v0.97, 24 November 2001
READ THIS IMPORTANT SECTION !!!
I hereby disclaim all responsibility for your use of this hack. If it backfires on you in any way whatsoever, that's the breaks. Not my fault. If you don't understand the risks inherent in doing this, don't do it. If you use this hack and it allows vicious vandals to break into your company's computers and costs you your job and your company millions of dollars, well that's just tough nuggies. Don't come crying to me.
Copyright © 1998-2001 by François-René Rideau.
This document is free software published under the bugroff license.
To ease their task, it has also been released to the LDP maintainers under the GNU Free Documentation License.
I have stopped actively developing this mini-HOWTO, although I'm still maintaining it. I'm looking for a maintainer to take over this document, who would extend it into a full-fledged HOWTO by expanding on the solutions whose existence I only mention, and who would maybe develop software to make it easier to pierce firewalls. I have a lot of ideas to expand this HOWTO and write according software, if anyone is interested. I also used to write a french version of this HOWTO, but no one has been maintaining it anymore for a long time.
Even though the only thing left is the disclaimers, this document owes a lot to the Term-Firewall mini-HOWTO by Barak Pearlmutter <firstname.lastname@example.org>. Barak's mini-HOWTO relies on an ancient and no-more-supported program named Term (a great program in its time, and maybe still useful in some unhappy circumstances), as well as on peculiarities of a not-so-standard telnet implementation, that is, many obsolete and non-portable facts. Nevertheless, there was a necessity for a mini-HOWTO about piercing firewalls, and despite the limitations of its hacks, this mini-HOWTO was a model and an encouragement.
The latest official LDP version of this document is on: http://www.linuxdoc.org/HOWTO/mini/Firewall-Piercing.html
The source of my latest official version of this document is on: http://fare.tunes.org/files/fwprc/
The source of my latest working draft of this document is on: http://tunes.org/cgi-bin/cvsweb/fare/fare/www/articles/Firewall-Piercing.en.sgml
This document has a moral. And the moral is: a firewall cannot protect a network against its own internal users, and should not even try to.
When an internal user asks you system administrator to open an outbound port to an external machine, or an inbound port to an internal machine, then you should do it for him. Of course you should help the user to make sure that his transactions are secure, and that his software is robust. But a flat out denial of service is plain incompetence. For unless he is so firewalled as to be completely cut from the outside world, with no ssh, no telnet, no web browsing, no email, no dns, no ping, no phone line, no radio, no nothing, then the user can and will use firewall piercing techniques to access the machines he wants nonetheless, and the net result for security will be an unaudited connection with the outside world. So either you trust your users, after proper training and selection, or you shouldn't grant them access to the network at all - but then again, the role of a network administrator is usually to serve its users, so your goal should be the former rather than the latter. You can and you shall protect them from the outside world; you can and you shall protect your critical services from them; but you can't and you shall not protect them from themselves.
Because there exists such things as system administrators who are either unresponsive, absent, overworked, plain incompetent, irresponsible, or more generally managed by incompetent people, it so happens that a user may find himself behind a firewall that he may cross, but only in awkward ways. This mini-HOWTO explains a generic and portable way to pierce tunnels into firewalls, by turning any thin, tiny trickle of bits into a full-fledged information superhighway, so the user can seamlessly use standard tools to access computers on the other side of the firewall. The very same technique can be used by competent system administrators to build virtual private networks (VPN).
Of course, if your sysadm has setup a firewall s/he might have a good reason, and you may have signed an agreement to not circumvent it. On the other hand, the fact that you can use telnet, the web, e-mail, or whatever other bidirectional information flux with the outside of the firewall (which is a prerequisite for the presented hacks to work) means that you are allowed to access external systems, and the fact that you can log into a particular external system somehow means you're allowed to do it, too.
So this is all a matter of conveniently using legal holes in a firewall, and allow generic programs to work from there with generic protocols, as opposed to requiring special or modified (and recompiled) programs going through lots of special-purpose proxies that be misconfigured by an uncaring or incompetent sysadm, or to installing lots of special-purpose converters to access each of your usual services (like e-mail) through ways supported by the firewall (like the web).
