servewaf always answers rproxy well within its time limit (closes #182)

A question waits on GeoJS at most 800ms, the new default of
SWWAF_LOOKUP_TIMEOUT. It does not wait at all when the checks will refuse
it whatever GeoJS answers. Just before the Core Rule Set, servewaf refuses
more than 50 cookies or 100 header fields with 431.

Shared default changed: SWWAF_LOOKUP_TIMEOUT also applies to smallwebwaf
with no command.
Judgement call: SWWAF_REPUTATION_TIMEOUT stays 2s, since no request waits
on the DNS blocklists or AbuseIPDB.

Model: opus-5-5

SPEC.md

@@ -1102,8 +1102,9 @@ metrics.
 
 ### The answer
 
-`servewaf` runs steps 3 to 12 of "What happens to a request", in that order.
-Every answer but the last below carries `X-Smallwebwaf-Disposition`.
+`servewaf` runs steps 3 to 12 of "What happens to a request", in that order,
+with the header limits just before step 12 (see "Answering in time"). Every
+answer but the last below carries `X-Smallwebwaf-Disposition`.
 
 - `X-Smallwebwaf-Disposition: allow`, status `200`, no body. `rproxy` passes the
   request to the app. With `SWWAF_ADD_LOOKUP_HEADERS` the answer also carries
@@ -1114,6 +1115,7 @@ Every answer but the last below carries `X-Smallwebwaf-Disposition`.
   status is one of:
     - `SWWAF_BAN_RESPONSE`, for a refusal;
     - `403`, for a `block` rule or the Core Rule Set;
+    - `431`, for a request over the header limits;
     - `404`, for a path under `/_smallwebwaf/` that passes the checks, so that
       the app never sees that prefix.
 - `X-Smallwebwaf-Disposition: close`, status `200`, no body. This takes the
@@ -1131,6 +1133,66 @@ The Core Rule Set inspects the body a question carries only while
 does not inspect a JSON or XML body of 64 KiB or more, which may have been cut
 short.
 
+### Answering in time
+
+`rproxy` stops waiting for `servewaf` after `SWRP_TIMEOUT_MS`, by default 2
+seconds, and `SWRP_DEFAULT_DISPOSITION` then decides. So at the defaults,
+`servewaf` answers a request well within that time, whatever the request and
+whatever its outside sources do.
+
+GeoJS is the only outside source a question can wait on. A question waits only
+when step 5 says so, and for at most `SWWAF_LOOKUP_TIMEOUT`, by default 800
+milliseconds. A client without an answer by then counts as unknown. No question
+waits on these, which run in the background:
+
+- the DNS blocklists and AbuseIPDB;
+- fetching the blocklists, the CrowdSec decision list and the file
+  `SWWAF_ASN_LIMIT_PERCENT_URL` names.
+
+A question does not wait for GeoJS when the checks will refuse it whatever GeoJS
+answers. That is a request:
+
+- over the header limits;
+- from a client on a fetched blocklist, while `SWWAF_BLOCKLIST_ACTION` is
+  `deny`;
+- from a client on the CrowdSec decision list;
+- that would take its client over a rate limit at 100% of it, since a percentage
+  can only lower a limit.
+
+Such a request still goes through the steps in their order. Its client's AS
+number and country stay unknown until GeoJS has answered. So while
+`SWWAF_EXCLUSIVELY_ALLOWED_COUNTRIES` is set, that list can refuse it first.
+
+The static lists and bans come before the lookup, so they never wait. These can
+still wait for GeoJS:
+
+- a country list, which needs the answer;
+- a rate limit a percentage has lowered, which needs the answer;
+- a DNS blocklist or AbuseIPDB, while `SWWAF_REPUTATION_ACTION` is `deny`.
+
+The header limits are fixed, with no setting. `servewaf` refuses a request over
+either one:
+
+- more than 50 cookies. Each part of its `Cookie` headers between semicolons
+  counts as one, unless it is empty.
+- more than 100 header fields. A header sent twice counts twice. `Host` and
+  `Transfer-Encoding` do not count. `X-Request-ID` and `X-Forwarded-Proto` do.
+
+Real clients never reach these limits. The answer is `deny` with `431`, and the
+log line's action is `too_large`. The refusal makes no ban and is no offence.
+The rate limits count the request as usual, so a client that keeps sending such
+requests breaks a limit.
+
+The limits apply on every path and in every `SWWAF_WAF_MODE`. A client in
+`SWWAF_ALLOW_NETS` skips them, as it skips the Core Rule Set. In `observe` mode
+such a request is answered `allow` with `would_action`, and the Core Rule Set
+does not read it.
+
+The Core Rule Set runs each of its rules on each cookie, so thousands of small
+cookies would keep it busy for over a second. Within the limits, one long cookie
+or query string can still keep it busy for a fraction of a second. So enough
+such requests at once can overload `servewaf` past `SWRP_TIMEOUT_MS`.
+
 ### The report
 
 Once the visitor's answer has ended, `rproxy` sends `POST /_smallwebwaf/report`

docs/metrics.md

@@ -17,7 +17,8 @@ label as `exported_instance` unless the scrape sets `honor_labels: true`.
 
 - `smallwebwaf_rate_limit_hits_total`, by `window` and by `kind` (`requests`,
   `bytes` or `refusals`);
-- `smallwebwaf_size_and_time_limit_hits_total`, by `limit`, the setting passed;
+- `smallwebwaf_size_and_time_limit_hits_total`, by `limit`, the setting passed.
+  `servewaf`'s header limits have no setting, so it leaves them out;
 - `smallwebwaf_offences_total`, by `kind`;
 - `smallwebwaf_bans_made_total`, by `cause`;
 - `smallwebwaf_active_bans` and `smallwebwaf_permanent_bans`.

