On Sat, Oct 24, 2020 at 07:45:36AM +1100, Herbert Xu wrote: > On Fri, Oct 23, 2020 at 08:56:04AM -0700, Eric Biggers wrote: > > > > When clearing memory because "it may be sensitive" rather than "it's needed > > for > > the code to behave correctly", I think it's best to use memzero_explicit() > > to > > make the intent clear, even if it seems that memset() is sufficient. Also > > keep > > in mind that support for compiling the kernel with LTO (link-time > > optimization) > > is being worked on (and some people already do it), which results in more > > code > > being optimized out. > > The rule up until now has been that we only use memzero_explicit for > stack variables. At this point please don't convert anything else > as it will cause too much churn. > > If LTO did arrive we should do a global conversion. >
LTO is actively being worked on, now up to v6: https://lkml.kernel.org/lkml/20201013003203.4168817-1-samitolva...@google.com/ And in the real world it's already being used; the Android Compatibility Definition Document strongly recommends enabling CFI, which depends on LTO. It's doubtful that anyone will do a global conversion from memset() to memzero_explicit(), as it's too hard to find all the places that should be converted. They are in lots of different subsystems; the crypto subsystem will have the most, but not all. We just need to fix as many as we can. If you'd like to do something more comprehensive than this patch, that would be great, but I hope we don't wait forever for a global conversion that never happens. FWIW, kfree_sensitive() (formerly kzfree) already got converted by https://git.kernel.org/linus/8982ae527fbef170, and it wasn't really controversial. Some people even wanted Cc stable (which I disagreed with, but it apparently made the final version). - Eric
