rat - ReDoS Abstract Tester 🐁

rat is a sound tool to detect exponential Regular Expression Denial of Service (ReDoS) attacks. Since the algorithm is proved to be sound, the tool cannot raise false negatives. This means that if rat determines that a regular expression is safe, it is impossible for it to be exploited by an attacker. Furthermore, rat takes advantage of efficient data structures to be particularly fast.

The paper is available here. The preprint PDF version is here.

Build, Install and Run

To compile the project you need opam installed, with a version of the OCaml compiler >= 4.08.0.

make deps      # Install the dependencies.
make           # Build the project.
make test      # Run the tests.
make install   # Install the executable.

To run rat without installing it:

dune exec rat -- <args>

Run with --help to print the help message. If there are no arguments, the command runs the interactive interpreter.

To run rat inside Docker:

# Build (you just need to do this once)
docker build . -t rat:latest

# Run rat in interactive mode
docker run -it rat:latest

# Example of running rat with arguments
docker run rat:latest rat --regex '(a|a)*b'

Example Analysis

Assume we have rat installed on the machine. The following command analyzes the regular expression (a|a)*b:

rat --regex '(a|a)*b'

The output should be similar to:

~ ⟦🐁⟧ ~
Exponential ReDoS: true
Exploit string: {
  prefix = ''
  pump = 'a'
  suffix = ''
}
Example: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
Runtime(ms): 0.399

The analyzer reports that the regular expression is vulnerable to a ReDoS attack, and prints an exploit string. The runtime includes only the time to run the analysis, and does not take into account the time to create a printable exploit string. Since in order to do this we must run an exponential algorithm (to build the minimal DFA recognizing the attack language), this can take a lot of time. The implementation of this algorithm can be improved considerably, and building the exploit string is not a necessary part of the analysis.

Analysis Options

It is possible to set the semantics of the matching with the option --semantics. If you are not sure, use the default option, which is match. In this case, the analyzer assumes that the input regular expressions match an input string even if just a prefix of the input string matches the regular expression. It is the default behaviour of matching engines, and corresponds to the behaviour of the function match in the re module in Python. If the fullmatch semantics is used, then the analyzer assumes that matching engines compute the language membership. The fullmatch semantics is the one used in the fullmatch function in Python. If you read the paper about rat (available soon!) and you are interested in trying the examples, you should use --semantics fullmatch, in all other cases the match semantics is more appropriate.

It is possible to print the full attack language for a regular expression with the option --show-lang. Since this language is the specification of the attack words, it becomes quickly difficult to read it.

Limitations

rat cannot analyze a wide variety of non-regular constructs in regular expressions, most notably backreferences and lookarounds. Some non-regular constructs, such as $, are supported in a limited form. Future improvements might fix this.

Benchmarking

If you want to benchmark rat against other detectors, please use an OCaml switch with flambda enabled. For instance, this can be done with the following:

# Create a with with OCaml 4.14 and flambda.
opam switch create 4.14.0+flambda --package=ocaml-variants.4.14.0+options,ocaml-option-flambda

# Install dependencies, build, and install.
make deps
make
make install

phreppo/rat