attachment_service: JWT auth enforcement (#6897)

## Problem
Attachment service does not do auth based on JWT scopes.

## Summary of changes
Do JWT based permission checking for requests coming into the attachment
service.

Requests into the attachment service must use different tokens based on
the endpoint:
* `/control` and `/debug` require `admin` scope
* `/upcall` requires `generations_api` scope
* `/v1/...` requires `pageserverapi` scope

Requests into the pageserver from the attachment service must use
`pageserverapi` scope.

control_plane/attachment_service/src/auth.rs

@@ -0,0 +1,9 @@
+use utils::auth::{AuthError, Claims, Scope};
+
+pub fn check_permission(claims: &Claims, required_scope: Scope) -> Result<(), AuthError> {
+    if claims.scope != required_scope {
+        return Err(AuthError("Scope mismatch. Permission denied".into()));
+    }
+
+    Ok(())
+}

control_plane/attachment_service/src/http.rs

@@ -10,8 +10,8 @@ use pageserver_api::shard::TenantShardId;
 use pageserver_client::mgmt_api;
 use std::sync::Arc;
 use std::time::{Duration, Instant};
-use utils::auth::SwappableJwtAuth;
-use utils::http::endpoint::{auth_middleware, request_span};
+use utils::auth::{Scope, SwappableJwtAuth};
+use utils::http::endpoint::{auth_middleware, check_permission_with, request_span};
 use utils::http::request::{must_get_query_param, parse_request_param};
 use utils::id::{TenantId, TimelineId};
 
@@ -64,6 +64,8 @@ fn get_state(request: &Request<Body>) -> &HttpState {
 
 /// Pageserver calls into this on startup, to learn which tenants it should attach
 async fn handle_re_attach(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
+    check_permissions(&req, Scope::GenerationsApi)?;
+
     let reattach_req = json_request::<ReAttachRequest>(&mut req).await?;
     let state = get_state(&req);
     json_response(StatusCode::OK, state.service.re_attach(reattach_req).await?)
@@ -72,6 +74,8 @@ async fn handle_re_attach(mut req: Request<Body>) -> Result<Response<Body>, ApiE
 /// Pageserver calls into this before doing deletions, to confirm that it still
 /// holds the latest generation for the tenants with deletions enqueued
 async fn handle_validate(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
+    check_permissions(&req, Scope::GenerationsApi)?;
+
     let validate_req = json_request::<ValidateRequest>(&mut req).await?;
     let state = get_state(&req);
     json_response(StatusCode::OK, state.service.validate(validate_req))
@@ -81,6 +85,8 @@ async fn handle_validate(mut req: Request<Body>) -> Result<Response<Body>, ApiEr
 /// (in the real control plane this is unnecessary, because the same program is managing
 ///  generation numbers and doing attachments).
 async fn handle_attach_hook(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
+    check_permissions(&req, Scope::Admin)?;
+
     let attach_req = json_request::<AttachHookRequest>(&mut req).await?;
     let state = get_state(&req);
 
@@ -95,6 +101,8 @@ async fn handle_attach_hook(mut req: Request<Body>) -> Result<Response<Body>, Ap
 }
 
 async fn handle_inspect(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
+    check_permissions(&req, Scope::Admin)?;
+
     let inspect_req = json_request::<InspectRequest>(&mut req).await?;
 
     let state = get_state(&req);
@@ -106,6 +114,8 @@ async fn handle_tenant_create(
     service: Arc<Service>,
     mut req: Request<Body>,
 ) -> Result<Response<Body>, ApiError> {
+    check_permissions(&req, Scope::PageServerApi)?;
+
     let create_req = json_request::<TenantCreateRequest>(&mut req).await?;
     json_response(
         StatusCode::CREATED,
@@ -164,6 +174,8 @@ async fn handle_tenant_location_config(
     mut req: Request<Body>,
 ) -> Result<Response<Body>, ApiError> {
     let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
+    check_permissions(&req, Scope::PageServerApi)?;
+
     let config_req = json_request::<TenantLocationConfigRequest>(&mut req).await?;
     json_response(
         StatusCode::OK,
@@ -178,6 +190,8 @@ async fn handle_tenant_time_travel_remote_storage(
     mut req: Request<Body>,
 ) -> Result<Response<Body>, ApiError> {
     let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
+    check_permissions(&req, Scope::PageServerApi)?;
+
     let time_travel_req = json_request::<TenantTimeTravelRequest>(&mut req).await?;
 