Moreover, the use of a user-level IP emulator such as SLiRP should still prevent external attackers from piercing the firewall back in the other way, unless explicitly permitted by you (or they are clever and wicked, and root or otherwise able to spy you on the server host).
All in all, the presented hack should be relatively safe. However, it all depends on the particular circumstances in which you set things up, and I can give no guarantee about this hack. Lots of things are intrinsically unsafe about any Internet connection, be it with this hack or not, so don't you assume anything is safe unless you have good reasons, and/or use some kind of encryption all the way.
Let's repeat the basics of networking security: you cannot trust anything about a connection more than you trust the hosts that can handle the unencrypted data, including hosts on both ends of the connection, and all hosts that can intercept the communication, unless the communication is properly encrypted with secret keys. If you misplace your trust, your passwords may be stolen and used against you, your credit card number may be stolen and used against you, and you may be fired from your work for endangering the whole company. Tough nuggies.
To sum it up, don't use this hack unless you know what you're doing. Re-read the disclaimer above.
It is assumed that you know what you're doing, that you know about configuring a network connection, that in case of doubt, you will have read all relevant documentation (HOWTOs, manual pages, web pages, mailing-list archives, RFCs, courses, tutorials).
It is assumed that you have shell accounts on both sides of the firewall, that you can somehow transmit packets of information both ways across the firewall (with telnet, ssh, e-mail, and the web being the ways currently known to work), and that you can let a daemon run as a background task on the server site (or benefit from and existing daemon, sshd, telnetd, or sendmail/procmail).
It is assumed that you know or are willing to learn how to configure an IP emulator (pppd, slirp) or an Internet access daemon and its associated library (SOCKS, Term) on each side, according to your needs in terms of connectivity and to your access rights, with your recompiling some software if needed.
Last but not least, so that you can use the hacks described in this document, it is assumed that you are root on the side of the firewall that needs full transparent IP access to the other side. Indeed, you'll want to run the PPP daemon on this side which allows for use the normal kernel packet routing facilities. In case you're not root on this side, your case is not desperate though: indeed, Barak Pearlmutter's Term-Firewall mini-HOWTO describes how to use Term, a purely userland program, to the end of piercing firewalls. Although there's no HOWTO, I suspect SOCKS could also be used as a way to pierce firewalls without have root privilege; I will gladly accept patches to this HOWTO that describe such a method of piercing firewalls.
Most software named in this HOWTO should be available from your standard Linux distribution, possibly among contrib's. At least, the four first below are available in as .rpm and .deb packages. In case you want to fetch the latest sources (after all, one of the ends of the connection may not be running under Linux), use the addresses below:
Understanding a problem is the first half of the path to solving it.
If you want this hack to work for you, you'll have to get an idea of how it works, so that in case anything breaks, you know where to look for.
The first step toward understanding the problem is to give a name to relevant concepts.
As usual, we'll herein call "client" the machine that decides to initiate the connection, as well as programs and files on that machine. Conversely, we'll call "server" that waits for connections and accepts them, as well as programs and files on that machine. Firewall piercing is useful when the two machines are separated by a firewall, such that it is possible for the server to accept some kind of connections, whereas the client might or might not be able to accept any. A tunnel will be created between the two machines that allows full IP traffic despite the firewall.
Usually, when piercing firewalls, the client is the machine behind a firewall: it has limited access to the internet, but can somehow open some kind of connection to the server. The server is a machine with full internet access, that will serve as a proxy for the client to access all of the internet. In a VPN, the firewall the roles might be reversed, with the client being on the internet, and the server serving as a proxy for the client to access some private network.
The main problem with firewall piercing is to create a tunnel: a continuous connection from the client machine to a server machine on the other side of the firewall, that allows for bidirectional exchange of information. Optionally, this connection should be a secure one. The secondary problem is to transform this connection into a full IP access for normal programs to use transparently.
For the main problem, we'll assume that either (1) you can establish normal TCP/IP connections from the client side of the firewall to some port on a server machine where a sshd runs or can be set to run, or (2) you can somehow establish a telnet connection through a telnet proxy. In case you cannot, we will give you pointers to other software that allows you to pierce a tunnel accross a firewall. Although we only give a secure solution in the first case, you can hack your own secure solution in the other cases, if you understand the principle (if you don't, someone, e.g. I, can do it for you in exchange for money).
