1. Handling of TFO_SERVER_COOKIE_NOT_CHKED flag was removed back in 2014, but this flag is still mentioned in the documentation:
Documentation/networking/ip-sysctl.txt:
0x100: Accept SYN data w/o validating the cookie.
2. There is no explanation how TFO_SERVER_WO_SOCKOPT1 and TFO_SERVER_WO_SOCKOPT2 differ, the docs only say:
Documentation/networking/ip-sysctl.txt:
0x400/0x800: Enable Fast Open on all listeners regardless of the
TCP_FASTOPEN socket option. The two different flags designate two
different ways of setting max_qlen without the TCP_FASTOPEN socket
option.
3. When TFO_SERVER_WO_SOCKOPT2 flag is set, the fastopenq.max_qlen is set to the value of sysctl bitmap containing flags (sysctl_tcp_fastopen), what is (at least for me) completely irrational and I believe is a bug:
net/ipv4/af_inet.c:
225 else if ((sysctl_tcp_fastopen &
226 TFO_SERVER_WO_SOCKOPT2) != 0)
227 fastopen_queue_tune(sk,
228 ((uint)sysctl_tcp_fastopen) >> 16);
On Sat, Aug 13, 2016 at 4:05 PM, Piotr Jurkiewicz
[off-list ref] wrote:
1. Handling of TFO_SERVER_COOKIE_NOT_CHKED flag was removed back in 2014,
but this flag is still mentioned in the documentation:
Documentation/networking/ip-sysctl.txt:
0x100: Accept SYN data w/o validating the cookie.
2. There is no explanation how TFO_SERVER_WO_SOCKOPT1 and
TFO_SERVER_WO_SOCKOPT2 differ, the docs only say:
Documentation/networking/ip-sysctl.txt:
0x400/0x800: Enable Fast Open on all listeners regardless of the
TCP_FASTOPEN socket option. The two different flags designate two
different ways of setting max_qlen without the TCP_FASTOPEN
socket
option.
3. When TFO_SERVER_WO_SOCKOPT2 flag is set, the fastopenq.max_qlen is set to
the value of sysctl bitmap containing flags (sysctl_tcp_fastopen), what is
(at least for me) completely irrational and I believe is a bug:
net/ipv4/af_inet.c:
225 else if ((sysctl_tcp_fastopen &
226 TFO_SERVER_WO_SOCKOPT2) != 0)
227 fastopen_queue_tune(sk,
228 ((uint)sysctl_tcp_fastopen) >> 16);
Thanks for pointing these issues out. The document indeed needs work.
TFO_SERVERS_WO_SOCKEOPT2 was used for debugging purposes that we
forget to clean up before upstream. Here is my plan:
1) send a patch to remove this opaque/buggy internal debugging option
2) update the ip-sysctl, perhaps update tcp option man pages
How does that sound?