[PATCH]: Set SOCK_DONE when TCP socket receives FIN

[PATCH]: Set SOCK_DONE when TCP socket receives FIN

[Jamie Lokier]

The SOCK_DONE flag is always clear and never set in 2.5.75.  Yet there
is code which tests it, and if it's clear, will return -ENOTCONN.
Admittedly I am confused as to how this is not noticed :)

This small patch sets it where it looks like it was intended.
Please check.

Enjoy,
-- Jamie

diff -ur orig-2.5.75/net/ipv4/tcp_input.c build-2.5.75/net/ipv4/tcp_input.c
--- orig-2.5.75/net/ipv4/tcp_input.c	2003-07-12 17:57:38.000000000 +0100
+++ build-2.5.75/net/ipv4/tcp_input.c	2003-07-14 02:29:15.000000000 +0100
@@ -2373,7 +2373,7 @@
 	tcp_schedule_ack(tp);
 
 	sk->sk_shutdown |= RCV_SHUTDOWN;
-	sock_reset_flag(sk, SOCK_DONE);
+	sock_set_flag(sk, SOCK_DONE);
 
 	switch (sk->sk_state) {
 		case TCP_SYN_RECV:

Re: [PATCH]: Set SOCK_DONE when TCP socket receives FIN

[David S. Miller]

On Mon, 14 Jul 2003 03:41:00 +0100
Jamie Lokier <jamie@shareable.org> wrote:

> The SOCK_DONE flag is always clear and never set in 2.5.75.  Yet there
> is code which tests it, and if it's clear, will return -ENOTCONN.
> Admittedly I am confused as to how this is not noticed :)
> 
> This small patch sets it where it looks like it was intended.
> Please check.

Indeed, good catch.  This bug got introduced when we changed
all of those volatile char things into a flags bitmask
(ChangeSet 1.889.291.25)

Patch applied, thanks.