installing and cell setup on rpm based linux

[openafs-wiki.git] / InstallingOpenAFSonRHEL.mdwn

This is a step-by-step guide to installing OpenAFS and setting up an AFS cell
on RedHat Enterprise Linux, Fedora, or CentOS.  This guide is current as of
OpenAFS version 1.6.9 on CentOS 6.5.

This document is based on Steven Pelley's guide for installing OpenAFS on
Fedora.

[[!toc levels=1]]

## Differences from older guides

### Non-DES Kerberos service key

As of OpenAFS version 1.6.5, non-DES encryption keys are supported and should
be used for the long term AFS service key. The `asetkey` program is no longer
needed to import the Kerberos service key. The OpenAFS servers read a Kerberos
keytab file directly. Special changes to the Kerberos configuration files to
allow weak crypto and weak enctypes are no longer needed.

The OpenAFS `kaserver` should not be used in new installations. This guide
shows how to install MIT Kerberos.

### -noauth server mode

Older versions of OpenAFS required the `-noauth` flag to be used when first
setting up the system. This entailed starting the `bosserver` in `-noauth` mode
to disable authorization during the initial setup.  Modern versions of OpenAFS
can be installed without the `-noauth`, making the initial setup more secure
and without the need to restart the server processes.

### -dynroot client mode

This guide assumes the OpenAFS cache manager is started with `-dynroot` mode
enabled, which is the default these days.  This guide shows how to create the
initial top level AFS volumes when the cache manager is in `-dynroot` mode.

### Recommended options

This guide recommends using the `bosserver` restricted mode to improve security.

The default options for the fileserver are rather out of date for modern
hardware. This guide provides some examples of more suitable options for
the fileserver.

This guide shows how to setup the Demand Attach Fileserver.

## Naming conventions

When setting up an AFS cell on the internet, the convention is to user your
internet domain name for your Kerberos realm and AFS cell name.  The Kerberos
realm name should be uppercase and the AFS cell name should be lowercase.

Note, it is possible to create a AFS cell with a different name than the
Kerberos realm (or even use a single Kerberos realm in multiple cells).  See
the documentation for the OpenAFS `krb.conf` server configuration file for
details on mapping realms to cell names.

## Server setup

A minimal OS install is sufficient.

For a simple installation, you may use a single server to host the Kerberos
KDC, OpenAFS database server, and OpenAFS fileserver.  For a production
environment, it is recommended that the Kerberos KDC be deployed on a
dedicated, secure server, the OpenAFS database servers be deployed on three
separate machines, and multiple file servers deployed as needed.

### Disk Partitions

An important thing to keep in mind is that you'll need at least one partition
on the file server to store volumes for AFS.  This will be mounted at
`/vicepa`. If you have multiple partitions they can be mounted at `/vicepb`,
`/vicepc`, etc.  The file server uses file-based storage (not block based).
`ext3`, `ext4`, and `xfs` are commonly used filesystems on the vicep
partitions.

Clients should have a dedicated partition for the file cache. The cache
partition is traditionally mounted at `/usr/vice/cache`.

### Networking

DNS should be working correctly for forward and reverse name lookups before you
begin the Kerberos and OpenAFS installation.  `bind` can be installed if you
need a local DNS server.  Use `system-config-bind` to add a zone and entries.

Servers need at least one IPv4 interface that is accessible by the AFS clients.
IPv6 interfaces are not yet supported.

### Time keeping

Kerberos, and therefore OpenAFS, requires good clock synchronization between
clients and servers. Be sure to install and configure `ntp` and verify the
clocks are automatically kept in sync.

    # yum install -y ntp
    # service ntpd start
    # chkconfig ntpd on

### Firewall

The default firewall settings on RHEL will block the network ports used by
Kerberos and OpenAFS.  You will need to adjust the firewall rules on the
servers to allow traffic on these ports.

