fix(chatd): extract session token from cookie for relay header (#22649)

## Problem

When a browser connects to the chat stream via WebSocket, it
authenticates using cookies only — the native WebSocket API cannot set
custom headers like `Coder-Session-Token`. The relay between replicas
copies the original request's `Cookie` header but did **not** set the
`Coder-Session-Token` header as a fallback.

This causes a **401 on the worker replica** when `EnableHostPrefix` is
enabled, because the `HTTPCookies.Middleware` strips bare
`coder_session_token` cookies (expecting the `__Host-` prefix). Without
a `Coder-Session-Token` header fallback, `apiKeyMiddleware` finds no
valid credentials.

### Root Cause

The data flow:
1. Browser → subscriber replica: `Cookie:
__Host-coder_session_token=xxx` (browser sends prefixed cookie)
2. Subscriber's `HTTPCookies.Middleware` normalizes: `Cookie:
coder_session_token=xxx` (strips prefix)
3. `relayHeaders()` copies `Cookie: coder_session_token=xxx` to relay
request
4. Worker replica's `HTTPCookies.Middleware` sees bare
`coder_session_token` → **strips it** (expects `__Host-` prefix)
5. `apiKeyMiddleware` → `APITokenFromRequest`: no cookie, no header →
**401**

## Fix

Modified `relayHeaders()` to extract the session token value from the
`Cookie` header and set it as the `Coder-Session-Token` header when no
explicit session token header is already present. The header is never
stripped by middleware, so the worker replica can always authenticate.

## Testing

- **`TestRelayHeaders`**: Unit tests for the updated `relayHeaders()`
function covering all scenarios (cookie-only, header+cookie, no auth,
nil source)
- **`TestExtractSessionTokenFromCookieHeader`**: Unit tests for the
helper function
- **`TestChatStreamRelay/RelayCookieOnlyAuth`**: Integration test with
plain HTTP, cookie-only WebSocket auth
- **`TestChatStreamRelay/RelayCookieOnlyAuthWithHostPrefix`**:
Integration test with `EnableHostPrefix=true`, confirming the 401 is
fixed
- **`cookieOnlySessionTokenProvider`**: Test helper that simulates
browser WebSocket behavior (sets Cookie header only on WebSocket dials,
no custom headers)