docs/request-log.md

@@ -67,7 +67,8 @@ Every line has:
       CrowdSec decision, each of which bans the client;
     - `country_denied`, `rate_limited`, `rule_blocked` and `waf_blocked`:
       refused by that check;
-    - `too_large` and `timed_out`: a size or time limit was passed;
+    - `too_large`: a size limit was passed, or a header limit in `servewaf`;
+    - `timed_out`: a time limit was passed;
     - `upstream_error`: the app could not be reached, or its answer broke off;
     - `admin`: answered at one of `smallwebwaf`'s own endpoints.
 - `would_action`: in `observe` mode, what `enforce` mode would have done with a

docs/settings.md

@@ -157,7 +157,7 @@ that take `off`, the `_FILE` form, and what each group does and why.
     - an anomaly threshold per AS number.
 - `SWWAF_LOOKUP_DB_PATH` (default empty): the IPinfo Lite database file, in its
   `.mmdb` form, for `SWWAF_LOOKUP_SOURCE=file`.
-- `SWWAF_LOOKUP_TIMEOUT` (default `1s`): how long a request waits for its
+- `SWWAF_LOOKUP_TIMEOUT` (default `800ms`): how long a request waits for its
   client's first answer from GeoJS, and how long a request to GeoJS may take. A
   duration; never `off`.
 - `SWWAF_ADD_LOOKUP_HEADERS` (default `false`): `true` passes the app the
@@ -340,7 +340,10 @@ A few limits are fixed rather than settings:
 - at most 1,000 DNS blocklist queries are under way at once;
 - at most 20,000 anomaly counters are kept;
 - each alert destination's queue holds at most 1,000 alerts;
-- `servewaf` holds at most 10,000 answered requests awaiting their reports.
+- `servewaf` holds at most 10,000 answered requests awaiting their reports;
+- `servewaf` refuses a request with more than 50 cookies or more than 100 header
+  fields (see "Answering in time" under "`servewaf`" in
+  [`SPEC.md`](../SPEC.md)).
 
 ## `servewaf`
 
@@ -358,3 +361,9 @@ at start still shows them.
 request to an app. `SWWAF_WAF_BODY_LIMIT` has an effect up to 64 KiB less one
 byte, whatever its value. A question carries no more of a body than 64 KiB (see
 "`servewaf`" in [`SPEC.md`](../SPEC.md)).
+
+`SWWAF_LOOKUP_TIMEOUT` is the longest a question waits on an outside source.
+Keep it well under `rproxy`'s `SWRP_TIMEOUT_MS`, as the default is. No question
+waits for `SWWAF_REPUTATION_TIMEOUT`, since the DNS blocklists and AbuseIPDB are
+asked in the background (see "Answering in time" under "`servewaf`" in
+[`SPEC.md`](../SPEC.md)).

internal/config/config.go

@@ -516,7 +516,7 @@ func FromEnvironment(lookupEnv func(string) (string, bool)) (*Config, error) {
 		BytesCount:              env.bytesCount("SWWAF_BYTES_COUNT", "both"),
 		LookupSource:            env.lookupSource("SWWAF_LOOKUP_SOURCE", "geojs"),
 		LookupDBPath:            env.value("SWWAF_LOOKUP_DB_PATH", ""),
-		LookupTimeout:           env.durationNotOff("SWWAF_LOOKUP_TIMEOUT", "1s"),
+		LookupTimeout:           env.durationNotOff("SWWAF_LOOKUP_TIMEOUT", "800ms"),
 		AddLookupHeaders:        env.boolean("SWWAF_ADD_LOOKUP_HEADERS", "false"),
 		DeniedCountries:         env.countries("SWWAF_DENIED_COUNTRIES", ""),
 		ExclusivelyAllowedCountries: env.countries(

internal/config/config_test.go

@@ -281,7 +281,7 @@ func TestDefaults(t *testing.T) {
 		RulesEnabled:                true,
 	})
 	wantLookupSettings(t, cfg, config.Config{
-		LookupSource: defaultLookupSource, LookupTimeout: time.Second,
+		LookupSource: defaultLookupSource, LookupTimeout: 800 * time.Millisecond,
 	})
 	wantByteLimitSettings(t, cfg, config.Config{
 		BytesLimitPerMinute: 4 << 30, BytesLimitPerHour: 20 << 30,
@@ -1165,7 +1165,7 @@ func TestLookupSettingsAsSet(t *testing.T) {
 
 	cfg = fromEnvironment(t, environment{lookupSource: off})
 	wantLookupSettings(t, cfg, config.Config{
-		LookupSource: off, LookupTimeout: time.Second,
+		LookupSource: off, LookupTimeout: 800 * time.Millisecond,
 	})
 }
 
@@ -2385,7 +2385,7 @@ func TestLogsEachSettingWithItsValue(t *testing.T) {
 		bytesCount:              "both",
 		lookupSource:            defaultLookupSource,
 		lookupDBPath:            "",
-		lookupTimeout:           "1s",
+		lookupTimeout:           "800ms",
 		addLookupHeaders:        "false",
 		deniedCountries:         "",
 		allowedCountries:        "",

internal/lookup/lookup.go

@@ -179,7 +179,7 @@ func New(params Params) *GeoJS {
 // own address for IPv4, and an address in the same place for an IPv6
 // netblock.
 func (g *GeoJS) LookUp(ctx context.Context, client netip.Prefix) Answer {
-	answer, asked := g.answerOrWait(ctx, client)
+	answer, asked := g.answerOrWait(ctx, client, g.wait)
 	if asked == nil {
 		return answer
 	}
@@ -209,6 +209,14 @@ func (g *GeoJS) LookUp(ctx context.Context, client netip.Prefix) Answer {
 	return answer
 }
 
+// LookUpWithoutWaiting is LookUp for a request that does not need the answer,
+// whatever Wait says: it returns the zero Answer at once when none is kept.
+func (g *GeoJS) LookUpWithoutWaiting(ctx context.Context, client netip.Prefix) Answer {
+	answer, _ := g.answerOrWait(ctx, client, false)
+
+	return answer
+}
+
 // Kept returns client's answer, if one is kept, without asking GeoJS.
 func (g *GeoJS) Kept(client netip.Prefix) (Answer, bool) {
 	g.mu.Lock()
@@ -263,9 +271,9 @@ func (g *GeoJS) Load(answers []Answer) {
 // answerOrWait returns client's kept answer if it has one. Otherwise it
 // puts the client among those waiting if there is room, has GeoJS asked
 // about them if it can be, and returns what to wait on for the answer, or
-// nil when there is nothing to wait for.
+// nil when there is nothing to wait for or mustWait is false.
 func (g *GeoJS) answerOrWait(
-	ctx context.Context, client netip.Prefix,
+	ctx context.Context, client netip.Prefix, mustWait bool,
 ) (Answer, <-chan struct{}) {
 	g.mu.Lock()
 	defer g.mu.Unlock()
@@ -283,7 +291,7 @@ func (g *GeoJS) answerOrWait(
 
 	g.ask(ctx)
 
