secubox-openwrt/package/secubox/secubox-base/files/usr/sbin/secubox-network-health

#!/bin/sh
# SecuBox Network Health Monitor
# Detects CRC errors, link flapping, and interface issues

. /usr/share/libubox/jshn.sh

DMESG_LINES=500
FLAP_THRESHOLD=5  # Number of link changes to consider flapping
CRC_THRESHOLD=10  # CRC errors to consider problematic

check_interface_health() {
    local iface="$1"
    local status="ok"
    local issues=""
    local crc_count=0
    local link_changes=0
    local current_state="unknown"

    # Get current link state
    if [ -d "/sys/class/net/$iface" ]; then
        current_state=$(cat /sys/class/net/$iface/operstate 2>/dev/null || echo "unknown")
    fi

    # Count CRC errors from dmesg (last N lines)
    crc_count=$(dmesg | tail -n $DMESG_LINES | grep -c "$iface.*crc error" 2>/dev/null || echo 0)

    # Count link state changes from dmesg
    link_up=$(dmesg | tail -n $DMESG_LINES | grep -c "$iface: Link is Up" 2>/dev/null || echo 0)
    link_down=$(dmesg | tail -n $DMESG_LINES | grep -c "$iface: Link is Down" 2>/dev/null || echo 0)
    link_changes=$((link_up + link_down))

    # Determine status
    if [ "$crc_count" -ge "$CRC_THRESHOLD" ]; then
        status="critical"
        issues="${issues}CRC errors detected ($crc_count); "
    fi

    if [ "$link_changes" -ge "$FLAP_THRESHOLD" ]; then
        if [ "$status" = "ok" ]; then
            status="warning"
        fi
        issues="${issues}Link flapping detected ($link_changes changes); "
    fi

    # Get interface stats
    local rx_errors=0 tx_errors=0 rx_dropped=0 tx_dropped=0
    if [ -f "/sys/class/net/$iface/statistics/rx_errors" ]; then
        rx_errors=$(cat /sys/class/net/$iface/statistics/rx_errors)
        tx_errors=$(cat /sys/class/net/$iface/statistics/tx_errors)
        rx_dropped=$(cat /sys/class/net/$iface/statistics/rx_dropped)
        tx_dropped=$(cat /sys/class/net/$iface/statistics/tx_dropped)
    fi

    if [ "$rx_errors" -gt 1000 ] || [ "$tx_errors" -gt 1000 ]; then
        if [ "$status" = "ok" ]; then
            status="warning"
        fi
        issues="${issues}High error count (rx:$rx_errors tx:$tx_errors); "
    fi

    # Output JSON for this interface
    json_add_object "$iface"
    json_add_string "status" "$status"
    json_add_string "state" "$current_state"
    json_add_int "crc_errors" "$crc_count"
    json_add_int "link_changes" "$link_changes"
    json_add_int "rx_errors" "$rx_errors"
    json_add_int "tx_errors" "$tx_errors"
    json_add_int "rx_dropped" "$rx_dropped"
    json_add_int "tx_dropped" "$tx_dropped"
    json_add_string "issues" "${issues%%; }"
    json_close_object
}

get_network_health() {
    json_init
    json_add_string "timestamp" "$(date -Iseconds)"
    json_add_object "interfaces"

    # Check all physical interfaces
    for iface in /sys/class/net/eth* /sys/class/net/wan* /sys/class/net/lan*; do
        [ -d "$iface" ] || continue
        iface_name=$(basename "$iface")
        # Skip virtual interfaces (must have device link)
        [ -d "$iface/device" ] || continue
        check_interface_health "$iface_name"
    done

    json_close_object

    # Overall status
    local overall="healthy"
    local critical_count=0
    local warning_count=0

    # Re-scan for overall status
    for iface in /sys/class/net/eth* /sys/class/net/wan* /sys/class/net/lan*; do
        [ -d "$iface" ] || continue
        [ -d "$iface/device" ] || continue
        iface_name=$(basename "$iface")

        crc=$(dmesg | tail -n $DMESG_LINES | grep -c "$iface_name.*crc error" 2>/dev/null || echo 0)
        if [ "$crc" -ge "$CRC_THRESHOLD" ]; then
            critical_count=$((critical_count + 1))
        fi

        link_up=$(dmesg | tail -n $DMESG_LINES | grep -c "$iface_name: Link is Up" 2>/dev/null || echo 0)
        link_down=$(dmesg | tail -n $DMESG_LINES | grep -c "$iface_name: Link is Down" 2>/dev/null || echo 0)
        if [ $((link_up + link_down)) -ge "$FLAP_THRESHOLD" ]; then
            warning_count=$((warning_count + 1))
        fi
    done

    if [ "$critical_count" -gt 0 ]; then
        overall="critical"
    elif [ "$warning_count" -gt 0 ]; then
        overall="warning"
    fi

    json_add_string "overall" "$overall"
    json_add_int "critical_interfaces" "$critical_count"
    json_add_int "warning_interfaces" "$warning_count"

    # Add recommendations if issues found
    if [ "$overall" != "healthy" ]; then
        json_add_array "recommendations"
        if [ "$critical_count" -gt 0 ]; then
            json_add_string "" "Check/replace Ethernet cables on affected interfaces"
            json_add_string "" "Try different port on switch/modem"
            json_add_string "" "Inspect RJ45 connectors for damage"
        fi
        if [ "$warning_count" -gt 0 ]; then
            json_add_string "" "Monitor link stability"
            json_add_string "" "Check for EMI interference near cables"
        fi
        json_close_array
    fi

    json_dump
}

get_interface_detail() {
    local iface="$1"

    if [ ! -d "/sys/class/net/$iface" ]; then
        echo '{"error": "Interface not found"}'
        return 1
    fi

    json_init
    json_add_string "interface" "$iface"
    json_add_string "state" "$(cat /sys/class/net/$iface/operstate 2>/dev/null)"
    json_add_string "mac" "$(cat /sys/class/net/$iface/address 2>/dev/null)"
    json_add_int "mtu" "$(cat /sys/class/net/$iface/mtu 2>/dev/null)"

    # Recent dmesg entries for this interface
    json_add_array "recent_events"
    dmesg | tail -n 100 | grep "$iface" | tail -n 10 | while read line; do
        json_add_string "" "$line"
    done
    json_close_array

    json_dump
}

case "$1" in
    status|health)
        get_network_health
        ;;
    detail)
        get_interface_detail "$2"
        ;;
    *)
        echo "Usage: $0 {status|health|detail <interface>}"
        exit 1
        ;;
esac