Skip to content

Blackbox functions return tainted value (false positives) #45

Closed

Description

@KevinHock

So, there are 2 commented out tests that I added during my interprocedural PR here and here.

I was going to fix these via doing something in an else of this if in visit_call, marking node.something as blackbox but I'm wondering what the best place to check something at, maybe in reaching_definitions_base, or adding them to the sanitizers dictionary so they can be marked as sanitized here.

Thoughts?

Activity

  1. self-assigned this
    on May 18, 2017
  2. Thalmann commented on May 21, 2017

    @Thalmann
    Contributor

    Hmm, i do not know if this even makes much sense.

    The first example about absolute import will not even run(ImportError).
    The second one will fail when we make the post.

    So i do not think we need to support analysis of code the python interpreter does not even understand.

  3. StefanMich commented on May 21, 2017

    @StefanMich
    Contributor

    In my opinion the best solution would be to mark it as a vulnerability and in some way indicate that it is due to the name not being imported. That way the developer is able to disregard the vulnerability if he knows that he uses some dynamic magic to set the name

  4. KevinHock commented on May 21, 2017

    @KevinHock
    CollaboratorAuthor

    I probably could have explained this better, say

    import subprocess
    from flask import Flask, render_template, request
    
    # This is a lib we can't possibly see inside of
    import scrypt
    
    
    app = Flask(__name__)
    
    @app.route('/menu', methods=['GET'])
    def menu():
        param = request.args.get('suggestion')
    
        # This is a function we can't possibly see inside of
        foobar = scrypt.encrypt('echo ' + param + ' >> ' + 'menu.txt', 'password')
        command = scrypt.encrypt('echo ' + param + ' >> ' + 'menu.txt', 'password')
        hey = command
        subprocess.call(hey, shell=True)
    
        with open('menu.txt','r') as f:
            menu = f.read()
    
        return render_template('command_injection.html', menu=menu)
    
    if __name__ == '__main__':
        app.run(debug=True)

    scrypt.encrypt could be any call from a library that we can't possibly know e.g. If all tainted args means a tainted return value.

    So we shouldn't report it as a vulnerability with as much confidence as everything else. We can talk more about the UI part when I make the PR on Monday or Tuesday.

    The way I did this was, to put the call in a set called blackbox_calls and then check in the assignment_with_call type function if the call is in blackbox calls. If it is, I add the assignment to blackbox assignments, which can be checked when sanitation is checked in vulnerabilities.py. This will get more complicated when I handle nested function calls but I'll write tests and incrementally handle things.

  5. KevinHock commented on May 21, 2017

    @KevinHock
    CollaboratorAuthor

    Maybe one day we'll do it like this awesome paper but, we're not there yet.

  6. KevinHock commented on May 21, 2017

    @KevinHock
    CollaboratorAuthor

    Maybe this update to the example file is better, thoughts? (Why would you exec an scrypt hash is a valid question but I didn't feel like running pyt until I find a better example.)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions