gokkaya46 / pypi-scan

Scan pypi for typosquatting

pypi-scan

Scan PyPI for typosquatting

There are four anti-typosquatting functionalities in pypi-scan:

• Check if there are potential typosquatters on a package you care about.
• Check if there are potential typosquatters on the most downloaded PyPI packages.
• Check if packages newly added to PyPI are potential typosquatters.
• Create list of package names that a typosquatter might use.

PyPI (the Python Package Index) is a repository for Python packages. It's like a store where anybody with an internet connection can download (for free) Python packages. Typosquatting is a practice in which someone chooses a package name that is similar to an existing package and places a malicious package in this deceptively titled namespace. Imagine you want to download the package 'numpy' but you mistype that name and spell 'nunpy' instead. You then download a malicious package. Of course, not all packages with similar names are malicious, but some might be.

To determine if a package is a potential typosquatter, pypi-scan checks for "misspelling attacks" by employing Levenshtein distance, which is, roughly speaking, a measure of how many edits are required to transform one string of characters into another. The default edit distance is one. pypi-scan also checks for "confusion attacks." This includes "order attacks," e.g. switching the order of words around a dash or underscore, and "homophone attacks," packages whose names are spelled quite differently but sound the same phonetically. For an example of an order attack, imagine confusing python-nmap with nmap-python. Finally, when scanning the most downloaded packages, pypi-scan checks PyPI package metadata to check if there are identical fields between the real package and the potential typosquatter; any packages with identical fields, which could indicate an attempt to confuse an unsuspecting user, are outputted in red characters.

pypi-scan has been to used to identify and report these packages:

• pandar (which was squatting on pandas) Removed from PyPI.
• prompt-tool-kit (which is squatting on prompt-toolkit) Reported to maintainer.
• requestsaa (which is squatting on requests) Reported to maintainer.

For additional analysis related to typosquatting and PyPI, see:

John Speed Meyers and Bentz Tozer, "Bewear: Python Typosquatting is About More than Typos," IQT Blog, September 30, 2020. Link

John Speed Meyers and Bentz Tozer, "pypi-scan: A Tool for Scanning the Python Package Index for Typosquatters," IQT Blog, October 16, 2020. Link

Usage

(Requires an internet connection.)

List potential typosquatters on a specific package (e.g. numpy):

>>> python main.py -m numpy
Checking numpy for typosquatting candidates.
0: bumpy
1: dumpy
2: gnumpy
...

Timing info: ~10 seconds

List potential typosquatters on top packages:

>>> python main.py -o top-mods
Number of top packages to examine: 43
urllib3 : ['urllib4', 'urllib5']
botocore : ['kotocore']
...

Timing info: ~15 seconds Note I: This command generates a .json report file in the 'results' directory. Note II: Users can whitelist known good packages by adding package names in whitelist.txt; there should be one package name per line.

Advanced usage includes use of several switches:

# Search top 100 pypi packages, if the package name is of at least length
# 4 and for all typosquatters within an edit distance of two. See the help
# section below for further explanation of switches.
>>> python main.py -o top-mods -e 2 -n 100 -l 4
...

Timing info: ~20 seconds

List packages recently added to PyPI and any other packages that these new packages might be typosquatting. This functionality is new and still under development.

>>> python main.py -o recent-scan
...

List potential names that typosquatters might choose for a particular package. A user could then defensively register these names on pypi. This functionality uses QWERTY distance, not Levenshtein distance. QWERTY distance measures distances by the distance between keys on a computer keyboard.

>>> python main.py -o defend-name -m pandas
Here is a list of similar names--measured by keyboard distance--to "pandas":
0: pabdas
1: panfas
2: pansas
...(output shortened)

Timing info: ~1 second NOTE: One colleague has asked me if registering similar namespaces as a defensive protection against typosquatting is ethical. My own review of Pypi suggests the practice is common among top-downloaded packages. But is it ethical? I'm not sure.

Alternatively, to build and run a container via Docker:

docker build -t pypi-scan .
# To open bash shell inside container
docker run -it test_app bash

All commands can then be run from the command line inside the container.

Installation

Download to your local machine via git:

git clone https://github.com/IQTLabs/pypi-scan

After navigating to folder that contains main.py, ensure dependencies are installed and then run tests:

cd pypiscan/pypiscan
pip install -r requirements.txt
python -m unittest

Help

Get help on command line options by navigating to folder that contains main.py and running this command:

>>> python main.py -h
usage: main.py [-h] [-o {mod-squatters,top-mods,defend-name,scan-recent}]
               [-m MODULE_NAME] [-e EDIT_DISTANCE] [-n NUMBER_PACKAGES]
               [-l LEN_PACKAGE_NAME] [-s] [--save]

optional arguments:
  -h, --help            show this help message and exit
  -o {mod-squatters,top-mods,defend-name,scan-recent}, --operation {mod-squatters,top-mods,defend-name,scan-recent}
                        Specify operation to perform. (default: mod-squatters)
  -m MODULE_NAME, --module_name MODULE_NAME
                        Module name to check for typosquatters. (default:
                        None)
  -e EDIT_DISTANCE, --edit_distance EDIT_DISTANCE
                        Maximum edit distance to check. (default: 1)
  -n NUMBER_PACKAGES, --number_packages NUMBER_PACKAGES
                        Specify number of top packages to scan (default: 50)
  -l LEN_PACKAGE_NAME, --len_package_name LEN_PACKAGE_NAME
                        Specify minimum length of package name (default: 5)
  -s, --stored_json     Use a stored top package list (default: False)
  --save                When using scan-recent, save newly created package
                        list (default: False)

NOTE: This command line interface is under development and could have changed.

Contributing

Pull requests are welcome! Also, you can open issues or submit bugs and security issues in the Github issue tracker. If you submit new functionality, please add corresponding tests.

Needed Improvements and Potential Extensions

• Check for forms of typosquatting other than "fat finger" attacks (open-cv vs. open-vc) and and order attacks (open-cv vs. cv-open). For instance, what about when there are real words that are joined and the attackers add a dash or underscore (coolpkg vs. cool-pkg)?
• Is there a way to use visualization to explore for typosquatting attacks?
• How should package metadata be used to search for potential typosquatters? The current version uses metadata crudely: pypi-scan simply changes the coloring of outputted text when a package has any identical metadata to the real package. There are likely improvements to be made.
• Should pypi-scan try to find attacks other than typosquatting? What types of attacks? How?

Again, issues, comments, discussion, and questions are welcome!

More coming soon...