Simon Levermann commented on CODEC-245:

Though I suppose it's still *somewhat* incorrect: Apparently, *all bytes* are 
valid salts, not just base64 characters. For example, invoking the libc crypt 
function like this:

char *foo = crypt("asdfasdf", "$6$mounds=1000$asdf$");
results in the following hash:
That is: "mounds=1000" is the literal salt here, and the actually supplied 
salt, "asdf" is discarded. Apparently the specification does NOT mandate that 
salts must be B64-encoded. It seems like any bytes (even non-printable 
characters) are valid for salts, but only ever the first 16 bytes are used. 
Thoughts on how to continue here?

> RegEx for verifying salts in ShaCrypt is incorrect
> --------------------------------------------------
>                 Key: CODEC-245
>                 URL: https://issues.apache.org/jira/browse/CODEC-245
>             Project: Commons Codec
>          Issue Type: Bug
>    Affects Versions: 1.11
>            Reporter: Simon Levermann
>            Priority: Major
> The regex/code that extracts the salt from a given salt string in Sha2Crypt 
> treats some invalid salt formats as valid:
> {code:java}
> ^\$([56])\$(rounds=(\d+)\$)?([\.\/a-zA-Z0-9]{1,16}).*
> {code}
> The code then goes on to use capture group 3 (the round count) to determine 
> how many rounds are used, and capture group 4 (the actual salt) to use as 
> salt data.
> However, for an input that contains an invalid salt specification like this:
> {code:java}
> $5$notrounds=1000$asdfasdf
> {code}
> This string is treated as valid. The operation then uses "notrounds" as the 
> salt for hashing:
> {code:java}
> System.out.println(Sha2Crypt.sha256Crypt(new byte[100], 
> "$5$notrounds=1000$asdfasdf"));
> {code}
> The above code prints
> {code:java}
> $5$notrounds$aAEx6EzUBfc.aIIeiItVjUREbj/ar4xlb/qcMxD90NA
> {code}
> This code should probably throw an exception. Additionally, other invalid 
> salt strings like
> {code:java}
> $6$rounds=1000$äöüäöü
> {code}
> Result in hashes like:
> {code:java}
> $5$rounds$ZIRdMgi25kK5Zmi1whKjcnFN/Pe8QCAOlTmxMfwAoq5
> {code}
> Completely ignoring the rounds parameter, and using the literal string 
> "rounds" as the salt for hashing.

This message was sent by Atlassian JIRA

Reply via email to