| ==Phrack Inc.== | |
| Volume Two, Issue 23, File 8 of 12 | |
| ____________________________________ | |
| | | | |
| | Getting Serious About VMS Hacking | | |
| | | | |
| | by VAXbusters International | | |
| | | | |
| | January 1989 | | |
| |____________________________________| | |
| The VAX/VMS operating system is said to be one of the most secure systems | |
| currently available. It has been massively extended in the past to provide | |
| features which can help system managers getting their machines locked up to | |
| abusers and to trace back any attempts to indiscriminate system security. As | |
| such, it is not easy getting into VMS machines now without having insider | |
| information, and it's even harder to stay in. | |
| The following article describes some of the internals which make up the VMS | |
| security features, and tries to give hints what to do to remain undiscovered. | |
| The reader should be familiar with the VMS system from the programmer's point | |
| of view. | |
| Some of the things mentioned are closely related to the internal workings of | |
| the VAX/VMS operating system. All descriptions are held as general as | |
| possible. It is tried to point out where weak points in the system are | |
| located, not to give step-by-step instructions on how to hack VMS machines. | |
| The main reason for this is, that it is very hard to remain undiscovered in a | |
| VMS system without having good knowledge of the whole system. This knowledge | |
| is only aquirable by experience. | |
| To use some of the techniques described herein, some literature is recommended: | |
| "The VAX Architecture Handbook," published by DEC. This book describes | |
| the VAX processor, it's instruction set and it's hardware. It is a good | |
| book to have on your desk, since it costs nothing (just go to your local | |
| DEC store and ask for it) and is only in paperback format. | |
| "MACRO and Instruction Set," part of the VMS documentation kit. This is | |
| needed only if you want to program bigger things in MACRO. It's | |
| recommended reading, but you don't need to have it on your own normally. | |
| "VAX/VMS Internals and Data Structures" by L.Kenah and S.Bate. This is | |
| the bible for VMS hackers. It describes the inner workings of the system | |
| as well as most of the data structures used within the kernel. The | |
| Version published always is one version number behind the current VMS | |
| release, but as the VAX architecture doesn't change, it is the best source | |
| on a description how the system works. After you've read and understood | |
| this book, the VAX won't look more mysterious than your C64. You can | |
| order this book from DEC, the order number for the V3.0 version of the | |
| book is EY-00014-DP. The major drawback is the price, which is around | |
| $70-$100. | |
| A good source of information naturally is the source code of the VMS system. | |
| The easiest way to snoop around in it is to get the microfiche set, which is | |
| delivered by DEC to all bigger customers of the system. The major disadvantage | |
| is that you need a fiche reader to use it. The fiche is needed if | |
| modifications to the system code are intended, unless you plan to disassemble | |
| everything you need. The VMS system is written in BLISS-32 and FORTRAN. BLISS | |
| is quite readable, but it might be worthwhile having a FORTRAN hacker around if | |
| you intend to do patch any of the programs implemented in FORTRAN. The source | |
| fiche always contains the current release, so it's useful to check if the | |
| information in "Internals and Data Structures" is still valid. | |
| Hacker's Tools | |
| ~~~~~~~~~~~~~~ | |
| There are several programs which are useful when snooping around on a VMS | |
| system. | |
| The most important utility might be the System Dump Analyzer (SDA), which is | |
| started with the command ANALYZE/SYSTEM. Originally, SDA was developed to | |
| analyze system crash dumps, which are created every time the machine crashes in | |
| a 'controlled' manner (bugcheck or opcrash). SDA can also be used to analyze | |
| the running system, which is the more useful function. A process which wants | |
| to run SDA needs the CMKRNL privilege. With SDA, you can examine any process | |
| and find out about accessed files and devices, contents of virtual memory (like | |
| typeahead and recall buffers), process status and more. SDA is a watching | |
| tool, so you normally can't destroy anything with it. | |
| Another helpful tool is the PATCH utility, called up by the command PATCH. As | |
