Record the 1.0.0 milestone as complete in TODO.md

The Status section claimed next held the completed 1.0.0 milestone with
every issue closed. Four are open (#176, #178, #186, #187), so a merge
of next to main would have shipped that claim to main.

Next Step still named #115 and #125 as open owner decisions; both
landed. It now names the real open items, #150 and #112, and the forced
merge order for the remaining four.

Completed Steps was seventeen units behind, back to 2026-08-12.

README.md

@@ -113,7 +113,7 @@ TTY detection, and security headers are always applied.
 | `RETENTION_SWEEP_INTERVAL` | How often the retention reaper and archive sweeper run (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 (10x that per IP across the route) | `120` |
-| `TRUSTED_PROXIES`       | CIDRs whose forwarded headers are trusted (unset: all clients behind a proxy share one rate-limit bucket) | `""` (none) |
+| `TRUSTED_PROXIES`       | CIDRs whose forwarded headers are trusted (unset: all clients behind a proxy share one rate-limit bucket; a correct login password is never throttled either way) | `""` (none) |
 
 #### Trusted proxies
 
@@ -136,13 +136,15 @@ That default is safe against forged headers, but leaving it unset in
 production has a cost you must know about. Production runs behind a
 TLS-terminating reverse proxy, so with `TRUSTED_PROXIES` unset every
 request keys on the proxy's own address and all clients share a single
-bucket per limit. For the login and password-change limits that is a
-denial of service anyone can perform: a steady five POSTs per minute
-from any address on the internet keeps the shared login bucket full,
-and the operator's own login then returns HTTP 429 for as long as the
-trickle continues. There is no second administrative path and no
-bypass. Restarting the service clears the in-memory buckets, but a
-sustained trickle re-locks them immediately.
+bucket per limit. The receiver limits become service-wide ceilings,
+and the login endpoint's failure counting collapses onto one key, so a
+stranger's wrong passwords throttle every other client's wrong
+passwords.
+
+What it cannot do is lock the operator out. The login endpoint
+verifies credentials **before** it consults any limit and charges only
+failures, so a correct password is never throttled no matter how full
+the bucket is. See [Rate Limiting](#rate-limiting).
 
 The remedy is to set `TRUSTED_PROXIES` to your reverse proxy's
 address, which restores per-client buckets. webhooker logs a warning
@@ -382,11 +384,12 @@ It uses:
 - **[gorilla/csrf](https://github.com/gorilla/csrf)** for CSRF
   protection (cookie-based double-submit tokens)
 - **[go-chi/httprate](https://github.com/go-chi/httprate)** for
-  sliding-window rate limiting of the login, password-change and
-  webhook receiver endpoints. The bucket is per client IP only when
+  sliding-window rate limiting of the password-change and webhook
+  receiver endpoints. The bucket is per client IP only when
   `TRUSTED_PROXIES` names the reverse proxy; unset, every client
-  behind that proxy shares one bucket per limit (see
-  [Rate Limiting](#rate-limiting))
+  behind that proxy shares one bucket per limit. The login endpoint
+  counts failed attempts itself instead, so that a correct password is
+  never throttled (see [Rate Limiting](#rate-limiting))
 - **[Prometheus](https://prometheus.io)** for metrics, served at
   `/metrics` behind basic auth
 - **[Sentry](https://sentry.io)** for optional error reporting
@@ -991,15 +994,16 @@ 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`
+limiter is the loud one: it 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`,
-served or rejected alike.
+the aggregate limit is what bounds the _number_ of 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. Their
+_width_ is bounded by the field budgets below, the same ones the access
+log spends. The access log is bounded by neither limit: every request
+is recorded once at `INFO`, served or rejected alike.
 
 What the access log does bound is the _content_ of those lines. A 3xx
 or 4xx response logs the chi route pattern — `/webhook/{uuid}`,
@@ -1022,6 +1026,92 @@ buy the same amplification as an invented path. Nothing debuggable is
 lost: `page`, on the authenticated pagination links, is the only query
 parameter this service reads.
 
+Client-supplied request content does not leave the host by the other
+route either. The Sentry SDK attaches the request to every event it
+captures, independently of the access log, and `SendDefaultPII=false`
+does not cover all of what it copies: the raw query string and the
+first 10 KiB of the request body are both taken unconditionally, the
+body precisely because these handlers call `ParseForm`. A `BeforeSend`
+hook therefore replaces the query string and the body with
+`(redacted)`, drops cookies and the remote-address environment, and
+reduces the headers to a fixed allowlist — `Accept`, `Content-Length`,
+`Content-Type`, `Host`, `Origin`, `Referer`, `User-Agent` and
+`X-Request-Id`.
+
+The same hook rewrites the request URL. The SDK builds it as
+`scheme://host/path` from the concrete path, which on the receiver
+route is `/webhook/<uuid>` in full — and that UUID is a write
+capability, not an identifier: anyone holding it can post events this
+service accepts and its targets then deliver. A tracker has its own
+retention, access control and deletion policy, so the rule the access
+log follows above does not carry across that boundary. What is sent is
+the chi route pattern instead: `http://host/webhook/{uuid}`.
+
+The scheme and the host are kept, and everything else in the URL is
+discarded rather than edited, so a future SDK version that starts
+appending a query string cannot widen this. The scheme has to survive
+for the reason given below. The host is whatever the request's `Host`
+header carried — this service validates no hostname, so on a directly
+exposed deployment a client sets it — and that same header is on the
+allowlist above, so scrubbing the host out of the URL would withhold
+nothing that is not sent anyway.
+
+The body, the query string and the URL are all handled on every route
+rather than filtered by route. For the URL that is also what keeps the
+event locatable: an error event is grouped by its exception and stack
+trace, not by its URL, so replacing the path with the pattern costs no
+grouping and the pattern still names the route in the UI. And an
+unconditional rule cannot leak on a route somebody forgets to add to
+it, which a route-conditional one can. For the body there is a second
+reason: nothing debuggable is lost, because every handler reads its
+fields with `PostFormValue`, so the body is exactly where the
+credentials are — the target destination URL, the login password, both
+password-change fields — and the one route whose body is genuine
+signal is the receiver, whose body is already stored on the event and
+served from the UI, so a tracker is not where anyone reads it.
+
+The route is reachable from the hook only on the error dispatch.
+`sentryhttp`'s recover path puts the request on the context it hands
+to `RecoverWithContext`, and the SDK carries that context through to
+`BeforeSend` as `hint.Context`, so
+`hint.Context.Value(sentry.RequestContextKey)` yields the live request
+and chi's `RoutePattern()` yields the matched pattern off it. The
+transaction dispatch has no such request: a finished span captures
+with a nil hint, which the client replaces with an empty one, so
+`BeforeSendTransaction` sees no context at all. Tracing is off in this
+service, so no transaction event is produced today, but the hook is
+installed on both dispatches as a floor.
+
+Where the pattern is out of reach — the transaction dispatch, an event
+captured outside the router, or a request that matched no route — the
+fallback is never the concrete path. The path becomes the literal
+`/(redacted)`, so the URL reads `http://host/(redacted)`; a URL the
+rewrite cannot parse into a scheme is withheld whole. A transaction
+event additionally carries the SDK's own `METHOD /path` name, built
+from the concrete path as well; it is rewritten on the same terms, to
+`POST /webhook/{uuid}` where the pattern is known and `POST
+/(redacted)` where it is not.
+
+The headers are an allowlist for the same reason the rules above are
+unconditional: the SDK's own filter removes four names and passes
+everything else, which would ship `X-CSRF-Token` and the shared
+secrets senders put on the receiver route. What survives still names
+the failing route — scheme, host, route pattern, method — and
+`X-Request-Id` ties the event to the local access log line that holds
+the rest. Nothing dropped is needed for the likeliest use, debugging a
+CSRF rejection. Its three inputs are the TLS decision, `Origin` and
+`Referer`; the latter two are kept, and the first is the scheme of the
+retained URL, because the SDK derives that scheme from
+`r.TLS != nil || r.Header.Get("X-Forwarded-Proto") == "https"` — byte
+for byte the predicate `internal/middleware/csrf.go` uses to choose
+between the `csrf.Secure(true)` and `csrf.Secure(false)` handlers.
+That is what the rewrite above preserves it for, and it is why
+dropping `X-Forwarded-Proto` costs nothing. The dropped provider
+headers (`X-GitHub-Event`, `X-Gitlab-Event` and the like) are real
+signal but are recorded locally on the event, and
+`Sentry-Trace`/`Baggage` are already reflected in the event's trace
+context.
+
 The remaining client-supplied fields are truncated rather than dropped,
 each to a fixed budget: 512 bytes for `url`, `useragent` and `referer`,
 128 for `request_id` (chi passes an inbound `X-Request-Id` header
@@ -1051,18 +1141,197 @@ the figure has headroom. `internal/middleware/accesslog_test.go`
 asserts it against 8 KB of client-chosen text in the path, in the
 query, and in each of `User-Agent`, `Referer` and `X-Request-Id`,
 including cases built from the characters the handlers escape, and
-against the widest line the service can be made to write: a 5xx that
-keeps its concrete path while all three header fields are also at their
-budget. Every case runs through both handlers `internal/logger` can
-select — the JSON one and the text one it installs on a tty — since the
-two do not escape alike and the ceiling is quoted unqualified. Measured
-over a real connection, the widest line is 1,972 bytes.
+against the widest access log line the service can be made to write: a
+5xx that keeps its concrete path while all three header fields are also
+at their budget. Every case runs through both handlers
+`internal/logger` can select — the JSON one and the text one it installs
+on a tty — since the two do not escape alike and the ceiling is quoted
+unqualified. Measured over a real connection, the widest access log line
+is 1,972 bytes.
 
 Multiply that ceiling by the request rate to size log storage. Note
 that the rate is not bounded by the limits above on every route:
 `/.well-known/healthcheck` and `/s/*` sit behind no limiter, so there
 the multiplier is whatever the deployment will serve.
 
+**The same ceiling covers every other line the service writes through
+`slog` that carries text an unauthenticated client supplies**, with one
+exception stated below it: the recovered-panic record, which carries a
+whole goroutine stack alongside its client-supplied fields and so has
+its own wider ceiling. The access log is not the only line a client can
+put its own text into, and a budget that held for one line and not the
+others would be worse than no stated budget at all. Every `slog` call an
+unauthenticated request can reach spends the same per-field budget
+through `internal/logfield`, and each carries strictly fewer
+client-supplied fields than the access log does, so none of them can be
+wider than it:
+
+| Log line                                   | Level   | Client-chosen value | Reachable unauthenticated                 |
+| ------------------------------------------ | ------- | ------------------- | ----------------------------------------- |
+| `request body exceeds limit` (413)         | `WARN`  | path, method        | yes — `MaxBodySize` precedes `RequireAuth` |
+| `csrf: token validation failed` (403)      | `WARN`  | path, method        | yes — `CSRF` precedes `RequireAuth`        |
+| `... rate limit exceeded` (429)            | `WARN`  | path                | yes, on the receiver                       |
+| `auth middleware: unauthenticated request` | `DEBUG` | path, method        | yes, by definition                         |
+| `entrypoint not found`                     | `DEBUG` | entrypoint UUID     | yes, on the receiver                       |
+| `user not found` / `invalid password`      | `DEBUG` | username            | yes, on the login form                     |
+| `login failure limit exceeded` (429)       | `WARN`  | path                | yes, on the login form                     |
+| `password verification capacity exhausted` | `WARN`  | path                | yes, on the login form                     |
+
+`DEBUG` being off by default is not a bound. An operator turning it on
+to diagnose a flood must not thereby hand the flood an unbounded write,
+so those lines are capped too.
+
+The last two rows are capped defensively rather than against a
+demonstrated width: chi routes `POST /pages/login` on a static pattern,
+so `r.URL.Path` there is the 12-byte constant `/pages/login` and each
+line lands near 120 bytes. `RecordLoginFailure` is nonetheless an
+exported method taking any `*http.Request`, and a future caller on a
+route with a URL parameter would widen the line. Since no request
+through the mux can, both caps are pinned by tests that call those two
+entry points directly with the path such a caller would supply.
+Removing either cap fails 14 subtests.
+
+`internal/middleware/logbound_test.go` and
+`internal/handlers/logbound_test.go` drive 8 KB of client-chosen text
+at each of these — 1 KB at `invalid password`, whose accounts are
+shared with the successful-login line, where a username past 4 KB
+overflows the session cookie and answers 500 before that line is
+written — through both handlers, and through seven fills: plain text
+as the baseline, and then the quotation mark, backslash, tab, newline,
+C0 control and astral non-printable, six characters the wider of the
+two handlers spends more on than the client spent sending them. Every
+case holds each line to the 2,560-byte ceiling. That per-line ceiling
+is what the figure above states, and every row establishes it.
+
+Three of the sites go further and bound the whole flood's output — the
+total bytes a run of distinct invented values wrote, which is the
+shape an operator sizing storage cares about. They are
+`request body exceeds limit`
+(`TestMaxBodySize_FloodOfOversizePathsDoesNotGrowTheLog`),
+`entrypoint not found` and `user not found` (the last two through
+`assertBoundedFlood`). The other rows carry no aggregate assertion;
+the per-line ceiling is what they establish.
+
+`internal/logfield/logfield_test.go` measures the per-rune charge
+against what the handlers really emit, over roughly 3,000 code points on
+each, so an undercharged rune fails a test rather than quietly
+falsifying the ceiling.
+
+**It covers GORM's statement logging as well.** GORM's own default
+logger printed the fully interpolated SQL — parameters and all — to
+standard output on every statement that returned an error, including a
+plain record-not-found, at a level no operator setting reached. Two of
+this service's lookups miss by design on unauthenticated routes: the
+entrypoint lookup behind `/webhook/{uuid}` and the user lookup behind
+the login form, whose path segment and submitted username the client
+picks outright. Every
+`gorm.Open` in the service now installs the adapter in
+`internal/gormlog` instead. It writes through the same `slog` logger as
+everything else, so its lines take the level the operator set and the
+handler `internal/logger` selected, and every value it emits is spent
+through the same `internal/logfield` budget. A record-not-found is not
+logged as an error: it is the expected outcome on both of those paths,
+and each handler already records its own miss at `DEBUG` — bounded, per
+the table above — without the SQL. Slow statements are kept, at `WARN`,
+above the same 200 ms threshold GORM used and with the statement
+bounded, because that report is the one thing GORM's logger gave an
+operator that nothing else here does. The adapter orders its cases
+exactly as GORM's own `Trace` orders them, so a statement that both
+missed and ran slow is still reported as slow, and dropping the miss
+costs an operator no report `IgnoreRecordNotFoundError` would have
+kept. A GORM line spends at most two of those budgets — the statement
+and the driver error — against a smaller fixed portion than the access
+log's, and `internal/gormlog/gormlog_test.go` asserts each line against
+`MaxAccessLogLineBytes` directly rather than leaving it as arithmetic.
+
+What that ceiling does **not** cover, stated here so the figure is not
+read as more than it is:
+
+- **Lines carrying an authenticated operator's own input**, which are
+  not truncated at all. `webhook created` logs the submitted `name`
+  verbatim and `target URL blocked by SSRF protection` logs the target
+  host (both `internal/handlers/source_management.go`), as do the
+  `target_name` lines in `internal/delivery/engine.go` and
+  `internal/delivery/target_http.go`. The only bound on any of them is
+  the 1 MB form body cap, so a 100 KB `name` writes a single line of
+  roughly 600 KB — measured. This is deliberate: every one of these
+  requires an authenticated operator on a service with no
+  self-registration, and truncating the operator's own configuration
+  echoed back would cost debuggability against no adversary. It does
+  mean the 2,560-byte figure sizes unauthenticated traffic, not the
+  operator's own administrative requests.
+- The **`log` delivery target**, which writes the whole inbound event —
+  headers and body — to the log. This one is deliberate: capping it
+  would defeat the target, since emitting the payload is the delivery.
+  It costs nothing unless an authenticated operator creates a target of
+  that type on a specific webhook, and each line it writes is bounded
+  per event by the 1 MB receiver body cap. Adding one is a decision to
+  spend log volume on that webhook's payloads.
+- **Two writers that do not go through `internal/logger` at all**, both
+  on standard error. `fx` prints the dependency graph and the lifecycle
+  hooks through its default console logger at startup and shutdown —
+  nothing calls `fx.WithLogger`, and `fx.New` builds that logger over
+  `os.Stderr`. The Go runtime writes a panic or a fatal error itself; a
+  panic in a background worker rather than in a request handler is the
+  case that reaches it, since nothing recovers those. Neither carries a
+  client-chosen value at a client-chosen length: the five `panic` calls
+  in this service are invariant guards over constants and over
+  `crypto/rand`.
+- **`net/http`'s own faults**, which are _not_ a separate writer.
+  `internal/server/http.go` builds its server with a nil `ErrorLog`, so
+  `net/http` falls back to the `log` package's default logger — and
+  `internal/logger` calls `slog.SetDefault`, which redirects that logger
+  into whichever handler it installed. Those lines therefore arrive on
+  standard output, shaped like every other line, at `INFO`. They are not
+  truncated. A handler panic is no longer one of them: the recover
+  middleware below answers it and writes it as the bounded record
+  described there instead, and `internal/server/recoverer_test.go`
+  requires that `http: panic serving` appear in neither of the process's
+  two streams when a panic is driven through the production router. The
+  one panic still handed back to `net/http` is `http.ErrAbortHandler`,
+  which it special-cases and does not log at all. What is left on this
+  path is `net/http`'s own diagnostics, whose values are the runtime's,
+  not a client's.
+
+Wider than that 2,560-byte ceiling, and stated separately rather than
+carved out of it: the record a recovered panic produces. The recover
+middleware in `internal/middleware` answers `500` and writes one `ERROR`
+record through `internal/logger` carrying the panic value, the stack and
+the request id — the same `request_id` the access log line for that
+request carries, which is how the two are joined. It replaced chi's
+`middleware.Recoverer`, which on a current Go release crashed inside its
+own stack pretty-printer: the connection was dropped rather than
+answered, and what reached the operator described that crash rather than
+the fault behind it.
+
+That record is bounded the same way, in the same encoded bytes and
+through the same `internal/logfield` budget: 512 for the panic value,
+because a handler is free to build one out of the request, 128 for the
+request id, which a client supplies outright through `X-Request-Id`,
+and 8,192 for the stack, cut at its far end so that the panic site
+survives a cut and `net/http`'s accept frames are what is lost. Net:
+**at most 10,240 bytes, once per recovered panic** — 9,121 by the
+arithmetic (523 + 8,203 + 139 + a 256-byte fixed portion), stated at
+10,240 for headroom.
+
+Those two numbers are the claim; the measurements below only
+illustrate it. `internal/middleware/recoverer_test.go` drives all
+three growable fields past their budgets on one record, over both
+handlers, and measured 9,009 bytes on the JSON handler and
+8,982–8,983 on the text one in one checkout. Neither is an invariant:
+the stack's own content decides where its cut lands, so the figures
+move by a byte or so between runs. The real case is far below both —
+through the shipped middleware chain the whole record measures
+roughly 3,960 bytes over a roughly 3,690-byte stack, taken by
+`internal/server/recoverer_test.go` from the process's own file
+descriptors while driving a panic through the production router over
+a real server in a subprocess. That pair moves further still, since
+`debug.Stack()` embeds absolute source paths and so depends on where
+the tree is checked out: four checkouts have reported 3,959, 3,961,
+3,984 and 4,026. What the tests assert is the ceiling, that every
+client-supplied field was cut, and that the shipped chain's stack
+arrived uncut — never the numbers.
+
 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
@@ -1091,14 +1360,117 @@ opposite directions:
   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`.
-- For the **login and password-change** limits it costs availability of
-  the only administrative path, which is not safe at all. Five POSTs
-  per minute from any address on the internet keeps the single shared
-  login bucket full, and the operator's own login returns HTTP 429 for
-  as long as that trickle continues. A restart clears the in-memory
-  buckets and a resumed trickle re-locks them. Production deployments
-  must set `TRUSTED_PROXIES`; webhooker warns at startup whenever it is
-  empty, in any environment.
+- For the **login and password-change** limits it costs precision, not
+  availability. Login failures from every client land in one counter,
+  so a stranger's wrong passwords make the operator's own wrong
+  passwords answer `429` sooner; the operator's _correct_ password is
+  never affected, because it is never counted. Production deployments
+  should still set `TRUSTED_PROXIES`; webhooker warns at startup
+  whenever it is empty, in any environment.
+
+#### The login endpoint
+
+The login `POST` is the one endpoint with no pre-emptive limiter in
+front of it, and that is deliberate. A limiter that spends budget on
+arrival is a lockout in this deployment shape: sharing one bucket, a
+stranger sending five POSTs a minute — about 0.08 requests per second,
+from anywhere — keeps it permanently full, and the operator has no
+second administrative path. So the handler inverts the order:
+
+1. **Credentials are verified first, and only a failed attempt spends
+   budget.** A correct password is never rate-limited, whatever the
+   counters hold. This is what guarantees the admin UI stays
+   reachable.
+2. **Failures are counted per (client bucket, submitted username)**,
+   five per minute, after which further _failures_ from that pair are
+   answered `429` with a `Retry-After`. That `429` is a label on the
+   response, not a gate in front of the work: the credential check has
+   already run by the time the counter is consulted, so a throttled
+   client's guess is still evaluated. See the guessing rate below. A
+   successful login clears the counter, so mistyping a few times and
+   then getting it right leaves you unthrottled. Because the submitted
+   username is attacker-controlled, at most 1024 username counters and
+   1024 fallback address counters are tracked; past the first cap
+   failures fall back to the address counter, and past both they are
+   answered as throttled without being recorded. Total tracked state
+   is under half a megabyte and does not grow with the number of
+   usernames an attacker invents.
+3. **Concurrent password verifications are capped at two, and the
+   queue for them at 16.** Verifying before counting means every login
+   request costs an Argon2id hash, and Argon2id here is 64 MB per
+   hash — two slots is a 128 MB ceiling on password hashing. Every
+   endpoint that hashes a password takes a slot, including the
+   password-change endpoint, which holds one across both the
+   verification and the new hash. A request that waits five seconds
+   without getting a slot is answered `503 Service Unavailable` and no
+   hash is computed for it. The wait alone does not bound memory, only
+   how long one request holds some, so the number of waiters is capped
+   as well. Size the queue from what a parked waiter actually retains,
+   not from the 1 MB body cap: that caps the raw body read, while the
+   body-cap, CSRF and form-parsing middleware all run before the
+   guard, so a waiter holds its parsed form plus its request header
+   block for the whole wait. Measured on the pinned Go 1.26.1
+   toolchain, as the heap delta with 64 waiters parked in the handler,
+   an ordinary two-field login form retains ~0 MB, a 1 MB urlencoded
+   body at Go's 10,000-parameter parse cap retains 2.82 MB (3.09 MB
+   with `%41` escapes), and the ~0.9 MB of headers the 1 MB header cap
+   allows takes it to **4.18 MB** — the retained parse and the headers
+   dominate, not the raw body. So the cap is 16 waiters: 16 x 4.18 MB
+   is about 67 MB of committed queue memory, and two slots drain a
+   full 16-deep queue in roughly 0.6 s, far inside the five-second
+   deadline. A request arriving past the cap is shed with `503`
+   immediately instead of joining the queue. **Peak commitment for the
+   endpoint is therefore about 203 MB**: 128 MB of Argon2id, plus the
+   18 requests holding a parsed form — 16 queued and the 2 being
+   hashed — at about 75 MB. That 203 MB is _live_ commitment, not
+   resident size: the Go collector lets the heap reach roughly twice
+   the live set before collecting, with transient parse garbage on top
+   of it. The independent review of this endpoint fired 18 adversarial
+   requests at an idle guard and measured a peak `HeapAlloc` of
+   392 MB. **Provision on the order of 400 MB**, not for the 203 MB
+   itemised here and not for the hashing budget alone.
+
+An unknown username is verified against a dummy hash rather than
+rejected early, so a nonexistent account costs the same time as a real
+one and the response cannot be used to enumerate usernames.
+
+**This raises online guessing throughput by about 300x, and that is
+the trade.** Because the credential check always precedes the counter,
+what bounds online brute force is the semaphore, not the failure
+counter. Two slots at the cost of one Argon2id verification is on the
+order of **27 guesses per second, about 2.3 million per day**, against
+5 per minute under the pre-emptive limiter this replaced. Treat that
+figure as a lower bound rather than a ceiling: it was measured with
+Go's race detector enabled, so real hardware verifies faster and
+guesses faster. Choose the admin password to survive millions of
+online guesses per day — a long random passphrase, not a memorable
+one. Rate-limiting `POST /pages/login` at the reverse proxy, where the
+real client address is visible, is the way to put a cheaper bound back
+on top.
+
+The residual exposure is a bounded, self-clearing loss of login
+**availability** — not merely of latency. A flood can keep both
+verification slots busy, and a request that neither gets a slot within
+five seconds nor finds room in the queue is answered `503`. Above
+roughly 27 requests per second the operator is not served slowly, it
+is shed: its chance per attempt is about the ratio of service rate to
+flood rate, so at 400 requests per second it is roughly one attempt in
+fourteen. A sufficiently determined flood still denies login for as
+long as it runs.
+
+What changed is the price and the aftermath. Denying login used to
+cost an attacker 0.08 requests per second from anywhere; it now costs
+30 or more sustained, about 400 times as much. Nothing accumulates
+while the flood runs, nothing needs resetting when it stops, and the
+operator's correct password succeeds on the first attempt afterwards.
+Restarting the service is **not** a remedy: a restart clears the
+failure counters, which are not what is saturated, and the flood
+re-fills both verification slots on its first two requests. The
+remedies are to block the source at the reverse proxy, or to
+rate-limit `POST /pages/login` there — the one place a limit can be
+applied without reintroducing the lockout, because the proxy sees the
+real client address. Setting `TRUSTED_PROXIES` does not stop the
+saturation, but it makes the source visible in the failure logs.
 
 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
@@ -1119,8 +1491,8 @@ abuse limit later; they are tracked as future work.
 
 | Method | Path            | Description |
 | ------ | --------------- | ----------- |
-| `GET`  | `/pages/login`  | Login page (not rate limited; the limiter applies to POST only) |
-| `POST` | `/pages/login`  | Login form submission (5 per minute per bucket, then 429) |
+| `GET`  | `/pages/login`  | Login page (not rate limited) |
+| `POST` | `/pages/login`  | Login form submission. Credentials are verified before any limit is consulted, so a correct password is never throttled; 5 FAILED attempts per minute per bucket per submitted username, then `429`. `503` if no verification slot frees up within 5s, or immediately if 16 requests are already queued for one (see [Rate Limiting](#rate-limiting)) |
 | `POST` | `/pages/logout` | Logout (destroys session) |
 
 #### Authenticated Endpoints
@@ -1128,7 +1500,7 @@ abuse limit later; they are tracked as future work.
 | Method | Path                     | Description |
 | ------ | ------------------------ | ----------- |
 | `GET`  | `/user/{username}`       | User profile page |
-| `POST` | `/user/{username}/password` | Change the user's password (5 per minute per bucket, then 429) |
+| `POST` | `/user/{username}/password` | Change the user's password (5 per minute per bucket, then `429`; `503` if no verification slot frees up within 5s, or immediately if 16 requests are already queued for one) |
 | `GET`  | `/sources`               | List user's webhooks |
 | `GET`  | `/sources/new`           | Create webhook form |
 | `POST` | `/sources/new`           | Create webhook submission |
@@ -1198,6 +1570,10 @@ webhooker/
 │   │   └── webhook_db_manager.go  # Per-webhook DB lifecycle manager
 │   ├── globals/
 │   │   └── globals.go         # Build-time variables (appname, version, arch)
+│   ├── gormlog/
+│   │   └── gormlog.go         # GORM's logger.Interface on top of slog, bounded
+│   ├── logfield/
+│   │   └── logfield.go        # Encoded-byte budget for client-supplied log values
 │   ├── delivery/
 │   │   ├── engine.go          # Event-driven delivery engine (channel + timer based)
 │   │   ├── circuit_breaker.go # Per-target circuit breaker for http/slack targets with retries
@@ -1230,6 +1606,7 @@ webhooker/
 │   │   ├── middleware.go      # Logging, CORS, Auth, Metrics, MetricsAuth, SecurityHeaders, MaxBodySize
 │   │   ├── csrf.go            # CSRF protection middleware (gorilla/csrf)
 │   │   ├── ratelimit.go       # Per-IP rate limiting middleware (go-chi/httprate)
+│   │   ├── loginguard.go      # Login failure counters and the Argon2id verification semaphore
 │   │   └── testing.go         # NewForTest: Middleware without the fx lifecycle
 │   ├── server/
 │   │   ├── server.go          # Server struct, fx lifecycle, signal handling
@@ -1292,19 +1669,32 @@ to record results.
 
 Applied to all routes in this order:
 
-1. **Recoverer** — Panic recovery (chi built-in)
-2. **RequestID** — Generate unique request IDs (chi built-in)
-3. **SecurityHeaders** — Production security headers on every response
+1. **RequestID** — Generate unique request IDs (chi built-in)
+2. **SecurityHeaders** — Production security headers on every response
    (HSTS, X-Content-Type-Options, X-Frame-Options, CSP, Referrer-Policy,
    Permissions-Policy)
-4. **Logging** — Structured request logging (method, URL, status,
+3. **Logging** — Structured request logging (method, URL, status,
    latency, remote IP, user agent, request ID)
-5. **Metrics** — Prometheus HTTP metrics (if `METRICS_USERNAME` is set)
-6. **CORS** — Cross-origin resource sharing headers
-7. **Timeout** — 60-second request timeout
+4. **Metrics** — Prometheus HTTP metrics (if `METRICS_USERNAME` is set)
+5. **CORS** — Cross-origin resource sharing headers
+6. **Timeout** — 60-second request timeout
+7. **Recoverer** — Panic recovery: one `ERROR` record through
+   `internal/logger` and a `500`
 8. **Sentry** — Error reporting to Sentry (if `SENTRY_DSN` is set;
    configured with `Repanic: true` so panics still reach Recoverer)
 
+Recoverer sits seventh rather than first, and both neighbours are the
+reason. It runs **inside** everything that observes the response, so
+the `500` it writes for a panicking handler is the status the access
+log records and the metrics count; registered first, as chi's own
+`middleware.Recoverer` was, the same request was logged as a `200` that
+the client never received. It runs **outside** the Sentry handler, so
+`Repanic: true` has something to re-raise into: an operator with
+`SENTRY_DSN` set keeps the report, and one without it now gets the
+local record instead of nothing. What that placement gives up is
+recovery of a panic in the six entries above it, none of which does
+more than set a header or start a timer.
+
 Additionally, form endpoints (`/pages`, `/user/*`, `/sources`,
 `/source/*`) apply a **MaxBodySize** middleware that limits
 POST/PUT/PATCH request bodies to 1 MB. It is registered ahead of the
@@ -1327,9 +1717,11 @@ one that lies about its length, is hard-capped by
 Those same four route groups then apply **CSRF** and **NoCache**
 (`Cache-Control: no-store`, `Pragma: no-cache`), and every group except
 `/pages` applies **RequireAuth**. The rate limiters are per-route
-rather than global: **LoginRateLimit** on `/pages/login`,
-**PasswordChangeRateLimit** on `/user/{username}/password`, and
-**ReceiverRateLimit** on `/webhook/{uuid}`.
+rather than global: **PasswordChangeRateLimit** on
+`/user/{username}/password` and **ReceiverRateLimit** on
+`/webhook/{uuid}`. There is deliberately none on `/pages/login` — that
+endpoint counts failures inside the handler, after the credential
+check, see [The login endpoint](#the-login-endpoint).
 
 ### Authentication
 
@@ -1367,14 +1759,27 @@ rather than global: **LoginRateLimit** on `/pages/login`,
   both at target creation time (URL validation) and at delivery time
   (custom HTTP transport with SSRF-safe dialer that validates resolved
   IPs before connecting, preventing DNS rebinding attacks)
-- **Login rate limiting** via [go-chi/httprate](https://github.com/go-chi/httprate):
-  sliding-window rate limiter on the login endpoint, 5 POST attempts
-  per minute per bucket, to slow brute-force attacks. GET requests to
-  the login page are not limited. The password-change endpoint carries
-  the same 5-per-minute limit. The bucket is per client IP only when
+- **Login limiting is inverted, deliberately.** The login `POST` has
+  no pre-emptive rate limiter in front of it. Credentials are
+  verified first and only a _failed_ attempt spends budget, so a
+  correct password is never throttled and no flood of wrong ones can
+  deny the operator the only administrative path. Failures are
+  counted per (bucket, submitted username), five per minute, after
+  which further failures are answered `429` with a `Retry-After`.
+  What bounds brute force is not that counter but the cap of two
+  concurrent Argon2id verifications: a throttled client's guess is
+  still evaluated, so roughly 27 guesses a second get through and the
+  admin password has to carry that load (see
+  [The login endpoint](#the-login-endpoint)). `GET` requests to the
+  login page are not limited
+- **Password-change rate limiting** via [go-chi/httprate](https://github.com/go-chi/httprate):
+  sliding-window rate limiter, 5 POST attempts per minute per bucket.
+  It runs behind session auth, so only a client already holding a
+  valid session reaches it, and an operator throttled out of changing
+  a password can still log in. The bucket is per client IP only when
   `TRUSTED_PROXIES` names the reverse proxy; unset, every client
-  shares one bucket and the login becomes remotely deniable (see
-  [Rate Limiting](#rate-limiting)). webhooker warns at startup
+  shares one bucket, which costs precision rather than availability
+  (see [Rate Limiting](#rate-limiting)). webhooker warns at startup
   whenever `TRUSTED_PROXIES` is empty
 - Prometheus metrics behind basic auth
 - Static assets embedded in binary (no filesystem access needed at

REPO_POLICIES.md

@@ -1,6 +1,6 @@
 ---
 title: Repository Policies
-last_modified: 2026-07-06
+last_modified: 2026-08-07
 ---
 
 This document covers repository structure, tooling, and workflow standards. Code
@@ -189,8 +189,13 @@ style conventions are in separate documents:
   module under test to verify it compiles/parses. There is no excuse for
   `make test` to be a no-op.
 
-- `make test` must complete in under 20 seconds. Add a 30-second timeout in the
-  Makefile.
+- `make test` must complete in under 60 seconds. That is the hard cap, and a
+  suite that exceeds it fails. Under 20 seconds is the target. A suite between
+  20 and 60 seconds is still green, but the overage must be filed as an
+  improvement bug against that repo. Add a 90-second timeout to the test
+  invocation in the Makefile (`go test -timeout 90s`). The backstop deliberately
+  sits above the hard cap so that it catches a genuinely hung test rather than a
+  merely slow one.
 
 - **`make test` should use the conditional verbose rerun pattern.** Run tests
   without `-v` (verbose) first. If tests fail, automatically rerun with `-v` to
@@ -209,9 +214,9 @@ style conventions are in separate documents:
 
     ```makefile
     test:
-    	@go test -timeout 30s -race -cover ./... || \
+    	@go test -timeout 90s -race -cover ./... || \
     		{ echo "--- Rerunning with -v for details ---"; \
-    		  go test -timeout 30s -race -v ./...; exit 1; }
+    		  go test -timeout 90s -race -v ./...; exit 1; }
     ```
 
     Python example:
@@ -260,7 +265,10 @@ style conventions are in separate documents:
 
 - `.golangci.yml` is standardized and must _NEVER_ be modified by an agent, only
   manually by the user. Fetch from
-  `https://git.eeqj.de/sneak/prompts/raw/branch/main/.golangci.yml`.
+  `https://git.eeqj.de/sneak/prompts/raw/branch/main/.golangci.yml`. The
+  canonical golangci-lint version is v2.12.2 (released 2026-05-06), installed
+  commit-pinned via
+  `go install github.com/golangci/golangci-lint/v2/cmd/golangci-lint@c0d3ddc9cf3faa61a4e378e879ece580256d76e5`.
 
 - When pinning images or packages by hash, add a comment above the reference
   with the version and date (YYYY-MM-DD).

TODO.md

@@ -24,30 +24,142 @@ event retention (#63), the database archiving target (#43), the admin
 password change flow (#65), policy compliance (#6), pinned lint tooling
 (#55), and fail-loud configuration parsing (#80).
 
-`next` holds the completed 1.0.0 milestone: every issue in it is closed,
-and it is verified green by cache-defeated container runs
-(`docker build --no-cache-filter=lint --no-cache-filter=builder`). The
-CI status is not independently claimed here: a superseded run is
-recorded as `skipped` and still rolls up green, so a commit status on
-`next` does not by itself evidence an executed check (#152). Before
-#119, a warm layer cache also let the gate report success without
-executing anything, and replayed the previous build's console log so
-the lie looked like a real run. Note: `TODO.md` was deliberately
+`next` holds the **complete 1.0.0 milestone**: every issue in it is
+closed, and it is verified green both by CI and by cache-defeated
+container runs (`docker build --no-cache-filter=lint
+--no-cache-filter=builder`).
+
+One caveat on reading a green check, narrower than it used to be. A
+docs-only commit deliberately replays from the layer cache (#119), so a
+green status on such a commit evidences a replay rather than an executed
+run; a code commit invalidates the `COPY` layer and genuinely executes.
+Superseded runs are no longer the hazard they were: before #152 they
+were recorded as `skipped` and rolled up green, and before #119 a warm
+layer cache let the gate report success without executing anything,
+replaying the previous build's console log so the lie looked like a real
+run. Both are fixed. Note: `TODO.md` was deliberately
 deleted from this repo in f9a9569 (2026-03-01, #6); its content was
 folded into the README TODO section, which this draft reconstructs as
 of 2026-07-06.
 
 # Next Step
 
-Merge the milestone PR to `main` and tag 1.0.0 from it.
+Merge the milestone PR (#111) to `main` and tag 1.0.0 from it. The
+milestone is empty and `next` is green; nothing else blocks the tag.
 
-Two decisions are open and belong to the owner, neither blocking the
-tag: #115 (mask the `http` target's destination URL, implemented
-speculatively and awaiting a yes or no) and #125 (whether IPv6
-rate-limit keys should bucket by `/64`).
+Three items belong to the owner, none of them blocking. #150 was decided
+by the manager rather than left to stall the queue and is flagged on the
+issue for reversal if that call was wrong. #112 (whether `Completed
+Steps` should exist at all, given it once conflicted on every unit) is
+unanswered; the provisional ruling in force is that issue branches do
+not touch this file. #198 records that `make test` is past the org 20s
+target — 46s of test execution inside a 62.8s CI layer — and turns on
+which quantity the 60s hard cap governs; it is scoped as the improvement
+bug the 20-60s band requires, and should be milestoned instead if the
+cap is read as covering the whole invocation.
+
+After the tag, the largest open cluster is the unmilestoned follow-up
+backlog these units generated: #183, #184, #185, #190, #191, #193 and
+#198.
 
