(Referenced at http://www.viacorp.com/crypto.html#hack and mirrored here with permission.)
by Liraz Siri liraz AT stop-SPAM.liraz.org === Introduction Today, when too many people think of security on the Internet, they think of individual hosts and networks. Alone. Got a problem? Damn! Must be those damn hacker punks. Again. Keep it to yourself. Call the Feds, call the New York Times. Make sure we don't get it. Didn't keep your systems patched? Moron. Don't make us sue you. With the growing irrelevance of security organizations like CERT and law enforcement on the Internet, an ever growing number of attacks are handled in isolation. Hundreds of millions of Internet users around the world have become accustomed to an Internet beyond boundaries. One site flows to the next, a jungle of software, protocols, media and people connecting, signal, noise, mixing, evolving, together. It seems silly to ignore the security of the system _as a whole_, but we still do. A helpful analogy might be to consider the Internet more a living organism then a neighborhood. A security compromise is can behave more like a disease then a "breakin". It is often contagious, and can spread. Remotely exploitable security vulnerabilities are like the natural wounds of the skin. They are relatively rare, sometimes difficult to squirm through, but once inside, infection can begin. This article describes the efforts of a small, independent, security research group to audit some 36 million hosts connected to the Internet, for commonly known security vulnerabilities in an unfocused low-res scan. === Why? Because we're a curious bunch, because we've been speculating (rather academicly) over the results for several years, and of course, because we can. === Why are we publishing now, Why haven't we published before? We know other groups, working for everyone from the UKUSA SIGINT agencies, foreign intelligence, private corporations and organized crime are not likely, for many obvious reasons, to disclose any "privileged information" to the general public. We feel this is not A Good Thing, and would like to do what we can to help level the playing field. We don't have any money, resources or academic prestige to back us up, but we do have a few, humble insights to share, and we hope these can speak for themselves. Besides, wouldn't it be a shame to keep all of our busy work to ourselves, when it could be reaching a much wider audience, spark debate, and maybe even making a difference? Up until now, a couple of issues have held us back. First of all, the timeless responsibility factor. We could not avoid the possibility (certainty?) that our work would be abused by malicious parties and we've all seen before how easy it is for people to point the finger. Secondly, we've been busy and publishing involves a significant investment in time writing articles, cleaning code, reaching the potential audience and reading (sometimes answering) endless e-mails. === Walk forth in dread So you want to scan the millions of computers on the Internet from Japan to Egypt to Florida? Reach out and audit the networks of Internet Service Providers, corporations, universities, government facilities, banks and sensitive military installations? First, take another moment to think about it. Many people get nervous on the receiving end of an uninvited security audit, and you'll eventually step on quite a few toes. In some countries, you can even expect unpleasant house-calls from local law enforcement which will brand you a criminal for your unusual efforts. Citizens of a large democracy with many three letter agencies should be aware that a fully-equipped SWAT team is likely to tag along. While this may deter, possibly comfort law-abiding readers, a criminally inclined party is not without it's options. Resources are abundant on the Internet, and many suitable, unsuspecting, high-bandwidth volunteers are not hard to find, with the modest help of your favorite bulk auditing software. Not intimidated? That's the spirit! === Quick & Dirty Overview Let's take a look at some of the basic ingredients we're going to need: 1) Some wheels. (BASS, a Bulk Auditing Security Scanner) 2) A map. (address search space) 3) Fuel. (resources) Although they are not required, logistical management skills, competence and patience can also come in real handy. === Wheels The Internet is getting rather big these days, and exploring it's tens of millions of unique hosts is by no means an easy task. Manually, we could never get the job done. Fortunately, we can let a computer (or several) do most of the dirty work, allowing us to concentrate on coordination and management. Assuming of course, we have the right software. In this case, we're going to need a robust bulk security scanner that can monotonicly run for weeks, even months at a time, efficiently processing millions of addresses, generating gigabytes of traffic and surviving everything from broken routing, to system shutdowns and unfriendly sysadmins. Since we've never liked re-inventing the wheel, the first thing we did, (circa Sep 1998) was take a look at existing scanning software. We were disappointed. There was no shortage of software, from Satan, to Nessus, with a jungle of (often silly) cracker tools in between, but none of them would do. Nessus was impressive, but clearly not designed with bulk in mind. Most of the rest were unreliable, poorly written, slow and inextensible. Primitive, specialized scanners (foobar-scan) were also common, and equally useless. So, it looked like we'd need to write "Yet Another Security Scanner" ourselves. During development, we were careful not to complicate the design and code any more then we had to, aware of the many virtues of simplicity (especially in security software). Our goal was producing a scanner which was reliable, efficient and extensible. After a several weeks of on-off programming, the first alpha version of BASS, the Bulk Auditing Security Scanner was ready for it's first test run. Israel was the first target in a series of trials. At this point (Sep-Oct 98) BASS could only identify 4 common security vulnerabilities, but adding more later was a simple matter. What we really needed to evaluate was how well the multi-threaded scanning architecture worked. "beware the bugs that bite beta programs" It didn't. Even with a small target like Israel, the scan came to a final halt after about 18,000 addresses. It seemed threads would occasionally freeze, waiting for service from a host they knew was online, but behind a misconfigured firewall, or a broken router. The frozen threads were rare but persistent. They would build up in BASS's scheduler over time, eventually choking the scanner to a grinding halt. A fail-safe timeout circuit fixed the problem, and we tried again. This time, the scan finished on schedule. 