Re: Stack Smash Protection disabled (was HEADS-UP: Stack Smash Protection enabled by default for amd64 and i386)

[Ty Sarna <ty%sarna.org@localhost>]

On Nov 13, 2009, at 11:36 AM, Thor Lancelot Simon wrote:

> On Fri, Nov 13, 2009 at 05:01:34AM +0000, David Holland wrote:
> > It's been noted elsewhere that theoretically the overhead of SSP is
> > not supposed to be 5%; it's supposed to be negligible. Where is this
> > 5% overhead coming from?
>
> One possibility is our non-default settings for the stack  
> protector.  We
> tell GCC to protect access to all objects -- not just objects large  
> enough
> to contain an address, which is the default.

So, perhaps we can enable full-on protection for some things, as  
before, but use default settings for everything else and get most of  
the benefit with a smaller overhead? Or maybe we should use default  
settings everywhere?

This seems insufficiently researched to me. I think we ought to  
understand what's going on better before accepting a 5% penalty that  
we may not need to, or conversely accepting the risk of doing without  
SSP when the cost might be easily reduced. Not to mention having a  
flamewar either way...