Bound the receiver rate limit per client IP across /webhook/* (closes #139)

The receiver limiter keyed buckets on (client IP, request path). The
route pattern /webhook/{uuid} matches any single segment, so a client
that invented a fresh path per request minted a fresh bucket per
request and never refilled one: its aggregate rate against the only
unauthenticated, internet-exposed endpoint was unbounded, and every
one of those requests reached an entrypoint lookup before it 404ed.

Put a second limiter in front of it, keyed on the client IP alone and
covering the whole route at ten times the configured per-entrypoint
limit (1200/min by default). The per-entrypoint limit is unchanged and
still wanted; it just bounds nothing in aggregate on its own. Ten
entrypoints' worth of headroom lets one sender address drive several
entrypoints at full rate while still capping what one address costs
the receiver. The multiplication saturates rather than wrapping, since
nothing bounds RECEIVER_RATE_LIMIT from above and a negative limit
would reject every request.

The aggregate limiter is the outer one, so it counts the requests the
per-entrypoint limiter rejects; a test pins that order by exhausting
one path against the inner limit and then requiring a request to an
unused path to be rejected.

Move the handler's INFO line for an incoming webhook below the
entrypoint lookup. The UUID is attacker-controlled path text, so
logging it first let a client write an INFO line per invented path; a
miss is already logged at DEBUG and the request is already in the
access log.

Give the aggregate limiter its own 429 handler that logs at DEBUG and
without the path, rather than the shared one that logs at WARN with
it. Its rejections are one line per request of the very flood it
exists to bound, so the shared handler would have let a client write
its own text into the operator's log at an alerting level once per
request. What this limiter bounds is the database work an invented
path costs; the access log still records every request once at INFO,
and the README now says so instead of claiming DEBUG-only logging.

.dockerignore

@@ -1,3 +1,6 @@
+# .ci-fingerprint is deliberately NOT excluded: it is the CI cache barrier
+# that keeps the check stages from replaying a cached pass. See the lint
+# stage of the Dockerfile.
 .git/
 bin/
 *.md

.gitea/workflows/check.yml

@@ -11,5 +11,53 @@ jobs:
     steps:
       - name: Checkout
         uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2 2024-10-23
+        with:
+          # The fingerprint step below needs history to find the last commit
+          # that touched the Docker build context.
+          fetch-depth: 0
+
+      - name: Neutralize superseded run statuses
+        # Gitea cancels the in-flight run when another commit is pushed to the
+        # same branch and records the cancellation as `failure`, so a commit
+        # that was never tested reads red. The cancellation is unconditional
+        # server-side for push events and cannot be disabled from a workflow
+        # file, so the superseding run rewrites those statuses to `skipped`.
+        # Only the exact cancellation status is touched; a real failure is
+        # left alone.
+        env:
+          GITEA_TOKEN: ${{ secrets.GITEA_TOKEN }}
+        run: |
+          set -eu
+          api="${GITHUB_API_URL}/repos/${GITHUB_REPOSITORY}"
+          ctx='check / check (push)'
+          for sha in $(git rev-list --max-count=20 "${GITHUB_SHA}^" || true); do
+            latest="$(curl -sf "${api}/commits/${sha}/status" | jq -r \
+              --arg c "$ctx" \
+              '[.statuses[] | select(.context == $c)][0] // empty
+               | "\(.status)|\(.description)"')" || continue
+            [ "$latest" = 'failure|Has been cancelled' ] || continue
+            curl -sf -X POST "${api}/statuses/${sha}" \
+              -H "Authorization: token ${GITEA_TOKEN}" \
+              -H 'Content-Type: application/json' \
+              -d "$(jq -nc --arg c "$ctx" '{
+                context: $c,
+                state: "skipped",
+                description: "Superseded by a newer commit; never tested"
+              }')" >/dev/null
+            echo "neutralized superseded status on ${sha}"
+          done
+
+      - name: Fingerprint the build context
+        # `.dockerignore` keeps docs out of the build context, so a docs-only
+        # commit legitimately replays the whole image from cache and stays
+        # cheap. Every other commit writes a new fingerprint into the context,
+        # which invalidates the `COPY . .` layer of both check stages: a
+        # commit that was never linted, formatted-checked, tested and built
+        # cannot report success from cache.
+        run: |
+          set -eu
+          fp="$(git log -1 --format=%H -- . ':!*.md' ':!LICENSE' ':!.editorconfig')"
+          printf '%s\n' "${fp:-$GITHUB_SHA}" > .ci-fingerprint
+
       - name: Build Docker image (runs make check)
         run: script/cibuild

