Changes/LabeledNFS

Enable SELinux Labeled NFS Support

Summary

The Linux Kernel has grown support for passing SELinux labels between a client and server using NFS.

Owner

• Name: Daniel Walsh
• Name: Steve Dickson

• Email: <dwalsh@redhat.com>
• Email: <steved@redhat.com>

Current status

• Targeted release: [Fedora 20]
• Last updated: Jul 24 2013
• Tracker bug: #998566
  • selinux-policy fixes are in Fedora 20.
  • Labeled NFS Support is in 3.11.0-0.rc0.git7.1.fc20.x86_64 kernel
  • nfs-utils support
    • https://bugzilla.redhat.com/show_bug.cgi?id=984718
  • mount support

Detailed Description

We have always needed to treat NFS mounts with a single label usually something like nfs_t. Or at best allow an administrator to override the default with a label using the mount --context option. With this change we have lots of different Labels supported on an NFS share.

Benefit to Fedora

There are two huge benefits for Fedora, in that currently we can not differentiate different labels on a single NFS mount point. Applications like Secure Virtualization as launched by libvirt, can not set the label of an image file on an NFS share, so sVirt separation is severely weakened. Similarly if you setup home directories on an NFS share, then any confined application that needs to write a file in a home directory now can write any file on an NFS Share.

With labeled NFS this vulnerability goes away.

Scope

• Proposal owners:
  • Steve Dickson needs to make sure nfs-utils works properly.
  • Dan Walsh needs to make sure selinux-policy works properly in all use cases.
• Other developers: Kernel

Turn on Labeled NFS in the Fedora Kernel, Fix any policy issues that arise because of this. I believe this is mainly a testing issue, and that the functionality is comeplete.

• Release engineering: N/A (No changes for Release Engineering)

• Policies and guidelines: N/A (not affected)

Upgrade/compatibility impact

In the beginning this will be turned off by default so updating a system from an older Fedora should continue to run, with nfs mounts treated as nfs_t, by SELinux.

How To Test

Do to a bug in nfs-utils:

• Server Side

Add RPCNFSDARGS="-V4.2" to /etc/sysconfig/nfs Start nfs service

   systemctl restart nfs

• Client Side

   mount  -o v4.2 server:mntpoint localmountpoint

There are many different scenarios that have to be tested with this new functionality.

Basically with Labeled NFS we need to test with client and servers supporting LNFS and SELinux

SELinux Testing

• SELinux Client LNFS - SELinux Server LNFS
• SELinux Client LNFS - SELinux Server No LNFS
• SELinux CLient LNFS - Server LNFS
• SELinux CLient LNFS - Server No LNFS
• Client LNFS - SELinux Server LNFS
• Client LNFS - SELInux Server No LNFS
• Client LNFS - Server LNFS
• Client LNFS - Server no LNFS
• Client no LNFS - SELinux Server LNFS
• Client no LNFS - SELInux Server No LNFS
• Client no LNFS - Server LNFS
• Client no LNFS - Server no LNFS

Also need testing on three way. IE You need two clients that support SELinux CLient NFS and change the label on one client, and make sure the other client sees the change.

• Application integration testing
  • Validate that libvirt/sVirt can store labeled VMs on NFS.
  • Validate that Secure Linux Applications can store content on NFS
  • Setup Users with NFS based homedirs and make sure tools like apache/sshd work without the use_nfs_home_dirs boolean turned on.

--

User Experience

If you turn on Labeled NFS for homedirs or other content, the content will be labeled the same was as if it was on a EXT* file system. There is potential errors on mount between Systems at different versions of NFS. IE Attempting to mount a server which does not support Labelling, could cause an error, but the user/administrator should be able to handle this.

Dependencies

This will only work on the 3.11 kernel. If you boot with an older kernel NFS will fall back to the old behaviour.

Contingency Plan

We can continue using what we always did, all clients labeled the same

Comments and Discussion

• See Talk:Features/LabeledNFS