[OE-core] [PATCH 2/5] security_flags: pass ssp-buffer-size param to stack protector

[Joshua Lock]

On Mon, 2016-08-22 at 09:02 +0100, André Draszik wrote:
> On Fr, 2016-08-19 at 19:46 +0100, Joshua G Lock wrote:
> > 
> > On Fri, 2016-08-19 at 10:07 -0700, Khem Raj wrote:
> > > 
> > > 
> > > > 
> > > > 
> > > > 
> > > > On Aug 19, 2016, at 8:34 AM, Joshua Lock <joshua.g.lock at intel.c
> > > > om>
> > > > wrote:
> > > > 
> > > > This tells the compiler to use a canary to protect any function
> > > > which
> > > > declares a character array of 4 or more bytes on its stack,
> > > > rather
> > > > than the default of 8 or more bytes.
> > > 
> > > Thats fine, however, it slows down the code, strong option was a
> > > compromise
> > > otherwise we could just use fstack-protector-all
> > 
> > It's my understanding that the ssp-buffer-size parameter changes
> > the
> > size of buffer the base, fstack-protector, protections affect and
> > that
> > the performance impact is less significant than adding protections
> > to
> > all functions via stack-protector-all?
> 
> I understand it as follows instead:
> 
> --param=ssp-buffer-size=X only makes sense together with -fstack-
> protector, as -fstack-protector can to be configured for the minimum
> size of arrays to protect (8 by default, if --param=ssp-buffer-size=
> is not given).
> 
> --param=ssp-buffer-size=X does not make sense with -fstack-protector-
> strong as this version protects arrays of *any* size anyway.
> 
> https://gcc.gnu.org/ml/gcc-patches/2012-06/msg00974.html
>   -> This also has the design doc towards the end.
> https://lwn.net/Articles/584225/
> 
> So I don't think this patch is needed at all...

Thanks for the link, having read the patch and design doc I'm inclined
to agree.

> > 
> > FWIW, the related options in Fedora and Ubuntu:
> > 
> > * Ubuntu: -fstack-protector --param=ssp-buffer-size=4 (default in
> > hardened builds)
> > * Fedora: -fstack-protector-strong --param=ssp-buffer-size=4
> > (default
> > in all builds)
> 
> Debian (sid) uses -fstack-protector-strong (without ssp-buffer-size).
> 

Thanks, I couldn't find a Debian system on Friday evening.

Regards,

Joshua