[Python] Tracking only the most updated data flow

[elManto]

Hi Everyone,

Probably a silly question, but I couldn't find much around. I need to track data flow in Python scripts. The goal is to track the expressions that flow into an open() call. For now I'm using this sample query:

module FilesystemTaint implements DataFlow::ConfigSig {
    predicate isSource(DataFlow::Node source) {
        exists( Expr e | source.asExpr() = e)
    }
    predicate isSink(DataFlow::Node sink) {
        exists (DataFlow::CallCfgNode call, FileSystemAccess fs | 
            sink = call.getArg(0) and call.getArg(0) = fs.getAPathArgument())
    }
}

module MyFlow = TaintTracking::Global<FilesystemTaint>;

from DataFlow::Node source, DataFlow::Node sink
where MyFlow::flow(source, sink) 
select source, sink, source.getLocation(), sink.getLocation()

And this sample testcase:

def get_input_from_stdin():
    return sys.argv[1]

def get_hardcoded_filename():
    return "output.txt"

def combine_input_and_hardcoded():
    user_input = get_input_from_stdin()
    return f"{user_input}_log.txt"

def intermediate_variable():
    part1 = get_input_from_stdin()
    part2 = "_data"
    filename = part1 + part2 + ".txt"
    return filename

def filename_from_multiple_functions():
    prefix = get_input_from_stdin()
    suffix = get_hardcoded_filename()
    return prefix + "_" + suffix

def nested_function_call():
    def inner():
        return get_input_from_stdin()
    filename = inner() + "_nested.txt"
    return filename


def custom_open(filepath):
    with open(filepath, 'w') as f:
        f.write("This is a test file.\n")

def main():
    custom_open(get_hardcoded_filename())
    custom_open(combine_input_and_hardcoded())
    custom_open(intermediate_variable())
    custom_open(filename_from_multiple_functions())
    custom_open(nested_function_call())

if __name__ == "__main__":
    main()

As expected CodeQL returns me one path for each source Expr . However, I would like to keep only the most updated one. For instance, for the function intermediate_variable() , the result will report that all the expressions flow into the open() API, namely, part1, part2, get_input_from_stdin(), part1 + part2 and the literals . However, I would like to keep only the most updated source, which is filename .

Obviously I could solve this by checking the line number but that's no going to fly for slightly more complex scenarios. Another solution I was thinking is to have an other DataFlow tracking to check intra-procedural flows among local expressions but before doing that I'm curious if there's a more "CodeQL"-like way to do this. Thanks.

[redsun82]

👋 @elManto

As far as I can tell, what you are seeing is a result of how you defined isSource, where you marked every expression as a source. This means that CodeQL will show you all paths from any expression to a file system access, which in the intermediate_variable function does indeed mean all of the expressions in that function.

Once you restrict isSource to what you really want to mark as a source of tainted data, you will get more sensible results. In taint flow analysis, there is no such thing as a "most updated source": there are original sources, sinks, and nodes on the paths from the former to the latter. In this terms filename won't probably be a source, but rather a node on the path from the source to the sink.

For example, if you restrict isSource to sys.argv elements (provided you're using global taint flow which I think you already are), then you'll get one path through intermediate_variable, roughly consisting in

sys.argv[1] -> return -> get_input_from_stdin() ->
  part1 -> part1 + part2 + ".txt" -> filename -> return ->
    intermediate_variable() -> custom_open arg

[elManto]

Hi @redsun82 thanks for your input

I understand what you say, but the problem is that I don't only want to track the user input coming from sys.argv . In your example I would end up having sys.argv[1] as the source whereas I would like to obtain part1 + part2 + ".txt" cause that's the actual path that is being accessed .

Second, the input flowing into custom_open(arg) might come from a source that is a concatenation of strings that don't necessarily depend on sys.argv (e.g., they might be concatenated to hardcoded strings or other functions' return values) -- for my project I would like to have all those "concatenated" strings, that's why I tried to use 'Expression' as a source in the attempt to capture all the possible sources