.gitignore

@@ -41,4 +41,7 @@ data/
 
 # Temporary files
 tmp/
-temp/ 
+temp/
+
+# CI cache barrier, written into the build context by the check workflow
+.ci-fingerprint 

Dockerfile

@@ -12,7 +12,11 @@ WORKDIR /src
 COPY go.mod go.sum ./
 RUN go mod download
 
-# Copy source code
+# Copy source code. In CI the context also carries .ci-fingerprint, whose
+# value changes with every commit that touches the build context (see
+# .gitea/workflows/check.yml). That invalidates this layer, so the checks
+# below cannot report success by replaying a cached pass. Do not add it to
+# .dockerignore.
 COPY . .
 
 # Run formatting check and linter
@@ -36,7 +40,8 @@ WORKDIR /build
 COPY go.mod go.sum ./
 RUN go mod download
 
-# Copy source code
+# Copy source code, including the .ci-fingerprint cache barrier described in
+# the lint stage above.
 COPY . .
 
 # Run tests and build

README.md

@@ -95,7 +95,7 @@ TTY detection, and security headers are always applied.
 | `SENTRY_DSN`            | Sentry error reporting DSN          | `""`     |
 | `RETENTION_SWEEP_INTERVAL` | Retention reaper period (Go duration, must be positive) | `1h` |
 | `SESSION_IDLE_TIMEOUT`  | Idle session timeout (Go duration)  | `24h`    |
-| `RECEIVER_RATE_LIMIT`   | Receiver requests/minute per IP per entrypoint | `120` |
+| `RECEIVER_RATE_LIMIT`   | Receiver requests/minute per IP per entrypoint (10x that per IP across the route) | `120` |
 | `TRUSTED_PROXIES`       | CIDRs whose forwarded headers are trusted | `""` (none) |
 
 #### Trusted proxies
@@ -856,6 +856,31 @@ legitimate webhook senders). Requests over the limit receive HTTP 429
 with a `Retry-After` header. A set-but-invalid `RECEIVER_RATE_LIMIT`
 value aborts startup rather than silently falling back to the default.
 
