86 lines
8.1 KiB
Markdown
86 lines
8.1 KiB
Markdown
# StellaOps Vexer Architecture
|
||
|
||
Vexer is StellaOps' vulnerability-exploitability (VEX) platform. It ingests VEX statements from multiple providers, normalizes them into canonical claims, projects trust-weighted consensus, and delivers deterministic export artifacts with signed attestations. This document summarizes the target architecture and how the current implementation maps to those goals.
|
||
|
||
## 1. Solution topology
|
||
|
||
| Module | Purpose | Key contracts |
|
||
| --- | --- | --- |
|
||
| `StellaOps.Vexer.Core` | Domain models (`VexClaim`, `VexConsensus`, `VexExportManifest`), deterministic JSON helpers, shared abstractions (connectors, exporters, attestations). | `IVexConnector`, `IVexExporter`, `IVexAttestationClient`, `VexCanonicalJsonSerializer` |
|
||
| `StellaOps.Vexer.Policy` | Loads operator policy (weights, overrides, justification gates) and exposes snapshots for consensus. | `IVexPolicyProvider`, `IVexPolicyEvaluator`, `VexPolicyOptions` |
|
||
| `StellaOps.Vexer.Storage.Mongo` | Persistence layer for providers, raw docs, claims, consensus, exports, cache. | `IVexRawStore`, `IVexExportStore`, Mongo class maps |
|
||
| `StellaOps.Vexer.Export` | Orchestrates export pipeline (query signature → cache lookup → snapshot build → attestation handoff). | `IExportEngine`, `IVexExportDataSource` |
|
||
| `StellaOps.Vexer.Attestation` *(planned)* | Builds in-toto/DSSE envelopes and communicates with Sigstore/Rekor. | `IVexAttestationClient` |
|
||
| `StellaOps.Vexer.WebService` *(planned)* | Minimal API host for ingest/export endpoints. | `AddVexerWebService()` |
|
||
| `StellaOps.Vexer.Worker` *(planned)* | Background executor for scheduled pulls, verification, reconciliation, cache GC. | Hosted services |
|
||
|
||
All modules target .NET 10 preview and follow the same deterministic logging and serialization conventions as Feedser.
|
||
|
||
## 2. Data model
|
||
|
||
MongoDB acts as the canonical store; collections (with logical responsibilities) are:
|
||
|
||
- `vex.providers` – provider metadata, trust tiers, discovery endpoints, and cosign/PGP details.
|
||
- `vex.raw` – immutable raw documents (CSAF, CycloneDX VEX, OpenVEX, OCI attestations) with digests, retrieval metadata, and signature state.
|
||
- `vex.claims` – normalized `VexClaim` rows; deduped on `(providerId, vulnId, productKey, docDigest)`.
|
||
- `vex.consensus` – consensus projections per `(vulnId, productKey)` capturing rollup status, source weights, conflicts, and policy revision.
|
||
- `vex.exports` – export manifests containing artifact digests, cache metadata, and attestation pointers.
|
||
- `vex.cache` – index from `querySignature`/`format` to export digest for fast reuse.
|
||
- `vex.migrations` – tracks applied storage migrations (index bootstrap, future schema updates).
|
||
|
||
GridFS is used for large raw payloads when necessary, and artifact stores (S3/MinIO/file) hold serialized exports referenced by `vex.exports`.
|
||
|
||
## 3. Ingestion and reconciliation flow
|
||
|
||
1. **Discovery & configuration** – connectors load YAML/JSON settings via `StellaOps.Vexer.Policy` (provider enablement, trust overrides).
|
||
2. **Fetch** – each `IVexConnector` pulls source windows, writing raw documents through `IVexRawDocumentSink` (Mongo-backed) with dedupe on digest.
|
||
3. **Verification** – signatures/attestations validated through `IVexSignatureVerifier`; metadata stored alongside raw records.
|
||
4. **Normalization** – format-specific `IVexNormalizer` instances translate raw payloads to canonical `VexClaim` batches.
|
||
5. **Consensus** – `VexConsensusResolver` (Core) consumes claims with policy weights supplied by `IVexPolicyEvaluator`, producing deterministic consensus entries and conflict annotations.
|
||
6. **Export** – query requests pass through `VexExportEngine`, generating `VexExportManifest` instances, caching by `VexQuerySignature`, and emitting artifacts for attestation/signature.
|
||
7. **Attestation & transparency** *(planned)* – `IVexAttestationClient` signs exports (in-toto/DSSE) and records bundles in Rekor v2.
