Slashdot Log In
Virtualization Decreases Security
Posted by
kdawson
on Thu Oct 25, 2007 11:54 AM
from the more-chances-to-blow-it dept.
from the more-chances-to-blow-it dept.
ParaFan writes "In a fascinating story on KernelTrap, Theo de Raadt asserts that while virtualization can increase hardware utilization, it does not in any way improve security. In fact, he contends the exact opposite is true: 'You are absolutely deluded, if not stupid, if you think that a worldwide collection of software engineers who can't write operating systems or applications without security holes, can then turn around and suddenly write virtualization layers without security holes.' de Raadt argues that the lack of support for process isolation on x86 hardware combined with numerous bugs in the architecture are a formula for virtualization decreasing overall security, not increasing it."
Related Stories
Firehose:Virtualization Decreases Security by Anonymous Coward
This discussion has been archived.
No new comments can be posted.
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
Full
Abbreviated
Hidden
Loading... please wait.
Uh oh (Score:5, Funny)
Re:Uh oh (Score:4, Funny)
Parent
It's easy to defeat Theo's argument (Score:5, Insightful)
There doesn't need to be a flame war, because in this particular instance Theo's argument has a gaping hole in it. Consider the following two system architectures:
1) An ordinary multi-function Unix-type system which also runs a non-trivial component that is exposed to the world (all non-trivial components have bugs, as Theo is right to point out, and hence are attack vectors).
2) A machine running 2-guest virtualization, in which the non-trivial component runs in one guest, and the rest of the functions run in another.
Now consider what happens when the world-facing component gets compromised, and by one of many methods (because sysadmins are fallible) the attack gets promoted to root privilege. Security has failed in one guest, but has it failed in the other? Not necessarily, depending on whether the sysadmin has made repeated blunders and not just one. (Eg. a fool might be keeping ssh keys on the public-facing guest
In this scenario, the isolation created by virtualization has given the syadmin an additional bulkhead against his own fallibility, and that is worthwhile for security, not only for better hardware utilization. The partitioning of the application and O/S space has reduced the cross-section of software open to attack.
Theo's argument also doesn't bear scrutiny at the hypervisor level, because while an O/S in dom0 is just as fragile as the one in domU that runs an exposed application, the instance in the hypervisor isn't exposed to attack. Theo seems to miss the distinction between endpoint fallibility and fallibility in the conveyance and resourcing that is done by hypervisors. They're different.
I like Theo's hard stance on security, but on this issue he's handwaving.
Parent
Re:It's easy to defeat Theo's argument (Score:5, Insightful)
Chris Mattern
Parent
Re:Uh oh (Score:5, Insightful)
They consider him a brilliant man, and excellent programmer, and generous to let people download his code. They consider him a hero for taking on and beating the US government. They consider him a jerk. I've never heard anybody call him a leader of the Free Software Movement. I've never even heard his license-free software to be considered Free Software.
As an aside, many people call him a jerk for his style of writing information and documentation. I had to install a DNS server, and I found his you-must-be-a-moron-so-I-will-explain-everything-in-very-simple-terms documentation very informative, clear, and helpful. The security advantage is nice, but to me, tinydns' greatest advantage was the DJB's documentation.
Parent
Re:Uh oh (Score:5, Insightful)
After reading vitriolic posts by these two fools, RMS doesn't seem all that bad.
I've been on the fence about virtualization for a very long time now. Sure, it's quite convenient to install VMware, load up a guest OS, and tinker with new features. But to load up a server with multiple instances of the same operating system is ludicrous. It certainly doesn't scale well at all. And the marketing teams are incredibly good at making people believe that by installing their virtualization software, you'll suddenly have a bunch of "virtual" servers with the same capabilities as a single server. Sure, they all have the same capabilities from an OS standpoint, but performance isn't going to be anything close to a standalone server..
And as far as security goes, it's nonsense. Ok, so I install 5 copies of RHEL 5.0 on my virtual server. If the virtualization software itself is attacked and compromised, all 5 servers go down. If an OS level attack is successful, then all 5 virtual servers are likely vulnerable because it's an OS level attack. The only security "benefit" I can see is if a single virtual server is compromised through something like a web application. That application may not exist on the other virtual servers, so they're "safe".. However, once you get into that one server, DDoS attacks aren't far behind. At the very least, you'll take up resources and you can potentially impact the operation of the other virtual servers.
