A REPORT ON THE INTERNET WORM Bob Page University of Lowell Computer Science Department November 7, 1988 [Because of the many misquotes the media have been giving, this report is Copyright (c) Bob Page, all rights reserved. Permission is granted to republish this ONLY if you republish it in its entirety.] Here's the scoop on the "Internet Worm". Actually it's not a virus - a virus is a piece of code that adds itself to other programs, including operating systems. It cannot run independently, but rather requires that its "host" program be run to activate it. As such, it has a clear analog to biologic viruses -- those viruses are not considered live, but they invade host cells and take them over, making them produce new viruses. A worm is a program that can run by itself and can propagate a fully working version of itself to other machines. As such, what was loosed on the Internet was clearly a worm. This data was collected through an emergency mailing list set up by Gene Spafford at Purdue University, for administrators of major Internet sites - some of the text is included verbatim from that list. Mail was heavy since the formation of the list; it continues to be on Monday afternoon - I get at least 2-3 messages every hour. It's possible that some of this information is incomplete, but I thought you'd like to know what I know so far. The basic object of the worm is to get a shell on another machine so it can reproduce further. There are three ways it attacks: sendmail, fingerd, and rsh/rexec. THE SENDMAIL ATTACK: In the sendmail attack, the worm opens a TCP connection to another machine's sendmail (the SMTP port), invokes debug mode, and sends a RCPT TO that requests its data be piped through a shell. That data, a shell script (first-stage bootstrap) creates a temporary second-stage bootstrap file called x$$,l1.c (where '$$' is the current process ID). This is a small (40-line) C program. The first-stage bootstrap compiles this program with the local cc and executes it with arguments giving the Internet hostid/socket/password of where it just came from. The second-stage bootstrap (the compiled C program) sucks over two object files, x$$,vax.o and x$$,sun3.o from the attacking host. It has an array for 20 file names (presumably for 20 different machines), but only two (vax and sun) were compiled in to this code. It then figures out whether it's running under BSD or SunOS and links the appropriate file against the C library to produce an executable program called /usr/tmp/sh - so it looks like the Bourne shell to anyone who looked there. THE FINGERD ATTACK: In the fingerd attack, it tries to infiltrate systems via a bug in fingerd, the finger daemon. Apparently this is where most of its success was (not in sendmail, as was originally reported). When fingerd is connected to, it reads its arguments from a pipe, but doesn't limit how much it reads. If it reads more than the internal 512-byte buffer allowed, it writes past the end of its stack. After the stack is a command to be executed ("/usr/ucb/finger") that actually does the work. On a VAX, the worm knew how much further from the stack it had to clobber to get to this command, which it replaced with the command "/bin/sh" (the bourne shell). So instead of the finger command being executed, a shell was started with no arguments. Since this is run in the context of the finger daemon, stdin and stdout are connected to the network socket, and all the files were sucked over just like the shell that sendmail provided. THE RSH/REXEC ATTACK: The third way it tried to get into systems was via the .rhosts and /etc/hosts.equiv files to determine 'trusted' hosts where it might be able to migrate to. To use the .rhosts feature, it needed to actually get into people's accounts - since the worm was not running as root (it was running as daemon) it had to figure out people's passwords. To do this, it went through the /etc/passwd file, trying to guess passwords. It tried combinations of: the username, the last, first, last+first, nick names (from the GECOS field), and a list of special "popular" passwords: aaa cornelius guntis noxious simon academia couscous hacker nutrition simple aerobics creation hamlet nyquist singer airplane creosote handily oceanography single albany cretin happening ocelot smile albatross daemon harmony olivetti smiles albert dancer harold olivia smooch alex daniel harvey oracle smother alexander danny hebrides orca snatch algebra dave heinlein orwell snoopy aliases december hello osiris soap alphabet defoe help outlaw socrates ama deluge herbert oxford sossina amorphous desperate hiawatha pacific sparrows analog develop hibernia