For the secondary problem, IP emulators (pppd or SLiRP) are run on each side of the tunnel.
On the side that wants full IP access to the other side, you'll want to run pppd. On the other side, you want to run pppd if you also want full IP access to the first side, or SLiRP if you want to prevent any access. Go to your usual pppd or SLiRP documentation for more information, if you have specific needs not covered by the examples given below.
Although this is conceptually trivial, it nonetheless requires a few silly tricks, so as to work, since (a) in case you're using some kind of programmed interactive shell session to start the server's IP emulator on either side, you need to correctly synchronize the start of the IP emulator on the other side, so as not to send garbage into the shell session, and (b) IP emulators are designed to be run on a "tty" interface so you have to convert your tunnel's interface into a tty one.
Issue (a) is just your usual synchronization problem, and doesn't even exist if you use ssh, that transparently handles server's command launching.
Issue (b) requires the use of a simple external utility. We wrote one, cotty just for that purpose.
Among the silly problems caused by pppd maintainers' shortmindedness (no more true in recent Linux versions), you can only run it through either a device in /dev or the current tty. You cannot run it through a pair of pipe (which would be the obvious design). This is fine for the server's pppd if any, as it can use the telnet or ssh session's tty; but for the client's pppd, this conflicts with the possible use of telnet as a way to establish a connection.
Indeed, telnet, too wants to be on a tty; it behaves almost correctly with a pair of pipe, except that it will still insist on doing ioctl's to the current tty, with which it will interfere; using telnet without a tty also causes race conditions, so that the whole connection will fail on "slow" computers (fwprc 0.1 worked perfectly on a P/MMX 233, one time out of 6 on a 6x86-P200+, and never on a 486dx2/66). All in all, when using telnet, you need cotty to run as a daemon to copy output from one tty on which runs pppd into another tty on which runs telnet, and conversely.
If I find the sucker (probably a MULTICS guy, though there must have been UNIX people stupid enough to copy the idea) who invented the principle of "tty" devices by which you read and write from a "same" pseudo-file, instead of having clean pairs of pipes, I strangle him!
Let's assume that your firewall administrator allows transparent TCP connections to some port on some server machine on the other side of the firewall (be it the standard SSH port 22, or an alternate destination port, like the HTTP port 80 or whatever), or that you somehow managed to get some port in one side of the firewall to get redirected to a port on the other side (using httptunnel, mailtunnel, some tunnel over telnet, or whatelse).
Then, you can run an sshd on the server side port, and connect to it with an ssh on the client side port. On both sides of the ssh connection, you run IP emulators ( pppd), and there you have your VPN, Virtual Public Network, that circumvents the stupid firewall limitations, with the added bonus of being encrypted for privacy (beware: the firewall administrator still knows the other end of the tunnel, and whatever authentication information you might have sent before to run ssh).
The exact same technology can be used to build a VPN, Virtual Private Network, whereby you securely join physical sites into a one logical network without sacrificing security with respect to the transport network between the sites.
Below is a sample script for you to adapt to your needs. It uses the array feature of zsh, but you may easily adapt it to your favorite shell. Use option -p for ssh to try another port than port 22 (but then, be sure to run sshd on same port).
Note that the script supposes that ssh can login without your having to interactively type your password (indeed, it's controlling tty will be connected to pppd, so if it asks for a password, you lose). This can be done either by ssh keys in your ˜/.ssh/authorized_keys that either do not require a password, or that you unlock using ssh-agent or ssh-askpass. See your SSH documentation. Actually, you might also use a chat script to enter your password, but this is definitely not the Right Thing.
If you are not root on the server end, or simply if want to screen your client's network from outbound connections, you can use slirp instead of pppd as the server's PPP emulator. Just uncomment the relevant line.
Note that default options from your /etc/ppp/options or ˜/.slirprc may break this script, so remove any unwanted option from there.
Also note that 10.0.2.2 is the default setting for slirp, which might or not fit your specific setup. In any case, you should most likely be using some address in one of the ranges reserved by RFC 1918 for private networks: 10.0.0.0/8, 172.16.0.0/12 or 192.168.0.0/16. The firewall-protected LAN might already be using some of them, and avoiding clashes is your responsibility. For more customization, please read the appropriate documentation.