On the Kerberos server, open udp ports 88 and 646:

    # iptables -A INPUT -p udp --dport 88 -j ACCEPT
    # iptables -A INPUT -p udp --dport 646 -j ACCEPT
    # service iptables save

On the OpenAFS database servers, open the udp ports 7002, 7003, and 7007:

    # iptables -A INPUT -p udp --dport 7002 -j ACCEPT
    # iptables -A INPUT -p udp --dport 7003 -j ACCEPT
    # iptables -A INPUT -p udp --dport 7007 -j ACCEPT
    # service iptables save

On the OpenAFS file servers, open the udp ports 7000, 7005, and 7007:

    # iptables -A INPUT -p udp --dport 7000 -j ACCEPT
    # iptables -A INPUT -p udp --dport 7005 -j ACCEPT
    # iptables -A INPUT -p udp --dport 7007 -j ACCEPT
    # service iptables save

OpenAFS clients use udp port 7001. Open udp port 7001 on any system that will
have the OpenAFS client installed.

    # iptables -A INPUT -p udp --dport 7001 -j ACCEPT
    # service iptables save

### SELinux

Ideally, SELinux should be kept in enforcing mode. SELinux may be set to
enforcing for the OpenAFS client, but unfortunately, for the servers, there are
still several [issues][1] remaining before the servers can run out the box with
SELinux in enforcing mode.

For test installations, SELinux can be set into permissive mode, which will
print warnings instead of blocking:

    # setenforce 0
    # sed -i -e 's/SELINUX=enforcing/SELINUX=permissive/' /etc/selinux/config

See [this report][1] for workarounds if you need to run in enforcing mode.

[1]: https://bugzilla.redhat.com/show_bug.cgi?id=1136396

## Installing Kerberos

Install the Kerberos server and client packages with the command:

    # yum install -y krb5-server krb5-workstation krb5-libs

Replace every instance of `EXAMPLE.COM` with your realm name in the following
configuration files:

  * `/etc/krb5.conf`
  * `/var/kerberos/krb5kdc/kdc.conf`
  * `/var/kerberos/krb5kdc/kadm5.acl`

Replace every instance of the example hostname `kerberos.example.com` with
the actual hostname of your Kerberos server in the file `/etc/krb5.conf`.

Create the Kerberos database using the `krb5_util` command. You will be
prompted for a master principal password. Choose a password, keep it secret,
and keep it safe.

    # /usr/sbin/kdb5_util create -s

Start the Kerberos servers.

    # service krb5kdc start
    # service kadmin start
    # chkconfig krb5kdc on
    # chkconfig kadmin on

## Installing OpenAFS servers

### Installing servers

OpenAFS RPM packages for RHEL 6 are available on the [[OpenAFS]] website. The
packages may be installed using `yum`.  First, install the repository
definition with

    # version=<version>
    # rpm -i http://openafs.org/dl/openafs/${version}/openafs-release-rhel-${version}-1.noarch.rpm

where `<version>` is the OpenAFS version you wish to install, e.g. "1.6.9".

Use `yum` to install the OpenAFS server packages:

    # yum install -y openafs openafs-server openafs-docs openafs-krb5

Create the Kerberos AFS service key and export it to a keytab file:

    # cellname=<cellname>
    # kadmin.local -q "addprinc -randkey afs/${cellname}"
    # kadmin.local -q "ktadd -k /usr/afs/etc/rxkad.keytab afs/${cellname}"

where `<cellname>` is the name of your cell.  Start the OpenAFS servers:

    # service openafs-server start

Check the server log `/usr/afs/logs/BosLog` to verify the OpenAFS `bosserver`
process started.  Set the cell name with the command:

    # bos setcellname localhost ${cellname} -localauth

### Starting the database services

The `ptserver` process stores the AFS users and group names in your cell. The
`vlserver` process stores the file server locations of the AFS volumes in your
cell.  Start the OpenAFS database processes with the commands:

    # bos create localhost ptserver simple -cmd /usr/afs/bin/ptserver -localauth
    # bos create localhost vlserver simple -cmd /usr/afs/bin/vlserver -localauth

Check the log files `BosLog`, `PTLog`, `VLLog` in the `/usr/afs/logs`
directory to verify the `ptserver` and `vlserver` started.

### Starting the file server

Start the file server. This is a rather long command line.

    # options="-d 1"  TODO: list some practical options
    # bos create localhost \
       dafs dafs \
       -cmd "/usr/afs/bin/dafileserver ${options}" \
       "/usr/afs/bin/davolserver -log" \
       "/usr/afs/bin/salvageserver" \
       "/usr/afs/bin/dasalvager" \
       -localauth

Check the servers logs `BosLog`, `FileLog`, `VolserLog`, and `SalsrvLog`, in
`/usr/afs/logs' to verify the file server started.  At this point the OpenAFS
server processes should be running.