110,000 addresses in under 4 hours, on a dual ISDN 128k connection. We selected the United Kingdom, with an address space of 1.4 million, for our next trial. If there were any further bugs, they were going to show, and they did. Around a million UK addresses later, BASS broke down and was dragging the entire system down with it. This time, several obscure memory leaks had slowly inflated BASS to monstrous proportions, consuming all available system memory. Several further painful debugging sessions were needed to bring the scanner up to par, during which 5 million addresses around the world had been scanned. Now that the architecture was stable, we proceeded to familiarizing BASS with the wonders of CGI and RPC, allowing the scanner to test for up to 18 widely known security vulnerabilities (see detailed listing in suffix item 1). The tests were designed to reduce false positives and false negatives to a minimum, combining passive (server's version header) and interactive (server's response to ill-formed input: a buffer-overflow, sneaky characters) implementation signatures to determine vulnerability. So now we could sit back, feed BASS a really big map of the Internet, and wait a few months (or weeks, depending on our resources) for results. === A map - A map you say? Yeah, well what I really mean is a really long list of "all" the computers connected to the Internet. Please note the term "all" is used loosely ("most" or the "majority" would probably be more accurate). - How many of them are there anyway? Reader, that's a tougher question then you might think. An Internet Protocol address, or IP for short, is a 32 bit integer. This means are there 2^32 (4.3 billion) possible unique IPs, the IP address space. In practice, only a very small fraction of this space is really used. Due to the anarchic nature of the Internet, nobody has any exact figures on usage statistics, but most estimates (circa Jan 1999) settle around 100 million users worldwide. The number of computers online is more around an order of a magnitude lower (15 million). This is because most users still access the Internet dynamicly, by dialup, over phone lines. ISPs (Internet Service Providers) can often manage to provide service with an address pool 4 to 10 times smaller then their customer base. Ideally, since BASS is (currently) Unix oriented, we would like to eliminate any non-unix computers (not that non-unix's are any more secure, quite the contrary) from our Really Big List. We would also want to skip any dynamic IP pools. In a perfect world, this would be a good idea. In ours, eliminating poor scanning candidates in advance would actually take longer then the scan itself. Optimizing a scan this way is only useful if you plan on repeating it frequently. - I'm confused, how many IPs are we going to end up scanning? That depends,.. In our case, we ended up scanning around 36 million IPs, which we estimates covered 85 percent of the active address space at the time. Keep in mind, however, that the Internet is growing very quickly, so these numbers will get bigger by the time you try this out yourself. Search for "Internet Surveys" on the web, and get an updated figure. - Wait, what's with the 85 percent? Calm down, mapping the entire used IP space is nearly impossible, even assuming you can agree with anyone else (try Usenet folks first!) on what "used" should mean. The main problem is using an IP is an internal decision organizations with an allocated slice of the address space makes for themselves. All those slices add up to 300 million IP addresses, of which only 5 percent have a computer at the other end, so we need to narrow down our search space. This is where the Domain Name System (DNS) comes to the rescue. The DNS is a tree structured lookup directory used (primarily) to map a hostname to an IP and vice-versa (www.nsa.gov <=> 126.96.36.199). By convention, most of the Internet's active addresses are registered with the DNS, although this is a not a mandatory requirement. - So we can just download the DNS's records from the Internet? Yes, and no. The DNS protocol has an "AXFR zone transfer" mechanism designed to allow one DNS server to mirror the contents of another, by requesting an AXFR zone transfer, you can download a server's records. This is helpful in providing for redundant backups, should the primary server fail. Unfortunately, since the DNS is a distributed system, we can't just download it's complete contents from any central authority. To make matters worse, many DNS servers nowadays (40 percent) refuse zone transfer requests, due to several (misunderstood) concerns over it's security implications. - Sounds rough. Well if you're going the do-it-yourself way, it's not going to be easy, but isn't as difficult as it sounds. Let's take a look at some of our options (If you aren't the do-it-yourself type, skip to item 4): 1) A top - down recursive download of the DNS. Using the DNS protocol's AXFR zone-transfer mechanism it is possible to recursively download the DNS's contents one zone at a time. In practice however, this method is usually reserved for mapping a known target that has not explicitly restricted zone-transfers. Trying to map the DNS this way has the disadvantage of being slow, unreliable and incomplete. A description of process is available in RFC1296. 2) Exploiting in-addr.arpa. We start off by recursively downloading the DNS's relatively small in-addr.arpa. domain. This will give us the allocated address space (300 million IPs). Most of the active addresses (the ones we want) in this space will have a PTR record somewhere in the in-addr.arpa domain. (so they can be mapped in reverse from IP numbers to hostnames). Many Internet protocols and applications rely on this pointer, by convention, so it is not likely to be absent on purpose. Unless the address isn't being used, of course, but we don't want any of those anyway. By checking to see which IPs in the allocated address space have a pointer in the in-addra.arpa. domain, we can narrow down the search space to about 13 percent (45 million IPs). This process demonstrates that the ever popular practice of blocking zone-transfers will not hide a network's topology. People relying on this method to obscure their security problems are begging for trouble. BTW, 'Network Wizards' are doing their Internet Survey this way, since the beginning of 1998, check them out. (http://www.nw.com/) The job is likely to take between a week, and a month (or several), depending on how much available bandwidth you have, and the quality of the software your using to get it done. 3) Scavenging Network Information Centers for pre-compiled lists. It turns out some NICs have precompiled data files available over anonymous FTP. Getting the data this way is much easier, faster and more reliable then slowly milking the DNS through the traditional AXFR zone-transfer protocol. As of Nov 1998, RIPE (ftp.ripe.net) was offering raw output files from it's recursive hostcount (Covering Europe, Russia and others. 