nixpkgs/nixos/modules/services/networking/sslh.nix
stuebinm 6afb255d97 nixos: remove all uses of lib.mdDoc
these changes were generated with nixq 0.0.2, by running

  nixq ">> lib.mdDoc[remove] Argument[keep]" --batchmode nixos/**.nix
  nixq ">> mdDoc[remove] Argument[keep]" --batchmode nixos/**.nix
  nixq ">> Inherit >> mdDoc[remove]" --batchmode nixos/**.nix

two mentions of the mdDoc function remain in nixos/, both of which
are inside of comments.

Since lib.mdDoc is already defined as just id, this commit is a no-op as
far as Nix (and the built manual) is concerned.
2024-04-13 10:07:35 -07:00

228 lines
9.1 KiB
Nix

{ config, lib, pkgs, ... }:
with lib;
let
cfg = config.services.sslh;
user = "sslh";
configFormat = pkgs.formats.libconfig {};
configFile = configFormat.generate "sslh.conf" cfg.settings;
in
{
imports = [
(mkRenamedOptionModule [ "services" "sslh" "listenAddress" ] [ "services" "sslh" "listenAddresses" ])
(mkRenamedOptionModule [ "services" "sslh" "timeout" ] [ "services" "sslh" "settings" "timeout" ])
(mkRenamedOptionModule [ "services" "sslh" "transparent" ] [ "services" "sslh" "settings" "transparent" ])
(mkRemovedOptionModule [ "services" "sslh" "appendConfig" ] "Use services.sslh.settings instead")
(mkChangedOptionModule [ "services" "sslh" "verbose" ] [ "services" "sslh" "settings" "verbose-connections" ]
(config: if config.services.sslh.verbose then 1 else 0))
];
meta.buildDocsInSandbox = false;
options.services.sslh = {
enable = mkEnableOption "sslh, protocol demultiplexer";
method = mkOption {
type = types.enum [ "fork" "select" "ev" ];
default = "fork";
description = ''
The method to use for handling connections:
- `fork` forks a new process for each incoming connection. It is
well-tested and very reliable, but incurs the overhead of many
processes.
- `select` uses only one thread, which monitors all connections at once.
It has lower overhead per connection, but if it stops, you'll lose all
connections.
- `ev` is implemented using libev, it's similar to `select` but
scales better to a large number of connections.
'';
};
listenAddresses = mkOption {
type = with types; coercedTo str singleton (listOf str);
default = [ "0.0.0.0" "[::]" ];
description = "Listening addresses or hostnames.";
};
port = mkOption {
type = types.port;
default = 443;
description = "Listening port.";
};
settings = mkOption {
type = types.submodule {
freeformType = configFormat.type;
options.timeout = mkOption {
type = types.ints.unsigned;
default = 2;
description = "Timeout in seconds.";
};
options.transparent = mkOption {
type = types.bool;
default = false;
description = ''
Whether the services behind sslh (Apache, sshd and so on) will see the
external IP and ports as if the external world connected directly to
them.
'';
};
options.verbose-connections = mkOption {
type = types.ints.between 0 4;
default = 0;
description = ''
Where to log connections information. Possible values are:
0. don't log anything
1. write log to stdout
2. write log to syslog
3. write log to both stdout and syslog
4. write to a log file ({option}`sslh.settings.logfile`)
'';
};
options.numeric = mkOption {
type = types.bool;
default = true;
description = ''
Whether to disable reverse DNS lookups, thus keeping IP
address literals in the log.
'';
};
options.protocols = mkOption {
type = types.listOf configFormat.type;
default = [
{ name = "ssh"; host = "localhost"; port = "22"; service= "ssh"; }
{ name = "openvpn"; host = "localhost"; port = "1194"; }
{ name = "xmpp"; host = "localhost"; port = "5222"; }
{ name = "http"; host = "localhost"; port = "80"; }
{ name = "tls"; host = "localhost"; port = "443"; }
{ name = "anyprot"; host = "localhost"; port = "443"; }
];
description = ''
List of protocols sslh will probe for and redirect.
Each protocol entry consists of:
- `name`: name of the probe.
- `service`: libwrap service name (see {manpage}`hosts_access(5)`),
- `host`, `port`: where to connect when this probe succeeds,
- `log_level`: to log incoming connections,
- `transparent`: proxy this protocol transparently,
- etc.
See the documentation for all options, including probe-specific ones.
