fix vps_init

vps_init.sh

@@ -225,6 +225,59 @@ handle_error() {
 # 设置错误跟踪
 trap 'handle_error $LINENO' ERR
 
+# ===========================================
+# 安全更新软件源：自动检测并禁用失效的软件源
+# （例如 Debian 版本归档后失效的 *-backports 源）后重试
+# ===========================================
+apt_update_safe() {
+    local tmp retry=0
+    tmp=$(mktemp)
+    while :; do
+        # 执行更新，忽略退出码（失效的源会让 apt 返回非0）
+        apt-get update > "$tmp" 2>&1 || true
+        cat "$tmp" >> "$LOG_FILE"
+        cat "$tmp"
+
+        # 没有检测到失效源，更新成功
+        if ! grep -qiE "no longer has a Release file|404 +Not Found|Failed to fetch" "$tmp"; then
+            rm -f "$tmp"
+            return 0
+        fi
+
+        # 已尝试修复过一次，仍有问题则忽略并继续，避免死循环
+        if [ "$retry" -ge 1 ]; then
+            log "${YELLOW}部分软件源仍不可用，将忽略失效源并继续${NC}"
+            rm -f "$tmp"
+            return 0
+        fi
+
+        log "${YELLOW}检测到失效的软件源，正在自动禁用...${NC}"
+
+        # 从 apt 输出中提取失效的套件名（位于 "<url> <suite> Release" 中）
+        local suites suite
+        suites=$(grep -oE "The repository '[^']+ Release'" "$tmp" \
+            | sed -E "s/The repository '(.*) Release'/\1/" \
+            | awk '{print $NF}' | sort -u)
+
+        if [ -z "$suites" ]; then
+            log "${YELLOW}无法解析失效套件名，将忽略失效源并继续${NC}"
+            rm -f "$tmp"
+            return 0
+        fi
+
+        for suite in $suites; do
+            [ -n "$suite" ] || continue
+            log "${YELLOW}  - 禁用失效套件: ${suite}${NC}"
+            for f in /etc/apt/sources.list $(ls /etc/apt/sources.list.d/*.list 2>/dev/null); do
+                [ -f "$f" ] || continue
+                # 注释掉引用该失效套件的未注释行
+                sed -i -E "s|^([^#].*[[:space:]]${suite}[[:space:]].*)$|#\1|g" "$f"
+            done
+        done
+        retry=$((retry + 1))
+    done
+}
+
 # ===========================================
 # 1. 系统更新
 # ===========================================
@@ -238,7 +291,7 @@ fi
 
 # 更新系统包
 if [ "$OS_TYPE" = "debian" ]; then
-    apt update -y || log "${RED}更新软件源失败${NC}"
+    apt_update_safe
     DEBIAN_FRONTEND=noninteractive apt full-upgrade -y || log "${RED}系统升级失败${NC}"
     apt install -y wget curl sudo vim git ufw net-tools htop iftop || log "${RED}安装基础软件包失败${NC}"
     log "${GREEN}系统更新完成，安装了常用工具${NC}"
@@ -378,22 +431,32 @@ if command -v docker &> /dev/null; then
 else
     # 安装Docker
     if [ "$OS_TYPE" = "debian" ]; then
+        # 先确保软件源可用（官方脚本内部也会执行 apt update）
+        apt_update_safe
+
         # 使用官方安装脚本
         curl -fsSL https://get.docker.com -o get-docker.sh
-        sh get-docker.sh
+        sh get-docker.sh || log "${RED}Docker安装脚本执行出现问题${NC}"
-        rm get-docker.sh
+        rm -f get-docker.sh
-        systemctl enable docker
+
-        
+        # 校验Docker是否真正安装成功
-        # 安装Docker Compose
+        if command -v docker &> /dev/null; then
-        if ! command -v docker-compose &> /dev/null; then
+            systemctl enable docker 2>/dev/null || log "${YELLOW}无法设置Docker开机自启${NC}"
-            curl -L "https://github.com/docker/compose/releases/latest/download/docker-compose-$(uname -s)-$(uname -m)" -o /usr/local/bin/docker-compose
+            systemctl start docker 2>/dev/null || true
-            chmod +x /usr/local/bin/docker-compose
+
-            log "${GREEN}Docker Compose 安装完成${NC}"
+            # 安装Docker Compose
+            if ! command -v docker-compose &> /dev/null; then
+                curl -L "https://github.com/docker/compose/releases/latest/download/docker-compose-$(uname -s)-$(uname -m)" -o /usr/local/bin/docker-compose
+                chmod +x /usr/local/bin/docker-compose
+                log "${GREEN}Docker Compose 安装完成${NC}"
+            fi
+
+            log "${GREEN}Docker安装完成，版本信息:${NC}"
+            docker --version
+            docker-compose --version 2>/dev/null || docker compose version 2>/dev/null || true
+        else
+            log "${RED}Docker安装失败，请检查网络或软件源后重试，可手动执行: curl -fsSL https://get.docker.com | sh${NC}"
         fi
-        
-        log "${GREEN}Docker安装完成，版本信息:${NC}"
-        docker --version
-        docker-compose --version
     else
         log "${YELLOW}非Debian系统，请手动安装Docker${NC}"
     fi
@@ -406,7 +469,8 @@ log "${BLUE}[6/10] 防火墙设置开始...${NC}"
 
