Re: crypto: user - crypto_alg_match removal

[Stephan Mueller <smueller@chronox.de>]

Am Montag, 24. November 2014, 08:22:46 schrieb Steffen Klassert:

Hi Steffen,

> Hi Stephan.
> 
> On Sat, Nov 22, 2014 at 11:25:05PM +0100, Stephan Mueller wrote:
> > Hi Steffen, Herbert,
> > 
> > may I ask for the reasons why crypto_alg_match exists? Doesn't it
> > implement crypto_alg_lookup -- and that not even complete? Is there a
> > particular reason why this exact flag of crypto_alg_match is really
> > needed in the context of crypto_user?
> > 
> > Unless there is such valid reason, may I ask whether we can remove
> > crypto_alg_match and simply use crypto_alg_lookup in all instances where
> > crypto_alg_match is invoked using the following replacement:
> > 
> > alg = crypto_alg_lookup(p->cru_name, p->cru_type, p->cru_mask)
> > 
> > The only problem with this replacement is that p->cru_driver_name is not
> > considered with that.
> 
> With crypto_alg_lookup() we don't know whether the match is based on
> the driver or the algorithm name. That's why we have crypto_alg_match(),
> here we can ask for a driver or an algorithm match. In some situations it
> is important to have an exact match on the crypto driver name. For example
> if a user wants to instantiate or delete a certain inplementation of an
> algorithm. In this case we need to know whether this exact algorithm
> driver is registered in the system.

I understand. But going with the logic of the kernel crypto API, if one needs 
an exact match, you pick the driver name. Otherwise the generic name. 
crypto_alg_lookup returns the exact algo when you supply a driver name. It 
returns the algo with the highest prio when you supply a generic name.

I do not see a difference for the scenarios you describe. What I am worried 
about is that the logic of how a name versus a driver name is applied differs 
between the kernel crypto API and crypto_user.


> 
> > Do you think a double invocation of
> > crypto_alg_lookup should be done or that even the user space interface
> > should be changed such that cru_driver_name is removed from it?
> 
> Whatever we do, we can't remove cru_driver_name as this is exported
> to userspace and tools already use it.

That is definitely an issue. But the more I think about it, the more I see 
that we do not need to change the interface.

Something like that would work in the kernel:

if (p->cru_driver_name[0])
	alg = crypto_alg_lookup(p->cru_driver_name, p->cru_type, p->cru_mask)
else
	alg = crypto_alg_lookup(p->cru_name, p->cru_type, p->cru_mask)

-- 
Ciao
Stephan