98 countries in total) for download at ftp://ftp.ripe.net/ripe/hostcount. Update: On the 01/02/1999 they restricted anonymous FTP access to the raw hostcount output files. You now have to either convince RIPE you really need them at firstname.lastname@example.org (for saving the world, no less) or grab them at one of RIPE's many mirrors. Network Wizards, the guys doing the Internet Survey, offer (some) of the raw data from their older surveys, up to 1997, at "http://www.isc.org/ISC_HTML/domainsurvey/archive-data/". ARIN (http://www.arin.net), the American Registry for Internet Numbers, is an interesting site to look into. While your reading exciting new number policies, grab ftp.arin.net/domain/inaddr.zone over anonymous FTP. (doing a zone-transfer take's so much longer) There are hundreds of NICs, structured hierarchicly. Search the web for "Network Information Centers", and you'll find quite a few. APNIC (Asian Pacific) and JPNIC (Japan's NIC at NIC.ad.jp) are two you should really look into. Then there's InterNIC, run by Network Solutions (NSI, the "dot com" guys), in charge of the root servers, ([A-M].ROOT-SERVERS.NET), at the root of Internet's DNS, all the three letter top level domains (com, net, org, edu, gov and mil) and the top level in-addr.arpa. domain (for reverse lookups). InterNIC is the closest thing the Internet has to a central authority on anything, and is currently being run as a lucrative for-profit US-government sanctioned monopoly. InterNIC no longer provides anonymous FTP access to most of it's DNS records, with the exception of the top-level in-addr.arpa. domain, stating it is trying to prevent spammers and squatters (domain name speculators) from abusing the DNS. As such, InterNIC will only offer FTP access to "organizations that can demonstrate a technical need for the information". Fortunately, the information is already out there, available on several anonymous FTP sites hosted by InterNIC affiliates (government, military, educational,. etc) who share it's records, but do not enforce it's censorship policies. Personally, we downloaded the top level .com, .net, .org, .edu, .mil and .gov domains from ftp.nic.mil (the first NIC we tried) several minutes after a disappointing encounter with an almost empty 'domains' directory at ftp.internic.net. (Update: ftp.nic.mil no longer provides these records over anon FTP) 4) The Greener Path The Internet Software Consortium (http://www.isc.org), of the bi-annual "Internet Survey", is offering it's raw data sets for resale through MIDS, Matrix Information and Directory Services (http://www.mids.org) at $2500. Frankly, shelling the green is alot easier, faster and even less expensive then trying to compile the data yourself, especially if you don't already have the software, expertise and bandwidth to pull it off. - What about you guys? What did you do? We like banging our heads against the wall, so we went down the slippery do-it-yourself path. We started off by learning as much as we could about the DNS, reading any RFCs that were relevant to the protocol, browsed through the documentation of it's most popular implementation "BIND", downloaded a zoo of freely available DNS utilities from the major FTP sites and read lots of source code. Eventually we ended up hacking a couple of popular DNS utilities, wrote way too many ugly shell scripts, C application wrappers, and some pretty silly Perl filters, mixing alot of method 3 (scavenging), 2 (in-addr.arpa.) and just a bit of 1 (vanilla zone-transfers). If you have any good sense, you'll do otherwise. === Fuel Swarming the Internet with probes requires some resources, bandwidth mostly. How much of it you need depends on how flexible your schedule is. Generally speaking, You're likely to find you need a lot less of it then you might first imagine. The good news is that scans are easy to parallelize, so you can divide the load over as many different computers and networks as you have access to, to either get the scan finished faster, or to consume fewer resources from each participating scanning node. This is similar logisticly to the distributed computing effort used to break a cryptographic key challenge. The difference is that our effort consumes network bandwidth instead of CPU cycles, and is much much easier. How much easier? (Assuming a search space of 40 million IPs...) One workstation running BASS, with enough memory (to support hundreds of scanning threads), and a T3's equivalence in bandwidth, could probe the entire Internet in under a week at about 4500 JPM. (Jobs Per Minute, the scanner's schedule goal, set on the command line at the beginning of a scanning session, or during recovery). At the other extreme, a small disperse group, running BASS on 10 personal computers with dailup-strength connections, could probe the entire Internet in a month or so at a modest 90 JPM each. (around 2 kilobytes/second). === A minor detour, introducing IDDN. (the International Digital Defense Network) All of this brings us to an interesting idea we've been playing around with that could dramaticly influence Internet security for the good, if / when it is eventually implemented. Frankly, the idea deserves an article of it's own, but since we are so busy, we will introduce it here. Inspired by the high response to cryptographic key challenges, distributed.net and the SETI effort, we vision a non-profit foundation, which we like to ambitiously call IDDN, the International Digital Defense Network, working in the public interest to organize massively distributed scanning efforts which routinely probe the Internet for security vulnerabilities. 10,000 participants could finish a scan cycle every 2-3 days at an insignificant, single JPM each. At the end of a cycle, an automated system could draw the attention of administrators worldwide to some of their local security problems, and offer whatever information and solutions (bug-fixes, patches, workarounds) it has on database (patches, advisories, exploits). In our opinion, such an effort is highly practical and could contribute more to the stability and security of the Internet then the traditional (somewhat pointless?) bruteforce crypto key challenges. We believe organizing an Internet neighborhood-watch of sorts is in everyone's interests, especially the Internet's commercial industry which depend on the Internet to eventually fulfill it's potential for global electronic commerce. We do not have the time or resources to get the IDDN off the drawing board by ourselves and would be interested in the community's input on this issue. === Let the show begin Tuesday, 1 December 1998. We've installed BASS on 8 Unix boxes around the world, each with at least 512kbps bandwidth. 