painless spit anchor dieter honey pakistan spring andromache digital horse pam springer animals discovery horus papers squires answer disney hutchins password strangle anthropogenic dog imbroglio patricia stratford anvils drought imperial penguin stuttgart anything duncan include peoria subway aria eager ingres percolate success ariadne easier inna persimmon summer arrow edges innocuous persona super arthur edinburgh irishman pete superstage athena edwin isis peter support atmosphere edwina japan philip supported aztecs egghead jessica phoenix surfer azure eiderdown jester pierre suzanne bacchus eileen jixian pizza swearer bailey einstein johnny plover symmetry banana elephant joseph plymouth tangerine bananas elizabeth joshua polynomial tape bandit ellen judith pondering target banks emerald juggle pork tarragon barber engine julia poster taylor baritone engineer kathleen praise telephone bass enterprise kermit precious temptation bassoon enzyme kernel prelude thailand batman ersatz kirkland prince tiger beater establish knight princeton toggle beauty estate ladle protect tomato beethoven euclid lambda protozoa topography beloved evelyn lamination pumpkin tortoise benz extension larkin puneet toyota beowulf fairway larry puppet trails berkeley felicia lazarus rabbit trivial berliner fender lebesgue rachmaninoff trombone beryl fermat lee rainbow tubas beverly fidelity leland raindrop tuttle bicameral finite leroy raleigh umesh bob fishers lewis random unhappy brenda flakes light rascal unicorn brian float lisa really unknown bridget flower louis rebecca urchin broadway flowers lynne remote utility bumbling foolproof macintosh rick vasant burgess football mack ripple vertigo campanile foresight maggot robotics vicky cantor format magic rochester village cardinal forsythe malcolm rolex virginia carmen fourier mark romano warren carolina fred markus ronald water caroline friend marty rosebud weenie cascades frighten marvin rosemary whatnot castle fun master roses whiting cat fungible maurice ruben whitney cayuga gabriel mellon rules will celtics gardner merlin ruth william cerulean garfield mets sal williamsburg change gauss michael saxon willie charles george michelle scamper winston charming gertrude mike scheme wisconsin charon ginger minimum scott wizard chester glacier minsky scotty wombat cigar gnu moguls secret woodwind classic golfer moose sensor wormwood clusters gorgeous morley serenity yaco coffee gorges mozart sharks yang coke gosling nancy sharon yellowstone collins gouge napoleon sheffield yosemite commrades graham nepenthe sheldon zap computer gryphon ness shiva zimmerman condo guest network shivers cookie guitar newton shuttle cooper gumption next signature [I wouldn't have picked some of these as "popular" passwords, but then again, I'm not a worm writer. What do I know?] When everything else fails, it opens /usr/dict/words and tries every word in the dictionary. It is pretty successful in finding passwords, as most people don't choose them very well. Once it gets into someone's account, it looks for a .rhosts file and does an 'rsh' and/or 'rexec' to another host, it sucks over the necessary files into /usr/tmp and runs /usr/tmp/sh to start all over again. Between these three methods of attack (sendmail, fingerd, .rhosts) it was able to spread very quickly. THE WORM ITSELF: The 'sh' program is the actual worm. When it starts up it clobbers its argv array so a 'ps' will not show its name. It opens all its necessary files, then unlinks (deletes) them so they can't be found (since it has them open, however, it can still access the contents). It then tries to infect as many other hosts as possible - when it sucessfully connects to one host, it forks a child to continue the infection while the parent keeps on trying new hosts. One of the things it does before it attacks a host is connect to the telnet port and immediately close it. Thus, "telnetd: ttloop: peer died" in /usr/adm/messages means the worm attempted an attack. The worm's role in life is to reproduce - nothing more. To do that it needs to find other hosts. It does a 'netstat -r -n' to find local routes to other hosts & networks, looks in /etc/hosts, and uses the yellow pages distributed hosts file if it's available. Any time it finds a host, it tries to infect it through one of the three methods, see above. Once it finds a local network (like 129.63.nn.nn for ulowell) it sequentially tries every address in that range. If the system crashes or is rebooted, most system boot procedures clear /tmp and /usr/tmp as a matter of course, erasing any evidence. However, sendmail log files show mail coming in from user /dev/null