Datasets:
Devilishcode
/
hacking

Dataset card Data Studio Files
xet
 Community
text
stringlengths
0
1.99k
- LiteTracer
| A selective symbolic tracer that integrates Triton with an extended
| TinyInst debugger. It dynamically instruments only instructions that
| access guarded untrusted input, using a combination of OS-level
| page protection and AArch64 hardware watchpoints to trap memory
| accesses and invoke symbolic emulation on demand.
- MemoryGuard
| A runtime protection module that configures page permissions
| and manages AArch64 hardware watchpoints to monitor access to
| untrusted input with efficient, fine-grained precision.
- PageMap
| A refinement-aware allocator for guarded memory tracking.
| Implements the Strict Recent Access Bias strategy through a
| recursive bitmask refinement algorithm and dynamic watchpoint
| reallocation. Translates runtime traps and access patterns into
| efficient hardware configurations using only 4 available CPU
| watchpoints per core.
- E0_Sample_Runner_Functions
| Constraint-driven sample generation logic. Captures SMT constraints
| during tracing and solves for new inputs that drive unexplored paths.
- Patched TinyInst Debugger (macOS/debugger.cpp)
| A modified version of TinyInst's macOS debugger module that adds
| support for ARM64 hardware breakpoints and watchpoints. Also extends
| exception handling to catch and recover from guarded memory access.
This alpha release targets ARM64e macOS systems only, and is provided as
part of this Phrack issue. A full open-source release of the complete E0
framework, including the fuzzer layer, Python bindings, event pipeline,
multi-platform support, and LLM integration, is planned at a later date.
All code is released under the terms of the Apache License, Version 2.0.
--[ 14 - Acknowledgments
E0 builds upon the efforts of several outstanding open-source projects
whose design and implementation significantly shaped this work.
Special thanks to:
- Ivan Fratric and the TinyInst / Jackalope team
E0 reuses core components from Jackalope's sample delivery, mutation
logic, multithreading, and target monitoring. It also extends
TinyInst's macOS debugger to support hardware breakpoints,
watchpoints, and fine-grained memory access instrumentation.
- Jonathan Salwan and the Triton team
Triton serves as the symbolic execution engine at the core of E0's
tracing logic. Its efficient architecture and symbolic precision make
it the ideal foundation for runtime analysis and constraint solving.
We are also grateful to Richard Johnson for his invaluable guidance,
detailed feedback, and critical insights during the preparation of this
paper. His thoughtful reviews helped sharpen the structure and clarity of
our arguments.
After exploring several architectural paths and full rewrites, we
ultimately chose to avoid redundancy and instead build on top of solid,
actively maintained projects. This included following Jackalope's
structural patterns, extending Triton's instruction semantics to support
ARM64e, and extending TinyInst debugger modules.
We encourage others to adopt this integration model when combining
fuzzing, tracing, and symbolic execution in real-world workflows.
--[ 15 - Conclusion
We introduced a novel methodology that integrates selective symbolic
instrumentation, data-flow-guided fuzzing, and structured LLM-driven
analysis, realized in the implementation of E0. By precisely tracking
external inputs and instrumenting only essential symbolic operations,
E0 prioritizes native execution and engages symbolic emulation only when
necessary, achieving high performance without sacrificing symbolic depth.
Structured symbolic definitions (ARG, MEM, RET) serve as precise anchors
for LLM-guided analysis, enhanced by reverse prompt engineering, ensemble
consensus prompting, and on-demand context expansion. This modular design
decomposes vulnerability analysis into verifiable stages, improving
detection accuracy and interpretability while reducing false positives
and test-time compute.
Each component of E0 contributes to an integrated vulnerability discovery
process. This fusion arms LLMs with concrete, high-fidelity context,
reduces wasted inference cycles, and closes the loop with end-to-end
fuzzing. It unites discovery breadth, symbolic precision, and semantic
reasoning into one scalable, low-latency engine.
Future work will refine AI-driven feedback loops, optimize symbolic
instrumentation triggers, and expand E0's evaluation across diverse
closed-source targets. Ultimately, E0 provides not just a technical
improvement but a practical, scalable framework for advancing binary
vulnerability research.
--[ 16 - References
[1] Jonathan Salwan (2015). Triton: Dynamic Binary Analysis Framework.