-	if !g.wait {
+	if !mustWait {
 		return Answer{}, nil // the answer is not needed before the request goes on
 	}
 

internal/lookup/lookup_test.go

@@ -72,6 +72,39 @@ func TestKeptAnswerIsUsedFor7DaysThenAskedAgain(t *testing.T) {
 	})
 }
 
+func TestLookUpWithoutWaitingAsksGeoJSAndGoesOnAtOnce(t *testing.T) {
+	t.Parallel()
+
+	synctest.Test(t, func(t *testing.T) {
+		geojs, clock, g := start()
+		client := netip.MustParsePrefix("203.0.113.9/32")
+
+		// Wait is set, and GeoJS does not answer.
+		geojs.set(hanging)
+
+		began := time.Now()
+		answer := g.LookUpWithoutWaiting(t.Context(), client)
+
+		if waited := time.Since(began); waited != 0 || answer != (lookup.Answer{}) {
+			t.Errorf("got %+v after %s, want no answer at once", answer, waited)
+		}
+
+		waitForRequests(t, geojs, 1)
+		wantAsked(t, geojs, 0, "203.0.113.9")
+
+		// That request is abandoned, and has failed once this returns. GeoJS
+		// answers when it is asked again, and the answer is kept.
+		time.Sleep(timeout)
+		synctest.Wait()
+		geojs.set(answering)
+		waitForCountry(t, g, clock, client, germany)
+
+		if got := g.LookUpWithoutWaiting(t.Context(), client).Country; got != germany {
+			t.Errorf("the kept answer gives the country %q, want %q", got, germany)
+		}
+	})
+}
+
 func TestNewClientWaitsAtMostOneSecondThenCountsAsNotFound(t *testing.T) {
 	t.Parallel()
 
@@ -123,8 +156,8 @@ func TestRequestWaitsAsLongAsTheTimeoutSaysAndGeoJSIsAbandonedAfterIt(t *testing
 	t.Parallel()
 
 	synctest.Test(t, func(t *testing.T) {
-		// A timeout longer than the default second, and a GeoJS that does
-		// not answer.
+		// A timeout longer than the default, and a GeoJS that does not
+		// answer.
 		const longerTimeout = 3 * time.Second
 
 		m := metrics.New(1, "app")

