refactor(server::client::auth): removed state machine and added approval flow coordination

IMPLEMENTATION.md

@@ -4,6 +4,52 @@ This document covers concrete technology choices and dependencies. For the archi
 
 ---
 
+## Client Connection Flow
+
+### New Client Approval
+
+When a client whose public key is not yet in the database connects, all connected user agents are asked to approve the connection. The first agent to respond determines the outcome; remaining requests are cancelled via a watch channel.
+
+```mermaid
+flowchart TD
+    A([Client connects]) --> B[Receive AuthChallengeRequest]
+    B --> C{pubkey in DB?}
+
+    C -- yes --> D[Read nonce\nIncrement nonce in DB]
+    D --> G
+
+    C -- no --> E[Ask all UserAgents:\nClientConnectionRequest]
+    E --> F{First response}
+    F -- denied --> Z([Reject connection])
+    F -- approved --> F2[Cancel remaining\nUserAgent requests]
+    F2 --> F3[INSERT client\nnonce = 1]
+    F3 --> G[Send AuthChallenge\nwith nonce]
+
+    G --> H[Receive AuthChallengeSolution]
+    H --> I{Signature valid?}
+    I -- no --> Z
+    I -- yes --> J([Session started])
+```
+
+### Known Issue: Concurrent Registration Race (TOCTOU)
+
+Two connections presenting the same previously-unknown public key can race through the approval flow simultaneously:
+
+1. Both check the DB → neither is registered.
+2. Both request approval from user agents → both receive approval.
+3. Both `INSERT` the client record → the second insert silently overwrites the first, resetting the nonce.
+
+This means the first connection's nonce is invalidated by the second, causing its challenge verification to fail. A fix requires either serialising new-client registration (e.g. an in-memory lock keyed on pubkey) or replacing the separate check + insert with an `INSERT OR IGNORE` / upsert guarded by a unique constraint on `public_key`.
+
+### Nonce Semantics
+
+The `program_client.nonce` column stores the **next usable nonce** — i.e. it is always one ahead of the nonce last issued in a challenge.
+
+- **New client:** inserted with `nonce = 1`; the first challenge is issued with `nonce = 0`.
+- **Existing client:** the current DB value is read and used as the challenge nonce, then immediately incremented within the same exclusive transaction, preventing replay.
+
+---
+
 ## Cryptography
 
 ### Authentication

server/crates/arbiter-proto/src/transport.rs

@@ -83,6 +83,23 @@ use async_trait::async_trait;
 pub enum Error {
     #[error("Transport channel is closed")]
     ChannelClosed,
+    #[error("Unexpected message received")]
+    UnexpectedMessage,
+}
+
+/// Receives one message from `transport` and extracts a value from it using
+/// `extractor`. Returns [`Error::ChannelClosed`] if the transport closes and
+/// [`Error::UnexpectedMessage`] if `extractor` returns `None`.
+pub async fn expect_message<T, Inbound, Outbound, Target, F>(
+    transport: &mut T,
+    extractor: F,
+) -> Result<Target, Error>
+where
+    T: Bi<Inbound, Outbound> + ?Sized,
+    F: FnOnce(Inbound) -> Option<Target>,
+{
+    let msg = transport.recv().await.ok_or(Error::ChannelClosed)?;
+    extractor(msg).ok_or(Error::UnexpectedMessage)
 }
 
 /// Minimal bidirectional transport abstraction used by protocol code.

