Slashdot Mirror
The Slow Bruteforce Botnet(s) May Be Learning
badger.foo writes "We've seen stories about the slow bruteforcers — we've discussed it here — and based on the data, my colleague Egil Möller was the first to suggest that since we know the attempts are coordinated, it is not too far-fetched to assume that the controlling system measures the rates of success for each of the chosen targets and allocates resources accordingly. (The probes of my systems have slowed in the last month.) If Egil's assumption is right, we are seeing the bad guys adapting. And they're avoiding OpenBSD machines." For fans of raw data, here are all the log entries (3MB) that badger.foo has collected since noticing the slow bruteforce attacks.
The obvious solution is to use public/private key authentication and disallow password logins.
This is much safer anyways, since your private key and your passphrase stays on your local machine always, so even if the server is compromised and the SSHd is bugged, no one will have immediate access to your login token.
I swear, some of the most adaptive, sophisticated, and advanced techniques seem to be coming out of the Botnets.
It's my (admittedly probably crazy) idea that we WILL begin to see "emergent intelligence properties" out of some sophisticated system at some point in time, whether it be Google, an AGI lab, or a botnet. I shudder at the prospect of our first AI of power will have grown from one of these botnets.
NOTE: I'm not saying this will happen tomorrow, but extrapolating the current state of botnets relative to the current state of other systems leads me to believe, on a relative basis, systems may be complex relative to one another as they are today. If that is the case, well... that would be bad.
If you can read this... 01110101 01110010 00100000 01100001 00100000 01100111 01100101 01100101 01101011
The conclusions are a bit too speculative, nonetheless the research is interesting. I am not sure if a few hundred hosts are enough to conclude that the "bad guys" are coordinating and sharing attack output. And as far as avoiding OpenBSD, come on..."OpenBSD is a bitch." Why is this a surprise?? :)
In principle, OpenBSD is no more or less vulnerable to weak username/password pairs than is any other OS. I suspect that, on average, OpenBSD machines are more likely to be set up for keypair auth; but any that aren't are in the same boat as everybody else(since, after all, username/password guesses aren't OS weaknesses, OSes are supposed to respond to correct username/password pairs.)
There is still reason to avoid them, though. Because OpenBSD is something of a niche system, you can make plausible inferences about the systems running it. Specifically, they most likely have admins who are interested in security and are watching activity fairly closely, and are more likely than average to do something about it. If you are doing something illegal, why attract such attention?
Bots were knocking on my door to the point I was worry about performance degradation. I know there are many ways to defeat these but here was my solution.
In hosts.deny /var/www/html/allow.txt
-----------------
sshd:ALL EXCEPT
-----------------
Create a simple cgi-script (password protected and accessed via secret random url) that writes your browser IP address to the allow.txt file and all those nasty botnets and go to hell.
These people are a tremendous illness upon the world. If it were legal, I would contribute to a bounty on the lives of the people responsible for this stuff. These people make me beyond sick. I have said it many times and sometimes I actually mean it -- if I knew of someone involved in this sort of business close by, I would appear on the news shortly thereafter. And I am pretty sure I am not alone in this sentiment.
At the risk of being unpopular ..... Just turn off the Internet already!
How would the botnet know they are attacking an OpenBSD box (vs Linux or something else)?
Is there some sort of server signature involved (that I'm not aware of)
My (Linux) ssh server at home just responds with a password prompt. I don't see any easy way to determine the underlying system from that.
BTW. On my server at home I use Hashlimits to limit each IP to 1 attempt per minute (maximum). This has taken the attacks down from hundreds / thousands per day ( The most attacks I ever got was ~7,000 from one IP) to about 3 to 6. This is typically, 1 attempt each, they then get blocked, and then they go away.
Ever stop to think
Don't forget about the economies surrounding botnets. There are two sides, those that profit from the botnets (the operators), and those that profit fighting the botnets (the fighters). Additionally, there are those that profit from providing botnet remedial "solutions" whilst not being in either of the primary (operator or fighter) categories. If botnets ceased to exist, there would be a *lot* more lost on the fighter and solution side than on the operator side. So... like SPAM, this raises the question of just who actually benefits the most from botnet existing.
