Re: Crypto Fixes for 4.12

[Linus Torvalds]

On Thu, Jun 15, 2017 at 9:54 AM, Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx> wrote:
>
> This push fixes a bug on sparc where we may dereference freed stack
> memory.

Ugh, that's a particularly ugly fix for a random gcc bug on a random
architecture that almost nobody tests.

In other words, it's nasty. It's nasty because nobody sane will ever
realize this pattern, and the code will either bit-rot or just happen
again somewhere else.

I'd have been *much* happier if this had been some nicer abstraction
that is built up around the use of SHASH_DESC_ON_STACK(), and just
have some rule that "SHASH_DESC_ON_STACK()" needs to be paired with
retrieving the final value and then a SHASH_DESC_DEALLOC() or
whatever.

Then you *could* implement SHASH_DESC_ON_STACK() as a kmalloc, and
SHASH_DESC_DEALLOC() would be a kfree - but with an alloca()-like
allocation the SHASH_DESC_DEALLOC() would be that "barrier_data()".

At that point the interface would make _sense_ at some conceptual
level, rather than being a random hack for a small collection of
random users of this thing.

There's a fair number of SHASH_DESC_ON_STACK users, are all the others
safe for some random reason that just happens to be about code
generation? Did people actually verify that?

                Linus