## Files Changed

- `enterprise/coderd/chatd/chatd.go` — `relayHeaders()` fix +
`extractSessionTokenFromCookieHeader()` helper
- `enterprise/coderd/chatd/relay_headers_internal_test.go` — unit tests
(new file)
- `enterprise/coderd/chats_test.go` — integration tests + test helper
type
This commit is contained in:
Kyle Carberry
2026-03-05 05:11:07 +00:00
committed by GitHub
parent f609de860f
commit 94a2e440a8
2 changed files with 417 additions and 27 deletions
+42 -27
View File
@@ -5,6 +5,7 @@ import (
"math"
"net/http"
"net/url"
"strings"
"time"
"github.com/google/uuid"
@@ -450,8 +451,9 @@ func dialRelay(
relayCtx, relayCancel := context.WithCancel(ctx)
sdkClient := codersdk.New(baseURL)
sdkClient.HTTPClient = cfg.ReplicaHTTPClient
sdkClient.SessionTokenProvider = relayHeaderTokenProvider{
header: relayHeaders(requestHeader, replicaID),
sdkClient.SessionTokenProvider = relayTokenProvider{
token: extractSessionToken(requestHeader),
replicaID: replicaID,
}
sourceEvents, sourceStream, err := sdkClient.StreamChat(relayCtx, chatID, &codersdk.StreamChatOptions{
AfterID: ptr.Ref(int64(math.MaxInt64)),
@@ -532,44 +534,57 @@ drainInitial:
return snapshot, events, cancelFn, nil
}
type relayHeaderTokenProvider struct {
header http.Header
// relayTokenProvider authenticates relay requests to the worker
// replica using the session token extracted from the original
// browser request. It also stamps each request with the relay
// source header so the worker can identify it as an inter-replica
// call.
type relayTokenProvider struct {
token string
replicaID uuid.UUID
}
func (p relayHeaderTokenProvider) AsRequestOption() codersdk.RequestOption {
func (p relayTokenProvider) AsRequestOption() codersdk.RequestOption {
return func(req *http.Request) {
for key, values := range p.header {
for _, value := range values {
req.Header.Add(key, value)
}
}
req.Header.Set(codersdk.SessionTokenHeader, p.token)
req.Header.Set(RelaySourceHeader, p.replicaID.String())
}
}
func (p relayHeaderTokenProvider) SetDialOption(opts *websocket.DialOptions) {
func (p relayTokenProvider) SetDialOption(opts *websocket.DialOptions) {
if opts.HTTPHeader == nil {
opts.HTTPHeader = make(http.Header)
}
for key, values := range p.header {
for _, value := range values {
opts.HTTPHeader.Add(key, value)
}
}
opts.HTTPHeader.Set(codersdk.SessionTokenHeader, p.token)
opts.HTTPHeader.Set(RelaySourceHeader, p.replicaID.String())
}
func (p relayHeaderTokenProvider) GetSessionToken() string {
return p.header.Get(codersdk.SessionTokenHeader)
func (p relayTokenProvider) GetSessionToken() string {
return p.token
}
func relayHeaders(source http.Header, replicaID uuid.UUID) http.Header {
header := make(http.Header)
if source != nil {
for _, key := range []string{codersdk.SessionTokenHeader, authorizationHeader, cookieHeader} {
for _, value := range source.Values(key) {
header.Add(key, value)
}
// extractSessionToken returns the session token carried by the
// given request headers. It mirrors the priority order used by
// apiKeyMiddleware: cookie, then Coder-Session-Token header, then
// Authorization: Bearer header.
func extractSessionToken(header http.Header) string {
if header == nil {
return ""
}
// Cookie (browser WebSocket upgrade — most common relay case).
if raw := header.Get(cookieHeader); raw != "" {
r := &http.Request{Header: http.Header{cookieHeader: {raw}}}
if c, err := r.Cookie(codersdk.SessionTokenCookie); err == nil && c.Value != "" {
return c.Value
}
}
header.Set(RelaySourceHeader, replicaID.String())
return header
// Coder-Session-Token header (SDK / CLI callers).
if v := header.Get(codersdk.SessionTokenHeader); v != "" {
return v
}
// Authorization: Bearer <token>.
if v := header.Get(authorizationHeader); len(v) > 7 && strings.EqualFold(v[:7], "bearer ") {
return strings.TrimSpace(v[7:])
}
return ""
}
+375
View File
@@ -19,6 +19,7 @@ import (
"github.com/coder/coder/v2/enterprise/coderd/coderdenttest"
"github.com/coder/coder/v2/enterprise/coderd/license"
