Add `rb/weak-sensitive-data-hashing` query port
This is a replacement of https://github.com/github/codeql/pull/12782, porting py/weak-sensitive-data-hashing to Ruby.
As with the previous PR, the python version is a little more sophisticated in terms of its source modeling. Compared to before, we have moved towards a concept of SensitiveDataSources as sources rather than just any SensitiveNode. This accounts for cases like:
def login_with_password(passwd) # the `passwd` parameter here should be marked as the source
some_unremarkable_method(passwd) # the access to `passwd` is a `SensitiveNode`, but is not a useful source
Digest::MD5.hexdigest(passwd) # read of `passwd` here is a sink
end
where using SensitiveNodes as sources would result in the read of passwd as an argument to some_unremarkable_method also being marked as a source.
Ideally we would move this out to a shared file to match Python's SensitiveDataSources.qll, but I've kept this as a private module in SensitiveDataHashingCustomization.qll to restrict the impact of this change for now.
QHelp previews:
ruby/ql/src/queries/security/cwe-327/WeakSensitiveDataHashing.qhelp
Use of a broken or weak cryptographic hashing algorithm on sensitive data
Using a broken or weak cryptographic hash function can leave data vulnerable, and should not be used in security related code.
A strong cryptographic hash function should be resistant to:
- pre-image attacks: if you know a hash value
h(x), you should not be able to easily find the inputx. - collision attacks: if you know a hash value
h(x), you should not be able to easily find a different inputywith the same hash valueh(x) = h(y). In cases with a limited input space, such as for passwords, the hash function also needs to be computationally expensive to be resistant to brute-force attacks. Passwords should also have an unique salt applied before hashing, but that is not considered by this query.
As an example, both MD5 and SHA-1 are known to be vulnerable to collision attacks.
Since it's OK to use a weak cryptographic hash function in a non-security context, this query only alerts when these are used to hash sensitive data (such as passwords, certificates, usernames).
Use of broken or weak cryptographic algorithms that are not hashing algorithms, is handled by the rb/weak-cryptographic-algorithm query.
Recommendation
Ensure that you use a strong, modern cryptographic hash function:
- such as Argon2, scrypt, bcrypt, or PBKDF2 for passwords and other data with limited input space.
- such as SHA-2, or SHA-3 in other cases.
Example
The following example shows two functions for checking whether the hash of a certificate matches a known value -- to prevent tampering. The first function uses MD5 that is known to be vulnerable to collision attacks. The second function uses SHA-256 that is a strong cryptographic hashing function.
require 'openssl'
def certificate_matches_known_hash_bad(certificate, known_hash)
hash = OpenSSL::Digest.new('SHA1').digest certificate
hash == known_hash
end
def certificate_matches_known_hash_good(certificate, known_hash)
hash = OpenSSL::Digest.new('SHA256').digest certificate
hash == known_hash
end
Example
The following example shows two functions for hashing passwords. The first function uses SHA-256 to hash passwords. Although SHA-256 is a strong cryptographic hash function, it is not suitable for password hashing since it is not computationally expensive.
require 'openssl'
def get_password_hash(password, salt)
OpenSSL::Digest.new('SHA256').digest(password + salt) # BAD
end
The second function uses Argon2 (through the argon2 gem), which is a strong password hashing algorithm (and includes a per-password salt by default).
require 'argon2'
def get_initial_hash(password)
Argon2::Password.create(password)
end
def check_password(password, known_hash)
Argon2::Password.verify_password(password, known_hash)
end
References
- OWASP: Password Storage Cheat Sheet
- Common Weakness Enumeration: CWE-327.
- Common Weakness Enumeration: CWE-328.
- Common Weakness Enumeration: CWE-916.
In the performance evaluation I see an increase in 'String cache size' for a single project :thinking: I didn't notice any kind of string manipulation being introduced in this PR... do you have an explanation for this? (maybe I overlooked something) -- If this project had 1000 new alerts I might understand it from the additional alert messages, but it doesn't seem like it even has 1 new alert.
I don't quite understand the need to move SensitiveDataSources away from the
SensitiveActions.qllfile in f217de9? -- even whenrb/weak-sensitive-data-hashingis the only Ruby query to use it, would there be a problem from it existing in the file? (as I understand it, it wouldn't affect existing uses of SensitiveNode)
This was just to avoid committing to a new public interface at this point - it shouldn't change anything in terms of functionality of this or any other query. The new sources could potentially be useful for other future queries, at which point it would make sense to undo this change (and maybe tidy up the interface before making it public if necessary).
In the performance evaluation I see an increase in 'String cache size' for a single project 🤔 I didn't notice any kind of string manipulation being introduced in this PR... do you have an explanation for this? (maybe I overlooked something) -- If this project had 1000 new alerts I might understand it from the additional alert messages, but it doesn't seem like it even has 1 new alert.
Honestly, I don't have an explanation for this. The alert message concatenation should be the only new string manipulation in this PR.