Re: [PATCH v4 3/4] NFSD: handle GETATTR conflict with write delegation

[Jeff Layton <jlayton@kernel.org>]

On Sun, 2023-05-21 at 20:56 -0700, dai.ngo@oracle.com wrote:
> On 5/21/23 7:56 PM, dai.ngo@oracle.com wrote:
> > 
> > On 5/21/23 4:08 PM, Jeff Layton wrote:
> > > On Sat, 2023-05-20 at 14:36 -0700, Dai Ngo wrote:
> > > > If the GETATTR request on a file that has write delegation in effect
> > > > and the request attributes include the change info and size attribute
> > > > then the write delegation is recalled and NFS4ERR_DELAY is returned
> > > > for the GETATTR.
> > > > 
> > > > Signed-off-by: Dai Ngo <dai.ngo@oracle.com>
> > > > ---
> > > >   fs/nfsd/nfs4xdr.c | 45 +++++++++++++++++++++++++++++++++++++++++++++
> > > >   1 file changed, 45 insertions(+)
> > > > 
> > > > diff --git a/fs/nfsd/nfs4xdr.c b/fs/nfsd/nfs4xdr.c
> > > > index 76db2fe29624..e069b970f136 100644
> > > > --- a/fs/nfsd/nfs4xdr.c
> > > > +++ b/fs/nfsd/nfs4xdr.c
> > > > @@ -2920,6 +2920,46 @@ nfsd4_encode_bitmap(struct xdr_stream *xdr, 
> > > > u32 bmval0, u32 bmval1, u32 bmval2)
> > > >       return nfserr_resource;
> > > >   }
> > > >   +static struct file_lock *
> > > > +nfs4_wrdeleg_filelock(struct svc_rqst *rqstp, struct inode *inode)
> > > > +{
> > > > +    struct file_lock_context *ctx;
> > > > +    struct file_lock *fl;
> > > > +
> > > > +    ctx = locks_inode_context(inode);
> > > > +    if (!ctx)
> > > > +        return NULL;
> > > > +    spin_lock(&ctx->flc_lock);
> > > > +    if (!list_empty(&ctx->flc_lease)) {
> > > > +        fl = list_first_entry(&ctx->flc_lease,
> > > > +                    struct file_lock, fl_list);
> > > > +        if (fl->fl_type == F_WRLCK) {
> > > > +            spin_unlock(&ctx->flc_lock);
> > > > +            return fl;
> > > > +        }

One more issue here too. FL_LAYOUT file_locks live on this list too.
They shouldn't conflict with leases or delegations, so you probably just
want to skip them.

Longer term, I wonder if we ought to add a new list in the
file_lock_context for layouts? There's no reason to keep them all on the
same list.

> > > > +    }
> > > > +    spin_unlock(&ctx->flc_lock);
> > > > +    return NULL;
> > > > +}
> > > > +
> > > > +static __be32
> > > > +nfs4_handle_wrdeleg_conflict(struct svc_rqst *rqstp, struct inode 
> > > > *inode)
> > > > +{
> > > > +    __be32 status;
> > > > +    struct file_lock *fl;
> > > > +    struct nfs4_delegation *dp;
> > > > +
> > > > +    fl = nfs4_wrdeleg_filelock(rqstp, inode);
> > > > +    if (!fl)
> > > > +        return 0;
> > > > +    dp = fl->fl_owner;
> > > > +    if (dp->dl_recall.cb_clp == *(rqstp->rq_lease_breaker))
> > > > +        return 0;
> > > > +    refcount_inc(&dp->dl_stid.sc_count);
> > > Another question: Why are you taking a reference here at all?
> > 
> > This is same as in nfsd_break_one_deleg and revoke_delegation.
> > I think it is to prevent the delegation to be freed while delegation
> > is being recalled.
> > 
> > >   AFAICT,
> > > you don't even look at the delegation again after that point, so it's
> > > not clear to me who's responsible for putting that reference.
> > 
> > In v2, the sc_count is decrement by nfs4_put_stid. I forgot to do that
> > in V4. I'll add it back in v5.
> 
> Actually the refcount is decremented after the CB_RECALL is done
> by nfs4_put_stid in nfsd4_cb_recall_release. So we don't have to
> decrement it here. This is to prevent the delegation to be free
> while the recall is being sent.
> 

That's the put for the increment in nfsd_break_one_deleg.

You don't need to take an extra reference to a delegation to call
nfsd_open_break_lease. You might not even know which delegation is being
broken. There could even be more than one, after all.

I think that extra refcount_inc is likely to cause a leak.
-- 
Jeff Layton <jlayton@kernel.org>