"github.com/coder/coder/v2/testutil"
"github.com/coder/websocket"
)
func TestChatStreamRelay(t *testing.T) {
@@ -367,6 +368,343 @@ func TestChatStreamRelay(t *testing.T) {
close(streamingChunks)
})
// This test verifies that the relay works when the subscriber
// replica's incoming request authenticates via cookies only,
// exactly as a browser WebSocket upgrade does. Browsers cannot
// set custom headers (like Coder-Session-Token) on WebSocket
// connections, so the relay must forward the Cookie header and
// the worker replica must accept it.
//
// Previous tests used SetSessionToken() which sets the
// Coder-Session-Token header, masking this code path.
t.Run("RelayCookieOnlyAuth", func(t *testing.T) {
t.Parallel()
ctx := testutil.Context(t, testutil.WaitLong)
db, pubsub := dbtestutil.NewDB(t)
firstClient, _ := coderdenttest.New(t, &coderdenttest.Options{
Options: &coderdtest.Options{
Database: db,
Pubsub: pubsub,
},
LicenseOptions: &coderdenttest.LicenseOptions{
Features: license.Features{
codersdk.FeatureHighAvailability: 1,
},
},
})
secondClient, _ := coderdenttest.New(t, &coderdenttest.Options{
Options: &coderdtest.Options{
Database: db,
Pubsub: pubsub,
},
DontAddLicense: true,
DontAddFirstUser: true,
})
//nolint:gocritic // Test uses owner client session token for cookie-based relay auth.
sessionToken := firstClient.SessionToken()
// Configure the second client to authenticate via cookies // only for WebSocket dials, matching browser behavior.
// For regular HTTP API calls we still need the header.
secondClient.SetSessionToken(sessionToken)
secondClient.SessionTokenProvider = cookieOnlySessionTokenProvider{
token: sessionToken,
targetURL: secondClient.URL,
}
replicas, err := secondClient.Replicas(ctx)
require.NoError(t, err)
require.Len(t, replicas, 2)
firstReplicaID := replicaIDForClientURL(t, firstClient.URL, replicas)
secondReplicaID := replicaIDForClientURL(t, secondClient.URL, replicas)
streamingChunks := make(chan chattest.OpenAIChunk, 8)
chatStreamStarted := make(chan struct{}, 1)
openai := chattest.NewOpenAI(t, func(req *chattest.OpenAIRequest) chattest.OpenAIResponse {
if req.Stream {
select {
case chatStreamStarted <- struct{}{}:
default:
}
return chattest.OpenAIResponse{StreamingChunks: streamingChunks}
}
return chattest.OpenAINonStreamingResponse("ok")
})
//nolint:gocritic // Test uses owner client to configure providers.
provider, err := firstClient.CreateChatProvider(ctx, codersdk.CreateChatProviderConfigRequest{
Provider: "openai",
DisplayName: "OpenAI",
APIKey: "test",
BaseURL: openai,
})
require.NoError(t, err)
model, err := firstClient.CreateChatModelConfig(ctx, codersdk.CreateChatModelConfigRequest{
Provider: provider.Provider,
Model: "gpt-4",
DisplayName: "GPT-4",
ContextLimit: &[]int64{1000}[0],
CompressionThreshold: &[]int32{70}[0],
})
require.NoError(t, err)
chat, err := firstClient.CreateChat(ctx, codersdk.CreateChatRequest{
Content: []codersdk.ChatInputPart{{
Type: codersdk.ChatInputPartTypeText,
Text: "Test cookie-only relay",
}},
ModelConfigID: &model.ID,
})
require.NoError(t, err)
require.Equal(t, codersdk.ChatStatusPending, chat.Status)
var runningChat database.Chat
require.Eventually(t, func() bool {
current, getErr := db.GetChatByID(ctx, chat.ID)
if getErr != nil {
return false
}
if current.Status != database.ChatStatusRunning || !current.WorkerID.Valid {
return false
}
runningChat = current
return true
}, testutil.WaitLong, testutil.IntervalFast)
var localClient *codersdk.Client
var relayClient *codersdk.Client
switch runningChat.WorkerID.UUID {
case firstReplicaID:
localClient = firstClient
relayClient = secondClient
case secondReplicaID:
localClient = secondClient
relayClient = firstClient
default:
require.FailNowf(
t,
"worker replica was not recognized",
"worker %s was not one of %s or %s",
runningChat.WorkerID.UUID,
firstReplicaID,
secondReplicaID,
)
}
firstEvents, firstStream, err := localClient.StreamChat(ctx, chat.ID, nil)