 # Completed Steps
 
+- 2026-08-18 Raise `script/test`'s per-package timeout from 30s to 90s,
+  matching the org-wide backstop. `go test` applies `-timeout` per
+  package, and `internal/handlers` had grown past the old budget: a
+  cache-defeated build failed outright at `GOMAXPROCS=4`, and every run
+  under deliberate host load breached 30s. The measurement table lives
+  in the script (#194)
+- 2026-08-18 Re-sync `REPO_POLICIES.md` from `prompts`. The local copy
+  was stale and still mandated a 20s test target with a 30s timeout,
+  which the org replaced with a 60s cap and a 90s backstop. A synced
+  copy is not a source; reading it as one nearly produced a PR against
+  `prompts` proposing a change already merged there (#196)
+- 2026-08-18 Report handler panics through the logger and answer 500.
+  chi v1.5.5's `Recoverer` scans for a `panic(0x` frame the runtime no
+  longer emits, then indexes `pkg[-1:]`, so it panicked inside its own
+  stack printer before writing a byte: the recovery never ran, the
+  client got a dropped connection instead of a 500, and the original
+  panic was lost. A local middleware replaces it, bounded by
+  `MaxPanicLogLineBytes` (#187)
+- 2026-08-18 Route GORM's logger through `slog` and bound it. Every
+  `gorm.Open` left `logger.Default` in place at `Warn` with
+  `IgnoreRecordNotFoundError` false, so **every record-not-found
+  printed the fully interpolated SQL to stdout** — including the
+  client-chosen path on `/webhook/{uuid}` and the submitted username on
+  the login form, at no level the operator set and outside
+  `internal/logger` entirely. Three call sites, not the two the issue
+  named (#178)
+- 2026-08-18 Bound every `slog` line against client-chosen text. Eight
+  sites reachable unauthenticated, found by reading every `slog` call in
+  the tree rather than only the one reported; the budget moved to a
+  shared `internal/logfield` so no second truncation exists. `DEBUG`
+  being off by default is not a bound and is not treated as one (#176)
+- 2026-08-18 Stop a slow host turning a login-guard test into a
+  segfault. A non-fatal `assert` on an acquire result was dereferenced
+  on the next line, so one timing miss killed the whole
+  `internal/middleware` binary and reddened CI for unrelated PRs. The
+  fix also removed a real production race — `acquire` could shed a
+  request with a slot standing free, because Go picks uniformly among
+  ready `select` cases (#186)
+- 2026-08-18 Send the chi route pattern to Sentry rather than the
+  concrete path. The receiver's path carries the entrypoint capability
+  token, so every Sentry event from `/webhook/{uuid}` shipped a live
+  credential to a third party. Request `Data`, `QueryString`, `Cookies`
+  and `Env` are dropped and headers reduced to an allowlist (#179)
+- 2026-08-18 Read form fields from the POST body only. `r.FormValue`
+  merges the query string, so a login could be driven by URL parameters
+  — putting the password somewhere that lands in access logs, proxy
+  logs and browser history (#160)
+- 2026-08-18 Verify login credentials before spending rate-limit
+  budget, so a flood of wrong passwords cannot lock out the account it
+  is guessing at. The manager took this decision rather than stall the
+  queue; it is flagged on the issue for reversal (#150)
+- 2026-08-18 Run all linting in Docker via `Dockerfile.lint`. Host lint
+  was wrong in both directions from version skew and shared caches.
+  `script/lint` asserts the summary line, because `--no-cache-filter`
+  silently ignores a stage name it does not match — the flag that makes
+  the gate meaningful fails open (#109)
+- 2026-08-18 Serve an event's full stored body over HTTP. The list
+  query truncates for rendering, and that truncated value was the only
+  way to read a body, so the full payload was unreachable (#157)
+- 2026-08-18 Bound the access log line against client-chosen text.
+  `internal/logfield` budgets by *encoded* bytes, not runes, so a
+  handler's JSON escaping cannot multiply a field past its allowance
+  (#146)
+- 2026-08-18 Mark superseded CI commits `failure` rather than
+  `skipped`. A skipped run rolls up green, so a commit that was never
+  tested reported success (#152)
+- 2026-08-18 Set `fx.StopTimeout` inside the container stop grace, so
+  shutdown hooks are bounded by a deadline the orchestrator will
+  actually honour rather than being killed mid-flush (#134)
+- 2026-08-17 Bucket IPv6 rate-limit keys by `/64`. A single allocation
+  hands out 2^64 addresses, so per-address keying let one client mint
+  unlimited buckets. Manager decision, recorded on the issue (#125)
+- 2026-08-17 Correct release-blocking README and startup-warning
+  inaccuracies, including claims about behaviour the code does not have
+  (#151)
+- 2026-08-17 Fetch and verify Alpine.js at build time against
+  `static/vendor.sha256` instead of committing the minified blob, so
+  the dependency is pinned by hash rather than by trust (#145)
+- 2026-08-17 Bound the event log's rendered bodies in the query itself,
+  so a large stored payload cannot be read into memory just to be
+  truncated for display (#135)
+- 2026-08-17 Mask the `http` target's destination URL in the UI: it can
+  carry a bearer credential in its path or query, and was rendered
+  verbatim. Manager decision to mask unconditionally (#115)
+- 2026-08-14 Bound shutdown hooks by their stop context, so a hook that
+  hangs cannot hold the process past its grace period (#102)
+- 2026-08-14 Render templates via a buffer rather than the
+  `ResponseWriter`, so a template error part-way through cannot commit
+  a 200 and then fail — the response is written only once it is whole
+  (#123)
+- 2026-08-14 Align the session codec's max-age with the 7-day absolute
+  cap. The codec accepted cookies the session layer considered expired,
+  so the cap was enforced in one place and not the other (#108)
+- 2026-08-12 Warn when `TRUSTED_PROXIES` is empty in production, where
+  the safe default silently discards forwarded headers and every client
+  rate-limits as the proxy's address (#149)
 - 2026-08-12 Bound the receiver rate limit per client IP across the
   whole `/webhook/*` route. The existing limiter keyed on the request
   path and `/webhook/{uuid}` matches any single segment, so a client

internal/config/config.go

@@ -430,10 +430,14 @@ func loadFromEnv() (*Config, error) {
 // what is in front of the process, which this code cannot observe:
 // with nothing in front, the peer is the client and the limits are
 // per-client as intended; behind a reverse proxy the peer is the proxy
-// for every request, so all clients share one bucket per limiter. The
-// login limiter's bucket is the dangerous one: any remote client can
-// keep it full, which denies the only administrative login to everyone
-// until the process restarts.
+// for every request, so all clients share one bucket per limiter.
+//
+// The login endpoint no longer spends budget on arrival — it verifies
+// credentials first and charges only failures — so a shared bucket
+// cannot deny the operator a correct password. What it does collapse
+// is the failure counting: one client's wrong passwords throttle
+// everyone else's wrong passwords, and the receiver's limits become
+// service-wide ceilings.
 //
 // The warning is deliberately not gated on WEBHOOKER_ENVIRONMENT. That
 // variable defaults to dev, so gating on it would silence the warning
@@ -454,11 +458,11 @@ func (c *Config) warnSharedRateLimitBucket(log *slog.Logger) {
 			"this process that is the client itself and the limits "+
 			"are per-client as intended. Behind a reverse proxy the "+
 			"peer is the proxy on every request, so all clients "+
-			"share one bucket per limit and any remote client can "+
-			"keep the login limit full, denying the admin login — "+
-			"the only administrative path — until restart. If "+
-			"anything proxies to this process, set TRUSTED_PROXIES "+
-			"to its address.",
+			"share one bucket per limit: the receiver limits become "+
+			"service-wide ceilings, and one client's failed logins "+
+			"throttle every other client's failed logins — a "+
+			"correct password still gets in. If anything proxies to "+
+			"this process, set TRUSTED_PROXIES to its address.",
 		"environment", c.Environment,
 		"trustedProxies", len(c.TrustedProxies),
 	)

internal/config/config_test.go

@@ -629,8 +629,9 @@ func testTrustedProxiesSuccess(
 
 // TestSharedRateLimitBucketWarning covers the startup warning that
 // tells an operator a deployment behind a reverse proxy shares one
-// rate-limit bucket between every client, which makes the admin login
-// remotely deniable. It must fire whenever TRUSTED_PROXIES is empty,
+// rate-limit bucket between every client, which turns the receiver
+// limits into service-wide ceilings and collapses login failure
+// counting. It must fire whenever TRUSTED_PROXIES is empty,
 // in any environment: WEBHOOKER_ENVIRONMENT defaults to dev, so gating
 // on it would silence the warning for exactly the operator who never
 // configured the deployment. It stays quiet once proxies are named.
@@ -707,7 +708,15 @@ func TestSharedRateLimitBucketWarning(t *testing.T) {
 			assert.Contains(t, logged, `"level":"WARN"`)
 			assert.Contains(t, logged, "TRUSTED_PROXIES")
 			assert.Contains(t, logged, "share one bucket")
-			assert.Contains(t, logged, "denying the admin login")
+			assert.Contains(
+				t, logged, "throttle every other client's failed logins",
+			)
+			// The warning must not claim a lockout the login
+			// endpoint no longer permits: credentials are verified
+			// before any budget is spent.
+			assert.Contains(
+				t, logged, "a correct password still gets in",
+			)
 			// The text must stay accurate for a developer with
 			// nothing in front of the process, where an empty
 			// list costs nothing.

internal/database/database.go

@@ -17,6 +17,7 @@ import (
 	"gorm.io/gorm"
 	_ "modernc.org/sqlite" // Pure Go SQLite driver
 	"sneak.berlin/go/webhooker/internal/config"
+	"sneak.berlin/go/webhooker/internal/gormlog"
 	"sneak.berlin/go/webhooker/internal/logger"
 )
 
@@ -155,7 +156,10 @@ func (d *Database) connect() error {
 	// Then use it with GORM
 	db, err := gorm.Open(sqlite.Dialector{
 		Conn: sqlDB,
-	}, &gorm.Config{})
+	}, &gorm.Config{
+		// Never leave this at GORM's default. See internal/gormlog.
+		Logger: gormlog.New(d.log),
+	})
 	if err != nil {
 		d.log.Error(
 			"failed to connect to database",

internal/database/export_test.go

@@ -65,3 +65,9 @@ func (r *RetentionReaper) ExportWedgeLoop(
 func (r *RetentionReaper) ExportSetInterval(d time.Duration) {
 	r.interval = d
 }
+
+// DummyPasswordHashForTest exposes the encoded hash that unknown
+// usernames are verified against.
+func DummyPasswordHashForTest() string {
+	return dummyPasswordHash()
+}

internal/database/model_apikey.go

@@ -2,12 +2,16 @@ package database
 
 import "time"
 
-// APIKey represents an API key for a user
+// APIKey represents an API key for a user.
+//
+// Key is a bearer credential, so it is never marshalled with the
+// model. A creation handler that has to show it once returns it in its
+// own response type.
 type APIKey struct {
 	BaseModel
 
 	UserID      string     `gorm:"type:uuid;not null"   json:"userId"`
-	Key         string     `gorm:"uniqueIndex;not null" json:"key"`
+	Key         string     `gorm:"uniqueIndex;not null" json:"-"`
 	Description string     `json:"description"`
 	LastUsedAt  *time.Time `json:"lastUsedAt,omitempty"`
 

internal/database/model_secrets_test.go

@@ -0,0 +1,107 @@
+package database_test
+
+import (
+	"encoding/json"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/webhooker/internal/database"
+)
+
+// keptField is a non-secret value planted alongside each secret, so
+// the assertions below cannot pass by the model marshalling to nothing.
+const keptField = "keepme"
+
+// marshalModel encodes a model the way a future JSON handler would.
+func marshalModel(t *testing.T, v any) string {
+	t.Helper()
+
+	encoded, err := json.Marshal(v)
+	require.NoError(t, err)
+
+	return string(encoded)
+}
+
+// TestModelsDoNotMarshalTheirSecrets pins the barrier for the JSON
+// path. The /api/v1 route group exists and is empty; delivery's
+// TargetView masks the credential for the HTML path only, so without
+// these tags the first handler that marshals a model serialises the
+// secret with it. Each field below is a live credential:
+//
+//   - Target.Config holds an incoming-webhook URL whose path segments
+//     are the bearer token.
+//   - APIKey.Key is a bearer token outright.
+//   - Setting.Value holds the session encryption key.
+//   - User.Password holds the Argon2 hash, and was already tagged.
+func TestModelsDoNotMarshalTheirSecrets(t *testing.T) {
+	t.Parallel()
+
+	const marker = "QQMODELMARKERQQ"
+
+	cases := []struct {
+		name  string
+		model any
+	}{
+		{
+			name: "target config",
+			model: database.Target{
+				Name:   keptField,
+				Type:   database.TargetTypeSlack,
+				Config: `{"webhookUrl":"https://h/s/` + marker + `"}`,
+			},
+		},
+		{
+			name: "api key",
+			model: database.APIKey{
+				Description: keptField,
+				Key:         marker,
+			},
+		},
+		{
+			name: "setting value",
+			model: database.Setting{
+				Key:   keptField,
+				Value: marker,
+			},
+		},
+		{
+			name: "user password hash",
+			model: database.User{
+				Username: keptField,
+				Password: marker,
+			},
+		},
+	}
+
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			encoded := marshalModel(t, tc.model)
+
+			assert.NotContains(t, encoded, marker)
+			assert.Contains(t, encoded, keptField)
+		})
+	}
+}
+
+// TestWebhookMarshalsNoTargetConfig covers the nested case: a webhook
+// marshalled with its targets preloaded must not carry the credential
+// through the association either.
+func TestWebhookMarshalsNoTargetConfig(t *testing.T) {
+	t.Parallel()
+
+	const marker = "QQNESTEDMARKERQQ"
+
+	encoded := marshalModel(t, database.Webhook{
+		Name: keptField,
+		Targets: []database.Target{{
+			Name:   "slack",
+			Config: `{"webhookUrl":"https://h/s/` + marker + `"}`,
+		}},
+	})
+
+	assert.NotContains(t, encoded, marker)
+	assert.Contains(t, encoded, keptField)
+}

internal/database/model_setting.go

@@ -3,6 +3,9 @@ package database
 // Setting stores application-level key-value configuration.
 // Used for auto-generated values like the session encryption key.
 type Setting struct {
-	Key   string `gorm:"primaryKey"         json:"key"`
-	Value string `gorm:"type:text;not null" json:"value"`
+	Key string `gorm:"primaryKey" json:"key"`
+
+	// Value holds the session encryption key, so it is never
+	// marshalled with the model.
+	Value string `gorm:"type:text;not null" json:"-"`
 }

internal/database/model_target.go

@@ -20,8 +20,14 @@ type Target struct {
 	Type      TargetType `gorm:"not null"           json:"type"`
 	Active    bool       `gorm:"default:true"       json:"active"`
 
-	// Configuration fields (JSON stored based on type)
-	Config string `gorm:"type:text" json:"config"` // JSON configuration
+	// Configuration fields (JSON stored based on type).
+	//
+	// json:"-" because the blob holds the target's credential — a
+	// Slack incoming-webhook URL, or an http destination whose path
+	// segments are the secret. delivery.TargetView is the masking
+	// barrier for the HTML path; this tag is the barrier for any
+	// handler that marshals the model itself.
+	Config string `gorm:"type:text" json:"-"` // JSON configuration
 
 	// For HTTP targets (max_retries=0 means fire-and-forget,
 	// >0 enables retries with backoff)

internal/database/password.go

@@ -8,6 +8,7 @@ import (
 	"fmt"
 	"math/big"
 	"strings"
+	"sync"
 
 	"golang.org/x/crypto/argon2"
 )
@@ -29,6 +30,10 @@ const hashParts = 6
 // triggers per-character-class complexity enforcement.
 const minPasswordComplexityLen = 4
 
+// dummyPasswordLen is the length of the throwaway password behind
+// dummyPasswordHash.
+const dummyPasswordLen = 32
+
 // Sentinel errors returned by decodeHash.
 var (
 	errInvalidHashFormat    = errors.New("invalid hash format")
@@ -122,6 +127,38 @@ func VerifyPassword(
 	return subtle.ConstantTimeCompare(hash, otherHash) == 1, nil
 }
 
+// dummyPasswordHash is an encoded Argon2id hash of a random
+// password, computed once on first use. Nothing can match it: the
+// password it encodes is discarded as soon as it is hashed. It is
+// process-wide because building it per request would add a second
+// 64 MB Argon2id pass to every login for an unknown username.
+//
+//nolint:gochecknoglobals // computed once, see above
+var dummyPasswordHash = sync.OnceValue(func() string {
+	password, err := GenerateRandomPassword(dummyPasswordLen)
+	if err != nil {
+		panic(fmt.Sprintf("generating the dummy password: %v", err))
+	}
+
+	hash, err := HashPassword(password)
+	if err != nil {
+		panic(fmt.Sprintf("hashing the dummy password: %v", err))
+	}
+
+	return hash
+})
+
+// VerifyDummyPassword performs a credential verification that cannot
+// succeed, at the same cost as a real one.
+//
+// Login must charge an unknown username the same work as a known
+// one. Returning early for an account that does not exist answers in
+// microseconds where a real account takes tens of milliseconds, which
+// is a username oracle any client can read off the response time.
+func VerifyDummyPassword(password string) {
+	_, _ = VerifyPassword(password, dummyPasswordHash())
+}
+
 // decodeHash extracts parameters, salt, and hash from an
 // encoded hash string.
 func decodeHash(

internal/database/password_test.go

@@ -191,3 +191,41 @@ func TestHashPasswordUniqueness(t *testing.T) {
 		)
 	}
 }
+
+// TestVerifyDummyPassword_DoesRealWork covers the anti-enumeration
+// path. Login charges an unknown username a verification against a
+// dummy hash so that a nonexistent account is not answered in
+// microseconds where a real one takes tens of milliseconds. That only
+// works if the dummy hash is a real, decodable Argon2id hash: a
+// malformed one would make VerifyPassword fail on the decode and
+// return before hashing anything.
+func TestVerifyDummyPassword_DoesRealWork(t *testing.T) {
+	t.Parallel()
+
+	// Runs the OnceValue that builds the dummy hash, so a panic in
+	// it surfaces here rather than on a live login.
+	database.VerifyDummyPassword("whatever was submitted")
+
+	dummy := database.DummyPasswordHashForTest()
+
+	// A hash the verifier cannot decode would make VerifyPassword
+	// return on the decode error, before hashing anything — the
+	// timing oracle this path exists to close.
+	valid, err := database.VerifyPassword("whatever", dummy)
+	if err != nil {
+		t.Fatalf(
+			"the dummy hash must decode like a real one: %v", err,
+		)
+	}
+
+	if valid {
+		t.Error("nothing may authenticate against the dummy hash")
+	}
+
+	if !strings.HasPrefix(dummy, "$argon2id$") {
+		t.Errorf(
+			"the dummy hash must use the same algorithm as real "+
+				"hashes, got %q", dummy,
+		)
+	}
+}

internal/database/webhook_db_manager.go

@@ -14,6 +14,7 @@ import (
 	"gorm.io/driver/sqlite"
 	"gorm.io/gorm"
 	"sneak.berlin/go/webhooker/internal/config"
+	"sneak.berlin/go/webhooker/internal/gormlog"
 	"sneak.berlin/go/webhooker/internal/logger"
 )
 
@@ -248,7 +249,10 @@ func (m *WebhookDBManager) openDB(
 
 	db, err := gorm.Open(sqlite.Dialector{
 		Conn: sqlDB,
-	}, &gorm.Config{})
+	}, &gorm.Config{
+		// Never leave this at GORM's default. See internal/gormlog.
+		Logger: gormlog.New(m.log),
+	})
 	if err != nil {
 		_ = sqlDB.Close()
 

internal/database/webhook_db_manager_test.go

@@ -339,7 +339,12 @@ func TestWebhookDBManager_MultipleWebhooks(t *testing.T) {
 	var events []database.Event
 
 	require.NoError(t, db2.Find(&events).Error)
-	assert.Len(t, events, 1)
+
+	// require, not assert: this is exactly the regression the test
+	// guards, so the empty slice is the expected failure, and a
+	// non-fatal length check would index into it on the next line and
+	// panic the whole package test binary instead of failing here.
+	require.Len(t, events, 1)
 	assert.Equal(t, "PUT", events[0].Method)
 }
 

internal/delivery/target_database_archive.go

@@ -12,6 +12,7 @@ import (
 
 	"gorm.io/driver/sqlite"
 	"gorm.io/gorm"
+	"sneak.berlin/go/webhooker/internal/gormlog"
 )
 
 // archiveExpiryNever is the expiry sentinel (and default) that
@@ -282,7 +283,11 @@ func (w *archiveWriter) openMode(
 	}
 
 	gdb, err := gorm.Open(
-		sqlite.Dialector{Conn: sqlDB}, &gorm.Config{},
+		sqlite.Dialector{Conn: sqlDB}, &gorm.Config{
+			// Never leave this at GORM's default. See
+			// internal/gormlog.
+			Logger: gormlog.New(w.log),
+		},
 	)
 	if err != nil {
 		_ = sqlDB.Close()

internal/delivery/target_database_archive_gormlog_test.go

@@ -0,0 +1,155 @@
+package delivery_test
+
+import (
+	"bytes"
+	"log"
+	"log/slog"
+	"path/filepath"
+	"strings"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"gorm.io/gorm"
+	gormlogger "gorm.io/gorm/logger"
+	"sneak.berlin/go/webhooker/internal/delivery"
+	"sneak.berlin/go/webhooker/internal/middleware"
+)
+
+// archiveGORMTailMarker sits at the far end of the value this file
+// drives into an archive lookup. Its presence in a log line means the
+// whole value reached the log, so nothing truncated it.
+const archiveGORMTailMarker = "ENDOFCLIENTVALUE"
+
+// archiveGORMFillBytes is how much text the lookup carries. It is far
+// past every budget in play.
+const archiveGORMFillBytes = 8 << 10
+
+// gormDefaultBuf collects what GORM's package-level default logger
+// writes, if anything reaches it.
+type gormDefaultBuf struct {
+	mu sync.Mutex
+	b  bytes.Buffer
+}
+
+func (g *gormDefaultBuf) Write(p []byte) (int, error) {
+	g.mu.Lock()
+	defer g.mu.Unlock()
+
+	return g.b.Write(p)
+}
+
+func (g *gormDefaultBuf) String() string {
+	g.mu.Lock()
+	defer g.mu.Unlock()
+
+	return g.b.String()
+}
+
+// captureArchiveGORMDefault replaces GORM's package-level default
+// logger with one configured exactly as GORM configures its own,
+// writing to a buffer.
+//
+// This duplicates the detector in internal/handlers rather than
+// sharing it: a test helper cannot cross a package's test boundary
+// without exporting production code to carry it, and a logging
+// detector is not worth a production symbol. What it detects is the
+// third gorm.Open in this service, at
+// internal/delivery/target_database_archive.go — the archive writer,
+// whose type is unexported, so nothing outside this package can drive
+// it.
+func captureArchiveGORMDefault(t *testing.T) *gormDefaultBuf {
+	t.Helper()
+
+	buf := &gormDefaultBuf{}
+	orig := gormlogger.Default
+
+	gormlogger.Default = gormlogger.New(
+		log.New(buf, "", log.LstdFlags),
+		gormlogger.Config{
+			SlowThreshold:             200 * time.Millisecond,
+			LogLevel:                  gormlogger.Warn,
+			IgnoreRecordNotFoundError: false,
+			Colorful:                  false,
+		},
+	)
+
+	t.Cleanup(func() { gormlogger.Default = orig })
+
+	return buf
+}
+
+// TestArchiveWriter_NeverUsesGORMsDefaultLogger pins the archive
+// writer's gorm.Open to the adapter.
+//
+// Restore a bare &gorm.Config{} at
+// internal/delivery/target_database_archive.go and this fails: the
+// default logger prints the fully interpolated SELECT on every
+// ErrRecordNotFound, so the client-chosen event id below arrives whole
+// and unbounded on stdout, answering to no level the operator set.
+//
+// Not parallel: gormlogger.Default is process-global. Go runs every
+// non-parallel top-level test to completion before it resumes the
+// parallel ones.
+//
+//nolint:paralleltest // Deliberately sequential; see above.
+func TestArchiveWriter_NeverUsesGORMsDefaultLogger(t *testing.T) {
+	var captured bytes.Buffer
+
+	gormDefault := captureArchiveGORMDefault(t)
+
+	w := delivery.NewExportArchiveWriter(
+		filepath.Join(t.TempDir(), "archive.db"),
+		slog.New(slog.NewTextHandler(
+			&captured, &slog.HandlerOptions{Level: slog.LevelDebug},
+		)),
+		0,
+	)
+
+	require.NoError(t, w.Open(0))
+
+	t.Cleanup(w.Evict)
+
+	// A lookup that misses, carrying a value the size of an inbound
+	// event id. Under the default logger this is the line that gets
+	// interpolated and printed.
+	value := strings.Repeat("\x01", archiveGORMFillBytes) +
+		archiveGORMTailMarker
+
+	var row delivery.ExportArchivedEvent
+
+	err := w.DB().Where("event_id = ?", value).First(&row).Error
+	require.ErrorIs(t, err, gorm.ErrRecordNotFound)
+
+	got := gormDefault.String()
+	assert.Empty(
+		t, got,
+		"GORM's default logger wrote %d bytes, so the archive "+
+			"writer's gorm.Open is back on a bare &gorm.Config{}; "+
+			"the first of them: %s",
+		len(got), got[:min(len(got), 300)],
+	)
+
+	// The adapter drops a miss, so this should be silent too — and
+	// whatever it does write stays inside the stated ceiling.
+	out := captured.String()
+
+	assert.NotContains(
+		t, out, archiveGORMTailMarker,
+		"the far end of the client-chosen value reached the log",
+	)
+
+	for line := range strings.SplitSeq(strings.TrimRight(out, "\n"), "\n") {
+		if line == "" {
+			continue
+		}
+
+		assert.LessOrEqual(
+			t, len(line), middleware.MaxAccessLogLineBytes,
+			"log line exceeded its bound: %s",
+			line[:min(len(line), 300)],
+		)
+	}
+}

internal/delivery/target_log.go

@@ -11,6 +11,17 @@ import (
 // inbound webhook — the full request body and headers, plus
 // the method, content type, and the webhook and entrypoint
 // ids — then records a single successful attempt.
+//
+// This is the one log call in the service that deliberately writes
+// unbounded client-chosen bytes, so it is the one exception to the
+// per-field budgets in internal/logfield and to the ceiling stated on
+// middleware.MaxAccessLogLineBytes. Capping here would defeat the
+// target: emitting the payload IS the delivery. It costs nothing by
+// default — an authenticated operator has to create a target of this
+// type on a specific webhook before a single line is written — and the
+// bytes it writes are bounded per event by maxWebhookBodySize (1 MB).
+// An operator who adds one is choosing to spend log volume on the
+// payloads that webhook receives.
 type logTarget struct {
 	eng *Engine
 }

internal/gormlog/export_test.go

@@ -0,0 +1,17 @@
+package gormlog
+
+import (
+	"log/slog"
+	"time"
+)
+
+// ExportNewWithSlowThreshold builds a Logger whose slow-statement
+// threshold is d rather than DefaultSlowThreshold, so a test can pin
+// which arm of Trace it is exercising instead of racing the clock on a
+// loaded machine. The threshold is set at construction, like every
+// other field, so the type's concurrency guarantee still holds.
+func ExportNewWithSlowThreshold(
+	log *slog.Logger, d time.Duration,
+) *Logger {
+	return &Logger{log: log, slowThreshold: d}
+}

internal/gormlog/gormlog.go

@@ -0,0 +1,168 @@
+// Package gormlog adapts GORM's logger onto the service's slog
+// logger.
+//
+// GORM's own default logger is not usable here. It is built at package
+// init with log.New(os.Stdout, ...) at LogLevel Warn with
+// IgnoreRecordNotFoundError false, so it writes the fully interpolated
+// SQL — parameters and all — for every statement that returns an
+// error, including gorm.ErrRecordNotFound. Two of this service's
+// lookups miss by design on unauthenticated routes: the entrypoint
+// lookup on /webhook/{uuid}, whose path segment the client picks
+// outright, and the user lookup behind the login form, whose username
+// the client picks outright. Under the default logger each of those
+// misses printed an unbounded, attacker-chosen string, at no level the
+// operator can turn down, past every handler internal/logger installs.
+//
+// This adapter fixes all three properties at once: the lines get a
+// level the operator controls, they are shaped by whichever handler
+// internal/logger selected, and every value a client can influence is
+// spent through logfield.Truncate.
+package gormlog
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"log/slog"
+	"time"
+
+	gormlogger "gorm.io/gorm/logger"
+	"sneak.berlin/go/webhooker/internal/logfield"
+)
+
+// DefaultSlowThreshold is the duration at or above which a statement
+// is logged as slow. It is GORM's own default, kept deliberately: slow
+// SQL is the one thing GORM's logger reports that nothing else in this
+// service does, so silencing the logger outright would have cost real
+// observability to fix a log-volume defect.
+const DefaultSlowThreshold = 200 * time.Millisecond
+
+// Logger implements gormlogger.Interface on top of an *slog.Logger.
+//
+// It is safe for concurrent use: every field is set at construction
+// and never written again.
+type Logger struct {
+	log           *slog.Logger
+	slowThreshold time.Duration
+}
+
+// Interface compliance is asserted here rather than discovered at the
+// gorm.Open call sites.
+var _ gormlogger.Interface = (*Logger)(nil)
+
+// New returns a GORM logger that writes through log.
+func New(log *slog.Logger) *Logger {
+	return &Logger{
+		log:           log,
+		slowThreshold: DefaultSlowThreshold,
+	}
+}
+
+// LogMode returns the logger unchanged.
+//
+// GORM's LogLevel is deliberately not honoured. Level is the operator's
+// decision and it is expressed once, through LOG_LEVEL and the
+// slog.LevelVar internal/logger holds; a second level knob inside the
+// database layer could only disagree with it. The mapping from GORM's
+// four categories onto slog levels is fixed in Trace below.
+//
+//nolint:ireturn // The interface return is GORM's signature, not a choice.
+func (l *Logger) LogMode(gormlogger.LogLevel) gormlogger.Interface {
+	return l
+}
+
+// Info logs one of GORM's own informational messages.
+func (l *Logger) Info(
+	ctx context.Context, msg string, data ...any,
+) {
+	l.log.InfoContext(ctx, "gorm", "message", format(msg, data...))
+}
+
+// Warn logs one of GORM's own warnings.
+func (l *Logger) Warn(
+	ctx context.Context, msg string, data ...any,
+) {
+	l.log.WarnContext(ctx, "gorm", "message", format(msg, data...))
+}
+
+// Error logs one of GORM's own errors.
+func (l *Logger) Error(
+	ctx context.Context, msg string, data ...any,
+) {
+	l.log.ErrorContext(ctx, "gorm", "message", format(msg, data...))
+}
+
+// Trace reports the outcome of a single statement. GORM calls it for
+// every statement it runs, so the cheap paths stay cheap: fc()
+// renders the interpolated SQL and is called only on a branch that
+// will actually emit.
+//
+// The arms are ordered exactly as GORM's own Trace orders them —
+// non-record-not-found error, then slow, then the routine case — so
+// that a statement which both misses and runs slow is still reported
+// as slow. A miss is the likeliest statement to be slow, since it is
+// the one that scans without finding a row, and ordering the drop
+// ahead of the slow arm would have made this adapter less observant
+// than the IgnoreRecordNotFoundError option it was chosen over.
+func (l *Logger) Trace(
+	ctx context.Context,
+	begin time.Time,
+	fc func() (string, int64),
+	err error,
+) {
+	elapsed := time.Since(begin)
+
+	switch {
+	case err != nil && !errors.Is(err, gormlogger.ErrRecordNotFound):
+		sql, rows := fc()
+		l.log.ErrorContext(ctx, "sql statement failed",
+			"error", logfield.Truncate(err.Error(), logfield.MaxBytes),
+			"sql", logfield.Truncate(sql, logfield.MaxBytes),
+			"rows", rows,
+			"elapsed_ms", elapsed.Milliseconds(),
+		)
+
+	case l.slowThreshold > 0 && elapsed >= l.slowThreshold:
+		sql, rows := fc()
+		l.log.WarnContext(ctx, "slow sql statement",
+			"sql", logfield.Truncate(sql, logfield.MaxBytes),
+			"rows", rows,
+			"elapsed_ms", elapsed.Milliseconds(),
+			"threshold_ms", l.slowThreshold.Milliseconds(),
+		)
+
+	case err != nil:
+		// gorm.ErrRecordNotFound is not an error on the paths that
+		// produce it here: an invented entrypoint UUID and an unknown
+		// username are the expected outcome of an unauthenticated
+		// request, not a fault. This is the IgnoreRecordNotFoundError
+		// behaviour, and it is unconditional rather than configurable
+		// because no caller in this service wants the other one — the
+		// two handlers that care already record the miss themselves,
+		// at DEBUG, without the SQL. A miss that ran slow has already
+		// been reported by the arm above.
+		return
+
+	case l.log.Enabled(ctx, slog.LevelDebug):
+		sql, rows := fc()
+		l.log.DebugContext(ctx, "sql statement",
+			"sql", logfield.Truncate(sql, logfield.MaxBytes),
+			"rows", rows,
+			"elapsed_ms", elapsed.Milliseconds(),
+		)
+	}
+}
+
+// format renders one of GORM's printf-style internal messages and
+// bounds it. GORM builds these itself, but they can quote a value the
+// statement carried, so they are spent through the same budget as
+// everything else rather than trusted.
+func format(msg string, data ...any) string {
+	if len(data) == 0 {
+		return logfield.Truncate(msg, logfield.MaxBytes)
+	}
+
+	return logfield.Truncate(
+		fmt.Sprintf(msg, data...), logfield.MaxBytes,
+	)
+}

internal/gormlog/gormlog_test.go

@@ -0,0 +1,436 @@
+package gormlog_test
+
+import (
+	"bytes"
+	"context"
+	"database/sql"
+	"fmt"
+	"log/slog"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"gorm.io/driver/sqlite"
+	"gorm.io/gorm"
+	_ "modernc.org/sqlite" // Pure Go SQLite driver.
+	"sneak.berlin/go/webhooker/internal/gormlog"
+	"sneak.berlin/go/webhooker/internal/middleware"
+)
+
+// fillBytes is how much client-chosen text each case drives into the
+// statement. It is well past every budget in play, so a value that
+// arrives short arrived short because something cut it.
+const fillBytes = 8 << 10
+
+// tailMarker sits at the far end of every generated value. A line that
+// contains it carried the whole value, which means nothing cut it — so
+// a value that merely happened to be short cannot pass for a truncated
+// one.
+const tailMarker = "ENDOFCLIENTVALUE"
+
+// fills are the characters a client can drive into a SQL parameter,
+// chosen for what the log handlers charge for them rather than for
+// looking dangerous.
+//
+// The C0 control is the one that matters. Both handlers spell U+0001
+// as a six-byte escape for the single byte it costs a client to send,
+// which is the widest multiplier available in the basic multilingual
+// plane and the case a raw-byte budget breaks on first. The astral
+// non-printable costs ten under the text handler, four more than the
+// JSON handler ever spends.
+func fills() []struct {
+	name string
+	fill string
+} {
+	return []struct {
+		name string
+		fill string
+	}{
+		{"plain", "x"},
+		{"quote", `"`},
+		{"backslash", `\`},
+		{"tab", "\t"},
+		{"newline", "\n"},
+		{"c0_control", "\x01"},
+		{"astral_nonprintable", "\U0001000C"},
+	}
+}
+
+// clientValue builds a value of at least fillBytes raw bytes out of
+// fill, ending in tailMarker.
+func clientValue(fill string) string {
+	var b strings.Builder
+
+	for b.Len() < fillBytes {
+		b.WriteString(fill)
+	}
+
+	b.WriteString(tailMarker)
+
+	return b.String()
+}
+
+// handlers are the two slog handlers internal/logger can install. The
+// ceiling is quoted to operators unqualified, so every case is
+// asserted under both.
+func handlers() []struct {
+	name string
+	make func(*bytes.Buffer) slog.Handler
+} {
+	opts := &slog.HandlerOptions{Level: slog.LevelDebug}
+
+	return []struct {
+		name string
+		make func(*bytes.Buffer) slog.Handler
+	}{
+		{"json", func(b *bytes.Buffer) slog.Handler {
+			return slog.NewJSONHandler(b, opts)
+		}},
+		{"text", func(b *bytes.Buffer) slog.Handler {
+			return slog.NewTextHandler(b, opts)
+		}},
+	}
+}
+
+type thing struct {
+	ID   string `gorm:"primaryKey"`
+	Name string
+}
+
+// neverSlow is a slow-statement threshold no statement in this file
+// can reach. Cases that are about a non-slow arm of Trace set it, so
+// that a machine under load cannot turn a miss into a slow report and
+// decide the outcome for them.
+const neverSlow = time.Hour
+
+// alwaysSlow makes every statement count as slow, so the slow arm is
+// reached without the test waiting for it.
+const alwaysSlow = time.Nanosecond
+
+// openDB opens a real SQLite database behind the adapter under test,
+// so every assertion below is made against SQL that GORM actually
+// rendered rather than against a string a test wrote by hand. slow is
+// the adapter's slow-statement threshold.
+func openDB(
+	t *testing.T, buf *bytes.Buffer, h slog.Handler, slow time.Duration,
+) *gorm.DB {
+	t.Helper()
+
+	sqlDB, err := sql.Open("sqlite", fmt.Sprintf(
+		"file:%s?mode=rwc",
+		filepath.Join(t.TempDir(), "gormlog.db"),
+	))
+	require.NoError(t, err)
+
+	t.Cleanup(func() { _ = sqlDB.Close() })
+
+	gl := gormlog.ExportNewWithSlowThreshold(slog.New(h), slow)
+
+	gdb, err := gorm.Open(
+		sqlite.Dialector{Conn: sqlDB},
+		&gorm.Config{Logger: gl},
+	)
+	require.NoError(t, err)
+
+	require.NoError(t, gdb.AutoMigrate(&thing{}))
+
+	// Migration chatter is not what any of these cases is about.
+	buf.Reset()
+
+	return gdb
+}
+
+// assertBounded holds every line the adapter wrote to the stated
+// ceiling and proves each was cut rather than merely short.
+func assertBounded(t *testing.T, out string) {
+	t.Helper()
+
+	assert.NotContains(
+		t, out, tailMarker,
+		"the far end of the client value reached the log, so "+
+			"nothing truncated it",
+	)
+
+	for line := range strings.SplitSeq(
+		strings.TrimRight(out, "\n"), "\n",
+	) {
+		if line == "" {
+			continue
+		}
+
+		assert.LessOrEqual(
+			t, len(line), middleware.MaxAccessLogLineBytes,
+			"log line exceeded its bound: %s",
+			line[:min(len(line), 300)],
+		)
+	}
+}
+
+// TestRecordNotFound_WritesNothing is the defect itself. GORM's own
+// default logger prints the fully interpolated SELECT on every
+// ErrRecordNotFound, and on this service's two unauthenticated
+// lookups the interpolated parameter is whatever the client sent.
+func TestRecordNotFound_WritesNothing(t *testing.T) {
+	t.Parallel()
+
+	for _, h := range handlers() {
+		for _, f := range fills() {
+			t.Run(h.name+"/"+f.name, func(t *testing.T) {
+				t.Parallel()
+
+				var buf bytes.Buffer
+
+				gdb := openDB(t, &buf, h.make(&buf), neverSlow)
+
+				var got thing
+
+				err := gdb.Where(
+					"id = ?", clientValue(f.fill),
+				).First(&got).Error
+				require.ErrorIs(t, err, gorm.ErrRecordNotFound)
+
+				assert.Empty(
+					t, buf.String(),
+					"a miss on a client-chosen key must not "+
+						"write a log line",
+				)
+			})
+		}
+	}
+}
+
+// TestSlowRecordNotFound_IsStillReportedSlow pins the arm ordering in
+// Trace against the drop above.
+//
+// GORM's own Trace orders its cases error-that-is-not-a-miss, then
+// slow, then routine, so IgnoreRecordNotFoundError: true — the cheap
+// option this adapter was chosen over — still reports a miss that ran
+// slow. An adapter that dropped the miss first would be strictly less
+// observant than the option it replaced, on exactly the two lookups
+// this package exists for. A miss is also the statement most likely to
+// be slow, since it is the one that scans without finding a row.
+func TestSlowRecordNotFound_IsStillReportedSlow(t *testing.T) {
+	t.Parallel()
+
+	for _, h := range handlers() {
+		for _, f := range fills() {
+			t.Run(h.name+"/"+f.name, func(t *testing.T) {
+				t.Parallel()
+
+				var buf bytes.Buffer
+
+				gdb := openDB(t, &buf, h.make(&buf), alwaysSlow)
+
+				var got thing
+
+				err := gdb.Where(
+					"id = ?", clientValue(f.fill),
+				).First(&got).Error
+				require.ErrorIs(t, err, gorm.ErrRecordNotFound)
+
+				assert.Contains(
+					t, buf.String(), "slow sql statement",
+					"a slow statement that missed was not "+
+						"reported as slow",
+				)
+				assertBounded(t, buf.String())
+			})
+		}
+	}
+}
+
+// TestRecordNotFoundFlood_DoesNotGrowWithInput states the definition
+// of done directly: a flood of misses at two input sizes 64 times
+// apart must cost the same number of bytes of log.
+func TestRecordNotFoundFlood_DoesNotGrowWithInput(t *testing.T) {
+	t.Parallel()
+
+	const requests = 50
+
+	flood := func(t *testing.T, size int) int {
+		t.Helper()
+
+		var buf bytes.Buffer
+
+		gdb := openDB(
+			t, &buf,
+			slog.NewJSONHandler(&buf, &slog.HandlerOptions{
+				Level: slog.LevelDebug,
+			}),
+			neverSlow,
+		)
+
+		value := strings.Repeat("\x01", size)
+
+		for range requests {
+			var got thing
+
+			_ = gdb.Where("id = ?", value).First(&got).Error
+		}
+
+		return buf.Len()
+	}
+
+	small := flood(t, 128)
+	big := flood(t, 128*64)
+
+	assert.Equal(
+		t, small, big,
+		"log volume tracked the size of the client's input",
+	)
+}
+
+// TestStatementError_LineIsBounded covers the branch that does log.
+// A driver error is not ErrRecordNotFound, so the interpolated
+// statement is written — and on an insert the interpolated value is
+// still whatever the client supplied.
+func TestStatementError_LineIsBounded(t *testing.T) {
+	t.Parallel()
+
+	for _, h := range handlers() {
+		for _, f := range fills() {
+			t.Run(h.name+"/"+f.name, func(t *testing.T) {
+				t.Parallel()
+
+				var buf bytes.Buffer
+
+				gdb := openDB(t, &buf, h.make(&buf), neverSlow)
+
+				row := thing{ID: clientValue(f.fill), Name: "a"}
+
+				require.NoError(t, gdb.Create(&row).Error)
+
+				buf.Reset()
+
+				// The same primary key a second time: a UNIQUE
+				// constraint failure, which is an error GORM logs.
+				err := gdb.Create(&thing{
+					ID: row.ID, Name: "b",
+				}).Error
+				require.Error(t, err)
+
+				assert.Contains(
+					t, buf.String(), "sql statement failed",
+				)
+				assertBounded(t, buf.String())
+			})
+		}
+	}
+}
+
+// TestSucceedingStatement_LineIsBoundedOnEitherArm covers the two
+// arms a statement that returns no error can take, over the same
+// query, so neither can be bounded by accident of the other.
+//
+//   - slow. Silencing GORM outright would have been the cheaper fix
+//     and would have cost this report, which is the one thing GORM's
+//     logger gave an operator that nothing else in this service does.
+//   - routine. The branch an operator reaches by turning the level
+//     down to DEBUG: every statement is reported, so every
+//     statement's interpolated parameters have to be bounded too.
+func TestSucceedingStatement_LineIsBoundedOnEitherArm(t *testing.T) {
+	t.Parallel()
+
+	// "sql statement" is a substring of "slow sql statement", so the
+	// routine arm carries notWant as well: Contains alone cannot tell
+	// the two arms apart in that direction.
+	arms := []struct {
+		name    string
+		slow    time.Duration
+		want    string
+		notWant string
+	}{
+		{"slow", alwaysSlow, "slow sql statement", ""},
+		{"routine", neverSlow, "sql statement", "slow sql statement"},
+	}
+
+	for _, a := range arms {
+		for _, h := range handlers() {
+			for _, f := range fills() {
+				name := a.name + "/" + h.name + "/" + f.name
+
+				t.Run(name, func(t *testing.T) {
+					t.Parallel()
+
+					var buf bytes.Buffer
+
+					gdb := openDB(t, &buf, h.make(&buf), a.slow)
+
+					var got []thing
+
+					require.NoError(t, gdb.Where(
+						"name = ?", clientValue(f.fill),
+					).Find(&got).Error)
+
+					assert.Contains(t, buf.String(), a.want)
+
+					if a.notWant != "" {
+						assert.NotContains(
+							t, buf.String(), a.notWant,
+						)
+					}
+
+					assertBounded(t, buf.String())
+				})
+			}
+		}
+	}
+}
+
+// TestGORMOwnMessages_AreBounded covers the three printf-style
+// entry points. GORM builds these itself, but nothing stops one of
+// them quoting a value the statement carried.
+func TestGORMOwnMessages_AreBounded(t *testing.T) {
+	t.Parallel()
+
+	for _, h := range handlers() {
+		for _, f := range fills() {
+			t.Run(h.name+"/"+f.name, func(t *testing.T) {
+				t.Parallel()
+
+				var buf bytes.Buffer
+
+				gl := gormlog.New(slog.New(h.make(&buf)))
+				ctx := context.Background()
+				value := clientValue(f.fill)
+
+				gl.Info(ctx, "%s", value)
+				gl.Warn(ctx, "%s", value)
+				gl.Error(ctx, "%s", value)
+
+				// The no-argument form, which is how GORM reports
+				// most of its own conditions. Reached through a
+				// function value so the vet printf check does not
+				// read the message as a format string — which is
+				// also why the adapter does not.
+				noArgs := func(
+					f func(context.Context, string, ...any),
+					msg string,
+				) {
+					f(ctx, msg)
+				}
+				noArgs(gl.Info, value)
+
+				assertBounded(t, buf.String())
+			})
+		}
+	}
+}
+
+// TestLogMode_KeepsTheOperatorsLevel records that GORM's own level
+// knob is deliberately inert: level belongs to LOG_LEVEL, and a
+// second one inside the database layer could only disagree with it.
+func TestLogMode_KeepsTheOperatorsLevel(t *testing.T) {
+	t.Parallel()
+
+	var buf bytes.Buffer
+
+	gl := gormlog.New(slog.New(slog.NewJSONHandler(
+		&buf, &slog.HandlerOptions{Level: slog.LevelDebug},
+	)))
+
+	assert.Same(t, gl, gl.LogMode(0))
+}

internal/handlers/auth.go

@@ -2,8 +2,10 @@ package handlers
 
 import (
 	"net/http"
+	"strconv"
 
