Permitting SFTP file access while denying shell login

When you google on how to configure remote file access on linux most tutorials and how-tos will guide you towards ftp (file transer protocol) which has existed for ages and is well supported across all devices from mobile to desktop. However using FTP in these day and age without securing it using SSL is kind a naive. User credentials are sent over in clear-text and I don’t have to tell you that sooner or later the credentials will be compromised and your data might be at risk.

FTP similar to Telnet was never a protocol designed with security in mind. There are several solutions to FTP’s shortcomings. As mentioned, you can use FTP in combination with SSL or you can use SSH’s File Transfer Protocol SFTP for securily transferring data using the SSH network protocol.

In general, SSH is used for securely logging into a system remotely. By default, every user with SSH access automatically also has SFTP file access. The issue with simply adding a user to the system for SFTP access is that they also automatically receive shell-access. This grants them the right to run processes and potentially administer your system you may not wish them to run on your server. This tutorial will show you how to add a user to your system without granting them shell permissions but still allowing them file-access.

First, if you haven’t yet installed the openssh server yet, do so now (prepend every command with “sudo” if not logged-in as root):

# apt-get install openssh-server

Once the ssh-server is installed we need to add a user we wish to only allow SFTP access. Add a new usegroup to consolidate multiple users that should only receive SFTP access:

# addgroup sftponly

Next, edit the ssh-daemon configuration file: /etc/ssh/sshd_config (you can use vim or whaever editor you prefer)
Look for the following line:

Subsystem sftp /usr/lib/openssh/sftp-server

and replace the line with this instruction (you can also simply comment out the above line using # and insert the following line below):

Subsystem sftp internal-sftp

At the end of the same file add the following lines to restrict the access of the users bleonging to the sftponly user-group:

# Rules for sftponly group
Match group sftponly
ChrootDirectory %h
X11Forwarding no
AllowTcpForwarding no
ForceCommand internal-sftp

Save the file and restart the ssh-daemon via:

# /etc/init.d/ssh restart

Next, create the directory which the users will be jailed too:

# mkdir /srv/sftponly/

Optionally you can use a mount-bind to map a specific filesystem into that folder. My share is located at /mnt/fs and I use mount-bind to map the shared-folder to sftponly:

# mount --bind /mnt/fs /srv/sftponly/

Add a new user to the system and specify the previously created group as the user’s primary group. Disable shell-login and don’t create a home directory. See man-pages about adduser for more information about the available command-line switches:

# adduser --home /srv/nfs4/fs/ --no-create-home --ingroup sftponly --disabled-login <username>

Set a password for the new user:

# passwd <username>

In order for the user to be chrooted (jailed) to the specified directory the user’s home directory must be owned by root as well as only be writable by root:

# chown root:root /srv/sftponly/

In case you are not able to log-in using the newly added user or even worse if the user is not getting jailed it’s best to check the auth.log in /var/log/auth.log

Apr 20 12:34:04 fs sshd[12015]: Server listening on 0.0.0.0 port 22.
Apr 20 12:34:04 fs sshd[12015]: Server listening on :: port 22.
Apr 20 12:34:09 fs sshd[12018]: Accepted password for <username> from 10.1.1.185 port 54493 ssh2
Apr 20 12:34:09 fs sshd[12018]: pam_unix(sshd:session): session opened for user <username> by (uid=0)
Apr 20 12:34:09 fs sshd[12023]: fatal: bad ownership or modes for chroot directory "/srv/sftponly"
Apr 20 12:34:09 fs sshd[12018]: pam_unix(sshd:session): session closed for user<username>
Apr 20 12:34:25 fs sshd[12029]: Accepted password for <username> from 10.1.1.185 port 54494 ssh2
Apr 20 12:34:25 fs sshd[12029]: pam_unix(sshd:session): session opened for user username by (uid=0)
Apr 20 12:34:25 fs sshd[12037]: fatal: bad ownership or modes for chroot directory "/srv/sftponly"
Apr 20 12:34:25 fs sshd[12029]: pam_unix(sshd:session): session closed for user <username>
Apr 20 12:37:02 fs sshd[12076]: Accepted password for <username> from 10.1.1.185 port 54516 ssh2
Apr 20 12:37:02 fs sshd[12076]: pam_unix(sshd:session): session opened for user <username> by (uid=0)
Apr 20 12:37:02 fs sshd[12081]: fatal: bad ownership or modes for chroot directory "/srv/sftponly"
Apr 20 12:37:02 fs sshd[12076]: pam_unix(sshd:session): session closed for user <username>

In my case the specific user wasn’t getting jailed because the permissions on the root-folder were not correctly set. Make sure the home directory is owned by root and not the user itself.

Remove write permissions from others:

# chmod go-w /srv/sftponly/

NFSv4 performance woes on Debian

Over the weekend I took the opportunity to reinstall the OS on my workstation at home. Instead of choosing Debian unstable I decided to go with Debian stable + backports this time. The reason is simply to not deal with package upgrades braking the current system again. Manually fixing my local gitlab installation after a ruby upgrade annoyed the hell out of me and I simply can’t be bothered with manually going through package management and pinning down a working version of ruby and ruby-gems just to keep my current setup working.

When I started to configure system services such as ssh, nfsv4, ftpd etc. I noticed that mounting NFSv4 shares on Debian 7 (Wheezy) was quite slow (around 11 to 15 seconds). Oddly enough, once the NFS share was mounted into my local filesystem I was able to copy files as quickly as usual.

I checked dmesg and /var/log/messages after mounting the NFS share and noticed the following entries:

[17603.799651] Key type dns_resolver registered
[17603.804909] NFS: Registering the id_resolver key type
[17603.804918] Key type id_resolver registered
[17603.804918] Key type id_legacy registered
[17618.828292] RPC: AUTH_GSS upcall timed out.
[17618.828292] Please check user daemon is running.

According to the following thread, rpc-svcgssd is responsible for the nfs Kerberos authentication at the server side. I have no intention of using Kerberos at home as the time required for the set-up outweigh the benefits of single-sign-on for just a single user. I can simply set-up the same account with the same credentials on my computers. Therefore kerberos authentication should also be optional at the client side. Well it actually is, but the implementation at the moment is rather tedious. It waits for a timeout that occurs after roughly 15 seconds which introduces the delay when mounting a share.

On Redhat’s bugtracker the following entry was posted: Bug 1001934 – 15 sec timeout when mounting with nfs4:
Marcindulak provided the following workaround to fix the issue until the client is updated

blacklist the rpcsec_gss_krb5 module on the client and reboot or simply unload the kernel module

echo "blacklist rpcsec_gss_krb5" > /etc/modprobe.d/blacklist-nfs-gss-krb5.conf
reboot

or instead of rebooting you can simply use rmmod to unload the currently loaded module:

rmmod rpcsec_gss_krb5

This should eliminate the delay when mounting NFSv4 shares on Debian.