Fnord is a pattern extractor for obfuscated code
Fnord has two main functions:
- Extract byte sequences and create some statistics
- Use these statistics, combine length, number of occurrences, similarity and keywords to create a YARA rule
Fnord processes the file with a sliding window of varying size to extract all sequences of with a minimum length
-m X (default: 4) up to a maximum length
-x X (default: 40). For each length, Fnord will present the most frequently occurring sequences
-t X (default: 3) in a table.
Each line in the table contains:
- Number of occurrences
- Sequence (string)
- Formatted (ascii/wide/hex)
- Hex encoded form
2. YARA Rule Creation
Fnord also generates an experimental YARA rule. During YARA rule creation it will calculate a score based in the length of the sequence and the number of occurrences (length * occurrences). It will then process each sequences by removing all non-letter characters and comparing them with a list of keywords (case-insensitive) to detect sequences that are more interesting than others. Before writing each string to the rule Fnord calculates a Levenshtein distance and skips sequences that are too similar to sequences that have already been integrated in the rule.
[Experimental] Fnord was created a few days ago and I have tested it with a handful of samples. My guess is that I'll adjust the defaults in the coming weeks and add some more keywords, filters, scoring options.
Improve the Results
If you've found obfuscated code in a sample, use a hex editor to extract the obfuscated section of the sample and save to a new file. Use that new file for the analysis.
Play with the flags
Please send me samples that produce weak YARA rules that could be better.
____ __ / __/__ ___ _______/ / / _// _ \/ _ \/ __/ _ / /_/ /_//_/\___/_/ \_,_/ Pattern Extractor for Obfuscated Code v0.6, Florian Roth usage: fnord.py [-h] [-f file] [-m min] [-x max] [-t top] [-n min-occ] [-e min-entropy] [--strings] [--include-padding] [--debug] [--noyara] [-s similarity] [-k keywords-multiplier] [-r structure-multiplier] [-c count-limiter] [--yara-exact] [--yara-strings max] [--show-score] [--show-count] [--author author] Fnord - Pattern Extractor for Obfuscated Code optional arguments: -h, --help show this help message and exit -f file File to process -m min Minimum sequence length -x max Maximum sequence length -t top Number of items in the Top x list -n min-occ Minimum number of occurrences to show -e min-entropy Minimum entropy --strings Show strings only --include-padding Include 0x00 and 0x20 in the extracted strings --debug Debug output YARA Rule Creation: --noyara Do not generate an experimental YARA rule -s similarity Allowed similarity (use values between 0.1=low and 10=high, default=1.5) -k keywords-multiplier Keywords multiplier (multiplies score of sequences if keyword is found) (best use values between 1 and 5, default=2.0) -r structure-multiplier Structure multiplier (multiplies score of sequences if it is identified as code structure and not payload) (best use values between 1 and 5, default=2.0) -c count-limiter Count limiter (limts the impact of the count by capping it at a certain amount) (best use values between 5 and 100, default=20) --yara-exact Add magic header and magic footer limitations to the rule --yara-strings max Maximum sequence length --show-score Show score in comments of YARA rules --show-count Show count in sample in comments of YARA rules --author author YARA rule author
git clone https://github.com/Neo23x0/Fnord.gitand
pip3 install -r ./requirements.txt
python3 ./fnord.py --help
python3 fnord.py -f ./test/wraeop.sct --yara-strings 10
python3 fnord.py -f ./test/vbs.txt --show-score --show-count -t 1 -x 20
python3 fnord.py -f ./test/inv-obf.txt --show-score --show-count -t 1 --yara-strings 4 --yara-exact
Why didn't you integrate Fnord in yarGen?
yarGen uses a white-listing approach to filter the strings that are best for the creation of a YARA rule. yarGen applies some regular expressions to adjust scores of strings before creating the YARA rules. But its approach is very different to the method used by Fnord, which calculates the score of the byte sequences based on statistics.
While yarGen is best used for un-obfuscated code. Fnord is for obfuscated code only and should produce much better results than yarGen.
Follow me on Twitter for updates @cyb3rops