 	"sneak.berlin/go/webhooker/internal/database"
+	"sneak.berlin/go/webhooker/internal/logfield"
 )
 
 // HandleLoginPage returns a handler for the login page (GET)
@@ -39,8 +41,10 @@ func (h *Handlers) HandleLoginSubmit() http.HandlerFunc {
 			return
 		}
 
-		username := r.FormValue("username")
-		password := r.FormValue("password")
+		// PostFormValue, not FormValue: the credential must come
+		// from the body, never from the query string.
+		username := r.PostFormValue("username")
+		password := r.PostFormValue("password")
 
 		// Validate input
 		if username == "" || password == "" {
@@ -67,7 +71,9 @@ func (h *Handlers) HandleLoginSubmit() http.HandlerFunc {
 
 		h.log.Info(
 			"user logged in",
-			"username", username,
+			"username", logfield.Truncate(
+				username, logfield.MaxBytes,
+			),
 			"user_id", user.ID,
 		)
 
@@ -93,6 +99,16 @@ func (h *Handlers) renderLoginError(
 
 // authenticateUser looks up and verifies a user's credentials.
 // On failure it writes an HTTP response and returns an error.
+//
+// The credential check runs BEFORE any rate-limit budget is
+// consulted, and only a failed check spends budget. That is what
+// keeps the single administrative path reachable: behind the reverse
+// proxy this deployment requires, with TRUSTED_PROXIES unset, every
+// client shares one bucket, so a limiter spent on arrival lets any
+// stranger deny the operator's own correct password indefinitely.
+//
+// Verifying first means every login POST costs an Argon2id hash, so
+// the work is taken under a bounded number of verification slots.
 func (h *Handlers) authenticateUser(
 	w http.ResponseWriter,
 	r *http.Request,
@@ -100,16 +116,49 @@ func (h *Handlers) authenticateUser(
 ) (database.User, error) {
 	var user database.User
 
+	release, ok := h.mw.BeginPasswordVerification(r.Context())
+	if !ok {
+		h.log.Warn(
+			"password verification capacity exhausted",
+			"path", logfield.Truncate(
+				r.URL.Path, logfield.MaxBytes,
+			),
+		)
+		h.renderLoginError(
+			w, r,
+			"The server is busy verifying credentials. "+
+				"Please try again.",
+			http.StatusServiceUnavailable,
+		)
+
+		return user, errVerificationBusy
+	}
+
+	defer release()
+
 	err := h.db.DB().Where(
 		"username = ?", username,
 	).First(&user).Error
 	if err != nil {
-		h.log.Debug("user not found", "username", username)
-		h.renderLoginError(
-			w, r,
-			"Invalid username or password",
-			http.StatusUnauthorized,
+		// A username that does not exist is charged the same work
+		// as one that does. Skipping the hash here would answer in
+		// microseconds where a real account takes tens of
+		// milliseconds, handing every client a username oracle.
+		h.dummyVerifications.Add(1)
+		database.VerifyDummyPassword(password)
+
+		// Login is unauthenticated, and the submitted username is
+		// a form field the client fills to any length the 1 MB
+		// body cap allows. On this branch it matched no row, so
+		// nothing else bounds it. The rate limiter caps how often
+		// the line is written, not how wide it is.
+		h.log.Debug(
+			"user not found",
+			"username", logfield.Truncate(
+				username, logfield.MaxBytes,
+			),
 		)
+		h.rejectLogin(w, r, username)
 
 		return user, err
 	}
@@ -126,17 +175,60 @@ func (h *Handlers) authenticateUser(
 	}
 
 	if !valid {
-		h.log.Debug("invalid password", "username", username)
+		// Reached only once the username matched a stored row, so
+		// it is bounded by the operator's own data. Capped anyway,
+		// so that every username this unauthenticated endpoint
+		// logs is capped and no reader has to work out which
+		// branch narrowed it.
+		h.log.Debug(
+			"invalid password",
+			"username", logfield.Truncate(
+				username, logfield.MaxBytes,
+			),
+		)
+		h.rejectLogin(w, r, username)
+
+		return user, errInvalidPassword
+	}
+
+	// The password was correct, so forgive whatever failures this
+	// client accumulated: an operator who mistypes a few times and
+	// then gets it right must not stay throttled afterwards.
+	h.mw.ForgiveLoginFailures(r, username)
+
+	return user, nil
+}
+
+// rejectLogin counts one failed credential verification and answers
+// it: 401 while this client still has failure budget against the
+// submitted username, 429 with a Retry-After once it is spent.
+//
+// The 429 throttles wrong passwords only. A correct one never
+// reaches here, so no amount of failure — from this client or any
+// other sharing its bucket — can keep the operator out.
+func (h *Handlers) rejectLogin(
+	w http.ResponseWriter,
+	r *http.Request,
+	username string,
+) {
+	if !h.mw.RecordLoginFailure(r, username) {
 		h.renderLoginError(
 			w, r,
 			"Invalid username or password",
 			http.StatusUnauthorized,
 		)
 
-		return user, errInvalidPassword
+		return
 	}
 
-	return user, nil
+	w.Header().Set("Retry-After", strconv.Itoa(int(
+		h.mw.LoginFailureInterval().Seconds(),
+	)))
+	h.renderLoginError(
+		w, r,
+		"Too many failed login attempts. Please try again later.",
+		http.StatusTooManyRequests,
+	)
 }
 
 // createAuthenticatedSession regenerates the session and stores

internal/handlers/auth_test.go

@@ -0,0 +1,455 @@
+package handlers_test
+
+import (
+	"context"
+	"fmt"
+	"net/http"
+	"net/http/httptest"
+	"net/url"
+	"strings"
+	"sync"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/webhooker/internal/database"
+	"sneak.berlin/go/webhooker/internal/handlers"
+	"sneak.berlin/go/webhooker/internal/session"
+)
+
+const (
+	// operatorUser and operatorPassword are the single admin account
+	// these tests defend.
+	operatorUser     = "admin"
+	operatorPassword = "correct horse battery staple"
+
+	// sharedProxyPeer is the whole point of this file. Production is
+	// required to run behind a TLS-terminating reverse proxy, and
+	// TRUSTED_PROXIES defaults to empty, so every client — attacker
+	// and operator alike — reaches the process from the proxy's
+	// address and shares one rate-limit bucket. Both parties in
+	// these tests therefore use the same RemoteAddr.
+	sharedProxyPeer = "10.0.0.1:44444"
+
+	// loginFailureLimit is the failure budget one client has against
+	// one submitted username. Restated here rather than imported
+	// from the middleware package, so that changing the production
+	// limit fails these tests instead of silently moving with them.
+	loginFailureLimit = 5
+)
+
+// seedOperator gives the bootstrapped admin account a password these
+// tests know. The account itself is created at startup with a random
+// password, which is exactly why its username is predictable to an
+// attacker and why keying failures by username alone does not fix
+// this issue.
+func seedOperator(t *testing.T, db *database.Database) {
+	t.Helper()
+
+	hash, err := database.HashPassword(operatorPassword)
+	require.NoError(t, err)
+
+	result := db.DB().Model(&database.User{}).
+		Where("username = ?", operatorUser).
+		Update("password", hash)
+
+	require.NoError(t, result.Error)
+	require.EqualValues(
+		t, 1, result.RowsAffected,
+		"the bootstrap admin account must exist",
+	)
+}
+
+// loginPost builds a login form POST arriving from peer.
+func loginPost(peer, username, password string) *http.Request {
+	form := url.Values{}
+	form.Set("username", username)
+	form.Set("password", password)
+
+	req := httptest.NewRequestWithContext(
+		context.Background(),
+		http.MethodPost,
+		"/pages/login",
+		strings.NewReader(form.Encode()),
+	)
+	req.Header.Set(
+		"Content-Type", "application/x-www-form-urlencoded",
+	)
+	req.RemoteAddr = peer
+
+	return req
+}
+
+// submitLogin drives one login POST through the handler.
+func submitLogin(
+	h *handlers.Handlers, peer, username, password string,
+) *httptest.ResponseRecorder {
+	w := httptest.NewRecorder()
+	h.HandleLoginSubmit().ServeHTTP(w, loginPost(
+		peer, username, password,
+	))
+
+	return w
+}
+
+// floodFailures sends attempts wrong-password logins for username
+// from peer, which is what an attacker does.
+func floodFailures(
+	t *testing.T,
+	h *handlers.Handlers,
+	peer, username string,
+	attempts int,
+) {
+	t.Helper()
+
+	for i := range attempts {
+		w := submitLogin(h, peer, username, fmt.Sprintf("guess-%d", i))
+		require.NotEqual(
+			t, http.StatusSeeOther, w.Code,
+			"attempt %d must not authenticate", i,
+		)
+	}
+}
+
+// TestLogin_StrangersFloodCannotLockOutTheOperator is the
+// done-criterion of https://git.eeqj.de/sneak/webhooker/issues/150.
+//
+// The attacker and the operator share one rate-limit bucket, because
+// behind the mandated reverse proxy with TRUSTED_PROXIES unset every
+// client keys on the proxy's address. The attacker floods the
+// operator's own username — a single-admin product has a predictable
+// one — far past the failure limit. The operator must still be able
+// to log in with the correct password.
+//
+// This fails if credentials stop being verified ahead of the limiter.
+func TestLogin_StrangersFloodCannotLockOutTheOperator(t *testing.T) {
+	t.Parallel()
+
+	var (
+		h  *handlers.Handlers
+		db *database.Database
+	)
+
+	app := newTestApp(t, &h, &db)
+	app.RequireStart()
+
+	t.Cleanup(app.RequireStop)
+
+	seedOperator(t, db)
+
+	// Well past the limit, and from the same bucket the operator
+	// will arrive in.
+	floodFailures(
+		t, h, sharedProxyPeer, operatorUser,
+		loginFailureLimit*2,
+	)
+
+	w := submitLogin(
+		h, sharedProxyPeer, operatorUser, operatorPassword,
+	)
+
+	assert.Equal(
+		t, http.StatusSeeOther, w.Code,
+		"a correct password must never be throttled: the operator "+
+			"has no second administrative path",
+	)
+	assert.Equal(t, "/", w.Header().Get("Location"))
+}
+
+// TestLogin_StrangersFloodCannotDenyAnotherAccount is the
+// cross-account half: flooding one username must not spend another
+// account's budget, even from the same shared bucket.
+func TestLogin_StrangersFloodCannotDenyAnotherAccount(t *testing.T) {
+	t.Parallel()
+
+	var (
+		h  *handlers.Handlers
+		db *database.Database
+	)
+
+	app := newTestApp(t, &h, &db)
+	app.RequireStart()
+
+	t.Cleanup(app.RequireStop)
+
+	seedOperator(t, db)
+
+	floodFailures(
+		t, h, sharedProxyPeer, "someone-else",
+		loginFailureLimit*2,
+	)
+
+	w := submitLogin(h, sharedProxyPeer, operatorUser, "wrong")
+
+	assert.Equal(
+		t, http.StatusUnauthorized, w.Code,
+		"a flood against one username must not spend another "+
+			"account's failure budget",
+	)
+}
+
+// TestLogin_RepeatedWrongPasswordsAreThrottled is the brute-force
+// half. Verifying before counting must not remove the throttle:
+// repeated wrong passwords for one username from one client key run
+// out of budget and are answered 429 with a Retry-After.
+func TestLogin_RepeatedWrongPasswordsAreThrottled(t *testing.T) {
+	t.Parallel()
+
+	var (
+		h  *handlers.Handlers
+		db *database.Database
+	)
+
+	app := newTestApp(t, &h, &db)
+	app.RequireStart()
+
+	t.Cleanup(app.RequireStop)
+
+	seedOperator(t, db)
+
+	for i := range loginFailureLimit - 1 {
+		w := submitLogin(
+			h, sharedProxyPeer, operatorUser,
+			fmt.Sprintf("guess-%d", i),
+		)
+		assert.Equal(
+			t, http.StatusUnauthorized, w.Code,
+			"attempt %d is still inside the budget", i,
+		)
+	}
+
+	w := submitLogin(h, sharedProxyPeer, operatorUser, "guess-last")
+	assert.Equal(
+		t, http.StatusTooManyRequests, w.Code,
+		"wrong passwords must still run out of budget",
+	)
+	assert.NotEmpty(
+		t, w.Header().Get("Retry-After"),
+		"a throttled login must say when to come back",
+	)
+}
+
+// TestLogin_SuccessForgivesEarlierMistakes covers the operator who
+// mistypes several times and then gets it right: the successful
+// attempt clears the counter, so the next mistake is answered 401
+// rather than 429.
+func TestLogin_SuccessForgivesEarlierMistakes(t *testing.T) {
+	t.Parallel()
+
+	var (
+		h  *handlers.Handlers
+		db *database.Database
+	)
+
+	app := newTestApp(t, &h, &db)
+	app.RequireStart()
+
+	t.Cleanup(app.RequireStop)
+
+	seedOperator(t, db)
+
+	floodFailures(
+		t, h, sharedProxyPeer, operatorUser,
+		loginFailureLimit,
+	)
+
+	require.Equal(
+		t, http.StatusSeeOther,
+		submitLogin(
+			h, sharedProxyPeer, operatorUser, operatorPassword,
+		).Code,
+	)
+
+	w := submitLogin(h, sharedProxyPeer, operatorUser, "typo")
+	assert.Equal(
+		t, http.StatusUnauthorized, w.Code,
+		"a success must forgive the failures before it",
+	)
+}
+
+// TestLogin_UnknownUsernameCostsTheSameVerification is the
+// username-enumeration guard. Verifying credentials before the
+// limiter means response time is observable per attempt, so an
+// unknown username must be charged an equivalent-cost verification
+// against a dummy hash rather than returning early.
+//
+// The assertion is on the code path, not on wall-clock time: timing
+// assertions are flaky, and what actually has to hold is that the
+// hash is computed.
+func TestLogin_UnknownUsernameCostsTheSameVerification(t *testing.T) {
+	t.Parallel()
+
+	var (
+		h  *handlers.Handlers
+		db *database.Database
+	)
+
+	app := newTestApp(t, &h, &db)
+	app.RequireStart()
+
+	t.Cleanup(app.RequireStop)
+
+	seedOperator(t, db)
+
+	require.Zero(t, h.DummyVerificationsForTest())
+
+	// A username that exists, with the wrong password: a real
+	// Argon2id verification runs, and no dummy is needed.
+	require.Equal(
+		t, http.StatusUnauthorized,
+		submitLogin(h, sharedProxyPeer, operatorUser, "wrong").Code,
+	)
+	assert.Zero(
+		t, h.DummyVerificationsForTest(),
+		"a known username verifies against its own hash",
+	)
+
+	// A username that does not exist: indistinguishable response,
+	// and the equivalent-cost verification must have run.
+	require.Equal(
+		t, http.StatusUnauthorized,
+		submitLogin(h, sharedProxyPeer, "nosuchuser", "wrong").Code,
+	)
+	assert.Equal(
+		t, uint64(1), h.DummyVerificationsForTest(),
+		"an unknown username must still pay for a hash, or the "+
+			"response time says whether the account exists",
+	)
+}
+
+// TestLogin_ConcurrentLoginsAreAllAnswered covers the login path
+// under the verification bound. The bound itself is pinned in the
+// middleware package; what matters here is that funnelling every
+// login through two slots does not lose or wedge a request — each one
+// is answered, whether it got a slot or was shed with 503.
+func TestLogin_ConcurrentLoginsAreAllAnswered(t *testing.T) {
+	t.Parallel()
+
+	const workers = 4
+
+	var (
+		h  *handlers.Handlers
+		db *database.Database
+	)
+
+	app := newTestApp(t, &h, &db)
+	app.RequireStart()
+
+	t.Cleanup(app.RequireStop)
+
+	seedOperator(t, db)
+
+	var (
+		wg      sync.WaitGroup
+		mu      sync.Mutex
+		answers = map[int]int{}
+	)
+
+	for i := range workers {
+		wg.Go(func() {
+			w := submitLogin(
+				h, fmt.Sprintf("203.0.113.%d:5000", i),
+				operatorUser, fmt.Sprintf("guess-%d", i),
+			)
+
+			mu.Lock()
+			answers[w.Code]++
+			mu.Unlock()
+		})
+	}
+
+	wg.Wait()
+
+	mu.Lock()
+	defer mu.Unlock()
+
+	assert.Zero(
+		t, answers[http.StatusInternalServerError],
+		"concurrent logins must not error",
+	)
+	assert.Equal(
+		t, workers,
+		answers[http.StatusUnauthorized]+
+			answers[http.StatusTooManyRequests]+
+			answers[http.StatusServiceUnavailable],
+		"every concurrent login must be answered, whether it got "+
+			"a verification slot or was shed with 503",
+	)
+}
+
+// TestLogin_MissingCredentialsRejectedBeforeAnyHash pins that the
+// empty-field check still runs ahead of the verification slot, so a
+// client sending nothing cannot occupy one.
+func TestLogin_MissingCredentialsRejectedBeforeAnyHash(t *testing.T) {
+	t.Parallel()
+
+	var (
+		h  *handlers.Handlers
+		db *database.Database
+	)
+
+	app := newTestApp(t, &h, &db)
+	app.RequireStart()
+
+	t.Cleanup(app.RequireStop)
+
+	seedOperator(t, db)
+
+	w := submitLogin(h, sharedProxyPeer, "", "")
+
+	assert.Equal(t, http.StatusBadRequest, w.Code)
+	assert.Zero(
+		t, h.DummyVerificationsForTest(),
+		"an empty submission must not cost a hash",
+	)
+}
+
+// TestLogin_SuccessCreatesSession is the control for the tests above:
+// the success path they assert on really does authenticate.
+func TestLogin_SuccessCreatesSession(t *testing.T) {
+	t.Parallel()
+
+	var (
+		h    *handlers.Handlers
+		db   *database.Database
+		sess *session.Session
+	)
+
+	app := newTestApp(t, &h, &db, &sess)
+	app.RequireStart()
+
+	t.Cleanup(app.RequireStop)
+
+	seedOperator(t, db)
+
+	w := submitLogin(
+		h, sharedProxyPeer, operatorUser, operatorPassword,
+	)
+
+	require.Equal(t, http.StatusSeeOther, w.Code)
+	require.NotEmpty(
+		t, w.Result().Cookies(), "a session cookie must be issued",
+	)
+
+	next := httptest.NewRequestWithContext(
+		context.Background(), http.MethodGet, "/", nil,
+	)
+
+	// Login regenerates the session, so the response carries two
+	// Set-Cookie headers under the same name: one expiring the
+	// pre-login cookie and one issuing the new one. A browser keeps
+	// only the second, so replay only the one that is not an
+	// expiry.
+	for _, c := range w.Result().Cookies() {
+		if c.MaxAge >= 0 {
+			next.AddCookie(c)
+		}
+	}
+
+	s, err := sess.Get(next)
+	require.NoError(t, err)
+	assert.True(
+		t, sess.IsAuthenticated(s),
+		"the issued cookie must carry an authenticated session",
+	)
+}

internal/handlers/export_test.go

@@ -2,15 +2,31 @@ package handlers
 
 import (
 	"html/template"
+	"log/slog"
 	"net/http"
 
 	"sneak.berlin/go/webhooker/internal/database"
 )
 
+// SetLogForTest replaces the handler's logger, so the handlers_test
+// package can assert on what a log line actually contains rather than
+// on what it is meant to contain.
+func (s *Handlers) SetLogForTest(log *slog.Logger) {
+	s.log = log
+}
+
 // MaxRenderedBodyBytesForTest exposes the event log's body cap
 // to the handlers_test package.
 const MaxRenderedBodyBytesForTest = maxRenderedBodyBytes
 
+// DummyVerificationsForTest reports how many equivalent-cost
+// verifications were charged for usernames that do not exist. It
+// lets a test prove the anti-enumeration path ran without timing
+// anything.
+func (s *Handlers) DummyVerificationsForTest() uint64 {
+	return s.dummyVerifications.Load()
+}
+
 // TrimPartialRuneForTest exposes trimPartialRune for use in the
 // handlers_test package.
 func TrimPartialRuneForTest(b []byte) []byte {

internal/handlers/gormlogbound_test.go

@@ -0,0 +1,462 @@
+package handlers_test
+
+import (
+	"bytes"
+	"context"
+	"io"
+	"log"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"strconv"
+	"strings"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/go-chi/chi"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"gorm.io/gorm"
+	gormlogger "gorm.io/gorm/logger"
+	"sneak.berlin/go/webhooker/internal/database"
+	"sneak.berlin/go/webhooker/internal/handlers"
+	"sneak.berlin/go/webhooker/internal/middleware"
+)
+
+// gormBoundTailMarker sits at the far end of every client-chosen value
+// this file sends. Its presence in the log means the whole value
+// reached the log, so a value that merely happened to be short cannot
+// pass for a truncated one.
+const gormBoundTailMarker = "ENDOFCLIENTVALUE"
+
+// gormBoundFills are the characters a client can drive through the
+// receiver path segment and the login username, chosen for what a log
+// handler charges for them.
+//
+// The bare C0 control is the one that matters: both handlers spell
+// U+0001 as a six-byte escape for the one byte it costs to send, the
+// widest multiplier available below U+10000 and the case a raw-byte
+// budget breaks on first. GORM's default logger applies no budget at
+// all, so under the mutation every one of these arrives whole.
+func gormBoundFills() []struct {
+	name string
+	fill string
+} {
+	return []struct {
+		name string
+		fill string
+	}{
+		{"plain", "x"},
+		{"quote", `"`},
+		{"backslash", `\`},
+		{"tab", "\t"},
+		{"newline", "\n"},
+		{"c0_control", "\x01"},
+		{"astral_nonprintable", "\U0001000C"},
+	}
+}
+
+// syncBuf collects captured output from the goroutine draining the
+// pipe.
+type syncBuf struct {
+	mu sync.Mutex
+	b  bytes.Buffer
+}
+
+func (s *syncBuf) Write(p []byte) (int, error) {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	return s.b.Write(p)
+}
+
+func (s *syncBuf) String() string {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	return s.b.String()
+}
+
+func (s *syncBuf) reset() {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+
+	s.b.Reset()
+}
+
+// stdoutCapture redirects os.Stdout for the duration of a test.
+//
+// internal/logger builds its handler over os.Stdout at construction
+// time, so redirecting the variable before the application is built
+// captures everything the service logger — and therefore the GORM
+// adapter, which writes through it — emits.
+type stdoutCapture struct {
+	buf  *syncBuf
+	r    *os.File
+	w    *os.File
+	orig *os.File
+	done chan struct{}
+	seq  int
+}
+
+func captureStdout(t *testing.T) *stdoutCapture {
+	t.Helper()
+
+	r, w, err := os.Pipe()
+	require.NoError(t, err)
+
+	c := &stdoutCapture{
+		buf:  &syncBuf{},
+		r:    r,
+		w:    w,
+		orig: os.Stdout,
+		done: make(chan struct{}),
+	}
+
+	os.Stdout = w
+
+	go func() {
+		defer close(c.done)
+
+		_, _ = io.Copy(c.buf, r)
+	}()
+
+	t.Cleanup(func() {
+		os.Stdout = c.orig
+		_ = w.Close()
+
+		<-c.done
+
+		_ = r.Close()
+	})
+
+	return c
+}
+
+// drain returns everything written since the previous drain and
+// clears the buffer.
+//
+// A sentinel is pushed through the same pipe and waited for, so the
+// draining goroutine is known to have caught up before the buffer is
+// read. Without it the comparison below would race the reader rather
+// than measure the writers.
+func (c *stdoutCapture) drain(t *testing.T) string {
+	t.Helper()
+
+	c.seq++
+
+	sentinel := "\n<<drain-" + strconv.Itoa(c.seq) + ">>\n"
+
+	_, err := c.w.WriteString(sentinel)
+	require.NoError(t, err)
+
+	deadline := time.Now().Add(10 * time.Second)
+	for !strings.Contains(c.buf.String(), sentinel) {
+		require.False(
+			t, time.Now().After(deadline),
+			"timed out waiting for captured output",
+		)
+		time.Sleep(time.Millisecond)
+	}
+
+	out := strings.Replace(c.buf.String(), sentinel, "", 1)
+
+	c.buf.reset()
+
+	return out
+}
+
+// teeStdout writes to a buffer and to whatever os.Stdout is at the
+// moment of the write.
+//
+// The second half is the point. GORM's package-level default logger
+// resolves os.Stdout once, at package init, so a logger built over the
+// variable would keep writing to the real terminal no matter what a
+// test redirects. Resolving it per write puts the bytes a defaulted
+// gorm.Config would cost in production into the same capture as
+// everything else internal/logger emits, which is what lets the volume
+// assertions below measure the whole writer set rather than one member
+// of it.
+type teeStdout struct {
+	buf *syncBuf
+}
+
+func (w teeStdout) Write(p []byte) (int, error) {
+	_, _ = os.Stdout.Write(p)
+
+	return w.buf.Write(p)
+}
+
+// captureGORMDefault replaces GORM's package-level default logger with
+// one configured exactly as GORM configures its own, writing to a
+// buffer and to os.Stdout.
+//
+// This is the mutation detector. gormlogger.Default is what a bare
+// &gorm.Config{} installs, and its config here is GORM's verbatim —
+// Warn, IgnoreRecordNotFoundError false — so a reverted call site
+// behaves as it would in production rather than as a test dialed it.
+// With every gorm.Open in this service naming its own logger, nothing
+// consults this value and the buffer stays empty; revert any one of
+// the three and the interpolated SQL lands here.
+func captureGORMDefault(t *testing.T) *syncBuf {
+	t.Helper()
+
+	buf := &syncBuf{}
+	orig := gormlogger.Default
+
+	gormlogger.Default = gormlogger.New(
+		log.New(teeStdout{buf: buf}, "", log.LstdFlags),
+		gormlogger.Config{
+			SlowThreshold:             200 * time.Millisecond,
+			LogLevel:                  gormlogger.Warn,
+			IgnoreRecordNotFoundError: false,
+			Colorful:                  false,
+		},
+	)
+
+	t.Cleanup(func() { gormlogger.Default = orig })
+
+	return buf
+}
+
+// floodUnauthenticated drives reps requests at each of the two
+// unauthenticated lookups that miss by design, for every fill, with a
+// client-chosen value of size raw bytes.
+func floodUnauthenticated(
+	t *testing.T, h *handlers.Handlers, size, reps int,
+) int {
+	t.Helper()
+
+	requests := 0
+
+	for _, f := range gormBoundFills() {
+		var b strings.Builder
+
+		for b.Len() < size {
+			b.WriteString(f.fill)
+		}
+
+		b.WriteString(gormBoundTailMarker)
+
+		value := b.String()
+
+		for range reps {
+			postWebhook(t, h, value)
+			postUnknownLogin(t, h, value)
+
+			requests += 2
+		}
+	}
+
+	return requests
+}
+
+// floodPerWebhook drives the same client-chosen values at the second
+// gorm.Open site, the per-webhook database internal/database's
+// WebhookDBManager opens.
+//
+// That site is behind authentication in production, so this is not
+// part of the unauthenticated flood above and is counted separately.
+// It is here because the ceiling the README states covers every
+// writer, and the manager is one of them: with nothing driving it, a
+// bare &gorm.Config{} could be restored at
+// internal/database/webhook_db_manager.go and the whole suite would
+// stay green.
+func floodPerWebhook(
+	t *testing.T, mgr *database.WebhookDBManager, size, reps int,
+) int {
+	t.Helper()
+
+	requests := 0
+
+	for _, f := range gormBoundFills() {
+		var b strings.Builder
+
+		for b.Len() < size {
+			b.WriteString(f.fill)
+		}
+
+		b.WriteString(gormBoundTailMarker)
+
+		value := b.String()
+
+		db, err := mgr.GetDB("pin-" + f.name)
+		require.NoError(t, err)
+
+		for range reps {
+			var got database.Event
+
+			err = db.Where("id = ?", value).First(&got).Error
+			require.ErrorIs(t, err, gorm.ErrRecordNotFound)
+
+			requests++
+		}
+	}
+
+	return requests
+}
+
+// postWebhook drives the receiver with an invented entrypoint path.
+// The route pattern matches any single segment, so every byte of the
+// value is the client's, and the lookup behind it misses by design.
+func postWebhook(
+	t *testing.T, h *handlers.Handlers, entrypoint string,
+) {
+	t.Helper()
+
+	req := httptest.NewRequestWithContext(
+		context.Background(), http.MethodPost, "/webhook/x",
+		strings.NewReader("{}"),
+	)
+
+	rctx := chi.NewRouteContext()
+	rctx.URLParams.Add("uuid", entrypoint)
+
+	req = req.WithContext(context.WithValue(
+		req.Context(), chi.RouteCtxKey, rctx,
+	))
+
+	w := httptest.NewRecorder()
+	h.HandleWebhook().ServeHTTP(w, req)
+
+	require.Equal(t, http.StatusNotFound, w.Code)
+}
+
+// postUnknownLogin submits the login form with an unknown username,
+// through the postLogin helper in logbound_test.go. The field is
+// bounded only by the 1 MB body cap, and the lookup behind it misses
+// by design.
+func postUnknownLogin(
+	t *testing.T, h *handlers.Handlers, username string,
+) {
+	t.Helper()
+
+	// 401 while the client still has failure budget against this
+	// username, 429 once the login guard has taken it away. Both
+	// outcomes sit behind the user lookup, which is the query this
+	// test is here to drive.
+	require.Contains(
+		t,
+		[]int{http.StatusUnauthorized, http.StatusTooManyRequests},
+		postLogin(t, h, username),
+	)
+}
+
+// assertFloodBounded holds every captured line to the stated ceiling
+// and proves nothing carried a whole client value.
+func assertFloodBounded(t *testing.T, label, out string) {
+	t.Helper()
+
+	assert.NotContains(
+		t, out, gormBoundTailMarker,
+		"%s: the far end of a client-chosen value reached the "+
+			"log, so nothing truncated it", label,
+	)
+
+	for line := range strings.SplitSeq(
+		strings.TrimRight(out, "\n"), "\n",
+	) {
+		if line == "" {
+			continue
+		}
+
+		assert.LessOrEqual(
+			t, len(line), middleware.MaxAccessLogLineBytes,
+			"%s: log line exceeded its bound: %s",
+			label, line[:min(len(line), 300)],
+		)
+	}
+}
+
+// TestFlood_NoWriterGrowsWithTheInput is the definition of done for
+// the GORM logger defect, stated over every writer at once, for two of
+// this service's three gorm.Open sites: the main database behind the
+// two unauthenticated lookups, and the per-webhook database the
+// WebhookDBManager opens. The third, the archive writer, is pinned in
+// internal/delivery, where its type lives.
+//
+// What each assertion is worth, since two of the three would pass
+// against a service that had never been fixed if the capture were set
+// up differently:
+//
+//   - The gormDefault check is the sharp one. It fires the moment any
+//     gorm.Open in this service goes back to a bare &gorm.Config{}.
+//   - The volume and per-line checks bite only because the replaced
+//     default logger tees into os.Stdout, so a reverted call site
+//     shows up in the same capture as everything internal/logger
+//     writes — the way it would in production. Without that tee both
+//     were vacuous: at INFO the two handler misses log at DEBUG and
+//     the adapter drops the record-not-found, so the capture holds
+//     nothing but fixed-string warnings.
+//
+// The level is left where newTestApp leaves it, at INFO: the level an
+// operator runs at by default, and the one the defect was visible at.
+// The handlers' own miss lines sit at DEBUG and spend the same
+// logfield budget as everything else, so they are not what makes
+// either assertion above bite at any level.
+//
+// It is deliberately not parallel: it redirects os.Stdout and replaces
+// gormlogger.Default, both of which are process-global. Go runs every
+// non-parallel top-level test to completion before it resumes the
+// parallel ones, so nothing else in this package is running while the
+// capture is installed.
+//
+//nolint:paralleltest // Deliberately sequential; see above.
+func TestFlood_NoWriterGrowsWithTheInput(t *testing.T) {
+	const (
+		smallBytes = 128
+		bigBytes   = 8 << 10
+		reps       = 5
+	)
+
+	gormDefault := captureGORMDefault(t)
+	capture := captureStdout(t)
+
+	var (
+		h   *handlers.Handlers
+		mgr *database.WebhookDBManager
+	)
+
+	app := newTestApp(t, &h, &mgr)
+	app.RequireStart()
+
+	t.Cleanup(app.RequireStop)
+
+	// Startup chatter is not what this test measures.
+	capture.drain(t)
+
+	floodUnauthenticated(t, h, smallBytes, reps)
+	floodPerWebhook(t, mgr, smallBytes, reps)
+
+	small := capture.drain(t)
+
+	requests := floodUnauthenticated(t, h, bigBytes, reps)
+	requests += floodPerWebhook(t, mgr, bigBytes, reps)
+	big := capture.drain(t)
+
+	assertFloodBounded(t, "small flood", small)
+	assertFloodBounded(t, "big flood", big)
+
+	// GORM's default logger is what the defect was. Nothing in this
+	// service may reach it.
+	got := gormDefault.String()
+	assert.Empty(
+		t, got,
+		"GORM's default logger wrote %d bytes; the first of them: %s",
+		len(got), got[:min(len(got), 300)],
+	)
+
+	// The same flood, with 64 times the client-chosen input, must not
+	// buy 64 times the log. A few bytes of slack covers a latency
+	// field changing width; the input grew by roughly half a megabyte.
+	const slackPerRequest = 64
+
+	assert.LessOrEqual(
+		t, len(big), len(small)+slackPerRequest*requests,
+		"log volume tracked the size of the client's input: "+
+			"%d bytes at %d bytes of input per request, %d bytes "+
+			"at %d",
+		len(small), smallBytes, len(big), bigBytes,
+	)
+}

internal/handlers/handlers.go

@@ -10,6 +10,7 @@ import (
 	"html/template"
 	"log/slog"
 	"net/http"
+	"sync/atomic"
 
 	"go.uber.org/fx"
 	"sneak.berlin/go/webhooker/internal/database"
@@ -39,6 +40,12 @@ const (
 // errInvalidPassword is returned when a password does not match.
 var errInvalidPassword = errors.New("invalid password")
 
+// errVerificationBusy is returned when no password-verification slot
+// became free before the wait elapsed, so no password was verified.
+var errVerificationBusy = errors.New(
+	"password verification capacity exhausted",
+)
+
 //nolint:revive // HandlersParams is a standard fx naming convention.
 type HandlersParams struct {
 	fx.In
@@ -49,6 +56,7 @@ type HandlersParams struct {
 	WebhookDBMgr *database.WebhookDBManager
 	Healthcheck  *healthcheck.Healthcheck
 	Session      *session.Session
+	Middleware   *middleware.Middleware
 	Notifier     delivery.Notifier
 	Evictor      delivery.WebhookEvictor
 }
@@ -62,9 +70,15 @@ type Handlers struct {
 	db        *database.Database
 	dbMgr     *database.WebhookDBManager
 	session   *session.Session
+	mw        *middleware.Middleware
 	notifier  delivery.Notifier
 	evictor   delivery.WebhookEvictor
 	templates map[string]*template.Template
+
+	// dummyVerifications counts the equivalent-cost verifications
+	// charged for usernames that do not exist. It exists so a test
+	// can prove that path runs without measuring wall-clock time.
+	dummyVerifications atomic.Uint64
 }
 
 // parsePageTemplate parses a page-specific template set from the
@@ -97,6 +111,7 @@ func New(
 	s.db = params.Database
 	s.dbMgr = params.WebhookDBMgr
 	s.session = params.Session
+	s.mw = params.Middleware
 	s.notifier = params.Notifier
 	s.evictor = params.Evictor
 

internal/handlers/handlers_test.go

@@ -20,6 +20,7 @@ import (
 	"sneak.berlin/go/webhooker/internal/handlers"
 	"sneak.berlin/go/webhooker/internal/healthcheck"
 	"sneak.berlin/go/webhooker/internal/logger"
+	"sneak.berlin/go/webhooker/internal/middleware"
 	"sneak.berlin/go/webhooker/internal/session"
 )
 
