netdev - [PATCH net-next] tcp: avoid retransmits of TCP packets hanging in host queues

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-ID: <1398041897.19600.26.camel@edumazet-glaptop2.roam.corp.google.com>
Date:	Sun, 20 Apr 2014 17:58:17 -0700
From:	Eric Dumazet <eric.dumazet@...il.com>
To:	David Miller <davem@...emloft.net>
Cc:	Yuchung Cheng <ycheng@...gle.com>,
	Neal Cardwell <ncardwell@...gle.com>,
	netdev <netdev@...r.kernel.org>
Subject: [PATCH net-next] tcp: avoid retransmits of TCP packets hanging in
 host queues

From: Eric Dumazet <edumazet@...gle.com>

In commit 0e280af026a5 ("tcp: introduce TCPSpuriousRtxHostQueues SNMP
counter") we added a logic to detect when a packet was retransmitted
while the prior clone was still in a qdisc or driver queue.

We are now confident we can do better, and catch the problem before
we fragment a TSO packet before retransmit, or in TLP path.

This patch fully exploits the logic by simply canceling the spurious
retransmit.
Original packet is in a queue and will eventually leave the host.

This helps to avoid network collapses when some events make the RTO
estimations very wrong, particularly when dealing with huge number of
sockets with synchronized blast.

Signed-off-by: Eric Dumazet <edumazet@...gle.com>
Signed-off-by: Neal Cardwell <ncardwell@...gle.com>
Signed-off-by: Yuchung Cheng <ycheng@...gle.com>
---
Google-Bug-Id: 8584703

 net/ipv4/tcp_output.c |   34 ++++++++++++++++++++++++++--------
 1 file changed, 26 insertions(+), 8 deletions(-)

diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index 025e25093984..21ad5fd193c9 100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -878,15 +878,8 @@ static int tcp_transmit_skb(struct sock *sk, struct sk_buff *skb, int clone_it,
 	BUG_ON(!skb || !tcp_skb_pcount(skb));
 
 	if (clone_it) {
-		const struct sk_buff *fclone = skb + 1;
-
 		skb_mstamp_get(&skb->skb_mstamp);
 
-		if (unlikely(skb->fclone == SKB_FCLONE_ORIG &&
-			     fclone->fclone == SKB_FCLONE_CLONE))
-			NET_INC_STATS(sock_net(sk),
-				      LINUX_MIB_TCPSPURIOUS_RTX_HOSTQUEUES);
-
 		if (unlikely(skb_cloned(skb)))
 			skb = pskb_copy(skb, gfp_mask);
 		else
@@ -2039,6 +2032,25 @@ bool tcp_schedule_loss_probe(struct sock *sk)
 	return true;
 }
 
+/* Thanks to skb fast clones, we can detect if a prior transmit of
+ * a packet is still in a qdisc or driver queue.
+ * In this case, there is very little point doing a retransmit !
+ * Note: This is called from BH context only.
+ */
+static bool skb_still_in_host_queue(const struct sock *sk,
+				    const struct sk_buff *skb)
+{
+	const struct sk_buff *fclone = skb + 1;
+
+	if (unlikely(skb->fclone == SKB_FCLONE_ORIG &&
+		     fclone->fclone == SKB_FCLONE_CLONE)) {
+		NET_INC_STATS_BH(sock_net(sk),
+				 LINUX_MIB_TCPSPURIOUS_RTX_HOSTQUEUES);
+		return true;
+	}
+	return false;
+}
+
 /* When probe timeout (PTO) fires, send a new segment if one exists, else
  * retransmit the last segment.
  */
@@ -2064,6 +2076,9 @@ void tcp_send_loss_probe(struct sock *sk)
 	if (WARN_ON(!skb))
 		goto rearm_timer;
 
+	if (skb_still_in_host_queue(sk, skb))
+		goto rearm_timer;
+
 	pcount = tcp_skb_pcount(skb);
 	if (WARN_ON(!pcount))
 		goto rearm_timer;
@@ -2385,6 +2400,9 @@ int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
 	    min(sk->sk_wmem_queued + (sk->sk_wmem_queued >> 2), sk->sk_sndbuf))
 		return -EAGAIN;
 
+	if (skb_still_in_host_queue(sk, skb))
+		return -EBUSY;
+
 	if (before(TCP_SKB_CB(skb)->seq, tp->snd_una)) {
 		if (before(TCP_SKB_CB(skb)->end_seq, tp->snd_una))
 			BUG();
@@ -2478,7 +2496,7 @@ int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb)
 		 * see tcp_input.c tcp_sacktag_write_queue().
 		 */
 		TCP_SKB_CB(skb)->ack_seq = tp->snd_nxt;
-	} else {
+	} else if (err != -EBUSY) {
 		NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPRETRANSFAIL);
 	}
 	return err;


--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html