     let timestamp_raw = must_get_query_param(&req, "travel_to")?;
@@ -211,6 +225,7 @@ async fn handle_tenant_delete(
     req: Request<Body>,
 ) -> Result<Response<Body>, ApiError> {
     let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
+    check_permissions(&req, Scope::PageServerApi)?;
 
     deletion_wrapper(service, move |service| async move {
         service.tenant_delete(tenant_id).await
@@ -223,6 +238,8 @@ async fn handle_tenant_timeline_create(
     mut req: Request<Body>,
 ) -> Result<Response<Body>, ApiError> {
     let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
+    check_permissions(&req, Scope::PageServerApi)?;
+
     let create_req = json_request::<TimelineCreateRequest>(&mut req).await?;
     json_response(
         StatusCode::CREATED,
@@ -237,6 +254,8 @@ async fn handle_tenant_timeline_delete(
     req: Request<Body>,
 ) -> Result<Response<Body>, ApiError> {
     let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
+    check_permissions(&req, Scope::PageServerApi)?;
+
     let timeline_id: TimelineId = parse_request_param(&req, "timeline_id")?;
 
     deletion_wrapper(service, move |service| async move {
@@ -250,6 +269,7 @@ async fn handle_tenant_timeline_passthrough(
     req: Request<Body>,
 ) -> Result<Response<Body>, ApiError> {
     let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
+    check_permissions(&req, Scope::PageServerApi)?;
 
     let Some(path) = req.uri().path_and_query() else {
         // This should never happen, our request router only calls us if there is a path
@@ -293,11 +313,15 @@ async fn handle_tenant_locate(
     service: Arc<Service>,
     req: Request<Body>,
 ) -> Result<Response<Body>, ApiError> {
+    check_permissions(&req, Scope::Admin)?;
+
     let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
     json_response(StatusCode::OK, service.tenant_locate(tenant_id)?)
 }
 
 async fn handle_node_register(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
+    check_permissions(&req, Scope::Admin)?;
+
     let register_req = json_request::<NodeRegisterRequest>(&mut req).await?;
     let state = get_state(&req);
     state.service.node_register(register_req).await?;
@@ -305,17 +329,23 @@ async fn handle_node_register(mut req: Request<Body>) -> Result<Response<Body>,
 }
 
 async fn handle_node_list(req: Request<Body>) -> Result<Response<Body>, ApiError> {
+    check_permissions(&req, Scope::Admin)?;
+
     let state = get_state(&req);
     json_response(StatusCode::OK, state.service.node_list().await?)
 }
 
 async fn handle_node_drop(req: Request<Body>) -> Result<Response<Body>, ApiError> {
+    check_permissions(&req, Scope::Admin)?;
+
     let state = get_state(&req);
     let node_id: NodeId = parse_request_param(&req, "node_id")?;
     json_response(StatusCode::OK, state.service.node_drop(node_id).await?)
 }
 
 async fn handle_node_configure(mut req: Request<Body>) -> Result<Response<Body>, ApiError> {
+    check_permissions(&req, Scope::Admin)?;
+
     let node_id: NodeId = parse_request_param(&req, "node_id")?;
     let config_req = json_request::<NodeConfigureRequest>(&mut req).await?;
     if node_id != config_req.node_id {
@@ -335,6 +365,8 @@ async fn handle_tenant_shard_split(
     service: Arc<Service>,
     mut req: Request<Body>,
 ) -> Result<Response<Body>, ApiError> {
+    check_permissions(&req, Scope::Admin)?;
+
     let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
     let split_req = json_request::<TenantShardSplitRequest>(&mut req).await?;
 
