On 2/10/20 6:29 PM, Kurt Roeckx wrote: > On Mon, Feb 10, 2020 at 04:19:20PM +0000, Matt Caswell wrote: >> >> >> On 10/02/2020 00:15, SHANE LONTIS wrote: >>> With the new architecture changes there are quite a few new calls to >>> >>> CRYPTO_UP_REF() >>> CRYPTO_DOWN_REF() >>> >>> These methods return an int that is not being checked in lots of places. >>> >>> This return value only seems to affect fallback code that calls >>> CRYPTO_atomic_add (which can return 0 on lock or unlock failure) >>> >>> SO the question is should we be checking this return value? >> >> Yes, I think we should be. > > I think that as long as we have that fallback code, that it should > be checked. > >
Yes, although I wonder why a function that checks the return value of CRYPTO_THREAD_write_lock and CRYPTO_THREAD_unlock does not check for possible overflow of the addition, which is far more likely to happen. Bernd.