fix(networking/usque): disconnect now always works successfully.HEADmaster

1 files changed, 109 insertions, 82 deletions

diff --git a/modules/networking/usque/usque-warp.sh b/modules/networking/usque/usque-warp.sh
index 1b70c12..5609589 100644
--- a/modules/networking/usque/usque-warp.sh
+++ b/modules/networking/usque/usque-warp.sh
@@ -32,59 +32,112 @@ detect_iface() {
   list_tun_ifaces | head -n1
 }
 
+# True if the PID in PID_FILE is a live usque process. Uses /proc rather
+# than `sudo kill -0` so it never prompts for a password (matters for
+# Waybar's status polling) and can't be fooled by PID reuse.
 is_running() {
   [[ -f $PID_FILE ]] || return 1
   local pid
   pid=$(cat "$PID_FILE" 2>/dev/null || true)
-  [[ -n $pid ]] || return 1
-  sudo kill -0 "$pid" 2>/dev/null
+  [[ $pid =~ ^[0-9]+$ ]] || return 1
+  [[ $(cat "/proc/$pid/comm" 2>/dev/null || true) == usque ]]
+}
+
+# True if ANY usque process exists, tracked by the PID file or not.
+any_usque_running() {
+  pgrep -x usque >/dev/null 2>&1
+}
+
+# Stop every usque process (tracked or orphaned): SIGTERM, wait, then SIGKILL.
+stop_usque() {
+  if any_usque_running; then
+    info "Stopping usque..."
+    sudo pkill -x usque 2>/dev/null || true
+    for _ in {1..25}; do
+      any_usque_running || break
+      sleep 0.2
+    done
+    if any_usque_running; then
+      warn "usque did not stop on SIGTERM, forcing kill..."
+      sudo pkill -9 -x usque 2>/dev/null || true
+      sleep 0.5
+    fi
+  fi
+  rm -f "$PID_FILE"
 }
 
 ensure_config() {
+  # Only register when there is no config; registering on every run
+  # creates a brand-new WARP account each time.
+  if [[ -f $CONFIG ]]; then
+    return
+  fi
   info "Creating $CONFIG_DIR..."
   mkdir -p "$CONFIG_DIR"
   info "Registering Cloudflare WARP account..."
   usque -c "$CONFIG" register < <(yes)
-  if [[ ! -f $CONFIG ]]; then
-    die "Failed to create config file: $CONFIG"
-  fi
+  [[ -f $CONFIG ]] || die "Failed to create config file: $CONFIG"
   info "Config created successfully."
 }
 
 remove_tun_default_routes() {
-  local dev="$1"
+  local dev="$1" route
   while ip route show | grep -qE "^default .*dev $dev"; do
-    ROUTE=$(ip route show | grep -E "^default .*dev $dev" | head -n1)
-    info "Removing route: $ROUTE"
-    sudo ip route del "$ROUTE" || break
+    route=$(ip route show | grep -E "^default .*dev $dev" | head -n1)
+    info "Removing route: $route"
+    # Unquoted on purpose: the route must be split into separate arguments.
+    # shellcheck disable=SC2086
+    sudo ip route del $route || break
   done
 }
 
+CONNECT_OK=false
+
+# EXIT trap for connect: if we leave for any reason (die, a failing command,
+# Ctrl-C) before reaching "Connected.", tear everything down again.
+cleanup_failed_connect() {
+  local rc=$?
+  trap - EXIT INT TERM
+  if [[ $CONNECT_OK != true ]]; then
+    warn "connect failed, cleaning up..."
+    disconnect || true
+  fi
+  exit "$rc"
+}
+
 connect() {
   sudo -v
-  ensure_config
-  if [[ -f $PID_FILE ]]; then
-    OLD_PID=$(cat "$PID_FILE")
-    if sudo kill -0 "$OLD_PID" 2>/dev/null; then
-      die "usque-warp is already running (PID $OLD_PID)"
-    else
-      info "Removing stale PID file..."
-      rm -f "$PID_FILE"
-    fi
+
+  # Refuse to start if any usque is already running, not just a tracked one.
+  # This runs before the cleanup trap is armed, so we never kill a tunnel
+  # we didn't start.
+  if any_usque_running; then
+    die "usque is already running (run '$0 disconnect' first)"
   fi
+  rm -f "$PID_FILE" "$LOG_FILE" "$STATE_FILE" "$IFACE_FILE"
+
+  ensure_config
+
+  trap cleanup_failed_connect EXIT
+  trap 'exit 130' INT TERM
 