|
||
|
||
The Worker coordinates the long-running steps (fetch/verify/normalize/export), while the WebService exposes synchronous APIs for on-demand operations and status lookups.
|
||
|
||
## 4. Policy semantics
|
||
|
||
- **Weights** – default tiers (`vendor=1.0`, `distro=0.9`, `platform=0.7`, `hub=0.5`, `attestation=0.6`) loaded via `VexPolicyOptions.Weights`, with per-provider overrides.
|
||
- **Justification gates** – policy enforces that `not_affected` claims must provide a recognized justification; rejected claims are preserved as conflicts with reason metadata.
|
||
- **Diagnostics** – policy snapshots carry structured issues for misconfigurations (out-of-range weights, empty overrides) surfaced to operators via logs and future CLI/Web endpoints.
|
||
|
||
Policy snapshots are immutable and versioned so consensus records capture the policy revision used during evaluation.
|
||
|
||
## 5. Determinism & caching
|
||
|
||
- JSON serialization uses `VexCanonicalJsonSerializer`, enforcing property ordering and camelCase naming for reproducible snapshots and test fixtures.
|
||
- `VexQuerySignature` produces canonical filter/order strings and SHA-256 digests, enabling cache keys shared across services.
|
||
- Export manifests reuse cached artifacts when the same signature/format is requested unless `ForceRefresh` is explicitly set.
|
||
- For scorring multiple sources on same VEX topic use - `VEXER_SCORRING.md`
|
||
|
||
## 6. Observability & offline posture
|
||
|
||
- Structured logs (`ILogger`) capture correlation IDs, query signatures, provider IDs, and policy revisions. Metrics/OTel instrumentation will mirror Feedser once tracing hooks are added.
|
||
- Offline-first: connectors, policy bundles, and export caches can be bundled inside the Offline Kit; no mandatory outbound calls beyond configured provider allowlists.
|
||
- Operator tooling (CLI/WebService) will expose diagnostics (policy issues, verification failures, cache status) so air-gapped deployments maintain visibility without external telemetry.
|
||
|
||
## 7. Roadmap highlights
|
||
|
||
- Complete storage mappings for providers/consensus/cache and add migrations/indices per collection.
|
||
- Implement Rekor/in-toto attestation clients and wire export engine to produce signed bundles.
|
||
- Build WebService endpoints (`/vexer/status`, `/vexer/claims`, `/vexer/exports`) plus CLI verbs mirroring Feedser patterns.
|
||
- Provide CSAF, CycloneDX VEX, and OpenVEX normalizers along with vendor-specific connectors (Red Hat, Cisco, SUSE, MSRC, Oracle, Ubuntu, OCI attestation).
|
||
- Extend policy diagnostics with schema validation, change tracking, and operator-facing diff reports.
|
||
- Mongo bootstrapper runs ordered migrations (`vex.migrations`) to ensure indexes for raw documents, providers, consensus snapshots, exports, and cache entries.
|
||
|
||
## Appendix A – Policy diagnostics workflow
|
||
|
||
- `StellaOps.Vexer.Policy` now exposes `IVexPolicyDiagnostics`, producing deterministic diagnostics reports with timestamp, severity counts, active provider overrides, and the full issue list surfaced by `IVexPolicyProvider`.
|
||
- CLI/WebService layers should call `IVexPolicyDiagnostics.GetDiagnostics()` to display operator-friendly summaries (`vexer policy diagnostics` and `/vexer/policy/diagnostics` are the planned entry points).
|
||
- Recommendations in the report guide operators to resolve blocking errors, review warnings, and audit override usage before consensus runs—embed them directly in UX copy instead of re-deriving logic.
|
||
- Export/consensus telemetry should log the diagnostic `Version` alongside `policyRevisionId` so dashboards can correlate policy changes with consensus decisions.
|
||
- Offline installations can persist the diagnostics report (JSON) in the Offline Kit to document policy headroom during audits; the output is deterministic and diff-friendly.
|
||
- Use `VexPolicyBinder` when ingesting operator-supplied YAML/JSON bundles; it normalizes weight/override values, reports deterministic issues, and returns the consensus-ready `VexConsensusPolicyOptions` used by `VexPolicyProvider`.
|
||
- Reload telemetry emits `vex.policy.reloads` (tags: `revision`, `version`, `issues`) whenever a new digest is observed—feed this into dashboards to correlate policy changes with consensus outcomes.
|
||
|
||
This architecture keeps Vexer aligned with StellaOps' deterministic, offline-operable design while layering VEX-specific consensus and attestation capabilities on top of the Feedser foundations.
|