On Wed, 23 Jan 2008, Ilpo Järvinen wrote:
> On Wed, 23 Jan 2008, Krishna Kumar2 wrote:
>
> > Hi Ilpo,
> >
> > > It's almost impossible to know which of these is the main cause and the
> > > first occuring due to reasons I'll not copy here. What a strange thing
> > > that it has been super quiet on this front until now everybody is seeing
> > > it, could there be something unrelated to TCP which has broken it all
> > > recently?
> >
> > I have been getting this for atleast 3 weeks but I was quiet since those
> > were kernels that I had modified.
>
> Since you can easily reproduce it, lets just figure out what's causing it
> hard way, rather than digging the endless git-logs... :-)
Hmm, perhaps it could be something related to this (and some untested
path somewhere which is now exposed):
commit 4a55b553f691abadaa63570dfc714e20913561c1
Author: Ilpo Järvinen <[EMAIL PROTECTED]>
Date: Thu Dec 20 20:36:03 2007 -0800
[TCP]: Fix TSO deferring
Dave, what do you think? Wouldn't explain the one -rc only report though
from Denys. Another one I'm a bit unsure of is this:
commit 757c32944b80fd95542bd66f06032ab773034d53
Author: Ilpo Järvinen <[EMAIL PROTECTED]>
Date: Thu Jan 3 20:39:01 2008 -0800
[TCP]: Perform setting of common control fields in one place
->sacked field is cleared in tcp_retransmit_skb due to a subtle change,
which might be buggy.... However, I find it rather unlikely that this
would explain Kumar's case. Anyway, here's the one that reverts the
problematic part of it. ...and this is net-2.6.25 as well so it won't
solve Denys' case either.
--
i.
--
[PATCH] [TCP]: Revert part of "[TCP]: Perform setting of common control..."
This commit reverts part of 757c32944b80fd95542bd66f06032ab773034d53
([TCP]: Perform setting of common control fields in one place)
because it's not valid to clear ->sacked field like that without
additional precautions with counters, mostly lost_out, sacked_out
should not be set due to reneging which kicks in much earlier than
this.
Signed-off-by: Ilpo Järvinen <[EMAIL PROTECTED]>
---
net/ipv4/tcp_output.c | 8 +++++---
1 files changed, 5 insertions(+), 3 deletions(-)
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 648340f..f6cbc1f 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -1887,10 +1887,12 @@ int tcp_retransmit_skb(struct sock *sk, struct sk_buff
*skb)
(TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN) &&
tp->snd_una == (TCP_SKB_CB(skb)->end_seq - 1)) {
if (!pskb_trim(skb, 0)) {
- /* Reuse, even though it does some unnecessary work */
- tcp_init_nondata_skb(skb, TCP_SKB_CB(skb)->end_seq - 1,
- TCP_SKB_CB(skb)->flags);
+ TCP_SKB_CB(skb)->seq = TCP_SKB_CB(skb)->end_seq - 1;
+ skb_shinfo(skb)->gso_segs = 1;
+ skb_shinfo(skb)->gso_size = 0;
+ skb_shinfo(skb)->gso_type = 0;
skb->ip_summed = CHECKSUM_NONE;
+ skb->csum = 0;
}
}
--
1.5.2.2
