fix(linux): stop chmod'ing resolv.conf to 0444 in direct-write path (EACCES crash loop on container restart)

src/linux.rs

@@ -316,7 +316,6 @@ fn write_buffer_to_fd(fd: std::os::fd::BorrowedFd<'_>, data: &[u8]) -> Result<()
 fn write_nameserver(fd: std::os::fd::BorrowedFd<'_>, tun_gateway: Option<IpAddr>) -> Result<()> {
     let tun_gateway = tun_gateway.unwrap_or_else(|| "198.18.0.1".parse().unwrap());
     let data = format!("nameserver {tun_gateway}\n");
-    nix::sys::stat::fchmod(fd.as_fd(), nix::sys::stat::Mode::from_bits(0o444).unwrap())?;
     write_buffer_to_fd(fd, data.as_bytes())?;
     Ok(())
 }
@@ -368,8 +367,21 @@ fn setup_resolv_conf(restore: &mut TproxyStateInner) -> Result<()> {
 
         restore.restore_resolvconf_content = Some(fs::read(ETC_RESOLV_CONF_FILE)?);
 
+        // Older versions chmod'ed /etc/resolv.conf to 0444 on this path, which made the
+        // write here fail with EACCES on every subsequent start in containers that
+        // drop CAP_DAC_OVERRIDE (e.g. Docker's bind-mounted resolv.conf). Restore
+        // writability before reopening.
+        let write_mode = nix::sys::stat::Mode::from_bits(0o644).unwrap();
         let flags = nix::fcntl::OFlag::O_WRONLY | nix::fcntl::OFlag::O_CLOEXEC | nix::fcntl::OFlag::O_TRUNC;
-        let fd = nix::fcntl::open(ETC_RESOLV_CONF_FILE, flags, nix::sys::stat::Mode::from_bits(0o644).unwrap())?;
+        let fd = match nix::fcntl::open(ETC_RESOLV_CONF_FILE, flags, write_mode) {
+            Ok(fd) => fd,
+            Err(nix::errno::Errno::EACCES) => {
+                // Recover from the 0444 mode left behind by older versions (see comment above).
+                fs::set_permissions(ETC_RESOLV_CONF_FILE, Permissions::from_mode(0o644))?;
+                nix::fcntl::open(ETC_RESOLV_CONF_FILE, flags, write_mode)?
+            }
+            Err(err) => return Err(err.into()),
+        };
         write_nameserver(fd.as_fd(), tun_gateway)?;
     }
     Ok(())

test-restart-eacces.sh

@@ -0,0 +1,61 @@
+#!/usr/bin/env bash
+# Integration test: tun2proxy container restart EACCES (tproxy-config issue)
+#
+# Validates two properties of the tproxy-config resolv.conf direct-write path:
+#   1. After a fresh container start, the host-side resolv.conf file keeps mode 0644
+#      (old versions chmod'ed it to 0444, which broke every subsequent restart).
+#   2. If the host-side file was left at 0444 by an older version (upgrade case),
+#      a restart must heal it back to 0644 and come back up healthy.
+#
+# Usage: IMAGE=<tun2proxy image> [PROXY=<ip:port>] ./test-restart-eacces.sh
+set -euo pipefail
+
+IMAGE="${IMAGE:?set IMAGE to the tun2proxy image under test}"
+PROXY="${PROXY:-192.0.2.1:1080}"  # dummy proxy target; reachability irrelevant for this test
+NAME="t2p-restart-test"
+
+cleanup() { docker rm -f "$NAME" >/dev/null 2>&1 || true; }
+trap cleanup EXIT
+
+fail() { echo "FAIL: $*" >&2; exit 1; }
+pass() { echo "PASS: $*"; }
+
+# Phase 1: fresh create must start (tun2proxy exits on fatal error if setup fails)
+docker run -d --name "$NAME" --device /dev/net/tun \
+  --cap-drop ALL --cap-add NET_ADMIN \
+  "$IMAGE" --proxy "http://$PROXY" --dns virtual --exit-on-fatal-error \
+  >/dev/null
+
+# Wait up to 15s for either a healthy run or a crash
+sleep 5
+STATE1=$(docker inspect "$NAME" --format '{{.State.Status}} {{.State.ExitCode}}' 2>/dev/null || echo "gone")
+echo "after create: $STATE1"
+[ "$STATE1" = "running 0" ] || [ "${STATE1%% *}" = "running" ] || fail "fresh container not running: $STATE1"
+
+# Phase 2: host-side resolv.conf mode must still be 0644 (not poisoned to 0444)
+DOCKER_ROOT=$(docker info --format '{{.DockerRootDir}}')
+CID=$(docker inspect "$NAME" --format '{{.Id}}')
+HOST_RESOLV="$DOCKER_ROOT/containers/$CID/resolv.conf"
+MODE=$(stat -c %a "$HOST_RESOLV")
+echo "host resolv.conf mode after first run: $MODE"
+[ "$MODE" = "644" ] || fail "resolv.conf poisoned: mode $MODE (expected 644)"
+
+# Phase 3: poison host file to 0444 (upgrade scenario) then restart;
+# the EACCES heal path must restore 0644 and the container must come back up
+chmod 0444 "$HOST_RESOLV" 2>/dev/null || sudo chmod 0444 "$HOST_RESOLV"
+docker restart "$NAME" >/dev/null
+sleep 5
+STATE2=$(docker inspect "$NAME" --format '{{.State.Status}} {{.State.ExitCode}}' 2>/dev/null || echo "gone")
+echo "after restart: $STATE2"
+[ "${STATE2%% *}" = "running" ] || fail "restart failed: $STATE2 (EACCES crash loop?)"
+
+MODE2=$(stat -c %a "$HOST_RESOLV")
+echo "host resolv.conf mode after restart: $MODE2"
+[ "$MODE2" = "644" ] || fail "heal failed: mode $MODE2 (expected 644)"
+
+# Phase 4: double-check no EACCES in logs
+if docker logs "$NAME" 2>&1 | grep -q "EACCES"; then
+  fail "EACCES present in logs after restart"
+fi
+
+pass "create healthy, resolv.conf 0644 preserved, restart heals 0444->0644, no EACCES"