<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE html>
<html xmlns="http://www.w3.org/1999/xhtml">
<body>
<h1>Connection authentication</h1>
<p>
When connecting to libvirt, some connections may require client
authentication before allowing use of the APIs. The set of possible
authentication mechanisms is administrator controlled, independent
of applications using libvirt. Once authenticated, libvirt can apply
fine grained <a href="acl.html">access control</a> to the operations
performed by a client.
</p>
<ul id="toc"></ul>
<h2><a id="Auth_client_config">Client configuration</a></h2>
<p>
When connecting to a remote hypervisor which requires authentication,
most libvirt applications will prompt the user for the credentials. It is
also possible to provide a client configuration file containing all the
authentication credentials, avoiding any interaction. Libvirt will look
for the authentication file using the following sequence:
</p>
<ol>
<li>The file path specified by the $LIBVIRT_AUTH_FILE environment
variable.</li>
<li>The file path specified by the "authfile=/some/file" URI
query parameter</li>
<li>The file $XDG_CONFIG_HOME/libvirt/auth.conf</li>
<li>The file /etc/libvirt/auth.conf</li>
</ol>
<p>
The auth configuration file uses the traditional <code>".ini"</code>
style syntax. There are two types of groups that can be present in
the config. First there are one or more <strong>credential</strong>
sets, which provide the actual authentication credentials. The keys
within the group may be:
</p>
<ul>
<li><code>username</code>: the user login name to act as. This
is relevant for ESX, Xen, HyperV and SSH, but probably not
the one you want to libvirtd with SASL.</li>
<li><code>authname</code>: the name to authorize as. This is
what is commonly required for libvirtd with SASL.</li>
<li><code>password</code>: the secret password</li>
<li><code>realm</code>: the domain realm for SASL, mostly
unused</li>
</ul>
<p>
Each set of credentials has a name, which is part of the group
entry name. Overall the syntax is
</p>
<pre>
[credentials-$NAME]
credname1=value1
credname2=value2</pre>
<p>
For example, to define two sets of credentials used for production
and test machines, using libvirtd, and a further ESX server for dev:
</p>
<pre>
[credentials-test]
authname=fred
password=123456
[credentials-prod]
authname=bar
password=letmein
[credentials-dev]
username=joe
password=hello
[credentials-defgrp]
username=defuser
password=defpw</pre>
<p>
The second set of groups provide mappings of credentials to
specific machine services. The config file group names compromise
the service type and host:
</p>
<pre>
[auth-$SERVICE-$HOSTNAME]
credentials=$CREDENTIALS</pre>
<p>
For example, following the previous example, here is how to
map some machines. For convenience libvirt supports a default
mapping of credentials to machines:
</p>
<pre>
[auth-libvirt-test1.example.com]
credentials=test
[auth-libvirt-test2.example.com]
credentials=test
[auth-libvirt-demo3.example.com]
credentials=test
[auth-libvirt-prod1.example.com]
credentials=prod
[auth-libvirt-default]
credentials=defgrp
[auth-esx-dev1.example.com]
credentials=dev
[auth-esx-default]
credentials=defgrp</pre>
<p>
The following service types are known to libvirt:
</p>
<ul>
<li><code>esx</code> - used for connections to an ESX or
VirtualCenter server</li>
<li><code>hyperv</code> - used for connections to an HyperV
server</li>
<li><code>libvirt</code> - used for connections to a libvirtd
server, which is configured with SASL auth</li>
<li><code>ssh</code> - used for connections to a remote QEMU driver
over SSH</li>
</ul>
<p>
Applications using libvirt are free to use this same configuration
file for storing other credentials. For example, it can be used
to storage VNC or SPICE login credentials
</p>
<h2><a id="ACL_server_config">Server configuration</a></h2>
<p>
The libvirt daemon allows the administrator to choose the authentication
mechanisms used for client connections on each network socket independently.
This is primarily controlled via the libvirt daemon master config file in
<code>/etc/libvirt/libvirtd.conf</code>. Each of the libvirt sockets can
have its authentication mechanism configured independently. There is
currently a choice of <code>none</code>, <code>polkit</code>, and <code>sasl</code>.
The SASL scheme can be further configured to choose between a large
number of different mechanisms.
</p>
<h2><a id="ACL_server_unix_perms">UNIX socket permissions/group</a></h2>
<p>
If libvirt does not contain support for PolicyKit, then access control for
the UNIX domain socket is done using traditional file user/group ownership
and permissions. There are 2 sockets, one for full read-write access, the
other for read-only access. The RW socket will be restricted (mode 0700) to
only allow the <code>root</code> user to connect. The read-only socket will
be open access (mode 0777) to allow any user to connect.
