#!/bin/bash

VERSION="1.1.0"

###################
# Color Definitions
###################
NC="\033[00m"
GREEN="\033[01;32m"
RED="\033[01;31m"
YELLOW="\033[01;33m"

###################
# Utility Functions
###################

print_header() {
    echo "
    ____                  ____              
   / __ \_________  _  __/ __ \____  _____ 
  / /_/ / ___/ __ \| |/_/ / / / __ \/ ___/ 
 / ____/ /  / /_/ />  </ /_/ / /_/ / /__   
/_/   /_/   \____/_/|_/_____/\____/\___/   
                                           
 The Proxmox System Doctor
 Version ${VERSION}
======================================"
    echo "Started at: $(date '+%Y-%m-%d %H:%M:%S')"
    echo "Examining patient: $(hostname)"
    echo "======================================"
}


handle_error() {
    echo -e "${RED}Error: $1${NC}"
    exit 1
}

log_message() {
    local level=$1
    local message=$2
    case $level in
        info)    echo -e "${GREEN}[INFO]${NC} $message" ;;
        warn)    echo -e "${YELLOW}[WARN]${NC} $message" ;;
        error)   echo -e "${RED}[ERROR]${NC} $message" ;;
    esac
}

check_requirements() {
    log_message info "Checking medical equipment..."
    local tools=("dmidecode" "lscpu" "ip" "smartctl" "sensors" "lspci" "bc")
    local missing=()
    
    for tool in "${tools[@]}"; do
        if ! command -v "$tool" >/dev/null 2>&1; then
            missing+=("$tool")
        fi
    done
    
    if [[ ${#missing[@]} -gt 0 ]]; then
        handle_error "Missing tools: ${missing[*]}\n Please install with 'curl -s http://10.10.10.63:3000/LotusGuild/freshStartScript/raw/branch/main/freshStart.sh | bash'"
    fi
}


checkIfOnHypervisor() {
    command -v pveversion >/dev/null 2>&1
}

###################
# System Information Functions
###################
get_system_info() {
    echo -e "\n${GREEN}=== System Information ===${NC}"
    echo -e "\n${GREEN}=== Diagnostic Run: $(date '+%Y-%m-%d %H:%M:%S') ===${NC}"
    echo -e "${GREEN}Hostname:${NC} $(uname -n)"
    echo -e "${GREEN}Kernel:${NC} $(uname -r)"
    
    if checkIfOnHypervisor; then
        echo -e "\n${GREEN}=== Proxmox Version ===${NC}"
        pveversion
    else
        echo -e "\n${GREEN}=== OS Information ===${NC}"
        if [[ -f /etc/os-release ]]; then
            source /etc/os-release
            echo -e "${GREEN}Distribution:${NC} $PRETTY_NAME"
        else
            echo "OS information not available"
        fi
    fi
}

get_temp_info() {
    echo -e "\n${GREEN}=== Temperature Information ===${NC}"
    if command -v sensors >/dev/null 2>&1; then
        sensors
    else
        log_message warn "sensors command not found. Install lm-sensors package for temperature monitoring"
    fi
}

get_disk_health() {
    echo -e "\n${GREEN}=== Disk Health Status ===${NC}"
    if command -v smartctl >/dev/null 2>&1; then
        for disk in $(lsblk -d -o name | grep -E '^sd|^nvme'); do
            echo -e "\nChecking /dev/$disk:"
            smartctl -H /dev/$disk
        done
    else
        log_message warn "smartctl not found. Install smartmontools for disk health monitoring"
    fi
}

get_cpu_info() {
    cpu_info=$(grep -m 1 -w 'model name' /proc/cpuinfo | awk -F: '{print $2}' | xargs) || {
        echo -e "${RED}Failed to retrieve CPU model information.${NC}"
    }
    cpu_cores=$(lscpu | grep '^CPU(s):' | awk '{print $2}')
    cpu_mhz=$(lscpu | grep 'MHz' | awk '{print $4}')
    
    echo -e "${GREEN}CPU Model:${NC} $cpu_info"
    echo -e "${GREEN}CPU Cores:${NC} $cpu_cores"
    echo -e "${GREEN}CPU MHz:${NC} $cpu_mhz"
}

get_ram_info() {
    ram_total=$(free -h | grep 'Mem:' | awk '{print $2}')
    ram_used=$(free -h | grep 'Mem:' | awk '{print $3}')
    ram_free=$(free -h | grep 'Mem:' | awk '{print $4}')
    
