Re: [PATCH v2] crypto: DRBG - always try to free Jitter RNG instance

[Eric Biggers <ebiggers@kernel.org>]

On Fri, Jun 05, 2020 at 08:52:57AM +0200, Stephan Mueller wrote:
> Am Freitag, 5. Juni 2020, 08:16:46 CEST schrieb Eric Biggers:
> 
> Hi Eric,
> 
> > On Fri, Jun 05, 2020 at 07:58:15AM +0200, Stephan Mueller wrote:
> > > Am Freitag, 5. Juni 2020, 02:43:36 CEST schrieb Eric Biggers:
> > > 
> > > Hi Eric,
> > > 
> > > > On Thu, Jun 04, 2020 at 08:41:00AM +0200, Stephan Müller wrote:
> > > > > The Jitter RNG is unconditionally allocated as a seed source follwoing
> > > > > the patch 97f2650e5040. Thus, the instance must always be deallocated.
> > > > > 
> > > > > Reported-by: syzbot+2e635807decef724a1fa@syzkaller.appspotmail.com
> > > > > Fixes: 97f2650e5040 ("crypto: drbg - always seeded with SP800-90B
> > > > > ...")
> > > > > Signed-off-by: Stephan Mueller <smueller@chronox.de>
> > > > > ---
> > > > > 
> > > > >  crypto/drbg.c | 3 +++
> > > > >  1 file changed, 3 insertions(+)
> > > > > 
> > > > > diff --git a/crypto/drbg.c b/crypto/drbg.c
> > > > > index 37526eb8c5d5..8a0f16950144 100644
> > > > > --- a/crypto/drbg.c
> > > > > +++ b/crypto/drbg.c
> > > > > @@ -1631,6 +1631,9 @@ static int drbg_uninstantiate(struct drbg_state
> > > > > *drbg)>
> > > > > 
> > > > >  	if (drbg->random_ready.func) {
> > > > >  	
> > > > >  		del_random_ready_callback(&drbg->random_ready);
> > > > >  		cancel_work_sync(&drbg->seed_work);
> > > > > 
> > > > > +	}
> > > > > +
> > > > > +	if (!IS_ERR_OR_NULL(drbg->jent)) {
> > > > > 
> > > > >  		crypto_free_rng(drbg->jent);
> > > > >  		drbg->jent = NULL;
> > > > >  	
> > > > >  	}
> > > > 
> > > > It it okay that ->jent can be left as an ERR_PTR() value?
> > > > 
> > > > Perhaps it should always be set to NULL?
> > > 
> > > The error value is used in the drbg_instantiate function. There it is
> > > checked whether -ENOENT (i.e. the cipher is not available) or any other
> > > error is present. I am not sure we should move that check.
> > > 
> > > Thanks for the review.
> > 
> > drbg_seed() and drbg_async_seed() check for drbg->jent being NULL.
> > 
> > Will that now break due it drbg->jent possibly being an ERR_PTR()?
> > 
> > Hence why I'm asking whether drbg_uninstantiate() should set it to NULL.
> 
> The allocation happens in drbg_prepare_hrng that is only invoked by 
> drbg_instantiate.
> 
> drbg_instantiate checks for the ERR_PTR and sets it to NULL in case the error 
> is deemed ok.
> 
> Thus, any subsequent functions would see either a valid pointer or NULL. The 
> only exception is drbg_uninstantiate when invoked from the error case 
> 
>                 ret = drbg_prepare_hrng(drbg);
>                 if (ret)
>                         goto free_everything;
> 
> Thus, I think that the two functions you mention will never see any values 
> other than NULL or a valid pointer.
> 

To be concrete, I'm suggesting:

	if (!IS_ERR_OR_NULL(drbg->jent))
		crypto_free_rng(drbg->jent);
	drbg->jent = NULL;

This would be similar to how drbg_dealloc_state() sets lots of other fields of
the drbg_state to NULL.

It's your call though.  I haven't properly read this code; the above is just
what makes sense to me at first glance...

- Eric