Re: [PATCH] proposal adding explicit_bzero

[Rich Felker <dalias@libc.org>]

On Tue, Jun 26, 2018 at 09:48:47PM +0100, David CARLIER wrote:
> Yes true sorry :-|.

No problem. Do you agree with my proposed fix and want me to make that
change and apply it?

Rich


> On Tue, 26 Jun 2018 at 21:43, Rich Felker <dalias@libc.org> wrote:
> 
> > On Fri, Jun 15, 2018 at 01:37:43PM +0000, David CARLIER wrote:
> > > Hi dear lists,
> > > this is my first message so forgive me if this idea had already been
> > rejected.
> >
> > It's definitely not rejected outright, and I think the consensus is
> > to adopt it. But..
> >
> > > From c0a16cf96b96b009097d6ed656a2a7b8969e8399 Mon Sep 17 00:00:00 2001
> > > From: David Carlier <dcarlier@afilias.info>
> > > Date: Fri, 15 Jun 2018 13:30:09 +0000
> > > Subject: [PATCH] string: adding simple explicit_bzero implementation.
> > >
> > > glibc implementing it and modern security based code starting
> > > using it widely, here a simple implementation using memory barrier.
> > > ---
> > >  include/string.h            | 1 +
> > >  src/string/explicit_bzero.c | 8 ++++++++
> > >  2 files changed, 9 insertions(+)
> > >  create mode 100644 src/string/explicit_bzero.c
> > >
> > > diff --git a/include/string.h b/include/string.h
> > > index ce1dc300..795a2abc 100644
> > > --- a/include/string.h
> > > +++ b/include/string.h
> > > @@ -82,6 +82,7 @@ void *memccpy (void *__restrict, const void
> > *__restrict, int, size_t);
> > >  char *strsep(char **, const char *);
> > >  size_t strlcat (char *, const char *, size_t);
> > >  size_t strlcpy (char *, const char *, size_t);
> > > +void explicit_bzero (void *, size_t);
> > >  #endif
> > >
> > >  #ifdef _GNU_SOURCE
> > > diff --git a/src/string/explicit_bzero.c b/src/string/explicit_bzero.c
> > > new file mode 100644
> > > index 00000000..47dba3c7
> > > --- /dev/null
> > > +++ b/src/string/explicit_bzero.c
> > > @@ -0,0 +1,8 @@
> > > +#define _BSD_SOURCE
> > > +#include <string.h>
> > > +
> > > +void explicit_bzero(void *d, size_t n)
> > > +{
> > > +     memset(d, 0, n);
> > > +     __asm__ volatile("": "r="(d) :: "memory");
> > > +}
> > > --
> >
> > The constraint here looks wrong. Normally = is written before the
> > type, not after; I'm not sure if all compiler versions accept the
> > unusual form with it after. But more importantly you have it as an
> > output constraint, where it's essentially a dead store, such that the
> > asm block does nothing to make explicit_bzero force the memset to
> > happen.
> >
> > I think you meant for the constraint to be an input constraint "r"(d).
> > Does that sound right?
> >
> > Rich
> >