kernel-hardening - Re: Re: [RFC v4 PATCH 00/13] HARDENED

Follow @Openwall on Twitter for new release announcements and other news

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20161110222744.GD8086@kroah.com>
Date: Thu, 10 Nov 2016 23:27:44 +0100
From: Greg KH <gregkh@...uxfoundation.org>
To: kernel-hardening@...ts.openwall.com
Cc: Will Deacon <will.deacon@....com>,
	Elena Reshetova <elena.reshetova@...el.com>, keescook@...omium.org,
	arnd@...db.de, tglx@...utronix.de, mingo@...hat.com,
	h.peter.anvin@...el.com, linux-kernel@...r.kernel.org
Subject: Re: Re: [RFC v4 PATCH 00/13] HARDENED_ATOMIC

On Thu, Nov 10, 2016 at 10:13:10PM +0100, Peter Zijlstra wrote:
> On Thu, Nov 10, 2016 at 08:48:38PM +0000, Will Deacon wrote:
> > > That said, I still don't much like this.
> > > 
> > > I would much rather you make kref useful and use that. It still means
> > > you get to audit all refcounts in the kernel, but hey, you had to do
> > > that anyway.
> > 
> > What needs to happen to kref to make it useful? Like many others, I've
> > been guilty of using atomic_t for refcounts in the past.
> 
> As it stands kref is a pointless wrapper. If it were to provide
> something actually useful, like wrap protection, then it might actually
> make sense to use it.

It provides the correct cleanup ability for a reference count and the
object it is in, so it's not all that pointless :)

But I'm always willing to change it to make it work better for people,
if kref did the wrapping protection (i.e. used a non-wrapping atomic
type), then you would have that.  I thought that was what this patchset
provided...

And yes, this is a horridly large patchset.  I've looked at these
changes, and in almost all of them, people are using atomic_t as merely
a "counter" for something (sequences, rx/tx stats, etc), to get away
without having to lock it with an external lock.

So, does it make more sense to just provide a "pointless" api for this
type of "counter" pattern:
	counter_inc()
	counter_dec()
	counter_read()
	counter_set()
	counter_add()
	counter_subtract()
Those would use the wrapping atomic type, as they can wrap all they want
and no one really is in trouble.  Once those changes are done, just make
atomic_t not wrap and all should be fine, no other code should need to
be changed.

We can bikeshed on the function names for a while, to let everyone feel
they contributed (counter, kcount, ksequence, sequence_t, cnt_t, etc.)...

And yes, out-of-tree code will work differently, but really, the worse
that could happen is their "sequence number" stops wrapping :)

Would that be a better way to implement this?

thanks,

greg k-h

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.