Re: [bpf-next 1/2] tcp: replace SOCK_DEBUG() with tcp_stats()

[Yafang Shao]

On Wed, Feb 13, 2019 at 10:15 AM Eric Dumazet <edumazet@xxxxxxxxxx> wrote:
>
> On Tue, Feb 12, 2019 at 6:07 PM Yafang Shao <laoar.shao@xxxxxxxxx> wrote:
> >
>
> > Let me explain the background for you.
> > I want to track some TCP abnormal  behavior in TCP/IP stack. But I
> > find there's no good way to do it.
> > The current MIBs are per net, other than per socket, that makes it not
> > very powerful.
> > And the ancient SOCK_DEBUG is not good as well.
> > So we think why not cleanup this ancient SOCK_DEBUG() and introduce a
> > more powerful method.
>
>
> I am all for it, but this more powerful method does nothing at all in
> the current patches.
>
> I can not accept patches just because they seem to be harmless,
> knowing that  the next patches
> will be pushed later changing more stuff, just because the new
> infrastructure is there "and can be used"
>
> Just remove all SOCK_DEBUG() calls, there are leftovers of very ancient times.
>

OK. I will send a patch for it.

> Do not add more debugging stuff unless you can demonstrate
> they actually allowed you to find a real bug and that you sent a
> public fix for it.
>

Sure.

> Just adding "cool stuff" in TCP stack does not please me, it is only
> more complexity for unproven gain.
>
> Otherwise, I am tempted to think that these BPF hooks are there only
> so that a company can more
> easily build a private variant of TCP, yet letting the community
> maintaining the hard part of TCP stack.

:-)

>
> Thank you.


Thanks
Yafang