RE: [PATCH v2 3/4] secure_seq: use siphash24 instead of md5_transform

From: David Laight
Date: Wed Dec 14 2016 - 09:47:27 EST

From: Jason A. Donenfeld
> Sent: 14 December 2016 13:44
> To: Hannes Frederic Sowa
> > __packed not only removes all padding of the struct but also changes the
> > alignment assumptions for the whole struct itself. The rule, the struct
> > is aligned by its maximum alignment of a member is no longer true. That
> > said, the code accessing this struct will change (not on archs that can
> > deal efficiently with unaligned access, but on others).
> That's interesting. There currently aren't any alignment requirements
> in siphash because we use the unaligned helper functions, but as David
> pointed out in another thread, maybe that too should change. In that
> case, we'd have an aligned-only version of the function that requires
> 8-byte aligned input. Perhaps the best way to go about that would be
> to just mark the struct as __packed __aligned(8). Or, I guess, since
> 64-bit accesses gets split into two on 32-bit, that'd be best descried
> as __packed __aligned(sizeof(long)). Would that be an acceptable
> solution?

Just remove the __packed and ensure that the structure is 'nice'.
This includes ensuring there is no 'tail padding'.
In some cases you'll need to put the port number into a 32bit field.

I'd also require that the key be aligned.
It probably ought to be a named structure type with two 64bit members
(or with an array member that has two elements).