[dpdk-users] checksums?

[dpdk-users] checksums?

[William Herrin]

Hi folks,

I've recently done a deep dive in to IP checksums and I've run in to
something I don't understand. Any insight would be helpful.

rte_ipv4_cksum() is implemented as:

return (cksum == 0xffff) ? cksum : (uint16_t)~cksum;

Which means: if the sum is zero, return -0 (0xffff) never +0 (0x0000).
Welcome to the wonderful world of 1's complement arithmetic.

RFC 1624, on the other hand, says:

"In one's complement, there are two representations of zero: the all zero
and the all one bit values, often referred to as +0 and -0. One's
complement addition of non-zero inputs can produce -0 as a result, but
never +0.  Since there is guaranteed to be at least one non-zero field in
the IP header, and the checksum field in the protocol header is the
complement of the sum, the checksum field can never contain ~(+0), which is
-0 (0xFFFF).  It can, however, contain ~(-0), which is +0 (0x0000)."

Which I understand to mean that +0 (0x0000) is a legal value in an IPv4
checksum field, but -0 (0xffff) is not.

Is this a bug? Is there a more authoritative source for which zero is
correct in an IPv4 header? Please help me find the error in my
understanding.

Thanks,
Bill Herrin


-- 
William Herrin ................ herrin@dirtside.com  bill@herrin.us
Dirtside Systems ......... Web: <http://www.dirtside.com/>

Re: [dpdk-users] checksums?

[N. Benes]

Hi,

William Herrin:
> Hi folks,
> 
> I've recently done a deep dive in to IP checksums and I've run in to
> something I don't understand. Any insight would be helpful.
> 
> rte_ipv4_cksum() is implemented as:
> 
> return (cksum == 0xffff) ? cksum : (uint16_t)~cksum;
> 
> Which means: if the sum is zero, return -0 (0xffff) never +0 (0x0000).
> Welcome to the wonderful world of 1's complement arithmetic.
> 
> RFC 1624, on the other hand, says:
> 
> "In one's complement, there are two representations of zero: the all zero
> and the all one bit values, often referred to as +0 and -0. One's
> complement addition of non-zero inputs can produce -0 as a result, but
> never +0.  Since there is guaranteed to be at least one non-zero field in
> the IP header, and the checksum field in the protocol header is the
> complement of the sum, the checksum field can never contain ~(+0), which is
> -0 (0xFFFF).  It can, however, contain ~(-0), which is +0 (0x0000)."
> 
> Which I understand to mean that +0 (0x0000) is a legal value in an IPv4
> checksum field, but -0 (0xffff) is not.
> 
> Is this a bug? Is there a more authoritative source for which zero is
> correct in an IPv4 header? Please help me find the error in my
> understanding.
> 
> Thanks,
> Bill Herrin
> 
> 

Related threads:

https://mails.dpdk.org/archives/users/2019-March/004021.html
https://mails.dpdk.org/archives/dev/2019-April/128473.html

Cheers!

Re: [dpdk-users] checksums?

[William Herrin]

On Fri, Apr 5, 2019 at 2:31 AM N. Benes <nbenes@eso.org> wrote:
> William Herrin:
> > Which I understand to mean that +0 (0x0000) is a legal value in an IPv4
> > checksum field, but -0 (0xffff) is not.
> >
> > Is this a bug?
>
> Related threads:
>
> https://mails.dpdk.org/archives/users/2019-March/004021.html
> https://mails.dpdk.org/archives/dev/2019-April/128473.html

Ah, my google-fu failed me. Thanks for the refs!

So it is a bug and Olivier Matz is writing a patch. Few receivers stumble
when presented with -0 as a checksum so it shouldn't be causing much of a
problem but I did stumble across this long-ago conversation:

https://mailarchive.ietf.org/arch/msg/ietf/2EZyjq9tiQvyfyAXgPMKQ7MtToU
https://mailarchive.ietf.org/arch/msg/ietf/36lReuBvahPjxVTth1wpm7kt7zs

Regards,
Bill Herrin


--
William Herrin ................ herrin@dirtside.com  bill@herrin.us
Dirtside Systems ......... Web: <http://www.dirtside.com/>