public inbox for netdev@vger.kernel.org 
 help / color / mirror / Atom feed
From: Willy Tarreau <w@1wt•eu>
To: Eric Dumazet <eric.dumazet@gmail•com>
Cc: netdev@vger•kernel.org
Subject: Re: Stable regression with 'tcp: allow splice() to build full TSO packets'
Date: Thu, 17 May 2012 23:54:37 +0200	[thread overview]
Message-ID: <20120517215437.GQ14498@1wt.eu> (raw)
In-Reply-To: <1337290822.3403.47.camel@edumazet-glaptop>

On Thu, May 17, 2012 at 11:40:22PM +0200, Eric Dumazet wrote:
> > diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
> > index 63ddaee..231fe53 100644
> > --- a/net/ipv4/tcp.c
> > +++ b/net/ipv4/tcp.c
> > @@ -917,8 +917,7 @@ new_segment:
> >  wait_for_sndbuf:
> >  		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
> >  wait_for_memory:
> > -		if (copied)
> > -			tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH);
> > +		tcp_push(sk, flags & ~MSG_MORE, mss_now, TCP_NAGLE_PUSH);
> >  
> >  		if ((err = sk_stream_wait_memory(sk, &timeo)) != 0)
> >  			goto do_error;
> 
> 
> I dont understand why we should tcp_push() if we sent 0 bytes in this
> splice() call.
> 
> The push() should have be done already by prior splice() call, dont you
> think ?
> 
> out:
> 	if (copied && !(flags & MSG_SENDPAGE_NOTLAST))
> 		tcp_push(sk, flags, mss_now, tp->nonagle);
> 

I do think so, but my guess is that something fails below us at some point.
I tracked the tcp_push() chain down to tcp_write_xmit() which I *think* might
fail. I'm not certain about the conditions which can make this happen though.
However what I'm almost sure is that if we fail one tcp_write_xmit() from a
call to tcp_push(), next time we come here in do_tcp_sendpages(), the buffers
won't have moved and this second splice() will immediately go to out with
copied == 0.

Maybe I just worked around the consequence of a deeper issue and something
else needs to enable tcp_push() when we fail, but it's not very clear to me.
I felt like the call to tcp_check_probe_timer() in __tcp_push_pending_frames()
is there for this purpose, but I'm not sure.

Willy

  parent reply	other threads:[~2012-05-17 21:54 UTC|newest]

Thread overview: 61+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2012-05-17 12:18 Stable regression with 'tcp: allow splice() to build full TSO packets' Willy Tarreau
2012-05-17 15:01 ` Willy Tarreau
2012-05-17 15:43   ` Eric Dumazet
2012-05-17 15:56     ` Willy Tarreau
2012-05-17 16:33       ` Eric Dumazet
2012-05-17 16:40         ` Willy Tarreau
2012-05-17 16:47           ` Eric Dumazet
2012-05-17 16:49           ` Eric Dumazet
2012-05-17 17:22             ` Willy Tarreau
2012-05-17 17:34             ` [PATCH net-next] net: netdev_alloc_skb() use build_skb() Eric Dumazet
2012-05-17 17:45               ` Willy Tarreau
2012-06-04 12:39                 ` Michael S. Tsirkin
2012-06-04 12:44                   ` Willy Tarreau
2012-05-17 19:53               ` David Miller
2012-05-18  4:41                 ` Eric Dumazet
2012-06-04 12:37               ` Michael S. Tsirkin
2012-06-04 13:06                 ` Eric Dumazet
2012-06-04 13:41                   ` Michael S. Tsirkin
2012-06-04 14:01                     ` Eric Dumazet
2012-06-04 14:09                       ` Eric Dumazet
2012-06-04 14:17                       ` Michael S. Tsirkin
2012-06-04 15:01                         ` Eric Dumazet
2012-06-04 17:20                           ` Michael S. Tsirkin
2012-06-04 17:44                             ` Eric Dumazet
2012-06-04 18:16                               ` Michael S. Tsirkin
2012-06-04 19:24                                 ` Eric Dumazet
2012-06-04 19:48                                   ` Michael S. Tsirkin
2012-06-04 19:56                                     ` Eric Dumazet
2012-06-04 21:20                                       ` Michael S. Tsirkin
2012-06-05  2:50                                         ` Eric Dumazet
2012-06-04 18:16                           ` Michael S. Tsirkin
2012-06-04 19:29                             ` Eric Dumazet
2012-06-04 19:43                               ` Michael S. Tsirkin
2012-06-04 19:52                                 ` Eric Dumazet
2012-06-04 21:54                                   ` Michael S. Tsirkin
2012-06-05  2:46                                     ` Eric Dumazet
2012-06-04 19:56                                 ` Michael S. Tsirkin
2012-06-04 20:05                                   ` Eric Dumazet
2012-05-17 18:38       ` Stable regression with 'tcp: allow splice() to build full TSO packets' Ben Hutchings
2012-05-17 19:55   ` David Miller
2012-05-17 20:04     ` Willy Tarreau
2012-05-17 20:07       ` David Miller
2012-05-17 20:41 ` Eric Dumazet
2012-05-17 21:14   ` Willy Tarreau
2012-05-17 21:40     ` Eric Dumazet
2012-05-17 21:50       ` Eric Dumazet
2012-05-17 21:57         ` Willy Tarreau
2012-05-17 22:01         ` Eric Dumazet
2012-05-17 22:10           ` Eric Dumazet
2012-05-17 22:16           ` Willy Tarreau
2012-05-17 22:22             ` Eric Dumazet
2012-05-17 22:24               ` Willy Tarreau
2012-05-17 22:25                 ` David Miller
2012-05-17 22:30                   ` Willy Tarreau
2012-05-17 22:35                     ` David Miller
2012-05-17 22:49                       ` Willy Tarreau
2012-05-17 22:27               ` Joe Perches
2012-05-17 21:54       ` Willy Tarreau [this message]
2012-05-17 21:47     ` Willy Tarreau
2012-05-17 22:14     ` Eric Dumazet
2012-05-17 22:29       ` Willy Tarreau

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20120517215437.GQ14498@1wt.eu \
    --to=w@1wt$(echo .)eu \
    --cc=eric.dumazet@gmail$(echo .)com \
    --cc=netdev@vger$(echo .)kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox