citadel/backup/restore

#!/bin/bash

# Restic backup restore script
# Location: /home/citadel/backup/restore
# Usage: ./restore <service_name> [--test|--production]

set -euo pipefail  # Strict mode: exit on error, undefined vars, pipe failures

# Configuration
readonly BACKUP_DIR="/home/citadel/backup"
readonly SERVICES_DIR="/home/citadel/services"
readonly CONFIG_FILE="$BACKUP_DIR/restic.conf"
readonly SCRIPT_NAME=$(basename "$0")
readonly VERSION="2.1.0"

# Global variables
USER_CHOICE=""

# Colors
readonly RED='\033[0;31m'
readonly GREEN='\033[0;32m'
readonly YELLOW='\033[1;33m'
readonly BLUE='\033[0;34m'
readonly NC='\033[0m'

# Logging functions
log() { echo -e "${BLUE}[$(date '+%H:%M:%S')]${NC} $1"; }
error() { echo -e "${RED}[ERROR]${NC} $1" >&2; exit 1; }
success() { echo -e "${GREEN}[SUCCESS]${NC} $1"; }
warning() { echo -e "${YELLOW}[WARNING]${NC} $1"; }

# Check dependencies
check_dependencies() {
    local deps=("restic" "jq" "awk" "grep")
    local missing=()

    for dep in "${deps[@]}"; do
        if ! command -v "$dep" &>/dev/null; then
            missing+=("$dep")
        fi
    done

    if [ ${#missing[@]} -ne 0 ]; then
        error "Missing dependencies: ${missing[*]}"
    fi
}

# Show help
show_help() {
    cat << EOF
Usage: $SCRIPT_NAME <service_name> [OPTIONS]

Options:
    --test          Restore to test instance (default)
    --production    Restore to production instance
    -h, --help      Show this help message
    -v, --version   Show version

Examples:
    $SCRIPT_NAME paperless              # Restore in test mode (default)
    $SCRIPT_NAME paperless --test       # Restore to test instance
    $SCRIPT_NAME paperless --production # Restore to production (requires confirmation)

Available services:
EOF

    if [ -f "$CONFIG_FILE" ]; then
        # Secure validation of config file
        if grep -qE '^[^#]*export\s+RESTIC_' "$CONFIG_FILE"; then
            # shellcheck source=/dev/null
            source "$CONFIG_FILE"

            # List services from snapshots
            if restic snapshots &>/dev/null; then
                restic snapshots --json 2>/dev/null | \
                    jq -r '.[].tags[]' 2>/dev/null | \
                    grep -v -E '^(daily|weekly|monthly|yearly)$' | \
                    sort -u | sed 's/^/  /' || echo "  (cannot list services)"
            else
                echo "  (cannot access repository)"
            fi
        else
            error "Invalid configuration file format"
        fi
    else
        echo "  (no config found)"
    fi
}

# Check prerequisites
check_prerequisites() {
    # Check configuration file
    if [ ! -f "$CONFIG_FILE" ]; then
        error "Config file not found: $CONFIG_FILE"
    fi

    # Check config file permissions
    if [ ! -r "$CONFIG_FILE" ]; then
        error "Cannot read config file: $CONFIG_FILE"
    fi

    # Secure validation before source
    if ! grep -qE '^[^#]*export\s+RESTIC_' "$CONFIG_FILE"; then
        error "Invalid configuration file format"
    fi

    # Source config
    # shellcheck source=/dev/null
    source "$CONFIG_FILE"

    # Check required variables
    if [ -z "${RESTIC_REPOSITORY:-}" ]; then
        error "RESTIC_REPOSITORY not defined in config"
    fi

    # Test restic access
    if ! restic snapshots &>/dev/null; then
        error "Cannot access Restic repository"
    fi

    log "Prerequisites OK"
}

# Show available snapshots
show_snapshots() {
    local service="$1"

    # Validate service name
    if [[ ! "$service" =~ ^[a-zA-Z0-9_-]+$ ]]; then
        error "Invalid service name format"
    fi

    echo ""
    log "Available snapshots for $service:"
    echo ""

    # Get snapshots with error handling
    local snapshots_json
    if ! snapshots_json=$(restic snapshots --tag "$service" --json 2>/dev/null); then
        error "Failed to retrieve snapshots"
    fi

    if [ -z "$snapshots_json" ] || [ "$snapshots_json" = "[]" ]; then
        warning "No snapshots found for service: $service"
        return 1
    fi