   info "Saving current default route..."
-  DEFAULT_ROUTE=$(ip route show default | grep -vE 'dev tun[0-9]+' | head -n1)
+  DEFAULT_ROUTE=$(ip route show default | grep -vE 'dev tun[0-9]+' | head -n1 || true)
   if [[ -z $DEFAULT_ROUTE ]]; then
     die "Could not determine current default route"
   fi
+  GATEWAY=$(echo "$DEFAULT_ROUTE" | awk '{for(i=1;i<=NF;i++) if($i=="via") print $(i+1)}')
+  INTERFACE=$(echo "$DEFAULT_ROUTE" | awk '{for(i=1;i<=NF;i++) if($i=="dev") print $(i+1)}')
+  if [[ -z $GATEWAY || -z $INTERFACE ]]; then
+    die "Cannot determine gateway/interface from: $DEFAULT_ROUTE"
+  fi
   echo "$DEFAULT_ROUTE" >"$STATE_FILE"
 
   info "Recording pre-existing tun interfaces..."
   BEFORE_IFACES=$(list_tun_ifaces)
 
   info "Starting usque..."
-  # sudo does not affect redirects — the outer shell would open "$LOG_FILE"
+  # sudo does not affect redirects: the outer shell would open "$LOG_FILE"
   # as the unprivileged user. Run the redirection inside sudo (via sh) so the
   # log is opened as root. $$ inside sh is usque's PID (sh exec's usque, keeping
   # the PID stable regardless of how sudo forks internally), written up front.
@@ -94,18 +147,16 @@ connect() {
   info "Waiting for MASQUE connection..."
   MASQUE_IP=""
   for _ in {1..30}; do
-    MASQUE_IP=$(grep -oP 'MASQUE connection to \K[0-9.]+(?=:443)' "$LOG_FILE" 2>/dev/null || true)
-    if [[ -n $MASQUE_IP ]]; then
-      break
+    # head -n1: usque may log this line more than once; we need exactly one IP.
+    MASQUE_IP=$(grep -oP 'MASQUE connection to \K[0-9.]+(?=:443)' "$LOG_FILE" 2>/dev/null | head -n1 || true)
+    [[ -n $MASQUE_IP ]] && break
+    if [[ -f $PID_FILE ]] && ! is_running; then
+      die "usque exited unexpectedly; see $LOG_FILE"
     fi
     sleep 1
   done
-  if [[ -z $MASQUE_IP ]]; then
-    error "Failed to detect MASQUE endpoint"
-    sudo kill "$(cat "$PID_FILE")" 2>/dev/null || true
-    rm -f "$PID_FILE"
-    exit 1
-  fi
+  [[ -n $MASQUE_IP ]] || die "Failed to detect MASQUE endpoint; see $LOG_FILE"
+  [[ $MASQUE_IP =~ ^[0-9]+(\.[0-9]+){3}$ ]] || die "Unexpected MASQUE endpoint: '$MASQUE_IP'"
 
   info "Waiting for usque interface..."
   TUN_DEV=""
@@ -115,69 +166,42 @@ connect() {
     [[ -n $TUN_DEV ]] && break
     sleep 1
   done
-  if [[ -z $TUN_DEV ]]; then
-    error "Failed to detect usque interface"
-    sudo kill "$(cat "$PID_FILE")" 2>/dev/null || true
-    rm -f "$PID_FILE"
-    exit 1
-  fi
+  [[ -n $TUN_DEV ]] || die "Failed to detect usque interface"
   echo "$TUN_DEV" >"$IFACE_FILE"
   info "Detected interface: $TUN_DEV"
 
-  GATEWAY=$(echo "$DEFAULT_ROUTE" | awk '{for(i=1;i<=NF;i++) if($i=="via") print $(i+1)}')
-  INTERFACE=$(echo "$DEFAULT_ROUTE" | awk '{for(i=1;i<=NF;i++) if($i=="dev") print $(i+1)}')
-  if [[ -z $GATEWAY || -z $INTERFACE ]]; then
-    die "Cannot determine gateway/interface"
-  fi
   echo "MASQUE_IP=$MASQUE_IP GATEWAY=$GATEWAY INTERFACE=$INTERFACE" >>"$STATE_FILE"
 
