Subject: Re: fixing send(2) semantics (kern/29750)
To: Emmanuel Dreyfus <manu@netbsd.org>
From: Thor Lancelot Simon <tls@rek.tjls.com>
List: tech-kern
Date: 03/27/2005 15:59:55
On Sun, Mar 27, 2005 at 10:51:05PM +0200, Emmanuel Dreyfus wrote:
> Jonathan Stone <jonathan@dsg.stanford.edu> wrote:
>
> > So, please stop saying that NetBSD's behaviour is "broken".
>
> All right, so if everyone agree there is no bug to fix, then let's close
> the PR and forget about it. The "oddity" (am I allowed to call it an
> oddity?) may deserve an explanation in the man page, though.
I still don't think that ENOBUFS should be returned. I think that if we
run out of space the interface queue packets for non-reliable protocols
should just be silently dropped. I'm not really able to read the manual
page or the quoted parts of the specification as allowing ENOBUFS in this
case but I think blocking is also wrong.
I still believe it's possible for your application to get what it wants
by using a low-water mark on its socket buffer, nonblocking I/O, and
*possibly* an increase in net.inet.ip.ifq.maxlen with sysctl.
Thor