When using syscookies initial receive window calculation uses wrong MSS
Subsystems (3)
PATCH 1/1 When using syscookies initial receive window calculation uses wrong MSS
When using syncookies we get the mss variable from the two bits encoded in the ACK number. This MSS though, represents the client/sender MSS.
When calculating initial receive window with tcp_select_initial_window, we feed it MSS so that the receive buffer can be MSS aligned. But this MSS is supposed to be our receiver/server advertised MSS, not MSS from the remote party. The code got it wrong.
Usually it's not a big deal but could yield weird/wrong rcv_ssthresh initial values. Consider this:
$ sysctl net.ipv4.tcp_syncookies=2 $ ip route change local 127.0.0.1/32 dev lo initrwnd 1 $ ip -6 route change local ::1 dev lo initrwnd 1
Flags [S], seq #, win 32767, options [mss 65495,nop,nop,sackOK,nop,wscale 10], length 0 Flags [S.], seq #, ack #, win 128, options [mss 128], length 0
Flags [.], ack 1, win 32767, length 0
Quoted 1 line
client_tcpi_snd_wnd:128 server_tcpi_rcv_ssthresh:128
Flags [P.], seq 1:2, ack 1, win 536, length 1
Flags [.], ack 2, win 32767, length 0
Quoted 1 line
client_snd_wnd:536 server_rcv_ssthresh:128
After the handshake we see the client reporting send window=128 and this is visible as tcpi_snd_wnd on client and rcv_ssthresh on server. Then, after 1 byte exchange, to get an ACK from server, we see the window to surprisingly grow to 536 bytes. This is surprising, since the server window should not be related to syncookie-derived MSS from the client. What is even more surprising, the window is 536 but the rcv_ssthresh on server is still 128.
Flags [S], seq #, win 32767, options [mss 65476,nop,nop,sackOK,nop,wscale 10], length 0 Flags [S.], seq #, ack #, win 128, options [mss 128], length 0
Flags [.], ack 1, win 32767, length 0
Quoted 1 line
client_snd_wnd:128 server_rcv_ssthresh:128
Flags [P.], seq 1:2, ack 1, win 1220, length 1
Flags [.], ack 2, win 32767, length 0
Quoted 1 line
client_tcpi_snd_wnd:1220 server_tcpi_rcv_ssthresh:128
Identical situation for IPv6.
After the patch we get more sensible values, and sender window is kept in sync with rcv_ssthresh:
Flags [S], seq #, win 65495, options [mss 65495,nop,nop,sackOK,nop,wscale 4], length 0 Flags [S.], seq #, ack #, win 128, options [mss 128], length 0
Flags [.], ack 1, win 65495, length 0
Quoted 1 line
client_snd_wnd:128 server_rcv_ssthresh:128
Flags [P.], seq 1:2, ack 1, win 128, length 1
Flags [.], ack 2, win 65494, length 0
Quoted 1 line
client_snd_wnd:128 server_rcv_ssthresh:128
Flags [S], seq #, win 65476, options [mss 65476,nop,nop,sackOK,nop,wscale 4], length 0 Flags [S.], seq #, ack #, win 128, options [mss 128], length 0
Flags [.], ack 1, win 65476, length 0
Quoted 1 line
client_snd_wnd:128 server_rcv_ssthresh:128
Flags [P.], seq 1:2, ack 1, win 128, length 1
Flags [.], ack 2, win 65475, length 0
Quoted 1 line
client_snd_wnd:128 server_rcv_ssthresh:128
--- net/ipv4/syncookies.c | 7 ++++++- net/ipv6/syncookies.c | 7 ++++++- 2 files changed, 12 insertions(+), 2 deletions(-)
net/ipv4/syncookies.c +6 -1
@@ -338,6 +338,8 @@ struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb)struct rtable *rt;__u8 rcv_wscale;struct flowi4 fl4; +u32 adj_mss; +u32 advmss;u32 tsoff =0 ; if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies) ||
@@ -430,6 +432,9 @@ struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb) goto out; } + advmss =tcp_mss_clamp (tp,dst_metric_advmss (&rt->dst)); + adj_mss = advmss - (ireq->tstamp_ok ?TCPOLEN_TSTAMP_ALIGNED :0 ); +/* Try to redo what tcp_v4_send_synack did. */ req->rsk_window_clamp = tp->window_clamp ? :dst_metric (&rt->dst,RTAX_WINDOW );/* limit the window selection if the user enforce a smaller rx buffer */
@@ -438,7 +443,7 @@ struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb) (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0)) req->rsk_window_clamp = full_space; -tcp_select_initial_window (sk, full_space, req->mss, +tcp_select_initial_window (sk, full_space, adj_mss, &req->rsk_rcv_wnd, &req->rsk_window_clamp, ireq->wscale_ok, &rcv_wscale,dst_metric (&rt->dst,RTAX_INITRWND ));
net/ipv6/syncookies.c +6 -1
@@ -139,6 +139,8 @@ struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb)int full_space, mss;struct dst_entry *dst;__u8 rcv_wscale; +u32 adj_mss; +u32 advmss;u32 tsoff =0 ; if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies) ||
@@ -249,7 +251,10 @@ struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb) (req->rsk_window_clamp > full_space || req->rsk_window_clamp == 0)) req->rsk_window_clamp = full_space; -tcp_select_initial_window (sk, full_space, req->mss, + advmss =tcp_mss_clamp (tp,dst_metric_advmss (dst)); + adj_mss = advmss - (ireq->tstamp_ok ?TCPOLEN_TSTAMP_ALIGNED :0 ); + +tcp_select_initial_window (sk, full_space, adj_mss, &req->rsk_rcv_wnd, &req->rsk_window_clamp, ireq->wscale_ok, &rcv_wscale,dst_metric (dst,RTAX_INITRWND ));
Signature · 1 line
2.25.1
Jakub Kicinski
maintainerQuoted 2 lines · On Fri, 29 Jul 2022 18:19:32 +0200 Marek Majkowski wrote:
Usually it's not a big deal but could yield weird/wrong
rcv_ssthreshinitial values.
Feels like we should either make it a proper fix or fold into the RTAX_INITRWND series?