fix(gateway): Remove /dev/net/tun requirement and clean up upgrade script (#3392)

* Clean up gateway upgrade script
* Fixes #3226 to remove another place where things can go wrong when
upgrading gateways

docker-compose.yml

@@ -128,8 +128,6 @@ services:
       - NET_ADMIN
     sysctls:
       - net.ipv6.conf.all.disable_ipv6=0
-    devices:
-      - "/dev/net/tun:/dev/net/tun"
     depends_on:
       gateway:
         condition: "service_healthy"
@@ -169,8 +167,6 @@ services:
       - net.ipv6.conf.all.disable_ipv6=0
       - net.ipv6.conf.all.forwarding=1
       - net.ipv6.conf.default.forwarding=1
-    devices:
-      - "/dev/net/tun:/dev/net/tun"
     depends_on:
       api:
         condition: "service_healthy"

elixir/apps/web/lib/web/live/relay_groups/new_token.ex

@@ -265,7 +265,6 @@ defmodule Web.RelayGroups.NewToken do
       "--sysctl net.ipv6.conf.all.disable_ipv6=0",
       "--sysctl net.ipv6.conf.all.forwarding=1",
       "--sysctl net.ipv6.conf.default.forwarding=1",
-      "--device=\"/dev/net/tun:/dev/net/tun\"",
       Enum.map(env, fn {key, value} -> "--env #{key}=\"#{value}\"" end),
       "--env FIREZONE_NAME=$(hostname)",
       "#{Domain.Config.fetch_env!(:domain, :docker_registry)}/relay:#{major_minor_version()}"

elixir/apps/web/lib/web/live/sites/new_token.ex

@@ -180,7 +180,6 @@ defmodule Web.Sites.NewToken do
       "--sysctl net.ipv6.conf.all.disable_ipv6=0",
       "--sysctl net.ipv6.conf.all.forwarding=1",
       "--sysctl net.ipv6.conf.default.forwarding=1",
-      "--device=\"/dev/net/tun:/dev/net/tun\"",
       Enum.map(env ++ [{"FIREZONE_ENABLE_MASQUERADE", "1"}], fn {key, value} ->
         "--env #{key}=\"#{value}\""
       end),

rust/connlib/tunnel/src/device_channel/tun_linux.rs

@@ -5,16 +5,21 @@ use futures_util::future::BoxFuture;
 use futures_util::FutureExt;
 use ip_network::IpNetwork;
 use libc::{
-    close, fcntl, open, F_GETFL, F_SETFL, IFF_MULTI_QUEUE, IFF_NO_PI, IFF_TUN, O_NONBLOCK, O_RDWR,
+    close, fcntl, makedev, mknod, open, F_GETFL, F_SETFL, IFF_MULTI_QUEUE, IFF_NO_PI, IFF_TUN,
+    O_NONBLOCK, O_RDWR, S_IFCHR,
 };
 use netlink_packet_route::RT_SCOPE_UNIVERSE;
 use parking_lot::Mutex;
 use rtnetlink::{new_connection, Error::NetlinkError, Handle};
 use std::net::IpAddr;
+use std::path::Path;
 use std::task::{Context, Poll};
 use std::{
-    fmt, io,
-    os::fd::{AsRawFd, RawFd},
+    fmt, fs, io,
+    os::{
+        fd::{AsRawFd, RawFd},
+        unix::fs::PermissionsExt,
+    },
 };
 use tokio::io::unix::AsyncFd;
 
@@ -25,6 +30,8 @@ pub(crate) const SIOCGIFMTU: libc::c_ulong = libc::SIOCGIFMTU;
 const IFACE_NAME: &str = "tun-firezone";
 const TUNSETIFF: libc::c_ulong = 0x4004_54ca;
 const TUN_FILE: &[u8] = b"/dev/net/tun\0";
+const TUN_DEV_MAJOR: u32 = 10;
+const TUN_DEV_MINOR: u32 = 200;
 const RT_PROT_STATIC: u8 = 4;
 const DEFAULT_MTU: u32 = 1280;
 const FILE_ALREADY_EXISTS: i32 = -17;
@@ -82,6 +89,8 @@ impl Tun {
     }
 
     pub fn new(config: &InterfaceConfig, _: Vec<IpAddr>, _: &impl Callbacks) -> Result<Self> {
+        create_tun_device()?;
+
         let fd = match unsafe { open(TUN_FILE.as_ptr() as _, O_RDWR) } {
             -1 => return Err(get_last_error()),
             fd => fd,
@@ -232,6 +241,31 @@ fn set_non_blocking(fd: RawFd) -> Result<()> {
     }
 }
 
