[Swan-dev] [libreswan/libreswan] `make rpm` in FIPS mode on Fedora 32 fails (#351) (fwd)

[Paul Wouters]

On Mon, 6 Jul 2020, Andrew Cagney wrote:

> 
> This provides more info:
>
> diff --git a/programs/cavp/Makefile b/programs/cavp/Makefile
> index 90b724ee1c..ca4a8a7120 100644
> --- a/programs/cavp/Makefile
> +++ b/programs/cavp/Makefile
> @@ -67,7 +67,7 @@ test.cavp.%.fax: | $(builddir)
>
> .PRECIOUS: %.out
> test.cavp.%.out: test.cavp.%.fax $(PROGRAM)
> -       $(builddir)/cavp \
> +       $(builddir)/cavp -fips -v \
>                $(builddir)/test.cavp.$*.fax \
>                > $(builddir)/test.cavp.$*.tmp
>        mv  $(builddir)/test.cavp.$*.tmp  $(builddir)/test.cavp.$*.out

I guess the spec file should detect if we are in FIPS mode and add that
flag in we are building in FIPS mode? But then it should skip any
non-allowed test vectors. And I think really we should not have non-FIPS
test vectors in these files because they came from NIST? :)

Maybe we should just remove those tests from the CAVP files?

Since NIST moved to CMVP, I guess these CAVP files won't see any more
updates.

And another interesting question would be, shouldn't we make this
requirement forced in non-FIPS too?

>> From memory, the key needs to be >= digest size when fips.  128 bits
> falls short of 384 and 512.

Yeah, it should be half the keysize atleast. So for SHA-384 you need 192
bits so nonces of 96 bits.

Paul