Docker --privileged

Reading time: 7 minutes

What Affects

When you run a container as privileged these are the protections you are disabling:

Mount /dev

In a privileged container, all the devices can be accessed in /dev/. Therefore you can escape by mounting the disk of the host.

# docker run --rm -it alpine sh
ls /dev
console  fd       mqueue   ptmx     random   stderr   stdout   urandom
core     full     null     pts      shm      stdin    tty      zero

# docker run --rm --privileged -it alpine sh
ls /dev
cachefiles       mapper           port             shm              tty24            tty44            tty7
console          mem              psaux            stderr           tty25            tty45            tty8
core             mqueue           ptmx             stdin            tty26            tty46            tty9
cpu              nbd0             pts              stdout           tty27            tty47            ttyS0
[...]

Read-only kernel file systems

Kernel file systems provide a mechanism for a process to modify the behavior of the kernel. However, when it comes to container processes, we want to prevent them from making any changes to the kernel. Therefore, we mount kernel file systems as read-only within the container, ensuring that the container processes cannot modify the kernel.

# docker run --rm -it alpine sh
mount | grep '(ro'
sysfs on /sys type sysfs (ro,nosuid,nodev,noexec,relatime)
cpuset on /sys/fs/cgroup/cpuset type cgroup (ro,nosuid,nodev,noexec,relatime,cpuset)
cpu on /sys/fs/cgroup/cpu type cgroup (ro,nosuid,nodev,noexec,relatime,cpu)
cpuacct on /sys/fs/cgroup/cpuacct type cgroup (ro,nosuid,nodev,noexec,relatime,cpuacct)

# docker run --rm --privileged -it alpine sh
mount  | grep '(ro'

Masking over kernel file systems

The /proc file system is selectively writable but for security, certain parts are shielded from write and read access by overlaying them with tmpfs, ensuring container processes can't access sensitive areas.

[!NOTE] > tmpfs is a file system that stores all the files in virtual memory. tmpfs doesn't create any files on your hard drive. So if you unmount a tmpfs file system, all the files residing in it are lost for ever.

# docker run --rm -it alpine sh
mount  | grep /proc.*tmpfs
tmpfs on /proc/acpi type tmpfs (ro,relatime)
tmpfs on /proc/kcore type tmpfs (rw,nosuid,size=65536k,mode=755)
tmpfs on /proc/keys type tmpfs (rw,nosuid,size=65536k,mode=755)

# docker run --rm --privileged -it alpine sh
mount  | grep /proc.*tmpfs

Linux capabilities

Container engines launch the containers with a limited number of capabilities to control what goes on inside of the container by default. Privileged ones have all the capabilities accesible. To learn about capabilities read:

Linux Capabilities

# docker run --rm -it alpine sh
apk add -U libcap; capsh --print
[...]
Current: cap_chown,cap_dac_override,cap_fowner,cap_fsetid,cap_kill,cap_setgid,cap_setuid,cap_setpcap,cap_net_bind_service,cap_net_raw,cap_sys_chroot,cap_mknod,cap_audit_write,cap_setfcap=eip
Bounding set =cap_chown,cap_dac_override,cap_fowner,cap_fsetid,cap_kill,cap_setgid,cap_setuid,cap_setpcap,cap_net_bind_service,cap_net_raw,cap_sys_chroot,cap_mknod,cap_audit_write,cap_setfcap
[...]

# docker run --rm --privileged -it alpine sh
apk add -U libcap; capsh --print
[...]
Current: =eip cap_perfmon,cap_bpf,cap_checkpoint_restore-eip
Bounding set =cap_chown,cap_dac_override,cap_dac_read_search,cap_fowner,cap_fsetid,cap_kill,cap_setgid,cap_setuid,cap_setpcap,cap_linux_immutable,cap_net_bind_service,cap_net_broadcast,cap_net_admin,cap_net_raw,cap_ipc_lock,cap_ipc_owner,cap_sys_module,cap_sys_rawio,cap_sys_chroot,cap_sys_ptrace,cap_sys_pacct,cap_sys_admin,cap_sys_boot,cap_sys_nice,cap_sys_resource,cap_sys_time,cap_sys_tty_config,cap_mknod,cap_lease,cap_audit_write,cap_audit_control,cap_setfcap,cap_mac_override,cap_mac_admin,cap_syslog,cap_wake_alarm,cap_block_suspend,cap_audit_read
[...]

You can manipulate the capabilities available to a container without running in --privileged mode by using the --cap-add and --cap-drop flags.

Seccomp

Seccomp is useful to limit the syscalls a container can call. A default seccomp profile is enabled by default when running docker containers, but in privileged mode it is disabled. Learn more about Seccomp here:

Seccomp

# docker run --rm -it alpine sh
grep Seccomp /proc/1/status
Seccomp:	2
Seccomp_filters:	1

# docker run --rm --privileged -it alpine sh
grep Seccomp /proc/1/status
Seccomp:	0
Seccomp_filters:	0

# You can manually disable seccomp in docker with
--security-opt seccomp=unconfined

Also, note that when Docker (or other CRIs) are used in a Kubernetes cluster, the seccomp filter is disabled by default

AppArmor

AppArmor is a kernel enhancement to confine containers to a limited set of resources with per-program profiles. When you run with the --privileged flag, this protection is disabled.

AppArmor

# You can manually disable seccomp in docker with
--security-opt apparmor=unconfined

SELinux

Running a container with the --privileged flag disables SELinux labels, causing it to inherit the label of the container engine, typically unconfined, granting full access similar to the container engine. In rootless mode, it uses container_runtime_t, while in root mode, spc_t is applied.

SELinux

# You can manually disable selinux in docker with
--security-opt label:disable

What Doesn't Affect

Namespaces

Namespaces are NOT affected by the --privileged flag. Even though they don't have the security constraints enabled, they do not see all of the processes on the system or the host network, for example. Users can disable individual namespaces by using the --pid=host, --net=host, --ipc=host, --uts=host container engines flags.

# docker run --rm --privileged -it alpine sh
ps -ef
PID   USER     TIME  COMMAND
    1 root      0:00 sh
   18 root      0:00 ps -ef

# docker run --rm --privileged --pid=host -it alpine sh
ps -ef
PID   USER     TIME  COMMAND
    1 root      0:03 /sbin/init
    2 root      0:00 [kthreadd]
    3 root      0:00 [rcu_gp]ount | grep /proc.*tmpfs
[...]

User namespace

By default, container engines don't utilize user namespaces, except for rootless containers, which require them for file system mounting and using multiple UIDs. User namespaces, integral for rootless containers, cannot be disabled and significantly enhance security by restricting privileges.

References

• https://www.redhat.com/sysadmin/privileged-flag-container-engines