#!/bin/bash
# Usage: ./check_nic_errors [-w WARNING] [-c CRITICAL] [-e EXCLUDE]
#Exclude is used if you want to exclude any of the ifaces, note that virtual ones are already excluded.

WARNING_THRESHOLD=0
CRITICAL_THRESHOLD=0
EXCLUDE_INTERFACES=""

STATE_OK=0
STATE_WARNING=1
STATE_CRITICAL=2
STATE_UNKNOWN=3

while getopts "w:c:e:h" opt; do
    case $opt in
        w) WARNING_THRESHOLD=$OPTARG ;;
        c) CRITICAL_THRESHOLD=$OPTARG ;;
        e) EXCLUDE_INTERFACES=$OPTARG ;;
        h)
            echo "Usage: $0 [-w WARNING] [-c CRITICAL] [-e EXCLUDE]"
            echo "  -w WARNING    Warning threshold for error count (default: 10)"
            echo "  -c CRITICAL   Critical threshold for error count (default: 100)"
            echo "  -e EXCLUDE    Comma-separated list of interfaces to exclude (e.g., lo,docker0)"
            exit $STATE_OK
            ;;
        \?)
            echo "Invalid option: -$OPTARG"
            exit $STATE_UNKNOWN
            ;;
    esac
done

is_excluded() {
    local iface=$1
    if [ -z "$EXCLUDE_INTERFACES" ]; then
        return 1
    fi

    IFS=',' read -ra EXCLUDED <<< "$EXCLUDE_INTERFACES"
    for excl in "${EXCLUDED[@]}"; do
        if [ "$iface" = "$excl" ]; then
            return 0
        fi
    done
    return 1
}

if [ ! -f /proc/net/dev ]; then
    echo "UNKNOWN - /proc/net/dev not found"
    exit $STATE_UNKNOWN
fi

exit_code=$STATE_OK
output_msg="OK - All NICs healthy"
problem_nics=()
perfdata=""

while IFS= read -r line; do

    if [[ $line =~ ^[[:space:]]*Inter- ]] || [[ $line =~ ^[[:space:]]*face ]]; then
        continue
    fi

    if [[ $line =~ ^[[:space:]]*([^:]+):[[:space:]]*(.+)$ ]]; then
        iface="${BASH_REMATCH[1]}"
        stats="${BASH_REMATCH[2]}"

        if [ "$iface" = "lo" ] || is_excluded "$iface"; then
            continue
        fi

        if [[ $iface =~ ^(bond|vmbr|vlan|tun|tap|veth|docker|br-|virbr) ]] || [[ $iface =~ \. ]]; then
            continue
        fi

        if [ ! -d "/sys/class/net/$iface" ]; then
            continue
        fi

        operstate=$(cat "/sys/class/net/$iface/operstate" 2>/dev/null)
        if [ "$operstate" != "up" ]; then
            continue
        fi

        read -r rx_bytes rx_packets rx_errs rx_drop rx_fifo rx_frame rx_compressed rx_multicast \
                tx_bytes tx_packets tx_errs tx_drop tx_fifo tx_colls tx_carrier tx_compressed \
                <<< "$stats"

        total_rx_errors=$((rx_errs + rx_fifo + rx_frame))
        total_tx_errors=$((tx_errs + tx_fifo + tx_carrier))
        total_errors=$((total_rx_errors + total_tx_errors))

        perfdata="${perfdata}'${iface}_rx_errors'=${rx_errs} '${iface}_rx_fifo'=${rx_fifo} '${iface}_rx_frame'=${rx_frame} "
        perfdata="${perfdata}'${iface}_tx_errors'=${tx_errs} '${iface}_tx_fifo'=${tx_fifo} '${iface}_tx_carrier'=${tx_carrier} "

        if [ $total_errors -ge $CRITICAL_THRESHOLD ]; then
            problem_nics+=("$iface: CRITICAL (RX_errs:$rx_errs RX_fifo:$rx_fifo RX_frame:$rx_frame TX_errs:$tx_errs TX_fifo:$tx_fifo TX_carrier:$tx_carrier)")
            exit_code=$STATE_CRITICAL
        elif [ $total_errors -ge $WARNING_THRESHOLD ] && [ $exit_code -lt $STATE_CRITICAL ]; then
            problem_nics+=("$iface: WARNING (RX_errs:$rx_errs RX_fifo:$rx_fifo RX_frame:$rx_frame TX_errs:$tx_errs TX_fifo:$tx_fifo TX_carrier:$tx_carrier)")
            exit_code=$STATE_WARNING
        fi
    fi
done < /proc/net/dev

if [ ${#problem_nics[@]} -gt 0 ]; then
    if [ $exit_code -eq $STATE_CRITICAL ]; then
        output_msg="CRITICAL - NIC errors detected: "
    else
        output_msg="WARNING - NIC errors detected: "
    fi

    for ((i=0; i<${#problem_nics[@]}; i++)); do
        if [ $i -eq 0 ]; then
            output_msg="${output_msg}${problem_nics[$i]}"
        else
            output_msg="${output_msg}; ${problem_nics[$i]}"
        fi
    done
fi

echo "${output_msg} | ${perfdata}"
exit $exit_code