Files
mxaccessgw/docs/Authorization.md
T
Joseph Doherty acebe18773 fix(SEC-31,SEC-32): make probe admission atomic and stop Reset clearing a shared fallback partition
Two defects found in code review of the limiter rework.

Probe admission was check-then-act across two lock scopes: Check() read
"probe due" under lock(state), released it, then re-acquired to advance
NextProbeAtTicks. A burst of requests arriving together at an interval boundary
could therefore all observe the slot as due and all be admitted, handing the
verifier the very burst the interval exists to bound. The claim is now a single
critical section (TryConsumeProbe). The two layers are still claimed one at a
time — holding two per-state locks at once would need a global lock ordering to
stay deadlock-free — so a slot claimed on the composite partition is compensated
via ReleaseProbe when the aggregate then refuses, which otherwise silently spent
the partition's next slot and pushed the legitimate holder out by a full
interval.

Reset() removed whatever partition the caller resolved to, including the
address's shared fallback partition when the caller's key id had been collapsed
into it by the per-peer cap (or when the token was junk-shaped). That bucket also
carries failures contributed by other key ids from the same address, so one
successful authentication became a reset button for an in-progress spray. Reset
now clears only a partition the caller owns (effectiveKeyId == presented key id);
the shared bucket decays by window expiry instead, and the caller still recovers
through probe admission. The key's aggregate is cleared either way, as designed.

Also applied from the review: closure-free GetOrAdd overload on _partitions, and
a remarks paragraph acknowledging the best-effort O(n) eviction scan under
sustained overflow. Threading the resolved partition key from Check through to
RecordFailure/Reset was declined: Check resolves with mint:false and RecordFailure
with mint:true, and the two can legitimately differ when a concurrent caller fills
the per-peer cap in between — reusing Check's key would record into the wrong
partition and bypass the cap, which is not worth saving one string concat.

Tests (limiter suite 11 -> 14): ProbeAdmission_UnderConcurrentArrivals_
GrantsExactlyOneSlot (200 rounds x 8 barrier-released threads at the boundary),
ProbeAdmission_WhenAggregateRefuses_ReturnsTheClaimedPeerSlot, and
Reset_WithOverCapKeyId_DoesNotClearSharedFallbackPartition. The latter two were
confirmed as genuine reds against the unfixed code; the concurrency test is a
guard — it is deterministically green on the fixed structure but did not
reproduce the original nanosecond-wide window on its own.
2026-08-07 05:57:26 -04:00

294 lines
19 KiB
Markdown

# Gateway gRPC Authorization
The authorization subsystem has two layers. The gRPC interceptor enforces the
verb scope required by the RPC. Service-layer constraint checks then narrow
what an authenticated API key can browse, read, or write inside the Galaxy.
## Overview
Authorization runs as a single gRPC server interceptor registered for every call on the gateway. It pulls the authenticated identity for the current request, derives the scope that the request type requires, and either lets the call continue or fails the call with a gRPC status. The pipeline keeps service classes free of cross-cutting checks, which matches the `gateway.md` "thin gRPC layer" rule that service handlers translate between contracts and domain code without owning policy.
The participating types live under `src/ZB.MOM.WW.MxGateway.Server/Security/Authorization/`:
- `GatewayGrpcAuthorizationInterceptor` runs the authenticate-then-authorize pipeline for unary and server-streaming calls.
- `GatewayGrpcScopeResolver` maps a request message (and, for `MxCommandRequest`, the inner `MxCommandKind`) to the scope string that must be present on the caller.
- `GatewayScopes` exposes the canonical scope constants used by the resolver and any downstream consumer.
- `GatewayRequestIdentityAccessor` and `IGatewayRequestIdentityAccessor` expose the verified identity to handlers and any service code that runs inside the call.
- `IConstraintEnforcer` applies optional API-key constraints against the
cached Galaxy hierarchy from service bodies.
- `GrpcAuthorizationServiceCollectionExtensions` wires the components into the DI container and the gRPC pipeline.
The `ApiKeyIdentity` consumed here is produced by the authentication layer; see [Authentication](./Authentication.md) for how it is built and how scopes are persisted.
## Why an Interceptor
Centralizing the policy in `GatewayGrpcAuthorizationInterceptor` produces three concrete benefits:
1. Every RPC defined in `MxAccessGatewayService` is covered by construction. A new RPC inherits the check the moment its request type is added to `GatewayGrpcScopeResolver`, instead of relying on each service method to remember to call an authorization helper.