 # 安装UFW
 if [ "$OS_TYPE" = "debian" ]; then
-    apt update -y && apt install -y ufw net-tools lsof
+    apt_update_safe
+    apt install -y ufw net-tools lsof || log "${RED}安装防火墙相关软件包(ufw/net-tools/lsof)失败${NC}"
 
     # 确保防火墙默认策略
     ufw default deny incoming
@@ -441,8 +505,8 @@ if [ "$OS_TYPE" = "debian" ]; then
     LISTENING_PORTS=$(netstat -tlnp 2>/dev/null | grep "LISTEN" | awk '{print $4}' | awk -F: '{print $NF}' | sort -n | uniq)
     
     # 使用lsof作为备选方法
-    if [ -z "$LISTENING_PORTS" ]; then
+    if [ -z "$LISTENING_PORTS" ] && command -v lsof &> /dev/null; then
-        LISTENING_PORTS=$(lsof -i -P -n | grep LISTEN | awk '{print $9}' | awk -F: '{print $NF}' | sort -n | uniq)
+        LISTENING_PORTS=$(lsof -i -P -n 2>/dev/null | grep LISTEN | awk '{print $9}' | awk -F: '{print $NF}' | sort -n | uniq)
     fi
     
     # 如果仍然为空，提示手动检查
@@ -457,7 +521,10 @@ if [ "$OS_TYPE" = "debian" ]; then
                   "$PORT" -lt "65535" && "$PORT" -gt "1024" ]]; then
                 
                 # 尝试找出服务名称
-                SERVICE=$(lsof -i:$PORT -sTCP:LISTEN | grep -v "COMMAND" | awk '{print $1}' | head -1)
+                SERVICE=""
+                if command -v lsof &> /dev/null; then
+                    SERVICE=$(lsof -i:$PORT -sTCP:LISTEN 2>/dev/null | grep -v "COMMAND" | awk '{print $1}' | head -1)
+                fi
                 if [ -z "$SERVICE" ]; then
                     SERVICE=$(netstat -tlnp 2>/dev/null | grep ":$PORT" | awk '{print $7}' | cut -d"/" -f2 | head -1)
                 fi
@@ -579,18 +646,26 @@ log "${BLUE}[9/10] Fail2ban安装开始...${NC}"
 