internal/proxy/answertime_test.go

@@ -0,0 +1,472 @@
+package proxy_test
+
+import (
+	"fmt"
+	"maps"
+	"net"
+	"net/http"
+	"net/http/httptest"
+	"net/netip"
+	"slices"
+	"strconv"
+	"strings"
+	"testing"
+	"testing/synctest"
+	"time"
+
+	"sneak.berlin/go/smallwebwaf/internal/bans"
+	"sneak.berlin/go/smallwebwaf/internal/proxy"
+	"sneak.berlin/go/smallwebwaf/internal/ratelimit"
+	"sneak.berlin/go/smallwebwaf/internal/reputation"
+	"sneak.berlin/go/smallwebwaf/internal/requestlog"
+)
+
+const (
+	// answerTimeLimit is the default of rproxy's SWRP_TIMEOUT_MS, the time
+	// servewaf has to answer.
+	answerTimeLimit = 2 * time.Second
+	// defaultLookupWait is the default of SWWAF_LOOKUP_TIMEOUT.
+	defaultLookupWait = 800 * time.Millisecond
+	// The header limits SPEC.md gives.
+	cookieLimit      = 50
+	headerFieldLimit = 100
+	// cookieHeader is the header the cookie limit counts.
+	cookieHeader = "Cookie"
+	// tooLarge is how servewaf refuses a request over the header limits.
+	tooLarge = http.StatusRequestHeaderFieldsTooLarge
+	// silentAbuseIPDBURL is where AbuseIPDB never answers: at the stand-in
+	// that unansweredGeoJSURL names too.
+	silentAbuseIPDBURL = "unanswered://abuseipdb/api/v2/check"
+	// wholeLimit is the percentage of a rate limit that lowers it not at all.
+	wholeLimit = 100
+	// clientsDecision is the CrowdSec decision list holding client alone.
+	clientsDecision = `[{"duration": "4h0m0s", "scenario": "crowdsecurity/ssh-bf", ` +
+		`"scope": "Ip", "type": "ban", "value": "` + client + `"}]`
+)
+
+// cookies returns a Cookie header's value with count cookies.
+func cookies(count int) string {
+	parts := make([]string, 0, count)
+	for i := range count {
+		parts = append(parts, "c"+strconv.Itoa(i)+"=1")
+	}
+
+	return strings.Join(parts, "; ")
+}
+
+// fillHeaders adds headers to a question until the checks see total header
+// fields: all but ClientHeader and PeerHeader, which servewaf removes. Go's
+// client sends nothing else but Host, which the limit does not count, once
+// User-Agent is set.
+func fillHeaders(req *http.Request, total int) {
+	req.Header.Set("User-Agent", "test")
+
+	for i := len(req.Header) - 2; i < total; i++ {
+		req.Header.Set("X-Filler-"+strconv.Itoa(i), "1")
+	}
+}
+
+func TestSixteenThousandCookiesAreRefusedWellWithinTheTimeLimit(t *testing.T) {
+	t.Parallel()
+
+	s := startServeWAF(t, "", map[string]string{wafMode: block})
+	req := s.question(siteA, client, "/")
+	req.Header.Set(cookieHeader, strings.Repeat("a;", 16000))
+
+	began := time.Now()
+	got := do(t, req)
+
+	if took := time.Since(began); took >= answerTimeLimit/2 {
+		t.Errorf("answered after %s, want well within %s", took, answerTimeLimit)
+	}
+
+	wantDisposition(t, got, tooLarge, proxy.DispositionDeny)
+
+	line := s.out.requestLine(t)
+	wantLine(t, line, tooLarge, requestlog.ActionTooLarge)
+
+	if line.DurationWAF != nil {
+		t.Errorf("the Core Rule Set read the request, for %f ms", *line.DurationWAF)
+	}
+}
+
+func TestEachHeaderLimitRefusesARequestOverIt(t *testing.T) {
+	t.Parallel()
+
+	repeated := make([]string, headerFieldLimit)
+	for i := range repeated {
+		repeated[i] = "1"
+	}
+
+	for _, tc := range []struct {
+		name string
+		make func(*http.Request)
+		// refused is true for a request over a limit, and false for one at it.
+		refused bool
+	}{
+		{"cookies at the limit", func(req *http.Request) {
+			req.Header.Set(cookieHeader, cookies(cookieLimit))
+		}, false},
+		{"cookies over the limit", func(req *http.Request) {
+			req.Header.Set(cookieHeader, cookies(cookieLimit+1))
+		}, true},
+		{"empty parts are no cookies", func(req *http.Request) {
+			req.Header.Set(cookieHeader, cookies(cookieLimit)+"; ;;")
+		}, false},
+		{"cookies over the limit across two headers", func(req *http.Request) {
+			req.Header[cookieHeader] = []string{cookies(cookieLimit), "last=1"}
+		}, true},
+		{"cookies without a value count", func(req *http.Request) {
+			req.Header.Set(cookieHeader, strings.Repeat("a;", cookieLimit+1))
+		}, true},
+		{"header fields at the limit", func(req *http.Request) {
+			fillHeaders(req, headerFieldLimit)
+		}, false},
+		{"header fields over the limit", func(req *http.Request) {
+			fillHeaders(req, headerFieldLimit+1)
+		}, true},
+		{"a header sent many times counts each time", func(req *http.Request) {
+			req.Header["X-Filler"] = repeated
+		}, true},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			s := startServeWAF(t, "", map[string]string{wafMode: block})
+			req := s.question(siteA, client, "/")
+			tc.make(req)
+
+			got := do(t, req)
+			line := s.out.requestLine(t)
+
+			if tc.refused {
+				wantDisposition(t, got, tooLarge, proxy.DispositionDeny)
+				wantLine(t, line, tooLarge, requestlog.ActionTooLarge)
+
+				return
+			}
+
+			wantDisposition(t, got, http.StatusOK, proxy.DispositionAllow)
+			wantLine(t, line, http.StatusOK, requestlog.ActionForward)
+
+			if line.DurationWAF == nil {
+				t.Error("the Core Rule Set did not read a request within the limits")
+			}
+		})
+	}
+}
+
+func TestHeaderLimitsHoldWhateverTheCoreRuleSetSettingsSay(t *testing.T) {
+	t.Parallel()
+
+	for _, env := range []map[string]string{
+		{wafMode: off},
+		{wafMode: "detect"},
+		{wafMode: block, "SWWAF_WAF_EXEMPT_PATHS": "/"},
+	} {
+		t.Run(fmt.Sprint(env), func(t *testing.T) {
+			t.Parallel()
+
+			s := startServeWAF(t, "", env)
+			req := s.question(siteA, client, "/")
+			req.Header.Set(cookieHeader, cookies(cookieLimit+1))
+
+			wantDisposition(t, do(t, req), tooLarge, proxy.DispositionDeny)
+		})
+	}
+}
+
+func TestClientInTheAllowNetsSkipsTheHeaderLimits(t *testing.T) {
+	t.Parallel()
+
+	s := startServeWAF(t, "", map[string]string{wafMode: block, allowNets: client})
+	req := s.question(siteA, client, "/")
+	req.Header.Set(cookieHeader, cookies(cookieLimit+1))
+
+	wantDisposition(t, do(t, req), http.StatusOK, proxy.DispositionAllow)
+}
+
+func TestObserveModeAllowsARequestOverTheHeaderLimitsUnread(t *testing.T) {
+	t.Parallel()
+
+	s := startServeWAF(t, "", map[string]string{mode: observe, wafMode: block})
+	req := s.question(siteA, client, "/")
+	req.Header.Set(cookieHeader, strings.Repeat("a;", 16000))
+
+	began := time.Now()
+	got := do(t, req)
+
+	if took := time.Since(began); took >= answerTimeLimit/2 {
+		t.Errorf("answered after %s, want well within %s", took, answerTimeLimit)
+	}
+
+	wantDisposition(t, got, http.StatusOK, proxy.DispositionAllow)
+
+	line := s.out.requestLine(t)
+	if line.WouldAction != requestlog.ActionTooLarge || line.DurationWAF != nil {
+		t.Errorf("would_action %q and duration_waf %v, want %q and none",
+			line.WouldAction, line.DurationWAF, requestlog.ActionTooLarge)
+	}
+}
+
+func TestRequestsOverTheHeaderLimitsCountTowardTheRateLimits(t *testing.T) {
+	t.Parallel()
+
+	s := startServeWAF(t, "", map[string]string{rateLimitPerMinute: "1"})
+	ask := func() answer {
+		req := s.question(siteA, client, "/")
+		req.Header.Set(cookieHeader, cookies(cookieLimit+1))
+
+		return do(t, req)
+	}
+
+	wantDisposition(t, ask(), tooLarge, proxy.DispositionDeny)
+	wantDisposition(t, ask(), http.StatusForbidden, proxy.DispositionDeny)
+
+	lines := s.out.requestLines(t, 2)
+	wantLine(t, lines[1], http.StatusForbidden, requestlog.ActionRateLimited)
+}
+
+func TestWithoutACommandTheHeaderLimitsDoNotApply(t *testing.T) {
+	t.Parallel()
+
+	app := startApp(t, func(http.ResponseWriter, *http.Request) {})
+	addr, out := startProxy(t, app.URL, nil)
+
+	req := newRequest(t, http.MethodGet, addr, "/", http.NoBody)
+	req.Header.Set(cookieHeader, cookies(cookieLimit+1))
+	fillHeaders(req, headerFieldLimit+1)
+
+	wantStatus(t, do(t, req), http.StatusOK)
+	wantLine(t, out.requestLine(t), http.StatusOK, requestlog.ActionForward)
