Name every rate limit that shares the client key

"Trusted proxies" now says a block covering clients makes every rate
limit bypassable, not "all three"; Rate Limiting lists delivery replay
and event resubmit among the limiters sharing the key. The
Configuration table and the TrustedProxies comment say a private
client behind a trusted proxy chooses its key, not behind any proxy.

Model: opus-5-5

README.md

@@ -147,7 +147,7 @@ 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. A set value replaces the default. Under the default, any client with a private address, whether it connects directly or through the proxy, can choose its own rate-limit key by sending its own `X-Forwarded-For`; if any clients have private addresses, set it to the proxy's address alone. See [Trusted proxies](#trusted-proxies) | `10.0.0.0/8,172.16.0.0/12,192.168.0.0/16` (RFC 1918) |
+| `TRUSTED_PROXIES`       | CIDRs whose forwarded headers are trusted. A set value replaces the default. Under the default, any client with a private address, whether it connects directly or through a trusted proxy, can choose its own rate-limit key by sending its own `X-Forwarded-For`; if any clients have private addresses, set it to the proxy's address alone. See [Trusted proxies](#trusted-proxies) | `10.0.0.0/8,172.16.0.0/12,192.168.0.0/16` (RFC 1918) |
 | `ALLOWED_EGRESS_CIDRS`  | CIDRs that delivery targets may reach despite the SSRF blocklist. Read [Allowing egress to your own network](#allowing-egress-to-your-own-network) before setting it | `""` (none) |
 
 #### Allowing egress to your own network
@@ -447,8 +447,8 @@ Two operator requirements follow:
   it can name a different address on every request to get a fresh
   bucket each time, or name another client's address to drain that
   client's bucket. A block that also covers clients — the default, on
-  a network where clients have private addresses — makes all three
+  a network where clients have private addresses — makes every rate
-  limits, including the unauthenticated webhook receiver, silently
+  limit, including the unauthenticated webhook receiver's, silently
   bypassable by every client in the block.
 
 #### Sessions
@@ -1455,7 +1455,7 @@ A registered user of the webhooker service.
 | Field      | Type     | Description |
 | ---------- | -------- | ----------- |
 | `id`       | UUID     | Primary key |
-| `username` | string   | Unique login name |
+| `username` | string   | Unique login name, at most 1024 bytes so that it fits in the session cookie |
 | `password` | string   | Argon2id hash (never exposed via API) |
 
 **Relations:** Has many Webhooks. Has many APIKeys.
@@ -2395,14 +2395,14 @@ 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
+at each of these — just under 1 KB at `invalid password`, whose
-shared with the successful-login line, where a username past 4 KB
+accounts are shared with the successful-login line and so must stay
-overflows the session cookie and answers 500 before that line is
+within the 1024-byte username limit — through both handlers, and
-written — through both handlers, and through seven fills: plain text
+through seven fills: plain text as the baseline, and then the
-as the baseline, and then the quotation mark, backslash, tab, newline,
+quotation mark, backslash, tab, newline, C0 control and astral
-C0 control and astral non-printable, six characters the wider of the
+non-printable, six characters the wider of the two handlers spends
-two handlers spends more on than the client spent sending them. Every
+more on than the client spent sending them. Every case holds each
-case holds each line to the 2,560-byte ceiling. That per-line ceiling
+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
@@ -2552,12 +2552,12 @@ the tree is checked out: four checkouts have reported 3,959, 3,961,
 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
+Every limiter here — receiver, login, password change, delivery replay
-the client the same way, through one shared key function: the
+and event resubmit — identifies the client the same way, through one
-connection's own address, unless the peer is inside
+shared key function: the connection's own address, unless the peer is
-`TRUSTED_PROXIES`, in which case the forwarded client address is used
+inside `TRUSTED_PROXIES`, in which case the forwarded client address is
-instead. That address becomes a bucket by family: IPv4 keys on the full
+used instead. That address becomes a bucket by family: IPv4 keys on
-address, IPv6 on its `/64` prefix. A routed `/64` is the normal
+the full address, IPv6 on its `/64` prefix. A routed `/64` is the normal
 residential and mobile IPv6 allocation, so keying IPv6 per address would
 let one subscriber rotate source addresses and mint a fresh bucket per
 request, evading these limits at the network layer without spoofing

internal/config/config.go

@@ -182,9 +182,9 @@ type Config struct {
 	// Other peers' forwarded headers are ignored and they are
 	// identified by the connection's own address. Under the
 	// default any client with a private address, directly or
-	// through a proxy, can choose its own rate-limit key, so
+	// through a trusted proxy, can choose its own rate-limit
-	// where any clients have private addresses this must be set
+	// key, so where any clients have private addresses this must
-	// to the proxy hosts alone.
+	// be set to the proxy hosts alone.
 	TrustedProxies []netip.Prefix
 
 	// AllowedEgressCIDRs is the set of networks a delivery target

internal/database/model_user.go

@@ -1,13 +1,58 @@
 package database
 
+import (
+	"errors"
+	"fmt"
+
+	"gorm.io/gorm"
+)
+
+// MaxUsernameBytes is the longest username, in bytes, that a user may
+// have. The same number appears in the check constraint on
+// User.Username, because a struct tag cannot reference a constant.
+//
+// A login stores the username in the session cookie, and both
+// securecookie and browsers refuse a cookie value past about 4096
+// bytes. That value is the session base64-encoded twice, so it holds
+// 4096 × 3/4 × 3/4 = 2304 bytes of session, and the signature,
+// timestamp and the session's other values take about 270 of those: a
+// username longer than about 2030 bytes can never log in. The limit is
+// about half that, so the session can carry more values later without
+// locking out an account whose username is already at the limit.
+const MaxUsernameBytes = 1024
+
+// ErrUsernameTooLong is returned when a user is saved with a username
+// longer than MaxUsernameBytes.
+var ErrUsernameTooLong = errors.New("username is too long")
+
 // User represents a user of the webhooker service
+//
+//nolint:lll // a struct tag cannot wrap
 type User struct {
 	BaseModel
 