@@ -82,6 +83,7 @@ func newTestApp(
 			func(r *recordingEvictor) delivery.WebhookEvictor {
 				return r
 			},
+			middleware.New,
 			handlers.New,
 		),
 		fx.Populate(targets...),

internal/handlers/logbound_test.go

@@ -0,0 +1,543 @@
+package handlers_test
+
+// The handler-side half of the log-field audit. Two slog calls in
+// this package reach a value an UNAUTHENTICATED client picks outright
+// and of a length it picks outright:
+//
+//   - the unknown-entrypoint DEBUG line on /webhook/{uuid}, whose
+//     path segment matched no stored entrypoint and so is bounded by
+//     nothing;
+//   - the failed-login DEBUG lines, whose username is a form field.
+//
+// Both are at DEBUG, which is off in production by default. That is
+// not a bound: an operator turning DEBUG on to diagnose a flood must
+// not thereby hand the flood an unbounded write. Both spend the same
+// internal/logfield budget as the access log, and both are held here
+// to middleware.MaxAccessLogLineBytes.
+//
+// The two login lines past the username lookup — "invalid password"
+// and "user logged in" — carry the same cap without needing it, since
+// by then the value is a stored row rather than the client's. They are
+// pinned here too, so the caps cannot be dropped silently.
+//
+// So is the "password verification capacity exhausted" WARN line,
+// whose path chi pins to the constant "/pages/login" on the one route
+// that reaches it. Its cap is defensive, and the test below drives the
+// handler directly with the path a parameterised route would give it,
+// because an unasserted cap is one a later edit removes for free.
+
+import (
+	"bytes"
+	"context"
+	"io"
+	"log/slog"
+	"net/http"
+	"net/http/httptest"
+	"net/url"
+	"strings"
+	"testing"
+
+	"github.com/go-chi/chi"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/webhooker/internal/database"
+	"sneak.berlin/go/webhooker/internal/handlers"
+	"sneak.berlin/go/webhooker/internal/middleware"
+)
+
+// floodRequests is the number of distinct invented values each flood
+// drives through the call site under test.
+const floodRequests = 32
+
+// oversizedFillBytes is the length of the single client-chosen value
+// used to show that line size does not track input size.
+const oversizedFillBytes = 8192
+
+// attackerMarker and tailMarker sit at the END of every oversized
+// value, past every budget. Their absence from the log is what
+// proves the value was cut rather than merely being short.
+const (
+	attackerMarker = "QQATTACKERTEXTQQ"
+	tailMarker     = "QQTRUNCATEDTAILQQ"
+)
+
+// escapeFills are the characters the log handlers escape, so a value
+// built out of them costs more on the line than it did on the wire. A
+// budget counted in raw bytes passes the plain case and fails these.
+//
+// U+1000C is unassigned, hence non-printable, and strconv.Quote
+// spells it as a ten-byte \UXXXXXXXX while the JSON handler passes
+// its four UTF-8 bytes through; only the text shape of these tests
+// reaches that charge.
+func escapeFills() map[string]string {
+	return map[string]string{
+		"plain":     "x",
+		"quote":     `"`,
+		"backslash": `\`,
+		"tab":       "\t",
+		"newline":   "\n",
+		// A C0 control neither handler has a short escape for, so
+		// each one costs six bytes on the line against the single
+		// byte it cost to send: the widest multiplier a client can
+		// drive, and the case a raw-byte budget breaks on first.
+		//
+		// This fill is load-bearing, not decoration. Budgeting raw
+		// bytes instead of encoded is caught by this fill alone,
+		// and only under the JSON handler, at 3,072 bytes against
+		// the 2,560 ceiling. Drop it and that mutation passes.
+		"control": "\x01",
+		"astral":  "\U0001000C",
+	}
+}
+
+// logHandlers are the two handlers internal/logger can install.
+func logHandlers() map[string]func(
+	io.Writer, *slog.HandlerOptions,
+) slog.Handler {
+	return map[string]func(
+		io.Writer, *slog.HandlerOptions,
+	) slog.Handler{
+		"json": func(
+			w io.Writer, o *slog.HandlerOptions,
+		) slog.Handler {
+			return slog.NewJSONHandler(w, o)
+		},
+		"text": func(
+			w io.Writer, o *slog.HandlerOptions,
+		) slog.Handler {
+			return slog.NewTextHandler(w, o)
+		},
+	}
+}
+
+// oversizedFill builds an 8 KB client-chosen value out of
+// repetitions of ch, with both markers at its far end.
+func oversizedFill(ch string) string {
+	return "x" + strings.Repeat(ch, oversizedFillBytes) +
+		attackerMarker + tailMarker
+}
+
+// capturingHandlers builds a Handlers whose log is captured into the
+// returned buffer at DEBUG through the named handler.
+//
+// extra is passed to fx.Populate alongside the Handlers, for the call
+// sites that also need the database the client's value is looked up
+// in, or the Middleware whose resource has to be exhausted before the
+// branch under test is reached.
+func capturingHandlers(
+	t *testing.T,
+	newHandler func(io.Writer, *slog.HandlerOptions) slog.Handler,
+	extra ...any,
+) (*handlers.Handlers, *bytes.Buffer) {
+	t.Helper()
+
+	var h *handlers.Handlers
+
+	app := newTestApp(t, append([]any{&h}, extra...)...)
+	app.RequireStart()
+
+	t.Cleanup(app.RequireStop)
+
+	buf := new(bytes.Buffer)
+	h.SetLogForTest(slog.New(newHandler(
+		buf, &slog.HandlerOptions{Level: slog.LevelDebug},
+	)))
+
+	return h, buf
+}
+
+// logLines splits the captured buffer into non-empty lines, holding
+// each to the stated per-line ceiling.
+func logLines(t *testing.T, buf *bytes.Buffer) []string {
+	t.Helper()
+
+	var lines []string
+
+	for line := range strings.SplitSeq(
+		strings.TrimSpace(buf.String()), "\n",
+	) {
+		if line == "" {
+			continue
+		}
+
+		require.LessOrEqual(
+			t, len(line), middleware.MaxAccessLogLineBytes,
+			"log line exceeded its bound: %s", line,
+		)
+
+		lines = append(lines, line)
+	}
+
+	return lines
+}
+
+// assertNoClientText fails if the far end of the client-chosen input
+// survived into the log.
+func assertNoClientText(t *testing.T, buf *bytes.Buffer) {
+	t.Helper()
+
+	assert.NotContains(
+		t, buf.String(), attackerMarker,
+		"log carried attacker-chosen text",
+	)
+	assert.NotContains(
+		t, buf.String(), tailMarker,
+		"log carried the tail of the attacker-chosen text",
+	)
+}
+
+// receiverRouter mounts the real receiver handler at the production
+// route pattern.
+func receiverRouter(h *handlers.Handlers) *chi.Mux {
+	router := chi.NewRouter()
+	router.Post("/webhook/{uuid}", h.HandleWebhook())
+
+	return router
+}
+
+// postReceiver sends one POST at /webhook/<segment>.
+//
+// RawPath is cleared after parsing so chi routes on the decoded path
+// and the handler sees the raw bytes rather than their percent-escaped
+// spelling. That is the harder case for the budget: the escaped
+// spelling is plain ASCII, which costs one byte per byte, while the
+// decoded bytes are what the log handler has to escape.
+func postReceiver(
+	t *testing.T, router *chi.Mux, segment string,
+) int {
+	t.Helper()
+
+	req := httptest.NewRequestWithContext(
+		context.Background(),
+		http.MethodPost,
+		"/webhook/"+url.PathEscape(segment),
+		strings.NewReader(""),
+	)
+	req.URL.RawPath = ""
+
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	return w.Code
+}
+
+// postLogin submits the login form with the given username and a
+// non-empty password.
+func postLogin(
+	t *testing.T, h *handlers.Handlers, username string,
+) int {
+	t.Helper()
+
+	return postLoginWithPassword(t, h, username, "not-the-password")
+}
+
+// postLoginWithPassword submits the login form with both credentials
+// chosen by the caller, so a test can reach the branches past the
+// username lookup.
+func postLoginWithPassword(
+	t *testing.T, h *handlers.Handlers, username, password string,
+) int {
+	t.Helper()
+
+	form := url.Values{
+		"username": {username},
+		"password": {password},
+	}
+
+	req := httptest.NewRequestWithContext(
+		context.Background(),
+		http.MethodPost,
+		"/pages/login",
+		strings.NewReader(form.Encode()),
+	)
+	req.Header.Set(
+		"Content-Type", "application/x-www-form-urlencoded",
+	)
+
+	w := httptest.NewRecorder()
+	h.HandleLoginSubmit().ServeHTTP(w, req)
+
+	return w.Code
+}
+
+// TestUnknownEntrypoint_LogLineDoesNotTrackPathSize drives 8 KB of
+// client-chosen path at the unauthenticated receiver's
+// unknown-entrypoint DEBUG line and holds it to the same ceiling the
+// access log states.
+func TestUnknownEntrypoint_LogLineDoesNotTrackPathSize(t *testing.T) {
+	t.Parallel()
+
+	for handlerName, newHandler := range logHandlers() {
+		for fillName, fill := range escapeFills() {
+			t.Run(handlerName+"/"+fillName, func(t *testing.T) {
+				t.Parallel()
+
+				h, buf := capturingHandlers(t, newHandler)
+				router := receiverRouter(h)
+
+				for i := range floodRequests {
+					assert.Equal(
+						t,
+						http.StatusNotFound,
+						postReceiver(
+							t, router,
+							oversizedFill(fill)+
+								strings.Repeat("y", i),
+						),
+					)
+				}
+
+				lines := logLines(t, buf)
+				require.Len(t, lines, floodRequests)
+				assertNoClientText(t, buf)
+
+				assertBoundedFlood(t, buf.Len())
+			})
+		}
+	}
+}
+
+// TestFailedLogin_LogLineDoesNotTrackUsernameSize drives 8 KB of
+// client-chosen username at the unauthenticated login endpoint's
+// DEBUG line and holds it to the same ceiling.
+func TestFailedLogin_LogLineDoesNotTrackUsernameSize(t *testing.T) {
+	t.Parallel()
+
+	for handlerName, newHandler := range logHandlers() {
+		for fillName, fill := range escapeFills() {
+			t.Run(handlerName+"/"+fillName, func(t *testing.T) {
+				t.Parallel()
+
+				h, buf := capturingHandlers(t, newHandler)
+
+				for i := range floodRequests {
+					assert.Equal(
+						t,
+						http.StatusUnauthorized,
+						postLogin(
+							t, h,
+							oversizedFill(fill)+
+								strings.Repeat("y", i),
+						),
+					)
+				}
+
+				lines := logLines(t, buf)
+				require.Len(t, lines, floodRequests)
+				assertNoClientText(t, buf)
+
+				assertBoundedFlood(t, buf.Len())
+			})
+		}
+	}
+}
+
+// storedUserPassword is the password held by the oversize accounts
+// the test below creates.
+const storedUserPassword = "correct-horse-battery-staple"
+
+// storedFillBytes is the raw length of the client-chosen value in
+// those accounts' usernames. It is well past the 512-byte field
+// budget, so the line is still truncated, but short enough that the
+// session cookie a successful login writes stays inside
+// securecookie's 4 KB limit: the cookie is written BEFORE the
+// "user logged in" line, so an 8 KB username answers 500 and never
+// reaches it.
+const storedFillBytes = 1024
+
+// storedFill builds a username fill of storedFillBytes raw bytes out
+// of repetitions of ch, with both markers at its far end.
+func storedFill(ch string) string {
+	return "x" + strings.Repeat(ch, storedFillBytes/len(ch)) +
+		attackerMarker + tailMarker
+}
+
+// TestStoredUsername_LogLinesDoNotTrackUsernameSize pins the two
+// login lines that are reached only AFTER the username matched a
+// stored row: "invalid password" and "user logged in". Neither
+// strictly needs its cap — the value is the operator's own data by
+// then, not the client's — but both carry one so that every username
+// this unauthenticated endpoint logs is capped, and an unasserted cap
+// is one a later edit removes for free.
+//
+// One app per handler with the accounts created inside it, and no
+// parallelism below that level: every account costs an Argon2id hash
+// and every attempt costs a verification.
+func TestStoredUsername_LogLinesDoNotTrackUsernameSize(t *testing.T) {
+	t.Parallel()
+
+	for handlerName, newHandler := range logHandlers() {
+		t.Run(handlerName, func(t *testing.T) {
+			t.Parallel()
+
+			var db *database.Database
+
+			h, buf := capturingHandlers(t, newHandler, &db)
+
+			hash, err := database.HashPassword(storedUserPassword)
+			require.NoError(t, err)
+
+			fills := escapeFills()
+			for fillName, fill := range fills {
+				username := storedFill(fill) + fillName
+
+				require.NoError(t, db.DB().Create(&database.User{
+					Username: username,
+					Password: hash,
+				}).Error)
+
+				// Matched the row, wrong secret: "invalid
+				// password".
+				assert.Equal(
+					t, http.StatusUnauthorized,
+					postLoginWithPassword(
+						t, h, username, "not-the-password",
+					),
+				)
+
+				// Matched the row, right secret: "user logged
+				// in".
+				assert.Equal(
+					t, http.StatusSeeOther,
+					postLoginWithPassword(
+						t, h, username, storedUserPassword,
+					),
+				)
+			}
+
+			lines := logLines(t, buf)
+			require.Len(t, lines, 2*len(fills))
+			assertNoClientText(t, buf)
+		})
+	}
+}
+
+// maxVerificationSlots bounds how many slots the loop below will
+// take before it gives up, so a semaphore that never fills fails the
+// test instead of hanging it. It is deliberately larger than the
+// real concurrency bound, which is not exported to this package.
+const maxVerificationSlots = 64
+
+// canceledContext returns a context that is already done. A
+// verification request carrying one takes the ctx.Done() branch of
+// the semaphore's bounded wait immediately, so these cases turn on
+// the semaphore being full rather than on a five-second timer firing.
+// Nothing here is timing-dependent.
+func canceledContext() context.Context {
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel()
+
+	return ctx
+}
+
+// holdEveryVerificationSlot takes verification slots until one is
+// refused, and releases them when the test ends. A free slot is
+// handed out before any context is consulted, so a canceled context
+// cannot make this loop stop early: it stops exactly when the slots
+// are gone.
+func holdEveryVerificationSlot(
+	t *testing.T, mw *middleware.Middleware,
+) {
+	t.Helper()
+
+	for range maxVerificationSlots {
+		release, ok := mw.BeginPasswordVerification(canceledContext())
+		if !ok {
+			return
+		}
+
+		t.Cleanup(release)
+	}
+
+	require.Fail(t, "the verification semaphore never filled")
+}
+
+// postLoginAtPath submits the login form at a path of the caller's
+// choosing, with a canceled context.
+func postLoginAtPath(
+	t *testing.T, h *handlers.Handlers, path string,
+) int {
+	t.Helper()
+
+	form := url.Values{
+		"username": {"someone"},
+		"password": {"not-the-password"},
+	}
+
+	req := httptest.NewRequestWithContext(
+		canceledContext(),
+		http.MethodPost,
+		path,
+		strings.NewReader(form.Encode()),
+	)
+	req.Header.Set(
+		"Content-Type", "application/x-www-form-urlencoded",
+	)
+
+	w := httptest.NewRecorder()
+	h.HandleLoginSubmit().ServeHTTP(w, req)
+
+	return w.Code
+}
+
+// TestVerificationCapacity_LogLineDoesNotTrackPathSize pins the cap
+// on the "password verification capacity exhausted" WARN line.
+//
+// The one route that reaches it is chi's static "/pages/login", so no
+// request through the mux can widen the line; the handler is driven
+// directly here with the path a parameterised route would give it,
+// which is what that cap exists for. Without this test, removing the
+// logfield.Truncate there fails nothing.
+func TestVerificationCapacity_LogLineDoesNotTrackPathSize(
+	t *testing.T,
+) {
+	t.Parallel()
+
+	for handlerName, newHandler := range logHandlers() {
+		for fillName, fill := range escapeFills() {
+			t.Run(handlerName+"/"+fillName, func(t *testing.T) {
+				t.Parallel()
+
+				var mw *middleware.Middleware
+
+				h, buf := capturingHandlers(t, newHandler, &mw)
+
+				holdEveryVerificationSlot(t, mw)
+
+				assert.Equal(
+					t,
+					http.StatusServiceUnavailable,
+					postLoginAtPath(
+						t, h,
+						"/source/"+url.PathEscape(
+							oversizedFill(fill),
+						)+"/login",
+					),
+				)
+
+				lines := logLines(t, buf)
+				require.Len(t, lines, 1)
+				assertNoClientText(t, buf)
+			})
+		}
+	}
+}
+
+// assertBoundedFlood holds the whole flood's log output to what the
+// stated per-line ceiling allows. The flood sent
+// floodRequests * oversizedFillBytes bytes of client-chosen text;
+// this is the assertion that the log did not grow with it.
+func assertBoundedFlood(t *testing.T, got int) {
+	t.Helper()
+
+	sent := floodRequests * oversizedFillBytes
+
+	require.Less(
+		t, got, sent/2,
+		"log volume tracked the size of the flood's input",
+	)
+	require.LessOrEqual(
+		t, got,
+		floodRequests*middleware.MaxAccessLogLineBytes,
+	)
+}

internal/handlers/profile.go

@@ -1,6 +1,7 @@
 package handlers
 
 import (
+	"context"
 	"net/http"
 
 	"github.com/go-chi/chi"
@@ -42,11 +43,14 @@ func (h *Handlers) HandlePasswordChange() http.HandlerFunc {
 		}
 
 		successMessage, errorMessage, handled := h.applyPasswordChange(
+			r.Context(),
 			w,
 			sessionUsername,
-			r.FormValue("current_password"),
-			r.FormValue("new_password"),
-			r.FormValue("confirm_password"),
+			// PostFormValue, not FormValue: the credential must
+			// come from the body, never from the query string.
+			r.PostFormValue("current_password"),
+			r.PostFormValue("new_password"),
+			r.PostFormValue("confirm_password"),
 		)
 		if !handled {
 			return
@@ -66,9 +70,30 @@ func (h *Handlers) HandlePasswordChange() http.HandlerFunc {
 // 500 response itself and returns handled=false, signalling the caller
 // to stop without re-rendering the page.
 func (h *Handlers) applyPasswordChange(
+	ctx context.Context,
 	w http.ResponseWriter,
 	username, currentPassword, newPassword, confirmPassword string,
 ) (string, string, bool) {
+	// This endpoint verifies one password and hashes another, at
+	// 64 MB each, so it takes a slot from the same bound the login
+	// endpoint uses. The bound is per hash, not per endpoint: leaving
+	// this path outside it would leave a hole in it. The slot is held
+	// across both hashes.
+	release, ok := h.mw.BeginPasswordVerification(ctx)
+	if !ok {
+		h.log.Warn("password verification capacity exhausted")
+		http.Error(
+			w,
+			"The server is busy verifying credentials. "+
+				"Please try again.",
+			http.StatusServiceUnavailable,
+		)
+
+		return "", "", false
+	}
+
+	defer release()
+
 	// Load the user row so we can verify the current password and
 	// persist the new hash.
 	var user database.User

internal/handlers/source_management.go

@@ -227,9 +227,9 @@ func (h *Handlers) HandleSourceCreateSubmit() http.HandlerFunc {
 			return
 		}
 
-		name := r.FormValue("name")
-		description := r.FormValue("description")
-		retentionStr := r.FormValue("retention_days")
+		name := r.PostFormValue("name")
+		description := r.PostFormValue("description")
+		retentionStr := r.PostFormValue("retention_days")
 
 		if name == "" {
 			w.WriteHeader(http.StatusBadRequest)
@@ -509,7 +509,7 @@ func (h *Handlers) applyWebhookEdit(
 ) {
 	// The body size cap is enforced by the MaxBodySize middleware,
 	// which runs before CSRF parses the form.
-	name := r.FormValue("name")
+	name := r.PostFormValue("name")
 	if name == "" {
 		data := map[string]any{
 			tmplKeyWebhook: webhook,
@@ -523,12 +523,12 @@ func (h *Handlers) applyWebhookEdit(
 	}
 
 	webhook.Name = name
-	webhook.Description = r.FormValue("description")
+	webhook.Description = r.PostFormValue("description")
 
 	// An empty field falls back to the stored value, so submitting the
 	// form without touching retention leaves the policy alone.
 	retentionDays, retErr := parseRetentionDays(
-		r.FormValue("retention_days"), webhook.RetentionDays,
+		r.PostFormValue("retention_days"), webhook.RetentionDays,
 	)
 	if retErr != nil {
 		data := map[string]any{
@@ -950,7 +950,7 @@ func (h *Handlers) HandleEntrypointCreate() http.HandlerFunc {
 			return
 		}
 
-		description := r.FormValue("description")
+		description := r.PostFormValue("description")
 
 		entrypoint := &database.Entrypoint{
 			WebhookID:   webhook.ID,
@@ -1020,11 +1020,18 @@ func (h *Handlers) processTargetCreate(
 ) {
 	// The body size cap is enforced by the MaxBodySize middleware,
 	// which runs before CSRF parses the form.
-	name := r.FormValue("name")
-	targetType := database.TargetType(r.FormValue("type"))
-	targetURL := r.FormValue("url")
-	maxRetriesStr := r.FormValue("max_retries")
-	expiry := r.FormValue("expiry")
+	//
+	// Every field here is read with PostFormValue, not FormValue.
+	// FormValue falls back to the query string, which would let
+	// `POST /source/{id}/targets?url=https://hooks.slack.com/...`
+	// configure a target from a value the request line carries — and
+	// the request line, unlike the body, is what logs, proxies,
+	// Referer headers and error trackers record.
+	name := r.PostFormValue("name")
+	targetType := database.TargetType(r.PostFormValue("type"))
+	targetURL := r.PostFormValue("url")
+	maxRetriesStr := r.PostFormValue("max_retries")
+	expiry := r.PostFormValue("expiry")
 
 	if name == "" {
 		http.Error(

internal/handlers/target_create_query_test.go

@@ -0,0 +1,206 @@
+package handlers_test
+
+import (
+	"bytes"
+	"context"
+	"log/slog"
+	"net/http"
+	"net/http/httptest"
+	"net/url"
+	"strings"
+	"testing"
+
+	"github.com/go-chi/chi"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/webhooker/internal/config"
+	"sneak.berlin/go/webhooker/internal/database"
+	"sneak.berlin/go/webhooker/internal/middleware"
+)
+
+// targetSecretSegments are the path segments of an incoming-webhook
+// URL. For Slack, Discord and Teams the path IS the bearer credential,
+// so this string must not reach storage or the access log by way of
+// the request line.
+const targetSecretSegments = "T00000000/B00000000/QQTARGETSECRETQQ"
+
+// targetSecretURL is a destination whose secret lives in its path. It
+// uses a literal public address rather than a hostname so the SSRF
+// check resolves nothing: with a hostname, a sandbox without DNS would
+// reject the URL for the wrong reason and the test would pass even
+// with the defect reintroduced.
+const targetSecretURL = "https://93.184.216.34/services/" +
+	targetSecretSegments
+
+// targetsForWebhook returns every target stored against a webhook.
+func targetsForWebhook(
+	t *testing.T,
+	db *database.Database,
+	webhookID string,
+) []database.Target {
+	t.Helper()
+
+	var targets []database.Target
+
+	require.NoError(
+		t,
+		db.DB().Where("webhook_id = ?", webhookID).
+			Find(&targets).Error,
+	)
+
+	return targets
+}
+
+// postTargetCreate drives HandleTargetCreate through the production
+// access-log middleware and a chi route, so the logged url field is
+// produced exactly as it ships, and returns the recorder plus the
+// captured log.
+func postTargetCreate(
+	t *testing.T,
+	env *sourceTestEnv,
+	webhookID string,
+	query string,
+	form url.Values,
+) (*httptest.ResponseRecorder, string) {
+	t.Helper()
+
+	logBuf := new(bytes.Buffer)
+	mw := middleware.NewForTest(
+		slog.New(slog.NewJSONHandler(
+			logBuf, &slog.HandlerOptions{Level: slog.LevelInfo},
+		)),
+		&config.Config{Environment: config.EnvironmentDev},
+		nil,
+	)
+
+	router := chi.NewRouter()
+	router.Use(mw.Logging())
+	router.Post(
+		"/source/{sourceID}/targets",
+		env.handlers.HandleTargetCreate(),
+	)
+
+	target := "/source/" + webhookID + "/targets"
+	if query != "" {
+		target += "?" + query
+	}
+
+	body := ""
+	if form != nil {
+		body = form.Encode()
+	}
+
+	req := httptest.NewRequestWithContext(
+		context.Background(),
+		http.MethodPost,
+		target,
+		strings.NewReader(body),
+	)
+	req.Header.Set(
+		"Content-Type", "application/x-www-form-urlencoded",
+	)
+
+	for _, c := range env.cookies {
+		req.AddCookie(c)
+	}
+
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	return w, logBuf.String()
+}
+
+// TestHandleTargetCreate_QueryStringURLDoesNotConfigureATarget is the
+// regression test for the ingress leak. r.FormValue falls back to the
+// query string when a field is absent from the POST body, so
+//
+//	POST /source/{id}/targets?url=https://hooks.slack.com/services/...
+//
+// with an empty url field used to create a working target from a value
+// carried on the request line — where logs, proxies, Referer headers
+// and error trackers record it. The handler reads the body only, so
+// the request is rejected for a missing URL and stores nothing.
+//
+// name and type are sent in the BODY on purpose: the request has to
+// get past those two validations for the assertion to be about the url
+// read specifically.
+func TestHandleTargetCreate_QueryStringURLDoesNotConfigureATarget(
+	t *testing.T,
+) {
+	t.Parallel()
+
+	env := setupSourceTest(t)
+	webhook := seedWebhookWithRetention(t, env.db, 30)
+
+	body := url.Values{}
+	body.Set("name", "leaky")
+	body.Set("type", string(database.TargetTypeSlack))
+
+	w, logged := postTargetCreate(
+		t, env, webhook.ID,
+		"url="+url.QueryEscape(targetSecretURL),
+		body,
+	)
+
+	assert.Equal(t, http.StatusBadRequest, w.Code)
+
+	targets := targetsForWebhook(t, env.db, webhook.ID)
+	assert.Empty(
+		t, targets,
+		"a query-string value must not populate a target config",
+	)
+
+	assert.NotContains(t, logged, targetSecretSegments)
+	assert.NotContains(t, logged, "93.184.216.34")
+	assert.NotEmpty(t, logged, "the access log line must still be written")
+}
+
+// TestHandleTargetCreate_BodyURLStillCreatesTheTarget is the positive
+// control for the test above: the rejection has to come from where the
+// value was read, not from the handler being broken.
+func TestHandleTargetCreate_BodyURLStillCreatesTheTarget(t *testing.T) {
+	t.Parallel()
+
+	env := setupSourceTest(t)
+	webhook := seedWebhookWithRetention(t, env.db, 30)
+
+	body := url.Values{}
+	body.Set("name", "legit")
+	body.Set("type", string(database.TargetTypeSlack))
+	body.Set("url", targetSecretURL)
+
+	w, logged := postTargetCreate(t, env, webhook.ID, "", body)
+
+	assert.Equal(t, http.StatusSeeOther, w.Code)
+
+	targets := targetsForWebhook(t, env.db, webhook.ID)
+	require.Len(t, targets, 1)
+	assert.Contains(t, targets[0].Config, targetSecretSegments)
+
+	// The body carried the credential, so the access log must still
+	// not have it: the log records the request line only.
+	assert.NotContains(t, logged, targetSecretSegments)
+}
+
+// TestHandleTargetCreate_QueryStringCannotSupplyNameOrType covers the
+// rest of the converted reads on this handler in one request: with an
+// empty body, nothing the query carries is visible to it.
+func TestHandleTargetCreate_QueryStringCannotSupplyNameOrType(
+	t *testing.T,
+) {
+	t.Parallel()
+
+	env := setupSourceTest(t)
+	webhook := seedWebhookWithRetention(t, env.db, 30)
+
+	w, _ := postTargetCreate(
+		t, env, webhook.ID,
+		"name=leaky&type=slack&max_retries=9&expiry=30d&url="+
+			url.QueryEscape(targetSecretURL),
+		url.Values{},
+	)
+
+	assert.Equal(t, http.StatusBadRequest, w.Code)
+	assert.Contains(t, w.Body.String(), "Name is required")
+	assert.Empty(t, targetsForWebhook(t, env.db, webhook.ID))
+}

internal/handlers/webhook.go

@@ -9,6 +9,7 @@ import (
 	"gorm.io/gorm"
 	"sneak.berlin/go/webhooker/internal/database"
 	"sneak.berlin/go/webhooker/internal/delivery"
+	"sneak.berlin/go/webhooker/internal/logfield"
 )
 
 const (
@@ -125,9 +126,16 @@ func (h *Handlers) lookupEntrypoint(
 		"path = ?", entrypointUUID,
 	).First(&entrypoint)
 	if result.Error != nil {
+		// The receiver is unauthenticated and /webhook/{uuid}
+		// matches any single segment, so this value is entirely
+		// client-chosen on exactly the branch where the lookup
+		// failed. DEBUG is off by default; the cap is what keeps
+		// turning it on from restoring an unbounded write.
 		h.log.Debug(
 			"entrypoint not found",
-			"path", entrypointUUID,
+			"path", logfield.Truncate(
+				entrypointUUID, logfield.MaxBytes,
+			),
 		)
 		http.NotFound(w, r)
 

internal/logfield/logfield.go

@@ -0,0 +1,143 @@
+// Package logfield bounds the client-supplied values this service
+// writes into its logs.
+//
+// Any log field whose content a client picks is spent against a budget
+// here, in ENCODED bytes rather than in the bytes the client sent, so
+// that escaping cannot multiply a field past its nominal size. One
+// budget and one implementation serves the access log in
+// internal/middleware and every other slog call that reaches a
+// client-chosen path, header or form value; a second, ad-hoc
+// truncation somewhere else in the tree is the thing this package
+// exists to prevent.
+package logfield
+
+import (
+	"strings"
+	"unicode"
+	"unicode/utf8"
+)
+
+const (
+	// MaxBytes is the default budget for a log field whose value the
+	// client supplies outright: a URL, a path, a header, a form value.
+	// The budget is spent in ENCODED bytes (see Truncate), so 512 still
+	// holds a real browser's User-Agent whole — those are plain ASCII,
+	// which encodes one byte for one — while a value built from
+	// characters the encoder escapes keeps a shorter prefix. That is
+	// the intended trade: 500 quotation marks are not a debugging
+	// asset.
+	MaxBytes = 512
+
+	// TruncationMarker is appended to any field that was cut, so a
+	// short value and a truncated one cannot be confused. It is charged
+	// on top of the budget, not inside it.
+	TruncationMarker = "[truncated]"
+)
+
+// EncodedBytes is what r costs on the line once the log handler has
+// escaped it, taking the worse of the two handlers internal/logger
+// configures.
+//
+// slog's JSON handler escapes quote, backslash, newline, carriage
+// return and tab to two bytes each, and every other C0 control plus
+// LINE SEPARATOR and PARAGRAPH SEPARATOR to a six-byte \u escape; it
+// passes every other rune through as its own UTF-8. Its text handler
+// quotes with strconv.Quote, which spells a non-printable rune below
+// U+10000 as \uXXXX but one at or above U+10000 as \UXXXXXXXX — ten
+// bytes, not six. The text handler is therefore the worse of the two
+// for every non-printable rune, and by four bytes apiece for the
+// 955,086 unassigned, private-use and format code points on planes 1
+// to 16.
+//
+// Charging ten there is what makes the stated line ceilings hold for
+// the tty handler as well: U+1000C encodes as F0 90 80 8C, every byte
+// >= 0x80, which httpguts.ValidHeaderFieldValue accepts and
+// net/textproto does not strip, so a header can be filled with them.
+//
+// Both handlers pass printable runes through as their own UTF-8, so
+// unicode.IsPrint separates the escaped cases from the plain ones for
+// either handler.
+func EncodedBytes(r rune) int {
+	const (
+		// A backslash and the character itself.
+		shortEscapeBytes = 2
+		// \uXXXX, which is also the width of \u00XX.
+		escapedRuneBytes = 6
+		// \UXXXXXXXX, strconv.Quote's spelling of a non-printable
+		// rune outside the basic multilingual plane.
+		escapedAstralRuneBytes = 10
+		// The first code point strconv.Quote spells with \U.
+		firstAstralRune = 0x10000
+	)
+
+	switch {
+	case r == '"' || r == '\\' || r == '\n' || r == '\r' || r == '\t':
+		return shortEscapeBytes
+	case !unicode.IsPrint(r) && r >= firstAstralRune:
+		return escapedAstralRuneBytes
+	case !unicode.IsPrint(r):
+		return escapedRuneBytes
+	default:
+		return utf8.RuneLen(r)
+	}
+}
+
+// Truncate caps s at maxBytes of ENCODED output, marking the value
+// when it cuts.
+//
+// Budgeting raw bytes would not bound the line. Escaping only ever
+// grows a value, so a raw budget spent on characters the encoder
+// escapes buys a field several times its nominal size — and the line
+// is the thing an operator is told to multiply by their request rate.
+// Charging each rune what it will actually cost is what makes a stated
+// ceiling true rather than merely larger. The visible consequence is
+// that an escape-heavy value keeps a shorter prefix than a plain one,
+// which is the correct trade.
+//
+// The result is always valid UTF-8. A cut on a byte boundary can split
+// a multi-byte rune, and a header can carry bytes that were never
+// valid UTF-8 to begin with; both are dropped rather than kept, since
+// an encoder would otherwise spend six bytes replacing each one.
+func Truncate(s string, maxBytes int) string {
+	// No rune encodes to fewer bytes than it occupies, so nothing past
+	// maxBytes raw can fit the budget. Slicing first bounds the scan
+	// below to the budget rather than to the size of the header the
+	// client sent.
+	window, cut := s, false
+	if len(window) > maxBytes {
+		window, cut = window[:maxBytes], true
+	}
+
+	var (
+		kept  strings.Builder
+		spent int
+	)
+
+	for i := 0; i < len(window); {
+		r, size := utf8.DecodeRuneInString(window[i:])
+		if r == utf8.RuneError && size == 1 {
+			i += size
+
+			continue
+		}
+
+		cost := EncodedBytes(r)
+		if spent+cost > maxBytes {
+			cut = true
+
+			break
+		}
+
+		spent += cost
+
+		kept.WriteString(window[i : i+size])
+
+		i += size
+	}
+
+	if !cut {
+		return kept.String()
+	}
+
+	return kept.String() + TruncationMarker
+}

internal/logfield/logfield_test.go

@@ -0,0 +1,321 @@
+package logfield_test
+
+import (
+	"bytes"
+	"io"
+	"log/slog"
+	"strings"
+	"testing"
+	"unicode/utf8"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/webhooker/internal/logfield"
+)
+
+// budget is the field budget these tests spend. Small enough that a
+// cut is unambiguous, large enough to hold several runes of every
+// width.
+const budget = 64
+
+// sampleRunes is how many runes wide the values in the charge test
+// are. The handlers add a constant per field — a pair of quotes when
+// the value needs quoting — so the per-rune charge is only visible
+// once it is amortised over a run of them.
+const sampleRunes = 64
+
+// quotingSlack is that constant: the pair of quotes a handler adds to
+// a value that needs them and omits from one that does not.
+const quotingSlack = 2
+
+// newHandlers are the two handlers internal/logger can install. Time
+// is dropped so a line's width is a function of its value alone —
+// RFC3339Nano trims trailing zeros, so two consecutive timestamps do
+// not render to the same number of bytes.
+func newHandlers() map[string]func(io.Writer) slog.Handler {
+	opts := &slog.HandlerOptions{
+		Level: slog.LevelDebug,
+		ReplaceAttr: func(_ []string, a slog.Attr) slog.Attr {
+			if a.Key == slog.TimeKey {
+				return slog.Attr{}
+			}
+
+			return a
+		},
+	}
+
+	return map[string]func(io.Writer) slog.Handler{
+		"json": func(w io.Writer) slog.Handler {
+			return slog.NewJSONHandler(w, opts)
+		},
+		"text": func(w io.Writer) slog.Handler {
+			return slog.NewTextHandler(w, opts)
+		},
+	}
+}
+
+// renderedWidth is the number of bytes a handler writes for a line
+// carrying value in a single attribute.
+func renderedWidth(
+	newHandler func(io.Writer) slog.Handler,
+	value string,
+) int {
+	buf := new(bytes.Buffer)
+	slog.New(newHandler(buf)).Info("m", "v", value)
+
+	return buf.Len()
+}
+
+// chargeTestRunes is the set of code points the charge test measures:
+// every rune in the first two planes' worth of the BMP that the
+// handlers are most likely to treat specially, the separators that
+// only slog's JSON handler escapes, and a stratified sample across
+// the rest of Unicode so the astral charge is exercised on more than
+// one hand-picked rune.
+func chargeTestRunes() []rune {
+	const (
+		denseCeiling = 0x800
+		stride       = 1021
+		surrogateLo  = 0xD800
+		surrogateHi  = 0xDFFF
+	)
+
+	var runes []rune
+
+	keep := func(r rune) {
+		if r >= surrogateLo && r <= surrogateHi {
+			return
+		}
+
+		runes = append(runes, r)
+	}
+
+	for r := range rune(denseCeiling) {
+		keep(r)
+	}
+
+	for _, r := range []rune{
+		0x2028, 0x2029, 0x200B, 0x4E00, 0xE000, 0xFFFD,
+		0x1000C, 0x1F600, 0xE0001, 0x10FFFF,
+	} {
+		keep(r)
+	}
+
+	for r := rune(denseCeiling); r <= utf8.MaxRune; r += stride {
+		keep(r)
+	}
+
+	return runes
+}
+
+// TestEncodedBytes_ChargesAtLeastWhatTheHandlersEmit is the property
+// the whole capping scheme rests on: a rune may not cost more on the
+// line than the budget was charged for it. An undercharged rune is
+// how a stated ceiling becomes false without any test noticing, so
+// the charge is measured against what the handlers actually write
+// rather than against the escaping rules as read.
+func TestEncodedBytes_ChargesAtLeastWhatTheHandlersEmit(t *testing.T) {
+	t.Parallel()
+
+	for name, newHandler := range newHandlers() {
+		t.Run(name, func(t *testing.T) {
+			t.Parallel()
+
+			// 'a' is a printable ASCII rune, charged exactly one
+			// byte, so it is the zero point the other runes are
+			// measured against.
+			base := renderedWidth(
+				newHandler, strings.Repeat("a", sampleRunes),
+			)
+
+			for _, r := range chargeTestRunes() {
+				got := renderedWidth(
+					newHandler,
+					strings.Repeat(string(r), sampleRunes),
+				)
+				charged := sampleRunes *
+					(logfield.EncodedBytes(r) - 1)
+
+				require.LessOrEqual(
+					t, got-base, charged+quotingSlack,
+					"U+%04X costs more on the line than "+
+						"EncodedBytes charges for it",
+					r,
+				)
+			}
+		})
+	}
+}
+
+// TestTruncate_SpendsNoMoreThanTheBudget holds the result to the
+// budget in ENCODED bytes, which is the unit the budget is stated in.
+// A raw-byte cap passes the ASCII case here and fails every other
+// one.
+func TestTruncate_SpendsNoMoreThanTheBudget(t *testing.T) {
+	t.Parallel()
+
+	for name, fill := range map[string]string{
+		"plain":     "x",
+		"quote":     `"`,
+		"backslash": `\`,
+		"tab":       "\t",
+		"newline":   "\n",
+		"control":   "\x01",
+		"astral":    "\U0001000C",
+		// U+4E00, a printable multi-byte rune, charged its three
+		// UTF-8 bytes rather than an escape. Spelled numerically
+		// because gosmopolitan rejects Han in a string literal.
+		"cjk": string(rune(0x4E00)),
+	} {
+		t.Run(name, func(t *testing.T) {
+			t.Parallel()
+
+			got := logfield.Truncate(
+				strings.Repeat(fill, budget*8), budget,
+			)
+
+			require.True(
+				t, strings.HasSuffix(
+					got, logfield.TruncationMarker,
+				),
+				"an oversized value must be marked as cut",
+			)
+
+			spent := 0
+			for _, r := range strings.TrimSuffix(
+				got, logfield.TruncationMarker,
+			) {
+				spent += logfield.EncodedBytes(r)
+			}
+
+			assert.LessOrEqual(t, spent, budget)
+			assert.True(t, utf8.ValidString(got))
+		})
+	}
+}
+
+// TestTruncate_LeavesShortValuesAlone keeps the marker meaningful: a
+// value that fits comes back byte for byte, so a marked value is
+// always a cut one.
+func TestTruncate_LeavesShortValuesAlone(t *testing.T) {
+	t.Parallel()
+
+	for _, s := range []string{
+		"", "GET", "/source/abc/edit", "Mozilla/5.0 (X11)",
+	} {
+		assert.Equal(t, s, logfield.Truncate(s, budget))
+	}
+}
+
+// TestTruncate_DropsInvalidUTF8 covers the bytes a header can carry
+// that were never valid UTF-8. Keeping them would make the encoder
+// spend six bytes apiece replacing them, which is exactly the
+// amplification the budget exists to prevent.
+func TestTruncate_DropsInvalidUTF8(t *testing.T) {
+	t.Parallel()
+
+	got := logfield.Truncate("a\xffb\xfe\xfec", budget)
+
+	assert.Equal(t, "abc", got)
+	assert.True(t, utf8.ValidString(got))
+}
+
+// encodedCost is what a whole string costs on a line, by the same
+// accounting Truncate spends its budget with.
+func encodedCost(s string) int {
+	total := 0
+	for _, r := range s {
+		total += logfield.EncodedBytes(r)
+	}
+
+	return total
+}
+
+// TestTruncate_SpendsEncodedBytesNotRawBytes is the zero-headroom
+// version of TestTruncate_SpendsNoMoreThanTheBudget above, and of the
+// line-length assertions elsewhere.
+//
+// A line ceiling has slack in it by construction, and a LessOrEqual
+// against the budget cannot tell a budget spent exactly from one
+// spent under. Here the budget is checked against exactly what it
+// bought: a value built from a single rune must keep exactly
+// MaxBytes/EncodedBytes(r) of them, with nothing spare. A raw-byte
+// budget — cost := utf8.RuneLen(r) — fails this for every rune the
+// handlers escape.
+func TestTruncate_SpendsEncodedBytesNotRawBytes(t *testing.T) {
+	t.Parallel()
+
+	for name, r := range map[string]rune{
+		"plain":           'x',
+		"quote":           '"',
+		"backslash":       '\\',
+		"tab":             '\t',
+		"newline":         '\n',
+		"carriage_return": '\r',
+		"c0_control":      '\x01',
+		"del":             '\x7f',
+		// U+2028 LINE SEPARATOR, which only the JSON handler
+		// escapes.
+		"line_separator":      '
',
+		"astral_nonprintable": '\U0001000C',
+		"multibyte_printable": 'é',
+		// U+20AC, a three-byte printable rune, charged its own
+		// UTF-8 bytes rather than an escape.
+		"three_byte_printable": '€',
+		"emoji_printable":      '\U0001F600',
+	} {
+		t.Run(name, func(t *testing.T) {
+			t.Parallel()
+
+			cost := logfield.EncodedBytes(r)
+			want := logfield.MaxBytes / cost
+
+			// Far past the budget under either accounting.
+			in := strings.Repeat(string(r), logfield.MaxBytes*2)
+
+			got := logfield.Truncate(in, logfield.MaxBytes)
+
+			require.True(
+				t, strings.HasSuffix(
+					got, logfield.TruncationMarker,
+				),
+				"a value past the budget must be marked",
+			)
+
+			kept := strings.TrimSuffix(
+				got, logfield.TruncationMarker,
+			)
+
+			assert.Equal(
+				t, want, utf8.RuneCountInString(kept),
+				"budget bought the wrong number of runes at "+
+					"%d encoded bytes each", cost,
+			)
+			assert.LessOrEqual(
+				t, encodedCost(kept), logfield.MaxBytes,
+			)
+		})
+	}
+}
+
+// TestTruncate_NeverSplitsARune covers a cut landing inside a
+// multi-byte encoding rather than between two of them.
+func TestTruncate_NeverSplitsARune(t *testing.T) {
+	t.Parallel()
+
+	// U+20AC, three bytes and printable, so a small budget lands
+	// inside an encoding rather than on a boundary.
+	in := strings.Repeat("€", logfield.MaxBytes)
+
+	for b := 1; b <= 16; b++ {
+		got := strings.TrimSuffix(
+			logfield.Truncate(in, b), logfield.TruncationMarker,
+		)
+
+		assert.True(
+			t, utf8.ValidString(got),
+			"budget %d produced invalid UTF-8", b,
+		)
+		assert.LessOrEqual(t, encodedCost(got), b)
+	}
+}

internal/middleware/accesslog_test.go

@@ -376,7 +376,7 @@ func lineSizeCases() map[string]sizeCase {
 
 	// The url field on a 5xx keeps the concrete path, so it reaches its
 	// own budget on the same line as the three header fields. That is
-	// the widest line the service can be made to write.
+	// the widest access log line the service can be made to write.
 	longPath := "/boom/" + strings.Repeat("x", oversizedSegmentBytes)
 	wantLongURL := longPath[:maxFieldBytes] + truncationSuffix
 

internal/middleware/csrf.go

@@ -4,6 +4,7 @@ import (
 	"net/http"
 
