8.0 KiB

Raw Permalink Blame History

Skip to recipes for quick setup instructions

components

ss-local provides SOCKS5 proxy with UDP associate support.

 socks5                                     ss              plain
--------> tcp:local_address:local_port ----> ss server -------> dest

ss-redir. The REDIRECT and TPROXY part are to be provided by ss-rules script. REDIRECT is for tcp traffic (SO_ORIGINAL_DST only supports TCP). TPROXY is for udp messages, but it's only available in the PREROUTING chain and as such cannot proxy local out traffic.

  plain             plain                                 ss              plain
---------> REDIRECT ------> tcp:local_address:local_port ----> ss server -----> original dest

  plain            plain                                 ss              plain
---------> TPROXY -------> udp:local_address:local_port -----> ss server -----> original dest

ss-tunnel provides ssh -L local-forwarding-like tunnel. Typically it's used to tunnel DNS traffic to the remote.

  plain                                       ss               plain
---------> tcp|udp:local_address:local_port ------> ss server -------> tunnel_address

ss-server, the "ss server" in the above diagram

uci

Option names are the same as those used in json config files. Check validate_xxx func definition of the service script and shadowsocks-libev's own documentation for supported options and expected value types. A sample config file is also provided for reference.

Every section have a disabled option to temporarily turn off the component instance or component instances referring to it.

Section type server is for definition of remote shadowsocks servers. They will be referred to from other component sections and as such should be named (as compared to anonymous section).

Section type ss_local, ss_redir, ss_tunnel are for specification of shadowsocks-libev components. They share mostly a common set of options like local_port, verbose, fast_open, timeout, etc.

Plugin options should be specified in server section and will be inherited by other compoenents referring to it.

We can have multiple instances of component and server sections. The relationship between them is many-to-one. This will have the following implications

It's possible to have both ss_local and ss_redir referring to the same server definition
It's possible to have multiple instances of ss_redir listening on the same address:port with reuse_port enabled referring to the same or different server sections

ss_rules section is for configuring the behaviour of ss-rules script. There can only exist at most one such section with the name also being ss_rules

redir_tcp		name of ss_redir section with mode tcp_only or tcp_and_udp
redir_udp		name of ss_redir section with mode udp_only or tcp_and_udp
ifnames			only apply rules on packets from these ifnames

--- for incoming packets having source address in

src_ips_bypass		will bypass the redir chain
src_ips_forward		will always go through the redir chain
src_ips_checkdst	will continue to have their destination addresses checked

--- otherwise, the default action can be specified with

src_default		bypass, forward, [checkdst]

--- if the previous check result is checkdst,
--- then packets having destination address in

dst_ips_bypass_file
dst_ips_bypass		will bypass the redir chain
dst_ips_forward_file
dst_ips_forward		will go through the redir chain

--- otherwise, the default action can be specified with

dst_default		[bypass], forward

--- for local out tcp packets, the default action can be specified with

local_default		[bypass], forward, checkdst

ss-rules now uses nft set for storing addresses/networks. Those set names are also part of the API and can be populated by other programs, e.g. dnsmasq with builtin nft set support. Note that while nftables set supports storing cidr networks when interval flag is on, it rejects elements with overlaping intervals.

Extra nftables expressions can be specified with nft_tcp_extra and nft_udp_extra to apply ss_rules only to selected tcp/udp traffics. E.g. tcp dport { 80, 443 }, udp dport 53, etc.

incompatible changes

Commit date	Commit ID	Subject	Comment
2022-03-01	`fdaf2de2a`	shadowsocks-libev: ss-rules: convert to using nft	ss-rules now uses nftables. UCI option ipt_args and dst_forward_recentrst are now deprecated and removed
2020-08-03	`7d7cbae75`	shadowsocks-libev: support ss-server option local_address_{v4,v6}	ss_server bind_address now deprecated, use local_address
2019-05-09	`afe7d3424`	shadowsocks-libev: move plugin options to server section	This is a revision against `c19e949` committed 2019-05-06
2017-07-02	`b61af9703`	shadowsocks-libev: rewrite	Packaging of shadowsocks-libev was rewritten from scratch

notes and faq

Useful paths and commands for debugging

# check current running status
ubus call service list '{"name": "shadowsocks-libev"}'
ubus call service list '{"name": "shadowsocks-libev", "verbose": true}'

# dump validate definition
ubus call service validate '{"package": "shadowsocks-libev"}'
ubus call service validate '{"package": "shadowsocks-libev"}' \
	| jsonfilter -e '$["shadowsocks-libev"]["ss_tunnel"]'

# check json config
ls -l /var/etc/shadowsocks-libev/

# set uci config option verbose to 1, restart the service and follow the log
logread -f

ss-redir needs to open a new socket and setsockopt IP_TRANSPARENT when sending udp reply to client. This requires CAP_NET_ADMIN and as such the process cannot run as nobody

ss-local, ss-redir, etc. supports specifying an array of remote ss server, but supporting this in uci seems to be overkill. The workaround can be defining multiple server sections and multiple ss-redir instances with reuse_port enabled

recipes

forward all

This will setup firewall rules to forward almost all incoming tcp/udp and locally generated tcp traffic (excluding those to private addresses like 192.168.0.0/16 etc.) through remote shadowsocks server

Install components. Retry each command till it succeed

opkg install shadowsocks-libev-ss-redir
opkg install shadowsocks-libev-ss-rules
opkg install shadowsocks-libev-ss-tunnel

Edit uci config /etc/config/shadowsocks-libev. Replace config server 'sss0' section with parameters of your own remote shadowsocks server. As for other options, change them only when you know the effect.

config server 'sss0'
	option disabled 0
	option server '_sss_addr_'
	option server_port '_sss_port_'
	option password '********'
	option method 'aes-256-cfb'

config ss_tunnel
	option disabled 0
	option server 'sss0'
	option local_address '0.0.0.0'
	option local_port '8053'
	option tunnel_address '8.8.8.8:53'
	option mode 'tcp_and_udp'

config ss_redir ssr0
	option disabled 0
	option server 'sss0'
	option local_address '0.0.0.0'
	option local_port '1100'
	option mode 'tcp_and_udp'
	option reuse_port 1

config ss_rules 'ss_rules'
	option disabled 0
	option redir_tcp 'ssr0'
	option redir_udp 'ssr0'
	option src_default 'checkdst'
	option dst_default 'forward'
	option local_default 'forward'

Restart shadowsocks-libev components

/etc/init.d/shadowsocks-libev restart

Check if things are in place

nft list ruleset | sed -r -n '/^\t[a-z]+ ss_rules[^ ]+ \{/,/^\t\}/p'
netstat -lntp | grep -E '8053|1100'
ps ww | grep ss-

Edit /etc/config/dhcp, making sure options are present in the first dnsmasq section like the following to let it use local tunnel endpoint for upstream dns query. Option noresolv instructs dnsmasq to not use other dns servers like advertised by local isp. Option localuse intends to make sure the device you are configuring also uses this dnsmasq instance as the resolver, not the ones from other sources.

config dnsmasq
	...
	list server '127.0.0.1#8053'
	option noresolv 1
	option localuse 1

Restart dnsmasq

/etc/init.d/dnsmasq restart

Check network on your computer

nslookup www.google.com
curl -vv https://www.google.com

8.0 KiB Raw Permalink Blame History