    echo -e "${GREEN}Total RAM:${NC} $ram_total"
    echo -e "${GREEN}Used RAM:${NC} $ram_used"
    echo -e "${GREEN}Free RAM:${NC} $ram_free"
}

get_storage_info() {
    echo -e "${GREEN}Storage Information:${NC}"
    df -h --output=source,size,used,avail,pcent | grep '^/dev'
    
    if command -v zpool >/dev/null 2>&1; then
        echo -e "\n${GREEN}=== ZFS Pool Status ===${NC}"
        zpool status
    fi
}

get_network_info() {
    default_gateway=$(ip route | grep default | awk '{print $3}')
    ip_addresses=$(hostname -I | xargs)
    echo -e "${GREEN}Default Gateway:${NC} $default_gateway"
    echo -e "${GREEN}IP Addresses:${NC} $ip_addresses"
}

get_detailed_network() {
    echo -e "\n${GREEN}=== Network Interface Statistics ===${NC}"
    
    local iface
    while IFS= read -r iface; do
        [[ -z "$iface" ]] && continue
        ip -s link show "$iface" 2>/dev/null
    done < <(get_physical_interfaces)
    
    echo -e "\n${GREEN}=== Network Statistics ===${NC}"
    if command -v ss >/dev/null 2>&1; then
        ss -s
    elif command -v netstat >/dev/null 2>&1; then
        netstat -i
    else
        log_message warn "netstat/ss not found for network statistics"
    fi
}

get_hardware_info() {
    echo -e "${GREEN}BIOS Version:${NC} $(dmidecode -s bios-version)"
    echo -e "\n${GREEN}=== PCI Devices ===${NC}"
    # Show interesting devices, exclude bridges, infrastructure, and integrated motherboard devices
    lspci | grep -v -E "Host bridge|PCI bridge|ISA bridge|SMBus|IOMMU|Dummy|USB controller|Audio device|Encryption controller|Multimedia controller"
}

get_motherboard_info() {
    echo -e "\n${GREEN}=== Motherboard Information ===${NC}"
    echo -e "${GREEN}Manufacturer:${NC} $(dmidecode -s baseboard-manufacturer)"
    echo -e "${GREEN}Product Name:${NC} $(dmidecode -s baseboard-product-name)"
    echo -e "${GREEN}Version:${NC} $(dmidecode -s baseboard-version)"
    echo -e "${GREEN}Serial Number:${NC} $(dmidecode -s baseboard-serial-number)"
    echo -e "${GREEN}System Manufacturer:${NC} $(dmidecode -s system-manufacturer)"
    echo -e "${GREEN}System Product:${NC} $(dmidecode -s system-product-name)"
    echo -e "${GREEN}System Serial:${NC} $(dmidecode -s system-serial-number)"
}

get_memory_details() {
    echo -e "\n${GREEN}=== Memory DIMM Information ===${NC}"
    dmidecode -t memory | awk '
        /Memory Device/,/^$/ {
            if (/Size:/ && !/No Module Installed/) {
                size=$2" "$3
            }
            if (/Type:/ && !/Unknown/ && !/Error/) {
                type=$2
            }
            if (/Speed:/ && !/Unknown/ && $2 != "Unknown") {
                speed=$2" "$3
            }
            if (/Manufacturer:/ && !/Unknown/ && $2 != "Unknown") {
                mfr=$2
            }
            if (/Part Number:/ && !/Unknown/) {
                part=$3
            }
            if (/Locator:/ && !/Bank/) {
                loc=$2
                if (size && size !~ /No/) {
                    printf "%-12s %-10s %-8s %-12s %-20s\n", loc, size, type, speed, mfr
                    size=""; type=""; speed=""; mfr=""; part=""
                }
            }
        }
    '
    echo -e "\n${GREEN}Memory Summary:${NC}"
    # Count actual DIMM slots by looking for Locator entries with slot-like names (DIMM, BANK, ChannelA, etc.)
    # Filter out Bank Locator lines and count unique slot names
    local total_slots=$(dmidecode -t memory | grep -E "^\s+Locator:" | grep -v "Bank Locator" | wc -l)
    # Count populated slots - those with actual size values (not "No Module Installed" or "Not Installed")
    local populated=$(dmidecode -t memory | grep -E "^\s+Size:" | grep -v -E "No Module|Not Installed" | wc -l)
    echo -e "  Total Slots: $total_slots"
    echo -e "  Populated:   $populated"
    echo -e "  Max Capacity: $(dmidecode -t memory | grep "Maximum Capacity" | head -1 | awk '{print $3" "$4}')"
}

get_nic_details() {
    echo -e "\n${GREEN}=== Network Interface Details ===${NC}"
    
    local iface
    while IFS= read -r iface; do
        [[ -z "$iface" ]] && continue
        
        echo -e "\n${GREEN}Interface: $iface${NC}"

