Potential misuse of linear-time or constant-time password comparisons

[brandond]

h/t to @porcupineyhairs

There are several places in the codebase where passwords are either directly compared, or where subtle.ConstantTimeCompare is misused to directly compare passwords. In places where the password length is variable, we should instead only use constant-time comparisons of password hashes, as per the discussion at golang/go#47001

We do not believe that this constitutes a vulnerability, as the current deficiencies only provide a theoretical vector by which password lengths could be derived via an unknown side-channel attack, and the lengths of the affected passwords (the cluster join token, and the node secret) are already well-known in most cases due to their having hard-coded default sizes when K3s is left to generate them for itself.

[fmoral2]

*no specific tests to be run

But i have validated cluster joins with different roles and tokens.

All OK.