mirror of
https://github.com/trezor/trezor-firmware.git
synced 2024-12-24 07:18:09 +00:00
cf3c57d0ae
Introduce fuzzing harnesses for zkp* functions and adapt some differential fuzzing Additional documentation and minor cleanup Add temporary workaround for clang-14 and more explicit Makefile behavior
50 lines
2.2 KiB
Bash
Executable File
50 lines
2.2 KiB
Bash
Executable File
#!/usr/bin/env bash
|
|
|
|
# usage: script.sh target-dictionary-filename
|
|
|
|
# This script searches for interesting strings in the source code and converts
|
|
# them into a standard fuzzer dictionary file.
|
|
#
|
|
# Note that this script is phased out in favor of the more sophisticated
|
|
# extract_fuzzer_dictionary.py program
|
|
|
|
# TODO known issues: the end result has some duplicates in it
|
|
|
|
TARGET_DIR=../tests
|
|
OUTPUT_FILE=${1:-fuzzer_crypto_tests_strings_dictionary1.txt}
|
|
|
|
multiline_string_search() {
|
|
# TODO the `find` regex behavior is Linux-specific
|
|
find $TARGET_DIR -type f -regextype posix-extended -regex '.*\.(c|h|py|json|java|js)' | xargs cat | perl -p0e 's/"\s*\n\s*\"//smg'
|
|
}
|
|
|
|
# ensure empty file
|
|
echo -n "" > $OUTPUT_FILE
|
|
|
|
# strip multiline strings and extract them
|
|
# exclude some hex strings, but allow hex strings with mixed capitalization (Ethereum, rskip60)
|
|
multiline_string_search | grep -P -o "\"[\w ]+\"" | grep -v -P "\"(([0-9a-f][0-9a-f])+|([0-9A-F][0-9A-F])+)\"" | sort | uniq | while read -r line ; do
|
|
echo "$line" >> $OUTPUT_FILE
|
|
done
|
|
|
|
# extract individual BIP39 and SLIP39 words
|
|
# TODO are those actually valuable as fuzzer dictionary input?
|
|
# grep -r -P -o -h "\"\w+\"" ../slip39_wordlist.h ../bip39_english.h | sort | uniq >> fuzzer_crypto_tests_strings_dictionary1.txt
|
|
|
|
# extract and convert binary input data from the unit tests
|
|
# find each file, cat it, concatenate multiline strings, look for hex strings in quotes
|
|
# note that this returns multiple megabyte of result strings due to the large amount
|
|
# of test cases in the wycheproof project subfolder
|
|
multiline_string_search | grep -P -o "\"([0-9a-fA-F][0-9a-fA-F])+\"" | grep -P -o "([0-9a-fA-F][0-9a-fA-F])+" | sort | uniq | while read -r line ; do
|
|
# turn ascii hex strings AA into \xaa for the fuzzer format and add quotes
|
|
# extra backslash escape due to the bash nesting
|
|
escaped_hex=`echo $line | sed -e 's/../\\\\x&/g'`
|
|
echo "\"$escaped_hex\"" >> $OUTPUT_FILE
|
|
done
|
|
|
|
# search and reassemble BIP39 test seeds that span multiple lines
|
|
# find each file, cat it, concatenate multiline strings, look for BIP39 seed combinations with reasonable length
|
|
multiline_string_search | grep -Po "(\w{3,10} ){11,23}(\w{3,10})" | sort | uniq | while read -r line ; do
|
|
echo "\"$line\"" >> $OUTPUT_FILE
|
|
done
|