+A second limit sits in front of that one, keyed on the client IP alone
+and covering the whole route at ten times `RECEIVER_RATE_LIMIT` requests
+per minute (default 1200). The per-entrypoint limit needs it: the route
+pattern matches any single path segment, so a client that invents a
+fresh path per request gets a fresh per-entrypoint bucket every time and
+would otherwise have no aggregate limit at all — while each of those
+requests still costs an entrypoint lookup before it 404s. The aggregate
+limit leaves room for one address to drive several entrypoints at their
+full rate, and it is not configurable separately.
+
+What that aggregate limit bounds is the database work an invented path
+costs; log volume it caps rather than eliminates. A path that names no
+entrypoint is recorded by the handler at `DEBUG`, and the aggregate
+limiter logs its own rejections at `DEBUG` and without the path, so
+neither appears at all under the default level. The per-entrypoint
+limiter is the loud one: it still logs every rejection at `WARN` with
+the request path, which on this route is attacker-controlled text. A
+client hammering a single invented path is served `RECEIVER_RATE_LIMIT`
+requests and has the rest of its aggregate budget rejected there, so
+the aggregate limit is what bounds those `WARN` lines — to under ten
+times `RECEIVER_RATE_LIMIT` per minute per client IP, 1080 at the
+defaults, where before it there was no bound at all. The access log is
+bounded by neither limit: every request is recorded once at `INFO` with
+its full URL, served or rejected alike.
+
 Every limiter here — receiver, login, and password change — identifies
 the client the same way, through one shared key function: the
 connection's own address, unless the peer is listed in
@@ -864,7 +889,14 @@ instead. See [Trusted proxies](#trusted-proxies). Deployed without that
 variable set, a client behind a reverse proxy shares one bucket with
 every other client behind the same proxy, which is the safe direction
 to be wrong in: set `TRUSTED_PROXIES` to the proxy's address to get
-per-client limits back.
+per-client limits back. That shared bucket matters more for the
+aggregate limit than for the per-entrypoint one: with `TRUSTED_PROXIES`
+unset behind the reverse proxy a production deployment is required to
+run behind, every request keys on the proxy, so the aggregate limit
+becomes a service-wide ceiling of 1200 requests per minute across all
+senders and all entrypoints, where the per-entrypoint limit's capacity
+still grows with the number of entrypoints. Any deployment with more
+than a handful of busy entrypoints must set `TRUSTED_PROXIES`.
 
 Finer-grained per-webhook rate limits (configured in the web UI and
 enforced in the webhook handler) can layer on top of this env-level
@@ -1109,6 +1141,34 @@ binary is statically linked and runs on Alpine.
 `docker build .` is the CI gate — if it passes, the code is formatted,
 linted, tested, and compiled.
 
+#### CI gate honesty
+
+A layer cache lets `docker build .` exit 0 in seconds with the lint and
+test stages replayed rather than executed, which would make a green
+check meaningless. The `check` workflow therefore writes
+`.ci-fingerprint` into the build context before building. Its value is
+the hash of the last commit that touched the build context, so:
+
+- Any commit that changes code (including a squash merge whose tree
+  matches an already-built branch) gets a new fingerprint, invalidates
+  the `COPY . .` layer of both check stages, and really runs
+  `make fmt-check`, `make lint`, `make test`, and `make build`. A run
+  that reports success ran them.
+- A docs-only commit leaves the fingerprint unchanged — `.dockerignore`
+  excludes `*.md` and `LICENSE` from the context anyway — so the image
+  replays from cache and costs seconds.
+
+The module download layer sits above `COPY . .` and stays cached either
+way.
+
+The workflow's first step covers a second way the gate lied: Gitea
+cancels an in-flight run when a newer commit lands on the same branch
+and records that cancellation as a `failure` status, marking a commit
+red that was never tested. Cancellation is unconditional server-side for
+push events, so the superseding run rewrites the exact
+`Has been cancelled` status to `skipped`. Genuine failures are never
+touched.
+
 ## TODO
 
 See [TODO.md](TODO.md).

internal/handlers/webhook.go

@@ -39,12 +39,6 @@ func (h *Handlers) HandleWebhook() http.HandlerFunc {
 			return
 		}
 
-		h.log.Info("webhook request received",
-			"entrypoint_uuid", entrypointUUID,
-			"method", r.Method,
-			"remote_addr", r.RemoteAddr,
-		)
-
 		entrypoint, ok := h.lookupEntrypoint(
 			w, r, entrypointUUID,
 		)
@@ -52,6 +46,18 @@ func (h *Handlers) HandleWebhook() http.HandlerFunc {
 			return
 		}
 