server/crates/arbiter-server/src/actors/client/auth.rs

@@ -0,0 +1,251 @@
+use arbiter_proto::{
+    format_challenge,
+    proto::client::{
+        AuthChallenge, AuthChallengeSolution, ClientConnectError, ClientRequest, ClientResponse,
+        client_connect_error::Code as ConnectErrorCode,
+        client_request::Payload as ClientRequestPayload,
+        client_response::Payload as ClientResponsePayload,
+    },
+    transport::expect_message,
+};
+use diesel::{
+    ExpressionMethods as _, OptionalExtension as _, QueryDsl as _, dsl::insert_into, update,
+};
+use diesel_async::RunQueryDsl as _;
+use ed25519_dalek::VerifyingKey;
+use kameo::error::SendError;
+use tracing::error;
+
+use crate::{
+    actors::{client::ClientConnection, router::{self, RequestClientApproval}},
+    db::{self, schema::program_client},
+};
+
+use super::session::ClientSession;
+
+#[derive(thiserror::Error, Debug, Clone, PartialEq, Eq)]
+pub enum Error {
+    #[error("Unexpected message payload")]
+    UnexpectedMessagePayload,
+    #[error("Invalid client public key length")]
+    InvalidClientPubkeyLength,
+    #[error("Invalid client public key encoding")]
+    InvalidAuthPubkeyEncoding,
+    #[error("Database pool unavailable")]
+    DatabasePoolUnavailable,
+    #[error("Database operation failed")]
+    DatabaseOperationFailed,
+    #[error("Invalid challenge solution")]
+    InvalidChallengeSolution,
+    #[error("Client approval request failed")]
+    ApproveError(#[from] ApproveError),
+    #[error("Internal error")]
+    InternalError,
+    #[error("Transport error")]
+    Transport,
+}
+
+#[derive(thiserror::Error, Debug, Clone, PartialEq, Eq)]
+pub enum ApproveError {
+    #[error("Internal error")]
+    Internal,
+    #[error("Client connection denied by user agents")]
+    Denied,
+    #[error("Upstream error: {0}")]
+    Upstream(router::ApprovalError),
+}
+
+/// Atomically reads and increments the nonce for a known client.
+/// Returns `None` if the pubkey is not registered.
+async fn get_nonce(db: &db::DatabasePool, pubkey: &VerifyingKey) -> Result<Option<i32>, Error> {
+    let pubkey_bytes = pubkey.as_bytes().to_vec();
+
+    let mut conn = db.get().await.map_err(|e| {
+        error!(error = ?e, "Database pool error");
+        Error::DatabasePoolUnavailable
+    })?;
+
+    conn.exclusive_transaction(|conn| {
+        let pubkey_bytes = pubkey_bytes.clone();
+        Box::pin(async move {
+            let Some(current_nonce) = program_client::table
+                .filter(program_client::public_key.eq(&pubkey_bytes))
+                .select(program_client::nonce)
+                .first::<i32>(conn)
+                .await
+                .optional()?
+            else {
+                return Result::<_, diesel::result::Error>::Ok(None);
+            };
+
+            update(program_client::table)
+                .filter(program_client::public_key.eq(&pubkey_bytes))
+                .set(program_client::nonce.eq(current_nonce + 1))
+                .execute(conn)
+                .await?;
+
+            Ok(Some(current_nonce))
+        })
+    })
+    .await
+    .map_err(|e| {
+        error!(error = ?e, "Database error");
+        Error::DatabaseOperationFailed
+    })
+}
+
+async fn approve_new_client(
+    actors: &crate::actors::GlobalActors,
+    pubkey: VerifyingKey,
+) -> Result<(), Error> {
+    let result = actors
+        .router
+        .ask(RequestClientApproval { client_pubkey: pubkey })
+        .await;
+
+    match result {
+        Ok(true) => Ok(()),
+        Ok(false) => Err(Error::ApproveError(ApproveError::Denied)),
+        Err(SendError::HandlerError(e)) => {
+            error!(error = ?e, "Approval upstream error");
+            Err(Error::ApproveError(ApproveError::Upstream(e)))
+        }
+        Err(e) => {
+            error!(error = ?e, "Approval request to router failed");
+            Err(Error::ApproveError(ApproveError::Internal))
+        }
+    }
+}
+
+async fn insert_client(db: &db::DatabasePool, pubkey: &VerifyingKey) -> Result<(), Error> {