I'll stick with standalone servers for now.. At least until there's a better solution, of which I don't see one coming anytime soon...
Parent
Re:Uh oh (Score:5, Informative)
Parent
Re:Uh oh (Score:5, Insightful)
Performance will take a hit from the overhead involved, but availability should increase. Most server applications don't fully utilize the CPU anyway, so sacrificing some cycles to run the apps in a virtualized environment is not really a big deal. Where virtualization shines is availability. If a server is malfunctioning or overburdened, the virtualized environment can migrate to another server without the server clients knowing this has taken place (other than some latency caused by the migration). This is actually the coolest part of this technology.
I never thought about using virtual servers to increase security. Except for running windows within Mac OS X, I really don't see virtualization making anything more secure.
I think this is much ado about nothing. It is only here because Theo is getting upset...
Parent
Perhaps a Different Train of Thought (Score:5, Insightful)
You've been smoking something really mind altering, and I think you should share it.
x86 virtualization is about basically placing another nearly full kernel, full of new bugs, on top of a nasty x86 architecture which barely has correct page protection. Then running your operating system on the other side of this brand new pile of shit.
You are absolutely deluded, if not stupid, if you think that a worldwide collection of software engineers who can't write operating systems or applications without security holes, can then turn around and suddenly write virtualization layers without security holes.
You've seen something on the shelf, and it has all sorts of pretty colours, and you've bought it.
That's all x86 virtualization is.
However, his technical argument is
I'm going to point out some other things I know about coding. Although more lines of code usually means more bugs, this is not always the case. Correlation does not equal causation. It is correlated but only because the more lines of code, the more probability that more people contributed to the project which means it is highly probable one of them was a bad coder. Also, if you plan things out and follow a rigorous model, it is within your power to make very fully functional, very nice software.
My second point is a different way of looking at the problem. Let's take the naive approach of assuming a primary job of the operating system is to protect the user (and applications) from completely fouling things up in the hardware & memory realm. So it does an 'ok' job at this but, as Theo noted, some bugs still exist. Let's say it's something really bad like they don't stop programs from altering a very sensitive range of memory that is very vital to the correct execution of the operating system itself. Now, hypothetically, the virtualized layer on top of this would give coders a chance to catch this and correct it and protect the user from bringing down the operating system. In this way of looking at things you have two nets. Alone one lets many things pass through so you double it up and now you're catching more fish.
But my analogy is probably very flawed and I must confess I have coded neither of these pieces of software so I cannot confirm or deny this. I am quite shocked that Mr. de Raadt would react so abusively to a post where someone was merely saying that they 'appeared' to be receiving some amount of additional security from virtualization.
As for the very last comment Mr. de Raadt makes, I am confused. My employer uses virtualization on a mass scale to more effectively utilize hardware. I believe it has more uses than just bright shiny colors and wrapping--in fact I am interested in its potentials for hosting web OSs and other neat applications to users. It might not be the future like some people think it is but I think Mr. de Raadt was suffering a moment of frustration or dealing with irritable people when he authored this.
I do wish he were open to more ideas. The second you start to just outright dismiss all your options because they don't satisfy you on the surface you will find you are left with none and often miss the best.
Re:Perhaps a Different Train of Thought (Score:5, Informative)
Parent
Risk profiles (Score:5, Insightful)
Virtualization is no doubt a complex problem to get right, but it's only one problem. There is a relatively fixed set of hardware any virtualization system claims to support. A reasonably complete virtualization system can be frozen at some level of functionality. An operating system can not; it must, by nature, constantly evolve to new requirements. Hardware, in contrast, is relatively more stable.
Operating systems running on virtualized systems also have the advantages of operating systems running any fixed configuration. While not quite as consistent as a completely emulated environment, virtualization gets most of the benefits, under reasonable assumptions.
So, in short, virtualization has the same sort of benefits microkernels were supposed to provide, albeit with a much more heavyweight solution: smaller core that's easier to secure. Virtualization has been used in the mainframe community for years. Virtualization is an even stronger form of process isolation than what operating systems provide.