 if [ "$OS_TYPE" = "debian" ]; then
     # 安装Fail2ban
-    apt update -y && apt install -y fail2ban
+    apt_update_safe
-    systemctl start fail2ban
+    apt install -y fail2ban || log "${RED}安装Fail2ban失败${NC}"
-    systemctl enable fail2ban
+
-    
+    # 校验是否真正安装成功
-    # 配置Fail2ban
+    if ! command -v fail2ban-client &> /dev/null && ! dpkg -l fail2ban 2>/dev/null | grep -q "^ii"; then
-    cp /etc/fail2ban/jail.conf /etc/fail2ban/jail.local
+        log "${RED}Fail2ban未成功安装，跳过其配置（请检查软件源后重试）${NC}"
-    
+    else
-    # 清理任何现有配置
+        systemctl enable fail2ban 2>/dev/null || log "${YELLOW}无法设置Fail2ban开机自启${NC}"
-    rm -rf /etc/fail2ban/jail.d/* 2>/dev/null || true
+
-    
+        # 配置Fail2ban
-    # 创建SSH防护配置
+        mkdir -p /etc/fail2ban/jail.d
-    cat <<EOF > /etc/fail2ban/jail.d/sshd.local
+        if [ -f /etc/fail2ban/jail.conf ]; then
+            cp /etc/fail2ban/jail.conf /etc/fail2ban/jail.local
+        fi
+
+        # 清理任何现有配置
+        rm -rf /etc/fail2ban/jail.d/* 2>/dev/null || true
+
+        # 创建SSH防护配置
+        cat <<EOF > /etc/fail2ban/jail.d/sshd.local
 [sshd]
 enabled = true
 mode = normal
@@ -601,45 +676,46 @@ maxretry = 5
 bantime = 1h
 findtime = 10m
 EOF
-    
+
-    # 重启Fail2ban
+        # 重启Fail2ban
-    log "${YELLOW}重启Fail2ban服务...${NC}"
+        log "${YELLOW}重启Fail2ban服务...${NC}"
-    systemctl restart fail2ban
+        systemctl restart fail2ban 2>/dev/null || systemctl start fail2ban 2>/dev/null || log "${RED}Fail2ban服务启动失败${NC}"
-    
+
-    # 等待服务启动完成
+        # 等待服务启动完成
-    log "${YELLOW}等待Fail2ban服务完全启动...${NC}"
+        log "${YELLOW}等待Fail2ban服务完全启动...${NC}"
-    sleep 5
+        sleep 5
-    
+
-    # 检查服务状态
+        # 检查服务状态
-    if systemctl is-active --quiet fail2ban; then
+        if systemctl is-active --quiet fail2ban; then
-        log "${GREEN}Fail2ban服务已成功启动${NC}"
+            log "${GREEN}Fail2ban服务已成功启动${NC}"
-        
+
-        # 显示Fail2ban状态（使用错误处理避免脚本终止）
+            # 显示Fail2ban状态（使用错误处理避免脚本终止）
-        log "${YELLOW}获取Fail2ban状态信息...${NC}"
+            log "${YELLOW}获取Fail2ban状态信息...${NC}"
-        
+
-        # 尝试获取fail2ban状态，忽略可能的错误
+            # 尝试获取fail2ban状态，忽略可能的错误
-        fail2ban-client status >/dev/null 2>&1 || log "${YELLOW}无法获取fail2ban综合状态，但这不影响功能${NC}"
+            fail2ban-client status >/dev/null 2>&1 || log "${YELLOW}无法获取fail2ban综合状态，但这不影响功能${NC}"
-        
+
-        # 尝试获取sshd监狱状态
+            # 尝试获取sshd监狱状态
-        if fail2ban-client status sshd >/dev/null 2>&1; then
+            if fail2ban-client status sshd >/dev/null 2>&1; then
-            log "${GREEN}SSH防护已成功配置${NC}"
+                log "${GREEN}SSH防护已成功配置${NC}"
-            # 只有在前面成功的情况下才显示详细信息
+                # 只有在前面成功的情况下才显示详细信息
-            fail2ban-client status sshd
+                fail2ban-client status sshd
+            else
+                log "${YELLOW}无法获取SSH监狱状态，这可能是因为服务刚刚启动或配置需要更多时间生效${NC}"
+                log "${YELLOW}如果在重启后仍有问题，请检查 /var/log/fail2ban.log${NC}"
+            fi
+
+            # 显示服务状态
+            systemctl status fail2ban --no-pager || true
         else
-            log "${YELLOW}无法获取SSH监狱状态，这可能是因为服务刚刚启动或配置需要更多时间生效${NC}"
+            log "${RED}Fail2ban服务启动失败，请检查错误日志${NC}"
-            log "${YELLOW}如果在重启后仍有问题，请检查 /var/log/fail2ban.log${NC}"
+            log "${YELLOW}尝试查看Fail2ban日志获取错误详情:${NC}"
+            tail -n 20 /var/log/fail2ban.log 2>/dev/null || log "${RED}无法读取Fail2ban日志${NC}"
         fi
-        
+
-        # 显示服务状态
+        log "${GREEN}Fail2ban安装和配置完成${NC}"
-        systemctl status fail2ban --no-pager || true
+        log "${YELLOW}如果出现临时错误，服务器重启后通常会正常工作${NC}"
-    else
-        log "${RED}Fail2ban服务启动失败，请检查错误日志${NC}"
-        log "${YELLOW}尝试查看Fail2ban日志获取错误详情:${NC}"
-        tail -n 20 /var/log/fail2ban.log 2>/dev/null || log "${RED}无法读取Fail2ban日志${NC}"
     fi
-    
-    log "${GREEN}Fail2ban安装和配置完成${NC}"
-    log "${YELLOW}如果出现临时错误，服务器重启后通常会正常工作${NC}"
 else
     log "${YELLOW}非Debian系统，请手动安装Fail2ban${NC}"
 fi