#!/bin/bash
################################################################################
# Script Name: crowdsec-exporter.sh
# Version: 1.0
# Description: Prometheus exporter for CrowdSec providing supplementary
#              operational metrics from cscli commands — active decisions,
#              alerts, bouncers, machines, hub items, and threat analysis
#
# Author: Phil Connor
# Contact: contact@mylinux.work
# Website: https://mylinux.work
# License: MIT
#
# Note: CrowdSec has a built-in Prometheus endpoint at port 6060 for internal
#       metrics (bucket counts, parser hits, etc.). This exporter provides
#       SUPPLEMENTARY operational metrics from cscli commands.
#
# Prerequisites:
#   - CrowdSec installed and running
#   - cscli command available
#   - jq for JSON parsing
#   - Root/sudo access
#   - netcat (nc) for HTTP mode
#   - curl for --grab-local mode
#
# Usage:
#   # Output to stdout
#   sudo ./crowdsec-exporter.sh
#
#   # HTTP server mode
#   sudo ./crowdsec-exporter.sh --http -p 9192
#
#   # Textfile collector mode
#   sudo ./crowdsec-exporter.sh --textfile
#
# Metrics Exported:
#   - crowdsec_up - Exporter status (1=up, 0=down)
#   - crowdsec_info{version,exporter_version} - CrowdSec version info
#   - crowdsec_decisions_active - Total active decisions
#   - crowdsec_decisions_active_by_type{type} - Active decisions by type
#   - crowdsec_decisions_active_by_origin{origin} - Active decisions by origin
#   - crowdsec_decisions_active_by_scenario{scenario} - Active decisions by scenario
#   - crowdsec_alerts_total - Total alerts
#   - crowdsec_alerts_per_period{period} - Alerts in 1h/24h
#   - crowdsec_top_attacker_decisions{ip} - Top 5 IPs by decision count
#   - crowdsec_top_scenario_alerts{scenario} - Top 5 scenarios by alert count
#   - crowdsec_bouncer_up{name} - Per-bouncer registered status
#   - crowdsec_bouncer_last_pull_timestamp{name} - Per-bouncer last pull time
#   - crowdsec_machine_up{name} - Machine registration status
#   - crowdsec_lapi_up - LAPI health status
#   - crowdsec_hub_items{type} - Installed hub items per type
#   - crowdsec_exporter_duration_seconds - Script execution time
#   - crowdsec_exporter_last_run_timestamp - Last run timestamp
#
# Configuration:
#   Default HTTP port: 9192
#   Textfile directory: /var/lib/node_exporter
#
################################################################################

# ============================================================================
# CONFIGURATION VARIABLES
# ============================================================================

TEXTFILE_DIR="/var/lib/node_exporter"
OUTPUT_FILE=""
HTTP_MODE=false
HTTP_PORT=9192
GRAB_LOCAL=false
LOCAL_PORT=6060

# ============================================================================
# HELPER FUNCTIONS
# ============================================================================

show_usage() {
    cat <<EOF
Usage: $0 [OPTIONS]

Export CrowdSec operational statistics as Prometheus metrics.

MODES:
    --textfile      Write to node_exporter textfile collector
    --http          Run HTTP server on port $HTTP_PORT

OPTIONS:
    -p, --port        HTTP port (default: 9192)
    -o, --output      Output file path
    --grab-local      Also fetch built-in metrics from localhost:6060
    --local-port      Port for built-in metrics (default: 6060)

EXAMPLES:
    sudo $0 --textfile                    # Write to textfile collector
    sudo $0 --textfile --grab-local       # Include built-in 6060 metrics
    sudo $0 --http --port 9192            # Run HTTP server
    sudo $0 -o /tmp/crowdsec.prom         # Write to custom file

NOTE:
    CrowdSec exposes internal metrics (bucket counts, parser hits) on
    port 6060 by default. Use --grab-local to include those metrics in
    this exporter's output, avoiding the need for a separate scrape job.