| VMS is distributed in a binary-only fashion, system updates are normally | |
| distributed as patches to binaries. PATCHES can be entered as assembler | |
| statements directly. Combined with the source fiche, PATCH is a powerful tool | |
| for your modifications and improvements to the VMS operating system. | |
| Privileges | |
| ~~~~~~~~~~ | |
| To do interesting things on the VMS system, you normally need privileges. The | |
| following lists describes some of the privileges which are useful in the | |
| onliner's daily life. | |
| CMKRNL | |
| CMEXEC These two privileges enable a user to execute arbitrary routines with | |
| KERNEL and EXECUTIVE access mode. These privileges are needed when one | |
| plans to access kernel data structures directly. CMKRNL is the most | |
| powerful privilege available, everything which is protected can be | |
| accessed utilizing it. | |
| SYSPRV A process which holds this privilege can access objects via the system | |
| protection. A process holding the this privilege has the same access | |
| rights as a process running under a system UIC. | |
| SHARE This allows a process to assign channels to nonshareable devices which | |
| already have channels assigned to them. This can be used to prevent | |
| terminal hangups and to assign channels to system mailboxes. | |
| Process States And The Process Control Block | |
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ | |
| When you get into kernel hacking, you should pay special attention to the field | |
| PCB$L_STS. This field tells about the process status. Interesting bits are | |
| PCB$V_DELPEN, PCB$V_NOACNT and PCB$V_BATCH. There can be achieved astonishing | |
| effects by setting these bits. | |
| Hideout | |
| ~~~~~~~ | |
| A nice possibility to have is to be unseen by a system manager. There are many | |
| ways to get invisible to SHOW USERS, but hiding from SHOW SYSTEM is another | |
| story, as it doesn't even use standard system calls to get a list of the | |
| currently running processes. And in fact, hiding from SDA is even harder, | |
| since it directly peeks kernel data structures. Anyway, being invisible to | |
| SHOW USERS is useful on small systems, where one user more could ring the alarm | |
| bell of the system operator. | |
| One possibility to do this is to become a subprocess of some non-interactive | |
| job (like a BATCH or NETWORK process). The other way is to patch the PCB to | |
| become a BATCH process or to delete the terminal name (which makes SHOW USERS | |
| think you are non-interactive as well). Patching the PCB has a disadvantage: | |
| The system global variable SYS$GW_IJOBCNT which contains the number of | |
| interactive users must be directly decremented before you hide, and MUST be | |
| incremented before you log out. | |
| If you forget this, the interactive job count will be wrong. If it becomes | |
| negative, strange effects will show up, which will confuse every system | |
| manager. | |
| Accounting And Audits | |
| ~~~~~~~~~~~~~~~~~~~~~ | |
| The most nasty thing about VMS since release 4.2 is the security auditing | |
| feature. It enables the system manager to log almost every security relevant | |
| event he desires. For example, access to files, login failures and | |
| modification user authorization data base can all be monitored, logged and | |
| written to the system printer. The first thing to find out in a new, unknown | |
| system is the awareness of the system management. The status of the accounting | |
| system is easily determinable by the command SHOW ACCOUNTING. Normally, | |
| everything except IMAGE accounting is enabled. When IMAGE accounting is also | |
| enabled, this is the first hint to be careful. The second thing to check out | |
| is the status of the security auditing system. You need the SECURITY privilege | |
| to execute the command SHOW AUDIT. | |
| If no audits are enabled, and image accounting is not turned on, the system | |
| normally is not set up to be especially secure. Such systems are the right | |
| playground for a system hacker, since one doesn't have to be as careful as one | |
| has to be on a correctly managed system. | |
| Accounting | |
| ~~~~~~~~~~ | |
| The main intention for running accounting on a system is the need to charge | |
| users for resources (cpu time, printer usage etc.) they use on the machine. On | |
| the other hand, accounting can be very useful to track down invaders. Luckily, | |
| accounting information is being logged in the normal file system, and as such | |
| one can edit out information which isn't supposed to be seen by sneaky eyes. | |
