Fix CryptoJS argument passing in DeriveEVPKey #1767
Conversation
CryptoJS treats strings as Utf8, so for binary strings, Latin1 needs to be used.
| @@ -62,11 +62,13 @@ class DeriveEVPKey extends Operation { | |||
| * @returns {string} | |||
| */ | |||
| run(input, args) { | |||
| const passphrase = Utils.convertToByteString(args[0].string, args[0].option), | |||
| const passphrase = CryptoJS.enc.Latin1.parse( | |||
There was a problem hiding this comment.
This feels a little contrived at the moment. Using Latin1 as the encoding scheme for the string does solve the issue, but doesn't feel optimal. Could we instead replace this with a conversion to a byte array and then create a crypto word array from that? E.g:
// Note: Entirely untested.
CryptoJS.lib.WordArray.create(new UInt8Array(Utils.convertToByteArray(args[0].string, args[0].option)))There was a problem hiding this comment.
That being said, this is a clear improvement over the current broken implementation, so if this isn't a two minute replacement I'm happy to approve and merge in this pull request.
There was a problem hiding this comment.
Doesn't WordArray expect an array of 32bit integers? I'm not sure whether new Uint32Array(new Uint8Array(byteArray).buffer) would have the correct endianness
There was a problem hiding this comment.
Looks like WordArray expects BigEndian version of the buffer.. The Latin1 parse code - at least to me - looks like something as direct as possible given we need an array of 32bit integers represending BE view of the string
There was a problem hiding this comment.
That's fair, I'll approve this then! 👍
As pointed out in #1593, CryptoJS treats strings as Utf8, so for binary strings, Latin1 needs to be used.
This PR should fix that issue, and there are no other instances of raw strings being passed to CryptoJS methods.