EOF
    exit 0
}

parse_args() {
    while [[ $# -gt 0 ]]; do
        case $1 in
            -h|--help) show_usage ;;
            --textfile) OUTPUT_FILE="$TEXTFILE_DIR/crowdsec.prom"; shift ;;
            --http) HTTP_MODE=true; shift ;;
            -p|--port) HTTP_PORT="$2"; shift 2 ;;
            -o|--output) OUTPUT_FILE="$2"; shift 2 ;;
            --grab-local) GRAB_LOCAL=true; shift ;;
            --local-port) LOCAL_PORT="$2"; shift 2 ;;
            *) echo "Unknown option: $1" >&2; exit 1 ;;
        esac
    done
}

# Check if CrowdSec is installed and responding
# Returns: 0 if OK, 1 if error
check_crowdsec() {
    if ! command -v cscli >/dev/null 2>&1; then
        echo "ERROR: cscli command not found" >&2
        return 1
    fi

    if ! command -v jq >/dev/null 2>&1; then
        echo "ERROR: jq not found (required for JSON parsing)" >&2
        return 1
    fi

    # Verify LAPI is responding
    if ! cscli lapi status >/dev/null 2>&1; then
        echo "ERROR: CrowdSec LAPI not responding" >&2
        return 1
    fi

    return 0
}

# Escape special characters in Prometheus label values
# Args: $1 - string to escape
# Returns: escaped string safe for Prometheus labels
prom_escape() {
    local val="$1"
    val="${val//\\/\\\\}"
    val="${val//\"/\\\"}"
    val="${val//$'\n'/}"
    echo "$val"
}

# Get CrowdSec version string
# Returns: version string (e.g., "1.5.4")
get_crowdsec_version() {
    cscli version 2>/dev/null | grep -oE '[0-9]+\.[0-9]+\.[0-9]+' | head -1
}

# Get active decisions as JSON
# Returns: JSON array of active decisions, or "null" on error
get_decisions_json() {
    cscli decisions list -o json 2>/dev/null
}

# Get alerts as JSON
# Args: $1 - optional --since parameter (e.g., "1h")
# Returns: JSON array of alerts, or "null" on error
get_alerts_json() {
    local since="$1"
    if [ -n "$since" ]; then
        cscli alerts list --since "$since" -o json 2>/dev/null
    else
        cscli alerts list -o json 2>/dev/null
    fi
}

# Get bouncers as JSON
# Returns: JSON array of bouncers
get_bouncers_json() {
    cscli bouncers list -o json 2>/dev/null
}

# Get machines as JSON
# Returns: JSON array of machines
get_machines_json() {
    cscli machines list -o json 2>/dev/null
}

# Check LAPI health
# Returns: 1 if healthy, 0 if not
get_lapi_status() {
    if cscli lapi status >/dev/null 2>&1; then
        echo "1"
    else
        echo "0"
    fi
}

# Get hub items as JSON
# Returns: JSON output from cscli hub list
get_hub_json() {
    cscli hub list -o json 2>/dev/null
}

# ============================================================================
# METRIC GENERATION
# ============================================================================

# Generate all Prometheus metrics
# Returns: Prometheus text format metrics on stdout
generate_metrics() {
    local script_start
    script_start=$(date +%s)

    # Check CrowdSec status first
    if ! check_crowdsec; then
        cat <<EOF
# HELP crowdsec_up CrowdSec exporter status
# TYPE crowdsec_up gauge
crowdsec_up 0
EOF
        return
    fi

    cat <<EOF
# HELP crowdsec_up CrowdSec exporter status
# TYPE crowdsec_up gauge
crowdsec_up 1

# HELP crowdsec_info CrowdSec version information
# TYPE crowdsec_info gauge
crowdsec_info{version="$(get_crowdsec_version)",exporter_version="1.0"} 1
EOF

    echo ""

    # ========================================================================
    # DECISIONS METRICS
    # ========================================================================

    local decisions_json
    decisions_json=$(get_decisions_json)

    # Handle "null" or empty output from cscli (means no active decisions)
    local total_decisions=0
    if [ -n "$decisions_json" ] && [ "$decisions_json" != "null" ]; then
        total_decisions=$(echo "$decisions_json" | jq 'length' 2>/dev/null)
        total_decisions=${total_decisions:-0}
    fi

    cat <<EOF
# HELP crowdsec_decisions_active Total active decisions
# TYPE crowdsec_decisions_active gauge
crowdsec_decisions_active $total_decisions
EOF

    echo ""

    # Decisions by type (ban, captcha, throttle)
    cat <<EOF
# HELP crowdsec_decisions_active_by_type Active decisions by type
# TYPE crowdsec_decisions_active_by_type gauge
EOF

    if [ "$total_decisions" -gt 0 ] 2>/dev/null; then
        echo "$decisions_json" | jq -r '
            group_by(.type) | .[] |
            "\(.[0].type) \(length)"
        ' 2>/dev/null | while read -r dtype count; do
            [ -z "$dtype" ] && continue
            echo "crowdsec_decisions_active_by_type{type=\"$(prom_escape "$dtype")\"} $count"
        done
    fi

    echo ""

