This updates tcp_acceptable_seq(), one of the oldest functions since 2.4.0, by preventing sending out a left-shifted sequence number from a Linux sender in response to a peer's shrunk receive-window caused by losing least significant bits in window-scaling.
RFC7323 page 10 (Chapter 2.4. Addressing Window Retraction) specifies sender side responsibility to handle the sequence number out of window: > On the sender side: > > 3) The initial transmission MUST be within the window announced by > the most recent <ACK>. Some related discussion can be found at the IETF [tcpm] mailing list: https://mailarchive.ietf.org/arch/msg/tcpm/pPO7cYxtky27Qto9b30eaHB_RQI The issue has been reproduced and the patch has been verified by scp a 20GB file from a Linux box using kernel version 4.4.48 to a FreeBSD 11.0 box. [ I mainly want feedback to see if everyone is OK with the approach. ] Cc: "David S. Miller" <da...@davemloft.net> Cc: Alexey Kuznetsov <kuz...@ms2.inr.ac.ru> Cc: James Morris <jmor...@namei.org> Cc: Hideaki YOSHIFUJI <yoshf...@linux-ipv6.org> Cc: Patrick McHardy <ka...@trash.net> Signed-off-by: Cheng Cui <cheng....@netapp.com> --- diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c index 1d5331a..29d736d 100644 --- a/net/ipv4/tcp_output.c +++ b/net/ipv4/tcp_output.c @@ -85,7 +85,8 @@ static void tcp_event_new_data_sent(struct sock *sk, const struct sk_buff *skb) tcp_skb_pcount(skb)); } -/* SND.NXT, if window was not shrunk. +/* SND.NXT, if window was not shrunk or the amount of shrunk was less than one + * window scaling factor due to loss of precision. * If window has been shrunk, what should we make? It is not clear at all. * Using SND.UNA we will fail to open window, SND.NXT is out of window. :-( * Anything in between SND.UNA...SND.UNA+SND.WND also can be already @@ -95,7 +96,8 @@ static inline __u32 tcp_acceptable_seq(const struct sock *sk) { const struct tcp_sock *tp = tcp_sk(sk); - if (!before(tcp_wnd_end(tp), tp->snd_nxt)) + if ((!before(tcp_wnd_end(tp), tp->snd_nxt)) || (tp->rx_opt.wscale_ok && + ((tp->snd_nxt - tcp_wnd_end(tp)) < (1 << tp->rx_opt.rcv_wscale)))) return tp->snd_nxt; else return tcp_wnd_end(tp); -- 2.9.3 (Apple Git-75)
