinct
+// ports → 429 on attempt 6.
+func TestEscrowReauthRateLimit_SharesFixedKey(t *testing.T) {
+	s := rateLimitTestServer(t)
+	var code int
+	for i := 1; i <= 6; i++ {
+		form := url.Values{"password": {"wrong"}}
+		r := httptest.NewRequest(http.MethodPost, "/api/escrow/start", strings.NewReader(form.Encode()))
+		r.Header.Set("Content-Type", "application/x-www-form-urlencoded")
+		r.RemoteAddr = fmt.Sprintf("127.0.0.1:%d", 6000+i)
+		w := httptest.NewRecorder()
+		s.escrowStartAPIHandler(w, r)
+		code = w.Code
+	}
+	if code != http.StatusTooManyRequests {
+		t.Fatalf("escrow re-auth attempt 6 (distinct ports, no XFF) MUST be 429; got %d", code)
+	}
+}
+
+// Scenario E: a successful login clears the counter for that host key.
+func TestLoginRateLimit_SuccessClearsCounter(t *testing.T) {
+	s := rateLimitTestServer(t)
+	for i := 1; i <= 4; i++ {
+		doLogin(s, fmt.Sprintf("127.0.0.1:%d", 7000+i), "", "wrong")
+	}
+	if s.loginAttempts["127.0.0.1"] == nil || s.loginAttempts["127.0.0.1"].count != 4 {
+		t.Fatalf("expected 4 accrued failures on the shared key before success")
+	}
+	doLogin(s, "127.0.0.1:7099", "", "correct-pass") // success
+	if s.loginAttempts["127.0.0.1"] != nil {
+		t.Fatalf("a successful login must clear the host's failure counter")
+	}
+}
+
+// clientIP unit: port stripped; XFF first-hop wins; no-port and IPv6 handled.
+func TestClientIP_StripsPort(t *testing.T) {
+	cases := []struct{ remote, xff, want string }{
+		{"127.0.0.1:5001", "", "127.0.0.1"},
+		{"127.0.0.1:5002", "203.0.113.9", "203.0.113.9"},
+		{"[::1]:443", "", "::1"},
+		{"192.168.0.5", "", "192.168.0.5"}, // no port → raw
+		{"10.0.0.1:80", "198.51.100.7, 203.0.113.9", "198.51.100.7"},
+	}
+	for _, c := range cases {
+		r := httptest.NewRequest(http.MethodGet, "/", nil)
+		r.RemoteAddr = c.remote
+		if c.xff != "" {
+			r.Header.Set("X-Forwarded-For", c.xff)
+		}
+		if got := clientIP(r); got != c.want {
+			t.Errorf("clientIP(remote=%q xff=%q) = %q, want %q", c.remote, c.xff, got, c.want)
+		}
+	}
+}