    # Decisions by origin (crowdsec, cscli, CAPI)
    cat <<EOF
# HELP crowdsec_decisions_active_by_origin Active decisions by origin
# TYPE crowdsec_decisions_active_by_origin gauge
EOF

    if [ "$total_decisions" -gt 0 ] 2>/dev/null; then
        echo "$decisions_json" | jq -r '
            group_by(.origin) | .[] |
            "\(.[0].origin) \(length)"
        ' 2>/dev/null | while read -r origin count; do
            [ -z "$origin" ] && continue
            echo "crowdsec_decisions_active_by_origin{origin=\"$(prom_escape "$origin")\"} $count"
        done
    fi

    echo ""

    # Decisions by scenario
    cat <<EOF
# HELP crowdsec_decisions_active_by_scenario Active decisions by scenario
# TYPE crowdsec_decisions_active_by_scenario gauge
EOF

    if [ "$total_decisions" -gt 0 ] 2>/dev/null; then
        echo "$decisions_json" | jq -r '
            group_by(.scenario) | .[] |
            "\(.[0].scenario) \(length)"
        ' 2>/dev/null | while read -r scenario count; do
            [ -z "$scenario" ] && continue
            echo "crowdsec_decisions_active_by_scenario{scenario=\"$(prom_escape "$scenario")\"} $count"
        done
    fi

    echo ""

    # Top 5 attackers by decision count
    cat <<EOF
# HELP crowdsec_top_attacker_decisions Top attacking IPs by active decision count
# TYPE crowdsec_top_attacker_decisions gauge
EOF

    if [ "$total_decisions" -gt 0 ] 2>/dev/null; then
        echo "$decisions_json" | jq -r '
            group_by(.value) | map({ip: .[0].value, count: length}) |
            sort_by(-.count) | .[0:5] | .[] |
            "\(.ip) \(.count)"
        ' 2>/dev/null | while read -r ip count; do
            [ -z "$ip" ] && continue
            echo "crowdsec_top_attacker_decisions{ip=\"$(prom_escape "$ip")\"} $count"
        done
    fi

    echo ""

    # ========================================================================
    # ALERTS METRICS
    # ========================================================================

    local alerts_json
    alerts_json=$(get_alerts_json)

    local total_alerts=0
    if [ -n "$alerts_json" ] && [ "$alerts_json" != "null" ]; then
        total_alerts=$(echo "$alerts_json" | jq 'length' 2>/dev/null)
        total_alerts=${total_alerts:-0}
    fi

    cat <<EOF
# HELP crowdsec_alerts_total Total alerts
# TYPE crowdsec_alerts_total gauge
crowdsec_alerts_total $total_alerts
EOF

    echo ""

    # Alerts per period (1h, 24h)
    local alerts_1h_json alerts_24h_json
    alerts_1h_json=$(get_alerts_json "1h")
    alerts_24h_json=$(get_alerts_json "24h")

    local alerts_1h=0
    if [ -n "$alerts_1h_json" ] && [ "$alerts_1h_json" != "null" ]; then
        alerts_1h=$(echo "$alerts_1h_json" | jq 'length' 2>/dev/null)
        alerts_1h=${alerts_1h:-0}
    fi

    local alerts_24h=0
    if [ -n "$alerts_24h_json" ] && [ "$alerts_24h_json" != "null" ]; then
        alerts_24h=$(echo "$alerts_24h_json" | jq 'length' 2>/dev/null)
        alerts_24h=${alerts_24h:-0}
    fi

    cat <<EOF
# HELP crowdsec_alerts_per_period Alerts in time period
# TYPE crowdsec_alerts_per_period gauge
crowdsec_alerts_per_period{period="1h"} $alerts_1h
crowdsec_alerts_per_period{period="24h"} $alerts_24h
EOF

    echo ""

    # Top 5 scenarios by alert count
    cat <<EOF
# HELP crowdsec_top_scenario_alerts Top scenarios by alert count
# TYPE crowdsec_top_scenario_alerts gauge
EOF

    if [ "$total_alerts" -gt 0 ] 2>/dev/null; then
        echo "$alerts_json" | jq -r '
            group_by(.scenario) | map({scenario: .[0].scenario, count: length}) |
            sort_by(-.count) | .[0:5] | .[] |
            "\(.scenario) \(.count)"
        ' 2>/dev/null | while read -r scenario count; do
            [ -z "$scenario" ] && continue
            echo "crowdsec_top_scenario_alerts{scenario=\"$(prom_escape "$scenario")\"} $count"
        done
    fi

    echo ""

