PolyFile

A utility to identify and map the semantic structure of files, including polyglots, chimeras, and schizophrenic files. It can be used in conjunction with its sister tool PolyTracker for Automated Lexical Annotation and Navigation of Parsers, a backronym devised solely for the purpose of collectively referring to the tools as The ALAN Parsers Project.

Quickstart

You can install the latest stable version of PolyFile from PyPI:

pip3 install polyfile

To install PolyFile from source, in the same directory as this README, run:

pip3 install -e .

This will automatically install the polyfile and polymerge executables in your path.

Usage

usage: polyfile [-h] [--filetype FILETYPE] [--list] [--html HTML]
                [--only-match-mime] [--only-match] [--require-match]
                [--max-matches MAX_MATCHES] [--debug] [--trace] [--debugger]
                [--no-debug-python] [--quiet] [--version] [-dumpversion]
                [FILE]

A utility to recursively map the structure of a file.

positional arguments:
  FILE                  the file to analyze; pass '-' or omit to read from
                        STDIN

optional arguments:
  -h, --help            show this help message and exit
  --filetype FILETYPE, -f FILETYPE
                        explicitly match against the given filetype or
                        filetype wildcard (default is to match against all
                        filetypes)
  --list, -l            list the supported filetypes (for the `--filetype`
                        argument) and exit
  --html HTML, -t HTML  path to write an interactive HTML file for exploring
                        the PDF
  --only-match-mime, -I
                        just print out the matching MIME types for the file,
                        one on each line
  --only-match, -m      do not attempt to parse known filetypes; only match
                        against file magic
  --require-match       if no matches are found, exit with code 127
  --max-matches MAX_MATCHES
                        stop scanning after having found this many matches
  --debug, -d           print debug information
  --trace, -dd          print extra verbose debug information
  --debugger, -db       drop into an interactive debugger for libmagic file
                        definition matching and PolyFile parsing
  --no-debug-python     by default, the `--debugger` option will break on
                        custom matchers and prompt to debug using PDB. This
                        option will suppress those prompts.
  --quiet, -q           suppress all log output (overrides --debug)
  --version, -v         print PolyFile's version information to STDERR
  -dumpversion          print PolyFile's raw version information to STDOUT and
                        exit

To generate a JSON mapping of a file, run:

polyfile INPUT_FILE > output.json

You can optionally have PolyFile output an interactive HTML page containing a labeled, interactive hexdump of the file:

polyfile INPUT_FILE --html output.html > output.json

Interactive Debugger

PolyFile has an interactive debugger both for its file matching and parsing. It can be used to debug a libmagic pattern definition, determine why a specific file fails to be classified as the expected MIME type, or step through a parser. You can run PolyFile with the debugger enabled using the -db option.

File Support

PolyFile has a cleanroom, pure Python implementation of the libmagic file classifier, and supports all 263 MIME types that it can identify.

It currently has support for parsing and semantically mapping the following formats:

• PDF, using an instrumented version of Didier Stevens' public domain, permissive, forensic parser
• ZIP, including recursive identification of all ZIP contents
• JPEG/JFIF, using its Kaitai Struct grammar
• iNES
• Any other format specified in a KSY grammar

For an example that exercises all of these file formats, run:

curl -v --silent https://www.sultanik.com/files/ESultanikResume.pdf | polyfile --html ESultanikResume.html - > ESultanikResume.json

Prior to PolyFile version 0.3.0, it used the TrID database for file identification rather than the libmagic file definitions. This proved to be very slow (since TrID has many duplicate entries) and prone to false positives (since TrID's file definitions are much simpler than libmagic's). The original TrID matching code is still shipped with PolyFile and can be invoked programmatically, but it is not used by default.

Output Format

PolyFile outputs its mapping in an extension of the SBuD JSON format described in the documentation.

PolyFile can also emit a standalone HTML document that contains an interactive hex viewer as well as syntax trees for the discovered file formats. Simply pass the --html argument to PolyFile with an output path:

$ polyfile input_file --html output.html

libMagic Implementation

PolyFile has a cleanroom implementation of libmagic (used in the file command). It can be invoked programmatically by running:

from polyfile.magic import MagicMatcher

with open("file_to_test", "rb") as f:
    # the default instance automatically loads all file definitions
    for match in MagicMatcher.DEFAULT_INSTANCE.match(f.read()):
        for mimetype in match.mimetypes:
            print(f"Matched MIME: {mimetype}")
        print(f"Match string: {match!s}")

To load a specific or custom file definition:

list_of_paths_to_definitions = ["def1", "def2"]
matcher = MagicMatcher.parse(*list_of_paths_to_definitions)
with open("file_to_test", "rb") as f:
    for match in matcher.match(f.read()):
        ...

Debugging the libmagic DSL

libmagic has an esoteric, poorly documented doman-specific language (DSL) for specifying its matching signatures. You can read the minimal and—as we have discovered in our cleanroom implementation—incomplete documentation by running man 5 magic. PolyFile implements an interactive debugger for stepping through the DSL specifications, modeled after GDB. You can enter this debugger by passing the --debugger or -db argument to PolyFile. It is useful for both implementing new libmagic DSLs, as well as figuring out why an existing DSL fails to match against a given file.

