A Look at BSD Rootkits

blackbearnh writes "Windows has a reputation for being easily exploited by rootkits, but just because you're using Linux or BSD doesn't mean you're safe from infection. In an interview on O'Reilly's ONLamp site, Joseph Kong (author of Designing BSD Rootkits ), talks about how to build and defend against Rootkits under BSD. 'I know a lot of people who refer to rootkits and rootkit-detectors as being in a big game of cat and mouse. However, it's really more like follow the leader — with rootkit authors always being the leader. Kind of grim, but that's really how it is. Until someone reveals how a specific (or certain class of) rootkit works, nobody thinks about protecting that part of the system. And when they do, the rootkit authors just find a way around it. This is what I meant earlier when I said rootkit hunting is hard — as you really have to validate the integrity of the entire system.'"

Comments

Illegal Book?

[Numbah One]

is this book illegal in Germany?

Run your system off of CD

[Nom du Keyboard]

Run your system off of a bootable CD. A little slower to boot, but once it's in memory...

Re:Run your system off of CD

[AKAImBatman]

And what do you do if you need your CD-ROM drive back? Also, some forms of malware install at the BIOS/hypervisor level. You can't even *detect* that from inside the OS! Some malware dynamically patches the kernel at runtime. So if you access settings on the hard disk or flash drive at all (and how could you not?) the malware can simply install itself after you boot.

The big question is how this malware gets there in the first place. The "towards verifiable systems" presentation linked to from the article listed such options as users who run attachments (merde), malicious employees who intentionally install kits (!), and use of a stolen password. These are all problems that can't be stopped, only mitigated. A malicious employee with physical access to a machine has everything they need, and you can't stop them. You can mitigate the problem by checking for things like tampering with the case and BIOS resets (to clear the password), but these are not foolproof solutions. Same with a stolen password. If you don't know its stolen, a window will always exist in which it can be used.

It *is* possible to build technology that does not suffer from trivial problems like buffer overflows, but you can't stop someone who has clear access to a machine. Authority is authority, and there will always be methods to steal and/or abuse that authority over a machine.

Re:Pardon me, but I'm not surprised

[jalet]

> based on my penetration testing and signature analysis.
> E. Wyatt Tomlinson

OK, so we finally analyzed your signature above, and now we would like to proceed with the penetration testing of you.

Please advise.

Re:*BSD developers leave behind trail of corpses

[BosstonesOwn]

Fact: DragonflyBSD, yet another offshoot of the beleaguered FreeBSD "project", is already collapsing under the weight of internal power struggles and in-fighting. "They haven't done a single decent release," notes Mark Baron, an industry watcher and columnist. "Their mailing lists read like an online version of a Jerry Springer episode, complete with food fights, swearing, name-calling, and chair-throwing." Netcraft reports that DragonflyBSD is run on exactly 0% of internet servers.

Since when did Steve Balmer start working at dragonflyBSD ?

rootkit detectors

[Random Walk]

He says that the only detectors he is aware of are chrootkit and Rootkit Hunter. As he correctly points out, these are signature based - they look e.g. for specific files in userspace (chrootkit also does a generic check for hidden processes). However, the samhain integrity checker also has two different modules to check for kernel-mode rootkits. First, it can check the kernel syscall table to detect syscall redirection within the kernel (on FreeBSD and Linux), and second, it can detect hidden processes (basically in the same way as chrootkit). Also, if checking file integrity, samhain compares the number of hardlinks of a directory with the number of subdirectories to detect hidden subdirectories.

There is no fundamental reason

[jd]

...why rootkit writers should ever be ahead of the game. First off, installing code into the kernel should be regulated such that rogue code can't gain all of the necessary rights. Yes, that means legit modules can't auto-install unless you have some multi-key system, such as having modules individually pre-approved, signed by the builder and countersigned by the system admin. Never said it would make your life easier. In fact, making things too easy for you makes it too easy for those you don't want to give access rights to.

Second, all modules should be exposed and visible at all times, along with the connections between modules. This is not hard. You have a hypervisor which doesn't do virtualization, rather it simply queries the OS on what the OS is currently doing. Why have something below the kernel? Because then it cannot be written to by the kernel. It is outside the kernel's memory space. It cannot be trapped, descheduled, replaced, modified or even sniffed by the OS or anything the OS is running. This gives you a covert channel to a monitoring tool that the rootkit cannot disable or interfere with. The OS merely needs to provide some sort of API that the hypervisor can use to supervise the kernel's activities and intercept the information needed to establish the covert channel with the monitoring tool.