If your client's pppd is old or non-linux (e.g. BSD) and hasn't got the pty option, use
Automatic reconnection is left as an exercise to the reader (hint: the nodetach option from pppd might help for that).
If all you can do is telnet (because of a telnet proxy), then this solution might be fit for you.
The firewall-piercing program, fwprc, will use a "tty proxy", cotty, that opens two pseudo-tty devices, launches some command on each of those devices' slaves, and stubbornly copies every character that one outputs to the tty that serves as input of the other command. One command will be telnet connection to server site, and the other will be the client's pppd. pppd can then open and control the telnet session with a chat script as usual.
Actually, if your telnet proxy allows connection to an arbitrary port, and if you can reliably run a daemon on the server host (with a cron job to relaunch it in case of breakage), then you'd better write some program that will just connect a client side port to the server side port through the proxy, so you can use the above secure solution, possibly using some variant of
Note: if you must use the unsecure telnet-based solution, be sure that nothing lies in your target account that you want to keep secret or untampered, since the password will be sent in clear text accross the Internet. If you can control these things, a one-time-password system, or an explicit cryptographic challenge system will enhance your security, although it will make automated connection scripts much more complex.
I wrote a very well self-documented script to pierce firewalls, fwprc, available from my site, together with cotty (which is required by fwprc 0.2 and later). At the time of my writing these lines, latest versions are fwprc 0.3e and cotty 0.4c.
The name "fwprc" is voluntarily made unreadable and unpronounceable, so that it will confuse the incompetent paranoid sysadm who might be the cause of the firewall that annoys you (of course, there can be legitimate firewalls, too, and even indispensable ones; security is all a matter of correct configuration). If you must read it aloud, choose the worst way you can imagine.
CONTEST! CONTEST! Send me an audio file with a digital audio recording of how you pronounce "fwprc". The worst entry will win a free upgrade and his name on the fwprc 1.0 page!
I tested the program in several settings, by configuring it through resource files. But of course, by Murphy's law, it will break for you. Feel free to contribute enhancements that will make life easier to other people who'll configure it after you.
fwprc can be customized through a file .fwprcrc meant to be the same on both sides of the firewall. Having several alternate configurations to choose from is sure possible (for instance, I do it), and is left as an exercise to the reader.
To begin with, copy the appropriate section of fwprc (the previous to last) into a file named .fwprcrc in your home directory. Then replace variable values with stuff that fits your configuration. Finally, copy to the other host, and test.
Default behavior is to use pppd on the client, and slirp on the server. To modify that, you can redefine the appropriate function in your .fwprcrc with such a line as:
Note that SLiRP is safer than pppd, and easier to have access to, since it does not require being root on the server machine, and needn't additional firewall configuration to prevent connections from the outside world into the firewalled network. The basic functionality in SLiRP works quite well, but I haven't managed to get some advertised pluses to work (like run-time controllability). Of course, since it is free software, feel free to hack the source so as to actually implement or fix whichever feature you need.
Piercing the firewall is not everything. You must also route the packets from the client side of the firewall to the server side. This section tackles the basic settings specific about routing accross a tunnel. For more detailed explanations of routing, see the relevant HOWTOs and man pages about networking, routing and masquerading.
The catch is that although your network administration would tell you to setup your some router on your client side's as the default route, (this may be relevant if you want to have a specific route to the networks on the client of the firewall), you should setup PPP link as the route to the networks on the server side.
In other words, your default route should point to a router on whichever side of the tunnel that gives you access to the Internet.
Most importantly, packets sent to the server host as part of running the tunnel should be routed through your usual network (e.g. your default ethernet router); otherwise, your kernel will have problems, as it tries to route through the inside the tunnel the very packets that ought to constitute the outside of the tunnel.
Thus, you'll have to setup correct routes in your network startup configuration. The precise location of your routing configuration data depends on your distribution, but it is typically under /etc/init.d/network or /etc/network/; similarly, your PPP configuration is typically in /etc/ppp/, and the proper place to configure its routes is usually in ip-up or ip-up.d/. (Tip: to identify your distribution-specific file locations, you must read the documentation of your distribution and otherwise RTFM; alternatively use grep recursively on your /etc; at worst, trace what happens at boot time, as configured in your /etc/inittab.)
