NFSv4: Fix a potential state manager deadlock when returning delegations

The nfsv4 state manager could potentially deadlock inside
__nfs_inode_return_delegation() if the server reboots, so that the calls to
nfs_msync_inode() end up waiting on state recovery to complete.

Also ensure that if a server reboot or network partition causes us to have
to stop returning delegations, that NFS4CLNT_DELEGRETURN is set so that
the state manager can resume any outstanding delegation returns after it
has dealt with the state recovery situation.

Finally, ensure that the state manager doesn't wait for the DELEGRETURN
call to complete. It doesn't need to, and that too can cause a deadlock.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>

1 files changed, 28 insertions, 14 deletions

diff --git a/fs/nfs/delegation.c b/fs/nfs/delegation.c
index 6dd48a4..eeecd69 100644
--- a/fs/nfs/delegation.c
+++ b/fs/nfs/delegation.c
@@ -92,7 +92,7 @@ out:
 	return status;
 }
 
-static void nfs_delegation_claim_opens(struct inode *inode, const nfs4_stateid *stateid)
+static int nfs_delegation_claim_opens(struct inode *inode, const nfs4_stateid *stateid)
 {
 	struct nfs_inode *nfsi = NFS_I(inode);
 	struct nfs_open_context *ctx;
@@ -116,10 +116,11 @@ again:
 			err = nfs_delegation_claim_locks(ctx, state);
 		put_nfs_open_context(ctx);
 		if (err != 0)
-			return;
+			return err;
 		goto again;
 	}
 	spin_unlock(&inode->i_lock);
+	return 0;
 }
 
 /*
@@ -261,30 +262,34 @@ static void nfs_msync_inode(struct inode *inode)
 /*
  * Basic procedure for returning a delegation to the server
  */
-static int __nfs_inode_return_delegation(struct inode *inode, struct nfs_delegation *delegation)
+static int __nfs_inode_return_delegation(struct inode *inode, struct nfs_delegation *delegation, int issync)
 {
 	struct nfs_inode *nfsi = NFS_I(inode);
+	int err;
 
-	nfs_msync_inode(inode);
 	/*
 	 * Guard against new delegated open/lock/unlock calls and against
 	 * state recovery
 	 */
 	down_write(&nfsi->rwsem);
-	nfs_delegation_claim_opens(inode, &delegation->stateid);
+	err = nfs_delegation_claim_opens(inode, &delegation->stateid);
 	up_write(&nfsi->rwsem);
-	nfs_msync_inode(inode);
+	if (err)
+		goto out;
 
-	return nfs_do_return_delegation(inode, delegation, 1);
+	err = nfs_do_return_delegation(inode, delegation, issync);
+out:
+	return err;
 }
 
 /*
  * Return all delegations that have been marked for return
  */
-void nfs_client_return_marked_delegations(struct nfs_client *clp)
+int nfs_client_return_marked_delegations(struct nfs_client *clp)
 {
 	struct nfs_delegation *delegation;
 	struct inode *inode;
+	int err = 0;
 
 restart:
 	rcu_read_lock();
@@ -298,12 +303,18 @@ restart:
 		delegation = nfs_detach_delegation_locked(NFS_I(inode), NULL);
 		spin_unlock(&clp->cl_lock);
 		rcu_read_unlock();
-		if (delegation != NULL)
-			__nfs_inode_return_delegation(inode, delegation);
+		if (delegation != NULL) {
+			filemap_flush(inode->i_mapping);
+			err = __nfs_inode_return_delegation(inode, delegation, 0);
+		}
 		iput(inode);
-		goto restart;
+		if (!err)
+			goto restart;
+		set_bit(NFS4CLNT_DELEGRETURN, &clp->cl_state);
+		return err;
 	}
 	rcu_read_unlock();
+	return 0;
 }
 
 /*
@@ -338,8 +349,10 @@ int nfs_inode_return_delegation(struct inode *inode)
 		spin_lock(&clp->cl_lock);
 		delegation = nfs_detach_delegation_locked(nfsi, NULL);
 		spin_unlock(&clp->cl_lock);
-		if (delegation != NULL)
-			err = __nfs_inode_return_delegation(inode, delegation);
+		if (delegation != NULL) {
+			nfs_msync_inode(inode);
+			err = __nfs_inode_return_delegation(inode, delegation, 1);
+		}
 	}
 	return err;
 }
@@ -368,7 +381,8 @@ void nfs_super_return_all_delegations(struct super_block *sb)
 		spin_unlock(&delegation->lock);
 	}
 	rcu_read_unlock();
-	nfs_client_return_marked_delegations(clp);
+	if (nfs_client_return_marked_delegations(clp) != 0)
+		nfs4_schedule_state_manager(clp);
 }
 
 static void nfs_client_mark_return_all_delegations(struct nfs_client *clp)