Virtualization is much more costly to run than a standard operating system process. This should be a clue that it probably provides stronger isolation guarantees, even if you don't buy the rest of the argument.
I think it's a specious argument, as usual, to claim that securing the virtualization layer is no harder or easier than securing an operating system. I think securing the virtualization layer is going to be much easier, because while the problem itself is complex, it's still less complex than a complete operating system is.
A better argument would have been to point out that guest operating systems running under virtualization are no less vulnerable to being compromised than those running on real hardware. But then that would point the finger at operating system vendors, not virtualization ones.
Useless (Score:4, Interesting)
Ugh.
Re:Useless (Score:5, Interesting)
VT-x and SVM provide paths for rootkits to integrate and hide. New rootkits like Blue Pill [bluepillproject.org] and Vitriol [theta44.org] utilize SVM and VT-x to virtualize the host platform and remain undetected and immune from removal. They're not widespread, but an attack vector exists, which implies the security concerns over them.
Makes sense to me.
Parent
I'm Not Sure I Buy His Analysis (Score:5, Interesting)
However, I see this more as if the virtualization layer actually sits under the OS layer, then the actual security for remote intrusion would be, first, Y/OS(X), THEN Y/V(X), where Y is the number of people with the knowledge to exploit each vulnerability. Thus, someone who wanted to exploit the system would both have to be capable of exploiting an OS vulnerability, and THEN also exploiting a virtualization vulnerability.
(And we're talking about remote usage, because we all know it's virtually impossible to protect a system from anyone who has direct access to the hardware.)
I understand that reality may not be quite as tidy, but it still seems like a virtualized system would be much more secure that a non-virtualized system, if only because the increased level of knowledge involved means a smaller number of hackers capable of exploiting both layers. What am I missing?
Theo rocks, as his usual! (Score:4, Funny)
Theo's pessimism and where it comes from. (Score:5, Insightful)
This doesn't mean that OpenBSD won't get some kind of virtualization support, it just means that he's being careful and conservative and letting other people be the pioneers. I think this is a good thing, on balance... you don't want to be pulling arrows out of your back because your secure OS decided to take you through unknown territory.
Yeh, he's got an emphatic way of putting things. You just gotta deal with it. Several years ago I asked him about stack protection and his response was eerily similar to this. A few years later OpenBSD enabled stack protection by default.
I think he's got a point, but he's comparing running separate computers to running separate OS instances on the same computer. If that's how you're using VMs, then yes, the resulting system is less secure overall... and for Windows that's often how VMs get used because Windows tends to make it unreasonably hard to run multiple instances of the same application on the same computer. If you're replacing less extreme isolation mechanisms on the same computer with VMs, though, then you're adding an extra layer of defence. Think of it as a hierarchy...
* Same application instance (eg, web server modules)
* Separate applications (running multiple instances of apache)
* User level separation (multiple accounts for the separate instances)
* File system separation (multiple chrooted instances of apache)
* OS-level separation (eg, FreeBSD jails and I think Solaris domains)
* Hardware-assisted software virtualization (VMware, Xen)
* Hardware virtualization (IBM VM "penguin farms")
* Separate physical computers
It might be argued that IBM's virtual machines should be lumped with virtualization, or that separate computers should be split from blades, and things like NAS and SAN complicate things, but you get the idea.
Theo's looking at the hierarchy starting at the bottom, and seeing a reduction in security. Other people are starting at the top, and seeing an increase in security. Both sides are correct, it depends on where you start.
Security == managing complexity (Score:5, Interesting)
Here's the first truth of security: your ability to secure a system is INVERSELY PROPORTIONAL to the size of the interface to that system. Every interface point is a potential attack vector, whether direct (an attacker can exploit the interface) or indirect (something outside your control is loaded at interface A, then an attacker at interface B causes A to exploit something). Most security products try to reduce the size of interfaces (e.g. a firewall limits the number of open ports, then further excludes some types of traffic from those ports).
Look at a general purpose operating system kernel. There are hundreds of system calls (direct attack vectors), hundreds more driver interfaces (indirect attack vectors - driver interfaces are privileged and thus drivers must be bug-free), a few thousand more configuration points (Windows Registery, Linux /sys and /proc trees). Add the libraries that make up the rest of the operating system, and the number of APIs has exploded to thousands, if not tens of thousands.
