Enforce 192-bit key length for TripleDES

[alex]

TripleDES now only accepts 24-byte (192-bit) keys. Users needing single DES (1-key) or two-key Triple DES compatibility must expand the key themselves:

• Single DES (8 bytes): key + key + key
• Two-key 3DES (16 bytes): key + key[:8]

This change removes the automatic key expansion that was previously done in the TripleDES constructor, making the key length requirement explicit and preventing accidental use of weaker key configurations.

[reaperhulk]

Have we tagged our downstreams to let them know about this?

[alex]

No, I was going to let the warning play that role. (Since this is a warning and not an immediate breakage.)

[alex]

FYI @glyph, @gpotter2 -- this appears to trigger warnings in twisted and scapy.

For twisted I think there's some tests that just use a short key for no reason. For scapy I think there's code that really does want Single DES (in which case this behavior can be obtained by concatting the key 3x).

[glyph]

We definitely have a recurring issue where people just generate the default, or sometimes even the shortest available, key for tests. And we don't have good infrastructure for responsiveness to warnings 😬

[alex]

@glyph good news, the problem is far dumber: https://github.com/twisted/twisted/blob/506d17fb71abf213dc77cfe7b55eb1fafb5f2437/src/twisted/conch/test/test_transport.py#L702-L705 this assumes all keys have a size of 16 bytes, but that's wrong for 3des, it's key size is 24 bytes. If you fix that, I think twisted will be good.

[glyph]

OK, unless I hear otherwise, I assume Twisted is good now

[gpotter2]

Thanks a lot for the heads up, I've drafted a PR on our side to have this ready.

I think there's code that really does want Single DES

Heh yeah, for compat with ancient IPSEC, Kerberos and TLS cases.

[alex]

thanks for jumping on this!