+}
+
+// newServeWAFInBubble makes a servewaf for a synctest bubble: GeoJS never
+// answers, a setting needs its answer, and everything is at its default but
+// for env. No request to it may use the network.
+func newServeWAFInBubble(t *testing.T, env map[string]string) (*proxy.Server, *output) {
+	t.Helper()
+
+	settings := map[string]string{deniedCountries: "KP"}
+	maps.Copy(settings, env)
+
+	server, out, _ := newServerWith(t, proxy.NewServeWAF, "http://app.invalid",
+		unansweredGeoJSURL, time.Now, settings)
+
+	return server, out
+}
+
+// askDirectly hands a question about a request from client straight to server,
+// as change leaves it, and returns the answer's status and how long it took.
+func askDirectly(
+	t *testing.T, server *proxy.Server, change func(*http.Request),
+) (int, time.Duration) {
+	t.Helper()
+
+	req := httptest.NewRequestWithContext(t.Context(), http.MethodGet, "/", http.NoBody)
+	req.RemoteAddr = net.JoinHostPort(rproxyPeer, "1234")
+	req.Header.Set(proxy.ClientHeader, client)
+	req.Header.Set(requestIDHeader, "the-id")
+	change(req)
+
+	recorder := httptest.NewRecorder()
+	began := time.Now()
+
+	server.Handler.ServeHTTP(recorder, req)
+
+	return recorder.Code, time.Since(began)
+}
+
+// banClient bans client at now.
+func banClient(_ *testing.T, server *proxy.Server, now time.Time) {
+	server.Ledger.BanForLimit(netip.MustParsePrefix(client+"/32"), now, bans.Notes{})
+}
+
+// countClient counts one request from client at now for the rate limits.
+func countClient(_ *testing.T, server *proxy.Server, now time.Time) {
+	server.Limiter.Count(netip.MustParsePrefix(client+"/32"), now, wholeLimit)
+}
+
+// listClient puts client on the blocklist at dropURL.
+func listClient(t *testing.T, server *proxy.Server, now time.Time) {
+	t.Helper()
+
+	err := server.Lists.Load([]reputation.List{{
+		URL: dropURL, Tried: now, Fetched: now, Lines: []string{client},
+	}})
+	if err != nil {
+		t.Fatalf("load the blocklist: %v", err)
+	}
+}
+
+// decideOnClient puts client on the CrowdSec decision list.
+func decideOnClient(t *testing.T, server *proxy.Server, now time.Time) {
+	t.Helper()
+	loadDecisions(t, server, now, clientsDecision)
+}
+
+// leaveAsItIs prepares nothing.
+func leaveAsItIs(*testing.T, *proxy.Server, time.Time) {}
+
+func TestARefusedRequestDoesNotWaitForASlowGeoJS(t *testing.T) {
+	t.Parallel()
+
+	const forbidden = http.StatusForbidden
+
+	for _, tc := range []struct {
+		name string
+		env  map[string]string
+		// prepare gets the server ready, at now, before the question.
+		prepare func(t *testing.T, server *proxy.Server, now time.Time)
+		// cookie is the request's Cookie header.
+		cookie string
+		status int
+		action string
+	}{
+		{
+			name: "SWWAF_DENY_NETS", env: map[string]string{denyNets: client},
+			prepare: leaveAsItIs, status: forbidden, action: requestlog.ActionDenied,
+		},
+		{
+			name:    "a ban",
+			prepare: banClient, status: forbidden, action: requestlog.ActionBanned,
+		},
+		{
+			name: "the rate limits", env: map[string]string{rateLimitPerMinute: "1"},
+			prepare: countClient, status: forbidden, action: requestlog.ActionRateLimited,
+		},
+		{
+			name:    "the rate limits, with a biased threshold",
+			env:     map[string]string{rateLimitPerMinute: "1", unknownLimitPercent: "50"},
+			prepare: countClient, status: forbidden, action: requestlog.ActionRateLimited,
+		},
+		{
+			name: "a blocklist", env: map[string]string{blocklistURLs: dropURL},
+			prepare: listClient, status: forbidden, action: requestlog.ActionDenied,
+		},
+		{
+			name:    "the CrowdSec decision list",
+			env:     map[string]string{crowdSecURL: lapi, crowdSecKey: bouncerKey},
+			prepare: decideOnClient, status: forbidden, action: requestlog.ActionBanned,
+		},
+		{
+			name: "the header limits", cookie: cookies(cookieLimit + 1),
+			prepare: leaveAsItIs, status: tooLarge, action: requestlog.ActionTooLarge,
+		},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			// In a synctest bubble, as TestRequestWaitsAsLongAsTheLookupTimeoutSays
+			// says, so that a request that does not wait takes no time at all.
+			synctest.Test(t, func(t *testing.T) {
+				server, out := newServeWAFInBubble(t, tc.env)
+				tc.prepare(t, server, time.Now())
+
+				status, took := askDirectly(t, server, func(req *http.Request) {
+					req.Header.Set(cookieHeader, tc.cookie)
+				})
+
+				if status != tc.status || took != 0 {
+					t.Errorf("answered %d after %s, want %d at once; output:\n%s",
+						status, took, tc.status, out.text())
+				}
+
+				wantLine(t, out.requestLine(t), tc.status, tc.action)
+
+				// GeoJS's request is abandoned before the bubble ends.
+				time.Sleep(defaultLookupWait)
+			})
+		})
+	}
+}
+
+func TestAQuestionWaitsForASlowGeoJSOnlyAsLongAsTheDefaultSays(t *testing.T) {
+	t.Parallel()
+
+	for _, tc := range []struct {
+		name   string
+		env    map[string]string
+		status int
+		action string
+	}{
+		{"nothing refuses it", nil, http.StatusOK, requestlog.ActionForward},
+		{
+			"a country list needs the answer", map[string]string{allowedCountries: "DE"},
+			http.StatusForbidden, requestlog.ActionCountryDenied,
+		},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			synctest.Test(t, func(t *testing.T) {
+				server, out := newServeWAFInBubble(t, tc.env)
+				status, took := askDirectly(t, server, func(*http.Request) {})
+
+				if status != tc.status || took != defaultLookupWait {
+					t.Errorf("answered %d after %s, want %d after %s; output:\n%s",
+						status, took, tc.status, defaultLookupWait, out.text())
+				}
+
+				wantLine(t, out.requestLine(t), tc.status, tc.action)
+			})
+		})
+	}
+}
+
+func TestServeWAFAnswersWithinTheTimeLimitWhileEveryOutsideSourceHangs(t *testing.T) {
+	t.Parallel()
+
+	held, askedGeoJS, _ := startHeldGeoJS(t)
+
+	// A resolver that takes every query and answers none.
+	resolver, err := (&net.ListenConfig{}).ListenPacket(t.Context(), "udp", localhost+":0")
+	if err != nil {
+		t.Fatalf("listen: %v", err)
+	}
+
+	t.Cleanup(func() { _ = resolver.Close() })
+
+	server, out, _ := newServerWith(t, func(params proxy.Params) *proxy.Server {
+		params.AbuseIPDBURL = silentAbuseIPDBURL
+
+		return proxy.NewServeWAF(params)
+	}, "http://app.invalid", held, time.Now, map[string]string{
+		deniedCountries: "KP", wafMode: block,
+		dnsblZones: dnsblZone, dnsblResolver: resolver.LocalAddr().String(),
+		abuseIPDBKey:  accountKey,
+		blocklistURLs: held + "/drop.txt", asnLimitPercentURL: held + "/asn.txt",
+		crowdSecURL: held, crowdSecKey: bouncerKey,
+	})
+
+	// AbuseIPDB is asked only about a client with an offence.
+	group := netip.MustParsePrefix(client + "/32")
+	server.Limiter.AddToHistory(group, time.Now(), ratelimit.Request{Attack: true})
+
+	go server.Lists.Run(t.Context())
+
+	status, took := askDirectly(t, server, func(req *http.Request) {
+		req.Header.Set("User-Agent", "test")
+		req.Header.Set("Accept", "*/*")
+		req.Host = siteA
+	})
+
+	if status != http.StatusOK || took >= answerTimeLimit {
+		t.Errorf("answered %d after %s, want %d within %s; output:\n%s",
+			status, took, http.StatusOK, answerTimeLimit, out.text())
+	}
+
+	// Each outside source was asked, and none has answered. The stand-in for
+	// GeoJS is the lists' too, which name no address.
+	if !slices.Contains(askedGeoJS(), client) {
+		t.Errorf("GeoJS was asked about %q, not %s", askedGeoJS(), client)
+	}
+
+	if queries := server.DNSBL.Queries(dnsblZone); queries != 1 {
+		t.Errorf("%d queries to the DNS blocklist, want 1", queries)
+	}
+
+	if checked := server.AbuseIPDB.Checked(); checked != 1 {
+		t.Errorf("%d checks with AbuseIPDB, want 1", checked)
+	}
+}