@@ -348,6 +380,8 @@ async fn handle_tenant_shard_migrate(
     service: Arc<Service>,
     mut req: Request<Body>,
 ) -> Result<Response<Body>, ApiError> {
+    check_permissions(&req, Scope::Admin)?;
+
     let tenant_shard_id: TenantShardId = parse_request_param(&req, "tenant_shard_id")?;
     let migrate_req = json_request::<TenantShardMigrateRequest>(&mut req).await?;
     json_response(
@@ -360,22 +394,30 @@ async fn handle_tenant_shard_migrate(
 
 async fn handle_tenant_drop(req: Request<Body>) -> Result<Response<Body>, ApiError> {
     let tenant_id: TenantId = parse_request_param(&req, "tenant_id")?;
+    check_permissions(&req, Scope::PageServerApi)?;
+
     let state = get_state(&req);
 
     json_response(StatusCode::OK, state.service.tenant_drop(tenant_id).await?)
 }
 
 async fn handle_tenants_dump(req: Request<Body>) -> Result<Response<Body>, ApiError> {
+    check_permissions(&req, Scope::Admin)?;
+
     let state = get_state(&req);
     state.service.tenants_dump()
 }
 
 async fn handle_scheduler_dump(req: Request<Body>) -> Result<Response<Body>, ApiError> {
+    check_permissions(&req, Scope::Admin)?;
+
     let state = get_state(&req);
     state.service.scheduler_dump()
 }
 
 async fn handle_consistency_check(req: Request<Body>) -> Result<Response<Body>, ApiError> {
+    check_permissions(&req, Scope::Admin)?;
+
     let state = get_state(&req);
 
     json_response(StatusCode::OK, state.service.consistency_check().await?)
@@ -432,6 +474,12 @@ where
     .await
 }
 
+fn check_permissions(request: &Request<Body>, required_scope: Scope) -> Result<(), ApiError> {
+    check_permission_with(request, |claims| {
+        crate::auth::check_permission(claims, required_scope)
+    })
+}
+
 pub fn make_router(
     service: Arc<Service>,
     auth: Option<Arc<SwappableJwtAuth>>,

control_plane/attachment_service/src/lib.rs

@@ -1,6 +1,7 @@
 use serde::{Deserialize, Serialize};
 use utils::seqwait::MonotonicCounter;
 
+mod auth;
 mod compute_hook;
 pub mod http;
 pub mod metrics;

control_plane/src/attachment_service.rs

@@ -11,12 +11,12 @@ use pageserver_api::{
 use pageserver_client::mgmt_api::ResponseErrorMessageExt;
 use postgres_backend::AuthType;
 use serde::{de::DeserializeOwned, Deserialize, Serialize};
-use std::str::FromStr;
+use std::{fs, str::FromStr};
 use tokio::process::Command;
 use tracing::instrument;
 use url::Url;
 use utils::{
-    auth::{Claims, Scope},
+    auth::{encode_from_key_file, Claims, Scope},
     id::{NodeId, TenantId},
 };
 
@@ -24,7 +24,7 @@ pub struct AttachmentService {
     env: LocalEnv,
     listen: String,
     path: Utf8PathBuf,
-    jwt_token: Option<String>,
+    private_key: Option<Vec<u8>>,
     public_key: Option<String>,
     postgres_port: u16,
     client: reqwest::Client,
@@ -204,12 +204,11 @@ impl AttachmentService {
             .pageservers
             .first()
             .expect("Config is validated to contain at least one pageserver");
-        let (jwt_token, public_key) = match ps_conf.http_auth_type {
+        let (private_key, public_key) = match ps_conf.http_auth_type {
             AuthType::Trust => (None, None),
             AuthType::NeonJWT => {
-                let jwt_token = env
-                    .generate_auth_token(&Claims::new(None, Scope::PageServerApi))
-                    .unwrap();
+                let private_key_path = env.get_private_key_path();
+                let private_key = fs::read(private_key_path).expect("failed to read private key");
 
