obeone/nfs-server (forked from erichough/nfs-server)

A lightweight, robust, flexible, and containerized NFS server.

Nightly built containers are available for multiple arch, and now manifest are signed using cosign !

A Helm chart is available to ease deploy on Kubernetes

Why?

This is the only containerized NFS server that offers all of the following features:

• small (~15MB) Alpine Linux image
• NFS versions 3, 4, or both simultaneously
• clean teardown of services upon termination (no lingering nfsd processes on Docker host)
• flexible construction of /etc/exports
• extensive server configuration via environment variables
• human-readable logging (with a helpful debug mode)
• optional bonus features
  • Kerberos security
  • NFSv4 user ID mapping via idmapd
  • AppArmor compatibility

This forked version add the following features :

• Support for arch : linux/amd64, linux/arm64, linux/i386, linux/armhf, linux/armel
• Build everyday to stay up-to-date with upstream distribution (alpine)

Table of Contents

• Requirements
• Usage
  • Starting the server
  • Mounting filesystems from a client
• Optional features
  • Debug logging
  • Kerberos security
  • NFSv4 user ID mapping
  • AppArmor integration
• Advanced
  • automatically load required kernel modules
  • custom server ports
  • custom NFS versions offered
  • performance tuning
• Help!
• Remaining tasks
• Acknowledgements

Requirements

1. The Docker host kernel will need the following kernel modules

   • nfs
   • nfsd
   • rpcsec_gss_krb5 (only if Kerberos is used)

   You can manually enable these modules on the Docker host with:

   modprobe {nfs,nfsd,rpcsec_gss_krb5}

   or you can just allow the container to load them automatically.

2. The container will need to run with CAP_SYS_ADMIN (or --privileged). This is necessary as the server needs to mount several filesystems inside the container to support its operation, and performing mounts from inside a container is impossible without these capabilities.

3. The container will need local access to the files you'd like to serve via NFS. You can use Docker volumes, bind mounts, files baked into a custom image, or virtually any other means of supplying files to a Docker container.

Usage

Starting the server

Starting the ghcr.io/obeone/nfs-server image will launch an NFS server. You'll need to supply some information upon container startup, which we'll cover below, but briefly speaking your docker run command might look something like this:

docker run                                            \
  -v /host/path/to/shared/files:/some/container/path  \
  -v /host/path/to/exports.txt:/etc/exports:ro        \
  --cap-add SYS_ADMIN                                 \
  -p 2049:2049                                        \
  ghcr.io/obeone/nfs-server

Let's break that command down into its individual pieces to see what's required for a successful server startup.

1. Provide the files to be shared over NFS

   As noted in the requirements, the container will need local access to the files you'd like to share over NFS. Some ideas for supplying these files:

   • bind mounts (-v /host/path/to/shared/files:/some/container/path)
   • volumes (-v some_volume:/some/container/path)
   • files baked into custom image (e.g. in a Dockerfile: COPY /host/files /some/container/path)

   You may use any combination of the above, or any other means to supply files to the container.

2. Provide your desired NFS exports (/etc/exports)

   You'll need to tell the server which container directories to share. You have three options for this; choose whichever one you prefer:

   1. bind mount /etc/exports into the container

      docker run                                      \
        -v /host/path/to/exports.txt:/etc/exports:ro  \
        ...                                           \
        ghcr.io/obeone/nfs-server
      

   2. provide each line of /etc/exports as an environment variable

      The container will look for environment variables that start with NFS_EXPORT_ and end with an integer. e.g. NFS_EXPORT_0, NFS_EXPORT_1, etc.

      docker run                                                                       \
        -e NFS_EXPORT_0='/container/path/foo                  *(ro,no_subtree_check)'  \
        -e NFS_EXPORT_1='/container/path/bar 123.123.123.123/32(rw,no_subtree_check)'  \
        ...                                                                            \
        ghcr.io/obeone/nfs-server
      

   3. bake /etc/exports into a custom image

      e.g. in a Dockerfile:

      FROM ghcr.io/obeone/nfs-server
      ADD /host/path/to/exports.txt /etc/exports

3. Use --cap-add SYS_ADMIN or --privileged

   As noted in the requirements, the container will need additional privileges. So your run command will need either:

   docker run --cap-add SYS_ADMIN ... ghcr.io/obeone/nfs-server
   

   or

   docker run --privileged ... ghcr.io/obeone/nfs-server
   

   Not sure which to use? Go for --cap-add SYS_ADMIN as it's the lesser of two evils.

4. Expose the server ports

   You'll need to open up at least one server port for your client connections. The ports listed in the examples below are the defaults used by this image and most can be customized.

   • If your clients connect via NFSv4 only, you can get by with just TCP port 2049:

     docker run -p 2049:2049 ... ghcr.io/obeone/nfs-server
     

   • If you'd like to support NFSv3, you'll need to expose a lot more ports:

     docker run                          \
       -p 2049:2049   -p 2049:2049/udp   \
       -p 111:111     -p 111:111/udp     \
       -p 32765:32765 -p 32765:32765/udp \
       -p 32767:32767 -p 32767:32767/udp \
       ...                               \
       ghcr.io/obeone/nfs-server
     

If you pay close attention to each of the items in this section, the server should start quickly and be ready to accept your NFS clients.

Mounting filesystems from a client

# mount <container-IP>:/some/export /some/local/path

Optional Features

• Debug logging
• Kerberos security
• NFSv4 user ID mapping
• AppArmor integration

Advanced

• automatically load required kernel modules
• customizing which ports are used
• customizing NFS versions offered
• performance tuning

Help!

Please open an issue if you have any questions, constructive criticism, or can't get something to work.

Remaining tasks

• figure out why rpc.nfsd takes 5 minutes to startup/timeout unless rpcbind is running
• add more examples

Acknowledgements

This work was based on prior projects:

• f-u-z-z-l-e/docker-nfs-server
• sjiveson/nfs-server-alpine
• ehough/docker-nfs-server