[rds-devel] [PATCH v3 net-next 0/3] rds: IPv6 support

[David Miller]

From: Sowmini Varadhan <sowmini.varadhan at oracle.com>
Date: Wed, 18 Jul 2018 03:33:40 -0700

> On (07/18/18 15:19), Ka-Cheong Poon wrote:
>> >bind() and connect() are using the sa_family/ss_family to have
>> >the application signal to the kernel about whether ipv4 or ipv6 is
>> >desired. (and bind and connect are doing the right thing for
>> >v4mapped, so that doesnt seem to be a problem there)
>> >
>> >In this case you want the application to signal that info via
>> >the optlen.  (And the reason for this inconsistency is that you dont
>> >want to deal with the user->kernel copy in the same way?)
>> 
>> 
>> Because doing that can break existing RDS apps.  Existing code
>> does not check the address family in processing this socket
>> option.  It only cares about the address and port.  If the new
> 
> I'll leave this up to DaveM. Existing code only handles IPv4,
> 
> everywhere else, we always check the sa_family or ss_family
> first and verify the length afterward. This was DaveM's original
> point about bind/connect/sendmsg. I dont know why rds sockopts have
> to be special.

Yes, but the above point is valid.

If the code never verified the sa_family value before, it is a very
real possibility that code exists out that which is not initializing
it or setting it incorrectly.

Those apps have worked for a long time, and suddenly will break.

We often have to deal with unfortunate mistakes like this.

But for now, I guess the check can be added but we have to look out
for any regressions this causes and revert if necessary.

Thanks.