require.NoError(t, err)
defer firstStream.Close()
select {
case <-chatStreamStarted:
case <-ctx.Done():
require.FailNowf(
t,
"timed out waiting for OpenAI stream request",
"chat stream did not start: %v",
ctx.Err(),
)
}
firstChunkText := "cookie-relay-part-one"
streamingChunks <- chattest.OpenAITextChunks(firstChunkText)[0]
firstEvent := waitForStreamTextPart(ctx, t, firstEvents, firstChunkText)
require.Equal(t, "assistant", firstEvent.MessagePart.Role)
// Subscribe from the non-worker replica with cookie-only
// auth. This triggers the relay dial. If the relay doesn't
// correctly forward cookies, this fails with 401.
secondEvents, secondStream, err := relayClient.StreamChat(ctx, chat.ID, nil)
require.NoError(t, err)
defer secondStream.Close()
secondSnapshotEvent := waitForStreamTextPart(ctx, t, secondEvents, firstChunkText)
require.Equal(t, "assistant", secondSnapshotEvent.MessagePart.Role)
secondChunkText := "cookie-relay-part-two"
streamingChunks <- chattest.OpenAITextChunks(secondChunkText)[0]
waitForStreamTextPart(ctx, t, firstEvents, secondChunkText)
waitForStreamTextPart(ctx, t, secondEvents, secondChunkText)
close(streamingChunks)
})
// This test verifies that cookie-only relay auth works when
// EnableHostPrefix is true. When the subscriber replica's
// HTTPCookies.Middleware normalizes __Host-coder_session_token
// to coder_session_token, the relay forwards the bare cookie.
// On the worker replica, the same middleware must not strip it.
//
// The fix ensures relayHeaders also extracts the token value
// and sets the Coder-Session-Token header so the worker
// replica can authenticate regardless of cookie prefix config.
t.Run("RelayCookieOnlyAuthWithHostPrefix", func(t *testing.T) {
t.Parallel()
ctx := testutil.Context(t, testutil.WaitLong)
db, pubsub := dbtestutil.NewDB(t)
hostPrefixValues := coderdtest.DeploymentValues(t, func(dv *codersdk.DeploymentValues) {
dv.HTTPCookies.EnableHostPrefix = true
dv.HTTPCookies.Secure = true
})
firstClient, _ := coderdenttest.New(t, &coderdenttest.Options{
Options: &coderdtest.Options{
Database: db,
Pubsub: pubsub,
DeploymentValues: hostPrefixValues,
},
LicenseOptions: &coderdenttest.LicenseOptions{
Features: license.Features{
codersdk.FeatureHighAvailability: 1,
},
},
})
secondClient, _ := coderdenttest.New(t, &coderdenttest.Options{
Options: &coderdtest.Options{
Database: db,
Pubsub: pubsub,
DeploymentValues: hostPrefixValues,
},
DontAddLicense: true,
DontAddFirstUser: true,
})
//nolint:gocritic // Test uses owner client session token for cookie-based relay auth.
sessionToken := firstClient.SessionToken()
// Use cookie-only auth for WebSocket, as browsers do. // With EnableHostPrefix, the browser would have
// __Host-coder_session_token but the middleware
// normalizes it. The relay copies the normalized cookie.
secondClient.SetSessionToken(sessionToken)
secondClient.SessionTokenProvider = cookieOnlySessionTokenProvider{
token: sessionToken,
targetURL: secondClient.URL,
hostPrefix: true,
}
replicas, err := secondClient.Replicas(ctx)
require.NoError(t, err)
require.Len(t, replicas, 2)
firstReplicaID := replicaIDForClientURL(t, firstClient.URL, replicas)
secondReplicaID := replicaIDForClientURL(t, secondClient.URL, replicas)
streamingChunks := make(chan chattest.OpenAIChunk, 8)
chatStreamStarted := make(chan struct{}, 1)
openai := chattest.NewOpenAI(t, func(req *chattest.OpenAIRequest) chattest.OpenAIResponse {
if req.Stream {
select {
case chatStreamStarted <- struct{}{}:
default:
}
return chattest.OpenAIResponse{StreamingChunks: streamingChunks}
}
return chattest.OpenAINonStreamingResponse("ok")
})
//nolint:gocritic // Test uses owner client to configure providers.
provider, err := firstClient.CreateChatProvider(ctx, codersdk.CreateChatProviderConfigRequest{
Provider: "openai",
DisplayName: "OpenAI",
APIKey: "test",
BaseURL: openai,
})
require.NoError(t, err)
model, err := firstClient.CreateChatModelConfig(ctx, codersdk.CreateChatModelConfigRequest{