Posting as AC because the people running botnets can be nasty...I had most of their hosts banned two weeks ago and it got more interesting.
To the people who say: "Use fail2ban" --it won't work unless you jail the host on the first failed login forever. They'll be back once every six hours on my system.
After I had a week worth of logs, I added them to hosts.deny--and now things are getting interesting. I'm working on compiling the pattern now--but it looks like there's "micro wordlists" being thrown at it until they get picked up in fail2ban...two or three a day from new hosts.
ipchains is Linux's 2.2 kernel firewall protection. BSD uses 'IPF'.
No matter what system you're using, a closed port is a closed port.
I think the main selling point between the two would be that IPF is slightly better performing and that iptables has quite a few addons that make for niceness if you know about and how to use them.
Bye!
OpenBSD doesn't use ipchains -- it uses pf, which many people -- myself included -- like a lot. OpenBSD is secure and easy to get routing.
The end result is the same, but pf can be easily adapted to many tricks like this, automatically blocking SSH bruteforcing.
I'd give the beginners using Ubuntu a break. They're overwhelming sometimes, but the community growing is a good thing. I'm sure someone I've introduced to Linux has needed online help (badly!), but another friend I introduced to Linux really dug in and we're now both better developers because of it. You just don't know.
I've looked at the TFA and the hard data and it seems like admins are the ones making the IT mistakes. With so many attempts for root and none of the other users personally identifiable, I can personally just set up a Bot to run tracert routines on failed attempts and report them for trying to access Root or Admin.
When it comes to multi-user sites however public key auth is standard, but your user ID and password have to match. What I don't understand is why everyone immediately resorts to AI development.
Clearly musing, he is. AI means "Self-adapting code". Self-adaption is too slow in real time and is only controlled by small control variables in games. Botnets have a heard. IT's the ADMIN's fault for being hearded, but they can have a techie d/c the power cord to save the rest of the world. Theres no real threat to secure folks because physical disconnection is trivial over a router (I just disable my IP assignment and I'm disconnected until I get another techie to do it physically) but more of a threat to people who can't control it. People controlled by the law, such as big-time Admins.
Sure, sure, the server won't crash when you're watching it, sure. But how boring will that be?
Here's the real issue: Remote Access
There has to be a way for the slow bots to get into root or admin or a remote access. I usually disable root or admin from working outside the internal loopback - 127.0.0.1 - standard Class A IP Address. I could technically configure a Bot to run Tracert (traceROOT) routines on all of those people (yes, windows user here) and have them reported to the federal government. It can't mess up my personal account, nor can it mess up DNS servers with sheer volume. It's small-scale.
so, the solution is proper remote access protocols. I remember NEVER activating remote access but at the same time using public-key authorized third party demo services to make minor changes remotely, including shutting the system down. I used logmein.com, free demo version, pathetically, but it's actually more secure as long as I have no idea how why I should do it myself. Once I used the shutdown signal it could not boot itself up unless someone would physically press the button. I have to call a physical person in the house to do that myself, so unless demons from hell can use an on/offswitch and my BIOS password without my permission, it ain't starting on it's own nor does it listen for a restart signal until I sign into windows for the first time (Windows XP here). My system has never been breached before, but it constantly deadlocks to save itself from burning the CPU out. It has a thermosensor and cutoff only in the power supply unit, however. Stupid laptops weren't designed for gaming even though thats how its advertised. How do I pull an all nighter at this rate? I'll just remove the sensor in my power supply and WHAM there goes my processor for not having heat sensors. Stupid dell power supply. Rocket fish will at least deadlock my system without damaging my hard disk.
Here. I admit. I'm part of the so-called "whitehat guys" who profit from stoping the botnets. But since I have no ethics or morals, I dont really stop them, I just give them kickbacks to make it look like I'm stopping them.
Now excuse me while I go get a back massage on from the hot ladies serving me martinis on the beach in Tahiti. Me and my fellow whitehats are making millions off you poor fools. If you only knew!
(adjust your tinfoil good sir, you are blocking the wrong signals)
I do computer and network security for a university.
