Re: [PATCH v2 1/8] ceph: add global dentry lease metric support

[Xiubo Li <xiubli@xxxxxxxxxx>]

On 2020/1/9 21:52, Jeff Layton wrote:
> On Wed, 2020-01-08 at 05:41 -0500, xiubli@xxxxxxxxxx wrote:
[...]
> > @@ -1589,13 +1595,14 @@ static int ceph_d_revalidate(struct dentry *dentry, unsigned int flags)
> >   		}
> >   	}
> >   
> > +	mdsc = ceph_sb_to_client(dir->i_sb)->mdsc;
> >   	if (!valid) {
> > -		struct ceph_mds_client *mdsc =
> > -			ceph_sb_to_client(dir->i_sb)->mdsc;
> >   		struct ceph_mds_request *req;
> >   		int op, err;
> >   		u32 mask;
> >   
> > +		percpu_counter_inc(&mdsc->metric.d_lease_mis);
> > +
> >   		if (flags & LOOKUP_RCU)
> >   			return -ECHILD;
> >   
> So suppose we're doing an RCU walk, and call d_revalidate and the dentry
> is invalid. We'll bump the counter here, but then return -ECHILD and the
> kernel will do the d_revalidate all over again with refwalk, and we bump
> the counter again.
>
> Won't that end up double-counting the cache miss? Or is that intended
> here?

Yeah, we should count it once. Will fix it in the next version.

Thanks,