On Wed, 2017-11-01 at 14:32 -0700, Eric Dumazet wrote:
> On Wed, Nov 1, 2017 at 2:10 PM, Christoph Paasch <[email protected]> wrote:
> > Since commit 971f10eca186 ("tcp: better TCP_SKB_CB layout to reduce cache
> > line misses") IPCB/IP6CB is no more at the beginning of the tcp_skb_cb.
> > This means that on the output path, we need to make sure that it has
> > been correctly initialized to 0, as is done in tcp_transmit_skb.
> >
> > However, when going through the other code-path in TCP that can send an
> > skb (e.g., through tcp_v6_send_synack), we end up in a situation where
> > IP6CB has some of its fields set to unexpected values. Depending on the
> > layout of tcp_skb_cb across the different kernel-versions this can be
> > lastopt, flags,...
>
> Or not use tcp_init_nondata_skb() on non fast clones, since it adds
> unnecessary writes and clears.
>
> tcp_make_synack() really has no business using tcp_init_nondata_skb()
> and could simply set th->seq = htonl(tcp_rsk(req)->snt_isn);
Something like :
diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
index
69cfdead0cb49e4365158048a0d1a9bbdd55fa83..5502abc5307f0ce1de610d4b70f3a59c4d5383c5
100644
--- a/net/ipv4/tcp_output.c
+++ b/net/ipv4/tcp_output.c
@@ -3399,13 +3399,8 @@ struct sk_buff *tcp_make_synack(const struct sock *sk,
struct dst_entry *dst,
tcp_ecn_make_synack(req, th);
th->source = htons(ireq->ir_num);
th->dest = ireq->ir_rmt_port;
- /* Setting of flags are superfluous here for callers (and ECE is
- * not even correctly set)
- */
- tcp_init_nondata_skb(skb, tcp_rsk(req)->snt_isn,
- TCPHDR_SYN | TCPHDR_ACK);
-
- th->seq = htonl(TCP_SKB_CB(skb)->seq);
+ skb->ip_summed = CHECKSUM_PARTIAL;
+ th->seq = htonl(tcp_rsk(req)->snt_isn);
/* XXX data is queued and acked as is. No buffer/window check */
th->ack_seq = htonl(tcp_rsk(req)->rcv_nxt);