internal/proxy/biased_test.go

@@ -284,7 +284,7 @@ func TestClientWithoutAnAnswerInTimeHasTheUnknownLimitPercent(t *testing.T) {
 		server, out, _ := newProxy(t, "http://app.invalid", unansweredGeoJSURL,
 			time.Now, map[string]string{unknownLimitPercent: "0"})
 
-		// Once the second the request waits for its answer is up, the client
+		// Once the time the request waits for its answer is up, the client
 		// counts as without a country, and its zero allowance refuses the
 		// request before it reaches the app.
 		serveFromDE(t, server, http.MethodGet, http.NoBody)

internal/proxy/countries_test.go

@@ -284,9 +284,9 @@ func TestExclusiveListRefusesAPrivateAddressUnlessAllowed(t *testing.T) {
 // each in an AS of its own, and no other address. It returns its URL, and
 // what returns the addresses it has been asked about. A test that needs
 // the stand-in to be asked or to answer sets SWWAF_LOOKUP_TIMEOUT to an
-// hour, whether or not a request waits for the answer: on the default
-// second, a hold-up of the test process can abandon the request to the
-// stand-in, and leave the client unknown.
+// hour, whether or not a request waits for the answer: on the default, a
+// hold-up of the test process can abandon the request to the stand-in, and
+// leave the client unknown.
 func startGeoJS(t *testing.T) (string, func() []string) {
 	t.Helper()
 

internal/proxy/lookup.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"net/http"
 	"net/netip"
+	"time"
 
 	"sneak.berlin/go/smallwebwaf/internal/lookup"
 )
@@ -23,17 +24,23 @@ const (
 // SWWAF_LOOKUP_SOURCE is off or the client is on a private, loopback or
 // link-local address, which no lookup can place. The lookup database
 // answers at once. With GeoJS, while a setting needs the answer, such as a
-// country list or a biased threshold, a new client's request waits for it.
-// ctx is the request's own context.
-func (rq *request) lookUp(ctx context.Context) {
+// country list or a biased threshold, a new client's request waits for it,
+// unless servewaf will refuse the request whatever the answer says. ctx is
+// the request's own context.
+func (rq *request) lookUp(ctx context.Context, now time.Time) {
 	if rq.h.config.LookupSource == "off" || !canBePlaced(rq.client) {
 		return
 	}
 
-	if rq.h.config.LookupSource == "file" {
-		rq.lookupAnswer = rq.h.lookupFile.LookUp(rq.h.clientGroup(rq.client))
-	} else {
-		rq.lookupAnswer = rq.h.geojs.LookUp(ctx, rq.h.clientGroup(rq.client))
+	client := rq.h.clientGroup(rq.client)
+
+	switch {
+	case rq.h.config.LookupSource == "file":
+		rq.lookupAnswer = rq.h.lookupFile.LookUp(client)
+	case rq.h.answersQuestions() && rq.refusedWhateverTheAnswer(now):
+		rq.lookupAnswer = rq.h.geojs.LookUpWithoutWaiting(ctx, client)
+	default:
+		rq.lookupAnswer = rq.h.geojs.LookUp(ctx, client)
 	}
 
 	rq.lookedUp = true

internal/proxy/lookup_test.go

@@ -383,8 +383,8 @@ func TestRequestWaitsAsLongAsTheLookupTimeoutSays(t *testing.T) {
 	// network, which would keep that clock from moving on: the request is
 	// handed to the proxy's handler, and GeoJS is one that never answers.
 	synctest.Test(t, func(t *testing.T) {
-		// Not the default second. The exclusive list needs the answer, and
-		// the app is never reached.
+		// Not the default. The exclusive list needs the answer, and the app
+		// is never reached.
 		const timeout = 3 * time.Second
 
 		server, out, _ := newProxy(t, "http://app.invalid", unansweredGeoJSURL,

internal/proxy/metrics_test.go

@@ -424,8 +424,8 @@ func TestMetricsCountGeoJSRequestsAndFailures(t *testing.T) {
 	req.Header.Set(forwardedFor, fromDE)
 	wantStatus(t, do(t, req), http.StatusOK)
 