8 different geographicly located participants in 5 different countries: Israel(1), Mexico(1), Russia(2), Japan(2) and Brazil(2). Two machines have already proven their strength during the scanner's painful debugging sessions. Three more will join them for the first time when we begin. The others are backups, ready in case anything goes wrong, and frankly, we have some concerns. Mostly, we expect the scan to raise some complaints, especially passing through the Internet's sensitive military, government and private networks, where snooping around is nothing short of a shooting offense, the prelude to a fullblown attack. Our probes 'come in peace', so to speak, but how can they know? They'll perceive us as a threat and could very well retaliate. We want the scan over before the new year, so we've set BASS's schedulers to finish in 3 weeks, at 250 JPM x 5. If all goes well, we'll be going over the results in the last week of 1998. If not, we'll have an extra week (at least) to fix whatever comes up and still be on schedule. An interesting point to note is how we've constructed the search space. We'll cover the domains by size, starting with the smaller domains first, so by the first week we'll have finished scanning 216 of the 228 active domains in the DNS (*.org, *.gov, *.int, and 212 countries, from Afghanistan with 1 host to the UK with 1.4 million). We create the individual search space of each participant by dividing the global space the same way you would deal a deck of cards, so that the original scanning order is preserved. At 02:00 GMT, we flip the switch, so to speak, activating BASS on the five participating hosts. Since these have all been configured to automaticly recover from any power failure or unexpected system shutdowns, we really don't have much to do now, besides keeping a lazy eye on progress. === First week. There is definitely a response out there to the scan, but it's much friendlier then we anticipated. Harmless acts of mindless automata and mutual curiosity, mostly. Pings, traceroutes, telnet sessions and finger attempts. Four to eight portscans a day. An occasional TCP/IP stack exercise, an OS fingerprint, a few mostly polite e-mails asking why our network was "attacking" theirs, frequently warning us that crackers may be abusing our systems, suggesting we look into it. Very mild, we are running into much less hostility then we expected. People either don't realize the scope of the scan, or don't care. On an individual basis, one quick security probe isn't usually enough to get the local sysadmin to notice. Those who do are probably security conscious enough to keep their networks up to date anyway, and confident enough to keep their cool when yet another 13 year old punk (who else?) bangs on their network walls. Oh, did we mention the scanner is precisely on schedule? 12 million hosts scanned by the end of the week, covering the US government's *.gov domain, Canada, Australia, Europe, and a window to some of the most intriguing corners of the world: Hostile mind-control regimes like China and Iran for example, which suffocate their repressed population's access to free ideas and information, but are still paradoxicly connected (albeit, very poorly) to the Internet. Third world potentials like India (the world's largest democracy!) and the rapidly developing countries of the far east. Exotic paradise locations like the Cocos Islands, Bahamas, the Virgin Islands, Barbados, Fiji, and Micronesia All of them as close and accessible as if they were right across the street, and in a certain way even closer. Computer expertise is rare in many of these countries, security expertise even rarer. Cracking into a Chinese computer half a world away, for example, is usually easier, more interesting, and safer (assuming you are not in Chinese jurisdiction of course) then cracking into a comparable western computer. As a precaution, all eight participants have backed up the 13 MBs worth of precious results, to make sure an emergency relocation recovery is possible, should this become necessary. (I.e, in case of a small thermonuclear attack on one or more scanning participants, possibly effecting their performance. Caution, nuclear warfare can really ruin your entire scan) === Second week. We started the week off by scanning US Military networks. Admitingly, we were pretty nervous, and spent much of the day keeping an eye out for telltale signs of a pissed off military retaliation. (also known as "InfoWar" and "spooky shit" in professional terminology) In just under 24 hours it was all over, and while we did notice a significant increase in the number of probes we were getting, to say we were not impressed by the security of the military network is a big fat major understatement. This might not be a problem, since according to NSCS (National Computer Security Center) network security policies, none of the systems on the public *.mil network could qualify for the storage and handling of classified DoD (Department of Defense) information. How strictly these policies are adhered to is another matter. And even if they are (and this is a _big_ if), the DoD is still (justifiably) concerned that crackers might glue together classified information from the little pieces of unclassified information fragments lying around their *.mil network (in great abundance). So they have plenty of good reasons to keep their network secure, but are (un)?fortunately doing a pretty lousy job. = DoS six o'clock. Wednesday, our Russian scanner runs into trouble. A denial of service attack, 512kbps stream of packets amplified 120 times strong over an unsuspecting Canadian broadcast amplifier. Half a world a way, the packet storm brings a large Russian ISP to it's knees, overwhelming it's available bandwidth. Ouch. Apparently, we stepped on someone's toes. At first, we assumed this was somehow connected to yesterday's *.mil scan, but no, it was just some ill-tempered English fellow who didn't appreciate getting probed last Monday. He tried crashing our stack first, with some nasty DoS attacks for NT and Unix. That didn't work, so he blasted our ISP out of the sky. Clear and simple, he didn't want to, but we left him no choice. You can't have decent English folks being polked around at by some Russian punks ... The attack lasted 16 hours straight, and since it wasn't too difficult to track down where it was coming from, we were very tempted to return the favor, or at least give this trigger-happy netizen a free security audit. We didn't though, the net's resources are much too valuable to further waste on such brutish exhibition of ego (a "cyber" pissing contest?). Besides, an eye for an eye and everyone goes blind, right? Anyway, one of our backups (also in Russia) quickly substituted for the lost computer as soon as we noticed the attack 6 hours later at 255 JPM, with no other significant setbacks to our week's schedule. The rest of the week chugged along nicely, scanning the United States (or more precisely, the *.us domain), Japan (*.jp), and the educational networks (*.edu). Hmmm, Has anyone noticed how unsymmetricly biased the DNS is in favor of the United States? Dot gov, dot mil, dot org, dot edu. Being so homogeneously American, shouldn't these go under the *.us domain? = "You're gonna rot in jail" - the legal corner We've began receiving e-mail's this week by people with alot less tolerance for our activities, most in delayed response to last week's scans. Some of these were written by lawyers who informed us we were either supporting or perpetrating acts of computer crime against their clients. They had notified the authorities (CERT and the FBI were commonly cited) and threatened to take us to court if we did not offer our full cooperation in immediately identifying the attacking party. Right... It seems some organizations hire fulltime "security officers" known for exaggerating the significance of petty incidents to justify getting payed. Unfortunately, in certain parts of the worlds, charges like these can cost you a fortune in legal defense, and with the wrong judge, a conviction, and a sentence anywhere between a large fine, and a few years in jail. Fortunately, on the Internet, getting around this is as easy as scanning