+    let now = std::time::SystemTime::now()
+        .duration_since(std::time::UNIX_EPOCH)
+        .unwrap_or_default()
+        .as_secs() as i32;
+
+    let mut conn = db.get().await.map_err(|e| {
+        error!(error = ?e, "Database pool error");
+        Error::DatabasePoolUnavailable
+    })?;
+
+    insert_into(program_client::table)
+        .values((
+            program_client::public_key.eq(pubkey.as_bytes().to_vec()),
+            program_client::nonce.eq(1), // pre-incremented; challenge uses 0
+            program_client::created_at.eq(now),
+            program_client::updated_at.eq(now),
+        ))
+        .execute(&mut conn)
+        .await
+        .map_err(|e| {
+            error!(error = ?e, "Failed to insert new client");
+            Error::DatabaseOperationFailed
+        })?;
+
+    Ok(())
+}
+
+async fn challenge_client(
+    props: &mut ClientConnection,
+    pubkey: VerifyingKey,
+    nonce: i32,
+) -> Result<(), Error> {
+    let challenge = AuthChallenge {
+        pubkey: pubkey.as_bytes().to_vec(),
+        nonce,
+    };
+
+    props
+        .transport
+        .send(Ok(ClientResponse {
+            payload: Some(ClientResponsePayload::AuthChallenge(challenge.clone())),
+        }))
+        .await
+        .map_err(|e| {
+            error!(error = ?e, "Failed to send auth challenge");
+            Error::Transport
+        })?;
+
+    let AuthChallengeSolution { signature } = expect_message(
+        &mut *props.transport,
+        |req: ClientRequest| match req.payload? {
+            ClientRequestPayload::AuthChallengeSolution(s) => Some(s),
+            _ => None,
+        },
+    )
+    .await
+    .map_err(|e| {
+        error!(error = ?e, "Failed to receive challenge solution");
+        Error::Transport
+    })?;
+
+    let formatted = format_challenge(nonce, &challenge.pubkey);
+    let sig = signature.as_slice().try_into().map_err(|_| {
+        error!("Invalid signature length");
+        Error::InvalidChallengeSolution
+    })?;
+
+    pubkey.verify_strict(&formatted, &sig).map_err(|_| {
+        error!("Challenge solution verification failed");
+        Error::InvalidChallengeSolution
+    })?;
+
+    Ok(())
+}
+
+fn connect_error_code(err: &Error) -> ConnectErrorCode {
+    match err {
+        Error::ApproveError(ApproveError::Denied) => ConnectErrorCode::ApprovalDenied,
+        Error::ApproveError(ApproveError::Upstream(router::ApprovalError::NoUserAgentsConnected)) => {
+            ConnectErrorCode::NoUserAgentsOnline
+        }
+        _ => ConnectErrorCode::Unknown,
+    }
+}
+
+async fn authenticate(props: &mut ClientConnection) -> Result<VerifyingKey, Error> {
+    let Some(ClientRequest {
+        payload: Some(ClientRequestPayload::AuthChallengeRequest(challenge)),
+    }) = props.transport.recv().await
+    else {
+        return Err(Error::Transport);
+    };
+
+    let pubkey_bytes = challenge
+        .pubkey
+        .as_array()
+        .ok_or(Error::InvalidClientPubkeyLength)?;
+    let pubkey =
+        VerifyingKey::from_bytes(pubkey_bytes).map_err(|_| Error::InvalidAuthPubkeyEncoding)?;
+
+    let nonce = match get_nonce(&props.db, &pubkey).await? {
+        Some(nonce) => nonce,
+        None => {
+            approve_new_client(&props.actors, pubkey).await?;
+            insert_client(&props.db, &pubkey).await?;
+            0
+        }
+    };
+
+    challenge_client(props, pubkey, nonce).await?;
+
+    Ok(pubkey)
+}
+
+pub async fn authenticate_and_create(mut props: ClientConnection) -> Result<ClientSession, Error> {
+    match authenticate(&mut props).await {
+        Ok(pubkey) => Ok(ClientSession::new(props, pubkey)),
+        Err(err) => {
+            let code = connect_error_code(&err);
+            let _ = props
+                .transport
+                .send(Ok(ClientResponse {
+                    payload: Some(ClientResponsePayload::ClientConnectError(
+                        ClientConnectError { code: code.into() },
+                    )),
+                }))
+                .await;
+            Err(err)
+        }
+    }
+}