Provider: provider.Provider,
Model: "gpt-4",
DisplayName: "GPT-4",
ContextLimit: &[]int64{1000}[0],
CompressionThreshold: &[]int32{70}[0],
})
require.NoError(t, err)
chat, err := firstClient.CreateChat(ctx, codersdk.CreateChatRequest{
Content: []codersdk.ChatInputPart{{
Type: codersdk.ChatInputPartTypeText,
Text: "Test host-prefix relay",
}},
ModelConfigID: &model.ID,
})
require.NoError(t, err)
require.Equal(t, codersdk.ChatStatusPending, chat.Status)
var runningChat database.Chat
require.Eventually(t, func() bool {
current, getErr := db.GetChatByID(ctx, chat.ID)
if getErr != nil {
return false
}
if current.Status != database.ChatStatusRunning || !current.WorkerID.Valid {
return false
}
runningChat = current
return true
}, testutil.WaitLong, testutil.IntervalFast)
var localClient *codersdk.Client
var relayClient *codersdk.Client
switch runningChat.WorkerID.UUID {
case firstReplicaID:
localClient = firstClient
relayClient = secondClient
case secondReplicaID:
localClient = secondClient
relayClient = firstClient
default:
require.FailNowf(
t,
"worker replica was not recognized",
"worker %s was not one of %s or %s",
runningChat.WorkerID.UUID,
firstReplicaID,
secondReplicaID,
)
}
firstEvents, firstStream, err := localClient.StreamChat(ctx, chat.ID, nil)
require.NoError(t, err)
defer firstStream.Close()
select {
case <-chatStreamStarted:
case <-ctx.Done():
require.FailNowf(
t,
"timed out waiting for OpenAI stream request",
"chat stream did not start: %v",
ctx.Err(),
)
}
firstChunkText := "hostprefix-relay-part-one"
streamingChunks <- chattest.OpenAITextChunks(firstChunkText)[0]
firstEvent := waitForStreamTextPart(ctx, t, firstEvents, firstChunkText)
require.Equal(t, "assistant", firstEvent.MessagePart.Role)
// This subscribe triggers the relay. With the bug, the
// worker replica's HTTPCookies.Middleware strips the bare
// coder_session_token cookie and there's no fallback
// Coder-Session-Token header, causing a 401.
secondEvents, secondStream, err := relayClient.StreamChat(ctx, chat.ID, nil)
require.NoError(t, err)
defer secondStream.Close()
secondSnapshotEvent := waitForStreamTextPart(ctx, t, secondEvents, firstChunkText)
require.Equal(t, "assistant", secondSnapshotEvent.MessagePart.Role)
secondChunkText := "hostprefix-relay-part-two"
streamingChunks <- chattest.OpenAITextChunks(secondChunkText)[0]
waitForStreamTextPart(ctx, t, firstEvents, secondChunkText)
waitForStreamTextPart(ctx, t, secondEvents, secondChunkText)
close(streamingChunks)
})
t.Run("RelaySnapshotIncludesBufferedParts", func(t *testing.T) {
t.Parallel()
ctx := testutil.Context(t, testutil.WaitLong)
@@ -716,3 +1054,40 @@ func findChatModelConfigByID(
require.FailNowf(t, "missing model config", "model config %s not found", id)
return codersdk.ChatModelConfig{}
}
// cookieOnlySessionTokenProvider authenticates HTTP requests via the
// Coder-Session-Token header (for regular API calls) but
// authenticates WebSocket dials via Cookie only, matching how
// browsers behave (the native WebSocket constructor cannot set
// custom headers).
type cookieOnlySessionTokenProvider struct {
token string
targetURL *url.URL
// hostPrefix, when true, sends the cookie with the
// __Host- prefix as browsers do with secure cookies.
hostPrefix bool
}
func (p cookieOnlySessionTokenProvider) AsRequestOption() codersdk.RequestOption {
return func(req *http.Request) {
req.Header.Set(codersdk.SessionTokenHeader, p.token)
}
}
func (p cookieOnlySessionTokenProvider) GetSessionToken() string {
return p.token
}
func (p cookieOnlySessionTokenProvider) SetDialOption(opts *websocket.DialOptions) {
// Browsers send cookies automatically on WebSocket upgrades
// but cannot send custom headers. Simulate this by setting
// only the Cookie header.
if opts.HTTPHeader == nil {
opts.HTTPHeader = make(http.Header)
}
cookieName := codersdk.SessionTokenCookie
if p.hostPrefix {
cookieName = "__Host-" + cookieName
}
opts.HTTPHeader.Set("Cookie", cookieName+"="+p.token)
}