Copyright © 1996 by Barak Pearlmutter
Copyright © 2001 by David C. Merrill
|Revision 1.1||2001-07-14||Revised by: dcm|
|Cleaned up a bit, reorganized a bit, converted to DocBook SGML and relicensed under GFDL.|
|Revision 1.0||1996-07-15||Revised by: pb|
This document explains how to use the term program to pierce a firewall from the inside, even without root privileges.
Term is an old program that almost no one uses anymore, because the 7-bit serial lines it is meant to cross are nowhere to be found anymore, and full IP ppp access is dirt cheap.
Archived Document Notice: This document has been archived by the LDP because it does not apply to modern Linux systems. It is no longer being actively maintained.
While every precaution has been taken in the preparation of this document, the Linux Documentation Project and the author(s) assume no responsibility for errors or omissions, or for damages resulting from the use of the information contained herein.
The term program is usually used to provide host-to-host services over a modem or serial line. However, sometimes it is useful to establish a term connection between two machines that communicate via telnet. The most interesting example is connecting two hosts which are separated by ethernet firewalls or SOCKS servers. Such firewalls provide facilities for establishing a telnet connection through the firewall, typically by using the SOCKS protocol, to allow inside machines to get connections out, and requiring outside users to telnet first to a gateway machine which requires a one-time password. These firewalls make it impossible to, for instance, have X clients on an inside machine communicate with an X server on an outside machine. But, by setting up a term connection, these restrictions can all be bypassed quite conveniently, at the user level.
Setting up a term connection over a telnet substrate is a two-phase process. First your usual telnet client is used to set up a telnet connection and log in. Next, the telnet client is paused and control of the established telnet connection is given to term.
First, from a machine inside the firewall, telnet to a target machine outside the firewall and log in.
Unless you are under linux and will be using the proc filesystem (see below) make sure your shell is an sh style shell. Ie if your default shell is a csh variant, invoke telnet by:
setenv SHELL /bin/sh; telnet machine.outside
After logging in, on the remote (outside) machine invoke the command:
term -r -n off telnet
Now break back to the telnet prompt on the local (inside) machine, using ^] or whatever, and use the telnet shell escape command ! to invoke term:
telnet> ! term -n on telnet >&3 <&3
If you have a variant telnet, you might have to use some other file descriptor than 3; easy to check using strace. But three seems to work on all bsd descendent telnet clients I've tried, under both SunOS 4.x and the usual linux distributions.
Some telnet clients do not have the ! shell escape command. Eg the telnet client distributed with Slackware 3.0 is one such client. The sources that the Slackware telnet client is supposedly built from
A simple solution is therefore to obtain these sources and recompile them. This unfortunately is a task I have had no luck with. Plus, if you are running from inside a SOCKS firewall, you will need a SOCKSified telnet client anyway. To that end, I was able to compile a SOCKSified telnet client from:
or, if you're outside the USA,
Alternatively, under linux kernels up to 1.2.13, you can pause the telnet with ^]^z, figure out its pid, and invoke:
term -n on -v /proc/&,t;telnetpid>/fd/3 telnet
This doesn't work with kernels after 1.3.x, which closed some mysterious security hole by preventing access to these fd's by processes other than the owner process and its children.
It is a good idea to give the term socket an explicit name. This is the telnet; argument in the invocations of term above. Unless you have the TERMSERVER environment variable set to telnet as appropriate, you invoke term clients with the -t switch, e.g., trsh -t telnet.
I have checked line clarity using linecheck over this medium. I expected it to be completely transparent, but it is not. However, the only bad character seems to be 255. The ˜/.term/termrc.telnet I use (the .telnet is the name of the term connection, see above) contains:
baudrate off escape 255 ignore 255 timeout 600
Perhaps it could be improved by diddling, I am getting a throughput of only about 30k cps over a long-haul connection through a slow firewall. Ftp can move about 100k cps over the same route. A realistic baudrate might avoid some timeouts.
Obviously, if you are starting from outside the firewall and zitching in using a SecureID card or something, you will want to reverse the roles of the remote vs local servers given above. (If you don't understand what this means, perhaps you are not familiar enough with term to use the trick described in this file responsibly.)
This is not much more of a vulnerability than the current possibility of having a telnet connection hijacked on an unsecured outside machine. The primary additional risk comes from people being able to use the term socket you set up without you even being aware of it. So be careful out there. (Personally, I do this with an outside machine I know to be pretty secure, namely a linux laptop I maintain myself that does not accept any incoming connections.)
Another possibility is to add
to the remote ˜/.term/termrc.telnet file, or
add "-u off"
to the invocation of term. This prevents the socket from being hijacked from the remote end, with only a minor loss of functionality.
Be sure the remote telnetd is not in some nasty seven-bit mode. Or if it is, you have to tell term about it when you invoke term, by adding the -a switch at both ends. (I sometimes use >^] telnet> set outbin or set bin, or invoke telnet with a -8 switch to put the connection into eight-bit mode.)
The linecheck program has some problems checking telnet connections sometimes. This is sometimes because it doesn't check the return code of the read() call it makes. For network connections, this call to read() can return -1 with an EINTR (interrupted) or EAGAIN (try again) error code. Obviously this should be checked for.
There are a number of features that could ease the use of term over telnet. These primarily relate to an assumption that influenced the design of term, namely that the connection is low bandwidth, low latency, and somewhat noisy.
A telnet connection is in general high bandwidth, high latency, and error free. This means that the connection could be better utilized if (a) the maximum window size was raised, well above the limit imposed by term's N_PACKETS/2=16, (b) there was an option to turn off sending and checking packet checksums, and (c) larger packets were permitted when appropriate.
Also, to enhance security, it would be nice to have a term option to log all connections through the socket it monitors to a log file, or to stderr, or both. This would allow one to see if one's term connection is being subverted by nasty hackers on the outside insecure machine.
Some telnet clients and servers agree to encrypt their communications, to prevent eavesdropping on the connection. Unfortunately, the hack used above (using the network connection that the telnet client has set up while the telnet client is idle) won't work in that case. Instead, one really must go through the telnet client itself, so it can do its encryption. It seems like that requires a simple hack to the telnet client itself, to add a command that runs a process with its stdin and stdout are connected to the live telnet connection. This would also be useful for various bots, so perhaps someone has already hacked it up.
A vaguely related trick is to SOCKSify one's Term library. Details, including patches to SOCKS, are available from Steven Danz email@example.com.