Now look at a hypervisor stack. The hypervisor::guest interface is the CPU instruction set (extremely well documented and easy to programatically verify, especially when 99% of instructions can be verified to have no side effects!). Much narrower interface than a general-purpose API. The driver::hypervisor interface is narrower too, since the hypervisor only uses a lower-level interface (e.g. Xen's block device interface, VMware's SCSI interface) that happens to be simpler and better documented. Configuration API is smaller, since it only needs to manage virtual machines, not every possible combination of user-level program and device.
It's the old microkernel / monolithic kernel debate all over again, where a hypervisor is a microkernel and a general-purpose OS is a monolithic kernel, and the performance loss is small enough that companies are using it in production today. Microkernel have advantages in being easier to secure, more robust in the face of bugs ... monolithic kernels are faster. Is the smaller API (and increased security) worth the loss in performance?
Here's some security thoughts, based on actual experience with virtualization bugs.
He's right, in theory (Score:5, Insightful)
Theo's expertise, and indeed that of the entire OpenBSD project, is in the realm of provably correct security. Virtualization adds yet another layer where something can go wrong. Sure there are and will be bugs. We're finding them and fixing them, just as we've always done. From an absolute security standpoint, Theo's right.
Of course, most businesses couldn't care less. Businesses don't view security as an absolute thing, because human factors make it generally impossible. Businesses view security as a risk, with associated probabilities and costs, worst-case scenarios, likely scenarios, mitigation strategies, and ultimately, diminishing marginal returns. For businesses using virtualization to consolidate systems, it generally reduces risk because it makes it easier to implement policies that mitigate human factors.
To be precise, virtualization *technology* decreases security, but virtualization *solutions* increase security, at least when done well, which is much more practical than the technical absolute of "done right".
[/disclosure]
Re:History teaches once again... (Score:4, Informative)
The same thing with Virtualization, each VM will not be completely secure and will have holes in it but spreading will be reduced because only a smaller portion of application will use that OS to virtualize.
I don't think that analogy applies here. I think TA's point is that the hypervisor itself may not be any more secure than the OSes it virtualizes. So now you're hypervisor OR the OS it's running may get hacked.
Parent
Re:History teaches once again... (Score:4, Insightful)
Parent
Re:History teaches once again... (Score:5, Informative)
His position has many facets. As I understand it:
* programmers make buggy code, and now programmers are programming virtual hardware
* the hardware they are emulating (PC architecture) is a nightmare, they have to do crazy, unsafe crap to implement it.
* application flaws in the VM can compromise the guest OS.
* OS flaws in the guest OS can potentially compromise the host OS.
* virtualizing hardware is inherently less secure than the physical segmentation of using actual, separate machines, so when you consolidate many machines onto a VM system you have a net loss in security.
Parent
Re:History teaches once again... (Score:5, Insightful)
Actually Theo's argument was that software engineers can't write an OS without security holes, therefore they can't write a hypervisor without security holes.
The argument is, of course, full of shit. The hypervisor in question, Xen, is 50,000 lines of code. Compare this to the linux kernel which is about 6 million lines of code or Vista which is said to be 10s of millions. Theo also drags out his favorite attack about page protection. He is known for attacking a "vulnerability" in a C2D code segment limit/page accessed issue (AI90) as being "assuredly exploitable" in OSes other than OpenBSD, even though nobody has been able to propose a way to exploit it.
The problem with Theo attacking things is that he is so well respected in BSD-land that his word is taken for granted. Sometimes he gets it wrong, but unless someone equally high up wants to spend the time to rebut his ranting (a lot of work for no gain) everybody accepts what he says.
Parent
Re:History teaches once again... (Score:5, Insightful)
There's still a lesson in diversity and computer security to be learned here. But it includes the harsher lesson that human leaders often don't care about the necessity for diversity and the cost to security (and thus the IT department), and can impose a homogeneity that is even worse than an IT department that just didn't consider diversity to be important.
Parent
Re:History teaches once again... (Score:5, Funny)
Indeed. Implementing proper security is no small potatoes.
Parent
Re:Welcome to the rest of the IT world, Theo! (Score:5, Insightful)
Parent