Re: [PATCH net v2] net: Update window_clamp if SOCK_RCVBUF is set

From: Eric Dumazet
Date: Mon Nov 09 2020 - 06:41:35 EST


Packetdrill test would be :

// Force syncookies
`sysctl -q net.ipv4.tcp_syncookies=2`

0 socket(..., SOCK_STREAM, IPPROTO_TCP) = 3
+0 setsockopt(3, SOL_SOCKET, SO_REUSEADDR, [1], 4) = 0
+0 setsockopt(3, SOL_SOCKET, SO_RCVBUF, [2048], 4) = 0
+0 bind(3, ..., ...) = 0
+0 listen(3, 1) = 0

+0 < S 0:0(0) win 32792 <mss 1000,sackOK,TS val 100 ecr 0,nop,wscale 7>
+0 > S. 0:0(0) ack 1 <mss 1460,sackOK,TS val 4000 ecr 100,nop,wscale 0>
+.1 < . 1:1(0) ack 1 win 1024 <nop,nop,TS val 200 ecr 4000>
+0 accept(3, ..., ...) = 4
+0 %{ assert tcpi_snd_wscale == 0, tcpi_snd_wscale }%

On Mon, Nov 9, 2020 at 12:02 PM Eric Dumazet <edumazet@xxxxxxxxxx> wrote:
>
> On Mon, Nov 9, 2020 at 11:12 AM Mao Wenan <wenan.mao@xxxxxxxxxxxxxxxxx> wrote:
> >
> >
> >
> > 在 2020/11/9 下午5:56, Eric Dumazet 写道:
> > > On Mon, Nov 9, 2020 at 10:33 AM Mao Wenan <wenan.mao@xxxxxxxxxxxxxxxxx> wrote:
> > >>
> > >> When net.ipv4.tcp_syncookies=1 and syn flood is happened,
> > >> cookie_v4_check or cookie_v6_check tries to redo what
> > >> tcp_v4_send_synack or tcp_v6_send_synack did,
> > >> rsk_window_clamp will be changed if SOCK_RCVBUF is set,
> > >> which will make rcv_wscale is different, the client
> > >> still operates with initial window scale and can overshot
> > >> granted window, the client use the initial scale but local
> > >> server use new scale to advertise window value, and session
> > >> work abnormally.
> > >
> > > What is not working exactly ?
> > >
> > > Sending a 'big wscale' should not really matter, unless perhaps there
> > > is a buggy stack at the remote end ?
> > 1)in tcp_v4_send_synack, if SO_RCVBUF is set and
> > tcp_full_space(sk)=65535, pass req->rsk_window_clamp=65535 to
> > tcp_select_initial_window, rcv_wscale will be zero, and send to client,
> > the client consider wscale is 0;
> > 2)when ack is back from client, if there is no this patch,
> > req->rsk_window_clamp is 0, and pass to tcp_select_initial_window,
> > wscale will be 7, this new rcv_wscale is no way to advertise to client.
> > 3)if server send rcv_wind to client with window=63, it consider the real
> > window is 63*2^7=8064, but client consider the server window is only
> > 63*2^0=63, it can't send big packet to server, and the send-q of client
> > is full.
> >
>
> I see, please change your patches so that tcp_full_space() is used _once_
>
> listener sk_rcvbuf can change under us.
>
> I really have no idea how window can be set to 63, so please send us
> the packetdrill test once you have it.