Too complex? Ok, then how about a third solution: Have the compiler randomize the kernel's ABI. Totally. Absolutely zero predictability in how parameters are passed. The compiler just needs to make sure ALL calls to a specific function call that function the same way, whether the object is linked in or compiled as a module. It also has to make sure that out-of-tree builds also compile to the correct ordering for each function. Binary-only modules would need to be supplied with the source code for a glue layer that can map the binary's ABI with that of the kernel. Any unauthorized module will make calls that violate the ABI and therefore cannot stay running. They might crash the kernel, but that could be better than unauthorized and uncontrolled activity which might do far more damage.

Last, but not least, the kernel could support trusted computing modules and mandatory access controls for memory. It's the least secure, in some ways, but Trusted IRIX is pretty damn secure and I'd have some measure of faith in any integrity system that came close in design. (SGI released "Open B1" - OB1 - which was some of the code used to make IRIX as hard as it was. Dunno if anybody looked through the code for ideas - it wouldn't have mapped directly into Linux, but ideas are ideas and are OS-independent. Besides, OpenBSD at the time had no mandatory access control support. I don't know how far the other BSDs have got - I know there's some MAC and/or RBAC support in some, but are we talking basics or B1-equiv?)

Maybe a rootkit author could bypass all of these, but I doubt - seriously doubt - that it would be a trivial weekend exercise to bypass Trusted Computing or strong authentication/validation mechanisms.

Re:There is no fundamental reason

[AKAImBatman]

Maybe a rootkit author could bypass all of these, but I doubt - seriously doubt - that it would be a trivial weekend exercise to bypass Trusted Computing or strong authentication/validation mechanisms.

Step 1: Analyze NVidia or ATI graphics driver for buffer overflows or similar security issues.

Step 2: Construct an OpenGL call to exploit the issue and create an easy access point into the kernel.

Step 3: Use new access point to patch the kernel or BIOS code.

Step 4: Close the doors and clean up the mess so that there is no evidence of tampering. Just a regular kernel running regular modules and processes. No one knows that the kernel has actually been modified.

Step 5: ??? (Contact foreign terrorists? Skim partial pennies into a swiss bank account? Use for DDoS operations?)

Step 6: Profit!

Note that this potentially works for modules other than graphics modules. It's just that they're the most complex and therefore easier to exploit.

Have the compiler randomize the kernel's ABI. Totally. Absolutely zero predictability in how parameters are passed. The compiler just needs to make sure ALL calls to a specific function call that function the same way, whether the object is linked in or compiled as a module.

Two problems:

1. This would make binary modules impossible.

2. The current ABI must be documented in a machine-readable form somewhere on the system. The rootkit installer can modify the patch before installing it. Worst case, it can compile source code into a pristine binary that is compatible. (Since you'd be required to have a compiler on your system.)

bogus remarks

[Anonymous Coward]

And what do you do if you need your CD-ROM drive back?

Pardon me? Last time I checked I could pass some "toram" parameter to a lot of Live CDs, making the system run perfectly fine, entirely in memory, on my old P4 / 1 GB of ram. No problem to use other CD/DVDs.

Also, some forms of malware install at the BIOS/hypervisor level.

BIOS and hypervisor are two very different things. I seriously doubt that, today, a BIOS malware could be sufficiently advanced to act as a real root-kit. That is, a BIOS malware will not even be able to fool anti-virus running from inside the system. There simply ain't enough room in the BIOS to code such a beast. And you explain me how you remotely install a BIOS on a system that requires changing a jumper before you can flash the BIOS.

That blue pill paper was sensationalism at its worst. An "hypervisor rootkit" should provide a system capable of itself allowing another hypervisor to run. If it's not the case, it's super easy to detect... And even to counter: simply install another hypervisor, like the very fine Xen I'm running now and the hypervisor root-kit can't do anything.

Now another possibility: an hypervisor rootkit allowing another hypervisor to take place: the system would be so slow that the "timing attack" (an attack that has be proven to detect 100% of "hypervisor rootkits", should they ever come to exist) would simply be the user seeing its system so slow that it's clear that something fishy is going on.

Remember that you were replying to someone talking about running a system of a live CD. If the system has no hard disk, explain me where your hypothetical, urban legendary, hypervisor rootkit would reside? I seriously hope you're not implying the BIOS hold enough room to contain an hypervisor rootkit (come take a look at an hypervisor like Xen to see what I'm talking about). Not to mention that should the system have an hard disk and the hypervisor be prevent on the hard disk, I hardly see how a system configured to boot from the CD first would launch your urban legendary hypervisor...

You can't even *detect* that from inside the OS!

Anyone relying on scan ran from inside the OS to detect malware is a fool. An anti-virus running from inside the OS can find some malware and can prevent some kind of infections but thinking that because an anti-virus running from inside the OS reports nothing is a proof that there's no malware present on the system is a fool.