'';
};
};
description = "sslh configuration. See {manpage}`sslh(8)` for available settings.";
};
};
config = mkMerge [
(mkIf cfg.enable {
systemd.services.sslh = {
description = "Applicative Protocol Multiplexer (e.g. share SSH and HTTPS on the same port)";
after = [ "network.target" ];
wantedBy = [ "multi-user.target" ];
serviceConfig = {
DynamicUser = true;
User = "sslh";
PermissionsStartOnly = true;
Restart = "always";
RestartSec = "1s";
ExecStart = "${pkgs.sslh}/bin/sslh-${cfg.method} -F${configFile}";
KillMode = "process";
AmbientCapabilities = ["CAP_NET_BIND_SERVICE" "CAP_NET_ADMIN" "CAP_SETGID" "CAP_SETUID"];
PrivateTmp = true;
PrivateDevices = true;
ProtectSystem = "full";
ProtectHome = true;
};
};
services.sslh.settings = {
# Settings defined here are not supposed to be changed: doing so will
# break the module, as such you need `lib.mkForce` to override them.
foreground = true;
inetd = false;
listen = map (addr: { host = addr; port = toString cfg.port; }) cfg.listenAddresses;
};
})
# code from https://github.com/yrutschle/sslh#transparent-proxy-support
# the only difference is using iptables mark 0x2 instead of 0x1 to avoid conflicts with nixos/nat module
(mkIf (cfg.enable && cfg.settings.transparent) {
# Set route_localnet = 1 on all interfaces so that ssl can use "localhost" as destination
boot.kernel.sysctl."net.ipv4.conf.default.route_localnet" = 1;
boot.kernel.sysctl."net.ipv4.conf.all.route_localnet" = 1;
systemd.services.sslh = let
iptablesCommands = [
# DROP martian packets as they would have been if route_localnet was zero
# Note: packets not leaving the server aren't affected by this, thus sslh will still work
{ table = "raw"; command = "PREROUTING ! -i lo -d 127.0.0.0/8 -j DROP"; }
{ table = "mangle"; command = "POSTROUTING ! -o lo -s 127.0.0.0/8 -j DROP"; }
# Mark all connections made by ssl for special treatment (here sslh is run as user ${user})
{ table = "nat"; command = "OUTPUT -m owner --uid-owner ${user} -p tcp --tcp-flags FIN,SYN,RST,ACK SYN -j CONNMARK --set-xmark 0x02/0x0f"; }
# Outgoing packets that should go to sslh instead have to be rerouted, so mark them accordingly (copying over the connection mark)
{ table = "mangle"; command = "OUTPUT ! -o lo -p tcp -m connmark --mark 0x02/0x0f -j CONNMARK --restore-mark --mask 0x0f"; }
];
ip6tablesCommands = [
{ table = "raw"; command = "PREROUTING ! -i lo -d ::1/128 -j DROP"; }
{ table = "mangle"; command = "POSTROUTING ! -o lo -s ::1/128 -j DROP"; }
{ table = "nat"; command = "OUTPUT -m owner --uid-owner ${user} -p tcp --tcp-flags FIN,SYN,RST,ACK SYN -j CONNMARK --set-xmark 0x02/0x0f"; }
{ table = "mangle"; command = "OUTPUT ! -o lo -p tcp -m connmark --mark 0x02/0x0f -j CONNMARK --restore-mark --mask 0x0f"; }
];
in {
path = [ pkgs.iptables pkgs.iproute2 pkgs.procps ];
preStart = ''
# Cleanup old iptables entries which might be still there
${concatMapStringsSep "\n" ({table, command}: "while iptables -w -t ${table} -D ${command} 2>/dev/null; do echo; done") iptablesCommands}
${concatMapStringsSep "\n" ({table, command}: "iptables -w -t ${table} -A ${command}" ) iptablesCommands}
# Configure routing for those marked packets
ip rule add fwmark 0x2 lookup 100
ip route add local 0.0.0.0/0 dev lo table 100
'' + optionalString config.networking.enableIPv6 ''
${concatMapStringsSep "\n" ({table, command}: "while ip6tables -w -t ${table} -D ${command} 2>/dev/null; do echo; done") ip6tablesCommands}
${concatMapStringsSep "\n" ({table, command}: "ip6tables -w -t ${table} -A ${command}" ) ip6tablesCommands}
ip -6 rule add fwmark 0x2 lookup 100
ip -6 route add local ::/0 dev lo table 100
'';
postStop = ''
${concatMapStringsSep "\n" ({table, command}: "iptables -w -t ${table} -D ${command}") iptablesCommands}
ip rule del fwmark 0x2 lookup 100
ip route del local 0.0.0.0/0 dev lo table 100
'' + optionalString config.networking.enableIPv6 ''
${concatMapStringsSep "\n" ({table, command}: "ip6tables -w -t ${table} -D ${command}") ip6tablesCommands}
ip -6 rule del fwmark 0x2 lookup 100
ip -6 route del local ::/0 dev lo table 100
'';
};
})
];
}