[2643] in linux-net channel archive
Re: As 2.0 looms
daemon@ATHENA.MIT.EDU (really kuznet@ms2.inr.ac.ru)
Thu Apr 25 09:49:17 1996
From: inr-linux-net@ms2.inr.ac.ru (really kuznet@ms2.inr.ac.ru)
To: linux-net@vger.rutgers.edu
Date: 24 Apr 1996 19:38:40 +0400
Alan Cox (alan@lxorguk.UKuu.ORG.UK) wrote:
: Thirdly:
: A lot of people are having trouble with path mtu discovery if enabled.
: Almost all of these are people using NCSA telnet - its buggy, tough, go fix.
: Two however seem to be seeing a real problem where the Linux box gets a
: failure sequence of using 596 bytes getting an error, should be going to 576
: but tries 596 again ad infinitum. If people are getting hangs talking only
: to some remote sites and their near local links are 576/596 can they dump
: sessions and look to see if this is what is up.
596-576 = 20. Looks like I forgot sizeof(struct iphdr) or
sizeof(struct tcphdr) somewhere. Khm... it is good.
The first bug (please, check it. It's possible, I am wrong now. Cannot
understand, why I subtracted sizeof(struct iphdr) here???)
size = skb->len - (((unsigned char *) th) - skb->data);
#ifndef CONFIG_NO_PATH_MTU_DISCOVERY
#ifdef NO_ANK_FIX
if (size > sk->mtu - sizeof(struct iphdr))
#else
if (size > sk->mtu)
#endif
{
iph->frag_off &= ~htons(IP_DF);
ip_send_check(iph);
}
#endif
th->ack_seq = htonl(sk->acked_seq);
It should not cause a problem.
Apparently, real BUG is in another place. I do not see now, where.
Can someone resend me failed session dumps?
Alexey Kuznetsov.