Btw I'm running an hypervisor on several machines, I've got "snort" behind a physically passive tap on my LAN, etc. I think the GP's advice of running a system of a Live CD is a very good advice (even tough you still can get infected... Good luck for malware to persist on the machine) while I think your answer is completely offtopic.

Re:bogus remarks

[AKAImBatman]

Last time I checked I could pass some "toram" parameter to a lot of Live CDs, making the system run perfectly fine, entirely in memory, on my old P4 / 1 GB of ram.

This is a possibility, but you're assuming that the system contains enough RAM to store all the necessary applications and datasets for the operation of the computer. Your anecdote does not prove that every machine can afford to load a complete OS into memory.

I seriously doubt that, today, a BIOS malware could be sufficiently advanced to act as a real root-kit.

Like it or not, security researchers consider it a real threat.

And you explain me how you remotely install a BIOS on a system that requires changing a jumper before you can flash the BIOS.

If you have a physical block in place, then one would think that you should be safe. Not all systems have this jumper, or have it set to prevent flashing by default. Also, an attacker with physical access could change the jumper setting. (See my original post above.)

Remember that you were replying to someone talking about running a system of a live CD. If the system has no hard disk, explain me where your hypothetical, urban legendary, hypervisor rootkit would reside?

If you were paying attention, I addressed that issue. If the computer stores settings anywhere (either a hard drive OR removable flash drive), then it is vulnerable. And let's be honest. How many users are going to create a new system layout and reburn it every time they want to change their system? Unless we're talking about an appliance device, not many.

Re:bogus remarks

[AKAImBatman]

If the system has no hard disk, explain me where your hypothetical, urban legendary, hypervisor rootkit would reside? I seriously hope you're not implying the BIOS hold enough room to contain an hypervisor rootkit (come take a look at an hypervisor like Xen to see what I'm talking about).

I just spent a few minutes reading this paper from the same fellow who introduced BIOS rootkits. It's quite interesting:

Many PCI cards contain an expansion ROM that holds additional code required to initialise the card during execution of the system BIOS. This code is also responsible for carrying out the device-specific self-test and hooking required interrupts. The presence of an expansion ROM is determined via the Expansion ROM Base Address Register within the PCI function's Configuration Header.

It is worth noting that the expansion ROM does not necessarily hold x86 code nor does it have to contain a single ROM image. The code type field within the ROM data structure within the image specifies the presence of x86 code or OpenBoot interpretive code (documented in the Open Firmware standard).

The expansion ROM is stored on either an EPROM, or more commonly on an EEPROM. EPROMs require that the chip is removed from the card and erased via exposing it to strong ultraviolet light before it can be reprogrammed. EEPROMs, however, can be erased electrically, in-circuit, thus the card need not be removed from the system and can be re-flashed from the operating system.

In order to perform this, the user must have the SeTcbPrivilege and call the undocumented Native API function, NtSetInformationProcess with a process information class of ProcessUserModeIOPL. Once the user can perform unrestricted I/O, they can potentially re-flash the card without having to load a driver.

This raises the possibility of (1) a remote attack that yields LocalSystem privilege (such as the server service vulnerability patched in update MS06-040) being used to deploy a malicious expansion ROM, (2) a browser exploit, that, if the user is running under the administrative context, obtains SeTcbPrivilege and re-flashes a card.

The paper goes on to explain the *exact* steps necessary to implement such a rootkit. Ouch.

Once you're penetrated you're ****ed.

[argent]

Basically, once someone has gotten their code running on your system, they can do anything they want, and they can pretty much keep you from noticing that they're there. If you go looking for them, though, odds are you'll find them... but who's going to go looking?

There's no magical difference between "rootkits" and any other trick for hiding code in a system... it doesn't matter if it's a "virus", or a "rootkit" or even a "polymorphic perverse passive-agressive viral-enhanced trojan rootkit" (or whatever the cool terminology of the week is), the trick to hiding is to change the things you know the rootkit detectors or antivirus software is looking for so they look right. The trick to finding them is to look in more places, and look in ways that they haven't thought of covering up. But the real trick is keeping them out in the first place.

Security is like sex... once you're penetrated you're ****ed. If the basic software is designed to that when implemented as documented there's no mechanism for an attacker to use, then you're in pretty good shape. At least, you will be able to fix any holes that DO show up without breaking working software. And that's the main disadvantage Windows has... there's just too much everyday software and important APIs that are inherently insecure. Even when implemented as documented, there's attacks ... which is why they have all those security dialogs: those dialogs come down to "this program is about to do something that might be stupid, is that OK?".

At the very least, you need to cut that down to "you just asked to do something that might be stupid, do you mean it?".

Re:Once you're penetrated you're ****ed.

[turing_m]

"Security is like sex... once you're penetrated you're ****ed."

I think a car analogy would work better here... at least cars are something most people here have a passing familiarity with.