Re: [PATCH bpf-next v2 1/3] net: Parse IPv6 ext headers from TCP sock_ops

[Mathieu Jadin <mathjadin@gmail.com>]

On 10/12/2021 03:01, Jakub Kicinski wrote:
> On Tue,  7 Dec 2021 23:56:33 +0100 Mathieu Jadin wrote:
>> Add a flag that, if set, triggers the call of eBPF program for each
>> packet holding an IPv6 extension header. Also add a sock_ops operator
>> that identifies such call.
>>
>> This change uses skb_data and skb_data_end introduced for TCP options'
>> parsing but these pointer cover the IPv6 header and its extension
>> headers.
>>
>> For instance, this change allows to read an eBPF sock_ops program to
>> read complex Segment Routing Headers carrying complex messages in TLV or
>> observing its intermediate segments as soon as they are received.
> 
> Can you share example use cases this opens up?
> 
In the context of IPv6 Segment Routing, Host A could communicate to Host 
B the Segment Routing Header (SRH) that it should use. For instance, a 
server could alleviate the load on its load balancer by asking the other 
host to route data away from the load balancer.
This can be more generic: Host A can have better information about the 
paths to use than the Host B.

Host A could communicate that in two ways:

1) Host B could simply reverse the segment list of an SRH used by Host 
A. This would make Host B follow the same path as Host A.

2) Host A could use a custom TLV in its own SRH to give an SRH to Host B.

For both options, Host B needs another patch to actually set the SRH in 
the same eBPF program. I plan to submit that later.

As for the other IPv6 extension headers, this seemed like a good 
opportunity to enable users to parse their received extension headers in 
eBPF.

Mathieu