Responses are in-line.
(...)
There is a more substantive comment on the first paragraph of
section 1.
The text states:
If an implementation chooses to support one of the algorithms
discussed in this document, then the implementation MUST do so as
described in this document.
I believe the text should be:
If an implementation chooses to support one of the algorithms
discussed in this document, then the implementation MUST do so as
described in [SHS].
The algorithms aren't defined in this document only the OIDs and where they
go in CMS ... so I still think it's actually "as described in this
document". But, Spencer suggests removing the sentences because they're not
needed and I tend to agree with him.
Spencer wins on this one. The sentence doesn't make much sense in a
standards-track document.
It is important to know in which document the algorithms are described in
detail.
Deleting the sentence would not solve my concern.
>A small discussion in the security considerations section on
the advantages (in particular in terms of performances versus
security) of using one or another function from the SHA2
family would be helpful.
I'll see if I can't get something from NIST (or at least point to it).
I'll disagree with this. These are not security considerations; they
are performance considerations. And pretty obvious ones at that.
It is both performance and security.
The larger the hash is, the better is the security, but the performance may be
lower.
>While I welcome this draft, everybody should take into
consideration that, if the SHA2 family happens to be broken
then we will be at risk.
This should be mentioned into the security considerations section.
If an algorithm is cracked then isn't it obvious that we're in trouble? No
other algorithm document I could find says something like this so I'm
inclined to not include this in the security considerations section.
... or anywhere else. If any algorithm (hash, encryption, signing,
...) is broken, it is broken. Sean's right here.
The message is the following: if the SHA2 family is broken, then you had better
to use two hash algorithms from a different family (e.g. use Whirlpool).
We should also reference
http://www.ietf.org/internet-drafts/draft-ietf-smime-multisig-04.txt
which allows to use two different hash functions (from different families, if
possible).
Denis