Re: [PATCH] 9p/client: fix data race on req->status

[Christian Schoenebeck]

On Friday, December 9, 2022 12:50:46 AM CET Dominique Martinet wrote:
> Christian Schoenebeck wrote on Thu, Dec 08, 2022 at 04:51:27PM +0100:
> > Right, looks like most of it should be fine. Maybe p9_client_zc_rpc() needs a
> > barrier as well?
> 
> Good point, the request is used without any other lock after the
> wait_event on req->status in trans_virtio.c;
> I'll send a separate patch for it later today.
> 
> 
> > > I think we're just protecting against compiler
> > > reordering or if on some arch the store isn't actually atomic.
> > 
> > And access order within the same thread.
> 
> In this case afaik the barrier also does that? There would be no point
> if a write barrier allowed a write placed before the barrier to be
> reordered after it...

If it's about a single access, right. However when there are multiple accesses
(e.g. an expression) and plain-C access was used then the compiler was still
free to re-order the accesses in a different order than coded.

> > > This code path actually was broken before I added the barrier a while
> > > ago (2b6e72ed747f68a03), as I was observing some rare but very real
> > > errors on a big server so I'm fairly confident that for at least x86_64
> > > the generated code isn't too bad, but if KCSAN helps catching stuff I
> > > won't complain.
> > 
> > What about p9_tag_alloc()?
> 
> I think that one's ok: it happens during the allocation before the
> request is enqueued in the idr, so it should be race-free by defition.
> 
> tools/memory-model/Documentation/access-marking.txt says
> "Initialization-time and cleanup-time accesses" should use plain
> C-language accesses, so I stuck to that.

When it is allocated then it is safe, but the object may also come from a pool
here. It's probably not likely to cause an issue here, just saying.