 	"github.com/gorilla/csrf"
+	"sneak.berlin/go/webhooker/internal/logfield"
 )
 
 // CSRFToken retrieves the CSRF token from the request context.
@@ -42,9 +43,22 @@ func isClientTLS(r *http.Request) bool {
 // csrf.Secure option is set at creation time, not per-request.
 func (m *Middleware) CSRF() func(http.Handler) http.Handler {
 	csrfErrorHandler := http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		// CSRF is registered ahead of RequireAuth on every route
+		// group that uses it, so this WARN is reachable by an
+		// unauthenticated client: a POST with no token to
+		// /source/<any length of any text>/edit lands here. The
+		// method and path are capped against the same budgets as
+		// the access log. remote_addr is set by net/http from the
+		// accepted connection rather than by the client, and
+		// csrf.FailureReason returns one of gorilla/csrf's own
+		// fixed error values, so neither is client-sized.
 		m.log.Warn("csrf: token validation failed",
-			"method", r.Method,
-			"path", r.URL.Path,
+			"method", logfield.Truncate(
+				r.Method, maxLogMethodBytes,
+			),
+			"path", logfield.Truncate(
+				r.URL.Path, logfield.MaxBytes,
+			),
 			"remote_addr", r.RemoteAddr,
 			"reason", csrf.FailureReason(r),
 		)

internal/middleware/export_test.go

@@ -1,7 +1,9 @@
 package middleware
 
 import (
+	"context"
 	"net/http"
+	"time"
 )
 
 // NewLoggingResponseWriterForTest wraps newLoggingResponseWriter
@@ -35,9 +37,79 @@ func IsClientTLS(r *http.Request) bool {
 	return isClientTLS(r)
 }
 
-// LoginRateLimitConst exposes the loginRateLimit constant.
+// LoginRateLimitConst exposes the loginRateLimit constant: the
+// number of FAILED login attempts one client may make against one
+// submitted username per interval.
 const LoginRateLimitConst = loginRateLimit
 
+// LoginFailureMaxKeysConst exposes the cap on each of the login
+// guard's key sets.
+const LoginFailureMaxKeysConst = loginFailureMaxKeys
+
+// PasswordVerifyConcurrencyConst exposes the bound on concurrent
+// Argon2id verifications.
+const PasswordVerifyConcurrencyConst = passwordVerifyConcurrency
+
+// PasswordVerifyMaxWaitersConst exposes the bound on how many
+// requests may queue for a verification slot.
+const PasswordVerifyMaxWaitersConst = passwordVerifyMaxWaiters
+
+// LoginGuard is the login failure counter and verification
+// semaphore, exposed for direct testing.
+type LoginGuard = loginGuard
+
+// NewLoginGuardForTest builds a guard with test-sized parameters.
+func NewLoginGuardForTest(
+	limit int,
+	interval time.Duration,
+	maxKeys, concurrency, maxWaiters int,
+	wait time.Duration,
+) *LoginGuard {
+	return newLoginGuard(
+		limit, interval, maxKeys, concurrency, maxWaiters, wait,
+	)
+}
+
+// QueuedWaitersForTest reports how many requests are currently
+// queued for a verification slot.
+func (g *LoginGuard) QueuedWaitersForTest() int {
+	return len(g.queue)
+}
+
+// SetNowForTest replaces the guard's clock.
+func (g *LoginGuard) SetNowForTest(now func() time.Time) {
+	g.mu.Lock()
+	defer g.mu.Unlock()
+
+	g.now = now
+}
+
+// FailForTest exposes fail.
+func (g *LoginGuard) FailForTest(clientKey, username string) bool {
+	return g.fail(clientKey, username)
+}
+
+// SucceedForTest exposes succeed.
+func (g *LoginGuard) SucceedForTest(clientKey, username string) {
+	g.succeed(clientKey, username)
+}
+
+// AcquireForTest exposes acquire.
+func (g *LoginGuard) AcquireForTest(
+	ctx context.Context,
+) (func(), bool) {
+	return g.acquire(ctx)
+}
+
+// TrackedKeysForTest reports how many failure counters the guard
+// holds, per-username and per-address respectively.
+func (g *LoginGuard) TrackedKeysForTest() (int, int) {
+	g.mu.Lock()
+	defer g.mu.Unlock()
+
+	return len(g.byUser), len(g.byAddr)
+}
+
 // PasswordChangeRateLimitConst exposes the
 // passwordChangeRateLimit constant.
 const PasswordChangeRateLimitConst = passwordChangeRateLimit

internal/middleware/logbound_test.go

@@ -0,0 +1,544 @@
+package middleware_test
+
+// This file covers the log lines OUTSIDE the access log that carry a
+// client-chosen value. accesslog_test.go bounds the one INFO line the
+// Logging middleware writes; these are the separate slog calls that
+// were never in that sweep and so never got the budget:
+//
+//   - MaxBodySize's 413 rejection, at WARN, registered ahead of
+//     RequireAuth and therefore reachable unauthenticated at a URL of
+//     the client's choosing.
+//   - CSRF's 403 rejection, at WARN, also registered ahead of
+//     RequireAuth.
+//   - The rate limiters' 429 rejection, at WARN, on the
+//     unauthenticated receiver among others.
+//   - RequireAuth's own unauthenticated-request line, at DEBUG.
+//   - RecordLoginFailure's throttle rejection, at WARN. Its cap is
+//     defensive rather than load-bearing today: chi pins the one
+//     route that calls it to the constant path "/pages/login". The
+//     method is exported and takes any *http.Request, so the test
+//     below hands it the request a caller on a parameterised route
+//     would, which is what the cap exists for.
+//
+// Every case here holds the ENCODED line to
+// middleware.MaxAccessLogLineBytes, under both handlers
+// internal/logger can install, against 8 KB of client-chosen text
+// built out of the characters those handlers escape. A budget spent
+// in raw bytes passes the plain-ASCII cases and fails the rest.
+
+import (
+	"bytes"
+	"context"
+	"io"
+	"log/slog"
+	"net/http"
+	"net/http/httptest"
+	"net/url"
+	"strings"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/webhooker/internal/config"
+	"sneak.berlin/go/webhooker/internal/middleware"
+)
+
+// bodyLimitBytes is the MaxBodySize cap these tests install. Any
+// declared Content-Length above it takes the 413 branch.
+const bodyLimitBytes = 1024
+
+// declaredBodyBytes is the Content-Length an oversize request
+// declares. Nothing is actually sent: the 413 branch fires off the
+// declaration alone, which is what makes the attack free.
+const declaredBodyBytes = bodyLimitBytes * 2
+
+// receiverLimitPerMinute is the per-entrypoint receiver limit these
+// tests install. The aggregate limiter sits at ten times this, so a
+// flood stays under it and the rejections come from the
+// per-entrypoint limiter, which is the one that logs the path.
+const receiverLimitPerMinute = 8
+
+// escapeFills are the characters a client can put in a request that
+// the log handlers then escape, coming out wider than they went in.
+// A budget counted in raw bytes lets any of them buy a field several
+// times its nominal size.
+//
+// U+1000C is the case the JSON handler alone does not reach: it is
+// unassigned, so it is non-printable, and strconv.Quote spells a
+// non-printable rune at or above U+10000 as a ten-byte \UXXXXXXXX
+// while the JSON handler passes its four UTF-8 bytes through. Only
+// the text-handler shape of these tests holds that charge honest.
+func escapeFills() map[string]string {
+	return map[string]string{
+		"plain":     "x",
+		"quote":     `"`,
+		"backslash": `\`,
+		"tab":       "\t",
+		"newline":   "\n",
+		// A C0 control neither handler has a short escape for, so
+		// each one costs six bytes on the line against the single
+		// byte it cost to send. This is the widest multiplier a
+		// client can drive, and the case a raw-byte budget breaks
+		// on first.
+		//
+		// This fill is load-bearing, not decoration. Budgeting raw
+		// bytes instead of encoded is caught by this fill alone,
+		// and only under the JSON handler, at 3,072 bytes against
+		// the 2,560 ceiling. Drop it and that mutation passes.
+		"control": "\x01",
+		"astral":  "\U0001000C",
+	}
+}
+
+// logHandlers are the two handlers internal/logger can install: the
+// JSON one, and the text one it selects when stderr is a tty. They do
+// not escape alike, and MaxAccessLogLineBytes is quoted unqualified,
+// so every case runs through both.
+func logHandlers() map[string]func(
+	io.Writer, *slog.HandlerOptions,
+) slog.Handler {
+	return map[string]func(
+		io.Writer, *slog.HandlerOptions,
+	) slog.Handler{
+		"json": func(
+			w io.Writer, o *slog.HandlerOptions,
+		) slog.Handler {
+			return slog.NewJSONHandler(w, o)
+		},
+		"text": func(
+			w io.Writer, o *slog.HandlerOptions,
+		) slog.Handler {
+			return slog.NewTextHandler(w, o)
+		},
+	}
+}
+
+// oversizedPathSegment builds an 8 KB client-chosen path segment out
+// of repetitions of ch, percent-encoded so it survives URL parsing
+// into r.URL.Path the way it would arriving off a socket.
+//
+// Both markers sit at the END, past every budget, so their absence
+// from the log is what proves the value was cut rather than merely
+// being short. The leading 'x' keeps the segment non-empty for fills
+// that a parser might otherwise fold away.
+func oversizedPathSegment(ch string) string {
+	return url.PathEscape(
+		"x" + strings.Repeat(ch, oversizedSegmentBytes) +
+			attackerMarker + tailMarker,
+	)
+}
+
+// capturingLogger returns a logger at DEBUG writing into the returned
+// buffer through the named handler.
+func capturingLogger(
+	newHandler func(io.Writer, *slog.HandlerOptions) slog.Handler,
+) (*slog.Logger, *bytes.Buffer) {
+	buf := new(bytes.Buffer)
+	opts := &slog.HandlerOptions{Level: slog.LevelDebug}
+
+	return slog.New(newHandler(buf, opts)), buf
+}
+
+// capturingBoundMiddleware builds a Middleware with a real session
+// manager (CSRF needs its key, RequireAuth needs its store) whose log
+// is captured at DEBUG.
+func capturingBoundMiddleware(
+	t *testing.T,
+	newHandler func(io.Writer, *slog.HandlerOptions) slog.Handler,
+) (*middleware.Middleware, *bytes.Buffer) {
+	t.Helper()
+
+	log, buf := capturingLogger(newHandler)
+
+	cfg := &config.Config{
+		Environment:       config.EnvironmentDev,
+		ReceiverRateLimit: receiverLimitPerMinute,
+	}
+
+	sess := newTestSessionManager(cfg, log, nil)
+
+	return middleware.NewForTest(log, cfg, sess), buf
+}
+
+// unreachable is a next-handler that fails the test if the middleware
+// under test let the request through. Every site here rejects.
+func unreachable(t *testing.T) http.Handler {
+	t.Helper()
+
+	return http.HandlerFunc(func(http.ResponseWriter, *http.Request) {
+		assert.Fail(t, "rejected request reached the next handler")
+	})
+}
+
+// logSite is one non-access-log call site that logs a client-chosen
+// path. drive sends requests at it that all take the rejecting
+// branch; linesPerRequest is how many log lines one such request
+// produces there.
+type logSite struct {
+	// build wraps the site's middleware around a handler that must
+	// not be reached.
+	build func(
+		t *testing.T, m *middleware.Middleware,
+	) http.Handler
+	// send issues one request for the given client-chosen path and
+	// returns the status. Some sites need a warm-up request before
+	// they reject, which send performs itself.
+	send func(h http.Handler, path string) int
+	// wantStatus is the status the rejecting branch answers with.
+	wantStatus int
+}
+
+// postOversize sends a POST whose declared Content-Length exceeds the
+// body limit without sending a body, which is the whole cost of the
+// attack on the MaxBodySize branch.
+func postOversize(h http.Handler, path string) int {
+	req := httptest.NewRequestWithContext(
+		context.Background(), http.MethodPost, path, nil,
+	)
+	req.ContentLength = declaredBodyBytes
+	req.Header.Set(
+		"Content-Type", "application/x-www-form-urlencoded",
+	)
+
+	w := httptest.NewRecorder()
+	h.ServeHTTP(w, req)
+
+	return w.Code
+}
+
+// postNoToken sends a POST carrying no CSRF token and no session
+// cookie, which is what an unauthenticated client sends.
+func postNoToken(h http.Handler, path string) int {
+	req := httptest.NewRequestWithContext(
+		context.Background(), http.MethodPost, path,
+		strings.NewReader(""),
+	)
+	req.Header.Set(
+		"Content-Type", "application/x-www-form-urlencoded",
+	)
+
+	w := httptest.NewRecorder()
+	h.ServeHTTP(w, req)
+
+	return w.Code
+}
+
+// getNoSession sends a GET with no session cookie.
+func getNoSession(h http.Handler, path string) int {
+	req := httptest.NewRequestWithContext(
+		context.Background(), http.MethodGet, path, nil,
+	)
+
+	w := httptest.NewRecorder()
+	h.ServeHTTP(w, req)
+
+	return w.Code
+}
+
+// logSites enumerates the call sites under test.
+func logSites() map[string]logSite {
+	return map[string]logSite{
+		// The site this file exists for: WARN, on by default, and
+		// registered ahead of RequireAuth.
+		"maxbodysize 413": {
+			build: func(
+				t *testing.T, m *middleware.Middleware,
+			) http.Handler {
+				t.Helper()
+
+				return m.MaxBodySize(bodyLimitBytes)(
+					unreachable(t),
+				)
+			},
+			send:       postOversize,
+			wantStatus: http.StatusRequestEntityTooLarge,
+		},
+		// Also ahead of RequireAuth, also WARN.
+		"csrf 403": {
+			build: func(
+				t *testing.T, m *middleware.Middleware,
+			) http.Handler {
+				t.Helper()
+
+				return m.CSRF()(unreachable(t))
+			},
+			send:       postNoToken,
+			wantStatus: http.StatusForbidden,
+		},
+		// The per-entrypoint receiver limiter, unauthenticated. Its
+		// bucket is keyed on the path, so the first request through a
+		// fresh path is served and only the ones after it are
+		// rejected; sendUntilLimited absorbs that.
+		"receiver rate limit 429": {
+			build: func(
+				t *testing.T, m *middleware.Middleware,
+			) http.Handler {
+				t.Helper()
+
+				return m.ReceiverRateLimit()(okHandler())
+			},
+			send:       sendUntilLimited,
+			wantStatus: http.StatusTooManyRequests,
+		},
+		// RequireAuth's own line. DEBUG is off in production by
+		// default, but turning it on to diagnose a flood must not
+		// restore an unbounded write.
+		"requireauth redirect": {
+			build: func(
+				t *testing.T, m *middleware.Middleware,
+			) http.Handler {
+				t.Helper()
+
+				return m.RequireAuth()(unreachable(t))
+			},
+			send:       getNoSession,
+			wantStatus: http.StatusSeeOther,
+		},
+	}
+}
+
+// sendUntilLimited drives the per-entrypoint receiver limiter past
+// its allowance on one path and returns the status of the rejected
+// request. Every request before the last is served, and only the last
+// one logs.
+func sendUntilLimited(h http.Handler, path string) int {
+	code := http.StatusOK
+
+	for range receiverLimitPerMinute + 1 {
+		req := httptest.NewRequestWithContext(
+			context.Background(), http.MethodPost, path, nil,
+		)
+		req.RemoteAddr = "203.0.113.7:5555"
+
+		w := httptest.NewRecorder()
+		h.ServeHTTP(w, req)
+		code = w.Code
+	}
+
+	return code
+}
+
+// logLines splits the captured buffer into non-empty lines, holding
+// each to bound bytes.
+func logLines(t *testing.T, buf *bytes.Buffer, bound int) []string {
+	t.Helper()
+
+	var lines []string
+
+	for line := range strings.SplitSeq(
+		strings.TrimSpace(buf.String()), "\n",
+	) {
+		if line == "" {
+			continue
+		}
+
+		require.LessOrEqual(
+			t, len(line), bound,
+			"log line exceeded its bound: %s", line,
+		)
+
+		lines = append(lines, line)
+	}
+
+	return lines
+}
+
+// assertNoClientText fails if any marker from the far end of the
+// client-chosen input survived into the log. Their absence is what
+// distinguishes a real cut from a value that merely happened to be
+// short.
+func assertNoClientText(t *testing.T, buf *bytes.Buffer) {
+	t.Helper()
+
+	assert.NotContains(
+		t, buf.String(), attackerMarker,
+		"log carried attacker-chosen text",
+	)
+	assert.NotContains(
+		t, buf.String(), tailMarker,
+		"log carried the tail of the attacker-chosen text",
+	)
+}
+
+// TestLogLines_ClientChosenPathDoesNotSizeTheLine points 8 KB of
+// client-chosen path at each non-access-log call site that logs one,
+// through both handlers and through every character those handlers
+// escape, and holds the resulting line to MaxAccessLogLineBytes.
+//
+// Removing any one of the logfield.Truncate calls at those sites
+// fails this test: the line grows to roughly the size of the input,
+// or to several times it on the escaping fills.
+func TestLogLines_ClientChosenPathDoesNotSizeTheLine(t *testing.T) {
+	t.Parallel()
+
+	for siteName, site := range logSites() {
+		for handlerName, newHandler := range logHandlers() {
+			for fillName, fill := range escapeFills() {
+				name := siteName + "/" + handlerName + "/" + fillName
+
+				t.Run(name, func(t *testing.T) {
+					t.Parallel()
+
+					m, buf := capturingBoundMiddleware(
+						t, newHandler,
+					)
+
+					path := "/source/" +
+						oversizedPathSegment(fill) + "/edit"
+
+					assert.Equal(
+						t,
+						site.wantStatus,
+						site.send(site.build(t, m), path),
+					)
+
+					lines := logLines(
+						t, buf,
+						middleware.MaxAccessLogLineBytes,
+					)
+					require.NotEmpty(
+						t, lines,
+						"the site under test logged nothing, "+
+							"so the bound proves nothing",
+					)
+					assertNoClientText(t, buf)
+				})
+			}
+		}
+	}
+}
+
+// TestLoginThrottle_LogLineDoesNotTrackPathSize pins the cap on
+// RecordLoginFailure's "login failure limit exceeded" WARN line.
+//
+// That site does not fit logSites above: it is not a middleware
+// wrapping a handler but an exported method the login handler calls,
+// and the only route that calls it today is chi's static
+// "/pages/login", so no request through the mux can widen the line.
+// Driving the method directly is therefore the whole point rather
+// than a shortcut — it is exactly the call a second caller on a route
+// with a URL parameter would make, and without this test removing the
+// logfield.Truncate there fails nothing.
+func TestLoginThrottle_LogLineDoesNotTrackPathSize(t *testing.T) {
+	t.Parallel()
+
+	for handlerName, newHandler := range logHandlers() {
+		for fillName, fill := range escapeFills() {
+			t.Run(handlerName+"/"+fillName, func(t *testing.T) {
+				t.Parallel()
+
+				m, buf := capturingBoundMiddleware(
+					t, newHandler,
+				)
+
+				req := httptest.NewRequestWithContext(
+					context.Background(),
+					http.MethodPost,
+					"/source/"+
+						oversizedPathSegment(fill)+"/login",
+					nil,
+				)
+				req.RemoteAddr = "203.0.113.9:5555"
+
+				// The budget is spent per client and username,
+				// so one more failure than the budget allows is
+				// what takes the throttled branch.
+				var throttled bool
+				for range middleware.LoginRateLimitConst + 1 {
+					throttled = m.RecordLoginFailure(
+						req, "someone",
+					)
+				}
+
+				require.True(
+					t, throttled,
+					"the throttled branch never ran, so the "+
+						"bound proves nothing",
+				)
+
+				lines := logLines(
+					t, buf, middleware.MaxAccessLogLineBytes,
+				)
+				require.NotEmpty(t, lines)
+				assertNoClientText(t, buf)
+			})
+		}
+	}
+}
+
+// TestMaxBodySize_FloodOfOversizePathsDoesNotGrowTheLog is the
+// flood shape from the issue: an unauthenticated client posting
+// oversize declarations at invented 8 KB paths, as fast as it likes.
+//
+// It asserts the property directly rather than by proxy — the bytes
+// the flood writes to the operator's log do not track the bytes the
+// flood sent. The same flood at a one-character path is the control:
+// 8 KB of extra input per request buys at most the field budget, not
+// 8 KB of log.
+func TestMaxBodySize_FloodOfOversizePathsDoesNotGrowTheLog(
+	t *testing.T,
+) {
+	t.Parallel()
+
+	for handlerName, newHandler := range logHandlers() {
+		for fillName, fill := range escapeFills() {
+			t.Run(handlerName+"/"+fillName, func(t *testing.T) {
+				t.Parallel()
+
+				flood := func(segment func(i int) string) int {
+					m, buf := capturingBoundMiddleware(
+						t, newHandler,
+					)
+					h := m.MaxBodySize(bodyLimitBytes)(
+						unreachable(t),
+					)
+
+					for i := range floodRequests {
+						assert.Equal(
+							t,
+							http.StatusRequestEntityTooLarge,
+							postOversize(
+								h,
+								"/source/"+segment(i)+"/edit",
+							),
+						)
+					}
+
+					lines := logLines(
+						t, buf,
+						middleware.MaxAccessLogLineBytes,
+					)
+					require.Len(t, lines, floodRequests)
+					assertNoClientText(t, buf)
+
+					return buf.Len()
+				}
+
+				sent := oversizedSegmentBytes * floodRequests
+
+				oversize := flood(func(i int) string {
+					return oversizedPathSegment(fill) +
+						strings.Repeat("y", i)
+				})
+				control := flood(func(i int) string {
+					return "a" + strings.Repeat("y", i)
+				})
+
+				// The whole point: 8 KB per request of extra
+				// client-chosen input bought a bounded amount of
+				// log, not a proportional amount.
+				assert.Less(
+					t, oversize-control, sent/2,
+					"log volume tracked the size of the flood's "+
+						"input",
+				)
+				assert.LessOrEqual(
+					t,
+					oversize,
+					floodRequests*
+						middleware.MaxAccessLogLineBytes,
+				)
+			})
+		}
+	}
+}

internal/middleware/loginguard.go

@@ -0,0 +1,409 @@
+package middleware
+
+import (
+	"context"
+	"crypto/sha256"
+	"encoding/hex"
+	"net/http"
+	"sync"
+	"time"
+
+	"sneak.berlin/go/webhooker/internal/logfield"
+)
+
+const (
+	// loginFailureMaxKeys bounds how many distinct failure counters
+	// each of the guard's two key sets holds. The submitted username
+	// is part of a key, so the key set is attacker-influenced and
+	// needs a hard cap or the limiter becomes the memory
+	// amplification surface it exists to protect.
+	//
+	// A single-admin deployment has a handful of legitimate (client,
+	// username) pairs, so 1024 is three orders of magnitude of
+	// headroom before a real operator can be pushed onto the
+	// fallback. It costs little: a counter is a ~64-byte key string,
+	// a 32-byte window and map overhead, call it 170 bytes, so both
+	// key sets full is 2 * 1024 * 170 bytes, under 0.4 MB.
+	loginFailureMaxKeys = 1024
+
+	// passwordVerifyConcurrency bounds how many Argon2id
+	// verifications may run at once across every password-verifying
+	// endpoint. Because credentials are now verified before any
+	// limiter budget is spent, an attacker can force one hash per
+	// request, and each hash allocates argon2Memory — 64 MB. Two
+	// slots commit at most 128 MB to password hashing, which fits
+	// inside the smallest container this service is realistically
+	// given alongside its own working set; four would commit 256 MB
+	// and crowd it. A single-admin product needs no concurrent
+	// logins at all, so the second slot exists only so that one
+	// stalled request does not serialise the endpoint.
+	passwordVerifyConcurrency = 2
+
+	// passwordVerifyWait is how long a request waits for a
+	// verification slot before it is answered 503. Slots are handed
+	// out in arrival order, so a legitimate request queues behind
+	// the requests already waiting rather than behind the flood as a
+	// whole. The wait is well inside the 60s request timeout.
+	passwordVerifyWait = 5 * time.Second
+
+	// passwordVerifyMaxWaiters bounds how many requests may be
+	// queued for a slot at once. Past it, acquire sheds immediately
+	// with 503 instead of joining the queue.
+	//
+	// The wait bounds how long one request occupies memory; this
+	// bounds how many do so at the same time, and without it the
+	// 128 MB hashing budget above is the smaller half of the real
+	// footprint. At the 400 req/s a saturation attack can offer, an
+	// unbounded queue would park ~2000 requests for the full five
+	// seconds.
+	//
+	// A waiter costs far more than maxFormBodySize suggests: that
+	// caps the raw body read, not what the parse retains. MaxBodySize,
+	// CSRF and ParseForm all run before acquire, so a parked waiter
+	// holds r.Form plus r.PostForm plus its header block for the
+	// whole wait. Measured on the pinned go1.26.1 toolchain, as the
+	// HeapAlloc delta across two GCs with 64 waiters parked in the
+	// handler: an ordinary two-field login form retains ~0 MB, but a
+	// 1 MB urlencoded body at Go's 10,000-parameter parse cap retains
+	// 2.82 MB (3.09 MB with %41 escapes), and adding the ~0.9 MB of
+	// headers httpMaxHeaderBytes allows takes it to 4.18 MB. The
+	// retained parse and the header block dominate; the raw body does
+	// not.
+	//
+	// Arithmetic, from the measured 4.18 MB worst case: 16 waiters
+	// commit ~67 MB of queue memory, and peak commitment for the
+	// endpoint is 128 MB of Argon2id plus the 18 requests that retain
+	// a parsed form — 16 queued and the 2 being hashed — at
+	// 18 * 4.18 MB, so ~75 MB: about 203 MB in all. Cross-check
+	// against the deadline: two slots at the ~27 verifications/s
+	// measured on a review host (with the race detector on, so the
+	// real rate is higher) drain a full 16-deep queue in about 0.6 s,
+	// far inside passwordVerifyWait.
+	//
+	// Those 203 MB are live bytes, not resident bytes: the Go
+	// collector lets the heap reach roughly twice the live set before
+	// collecting, with transient parse garbage on top. The review
+	// measured a peak HeapAlloc of 392 MB against this guard under 18
+	// adversarial requests, so provision on the order of 400 MB rather
+	// than 203 MB.
+	passwordVerifyMaxWaiters = 16
+
+	// failureKeyHashBytes is how much of the username digest goes
+	// into a failure key. 64 bits over at most loginFailureMaxKeys
+	// live keys makes a collision negligible, and a collision would
+	// only merge two usernames' failure counters, which throttles
+	// sooner rather than later.
+	failureKeyHashBytes = 8
+)
+
+// failureWindow counts failed credential verifications for one
+// bucket, and records when that count lapses.
+type failureWindow struct {
+	count   int
+	resetAt time.Time
+}
+
+// loginGuard is what replaced the pre-emptive rate limiter on the
+// login POST.
+//
+// A limiter that spends budget on arrival cannot protect a
+// single-admin product: behind the reverse proxy the deployment
+// requires, with TRUSTED_PROXIES unset, every client keys on the
+// proxy, so a stranger trickling five POSTs a minute keeps the one
+// bucket full and the operator's own correct password is answered 429
+// forever. There is no second administrative path.
+//
+// So budget is spent only by a FAILED verification. A correct
+// password is never throttled, whatever the counters say, which is
+// the only shape that guarantees the operator can get in. Two
+// consequences follow and are handled here:
+//
+//   - Every login request now costs an Argon2id hash, so the number
+//     running concurrently is bounded by slots. Without that bound
+//     this trades an admin lockout for memory exhaustion, which is
+//     strictly worse.
+//   - Counting per (client, username) makes the key set
+//     attacker-influenced, so both key sets are capped. Beyond the
+//     per-username cap, failures fall back to a counter keyed on the
+//     client alone; beyond that cap too, a failure is answered as
+//     throttled without being recorded, since refusing to answer a
+//     wrong password costs the operator nothing.
+type loginGuard struct {
+	mu     sync.Mutex
+	byUser map[string]*failureWindow
+	byAddr map[string]*failureWindow
+
+	slots chan struct{}
+
+	// queue holds one token per request waiting for a slot. A token
+	// is taken non-blockingly, so a request that finds it full is
+	// shed rather than queued, and is given up as soon as the wait
+	// ends however it ends.
+	queue chan struct{}
+
+	limit    int
+	interval time.Duration
+	maxKeys  int
+	wait     time.Duration
+
+	// now is time.Now outside tests.
+	now func() time.Time
+}
+
+// newLoginGuard builds a guard with the given failure limit per
+// interval, key-set cap, verification concurrency, queue depth and
+// slot wait.
+func newLoginGuard(
+	limit int,
+	interval time.Duration,
+	maxKeys, concurrency, maxWaiters int,
+	wait time.Duration,
+) *loginGuard {
+	return &loginGuard{
+		byUser:   make(map[string]*failureWindow),
+		byAddr:   make(map[string]*failureWindow),
+		slots:    make(chan struct{}, concurrency),
+		queue:    make(chan struct{}, maxWaiters),
+		limit:    limit,
+		interval: interval,
+		maxKeys:  maxKeys,
+		wait:     wait,
+		now:      time.Now,
+	}
+}
+
+// acquire reserves a verification slot, waiting up to the guard's
+// wait for one. It reports false when the queue of waiters is
+// already full, when no slot became available in time, or when the
+// request was cancelled while waiting; the caller must then answer
+// 503 without verifying anything. The returned function releases the
+// slot and must be called exactly once.
+//
+// ctx is consulted only once the request has to wait: a slot that is
+// free on arrival is handed out without looking at it, so an
+// already-cancelled request can be granted one. That is deliberate
+// and matches lifecycle.waitDone — the caller abandons the work on
+// its own ctx and releases the slot immediately, so nothing is spent
+// on it, and refusing instead would mean shedding a request with
+// capacity standing free.
+func (g *loginGuard) acquire(ctx context.Context) (func(), bool) {
+	// A free slot is taken before any timer is armed, and before a
+	// queue place is claimed: a request that never waits is not a
+	// waiter. Without this preamble the bounded select below can find
+	// its slot send and an already-expired timer ready at the same
+	// time, and Go picks among ready cases uniformly at random — so a
+	// process descheduled for longer than the wait sheds a request
+	// with slots standing free, which is precisely when shedding is
+	// least defensible.
+	//
+	// This cannot let a late arrival barge past a queued waiter. A
+	// waiter can only be parked on a FULL buffer, and a release
+	// refills that buffer from the head of the send queue under the
+	// channel lock, so the buffer never appears non-full while anyone
+	// is parked and this send fails whenever there is a waiter.
+	select {
+	case g.slots <- struct{}{}:
+		return func() { <-g.slots }, true
+	default:
+	}
+
+	// Shedding past the queue depth is what keeps waiting memory
+	// bounded; the wait alone only bounds how long one waiter holds
+	// its parsed form, not how many hold one at once.
+	select {
+	case g.queue <- struct{}{}:
+	default:
+		return nil, false
+	}
+
+	// Held only for the wait. A request that gets a slot gives its
+	// queue token back before it starts hashing, so the depth is a
+	// bound on waiters rather than on requests in the handler.
+	defer func() { <-g.queue }()
+
+	timer := time.NewTimer(g.wait)
+	defer timer.Stop()
+
+	// The blocking send is deliberate: a receive on a full buffered
+	// channel hands the slot straight to the head of the send queue,
+	// so slots go out in arrival order and a later arrival cannot
+	// barge past a request already waiting.
+	select {
+	case g.slots <- struct{}{}:
+		return func() { <-g.slots }, true
+	case <-timer.C:
+		return nil, false
+	case <-ctx.Done():
+		return nil, false
+	}
+}
+
+// fail records one failed credential verification by clientKey
+// against username, and reports whether this client has now spent
+// its failure budget and should be answered 429.
+func (g *loginGuard) fail(clientKey, username string) bool {
+	g.mu.Lock()
+	defer g.mu.Unlock()
+
+	now := g.now()
+
+	window := g.window(
+		g.byUser, userFailureKey(clientKey, username), now,
+	)
+	if window == nil {
+		window = g.window(g.byAddr, clientKey, now)
+	}
+
+	if window == nil {
+		// Both key sets are full and neither already tracks this
+		// client, so nothing can be counted without unbounded
+		// growth. Answering the failure as throttled is the safe
+		// direction: it never touches a correct password.
+		return true
+	}
+
+	window.count++
+
+	return window.count >= g.limit
+}
+
+// succeed forgives clientKey's failures against username. A correct
+// password clears the counters, so an operator who mistypes several
+// times and then gets it right is not throttled afterwards.
+func (g *loginGuard) succeed(clientKey, username string) {
+	g.mu.Lock()
+	defer g.mu.Unlock()
+
+	delete(g.byUser, userFailureKey(clientKey, username))
+	delete(g.byAddr, clientKey)
+}
+
+// window returns the live counter for key in set, resetting a lapsed
+// one and creating a missing one when the cap allows. It returns nil
+// only when key is absent and set is full even after lapsed entries
+// are swept.
+func (g *loginGuard) window(
+	set map[string]*failureWindow,
+	key string,
+	now time.Time,
+) *failureWindow {
+	window, ok := set[key]
+	if ok {
+		if !now.Before(window.resetAt) {
+			window.count = 0
+			window.resetAt = now.Add(g.interval)
+		}
+
+		return window
+	}
+
+	if len(set) >= g.maxKeys {
+		sweepLapsed(set, now)
+	}
+
+	if len(set) >= g.maxKeys {
+		return nil
+	}
+
+	window = &failureWindow{resetAt: now.Add(g.interval)}
+	set[key] = window
+
+	return window
+}
+
+// sweepLapsed drops counters whose interval has elapsed.
+func sweepLapsed(set map[string]*failureWindow, now time.Time) {
+	for key, window := range set {
+		if !now.Before(window.resetAt) {
+			delete(set, key)
+		}
+	}
+}
+
+// userFailureKey identifies one (client, submitted username) pair.
+// The username is hashed rather than embedded: a submitted username
+// is attacker-controlled text of attacker-chosen length, and hashing
+// makes every key the same size whatever was sent.
+func userFailureKey(clientKey, username string) string {
+	sum := sha256.Sum256([]byte(username))
+
+	return clientKey + "|" +
+		hex.EncodeToString(sum[:failureKeyHashBytes])
+}
+
+// guard returns the middleware's login guard, building it on first
+// use so that every construction path — fx and the test constructor
+// alike — gets one.
+func (m *Middleware) guard() *loginGuard {
+	m.loginGuardOnce.Do(func() {
+		m.loginGuard = newLoginGuard(
+			loginRateLimit,
+			loginRateInterval,
+			loginFailureMaxKeys,
+			passwordVerifyConcurrency,
+			passwordVerifyMaxWaiters,
+			passwordVerifyWait,
+		)
+	})
+
+	return m.loginGuard
+}
+
+// BeginPasswordVerification reserves one of the bounded Argon2id
+// verification slots. It reports false when the queue of waiting
+// requests is already at passwordVerifyMaxWaiters, or when no slot
+// became free within passwordVerifyWait; in either case the caller
+// must answer 503 and must not verify a password. The returned
+// function releases the slot and must be called exactly once.
+//
+// Every endpoint that hashes a password on request must go through
+// this, or the bound has a hole: the memory is committed per hash,
+// not per endpoint.
+func (m *Middleware) BeginPasswordVerification(
+	ctx context.Context,
+) (func(), bool) {
+	return m.guard().acquire(ctx)
+}
+
+// RecordLoginFailure counts a failed credential verification for the
+// request's client against the submitted username, and reports
+// whether the response should be 429 rather than 401.
+func (m *Middleware) RecordLoginFailure(
+	r *http.Request,
+	username string,
+) bool {
+	throttled := m.guard().fail(m.clientKey(r), username)
+	if throttled {
+		// Truncated even though chi pins this route's path to
+		// the 12-byte constant "/pages/login": RecordLoginFailure
+		// is exported and takes any *http.Request, so a caller on
+		// a route with a URL parameter would otherwise widen this
+		// line. logbound_test.go pins the cap by making exactly
+		// that call, since no request through the mux can.
+		m.log.Warn(
+			"login failure limit exceeded",
+			"path", logfield.Truncate(
+				r.URL.Path, logfield.MaxBytes,
+			),
+		)
+	}
+
+	return throttled
+}
+
+// ForgiveLoginFailures clears the failure counters for the request's
+// client and the submitted username after a successful
+// authentication.
+func (m *Middleware) ForgiveLoginFailures(
+	r *http.Request,
+	username string,
+) {
+	m.guard().succeed(m.clientKey(r), username)
+}
+
+// LoginFailureInterval is how long a spent login failure budget
+// takes to refill, which is what a throttled login answers as
+// Retry-After.
+func (m *Middleware) LoginFailureInterval() time.Duration {
+	return m.guard().interval
+}