                 // If pageserver auth is enabled, this implicitly enables auth for this service,
                 // using the same credentials.
@@ -235,7 +234,7 @@ impl AttachmentService {
                 } else {
                     std::fs::read_to_string(&public_key_path).expect("Can't read public key")
                 };
-                (Some(jwt_token), Some(public_key))
+                (Some(private_key), Some(public_key))
             }
         };
 
@@ -243,7 +242,7 @@ impl AttachmentService {
             env: env.clone(),
             path,
             listen,
-            jwt_token,
+            private_key,
             public_key,
             postgres_port,
             client: reqwest::ClientBuilder::new()
@@ -397,7 +396,10 @@ impl AttachmentService {
         .into_iter()
         .map(|s| s.to_string())
         .collect::<Vec<_>>();
-        if let Some(jwt_token) = &self.jwt_token {
+        if let Some(private_key) = &self.private_key {
+            let claims = Claims::new(None, Scope::PageServerApi);
+            let jwt_token =
+                encode_from_key_file(&claims, private_key).expect("failed to generate jwt token");
             args.push(format!("--jwt-token={jwt_token}"));
         }
 
@@ -468,6 +470,20 @@ impl AttachmentService {
         Ok(())
     }
 
+    fn get_claims_for_path(path: &str) -> anyhow::Result<Option<Claims>> {
+        let category = match path.find('/') {
+            Some(idx) => &path[..idx],
+            None => path,
+        };
+
+        match category {
+            "status" | "ready" => Ok(None),
+            "control" | "debug" => Ok(Some(Claims::new(None, Scope::Admin))),
+            "v1" => Ok(Some(Claims::new(None, Scope::PageServerApi))),
+            _ => Err(anyhow::anyhow!("Failed to determine claims for {}", path)),
+        }
+    }
+
     /// Simple HTTP request wrapper for calling into attachment service
     async fn dispatch<RQ, RS>(
         &self,
@@ -493,11 +509,16 @@ impl AttachmentService {
         if let Some(body) = body {
             builder = builder.json(&body)
         }
-        if let Some(jwt_token) = &self.jwt_token {
-            builder = builder.header(
-                reqwest::header::AUTHORIZATION,
-                format!("Bearer {jwt_token}"),
-            );
+        if let Some(private_key) = &self.private_key {
+            println!("Getting claims for path {}", path);
+            if let Some(required_claims) = Self::get_claims_for_path(&path)? {
+                println!("Got claims {:?} for path {}", required_claims, path);
+                let jwt_token = encode_from_key_file(&required_claims, private_key)?;
+                builder = builder.header(
+                    reqwest::header::AUTHORIZATION,
+                    format!("Bearer {jwt_token}"),
+                );
+            }
         }
 
         let response = builder.send().await?;

control_plane/src/local_env.rs

@@ -412,14 +412,17 @@ impl LocalEnv {
 
     // this function is used only for testing purposes in CLI e g generate tokens during init
     pub fn generate_auth_token(&self, claims: &Claims) -> anyhow::Result<String> {
-        let private_key_path = if self.private_key_path.is_absolute() {
+        let private_key_path = self.get_private_key_path();
+        let key_data = fs::read(private_key_path)?;
+        encode_from_key_file(claims, &key_data)
+    }
+
+    pub fn get_private_key_path(&self) -> PathBuf {
+        if self.private_key_path.is_absolute() {
             self.private_key_path.to_path_buf()
         } else {
             self.base_data_dir.join(&self.private_key_path)
-        };
-
-        let key_data = fs::read(private_key_path)?;
-        encode_from_key_file(claims, &key_data)
+        }
     }
 
     //

control_plane/src/pageserver.rs

@@ -115,7 +115,7 @@ impl PageServerNode {
             if matches!(self.conf.http_auth_type, AuthType::NeonJWT) {
                 let jwt_token = self
                     .env
-                    .generate_auth_token(&Claims::new(None, Scope::PageServerApi))
+                    .generate_auth_token(&Claims::new(None, Scope::GenerationsApi))
                     .unwrap();
                 overrides.push(format!("control_plane_api_token='{}'", jwt_token));
             }