Mark Kettenis <[email protected]> wrote:

> Well, I believe that Scott was trying to fix a race condition that can
> only happen if code is using kbind(2) incorrectly, i.e. when the
> threads deliberately pass different cookies to kbind(2) or execute
> kbind(2) from different "text" addresses.
> 
> I still think the solution is simply to accept that race condition.

Right.

People are not calling kbind.  They are calling syscall(SYS_kbind

The man page says "don't do that".  No user serviceable parts inside.
Do not provide to children.

That said, Scott is about to share a diff he and I did a few cycles
around, to at least make the call-in transaction be a lock.

Reply via email to