internal/middleware/loginguard_test.go

@@ -0,0 +1,629 @@
+package middleware_test
+
+import (
+	"context"
+	"fmt"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/webhooker/internal/database"
+	"sneak.berlin/go/webhooker/internal/middleware"
+)
+
+// mib converts the Argon2id memory parameter, which is in KiB, to MB.
+const mib = 1024
+
+const (
+	// guardInterval is the failure window these tests use. It is
+	// long enough that nothing lapses mid-test on its own; tests
+	// that need a lapse drive the clock instead.
+	guardInterval = time.Minute
+
+	// guardWait is the slot wait for tests that expect to get a
+	// slot. Tests that expect to be refused set their own.
+	guardWait = 2 * time.Second
+
+	guardClient = "198.51.100.7"
+	guardUser   = "admin"
+
+	// racePasses is how many times a both-cases-ready select race is
+	// run. A pass can only go the wrong way once the zero-duration
+	// timer has fired, so the per-pass detection probability is
+	// somewhere below 1/2 rather than exactly it; the bound that
+	// matters is that passes are independent, so a regression that
+	// survives is exponentially unlikely in N. The test still waits
+	// on nothing.
+	racePasses = 1000
+)
+
+// newGuard builds a guard with production-shaped defaults and the
+// given key-set cap and verification concurrency.
+func newGuard(maxKeys, concurrency int) *middleware.LoginGuard {
+	return middleware.NewLoginGuardForTest(
+		middleware.LoginRateLimitConst,
+		guardInterval,
+		maxKeys,
+		concurrency,
+		middleware.PasswordVerifyMaxWaitersConst,
+		guardWait,
+	)
+}
+
+// TestLoginGuard_ThrottlesRepeatedFailures is the brute-force half:
+// wrong passwords for one username from one client key still run out
+// of budget and are answered 429.
+func TestLoginGuard_ThrottlesRepeatedFailures(t *testing.T) {
+	t.Parallel()
+
+	g := newGuard(middleware.LoginFailureMaxKeysConst, 1)
+
+	for i := range middleware.LoginRateLimitConst - 1 {
+		assert.False(
+			t, g.FailForTest(guardClient, guardUser),
+			"failure %d is still inside the budget", i,
+		)
+	}
+
+	assert.True(
+		t, g.FailForTest(guardClient, guardUser),
+		"the last failure of the budget must throttle",
+	)
+	assert.True(
+		t, g.FailForTest(guardClient, guardUser),
+		"failures past the budget must stay throttled",
+	)
+}
+
+// TestLoginGuard_SuccessForgivesFailures pins the forgiveness rule:
+// an operator who mistypes several times and then gets it right must
+// not be left throttled.
+func TestLoginGuard_SuccessForgivesFailures(t *testing.T) {
+	t.Parallel()
+
+	g := newGuard(middleware.LoginFailureMaxKeysConst, 1)
+
+	for range middleware.LoginRateLimitConst {
+		g.FailForTest(guardClient, guardUser)
+	}
+
+	g.SucceedForTest(guardClient, guardUser)
+
+	assert.False(
+		t, g.FailForTest(guardClient, guardUser),
+		"a success must reset the counter, so the next mistake "+
+			"starts a fresh budget",
+	)
+}
+
+// TestLoginGuard_FailuresAreKeyedPerUsername proves the second half
+// of the keying: one username's spent budget does not throttle
+// another's from the same client.
+func TestLoginGuard_FailuresAreKeyedPerUsername(t *testing.T) {
+	t.Parallel()
+
+	g := newGuard(middleware.LoginFailureMaxKeysConst, 1)
+
+	for range middleware.LoginRateLimitConst {
+		g.FailForTest(guardClient, guardUser)
+	}
+
+	assert.True(t, g.FailForTest(guardClient, guardUser))
+	assert.False(
+		t, g.FailForTest(guardClient, "someone-else"),
+		"a different submitted username must have its own budget",
+	)
+}
+
+// TestLoginGuard_WindowLapses covers the interval: a counter that has
+// gone quiet for the whole window starts again from zero.
+func TestLoginGuard_WindowLapses(t *testing.T) {
+	t.Parallel()
+
+	g := newGuard(middleware.LoginFailureMaxKeysConst, 1)
+
+	var now atomic.Int64
+
+	now.Store(time.Now().UnixNano())
+	g.SetNowForTest(func() time.Time {
+		return time.Unix(0, now.Load())
+	})
+
+	for range middleware.LoginRateLimitConst {
+		g.FailForTest(guardClient, guardUser)
+	}
+
+	assert.True(t, g.FailForTest(guardClient, guardUser))
+
+	now.Add(int64(guardInterval) + 1)
+
+	assert.False(
+		t, g.FailForTest(guardClient, guardUser),
+		"a lapsed window must start a fresh budget",
+	)
+}
+
+// TestLoginGuard_UsernameKeySetIsBounded is the memory bound. The
+// submitted username is attacker-controlled, so an attacker rotating
+// usernames must not be able to grow the guard without limit: past
+// the cap, tracking falls back to a counter keyed on the client
+// address alone.
+func TestLoginGuard_UsernameKeySetIsBounded(t *testing.T) {
+	t.Parallel()
+
+	const (
+		maxKeys  = 8
+		attempts = 500
+	)
+
+	g := newGuard(maxKeys, 1)
+
+	for i := range attempts {
+		g.FailForTest(guardClient, fmt.Sprintf("user-%d", i))
+	}
+
+	byUser, byAddr := g.TrackedKeysForTest()
+
+	assert.LessOrEqual(
+		t, byUser, maxKeys,
+		"the per-username key set must not grow past its cap",
+	)
+	assert.LessOrEqual(
+		t, byAddr, maxKeys,
+		"the fallback key set must not grow past its cap either",
+	)
+	assert.Positive(
+		t, byAddr,
+		"past the cap, failures must fall back to the address "+
+			"bucket rather than being dropped",
+	)
+	assert.Less(
+		t, byUser+byAddr, attempts,
+		"memory must not grow with the number of distinct "+
+			"usernames submitted",
+	)
+}
+
+// TestLoginGuard_BeyondBothCapsStaysThrottled covers the hard stop.
+// When both key sets are full of live counters and the client is in
+// neither, there is nothing to count without unbounded growth, so the
+// failure is answered as throttled. That costs the operator nothing:
+// a correct password never reaches this path.
+func TestLoginGuard_BeyondBothCapsStaysThrottled(t *testing.T) {
+	t.Parallel()
+
+	const maxKeys = 4
+
+	g := newGuard(maxKeys, 1)
+
+	// Fill the per-username set from one client, then fill the
+	// address set from distinct clients.
+	for i := range maxKeys {
+		g.FailForTest(guardClient, fmt.Sprintf("user-%d", i))
+	}
+
+	for i := range maxKeys {
+		g.FailForTest(fmt.Sprintf("203.0.113.%d", i), "whoever")
+	}
+
+	assert.True(
+		t, g.FailForTest("203.0.113.200", "brand-new"),
+		"a client that fits in neither full key set must be "+
+			"answered as throttled rather than tracked",
+	)
+
+	byUser, byAddr := g.TrackedKeysForTest()
+	assert.LessOrEqual(t, byUser, maxKeys)
+	assert.LessOrEqual(t, byAddr, maxKeys)
+}
+
+// TestLoginGuard_SemaphoreBoundsConcurrentVerifications is the memory
+// bound on the hashing itself. Verifying credentials before spending
+// limiter budget means an attacker can force one Argon2id hash per
+// request, and each allocates 64 MB; without this bound the fix for
+// an admin lockout would be a memory-exhaustion DoS instead.
+func TestLoginGuard_SemaphoreBoundsConcurrentVerifications(
+	t *testing.T,
+) {
+	t.Parallel()
+
+	const (
+		concurrency = 2
+		workers     = 12
+
+		// rendezvousDeadlock is the deadlock guard described below.
+		// It is orders of magnitude longer than any scheduling delay,
+		// so it never decides the result, and well inside script/test's
+		// 30s timeout, so a wedge fails on the assertion instead of
+		// blowing the package timeout.
+		rendezvousDeadlock = 5 * time.Second
+	)
+
+	g := newGuard(middleware.LoginFailureMaxKeysConst, concurrency)
+
+	var (
+		mu       sync.Mutex
+		inside   int
+		highest  int
+		wg       sync.WaitGroup
+		recorded sync.WaitGroup
+		once     sync.Once
+	)
+
+	// Slot holders rendezvous instead of sleeping, and they hold until
+	// every worker has been answered. A sleep only makes overlap
+	// likely — on a host loaded enough to deschedule a goroutine for
+	// longer than the sleep the workers serialise and the maximum
+	// observed comes back as 1 — so the rendezvous is what makes the
+	// overlap a fact rather than a race won.
+	//
+	// The barrier must not open at the concurrency-th holder, which
+	// would fix the lower bound at the cost of the upper one this test
+	// exists to enforce: holders would leave as soon as the count
+	// reached concurrency, so a guard admitting extra requests would
+	// let them arrive after the first holders had already left and
+	// highest would report concurrency however many were really let
+	// in. It opens instead once every worker's acquire has returned
+	// and any slot it won has been counted, so under a broken guard
+	// every admitted worker is inside simultaneously and highest is
+	// the true maximum. Under a correct guard the refused workers
+	// return within the guard's own wait, which decides nothing beyond
+	// how long that takes.
+	overlapped := make(chan struct{})
+	closeOverlapped := func() {
+		once.Do(func() { close(overlapped) })
+	}
+
+	// Deadlock guard, not a timing margin: no assertion depends on its
+	// length, and the only way to reach it is a worker that never
+	// returns from acquire at all. It is here so that such a wedge
+	// fails legibly on the assertion below instead of hanging until
+	// the package test timeout.
+	abandon := time.AfterFunc(rendezvousDeadlock, closeOverlapped)
+	defer abandon.Stop()
+
+	recorded.Add(workers)
+
+	go func() {
+		recorded.Wait()
+		closeOverlapped()
+	}()
+
+	for range workers {
+		wg.Go(func() {
+			release, ok := g.AcquireForTest(context.Background())
+			if !ok {
+				recorded.Done()
+
+				return
+			}
+
+			defer release()
+
+			mu.Lock()
+
+			inside++
+			if inside > highest {
+				highest = inside
+			}
+
+			mu.Unlock()
+
+			// Counted before signalling, so the barrier can never open
+			// while an admitted worker is still on its way to being
+			// counted.
+			recorded.Done()
+
+			<-overlapped
+
+			mu.Lock()
+			inside--
+			mu.Unlock()
+		})
+	}
+
+	wg.Wait()
+
+	mu.Lock()
+	defer mu.Unlock()
+
+	assert.Equal(
+		t, concurrency, highest,
+		"no more than %d verifications may run at once", concurrency,
+	)
+}
+
+// TestLoginGuard_SaturatedSemaphoreRefusesRatherThanQueueing pins
+// what happens when every slot is taken for longer than the wait: the
+// request is refused, so the caller answers 503 without allocating
+// another 64 MB hash.
+//
+// Neither half of this rides on the wait being long enough. The
+// refusal holds the only slot across the whole of the second call, so
+// there is no wait it could get lucky with — the wait fixes only how
+// long the refusal takes, not whether it happens. The reuse after
+// release is settled by acquire's non-blocking preamble, which is
+// pinned separately by TestLoginGuard_FreeSlotBeatsAnExpiredWait. So
+// the wait below is sized to keep the test quick, not to win a race.
+func TestLoginGuard_SaturatedSemaphoreRefusesRatherThanQueueing(
+	t *testing.T,
+) {
+	t.Parallel()
+
+	g := middleware.NewLoginGuardForTest(
+		middleware.LoginRateLimitConst,
+		guardInterval,
+		middleware.LoginFailureMaxKeysConst,
+		1,
+		middleware.PasswordVerifyMaxWaitersConst,
+		10*time.Millisecond,
+	)
+
+	release, ok := g.AcquireForTest(context.Background())
+	require.True(t, ok, "the first acquire must get the only slot")
+
+	_, ok = g.AcquireForTest(context.Background())
+	assert.False(
+		t, ok,
+		"with the only slot held, a second request must be refused "+
+			"rather than wait indefinitely",
+	)
+
+	release()
+
+	release, ok = g.AcquireForTest(context.Background())
+
+	// require, not assert: acquire returns a nil release alongside a
+	// false ok, so calling it after a non-fatal assertion turns one
+	// failed test into a segfault that takes the whole package test
+	// binary down. Every assertion whose value is dereferenced later
+	// has to stop the test.
+	require.True(
+		t, ok, "the slot must be reusable once released",
+	)
+
+	release()
+}
+
+// TestLoginGuard_FreeSlotBeatsAnExpiredWait is the determinism this
+// file used to lack. acquire selects over a slot send and a wait
+// timer, and Go chooses among ready cases uniformly at random, so a
+// call made after the timer had already fired was a coin flip: on a
+// loaded host the previous test's third acquire could be refused
+// with its slot standing free, and then dereference the nil release
+// it got back.
+//
+// The wait here is already elapsed on arrival, which is the worst
+// case that scheduling can produce, so a free slot must still be
+// granted every time. Without acquire's non-blocking preamble each
+// pass is an independent coin flip and the loop fails within a few
+// passes; with it the property holds by construction and no wall
+// clock is involved.
+func TestLoginGuard_FreeSlotBeatsAnExpiredWait(t *testing.T) {
+	t.Parallel()
+
+	g := middleware.NewLoginGuardForTest(
+		middleware.LoginRateLimitConst,
+		guardInterval,
+		middleware.LoginFailureMaxKeysConst,
+		1,
+		middleware.PasswordVerifyMaxWaitersConst,
+		0,
+	)
+
+	for pass := range racePasses {
+		release, ok := g.AcquireForTest(context.Background())
+		require.Truef(
+			t, ok,
+			"pass %d was refused a slot that was free; an expired "+
+				"wait must never beat an available slot",
+			pass,
+		)
+
+		release()
+	}
+}
+
+// TestLoginGuard_AcquireHonoursCancellation proves a client that
+// disconnects while queued frees its place immediately instead of
+// holding it for the full wait.
+func TestLoginGuard_AcquireHonoursCancellation(t *testing.T) {
+	t.Parallel()
+
+	g := newGuard(middleware.LoginFailureMaxKeysConst, 1)
+
+	release, ok := g.AcquireForTest(context.Background())
+	require.True(t, ok)
+
+	defer release()
+
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel()
+
+	_, ok = g.AcquireForTest(ctx)
+	assert.False(
+		t, ok, "a cancelled request must not wait for a slot",
+	)
+}
+
+// TestPasswordVerifyConcurrency_MatchesMemoryBudget pins the
+// concurrency constant to the arithmetic behind it: the number of
+// slots is the hashing budget divided by what one Argon2id hash
+// actually costs.
+//
+// The per-hash figure is read out of the shipped password
+// parameters rather than copied here. A guard that asserts a literal
+// against a literal cannot see the thing it guards: raising
+// argon2Memory would leave it green while the real ceiling doubled.
+func TestPasswordVerifyConcurrency_MatchesMemoryBudget(t *testing.T) {
+	t.Parallel()
+
+	// Memory is the real argon2Memory, in KiB.
+	perHashMB := int(database.DefaultPasswordConfig().Memory) / mib
+
+	require.Positive(
+		t, perHashMB,
+		"the Argon2id memory parameter must be readable in MB",
+	)
+
+	// The memory this service commits to password hashing.
+	const budgetMB = 128
+
+	assert.Equal(
+		t,
+		middleware.PasswordVerifyConcurrencyConst,
+		budgetMB/perHashMB,
+		"the verification concurrency must be the %d MB hashing "+
+			"budget divided by the %d MB one Argon2id hash costs; "+
+			"if the Argon2id parameters changed, the slot count "+
+			"must change with them",
+		budgetMB, perHashMB,
+	)
+}
+
+// TestLoginGuard_ShedsPastTheQueueCap pins the memory bound on
+// waiting, as distinct from the bound on hashing. A waiter arrives
+// with its form already parsed, and the retained parse plus its
+// header block cost several MB — far more than maxFormBodySize
+// suggests, since that caps only the raw body read — so an unbounded
+// queue would hold that much per waiting request for the whole wait;
+// past the cap the guard must refuse instantly rather than grow.
+func TestLoginGuard_ShedsPastTheQueueCap(t *testing.T) {
+	t.Parallel()
+
+	const (
+		maxWaiters = 2
+
+		// Long enough that a queued waiter never times out on its
+		// own, so anything the test observes leaving the queue left
+		// because it was shed.
+		neverElapses = time.Minute
+
+		// The probe carries its own deadline, so a guard that queues
+		// the probe instead of shedding it fails here rather than
+		// hanging until the package test timeout.
+		//
+		// This is a patience budget, not a margin to be won. A shed
+		// returns in microseconds and a probe that queued instead
+		// would not return for neverElapses, so the two are a whole
+		// minute apart and any budget between them separates them. It
+		// is set far above any scheduling stall a loaded host can
+		// produce, because the previous 200 ms — and the 100 ms
+		// elapsed-time assertion it fed — bounded the latency of a
+		// goroutine hand-off, which is a false red waiting to happen
+		// on the machine this suite runs on. What actually proves the
+		// probe was not queued is the queue depth asserted below.
+		probePatience = 5 * time.Second
+	)
+
+	g := middleware.NewLoginGuardForTest(
+		middleware.LoginRateLimitConst,
+		guardInterval,
+		middleware.LoginFailureMaxKeysConst,
+		1,
+		maxWaiters,
+		neverElapses,
+	)
+
+	// Occupy the only slot, so everything after this queues.
+	release, ok := g.AcquireForTest(context.Background())
+	require.True(t, ok)
+
+	defer release()
+	defer fillQueue(t, g, maxWaiters)()
+
+	granted, answered := probeQueueCap(g, probePatience)
+
+	require.True(
+		t, answered,
+		"a request arriving past the queue cap is still waiting to "+
+			"be queued; it must have been shed",
+	)
+	assert.False(
+		t, granted,
+		"a request arriving past the queue cap must be shed",
+	)
+	assert.Equal(
+		t, maxWaiters, g.QueuedWaitersForTest(),
+		"a shed request must not have grown the queue",
+	)
+}
+
+// fillQueue starts n waiters on g and returns once all of them are
+// queued for a slot. The returned function releases them and waits
+// for them to exit.
+func fillQueue(
+	t *testing.T,
+	g *middleware.LoginGuard,
+	n int,
+) func() {
+	t.Helper()
+
+	ctx, cancel := context.WithCancel(context.Background())
+
+	var wg sync.WaitGroup
+
+	for range n {
+		wg.Go(func() {
+			done, got := g.AcquireForTest(ctx)
+			if got {
+				done()
+			}
+		})
+	}
+
+	// Patience budget, not a margin: the waiters park in microseconds
+	// and nothing releases them, so the only way to exhaust this is a
+	// guard that never queues. One second is the same order as the
+	// scheduling stalls this suite has to survive, so it is not one.
+	require.Eventually(
+		t,
+		func() bool { return g.QueuedWaitersForTest() == n },
+		5*time.Second, time.Millisecond,
+		"the waiters must reach the queue before the cap is tested",
+	)
+
+	return func() {
+		cancel()
+		wg.Wait()
+	}
+}
+
+// probeQueueCap acquires from another goroutine. It reports, in
+// order, whether the call was granted a slot and whether it was
+// answered at all within wait; a call that never returned reports
+// false for both.
+//
+// It runs off the test goroutine deliberately. Joining a full queue
+// is not cancellable by context — refusing to join is the property
+// under test — so a guard that fails this would otherwise hang the
+// package until the test timeout instead of failing here.
+//
+// It reports no elapsed time. Timing a goroutine hand-off measures
+// the host, not the guard, and the caller distinguishes shedding from
+// queueing by the queue depth instead.
+func probeQueueCap(
+	g *middleware.LoginGuard,
+	wait time.Duration,
+) (bool, bool) {
+	probed := make(chan bool, 1)
+
+	go func() {
+		release, ok := g.AcquireForTest(context.Background())
+		if ok {
+			release()
+		}
+
+		probed <- ok
+	}()
+
+	select {
+	case result := <-probed:
+		return result, true
+	case <-time.After(wait):
+		return false, false
+	}
+}

internal/middleware/middleware.go

@@ -6,10 +6,8 @@ import (
 	"log/slog"
 	"net"
 	"net/http"
-	"strings"
+	"sync"
 	"time"
-	"unicode"
-	"unicode/utf8"
 
 	basicauth "github.com/99designs/basicauth-go"
 	"github.com/go-chi/chi"
@@ -21,6 +19,7 @@ import (
 	"go.uber.org/fx"
 	"sneak.berlin/go/webhooker/internal/config"
 	"sneak.berlin/go/webhooker/internal/globals"
+	"sneak.berlin/go/webhooker/internal/logfield"
 	"sneak.berlin/go/webhooker/internal/logger"
 	"sneak.berlin/go/webhooker/internal/session"
 )
@@ -43,16 +42,6 @@ const (
 	// pick the size of the line it writes.
 	redactedQuery = "?(redacted)"
 
-	// maxLogFieldBytes bounds each access log field whose value the
-	// client supplies outright: the URL, the User-Agent and the
-	// Referer. The budget is spent in ENCODED bytes (see
-	// truncateLogField), so 512 still holds a real browser's User-Agent
-	// whole — those are plain ASCII, which encodes one byte for one —
-	// while a value built from characters the encoder escapes keeps a
-	// shorter prefix. That is the intended trade: 500 quotation marks
-	// are not a debugging asset.
-	maxLogFieldBytes = 512
-
 	// maxLogRequestIDBytes bounds the request id, which is also
 	// client-supplied: chi's RequestID middleware passes an inbound
 	// X-Request-Id header through verbatim. Its generated form is an
@@ -65,15 +54,10 @@ const (
 	// is half this.
 	maxLogMethodBytes = 32
 
-	// truncationMarker is appended to any field the access log cut, so
-	// a short value and a truncated one cannot be confused. It is
-	// charged on top of the budget, not inside it.
-	truncationMarker = "[truncated]"
-
 	// MaxAccessLogLineBytes is the ceiling on one JSON access log line,
 	// and the number an operator multiplies by the request rate to size
 	// log storage. It is not an observation of a sample: it is the sum
-	// of the budgets above, each of which truncateLogField enforces in
+	// of the budgets above, each of which logfield.Truncate enforces in
 	// ENCODED bytes, plus the part of the line no client can influence.
 	//
 	//	url, useragent, referer   3*(512+11) = 1569
@@ -83,6 +67,9 @@ const (
 	//	                                        ----
 	//	                                        2087
 	//
+	// The 512 is logfield.MaxBytes; the 11 is the truncation marker,
+	// charged on top of each budget rather than inside it.
+	//
 	// The fixed portion is the JSON punctuation, the field names, the
 	// level and the message, both timestamps at their longest, an IPv6
 	// remoteIP with a zone, a three-digit status and a full-width int64
@@ -90,13 +77,68 @@ const (
 	// than sitting on the arithmetic.
 	//
 	// The tty text handler in internal/logger is covered by the same
-	// figure. encodedLogFieldBytes charges every rune at least what
+	// figure. logfield.EncodedBytes charges every rune at least what
 	// the wider of the two handlers emits for it — including the ten
 	// bytes strconv.Quote spends on a non-printable rune at or above
 	// U+10000, which is four more than the JSON handler ever spends —
 	// so each budget bounds the encoded field under either handler.
 	// The text handler's fixed portion is 286, the smaller of the two,
 	// which puts its worst case at 2037.
+	//
+	// It is also the ceiling on every OTHER line this service writes
+	// THROUGH SLOG that carries text an UNAUTHENTICATED client
+	// supplies. Those lines — the MaxBodySize rejection, the CSRF
+	// rejection, the rate-limit rejection, the unauthenticated-request
+	// and unknown-entrypoint DEBUG lines, the failed-login DEBUG
+	// lines, and the two login-throttle WARN lines ("login failure
+	// limit exceeded" in loginguard.go and "password verification
+	// capacity exhausted" in internal/handlers/auth.go) — spend the
+	// same per-field budgets, and each carries strictly fewer
+	// client-supplied fields than the access log does,
+	// so none of them can reach a width the access log cannot. That is
+	// asserted directly, per line and under both handlers, rather than
+	// left to the reasoning: see logbound_test.go in this package and
+	// in internal/handlers.
+	//
+	// The two login-throttle lines are capped defensively: chi pins
+	// their route to the constant path "/pages/login", so no request
+	// through the mux can widen either one. Their assertions call
+	// RecordLoginFailure and the login handler directly with the path
+	// a caller on a parameterised route would supply, which is the
+	// only way those caps can be pinned at all.
+	//
+	// One writer in that set is not a handler's own slog call. The
+	// GORM adapter in internal/gormlog logs the statement with the
+	// client-chosen parameter already interpolated into it, which on
+	// the receiver and login lookups is exactly the value the budgets
+	// above exist for. Its widest line spends two logfield.MaxBytes
+	// budgets — the statement and the driver error — against a fixed
+	// portion smaller than this one's, and
+	// internal/gormlog/gormlog_test.go asserts every line it emits
+	// against this constant directly rather than leaving it as
+	// arithmetic.
+	//
+	// What it does NOT cover, so that the figure above is not read as
+	// more than it is:
+	//
+	//   - Lines carrying an AUTHENTICATED operator's own input, which
+	//     are not truncated at all: the webhook name on "webhook
+	//     created" and the target host on "target URL blocked by SSRF
+	//     protection" (both internal/handlers/source_management.go),
+	//     and target_name in internal/delivery/engine.go and
+	//     target_http.go. Each is bounded only by the 1 MB form body
+	//     cap, so a 100 KB name writes one line of roughly 600 KB.
+	//     Deliberate: truncating the operator's own configuration
+	//     echoed back costs debuggability against no adversary.
+	//   - The "log" delivery target, which exists to write the whole
+	//     inbound event to the log. Deliberate; see
+	//     internal/delivery/target_log.go.
+	//   - The record a recovered panic writes, which is not an access
+	//     log line: its client-supplied fields are charged the same
+	//     budgets, but it carries a whole goroutine stack as well and
+	//     is wider than this figure. It has its own stated ceiling,
+	//     MaxPanicLogLineBytes in recoverer.go, and is written once
+	//     per recovered panic rather than once per request.
 	MaxAccessLogLineBytes = 2560
 )
 
@@ -116,6 +158,12 @@ type Middleware struct {
 	log     *slog.Logger
 	params  *MiddlewareParams
 	session *session.Session
+
+	// loginGuard counts failed credential verifications and bounds
+	// concurrent password hashing. It is built on first use so that
+	// every construction path gets one; see guard().
+	loginGuardOnce sync.Once
+	loginGuard     *loginGuard
 }
 
 // New creates a Middleware from the provided fx parameters.
@@ -167,114 +215,6 @@ func (lrw *loggingResponseWriter) WriteHeader(code int) {
 	lrw.ResponseWriter.WriteHeader(code)
 }
 
-// encodedLogFieldBytes is what r costs on the line once the log
-// handler has escaped it, taking the worse of the two handlers
-// internal/logger configures.
-//
-// slog's JSON handler escapes quote, backslash, newline, carriage
-// return and tab to two bytes each, and every other C0 control plus
-// LINE SEPARATOR and PARAGRAPH SEPARATOR to a six-byte \u escape; it
-// passes every other rune through as its own UTF-8. Its text handler
-// quotes with strconv.Quote, which spells a non-printable rune below
-// U+10000 as \uXXXX but one at or above U+10000 as \UXXXXXXXX — ten
-// bytes, not six. The text handler is therefore the worse of the two
-// for every non-printable rune, and by four bytes apiece for the
-// 955,086 unassigned, private-use and format code points on planes 1
-// to 16.
-//
-// Charging ten there is what makes MaxAccessLogLineBytes hold for the
-// tty handler as well: U+1000C encodes as F0 90 80 8C, every byte
-// >= 0x80, which httpguts.ValidHeaderFieldValue accepts and
-// net/textproto does not strip, so a header can be filled with them.
-//
-// Both handlers pass printable runes through as their own UTF-8, so
-// unicode.IsPrint separates the escaped cases from the plain ones for
-// either handler.
-func encodedLogFieldBytes(r rune) int {
-	const (
-		// A backslash and the character itself.
-		shortEscapeBytes = 2
-		// \uXXXX, which is also the width of \u00XX.
-		escapedRuneBytes = 6
-		// \UXXXXXXXX, strconv.Quote's spelling of a non-printable
-		// rune outside the basic multilingual plane.
-		escapedAstralRuneBytes = 10
-		// The first code point strconv.Quote spells with \U.
-		firstAstralRune = 0x10000
-	)
-
-	switch {
-	case r == '"' || r == '\\' || r == '\n' || r == '\r' || r == '\t':
-		return shortEscapeBytes
-	case !unicode.IsPrint(r) && r >= firstAstralRune:
-		return escapedAstralRuneBytes
-	case !unicode.IsPrint(r):
-		return escapedRuneBytes
-	default:
-		return utf8.RuneLen(r)
-	}
-}
-
-// truncateLogField caps s at maxBytes of ENCODED output, marking the
-// value when it cuts.
-//
-// Budgeting raw bytes would not bound the line. Escaping only ever
-// grows a value, so a raw budget spent on characters the encoder
-// escapes buys a field several times its nominal size — and the line
-// is the thing an operator is told to multiply by their request rate.
-// Charging each rune what it will actually cost is what makes
-// MaxAccessLogLineBytes true rather than merely larger. The visible
-// consequence is that an escape-heavy value keeps a shorter prefix
-// than a plain one, which is the correct trade.
-//
-// The result is always valid UTF-8. A cut on a byte boundary can split
-// a multi-byte rune, and a header can carry bytes that were never
-// valid UTF-8 to begin with; both are dropped rather than kept, since
-// an encoder would otherwise spend six bytes replacing each one.
-func truncateLogField(s string, maxBytes int) string {
-	// No rune encodes to fewer bytes than it occupies, so nothing past
-	// maxBytes raw can fit the budget. Slicing first bounds the scan
-	// below to the budget rather than to the size of the header the
-	// client sent.
-	window, cut := s, false
-	if len(window) > maxBytes {
-		window, cut = window[:maxBytes], true
-	}
-
-	var (
-		kept  strings.Builder
-		spent int
-	)
-
-	for i := 0; i < len(window); {
-		r, size := utf8.DecodeRuneInString(window[i:])
-		if r == utf8.RuneError && size == 1 {
-			i += size
-
-			continue
-		}
-
-		cost := encodedLogFieldBytes(r)
-		if spent+cost > maxBytes {
-			cut = true
-
-			break
-		}
-
-		spent += cost
-
-		kept.WriteString(window[i : i+size])
-
-		i += size
-	}
-
-	if !cut {
-		return kept.String()
-	}
-
-	return kept.String() + truncationMarker
-}
-
 // concreteLogURL renders the request's own URL for the access log
 // branches that keep it, with the query string replaced by a fixed
 // marker.
@@ -368,21 +308,21 @@ func (s *Middleware) Logging() func(http.Handler) http.Handler {
 				// line does not track the size of the request.
 				s.log.Info("http request",
 					"request_start", start,
-					"method", truncateLogField(
+					"method", logfield.Truncate(
 						r.Method, maxLogMethodBytes,
 					),
-					"url", truncateLogField(
+					"url", logfield.Truncate(
 						accessLogURL(r, lrw.statusCode),
-						maxLogFieldBytes,
+						logfield.MaxBytes,
 					),
-					"useragent", truncateLogField(
-						r.UserAgent(), maxLogFieldBytes,
+					"useragent", logfield.Truncate(
+						r.UserAgent(), logfield.MaxBytes,
 					),
-					"request_id", truncateLogField(
+					"request_id", logfield.Truncate(
 						requestID, maxLogRequestIDBytes,
 					),
-					"referer", truncateLogField(
-						r.Referer(), maxLogFieldBytes,
+					"referer", logfield.Truncate(
+						r.Referer(), logfield.MaxBytes,
 					),
 					"proto", r.Proto,
 					"remoteIP", ipFromHostPort(r.RemoteAddr),
@@ -450,10 +390,21 @@ func (s *Middleware) RequireAuth() func(http.Handler) http.Handler {
 			// session lands here and is sent back to the login
 			// page.
 			if !s.session.IsAuthenticated(sess) {
+				// This is the unauthenticated branch, so both
+				// fields are entirely client-chosen and neither
+				// is bounded by anything the router did. DEBUG
+				// is off by default, but turning it on to
+				// diagnose a problem must not hand a client an
+				// unbounded write into the log, so the same
+				// budgets apply here as in the access log.
 				s.log.Debug(
 					"auth middleware: unauthenticated request",
-					"path", r.URL.Path,
-					"method", r.Method,
+					"path", logfield.Truncate(
+						r.URL.Path, logfield.MaxBytes,
+					),
+					"method", logfield.Truncate(
+						r.Method, maxLogMethodBytes,
+					),
 				)
 				http.Redirect(
 					w, r, "/pages/login", http.StatusSeeOther,
@@ -613,10 +564,26 @@ func (s *Middleware) MaxBodySize(
 			}
 
 			if r.ContentLength > maxBytes {
+				// This runs ahead of RequireAuth (see
+				// setupUserRoutes and friends in
+				// internal/server/routes.go), so an
+				// unauthenticated client reaches it with a path
+				// of its own choosing and its own length —
+				// POST /source/<8 KB>/edit with an oversize
+				// declared Content-Length costs nothing to
+				// send. At WARN, on by default, that is a
+				// write into the operator's log sized by the
+				// attacker unless the path is capped. Same
+				// budgets as the access log, so this line
+				// cannot be wider than that one.
 				s.log.Warn(
 					"request body exceeds limit",
-					"method", r.Method,
-					"path", r.URL.Path,
+					"method", logfield.Truncate(
+						r.Method, maxLogMethodBytes,
+					),
+					"path", logfield.Truncate(
+						r.URL.Path, logfield.MaxBytes,
+					),
 					"content_length", r.ContentLength,
 					"limit", maxBytes,
 				)

internal/middleware/middleware_test.go

@@ -57,7 +57,21 @@ func testMiddlewareWithSessionClock(
 		SessionIdleTimeout: idleTimeout,
 	}
 
-	// Create a real session manager with a known key
+	sessManager := newTestSessionManager(cfg, log, clock)
+
+	m := middleware.NewForTest(log, cfg, sessManager)
+
+	return m, sessManager, clock
+}
+
+// newTestSessionManager builds the real session.Session the
+// middleware tests run against: an in-memory cookie store with a
+// known key, and optionally a manually advanced clock.
+func newTestSessionManager(
+	cfg *config.Config,
+	log *slog.Logger,
+	clock *fakeClock,
+) *session.Session {
 	key := make([]byte, testKeySize)
 
 	for i := range key {
@@ -79,11 +93,7 @@ func testMiddlewareWithSessionClock(
 		now = clock.Now
 	}
 
-	sessManager := session.NewForTest(store, cfg, log, key, now)
-
-	m := middleware.NewForTest(log, cfg, sessManager)
-
-	return m, sessManager, clock
+	return session.NewForTest(store, cfg, log, key, now)
 }
 
 // fakeClock is a manually advanced clock, so session expiry can be

internal/middleware/ratelimit.go

@@ -9,14 +9,19 @@ import (
 	"time"
 
 	"github.com/go-chi/httprate"
+	"sneak.berlin/go/webhooker/internal/logfield"
 )
 
 const (
-	// loginRateLimit is the maximum number of login attempts
-	// per interval.
+	// loginRateLimit is the maximum number of FAILED login attempts
+	// one client may make against one submitted username per
+	// interval before further failures are answered 429. Successful
+	// attempts are never counted and never throttled — see
+	// loginGuard.
 	loginRateLimit = 5
 
-	// loginRateInterval is the time window for the rate limit.
+	// loginRateInterval is the time window for the login failure
+	// limit.
 	loginRateInterval = 1 * time.Minute
 
 	// passwordChangeRateLimit is the maximum number of password
@@ -216,16 +221,27 @@ func (m *Middleware) clientKey(r *http.Request) string {
 	return bucketKey(peer)
 }
 
-// tooManyRequests returns the 429 handler used by the login,
+// tooManyRequests returns the 429 handler used by the
 // password-change and per-entrypoint receiver limiters: it logs the
 // rejection with logMessage and answers with responseMessage.
 // httprate adds the Retry-After header (RFC 6585). The aggregate
 // receiver limiter uses floodTooManyRequests instead.
+//
+// The path is capped against the same budget as the access log's url
+// field. The per-entrypoint receiver limiter is unauthenticated and
+// its path is a client-chosen segment of client-chosen length, so at
+// WARN an uncapped path would let a sender pick the size of the line
+// it writes — the same defect the access log capping closed.
 func (m *Middleware) tooManyRequests(
 	logMessage, responseMessage string,
 ) http.HandlerFunc {
 	return func(w http.ResponseWriter, r *http.Request) {
-		m.log.Warn(logMessage, "path", r.URL.Path)
+		m.log.Warn(
+			logMessage,
+			"path", logfield.Truncate(
+				r.URL.Path, logfield.MaxBytes,
+			),
+		)
 		http.Error(w, responseMessage, http.StatusTooManyRequests)
 	}
 }
@@ -255,26 +271,15 @@ func (m *Middleware) floodTooManyRequests(
 	}
 }
 
-// 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
-// form) pass through unaffected. When the rate limit is exceeded,
-// a 429 Too Many Requests response is returned. Clients are
-// identified by rateLimitKey.
-func (m *Middleware) LoginRateLimit() func(http.Handler) http.Handler {
-	return m.postRateLimit(
-		loginRateLimit,
-		loginRateInterval,
-		"login rate limit exceeded",
-		"Too many login attempts. Please try again later.",
-	)
-}
-
 // PasswordChangeRateLimit returns middleware that enforces
 // per-IP rate limiting on password change attempts. The change
 // endpoint verifies the current password, so without a limit a
-// stolen session could be used to brute-force it; the limit
-// matches the login endpoint's.
+// stolen session could be used to brute-force it.
+//
+// Unlike the login POST this limit is still spent on arrival, which
+// is safe here: RequireAuth runs ahead of it, so only a request
+// already carrying a valid session can reach the bucket, and an
+// operator locked out of changing a password can still log in.
 func (m *Middleware) PasswordChangeRateLimit() func(http.Handler) http.Handler {
 	return m.postRateLimit(
 		passwordChangeRateLimit,

internal/middleware/ratelimit_test.go

@@ -20,14 +20,14 @@ import (
 	"sneak.berlin/go/webhooker/internal/middleware"
 )
 
-func TestLoginRateLimit_AllowsGET(t *testing.T) {
+func TestPostRateLimit_AllowsGET(t *testing.T) {
 	t.Parallel()
 
 	m, _ := testMiddleware(t, config.EnvironmentDev)
 
 	var callCount int
 
-	handler := m.LoginRateLimit()(http.HandlerFunc(
+	handler := m.PasswordChangeRateLimit()(http.HandlerFunc(
 		func(w http.ResponseWriter, _ *http.Request) {
 			callCount++
 
@@ -39,7 +39,7 @@ func TestLoginRateLimit_AllowsGET(t *testing.T) {
 	for i := range 20 {
 		req := httptest.NewRequestWithContext(
 			context.Background(),
-			http.MethodGet, "/pages/login", nil,
+			http.MethodGet, "/user/admin/password", nil,
 		)
 		req.RemoteAddr = "192.168.1.1:12345"
 
@@ -110,20 +110,6 @@ func runPostLimitTest(
 	assert.Equal(t, limit, callCount)
 }
 
-func TestLoginRateLimit_LimitsPOST(t *testing.T) {
-	t.Parallel()
-
-	m, _ := testMiddleware(t, config.EnvironmentDev)
-
-	runPostLimitTest(
-		t,
-		m.LoginRateLimit(),
-		middleware.LoginRateLimitConst,
-		"/pages/login",
-		"10.0.0.1:12345",
-	)
-}
-
 func TestPasswordChangeRateLimit_LimitsPOST(t *testing.T) {
 	t.Parallel()
 
@@ -138,19 +124,19 @@ func TestPasswordChangeRateLimit_LimitsPOST(t *testing.T) {
 	)
 }
 
-func TestLoginRateLimit_IndependentPerIP(t *testing.T) {
+func TestPostRateLimit_IndependentPerIP(t *testing.T) {
 	t.Parallel()
 
 	m, _ := testMiddleware(t, config.EnvironmentDev)
 
-	handler := m.LoginRateLimit()(http.HandlerFunc(
+	handler := m.PasswordChangeRateLimit()(http.HandlerFunc(
 		func(w http.ResponseWriter, _ *http.Request) {
 			w.WriteHeader(http.StatusOK)
 		},
 	))
 
 	// Exhaust limit for IP1
-	for range middleware.LoginRateLimitConst {
+	for range middleware.PasswordChangeRateLimitConst {
 		req := httptest.NewRequestWithContext(
 			context.Background(),
 			http.MethodPost, "/pages/login", nil,
@@ -367,7 +353,14 @@ func TestReceiverRateLimit_CountsEveryMethod(t *testing.T) {
 }
 
 const (
-	loginPath  = "/pages/login"
+	// limitedPath is the endpoint these tests drive the shared POST
+	// rate limiter through. It is the password-change path: since
+	// the login POST verifies credentials before spending any
+	// budget, the password-change limiter is the only pre-emptive
+	// POST limiter left, and it is what pins the shared key
+	// function's behaviour here.
+	limitedPath = "/user/admin/password"
+
 	headerXFF  = "X-Forwarded-For"
 	headerReal = "X-Real-IP"
 	headerTrue = "True-Client-IP"
@@ -415,10 +408,10 @@ func assertSharedBucket(
 	m := rateLimitMiddleware(
 		t, &config.Config{TrustedProxies: proxies},
 	)
-	handler := m.LoginRateLimit()(okHandler())
+	handler := m.PasswordChangeRateLimit()(okHandler())
 
-	for i := range middleware.LoginRateLimitConst {
-		w := postWithHeaders(handler, peer, loginPath, headers(i))
+	for i := range middleware.PasswordChangeRateLimitConst {
+		w := postWithHeaders(handler, peer, limitedPath, headers(i))
 		assert.Equal(
 			t, http.StatusOK, w.Code,
 			"request %d should pass", i,
@@ -426,8 +419,8 @@ func assertSharedBucket(
 	}
 
 	w := postWithHeaders(
-		handler, peer, loginPath,
-		headers(middleware.LoginRateLimitConst),
+		handler, peer, limitedPath,
+		headers(middleware.PasswordChangeRateLimitConst),
 	)
 	assert.Equal(t, http.StatusTooManyRequests, w.Code, msg)
 }
@@ -549,24 +542,24 @@ func TestRateLimitKey_ForwardedHonouredFromTrustedPeer(
 	m := rateLimitMiddleware(t, &config.Config{
 		TrustedProxies: trustedProxies(trustedProxyCIDR),
 	})
-	handler := m.LoginRateLimit()(okHandler())
+	handler := m.PasswordChangeRateLimit()(okHandler())
 
 	const peer = trustedPeer
 
 	first := map[string]string{headerXFF: clientIPv4}
 
-	for range middleware.LoginRateLimitConst {
-		postWithHeaders(handler, peer, loginPath, first)
+	for range middleware.PasswordChangeRateLimitConst {
+		postWithHeaders(handler, peer, limitedPath, first)
 	}
 
-	w := postWithHeaders(handler, peer, loginPath, first)
+	w := postWithHeaders(handler, peer, limitedPath, first)
 	assert.Equal(
 		t, http.StatusTooManyRequests, w.Code,
 		"the forwarded client's own bucket must fill up",
 	)
 
 	w = postWithHeaders(
-		handler, peer, loginPath,
+		handler, peer, limitedPath,
 		map[string]string{headerXFF: clientIPv4Alt},
 	)
 	assert.Equal(
@@ -662,7 +655,7 @@ func TestRateLimitKey_LongChainAllocationIsBounded(t *testing.T) {
 	})
 
 	req := httptest.NewRequestWithContext(
-		context.Background(), http.MethodPost, loginPath, nil,
+		context.Background(), http.MethodPost, limitedPath, nil,
 	)
 	req.RemoteAddr = trustedPeer
 	req.Header.Set(
@@ -869,7 +862,7 @@ func clientKeyFor(
 	t.Helper()
 
 	req := httptest.NewRequestWithContext(
-		context.Background(), http.MethodPost, loginPath, nil,
+		context.Background(), http.MethodPost, limitedPath, nil,
 	)
 	req.RemoteAddr = remoteAddr
 
@@ -1019,23 +1012,23 @@ func TestRateLimitKey_UnparseablePeerKeepsDistinctBuckets(
 	)
 }
 
-// TestLoginRateLimit_IPv6SharesBucketWithinSlash64 is the behavioural
+// TestPostRateLimit_IPv6SharesBucketWithinSlash64 is the behavioural
 // half, and the regression test for the bypass itself: a client that
 // rotates source addresses inside its own routed /64 must stay in one
 // bucket. Reverting the masking makes this test fail, because each
 // rotated address would mint a fresh bucket and nothing would be
 // rejected.
-func TestLoginRateLimit_IPv6SharesBucketWithinSlash64(t *testing.T) {
+func TestPostRateLimit_IPv6SharesBucketWithinSlash64(t *testing.T) {
 	t.Parallel()
 
 	m := rateLimitMiddleware(t, &config.Config{})
-	handler := m.LoginRateLimit()(okHandler())
+	handler := m.PasswordChangeRateLimit()(okHandler())
 
-	for i := range middleware.LoginRateLimitConst {
+	for i := range middleware.PasswordChangeRateLimitConst {
 		w := postWithHeaders(
 			handler,
 			fmt.Sprintf("[2001:db8:1:2::%d]:44444", i+1),
-			loginPath, nil,
+			limitedPath, nil,
 		)
 		assert.Equal(
 			t, http.StatusOK, w.Code, "request %d should pass", i,
@@ -1043,7 +1036,7 @@ func TestLoginRateLimit_IPv6SharesBucketWithinSlash64(t *testing.T) {
 	}
 
 	w := postWithHeaders(
-		handler, "[2001:db8:1:2::ffff]:44444", loginPath, nil,
+		handler, "[2001:db8:1:2::ffff]:44444", limitedPath, nil,
 	)
 	assert.Equal(
 		t, http.StatusTooManyRequests, w.Code,
@@ -1052,23 +1045,23 @@ func TestLoginRateLimit_IPv6SharesBucketWithinSlash64(t *testing.T) {
 	)
 }
 
