Re: [PATCH] explicit_bzero final

[Florian Weimer <fweimer at redhat dot com>]

On 12/16/2016 12:31 AM, Zack Weinberg wrote:
> > > +/* This is the generic definition of __explicit_bzero_chk.  The
> > > +   __explicit_bzero_chk symbol is used as the implementation of
> > > +   explicit_bzero throughout glibc.  If this file is overriden by an
> > > +   architecture, both __explicit_bzero_chk and
> > > +   __explicit_bzero_chk_internal have to be defined (the latter not as
> > > +   an IFUNC).  */
> > >
> > > This file is not in sysdeps/generic, so it cannot be overridden (or is
> > > that no longer the case? If so, why do we still have sysdeps/generic?)
> > > and I don't think we need the capability to override it.  Better we
> > > should get libc-internal references to memset going to the proper ifunc
> > > for the architecture.
> > It can be overridden, and it has to be, otherwise you end up with
> > multiple definitions of the same symbol when linking libc.so.
> ??? Your patch doesn't have any overrides of it.

I meant it has to be overridden if you want to introduce 
architecture-specific implementations of explicit_bzero.  The patch 
doesn't contain any, but should be compatible with future development.

Due the transitive nature of IFUNCs and the desire to avoid indirect 
calls, all memset/__memset_chk variants need an 
explicit_bzero/__explicit_bzero_chk wrapper, so this is quite involved.

Thanks,
Florian