from places which are not known for overzealousness in regard to their definition of "computer crime". This is just another example of how poorly the local and international legal system deals with so called "computer crime" and the Internet. Under the (US) state of Oregon's computer crime law (164-377a), for example, we could definitely be defined as computer criminals, trailed and sent away to many years in prison. (But so could everyone else...) A chosen excerpt from the law: (4) Any person who knowingly and without authorization uses, accesses or attempts to access any computer, computer system, computer network, or any computer software, program, documentation or data contained in such computer, computer system or computer network, commits computer crime. As you can see, the law is unreasonably vague. "Criminal" or not, it all comes down to your definition of "authorization". But, having it would constitute some sort of prior agreement between a user and the owners of a computer, computer network or computer software. The Internet however is a public network, and the majority of it's services are used anonymously, by users with which there is no persistent relationship. In the physical world, any behavior is possible, so society enforces order by restricting behavior it finds unacceptable through the regulative government system, which is "programmed" by the code of the law. The computer world is pure code, instructions and information, none of which are capable of discrimination. The computer programmer is the god of a perfectly obedient universe. Like the artist, the canvas of his creation is as expressive or inexpressive of his will and intention as he has made it to be. This means software, like the law, can inherit the imperfections of it's creator. Poorly written computer and legal code can allow the system to behave in conflict with the original intentions of the men who wrote it. Legal loopholes and software bugs, Lawyers and Hackers, different sides of the same coin. The only way to really prevent the abuse of the system is to write better code. This is the reason we find most "computer crime" legislation so absurd. The laws try to protect computer systems from being misused, when the only definitive expression of what constitutes "acceptable use" is in the code itself, which may or may not be a precise manifestation of the author's intentions, depending on his competence as a programmer. If the public insists on "computer crime" legislation anyway, we believe most of the it's problems could be easily resolved by eliminating ambiguous wording, over generalization, and specificly breaking down what the law defines as acts of "computer crime": 1) knowingly exploiting a finite list of common misimplementations (bufferoverflow, a race condition, ...) 2) intentionally performing a Denial of Service attack. 3) wiretapping (sniffing a network, capturing keyboard strokes, screen content, etc.) 4) using a party's identification token[s] (username / password) without the party's permission. (logging into a system on someone elses account, reading someone else's email) 5) Spam. (death penalty for repeated offenders) Note that we've removed "attempted" attacks from the offense list, since these are hard to define, prove, and cause no damage. (If in the course of an attempted attack a system is damaged, in a denial of service attack for example, then we can prosecute this event as a separate incident, with nothing "attempted" about it) Interested readers are advised to read up on the Oregon vs. Randal L. Schwartz case, a good example as to why Draconian "computer crime" legislation should be fought with a vengeance. (http://www.lightlink.com/fors) === Third week: Last week. Only the mammoth *.com and half of the *.net domain left and we're done. = they're heeeere... Friday, our Japanese participants discover that a computer on their company network has been cracked into, one very secure Linux box running only SSH and Apache 1.3.4. Now this would definitely send a chill up your spine if you knew just how fanatic our friends are when it comes to network security. Furthermore, they only detected the intrusion three days after the fact, which is unbelievable when you consider the insane monitoring levels they've been keeping since they agreed to participate in the scan. They would have noticed any funny stuff, and in fact, they did, lots of it, but none of which came close enough to a security breach to raise any alarms. Readers should also note how although a key binary in the cracked machine had been modified, tripwire and an assortment of other booby traps failed to detect this had happened. Even a close-up manual inspection (comparing file contents with a trusted backup, playing with it's name) could not detect any odd behavior. This trick, and others equally spooky were achieved by clever manipulation of the OS's kernel code (dynamicly, through a module). Other characteristics of the attack which make it so eerily sophisticated: 1) The attacker (convincingly) masquerades as a local employee. The attacker knows the employee's username and password and is even connecting through the employee's Japanese ISP on the employee's account! (the phone company identified this was an untraceable overseas caller) This information could not have been sniffed, since network services are only provided over encrypted SSH sessions. Further investigation shows that this employee's personal NT box, connected over a dynamic dailup connection, had been cracked into 4 days earlier. His ssh client (TTSSH extension to TeraTerm) had been trojaned to transmit XOR garbled account information (hostname/username/password) over pseudo-DNS udp packets to a refurnished i486 Redhat v4.2 box used as a single-purpose cheap Samba fileserver in a small Australian ISP. The little box was every cracker's dream, a discrete, utopian crack haven, installed by a former Linux-savvy administrator, the last of it's kind in a homogeneous Unix-illiterate Microsoft environment. The ISP practicly ignored the box, which was running (up 270 days straight) so reliably none of them had even bothered to log in since mid 1997! So as long as the crackers kept Samba running, they would the box completely to themselves. How the NT box was cracked into in the first place is still a mystery. The logs weren't helpful (surprise! surprise!) and the only way we were even able to confirm this had happened was by putting a sniff on the NT's traffic (following a hunch) and catching those sneaky packets redhanded, transmitting our SSH identification down under. We never liked NT before, being generally suspicious of propriety blackbox OS, from a company with a long history of poor quality bloatware. But realizing just how helpless we were against an attacker that obviously knew the ins and outs of this can-of-worms OS, the company recognized that NT was a serious security hazard and changed it's security policies to keep it as far away from it's systems as possible, and this included restricting employees from using it from at home to log into the company network (even with SSH). 