2. Verb-scope policy stays centralized. Request-specific constraints still run
in service bodies because they need command payloads, item handles, and
Galaxy metadata that the interceptor should not inspect.
3. Authentication and authorization happen in one place, so the gRPC `Status` mapping is consistent. A failed key check always returns `Unauthenticated`, and a missing scope always returns `PermissionDenied` with the offending scope name.
## Interceptor Flow
`GatewayGrpcAuthorizationInterceptor` overrides both `UnaryServerHandler` and `ServerStreamingServerHandler`. Both call the same private `AuthenticateAndAuthorizeAsync` helper before invoking the continuation, then push the resolved identity onto the accessor for the duration of the call.
```csharp
public override async Task<TResponse> UnaryServerHandler<TRequest, TResponse>(
TRequest request,
ServerCallContext context,
UnaryServerMethod<TRequest, TResponse> continuation)
{
ApiKeyIdentity? identity = await AuthenticateAndAuthorizeAsync(request, context).ConfigureAwait(false);
IDisposable? identityScope = identity is null ? null : identityAccessor.Push(identity);
using (identityScope)
{
return await continuation(request, context).ConfigureAwait(false);
}
}
```
The shared helper performs the actual decision:
```csharp
if (options.Value.Authentication.Mode == AuthenticationMode.Disabled)
{
return null;
}
string? authorizationHeader = context.RequestHeaders.GetValue("authorization");
ApiKeyVerificationResult verificationResult = await apiKeyVerifier
.VerifyAsync(authorizationHeader, context.CancellationToken)
.ConfigureAwait(false);
if (!verificationResult.Succeeded || verificationResult.Identity is null)
{
throw new RpcException(new Status(
StatusCode.Unauthenticated,
"Missing or invalid API key."));
}
string requiredScope = scopeResolver.ResolveRequiredScope(request);
if (!verificationResult.Identity.Scopes.Contains(requiredScope))
{
throw new RpcException(new Status(
StatusCode.PermissionDenied,
$"API key is missing required scope '{requiredScope}'."));
}
return verificationResult.Identity;
```
The flow is:
1. If `GatewayOptions.Authentication.Mode` is `AuthenticationMode.Disabled`, the helper returns `null` immediately. No identity is pushed onto the accessor and the continuation runs without scope enforcement. This matches the `AuthenticationMode` enum, which only defines `ApiKey` and `Disabled`.
2. Otherwise, the `authorization` request header is read directly off `ServerCallContext.RequestHeaders` and handed to `IApiKeyVerifier.VerifyAsync`. A failed verification or a missing identity throws `RpcException` with `StatusCode.Unauthenticated`.
3. `GatewayGrpcScopeResolver.ResolveRequiredScope(request)` produces the scope string. If the identity's `Scopes` set does not contain it, the helper throws `RpcException` with `StatusCode.PermissionDenied` and embeds the missing scope name in `Status.Detail` so callers can diagnose the failure.
4. On success, the verified `ApiKeyIdentity` is returned and pushed onto `IGatewayRequestIdentityAccessor` for the lifetime of the call.
The status codes are deliberately distinct: `Unauthenticated` signals "we do not know who you are," and `PermissionDenied` signals "we know who you are, but you cannot do this." Treating the two as the same code would make troubleshooting harder for client implementations.
### Rate limiting the auth surface (SEC-11, SEC-31, SEC-32)
Before the verification store read, the helper asks a cheap in-process failure counter (`ApiKeyFailureLimiter`) whether the attempt may proceed, so online guessing of API-key secrets cannot spend a SQLite read (and, in a naive design, a cache miss) per attempt. The counter has two layers over one sliding `ApiKeyFailureWindowSeconds` window:
- **Composite `(transport peer, key id)` partitions.** Reaching `MxGateway:Security:ApiKeyFailureLimit` failures binds the throttle to the address that produced them. The key id alone is never the partition: key ids are not secret — they ride in every token and are listed on the dashboard — so keying on them let any network peer deny a key to its legitimate holder. The key id joins the partition only after a token-shape check (literal `mxgw` prefix, at least three non-empty `_` segments, key id of at most 64 characters), and one address may mint at most 32 key-id partitions before the overflow collapses onto that address's fallback partition.
