Slashdot Mirror
Device Drivers Filled with Flaws, Pose Risk
Gary W. Longsine writes "Security Focus describes device drivers as an untapped source of buffer overflows, posing substantial risk not typically considered as part of a standard risk assessment. The security risks of device drivers on both Windows and Linux, including network (remotely exploitable) and hardware drivers (typically only locally exploitable) are discussed in the article. I've noticed that software you wouldn't expect sometimes installs a device driver component. I can understand this as a component of an antivirus or host based firewall, but it seems to be an oddly common design pattern on Windows, which clearly poses substantial risk."
Could someone give me examples of this? Most software I use doesn't do this. It seems more of a design pattern for DRM stuff (like DVD audio).
Sending a modem user a ping with +++ATH0 embedded. As soon as it was returned, those modems with defective modem drivers that didn't filter anything would hang up. Quick simple DoS.
Surprisingly it still works on some systems more than 18 years after I first tried it.
This leads to many problems like stuff found recently in almost all Computer Associates eTrust Antivirus products. Because Zonealarm licenced the same software, they were affected, too.
This is just one example of many :
So many well known enterprice Antivurs/Firewall companys create drivers that lead to security flaws and it is not limited to Windows....
Video games' copy protection systems install device drivers like crazy to try to prevent CD-ROM emulators and such. Others install drivers to prevent cheating. When they do this, they often mess up the system involved and leave the system vulnerable to attack.
For example, a few months ago, the nProtect anti-cheat system, which installs device drivers, had a buffer overflow in it that allowed local privilege escalation.
Melissa
"Screw Sun, cross-platform will never work. Let's move on and steal the Java language." - Visual J++ Product Manager
seems these are almost everywhere these days. and with all the odd keys a lot of them Do need their own custom drivers for the extra keys and knobs and dials etc.
:)
whatever happend to the good old days when an IBM model M was all you needed
XML - A clever joke would be here if
To cite poor design as a source of security vulnerability is to state the obvious. We spend so many man hours fixing problems that didn't have to exist in the first place, that we cannot address the problems that came inevitably over the course of implementation of software packages and protocols.
The Crimson Dragon
let's say there is a driver and it allows a buffer overrun to execute some attacker's code. Well to get to the driver the attacker has to first go through a user application. So there is a problem when the combination user application/device driver both have a problem validating the same input. I am not saying this is impossible, but it would be more unlikely - there must be a great coincidence at work here. Besides normally user applications do not pass user input directly to the device drivers. The user applications translate input from user form to some implementation specific device driver input. So more likely than not there is a layer of input transformation between the user and device driver.
Now to go around this problem the attacker may need to get the user to execute some code on the machine and this could mean that if the code is executed - even on a Linux box for example, and the code exploits a device driver flaw, due to the monolythic kernel structure of Linux it is in principle possible to execute code that will say change user privileges to admin level. I guess this would be much more difficult with a microkernel approach like what Hurd is supposed to be, because even device drivers will run in managed memory mode.
You can't handle the truth.
Not only device drivers! :-)
The Tao of math: The numbers you can count are not the real numbers.
Well, ATI's drivers have always been nasty. Now I can call them "viral"? :)
1. Linus didn't say that, Raymond did. 2. By your own analysis, all famous open-source projects should be bug-free, right? Like Firefox, right?
Stop drinking the kool-aid. Open source is not a panacea for all software development problems, and Raymond made a lot of sweeping generalities in the book you're quoting, many which make for great sound bites but are absolutely irrelevant.
Fool.
Drivers that come with the OS are still drivers.
Who told the god damn noobies about slashdot?
I'll do the stupid thing first and then you shy people follow...
I've always tried to buy hardware which is supported by default in Windows - since XP-SP2 added a bunch of new drivers this has got a lot easier.
My reasons were so that a reinstall is a simpler affair, but it appears I may have been reaping security benefits too...
I have been a user for about 10 years. This ends Feb 2014. The site's been ruined. I'm off. Dice, FU
An individual instance of a given buffer overflow exploit in a device driver in and of itself is not really a bigger risk on Windows. It just seems to be a more common design pattern on Windows systems, thus creating more opportunities for exploit. (Several fine examples of questionable use of device drivers, and some associated known vulnerabilities are discussed by others here).
The referenced article at Security Focus points out that inspection of device drivers in Linux revealed similar defects in device drivers.
Device drivers are more interesting than user land software because they run in kernel space, allowing the exploit to be immediately useful to perform nasty things like install rootkits and trojans, log keystrokes, etc.