2) The attacker is using a custom built software penetration agent. This is only an hypothesis, but is strongly supported by the fact that the entire attack only lasted an incredible 8 seconds! During which the attacker manages to log on (over an employee's SSH account, no less), gain root privileges, backdoor the system, remove any (standard) traces of it's activity and log off. And they probably would have gotten away with it too, if it wasn't for those meddling kids! Who thoughtfully installed a crude old tty surveillance-camera hack that trapped IO calls to and from isatty(3) file descriptors, in realtime, saving them on file along with a timestamp for neato it's-almost-as-if-you-were-there playback qualities. And Wow! If there ever was a crack to appreciate for it's elegance, simplicity, and efficiency, this was it. First off this thing is smoking fast! Which puts the likelihood of any manual intervention at square zero. It's also mean and lean. Forget fumbling with an FTP client, leave that to the slow soft pink-bellied human cracker-weenies, real agents pump files directly through the shell (uuencode(1)'d at one end, uudecode(1)'d at the other). Extending privileges with an army of amateurish recipe-book Bugtraq exploits? I think not! Introducing the super-exploit, an all-in-one security penetration wonder which quickly identifies and exploits any local security vulnerabilities for that wholesome, crispy, UID zero flavor (we were vulnerable to a recent KDE buffer overflow). After promptly confirming it's shiny new root privileges, the agent transfers it's last archive (a cross between a self-installing feature-rich backdoor, and a clean-up-the-mess, we-were-never-here log doctor), executes it and logs off. After watching the attack on playback (at 1/8 of it's original speed) several times over, standard security-compromise ritual kicked in. We took the affected machine offline, remounted the disks read-only, fired up our trusty filesystem debugger, and slaved away to salvage whatever we could. Luckily, we found the attacker's transfered archives still intact, along with large fragments of the undoctored logs, allowing us to fill any still-missing details on the blitz attack. At the end of the day, when we finished playing with the cracked machine on loopback, we changed the compromised account's password, restored binary integrity, rebooted the system and put it back on the network, this time running a network dump of all it's incoming-outgoing traffic, just to be on the safe side. Whoever they were, they certainly knew what they were doing, and for the most part seemed very good at it. But being determined, clever, and sophisticated just doesn't cut it when you do battle with wizardly foes (that's us) yielding the great powers of the Universe to their command: Dumb luck and clinical paranoia. So who done it ??? Could it be ... (A government conspiracy I tell ya'!) Any one of the many press-savvy three letter agencies scrambling for a bigger slice of the US-government funding pie? They've got motive, but are they really sneaky, clue-full and competent enough to take the blame? How about the SIGINT spooks? The NSA (Information superiority for Americans!), or the GHCQ (Her Royal Majesty's Intelligence)? Someone working for the Chinese? The KGB? The Russian mob? The giant from Redmond? Elvis and Bigfoot?! Who knows ... They tried something spooky 2 nights later, when around 4 AM (Japanese time) our network dump captures several pseudo-DNS udp packets originating from a familiar Linux box in a small Australian ISP. We assume they were attempting to communicate with the software they left behind during their brisk first visit. Several minutes pass, and the attempt is followed by a "TCP ping" (a stealthy alternative to an ICMP ping), several more pseudo-DNS udp packets, and silence. To the best of my knowledge, we haven't heard from them since. How discrete. === End of the road That's it, it's over, on time, 10 days before the new year, 1999. Our success. Scattered across the world, from Japan to Russia, from the Middle East to Mexico to Brazil. We were all awake when the scanners calmed down, within an hour of each other, on Dec 21th, 1998 08:00 GMT. We celebrated the event at "the bunker" (see suffix item 2 for details), a discrete gathering corner where we hang out, meditate, plot, debate, and coordinate cr^H^Hhacking campaigns of mystical lore.
Most of the attention (not to mention conversation) concentrated around "iap-results.txt.gz", a humble 6.4 MB compressed (1:8 ratio) textfile which embodied the sum results of our 4 month long effort. In no time, people downloaded local copies of the post, and were reading, grepping, parsing, cross referencing and analyzing this, that and other. It was unbelievable non-stop fun the likes we had never before and never since enjoyed at the bunker. A very memorable un"real" moment. It's funny how close the Net can bring a group of people who have never "really" met, who've never "really" seen each other face to face. And it doesn't seem to "really" matter, it's just as "real", as "real" as anything else gets. "real" is really overrated these days anyway, I mean, really. "He's suffering from some sort of reality complex,.. obviously." Friendship, cooperation, common interests, goals and ideals. They're the same here, in this funny netherworld, "cyberspace", as anywhere else. Across the barriers of culture, language and geography. The universality of human kinship, the couple, the pact, the tribe, the organization, the community, gracefully extended into the online domain. It's all about having a medium, connecting people, communicating. Together we are better. === IAP cheat-sheet BEGIN TIME: 02:00, Dec 01, 1998 GMT END TIME: 08:00, Dec 21 1998 GMT Scanning nodes: 5 Jobs Per Minute: 250 Scan time: 20.24 days Vulnerabilities tested: 18 Domain count: 7 three letter domains, 214 national domains (see suffix item 3) Host count: 36,431,374 Vulnerability count: 730,213 Vulnerable host count: 450,000 Statistical output: service | vulnerability count, percentage -------------------------------------------------------- webdist | 5622 hosts counted, 0.77% from total wu_imapd | 113183 hosts counted, 15.5% from total qpopper | 90546 hosts counted, 12.4% from total innd | 3797 hosts counted, 0.52% from total tooltalk | 190585 hosts counted, 26.1% from total rpc_mountd | 78863 hosts counted, 10.8% from total bind | 132168 hosts counted, 18.1% from total wwwcount | 86165 hosts counted, 11.8% from total phf | 6790 hosts counted, 0.93% from total ews | 9346 hosts counted, 1.28% from total (other vulnerabilities which weren't common enough to generate statistics for) other: | 18K hosts counted, 2.42% from total === Conclusions A global fury of half a billion packets, digital signals zipping back and force across the planet at the speed of light. Above the Earth, across the land, under the sea, over satellite microwave, copper wiring, fiberoptics, wireless and undersea cable. Probing cyberspace. Pretty cool, the kind of power information technology puts in our hands these days. Seven hundred thousand vulnerabilities, gaping holes, wounds in the skin of our present and future information infrastructures, our dream for a free nexus of knowledge, a prosperous