libnids symbols sadness
Lukas Schwaighofer
lukas at schwaighofer.name
Sun Aug 27 21:21:40 UTC 2017
On Sun, 27 Aug 2017 23:01:38 +0200
Hilko Bengen <bengen at debian.org> wrote:
> * Lukas Schwaighofer:
>
> > I double checked that as well. While the linux tag seems
> > appropriate, the architecture should indeed be hardcoded with i386,
> > as the guard in the code is
> > #if ( __i386__ || __i386 )
> > and contains i386 assembler… so i386 is really the only architecture
> > which will see the csum_partial function, all others should see
> > ip_check_ext instead.
>
> Somehow I doubt that the hand-crafted bits of assembly code really
> make any difference with today's compilers and CPUs, so one might as
> well get rid of them.
>
> Do I remember correctly that those were only optimizied versions of IP
> or TCP calculations?
Yes, it's only for the functions which compute the IPv4/TCP/UDP
checksums.
I don't really have a strong opinion on removing that code. Reducing
the codebase is always a good thing (especially when upstream hasn't
made a release in 7 years).
However, the code has also been in production for quite a long time
without any problems, so unless I messed up the symbols file again we
might as well keep it.
Regards
Lukas
More information about the Pkg-security-team
mailing list