+fn create_tun_device() -> Result<()> {
+    let path = Path::new(std::str::from_utf8(TUN_FILE).unwrap());
+
+    if path.exists() {
+        return Ok(());
+    }
+
+    let parent_dir = path.parent().unwrap();
+    fs::create_dir_all(parent_dir)?;
+    let permissions = fs::Permissions::from_mode(0o751);
+    fs::set_permissions(parent_dir, permissions)?;
+    if unsafe {
+        mknod(
+            TUN_FILE.as_ptr() as _,
+            S_IFCHR,
+            makedev(TUN_DEV_MAJOR, TUN_DEV_MINOR),
+        )
+    } != 0
+    {
+        return Err(get_last_error());
+    }
+
+    Ok(())
+}
+
 /// Read from the given file descriptor in the buffer.
 fn read(fd: RawFd, dst: &mut [u8]) -> io::Result<usize> {
     // Safety: Within this module, the file descriptor is always valid.

scripts/gateway-docker-upgrade.sh

@@ -3,8 +3,6 @@
 set -e
 
 TARGET_IMAGE="ghcr.io/firezone/gateway:1"
-REPO=$(dirname "$TARGET_IMAGE")
-IMAGE=$(basename "$TARGET_IMAGE")
 
 CURRENTLY_RUNNING=$(docker ps --format "{{.Names}} {{.Image}}" | grep -e "$TARGET_IMAGE" | awk '{print $1}')
 if [ "$CURRENTLY_RUNNING" == "" ]; then
@@ -40,7 +38,6 @@ do
           --sysctl net.ipv6.conf.all.disable_ipv6=0 \
           --sysctl net.ipv6.conf.all.forwarding=1 \
           --sysctl net.ipv6.conf.default.forwarding=1 \
-          --device="/dev/net/tun:/dev/net/tun" \
           "$TARGET_IMAGE"
         rm variables.env
         echo "Container upgraded"

terraform/modules/aws/gateway/templates/cloud-init.yaml

@@ -24,7 +24,7 @@ write_files:
       ExecStartPre=-/usr/bin/docker stop ${container_name}
       ExecStartPre=-/usr/bin/docker rm ${container_name}
       ExecStartPre=/usr/bin/docker pull ${container_image}
-      ExecStart=/bin/sh -c 'docker run --rm --name=${container_name} --cap-add=NET_ADMIN --volume /etc/firezone --sysctl net.ipv4.ip_forward=1 --sysctl net.ipv4.conf.all.src_valid_mark=1 --sysctl net.ipv6.conf.all.disable_ipv6=0 --sysctl net.ipv6.conf.all.forwarding=1 --sysctl net.ipv6.conf.default.forwarding=1 --device="/dev/net/tun:/dev/net/tun" --env FIREZONE_NAME=$(hostname) --env FIREZONE_ID=$(echo $RANDOM$(hostname) | md5sum | head -c 20; echo;) --env-file="/etc/firezone-gateway/env" ${container_image}'
+      ExecStart=/bin/sh -c 'docker run --rm --name=${container_name} --cap-add=NET_ADMIN --volume /etc/firezone --sysctl net.ipv4.ip_forward=1 --sysctl net.ipv4.conf.all.src_valid_mark=1 --sysctl net.ipv6.conf.all.disable_ipv6=0 --sysctl net.ipv6.conf.all.forwarding=1 --sysctl net.ipv6.conf.default.forwarding=1 --env FIREZONE_NAME=$(hostname) --env FIREZONE_ID=$(echo $RANDOM$(hostname) | md5sum | head -c 20; echo;) --env-file="/etc/firezone-gateway/env" ${container_image}'
       ExecStop=/usr/bin/docker stop gateway
       ExecStopPost=/usr/bin/docker rm gateway
 

terraform/modules/gateway-google-cloud-compute/templates/cloud-init.yaml

@@ -107,7 +107,7 @@ write_files:
       TimeoutStartSec=0
       Restart=always
       ExecStartPre=/usr/bin/docker pull ${container_image}
-      ExecStart=/bin/sh -c 'docker run --rm --name=${container_name} --cap-add=NET_ADMIN --volume /etc/firezone --device="/dev/net/tun:/dev/net/tun" --env FIREZONE_NAME=$(hostname) --env FIREZONE_ID=$(echo $RANDOM$(hostname) | md5sum | head -c 20; echo;) --env-file="/etc/firezone-gateway/.env" ${container_image}'
+      ExecStart=/bin/sh -c 'docker run --rm --name=${container_name} --cap-add=NET_ADMIN --volume /etc/firezone --env FIREZONE_NAME=$(hostname) --env FIREZONE_ID=$(echo $RANDOM$(hostname) | md5sum | head -c 20; echo;) --env-file="/etc/firezone-gateway/.env" ${container_image}'
       ExecStop=/usr/bin/docker stop gateway
       ExecStopPost=/usr/bin/docker rm gateway