A 60MB travel router doesn't need xray — bare Shadowsocks with all-TCP DNS just works
My travel router is a LEDE 17.01 box (60MB RAM, MIPS) living behind the GFW. The "modern" stack — xray with DoH DNS and geoip routing, plus kcptun for UDP acceleration — ate ~45MB of RAM, lived on a USB stick, and kept zombifying after restarts. Meanwhile UDP on Chinese networks got worse every month: kcptun sessions died, UDP DNS relays never worked.
The fix was going backwards: bare shadowsocks-libev 3.2.1, global transparent proxy, DNS over TCP.
Global transparent proxy, no geoip — everything except private ranges rides the tunnel:
iptables -t nat -N SSRULE
for net in 0.0.0.0/8 10.0.0.0/8 127.0.0.0/8 169.254.0.0/16 \
172.16.0.0/12 192.168.0.0/16 224.0.0.0/4 240.0.0.0/4 \
100.64.0.0/10 <server-ip>/32; do
iptables -t nat -A SSRULE -d $net -j RETURN # server-IP entry avoids a redirect loop
done
iptables -t nat -A SSRULE -p tcp -j REDIRECT --to-ports 12345
iptables -t nat -A PREROUTING -i br-lan -p tcp -j SSRULE
iptables -t nat -A OUTPUT -p tcp -j SSRULE
DNS without UDP relay is the clever bit. Unpoisoned DNS has to go through the tunnel, which normally means ss-tunnel's UDP relay — which silently fails on this MIPS box: it logs "listening" but no socket ever appears in netstat. Instead, OpenWrt's tiny dns-forwarder listens on UDP locally and queries upstream over TCP:
dns-forwarder -b 127.0.0.1 -p 1053 -s 8.8.8.8:53
dnsmasq forwards to 127.0.0.1#1053 as usual; the forwarder's TCP:53 packets hit the same REDIRECT rule and ride the tunnel. You get real answers — NXDOMAIN passes through untouched, something poisoned DNS never gives you. No UDP ever leaves the router.
Total cost: 490KB of binaries, embedded in internal flash (verified by unmounting the USB stick mid-run and pulling google in 1.3s), RAM down to 28MB, procd respawns everything if it dies, and boot just works.
Sometimes the modern stack is the problem, not the solution.