Re: [PATCH] resolv: Fix tests by aligning hand crafted queries

[Stafford Horne <shorne@gmail.com>]

On Tue, Sep 28, 2021 at 06:00:04PM +0200, Florian Weimer wrote:
> * Adhemerval Zanella via Libc-alpha:
> 
> > On 14/06/2021 20:40, Stafford Horne via Libc-alpha wrote:
> >> When testing OpenRISC I get a bus error in res_send.  This is due to the
> >> buf being cast to a (HEADER *) and trying the res_send code trying to read
> >> different bits of the HEADER struct including 16-bit id etc.
> >> 
> >> On OpenRISC reads of 16-bits and 32-bits from structures need to be 2
> >> byte and 4 byte aligned, respectively.
> >> 
> >> To fix this we can align the hand crafted queries.
> >
> >
> > But the res_send() interface does specify that buffer is an 'unsigned char',
> > so I think the problem is in fact send_vc() (any any other code that consume
> > the buffer) where the cast is in fact undefined.  I am not sure why it has
> > not show any issue on architecture that trap on unaligned access, may guess
> > is the stack buffer alignment is the same as the HEADER.
> >
> > The issue is resolv code seems to abuse this...
> 
> Right, there is a separate bug for this:
> 
>   Misaligned access in res_query.c HEADER struct
>   <https://sourceware.org/bugzilla/show_bug.cgi?id=20243>
> 
> I really want to fix this, but each time I get side-tracked by other
> cleanups that appear to be necessary for this. 8-(

Thanks for the comments, in the mean time does that mean my patch should not go
in as it is?

-Stafford