-// TestLoginRateLimit_IPv6IndependentAcrossSlash64 is the other side
+// TestPostRateLimit_IPv6IndependentAcrossSlash64 is the other side
 // of the trade: bucketing by /64 must not merge separate allocations,
 // so a client in a different /64 keeps its own limit.
-func TestLoginRateLimit_IPv6IndependentAcrossSlash64(t *testing.T) {
+func TestPostRateLimit_IPv6IndependentAcrossSlash64(t *testing.T) {
 	t.Parallel()
 
 	m := rateLimitMiddleware(t, &config.Config{})
-	handler := m.LoginRateLimit()(okHandler())
+	handler := m.PasswordChangeRateLimit()(okHandler())
 
-	for range middleware.LoginRateLimitConst + 1 {
+	for range middleware.PasswordChangeRateLimitConst + 1 {
 		postWithHeaders(
-			handler, "[2001:db8:1:2::1]:44444", loginPath, nil,
+			handler, "[2001:db8:1:2::1]:44444", limitedPath, nil,
 		)
 	}
 
 	w := postWithHeaders(
-		handler, "[2001:db8:1:3::1]:44444", loginPath, nil,
+		handler, "[2001:db8:1:3::1]:44444", limitedPath, nil,
 	)
 	assert.Equal(
 		t, http.StatusOK, w.Code,
@@ -1076,23 +1069,23 @@ func TestLoginRateLimit_IPv6IndependentAcrossSlash64(t *testing.T) {
 	)
 }
 
-// TestLoginRateLimit_IPv4IndependentPerAddress guards against the
+// TestPostRateLimit_IPv4IndependentPerAddress guards against the
 // masking leaking into IPv4: two addresses one apart must still hold
 // separate buckets.
-func TestLoginRateLimit_IPv4IndependentPerAddress(t *testing.T) {
+func TestPostRateLimit_IPv4IndependentPerAddress(t *testing.T) {
 	t.Parallel()
 
 	m := rateLimitMiddleware(t, &config.Config{})
-	handler := m.LoginRateLimit()(okHandler())
+	handler := m.PasswordChangeRateLimit()(okHandler())
 
-	for range middleware.LoginRateLimitConst + 1 {
+	for range middleware.PasswordChangeRateLimitConst + 1 {
 		postWithHeaders(
-			handler, clientIPv4+":44444", loginPath, nil,
+			handler, clientIPv4+":44444", limitedPath, nil,
 		)
 	}
 
 	w := postWithHeaders(
-		handler, clientIPv4Alt+":44444", loginPath, nil,
+		handler, clientIPv4Alt+":44444", limitedPath, nil,
 	)
 	assert.Equal(
 		t, http.StatusOK, w.Code,
@@ -1112,7 +1105,7 @@ func forwardedKeyFor(
 	t.Helper()
 
 	req := httptest.NewRequestWithContext(
-		context.Background(), http.MethodPost, loginPath, nil,
+		context.Background(), http.MethodPost, limitedPath, nil,
 	)
 	req.RemoteAddr = trustedPeer
 	req.Header.Set(headerXFF, forwarded)
@@ -1177,11 +1170,77 @@ func TestRateLimitKey_ForwardedIPv6BucketsByPrefix(t *testing.T) {
 	}
 }
 
-// TestLoginRateLimit_ForwardedIPv6SharesBucketWithinSlash64 is the
+// TestRateLimitKey_TrustedPeerUnusableForwardedMasksPeer covers the
+// third bucketKey call site: the peer IS a trusted proxy, but the
+// forwarded chain cannot name a client, so the key falls back to the
+// peer address — and that fallback owes the same /64 masking every
+// other key gets.
+//
+// Every existing test of this fallback uses an IPv4 proxy, where
+// bucketKey is the identity function, so replacing the call with
+// peer.String() leaves the whole suite green. Only operator-listed
+// addresses reach this line and the fallback is fail-closed, so this
+// pins behaviour rather than fixing a defect.
+func TestRateLimitKey_TrustedPeerUnusableForwardedMasksPeer(
+	t *testing.T,
+) {
+	t.Parallel()
+
+	const (
+		proxyCIDR = "2001:db8:ffff::/48"
+		proxyPeer = "[2001:db8:ffff:1::5]:44444"
+		wantKey   = "2001:db8:ffff:1::/64"
+	)
+
+	m := rateLimitMiddleware(t, &config.Config{
+		TrustedProxies: trustedProxies(proxyCIDR),
+	})
+
+	for _, tc := range []struct {
+		name      string
+		forwarded string
+		about     string
+	}{{
+		name:  "absent",
+		about: "no X-Forwarded-For at all falls back to the peer",
+	}, {
+		name:      "unreadable-hop",
+		forwarded: "unknown",
+		about: "a hop that is not a bare address ends the walk " +
+			"and falls back to the peer",
+	}, {
+		name:      "all-hops-trusted",
+		forwarded: "2001:db8:ffff:2::9",
+		about: "a chain naming only trusted proxies names no " +
+			"client, so the peer is used",
+	}} {
+		t.Run(tc.name, func(t *testing.T) {
+			t.Parallel()
+
+			req := httptest.NewRequestWithContext(
+				context.Background(),
+				http.MethodPost, limitedPath, nil,
+			)
+			req.RemoteAddr = proxyPeer
+
+			if tc.forwarded != "" {
+				req.Header.Set(headerXFF, tc.forwarded)
+			}
+
+			assert.Equal(
+				t, wantKey,
+				middleware.ClientKeyForTest(m, req),
+				"%s, masked to its /64", tc.about,
+			)
+		})
+	}
+}
+
+// TestPostRateLimit_ForwardedIPv6SharesBucketWithinSlash64 is the
 // behavioural half on the production path: behind a trusted proxy, a
 // client rotating source addresses inside its own routed /64 must
 // stay in one bucket.
-func TestLoginRateLimit_ForwardedIPv6SharesBucketWithinSlash64(
+func TestPostRateLimit_ForwardedIPv6SharesBucketWithinSlash64(
 	t *testing.T,
 ) {
 	t.Parallel()
@@ -1198,10 +1257,10 @@ func TestLoginRateLimit_ForwardedIPv6SharesBucketWithinSlash64(
 	)
 }
 
-// TestLoginRateLimit_ForwardedIPv6IndependentAcrossSlash64 is the
+// TestPostRateLimit_ForwardedIPv6IndependentAcrossSlash64 is the
 // other side of that trade on the same path: bucketing by /64 must
 // not merge two allocations reaching the proxy.
-func TestLoginRateLimit_ForwardedIPv6IndependentAcrossSlash64(
+func TestPostRateLimit_ForwardedIPv6IndependentAcrossSlash64(
 	t *testing.T,
 ) {
 	t.Parallel()
@@ -1209,15 +1268,15 @@ func TestLoginRateLimit_ForwardedIPv6IndependentAcrossSlash64(
 	m := rateLimitMiddleware(t, &config.Config{
 		TrustedProxies: trustedProxies(trustedProxyCIDR),
 	})
-	handler := m.LoginRateLimit()(okHandler())
+	handler := m.PasswordChangeRateLimit()(okHandler())
 
 	spent := map[string]string{headerXFF: clientIPv6}
-	for range middleware.LoginRateLimitConst + 1 {
-		postWithHeaders(handler, trustedPeer, loginPath, spent)
+	for range middleware.PasswordChangeRateLimitConst + 1 {
+		postWithHeaders(handler, trustedPeer, limitedPath, spent)
 	}
 
 	w := postWithHeaders(
-		handler, trustedPeer, loginPath,
+		handler, trustedPeer, limitedPath,
 		map[string]string{headerXFF: clientIPv6Other},
 	)
 	assert.Equal(

internal/middleware/recoverer.go

@@ -0,0 +1,209 @@
+package middleware
+
+import (
+	"errors"
+	"fmt"
+	"net/http"
+	"runtime/debug"
+
+	"github.com/go-chi/chi/middleware"
+	"sneak.berlin/go/webhooker/internal/logfield"
+)
+
+const (
+	// maxPanicValueBytes bounds the recovered panic value. The value
+	// is our own text, but a handler is free to build one out of the
+	// request — panic(fmt.Sprintf("bad %q", r.URL.Path)) — so it is
+	// charged the same budget the access log gives a field the
+	// client supplies outright.
+	maxPanicValueBytes = logfield.MaxBytes
+
+	// maxPanicStackBytes bounds the stack, in the same ENCODED bytes
+	// logfield.Truncate charges everywhere else. Nothing a client
+	// sends chooses the depth of our own call stack, so this is not
+	// a safety limit; it is what makes MaxPanicLogLineBytes an
+	// arithmetic ceiling rather than an observation. A stack is cut
+	// at its far end, which is net/http's accept frames — the panic
+	// site and the handler that reached it are at the near end and
+	// are always kept.
+	//
+	// For scale: a handler panicking under the full shipped
+	// middleware chain produces a stack of roughly 3,690 bytes in a
+	// record of roughly 3,960, so this budget holds better than
+	// twice the depth that case reaches. Neither number is an
+	// invariant — debug.Stack() embeds absolute source paths, so
+	// both move with where the tree sits, and four checkouts have
+	// reported records of 3,959, 3,961, 3,984 and 4,026 bytes.
+	// internal/server's TestPanicThroughProductionRouter asserts the
+	// ceiling and that the stack arrived uncut, not the figures.
+	maxPanicStackBytes = 8192
+
+	// MaxPanicLogLineBytes is the ceiling on the single line a
+	// recovered panic writes. It is wider than
+	// MaxAccessLogLineBytes, which bounds a line written once per
+	// request, where this one is written once per recovered panic.
+	//
+	//	panic                     512+11 =  523
+	//	stack                    8192+11 = 8203
+	//	request_id                128+11 =  139
+	//	fixed portion                    =  256
+	//	                                    ----
+	//	                                    9121
+	//
+	// The fixed portion is the JSON punctuation, the field names,
+	// the level, the message, the timestamp at its longest and the
+	// response_committed boolean.
+	//
+	// Stated at 10240 so the figure carries headroom rather than
+	// sitting on the arithmetic, exactly as MaxAccessLogLineBytes
+	// is. Both handlers internal/logger can install are covered, for
+	// the reason given there: logfield.EncodedBytes charges every
+	// rune the wider of the two.
+	//
+	// The 9121 and the 10240 are the invariants here. What follows
+	// is illustration: with all three growable fields — the stack,
+	// the panic value and a client-supplied X-Request-Id — driven
+	// past their budgets at once, TestRecovererBoundsTheStack
+	// measured 9,009 bytes on the JSON handler and 8,982 to 8,983 on
+	// the text one in this checkout. Neither is fixed: the stack's
+	// own content decides where its cut lands, so the figures move by
+	// a byte or so between runs and with the checkout. The test
+	// asserts the ceiling and that every growable field was cut,
+	// never the figures.
+	MaxPanicLogLineBytes = 10240
+)
+
+// recoverResponseWriter records whether the response has been
+// committed, which is the one thing the recoverer cannot learn from
+// the panic itself: a handler that panics after writing a status has
+// already spent the response, and a second WriteHeader would only
+// draw net/http's "superfluous response.WriteHeader" complaint
+// without changing what the client received.
+type recoverResponseWriter struct {
+	http.ResponseWriter
+
+	committed bool
+}
+
+func (w *recoverResponseWriter) WriteHeader(code int) {
+	w.committed = true
+
+	w.ResponseWriter.WriteHeader(code)
+}
+
+func (w *recoverResponseWriter) Write(b []byte) (int, error) {
+	// An unheralded Write commits the response just as surely as
+	// WriteHeader does: net/http sends 200 in front of it.
+	w.committed = true
+
+	//nolint:wrapcheck // Pass the writer's own error through unchanged.
+	return w.ResponseWriter.Write(b)
+}
+
+// Unwrap lets http.ResponseController reach the writer underneath, so
+// a handler can still flush or set a write deadline through this
+// wrapper.
+func (w *recoverResponseWriter) Unwrap() http.ResponseWriter {
+	return w.ResponseWriter
+}
+
+// Recoverer returns middleware that turns a handler panic into one
+// structured ERROR record and a 500, rather than a dropped
+// connection.
+//
+// It replaces chi's middleware.Recoverer, which does neither on a
+// current Go release. chi v1.5.5's pretty-printer scans the stack for
+// a frame beginning "panic(0x", which the runtime has not emitted
+// since it started printing "panic({0x...}"; the scan therefore never
+// terminates early, every line reaches decorateFuncCallLine, and that
+// function slices pkg[strings.Index(pkg, "."):] without checking for
+// -1. The resulting second panic escapes chi's own deferred function,
+// so its WriteHeader(500) never runs and net/http closes the
+// connection reporting its own crash instead of the original one.
+// See https://git.eeqj.de/sneak/webhooker/issues/187.
+//
+// chi v5.3.1 has since fixed both halves of that — it scans for
+// "panic(" and guards the index — so upgrading would restore the 500.
+// It would not give what this does: v5 still writes an ANSI-coloured
+// pretty stack straight to os.Stderr, outside internal/logger, outside
+// any budget, at no level the operator set.
+//
+// Where this sits in the chain is load-bearing, and routes.go states
+// it: inside everything that observes the response, so the 500 is
+// what the access log records and the metrics count, and outside the
+// sentryhttp handler, whose Repanic option depends on something
+// further out recovering what it re-raises.
+func (s *Middleware) Recoverer() func(http.Handler) http.Handler {
+	return func(next http.Handler) http.Handler {
+		return http.HandlerFunc(func(
+			w http.ResponseWriter,
+			r *http.Request,
+		) {
+			rw := &recoverResponseWriter{ResponseWriter: w}
+
+			defer func() {
+				rvr := recover()
+				if rvr == nil {
+					return
+				}
+
+				// http.ErrAbortHandler is a handler stating that it
+				// is abandoning the connection on purpose, not a
+				// fault. net/http special-cases it, suppressing both
+				// the stack trace and any response, so it is passed
+				// straight back out rather than logged and answered.
+				err, isError := rvr.(error)
+				if isError &&
+					errors.Is(err, http.ErrAbortHandler) {
+					panic(rvr)
+				}
+
+				s.logPanic(r, rvr, rw.committed)
+
+				if rw.committed {
+					return
+				}
+
+				http.Error(
+					rw,
+					http.StatusText(
+						http.StatusInternalServerError,
+					),
+					http.StatusInternalServerError,
+				)
+			}()
+
+			next.ServeHTTP(rw, r)
+		})
+	}
+}
+
+// logPanic writes the record. Every field it can grow is truncated to
+// a fixed budget, so MaxPanicLogLineBytes holds.
+//
+// The request is identified by request_id alone rather than by
+// repeating the method, URL and address: the access log line for the
+// same request carries all of those, already bounded, and — because
+// the recoverer runs inside the logging middleware — now carries the
+// 500 as its status too. Repeating them here would double those
+// budgets against a line already wider than the access log's ceiling,
+// to say a second time what one join already says.
+func (s *Middleware) logPanic(
+	r *http.Request,
+	rvr any,
+	committed bool,
+) {
+	s.log.Error("handler panic",
+		"panic", logfield.Truncate(
+			fmt.Sprint(rvr), maxPanicValueBytes,
+		),
+		"stack", logfield.Truncate(
+			string(debug.Stack()), maxPanicStackBytes,
+		),
+		"request_id", logfield.Truncate(
+			middleware.GetReqID(r.Context()),
+			maxLogRequestIDBytes,
+		),
+		"response_committed", committed,
+	)
+}

internal/middleware/recoverer_test.go

@@ -0,0 +1,645 @@
+package middleware_test
+
+import (
+	"bytes"
+	"encoding/json"
+	"io"
+	"log"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"github.com/go-chi/chi"
+	chimw "github.com/go-chi/chi/middleware"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/webhooker/internal/middleware"
+)
+
+// panicMarker is the panic value the probe handlers raise. The
+// recoverer's whole job is to put this string, and not some second
+// panic's, in front of an operator.
+const panicMarker = "QQORIGINALPANICVALUEQQ"
+
+// probeFuncName appears in the stack of every panic raised below,
+// since that is the function raising it. Its presence is how these
+// tests tell a real stack from an empty field.
+const probeFuncName = "panicProbe"
+
+// committedStatus is the status a handler sends before panicking in
+// the already-committed case. It is deliberately not 200, so a test
+// cannot pass on net/http's implicit default.
+const committedStatus = http.StatusMultiStatus
+
+// recovererProbe is a test server carrying one panicking route,
+// behind the production recoverer.
+type recovererProbe struct {
+	server *httptest.Server
+
+	// logs holds every record the middleware wrote.
+	logs *bytes.Buffer
+
+	// serverErrors holds everything net/http wrote to its own error
+	// log. A working recoverer leaves it empty: net/http only reports
+	// a request when a panic escapes the handler chain, which is the
+	// failure this issue is about.
+	serverErrors *bytes.Buffer
+}
+
+// newRecovererProbe stands up a real HTTP server — a real listener, a
+// real connection, a real client — behind the production recoverer.
+//
+// A real server rather than an httptest.ResponseRecorder, because a
+// recorder cannot express the outcome that made this a defect: chi's
+// Recoverer left net/http to close the connection, which a recorder
+// records as an ordinary unwritten response while a client sees EOF.
+// The status a client actually receives is only observable over a
+// socket.
+func newRecovererProbe(
+	t *testing.T,
+	textHandler bool,
+	handler http.HandlerFunc,
+) *recovererProbe {
+	t.Helper()
+
+	newMiddleware := capturingMiddleware
+	if textHandler {
+		newMiddleware = capturingTextMiddleware
+	}
+
+	m, logs := newMiddleware(t)
+
+	router := chi.NewRouter()
+	// The registration order the production router uses: RequestID
+	// outside so the recoverer's record can name the request,
+	// Logging outside so the recovered 500 is the status it records.
+	router.Use(chimw.RequestID)
+	router.Use(m.Logging())
+	router.Use(m.Recoverer())
+	router.Get("/probe", handler)
+
+	serverErrors := new(bytes.Buffer)
+
+	server := httptest.NewUnstartedServer(router)
+	server.Config.ErrorLog = log.New(serverErrors, "", 0)
+	server.Start()
+	t.Cleanup(server.Close)
+
+	return &recovererProbe{
+		server:       server,
+		logs:         logs,
+		serverErrors: serverErrors,
+	}
+}
+
+// get drives one request at the probe route and returns the response,
+// or the transport error if the connection was dropped instead.
+func (p *recovererProbe) get(t *testing.T) (*http.Response, error) {
+	t.Helper()
+
+	return p.getWithRequestID(t, "")
+}
+
+// getWithRequestID drives the same request carrying a client-supplied
+// X-Request-Id. chi's RequestID middleware adopts that header verbatim
+// when it is present and only generates a value when it is absent, so
+// this is the third growable field on the panic record and the only
+// one a client fills outright.
+func (p *recovererProbe) getWithRequestID(
+	t *testing.T,
+	requestID string,
+) (*http.Response, error) {
+	t.Helper()
+
+	req, err := http.NewRequestWithContext(
+		t.Context(), http.MethodGet, p.server.URL+"/probe", nil,
+	)
+	require.NoError(t, err)
+
+	if requestID != "" {
+		req.Header.Set(chimw.RequestIDHeader, requestID)
+	}
+
+	return p.server.Client().Do(req)
+}
+
+// wait shuts the server down and blocks until every in-flight request
+// has finished, which is what makes the log buffer safe to read.
+//
+// A client returns as soon as the response is complete — or, for a
+// deliberately aborted connection, as soon as it is closed — while the
+// access log line for the same request is still being written on the
+// server goroutine. It is idempotent, so a test may call it directly
+// before reading the buffer itself.
+func (p *recovererProbe) wait() {
+	p.server.Close()
+}
+
+// records decodes every JSON log line the probe captured.
+func (p *recovererProbe) records(t *testing.T) []map[string]any {
+	t.Helper()
+
+	p.wait()
+
+	var out []map[string]any
+
+	for line := range strings.SplitSeq(
+		strings.TrimSpace(p.logs.String()), "\n",
+	) {
+		if line == "" {
+			continue
+		}
+
+		record := map[string]any{}
+		require.NoError(t, json.Unmarshal([]byte(line), &record))
+
+		out = append(out, record)
+	}
+
+	return out
+}
+
+// panicRecord returns the single "handler panic" record, failing if
+// there is not exactly one.
+func (p *recovererProbe) panicRecord(t *testing.T) map[string]any {
+	t.Helper()
+
+	var found []map[string]any
+
+	for _, record := range p.records(t) {
+		if record["msg"] == "handler panic" {
+			found = append(found, record)
+		}
+	}
+
+	require.Len(
+		t, found, 1,
+		"exactly one panic record expected, log was:\n%s",
+		p.logs.String(),
+	)
+
+	return found[0]
+}
+
+// panicProbe panics with the marker. It is a named function so the
+// stack assertions have something to look for.
+func panicProbe(http.ResponseWriter, *http.Request) {
+	panic(panicMarker)
+}
+
+func TestRecovererAnswers500AndLogsTheOriginalPanic(t *testing.T) {
+	t.Parallel()
+
+	probe := newRecovererProbe(t, false, panicProbe)
+
+	resp, err := probe.get(t)
+	require.NoError(
+		t, err,
+		"a panicking handler must answer, not drop the connection",
+	)
+
+	defer func() { _ = resp.Body.Close() }()
+
+	body, err := io.ReadAll(resp.Body)
+	require.NoError(t, err)
+
+	assert.Equal(t, http.StatusInternalServerError, resp.StatusCode)
+	assert.Contains(t, string(body), "Internal Server Error")
+
+	record := probe.panicRecord(t)
+	assert.Equal(t, "ERROR", record["level"])
+	assert.Equal(t, panicMarker, record["panic"])
+	assert.Equal(t, false, record["response_committed"])
+
+	stack, ok := record["stack"].(string)
+	require.True(t, ok, "the record must carry a stack")
+	assert.Contains(
+		t, stack, probeFuncName,
+		"the stack must reach the function that panicked",
+	)
+	assert.NotContains(
+		t, stack, "slice bounds out of range",
+		"a secondary panic must not have occurred",
+	)
+
+	assert.Empty(
+		t, probe.serverErrors.String(),
+		"net/http must not have had to report anything",
+	)
+}
+
+// TestRecovererStatusReachesTheAccessLog pins the placement. The
+// recoverer runs inside the logging middleware precisely so the status
+// it writes is the one the access log records; registered outside it,
+// as chi's Recoverer was, the same request is logged as a 200 that the
+// client never received.
+func TestRecovererStatusReachesTheAccessLog(t *testing.T) {
+	t.Parallel()
+
+	probe := newRecovererProbe(t, false, panicProbe)
+
+	resp, err := probe.get(t)
+	require.NoError(t, err)
+	require.NoError(t, resp.Body.Close())
+	require.Equal(t, http.StatusInternalServerError, resp.StatusCode)
+
+	var access map[string]any
+
+	for _, record := range probe.records(t) {
+		if record["msg"] == "http request" {
+			access = record
+		}
+	}
+
+	require.NotNil(t, access, "the request must still be logged")
+	assert.EqualValues(
+		t, http.StatusInternalServerError, access["status"],
+		"the access log must record the status the client got",
+	)
+
+	// The panic record identifies its request by request_id alone,
+	// so that join has to work.
+	assert.Equal(
+		t, access["request_id"],
+		probe.panicRecord(t)["request_id"],
+	)
+	assert.NotEmpty(t, access["request_id"])
+}
+
+// TestRecovererRepanicsErrAbortHandler covers the one panic value that
+// must not be turned into a 500. net/http documents it as the way a
+// handler abandons a connection deliberately and special-cases it,
+// suppressing both the response and its own stack report.
+func TestRecovererRepanicsErrAbortHandler(t *testing.T) {
+	t.Parallel()
+
+	probe := newRecovererProbe(
+		t, false,
+		func(http.ResponseWriter, *http.Request) {
+			panic(http.ErrAbortHandler)
+		},
+	)
+
+	resp, err := probe.get(t)
+	if err == nil {
+		_ = resp.Body.Close()
+	}
+
+	require.Error(
+		t, err,
+		"an aborted handler must not answer with a status",
+	)
+
+	for _, record := range probe.records(t) {
+		assert.NotEqual(
+			t, "handler panic", record["msg"],
+			"a deliberate abort is not a fault to report",
+		)
+	}
+
+	assert.Empty(
+		t, probe.serverErrors.String(),
+		"net/http suppresses ErrAbortHandler; it must still see it",
+	)
+}
+
+// TestRecovererKeepsAnAlreadyCommittedResponse covers a handler that
+// panics after sending its status. The bytes are already on the wire,
+// so a second WriteHeader would change nothing the client sees and
+// would draw net/http's "superfluous response.WriteHeader" report.
+func TestRecovererKeepsAnAlreadyCommittedResponse(t *testing.T) {
+	t.Parallel()
+
+	probe := newRecovererProbe(
+		t, false,
+		func(w http.ResponseWriter, _ *http.Request) {
+			w.WriteHeader(committedStatus)
+			_, _ = w.Write([]byte("partial"))
+
+			panic(panicMarker)
+		},
+	)
+
+	resp, err := probe.get(t)
+	require.NoError(t, err)
+
+	defer func() { _ = resp.Body.Close() }()
+
+	body, err := io.ReadAll(resp.Body)
+	require.NoError(t, err)
+
+	assert.Equal(t, committedStatus, resp.StatusCode)
+	assert.Equal(t, "partial", string(body))
+
+	record := probe.panicRecord(t)
+	assert.Equal(t, panicMarker, record["panic"])
+	assert.Equal(
+		t, true, record["response_committed"],
+		"the record must say why no 500 was sent",
+	)
+
+	assert.NotContains(
+		t, probe.serverErrors.String(),
+		"superfluous response.WriteHeader",
+	)
+}
+
+// TestRecovererKeepsAnImplicitlyCommittedResponse is the same case
+// without an explicit WriteHeader: a bare Write commits the response
+// to 200 just as surely.
+func TestRecovererKeepsAnImplicitlyCommittedResponse(t *testing.T) {
+	t.Parallel()
+
+	probe := newRecovererProbe(
+		t, false,
+		func(w http.ResponseWriter, _ *http.Request) {
+			_, _ = w.Write([]byte("partial"))
+
+			panic(panicMarker)
+		},
+	)
+
+	resp, err := probe.get(t)
+	require.NoError(t, err)
+	require.NoError(t, resp.Body.Close())
+
+	assert.Equal(t, http.StatusOK, resp.StatusCode)
+	assert.Equal(
+		t, true, probe.panicRecord(t)["response_committed"],
+	)
+	assert.NotContains(
+		t, probe.serverErrors.String(),
+		"superfluous response.WriteHeader",
+	)
+}
+
+// panicLogHandler names one of the two handlers internal/logger can
+// install. The recoverer's probe selects between them with a bool
+// rather than by constructing one, which is why this does not reuse
+// logHandlers() the way the fills reuse escapeFills().
+type panicLogHandler struct {
+	name string
+	text bool
+}
+
+func panicLogHandlers() []panicLogHandler {
+	return []panicLogHandler{{"json", false}, {"text", true}}
+}
+
+// TestRecovererBoundsThePanicRecord holds the record to its stated
+// ceiling with a panic value the size of a request. A handler is free
+// to build a panic value out of what the client sent, so the value is
+// charged a client-sized budget even though the stack is not.
+func TestRecovererBoundsThePanicRecord(t *testing.T) {
+	t.Parallel()
+
+	for _, handler := range panicLogHandlers() {
+		// The fills are internal/middleware's own access log fills,
+		// shared rather than restated: plain text, the characters
+		// both handlers escape to two bytes, a bare C0 control, and
+		// an astral non-printable the text handler spells with a
+		// ten-byte \U escape.
+		for fillName, fillRune := range escapeFills() {
+			t.Run(handler.name+"/"+fillName, func(t *testing.T) {
+				t.Parallel()
+
+				value := strings.Repeat(
+					fillRune, oversizedSegmentBytes,
+				) + tailMarker
+
+				probe := newRecovererProbe(
+					t, handler.text,
+					func(http.ResponseWriter, *http.Request) {
+						panic(value)
+					},
+				)
+
+				resp, err := probe.get(t)
+				require.NoError(t, err)
+				require.NoError(t, resp.Body.Close())
+				require.Equal(
+					t, http.StatusInternalServerError,
+					resp.StatusCode,
+				)
+				probe.wait()
+
+				for line := range strings.SplitSeq(
+					strings.TrimSpace(probe.logs.String()), "\n",
+				) {
+					assert.LessOrEqual(
+						t, len(line),
+						middleware.MaxPanicLogLineBytes,
+						"log line exceeded its stated bound",
+					)
+					assert.NotContains(
+						t, line, tailMarker,
+						"the far end of the panic value reached "+
+							"the log, so nothing truncated it",
+					)
+				}
+			})
+		}
+	}
+}
+
+// deepPanic recurses to depth and then panics, so the stack itself
+// overruns its budget. It is the only way to exercise the stack cut:
+// the shipped middleware chain does not come close (see
+// TestPanicThroughProductionRouter in internal/server).
+func deepPanic(depth int, value string) int {
+	if depth == 0 {
+		panic(value)
+	}
+
+	return deepPanic(depth-1, value) + 1
+}
+
+// assertEveryFieldWasCut holds each of the record's three growable
+// fields to its own budget, which is what the ceiling is the sum of.
+// The stack is cut at its far end, so its near end — the panic site —
+// has to survive; the request id is the client's own bytes, so its
+// cut is the one that bounds an attacker rather than our own call
+// depth.
+func assertEveryFieldWasCut(t *testing.T, record map[string]any) {
+	t.Helper()
+
+	stack, ok := record["stack"].(string)
+	require.True(t, ok)
+	assert.True(
+		t, strings.HasSuffix(stack, truncationSuffix),
+		"an oversized stack must be marked as cut",
+	)
+	assert.Contains(
+		t, stack, "deepPanic",
+		"the near end of the stack must survive the cut",
+	)
+	assert.NotContains(
+		t, stack, "net/http.(*conn).serve",
+		"the far end is what a cut discards",
+	)
+
+	id, ok := record["request_id"].(string)
+	require.True(t, ok)
+	assert.True(
+		t, strings.HasSuffix(id, truncationSuffix),
+		"an oversized request id must be marked as cut",
+	)
+	assert.LessOrEqual(
+		t, len(id), maxRequestIDBytes+len(truncationSuffix),
+		"the request id must be held to its own budget",
+	)
+
+	value, ok := record["panic"].(string)
+	require.True(t, ok)
+	assert.True(
+		t, strings.HasSuffix(value, truncationSuffix),
+		"an oversized panic value must be marked as cut",
+	)
+}
+
+// TestRecovererBoundsTheStack drives every growable field on the
+// record past its budget at once — an oversized stack, an oversized
+// panic value and an oversized client-supplied X-Request-Id — over
+// both log handlers. It holds that line to the stated ceiling and
+// reports what it measured, and it pins that a cut stack keeps its
+// near end — the panic site — rather than its far one.
+//
+// The two fields the test picks the content of — the panic value and
+// the request id — are filled with the quotation mark. Both handlers
+// escape it to two bytes, which is exactly what logfield charges for
+// it, so each of those fields emits every byte of its budget; no fill
+// emits more, since logfield charges each rune the wider of the two
+// handlers and a field can therefore never emit more than it spent.
+// The stack is not a fill: recursion drives it past its budget and
+// the cut lands wherever its own content puts it, which is why the
+// measured widths move by a byte between runs.
+func TestRecovererBoundsTheStack(t *testing.T) {
+	t.Parallel()
+
+	for _, handler := range panicLogHandlers() {
+		t.Run(handler.name, func(t *testing.T) {
+			t.Parallel()
+
+			value := strings.Repeat(`"`, oversizedSegmentBytes) +
+				tailMarker
+			requestID := strings.Repeat(`"`, oversizedSegmentBytes) +
+				tailMarker
+
+			probe := newRecovererProbe(
+				t, handler.text,
+				func(http.ResponseWriter, *http.Request) {
+					_ = deepPanic(512, value)
+				},
+			)
+
+			resp, err := probe.getWithRequestID(t, requestID)
+			require.NoError(t, err)
+			require.NoError(t, resp.Body.Close())
+			require.Equal(
+				t, http.StatusInternalServerError, resp.StatusCode,
+			)
+			probe.wait()
+
+			// The text handler does not emit JSON, so the field-level
+			// assertions run on the JSON one; the line bound below
+			// is asserted on both, which is the point of the sweep.
+			if !handler.text {
+				assertEveryFieldWasCut(t, probe.panicRecord(t))
+			}
+
+			widest := 0
+
+			for line := range strings.SplitSeq(
+				strings.TrimSpace(probe.logs.String()), "\n",
+			) {
+				assert.LessOrEqual(
+					t, len(line), middleware.MaxPanicLogLineBytes,
+				)
+				assert.NotContains(t, line, tailMarker)
+
+				widest = max(widest, len(line))
+			}
+
+			t.Logf(
+				"widest line measured: %d bytes (ceiling %d)",
+				widest, middleware.MaxPanicLogLineBytes,
+			)
+		})
+	}
+}
+
+// TestRecovererIgnoresANonPanickingHandler is the negative control:
+// the middleware must be inert on the ordinary path.
+func TestRecovererIgnoresANonPanickingHandler(t *testing.T) {
+	t.Parallel()
+
+	probe := newRecovererProbe(
+		t, false,
+		func(w http.ResponseWriter, _ *http.Request) {
+			w.WriteHeader(http.StatusTeapot)
+		},
+	)
+
+	resp, err := probe.get(t)
+	require.NoError(t, err)
+	require.NoError(t, resp.Body.Close())
+
+	assert.Equal(t, http.StatusTeapot, resp.StatusCode)
+
+	for _, record := range probe.records(t) {
+		assert.NotEqual(t, "handler panic", record["msg"])
+	}
+}
+
+// TestRecovererKeepsResponseControllerWorking pins the Unwrap method.
+// The middleware wraps the ResponseWriter to learn whether the
+// response was committed, and a wrapper without Unwrap hides
+// net/http's own writer from http.ResponseController, so a handler
+// that flushes or sets a deadline starts failing.
+//
+// The recoverer is the only middleware in the chain here. The access
+// logger's own wrapper does not implement Unwrap, so a chain
+// containing it fails this regardless of what the recoverer does;
+// what is being pinned is that the recoverer adds no such opacity of
+// its own.
+func TestRecovererKeepsResponseControllerWorking(t *testing.T) {
+	t.Parallel()
+
+	m, _ := capturingMiddleware(t)
+
+	handler := m.Recoverer()(http.HandlerFunc(
+		func(w http.ResponseWriter, _ *http.Request) {
+			_, _ = w.Write([]byte("chunk"))
+
+			flushErr := http.NewResponseController(w).Flush()
+			if flushErr != nil {
+				http.Error(
+					w, "flush failed",
+					http.StatusInternalServerError,
+				)
+
+				return
+			}
+		},
+	))
+
+	server := httptest.NewServer(handler)
+	t.Cleanup(server.Close)
+
+	req, err := http.NewRequestWithContext(
+		t.Context(), http.MethodGet, server.URL, nil,
+	)
+	require.NoError(t, err)
+
+	resp, err := server.Client().Do(req)
+	require.NoError(t, err)
+
+	defer func() { _ = resp.Body.Close() }()
+
+	body, err := io.ReadAll(resp.Body)
+	require.NoError(t, err)
+
+	assert.Equal(t, http.StatusOK, resp.StatusCode)
+	assert.Equal(t, "chunk", string(body))
+}

internal/server/export_test.go

@@ -4,6 +4,7 @@ import (
 	"log/slog"
 	"net/http"
 
+	"github.com/getsentry/sentry-go"
 	"sneak.berlin/go/webhooker/internal/config"
 	"sneak.berlin/go/webhooker/internal/handlers"
 	"sneak.berlin/go/webhooker/internal/middleware"
@@ -13,6 +14,25 @@ import (
 // build requests that sit exactly at, below, and above it.
 const MaxFormBodySizeForTest = maxFormBodySize
 