server/crates/arbiter-server/src/actors/client/auth/mod.rs

@@ -1,101 +0,0 @@
-use arbiter_proto::proto::client::{
-    AuthChallengeRequest, AuthChallengeSolution, ClientRequest,
-    client_request::Payload as ClientRequestPayload,
-};
-use ed25519_dalek::VerifyingKey;
-use tracing::error;
-
-use crate::actors::client::{
-    ClientConnection,
-    auth::state::{AuthContext, AuthStateMachine},
-    session::ClientSession,
-};
-
-#[derive(thiserror::Error, Debug, Clone, PartialEq, Eq)]
-pub enum Error {
-    #[error("Unexpected message payload")]
-    UnexpectedMessagePayload,
-    #[error("Invalid client public key length")]
-    InvalidClientPubkeyLength,
-    #[error("Invalid client public key encoding")]
-    InvalidAuthPubkeyEncoding,
-    #[error("Database pool unavailable")]
-    DatabasePoolUnavailable,
-    #[error("Database operation failed")]
-    DatabaseOperationFailed,
-    #[error("Public key not registered")]
-    PublicKeyNotRegistered,
-    #[error("Invalid signature length")]
-    InvalidSignatureLength,
-    #[error("Invalid challenge solution")]
-    InvalidChallengeSolution,
-    #[error("Transport error")]
-    Transport,
-}
-
-mod state;
-use state::*;
-
-fn parse_auth_event(payload: ClientRequestPayload) -> Result<AuthEvents, Error> {
-    match payload {
-        ClientRequestPayload::AuthChallengeRequest(AuthChallengeRequest { pubkey }) => {
-            let pubkey_bytes = pubkey.as_array().ok_or(Error::InvalidClientPubkeyLength)?;
-            let pubkey = VerifyingKey::from_bytes(pubkey_bytes)
-                .map_err(|_| Error::InvalidAuthPubkeyEncoding)?;
-            Ok(AuthEvents::AuthRequest(ChallengeRequest {
-                pubkey: pubkey.into(),
-            }))
-        }
-        ClientRequestPayload::AuthChallengeSolution(AuthChallengeSolution { signature }) => {
-            Ok(AuthEvents::ReceivedSolution(ChallengeSolution {
-                solution: signature,
-            }))
-        }
-    }
-}
-
-pub async fn authenticate(props: &mut ClientConnection) -> Result<VerifyingKey, Error> {
-    let mut state = AuthStateMachine::new(AuthContext::new(props));
-
-    loop {
-        let transport = state.context_mut().conn.transport.as_mut();
-        let Some(ClientRequest {
-            payload: Some(payload),
-        }) = transport.recv().await
-        else {
-            return Err(Error::Transport);
-        };
-
-        let event = parse_auth_event(payload)?;
-
-        match state.process_event(event).await {
-            Ok(AuthStates::AuthOk(key)) => return Ok(key.clone()),
-            Err(AuthError::ActionFailed(err)) => {
-                error!(?err, "State machine action failed");
-                return Err(err);
-            }
-            Err(AuthError::GuardFailed(err)) => {
-                error!(?err, "State machine guard failed");
-                return Err(err);
-            }
-            Err(AuthError::InvalidEvent) => {
-                error!("Invalid event for current state");
-                return Err(Error::InvalidChallengeSolution);
-            }
-            Err(AuthError::TransitionsFailed) => {
-                error!("Invalid state transition");
-                return Err(Error::InvalidChallengeSolution);
-            }
-
-            _ => (),
-        }
-    }
-}
-
-pub async fn authenticate_and_create(
-    mut props: ClientConnection,
-) -> Result<ClientSession, Error> {
-    let key = authenticate(&mut props).await?;
-    let session = ClientSession::new(props, key);
-    Ok(session)
-}

