steg86

steg86 is a format-agnostic steganographic tool for x86 and AMD64 binaries. You can use it to hide information in compiled programs, regardless of executable format (PE, ELF, Mach-O, raw, &c). It has no performance or size impact on the files that it modifies (adding a message does not increase binary size or decrease execution speed).

For more details on how steg86 works, see the Theory of Operation section.

Installation

steg86 can be installed via cargo:

$ cargo install steg86

Alternatively, you can build it in this repository with cargo build:

$ cargo build

Usage

See steg86 --help for a full list of flags and subcommands.

Profiling

To profile a binary for steganographic suitability:

$ steg86 profile /bin/bash
Summary for /bin/bash:
  175828 total instructions
  27957 potential semantic pairs
  27925 bits of information capacity (approx. 3KB)

Embedding

To embed a message into a binary:

$ steg86 embed /bin/bash ./bash.steg <<< "here is my secret message"

By default, steg86 embed writes its output to $input.steg. For example, /lib64/ld-linux-x86-64.so.2 would become /lib64/ld-linux-x86-64.so.2.steg.

steg86 embed will exit with a non-zero status if the message cannot be embedded (e.g., if it's too large).

Extraction

To extract a message from a binary:

$ steg86 extract bash.steg > my_message
$ cat message
here is my secret message

steg86 extract will exit with a non-zero status if a message cannot be extracted (e.g., if it can't find one).

Theory of Operation

steg86 takes advantage of one of x86's encoding peculiarities: the R/M field of the ModR/M byte:

  7  6  5  4  3  2  1  0
 -------------------------
 | MOD |  REG  |   R/M   |
 -------------------------

The ModR/M byte is normally used to support both register-to-memory and memory-to-register variants of the same instruction. For example, the MOV instruction has the following variants (among many others):

Because the ModR/M field can encode either a memory addressing operation or a bare register, opcodes that support both register-to-memory and memory-to-register operations also support multiple encodings of register-to-register operations.

For example, mov eax, ebx can be encoded as either 89 d8 or 8b c3 without any semantic changes. This gives us one bit of information per duplicated instruction semantic. Given enough register-to-register instructions with multiple encodings, we can hide entire messages with those bits.

Additionally, because these semantically identical encodings are frequently the same size, we can modify preexisting binaries without having to fix relocations or RIP-relative addressing.

steg86 does primitive binary translation to accomplish these goals. It uses iced-x86 for encoding and decoding, and goblin for binary format wrangling.

Prior work

The inspiration for steg86 came from @inventednight, who described it as an adaptation of a similar idea (also theirs) for RISC-V binaries.

The technique mentioned above is discussed in detail in Hydan: Hiding Information in Program Binaries (2004).

steg86 constitutes a separate discovery of Hydan's technique and was written entirely independently; the refinements discussed in the paper may or may not be more optimal than the ones implemented in steg86.

Future improvements

• steg86 currently limits the embedded message to 16KB. This is a purely artificial limitation that could be resolved with some small format changes.

• x86 (and AMD64) both have multi-byte NOPs, for alignment purposes. Additional information can be hidden in these in a few ways:

  • The OF 1F /0 multi-byte NOP can be up to 9 bytes, of which up to 5 are free (SIB + 4-byte displacement).
  • There are longer NOPs (11, 15 bytes) that may also be usable.

• Going beyond register-to-register duals and rewriting add/sub, as Hydan does.