On Thu, Aug 29, 2024 at 05:33:54PM +0200, Alexander Bluhm wrote:
> On Thu, Aug 29, 2024 at 09:00:03AM -0600, Theo de Raadt wrote:
> > Claudio Jeker <cjeker@diehard.n-r-g.com> wrote:
> > 
> > > On Thu, Aug 29, 2024 at 04:21:01PM +0200, Alexander Bluhm wrote:
> > > > Hi,
> > > > 
> > > > m_pulldown() copies memory between different mbufs.  So they cannot
> > > > overlap and memcpy() should be enough.
> > > > 
> > > > ok?
> > > 
> > > Indeed. The first memcpy is save because n is not a read-only buffer and
> > > so n and n->m_next have do be different mbufs or clusters.
> > > The 2nd memcpy goes to freshly allocated buffer o. So that is safe too.
> > > 
> > > OK claudio@
> > 
> > I agree.
> > 
> > I don't recall the situation backwards copy.  Does the kernel not check
> > for that, is that only in userland?  I did that so long ago..
> 
> lib/libkern/memmove.c checks direction, lib/libkern/memcpy.c does
> not.  I think we removed the direction check from all memcpy a while
> ago.  Argument was, that memcpy caller should do it correctly.
> 
> ----------------------------
> revision 1.3
> date: 2013/06/12 16:44:22;  author: deraadt;  state: Exp;  lines: +4 -16;
> From now on, the MI libkern memcpy should not do overlap handling.
> ----------------------------
> 

Also we use __builtin for most of those functions and only fall
back to the libkern version if the compiler does so.

-- 
:wq Claudio