+		// Logged only once the UUID is known to name a real
+		// entrypoint. The UUID comes straight out of the path on
+		// the one unauthenticated endpoint, so logging it before
+		// the lookup let a client write an INFO line per invented
+		// path; the request itself is already in the access log
+		// and a miss is already logged at DEBUG.
+		h.log.Info("webhook request received",
+			"entrypoint_uuid", entrypointUUID,
+			"method", r.Method,
+			"remote_addr", r.RemoteAddr,
+		)
+
 		if !entrypoint.Active {
 			http.Error(w, "Gone", http.StatusGone)
 

internal/middleware/export_test.go

@@ -41,3 +41,13 @@ const LoginRateLimitConst = loginRateLimit
 // PasswordChangeRateLimitConst exposes the
 // passwordChangeRateLimit constant.
 const PasswordChangeRateLimitConst = passwordChangeRateLimit
+
+// ReceiverAggregateMultiplierConst exposes the
+// receiverAggregateMultiplier constant.
+const ReceiverAggregateMultiplierConst = receiverAggregateMultiplier
+
+// ReceiverAggregateLimitForTest exposes receiverAggregateLimit for
+// testing.
+func ReceiverAggregateLimitForTest(perEntrypoint int) int {
+	return receiverAggregateLimit(perEntrypoint)
+}

internal/middleware/ratelimit.go

@@ -1,6 +1,7 @@
 package middleware
 
 import (
+	"math"
 	"net/http"
 	"net/netip"
 	"slices"
@@ -33,6 +34,14 @@ const (
 	// requests per minute.
 	receiverRateInterval = 1 * time.Minute
 
+	// receiverAggregateMultiplier scales the configured
+	// per-entrypoint receiver limit into the aggregate limit one
+	// client IP may spend across the whole /webhook/* route. Ten
+	// entrypoints' worth lets a single sender address drive several
+	// entrypoints at their full rate, while still capping what one
+	// address costs the unauthenticated receiver.
+	receiverAggregateMultiplier = 10
+
 	// maxForwardedHops bounds how many X-Forwarded-For entries the
 	// chain walk examines. Real chains are one to three hops, but a
 	// client can pad the header up to MaxHeaderBytes, so without a
@@ -162,9 +171,11 @@ func (m *Middleware) clientKey(r *http.Request) string {
 	return peer.String()
 }
 
-// tooManyRequests returns the 429 handler shared by every limiter:
+// tooManyRequests returns the 429 handler used by the login,
-// it logs the rejection with logMessage and answers with
+// password-change and per-entrypoint receiver limiters: it logs the
-// responseMessage. httprate adds the Retry-After header (RFC 6585).
+// rejection with logMessage and answers with responseMessage.
+// httprate adds the Retry-After header (RFC 6585). The aggregate
+// receiver limiter uses floodTooManyRequests instead.
 func (m *Middleware) tooManyRequests(
 	logMessage, responseMessage string,
 ) http.HandlerFunc {
@@ -174,6 +185,31 @@ func (m *Middleware) tooManyRequests(
 	}
 }
 
+// floodTooManyRequests returns the 429 handler for a limiter whose
+// rejections are themselves the flood: it logs at DEBUG and without
+// the path, then answers with responseMessage.
+//
+// The aggregate receiver limiter trips exactly when one address is
+// sending faster than the receiver wants to serve, so its rejection
+// log is one line per request of that flood. At WARN with "path" that
+// hands a client a way to write its own text into the operator's log,
+// at a level that trips alerting, once per request — the log-volume
+// problem this limiter exists to bound. DEBUG is off in production by
+// default, so a flood costs nothing here; the path is dropped so that
+// turning DEBUG on to diagnose one does not restore the problem.
+//
+// This limiter bounds the database work an invented path costs, not
+// the number of log lines it produces: the access log in
+// middleware.go still records every request, served or rejected.
+func (m *Middleware) floodTooManyRequests(
+	logMessage, responseMessage string,
+) http.HandlerFunc {
+	return func(w http.ResponseWriter, _ *http.Request) {
+		m.log.Debug(logMessage)
+		http.Error(w, responseMessage, http.StatusTooManyRequests)
+	}
+}
+
 // LoginRateLimit returns middleware that enforces per-IP rate
 // limiting on login attempts using go-chi/httprate. Only POST
 // requests are rate-limited; GET requests (rendering the login
@@ -242,15 +278,26 @@ func (m *Middleware) postRateLimit(
 	}
 }
 
