Files
alknet/docs/architecture/crates/core/auth.md
glm-5.2 f224ea998c docs(arch): ADR-030..033 — repo/adapter pattern, PeerEntry, CredentialStore, forwarded-for
Land the storage and auth strategy research (findings.md) as four
accepted ADRs and amend the core and call specs to match:

- ADR-030: PeerEntry and Identity.id decoupling. Replaces
  authorized_fingerprints with peers: Vec<PeerEntry>; Identity.id becomes
  the stable peer_id, decoupled from the rotating fingerprint. Supersedes
  ADR-029 Assumption 1's UUID source (one-way door preserved, source
  changes). Resolves OQ-33 and the storage-boundary half of OQ-34. Records
  the API-key asymmetry as deliberate (OQ-35).

- ADR-031: CredentialStore repo trait + InMemoryCredentialStore default
  adapter in core. Second repo trait alongside IdentityProvider. Vault
  encrypts; the store persists the EncryptedData blob; assembly layer
  loads into Capabilities. EncryptedData core mirror includes salt for
  wire-format compat.

- ADR-032: Forwarded-for identity. forwarded_for field on call.requested
  and OperationContext — metadata only, never read by AccessControl::check
  (enforced structurally via the check signature). The from_call handler
  populates it. Wire-format one-way door, folded into the ADR-029
  migration window.

- ADR-033: Storage boundary and repo/adapter pattern. Core defines repo
  traits + in-memory defaults; persistence adapters are separate crates;
  assembly layer wires. Resolves OQ-34. Concrete adapter shapes deferred
  for exploration (OQ-36).

Amends auth.md, config.md, operation-registry.md, client-and-adapters.md,
open-questions.md, README.md, crates/core/README.md. Marks ADR-029
Accepted (Assumption 1 carries the ADR-030 superseded note). Marks the
research findings doc reviewed.
2026-06-27 12:12:25 +00:00

20 KiB
Raw Blame History

status, last_updated
status last_updated
draft 2026-06-27

Authentication

AuthContext, Identity, IdentityProvider, AuthToken, and the resolution flow.

See ADR-004 and ADR-011 for rationale.

AuthContext

Created by the endpoint for each incoming connection. Passed to ProtocolHandler::handle() as an immutable reference.

#[derive(Clone)]
pub struct AuthContext {
    /// The peer's authenticated identity, if resolved by the endpoint.
    /// None means the endpoint has no identity information for this connection.
    pub identity: Option<Identity>,

    /// The negotiated ALPN for this connection. Always present.
    pub alpn: Vec<u8>,

    /// The peer's remote address, if available. Informational (NAT/proxy).
    pub remote_addr: Option<SocketAddr>,

    /// SHA-256 fingerprint of the TLS client certificate, if presented.
    /// Set by the endpoint during TLS handshake. Handlers may use this for
    /// fingerprint-based auth even when IdentityProvider returns None.
    pub tls_client_fingerprint: Option<String>,
}

Construction by the endpoint

The endpoint constructs AuthContext from the QUIC connection:

  1. alpn: From connection.alpn() — always present after TLS handshake.
  2. remote_addr: From connection.remote_addr() — may be None for iroh connections.
  3. tls_client_fingerprint: Extracted from the TLS session's client certificate, if one was presented.
  4. identity: If a TLS client fingerprint is available, the endpoint calls IdentityProvider::resolve_from_fingerprint(). If it resolves, identity = Some(resolved). If not, identity = None.

Handler-level resolution

Handlers that require authentication extract protocol-specific credentials and call IdentityProvider inside handle(). When identity is resolved, the handler stores it on the Connection for observability:

// Example: CallAdapter extracting an AuthToken from the first frame
async fn handle(&self, connection: Connection, auth: &AuthContext) -> Result<(), HandlerError> {
    let identity = match &auth.identity {
        Some(id) => id.clone(),  // Endpoint already resolved identity
        None => {
            let stream = connection.accept_bi().await?;
            let token = extract_auth_token(stream).await?;
            self.identity_provider
                .resolve_from_token(&token)
                .ok_or(HandlerError::AuthRequired)?
        }
    };
    connection.set_identity(identity);  // Store for observability (OQ-11)
    // ... proceed with authenticated identity
}

Handlers that don't require authentication (e.g., DNS resolver, health check) can ignore auth.identity entirely and don't call set_identity.

Two Identity Scopes

There are two distinct identity scopes that must not be conflated:

Scope Where it's set Where it's stored What it represents Used for
Connection-level Handler in handle() Connection (via set_identity) Who opened this QUIC connection Observability, logging, audit
Per-request CallAdapter per call.requested OperationContext.identity Who is making this specific call ACL (ADR-015)

The connection-level identity is stable — set once when the handler resolves it. The per-request identity is dynamic — resolved per call.requested, potentially different across requests on the same connection (if different auth tokens are used). The per-request identity takes precedence for ACL on OperationContext; the connection-level identity is for observability only, not for ACL.