for user /bin/sed, which is a tipoff that the worm entered. Each time the worm is started, there is a 1/15 chance (it calls random()) that it sends a single byte to ernie.berkeley.edu on some magic port, apparently to act as some kind of monitoring mechanism. THE CRACKDOWN: Three main 'swat' teams from Berkeley, MIT and Purdue found copies of the VAX code (the .o files had all the symbols intact with somewhat meaningful names) and disassembled it into about 3000 lines of C. The BSD development team poked fun at the code, even going so far to point out bugs in the code and supplying source patches for it! They have not released the actual source code, however, and refuse to do so. That could change - there are a number of people who want to see the code. Portions of the code appear incomplete, as if the program development was not yet finished. For example, it knows the offset needed to break the BSD fingerd, but doesn't know the correct offset for Sun's fingerd (which causes it to dump core); it also doesn't erase its tracks as cleverly as it might; and so on. The worm uses a variable called 'pleasequit' but doesn't correctly initialize it, so some folks added a module called _worm.o to the C library, which is produced from: int pleasequit = -1; the fact that this value is set to -1 will cause it to exit after one iteration. The close scrutiny of the code also turned up comments on the programmer's style. Verbatim from someone at MIT: From disassembling the code, it looks like the programmer is really anally retentive about checking return codes, and, in addition, prefers to use array indexing instead of pointers to walk through arrays. Anyone who looks at the binary will not see any embedded strings - they are XOR'ed with 81 (hex). That's how the shell commands are imbedded. The "obvious" passwords are stored with their high bit set. Although it spreads very fast, it is somewhat slowed down by the fact that it drives the load average up on the machine - this is due to all the encryptions going on, and the large number of incoming worms from other machines. [Initially, the fastest defense against the worm is is to create a directory called /usr/tmp/sh. The script that creates /usr/tmp/sh from one of the .o files checks to see if /usr/tmp/sh exists, but not to see if it's a directory. This fix is known as 'the condom'.] NOW WHAT? None of the ULowell machines were hit by the worm. When BBN staffers found their systems infected, they cut themselves off from all other hosts. Since our connection to the Internet is through BBN, we were cut off as well. Before we were cut off, I received mail about the sendmail problem and installed a patch to disable the feature the worm uses to get in through sendmail. I had made local modifications to fingerd which changed the offsets, so any attempt to scribble over the stack would probably have ended up in a core dump. Most Internet systems running 4.3BSD or SunOS have installed the necessary patches to close the holes and have rejoined the Internet. As you would expect, there is a renewed interest in system/network security, finding and plugging holes, and speculation over what will happen to the worm's creator. If you haven't read or watched the news, various log files have named the responsible person as Robert Morris Jr., a 23-year old doctoral student at Cornell. His father is head of the National Computer Security Center, the NSA's public effort in computer security, and has lectured widely on security aspects of UNIX. Associates of the student claim the worm was a 'mistake' - that he intended to unleash it but it was not supposed to move so quickly or spread so much. His goal (from what I understand) was to have a program 'live' within the Internet. If the reports that he intended it to spread slowly are true, then it's possible that the bytes sent to ernie.berkeley.edu were intended to monitor the spread of the worm. Some news reports mentioned that he panicked when, via some "monitoring mechanism" he saw how fast it had propagated. A source inside DEC reports that although the worm didn't make much progress there, it was sighted on several machines that wouldn't be on its normal propagation path, i.e. not gateways and not on the same subnet. These machines are not reachable from the outside. Morris was a summer intern at DEC in '87. He might have included names or addresses he remembered as targets for infesting hidden internal networks. Most of the DEC machines in question belong to the group he worked in. The final word has not been written - I don't think the FBI have even met with this guy yet. It will be interesting to see what happens.