digital economy, where we learn, work, play and live our lives. Easy pickings, at the fingerprints of anyone who follows in our footsteps, friend or foe. These open points of penetration immediately threaten the security of their affiliated networks, putting many millions of systems in commercial, academic, government and military organizations at a high compromise risk. Ironicly, the sheer mass of vulnerable hosts on the Internet offers it's members a primitive form of protection, that is, in a you-can-eat-the-other-guy school of fish sort of way. Unfortunately, this doesn't work when you're flashing bright colors and look tasty. If you show up when a shark greps your school for "bank", you're in really bad shape. As this is *not* an example. We were stunned to find just how many networks you would expect to be ultra secure were wide open to attack. Banks, billion dollar commerce sites, computer security companies, even nuclear weapon research centers, goddamit! You'd think people would have some good sense and _at least_ patch their systems when an advisory comes out. "Computers are unreliable, but humans are even more unreliable. Any system which depends on human reliability is unreliable." - Gilb Looking at the big picture, the problem gets worse. A catastrophe in the works. So far, we've been pretty lucky. Consider the power these unsecure networks represent _together_. Penetrating and controlling millions of hosts? You couldn't do it manually, but with the right software, you could automate most of the dirty work. You'd need a careful network worm (suffix item 4), stealthy remote administration software (suffix item 5) and a self organizing network nervous system by which you could propagate control. Imagine the implications if this sort of capability ever fell into the wrong hands. A government (China perhaps), a political terrorist group or organized crime. On bandwidth alone they could shut down any part (or all) of the Internet in mammoth DoS attacks. A country, a portal, a news site, or maybe just InterNIC. Leverage and attention, for fun and profit. They could "build" the world's largest distributed supercomputer, or construct an Intelligence network rivalled only by the NSA's Echelon. Of course, who says only one group can play the game? Struggles for power in the digital domain could very well develop into the world's first real information war, with the very future of the Internet as a free unregulated supernetwork caught in the cross fire. Unlikely? Far fetched? We hope so. Still, with all the hype Y2K is getting, it seems ludicrous that the most serious _real_ threat to information technology is consistently ignored. The only thing necessary for the triumph of evil is for good men to do nothing. Wake up fellow countrymen. Let's get to work. Everywhere you go you'll see them searching, Everywhere you turn you'll feel the pain, Everyone is looking for the answer, Well look again. -- Moody Blues, "Lost in a Lost World" === SUFFIX = [item 1] Vulnerabilities BASS can test for (as of version 1.0.7): General: bind CA-98.05 wu_imapd CA-98.09 innd CA-97.08 qpopper CA-98.08 RPC: rpc.mountd CA-98.12 tooltalk CA-98.11 CGI: wwwcount phf php handler compas faxsurvey webdist ews glimpse info2www webgais websendmail = [item 2] "the bunker" - a technical reference guide "The bunker" was hacked together by a friend who noticed how badly the group needed a realtime, secure communication forum. Our configuration combines an unmodified IRC server, SSH, a firewall and a Linux box (or two). There are two possible implementations, one more secure then the other but also (slightly) more expensive (you'll need another cheap i86 box). We'll start with our (secure) configuration. We take a cheap Linux box (i486, 8mb RAM, 500mb diskspace, two $15 Ethernet cards), with the bare minimum Debian installation, remove any "privilege relays" (network services, daemons (crond), suid files) and configure the kernel _with_ firewall support and _without_ IP forwarding. We then installed the SSH suite, and double check to make sure the *only* available network service is sshd's port 22 (ICMP / UDP included). As an additional layer of security, we enforce our SSH only policy at the OS level, by setting up the kernel's IP firewall to reject *all* incoming and outgoing _Internet_ packet traffic by default, except what we explicitly need to maintain *incoming* SSH sessions. incoming rules: * default policy: deny * accept TCP packets from any source to thebunker.com port SSH(22) outgoing rules: * default policy: deny * accept TCP packets from thebunker.com port SSH(22) to any destination An example implementation (Our ipfwadm(8) bootup configuration): #!/etc/ipfw/ipfw-setup # * eth0 interfaces the Internet, and eth1 interfaces the private IRC # server. # # * On 2.2.X kernels and higher the IP firewalling code has been replaced, # so ipfwadm (and this configuration) will no longer work. ipchains(8) # should be used instead. # * Since we are not forwarding between interfaces, 0.0.0.0/0 can be used # as a safe (portable) alternative to our IP address. Those of you # who would rather be specific should put their IP here with a mask of 32. # (For example: 188.8.131.52/32) ipfwadm -I -f ipfwadm -I -p deny ipfwadm -I -a accept -W eth1 ipfwadm -I -a accept -W eth0 -P tcp -D 0.0.0.0/0 22 ipfwadm -O -f ipfwadm -O -p deny ipfwadm -O -a accept -W eth1 ipfwadm -O -a accept -W eth0 -P tcp -S 0.0.0.0/0 22 ---[ EOF ]--- A simple, airtight firewall. One interface faces the Internet, and the other jacks straight into the safehouse (our IRC server), which should *not* be capable of accessing the Internet directly and vice versa. The safehouse is a similarly configured bare metal, secure Linux configuration running _only_ Ircd (_not_ as root!) and sshd. General purpose use of the safehouse is strongly discouraged. User accounts on the firewall are opened for authorized members of the group, but despite trusting the system's users, access to administrative account must be strictly limited. This is to insulate the system from the possible security problems of its users, with the added benefit of protecting a user from coercion (they couldn't compromise security if their life depended on it). The second configuration may be less secure, depending on your risk model, but is also less expensive. You would only need one Linux box, and one Ethernet card. We eliminate the "safehouse" and trust the firewall to run the Ircd server safely on loopback (_not_ as root!), while isolating it from the Internet. In this case, the security of the system _depends_ on correctly enforcing the strict IP firewall filters, and these are not merely an additional layer of security. Because we are running a service on loopback, the IP firewall must be set up to allow packets to and from the server on the local interface. While this setup is theoreticly secure "enough", it leaves a larger margin for error and malice. In a nostalgic tribute to the old BBS days, "the bunker" features a black and white (green), menu driven default login shell (based on pdmenu), which greets users with the message of the day, announces events, and offers a consistent customizable UI to local