Connection exposes set_identity via interior mutability — the handler sets it once when resolved, the endpoint and observability layers read it. The identity is write-once-read-many.

AuthContext is Clone and immutable

  • derive(Clone) allows handlers to clone AuthContext for per-stream or per-channel contexts.
  • handle() receives &AuthContext — immutable. Handlers that resolve identity create local variables, they don't mutate the shared context. This prevents cross-contamination between streams on the same connection.

Identity

The authenticated peer identity. Carries authorization information.

#[derive(Debug, Clone, PartialEq)]
pub struct Identity {
    /// Stable logical identifier. On the fingerprint path, this is the
    /// `PeerEntry.peer_id` (stable across key rotation, ADR-030). On the
    /// API-key path, this is the key prefix (changes with the key — see
    /// "API keys vs peer entries" below). On the composition path, this
    /// is the `CompositionAuthority` label (ADR-022).
    pub id: String,

    /// Authorization scopes. e.g., ["relay:connect", "secrets:derive"]
    pub scopes: Vec<String>,

    /// Named resource lists. e.g., {"service": ["gitea", "registry"]}
    /// Populated from `PeerEntry.resources` on the fingerprint path
    /// (ADR-030), from `CompositionAuthority.resources` on the
    /// composition path (ADR-022), and empty on the API-key path.
    pub resources: HashMap<String, Vec<String>>,
}

This is the same structure as the reference implementation (alknet-main/crates/alknet-core/src/auth/identity.rs), minus the russh dependency. The id field is ALPN-agnostic:

  • SSH key / TLS cert auth (fingerprint path): the PeerEntry.peer_id (ADR-030) — a stable logical name like "worker-a", not the fingerprint. The fingerprint is the credential; the peer_id is the identity. Decoupling them means key rotation changes the credential but not the identity, so ACL entries and routing references stay stable.
  • API key auth: "alk_test" (key prefix) — the prefix IS the identity; rotation = new identity (see "API keys vs peer entries" below).
  • Composition path: the CompositionAuthority label (ADR-022) — e.g., "agent-chat".

API keys vs peer entries

The fingerprint and API-key auth paths have different identity semantics, by design (ADR-030):

Axis Fingerprint (PeerEntry) API key (ApiKeyEntry)
Identity source TLS handshake / SSH key Bearer token in protocol frame
Key rotation Same logical node, new material New identity (revocation = new key)
Identity.id peer_id (stable across rotation) prefix (changes with the key)
Identity.resources Populated from PeerEntry.resources Empty (resources are composition-only)

An API key's prefix IS the identity — rotating the key means a new prefix and a new identity, by design (revocation is the rotation mechanism for API keys). Decoupling the API key identity from the prefix would solve a problem API keys don't have: they're bearer tokens, not node identities. The fingerprint path gets the PeerEntry treatment because node identity must survive key rotation; the API-key path doesn't because bearer-token identity IS the token. The asymmetry is deliberate, not an oversight — see ADR-030 §"API keys".

AuthToken

Opaque authentication token carried in protocol frames.

#[derive(Debug, Clone)]
pub struct AuthToken {
    pub raw: Vec<u8>,
}

Unchanged from the reference implementation. The handler that extracted it knows its encoding (UTF-8 string, binary token, etc.).

IdentityProvider

Trait for resolving credentials to identities. Implemented by ConfigIdentityProvider.

pub trait IdentityProvider: Send + Sync + 'static {
    fn resolve_from_fingerprint(&self, fingerprint: &str) -> Option<Identity>;
    fn resolve_from_token(&self, token: &AuthToken) -> Option<Identity>;
}
  • resolve_from_fingerprint(): Used by the endpoint (TLS client cert) and by SSH (key fingerprint).
  • resolve_from_token(): Used by call protocol (AuthToken in first frame) and HTTP (Bearer header).

Both methods return Option<Identity>None means the credential is not recognized.

ConfigIdentityProvider

The default implementation. Resolves identities from DynamicConfig:

pub struct ConfigIdentityProvider {
    dynamic: Arc<ArcSwap<DynamicConfig>>,
}

The "Config" prefix indicates that identities are resolved from configuration (as opposed to a database or external service). This reads from ArcSwap<DynamicConfig>, which is hot-reloadable — not from StaticConfig. An alternative name would be DynamicConfigIdentityProvider to make this clearer, but ConfigIdentityProvider is consistent with the reference implementation and the naming is unlikely to cause confusion in practice.

