• Uncategorized

About shell : How-to-check-if-a-process-is-running-inside-docker-container

Question Detail

[Updated1] I have a shell which will change TCP kernel parameters in some functions, but now I need to make this shell run in Docker container, that means, the shell need to know it is running inside a container and stop configuring the kernel.

Now I’m not sure how to achieve that, here is the contents of /proc/self/cgroup inside the container:

9:hugetlb:/
8:perf_event:/
7:blkio:/
6:freezer:/
5:devices:/
4:memory:/
3:cpuacct:/
2:cpu:/docker/25ef774c390558ad8c4e9a8590b6a1956231aae404d6a7aba4dde320ff569b8b
1:cpuset:/

Any flags above can I use to figure out if this process is running inside a container?

[Updated2]: I have also noticed Determining if a process runs inside lxc/Docker, but it seems not working in this case, the content in /proc/1/cgroup of my container is:

8:perf_event:/
7:blkio:/
6:freezer:/
5:devices:/
4:memory:/
3:cpuacct:/
2:cpu:/docker/25ef774c390558ad8c4e9a8590b6a1956231aae404d6a7aba4dde320ff569b8b
1:cpuset:/

No /lxc/containerid

Question Answer

Docker creates .dockerenv and .dockerinit (removed in v1.11) files at the top of the container’s directory tree so you might want to check if those exist.

Something like this should work.

#!/bin/bash
if [ -f /.dockerenv ]; then
    echo "I'm inside matrix ;(";
else
    echo "I'm living in real world!";
fi

To check inside a Docker container if you are inside a Docker container or not can be done via /proc/1/cgroup. As this post suggests you can to the following:

Outside a docker container all entries in /proc/1/cgroup end on / as you can see here:

[email protected]:~$ cat /proc/1/cgroup
11:name=systemd:/
10:hugetlb:/
9:perf_event:/
8:blkio:/
7:freezer:/
6:devices:/
5:memory:/
4:cpuacct:/
3:cpu:/
2:cpuset:/

Inside a Docker container some of the control groups will belong to Docker (or LXC):

[email protected]:~$ docker run busybox cat /proc/1/cgroup
11:name=systemd:/
10:hugetlb:/
9:perf_event:/
8:blkio:/
7:freezer:/
6:devices:/docker/3601745b3bd54d9780436faa5f0e4f72bb46231663bb99a6bb892764917832c2
5:memory:/
4:cpuacct:/
3:cpu:/docker/3601745b3bd54d9780436faa5f0e4f72bb46231663bb99a6bb892764917832c2
2:cpuset:/

We use the proc’s sched (/proc/$PID/sched) to extract the PID of the process. The process’s PID inside the container will differ then it’s PID on the host (a non-container system).

For example, the output of /proc/1/sched on a container
will return:

[email protected]:~# cat /proc/1/sched | head -n 1
bash (5276, #threads: 1)

While on a non-container host:

$ cat /proc/1/sched  | head -n 1
init (1, #threads: 1)

This helps to differentiate if you are in a container or not. eg you can do:

if [[ ! $(cat /proc/1/sched | head -n 1 | grep init) ]]; then {
    echo in docker
} else {
    echo not in docker
} fi

Thomas’ solution as code:

running_in_docker() {
  (awk -F/ '$2 == "docker"' /proc/self/cgroup | read non_empty_input)
}

Note

The read with a dummy variable is a simple idiom for Does this produce any output?. It’s a compact method for turning a possibly verbose grep or awk into a test of a pattern.

Additional note on read

Using Environment Variables

For my money, I prefer to set an environment variable inside the docker image that can then be detected by the application.

For example, this is the start of a demo Dockerfile config:

FROM node:12.20.1 as base
ENV DOCKER_RUNNING=true
RUN yarn install --production
RUN yarn build

The second line sets an envar called DOCKER_RUNNING that is then easy to detect. The issue with this is that in a multi-stage build, you will have to repeat the ENV line every time you FROM off of an external image. For example, you can see that I FROM off of node:12.20.1, which includes a lot of extra stuff (git, for example). Later on in my Dockerfile I then COPY things over to a new image based on node:12.20.1-slim, which is much smaller:

FROM node:12.20.1-slim as server
ENV DOCKER_RUNNING=true
EXPOSE 3000
COPY --from=base /build /build
CMD ["node", "server.js"]

Even though this image target server is in the same Dockerfile, it requires the ENV var to be defined again because it has a different base image.

If you make use of Docker-Compose, you could instead easily define an envar there. For example, your docker-compose.yml file could look like this:

version: "3.8"
services:
  nodeserver:
    image: michaeloryl/stackdemo
    environment:
      - NODE_ENV=production
      - DOCKER_RUNNING=true

What works for me is to check for the inode number of the ‘/.’
Inside the docker, its a very high number.
Outside the docker, its a very low number like ‘2’.
I reckon this approach would also depend on the FileSystem being used.