   info "Allowing MASQUE endpoint outside tunnel..."
-  sudo ip route replace \
-    "$MASQUE_IP" \
-    via "$GATEWAY" \
-    dev "$INTERFACE"
+  sudo ip route replace "$MASQUE_IP" via "$GATEWAY" dev "$INTERFACE"
 
   info "Removing old tun routes..."
   remove_tun_default_routes "$TUN_DEV"
 
   info "Switching default route to $TUN_DEV..."
   sudo ip route add default dev "$TUN_DEV" metric 1
+
+  CONNECT_OK=true
+  trap - EXIT INT TERM
   info "Connected."
 }
 
 disconnect() {
   sudo -v
   info "Disconnecting..."
-  local dev
-  if [[ -f $IFACE_FILE ]]; then
-    dev=$(cat "$IFACE_FILE")
-  else
-    dev=$(list_tun_ifaces | head -n1)
-  fi
 
-  # Kill usque FIRST so the kernel tears down tun0 (and every route
-  # bound to it) as a single atomic operation, instead of us racing
-  # it by pulling routes out from under a device that's still up.
-  if [[ -f $PID_FILE ]]; then
-    PID=$(cat "$PID_FILE")
-    if sudo kill -0 "$PID" 2>/dev/null; then
-      info "Stopping usque..."
-      sudo kill "$PID" 2>/dev/null || true
-      for _ in {1..25}; do
-        sudo kill -0 "$PID" 2>/dev/null || break
-        sleep 0.2
-      done
-      if sudo kill -0 "$PID" 2>/dev/null; then
-        warn "usque did not stop on SIGTERM, forcing kill..."
-        sudo kill -9 "$PID" 2>/dev/null || true
-      fi
-    fi
-    rm -f "$PID_FILE"
-  fi
+  local dev=""
+  [[ -f $IFACE_FILE ]] && dev=$(cat "$IFACE_FILE")
 
-  # Give the kernel a moment to tear the tunnel device down; only touch
-  # routes manually if it is not disappearing on its own.
-  if [[ -n ${dev:-} ]]; then
+  # Kill usque FIRST so the kernel tears down the tun device (and every
+  # route bound to it) in one step, instead of us racing it by pulling
+  # routes out from under a device that's still up. This also kills any
+  # orphaned usque processes the PID file doesn't know about.
+  stop_usque
+
+  # Give the kernel a moment to remove the device; only touch routes
+  # manually if it is not disappearing on its own.
+  if [[ -n $dev ]]; then
     info "Waiting for tunnel interface to go down..."
     for _ in {1..25}; do
       list_tun_ifaces | grep -qx "$dev" || break
@@ -191,25 +215,28 @@ disconnect() {
   fi
 
   if [[ -f $STATE_FILE ]]; then
-    MASQUE_IP=$(grep -oP 'MASQUE_IP=\K[0-9.]+' "$STATE_FILE" || true)
-    if [[ -n $MASQUE_IP ]]; then
-      info "Removing MASQUE route: $MASQUE_IP"
-      sudo ip route del "$MASQUE_IP" 2>/dev/null || true
+    local masque_ip original_default
+    masque_ip=$(grep -oP 'MASQUE_IP=\K[0-9.]+' "$STATE_FILE" | head -n1 || true)
+    if [[ -n $masque_ip ]]; then
+      info "Removing MASQUE route: $masque_ip"
+      sudo ip route del "$masque_ip" 2>/dev/null || true
     fi
 
     # Explicitly restore the pre-connect default route rather than
     # assuming it's still intact. 'replace' is idempotent.
-    ORIGINAL_DEFAULT=$(head -n1 "$STATE_FILE")
-    if [[ $ORIGINAL_DEFAULT == default* ]]; then
+    original_default=$(head -n1 "$STATE_FILE")
+    if [[ $original_default == default* ]]; then
       info "Restoring original default route..."
-      sudo ip route replace "$ORIGINAL_DEFAULT" ||
+      # Unquoted on purpose: the route must be split into separate arguments.
+      # shellcheck disable=SC2086
+      sudo ip route replace $original_default ||
         warn "could not restore original default route"
     fi
-
-    rm -f "$STATE_FILE"
+  elif [[ -z $(ip route show default) ]]; then
+    warn "no default route and no saved state; restore it manually (ip route add default via <gw> dev <if>)"
   fi
 
-  rm -f "$IFACE_FILE"
+  rm -f "$STATE_FILE" "$IFACE_FILE"
   info "Disconnected."
 }