TherapistEncrypter

A single-file, self-contained C++17 command-line encryptor. This repository implements a compact, audit-friendly double-pass cipher with a memory-hard key derivation and streaming file I/O. It is intended for local file encryption and a small interactive message mode.

See the implementation in main.cpp.

Current status

• Single-file C++17 implementation: main.cpp contains the full program (no external libraries required).
• Streaming chunked I/O: file-mode encryption/decryption streams data in 64 KiB chunks to avoid full-file buffering (encryptFileStreamToFile / decryptFileStreamToFile).
• Hardened KDF (V5): memory-hard key stretching using a 1 MiB scratch area and 131,072 iterations (deriveHardenedSchedule).
• Double-pass V5 encryption with independent salts (salt1/salt2) and legacy V4 decryption support.
• Incremental 256-bit MAC (4 × u64) with constant-time comparison; MAC is computed over the original plaintext and written into the header after streaming the ciphertext.
• Optimizations: 64-bit word operations, memcpy-based LE load/store, precomputed 64-bit S-box lane tables (SBoxPair::fwd64), 64-bit XOR hot-paths in CTR loops.

Build

Recommended (highest-performance, GCC/Clang):

g++ -std=c++17 -O3 -march=native -flto -DNDEBUG -o therapist main.cpp

Portable / quick build:

g++ -std=c++17 -O2 -pipe -o therapist main.cpp

Notes:

• On Windows with MinGW/MSYS these commands work similarly. For MSVC create a Visual Studio project or compile with cl and equivalent optimization flags.
• If you enable future threaded features (parallel CTR segmentation, thread pools), you may need to add -pthread on POSIX systems.
• For profile-guided optimizations (PGO), build with -fprofile-generate and then -fprofile-use as usual for your toolchain.

Usage

Non-interactive encrypt (simple):

./therapist input.bin "my passphrase"
# -> writes: input.bin.encrypt

Non-interactive decrypt:

./therapist decrypt input.bin.encrypt "my passphrase"
# -> writes: input.bin (or input.bin.decrypted if original name cannot be restored)

Interactive mode (menu, message mode):

./therapist

Interactive menu includes: encrypt, decrypt, and a message mode that stores short timestamped encrypted messages as files named file_<timestamp> next to the executable.

Behavior notes:

• Invoking with two arguments treats the first as an input path and the second as the passphrase (encrypt).
• Invoking with decrypt <path> <pass> performs decryption.

File format (V5)

• Header magic: 4 bytes {'T','P','C','5'}
• Version: 1 byte (5)
• Salt length: 1 byte (32)
• MAC length: 1 byte (32)
• Layout: [salt1 (32)] [salt2 (32)] [mac (32)] [ciphertext (N)]

V4 (legacy) blobs are still accepted for decryption.

Implementation highlights & where to look

• Source: main.cpp
• Key functions/types:
  • deriveHardenedSchedule -- memory-hard KDF (V5)
  • encryptPayloadV5 / decryptPayloadV5 -- in-memory double-pass flows
  • encryptFileStreamToFile / decryptFileStreamToFile -- streaming file-mode flows (64 KiB chunks)
  • applyEnhancedCipher / applyEnhancedCipherChunk -- CTR-style streaming cipher
  • computeHardenedMac, macInit, macFeedBuffer, macFinalize -- incremental MAC helpers
  • SBoxPair::fwd64 and applySBoxToWord -- precomputed 64-bit S-box lane tables

Security notes & threat model

• The implementation is designed for local file encryption with a focus on auditability and hardening via an expensive KDF. It is self-contained but implements custom primitives -- for high-assurance projects, prefer using vetted AEAD libraries (e.g., libsodium or OpenSSL) or subject this implementation to independent review.
• Default KDF parameters are intentionally expensive: 1 MiB memory and 131,072 iterations (compile-time constants in main.cpp). Adjust them in source if you need a different performance/security trade-off.
• The MAC covers the original plaintext; decryption verifies the MAC in constant time. On Windows the program performs optional anti-debug checks; these are defense-in-depth and not a substitute for secure runtime environments.

Roadmap / TODO (high-impact items)

• Vectorize / widen MAC computation (word-wise or SIMD). (not started)
• Align and prefetch KDF memory; consider mlock/platform APIs to reduce swapping. (not started)
• Expose KDF parameters via CLI/env for easier tuning. (not started)
• Add microbench harness and profiling hooks; enable PGO workflow. (not started)
• Add unit tests and CI including format tests and basic fuzzing. (not started)
• Consider optional SIMD S-box (AVX2/NEON) fallback for heavy CPU workloads. (not started)

If you'd like, I can implement the top-priority item (parallel CTR segmentation for multi-core throughput) next or maybe you could help me out with it.

Contributing

• Bug reports, patches, and PRs welcome. Changes that improve auditability, add tests, or integrate standard AEAD backends are especially appreciated.

License

MIT -- see LICENSE.