nfsd: Add a mutex to protect the NFSv4.0 open owner replay cache

We don't want to rely on the client_mutex for protection in the case of
NFSv4 open owners. Instead, we add a mutex that will only be taken for
NFSv4.0 state mutating operations, and that will be released once the
entire compound is done.

Also, ensure that nfsd4_cstate_assign_replay/nfsd4_cstate_clear_replay
take a reference to the stateowner when they are using it for NFSv4.0
open and lock replay caching.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>
Signed-off-by: Jeff Layton <jlayton@primarydata.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>

fs/nfsd/nfs4proc.c

@@ -469,11 +469,8 @@ out:
 		fh_put(resfh);
 		kfree(resfh);
 	}
-	nfsd4_cleanup_open_state(open, status);
+	nfsd4_cleanup_open_state(cstate, open, status);
-	if (open->op_openowner && !nfsd4_has_session(cstate))
-		cstate->replay_owner = &open->op_openowner->oo_owner;
 	nfsd4_bump_seqid(cstate, status);
-	if (!cstate->replay_owner)
 	nfs4_unlock_state();
 	return status;
 }
@@ -1395,10 +1392,7 @@ encode_op:
 			args->ops, args->opcnt, resp->opcnt, op->opnum,
 			be32_to_cpu(status));
 
-		if (cstate->replay_owner) {
+		nfsd4_cstate_clear_replay(cstate);
-			nfs4_unlock_state();
-			cstate->replay_owner = NULL;
-		}
 		/* XXX Ugh, we need to get rid of this kind of special case: */
 		if (op->opnum == OP_READ && op->u.read.rd_filp)
 			fput(op->u.read.rd_filp);

fs/nfsd/nfs4state.c

@@ -1069,7 +1069,7 @@ void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
 		return;
 
 	if (!seqid_mutating_err(ntohl(nfserr))) {
-		cstate->replay_owner = NULL;
+		nfsd4_cstate_clear_replay(cstate);
 		return;
 	}
 	if (!so)
@@ -2940,6 +2940,28 @@ static void init_nfs4_replay(struct nfs4_replay *rp)
 	rp->rp_status = nfserr_serverfault;
 	rp->rp_buflen = 0;
 	rp->rp_buf = rp->rp_ibuf;
+	mutex_init(&rp->rp_mutex);
+}
+
+static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
+		struct nfs4_stateowner *so)
+{
+	if (!nfsd4_has_session(cstate)) {
+		mutex_lock(&so->so_replay.rp_mutex);
+		cstate->replay_owner = so;
+		atomic_inc(&so->so_count);
+	}
+}
+
+void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
+{
+	struct nfs4_stateowner *so = cstate->replay_owner;
+
+	if (so != NULL) {
+		cstate->replay_owner = NULL;
+		mutex_unlock(&so->so_replay.rp_mutex);
+		nfs4_put_stateowner(so);
+	}
 }
 
 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
@@ -3855,7 +3877,8 @@ out:
 	return status;
 }
 
-void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
+void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
+			      struct nfsd4_open *open, __be32 status)
 {
 	if (open->op_openowner) {
 		struct nfs4_openowner *oo = open->op_openowner;
@@ -3869,6 +3892,8 @@ void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
 			} else
 				oo->oo_flags &= ~NFS4_OO_NEW;
 		}
+		if (open->op_openowner)
+			nfsd4_cstate_assign_replay(cstate, &oo->oo_owner);
 	}
 	if (open->op_file)
 		nfsd4_free_file(open->op_file);
@@ -4399,8 +4424,7 @@ nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
 	if (status)
 		return status;
 	stp = openlockstateid(s);
-	if (!nfsd4_has_session(cstate))
+	nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
-		cstate->replay_owner = stp->st_stateowner;
 
 	status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
 	if (!status)
@@ -4469,7 +4493,6 @@ put_stateid:
 	nfs4_put_stid(&stp->st_stid);
 out:
 	nfsd4_bump_seqid(cstate, status);
-	if (!cstate->replay_owner)
 	nfs4_unlock_state();
 	return status;
 }
@@ -4544,7 +4567,6 @@ put_stateid:
 	nfs4_put_stid(&stp->st_stid);
 out:
 	nfsd4_bump_seqid(cstate, status);
-	if (!cstate->replay_owner)
 	nfs4_unlock_state();
 	return status;
 }
@@ -4610,7 +4632,6 @@ nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
 	/* put reference from nfs4_preprocess_seqid_op */
 	nfs4_put_stid(&stp->st_stid);
 out:
-	if (!cstate->replay_owner)
 	nfs4_unlock_state();
 	return status;
 }
@@ -5071,7 +5092,6 @@ out:
 	if (status && new_state)
 		release_lock_stateid(lock_stp);
 	nfsd4_bump_seqid(cstate, status);
-	if (!cstate->replay_owner)
 	nfs4_unlock_state();
 	if (file_lock)
 		locks_free_lock(file_lock);
@@ -5236,7 +5256,6 @@ put_stateid:
 	nfs4_put_stid(&stp->st_stid);
 out:
 	nfsd4_bump_seqid(cstate, status);
-	if (!cstate->replay_owner)
 	nfs4_unlock_state();
 	if (file_lock)
 		locks_free_lock(file_lock);

fs/nfsd/nfs4xdr.c

@@ -3925,8 +3925,6 @@ status:
  * 
  * XDR note: do not encode rp->rp_buflen: the buffer contains the
  * previously sent already encoded operation.
- *
- * called with nfs4_lock_state() held
  */
 void
 nfsd4_encode_replay(struct xdr_stream *xdr, struct nfsd4_op *op)

fs/nfsd/state.h

@@ -328,6 +328,7 @@ struct nfs4_replay {
 	unsigned int		rp_buflen;
 	char			*rp_buf;
 	struct knfsd_fh		rp_openfh;
+	struct mutex		rp_mutex;
 	char			rp_ibuf[NFSD4_REPLAY_ISIZE];
 };
 

fs/nfsd/xdr4.h

@@ -599,7 +599,9 @@ extern __be32 nfsd4_process_open1(struct nfsd4_compound_state *,
 		struct nfsd4_open *open, struct nfsd_net *nn);
 extern __be32 nfsd4_process_open2(struct svc_rqst *rqstp,
 		struct svc_fh *current_fh, struct nfsd4_open *open);
-extern void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status);
+extern void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate);
+extern void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
+		struct nfsd4_open *open, __be32 status);
 extern __be32 nfsd4_open_confirm(struct svc_rqst *rqstp,
 		struct nfsd4_compound_state *, struct nfsd4_open_confirm *oc);
 extern __be32 nfsd4_close(struct svc_rqst *rqstp,
@@ -630,6 +632,7 @@ extern __be32 nfsd4_test_stateid(struct svc_rqst *rqstp,
 extern __be32 nfsd4_free_stateid(struct svc_rqst *rqstp,
 		struct nfsd4_compound_state *, struct nfsd4_free_stateid *free_stateid);
 extern void nfsd4_bump_seqid(struct nfsd4_compound_state *, __be32 nfserr);
+
 #endif
 
 /*