Re: [PATCH RFC tip/core/rcu 23/23] vhost: add __rcu annotations

From: Mathieu Desnoyers
Date: Tue May 18 2010 - 10:52:39 EST


* Paul E. McKenney (paulmck@xxxxxxxxxxxxxxxxxx) wrote:
> On Mon, May 17, 2010 at 09:35:28PM -0400, Mathieu Desnoyers wrote:
> > * Paul E. McKenney (paulmck@xxxxxxxxxxxxxxxxxx) wrote:
> > > On Mon, May 17, 2010 at 07:40:25PM -0400, Mathieu Desnoyers wrote:
> > > > * Paul E. McKenney (paulmck@xxxxxxxxxxxxxxxxxx) wrote:
> > > > > On Mon, May 17, 2010 at 06:00:25PM -0400, Mathieu Desnoyers wrote:
>
> [ . . . ]
>
> > > > > But perhaps we should be simply treating this as a use-after-free
> > > > > problem, so that RCU is not directly involved. Isn't that the standard
> > > > > use of debugobjects anyway?
> > > >
> > > > OK so we could tie "rcu_dereference" do debugobjects, and free would be
> > > > a standard free. Yes, I think it could be done. It looks a bit like the
> > > > memory allocation debugging code. If we know that a certain
> > > > rcu_dereference always access dynamically allocated memory, we could
> > > > probably add some checks there based on the memory allocator debug
> > > > objects.
> > >
> > > We probably need vhost to add code at the end of the relevant RCU
> > > read-side critical section checking that the pointers returned by
> > > any rcu_dereference() calls still point to valid memory. Don't get
> > > me wrong, your approach could find bugs in which someone forgot to
> > > remove the RCU-protected structure from a public list, but it could
> > > not detect failure to wait a grace period between the time of removal
> > > and the time of freeing.
> >
> > Good point too. So something like a new rcu_unreference() (or feel free
> > to find any better name) ;) that would be compiled out normally, but
> > would call into debugobjects might do the trick. We would have to add
> > these annotations to match every rcu_dereference() though, might means a
> > lot of new lines of code. On the plus side, that looks like a good audit
> > of RCU read-side use. ;)
>
> My first thought is that we have added quite a bit of RCU consistency
> check code in the past few months, so we should see what bugs they find
> and what bugs escape. It is all too easy to create consistency check
> code that is more trouble than it is worth.

Yes, although I expect that this new checking scheme will take some time
to implement and mainline anyway (implementation effort which I might
leave to someone else, as I have to focus on tracing at the moment).

> But in the meantime, let's see what would be required to check for
> failures to insert grace-period delays:
>
> o There would need to be something like rcu_unreference(),
> rcu_no_more_readers() or some such after the grace period.
> The update side would then become something like the following:
>
> oldp = rcu_dereference_protected(gp, &mylock);
> rcu_assign_pointer(gp, newp);
> synchronize_rcu();
> rcu_no_more_readers(oldp);
> kfree(oldp);

Replacing a kfree with a rcu_free(kfree, oldp) call that would include
both could lessen the amount of typing:

#define rcu_free(freefct, ptr) \
do { \
rcu_no_more_readers(ptr); \
freefct(ptr); \
} while (0)


> o There would need to be something to check all of the pointers
> traversed in the read-side critical sections:
>
> rcu_read_lock();
> ...
> p1 = rcu_dereference(gp1->field1);
> ...
> p2 = rcu_dereference(gp2->field2);
> ...
>
> rcu_validate(p1);
> rcu_validate(p2);

Hrm, isn't the goal of this "rcu_validate(p1)" just to keep track of
"p1" liveness ? Or do you plan to add a check there also ? I'm not sure
I figure out what you are planning to validate here. I was thinking more
in terms of

rcu_unreference(p1);
rcu_unreference(p1);

that would be symmetric with the rcu_dereference.

> rcu_read_unlock();
>
> One thing that bothers me about this is that we are forcing the developer
> to do a lot of extra typing. For example, rcu_no_more_readers() is in
> a truth-and-beauty sense redundant with kfree() -- why type both? The
> same could be said about rcu_validate() and rcu_read_unlock(), but nested
> RCU read-side critical sections make this difficult.

Ideally we'd like to add near-zero burden on developers, but I fear this
cannot be done easily for read-side C.S.. As for write-side, we have to
choose between tradeoff of genericity and less typing, e.g., between:

rcu_free(kfree, ptr);
and
rcu_kfree(ptr)

for the second, we would have to create a whole family of rcu_*free().

>
> Or am I misunderstanding what you are suggesting?

I'm only unsure about the "validate" part.

Thanks,

Mathieu

>
> Thanx, Paul

--
Mathieu Desnoyers
Operating System Efficiency R&D Consultant
EfficiOS Inc.
http://www.efficios.com
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/