</p>
<p>
To allow non-root users greater access, the <code>libvirtd.conf</code> file
can be edited to change the permissions via the <code>unix_sock_rw_perms</code>,
config parameter and to set a user group via the <code>unix_sock_group</code>
parameter. For example, setting the former to mode <code>0770</code> and the
latter <code>wheel</code> would let any user in the wheel group connect to
the libvirt daemon.
</p>
<h2><a id="ACL_server_polkit">UNIX socket PolicyKit auth</a></h2>
<p>
If libvirt contains support for PolicyKit, then access control options are
more advanced. The <code>auth_unix_rw</code> parameter will default to
<code>polkit</code>, and the file permissions will default to <code>0777</code>
even on the RW socket. Upon connecting to the socket, the client application
will be required to identify itself with PolicyKit. The default policy for the
RW daemon socket will require any application running in the current desktop
session to authenticate using the user's password. This is akin to <code>sudo</code>
auth, but does not require that the client application ultimately run as root.
Default policy will still allow any application to connect to the RO socket.
</p>
<p>
The default policy can be overridden by creating a new policy file in the
<code>/etc/polkit-1/rules.d</code> directory. Information on the options
available can be found by reading the <code>polkit(8)</code> man page. The
two libvirt actions are named <code>org.libvirt.unix.manage</code> for full
management access, and <code>org.libvirt.unix.monitor</code> for read-only
access.
</p>
<p>
As an example, creating <code>/etc/polkit-1/rules.d/80-libvirt-manage.rules</code>
with the following gives the user <code>fred</code> full management access
when accessing from an active local session:
</p>
<pre>polkit.addRule(function(action, subject) {
if (action.id == "org.libvirt.unix.manage" &&
subject.local && subject.active && subject.user == "fred") {
return polkit.Result.YES;
}
});</pre>
<p>
Older versions of PolicyKit used policy files ending with .pkla in the
local override directory <code>/etc/polkit-1/localauthority/50-local.d/</code>.
Compatibility with this older format is provided by <a
href="https://pagure.io/polkit-pkla-compat">polkit-pkla-compat</a>. As an
example, this gives the user <code>fred</code> full management access:
</p>
<pre>[Allow fred libvirt management permissions]
Identity=unix-user:fred
Action=org.libvirt.unix.manage
ResultAny=yes
ResultInactive=yes
ResultActive=yes</pre>
<h2><a id="ACL_server_sasl">SASL pluggable authentication</a></h2>
<p>
Libvirt integrates with the cyrus-sasl library to provide a pluggable authentication
system using the SASL protocol. SASL can be used in combination with libvirtd's TLS
or TCP socket listeners. When used with the TCP listener, the SASL mechanism is
rqeuired to provide session encryption in addition to authentication. Only a very
few SASL mechanisms are able to do this, and of those that can do it, only the
GSSAPI plugin is considered acceptably secure by modern standards:
</p>
<dl>
<dt>GSSAPI</dt>
<dd><strong>This is the current default mechanism to use with libvirtd</strong>.
It uses the Kerberos v5 authentication protocol underneath, and assuming
the Kerberos client/server are configured with modern ciphers (AES),
it provides strong session encryption capabilities.</dd>
<dt>DIGEST-MD5</dt>
<dd>This was previously set as the default mechanism to use with libvirtd.
It provides a simple username/password based authentication mechanism
that includes session encryption.
<a href="https://tools.ietf.org/html/rfc6331">RFC 6331</a>, however,
documents a number of serious security flaws with DIGEST-MD5 and as a
result marks it as <code>OBSOLETE</code>. Specific concerns are that
it is vulnerable to MITM attacks and the MD5 hash can be brute-forced
to reveal the password. A replacement is provided via the SCRAM mechanism,
however, note that this does not provide encryption, so the SCRAM
mechanism can only be used on the libvirtd TLS listener.
</dd>
<dt>PASSDSS-3DES-1</dt>
<dd>This provides a simple username/password based authentication
mechanism that includes session encryption. The current cyrus-sasl
implementation does not provide a way to validate the server's
public key identity, thus it is susceptible to a MITM attacker
impersonating the server. It is also not enabled in many OS
distros when building SASL libraries.</dd>
<dt>KERBEROS_V4</dt>
<dd>This uses the obsolete Kerberos v4 protocol to provide both authentication
and session encryption. Kerberos v4 protocol has been obsolete since the
early 1990's and has known security vulnerabilities so this will never be
used in practice.</dd>
</dl>
<p>
Other SASL mechanisms, not listed above, can only be used when the libvirtd
TLS or UNIX socket listeners.
</p>
<h3><a id="ACL_server_username">Username/password auth</a></h3>
<p>
As noted above, the DIGEST-MD5 mechanism is considered obsolete and should
not be used anymore. To provide a simple username/password auth scheme on
the libvirt UNIX socket or TLS listeners, however, it is possible to use
the SCRAM mechanism. The <code>auth_unix_ro</code>, <code>auth_unix_rw</code>,
<code>auth_tls</code> config params in <code>libvirt.conf</code> can be used
to turn on SASL auth in these listeners.
</p>
<p>
Since the libvirt SASL config file defaults to using GSSAPI (Kerberos), a
config change is required to enable plain password auth. This is done by
editting <code>/etc/sasl2/libvirt.conf</code> to set the <code>mech_list</code>
parameter to <code>scram-sha-1</code>.
</p>
<p>
Out of the box, no user accounts are defined, so no clients will be able to authenticate
on the TCP socket. Adding users and setting their passwords is done with the <code>saslpasswd2</code>
command. When running this command it is important to tell it that the appname is <code>libvirt</code>.
As an example, to add a user <code>fred</code>, run
</p>
<pre>
# saslpasswd2 -a libvirt fred
Password: xxxxxx
Again (for verification): xxxxxx
</pre>
<p>
To see a list of all accounts the <code>sasldblistusers2</code> command can be used.
This command expects to be given the path to the libvirt user database, which is kept
in <code>/etc/libvirt/passwd.db</code>
</p>
<pre>
# sasldblistusers2 -f /etc/libvirt/passwd.db
fred@t60wlan.home.berrange.com: userPassword
</pre>
<p>
Finally, to disable a user's access, the <code>saslpasswd2</code> command can be used
again:
</p>
<pre>
# saslpasswd2 -a libvirt -d fred
</pre>
<h3><a id="ACL_server_kerberos">GSSAPI/Kerberos auth</a></h3>
<p>
The plain TCP listener of the libvirt daemon defaults to using SASL for authentication.
The libvirt SASL config also defaults to GSSAPI, so there is no need to edit the
SASL config when using GSSAPI. If the libvirtd TLS or UNIX listeners are used,
then the Kerberos session encryption will be disabled since it is not required
in these scenarios - only the plain TCP listener needs encryption
</p>
<p>
Some operating systems do not install the SASL kerberos plugin by default. It
may be necessary to install a sub-package such as <code>cyrus-sasl-gssapi</code>.
To check whether the Kerberos plugin is installed run the <code>pluginviewer</code>
program and verify that <code>gssapi</code> is listed, e.g.:
</p>
<pre>
# pluginviewer
...snip...
Plugin "gssapiv2" [loaded], API version: 4
SASL mechanism: GSSAPI, best SSF: 56
security flags: NO_ANONYMOUS|NO_PLAINTEXT|NO_ACTIVE|PASS_CREDENTIALS|MUTUAL_AUTH
features: WANT_CLIENT_FIRST|PROXY_AUTHENTICATION|NEED_SERVER_FQDN
</pre>
<p>
Next it is necessary for the administrator of the Kerberos realm to
issue a principal for the libvirt server. There needs to be one
principal per host running the libvirt daemon. The principal should be
named <code>libvirt/full.hostname@KERBEROS.REALM</code>. This is
typically done by running the <code>kadmin.local</code> command on the
Kerberos server, though some Kerberos servers have alternate ways of
setting up service principals. Once created, the principal should be
exported to a keytab, copied to the host running the libvirt daemon
and placed in <code>/etc/libvirt/krb5.tab</code>
</p>
<pre>
# kadmin.local
kadmin.local: add_principal libvirt/foo.example.com
Enter password for principal "libvirt/foo.example.com@EXAMPLE.COM":
Re-enter password for principal "libvirt/foo.example.com@EXAMPLE.COM":
Principal "libvirt/foo.example.com@EXAMPLE.COM" created.
kadmin.local: ktadd -k /root/libvirt-foo-example.tab libvirt/foo.example.com@EXAMPLE.COM
Entry for principal libvirt/foo.example.com@EXAMPLE.COM with kvno 4, encryption type Triple DES cbc mode with HMAC/sha1 added to keytab WRFILE:/root/libvirt-foo-example.tab.
Entry for principal libvirt/foo.example.com@EXAMPLE.COM with kvno 4, encryption type ArcFour with HMAC/md5 added to keytab WRFILE:/root/libvirt-foo-example.tab.
Entry for principal libvirt/foo.example.com@EXAMPLE.COM with kvno 4, encryption type DES with HMAC/sha1 added to keytab WRFILE:/root/libvirt-foo-example.tab.
Entry for principal libvirt/foo.example.com@EXAMPLE.COM with kvno 4, encryption type DES cbc mode with RSA-MD5 added to keytab WRFILE:/root/libvirt-foo-example.tab.
kadmin.local: quit
# scp /root/libvirt-foo-example.tab root@foo.example.com:/etc/libvirt/krb5.tab
# rm /root/libvirt-foo-example.tab
</pre>
<p>
Any client application wishing to connect to a Kerberos enabled libvirt server
merely needs to run <code>kinit</code> to gain a user principal. This may well
be done automatically when a user logs into a desktop session, if PAM is set up
to authenticate against Kerberos.
</p>
</body>
</html>