If you mod me down, I shall become more powerful than you could possibly imagine.
oh god no!
I tried that once. it suggested an update for my network card and pretty much fucked my system.
never again. never.
One more reason to help groups trying to get documentation in order write their own drivers. Manufacturers seem more concerned with slowing down their rivals than with growing their customer base (for free!). Consider OpenBSD's recent problems with Adaptec.
-- "At Microsoft, quality is job 1.1" -- PC Magazine, Nov. 1994
The real cause of most reboots are attempts to replace active user-mode executables (EXE or DLL). Executable files cannot be replaced while they're running. This makes it practically impossible to update system DLL's without a reboot, since they're going to running in some process all the time.
Yup. This is a major design flaw in the Windows kernel that dates way back. It's a significant part of the reason that you don't have to reboot Linux for anything other than a new kernel, but Windows frequently needs to be rebooted for user-level application installations.
It's on my list of "stupid design decisions made in Windows" (where "Windows" == NT family, not 9x family).
Other goodies are:
* "pageable kernel memory pools" (sounds like a good idea, but significantly increases complexity of writing kernel code and ease of introducing lockup bugs) over Linux's approach of just unloading modules
* Microsoft's decision to not support "real" links, just shortcuts, in their user-mode software.
* Allowing application software to "block" a shutdown.
* Not allowing Windows to run without VM.
* Not designing Windows to be able to run off of read-only media.
* Godawful shell (not fundamental to the OS, and hopefully will fixed in Longhorn) and virtual terminal, to the point where most Windows users hate the terminal.
* Bad VM algorithms. I don't know what they use, but try running low on memory on a Windows box and the system becomes bloody unusable.
(From a developer standpoint)
* Poor sockets implementation (which is still the only reasonably portable networking API under Windows -- even IOCP lacks a IOCP-able connect() up until WinXP) with no way to kick select() awake, no local-domain sockets and lots of other flaws and irritations that have to be worked around by the Windows sockets programmer.
* Never precisely specifying API behavior -- MSDN is more of a tutorial or basic user guide to the API than a true specification. Look at a Linux man page and you generally have pretty strong guarantees on the behavior of the function provided, and that isn't even the specification (those which the function conforms to are listed and you can read a hard specification of guaranteed behavior).
Any program relying on (nontrivial) preemptive multithreading will be buggy.
Why do we even need device drivers at all? I've worked on (used, administered) two different kinds of major operating systems (and a couple more smaller ones) that did not use device drivers at all. The answer to thise question reflects a condition that those two major OSes did not have to deal with: lack of standardized hardware.
The original IBM System 360, released in the 1960s, effectively had relatively standardized hardware. That was because IBM made all the hardware. When other manufacturers eventually made their own hardware, they were forced to make that hardware compatible. A manufacturer of a disk drive had to make it accept every hardware command that IBM's own models accepted, or it would not work. No provision existed in the operating systems for these machine to install or load a special device driver, short of modifying the source directly (which was all in assembly code for the mainframe CPU architecture).
I/O operations in the original System 360, and to a great extend in the 370 and 390 that followed, is quite uniform compared to the PC architecture. Although IBM popularized this architecture, it was actually the design of the 8088 CPU that caused things to be quite non-uniform due to it's lack of any I/O architecture (it only had a simplistic in/out CPU instruction set, which effectively functioned like fetch/store instructions in a private address space). This meant every peripheral (like a serial port) had to operate its own way. Microsoft's DOS operating system furthered the dependency on device drivers being added by making it relatively easy to do. So by combining an architecture that was very poor at I/O, absent of an I/O standard, and an OS that made discrete device drivers easy, we have this become dependent on this.
A computer architecture could still be built that includes a standardized I/O infrastructure (e.g. all devices interface the same way) and standardized I/O command set (e.g. all operations of the same class operate identically), and would not need individual device drivers. Each different class of device (e.g. a hard drive is an example of one class) would have its own I/O handling code in the OS which can be referred to as the device driver, but it would be one set of code that handles every device of that class. A command from the "hard drive handler" code in the OS to read a specific sector of storage would be exactly identical regardless of the size of the drive (if it accesses a non-existant sector, it always gets a standard error), the maker of the drive, and whether or not there is a gateway controller to interconnect legacy hardware (e.g. SCSI, IDE, SATA, etc). The same principle would be applied for all other classes of devices. All random accessible devices could then be bootable with merely the issuance of a basic "read a sector from offset N" command generated by a very simple firmware system ... for some standardized value of N for booting purposes.
now we need to go OSS in diesel cars