Feature(socks5): support abstract namespace unix sock

[SuzukiHonoka]

On Linux, if the name starts with an at symbol (‘@') it is read as an abstract namespace socket: the leading ‘@' is replaced with a NUL byte before binding or connecting. For details, see unix(7)

In golang, we can add @ or use 0x0 as prefix to the unix addr for using Linux abstract namespace.
But the current unix processing logic can't not parse it, it only supports the filesystem socket type.
This PR should fix this.

[xjasonlyu]

Thanks for bringing the abstract namespace unix sock feature to the socks5 module!

It overall looks good to me. Just a small question.

[xjasonlyu]

Suggested change

	unix := len(addr) > 0 && addr[0] == '/'
	unix := len(addr) > 0 && (addr[0] == '/' || addr[0] == '@' || addr[0] == 0x00)

Just wondering why not ^?

[SuzukiHonoka]

You pass the url.path from url instance when url.host is empty. If we use the format like socks5:///@test, the address passed to NewSocks will be /@test, which requires additional processing logic. and for len(address)>2, it is only for a more safe in case the out of index error. The approach here checks if the first char after / is @ or null(0x00), and remove the leading /, since it will cause error.

[SuzukiHonoka]

And the reason we can’t avoid using socks5:/// is the string before @ will be parsed as username/password whenever is has valid user:pass or not. The @ char will be omitted by parser.

[xjasonlyu]

Oh I see. that makes sense. But I think such hack adjustment might be better to add here in the parseSocks5 function:

tun2socks/engine/parse.go

Lines 123 to 132 in ec85410

	func parseSocks5(u *url.URL) (proxy.Proxy, error) {
	address, username := u.Host, u.User.Username()
	password, _ := u.User.Password()
	// Socks5 over UDS
	if address == "" {
	address = u.Path
	}
	return proxy.NewSocks5(address, username, password)
	}

And NewSocks5 func should keep its straightforwardness. what do you think?

[SuzukiHonoka]

It makes sense, in this case, the unix check logic may also move to parser function as well. And I will need to make a new arg for unix flag for NewSocks5 , else it will need redundant code to test if the address is unix or not.

[xjasonlyu]

Yeah, it sounds good