[6] in linux-net channel archive
Re: .65-.72 to .73 problems go away
daemon@ATHENA.MIT.EDU (Linus Torvalds)
Mon Dec 19 12:31:59 1994
Date: Mon, 19 Dec 1994 18:14:39 +0200
From: Linus Torvalds <Linus.Torvalds@cs.helsinki.fi>
In-Reply-To: lilo's message as of Dec 19, 9:57
To: lilo <TaRDiS@mail.utexas.edu>, Brian Edmonds <edmonds@cs.ubc.ca>
Cc: linux-kernel@vger.rutgers.edu, linux-net@vger.rutgers.edu
lilo: "Re: .65-.72 to .73 problems go away" (Dec 19, 9:57):
> On Mon, 19 Dec 1994, Brian Edmonds wrote:
>
> > >>>>> On Sun, 18 Dec 1994, lilo <TaRDiS@mail.utexas.edu> said:
> >
> > l> Some of the tcp.c code was completely rewritten between 1.1.72 and
> > l> 1.1.73. I think Alan plans to rewrite more of it.
> >
> > I hope so. As someone else (I forget who) pointed out here, p73 is
> > resulting in a lot of hung sockets for me. This reminds of the old days
> > of 0.99.13 before I added Fred's net2e patches, when I had to reboot
> > every day to clean up the jammed sockets. Say what you will of Fred's
> > work, but it was a real lifesaver for me at the time -- kept me going
> > right up until I upgraded to 1.1.36...
>
> I think we'd probably have seen a patch much more quickly, but for the
> time of year. As I understand it, Alan is out until the beginning of
> the year, and I'm sure we'll see another patch sometime soon
> thereafter.... The jammed sockets are a known problem with 1.1.73....
We'll see what happens, but people seeing problems with 1.1.73 could try
this patch to tcp.c: it fixes a timer problem and a some details with
the TIME_WAIT socket re-use.
Linus
----------
diff -u --recursive --new-file v1.1.73/linux/net/inet/tcp.c linux/net/inet/tcp.c
--- v1.1.73/linux/net/inet/tcp.c Thu Dec 15 11:01:58 1994
+++ linux/net/inet/tcp.c Mon Dec 19 18:04:50 1994
@@ -3291,10 +3291,15 @@
* happens, we must ack the received FIN and
* enter the CLOSING state.
*
- * XXX timeout not set properly
+ * This causes a WRITE timeout, which will either
+ * move on to TIME_WAIT when we timeout, or resend
+ * the FIN properly (maybe we get rid of that annoying
+ * FIN lost hang). The TIME_WRITE code is already correct
+ * for handling this timeout.
*/
- reset_timer(sk, TIME_CLOSE, TCP_TIMEWAIT_LEN);
+ if(sk->timeout != TIME_WRITE)
+ reset_timer(sk, TIME_WRITE, sk->rto);
tcp_set_state(sk,TCP_CLOSING);
break;
case TCP_FIN_WAIT2:
@@ -4320,14 +4325,18 @@
if(sk->debug)
printk("Doing a BSD time wait\n");
tcp_statistics.TcpEstabResets++;
+ sk->rmem_alloc -= skb->mem_len;
+ skb->sk = NULL;
sk->err=ECONNRESET;
tcp_set_state(sk, TCP_CLOSE);
sk->shutdown = SHUTDOWN_MASK;
release_sock(sk);
- sk=get_sock(&tcp_prot, th->source, daddr, th->dest, saddr);
+ sk=get_sock(&tcp_prot, th->dest, saddr, th->source, daddr);
if(sk && sk->state==TCP_LISTEN)
{
sk->inuse=1;
+ skb->sk = sk;
+ sk->rmem_alloc += skb->mem_len;
tcp_conn_request(sk, skb, daddr, saddr,opt, dev,seq+128000);
release_sock(sk);
return 0;
----------
