Re: 2.6.6-mm1

[Andrew Morton <akpm@osdl.org>]

Christoph Hellwig <hch@infradead.org> wrote:
>
> On Mon, May 10, 2004 at 03:15:54PM -0700, Andrew Morton wrote:
> > It beats the alternatives which are floating about, which includes a sysctl
> > which defeats CAP_SYS_MLOCK system-wide.
> 
> That one might not be nice, but at least it doesn't randomly change
> the meaning of a group id.

If you don't set the sysctl there is no change in system behaviour.

> So yeah, although it's a hack too it's
> much much better than the junk that just went into Linus tree.

Untrue.  With the system-wide thing unprivileged local users can
trivially DoS the database.
 
> Why btw do we have a staging tree if such sensitive patches go into
> mainline without proper review after just one day?

It was discussed on lkml, then later was dicussed extensively off-list
and I lost track of how long it had been in -mm.  Sorry.

> > >  What happened to the patch rick promised
> > > to make mlock an rlimit?  This is the right approach and could be easily
> > > extended to hugetlb pages.
> > 
> > rlimits don't work for this.  shm segments persist after process exit and
> > aren't associated with a particular user.
> 
> When did shm segments come into the play?  I know we bolted hugetlb
> support onto the back of the already horrible sysv shm interface, but
> if people want additional interfaces ontop of that they should use
> the proper mmap api.

Rewriting Oracle isn't a practical alternative.