        # Get driver info
        if [[ -L "/sys/class/net/$iface/device/driver" ]]; then
            local driver
            driver=$(basename "$(readlink "/sys/class/net/$iface/device/driver")")
            echo -e "  Driver: $driver"
        fi

        # Get MAC address
        if [[ -f "/sys/class/net/$iface/address" ]]; then
            echo -e "  MAC: $(cat "/sys/class/net/$iface/address")"
        fi

        # Get link state
        if [[ -f "/sys/class/net/$iface/operstate" ]]; then
            echo -e "  State: $(cat "/sys/class/net/$iface/operstate")"
        fi

        # Use ethtool if available
        if command -v ethtool >/dev/null 2>&1; then
            # Get speed and duplex
            local link_info
            link_info=$(ethtool "$iface" 2>/dev/null | grep -E "Speed:|Duplex:|Link detected:")
            if [[ -n "$link_info" ]]; then
                echo "$link_info" | while IFS= read -r line; do
                    echo -e "  $line"
                done
            fi

            # Get firmware version
            local fw_ver
            fw_ver=$(ethtool -i "$iface" 2>/dev/null | grep "firmware-version" | awk '{print $2}')
            if [[ -n "$fw_ver" ]]; then
                echo -e "  Firmware: $fw_ver"
            fi
        fi
    done < <(get_physical_interfaces)
}

get_physical_interfaces() {
    local iface
    for iface in /sys/class/net/*; do
        # Skip if glob didn't match anything
        [[ -e "$iface" ]] || continue
        
        # Get just the interface name
        iface=$(basename "$iface")
        
        # Skip loopback
        [[ "$iface" == "lo" ]] && continue
        
        # Skip virtual/firewall interfaces
        [[ "$iface" =~ ^(veth|fwbr|fwln|fwpr|tap) ]] && continue
        
        # This is a physical interface
        echo "$iface"
    done
}

get_hba_info() {
    echo -e "\n${GREEN}=== HBA/Storage Controller Information ===${NC}"

    # Find RAID, SAS, SATA, SCSI, and storage controllers
    lspci -vmm 2>/dev/null | awk '
        BEGIN { RS=""; FS="\n" }
        /RAID|SAS|SATA|SCSI|Mass storage|Serial Attached|Fibre Channel|NVMe/ {
            for (i=1; i<=NF; i++) {
                if ($i ~ /^Slot:/) slot = substr($i, 7)
                if ($i ~ /^Class:/) class = substr($i, 8)
                if ($i ~ /^Vendor:/) vendor = substr($i, 9)
                if ($i ~ /^Device:/) device = substr($i, 9)
                if ($i ~ /^Rev:/) rev = substr($i, 6)
            }
            printf "\n%s\n", slot
            printf "  Class:  %s\n", class
            printf "  Vendor: %s\n", vendor
            printf "  Device: %s\n", device
            if (rev) printf "  Rev:    %s\n", rev
            slot=""; class=""; vendor=""; device=""; rev=""
        }
    '

    # Show detailed info for storage controllers
    echo -e "\n${GREEN}=== Storage Controller Details ===${NC}"
    for ctrl in $(lspci | grep -iE "RAID|SAS|SATA|SCSI|Mass storage|NVMe" | awk '{print $1}'); do
        echo -e "\n${GREEN}Controller $ctrl:${NC}"
        lspci -vvs "$ctrl" 2>/dev/null | grep -E "^\s+(Subsystem|LnkSta|Kernel driver)" | head -5
    done
}

get_system_status() {
    echo -e "\n${GREEN}=== System Load ===${NC}"
    uptime
    
    echo -e "\n${GREEN}=== Service Status ===${NC}"
    systemctl list-units --type=service --state=running | wc -l
    
    echo -e "\n${GREEN}=== Recent System Errors ===${NC}"
    journalctl -p err -n 5 --no-pager
}