server/crates/arbiter-server/src/actors/client/auth/state.rs

@@ -1,136 +0,0 @@
-use arbiter_proto::proto::client::{
-    AuthChallenge, ClientResponse,
-    client_response::Payload as ClientResponsePayload,
-};
-use diesel::{ExpressionMethods as _, OptionalExtension as _, QueryDsl, update};
-use diesel_async::RunQueryDsl;
-use ed25519_dalek::VerifyingKey;
-use tracing::error;
-
-use super::Error;
-use crate::{actors::client::ClientConnection, db::schema};
-
-pub struct ChallengeRequest {
-    pub pubkey: VerifyingKey,
-}
-
-pub struct ChallengeContext {
-    pub challenge: AuthChallenge,
-    pub key: VerifyingKey,
-}
-
-pub struct ChallengeSolution {
-    pub solution: Vec<u8>,
-}
-
-smlang::statemachine!(
-    name: Auth,
-    custom_error: true,
-    transitions: {
-        *Init + AuthRequest(ChallengeRequest) / async prepare_challenge = SentChallenge(ChallengeContext),
-        SentChallenge(ChallengeContext) + ReceivedSolution(ChallengeSolution) [async verify_solution] / provide_key = AuthOk(VerifyingKey),
-    }
-);
-
-async fn create_nonce(db: &crate::db::DatabasePool, pubkey_bytes: &[u8]) -> Result<i32, Error> {
-    let mut db_conn = db.get().await.map_err(|e| {
-        error!(error = ?e, "Database pool error");
-        Error::DatabasePoolUnavailable
-    })?;
-    db_conn
-        .exclusive_transaction(|conn| {
-            Box::pin(async move {
-                let current_nonce = schema::program_client::table
-                    .filter(schema::program_client::public_key.eq(pubkey_bytes.to_vec()))
-                    .select(schema::program_client::nonce)
-                    .first::<i32>(conn)
-                    .await?;
-
-                update(schema::program_client::table)
-                    .filter(schema::program_client::public_key.eq(pubkey_bytes.to_vec()))
-                    .set(schema::program_client::nonce.eq(current_nonce + 1))
-                    .execute(conn)
-                    .await?;
-
-                Result::<_, diesel::result::Error>::Ok(current_nonce)
-            })
-        })
-        .await
-        .optional()
-        .map_err(|e| {
-            error!(error = ?e, "Database error");
-            Error::DatabaseOperationFailed
-        })?
-        .ok_or_else(|| {
-            error!(?pubkey_bytes, "Public key not found in database");
-            Error::PublicKeyNotRegistered
-        })
-}
-
-pub struct AuthContext<'a> {
-    pub(super) conn: &'a mut ClientConnection,
-}
-
-impl<'a> AuthContext<'a> {
-    pub fn new(conn: &'a mut ClientConnection) -> Self {
-        Self { conn }
-    }
-}
-
-impl AuthStateMachineContext for AuthContext<'_> {
-    type Error = Error;
-
-    async fn verify_solution(
-        &self,
-        ChallengeContext { challenge, key }: &ChallengeContext,
-        ChallengeSolution { solution }: &ChallengeSolution,
-    ) -> Result<bool, Self::Error> {
-        let formatted_challenge =
-            arbiter_proto::format_challenge(challenge.nonce, &challenge.pubkey);
-
-        let signature = solution.as_slice().try_into().map_err(|_| {
-            error!(?solution, "Invalid signature length");
-            Error::InvalidChallengeSolution
-        })?;
-
-        let valid = key.verify_strict(&formatted_challenge, &signature).is_ok();
-
-        Ok(valid)
-    }
-
-    async fn prepare_challenge(
-        &mut self,
-        ChallengeRequest { pubkey }: ChallengeRequest,
-    ) -> Result<ChallengeContext, Self::Error> {
-        let nonce = create_nonce(&self.conn.db, pubkey.as_bytes()).await?;
-
-        let challenge = AuthChallenge {
-            pubkey: pubkey.as_bytes().to_vec(),
-            nonce,
-        };
-
-        self.conn
-            .transport
-            .send(Ok(ClientResponse {
-                payload: Some(ClientResponsePayload::AuthChallenge(challenge.clone())),
-            }))
-            .await
-            .map_err(|e| {
-                error!(?e, "Failed to send auth challenge");
-                Error::Transport
-            })?;
-
-        Ok(ChallengeContext {
-            challenge,
-            key: pubkey,
-        })
-    }
-
-    fn provide_key(
-        &mut self,
-        state_data: &ChallengeContext,
-        _: ChallengeSolution,
-    ) -> Result<VerifyingKey, Self::Error> {
-        Ok(state_data.key)
-    }
-}

server/crates/arbiter-server/src/actors/router/mod.rs

@@ -56,7 +56,7 @@ impl Actor for MessageRouter {
     }
 }
 
-#[derive(Debug, thiserror::Error)]
+#[derive(Debug, thiserror::Error, Clone, PartialEq, Eq, Hash)]
 pub enum ApprovalError {
     #[error("No user agents connected")]
     NoUserAgentsConnected,

server/crates/arbiter-server/src/actors/user_agent/session.rs

@@ -1,17 +1,14 @@
 use std::{ops::DerefMut, sync::Mutex};
 
