Add fallback-ap.sh

fallback-ap.sh

@@ -0,0 +1,560 @@
+#!/bin/bash
+
+set -e
+
+# Configuration =======================================================================================================
+
+WLAN0_IFACE="wlan0"
+
+AP_NAME="${HOSTNAME}"
+AP_PASSWORD='guesswhatitis'
+AP_ENCRYPTION="wpa-psk" # none | wpa-psk | sae
+
+AP_CIDR="10.10.10.1/24"
+DHCP_START="10.10.10.100"
+DHCP_END="10.10.10.250"
+
+# How often, while the AP has no clients, to stop the AP and give
+# NetworkManager a chance to reconnect to a known Wi-Fi network.
+RECOVERY_CHECK_INTERVAL=60
+
+# How long to leave the AP down during a recovery attempt before giving
+# up and bringing the AP back.
+RECOVERY_WINDOW=20
+
+# Small delay after changing wlan0 state, to let things settle before
+# we query it again.
+STATE_SETTLE_DELAY=1
+
+CONNECTION_NAME="Fallback AP"
+SERVICE_NAME="pi-fallback-ap"
+SERVICE_FILE="/etc/systemd/system/$SERVICE_NAME.service"
+
+STATE_DIR="/var/lib/pi-fallback-ap"
+LAST_WIFI_FILE="$STATE_DIR/last-wifi"
+LOCK_FILE="$STATE_DIR/lock"
+STATE_FILE="$STATE_DIR/state"
+
+# End of Configuration ================================================================================================
+
+
+# Resolve the actual location of this script.
+SCRIPT_PATH="$(readlink -f "$0")"
+
+if [ -z "$SCRIPT_PATH" ] || [ ! -f "$SCRIPT_PATH" ]; then
+    echo "Unable to determine script path."
+    exit 1
+fi
+
+
+log() {
+    echo "[$(date '+%H:%M:%S')] $*"
+}
+
+
+set_state() {
+    printf '%s\n' "$1" > "$STATE_FILE"
+}
+
+
+get_state() {
+    cat "$STATE_FILE" 2>/dev/null || echo "unknown"
+}
+
+
+is_ap_on() {
+    nmcli -t -f NAME connection show --active 2>/dev/null |
+        grep -Fxq "$CONNECTION_NAME"
+}
+
+
+is_wifi_connected() {
+    nmcli -t -f DEVICE,TYPE,STATE,CONNECTION device status 2>/dev/null |
+        awk -F: -v iface="$WLAN0_IFACE" -v ap="$CONNECTION_NAME" \
+        '$1 == iface && $2 == "wifi" && $3 == "connected" && $4 != ap {found=1} END {exit !found}'
+}
+
+
+get_wifi_connection() {
+    nmcli -t -f DEVICE,TYPE,STATE,CONNECTION device status 2>/dev/null |
+        awk -F: -v iface="$WLAN0_IFACE" \
+        '$1 == iface && $2 == "wifi" && $3 == "connected" {print $4; exit}'
+}
+
+
+remember_wifi() {
+    local connection
+
+    connection="$(get_wifi_connection)"
+
+    [ -z "$connection" ] && return
+    [ "$connection" = "$CONNECTION_NAME" ] && return
+
+    mkdir -p "$STATE_DIR"
+
+    if [ "$(cat "$LAST_WIFI_FILE" 2>/dev/null || true)" != "$connection" ]; then
+        log "Remembering Wi-Fi connection: $connection"
+        printf '%s\n' "$connection" > "$LAST_WIFI_FILE"
+    fi
+}
+
+
+get_last_wifi() {
+    [ -f "$LAST_WIFI_FILE" ] && cat "$LAST_WIFI_FILE"
+}
+
+
+get_client_count() {
+    iw dev "$WLAN0_IFACE" station dump 2>/dev/null |
+        grep -c '^Station ' || true
+}
+
+
+wait_for_ap_state() {
+    local wanted="$1"
+    local timeout="${2:-5}"
+
+    for ((i=0; i<timeout*10; i++)); do
+        if [ "$wanted" = "on" ] && is_ap_on; then
+            return 0
+        fi
+
+        if [ "$wanted" = "off" ] && ! is_ap_on; then
+            return 0
+        fi
+
+        sleep 0.1
+    done
+
+    return 1
+}
+
+
+wait_for_wifi() {
+    local timeout="$1"
+
+    for ((i=0; i<timeout*10; i++)); do
+        if is_wifi_connected; then
+            return 0
+        fi
+
+        sleep 0.1
+    done
+
+    return 1
+}
+
+
+start_ap() {
+    if is_ap_on; then
+        return 0
+    fi
+
+    log "Enabling AP: $AP_NAME"
+
+    nmcli connection up "$CONNECTION_NAME" ifname "$WLAN0_IFACE" >/dev/null 2>&1 || {
+        log "Failed to start AP"
+        return 1
+    }
+
+    if ! wait_for_ap_state on 5; then
+        log "AP failed to become active"
+        return 1
+    fi
+
+    sleep "$STATE_SETTLE_DELAY"
+
+    log "AP enabled"
+
+    return 0
+}
+
+
+stop_ap() {
+    if ! is_ap_on; then
+        return 0
+    fi
+
+    log "Disabling AP"
+
+    nmcli connection down "$CONNECTION_NAME" >/dev/null 2>&1 || true
+
+    if ! wait_for_ap_state off 5; then
+        log "AP failed to stop cleanly"
+        return 1
+    fi
+
+    sleep "$STATE_SETTLE_DELAY"
+
+    return 0
+}
+
+
+# ==========================================================================
+# One tick of the control loop. Called once per second, always while
+# holding the wlan0 lock, so everything here runs to completion before
+# the next tick starts - there is no interleaving to race against.
