Skip to content

Research Journal (2026)

  • Focus: Integrating Telos (Agent Security) into the main architecture graph.
  • Challenge: The graph was too vertical. Switched to graph TD with subgraph clusters to show the 4-phase pipeline.
  • Outcome: Successfully mapped the flow from Browser Eye -> Cortex -> Kernel.
  • Focus: Solving the ASLR and ROP problem in Sentinel-CC.
  • Achievement: Implemented “Deep CFI” using eBPF stack walking (bpf_get_stack).
  • Technical Win: Successfully mapped dynamic libraries using an LPM_TRIE (Map-of-Maps approach). Now Sentinel protects glibc calls, not just the main binary.
  • Outcome: The system now withstands ROP attacks that try to reuse valid syscall sites.
  • Focus: Verified XDP throughput on the new 10GbE cards.
  • Result: Hit 65.28 Gbps with full DPI enabled.
  • Note: The “Slight Increase” in throughput with DPI is weird. Suspect XDP is dropping packets so fast it clears the RX ring buffer, artificially inflating “processed” numbers. Need to investigate ethtool -S stats next week.
  • Focus: Blocking mmap(PROT_EXEC).
  • Code: Added security_mmap_file hook in sentinel_lsm.c.
  • Blocker: The JIT compiler for Java crashes because it needs RWX memory.
  • Fix: Whitelisted the JVM PID for now. Need a better solution for Phase 7.