Re: [PATCH 2/11] [NET] Convert sk_zapped into SOCK_ZAPPED flag

[Thomas Graf <tgraf@xxxxxxx>]

* Patrick McHardy <422F571E.2020708@xxxxxxxxx> 2005-03-09 21:05
> Thomas Graf wrote:
> >* Patrick McHardy <422F5461.4080008@xxxxxxxxx> 2005-03-09 20:54
> >>
> >>What about volatile ? sock_set_flag() uses __set_bit(), so its not
> >>the same.
> >
> >
> >I thought about this for a while but couldn't find a reason
> >why it shouldn't work. Actually I don't even see any reason for
> >having sk_zapped be volatile.
> 
> You're probably right. I believe this piece of code from 2.4 is the
> reason for it beeing volatile:
> 
> #ifdef TCP_DEBUG
>         if (sk->zapped) {
>                 printk(KERN_DEBUG "TCP: double destroy sk=%p\n", sk);
>                 sock_hold(sk);
>         }
>         sk->zapped = 1;
> #endif

Yes, this makes sense. I haven't spotted any places in 2.6 where
any of the flags I've converted would suffer from a reordering.
Since all the flags have been chars the missing atomicy shouldn't
be an issue either.