    # ========================================================================
    # BOUNCER METRICS
    # ========================================================================

    local bouncers_json
    bouncers_json=$(get_bouncers_json)

    cat <<EOF
# HELP crowdsec_bouncer_up Bouncer registration status (1=valid, 0=revoked)
# TYPE crowdsec_bouncer_up gauge
EOF

    if [ -n "$bouncers_json" ] && [ "$bouncers_json" != "null" ]; then
        echo "$bouncers_json" | jq -r '
            .[] | "\(.name) \(if .revoked then 0 else 1 end)"
        ' 2>/dev/null | while read -r name status; do
            [ -z "$name" ] && continue
            echo "crowdsec_bouncer_up{name=\"$(prom_escape "$name")\"} $status"
        done
    fi

    echo ""

    cat <<EOF
# HELP crowdsec_bouncer_last_pull_timestamp Bouncer last pull Unix timestamp
# TYPE crowdsec_bouncer_last_pull_timestamp gauge
EOF

    if [ -n "$bouncers_json" ] && [ "$bouncers_json" != "null" ]; then
        echo "$bouncers_json" | jq -r '
            .[] | "\(.name) \(.last_pull)"
        ' 2>/dev/null | while read -r name last_pull; do
            [ -z "$name" ] && continue
            # Convert ISO timestamp to Unix epoch
            local ts
            ts=$(date -d "$last_pull" +%s 2>/dev/null || echo "0")
            echo "crowdsec_bouncer_last_pull_timestamp{name=\"$(prom_escape "$name")\"} $ts"
        done
    fi

    echo ""

    # ========================================================================
    # MACHINE METRICS
    # ========================================================================

    local machines_json
    machines_json=$(get_machines_json)

    cat <<EOF
# HELP crowdsec_machine_up Machine registration status (1=validated, 0=not validated)
# TYPE crowdsec_machine_up gauge
EOF

    if [ -n "$machines_json" ] && [ "$machines_json" != "null" ]; then
        echo "$machines_json" | jq -r '
            .[] | "\(.machineId) \(if .isValidated then 1 else 0 end)"
        ' 2>/dev/null | while read -r name status; do
            [ -z "$name" ] && continue
            echo "crowdsec_machine_up{name=\"$(prom_escape "$name")\"} $status"
        done
    fi

    echo ""

    # ========================================================================
    # LAPI HEALTH
    # ========================================================================

    local lapi_status
    lapi_status=$(get_lapi_status)

    cat <<EOF
# HELP crowdsec_lapi_up LAPI health status (1=healthy, 0=unhealthy)
# TYPE crowdsec_lapi_up gauge
crowdsec_lapi_up $lapi_status
EOF

    echo ""

    # ========================================================================
    # HUB ITEMS
    # ========================================================================

    local hub_json
    hub_json=$(get_hub_json)

    cat <<EOF
# HELP crowdsec_hub_items Installed hub items by type
# TYPE crowdsec_hub_items gauge
EOF

    if [ -n "$hub_json" ] && [ "$hub_json" != "null" ]; then
        local collections parsers scenarios postoverflows

        collections=$(echo "$hub_json" | jq '[.collections // [] | .[] | select(.installed == true)] | length' 2>/dev/null)
        parsers=$(echo "$hub_json" | jq '[.parsers // [] | .[] | select(.installed == true)] | length' 2>/dev/null)
        scenarios=$(echo "$hub_json" | jq '[.scenarios // [] | .[] | select(.installed == true)] | length' 2>/dev/null)
        postoverflows=$(echo "$hub_json" | jq '[.postoverflows // [] | .[] | select(.installed == true)] | length' 2>/dev/null)

        echo "crowdsec_hub_items{type=\"collections\"} ${collections:-0}"
        echo "crowdsec_hub_items{type=\"parsers\"} ${parsers:-0}"
        echo "crowdsec_hub_items{type=\"scenarios\"} ${scenarios:-0}"
        echo "crowdsec_hub_items{type=\"postoverflows\"} ${postoverflows:-0}"
    else
        echo "crowdsec_hub_items{type=\"collections\"} 0"
        echo "crowdsec_hub_items{type=\"parsers\"} 0"
        echo "crowdsec_hub_items{type=\"scenarios\"} 0"
        echo "crowdsec_hub_items{type=\"postoverflows\"} 0"
    fi

    echo ""

