Finding an arbitrary file upload vulnerability in a filesharing script

[internetwache]

https://0day.work/finding-an-arbitrary-file-upload-vulnerability-in-a-filesharing-script/

[fake-name]

> I thought about a possible fix for a while, but in the end decided that the quickest and easiest fix would be to adjust the regular expressio

..... WHAT?

Python has a suite of facilities exactly for this very kind of problem.

Literally, the solution is "os.path.abspath(filename).startswith(os.path.abspath(dlfolder))"

This should, in all cases, return true if the filename is within the download folder directory, and false for any other case.

[jwilk → fake-name]

It doesn't do the right thing if dlfolder doesn't end with a slash.

[jwilk]

So I was curious what translate_path does:

    def translate_path(self, path):
        """Translate a /-separated PATH to the local filename syntax.

        Components that mean special things to the local file system
        (e.g. drive or directory names) are ignored.  (XXX They should
        probably be diagnosed.)

        """
        # abandon query parameters
        path = path.split('?',1)[0]
        path = path.split('#',1)[0]
        path = posixpath.normpath(urllib.unquote(path))
        words = path.split('/')
        words = filter(None, words)
        path = os.getcwd()
        for word in words:
            drive, word = os.path.splitdrive(word)
            head, word = os.path.split(word)
            if word in (os.curdir, os.pardir): continue
            path = os.path.join(path, word)
        return path

That... that doesn't make any sense.

[jwilk → jwilk]

Oh, and thanks to urllib.unquote, I can disguise slashes as "%2f", so the tightened regexp doesn't help.

[jwilk → jwilk]

Never mind me, I misread how this function is used.

Apparently this code was copied from stdlib, including a different directory traversal bug on Windows:

https://bugs.python.org/issue26657

[jwilk]

Apparently the upload directory is the same as the script directory, which is not good. If you upload a file named "cgi.py", it will be imported next time someone runs the script. No directory traversal needed to exploit this.

[gehaxelt → jwilk]

Hi,

I asssumed that the script would not be restarted, but you're totally right.

Thanks for the comment :)

[based2]

https://www.owasp.org/index.php/Unrestricted_File_Upload

[netsec_burn]

Nice find, I commented on your blog a method you can use to extend this vulnerability into RCE without admin interactivity.

[gruez → netsec_burn]

>Upload a .py file to the untrusted directory that has the same name of one of the imports used in the Python script. It will be executed the next time Python is called by the script, so you don't need to wait for interactivity on the admins part

I have a feeling that only works on python2 because of absolute imports on python3.

[gcr → gruez]

Upload a script that replaces the old file source and wait for gunicorn or whatever to reload the app server.

[gehaxelt → gcr]

The script seems to ship with a rundimentary webserver by default, so I don't see a reason to wrap gunicorn et al. around it?

However, you're right that overwriting the source is a possible attack vector.

[internetwache → netsec_burn]

Hi,

thanks for commenting!

I had cronjobs in mind, but as you said, they are not writeable.

I also thought about uploading a .py file with an import's name, but wouldn't that require a __init__.py (in a subdirectory?) to be regarded as a valid module/import by python?

[ecdh]

This is hilarious -- I found this exact script and noted the exact same vulnerability last year. Good on you for posting about it.

[gehaxelt → ecdh]

Nice catch :)

I thought the the amount of forks and stars warrants a short writeup.