-	// The client stops waiting for GeoJS after a second, so GeoJS's
-	// failure can come after its request has ended.
+	// The client stops waiting for GeoJS once SWWAF_LOOKUP_TIMEOUT is up, so
+	// GeoJS's failure can come after its request has ended.
 	deadline := time.Now().Add(waitLimit)
 	metrics := scrape(t, addr)
 

internal/proxy/request.go

@@ -222,12 +222,12 @@ func requestHeaders(r *http.Request, names []string) map[string]string {
 // is known, before anything reaches the app, and before its body is read,
 // but for the part the Core Rule Set reads. It returns nil to let the
 // request through. The checks of checkClient come first, answered with
-// SWWAF_BAN_RESPONSE, or 403 for a block rule or the Core Rule Set, and
-// then the size limit, so that a request the rate limits count is counted
-// even when it is refused for its size. In observe mode a request
-// checkClient refuses goes on to the size limit like any other. A size or
-// time limit the Core Rule Set's reading of the body meets ends the
-// request in either mode. ctx is the request's own context.
+// SWWAF_BAN_RESPONSE, 403 for a block rule or the Core Rule Set, or 431 for
+// servewaf's header limits, and then the size limit, so that a request the
+// rate limits count is counted even when it is refused for its size. In
+// observe mode a request checkClient refuses goes on to the size limit like
+// any other. A size or time limit the Core Rule Set's reading of the body
+// meets ends the request in either mode. ctx is the request's own context.
 func (rq *request) check(ctx context.Context) *refusal {
 	action := rq.checkClient(ctx)
 
@@ -243,6 +243,8 @@ func (rq *request) check(ctx context.Context) *refusal {
 		rq.line.WouldAction = action
 	case action == requestlog.ActionRuleBlocked || action == requestlog.ActionWAFBlocked:
 		return &refusal{status: http.StatusForbidden, action: action}
+	case action == requestlog.ActionTooLarge:
+		return &refusal{status: http.StatusRequestHeaderFieldsTooLarge, action: action}
 	default:
 		return rq.banResponse(action)
 	}
@@ -271,9 +273,10 @@ func (rq *request) check(ctx context.Context) *refusal {
 // rate limits, unless the client is in SWWAF_RATE_LIMIT_EXEMPT_NETS or the
 // request's path is exempt under SWWAF_RATE_LIMIT_EXEMPT_PATHS, so that
 // every other request is counted, each of them by the client's limit
-// percentages, then SWWAF_TRAP_PATHS, then the rule files, and last the
-// Core Rule Set. A request exempt from the rate limits is exempt from the
-// byte limits too. ctx is the request's own context.
+// percentages, then SWWAF_TRAP_PATHS, then the rule files, then, in
+// servewaf, the header limits, and last the Core Rule Set. A request exempt
+// from the rate limits is exempt from the byte limits too. ctx is the
+// request's own context.
 func (rq *request) checkClient(ctx context.Context) string {
 	cfg := rq.h.config
 	if isInside(rq.client, cfg.AllowNets) {
@@ -290,7 +293,7 @@ func (rq *request) checkClient(ctx context.Context) string {
 		return requestlog.ActionBanned
 	}
 
-	rq.lookUp(ctx)
+	rq.lookUp(ctx, now)
 
 	// The lookup can wait for GeoJS, so the checks after it read the time
 	// again: a request counted at the time it arrived could be counted
@@ -314,8 +317,7 @@ func (rq *request) checkClient(ctx context.Context) string {
 		return requestlog.ActionDenied
 	}
 
-	rq.counted = !isInside(rq.client, cfg.RateLimitExemptNets) &&
-		!pathExempt(rq.in, cfg.RateLimitExemptPaths)
+	rq.counted = rq.countedByRateLimits()
 	if rq.counted {
 		rq.limitPercent, rq.bytesPercent = rq.limitPercentages()
 		rq.line.LimitPercent, rq.line.LimitPercentSetting = rq.limitPercent.logged()
@@ -335,9 +337,23 @@ func (rq *request) checkClient(ctx context.Context) string {
 		return action
 	}
 
+	if rq.overHeaderLimits() {
+		return requestlog.ActionTooLarge
+	}
+
 	return rq.checkCoreRuleSet()
 }
 
+// countedByRateLimits reports whether the rate limits count the request: its
+// client is not in SWWAF_RATE_LIMIT_EXEMPT_NETS, and its path is not exempt
+// under SWWAF_RATE_LIMIT_EXEMPT_PATHS.
+func (rq *request) countedByRateLimits() bool {
+	cfg := rq.h.config
+
+	return !isInside(rq.client, cfg.RateLimitExemptNets) &&
+		!pathExempt(rq.in, cfg.RateLimitExemptPaths)
+}
+
 // pathExempt reports whether r is exempt under prefixes,
 // SWWAF_RATE_LIMIT_EXEMPT_PATHS or SWWAF_WAF_EXEMPT_PATHS: whether its
 // path as sent, the path the app receives, not percent-decoded, starts

internal/proxy/servewaf.go

@@ -53,6 +53,13 @@ const ReportPath = "/_smallwebwaf/report"
 // QuestionBodyMaxBytes is the most of a visitor's body a question carries.
 const QuestionBodyMaxBytes = 64 << 10
 
+// The header limits: the most cookies and header fields a request may have
+// when the checks reach the Core Rule Set. Real clients never reach them.
+const (
+	maxCookies      = 50
+	maxHeaderFields = 100
+)
+
 const (
 	// maxAwaitingReport is the most answered requests held until their
 	// reports come. Past it, the request answered longest ago is dropped.
@@ -146,6 +153,56 @@ func (h *handler) serveWAF(w http.ResponseWriter, r *http.Request) {
 	h.answerOwnEndpoint(w, r)
 }
 
