Re: Fwd: Segfaulting with Ubuntu

I tested your CVS update, and it worked great.  Thanks for getting
that in so quickly!

On Fri, Feb 5, 2010 at 12:43 PM, Felix von Leitner
<[email protected]> wrote:
>> That worked!  What exactly causes this issue? Certainly not a buffer
>> overflow in dietlibc?
>
> I added a workaround to the Makefile.
>
> The issue is this: Ubuntu adds -fstack-protector to the gcc default
> flags, without telling anyone.  On x86_64, this emits code like this:
>
>   b:   64 48 8b 04 25 28 00    mov    %fs:0x28,%rax
>  12:   00 00
>  14:   48 89 45 f8             mov    %rax,-0x8(%rbp)
>
> at the beginning of each function, and
>
>  24:   48 8b 55 f8             mov    -0x8(%rbp),%rdx
>  28:   64 48 33 14 25 28 00    xor    %fs:0x28,%rdx
>  2f:   00 00
>  31:   74 05                   je     38 <foo+0x38>
>  33:   e8 00 00 00 00          callq  38 <foo+0x38>
>                        34: R_X86_64_PC32       __stack_chk_fail+0xfffffffffffffffc
>
> at the end of each function.  Now, %fs is not initialized, and accessing
> %fs:0x28 segfaults.  The libc startup code is expected to set this up,
> so that that memory can be accessed and there is a random value at that
> address.  Now, since Ubuntu enables the stack protector globally, the
> function in dietlibc that sets up %fs and %fs:0x28 is ALSO compiled with
> this extra code, and the first part accesses %fs:0x28 before the
> function has a chance to set it up, causing a segfault.
>
> If you ask me, the Ubuntu people deserve a kick in the nuts for this.
>
> I added a workaround to the cvs Makefile.
>
> Please note that dietlibc does not support -fstack-protector on all
> platforms, because I do not have access to all platforms.  So you might
> still get segfaults on non-x86 platforms.  If you do, please contact me
> so we can add stack-protector support for your platform.
>
> Felix
>