Re: FIN_WAIT1 / TCP_CORK / 2.2 -- reproducible bug and test case

[Alexey Kuznetsov <kuznet@xxxxxxxxxxxxx>]

Hello!

> I can't reproduce it on 2.4.18 or .19.  It looked to me like
> tcp_snd_test() and related stuff had been substantially rewritten.

No, tcp_snd_test() in 2.2 is backport and it is accurate.
Apparently, the problem remained in another place, which was not backported.

I think this is tcp_send_fin(). It is obscure and apparently incorrect
at least on corked sockets. I would kill all the dubious "if" after

/* Special case to avoid Nagle bogosity....

and replaced it with straight tcp_push_pending_frames() like it was
made in 2.4. Please, try.

Alexey


BTW your tcpdump contains less than 25% of packets. And all the interesting
piece is absent completely. :-)