tcp: skip DSACKs with dubious sequence ranges

Currently, we use length of DSACKed range to compute number of
delivered packets. And if sequence range in DSACK is corrupted,
we can get bogus dsacked/acked count, and bogus cwnd.

This patch put bounds on DSACKed range to skip update of data
delivery and spurious retransmission information, if the DSACK
is unlikely caused by sender's action:
- DSACKed range shouldn't be greater than maximum advertised rwnd.
- Total no. of DSACKed segments shouldn't be greater than total
  no. of retransmitted segs. Unlike spurious retransmits, network
  duplicates or corrupted DSACKs shouldn't be counted as delivery.

Signed-off-by: Priyaranjan Jha <priyarjha@google.com>
Signed-off-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/include/uapi/linux/snmp.h b/include/uapi/linux/snmp.h
index cee9f8e..f84e7bc 100644 (file)
--- a/include/uapi/linux/snmp.h
+++ b/include/uapi/linux/snmp.h
@@ -288,6 +288,7 @@ enum
        LINUX_MIB_TCPTIMEOUTREHASH,             /* TCPTimeoutRehash */
        LINUX_MIB_TCPDUPLICATEDATAREHASH,       /* TCPDuplicateDataRehash */
        LINUX_MIB_TCPDSACKRECVSEGS,             /* TCPDSACKRecvSegs */
+       LINUX_MIB_TCPDSACKIGNOREDDUBIOUS,       /* TCPDSACKIgnoredDubious */
        __LINUX_MIB_MAX
 };
 

diff --git a/net/ipv4/proc.c b/net/ipv4/proc.c
index 1074df7..8d5e169 100644 (file)
--- a/net/ipv4/proc.c
+++ b/net/ipv4/proc.c
@@ -293,6 +293,7 @@ static const struct snmp_mib snmp4_net_list[] = {
        SNMP_MIB_ITEM("TcpTimeoutRehash", LINUX_MIB_TCPTIMEOUTREHASH),
        SNMP_MIB_ITEM("TcpDuplicateDataRehash", LINUX_MIB_TCPDUPLICATEDATAREHASH),
        SNMP_MIB_ITEM("TCPDSACKRecvSegs", LINUX_MIB_TCPDSACKRECVSEGS),
+       SNMP_MIB_ITEM("TCPDSACKIgnoredDubious", LINUX_MIB_TCPDSACKIGNOREDDUBIOUS),
        SNMP_MIB_SENTINEL
 };
 

diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c
index 184ea55..b1ce205 100644 (file)
--- a/net/ipv4/tcp_input.c
+++ b/net/ipv4/tcp_input.c
@@ -885,21 +885,34 @@ struct tcp_sacktag_state {
        struct rate_sample *rate;
 };
 
-/* Take a notice that peer is sending D-SACKs */
+/* Take a notice that peer is sending D-SACKs. Skip update of data delivery
+ * and spurious retransmission information if this DSACK is unlikely caused by
+ * sender's action:
+ * - DSACKed sequence range is larger than maximum receiver's window.
+ * - Total no. of DSACKed segments exceed the total no. of retransmitted segs.
+ */
 static u32 tcp_dsack_seen(struct tcp_sock *tp, u32 start_seq,
                          u32 end_seq, struct tcp_sacktag_state *state)
 {
        u32 seq_len, dup_segs = 1;
 
-       if (before(start_seq, end_seq)) {
-               seq_len = end_seq - start_seq;
-               if (seq_len > tp->mss_cache)
-                       dup_segs = DIV_ROUND_UP(seq_len, tp->mss_cache);
-       }
+       if (!before(start_seq, end_seq))
+               return 0;
+
+       seq_len = end_seq - start_seq;
+       /* Dubious DSACK: DSACKed range greater than maximum advertised rwnd */
+       if (seq_len > tp->max_window)
+               return 0;
+       if (seq_len > tp->mss_cache)
+               dup_segs = DIV_ROUND_UP(seq_len, tp->mss_cache);
+
+       tp->dsack_dups += dup_segs;
+       /* Skip the DSACK if dup segs weren't retransmitted by sender */
+       if (tp->dsack_dups > tp->total_retrans)
+               return 0;
 
        tp->rx_opt.sack_ok |= TCP_DSACK_SEEN;
        tp->rack.dsack_seen = 1;
-       tp->dsack_dups += dup_segs;
 
        state->flag |= FLAG_DSACKING_ACK;
        /* A spurious retransmission is delivered */
@@ -1153,6 +1166,11 @@ static bool tcp_check_dsack(struct sock *sk, const struct sk_buff *ack_skb,
        }
 
        dup_segs = tcp_dsack_seen(tp, start_seq_0, end_seq_0, state);
+       if (!dup_segs) {        /* Skip dubious DSACK */
+               NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPDSACKIGNOREDDUBIOUS);
+               return false;
+       }
+
        NET_ADD_STATS(sock_net(sk), LINUX_MIB_TCPDSACKRECVSEGS, dup_segs);
 
        /* D-SACK for already forgotten data... Do dumb counting. */