How it resolves:

  • Fingerprint: Look up in DynamicConfig::auth.peers for the matching PeerEntry (by fingerprint). If found and enabled, return Identity { id: peer.peer_id, scopes: peer.scopes, resources: peer.resources }. The Identity.id is the stable peer_id, not the fingerprint — key rotation changes the fingerprint but not the peer_id, so ACL entries and routing references stay stable (ADR-030).
  • Token: Parse as UTF-8. If it starts with alk_, look up in DynamicConfig::auth.api_keys by prefix match + SHA-256 hash. If found and not expired, return Identity { id: prefix, scopes: entry.scopes, resources: {} }. The Identity.id is the key prefix — API key rotation = new identity (see "API keys vs peer entries" above).

See ADR-030 for the PeerEntry model and the id-fingerprint decoupling rationale.

Resource-scoped ACLs

Identity.resources is populated on three paths:

Path Source of resources Use case
Fingerprint resolution (ConfigIdentityProvider) PeerEntry.resources (ADR-030) External fingerprint-authenticated callers with per-peer resource binding
API key resolution (ConfigIdentityProvider) Empty (by design) API keys grant scopes only; resource-scoped access is composition-only
Composition (CompositionAuthority::as_identity, ADR-015/022) CompositionAuthority.resources Internal composition calls with declared resource binding

An OperationSpec that declares resource_type/resource_action will return FORBIDDEN when the caller authenticated via API key (because Identity.resources is empty), but succeeds when the caller authenticated via fingerprint with matching PeerEntry.resources, or via composition with matching CompositionAuthority.resources. The API-key limitation is deliberate (see "API keys vs peer entries" above); the fingerprint path's resource binding is the ADR-030 change that lifts the pre-ADR-030 limitation.

Changes to DynamicConfig via ConfigReloadHandle are reflected immediately — ConfigIdentityProvider reads from ArcSwap on every call.

Fingerprint string format

tls_client_fingerprint and PeerEntry.fingerprint use a prefixed-hex format. The prefix identifies the key type; the body is the hex-encoded hash or raw key bytes. AuthPolicy::resolve_identity_from_fingerprint scans peers for a matching fingerprint field — no normalization — so the extractor and the operator config must use the same format.

Transport Source Format
quinn (X.509) leaf client cert DER SHA256:<hex of SHA-256(cert_der)>
iroh (raw Ed25519) peer NodeId ed25519:<lowercase hex of 32-byte pub key>

When no client cert is presented (the current default — server uses with_no_client_auth()), the fingerprint is None and identity remains unresolved at the endpoint layer. A follow-up task will switch the server config to request-but-not-require client certs so fingerprints flow for peers that present them.

Server-side client cert request

The quinn rustls::ServerConfig uses a custom AcceptAnyCertVerifier that requests client certs but does not require them and does not verify them against a CA. This is the "request-but-don't-require" mode: peers that present a cert (X.509 or RFC 7250 raw key) have their fingerprint extracted via peer_identity(); peers that don't present a cert connect normally with tls_client_fingerprint: None.

The verifier accepts any presented cert without CA verification because alknet's identity model is fingerprint-based, not PKI-based — the AuthPolicy::peers set is the trust anchor, not a root CA store. The cert bytes are extracted at the TLS layer and hashed to a fingerprint string; the fingerprint is then matched against the configured PeerEntry.fingerprint fields by IdentityProvider::resolve_from_fingerprint().

Resolution Flow

Endpoint-level (before handle())

QUIC connection arrives
  → TLS handshake (ALPN negotiation)
  → Extract TLS client certificate fingerprint (if presented)
  → If fingerprint present: IdentityProvider::resolve_from_fingerprint()
    → Some(identity): auth.identity = Some(identity)
    → None: auth.identity = None
  → Construct AuthContext { identity, alpn, remote_addr, tls_client_fingerprint }
  → Look up handler by alpn
  → tokio::spawn(handler.handle(connection, &auth))

Handler-level (inside handle())

Handler receives &AuthContext
  → If auth.identity is Some: use it (endpoint already resolved)
  → If auth.identity is None and handler requires auth:
    → Extract protocol-specific credential (AuthToken, SSH key, etc.)
    → Call IdentityProvider::resolve_from_token() or resolve_from_fingerprint()
    → If resolved: use the Identity
    → If not resolved: return HandlerError::AuthRequired
  → If handler doesn't require auth: proceed without identity

IdentityProvider Injection

Handlers need access to IdentityProvider to resolve credentials inside handle(). Since ProtocolHandler::handle() doesn't receive an IdentityProvider parameter, each handler must obtain it through constructor injection:

// Example: SshAdapter holds an Arc<dyn IdentityProvider>
pub struct SshAdapter {
    identity_provider: Arc<dyn IdentityProvider>,
    // ... other handler-specific state
}

#[async_trait]
impl ProtocolHandler for SshAdapter {
    fn alpn(&self) -> &'static [u8] { b"alknet/ssh" }

    async fn handle(&self, connection: Connection, auth: &AuthContext) -> Result<(), HandlerError> {
        let identity = match &auth.identity {
            Some(id) => id.clone(),
            None => {
                // Extract SSH key fingerprint, resolve via identity_provider
                let fingerprint = extract_ssh_fingerprint(&connection).await?;
                self.identity_provider
                    .resolve_from_fingerprint(&fingerprint)
                    .ok_or(HandlerError::AuthRequired)?
            }
        };
        // ...
    }
}

The CLI binary constructs each handler with Arc::clone(&identity_provider) and passes it when building the HandlerRegistry. This is the assembly pattern: the CLI (the only crate that depends on all handlers) wires dependencies together.

