OpenBSD manual page server

Manual Page Search Parameters

TCPD(8) System Manager's Manual TCPD(8)

tcpdtcp wrappers access control facility for internet services

The tcpd program can be set up to monitor incoming requests for telnet(1), finger(1), ftp(1), rsh(1), tftp(1), talk(1), comsat(8), and other services that have a one-to-one mapping onto executable files.

Operation is as follows: whenever a request for service arrives, the inetd(8) daemon is tricked into running the tcpd program instead of the desired server. tcpd logs the request and does some additional checks. When all is well, tcpd runs the appropriate server program and goes away.

Optional features are: pattern-based access control, client username lookups with the RFC 931 etc. protocol, protection against hosts that pretend to have someone else's host name, and protection against hosts that pretend to have someone else's network address.

Connections that are monitored by tcpd are reported through the syslog(3) facility. Each record contains a time stamp, the client host name and the name of the requested service. The information can be useful to detect unwanted activities, especially when logfile information from several hosts is merged.

In order to find out where your logs are going, examine the syslog configuration file, usually /etc/syslog.conf.

Optionally, tcpd supports a simple form of access control that is based on pattern matching. The access-control software provides hooks for the execution of shell commands when a pattern fires. For details, see the hosts_access(5) manual page.

The authentication scheme of some protocols (rsh(1)) relies on host names. Some implementations believe the host name that they get from any random name server; other implementations are more careful but use a flawed algorithm.

tcpd verifies the client host name that is returned by the address->name DNS server by looking at the host name and address that are returned by the name->address DNS server. If any discrepancy is detected, tcpd concludes that it is dealing with a host that pretends to have someone elses host name.

Optionally, tcpd disables source-routing socket options on every connection that it deals with. This will take care of most attacks from hosts that pretend to have an address that belongs to someone else's network. UDP services do not benefit from this protection. This feature must be turned on at compile-time.

When RFC 931 etc. lookups are enabled (compile-time option) tcpd will attempt to establish the name of the client user. This will succeed only if the client host runs an RFC 931-compliant daemon. Client user name lookups will not work for datagram-oriented connections, and may cause noticeable delays in the case of connections from PCs.

The default locations of the host access control tables are:

Access control table (allow list)
Access control table (deny list)

This example applies when tcpd expects that the network daemons are left in their original place, as it is configured within OpenBSD.

In order to monitor access to the finger(1) service, perform the following edits on the inetd(8) configuration file, /etc/inetd.conf:

finger  stream  tcp  nowait  nobody  /usr/libexec/fingerd  fingerd


finger  stream  tcp  nowait  nobody  /usr/libexec/tcpd     fingerd

Similar changes will be needed for the other services that are to be covered by tcpd. Send a `kill -HUP´ to the inetd(8) process to make the changes effective.

In the case of daemons that do not live in a common directory ("secret" or otherwise), edit the inetd(8) configuration file so that it specifies an absolute path name for the process name field. For example:

    ntalk  dgram  udp  wait  root  /usr/libexec/tcpd  /usr/local/lib/ntalkd

Only the last component (ntalkd) of the pathname will be used for access control and logging.

hosts_access(5), inetd.conf(5), syslog.conf(5)

Wietse Venema (,
Department of Mathematics and Computing Science,
Eindhoven University of Technology
Den Dolech 2, P.O. Box 513,
5600 MB Eindhoven, The Netherlands

Some UDP (and RPC) daemons linger around for a while after they have finished their work, in case another request comes in. In the inetd configuration file these services are registered with the wait option. Only the request that started such a daemon will be logged.

RPC broadcast requests (for example: rwall(1), rup(1), rusers(1)) always appear to come from the responding host. What happens is that the client broadcasts the request to all portmap(8) daemons on its network; each portmap(8) daemon forwards the request to a local daemon. As far as the rwalld(8) etc. daemons know, the request comes from the local host.

May 31, 2007 OpenBSD-5.1