Re: [PATCH v2] fs: binfmt_elf: Add ELF header consistency checks

[Geyslan Gregório Bem]

2013/11/21 Geyslan Gregório Bem <geyslan@xxxxxxxxx>:
> 2013/11/20 Al Viro <viro@xxxxxxxxxxxxxxxxxx>:
>> On Wed, Nov 20, 2013 at 09:34:31PM -0300, Geyslan G. Bem wrote:
>>> The member 'e_ehsize' that holds the ELF header size is compared
>>> with the elfhdr struct size. If not equal, goes out.
>>> If 'e_phoff' holds 0 the object has no program header table, so
>>> goes out.
>>> Ensures the file being loaded has the correct data encoding, checking
>>> 'e_ident[EI_DATA]' against 'ELF_DATA'.
>>>
>>> Besides the checks being in accordance with the ELF Specifications,
>>> they increase the binary consistency reducing the use of malformed ones.
>>
>> This is completely misguided.  We are allowed to reject such binaries,
>> but what's the point of doing that?
>
> Viro, First of all, thanks for reply.
>
> The security (or anti-security) guys are used to mess up with the not checked
> header fields for their "benefits": anti-debugging, injection and so on.
>
> Concerning to 'e_phoff': when it is 0 the check avoids that 'elf_phdr' been read
> from a erroneous offset (ELF header). I know that without this check the binary
> will goes out anyway. But it reduces wasting cpu cycles.
>
> Regarding 'e_ident[EI_DATA]': that check also prevents a farther code reading
> when the binary, although been the correct arch, is compiled with a different
> data encoding (MSB vs LSB).
>
> So checking besides increase the binary consistency, guarantee some mislead
> and fewer cpu cycles.
>
> --
> Regards,
>
> Geyslan G. Bem
> hackingbits.com

Another good reason is that ld does reject such binaries (I hex edit
one to hold MSB value in header):

uzumaki@hb ~ $ /lib/ld-linux-x86-64.so.2 ./a.out
./a.out: error while loading shared libraries: ./a.out: ELF file data
encoding not little-endian

After zeroing the phoff:

uzumaki@hb ~ $ /lib/ld-linux-x86-64.so.2 ./a.out
./a.out: error while loading shared libraries: ./a.out: object file
has no loadable segments

I really think that is a way to get a more robust binfmt consistency check.

What you think?

-- 
Regards,

Geyslan G. Bem
hackingbits.com
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/