The endpoint's AlknetEndpoint also holds Arc<dyn IdentityProvider> for endpoint-level auth resolution (TLS client certificate fingerprints), but handlers don't receive it from the endpoint — they receive it at construction time from the CLI.

Handler Credential source Resolution method
SshAdapter SSH public key handshake resolve_from_fingerprint()
CallAdapter AuthToken in first frame resolve_from_token()
HttpAdapter Authorization: Bearer header resolve_from_token()
DnsAdapter AuthToken in query labels resolve_from_token()
GitAdapter Signed push certificate resolve_from_fingerprint()
SftpAdapter SSH key (shares with SshAdapter) resolve_from_fingerprint()

Key Differences from Reference Implementation

Aspect Reference New Model
Auth resolution Inside SSH handler, before handle() Hybrid: endpoint resolves TLS-level, handler resolves protocol-level
AuthContext type None (just Arc<ArcSwap<DynamicConfig>> + IdentityProvider) Explicit struct with optional fields
Identity.id Always a fingerprint or API key prefix Same, but ALPN-agnostic documentation
ConfigIdentityProvider Depends on russh for PublicKey types No russh dependency; fingerprints stored as strings
Credential phases AD phases in CredentialProvider Two paths: fingerprint and token. No phases.

Design Decisions

Decision ADR Summary
Hybrid auth model ADR-004 Endpoint resolves TLS-level, handler resolves protocol-level
AuthContext with optional Identity ADR-011 Explicit None, not "partially authenticated"
AuthContext is immutable in handle() ADR-011 Handlers create local variables for resolved identity
Two resolution paths ADR-004 Fingerprint and token, not phased auth
Handler stores resolved identity on Connection OQ-11 (resolved) connection.set_identity() — write-once-read-many for observability
PeerEntry and Identity.id decoupling ADR-030 authorized_fingerprintspeers: Vec<PeerEntry>; Identity.id = peer_id (stable), not fingerprint; key rotation changes fingerprint, not identity
CredentialStore repo trait ADR-031 Second repo trait in core (alongside IdentityProvider); InMemoryCredentialStore default adapter
Storage boundary and repo/adapter pattern ADR-033 Core defines traits + in-memory defaults; persistence adapters are separate crates

Open Questions

  • OQ-35: API key identity vs peer identity — the asymmetry between the fingerprint path (gets PeerEntry id-decoupling) and the API-key path (doesn't) is deliberate. See ADR-030 §"API keys" and "API keys vs peer entries" above.

Security Constraints

These are security-critical implementation requirements, not architectural decisions (the architecture is locked by the ADRs above). They are documented here so implementation agents don't miss them.

  • Token entropy: generated alk_ tokens must have ≥128 bits of entropy. The prefix (first 8 chars) is for O(1) lookup and is not secret — it appears in logs by design. SHA-256 of the full token allows offline verification; this is safe only if the full token is high-entropy. The prefix alone must not be sufficient to authenticate.
  • Config reload must be authenticated: a reload that adds an authorized fingerprint or API key grants access immediately (see config.md). The reload trigger must be local-only (SIGHUP, file watch) or an admin-scoped call protocol operation. A malicious reload is equivalent to root-level privilege grant.
  • Connection-level identity is for observability only: Connection::set_identity stores the handler-resolved identity for logging/audit. Per-request identity (OperationContext.identity) takes precedence for ACL. See OQ-11.
  • Cryptographic nonces use OsRng: AES-GCM IVs and any other cryptographic nonces must use OsRng (or equivalent CSPRNG), not rand::random(). IV reuse under the same key is catastrophic for GCM (authenticity breaks, two-time-pad on plaintext). The vault implementation (crates/alknet-vault/src/encryption.rs) must use OsRng for IV generation.
  • Derived keys are zeroized on drop: cached derived keys (CachedKey) must derive Zeroize and ZeroizeOnDrop. When the cache evicts an entry (LRU) or the process exits without explicit lock(), derived private keys must not linger in freed heap memory. The cache must clear on drop, not just on explicit lock().
  • No unwrap() or expect() outside tests: poisoned lock recovery uses unwrap_or_else(|e| e.into_inner()) or explicit error propagation. A panic in one vault operation must not brick the vault for all other operations.