### Creating the admin account

Create a user account for Kerberos and AFS administration.

    # myname=<username>
    # kadmin.local -q "addprinc ${myname}/admin"
    Enter password: <password>
    Re-enter password: <password>
    # pts createuser ${myname}.admin -localauth
    # pts adduser ${myname}.admin system:administrators -localauth
    # bos adduser localhost ${myname}.admin -localauth

where `<myname>` is your user name and `<password>` is your chosen password.

The admin principal can be any name you want.  The recommended practice is to
create two principals for admins: one for your normal user, and an additional
admin account. For example, I may have `steve` and `steve/admin`. Note that for
Kerberos 5, the name is `steve/admin@REALM`, whereas in AFS, the name is
`steve.admin`. Use `steve.admin` for all AFS commands.  Since this is to be an
administrator we will also register it as such with the `bos` server. We can give
it administrator rights by adding it to the group `system:administrators`. This
is an AFS default group. The `pts membership` command will list all the groups
that your user is a member of. Verify that it lists `system:administrators`.

### Create the root volumes

At this point we need our `/vicepa` partition.  You should have done this when
installing the operating system. If you have not, do it now, then restart the
fileserver with `service openafs-server restart`.  (If this is only a test
system you may create a pseudo partition without needing to create an actual
separate filesystem. To do this, create an empty directory called `/vicepa` and
then create an empty file called `/vicepa/AlwaysAttach`, then restart the file
server with `service openafs-server restart`.)

Create the root volumes with the commands:

    # vos create localhost a root.afs -localauth
    # vos create localhost a root.cell -localauth

Check the volume location database to verify the two volumes are listed.

    # vos listvldb

Finally, now that the server configuration is done, put the `bosserver` into
the more secure restricted mode, which disables several bos commands which are
strictly not needed for normal operation.

    # bos setrestricted localhost -mode 1 -localauth

This completes the server side setup. At this point will need to install the
OpenAFS cache manager (client), setup the top level directories, and then start
adding files to your new cell.  The cache manager may be installed on a
separate machine (for example, your laptop.)  Also, you will no longer be using
the `root` user to run OpenAFS commands, but instead from this point forward
you should use your Kerberos credentials.

## Installing OpenAFS Client

### Kernel Module

If installing the cache manager on an OpenAFS server, first remove the symlinks
created by `bosserver`. These will be in the way if the client is installed.

    # test -h /usr/vice/etc/ThisCell && rm /usr/vice/etc/ThisCell
    # test -h /usr/vice/etc/CellServDB && rm /usr/vice/etc/CellServDB

Install the OpenAFS client with yum:

    # yum install -y openafs openafs-client openafs-krb5 kmod-openafs

You may need to reboot you system if the kernel version was updated.

The OpenAFS kernel module must match your kernel version.  If a pre-built
module is not available for the kernel version you are running, the yum install
of kmod-openafs will fail.  If this happens, then you may use
[DKMS to build and install|RpmClientInstallationWithDKMS]] a kernel module
that matches your kernel version.

    # yum install -y dkms gcc
    # yum install -y openafs openafs-client openafs-krb5 dkms-openafs

Alternately, you can build the kernel module from the source rpm.  You will
need to download the OpenAFS SRPM file from OpenAFS.org, install the
`kernel-devel` package that matches your kernel version, and the necessary
build tools.  Use the `rpmbuild` command to build the kernel module.

    # rpmbuild --rebuild -bb \
        --define 'build_userspace 0' \
        --define 'build_modules 1' \
        openafs-<version>-1.src.rpm

### Client side configuration

`/usr/afs/etc` is the location for the server files. We also need to configure
the client. The client files are located in `/usr/vice/etc`. RPM based OpenAFS
packages are set up in such a way that there are two `CellServDB` client
files in `/usr/vice/etc`: `CellServDB.dist` and `CellServDB.local`. We will
copy ours to the local list.

    # cp /usr/afs/etc/CellServDB /usr/vice/etc/CellServDB.local
    # cp /usr/afs/etc/ThisCell /usr/vice/etc/ThisCell

The RPM based `openafs-client` init script will combine the `CellServDB.dist`
and `CellServDB.local` files into the `CellServDB` file, which the cache
manager reads on startup.

### Start the cache manager

Start the cache manager with the command:

    # service openafs-client start

Run the `mount` command to verify the AFS filesystem is mounted at `/afs`.

