Re: unused in_cksum_update() ?

On Fri, Jan 10, 2014 at 02:31:40PM +0400, Gleb Smirnoff wrote:
> On Thu, Jan 09, 2014 at 08:21:14PM +0100, Luigi Rizzo wrote:
> L> a lot of arch-specific headers (sys/${ARCH}/include/in_cksum.h)
> L> have a lengthy definition for
> L> 
> L> 	in_cksum_update(struct ip *ip)
> L> 
> L> which seems completely unused in our source tree.
> L> Time to remove it perhaps ?
> L> 
> L> grep cannot find any use at least since stable/8
> 
> I'd prefer not to hurry with its removal. Might be that pf will use it.
> Since it lives in a header file, it doesn't add a single bit to kernel
> size.

we should care more about obfuscation and correcteness, and this is
a killer in both respects.
Depending on $arch the function is not even available or wrong:

In particular, the basic code follows the description in
http://tools.ietf.org/html/rfc1141A with ntohs/htons to deal
with endianness (note that the '256' should not be converted):

	tmp = ntohs(sum)+256;
	tmp = tmp + (tmp >> 16);
	sum = htons(tmp); // also truncates high bits

It is correctly implemented (but in a totally generic way, so no
point to have it in the arch-specific files) for  amd64, i386,
ia64, mips, powerpc; it is not implemented for arm, and it is wrong
for sparc64 (where the 256 is incorrectly replaced by a 1).

In terms of usage: the svn repo suggests that it was added in r15884
in 1996 (stable/2.2 is the first branch where it appears):

http://svnweb.freebsd.org/base/head/sys/i386/include/in_cksum.h?r1=15884&r2=15883&pathrev=15884

as far as i can tell never used anywhere, and copied from place to
place when we started to support different architectures.

Shall we wait until it becomes 18 ? :)

I am adding Garret to the list as he may have more details.

	cheers
	luigi