[Devel] Re: [RFC PATCH] Make AT_VECTOR_SIZE_ARCH 2 for x86-32

[Oren Laadan]

Serge E. Hallyn wrote:
> Quoting Oren Laadan (orenl at cs.columbia.edu):
>>
>> Serge E. Hallyn wrote:
>>> Quoting Serge E. Hallyn (serue at us.ibm.com):
>>>> [ RFC:  Am I on crack? ]
>>>>
>>>> Both x86-32 and x86-64 with 32-bit compat use ARCH_DLINFO_IA32,
>>>> which defines two saved_auxv entries.  But system.h only defines
>>>> AT_VECTOR_SIZE_ARCH as 2 for CONFIG_IA32_EMULATION, not for
>>>> CONFIG_X86_32.  Fix that.
>>> To be clear, this patch if right would be for pushing upstream
>>> immediately.  It still leaves open the question of what we want
>>> to do about saved_auxv.  We currently just write it out as a
>>> buffer, but since it is actually an array of longs, and therefore
>>> differently sized on x86-32 and x86-64-compat, we would need to
>>> write them out entry-by-entry (and validate no overflows for
>>> TIF_IA32 tasks).  Does that seem warranted?
>> Yes: iterate over entries and copy them.
>>
>> From a brief look at the code, I don't think the contents of the
>> saved_auxv is used anywhere inside the kernel (it's exported via
>> /proc), except for the reliance on a trailing AT_NULL record
>> which is easy to test for.
>>
>> Would it be wrong or insecure to export whatever garbage the user
>> may have put in that array (assuming it is null terminated) ?
> 
> I don't know which tools use the /proc/$$/auxv output, but I don't
> see why it would be unsafe so long as we (as I do) only copy
> AT_VECTOR_SIZE unsigned longs.
> 
> I suppose we could try and be more knowledgable about the internals
> and restore them to values that make sense, using code we'd share
> with fs/binfmt_elf.c...  

I'm quite certain that restart much validate at least that
the array ends with an AT_NULL. See for example fill_auxv_note()
in binmft_elf.c.

Oren.

_______________________________________________
Containers mailing list
Containers at lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers