.\" $OpenBSD: authpf.8,v 1.47 2009/01/06 03:11:50 mcbride Exp $
.\"
.\" Copyright (c) 1998-2007 Bob Beck (beck@openbsd.org>.  All rights reserved.
.\"
.\" Permission to use, copy, modify, and distribute this software for any
.\" purpose with or without fee is hereby granted, provided that the above
.\" copyright notice and this permission notice appear in all copies.
.\"
.\" THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
.\" WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
.\" MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
.\" ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
.\" WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
.\" ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
.\" OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
.\"
.Dd $Mdocdate: January 6 2009 $
.Dt AUTHPF 8
.Os
.Sh NAME
.Nm authpf ,
.Nm authpf-noip
.Nd authenticating gateway user shell
.Sh SYNOPSIS
.Nm authpf
.Nm authpf-noip
.Sh DESCRIPTION
.Nm
is a user shell for authenticating gateways.
It is used to change
.Xr pf 4
rules when a user authenticates and starts a session with
.Xr sshd 8
and to undo these changes when the user's session exits.
Typical use would be for a gateway that authenticates users before
allowing them Internet use, or a gateway that allows different users into
different places.
Combined with properly set up filter rules and secure switches,
.Nm
can be used to ensure users are held accountable for their network traffic.
It is meant to be used with users who can connect via
.Xr ssh 1
only, and requires the
.Xr pf 4
subsystem to be enabled.
.Pp
.Nm authpf-noip
is a user shell
which allows multiple connections to take
place from the same IP address.
It is useful primarily in cases where connections are tunneled via
the gateway system, and can be directly associated with the user name.
It cannot ensure accountability when
classifying connections by IP address;
in this case the client's IP address
is not provided to the packet filter via the
.Ar client_ip
macro or the
.Ar authpf_users
table.
Additionally, states associated with the client IP address
are not purged when the session is ended.
.Pp
To use either
.Nm
or
.Nm authpf-noip ,
the user's shell needs to be set to
.Pa /usr/sbin/authpf
or
.Pa /usr/sbin/authpf-noip .
.Pp
.Nm
uses the
.Xr pf.conf 5
syntax to change filter and translation rules for an individual
user or client IP address as long as a user maintains an active
.Xr ssh 1
session, and logs the successful start and end of a session to
.Xr syslogd 8 .
.Nm
retrieves the client's connecting IP address via the
.Ev SSH_CLIENT
environment variable and, after performing additional access checks,
reads a template file to determine what filter and translation rules
(if any) to add, and
maintains the list of IP addresses of connected users in the
.Ar authpf_users
table.
On session exit the same rules and table entries that were added at startup
are removed, and all states associated with the client's IP address are purged.
.Pp
Each
.Nm
process stores its rules in a separate ruleset inside a
.Xr pf 4
.Pa anchor
shared by all
.Nm
processes.
By default, the
.Pa anchor
name "authpf" is used, and the ruleset names equal the username and PID of the
.Nm
processes as "username(pid)".
The following rules need to be added to the main ruleset
.Pa /etc/pf.conf
in order to cause evaluation of any
.Nm
rules:
.Bd -literal -offset indent
nat-anchor "authpf/*"
rdr-anchor "authpf/*"
binat-anchor "authpf/*"
anchor "authpf/*"
.Ed
.Pp
The "/*" at the end of the anchor name is required for
.Xr pf 4
to process the rulesets attached to the anchor by
.Nm authpf .
.Sh FILTER AND TRANSLATION RULES
Filter and translation rules for
.Nm
use the same format described in
.Xr pf.conf 5 .
The only difference is that these rules may (and probably should) use
the macro
.Em user_ip ,
which is assigned the connecting IP address whenever
.Nm
is run.
Additionally, the macro
.Em user_id
is assigned the user name.
.Pp
Filter and translation rules are stored in a file called
.Pa authpf.rules .
This file will first be searched for in
.Pa /etc/authpf/users/$USER/
and then in
.Pa /etc/authpf/ .
Only one of these files will be used if both are present.
.Pp
Per-user rules from the
.Pa /etc/authpf/users/$USER/
directory are intended to be used when non-default rules
are needed on an individual user basis.
It is important to ensure that a user can not write or change
these configuration files.
.Pp
The
.Pa authpf.rules
file must exist in one of the above locations for
.Nm
to run.
.Sh CONFIGURATION
Options are controlled by the
.Pa /etc/authpf/authpf.conf
file.
If the file is empty, defaults are used for all
configuration options.
The file consists of pairs of the form
.Li name=value ,
one per line.
Currently, the allowed values are as follows:
.Bl -tag -width Ds
.It anchor=name
Use the specified
.Pa anchor
name instead of "authpf".
.It table=name
Use the specified
.Pa table
name instead of "authpf_users".
.El
.Sh USER MESSAGES
On successful invocation,
.Nm
displays a message telling the user he or she has been authenticated.
It will additionally display the contents of the file
.Pa /etc/authpf/authpf.message
if the file exists and is readable.
.Pp
There exist two methods for providing additional granularity to the control
offered by
.Nm
- it is possible to set the gateway to explicitly allow users who have
authenticated to
.Xr ssh 1
and deny access to only a few troublesome individuals.
This is done by creating a file with the banned user's login name as the
filename in
.Pa /etc/authpf/banned/ .
The contents of this file will be displayed to a banned user, thus providing
a method for informing the user that they have been banned, and where they can
go and how to get there if they want to have their service restored.
This is the default behaviour.
.Pp
It is also possible to configure
.Nm
to only allow specific users access.
This is done by listing their login names, one per line, in
.Pa /etc/authpf/authpf.allow .
A group of users can also be indicated by prepending "%" to the group name,
and all members of a login class can be indicated by prepending "@" to the
login class name.
If "*" is found on a line, then all usernames match.
If
.Nm
is unable to verify the user's permission to use the gateway, it will
print a brief message and die.
It should be noted that a ban takes precedence over an allow.
.Pp
On failure, messages will be logged to
.Xr syslogd 8
for the system administrator.
The user does not see these, but will be told the system is unavailable due to
technical difficulties.
The contents of the file
.Pa /etc/authpf/authpf.problem
will also be displayed if the file exists and is readable.
.Sh CONFIGURATION ISSUES
.Nm
maintains the changed filter rules as long as the user maintains an
active session.
It is important to remember however, that the existence
of this session means the user is authenticated.
Because of this, it is important to configure
.Xr sshd 8
to ensure the security of the session, and to ensure that the network
through which users connect is secure.
.Xr sshd 8
should be configured to use the
.Ar ClientAliveInterval
and
.Ar ClientAliveCountMax
parameters to ensure that a ssh session is terminated quickly if
it becomes unresponsive, or if arp or address spoofing is used to
hijack the session.
Note that TCP keepalives are not sufficient for
this, since they are not secure.
Also note that the various SSH tunnelling mechanisms,
such as
.Ar AllowTcpForwarding
and
.Ar PermitTunnel ,
should be disabled for
.Nm
users to prevent them from circumventing restrictions imposed by the
packet filter ruleset.
.Pp
.Nm
will remove state table entries that were created during a user's
session.
This ensures that there will be no unauthenticated traffic
allowed to pass after the controlling
.Xr ssh 1
session has been closed.
.Pp
.Nm
is designed for gateway machines which typically do not have regular
(non-administrative) users using the machine.
An administrator must remember that
.Nm
can be used to modify the filter rules through the environment in
which it is run, and as such could be used to modify the filter rules
(based on the contents of the configuration files) by regular
users.
In the case where a machine has regular users using it, as well
as users with
.Nm
as their shell, the regular users should be prevented from running
.Nm
by using the
.Pa /etc/authpf/authpf.allow
or
.Pa /etc/authpf/banned/
facilities.
.Pp
.Nm
modifies the packet filter and address translation rules, and because
of this it needs to be configured carefully.
.Nm
will not run and will exit silently if the
.Pa /etc/authpf/authpf.conf
file does not exist.
After considering the effect
.Nm
may have on the main packet filter rules, the system administrator may
enable
.Nm
by creating an appropriate
.Pa /etc/authpf/authpf.conf
file.
.Sh EXAMPLES
.Sy Control Files
\- To illustrate the user-specific access control
mechanisms, let us consider a typical user named bob.
Normally, as long as bob can authenticate himself, the
.Nm
program will load the appropriate rules.
Enter the
.Pa /etc/authpf/banned/
directory.
If bob has somehow fallen from grace in the eyes of the
powers-that-be, they can prohibit him from using the gateway by creating
the file
.Pa /etc/authpf/banned/bob
containing a message about why he has been banned from using the network.
Once bob has done suitable penance, his access may be restored by moving or
removing the file
.Pa /etc/authpf/banned/bob .
.Pp
Now consider a workgroup containing alice, bob, carol and dave.
They have a
wireless network which they would like to protect from unauthorized use.
To accomplish this, they create the file
.Pa /etc/authpf/authpf.allow
which lists their login ids, group prepended with "%", or login class
prepended with "@", one per line.
At this point, even if eve could authenticate to
.Xr sshd 8 ,
she would not be allowed to use the gateway.
Adding and removing users from
the work group is a simple matter of maintaining a list of allowed userids.
If bob once again manages to annoy the powers-that-be, they can ban him from
using the gateway by creating the familiar
.Pa /etc/authpf/banned/bob
file.
Though bob is listed in the allow file, he is prevented from using
this gateway due to the existence of a ban file.
.Pp
.Sy Distributed Authentication
\- It is often desirable to interface with a
distributed password system rather than forcing the sysadmins to keep a large
number of local password files in sync.
The
.Xr login.conf 5
mechanism in
.Ox
can be used to fork the right shell.
To make that happen,
.Xr login.conf 5
should have entries that look something like this:
.Bd -literal -offset indent
shell-default:shell=/bin/csh

default:\e
	...
	:shell=/usr/sbin/authpf

daemon:\e
	...
	:shell=/bin/csh:\e
	:tc=default:

staff:\e
	...
	:shell=/bin/csh:\e
	:tc=default:
.Ed
.Pp
Using a default password file, all users will get
.Nm
as their shell except for root who will get
.Pa /bin/csh .
.Pp
.Sy SSH Configuration
\- As stated earlier,
.Xr sshd 8
must be properly configured to detect and defeat network attacks.
To that end, the following options should be added to
.Xr sshd_config 5 :
.Bd -literal -offset indent
Protocol 2
ClientAliveInterval 15
ClientAliveCountMax 3
.Ed
.Pp
This ensures that unresponsive or spoofed sessions are terminated within a
minute, since a hijacker should not be able to spoof ssh keepalive messages.
.Pp
.Sy Banners
\- Once authenticated, the user is shown the contents of
.Pa /etc/authpf/authpf.message .
This message may be a screen-full of the appropriate use policy, the contents
of
.Pa /etc/motd
or something as simple as the following:
.Bd -literal -offset indent
This means you will be held accountable by the powers that be
for traffic originating from your machine, so please play nice.
.Ed
.Pp
To tell the user where to go when the system is broken,
.Pa /etc/authpf/authpf.problem
could contain something like this:
.Bd -literal -offset indent
Sorry, there appears to be some system problem. To report this
problem so we can fix it, please phone 1-900-314-1597 or send
an email to remove@bulkmailerz.net.
.Ed
.Pp
.Sy Packet Filter Rules
\- In areas where this gateway is used to protect a
wireless network (a hub with several hundred ports), the default rule set as
well as the per-user rules should probably allow very few things beyond
encrypted protocols like
.Xr ssh 1 ,
.Xr ssl 8 ,
or
.Xr ipsec 4 .
On a securely switched network, with plug-in jacks for visitors who are
given authentication accounts, you might want to allow out everything.
In this context, a secure switch is one that tries to prevent address table
overflow attacks.
.Pp
Example
.Pa /etc/pf.conf :
.Bd -literal
# by default we allow internal clients to talk to us using
# ssh and use us as a dns server.
internal_if="fxp1"
gateway_addr="10.0.1.1"
nat-anchor "authpf/*"
rdr-anchor "authpf/*"
binat-anchor "authpf/*"
block in on $internal_if from any to any
pass in quick on $internal_if proto tcp from any to $gateway_addr \e
      port = ssh
pass in quick on $internal_if proto udp from any to $gateway_addr \e
      port = domain
anchor "authpf/*"
.Ed
.Pp
.Sy For a switched, wired net
\- This example
.Pa /etc/authpf/authpf.rules
makes no real restrictions; it turns the IP address on and off, logging
TCP connections.
.Bd -literal
external_if = "xl0"
internal_if = "fxp0"

pass in log quick on $internal_if proto tcp from $user_ip to any
pass in quick on $internal_if from $user_ip to any
.Ed
.Pp
.Sy For a wireless or shared net
\- This example
.Pa /etc/authpf/authpf.rules
could be used for an insecure network (such as a public wireless network) where
we might need to be a bit more restrictive.
.Bd -literal
internal_if="fxp1"
ipsec_gw="10.2.3.4"

# rdr ftp for proxying by ftp-proxy(8)
rdr on $internal_if proto tcp from $user_ip to any port 21 \e
      -> 127.0.0.1 port 8021

# allow out ftp, ssh, www and https only, and allow user to negotiate
# ipsec with the ipsec server.
pass in log quick on $internal_if proto tcp from $user_ip to any \e
      port { 21, 22, 80, 443 }
pass in quick on $internal_if proto tcp from $user_ip to any \e
      port { 21, 22, 80, 443 }
pass in quick proto udp from $user_ip to $ipsec_gw port = isakmp
pass in quick proto esp from $user_ip to $ipsec_gw
.Ed
.Pp
.Sy Dealing with NAT
\- The following
.Pa /etc/authpf/authpf.rules
shows how to deal with NAT, using tags:
.Bd -literal
ext_if = "fxp1"
ext_addr = 129.128.11.10
int_if = "fxp0"
# nat and tag connections...
nat on $ext_if from $user_ip to any tag $user_ip -> $ext_addr
pass in quick on $int_if from $user_ip to any
pass out log quick on $ext_if tagged $user_ip
.Ed
.Pp
With the above rules added by
.Nm ,
outbound connections corresponding to each users NAT'ed connections
will be logged as in the example below, where the user may be identified
from the ruleset name.
.Bd -literal
# tcpdump -n -e -ttt -i pflog0
Oct 31 19:42:30.296553 rule 0.bbeck(20267).1/0(match): pass out on fxp1: \e
129.128.11.10.60539 > 198.137.240.92.22: S 2131494121:2131494121(0) win \e
16384 <mss 1460,nop,nop,sackOK> (DF)
.Ed
.Pp
.Sy Using the authpf_users table
\- Simple
.Nm
settings can be implemented without an anchor by just using the "authpf_users"
.Pa table .
For example, the following
.Xr pf.conf 5
lines will give SMTP and IMAP access to logged in users:
.Bd -literal
table <authpf_users> persist
pass in on $ext_if proto tcp from <authpf_users> \e
        to port { smtp imap }
.Ed
.Pp
It is also possible to use the "authpf_users"
.Pa table
in combination with anchors.
For example,
.Xr pf 4
processing can be sped up by looking up the anchor
only for packets coming from logged in users:
.Bd -literal
table <authpf_users> persist
anchor "authpf/*" from <authpf_users>
rdr-anchor "authpf/*" from <authpf_users>
.Ed
.Pp
.Sy Tunneled users
\- normally
.Nm
allows only one session per client IP address.
However in some cases, such as when connections are tunneled via
.Xr ssh 1
or
.Xr ipsec 4 ,
the connections can be authorized based on the userid of the user instead of
the client IP address.
In this case it is appropriate to use
.Nm authpf-noip
to allow multiple users behind a NAT gateway to connect.
In the
.Pa /etc/authpf/authpf.rules
example below, the remote user could tunnel a remote desktop session to their
workstation:
.Bd -literal
internal_if="bge0"
workstation_ip="10.2.3.4"

pass out on $internal_if from (self) to $workstation_ip port 3389 \e
       user $user_id
.Ed
.Sh FILES
.Bl -tag -width "/etc/authpf/authpf.conf" -compact
.It Pa /etc/authpf/authpf.conf
.It Pa /etc/authpf/authpf.allow
.It Pa /etc/authpf/authpf.rules
.It Pa /etc/authpf/authpf.message
.It Pa /etc/authpf/authpf.problem
.El
.Sh SEE ALSO
.Xr pf 4 ,
.Xr pf.conf 5 ,
.Xr securelevel 7 ,
.Xr ftp-proxy 8
.Sh HISTORY
The
.Nm
program first appeared in
.Ox 3.1 .
.Sh BUGS
Configuration issues are tricky.
The authenticating
.Xr ssh 1
connection may be secured, but if the network is not secured the user may
expose insecure protocols to attackers on the same network, or enable other
attackers on the network to pretend to be the user by spoofing their IP
address.
.Pp
.Nm
is not designed to prevent users from denying service to other users.