mail, project forums, IRC (directly into the official, often the only system channel), and an ever growing list of other system activities. ("just one more feature"!) The interface started out as a joke, and while it sounds out of date, with the current explosion of graphics, sound and video on the WWW, it's oddly cozy, and most of us have warmed around to it. (besides, when real work needs to get done, reaching emacs (or a shell) is just a key-press away) = [item 3] domains scanned 7 three letter domains: com - Commercial net - Networks edu - Educational mil - US Military org - Organizations gov - Government int - International Organizations 214 national domains (sorted by size, left right, top down): jp (Japan) us (United States) uk (United Kingdom) de (Germany) ca (Canada) au (Australia) nl (Netherlands) fi (Finland) fr (France) se (Sweden) it (Italy) no (Norway) tw (Taiwan, Province Of China) dk (Denmark) es (Spain) ch (Switzerland) br (Brazil) kr (Korea, Republic) be (Belgium) ru (Russian Federation) za (South Africa) at (Austria) nz (New Zealand) mx (Mexico) pl (Poland) il (Israel) hu (Hungary) hk (Hong Kong) cz (Czech Republic) sg (Singapore) ar (Argentina) ie (Ireland) gr (Greece) pt (Portugal) my (Malaysia) tr (Turkey) cl (Chile) ee (Estonia) is (Iceland) th (Thailand) su (Soviet Union) sk (Slovakia, Slovak Republic) ae (United Arab Emirates) si (Slovenia) cn (China) ro (Romania) co (Colombia) ua (Ukraine) id (Indonesia) uy (Uruguay) in (India) lv (Latvia) lt (Lithuania) ph (Philippines) ve (Venezuela) bg (Bulgaria) hr (Croatia 'Hrvatska') yu (Yugoslavia) lu (Luxembourg) kw (Kuwait) do (Dominican Republic) pe (Peru) cy (Cyprus) nu (Niue) cr (Costa Rica) pk (Pakistan) na (Namibia) lb (Lebanon) tt (Trinidad And Tobago) eg (Egypt) kg (Kyrgyzstan) to (Tonga) gl (Greenland) pr (Puerto Rico) ec (Ecuador) kz (Kazakhstan) bm (Bermuda) bn (Brunei Darussalam) py (Paraguay) zw (Zimbabwe) mt (Malta) gt (Guatemala) sv (El Salvador) cc (Cocos 'Keeling' Islands) cx (Christmas Island) pa (Panama) by (Belarus) ni (Nicaragua) ge (Georgia) ke (Kenya) om (Oman) bw (Botswana) bo (Bolivia) fo (Faroe Islands) bh (Bahrain) mu (Mauritius) ma (Morocco) lk (Sri Lanka) ad (Andorra) mk (Macedonia, Former Yugoslav) md (Moldova, Republic) bs (Bahamas) vi (Virgin Islands, US) ng (Nigeria) am (Armenia) ba (Bosnia And Herzegowina) jo (Jordan) ky (Cayman Islands) li (Liechtenstein) jm (Jamaica) sa (Saudi Arabia) gi (Gibraltar) zm (Zambia) pf (French Polynesia) sz (Swaziland) tm (Turkmenistan) bz (Belize) mc (Monaco) ir (Iran, Islamic Republic) ci (Cote D'Ivoire) uz (Uzbekistan) sm (San Marino) ai (Anguilla) fj (Fiji) sn (Senegal) gh (Ghana) bf (Burkina Faso) ag (Antigua And Barbuda) fm (Micronesia, Federated States) az (Azerbaijan) gp (Guadeloupe) np (Nepal) dm (Dominica) mo (Macau) mz (Mozambique) tz (Tanzania, United Republic) pg (Papua New Guinea) st (Sao Tome And Principe) ug (Uganda) nc (New Caledonia) gf (French Guiana) tg (Togo) mv (Maldives) gu (Guam) al (Albania) hn (Honduras) im (Isle of Man) aw (Aruba) cu (Cuba) vu (Vanuatu) tc (Turks And Caicos Islands) et (Ethiopia) tj (Tajikistan) hm (Heard And Mc Donald Islands) gy (Guyana) tn (Tunisia) mg (Madagascar) kh (Cambodia) ac (Ascension Island) as (American Samoa) nf (Norfolk Island) aq (Antarctica) io (British Indian Ocean Territory) ck (Cook Islands) bb (Barbados) gb (United Kingdom) je (Jersey) mq (Martinique) sh (St. Helena) bt (Bhutan) vn (Viet Nam) ms (Montserrat) lc (Saint Lucia) dz (Algeria) vg (Virgin Islands, British) ye (Yemen) sb (Solomon Islands) mn (Mongolia) ls (Lesotho) gg (Guernsey) ne (Niger) mr (Mauritania) mp (Northern Mariana Islands) gw (Guinea-Bissau) sl (Sierra Leone) qa (Qatar) tf (French Southern Territories) bj (Benin) va (Vatican City State) cd (Congo, Democratic Republic) an (Netherlands Antilles) km (Comoros) sc (Seychelles) gs (South Sandwich Islands) kn (Saint Kitts And Nevis) ly (Libyan Arab Jamahiriya) pn (Pitcairn) gd (Grenada) cm (Cameroon) tp (East Timor) mh (Marshall Islands) ws (Samoa) um (United States Minor Outlying Islands) tv (Tuvalu) sy (Syrian Arab Republic) re (Reunion) pw (Palau) mw (Malawi) mm (Myanmar) ml (Mali) lr (Liberia) cv (Cape Verde) cg (Congo, Republic) af (Afghanistan) = [item 4] Lukemia One of our first research projects (circa 1997) involved researching possible designs of a modern network worm. We even developed a prototype in C which implements some of our ideas. Today, we're pretty horrified by our choice of language (In C, everything is equally difficult, "help save the world!" -- use Perl) and the quality of the code (butt ugly). = [item 5] Portacelo "Local security subversion. Why human and (current) software (tripwire and others) host-based Intrusion Detection Systems are a bad idea." We did some research (right after the IAP was over) in this subject, and plan to release an article sometime in the near future. A fully-featured backdoor implementation is available, demonstrating the concept, which combines SSH ESP (suffix item 6), a kernel module, direct memory manipulation, and a good old fashioned binary trojan. = [item 6] SSH ESP A hacked SSH suite modified to implement ESP (Encapsulated encrypted STREAMS Protocol) at the application level. Notable features include: * piercing almost any current filter firewall. (ab-uses any available packet traffic: tcp, udp and icmp) * invisible at the operating system level. (netstat and friends will not register any activity) * practical. (ESP is almost as fast and reliable as TCP, including error correction) * military strength encryption. (thanks to SSH) = [iem 7] Note to the reader Christ, it took me, Liraz, over 2 weeks to write this silly article, during which I had to drop whatever I was doing, and devote the bulk of my time to writing this memorandum of the IAP in English, which is not my native language. (Disclaimer: Please excuse any errors in syntax, grammar or spelling. That felt good. Please forgive my bad writing, untasteful dramatics, poor sense of humor... I'll stop now...) In the process I had to convince my fellow project associates (some of them very strong willed) that documenting the IAP was A Good Thing, at least for posterity's sake... And all so I could offer you, dear reader, a chance to share some of my humble insights on computer security, and a taste of hacker culture. This is my first publication, I'm not too sure on how this is going to be accepted. Frankly, I prefer writing code, so I'm not sure I'll be writing any more articles soon. Whether or not that happens depends on the response I get from interested readers. If there is a good response, there will be more. But goddammit, they'll be shorter this time! I hope the article wasn't too technical for your tastes, but the project was mostly about overcoming technical and logistical difficulties, so that was hard to escape. Also, I am very short on time and resources, so if anyone is interested in sponsoring the material (an official SSR website for the rant and the software), that would be great. Oh, any takers on the IDDN front? We can start out with a (preferably archived) mailing list, find some interested people, get the ball rolling... All points of contact: email@example.com