On Mon, 2017-10-16 at 14:29 -0700, David Ahern wrote:
> Running perf in one window to capture tcp_retransmit_skb tracepoint:
> $ perf record -e tcp:tcp_retransmit_skb -a
>
> And causing a retransmission on an active TCP session (e.g., dropping
> packets in the receiver, changing MTU on the interface to 500 and back
> to 1500) triggers a panic:
...
> Check that daddr_cache is non-NULL before de-referencing.
>
> Fixes: e086101b150a ("tcp: add a tracepoint for tcp retransmission")
> Signed-off-by: David Ahern <dsah...@gmail.com>
> ---
> include/trace/events/tcp.h | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/include/trace/events/tcp.h b/include/trace/events/tcp.h
> index 3d1cbd072b7e..13e8ee8af2c8 100644
> --- a/include/trace/events/tcp.h
> +++ b/include/trace/events/tcp.h
> @@ -48,7 +48,8 @@ TRACE_EVENT(tcp_retransmit_skb,
> pin6 = (struct in6_addr *)__entry->saddr_v6;
> *pin6 = np->saddr;
> pin6 = (struct in6_addr *)__entry->daddr_v6;
> - *pin6 = *(np->daddr_cache);
> + if (np->daddr_cache)
> + *pin6 = *(np->daddr_cache);
> } else {
> pin6 = (struct in6_addr *)__entry->saddr_v6;
> ipv6_addr_set_v4mapped(inet->inet_saddr, pin6);
This is weird.
IPV6 TCP uses sk->sk_v6_daddr and sk->->sk_v6_rcv_saddr
So I would rather remove the need to fetch np = inet6_sk(sk) in the
first place, and look at sk->sk_family instead.
No need to get to np at all.