+#
+# `last_attempt` is intentionally not declared `local` here: it lives in
+# run()'s scope and this function updates it directly (bash resolves it
+# via the caller's scope as long as nothing else shadows it).
+# ==========================================================================
+tick() {
+    local now="$1"
+
+    if is_wifi_connected; then
+        remember_wifi
+
+        if is_ap_on; then
+            stop_ap
+        fi
+
+        set_state "wifi"
+        last_attempt="$now"
+        return
+    fi
+
+    if ! is_ap_on; then
+        start_ap
+        set_state "fallback"
+        last_attempt="$now"
+        return
+    fi
+
+    set_state "fallback"
+
+    local clients
+    clients="$(get_client_count)"
+
+    if [ "$clients" -gt 0 ]; then
+        # Someone is using the AP right now. Leave it alone and push the
+        # next recovery attempt back so we don't kick them off.
+        last_attempt="$now"
+        return
+    fi
+
+    if (( now - last_attempt >= RECOVERY_CHECK_INTERVAL )); then
+        last_attempt="$now"
+        attempt_recovery
+    fi
+}
+
+
+# Stop the AP, give NetworkManager a window to autoconnect to any known
+# network it can see, and bring the AP back if that doesn't happen.
+attempt_recovery() {
+    log "No AP clients, checking for known Wi-Fi (AP down for up to ${RECOVERY_WINDOW}s)"
+
+    set_state "checking"
+
+    stop_ap || true
+
+    if is_ap_on; then
+        log "AP did not go down, aborting recovery attempt"
+        start_ap
+        set_state "fallback"
+        return 1
+    fi
+
+    if wait_for_wifi "$RECOVERY_WINDOW"; then
+        sleep "$STATE_SETTLE_DELAY"
+
+        if is_wifi_connected; then
+            log "Wi-Fi reconnected"
+            remember_wifi
+            set_state "wifi"
+            return 0
+        fi
+    fi
+
+    log "No Wi-Fi available, restoring AP"
+
+    start_ap
+    set_state "fallback"
+
+    return 1
+}
+
+
+install() {
+    if [ "$EUID" -ne 0 ]; then
+        echo "Run this script as root."
+        exit 1
+    fi
+
+    log "Installing NetworkManager fallback AP"
+
+    apt-get update
+    apt-get install -y network-manager iw
+
+    systemctl enable NetworkManager
+    systemctl start NetworkManager
+
+    mkdir -p "$STATE_DIR"
+
+    if is_wifi_connected; then
+        remember_wifi
+    fi
+
+    create_connection
+
+    log "Creating recovery service"
+
+    cat > "$SERVICE_FILE" <<EOF
+[Unit]
+Description=NetworkManager Fallback AP
+After=NetworkManager.service
+Wants=NetworkManager.service
+
+[Service]
+Type=simple
+ExecStart=$SCRIPT_PATH run
+Restart=always
+RestartSec=2
+
+[Install]
+WantedBy=multi-user.target
+EOF
+
+    chmod 644 "$SERVICE_FILE"
+
+    systemctl daemon-reload
+    systemctl enable "$SERVICE_NAME"
+    systemctl restart "$SERVICE_NAME"
+
+    log "Installation complete"
+    log "Script: $SCRIPT_PATH"
+    log "SSID: $AP_NAME"
+    log "Interface: $WLAN0_IFACE"
+    log "Network: $AP_CIDR"
+    log "DHCP: $DHCP_START - $DHCP_END"
+    log "Encryption: $AP_ENCRYPTION"
+    log "Recovery check interval: ${RECOVERY_CHECK_INTERVAL}s"
+    log "Recovery window: ${RECOVERY_WINDOW}s"
+    log "Last Wi-Fi: $(get_last_wifi || echo none)"
+
+    echo
+    nmcli device status
+}
+
+
+create_connection() {
+    log "Creating NetworkManager AP connection..."
+
+    nmcli connection delete "$CONNECTION_NAME" >/dev/null 2>&1 || true
+
+    nmcli connection add \
+        type wifi \
+        ifname "$WLAN0_IFACE" \
+        con-name "$CONNECTION_NAME" \
+        autoconnect no \
+        ssid "$AP_NAME"
+
+    nmcli connection modify "$CONNECTION_NAME" \
+        connection.autoconnect no \
+        connection.autoconnect-priority -999 \
+        connection.autoconnect-retries 0 \
+        802-11-wireless.mode ap \
+        802-11-wireless.powersave 2 \
+        ipv4.method shared \
+        ipv4.addresses "$AP_CIDR" \
+        ipv4.shared-dhcp-range "$DHCP_START,$DHCP_END" \
+        ipv6.method disabled
+
+    case "$AP_ENCRYPTION" in
+        none)
+            nmcli connection modify "$CONNECTION_NAME" \
+                wifi-sec.key-mgmt none
+            ;;
+
+        wpa-psk)
+            if [ "${#AP_PASSWORD}" -lt 8 ] || [ "${#AP_PASSWORD}" -gt 63 ]; then
+                echo "WPA-PSK password must be between 8 and 63 characters."