-// ReceiverRateLimit returns middleware that rate-limits the
+// ReceiverRateLimit returns middleware that rate-limits the public
-// public webhook receiver endpoint per client IP per request
+// webhook receiver endpoint with two limits in series.
-// path (the path contains the entrypoint UUID, so each sender
+//
-// is limited per entrypoint without affecting other senders or
+// The inner limit is per client IP per request path: the path
-// other entrypoints). The limit is Config.ReceiverRateLimit
+// contains the entrypoint UUID, so each sender is limited per
-// requests per minute. Requests over the limit receive a 429.
+// entrypoint without affecting other senders or other entrypoints.
-// Clients are identified by rateLimitKey.
+// It is Config.ReceiverRateLimit requests per minute.
+//
+// That limit alone bounds nothing in aggregate. The route pattern
+// /webhook/{uuid} matches any single segment, so a client that
+// invents a fresh path per request mints a fresh bucket per request
+// and never refills one — and every such request still reaches the
+// handler's entrypoint lookup before it 404s. The outer limit is
+// therefore keyed on the client IP alone, capping what one address
+// can spend across the whole route however it varies the path.
+//
+// Requests over either limit receive a 429. Clients are identified
+// by rateLimitKey.
 func (m *Middleware) ReceiverRateLimit() func(http.Handler) http.Handler {
-	return httprate.Limit(
+	perEntrypoint := httprate.Limit(
 		m.params.Config.ReceiverRateLimit,
 		receiverRateInterval,
 		httprate.WithKeyFuncs(
@@ -262,4 +309,31 @@ func (m *Middleware) ReceiverRateLimit() func(http.Handler) http.Handler {
 			"Too many requests. Please slow down.",
 		)),
 	)
+
+	aggregate := httprate.Limit(
+		receiverAggregateLimit(m.params.Config.ReceiverRateLimit),
+		receiverRateInterval,
+		httprate.WithKeyFuncs(m.rateLimitKey),
+		httprate.WithLimitHandler(m.floodTooManyRequests(
+			"webhook receiver aggregate rate limit exceeded",
+			"Too many requests. Please slow down.",
+		)),
+	)
+
+	return func(next http.Handler) http.Handler {
+		return aggregate(perEntrypoint(next))
+	}
+}
+
+// receiverAggregateLimit is the per-IP aggregate limit derived from
+// the configured per-entrypoint limit. The operator sets the latter
+// and nothing bounds it from above, so the multiplication is
+// saturated rather than allowed to wrap into a negative limit that
+// would reject every request.
+func receiverAggregateLimit(perEntrypoint int) int {
+	if perEntrypoint > math.MaxInt/receiverAggregateMultiplier {
+		return math.MaxInt
+	}
+
+	return perEntrypoint * receiverAggregateMultiplier
 }

internal/middleware/ratelimit_test.go

@@ -4,6 +4,7 @@ import (
 	"context"
 	"fmt"
 	"log/slog"
+	"math"
 	"net/http"
 	"net/http/httptest"
 	"net/netip"
@@ -670,6 +671,129 @@ func TestRateLimitKey_LongChainAllocationIsBounded(t *testing.T) {
 	)
 }
 
