Re: [PATCH -next v2 0/4] mm: replace mmap_sem for mm->exe_file serialization

From: Davidlohr Bueso <dave@stgolabs.net>
To: Oleg Nesterov <oleg@redhat.com>
Cc: akpm@linux-foundation.org, viro@zeniv.linux.org.uk,
	gorcunov@openvz.org, koct9i@gmail.com, linux-mm@kvack.org,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH -next v2 0/4] mm: replace mmap_sem for mm->exe_file serialization
Date: Sun, 15 Mar 2015 08:42:05 -0700	[thread overview]
Message-ID: <1426434125.28068.100.camel@stgolabs.net> (raw)
In-Reply-To: <20150315152652.GA24590@redhat.com>

On Sun, 2015-03-15 at 16:26 +0100, Oleg Nesterov wrote:
> On 03/15, Davidlohr Bueso wrote:
> >
> > On Sun, 2015-03-15 at 15:21 +0100, Oleg Nesterov wrote:
> > > I didn't even read this version, but honestly I don't like it anyway.
> > >
> > > I leave the review to Cyrill and Konstantin though, If they like these
> > > changes I won't argue.
> > >
> > > But I simply can't understand why are you doing this.
> > >
> > >
> > >
> > > Yes, this code needs cleanups, I agree. Does this series makes it better?
> > > To me it doesn't, and the diffstat below shows that it blows the code.
> >
> > Looking at some of the caller paths now, I have to disagree.
> 
> And I believe you are wrong. But let me repeat, I leave this to Cyrill
> and Konstantin. Cleanups are always subjective.
> 
> > > In fact, to me it complicates this code. For example. Personally I think
> > > that MMF_EXE_FILE_CHANGED should die. And currently we can just remove it.
> >
> > How could you remove this?
> 
> Just remove this flag and the test_and_set_bit(MMF_EXE_FILE_CHANGED) check.
> Again, this is subjective, but to me it looks ugly. Why do we allow to
> change ->exe_file but only once?

Ok I think I am finally seeing where you are going. And I like it *a
lot* because it allows us to basically replace mmap_sem with rcu
(MMF_EXE_FILE_CHANGED being the only user that requires a lock!!), but
am afraid it might not be possible. I mean currently we have no rule wrt
to users that don't deal with prctl. 

Forbidding multiple exe_file changes to be generic would certainly
change address space semantics, probably for the better (tighter around
security), but changed nonetheless so users would have a right to
complain, no? So if we can get away with removing MMF_EXE_FILE_CHANGED
I'm all for it. Andrew?

Thanks,
Davidlohr

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>