lkml.org 
[lkml]   [2003]   [Aug]   [7]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [BROKEN PATCH] syscalls leak data via registers?
Hi!

> It looks to me like the syscall calling convention, on x86 at least,
> leaks kernel data out via the registers.
[scary stuff]
> When sys_foo returns, the value of %ebx has been changed to 77 on the
> stack, so when it returns to user-mode, the whole world can see that
> arg1 was assigned 77 at some point.
>
> It seems to me the bug is in restoring the register values on return to
> user-mode. As I understand it, the x86 ABI says that the called
> function owns the stack memory which contains the function's arguments,
> so it is completely within gcc's right to reuse the memory as spill
> space (or anything else) when generating code for that function.
> Therefore, the code in entry.S should not restore those values to
> registers - it should just trash all the registers (except %eax, of
> course) before returning.
>
> I tried writing a patch which replaces the RESTORE_ALL with the
> equivalent which simply skips %esp over the other registers, pops %eax
> and then assigns it to %ebx-%ebp (it makes as good a trash value as
> any), but this crashes when calibrating the delay loop. Hm, looks like
> the RESTORE_ALL on the syscall return path is also used by the interrupt
> return path - that probably shouldn't trash registers.

I believe userspace depends on registers to be preserved over system
call, except for eax. So what you found is not only security problem,
but also crasher bug.
Pavel

--
When do you have a heart between your knees?
[Johanka's followup: and *two* hearts?]
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:47    [W:0.071 / U:4.064 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site