- **A per-key-id aggregate** across all peers (`ApiKeyFailureAggregateLimit`, default 30), which bounds a distributed or source-rotating sprayer that never trips any single partition.
An over-limit state is a valve, not a wall: one request per `ApiKeyFailureProbeIntervalSeconds` (default 5 s) is admitted through to the real verifier, and everything else is refused with `StatusCode.ResourceExhausted` before the store read. The slot is claimed atomically, so a burst arriving together at an interval boundary still yields exactly one admission. A successful verification resets both layers — which is why the reset path stays reachable while a key is under active spray. One exception: when the caller's key id was collapsed into its address's shared fallback partition by the per-peer cap, a success clears the key's aggregate but leaves that shared partition alone, since it also holds failures contributed by other key ids from the same address. The tracked partitions form a bounded LRU (`ApiKeyFailureTrackedPeers`) whose eviction prefers fully expired windows and never removes an over-limit partition below a 2x transient overshoot ceiling, so the cap bounds memory without becoming a reset button for an active block. `ResourceExhausted` reveals only that throttling is in effect, not whether any particular secret was valid, preserving the opaque-failure property. Refusals increment `mxgateway.auth.throttled`, tagged `stage=peer|aggregate` and nothing else — `/metrics` is unauthenticated, so neither key ids nor peer addresses may appear there.
The dashboard login surface is throttled independently: `POST /auth/login` carries a fixed-window ASP.NET Core rate-limiter policy keyed per remote IP (`MxGateway:Security:LoginRateLimit*`), rejecting a burst with HTTP 429 before the LDAP bind is relayed to the directory. See [GatewayConfiguration](./GatewayConfiguration.md#security-options).
## Scope Resolution
`GatewayGrpcScopeResolver` is a stateless singleton that switches on the runtime request type. Top-level RPC requests map directly:
```csharp
public string ResolveRequiredScope(object request)
{
return request switch
{
OpenSessionRequest => GatewayScopes.SessionOpen,
CloseSessionRequest => GatewayScopes.SessionClose,
StreamEventsRequest => GatewayScopes.EventsRead,
MxCommandRequest commandRequest => ResolveCommandScope(commandRequest.Command?.Kind ?? MxCommandKind.Unspecified),
AcknowledgeAlarmRequest => GatewayScopes.InvokeWrite,
StreamAlarmsRequest => GatewayScopes.EventsRead,
QueryActiveAlarmsRequest => GatewayScopes.EventsRead,
TestConnectionRequest or
GetLastDeployTimeRequest or
DiscoverHierarchyRequest or
WatchDeployEventsRequest => GatewayScopes.MetadataRead,
_ => GatewayScopes.Admin
};
}
```
The `_ => GatewayScopes.Admin` fallback is intentional: any future request type that the resolver does not recognize fails closed, requiring the strongest scope until the resolver is updated. `AcknowledgeAlarm` is treated as a write — it mutates alarm state, mirroring `MxCommandKind.Write*` — and `StreamAlarms` and `QueryActiveAlarms` share the alarm/event surface with `StreamEvents` and `MxCommandKind.DrainEvents`, so they carry `events:read` (the active-alarm snapshot is the same data reachable through the event surface). All three alarm RPCs are session-less: the scope check is the only authorization gate, since there is no per-session ownership to enforce.
`MxCommandRequest` is special because it multiplexes many MxAccess operations through a single RPC. The resolver inspects the embedded `MxCommandKind` so each operation gets its own scope:
```csharp
private static string ResolveCommandScope(MxCommandKind kind)
{
return kind switch
{
MxCommandKind.Write or
MxCommandKind.Write2 or
MxCommandKind.WriteBulk or
MxCommandKind.Write2Bulk => GatewayScopes.InvokeWrite,
MxCommandKind.WriteSecured or
MxCommandKind.WriteSecured2 or
MxCommandKind.WriteSecuredBulk or
MxCommandKind.WriteSecured2Bulk or
MxCommandKind.AuthenticateUser => GatewayScopes.InvokeSecure,
MxCommandKind.ArchestraUserToId or
MxCommandKind.GetSessionState or
MxCommandKind.GetWorkerInfo => GatewayScopes.MetadataRead,
MxCommandKind.DrainEvents => GatewayScopes.EventsRead,
MxCommandKind.ShutdownWorker => GatewayScopes.Admin,
_ => GatewayScopes.InvokeRead
};
}
```
Reads (`Register`, `AddItem`, `Advise`, `ReadBulk`, and any other unspecified kind) fall through to `InvokeRead`, which keeps the matrix small while still separating reads from writes, secured writes, metadata lookups, event drains, and worker shutdown. The four bulk-write families (`WriteBulk`, `Write2Bulk`, `WriteSecuredBulk`, `WriteSecured2Bulk`) are mapped explicitly so a missing arm cannot silently demote a bulk write to a read scope.