| The most important utility to handle accounting files is, naturally, the | |
| ACCOUNTING utility. It has options to collect information which is stored in | |
| accounting files, print it in a human readable manner, and, most importantly, | |
| edit accounting files. That is, you can edit all information out of an | |
| accounting file which you don't want to appear in reports anymore. The | |
| important qualifier to the ACCOUNTING command is /BINARY. | |
| File Access Dates | |
| ~~~~~~~~~~~~~~~~~ | |
| One way for system managers to discover unwanted guests is to look out for | |
| modified system files. Fortunately, there are ways to modify the modification | |
| dates in a file's header. This can be done with RMS system calls, but there is | |
| no easy way to do that with pure DCL. There are several utilities to do this | |
| kind of things in the public domain, so look out in the DECUS catalog. | |
| OPCOM | |
| ~~~~~ | |
| OPCOM is a process which logs system and security relevant events (like tape | |
| and disk mount transactions, security auditing messages etc.). OPCOM receives | |
| messages via a mailbox device, formats them, logs the event in the operator | |
| logfile (SYS$MANAGER:OPERATOR.LOG) and notifies all operators. Additionally, | |
| it sends all messages to it's standard output, which normally is the system | |
| console device _OPA0:. When OPCOM is started, one message is sent to the | |
| standard output announcing that the operator logfile has been initialized. | |
| Thus, it's not recommended to kill OPCOM to remain undiscovered, since the | |
| system manager most likely will get suspicious if the operator logfile has been | |
| initialized without an obvious reason. The elegant solution to suspend OPCOM, | |
| for the time where no operator messages shall come through. While OPCOM is | |
| suspended, all messages will be buffered in the mailbox device, where every | |
| process with sufficient privilege can read them out, thus avoiding that OPCOM | |
| reads those messages after it is restarted. | |
| There is one problem with this solution though: OPCOM always has a read | |
| pending on that mailbox, and this read will be there even if the OPCOM process | |
| is suspended. Unless you're heavily into kernel hacking, there is no way to | |
| get rid of this read request. As such, the easy solution is to generate an | |
| unsuspicious operator message as soon as OPCOM is suspended. Afterwards, your | |
| own process (which can be a DCL procedure) reads all subsequent messages off | |
| the OPCOM mailbox until you feel save enough to have OPCOM resume it's work. By | |
| the way, the OPCOM message mailbox is temporary and has no logical name | |
| assigned to it. You'll need SDA to get information about the device name. | |
| Command Procedures | |
| ~~~~~~~~~~~~~~~~~~ | |
| Timely, you'll need DCL procedures to have some routine work done | |
| automatically. It is important not to have strange command procedures lying | |
| around on a foreign system, since they can be easily read by system managers. | |
| Fortunately, a command file may be deleted while someone is executing it. It | |
| is good practice to do so, utilizing the lexical function F$ENVIRONMENT. If | |
| you need access to the command file itself from the running procedure, just | |
| assign a channel to it with OPEN. | |
| Piggy-Backing | |
| ~~~~~~~~~~~~~ | |
| It's not normally a good idea to add new, possibly privileged accounts to a | |
| foreign system. The better approach is to to use accounts which have been | |
| unused for some months and to hide privileged programs or piggybacks which gain | |
| privilege to the caller by some mechanism. A piggyback is a piece of code | |
| which is added to a privileged system program, and which gives privileges | |
| and/or special capabilities to callers which have some kind of speciality (like | |
| a special process name, for example). Be careful not to change file sizes and | |
| dates, since this makes people suspicious. | |
| Conclusion | |
| ~~~~~~~~~~ | |
| This file just tries to give an impression how interesting VMS kernel hacking | |
| can be, and what possibilities there are. It of course is not complete, and | |
| many details have been left out. Hopefully, it has been useful and/or | |
| interesting lecture. | |
| (C)opyright 1989 by the VAXBusters International. | |
| You may give around this work as long as you don't pretend you wrote it. | |
| _______________________________________________________________________________ | |