On 06/10/2025 21:26, Eric Biggers wrote:

On Mon, Oct 06, 2025 at 09:11:41PM +0200, Vegard Nossum wrote:

quoted

The fact is that fips=1 is not useful if it doesn't actually result
something that complies with the standard; the only purpose of fips=1 is
to allow the kernel to be used and certified as a FIPS module.

Don't all the distros doing this actually carry out-of-tree patches to
fix up some things required for certification that upstream has never
done?  So that puts the upstream fips=1 support in an awkward place,
where it's always been an unfinished (and undocumented) feature.

I can't speak for all distros, but we have a handful of patches, around
6 or 7 I believe, most are fairly small. (We are, however, looking to
move to the standalone module I sent the RFC for, which has a lot more
patches...)

But yes, mainline fips=1 support is in a slightly awkward place. I see
no real reason for anybody to ever use it in production unless it's
actually a NIST certified build either.

That doesn't mean we shouldn't try to minimize the amount of downstream
patches, though. (IMHO, anyway.)

I would like to try to document what fips=1 is currently and how to use
it and how to program for it (if nobody -- however unlikely -- beats me
to it). I came across this thread from over 10 years ago where people
are asking about the kernel FIPS docs and we still don't have any:

https://mta.openssl.org/pipermail/openssl-users/2015-March/000904.html


Vegard