## Constraint Enforcement
`ApiKeyIdentity.Constraints` is optional. Empty constraints preserve the
previous behavior: the key is authorized only by its verb scopes. Non-empty
constraints are stored as JSON in `api_keys.constraints` and are applied by
`IConstraintEnforcer` after the interceptor succeeds.
Supported constraints are:
| Constraint | Meaning |
|------------|---------|
| `read_subtrees` | Contained-path globs allowed for read/subscription commands. |
| `write_subtrees` | Contained-path globs allowed for write commands. |
| `read_tag_globs` | Tag-address globs allowed for read/subscription commands. |
| `write_tag_globs` | Tag-address globs allowed for write commands. |
| `max_write_classification` | Maximum Galaxy attribute `security_classification` a key may write. |
| `browse_subtrees` | Contained-path globs used to filter Galaxy browse results and deploy-event counts. |
| `read_alarm_only` | Read/subscription commands must target objects with alarm-bearing attributes. |
| `read_historized_only` | Read/subscription commands must target objects with historized attributes. |
Glob matching is anchored, case-insensitive, and supports `*` and `?`.
Subtree and tag glob lists are alternatives: matching either list allows that
scope dimension. Empty lists mean unconstrained for that dimension.
Constraints are set when a key is created — through the `apikey create-key`
flags (see [Authentication](./Authentication.md)) or the dashboard API Keys
page create dialog (see
[Gateway Dashboard Design](./GatewayDashboardDesign.md#api-keys-page)). The
dashboard API Keys page also renders each key's effective constraints.
The service checks read constraints for `AddItem`, `AddItem2`, `AddItemBulk`,
`SubscribeBulk`, `AdviseItemBulk`, and `ReadBulk`. It checks write constraints
for `Write`, `Write2`, `WriteSecured`, `WriteSecured2`, `WriteBulk`,
`Write2Bulk`, `WriteSecuredBulk`, and `WriteSecured2Bulk`. Bulk commands run
through `BulkConstraintPlan` (`ReadBulkConstraintPlan`,
`WriteBulkConstraintPlan`, `SubscribeBulkConstraintPlan`), which preserves the
caller's input order: each entry is evaluated against the constraint surface,
and `BulkConstraintPlan.MergeDeniedInto` re-merges denied entries back into
their original index positions so the reply slot at `entries[i]` always
corresponds to the request slot at `entries[i]`. Successful item registrations
are tracked per session so later item-handle commands resolve back to the
original tag address. If a constrained key presents an unknown item handle,
the gateway fails closed.
Non-bulk constraint failures return gRPC `PermissionDenied`. Bulk read
commands preserve input order and return a failed `SubscribeResult` for each
denied item while still forwarding allowed items to the worker. Every denial
adds an `api_key_audit` entry with the key id, command kind, target, and
blocking constraint; secured values and raw credentials are never logged.