This distributed SSH password guessing is not a new tactic. We have seen and tracked this tactic off and on for over a year.
If this tactic was a game changer, we would have seen it ramp up before now. It would occur all the time. But it doesn't. It only seems to occur during holidays.
At it's heart, this tactic is not any more effective than non-distributed password guessing. Either way, the attacker has to enumerate the same number of guesses before finding a hit. If a machine is vulnerable, it will be successfully attacked by either approach to password guessing. If it is not vulnerable, neither approach will work.
Modern hacking is a economic activity. It must balance risk and reward. This attack doesn't offer any more reward than conventional password guessing. It's main feature is to try to change the risk side of the equation.
Conventional SSH password guessing is noisy. One machine will portscan for TCP/22. Then it rapidly guesses passwords against everything that responds. That one machine is usually lost to the attacker. Automated defense systems block it. Also, defenders report it to the owning ISP. The only way this works for the attacker is if he can harvest more that he loses.
The distributed guessing attack is also noisy, but in a different way. Currently, we see the attacker start by sacrificing 1 computer to do a TCP/22 portscan. At this point, he has already risked as much as a conventional password guessing attack. Then he feeds the results to a bunch of bots. Each bot then takes turns guessing passwords. Each bot guesses 1 password at a time. However, each bot guesses against multiple SSH servers at the same time.
This attack is inherently more risky that conventional password guessing. The attacker exposes many of his computers. If we can detect and respond, this attack is not as cost effective as conventional password guessing.
It is easy for my university to detect and respond to these attacks. We detect it in three different ways.
1) Each attacker has a distinctive network behavior pattern. We can automate detection by looking at aggregate Cisco netflow data.
2) It is trivial to pick off this attack using a SSH honeypot.
3) We use a network visualization tool to watch aggregate SSH activity. This password guessing is obvious on our visualization tool.
Once we have detected the attackers, we respond to them in the normal way. We block them. We inform our peer institutions and the authorities. We inform the owning ISP.
The main difference in this situation is that detection and response is easy if you have access to aggregate traffic or multiple SSH servers. It is difficult if you only manage 1 SSH server.
I don't expect this form of attack to last much longer. I am sure that everybody else is adapting. Once the defenders adapt, this tactic is too expensive to be used.
Miles
ipchains is Linux's 2.2 kernel firewall protection. BSD uses 'IPF'.
OpenBSD uses PF not IPF.
FreeBSD uses PF or IPF.
Linux uses iptables. It's not been ipchains since a few major kernel versions back.
Pf rules. It's far clearer, more sensible, and more configurable than iptables.
On my OpenBSD webserver I noticed a recent spike in hacking attempts. After checking with my clients with regards to where their web traffic and sales come from I discovered that virtually none needed to have their webpages displayed offshore.
I then blocked the entire Asia Pacific Network. I am talking about the entire CIDR range from the offending ISP. I also blocked select addresses in Russia, Turkey, Germany, Poland, Brazil, etc. Every few days I check the logs and add a few more blocks if need be.
While I freely admit this move is quite drastic in nature and not possible for everyone, the illegal activity has dropped off to virtually nil. My Bandwidth utilization is way down as well.
The way I see it, I am more than willing to accept the loss of 1% legitimate traffic for 99% that isn't. If these people can't play nice, why let them play at all? I am naive enough to think that if more and more people adopted this policy, perhaps the offending governments would stand up and take notice. They seem to be able to control whether or not their citizens are able to look at pro-democracy information. If they cared about the illegal activity as well, they could do something about it. Until then, they'll remained blocked and I sleep very well at night.
Would a bayesian filter work on this? The filter would match bad userids against the set of valid ones; bad userids that do not resemble any valid id by more than X% will score a demerit against the host that submitted the bad ID. Enough bad ids will probably identify an attacking bot, which can then be blocked. This is a slow defense, but the attack itself is slow and will probably statistically require far more attempts than a bayesian filter requires to identify the attacker.
Since the attacker doesn't know the set of valid userids on the target system, it's hard to see how this could be countered. Spam authors know how normal email looks, but still can't defeat bayesian spam filters.
Pavlov wouldn't be so famous if he'd used a can opener instead of a bell.