# TCP Fast Open # # Linux: # Both server and client-side TFO support must be enabled in the # kernel, 'sudo sh -c "echo 3 > /proc/sys/net/ipv4/tcp_fastopen"'. # # A packet capture on the loopback interface will show the TFO # option on the SYN, but the fast-output SMTP banner will not # be seen unless you also deliberately emulate a long path: # 'sudo tc qdisc add dev lo root netem delay 50ms' # You'll need iproute-tc installed, for the tc command. # You'll need kernel-modules-extra installed, or you get # an unhelpful error from RTNETLINK. # To tidy up: 'sudo tc qdisc delete dev lo root' # # MacOS: # The kernel seems to have TFO enabled both ways as default. # There is a net.inet.tcp.clear_tfocache parameter ## sysctl -w foo-val # # For network delays there is something called 'Network Link Conditioner' # which might do the job. But how to manipulate it? # # sudo perl system ("tc qdisc add dev lo root netem delay 50ms"); **** # # # Disable the TFO blackhole detection, as we seem to be running foul of it. # If bitten, we see the expected EINPROGRESS for sendto, yet no TFO cookie # option on the SYN. # sudo perl system ("[ -e /proc/sys/net/ipv4/tcp_fastopen_blackhole_timeout_sec ] && echo 0 > /proc/sys/net/ipv4/tcp_fastopen_blackhole_timeout_sec"); **** # # First time runs will see a TFO request option only; subsequent # ones should see the TFO cookie and fast-output SMTP banner # (currently on a separate packet after the server SYN,ACK but before # the client ACK). # # The client log => line should have a "TFO" element. # The server log <= line for a@test.ex should not. # # First clear any previously-obtained cookie: sudo perl system ("ip tcp_metrics delete 127.0.0.1"); **** # # # # FreeBSD: it looks like you have to compile a custom kernel, with # 'options TCP_RFC7413' in the config. Also set # 'net.inet.tcp.fastopen.enabled=1' in /etc/sysctl.conf # Untested. # exim -DSERVER=server -bd -oX PORT_D **** # exim a@test.ex Testing **** sleep 3 # # The server log <= line for b@test.ex should have a "TFO" element, but # this will only be obtained when the above delay is inserted into the # loopback net path. # exim b@test.ex Testing **** sleep 3 # # sudo perl system ("tc qdisc delete dev lo root"); system ("[ -e /proc/sys/net/ipv4/tcp_fastopen_blackhole_timeout_sec ] && echo 3600 > /proc/sys/net/ipv4/tcp_fastopen_blackhole_timeout_sec"); **** # killdaemon no_msglog_check