+                exit 1
+            fi
+
+            nmcli connection modify "$CONNECTION_NAME" \
+                wifi-sec.key-mgmt wpa-psk \
+                wifi-sec.psk "$AP_PASSWORD"
+            ;;
+
+        sae)
+            if [ "${#AP_PASSWORD}" -lt 8 ]; then
+                echo "WPA3-SAE password must be at least 8 characters."
+                exit 1
+            fi
+
+            nmcli connection modify "$CONNECTION_NAME" \
+                wifi-sec.key-mgmt sae \
+                wifi-sec.psk "$AP_PASSWORD"
+            ;;
+
+        *)
+            echo "Unsupported encryption: $AP_ENCRYPTION"
+            exit 1
+            ;;
+    esac
+}
+
+
+uninstall() {
+    if [ "$EUID" -ne 0 ]; then
+        echo "Run this script as root."
+        exit 1
+    fi
+
+    log "Uninstalling fallback AP"
+
+    systemctl disable --now "$SERVICE_NAME" >/dev/null 2>&1 || true
+
+    nmcli connection down "$CONNECTION_NAME" >/dev/null 2>&1 || true
+    nmcli connection delete "$CONNECTION_NAME" >/dev/null 2>&1 || true
+
+    rm -f "$SERVICE_FILE"
+    rm -rf "$STATE_DIR"
+
+    systemctl daemon-reload
+
+    log "Fallback AP removed"
+}
+
+
+start() {
+    mkdir -p "$STATE_DIR"
+
+    (
+        flock -n 9 || exit 1
+        start_ap
+    ) 9>"$LOCK_FILE"
+}
+
+
+stop() {
+    mkdir -p "$STATE_DIR"
+
+    (
+        flock -n 9 || exit 1
+        stop_ap
+    ) 9>"$LOCK_FILE"
+}
+
+
+status() {
+    echo
+    echo "=== Service ==="
+    systemctl status "$SERVICE_NAME" --no-pager || true
+
+    echo
+    echo "=== Devices ==="
+    nmcli device status
+
+    echo
+    echo "=== Active Connections ==="
+    nmcli connection show --active
+
+    echo
+    echo "=== Saved Wi-Fi Profiles ==="
+    nmcli -f NAME,TYPE,AUTOCONNECT,AUTOCONNECT-PRIORITY connection show
+
+    echo
+    echo "=== Fallback AP ==="
+    nmcli connection show "$CONNECTION_NAME" 2>/dev/null || echo "Not installed"
+
+    echo
+    echo "=== Last Known Wi-Fi ==="
+    get_last_wifi || echo "None"
+
+    echo
+    echo "=== State ==="
+    get_state
+
+    echo
+    echo "=== AP Clients ==="
+    get_client_count
+
+    echo
+    echo "=== Script ==="
+    echo "$SCRIPT_PATH"
+}
+
+
+run() {
+    local last_attempt=0
+    local now
+
+    mkdir -p "$STATE_DIR"
+
+    log "Fallback AP service started"
+    log "Script: $SCRIPT_PATH"
+
+    # Open the lock fd once for the life of the process. Using `flock`
+    # directly against this fd (instead of wrapping each tick in its own
+    # `( ... ) 9>"$LOCK_FILE"` subshell) matters: any variable a tick
+    # updates - like last_attempt below - has to survive into the next
+    # loop iteration, and a subshell would silently throw those updates
+    # away when it exits.
+    exec 9>"$LOCK_FILE"
+
+    flock 9
+
+    if is_wifi_connected; then
+        remember_wifi
+        set_state "wifi"
+        log "Wi-Fi already connected"
+    else
+        log "No Wi-Fi connection, starting fallback AP"
+        start_ap
+        set_state "fallback"
+    fi
+
+    flock -u 9
+
+    while true; do
+        now="$(date +%s)"
+
+        if flock -n 9; then
+            tick "$now"
+            flock -u 9
+        fi
+
+        sleep 1
+    done
+}
+
+
+if [ "$EUID" -ne 0 ]; then
+    echo "Run this script as root."
+    exit 1
+fi
+
+
+case "${1:-}" in
+    install)
+        install
+        ;;
+
+    uninstall)
+        uninstall
+        ;;
+
+    start)
+        start
+        ;;
+
+    stop)
+        stop
+        ;;
+
+    status)
+        status
+        ;;
+
+    run)
+        run
+        ;;
+
+    *)
+        echo "Usage: $SCRIPT_PATH {install|uninstall|start|stop|status|run}"
+        exit 1
+        ;;
+esac