Found a new malware? Great! Add a new sample and rule in 5 minutes.
- Add malware file
- Find a literal string or regex that identifies the malware
- Run tests to validate results
- Make Pull Request
- Profit!
Every malware should have at least 1 corresponding rule.
Every rule should have at least 1 corresponding malware.
After installing the requirements, malware can be added by following these steps.
sudo apt install -qy yara
# If you don't have EPEL yet, for CentOS 6:
wget https://dl.fedoraproject.org/pub/epel/epel-release-latest-6.noarch.rpm
sudo rpm -ivh epel-release-latest-6.noarch.rpm
sudo yum -y install yara
If you don't have Homebrew yet, follow instructions on https://brew.sh/
brew install coreutils python3 yara md5sha1sum gnu-sed
Save your malare sample in the corpus/incoming/frontend
folder for JavaScript samples, or corpus/incoming/backend
folder for PHP samples.
Create an MD5 hash of the file, rename it and move it to the appropriate corpus.
For PHP samples:
tools/backend_md5_to_incoming.sh
For JavaScript samples:
tools/frontend_md5_to_incoming.sh
Check if your sample isn't already detected:
tools/validate_signatures.py
If it's not recognized yet, it will show you a warning. Please proceed to contribute it!
Please edit frontend.txt
or backend.txt
in the rules
folder. You have two options here:
- Create a string literal: just copy/past a relevant portion of the malware. It will be scanned as-is (no backslash escaping etc).
- Create a regular expression. This is more powerful, but also slower. Please only use if absolutely necessary.
Malware signatures can be extremely specific (a file checksum) or too generic (check for suspicious eval()
). In case of doubt, choose specific. We try to avoid false positives at all costs!
Example malware
<?php @eval(stripslashes($_REQUEST[q]));
Example rule
# eval looks evil
@eval(stripslashes($_REQUEST[q]));
The above example is simple and sufficient. More advanced samples may require more complex signatures. See below for signature writing strategies. In short:
- One signature can cover multiple strains of malware.
- It's better to be specific and have less coverage, than broad coverage and possibly raise false positives.
- Prefer string match over regex match (for speed)
- If using regex, limit the use of unbound match operators (
+
or*
) as they might have to scan the whole file. Better:{,x}
to limit to x characters.
For more information on writing rules, refer to the documentation below
Re-run the validator to verify that your signature indeed identifies the new malware:
tools/validate_signatures.py
No warnings? Great! Please commit and PR.
Once you submit a PR, our Travis job will check your signature against a plethora of Magento installations, just in case. You can also run this locally:
The tools/mageffcheck.sh
bash script has a couple of commands available to get you started.
The idea is to download and extract various vanilla Magento packages and run the YARA tests against it. When these tests return output, it's most likely a false flag.
Running the following command will output some information:
tools/mageffcheck.sh
# OR
tools/mageffcheck.sh help
You can initialize the test setup by appending the init
argument
tools/mageffcheck.sh init
This will download and extract several Magento packages from the OpenMage Magento Mirror.
Note: When using an IDE
Downloading and extracting several Magento package may trigger the IDE's indexation. Due to the amount of code it may lag or even hang (for a while).
These packages will be stored in ./tmp
make sure to mark this directory as excluded before or right after running init
.
After this you could run the following to trigger the YARA tests against the vanilla code.
tools/mageffcheck.sh run
You should expect no output from running this command.
Malware signatures can be extremely specific (a file checksum) to generic (check for suspicious eval()
) or anything in between. As a signature author, you have to decide on a proper balance. Pro specific: no chance for false positives. Pro generic: less work, as one signature will catch multiple strains of malware.
In case of doubt, choose specific.
Taking a shortcut with a more generic signature might be tempting, but if it causes false positives (possibly in the future), you will have to deal with the fallout, plus have to repeat your malware analysis.
Remember, we can already win a lot by de-duplicating unique malware instance identification across organisations. Identifying future, or merely suspicious, malware is also desireable but requires a different strategy.
Other malware research suggests that malwares are quickly morphing, rendering the average lifespan of a unique malware instance to be only 58 seconds. But I haven't seen this behaviour in Magento specific malware yet. Let's enjoy the tide while it lasts.