When piercing a tunnel from a roaming laptop on the Internet into a protected network, the script getroute.pl (available from the fwprc distribution) gives the current route to the server host that is the other end of the tunnel.
Once you can route packets to the server side of the tunnel, you might want to setup your machine as a router for all your pals on the client side of the firewall, achieving a full-fledged shared VPN. This is not specific to Firewall-Piercing, so just you read the relevant HOWTOs about networking, routing and masquerading. Also, for security reasons, be sure to also setup a proper firewall on your machine, especially if you're going to be a router for other people.
Finally, be reminded that if you're using pppd on the server end of the tunnel (as opposed to user-mode slirp), you will have to configure proper routes and firewall rules on the server side of the tunnel, too.
In this example, your client machine is connected to a firewalled LAN through ethernet device eth0. Its IP address is 126.96.36.199; its network is 188.8.131.52/24; its router is 184.108.40.206.
Your network administrator may have told you to use 220.127.116.11 as default router, but you shouldn't. You should only use it as a route to the client side of the firewall.
Let's suppose the client side of your firewall is made of networks 18.104.22.168/16 and 22.214.171.124/16, and of host 126.96.36.199. To make them accessible through your client router, add these routes to your global network startup script:
Sometimes, only one side of the firewall can launch telnet sessions into the other side; however, some means of communication is possible (typically, through e-mail). Piercing the firewall is still possible, by triggering with whatever messaging capability is available a telnet connection from the ``right'' side of the firewall to the other.
fwprc includes code to trigger such connections from an OpenPGP-authentified email message; all you need is add fwprc as a procmail filter to messages using the protocol, (instructions included in fwprc itself). Note however, that if you are to launch pppd with appropriate privileges, you might need create your own suid wrapper to become root. Instructions enclosed in fwprc.
Also, authentified trigger does not remotely mean secure connection. You should really use ssh (perhaps over telnet) for secure connections. And then, beware of what happens between the triggering of a telnet connection, and ssh taking over that connection. Contribution in that direction welcome.
If you are firewalled, your mail may as well be in a central mailserver that doesn't do procmail filtering or allow telnet sessions. No problem! You can run fetchmail in daemon mode (or within a cron job) to poll your mailserver and deliver mail to your linux system which itself will have been configured to use procmail at delivery. Note that if you run fetchmail as a background daemon, it will lock away any other fetchmail that you'd like to run only at other times, like when you open a fwprc; of course, if you can also run a fetchmail daemon as a fake user. Too frequent a poll won't be nice to either the mailserver or your host. Too infrequent a poll means you'll have to wait before the message gets read and the reverse connection gets established. I use two-minute poll frequency.
Another way to poll for messages, when you don't have a mailbox, but do have outbound FTP access, is to use FTP tunnel.
A tool to maintain a permanent connection between a firewalled host and an external proxy, so as to export services from the host to the world, is firewall tunnel.
I have no idea how to pierce firewalls with lesser operating systems, but you can take one of these old disused computers (about anything with 8MB of RAM and an ethernet card should do), install Linux or BSD as on it, and pierce the firewall with it, while serving as a router for other machines running lesser OSes. See appropriate HOWTOs about routing, IP forwarding, NAT, etc.
I don't know the details, but a promising tool to pierce firewalls is Chris Mason's Bouncer, which acts as a SOCKS-proxy-over-SSL.
There are other kinds of firewalls than those that allow for direct ssh or telnet connections. As long as a continuous flow of packets may transmit information through a firewall in both directions, it is possible to pierce it; only the price of writing the piercer may be higher or lower.
In a very easy case, we saw that you can just launch ssh over a pty master and do some pppd in the slave tty. You may even want to do it without an adverse firewall, just so as to build a secure ``VPN'' (Virtual Private Network). The VPN mini-HOWTO gives all the details you need about this. We invite you, as an exercise, to modify fwprc so as to use this technique, or perhaps even so as to use it inside a previous non-secure fwprc session.