    # Format and display snapshots
    echo "$snapshots_json" | jq -r '.[] | "\(.short_id)  \(.time[:19])  \(.summary.total_files_processed // "?") files"' | \
        awk 'BEGIN {printf "%-12s %-20s %s\n", "ID", "DATE", "FILES"; print "----------------------------------------"}
             {printf "%-12s %-20s %s\n", $1, $2, $3" "$4}'

    echo ""
    return 0
}

# Get user choice
get_user_choice() {
    local max_attempts=3
    local attempt=0

    while [ $attempt -lt $max_attempts ]; do
        echo ""
        echo "Enter snapshot ID (or 'latest' for most recent, 'q' to quit):"
        echo -n "> "
        read -r USER_CHOICE

        echo ""

        # Basic validation
        if [ "$USER_CHOICE" = "q" ] || [ "$USER_CHOICE" = "quit" ]; then
            log "Operation cancelled by user"
            exit 0
        fi

        if [ -n "$USER_CHOICE" ]; then
            # Format validation (alphanumeric or "latest")
            if [[ "$USER_CHOICE" =~ ^[a-zA-Z0-9]+$ ]] || [ "$USER_CHOICE" = "latest" ]; then
                return 0
            else
                warning "Invalid snapshot ID format. Please use alphanumeric characters only."
            fi
        else
            warning "No selection made"
        fi

        ((attempt++))
    done

    error "Maximum attempts exceeded"
}

# Resolve snapshot ID
resolve_snapshot_id() {
    local service="$1"
    local choice="$2"

    if [ "$choice" = "latest" ]; then
        local latest_id
        latest_id=$(restic snapshots --tag "$service" --json 2>/dev/null | jq -r '.[-1].short_id // empty')

        if [ -z "$latest_id" ]; then
            error "No snapshots found for service: $service"
        fi

        echo "$latest_id"
    else
        echo "$choice"
    fi
}

# Verify snapshot
verify_snapshot() {
    local service="$1"
    local snapshot_id="$2"

    log "Verifying snapshot $snapshot_id for service $service"

    # Get snapshot info
    local snapshot_info
    snapshot_info=$(restic snapshots --tag "$service" --json 2>/dev/null | \
        jq -r ".[] | select(.short_id == \"$snapshot_id\")")

    if [ -z "$snapshot_info" ]; then
        error "Snapshot $snapshot_id not found for service $service"
    fi

    # Extract date
    local date
    date=$(echo "$snapshot_info" | jq -r '.time[:19]')

    # Check integrity (optional but recommended)
    if restic check --read-data-subset=1/10 "$snapshot_id" &>/dev/null; then
        success "Found snapshot $snapshot_id from $date (integrity check passed)"
    else
        warning "Found snapshot $snapshot_id from $date (integrity check skipped)"
    fi
}

# Check service capabilities
check_service_capabilities() {
    local service="$1"
    local service_dir="$SERVICES_DIR/$service"

    echo ""
    log "Checking service capabilities for $service"

    # Check service directory exists
    if [ ! -d "$service_dir" ]; then
        error "Service directory not found: $service_dir"
    fi

    # Check for custom restore script
    local custom_restore="$service_dir/restore"
    local has_custom_restore=false
    if [ -f "$custom_restore" ] && [ -x "$custom_restore" ]; then
        has_custom_restore=true
        success "Found custom restore script: $custom_restore"
    else
        log "No custom restore script found - will extract to temporary directory"
    fi

    echo "$has_custom_restore"
}

# Create secure temporary directory
create_secure_temp_dir() {
    local prefix="${1:-restore}"
    local temp_dir

    # Use mktemp to create a secure temporary directory
    temp_dir=$(mktemp -d "/tmp/${prefix}-$(date +%Y%m%d-%H%M%S)-XXXXXX")

    if [ ! -d "$temp_dir" ]; then
        error "Failed to create temporary directory"
    fi

    # Set restrictive permissions
    chmod 700 "$temp_dir"

    echo "$temp_dir"
}

# Extract snapshot to temporary directory
extract_snapshot_to_temp() {
    local snapshot_id="$1"
    local temp_dir

    temp_dir=$(create_secure_temp_dir "restore")

    log "Extracting snapshot $snapshot_id to temporary directory"
    log "Target directory: $temp_dir"

    # Extract with error handling
    if ! restic restore "$snapshot_id" --target "$temp_dir" 2>&1 | tee -a "$temp_dir/restore.log"; then
        error "Failed to extract snapshot"
    fi

    success "Snapshot extracted to: $temp_dir"
    echo ""
    echo "📁 Contents of the restore:"
    ls -la "$temp_dir" | head -20
    echo ""
    echo "💡 Manual restore instructions:"
    echo "   1. Review the extracted data in: $temp_dir"
    echo "   2. Copy the required files to their destination"
    echo "   3. Verify permissions and ownership"
    echo "   4. Remove the temporary directory: rm -rf $temp_dir"
    echo ""
    echo "⚠️  The directory will remain until you delete it manually"
}