+// TestReceiverRateLimit_LimitsAggregateAcrossInventedPaths is the
+// regression test for the per-path bucket key. The route pattern
+// matches any single segment, so a client that never reuses a path
+// never reuses a per-entrypoint bucket either, and its aggregate
+// rate against the receiver is whatever it likes — with every
+// request reaching an entrypoint lookup before it 404s. The IP-only
+// aggregate limiter is what bounds that, so this must fail if the
+// aggregate limiter is removed.
+func TestReceiverRateLimit_LimitsAggregateAcrossInventedPaths(
+	t *testing.T,
+) {
+	t.Parallel()
+
+	const (
+		limit = 3
+		ip    = "6.6.6.6:1234"
+	)
+
+	aggregate := limit * middleware.ReceiverAggregateMultiplierConst
+
+	handler := receiverLimitedHandler(t, limit)
+
+	// Every request goes to a path this client has never used, so
+	// none of them shares a per-entrypoint bucket with another.
+	for i := range aggregate {
+		w := receiverPost(
+			handler, ip, fmt.Sprintf("/webhook/invented-%d", i),
+		)
+		assert.Equal(
+			t, http.StatusOK, w.Code,
+			"request %d to a distinct path should pass", i,
+		)
+	}
+
+	w := receiverPost(
+		handler, ip, fmt.Sprintf("/webhook/invented-%d", aggregate),
+	)
+	assert.Equal(
+		t, http.StatusTooManyRequests, w.Code,
+		"a client must not be able to raise its aggregate rate "+
+			"against /webhook/* by varying the path",
+	)
+
+	// The aggregate limit is still per client IP: exhausting one
+	// address must not throttle another.
+	w = receiverPost(handler, "6.6.6.7:1234", "/webhook/invented-0")
+	assert.Equal(
+		t, http.StatusOK, w.Code,
+		"a different client IP must not be affected",
+	)
+}
+
+// TestReceiverRateLimit_RejectedRequestsCountTowardAggregate pins the
+// order the two limiters are chained in. The aggregate limiter has to
+// be the outer one, so that it counts requests the per-entrypoint
+// limiter rejects: those requests still arrive, and the aggregate
+// limit exists to bound what one address can make the receiver do.
+//
+// One path is hammered past the per-entrypoint limit, which alone
+// would leave the aggregate budget almost untouched; then a path the
+// client has never used must be rejected, which only the aggregate
+// limiter can do. Swap the two limiters and that last request is
+// served, because the rejected ones never reached the aggregate
+// limiter to be counted.
+func TestReceiverRateLimit_RejectedRequestsCountTowardAggregate(
+	t *testing.T,
+) {
+	t.Parallel()
+
+	const (
+		limit = 3
+		ip    = "6.6.6.8:1234"
+	)
+
+	aggregate := limit * middleware.ReceiverAggregateMultiplierConst
+
+	handler := receiverLimitedHandler(t, limit)
+
+	// Spend the whole aggregate budget on one path. Only the first
+	// limit requests are served; the rest are rejected by the
+	// per-entrypoint limiter but still count against the aggregate.
+	for i := range aggregate {
+		w := receiverPost(handler, ip, "/webhook/exhausted")
+
+		want := http.StatusTooManyRequests
+		if i < limit {
+			want = http.StatusOK
+		}
+
+		assert.Equal(
+			t, want, w.Code,
+			"request %d to the exhausted path", i,
+		)
+	}
+
+	w := receiverPost(handler, ip, "/webhook/never-used")
+	assert.Equal(
+		t, http.StatusTooManyRequests, w.Code,
+		"requests rejected per entrypoint must still count "+
+			"toward the aggregate limit, so the aggregate "+
+			"limiter has to run first",
+	)
+}
+
+// TestReceiverAggregateLimit_SaturatesOnOverflow covers the derived
+// aggregate limit for a configured per-entrypoint limit large enough
+// that multiplying it would wrap negative, which httprate would read
+// as a limit that rejects every request.
+func TestReceiverAggregateLimit_SaturatesOnOverflow(t *testing.T) {
+	t.Parallel()
+
+	assert.Equal(
+		t, 1200,
+		middleware.ReceiverAggregateLimitForTest(120),
+		"the default limit scales by the multiplier",
+	)
+	assert.Equal(
+		t, math.MaxInt,
+		middleware.ReceiverAggregateLimitForTest(math.MaxInt),
+		"an overflowing limit saturates instead of wrapping",
+	)
+}
+
 // TestReceiverRateLimit_IgnoresForwardedFromUntrustedPeer proves
 // the receiver limiter uses the same gated key function as the
 // POST limiters.