Dis86 is a decompiler for 16-bit real-mode x86 DOS binaries.

Dis86 has been built for doing reverse-engineering work such as analyzing and re-implementing old DOS video games from the early 1990s. The project is a work-in-progress and the development team makes no guarantees it will work or be useful out-of-the-box for any applications other than their own. Features and improvements are made on-demand as needed.

Goals:

• Support reverse-engineering 16-bit real-mode x86 DOS binaries
• Generate code that is semantically correct (in so far as practical)
• Generate code that integrates will with a hybrid-runtime system (Hydra) [currently unreleased]
• Avoid making many assumptions or using heuristics that can lead to broken decompiled code
• Be hackable and easy to extend as required
• Automate away common manual transformations and let a human reverser focus on the subjective tasks a computer cannot do well (e.g. naming things)

Non-goals:

• Output code beauty (semantic correctness is more important)
• Re-compilable to equivalent binaries

Also, we generally prefer manual configuration/annotation tables to flawed heuristics that will generate incorrect code.

Discussion of the internals will be published periodically on the author's blog: xorvoid

Assuming you have rust and cargo installed:

just build

Emit Disassembly:

./target/debug/dis86 --config <your_config.bsl> --binary <raw_text_segment> --name <function_name> --emit-dis <output-file>

Emit initial Intermediate Representation (IR):

./target/debug/dis86 --config <your_config.bsl> --binary <raw_text_segment> --name <function_name> --emit-ir-initial <output-file>

Emit final (optimized) Intermediate Representation (IR):

./target/debug/dis86 --config <your_config.bsl> --binary <raw_text_segment> --name <function_name> --emit-ir-final <output-file>

Visualize the control-flow graph with graphviz:

./target/debug/dis86 --config <your_config.bsl> --binary <raw_text_segment> --name <function_name> --emit-graph /tmp/ctrlflow.dot
dot -Tpng /tmp/ctrlflow.dot > /tmp/control_flow_graph.png
open /tmp/control_flow_graph.png

Emit inferred higher-level control-flow structure:

./target/debug/dis86 --config <your_config.bsl> --binary <raw_text_segment> --name <function_name> --emit-ctrlflow <output-file>

Emit an Abstract Syntax Tree (AST):

./target/debug/dis86 --config <your_config.bsl> --binary <raw_text_segment> --name <function_name> --emit-ast <output-file>

Emit C code:

./target/debug/dis86 --config <your_config.bsl> --binary <raw_text_segment> --name <function_name> --emit-code <output-file>

Primary development goal is to support an ongoing reverse-engineering and reimplementation project. The decompiler is also designed to emit code that integrates well with a hybrid runtime system (called Hydra) that is used to run a partially-decompiled / reimplemented project. As such, uses that fall out of this scope have been unconsidered and may have numerous unknown issues.

Some specific known limitations:

• The decompiler accepts only a flat binary region for the text segment. It doesn't handle common binary file-formats (e.g. MZ) at the moment.
• Handling of many 8086 opcodes are unimplemented in the assembly->ir build step. Implementations are added as needed.
• Handling of some IR ops are unimplemented in the ir->ast convert step. Implementations are added as needed.
• Control-flow synthesis is limited to while-loops, if-stmts, and switch-stmts. If-else is unimplemented.
• Block scheduling and placement is very unoptimal for more complicated control-flow.
• ... and many more ...

Feature wishlist

• Array accesses
• Compound types (struct and unions)
• Synthesizing struct/union member access
• If-else statements
• Pointer analysis and arithmetic
• More "u16 pair -> u32" fusing
• Improved type-aware IR
• Less verbose output C code patterns for common operations (e.g. passing pointer as a function call arg)

Dis86 began life as a simple disassembler and 1-to-1 instruction => C-statement decompiler that integrated well with the Hydra Runtime. Over time it gained complexity and it became difficult to implement more sophisticated transformations. So, it was rebuilt and rearchitected with a proper SSA IR.

The older versions remain in the repo under old/. In particular, old/v2 was much less sophisticated albeit more complete in terms of the input machine-code it could handle.

These versions remain as sometimes they are still useful when the latest version is missing some feature.