Re: [PATCH] In crypto_add_alg(), 'exact' wants to be initialized to 0

[Steffen Klassert <steffen.klassert@secunet.com>]

On Wed, Feb 01, 2012 at 09:21:39PM +0100, Jesper Juhl wrote:
> On Wed, 1 Feb 2012, devendra.aaru wrote:
> 
> > On Sun, Jan 29, 2012 at 5:39 PM, Jesper Juhl <jj@chaosbits.net> wrote:
> > > We declare 'exact' without initializing it and then do:
> > >
> > > [...]
> > >        if (strlen(p->cru_driver_name))
> > >                exact = 1;
> > >
> > >        if (priority && !exact)
> > >                return -EINVAL;
> > >
> > > [...]
> > >
> > > If the first 'if' is not true, then the second will test an
> > > uninitialized 'exact'.
> > 
> > not needed . as the cru_driver_name will always be present :).
> 
> If that is indeed the case, and we are guaranteed that, then it would seem 
> that a patch like the following would be what we want instead??
> 
> Please note that this patch is intended just for discussion, nothing else 
> (which is why I left out a Signed-off-by on purpose), since I've not 
> tested it beyond checking that it compiles, nor have I verified your claim 
> that cru_driver_name will always be present.
> 

We get cru_driver_name from a netlink message that a user sends us.
So it depends pretty much on the user whether cru_driver_name is
set or not. Usually it is set when a user wants to instantiate
a certain algorithm driver, like "cbc(aes-asm)". If the user just
wants to instantiate the system default of an algorithm, he can
set cru_name (e.g. to "cbc(aes)") instead of cru_driver_name.

Your first patch is correct.

Thanks,

Steffen