+// answersQuestions reports whether h is servewaf, which alone holds requests
+// for their reports.
+func (h *handler) answersQuestions() bool {
+	return h.awaitingReport != nil
+}
+
+// refusedWhateverTheAnswer reports whether the checks will refuse the request
+// at now whatever GeoJS answers about its client: it is over the header limits,
+// a blocklist lists its client while SWWAF_BLOCKLIST_ACTION is deny, the
+// CrowdSec decision list lists it, or it is over a whole rate limit. rproxy
+// lets a request through when servewaf answers too late, so such a request
+// does not wait for GeoJS.
+func (rq *request) refusedWhateverTheAnswer(now time.Time) bool {
+	lists := rq.h.lists
+	_, decided := lists.CrowdSecDecision(rq.client, now)
+
+	return decided || rq.overHeaderLimits() ||
+		(rq.h.config.BlocklistAction == deny && len(lists.ListedBy(rq.client)) > 0) ||
+		(rq.countedByRateLimits() &&
+			rq.h.limiter.OverARateLimit(rq.h.clientGroup(rq.client), now))
+}
+
+// overHeaderLimits reports whether servewaf refuses the request for its
+// headers: it has more than maxHeaderFields fields, each value of a header sent
+// twice being one, or more than maxCookies cookies, each part of a Cookie header
+// between semicolons that is not empty being one. The Core Rule Set runs each of
+// its rules on each cookie, so that thousands of small cookies keep it busy for
+// over a second.
+func (rq *request) overHeaderLimits() bool {
+	if !rq.h.answersQuestions() {
+		return false
+	}
+
+	fields, cookies := 0, 0
+
+	for _, values := range rq.in.Header {
+		fields += len(values)
+	}
+
+	for _, value := range rq.in.Header.Values("Cookie") {
+		for part := range strings.SplitSeq(value, ";") {
+			if strings.TrimSpace(part) != "" {
+				cookies++
+			}
+		}
+	}
+
+	return fields > maxHeaderFields || cookies > maxCookies
+}
+
 // answerQuestion runs the checks on the visitor's request a question
 // carries, and answers rproxy what to do with it. A denied request's counting
 // is finished here, and an allowed one's by its report.

internal/ratelimit/ratelimit.go

@@ -228,6 +228,30 @@ func (l *Limiter) Count(
 	return l.count(client, now, 1, 0, percent)
 }
 
+// OverARateLimit reports whether one more request from client at now would
+// take it over a whole rate limit. A client's percentage can only lower a
+// limit, so that request is over the limit at any percentage. It counts
+// nothing, and does not make the client the most recently seen.
+func (l *Limiter) OverARateLimit(client netip.Prefix, now time.Time) bool {
+	l.mu.Lock()
+	defer l.mu.Unlock()
+
+	c, seen := l.clients.Peek(client)
+	if !seen {
+		return false
+	}
+
+	buckets := c.buckets()
+
+	for i, w := range l.windows {
+		if w.limit > 0 && buckets[i].Add(now, w.length, 0)+1 > float64(w.limit) {
+			return true
+		}
+	}
+
+	return false
+}
+
 // CountBytes counts bytes, those of a request from client that has ended,
 // at now, in every window, and returns the client's counts in each window.
 // It reports whether the bytes take the client over a byte limit, of which

internal/ratelimit/ratelimit_test.go

@@ -26,6 +26,50 @@ const (
 	hour   = "hour"
 )
 
+func TestOverARateLimitSaysWhatTheNextRequestWouldDoAndCountsNothing(t *testing.T) {
+	t.Parallel()
+
+	client := netip.MustParsePrefix("203.0.113.9/32")
+	now := midnight()
+	limiter := ratelimit.New(ratelimit.Limits{PerMinute: limit, PerDay: limit + 1},
+		tableSize)
+
+	if limiter.OverARateLimit(client, now) {
+		t.Error("a client never seen is over a rate limit")
+	}
+
+	if limiter.Len() != 0 {
+		t.Error("asking about a client never seen added it to the table")
+	}
+
+	for range limit {
+		if limiter.OverARateLimit(client, now) {
+			t.Error("a client with room for another request is over a rate limit")
+		}
+
+		wantCount(t, limiter, client, now, "")
+	}
+
+	if !limiter.OverARateLimit(client, now) {
+		t.Error("a client at its limit is not over a rate limit")
+	}
+
+	held, _ := limiter.Client(client)
+	if held.Minute.Current != limit {
+		t.Errorf("the client's minute holds %d requests, want %d", held.Minute.Current, limit)
+	}
+
+	wantCount(t, limiter, client, now, minute)
+
+	// A limit that is off is never gone over.
+	off := ratelimit.New(ratelimit.Limits{}, tableSize)
+	off.Count(client, now, whole)
+
+	if off.OverARateLimit(client, now) {
+		t.Error("a client is over a rate limit that is off")
+	}
+}
+
 func TestEachWindowRefusesAtItsLimitAndLetsTheClientBack(t *testing.T) {
 	t.Parallel()
 

internal/requestlog/requestlog.go

@@ -18,7 +18,8 @@ import (
 const (
 	// ActionForward is a request passed to the app.
 	ActionForward = "forward"
-	// ActionTooLarge is a request or response over its size limit.
+	// ActionTooLarge is a request or response over its size limit, or, in
+	// servewaf, a request over the header limits.
 	ActionTooLarge = "too_large"
 	// ActionTimedOut is a request or response that ran out of time.
 	ActionTimedOut = "timed_out"
@@ -139,7 +140,8 @@ type Line struct {
 	// WouldAction is, in observe mode, the action enforce mode would have
 	// taken with a request it would have refused: ActionDenied,
 	// ActionBanned, ActionCountryDenied, ActionRateLimited,
-	// ActionRuleBlocked or ActionWAFBlocked.
+	// ActionRuleBlocked, ActionWAFBlocked, or, in servewaf, ActionTooLarge
+	// for the header limits.
 	WouldAction string `json:"would_action,omitempty"`
 	// LimitPercent and LimitPercentSetting are, for a request the rate
 	// limits counted whose client a biased threshold gives a percentage of