[TCP] FRTO: Clear frto_highmark only after process_frto that uses it
Ilpo Järvinen [Wed, 14 Nov 2007 23:55:09 +0000 (15:55 -0800)]
I broke this in commit 3de96471bd7fb76406e975ef6387abe3a0698149:

[TCP]: Wrap-safed reordering detection FRTO check

tcp_process_frto should always see a valid frto_highmark. An invalid
frto_highmark (zero) is very likely what ultimately caused a seqno
compare in tcp_frto_enter_loss to do the wrong leading to the LOST-bit
leak.

Having LOST-bits integry ensured like done after commit
23aeeec365dcf8bc87fae44c533e50d0bb4f23cc:

[TCP] FRTO: Plug potential LOST-bit leak

won't hurt. It may still be useful in some other, possibly legimate,
scenario.

Reported by Chazarain Guillaume <guichaz@yahoo.fr>.

Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@helsinki.fi>
Signed-off-by: David S. Miller <davem@davemloft.net>

net/ipv4/tcp_input.c

index 3f126ec..0f0c1c9 100644 (file)
@@ -3113,11 +3113,11 @@ static int tcp_ack(struct sock *sk, struct sk_buff *skb, int flag)
        /* See if we can take anything off of the retransmit queue. */
        flag |= tcp_clean_rtx_queue(sk, &seq_rtt, prior_fackets);
 
+       if (tp->frto_counter)
+               frto_cwnd = tcp_process_frto(sk, flag);
        /* Guarantee sacktag reordering detection against wrap-arounds */
        if (before(tp->frto_highmark, tp->snd_una))
                tp->frto_highmark = 0;
-       if (tp->frto_counter)
-               frto_cwnd = tcp_process_frto(sk, flag);
 
        if (tcp_ack_is_dubious(sk, flag)) {
                /* Advance CWND, if state allows this. */