###################
# DriveAtlas & Monitoring Functions
###################

get_drive_atlas() {
    echo -e "\n${GREEN}=== Drive Atlas - Physical Bay Mapping ===${NC}"
    if command -v curl >/dev/null 2>&1; then
        if ! bash <(curl -sL "http://10.10.10.63:3000/LotusGuild/driveAtlas/raw/branch/main/driveAtlas.sh") 2>/dev/null; then
            log_message warn "DriveAtlas failed to execute or server unavailable"
        fi
    else
        log_message warn "curl not installed - cannot fetch DriveAtlas"
    fi
}

get_ceph_health() {
    echo -e "\n${GREEN}=== Ceph Cluster Health ===${NC}"
    if command -v ceph >/dev/null 2>&1; then
        echo -e "${GREEN}Health Status:${NC}"
        ceph health detail 2>/dev/null || log_message warn "Cannot connect to Ceph cluster"

        echo -e "\n${GREEN}=== Ceph OSD Tree ===${NC}"
        ceph osd tree 2>/dev/null || true

        echo -e "\n${GREEN}=== Ceph Pool Usage ===${NC}"
        ceph df 2>/dev/null || true

        echo -e "\n${GREEN}=== Ceph OSD Usage ===${NC}"
        ceph osd df 2>/dev/null || true
    else
        log_message info "Ceph tools not installed on this node"
    fi
}

get_node_exporter_status() {
    echo -e "\n${GREEN}=== Node Exporter Status ===${NC}"
    if systemctl is-active --quiet node_exporter 2>/dev/null; then
        echo -e "${GREEN}Service:${NC} Running"
        local ip=$(hostname -I | awk '{print $1}')
        echo -e "${GREEN}Metrics URL:${NC} http://${ip}:9100/metrics"
        if ss -tlnp 2>/dev/null | grep -q ':9100'; then
            echo -e "${GREEN}Port 9100:${NC} Listening"
        else
            log_message warn "Port 9100 not listening"
        fi
    elif systemctl list-unit-files 2>/dev/null | grep -q node_exporter; then
        log_message warn "Node Exporter is installed but not running"
        echo -e "Start with: systemctl start node_exporter"
    else
        log_message info "Node Exporter not installed"
    fi
}

get_hwmon_status() {
    echo -e "\n${GREEN}=== hwmon Daemon Status ===${NC}"
    if systemctl is-active --quiet hwmon.timer 2>/dev/null; then
        echo -e "${GREEN}Timer:${NC} Active"
        systemctl list-timers hwmon.timer --no-pager 2>/dev/null
        echo -e "\n${GREEN}Last Run:${NC}"
        journalctl -u hwmon.service -n 3 --no-pager 2>/dev/null || true
    elif systemctl list-unit-files 2>/dev/null | grep -q hwmon.timer; then
        log_message warn "hwmon timer is installed but not active"
        echo -e "Enable with: systemctl enable --now hwmon.timer"
    else
        log_message info "hwmon daemon not installed"
    fi
}

quick_health_check() {
    echo -e "\n${GREEN}=== Quick Health Check ===${NC}"
    echo -e "Running quick health assessment...\n"

    # Services
    check_services

    # Temperatures
    get_temp_info

    # Disk health (quick)
    echo -e "\n${GREEN}=== Disk Health Summary ===${NC}"
    if command -v smartctl >/dev/null 2>&1; then
        for disk in $(lsblk -d -o name | grep -E '^sd|^nvme'); do
            health=$(smartctl -H /dev/$disk 2>/dev/null | grep -i "health" | awk -F: '{print $2}' | xargs)
            if [[ -n "$health" ]]; then
                echo -e "/dev/$disk: $health"
            fi
        done
    fi

    # Node Exporter
    get_node_exporter_status

    # Ceph quick status
    if command -v ceph >/dev/null 2>&1; then
        echo -e "\n${GREEN}=== Ceph Quick Status ===${NC}"
        ceph health 2>/dev/null || true
    fi
}

###################
# Proxmox Specific Functions
###################
check_services() {
    if ! checkIfOnHypervisor; then
        log_message warn "Not on Proxmox - skipping Proxmox service checks"
        return 0
    fi
    
    echo -e "${GREEN}Checking critical services:${NC}"
    local services=("pvedaemon" "pveproxy" "pvecluster" "pve-cluster" "corosync")
    for service in "${services[@]}"; do
        local status
        status=$(systemctl is-active "$service" 2>/dev/null || echo "not-found")
        echo -e "${GREEN}$service:${NC} $status"
    done
}

check_pve_version() {
    local min_version="6.0"
    local current_version=$(pveversion | grep -oP 'pve-manager/\K[0-9]+\.[0-9]+' || echo "0.0")
    if (( $(echo "$current_version < $min_version" | bc -l) )); then
        log_message warn "Proxmox VE version $current_version may not support all features"
    fi
}