Example

Inside the docker:

# ls -ali / | sed '2!d' |awk {'print $1'}
1565265

Outside the docker

$ ls -ali / | sed '2!d' |awk {'print $1'}
2

In a script:

#!/bin/bash
INODE_NUM=`ls -ali / | sed '2!d' |awk {'print $1'}`
if [ $INODE_NUM == '2' ];
then
        echo "Outside the docker"
else
        echo "Inside the docker"
fi

golang code, via the /proc/%s/cgroup to check a process in a docker,include the k8s cluster

func GetContainerID(pid int32) string {
    cgroupPath := fmt.Sprintf("/proc/%s/cgroup", strconv.Itoa(int(pid)))
    return getContainerID(cgroupPath)
}

func GetImage(containerId string) string {
    if containerId == "" {
        return ""
    }
    image, ok := containerImage[containerId]
    if ok {
        return image
    } else {
        return ""
    }
}
func getContainerID(cgroupPath string) string {
    containerID := ""
    content, err := ioutil.ReadFile(cgroupPath)
    if err != nil {
        return containerID
    }
    lines := strings.Split(string(content), "\n")
    for _, line := range lines {
        field := strings.Split(line, ":")
        if len(field) < 3 {
            continue
        }
        cgroup_path := field[2]
        if len(cgroup_path) < 64 {
            continue
        }
        // Non-systemd Docker
        //5:net_prio,net_cls:/docker/de630f22746b9c06c412858f26ca286c6cdfed086d3b302998aa403d9dcedc42
        //3:net_cls:/kubepods/burstable/pod5f399c1a-f9fc-11e8-bf65-246e9659ebfc/9170559b8aadd07d99978d9460cf8d1c71552f3c64fefc7e9906ab3fb7e18f69
        pos := strings.LastIndex(cgroup_path, "/")
        if pos > 0 {
            id_len := len(cgroup_path) - pos - 1
            if id_len == 64 {
                //p.InDocker = true
                // docker id
                containerID = cgroup_path[pos+1 : pos+1+64]
                // logs.Debug("pid:%v in docker id:%v", pid, id)
                return containerID
            }
        }
        // systemd Docker
        //5:net_cls:/system.slice/docker-afd862d2ed48ef5dc0ce8f1863e4475894e331098c9a512789233ca9ca06fc62.scope
        docker_str := "docker-"
        pos = strings.Index(cgroup_path, docker_str)
        if pos > 0 {
            pos_scope := strings.Index(cgroup_path, ".scope")
            id_len := pos_scope - pos - len(docker_str)
            if pos_scope > 0 && id_len == 64 {
                containerID = cgroup_path[pos+len(docker_str) : pos+len(docker_str)+64]
                return containerID
            }
        }
    }
    return containerID
}

We needed to exclude processes running in containers, but instead of checking for just docker cgroups we decided to compare /proc/<pid>/ns/pid to the init system at /proc/1/ns/pid. Example:

pid=$(ps ax | grep "[r]edis-server \*:6379" | awk '{print $1}')
if [ $(readlink "/proc/$pid/ns/pid") == $(readlink /proc/1/ns/pid) ]; then
   echo "pid $pid is the same namespace as init system"
else
   echo "pid $pid is in a different namespace as init system"
fi

Or in our case we wanted a one liner that generates an error if the process is NOT in a container

bash -c "test -h /proc/4129/ns/pid && test $(readlink /proc/4129/ns/pid) != $(readlink /proc/1/ns/pid)"

which we can execute from another process and if the exit code is zero then the specified PID is running in a different namespace.

Based on Dan Walsh’s comment about using SELinux ps -eZ | grep container_t, but without requiring ps to be installed:

$ podman run --rm fedora:31 cat /proc/1/attr/current
system_u:system_r:container_t:s0:c56,c299
$ podman run --rm alpine cat /proc/1/attr/current
system_u:system_r:container_t:s0:c558,c813
$ docker run --rm fedora:31 cat /proc/1/attr/current
system_u:system_r:container_t:s0:c8,c583
$ cat /proc/1/attr/current
system_u:system_r:init_t:s0

This just tells you you’re running in a container, but not which runtime.

Didn’t check other container runtimes but https://opensource.com/article/18/2/understanding-selinux-labels-container-runtimes provides more info and suggests this is widely used, might also work for rkt and lxc?

What works for me, as long as I know the system programs/scrips will be running on, is confirming if what’s running with PID 1 is systemd (or equivalent). If not, that’s a container.
And this should be true for any linux container, not only docker.

Had the need for this capability in 2022 on macOS and only the answer by @at0S still works from all the other options.

  • /proc/1/cgroup only has the root directory in a container unless configured otherwise
  • /proc/1/sched showed the same in-container process number. The name was different (bash) but that’s not very portable.
  • Environment variables work if you configure your container yourself, but none of the default environment variables helped

I did find an option not listed in the other answers: /proc/1/mounts included an overlay filesystem with “docker” in its path.

You may also like...

Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.