Try logging in to AFS. `kinit` logs you into Kerberos (this is the normal
Kerberos utility). `aklog` gets you an AFS token.  The `tokens` command lists
the tokens you have. You should see `afs@<cellname>`. If you run into problems, you
can use `klist` to list your Kerberos tickets, or `aklog` with the `-d` flag.

    $ kinit <username>/admin
    <password>
    $ aklog
    $ tokens

## Setting up the cell root directory

Now we will set up the root directories.  The root directory for the AFS
namespace is in the volume called `root.afs`. The root directory of your cell
should be in a volume called `root.cell`. You will need to set the ACLs for
these directories.  AFS access rights are rather different from those in UNIX.
I suggest reading the IBM documentation for this; it still applies.

The cache manager is started in `-dynroot` mode on RPM-based installations.
This allows the cache manager to mount the AFS filesystem without the need to
contact the OpenAFS servers. The side-effect of `-dynroot` is the `root.afs`
volume cannot be accessed directly.  Fortunately, we can use "magic" `.:mount`
directory to access the `root.afs` volume.

Set up the top level directories.

    $ cellname=$(cat /usr/vice/etc/ThisCell)
    
    $ cd /afs/.:mount/${cellname}:root.afs/
    $ fs mkmount ${cellname} root.cell -cell ${cellname}
    $ fs mkmount .${cellname} root.cell -cell ${cellname} -rw
    $ fs setacl . system:anyuser read
    
    $ cd /afs/.:mount/${cellname}:root.cell/
    $ fs setacl . system:anyuser read

Replicate the root volumes so that you have read only copies.  Later, if more
file servers are added to the cell, additional read-only copies should be made.

    $ server=<hostname of the fileserver>
    $ vos addsite ${server} a root.afs
    $ vos release root.afs
    $ vos addsite ${server} a root.cell
    $ vos release root.cell

## Adding users and volumes

Now that OpenAFS is installed, the site specific AFS volumes and directory
structure can be set up. Users should be made, along with their home
volumes. ACLs for the volume directories should be established.

This section provides an example setup. The names of the volumes and
directories can be specific to your needs.

You must first authenticate as a Kerberos/AFS admin to run the commands
shown in this section.

    $ kadmin <username>/admin
    $ aklog

### Creating user accounts

We can create a user by registering it to Kerberos and the `ptserver` database.
If you use integrated login, make sure that the users' UNIX uids and pts ids
match.

    $ kadmin -q "addprinc <username>"
    <enter password for username>
    $ pts createuser <username> -id <numeric uid>

If you use integrated login, make sure that you add an entry to /etc/passwd or
whatever means you use of distributing user information.

### Setting up volumes for users

First, we can make a top level volume to contain the mount points to volumes
for individuals. The IBM documentation suggests making a directory
`/afs/<cellname>/user` with volume name user for all of your AFS users. Some
sites have adopted the directory `home` instead of `user`.  If you use `home`,
your users may feel more comfortable, as this is the convention in Linux and
most UNIXes.

The following commands create the `home` volume and make a read-only replica:

    $ vos create <fileserver> a home
    $ cd /afs/.<cellname>
    $ fs mkmount home home
    $ vos addsite <fileserver> a home
    $ vos release root.cell
    $ vos release home

Now you can create directories for any of your users. We will not replicate
these volumes. By not replicating them, we force the cache manager to access a
read/write volume. This means that even if we access the cell through the
read-only volume we can still access our read/write user directories (this is
what you want). Maxquota 0 means that there is no size restriction to the
volume. You can give it a restriction if you like (the default is 5mb). Do
these commands for every directory you like.

    $ vos create <fileserver> a home.<uid> -maxquota 0
    $ cd /afs/.<cellname>/home
    $ fs mkmount <user> home.<uid>
    $ vos release home

The home volume is released to make the new directories are visible from the
read only mount point.

### Setting ACLs

Now that we have volumes, we should set some restrictions on those volumes.
If you trust the users not to make directories *world writable*, you
can give the user of the directory full rights.

    $ cd /afs/.<cellname>/home
    $ fs setacl -dir <user> -acl <user> all

To give the users read and write access, but not rights to change ACLs,

    $ cd /afs/.<cellname>/home
    $ fs setacl -dir <user> -acl <user> write