Another promising program for piercing through HTTP is Lars Brinkoff's httptunnel, a http server and client combination that achieves a TCP/IP tunnel connection through the proxy-friendly HTTP protocol. You should then be able to run fwprc (preferably over ssh) over that connection, although I haven't tried it yet. Could anyone test and report? Note that httptunnel is still under development, so you may help implement the features it currently lacks, like, having multiple connections, and/or serving fake pages so as to mislead suspicious adverse firewall administrators.
Whatever goes through your firewall, be it telnet, HTTP or other TCP/IP connections, or something real weird like DNS queries, ICMP packets, e-mail (see mailtunnel, icmptunnel), or whatelse, you can always write a tunnel client/server combination, and run a ssh and/or PPP connection through it. The performance mightn't be high, depending on the effective information communication rate after paying the overhead for coding around filters and proxies; but such a tunnel is still interesting as long as it's good enough to use fetchmail, suck, and other non-interactive programs.
If you need cross a 7-bit line, you'll want to use SLIP instead of PPP. I never tried, because lines are more or less 8-bit clean these days, but it shouldn't be difficult. If necessary, fall back to using the Term-Firewall mini-HOWTO.
If you have an 8-bit clean connection and you're root on linux both sides of the firewall, you might want to use ethertap for better performance, encapsulating raw ethernet communications on top of your connection. David Madore has written ethertap-over-TCP and ethertap-over-UDP tunneling ftp://quatramaran.ens.fr/pub/madore/misc/. There remains to write some ethertap-over-tty to combine with fwprc-like tools.
If you really need more performance than you can get while paying for a user-space sequential communication tunnel through which to run PPP, then you're in the very hard case where you might have to re-hack a weird IP stack, using (for instance) the Fox project's packet-protocol functors. You'll then achieve some direct IP-over-HTTP, IP-over-DNS, IP-over-ICMP, or such, which requires not only an elaborate protocol, but also an interface to an OS kernel, both of which are costly to implement.
Finally, if you're not fighting against an adverse firewall, but just building your own VPN, there is a large offer of VPN tools, and although the tricks I present are simple, work well, and might be enough for your needs, it could be a good idea to look at this evolving offer (that I do not know much about) for a solution that fits your requirements of performance and maintainability.
I felt it was necessary to write it, but I don't have that much time for that, so this mini-HOWTO is very rough. Thus will it stay, until I get enough feedback so as to know what sections to enhance, or better, until someone comes and takes over maintenance for the mini-HOWTO. Feedback welcome. Help welcome. mini-HOWTO maintenance take-over welcome.
In any case, the above sections have shown many problems whose solution is just a matter of someone (you?) spending some time (or money, by hiring someone else) to sit down and write it: nothing conceptually complicated, though the details might be burdensome or tricky.
Do not hesitate to contribute more problems, and hopefully more solutions, to this mini-HOWTO.
The LDP publishes many documents related to this mini-HOWTO. most notably the Linux Security Knowledge Base, the VPN HOWTO and the VPN mini-HOWTO. For more general questions about networking, routing and firewalling, start from the Networking Overview HOWTO. See also the Linux Firewall and Security site.
Then again, when facing a problem with some program, one reflex for any Linux user should be to RTFM: Read The Fscking Manual pages for the considered programs.
I've come to the conclusion that much like the need for Design Patterns came directly from the fact that people were using inferior languages like C++ or Java that don't allow to directly express higher-level programming constructs (whereas good languages such as LISP allow to express them), the need HOWTOs comes directly from the fact that Linux and UNIX systems are inferior operating systems that do not allow to directly express those simple tasks that people attempt to do with them.
If you think that all this mucking around with stupid scripts and silly HOWTOs is overly complicated and that a decent computer system ought to automate it all for you, then welcome with me among UNIX haters and other people who hate current low-level operating systems, and yearn for declarative computing systems that take care of the silly details and let us focus on things that matter. (Maybe have a peek at my own TUNES project).
"I hereby disclaim all responsibility for your use of this hack. If it backfires on you in any way whatsoever, that's the breaks. Not my fault. If you don't understand the risks inherent in doing this, don't do it. If you use this hack and it allows vicious vandals to break into your company's computers and costs you your job and your company millions of dollars, well that's just tough nuggies. Don't come crying to me."