-	Username string `gorm:"uniqueIndex;not null" json:"username"`
+	Username string `gorm:"uniqueIndex;not null;check:length(CAST(username AS BLOB)) <= 1024" json:"username"`
-	Password string `gorm:"not null"             json:"-"` // Argon2 hashed
+	Password string `gorm:"not null"                                                          json:"-"` // Argon2 hashed
 
 	// Relations
 	Webhooks []Webhook `json:"webhooks,omitempty"`
 	APIKeys  []APIKey  `json:"apiKeys,omitempty"`
 }
+
+// BeforeSave rejects a username longer than MaxUsernameBytes when a whole
+// User is created or saved, so those calls get ErrUsernameTooLong rather
+// than the database's constraint error. A column update such as
+// Update("username", ...) is caught only by the check constraint, as is
+// any path that writes the table without this model.
+func (u *User) BeforeSave(_ *gorm.DB) error {
+	if len(u.Username) > MaxUsernameBytes {
+		return fmt.Errorf(
+			"%w: %d bytes, limit is %d",
+			ErrUsernameTooLong,
+			len(u.Username),
+			MaxUsernameBytes,
+		)
+	}
+
+	return nil
+}

internal/database/model_user_test.go

@@ -0,0 +1,65 @@
+package database_test
+
+import (
+	"strings"
+	"testing"
+
+	"github.com/google/uuid"
+	"github.com/stretchr/testify/assert"
+	"github.com/stretchr/testify/require"
+	"sneak.berlin/go/webhooker/internal/database"
+)
+
+// usernameAtLimit is exactly MaxUsernameBytes long, built from a
+// two-byte character. A check that counted characters rather than bytes
+// would see half the length and let the one-byte-longer name through.
+func usernameAtLimit() string {
+	return strings.Repeat("é", database.MaxUsernameBytes/2)
+}
+
+func TestUserCreate_RejectsOverlongUsername(t *testing.T) {
+	t.Parallel()
+
+	db := startedTestDB(t)
+
+	err := db.Create(&database.User{
+		Username: usernameAtLimit() + "x",
+		Password: "hash",
+	}).Error
+
+	require.ErrorIs(t, err, database.ErrUsernameTooLong)
+}
+
+func TestUserCreate_AcceptsUsernameAtLimit(t *testing.T) {
+	t.Parallel()
+
+	db := startedTestDB(t)
+
+	require.NoError(t, db.Create(&database.User{
+		Username: usernameAtLimit(),
+		Password: "hash",
+	}).Error)
+}
+
+// TestUsersTable_EnforcesUsernameLimitWithoutTheModel inserts with raw
+// SQL, as a path that bypassed User.BeforeSave would, so only the
+// table's check constraint stands between it and an over-long
+// username. Accepting the name at the limit and refusing the next byte
+// also pins the constraint's number to MaxUsernameBytes.
+func TestUsersTable_EnforcesUsernameLimitWithoutTheModel(t *testing.T) {
+	t.Parallel()
+
+	db := startedTestDB(t)
+
+	insert := "INSERT INTO users (id, username, password) VALUES (?, ?, ?)"
+
+	require.NoError(t, db.Exec(
+		insert, uuid.New().String(), usernameAtLimit(), "hash",
+	).Error)
+
+	err := db.Exec(
+		insert, uuid.New().String(), usernameAtLimit()+"x", "hash",
+	).Error
+	require.Error(t, err)
+	assert.Contains(t, err.Error(), "CHECK constraint failed")
+}

internal/handlers/auth_test.go

@@ -453,3 +453,33 @@ func TestLogin_SuccessCreatesSession(t *testing.T) {
 		"the issued cookie must carry an authenticated session",
 	)
 }
+
+// TestLogin_UsernameAtLimitCanLogIn shows that a username of exactly
+// database.MaxUsernameBytes still fits in the session cookie. Past
+// what the cookie can carry, a correct login answers 500.
+func TestLogin_UsernameAtLimitCanLogIn(t *testing.T) {
+	t.Parallel()
+
+	var (
+		h  *handlers.Handlers
+		db *database.Database
+	)
+
+	app := newTestApp(t, &h, &db)
+	app.RequireStart()
+
+	t.Cleanup(app.RequireStop)
+
+	username := strings.Repeat("a", database.MaxUsernameBytes)
+
+	hash, err := database.HashPassword(operatorPassword)
+	require.NoError(t, err)
+	require.NoError(t, db.DB().Create(&database.User{
+		Username: username,
+		Password: hash,
+	}).Error)
+
+	w := submitLogin(h, sharedProxyPeer, username, operatorPassword)
+
+	assert.Equal(t, http.StatusSeeOther, w.Code)
+}

internal/handlers/logbound_test.go

@@ -339,11 +339,9 @@ 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
+// whole username, markers and fill name included, stays within
-// securecookie's 4 KB limit: the cookie is written BEFORE the
+// database.MaxUsernameBytes.
-// "user logged in" line, so an 8 KB username answers 500 and never
+const storedFillBytes = 960
-// 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.