# Run custom restore
run_custom_restore() {
    local service="$1"
    local snapshot_id="$2"
    local mode="$3"
    local service_dir="$SERVICES_DIR/$service"
    local custom_restore="$service_dir/restore"

    log "Running custom restore script for $service"
    log "Script: $custom_restore"
    log "Snapshot: $snapshot_id"
    log "Mode: $mode"

    # Create restore log
    local log_file="/tmp/restore-${service}-$(date +%Y%m%d-%H%M%S).log"

    # Change directory with verification
    if ! cd "$service_dir"; then
        error "Cannot change to service directory: $service_dir"
    fi

    # Execute restore script with log capture
    if "$custom_restore" "$snapshot_id" "--$mode" 2>&1 | tee "$log_file"; then
        success "Custom restore script completed"
        log "Restore log saved to: $log_file"
    else
        error "Custom restore script failed. Check log: $log_file"
    fi
}

# Main function
main() {
    # Check dependencies
    check_dependencies

    # Parse arguments
    if [ $# -eq 0 ]; then
        show_help
        exit 0
    fi

    local service=""
    local mode="test"  # Default mode: test (safer)

    # Parse arguments
    while [ $# -gt 0 ]; do
        case "$1" in
            --test)
                mode="test"
                shift
                ;;
            --production)
                mode="production"
                shift
                ;;
            -h|--help)
                show_help
                exit 0
                ;;
            -v|--version)
                echo "$SCRIPT_NAME version $VERSION"
                exit 0
                ;;
            -*)
                error "Unknown option: $1"
                ;;
            *)
                if [ -z "$service" ]; then
                    service="$1"
                else
                    error "Too many arguments"
                fi
                shift
                ;;
        esac
    done

    # Check service is specified
    if [ -z "$service" ]; then
        error "Service name required"
    fi

    # Validate service name format
    if [[ ! "$service" =~ ^[a-zA-Z0-9_-]+$ ]]; then
        error "Invalid service name format. Use only alphanumeric characters, hyphens, and underscores."
    fi

    # Check service directory exists
    if [ ! -d "$SERVICES_DIR/$service" ]; then
        error "Service directory not found: $SERVICES_DIR/$service"
    fi

    echo "=== Restore $service ==="
    log "Mode: $mode"

    # Step 1: Check prerequisites
    check_prerequisites

    # Step 2: Check service capabilities
    local has_custom_restore
    has_custom_restore=$(check_service_capabilities "$service")

    # Step 3: Show snapshots
    if ! show_snapshots "$service"; then
        exit 1
    fi

    # Step 4: Get user choice
    get_user_choice
    local choice="$USER_CHOICE"

    # Step 5: Resolve and verify snapshot
    local snapshot_id
    snapshot_id=$(resolve_snapshot_id "$service" "$choice")
    verify_snapshot "$service" "$snapshot_id"

    # Step 6: Display restore plan
    echo ""
    log "Restore plan:"
    echo "  Service:  $service"
    echo "  Snapshot: $snapshot_id"
    echo "  Mode:     $mode"

    if [ "$has_custom_restore" = "true" ]; then
        echo "  Method:   Custom restore script"
    else
        echo "  Method:   Extract to temporary directory"
    fi

    # Step 7: Final confirmation
    echo ""
    if [ "$mode" = "test" ]; then
        read -p "Proceed with TEST restore? [y/N]: " -r confirm
        if [[ ! $confirm =~ ^[Yy]$ ]]; then
            echo "Cancelled."
            exit 0
        fi
    else
        echo -e "${RED}⚠️  WARNING: This is a PRODUCTION restore!${NC}"
        echo -e "${RED}This operation may overwrite existing data.${NC}"
        echo ""
        read -p "Type 'yes-production' to proceed: " -r confirm
        if [ "$confirm" != "yes-production" ]; then
            echo "Cancelled."
            exit 0
        fi
    fi

    # Step 8: Execute restore
    echo ""
    if [ "$has_custom_restore" = "true" ]; then
        run_custom_restore "$service" "$snapshot_id" "$mode"
    else
        # No custom script, extract to temp for both test and production
        extract_snapshot_to_temp "$snapshot_id"
    fi

    echo ""
    success "Restore operation completed!"
}

# Trap to clean up on interruption
trap 'echo -e "\n${RED}Interrupted${NC}"; exit 130' INT TERM

# Entry point
main "$@"