list_vms() {
    if ! checkIfOnHypervisor; then
        log_message info "Not on Proxmox - skipping VM list"
        return 0
    fi
    
    if command -v qm >/dev/null 2>&1; then
        echo -e "${GREEN}Virtual Machine Status:${NC}"
        qm list
    else
        log_message warn "qm command not found"
    fi
}

list_containers() {
    if ! checkIfOnHypervisor; then
        log_message info "Not on Proxmox - skipping container list"
        return 0
    fi
    
    if command -v pct >/dev/null 2>&1; then
        echo -e "\n${GREEN}=== LXC Container Status ===${NC}"
        pct list
    else
        log_message warn "pct command not found"
    fi
}

###################
# Command Line Interface Functions
###################
help() {
    echo "ProxDoc - The Proxmox System Doctor v${VERSION}"
    echo ""
    echo "Usage: curl -sL \"http://10.10.10.63:3000/LotusGuild/proxDoc/raw/branch/main/proxDoc.sh\" | bash -s -- [OPTION]"
    echo ""
    echo "A comprehensive diagnostic tool for Proxmox server health checks."
    echo ""
    echo "Treatment Options:"
    echo "  --help           Show this prescription guide"
    echo "  --diags          Perform full system examination"
    echo "  --quick          Quick health check (services, temps, disks)"
    echo "  --drives         Show physical drive bay mapping (DriveAtlas)"
    echo "  --ceph           Check Ceph cluster health"
    echo "  --node-exporter  Check Node Exporter status"
    echo "  --hwmon          Check hwmon daemon status"
    echo "  --services       Check vital Proxmox services"
    echo "  --vm-list        Check VM vitals"
    echo "  --ct-list        Check container vitals"
    echo "  --backup         Review backup health"
    exit 0
}




###################
# Main Functions
###################
runDiags() {
    log_message info "Beginning system examination..."
    
    # Check if running on Proxmox
    local is_proxmox=false
    if checkIfOnHypervisor; then
        is_proxmox=true
        log_message info "Detected Proxmox VE hypervisor"
    else
        log_message warn "Not running on Proxmox VE - some checks will be skipped"
    fi
    
    echo ""
    
    log_message info "Checking system information..."
    get_system_info
    
    log_message info "Checking CPU..."
    get_cpu_info
    
    log_message info "Checking RAM..."
    get_ram_info
    
    log_message info "Checking memory details..."
    get_memory_details
    
    log_message info "Checking storage..."
    get_storage_info
    
    log_message info "Checking drive atlas..."
    get_drive_atlas
    
    log_message info "Checking network..."
    get_network_info
    get_detailed_network
    get_nic_details
    
    log_message info "Checking hardware..."
    get_hardware_info
    get_motherboard_info
    get_hba_info
    
    log_message info "Checking temperatures..."
    get_temp_info
    
    log_message info "Checking system status..."
    get_system_status
    
    log_message info "Checking monitoring services..."
    get_node_exporter_status
    get_hwmon_status
    
    # Only run Proxmox-specific checks if on Proxmox
    if [[ "$is_proxmox" == true ]]; then
        log_message info "Checking Ceph cluster..."
        get_ceph_health
        
        log_message info "Checking VMs..."
        list_vms
        
        log_message info "Checking containers..."
        list_containers
    fi
    
    echo ""
    log_message info "Examination complete"
}

checkForInput() {
    case $1 in
        --help)          help ;;
        --diags)         check_requirements; runDiags ;;
        --quick)         quick_health_check ;;
        --drives)        get_drive_atlas ;;
        --ceph)          get_ceph_health ;;
        --node-exporter) get_node_exporter_status ;;
        --hwmon)         get_hwmon_status ;;
        --services)      check_services ;;
        --vm-list)       list_vms ;;
        --ct-list)       list_containers ;;
        --backup)        echo -e "${GREEN}Backup Status:${NC}"; pvesm status 2>/dev/null || log_message warn "pvesm not available" ;;
        *)               echo -e "${RED}Invalid option: $1${NC}"; help ;;
    esac
}


###################
# Script Execution
###################
argOne=$1

# Show header
print_header

# Check root
if [[ $EUID -ne 0 ]]; then
    handle_error "This script must be run as root"
fi

# Set trap for interrupts
trap 'echo -e "${RED}Script interrupted.${NC}"; exit 1' INT TERM

if [[ -n $argOne ]]; then
    checkForInput "$argOne"
else
    help
fi