+// ScrubSentryRequestForTest exposes the BeforeSend hook that
+// enableSentry installs, so a test can assert on what it leaves in an
+// event without standing up a Sentry client.
+func ScrubSentryRequestForTest(
+	event *sentry.Event,
+	hint *sentry.EventHint,
+) *sentry.Event {
+	return scrubSentryRequest(event, hint)
+}
+
+// SentryClientOptionsForTest exposes the exact options enableSentry
+// initialises the SDK with, so a test can capture events through the
+// production hook wiring rather than a hand-built equivalent.
+func SentryClientOptionsForTest(
+	dsn, release string,
+) sentry.ClientOptions {
+	return sentryClientOptions(dsn, release)
+}
+
 // NewRouterForTest builds the real route tree via SetupRoutes with
 // the supplied middleware and handlers, bypassing the fx lifecycle
 // and the HTTP listener. Tests use it so that route-group middleware
@@ -34,3 +54,40 @@ func NewRouterForTest(
 
 	return s.router
 }
+
+// ProbePattern is the route NewRouterWithProbeForTest adds to the
+// production route tree.
+const ProbePattern = "/probe"
+
+// NewRouterWithProbeForTest builds the production route tree exactly
+// as NewRouterForTest does and then registers probe at ProbePattern,
+// so a test can drive a handler that panics through the shipped
+// global middleware chain rather than a hand-assembled one. Nothing
+// about the chain is rebuilt here: the probe is an extra leaf under
+// the same Use() registrations every other route gets.
+//
+// sentryEnabled selects whether the sentryhttp handler is registered,
+// which in production a configured SENTRY_DSN decides. It is a
+// parameter because the relationship between that handler's Repanic
+// option and the recoverer registered outside it is the thing a test
+// has to be able to pin.
+func NewRouterWithProbeForTest(
+	log *slog.Logger,
+	cfg *config.Config,
+	mw *middleware.Middleware,
+	h *handlers.Handlers,
+	sentryEnabled bool,
+	probe http.HandlerFunc,
+) http.Handler {
+	s := &Server{
+		log:           log,
+		mw:            mw,
+		h:             h,
+		params:        ServerParams{Config: cfg},
+		sentryEnabled: sentryEnabled,
+	}
+	s.SetupRoutes()
+	s.router.Handle(ProbePattern, probe)
+
+	return s.router
+}

internal/server/recoverer_test.go

@@ -0,0 +1,285 @@
+package server_test
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"os/exec"
+	"strings"
+	"testing"
+
+	"github.com/getsentry/sentry-go"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/webhooker/internal/middleware"
+	"sneak.berlin/go/webhooker/internal/server"
+)
+
+// panicProbeMarker is the value the probe handler panics with. The
+// defect this pins lost it entirely: what reached the operator was the
+// recoverer's own secondary panic, naming chi's decorateFuncCallLine
+// and nothing about the fault that caused it.
+const panicProbeMarker = "QQPRODUCTIONPANICVALUEQQ"
+
+// panicChildEnv, when set, tells the re-executed test binary to run
+// the child half of the fd-level probe below.
+const panicChildEnv = "WEBHOOKER_PANIC_PROBE_CHILD"
+
+// panicChildResultPrefix labels the child's own one-line report of
+// what the HTTP client saw, so the parent can find it among whatever
+// else lands on the child's standard output.
+const panicChildResultPrefix = "PANIC-PROBE-RESULT "
+
+// stackTruncationMarker mirrors what internal/middleware appends to a
+// field it cut. It is duplicated rather than exported, as the access
+// log's budgets are, so that changing it has to be restated here
+// deliberately.
+const stackTruncationMarker = "[truncated]"
+
+// TestPanicThroughProductionRouter drives a handler panic through the
+// shipped router, over a real server, in a subprocess whose actual
+// file descriptors are captured.
+//
+// Every part of that is load-bearing.
+//
+// A subprocess, because the question is what reaches fd 1 and fd 2 of
+// the process an operator runs. The defect's signature was 0 bytes on
+// standard error and a 2,772-byte record on standard output describing
+// chi's own crash, and neither is visible to a test that swaps the
+// logger for a buffer.
+//
+// A real server, because a panicking handler under chi's Recoverer
+// dropped the connection: the client got EOF, not a status. An
+// httptest.ResponseRecorder has no connection to drop and would have
+// recorded the same unwritten response either way, which is why this
+// defect survived the existing suite.
+//
+// The production router, because the placement of the recoverer among
+// the other global middleware is part of the fix.
+func TestPanicThroughProductionRouter(t *testing.T) {
+	t.Parallel()
+
+	if os.Getenv(panicChildEnv) != "" {
+		t.Skip("child half; run by the parent below")
+	}
+
+	//nolint:gosec // Re-executing this test binary, with a fixed arg.
+	cmd := exec.CommandContext(
+		t.Context(), os.Args[0],
+		"-test.run", "^TestPanicProbeChild$",
+	)
+
+	cmd.Env = append(os.Environ(), panicChildEnv+"=1")
+
+	var stdout, stderr bytes.Buffer
+
+	cmd.Stdout = &stdout
+	cmd.Stderr = &stderr
+
+	require.NoError(
+		t, cmd.Run(),
+		"child failed\nstdout:\n%s\nstderr:\n%s",
+		stdout.String(), stderr.String(),
+	)
+
+	assertPanicProbeOutput(t, stdout.String(), stderr.String())
+}
+
+// assertPanicProbeOutput holds the child's descriptors to what a
+// working recoverer produces.
+func assertPanicProbeOutput(t *testing.T, stdout, stderr string) {
+	t.Helper()
+
+	result := ""
+
+	var record map[string]any
+
+	for line := range strings.SplitSeq(stdout, "\n") {
+		if after, found := strings.CutPrefix(
+			line, panicChildResultPrefix,
+		); found {
+			result = after
+
+			continue
+		}
+
+		if !strings.HasPrefix(line, `{"time"`) {
+			continue
+		}
+
+		decoded := map[string]any{}
+		if json.Unmarshal([]byte(line), &decoded) != nil {
+			continue
+		}
+
+		if decoded["msg"] == "handler panic" {
+			require.Nil(
+				t, record, "one panic record expected, got two",
+			)
+
+			record = decoded
+
+			assert.LessOrEqual(
+				t, len(line), middleware.MaxPanicLogLineBytes,
+				"the panic record must hold its stated ceiling",
+			)
+
+			t.Logf(
+				"panic record through the shipped chain: %d bytes "+
+					"(ceiling %d)",
+				len(line), middleware.MaxPanicLogLineBytes,
+			)
+		}
+	}
+
+	// What the client got. Under the defect this read
+	// `status=0 err=... EOF`.
+	require.Equal(
+		t, "status=500 err=<nil>", result,
+		"the client must receive a 500, not a dropped connection",
+	)
+
+	// What the operator got. Under the defect there was no such
+	// record: standard output carried net/http reporting chi's own
+	// crash, at INFO, with the original panic value nowhere in it.
+	require.NotNil(
+		t, record,
+		"no structured panic record reached standard output",
+	)
+	assert.Equal(t, "ERROR", record["level"])
+	assert.Equal(t, panicProbeMarker, record["panic"])
+	assert.Equal(t, false, record["response_committed"])
+
+	stack, ok := record["stack"].(string)
+	require.True(t, ok)
+	assert.Contains(t, stack, "panicProbeHandler")
+	assert.NotContains(
+		t, stack, stackTruncationMarker,
+		"the shipped middleware chain's own stack must fit the "+
+			"stack budget without being cut",
+	)
+	t.Logf("stack through the shipped chain: %d bytes", len(stack))
+
+	// The secondary panic, in every form it took. net/http's report
+	// is the tell: it only logs a request when something escaped the
+	// handler chain.
+	assert.NotContains(t, stdout, "http: panic serving")
+	assert.NotContains(t, stdout, "slice bounds out of range")
+	assert.NotContains(t, stdout, "decorateFuncCallLine")
+	assert.Empty(
+		t, strings.TrimSpace(stderr),
+		"nothing may reach standard error",
+	)
+}
+
+// panicProbeHandler is the panicking route the child installs. It is a
+// named function so the stack assertion has something to look for.
+func panicProbeHandler(http.ResponseWriter, *http.Request) {
+	panic(panicProbeMarker)
+}
+
+// TestPanicProbeChild is the child half of the probe above. It runs
+// only when re-executed with panicChildEnv set; in an ordinary run it
+// returns immediately.
+//
+// It writes its result to standard output with a prefix rather than
+// asserting, because the assertions belong to the parent, which is the
+// only side that can see both descriptors.
+func TestPanicProbeChild(t *testing.T) {
+	t.Parallel()
+
+	if os.Getenv(panicChildEnv) == "" {
+		return
+	}
+
+	env := newTestEnv(t)
+
+	router := server.NewRouterWithProbeForTest(
+		env.log.Get(), env.cfg, env.mw, env.hnd,
+		false, panicProbeHandler,
+	)
+
+	srv := httptest.NewServer(router)
+	defer srv.Close()
+
+	req, err := http.NewRequestWithContext(
+		context.Background(), http.MethodGet,
+		srv.URL+server.ProbePattern, nil,
+	)
+	require.NoError(t, err)
+
+	status := 0
+
+	resp, err := srv.Client().Do(req)
+	if err == nil {
+		status = resp.StatusCode
+
+		_ = resp.Body.Close()
+	}
+
+	// Written to the descriptor rather than through the testing
+	// package's own output, because fd 1 is exactly what the parent
+	// is measuring.
+	_, writeErr := fmt.Fprintf(
+		os.Stdout, "%sstatus=%d err=%v\n",
+		panicChildResultPrefix, status, err,
+	)
+	require.NoError(t, writeErr)
+}
+
+// TestSentryStillSeesAPanic pins the relationship the recoverer's
+// placement has to preserve. sentryhttp is registered with
+// Repanic: true, inside the recoverer, so an operator with SENTRY_DSN
+// set keeps the report and the client still gets a 500. Registered the
+// other way round, the SDK would swallow the panic and the recoverer
+// would never see it.
+func TestSentryStillSeesAPanic(t *testing.T) {
+	t.Parallel()
+
+	env := newTestEnv(t)
+
+	transport := &captureTransport{}
+
+	opts := server.SentryClientOptionsForTest(
+		"https://public@sentry.invalid/1", "webhooker-test",
+	)
+	opts.Transport = transport
+
+	client, err := sentry.NewClient(opts)
+	require.NoError(t, err)
+
+	router := server.NewRouterWithProbeForTest(
+		env.log.Get(), env.cfg, env.mw, env.hnd,
+		true, panicProbeHandler,
+	)
+
+	req := httptest.NewRequestWithContext(
+		sentry.SetHubOnContext(
+			context.Background(),
+			sentry.NewHub(client, sentry.NewScope()),
+		),
+		http.MethodGet, server.ProbePattern, nil,
+	)
+
+	w := httptest.NewRecorder()
+	router.ServeHTTP(w, req)
+
+	assert.Equal(
+		t, http.StatusInternalServerError, w.Code,
+		"the recoverer must still answer what sentryhttp re-raised",
+	)
+
+	events := transport.events
+	require.Len(t, events, 1, "Sentry must still see the panic")
+	assert.Equal(t, sentry.LevelFatal, events[0].Level)
+	// The SDK renders a string panic value as the event message
+	// rather than an exception, so the whole payload is checked for
+	// the value rather than one field of it.
+	assert.Contains(
+		t, marshalEvent(t, events[0]), panicProbeMarker,
+	)
+}

internal/server/routes.go

@@ -14,6 +14,28 @@ import (
 // maxFormBodySize is the maximum allowed request body size (in
 // bytes) for form POST endpoints. 1 MB is generous for any form
 // submission while preventing abuse from oversized payloads.
+//
+// Every route group below installs MaxBodySize(maxFormBodySize) as
+// its FIRST middleware, ahead of both CSRF and RequireAuth. Both
+// orderings are deliberate.
+//
+// Ahead of CSRF because gorilla/csrf parses the form. The cap has to
+// be installed before anything reads the body, or the parse runs
+// under net/http's 10 MB default instead of this one.
+//
+// Ahead of RequireAuth because an oversize body should be refused
+// before the request buys a cookie decrypt, a session load and the
+// database read behind it. Rejecting first is the cheaper failure,
+// and it is the ordering that keeps an unauthenticated flood from
+// choosing how much session work the process does.
+//
+// What that ordering costs: the 413 branch is reachable
+// unauthenticated, at a URL of the client's choosing and of the
+// client's chosen length. So is the CSRF rejection, which sits in
+// front of RequireAuth for the same reason. Both log that path, so
+// both cap it — see the log calls in Middleware.MaxBodySize and
+// Middleware.CSRF, which spend the same per-field budget as the
+// access log.
 const maxFormBodySize int64 = 1 * 1024 * 1024 // 1 MB
 
 // requestTimeout is the maximum time allowed for a single HTTP
@@ -29,7 +51,6 @@ func (s *Server) SetupRoutes() {
 }
 
 func (s *Server) setupGlobalMiddleware() {
-	s.router.Use(middleware.Recoverer)
 	s.router.Use(middleware.RequestID)
 	s.router.Use(s.mw.SecurityHeaders())
 	s.router.Use(s.mw.Logging())
@@ -42,8 +63,21 @@ func (s *Server) setupGlobalMiddleware() {
 	s.router.Use(s.mw.CORS())
 	s.router.Use(middleware.Timeout(requestTimeout))
 
+	// Panic recovery, deliberately here rather than first. It has to
+	// run inside every middleware that observes the response, so the
+	// 500 it writes is the status the access log records and the
+	// metrics count, and outside the sentryhttp handler below, whose
+	// Repanic option needs something further out to catch what it
+	// re-raises. chi's own middleware.Recoverer held the first slot
+	// until it was measured: on a current Go release it crashes
+	// inside its stack pretty-printer instead of recovering, so the
+	// connection dropped and the original panic was never reported.
+	// See https://git.eeqj.de/sneak/webhooker/issues/187.
+	s.router.Use(s.mw.Recoverer())
+
 	// Sentry error reporting (if SENTRY_DSN is set). Repanic is
-	// true so panics still bubble up to the Recoverer middleware.
+	// true so panics still bubble up to the Recoverer middleware
+	// registered immediately above.
 	if s.sentryEnabled {
 		sentryHandler := sentryhttp.New(sentryhttp.Options{
 			Repanic: true,
@@ -90,17 +124,20 @@ func (s *Server) setupRoutes() {
 
 func (s *Server) setupPageRoutes() {
 	s.router.Route("/pages", func(r chi.Router) {
-		// MaxBodySize must precede CSRF: gorilla/csrf parses the
-		// form, so the cap has to be installed before it runs.
+		// MaxBodySize precedes CSRF and RequireAuth deliberately;
+		// see maxFormBodySize for why, and for what it costs.
 		r.Use(s.mw.MaxBodySize(maxFormBodySize))
 		r.Use(s.mw.CSRF())
 		r.Use(s.mw.NoCache())
 
-		r.Group(func(r chi.Router) {
-			r.Use(s.mw.LoginRateLimit())
-			r.Get("/login", s.h.HandleLoginPage())
-			r.Post("/login", s.h.HandleLoginSubmit())
-		})
+		// The login POST carries no pre-emptive rate limiter. Behind
+		// the reverse proxy production requires, with TRUSTED_PROXIES
+		// unset, every client shares one bucket, so a limiter spent
+		// on arrival lets any stranger deny the operator the only
+		// administrative path. The handler verifies credentials first
+		// and charges only failures; see Handlers.authenticateUser.
+		r.Get("/login", s.h.HandleLoginPage())
+		r.Post("/login", s.h.HandleLoginSubmit())
 
 		r.Post("/logout", s.h.HandleLogout())
 	})
@@ -108,8 +145,8 @@ func (s *Server) setupPageRoutes() {
 
 func (s *Server) setupUserRoutes() {
 	s.router.Route("/user/{username}", func(r chi.Router) {
-		// MaxBodySize must precede CSRF: gorilla/csrf parses the
-		// form, so the cap has to be installed before it runs.
+		// MaxBodySize precedes CSRF and RequireAuth deliberately;
+		// see maxFormBodySize for why, and for what it costs.
 		r.Use(s.mw.MaxBodySize(maxFormBodySize))
 		r.Use(s.mw.CSRF())
 		r.Use(s.mw.NoCache())
@@ -123,8 +160,8 @@ func (s *Server) setupUserRoutes() {
 
 func (s *Server) setupSourceRoutes() {
 	s.router.Route("/sources", func(r chi.Router) {
-		// MaxBodySize must precede CSRF: gorilla/csrf parses the
-		// form, so the cap has to be installed before it runs.
+		// MaxBodySize precedes CSRF and RequireAuth deliberately;
+		// see maxFormBodySize for why, and for what it costs.
 		r.Use(s.mw.MaxBodySize(maxFormBodySize))
 		r.Use(s.mw.CSRF())
 		r.Use(s.mw.NoCache())
@@ -135,8 +172,8 @@ func (s *Server) setupSourceRoutes() {
 	})
 
 	s.router.Route("/source/{sourceID}", func(r chi.Router) {
-		// MaxBodySize must precede CSRF: gorilla/csrf parses the
-		// form, so the cap has to be installed before it runs.
+		// MaxBodySize precedes CSRF and RequireAuth deliberately;
+		// see maxFormBodySize for why, and for what it costs.
 		r.Use(s.mw.MaxBodySize(maxFormBodySize))
 		r.Use(s.mw.CSRF())
 		r.Use(s.mw.NoCache())

internal/server/routes_test.go

@@ -52,6 +52,15 @@ type testEnv struct {
 	sess   *session.Session
 	db     *database.Database
 	dbMgr  *database.WebhookDBManager
+
+	// The collaborators the router was built from, kept so a test
+	// that needs a second router over the same graph — one carrying
+	// a panicking probe route, or one with Sentry registered — can
+	// build it without wiring the graph again.
+	log *logger.Logger
+	cfg *config.Config
+	mw  *middleware.Middleware
+	hnd *handlers.Handlers
 }
 
 // newTestEnv wires the dependency graph with fx and builds the
@@ -100,6 +109,10 @@ func newTestEnv(t *testing.T) *testEnv {
 		sess:   sess,
 		db:     db,
 		dbMgr:  dbMgr,
+		log:    log,
+		cfg:    cfg,
+		mw:     mw,
+		hnd:    hnd,
 	}
 }
 
@@ -420,6 +433,78 @@ func TestPagesLogin_UnderLimit_ValidToken_ReachesHandler(
 	)
 }
 
+// TestPagesLogin_CorrectPasswordSurvivesASpentBudget pins the
+// routing half of the fix, which every other login test misses by
+// driving the handler directly: no pre-emptive limiter sits in front
+// of POST /pages/login on the real route tree.
+//
+// A limiter registered there would answer the last request 429
+// however correct its password is, because the wrong passwords
+// before it have already spent the bucket — which is the lockout
+// this endpoint exists to not have. CSRF and the body cap still run,
+// since every request here carries a harvested token.
+func TestPagesLogin_CorrectPasswordSurvivesASpentBudget(
+	t *testing.T,
+) {
+	t.Parallel()
+
+	const (
+		username = "operator"
+		password = "correct-horse-battery-staple"
+	)
+
+	env := newTestEnv(t)
+	env.seedUser(t, username, password)
+
+	submit := func(t *testing.T, pw string) *httptest.ResponseRecorder {
+		t.Helper()
+
+		token, cookies := env.csrfFrom(t, "/pages/login", nil)
+
+		form := url.Values{}
+		form.Set("csrf_token", token)
+		form.Set("username", username)
+		form.Set("password", pw)
+
+		return env.post("/pages/login", form, cookies)
+	}
+
+	// Spend the failure budget against this username. The exact
+	// limit belongs to the middleware; this waits for the throttle
+	// to appear rather than restating it, under a ceiling well
+	// above it so a broken limiter fails the test instead of
+	// looping.
+	const maxAttempts = 20
+
+	spent := false
+
+	for range maxAttempts {
+		code := submit(t, "wrong").Code
+		if code == http.StatusTooManyRequests {
+			spent = true
+
+			break
+		}
+
+		require.Equal(
+			t, http.StatusUnauthorized, code,
+			"a wrong password must be rejected, not accepted",
+		)
+	}
+
+	require.True(
+		t, spent,
+		"repeated wrong passwords must eventually be throttled",
+	)
+
+	assert.Equal(
+		t, http.StatusSeeOther, submit(t, password).Code,
+		"a correct password must be accepted on the routed "+
+			"endpoint even with the failure budget spent: the "+
+			"operator has no second administrative path",
+	)
+}
+
 // --- /user/{username} group ---
 
 // TestPasswordChange_OversizeBody_RejectedAndPasswordUnchanged

internal/server/sentry.go

@@ -0,0 +1,236 @@
+package server
+
+import (
+	"net/http"
+	"net/url"
+	"strings"
+
+	"github.com/getsentry/sentry-go"
+	"github.com/go-chi/chi"
+)
+
+// sentryRedacted stands in for a withheld field on every event shipped
+// to Sentry. It is a marker rather than an empty string so a reader
+// can tell a suppressed value from an absent one.
+const sentryRedacted = "(redacted)"
+
+// sentryRedactedPath is what stands in for the request path when the
+// route pattern is not reachable. It is deliberately not the concrete
+// path: on the receiver route that path carries the entrypoint UUID,
+// which is a write capability rather than an identifier.
+const sentryRedactedPath = "/" + sentryRedacted
+
+// sentryClientOptions builds the options the SDK is initialised with.
+// It is its own function so a test can stand up a client wired exactly
+// as production is, with only the transport swapped.
+func sentryClientOptions(dsn, release string) sentry.ClientOptions {
+	return sentry.ClientOptions{
+		Dsn:     dsn,
+		Release: release,
+		// Both hooks, because the SDK runs one for error events
+		// and the other for transactions.
+		BeforeSend:            scrubSentryRequest,
+		BeforeSendTransaction: scrubSentryRequest,
+	}
+}
+
+// scrubSentryRequest strips client-supplied content from an event's
+// request context before it leaves the process.
+//
+// sentryhttp attaches the whole *http.Request to the scope
+// (sentryhttp.go:113), and Scope.ApplyToEvent fills the event's
+// Request from it inside prepareEvent, which runs before this hook.
+// Two of the fields it fills are copied with no SendDefaultPII guard:
+//
+//   - QueryString, verbatim from r.URL.RawQuery.
+//   - Data, the first 10 KiB of the request body, teed off r.Body by
+//     SetRequest and filled precisely because the handlers call
+//     ParseForm.
+//
+// Since every form field in this service is read with PostFormValue,
+// the body is the only place a credential is submitted: a target's
+// destination URL, whose path segments are the bearer token, plus the
+// login password and both password-change fields. None of that may
+// reach a third-party service.
+//
+// URL is the third such field. NewRequest builds it as
+// scheme://host/path (interfaces.go:183), and on the receiver route
+// that path is /webhook/<uuid> in full — a write capability, not an
+// identifier. It is rebuilt here from the chi route pattern, on every
+// route, keeping the scheme and the host.
+//
+// This hook is a floor, not a default: the fields it clears stay
+// cleared even if SendDefaultPII is ever turned on.
+func scrubSentryRequest(
+	event *sentry.Event,
+	hint *sentry.EventHint,
+) *sentry.Event {
+	if event == nil {
+		return event
+	}
+
+	pattern := sentryRoutePattern(hint)
+
+	// Only transaction events carry a Transaction name, and the SDK
+	// builds it from the concrete path too (sentryhttp.go:105 via
+	// tracing.go:553). Rewritten on the same terms.
+	if event.Transaction != "" {
+		event.Transaction = sentryTransactionName(
+			event.Transaction, pattern,
+		)
+	}
+
+	if event.Request == nil {
+		return event
+	}
+
+	req := event.Request
+
+	if req.URL != "" {
+		req.URL = sentryRouteURL(req.URL, pattern)
+	}
+
+	if req.QueryString != "" {
+		req.QueryString = sentryRedacted
+	}
+
+	if req.Data != "" {
+		req.Data = sentryRedacted
+	}
+
+	req.Cookies = ""
+	req.Env = nil
+	req.Headers = keptSentryHeaders(req.Headers)
+
+	return event
+}
+
+// sentryRoutePattern returns the chi route pattern for the request the
+// hint carries, or "" when it is not reachable.
+//
+// The request is reachable on the error dispatch only. sentryhttp's
+// recover path calls RecoverWithContext with the request on the
+// context under sentry.RequestContextKey (sentryhttp.go:124-125), and
+// the client copies that context onto the hint (client.go:484-485)
+// before handing it to BeforeSend (client.go:631). chi's routing
+// context is a pointer placed on the request context before the
+// middleware chain runs (chi mux.go:84) and filled in as the mux
+// routes, so by the time a handler panics it names the matched route.
+//
+// The transaction dispatch has no such request: Span.doFinish calls
+// hub.CaptureEvent (tracing.go:356), which passes a nil hint that the
+// client replaces with an empty one (client.go:620-622). The pattern
+// is therefore always "" there, and the callers fall back.
+func sentryRoutePattern(hint *sentry.EventHint) string {
+	if hint == nil || hint.Context == nil {
+		return ""
+	}
+
+	req, ok := hint.Context.Value(
+		sentry.RequestContextKey,
+	).(*http.Request)
+	if !ok || req == nil {
+		return ""
+	}
+
+	rctx := chi.RouteContext(req.Context())
+	if rctx == nil {
+		return ""
+	}
+
+	// Empty when no route matched, which is the fallback case too.
+	return rctx.RoutePattern()
+}
+
+// sentryRouteURL rebuilds an event's request URL with the route
+// pattern in place of the concrete path.
+//
+// The scheme is load-bearing and is kept: the SDK derives it from
+// r.TLS != nil || r.Header.Get("X-Forwarded-Proto") == "https"
+// (interfaces.go:180), byte for byte the predicate
+// internal/middleware/csrf.go uses, so it is the CSRF TLS decision and
+// the reason dropping X-Forwarded-Proto from the header allowlist
+// costs nothing. The host is parsed.Host of the SDK's
+// scheme://r.Host/path, so it is whatever the client's Host header
+// carried: this service validates no hostname. It is kept because that
+// same header is on the allowlist, so scrubbing it here would withhold
+// nothing that is not sent anyway.
+//
+// Everything else in the URL is discarded rather than edited, so a
+// future SDK that starts appending a query string cannot widen this.
+func sentryRouteURL(rawURL, pattern string) string {
+	parsed, err := url.Parse(rawURL)
+	if err != nil || parsed.Scheme == "" {
+		// Not a shape this can safely take apart.
+		return sentryRedacted
+	}
+
+	if pattern == "" {
+		pattern = sentryRedactedPath
+	}
+
+	return parsed.Scheme + "://" + parsed.Host + pattern
+}
+
+// sentryTransactionName rebuilds the SDK's "METHOD /path" transaction
+// name with the route pattern in place of the concrete path. Method is
+// kept for the same reason Request.Method is: net/http admits only a
+// bounded token there. A name in any other shape is withheld whole,
+// since nothing can be said about which part of it is a path.
+func sentryTransactionName(name, pattern string) string {
+	method, _, found := strings.Cut(name, " ")
+	if !found {
+		return sentryRedacted
+	}
+
+	if pattern == "" {
+		pattern = sentryRedactedPath
+	}
+
+	return method + " " + pattern
+}
+
+// keptSentryHeaders returns the subset of headers an event may carry
+// off-host. Dropping by allowlist rather than by blocklist is what
+// makes an unrecognised header safe: the SDK's own filter removes four
+// names and passes everything else, so X-Csrf-Token — which
+// gorilla/csrf accepts in place of the form field — and the shared
+// secrets senders put on the receiver route (X-Gitlab-Token and the
+// per-provider signature headers) would otherwise ship verbatim.
+func keptSentryHeaders(headers map[string]string) map[string]string {
+	if len(headers) == 0 {
+		return headers
+	}
+
+	kept := make(map[string]string, len(headers))
+
+	for name, value := range headers {
+		if sentryKeepsHeader(name) {
+			kept[name] = value
+		}
+	}
+
+	return kept
+}
+
+// sentryKeepsHeader reports whether a request header is routing or
+// content metadata rather than client-chosen payload. Referer is kept
+// on the reasoning that it is browser-set, that this service emits
+// only ?page= in its own links, and that Referrer-Policy is set to
+// strict-origin-when-cross-origin. X-Request-Id ties the event to the
+// local access log line, which holds the rest of the detail.
+func sentryKeepsHeader(name string) bool {
+	switch http.CanonicalHeaderKey(name) {
+	case "Accept",
+		"Content-Length",
+		"Content-Type",
+		"Host",
+		"Origin",
+		"Referer",
+		"User-Agent",
+		"X-Request-Id":
+		return true
+	default:
+		return false
+	}
+}

internal/server/sentry_test.go

@@ -0,0 +1,486 @@
+package server_test
+
+import (
+	"context"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"net/url"
+	"strings"
+	"sync"
+	"testing"
+	"time"
+
+	"github.com/getsentry/sentry-go"
+	sentryhttp "github.com/getsentry/sentry-go/http"
+	"github.com/go-chi/chi"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/webhooker/internal/server"
+)
+
+// The four markers below are the credentials a captured event could
+// carry off-host, one per field of sentry.Request that the SDK fills
+// from the request without a SendDefaultPII guard.
+const (
+	// sentryBodyMarker is submitted as a form value. Since every
+	// handler reads its fields with PostFormValue, the body is the
+	// only place a password or a target URL is ever supplied.
+	sentryBodyMarker = "QQSENTRYBODYMARKERQQ"
+
+	// sentryQueryMarker rides the request line.
+	sentryQueryMarker = "T00000000/B00000000/QQSENTRYQUERYMARKERQQ"
+
+	// sentryHeaderMarker rides X-Csrf-Token, which gorilla/csrf
+	// accepts in place of the form field.
+	sentryHeaderMarker = "QQSENTRYHEADERMARKERQQ"
+
+	// sentryReceiverUUID is the entrypoint identifier in the path of
+	// a receiver request. It is a write capability: anyone holding
+	// it can POST events this service accepts and its targets then
+	// deliver, so it may not reach a third-party tracker.
+	sentryReceiverUUID = "6d1f9c2a-3b7e-4f58-9a0d-c0ffeebadc0d"
+)
+
+// sentryKeptUserAgent is a non-secret header value planted so the
+// assertions below cannot pass by the event carrying no headers at
+// all.
+const sentryKeptUserAgent = "webhooker-test-agent"
+
+// captureTransport records events instead of shipping them, so a test
+// sees exactly the payload the SDK would have put on the wire.
+type captureTransport struct {
+	mu     sync.Mutex
+	events []*sentry.Event
+}
+
+func (c *captureTransport) Configure(sentry.ClientOptions) {}
+
+func (c *captureTransport) Flush(time.Duration) bool { return true }
+
+func (c *captureTransport) SendEvent(event *sentry.Event) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+
+	c.events = append(c.events, event)
+}
+
+// sentryCase drives one request through the real sentryhttp middleware
+// inside a real chi router and returns the events the SDK produced.
+//
+// Routing through a chi mux is load-bearing, not decoration. chi puts
+// its routing context on the request context before the middleware
+// chain runs and fills it in as it matches, so a hand-built request
+// carries no route pattern at all and could not distinguish the hook
+// working from the hook falling back.
+//
+// This is also the only construction path on which Request.Data
+// appears: sentryhttp calls Scope.SetRequest, which tees r.Body into a
+// 10 KiB buffer, ParseForm drains the tee, and Scope.ApplyToEvent
+// copies the buffer into the event inside prepareEvent — before
+// BeforeSend runs. A hand-built sentry.NewRequest never reads the body
+// and so cannot regress-test any of it.
+type sentryCase struct {
+	// scrub selects whether the production BeforeSend hooks are
+	// installed, so the same path shows both what the SDK collects
+	// and what survives.
+	scrub bool
+
+	// tracing enables the transaction dispatch, which the service
+	// leaves off. With it on, a served request produces a
+	// transaction event through BeforeSendTransaction.
+	tracing bool
+
+	// panics selects the error dispatch, via BeforeSend.
+	panics bool
+
+	// request builds the request to serve, given the client whose
+	// hub it must carry.
+	request func(*sentry.Client) *http.Request
+}
+
+func (c sentryCase) capture(t *testing.T) []*sentry.Event {
+	t.Helper()
+
+	transport := &captureTransport{}
+
+	opts := server.SentryClientOptionsForTest(
+		"https://public@sentry.invalid/1", "webhooker-test",
+	)
+	opts.Transport = transport
+
+	if !c.scrub {
+		opts.BeforeSend = nil
+		opts.BeforeSendTransaction = nil
+	}
+
+	if c.tracing {
+		opts.EnableTracing = true
+		opts.TracesSampleRate = 1.0
+	}
+
+	client, err := sentry.NewClient(opts)
+	require.NoError(t, err)
+
+	c.router().ServeHTTP(httptest.NewRecorder(), c.request(client))
+
+	return transport.events
+}
+
+// router mirrors the one ordering these tests depend on, over the two
+// route patterns they need: a recovering middleware outside, then the
+// sentryhttp handler registered with Use and Repanic set, exactly as
+// routes.go orders the two. The bare recover stands in for
+// Middleware.Recoverer, which holds that outer slot in production; it
+// is here only to keep panic stacks out of the test output. That the
+// production one really does catch what sentryhttp re-raises is
+// pinned separately, by TestSentryStillSeesAPanic.
+func (c sentryCase) router() http.Handler {
+	handler := func(_ http.ResponseWriter, r *http.Request) {
+		// This call is what drains the body tee and fills the
+		// buffer. Its success is asserted by the unscrubbed case
+		// below, which sees the body in the event.
+		_ = r.ParseForm()
+
+		if c.panics {
+			panic("boom")
+		}
+	}
+
+	router := chi.NewRouter()
+	router.Use(recoveringMiddleware)
+	router.Use(
+		sentryhttp.New(sentryhttp.Options{Repanic: true}).Handle,
+	)
+	router.HandleFunc("/pages/login", handler)
+	router.HandleFunc("/webhook/{uuid}", handler)
+
+	return router
+}
+
+func recoveringMiddleware(next http.Handler) http.Handler {
+	return http.HandlerFunc(
+		func(w http.ResponseWriter, r *http.Request) {
+			defer func() { _ = recover() }()
+
+			next.ServeHTTP(w, r)
+		},
+	)
+}
+
+// sentryLoginRequest builds the password POST most cases drive, with a
+// credential planted in the body, the query and a header.
+func sentryLoginRequest(client *sentry.Client) *http.Request {
+	form := url.Values{}
+	form.Set("username", "admin")
+	form.Set("password", sentryBodyMarker)
+
+	req := sentryRequest(
+		client,
+		"/pages/login?url=https://hooks.slack.com/services/"+
+			sentryQueryMarker,
+		form.Encode(),
+	)
+
+	req.Header.Set("X-Csrf-Token", sentryHeaderMarker)
+
+	return req
+}
+
+// sentryReceiverRequest builds a POST to the receiver route, whose
+// concrete path carries the entrypoint capability.
+func sentryReceiverRequest(client *sentry.Client) *http.Request {
+	return sentryRequest(
+		client, "/webhook/"+sentryReceiverUUID, "payload=hello",
+	)
+}
+
+func sentryRequest(
+	client *sentry.Client,
+	target, body string,
+) *http.Request {
+	req := httptest.NewRequestWithContext(
+		sentry.SetHubOnContext(
+			context.Background(),
+			sentry.NewHub(client, sentry.NewScope()),
+		),
+		http.MethodPost,
+		target,
+		strings.NewReader(body),
+	)
+
+	req.Header.Set(
+		"Content-Type", "application/x-www-form-urlencoded",
+	)
+	req.Header.Set("User-Agent", sentryKeptUserAgent)
+
+	return req
+}
+
+// marshalEvent encodes an event the way the transport does.
+func marshalEvent(t *testing.T, event *sentry.Event) string {
+	t.Helper()
+
+	encoded, err := json.Marshal(event)
+	require.NoError(t, err)
+
+	return string(encoded)
+}
+
+// onlyEvent asserts a single event was captured and returns it.
+func onlyEvent(t *testing.T, events []*sentry.Event) *sentry.Event {
+	t.Helper()
+
+	require.Len(t, events, 1)
+	require.NotNil(t, events[0].Request)
+
+	return events[0]
+}
+
+// TestSentryScrub_SDKCollectsTheRequestUnscrubbed pins the premise the
+// hook exists for. Without it the SDK ships the whole POST body, the
+// raw query, the CSRF header and the concrete request path, none of
+// which SendDefaultPII=false suppresses.
+func TestSentryScrub_SDKCollectsTheRequestUnscrubbed(t *testing.T) {
+	t.Parallel()
+
+	event := onlyEvent(t, sentryCase{
+		panics:  true,
+		request: sentryLoginRequest,
+	}.capture(t))
+
+	assert.Contains(
+		t, event.Request.Data, sentryBodyMarker,
+		"the SDK is expected to collect the POST body; if it no "+
+			"longer does, the scrub hook's premise changed",
+	)
+	assert.Contains(t, event.Request.QueryString, sentryQueryMarker)
+	assert.Contains(
+		t, marshalEvent(t, event), sentryHeaderMarker,
+	)
+
+	receiver := onlyEvent(t, sentryCase{
+		panics:  true,
+		request: sentryReceiverRequest,
+	}.capture(t))
+
+	assert.Contains(
+		t, receiver.Request.URL, sentryReceiverUUID,
+		"the SDK is expected to build Request.URL from the "+
+			"concrete path; if it no longer does, the route "+
+			"pattern rewrite's premise changed",
+	)
+}
+
+// TestSentryScrub_RedactsTheCapturedRequest is the regression test: no
+// byte of any planted credential may survive into the marshalled event
+// that leaves the process.
+func TestSentryScrub_RedactsTheCapturedRequest(t *testing.T) {
+	t.Parallel()
+
+	event := onlyEvent(t, sentryCase{
+		scrub:   true,
+		panics:  true,
+		request: sentryLoginRequest,
+	}.capture(t))
+
+	encoded := marshalEvent(t, event)
+
+	assert.NotContains(t, encoded, sentryBodyMarker)
+	assert.NotContains(t, encoded, sentryQueryMarker)
+	assert.NotContains(t, encoded, sentryHeaderMarker)
+	assert.NotContains(t, encoded, "hooks.slack.com")
+
+	assert.Equal(t, "(redacted)", event.Request.Data)
+	assert.Equal(t, "(redacted)", event.Request.QueryString)
+	assert.Empty(t, event.Request.Cookies)
+	assert.Empty(t, event.Request.Env)
+}
+
+// TestSentryScrub_ReplacesTheCapabilityPathWithTheRoutePattern is the
+// regression test for the receiver URL: the entrypoint UUID is a write
+// capability and may not reach the tracker, while the route it names
+// must still be readable there.
+func TestSentryScrub_ReplacesTheCapabilityPathWithTheRoutePattern(
+	t *testing.T,
+) {
+	t.Parallel()
+
+	event := onlyEvent(t, sentryCase{
+		scrub:   true,
+		panics:  true,
+		request: sentryReceiverRequest,
+	}.capture(t))
+
+	assert.NotContains(
+		t, marshalEvent(t, event), sentryReceiverUUID,
+	)
+	assert.Equal(
+		t, "http://example.com/webhook/{uuid}", event.Request.URL,
+	)
+}
+
+// TestSentryScrub_KeepsTheRoutingContext checks the hook does not cost
+// the debugging signal: the route, its scheme and host, the method and
+// the metadata headers still identify what failed. On a static route
+// the pattern is the path, so the URL is unchanged there.
+func TestSentryScrub_KeepsTheRoutingContext(t *testing.T) {
+	t.Parallel()
+
+	event := onlyEvent(t, sentryCase{
+		scrub:   true,
+		panics:  true,
+		request: sentryLoginRequest,
+	}.capture(t))
+
+	assert.Equal(
+		t, "http://example.com/pages/login", event.Request.URL,
+	)
+	assert.Equal(t, http.MethodPost, event.Request.Method)
+	assert.Equal(
+		t,
+		sentryKeptUserAgent,
+		event.Request.Headers["User-Agent"],
+	)
+	assert.Equal(
+		t,
+		"application/x-www-form-urlencoded",
+		event.Request.Headers["Content-Type"],
+	)
+}
+
+// TestSentryScrub_RedactsTheTransactionDispatch covers the other hook.
+// Span.doFinish captures with a nil hint, so BeforeSendTransaction
+// gets one with no context and no request: the route pattern is out of
+// reach and both the URL and the SDK-built transaction name have to
+// fall back. Tracing is off in this service, so no transaction event
+// is produced today; the hook is a floor against that changing.
+func TestSentryScrub_RedactsTheTransactionDispatch(t *testing.T) {
+	t.Parallel()
+
+	events := sentryCase{
+		scrub:   true,
+		tracing: true,
+		request: sentryReceiverRequest,
+	}.capture(t)
+
+	event := onlyEvent(t, events)
+	require.Equal(t, "transaction", event.Type)
+
+	assert.NotContains(
+		t, marshalEvent(t, event), sentryReceiverUUID,
+	)
+	assert.Equal(
+		t, "http://example.com/(redacted)", event.Request.URL,
+	)
+	assert.Equal(t, "POST /(redacted)", event.Transaction)
+}
+
+// TestSentryScrub_TransactionDispatchIsUnscrubbedWithoutTheHook pins
+// that dispatch's premise the same way, since it is the one the
+// service does not exercise today.
+func TestSentryScrub_TransactionDispatchIsUnscrubbedWithoutTheHook(
+	t *testing.T,
+) {
+	t.Parallel()
+
+	event := onlyEvent(t, sentryCase{
+		tracing: true,
+		request: sentryReceiverRequest,
+	}.capture(t))
+
+	require.Equal(t, "transaction", event.Type)
+	assert.Contains(t, event.Request.URL, sentryReceiverUUID)
+	assert.Contains(t, event.Transaction, sentryReceiverUUID)
+}
+
+// TestSentryScrub_FallsBackWithoutARoutePattern covers every way the
+// pattern can be missing. None of them may fall back to the concrete
+// path, and all of them keep the scheme, which is the CSRF TLS
+// decision.
+func TestSentryScrub_FallsBackWithoutARoutePattern(t *testing.T) {
+	t.Parallel()
+
+	concrete := "https://example.com/webhook/" + sentryReceiverUUID
+
+	// A request with no chi routing context on it at all, which is
+	// what an event captured outside the router would carry.
+	unrouted := httptest.NewRequestWithContext(
+		context.Background(), http.MethodPost, concrete, nil,
+	)
+
+	for name, hint := range map[string]*sentry.EventHint{
+		"no hint":    nil,
+		"no context": {},
+		"no request": {Context: context.Background()},
+		"unrouted request": {
+			Context: context.WithValue(
+				context.Background(),
+				sentry.RequestContextKey,
+				unrouted,
+			),
+		},
+	} {
+		t.Run(name, func(t *testing.T) {
+			t.Parallel()
+
+			event := sentry.NewEvent()
+			event.Request = &sentry.Request{URL: concrete}
+			event.Transaction = "POST /webhook/" +
+				sentryReceiverUUID
+
+			scrubbed := server.ScrubSentryRequestForTest(
+				event, hint,
+			)
+			require.NotNil(t, scrubbed)
+
+			assert.Equal(
+				t,
+				"https://example.com/(redacted)",
+				scrubbed.Request.URL,
+			)
+			assert.Equal(
+				t, "POST /(redacted)", scrubbed.Transaction,
+			)
+			assert.NotContains(
+				t,
+				marshalEvent(t, scrubbed),
+				sentryReceiverUUID,
+			)
+		})
+	}
+}
+
+// TestSentryScrub_WithholdsUnparseableValues covers the shapes the
+// rewrite cannot take apart. Withholding them whole is the safe
+// answer, since nothing can be said about which part is a path.
+func TestSentryScrub_WithholdsUnparseableValues(t *testing.T) {
+	t.Parallel()
+
+	event := sentry.NewEvent()
+	event.Request = &sentry.Request{
+		URL: "/webhook/" + sentryReceiverUUID,
+	}
+	event.Transaction = "/webhook/" + sentryReceiverUUID
+
+	scrubbed := server.ScrubSentryRequestForTest(event, nil)
+	require.NotNil(t, scrubbed)
+
+	assert.Equal(t, "(redacted)", scrubbed.Request.URL)
+	assert.Equal(t, "(redacted)", scrubbed.Transaction)
+}
+
+// TestSentryScrub_ToleratesEventsWithoutARequest covers the events the
+// hook sees outside an HTTP handler, where no request is attached.
+func TestSentryScrub_ToleratesEventsWithoutARequest(t *testing.T) {
+	t.Parallel()
+
+	scrubbed := server.ScrubSentryRequestForTest(
+		sentry.NewEvent(), nil,
+	)
+
+	require.NotNil(t, scrubbed)
+	assert.Nil(t, scrubbed.Request)
+	assert.Empty(t, scrubbed.Transaction)
+	assert.Nil(t, server.ScrubSentryRequestForTest(nil, nil))
+}

internal/server/server.go

@@ -141,14 +141,14 @@ func (s *Server) enableSentry() {
 		return
 	}
 
-	err := sentry.Init(sentry.ClientOptions{
-		Dsn: s.params.Config.SentryDSN,
-		Release: fmt.Sprintf(
+	err := sentry.Init(sentryClientOptions(
+		s.params.Config.SentryDSN,
+		fmt.Sprintf(
 			"%s-%s",
 			s.params.Globals.Appname,
 			s.params.Globals.Version,
 		),
-	})
+	))
 	if err != nil {
 		s.log.Error("sentry init failure", "error", err)
 		// Don't use fatal since we still want the service to run

script/test

@@ -1,12 +1,34 @@
 #!/bin/sh
 # script/test: run the test suite.
+#
+# -timeout is applied by `go test` per package, not to the run as a whole, so
+# it only has to clear the slowest single package. That is internal/handlers,
+# measured in a cache-defeated builder stage on the 48-core shared build host
+# (2026-08-18); load- and host-dependent, not invariants:
+#
+#   16.9s                  host load 5-20, GOMAXPROCS 48
+#   45.9s / 47.3s / 49.0s  three runs at deliberate host load 31-73
+#   30.6s / 39.7s          host load 5-20, GOMAXPROCS 6 / 4
+#   67.3s / 97.5s          host load 5-20, GOMAXPROCS 2 / 1
+#   67.3s                  GOMAXPROCS 4 at deliberate host load 52-68
+#
+# The old 30s budget was breached by every loaded run and by every GOMAXPROCS
+# at or below 6; at GOMAXPROCS 4 it failed outright ("panic: test timed out
+# after 30s"), reproduced on 33e4fa4 with no other change.
+#
+# 90s matches the org-wide backstop in REPO_POLICIES.md and is sized here
+# against the figures above: the worst case under native parallelism is 49.0s,
+# and the compound GOMAXPROCS-4-under-load case at 67.3s sits at 75% of it.
+# The one figure above 90s is GOMAXPROCS 1, a synthetic core floor rather than
+# a condition CI runs under. If a CPU-limited runner ever puts a real run near
+# 67s, that is the datum to revisit the org figure with.
 set -eu
 
 ROOT="$(cd "$(dirname "$0")/.." && pwd -P)"
 
 main() {
     cd "$ROOT"
-    go test -v -race -timeout 30s ./...
+    go test -v -race -timeout 90s ./...
 }
 
 main "$@"