[Enhancement] Policy Engine Implementation

[marctjones]

Summary

Implement configurable policy engine for fine-grained access control.

Priority: Medium - Security hardening
Depends on: #8 (Design: Program Types) - must decide on types first
Milestone: v0.3.0 - Production-Ready

---

Background

Currently, polkit auth grants full access to all eBPF operations. A policy engine would allow:

• Restricting program types
• Limiting attach points
• Path-based restrictions
• Per-user policies

Deliverables

Policy File Format

# ~/.config/ebpf-assist/policy.toml

[defaults]
# What to do when no rule matches
action = "allow"  # or "deny"

[program_types]
# Which program types are allowed
allow = ["kprobe", "kretprobe", "uprobe", "uretprobe", "tracepoint", "perf_event"]
warn = ["xdp", "tc", "socket_filter"]
deny = ["lsm", "struct_ops", "cgroup_*"]

[paths]
# Where programs can be loaded from
allow = ["~/.local/share/ebpf-assist/**", "/usr/share/ebpf/**"]
deny = ["/tmp/**"]  # Dangerous to load from /tmp

[attach_points]
# What can be attached to
allow = ["*"]  # All by default
deny = ["security_*"]  # No LSM hooks

CLI Commands

ebpf-assist policy show          # Show current policy
ebpf-assist policy check <file>  # Check if program would be allowed
ebpf-assist policy edit          # Open policy in editor

Implementation

• Policy file parser (TOML)
• Policy evaluation engine
• Integration with load/attach handlers
• Clear error messages on policy violations
• MCP tool: ebpf_policy_check

Blocked By

• [Design] Default Allowed Program Types #8 must be resolved first to know what program types to allow/deny by default

Success Criteria

• Policy violations return clear, actionable errors
• Users can customize restrictions per their security requirements
• AI gets helpful feedback when policy blocks an operation