    # ========================================================================
    # BUILT-IN METRICS (optional, via --grab-local)
    # ========================================================================

    if [ "$GRAB_LOCAL" = true ]; then
        local builtin_metrics
        builtin_metrics=$(curl -sf --max-time 5 "http://localhost:${LOCAL_PORT}/metrics" 2>/dev/null)

        if [ -n "$builtin_metrics" ]; then
            echo "# ================================================================"
            echo "# CrowdSec built-in metrics from localhost:${LOCAL_PORT}"
            echo "# ================================================================"
            echo "$builtin_metrics"
            echo ""
        else
            echo "# WARNING: Failed to fetch built-in metrics from localhost:${LOCAL_PORT}"
            echo ""
        fi
    fi

    # ========================================================================
    # EXPORTER RUNTIME
    # ========================================================================

    local script_end script_duration
    script_end=$(date +%s)
    script_duration=$((script_end - script_start))

    cat <<EOF
# HELP crowdsec_exporter_duration_seconds Time to generate all metrics
# TYPE crowdsec_exporter_duration_seconds gauge
crowdsec_exporter_duration_seconds $script_duration

# HELP crowdsec_exporter_last_run_timestamp Unix timestamp of last successful run
# TYPE crowdsec_exporter_last_run_timestamp gauge
crowdsec_exporter_last_run_timestamp $script_end
EOF

    echo ""
}

# ============================================================================
# HTTP SERVER MODE
# ============================================================================

# Run simple HTTP server using netcat
# Serves metrics on /metrics endpoint
run_http_server() {
    echo "Starting CrowdSec exporter on port $HTTP_PORT..." >&2

    if ! command -v nc >/dev/null 2>&1; then
        echo "ERROR: netcat (nc) required for HTTP mode" >&2
        exit 1
    fi

    # Infinite loop accepting HTTP requests
    while true; do
        {
            read -r request
            # Check if request is for /metrics endpoint
            if [[ "$request" =~ ^GET\ /metrics ]]; then
                echo -e "HTTP/1.1 200 OK\r\nContent-Type: text/plain; version=0.0.4\r\n\r"
                generate_metrics
            else  # Serve HTML landing page for other requests
                echo -e "HTTP/1.1 200 OK\r\nContent-Type: text/html\r\n\r"
                cat <<EOF
<!DOCTYPE html>
<html>
<head><title>CrowdSec Exporter v1.0</title></head>
<body>
<h1>CrowdSec Prometheus Exporter v1.0</h1>
<p><a href="/metrics">Metrics</a></p>
<p>Supplementary operational metrics from cscli commands.</p>
<p>For internal CrowdSec metrics (buckets, parsers), see port 6060.</p>
</body>
</html>
EOF
            fi
        } | nc -l -p "$HTTP_PORT" -q 1 2>/dev/null
    done
}

# ============================================================================
# MAIN EXECUTION
# ============================================================================

# Main entry point - routes to appropriate output mode
main() {
    parse_args "$@"

    if [ "$HTTP_MODE" = true ]; then
        # Run HTTP server (blocks until killed)
        run_http_server
    elif [ -n "$OUTPUT_FILE" ]; then
        # Textfile collector mode: write atomically using temp file
        local output_dir
        output_dir="$(dirname "$OUTPUT_FILE")"
        mkdir -p "$output_dir"

        # Create temp file in SAME directory for atomic rename (same filesystem)
        local temp_file
        temp_file=$(mktemp "${output_dir}/.crowdsec_metrics.XXXXXX")

        # Generate metrics to temp file
        if ! generate_metrics > "$temp_file" 2>/dev/null; then
            rm -f "$temp_file"
            echo "ERROR: Failed to generate metrics" >&2
            exit 1
        fi

        # Validate: file must exist, have content
        local file_lines
        file_lines=$(wc -l < "$temp_file" 2>/dev/null || echo 0)

        if [ "$file_lines" -lt 10 ]; then
            rm -f "$temp_file"
            echo "ERROR: Metrics file too small ($file_lines lines), keeping previous" >&2
            exit 1
        fi

        # Set permissions before move
        chmod 644 "$temp_file"

        # Atomic rename - no gap where file is missing
        mv -f "$temp_file" "$OUTPUT_FILE"

        echo "Metrics written to $OUTPUT_FILE ($file_lines lines)" >&2
    else
        # Default: output to stdout
        generate_metrics
    fi
}

# Execute main function with all script arguments
main "$@"