svn commit: r272886 - in head/sys: netinet netinet6

[Bjoern A. Zeeb]

> On 12 Jan 2015, at 15:51 , John Baldwin <john at baldwin.cx> wrote:
> 
> On Tuesday, January 06, 2015 07:07:11 PM Bryan Venteicher wrote:
>> On Tue, Jan 6, 2015 at 5:27 PM, Bryan Drewery <bdrewery at freebsd.org> wrote:
>>> On 1/6/2015 4:00 PM, Bryan Venteicher wrote:
>>>> On Tue, Jan 6, 2015 at 2:52 PM, John Nielsen <lists at jnielsen.net
>>>> 
>>>> <mailto:lists at jnielsen.net>> wrote:
>>>>    Bryan-
>>>> 
>>>>    On Oct 10, 2014, at 12:09 AM, Bryan Venteicher <bryanv at freebsd.org
>>>> 
>>>>    <mailto:bryanv at freebsd.org>> wrote:
>>>>> Author: bryanv
>>>>> Date: Fri Oct 10 06:08:59 2014
>>>>> New Revision: 272886
>>>>> URL: https://svnweb.freebsd.org/changeset/base/272886
>>>>> 
>>>>> Log:
>>>>> Add context pointer and source address to the UDP tunnel callback
>>>>> 
>>>>> These are needed for the forthcoming vxlan implementation. The
>>> 
>>> context
>>> 
>>>>> pointer means we do not have to use a spare pointer field in the
>>> 
>>> inpcb,
>>> 
>>>>> and the source address is required to populate vxlan's forwarding
>>> 
>>> table.
>>> 
>>>>> While I highly doubt there is an out of tree consumer of the UDP
>>>>> tunneling callback, this change may be a difficult to eventually
>>> 
>>> MFC.
>>> 
>>>>    I noticed this comment while doing an MFC of vxlan to my local tree.
>>>>    Do you think an MFC to 10-STABLE of this change (and vxlan
>>>>    generally) will be feasible? Is there precedent for ABI changes like
>>>>    this being sanctioned? Could symbol versioning help?
>>>> 
>>>> I'd like to get some consensus on whether this commit is OK to MFC. With
>>>> this commit, vxlan should be an easy to MFC.
>>> 
>>> Breaking ABI will potentially hurt packages. FreeBSD builds packages for
>>> the oldest supported release on a branch. If you break ABI in 10.2 while
>>> we are building packages for 10.1 then any packages using these
>>> interfaces may not work right or result in panics packages with kmods.
>>> Please consider that.
>> 
>> The only user visible change of this commit would be the addition of a
>> field at the end of 'struct udpcb'. I don't think that is a problem, at
>> least a similar change didn't prevent the MFC of UDP Lite.
>> 
>> The kernel part of this changes the UDP tunneling functions which I guess
>> there could be a 3rd party module out there, but I very highly doubt that,
>> based on how un-useful the previous interface was.
> 
> Userland should not be impacted by this at all.  (Nothing in userland cares
> about udpcb's internals.)  I think there was only ever one consumer for the 
> existing UDP tunneling code (bz@ knows what it is).  I'm not sure where it 
> lives.

If you are talking about u_tun_func then it came from SCTP over UDP tunneling.  tuexen and rrs are your friends.

I was wondering if it could be used similarly for IPsec UDPencap but I think that went nowhere back then.

— 
Bjoern A. Zeeb                                  Charles Haddon Spurgeon:
"Friendship is one of the sweetest joys of life.  Many might have failed
 beneath the bitterness of their trial  had they not found a friend."