Questions on sk_no_check implemenation

Questions on sk_no_check implemenation

[Tom Herbert]

Hi, I'm looking at sk_no_check implementation and am hoping to get
clarification on a couple of use cases.

1) In l2tp_core.c, if sk_no_check is set it looks like this will
completely bypass UDP checksum verification on RX even for packets
with nonzero checksums.

d2cf33616 (Benjamin LaHaise 2012-04-27 08:24:18 +0000 505) if
(sk->sk_no_check || skb_csum_unnecessary(skb))

I don't believe this is standards compliant and seems pretty risky
otherwise. What is the requirement here? Is this still needed?

2) The constant UDP_CSUM_NORCV was introduced for sunrpc and in fact
this looks like the only code that sets it. AFAICT the only effect
this would have would be to disable receive checksum verification in
L2TP as described above. What was the intent of UDP_CSUM_NORCV and why
was using this specific to sunrpc?

Thanks for  your help,
Tom

Re: Questions on sk_no_check implemenation

[David Miller]

From: Tom Herbert <therbert@google.com>
Date: Mon, 5 May 2014 19:34:40 -0700

> Hi, I'm looking at sk_no_check implementation and am hoping to get
> clarification on a couple of use cases.
> 
> 1) In l2tp_core.c, if sk_no_check is set it looks like this will
> completely bypass UDP checksum verification on RX even for packets
> with nonzero checksums.
> 
> d2cf33616 (Benjamin LaHaise 2012-04-27 08:24:18 +0000 505) if
> (sk->sk_no_check || skb_csum_unnecessary(skb))
> 
> I don't believe this is standards compliant and seems pretty risky
> otherwise. What is the requirement here? Is this still needed?

I think this is simply a thinko and the code should be checking non-zero
csums.

> 2) The constant UDP_CSUM_NORCV was introduced for sunrpc and in fact
> this looks like the only code that sets it. AFAICT the only effect
> this would have would be to disable receive checksum verification in
> L2TP as described above. What was the intent of UDP_CSUM_NORCV and why
> was using this specific to sunrpc?

I have no idea about this case.