Slashdot Mirror
Torvalds Creates Patch for Cross-Platform Virus
Newsforge is reporting that Linus Torvalds took a few minutes to review the cross-platform proof of concept virus covered yesterday and has proven that the virus does indeed not work with latest kernel version 2.6.16 and even released a patch in order to fix this "problem." From the article: "The reason that the virus is not propagating itself in the latest kernel versions is due to a bug in how GCC handles specific registers in a particular system call. [...] So the virus did a number of strange things to make this show up, but on the other hand the kernel does try to avoid touching user registers, even if we've never really _guaranteed_ that. So the 2.6.16 effect is a mis-feature, even if a _normal_ app would never care. It just happened to bite the infection logic of your virus thing."
I think you misunderstand. He fixed a flaw in the kernel that kept the virus from *working*. The patched systems should be vulnerable.
You say
Yes. The kernel patch works around a bug in gcc. The patched systems are now *vulnerable*
There are 11 types of people. Those who understand binary, those who don't and those who are sick of this lame joke.
Linus did not create a patch for the virus. Linus created a patch for the Linux kernel, to fix a bug which happened to have been discovered by looking at the virus.
Of course, if the story had been submitted with the correct title of "Linus fixes bug in Linux", it probably would never have been posted.
Tarsnap: Online backups for the truly paranoid
The virus in question apparently wasn't infecting system files--it didn't have an elevation-of-privileges feature, so it couldn't access /bin, /usr, etc. (And /etc, too, though that's not relevant.)
So if a 'virus' is using standard OS features that legitimate applications also use, and suddenly the virus stops working, there's obviously been a change, and it breaks those legitimate applications.
In short, Torvalds didn't want to remove a feature without prior discussion.
AFIAK, there is no actual exploit in the code provided. The virus only does things that a regular program should be able to do, given the correct permissions.
The virus, written in assembly, calls the kernel via a depreciated interface (int 0x80 instead of syscall). It happens to have a value in the ebx register that it needs after the (buggy) system call.
The bug in the kernel is due to the fact that gcc assumes the system call doesn't change user registers (which the kernel isn't suppossed to as a policy) so gcc forms code to make the system call in less time (less instructions, less overhead) by not caring about user registers. The fix for the bug simply restores the value of the ebx register to what it was before the system call, hence the bug now works (as it has the correct value in the ebx register).