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

[Mao Wenan <wenan.mao@linux.alibaba.com>]

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@linux.alibaba.com> 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.


> 
>>
>> Signed-off-by: Mao Wenan <wenan.mao@linux.alibaba.com>
>> ---
>>   v2: fix for ipv6.
>>   net/ipv4/syncookies.c | 4 ++++
>>   net/ipv6/syncookies.c | 5 +++++
>>   2 files changed, 9 insertions(+)
>>
>> diff --git a/net/ipv4/syncookies.c b/net/ipv4/syncookies.c
>> index 6ac473b..57ce317 100644
>> --- a/net/ipv4/syncookies.c
>> +++ b/net/ipv4/syncookies.c
>> @@ -427,6 +427,10 @@ struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb)
>>
>>          /* 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 */
>> +       if (sk->sk_userlocks & SOCK_RCVBUF_LOCK &&
>> +           (req->rsk_window_clamp > tcp_full_space(sk) || req->rsk_window_clamp == 0))
>> +               req->rsk_window_clamp = tcp_full_space(sk);
> 
> This seems not needed to me.
> 
> We call tcp_select_initial_window() with tcp_full_space(sk) passed as
> the 2nd parameter.
> 
> tcp_full_space(sk) will then apply :
> 
> space = min(*window_clamp, space);

if cookie_v4_check pass window_clamp=0 to tcp_select_initial_window, it 
will set window_clamp to max value.
(*window_clamp) = (U16_MAX << TCP_MAX_WSCALE);

but space will fetch from sysctl_rmem_max and sysctl_tcp_rmem[2] which 
is also big value.
space = max_t(u32, space, sock_net(sk)->ipv4.sysctl_tcp_rmem[2]);
space = max_t(u32, space, sysctl_rmem_max);

Then,space = min(*window_clamp, space) is a big value, lead wscale to 7,
is different from tcp_v4_send_synack.


> 
> Please cook a packetdrill test to demonstrate what you are seeing ?
> 
I have real environment and reproduce this case, this patch can fix 
that, i will try to use packetdrill with syn cookies and syn flood happen.