$ polyfile -db input_file
PolyFile 0.3.5
Copyright ©2021 Trail of Bits
Apache License Version 2.0 https://www.apache.org/licenses/

For help, type "help".
(polyfile) help
help ....... print this message
continue ... continue execution until the next breakpoint is hit
step ....... step through a single magic test
next ....... continue execution until the next test that matches
where ...... print the context of the current magic test (aliases: info stack and backtrace)
test ....... test the following libmagic DSL test at the current position
print ...... print the computed absolute offset of the following libmagic DSL offset
breakpoint . list the current breakpoints or add a new one
delete ..... delete a breakpoint
quit ....... exit the debugger

Merging Output From PolyTracker

PolyTracker is PolyFile’s sister utility for automatically instrumenting a parser to track the input byte offsets operated on by each function. The output of both tools can be merged to automatically label the semantic purpose of the functions in a parser. For example, given an instrumented black-box binary, we can quickly determine which functions in the program are responsible for parsing which parts of the input file format’s grammar. This is an area of active research intended to achieve fully automated grammar extraction from a parser.

A separate utility called polymerge is installed with PolyFile specifically designed to merge the output of both tools.

usage: polymerge [-h] [--cfg CFG] [--cfg-pdf CFG_PDF]
                 [--dataflow [DATAFLOW ...]] [--no-intermediate-functions]
                 [--demangle] [--type-hierarchy TYPE_HIERARCHY]
                 [--type-hierarchy-pdf TYPE_HIERARCHY_PDF] [--diff [DIFF ...]]
                 [--debug] [--quiet] [--version] [-dumpversion]
                 FILES [FILES ...]

A utility to merge the JSON output of `polyfile`
with a polytracker.json file from PolyTracker.

https://github.com/trailofbits/polyfile/
https://github.com/trailofbits/polytracker/

positional arguments:
  FILES                 Path to the PolyFile JSON output and/or the PolyTracker JSON output. Merging will only occur if both files are provided. The `--cfg` and `--type-hierarchy` options can be used if only a single file is provided, but no merging will occur.

optional arguments:
  -h, --help            show this help message and exit
  --cfg CFG, -c CFG     Optional path to output a Graphviz .dot file representing the control flow graph of the program trace
  --cfg-pdf CFG_PDF, -p CFG_PDF
                        Similar to --cfg, but renders the graph to a PDF instead of outputting the .dot source
  --dataflow [DATAFLOW ...]
                        For the CFG generation options, only render functions that participated in dataflow. `--dataflow 10` means that only functions in the dataflow related to byte 10 should be included. `--dataflow 10:30` means that only functions operating on bytes 10 through 29 should be included. The beginning or end of a range can be omitted and will default to the beginning and end of the file, respectively. Multiple `--dataflow` ranges can be specified. `--dataflow :` will render the CFG only with functions that operated on tainted bytes. Omitting `--dataflow` will produce a CFG containing all functions.
  --no-intermediate-functions
                        To be used in conjunction with `--dataflow`. If enabled, only functions in the dataflow graph if they operated on the tainted bytes. This can result in a disjoint dataflow graph.
  --demangle            Demangle C++ function names in the CFG (requires that PolyFile was installed with the `demangle` option, or that the `cxxfilt` Python module is installed.)
  --type-hierarchy TYPE_HIERARCHY, -t TYPE_HIERARCHY
                        Optional path to output a Graphviz .dot file representing the type hierarchy extracted from PolyFile
  --type-hierarchy-pdf TYPE_HIERARCHY_PDF, -y TYPE_HIERARCHY_PDF
                        Similar to --type-hierarchy, but renders the graph to a PDF instead of outputting the .dot source
  --diff [DIFF ...]     Diff an arbitrary number of input polytracker.json files, all treated as the same class, against one or more polytracker.json provided after `--diff` arguments
  --debug, -d           Print debug information
  --quiet, -q           Suppress all log output (overrides --debug)
  --version, -v         Print PolyMerge's version information and exit
  -dumpversion          Print PolyMerge's raw version information and exit

The output of polymerge is the same as PolyFile’s output format, augmented with the following:

1. For each semantic label in the hierarchy, a list of…
   1. …functions that operated on bytes tainted with that label; and
   2. …functions whose control flow was influenced by bytes tainted with that label.
2. For each type within the semantic hierarchy, a list of functions that are “most specialized” in processing that type. This process is described in the next section.

polymerge can also optionally emit a Graphviz .dot file or rendered PDF of the runtime control-flow graph recorded by PolyTracker.

Identifying Function Specializations

As mentioned above, polymerge attempts to match each semantic type of the input file to a set of functions that are “most specialized” in operating on that type. This is an active area of academic research and is likely to change in the future, but here is the current method employed by polymerge:

1. For each semantic type in the input file, collect the functions that operated on bytes from that type;
2. For each function, calculate the Shannon entropy of the different types on which that function operated;
3. Sort the functions by entropy, and select the functions in the smallest standard deviation; and
4. Keep the functions that are shallowest in the dominator tree of the runtime control-flow graph.

License and Acknowledgements

This research was developed by Trail of Bits with funding from the Defense Advanced Research Projects Agency (DARPA) under the SafeDocs program as a subcontractor to Galois. It is licensed under the Apache 2.0 license. © 2019, Trail of Bits.