Re: [TCP] Probable bug in icsk_probes_out handling ?

[Eric Dumazet <dada1@cosmosbay.com>]

David Miller a écrit :
> From: Eric Dumazet <dada1@cosmosbay.com>
> Date: Wed, 23 Jul 2008 15:35:19 +0200
> 
>> Apparently, we can in some situations reset TCP connections in a couple of seconds when some frames are lost.
> 
> I think it is some optimization that went in between 2.2.x and 2.4.x
> which took things a little bit too far.
> 
> Please try this patch:

Yes, this patch solves the problem. 
I tried a similar patch yesterday but was not sure about its validity.
Thanks a lot David.
 
> 
> diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
> index 1f5e604..75efd24 100644
> --- a/net/ipv4/tcp_input.c
> +++ b/net/ipv4/tcp_input.c
> @@ -3292,6 +3292,7 @@ static int tcp_ack(struct sock *sk, struct sk_buff *skb, int flag)
>  	 * log. Something worked...
>  	 */
>  	sk->sk_err_soft = 0;
> +	icsk->icsk_probes_out = 0;
>  	tp->rcv_tstamp = tcp_time_stamp;
>  	prior_packets = tp->packets_out;
>  	if (!prior_packets)
> @@ -3324,8 +3325,6 @@ static int tcp_ack(struct sock *sk, struct sk_buff *skb, int flag)
>  	return 1;
>  
>  no_queue:
> -	icsk->icsk_probes_out = 0;
> -
>  	/* If this ack opens up a zero window, clear backoff.  It was
>  	 * being used to time the probes, and is probably far higher than
>  	 * it needs to be for normal retransmission.