-use arbiter_proto::proto::{
-    client,
-    user_agent::{
+use arbiter_proto::proto::user_agent::{
         ClientConnectionCancel, ClientConnectionRequest, UnsealEncryptedKey, UnsealResult,
         UnsealStart, UnsealStartResponse, UserAgentRequest, UserAgentResponse,
         user_agent_request::Payload as UserAgentRequestPayload,
         user_agent_response::Payload as UserAgentResponsePayload,
-    },
-};
+    };
 use chacha20poly1305::{AeadInPlace, XChaCha20Poly1305, XNonce, aead::KeyInit};
 use ed25519_dalek::VerifyingKey;
-use kameo::{Actor, error::SendError, message, messages, prelude::Context};
+use kameo::{Actor, error::SendError, messages, prelude::Context};
 use memsafe::MemSafe;
 use tokio::{select, sync::watch};
 use tracing::{error, info};
@@ -62,7 +59,7 @@ impl UserAgentSession {
     async fn send_msg<Reply: kameo::Reply>(
         &mut self,
         msg: UserAgentResponsePayload,
-        ctx: &mut Context<Self, Reply>,
+        _ctx: &mut Context<Self, Reply>,
     ) -> Result<(), Error> {
         self.props
             .transport
@@ -111,7 +108,6 @@ impl UserAgentSession {
 
 #[messages]
 impl UserAgentSession {
-
     // TODO: Think about refactoring it to state-machine based flow, as we already have one 
     #[message(ctx)]
     pub async fn request_new_client_approval(

server/crates/arbiter-server/src/db/models.rs

@@ -1,12 +1,13 @@
 #![allow(unused)]
 #![allow(clippy::all)]
 
-use crate::db::schema::{self, aead_encrypted, arbiter_settings, root_key_history, tls_history};
-use diesel::{prelude::*, sqlite::Sqlite};
+use crate::db::{ schema::{self, aead_encrypted, arbiter_settings, root_key_history, tls_history}};
+use diesel::{prelude::*, sql_types::Bool, sqlite::Sqlite};
 use restructed::Models;
 
 pub mod types {
     use chrono::{DateTime, Utc};
+    use diesel::{deserialize::FromSql, expression::AsExpression, serialize::{IsNull, ToSql}, sql_types::Integer, sqlite::Sqlite};
     pub struct SqliteTimestamp(DateTime<Utc>);
 }
 
@@ -58,8 +59,8 @@ pub struct TlsHistory {
     pub id: i32,
     pub cert: String,
     pub cert_key: String, // PEM Encoded private key
-    pub ca_cert: String, // PEM Encoded certificate for cert signing
-    pub ca_key: String, // PEM Encoded public key for cert signing
+    pub ca_cert: String,  // PEM Encoded certificate for cert signing
+    pub ca_key: String,   // PEM Encoded public key for cert signing
     pub created_at: i32,
 }
 
@@ -68,10 +69,10 @@ pub struct TlsHistory {
 pub struct ArbiterSettings {
     pub id: i32,
     pub root_key_id: Option<i32>, // references root_key_history.id
-    pub tls_id: Option<i32>, // references tls_history.id
+    pub tls_id: Option<i32>,      // references tls_history.id
 }
 
-#[derive(Queryable, Debug)]
+#[derive(Queryable, Debug, Insertable, Selectable)]
 #[diesel(table_name = schema::program_client, check_for_backend(Sqlite))]
 pub struct ProgramClient {
     pub id: i32,

server/crates/arbiter-server/src/lib.rs

@@ -77,13 +77,12 @@ fn client_auth_error_status(value: &client::auth::Error) -> Status {
         Error::InvalidAuthPubkeyEncoding => {
             Status::invalid_argument("Failed to convert pubkey to VerifyingKey")
         }
-        Error::InvalidSignatureLength => Status::invalid_argument("Invalid signature length"),
-        Error::PublicKeyNotRegistered | Error::InvalidChallengeSolution => {
-            Status::unauthenticated(value.to_string())
-        }
+        Error::InvalidChallengeSolution => Status::unauthenticated(value.to_string()),
+        Error::ApproveError(_) => Status::permission_denied(value.to_string()),
         Error::Transport => Status::internal("Transport error"),
         Error::DatabasePoolUnavailable => Status::internal("Database pool error"),
         Error::DatabaseOperationFailed => Status::internal("Database error"),
+        Error::InternalError => Status::internal("Internal error"),
     }
 }