## Scope Catalog
`GatewayScopes` is the single source of truth for scope strings. Every entry is currently mapped by either the resolver or another security component:
| Constant | Value | Required For |
|----------|-------|--------------|
| `SessionOpen` | `session:open` | `OpenSessionRequest` |
| `SessionClose` | `session:close` | `CloseSessionRequest` |
| `EventsRead` | `events:read` | `StreamEventsRequest`, `StreamAlarmsRequest`, `QueryActiveAlarmsRequest`, `MxCommandKind.DrainEvents` |
| `InvokeRead` | `invoke:read` | `MxCommandRequest` for read-style command kinds (`Register`, `AddItem`, `Advise`, `ReadBulk`, and any kind not otherwise mapped) |
| `InvokeWrite` | `invoke:write` | `AcknowledgeAlarmRequest`, `MxCommandKind.Write`, `MxCommandKind.Write2`, `MxCommandKind.WriteBulk`, `MxCommandKind.Write2Bulk` |
| `InvokeSecure` | `invoke:secure` | `MxCommandKind.WriteSecured`, `MxCommandKind.WriteSecured2`, `MxCommandKind.WriteSecuredBulk`, `MxCommandKind.WriteSecured2Bulk`, `MxCommandKind.AuthenticateUser` |
| `MetadataRead` | `metadata:read` | `MxCommandKind.ArchestraUserToId`, `MxCommandKind.GetSessionState`, `MxCommandKind.GetWorkerInfo`, `GalaxyRepository.TestConnection`, `GalaxyRepository.GetLastDeployTime`, `GalaxyRepository.DiscoverHierarchy`, `GalaxyRepository.WatchDeployEvents` |
| `Admin` | `admin` | `MxCommandKind.ShutdownWorker`, the default for any unrecognized request type, and the dashboard authorization policy |
The `Admin` constant is also referenced by `DashboardAuthenticator` and `DashboardAuthorizationHandler` so that the dashboard and the gRPC layer agree on what "admin" means.
## Identity Access for Downstream Layers
Once authorization passes, `GatewayGrpcAuthorizationInterceptor` calls `identityAccessor.Push(identity)` and disposes the returned scope when the continuation completes. `GatewayRequestIdentityAccessor` stores the active identity in an `AsyncLocal<ApiKeyIdentity?>`, so the value flows across `await` boundaries and child tasks belonging to the same request.
```csharp
public sealed class GatewayRequestIdentityAccessor : IGatewayRequestIdentityAccessor
{
private readonly AsyncLocal<ApiKeyIdentity?> currentIdentity = new();
public ApiKeyIdentity? Current => currentIdentity.Value;
public IDisposable Push(ApiKeyIdentity identity)
{
ArgumentNullException.ThrowIfNull(identity);
ApiKeyIdentity? previousIdentity = currentIdentity.Value;
currentIdentity.Value = identity;
return new IdentityScope(this, previousIdentity);
}
}
```
The returned `IdentityScope` restores the previous value on dispose rather than clearing it. This makes the accessor safe for nested pushes, even though the current interceptor only pushes once per call. Disposing twice is a no-op because of the `disposed` guard inside `IdentityScope`.
Downstream code consumes the accessor through the `IGatewayRequestIdentityAccessor` interface:
```csharp
public interface IGatewayRequestIdentityAccessor
{
ApiKeyIdentity? Current { get; }
IDisposable Push(ApiKeyIdentity identity);
}
```
`MxAccessGatewayService` takes `IGatewayRequestIdentityAccessor` as a constructor dependency and reads `Current` whenever it needs to attach the calling identity to a domain operation, which keeps the service free of header parsing or scope checks.
When `AuthenticationMode.Disabled` is configured, no identity is pushed, so `Current` returns `null`. Downstream code must tolerate that, just as it tolerates the absence of a scope check.
## Registration
`GrpcAuthorizationServiceCollectionExtensions.AddGatewayGrpcAuthorization` is the single entry point that registers every component and inserts the interceptor into the gRPC pipeline:
```csharp
public static IServiceCollection AddGatewayGrpcAuthorization(this IServiceCollection services)
{
services.AddSingleton<GatewayGrpcScopeResolver>();
services.AddSingleton<IGatewayRequestIdentityAccessor, GatewayRequestIdentityAccessor>();
services.AddSingleton<GatewayGrpcAuthorizationInterceptor>();
services.AddGrpc(options => options.Interceptors.Add<GatewayGrpcAuthorizationInterceptor>());
return services;
}
```
Singleton lifetimes are appropriate because none of the three classes hold per-request state on instance fields; the request-scoped value lives inside the `AsyncLocal` on `GatewayRequestIdentityAccessor`. `GatewayApplication` calls `builder.Services.AddGatewayGrpcAuthorization()` during startup, and the call also performs `AddGrpc`, so the gateway never registers gRPC without the interceptor attached.
## Related Documentation
- [Authentication](./Authentication.md)
- [Gateway Dashboard Design](./GatewayDashboardDesign.md)
- [Grpc](./Grpc.md)
- [GatewayConfiguration](./GatewayConfiguration.md)
- [Galaxy Repository Browse](./GalaxyRepository.md)