NFS: Optimise NFS close()

Close-to-open cache consistency rules really only require us to flush out
writes on calls to close(), and require us to revalidate attributes on the
very last close of the file.

Currently we appear to be doing a lot of extra attribute revalidation
and cache flushes.

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

7 files changed, 47 insertions, 21 deletions

diff --git a/fs/nfs/file.c b/fs/nfs/file.c
index 1eab9c9..d451073 100644
--- a/fs/nfs/file.c
+++ b/fs/nfs/file.c
@@ -137,9 +137,6 @@ nfs_file_release(struct inode *inode, struct file *filp)
 			dentry->d_parent->d_name.name,
 			dentry->d_name.name);
 
-	/* Ensure that dirty pages are flushed out with the right creds */
-	if (filp->f_mode & FMODE_WRITE)
-		nfs_wb_all(dentry->d_inode);
 	nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
 	return nfs_release(inode, filp);
 }
@@ -231,7 +228,6 @@ nfs_file_flush(struct file *file, fl_owner_t id)
 	struct nfs_open_context *ctx = nfs_file_open_context(file);
 	struct dentry	*dentry = file->f_path.dentry;
 	struct inode	*inode = dentry->d_inode;
-	int		status;
 
 	dprintk("NFS: flush(%s/%s)\n",
 			dentry->d_parent->d_name.name,
@@ -241,11 +237,8 @@ nfs_file_flush(struct file *file, fl_owner_t id)
 		return 0;
 	nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
 
-	/* Ensure that data+attribute caches are up to date after close() */
-	status = nfs_do_fsync(ctx, inode);
-	if (!status)
-		nfs_revalidate_inode(NFS_SERVER(inode), inode);
-	return status;
+	/* Flush writes to the server and return any errors */
+	return nfs_do_fsync(ctx, inode);
 }
 
 static ssize_t
diff --git a/fs/nfs/inode.c b/fs/nfs/inode.c
index c40adc5..a834d1d 100644
--- a/fs/nfs/inode.c
+++ b/fs/nfs/inode.c
@@ -541,6 +541,32 @@ int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
 	return err;
 }
 
+/**
+ * nfs_close_context - Common close_context() routine NFSv2/v3
+ * @ctx: pointer to context
+ * @is_sync: is this a synchronous close
+ *
+ * always ensure that the attributes are up to date if we're mounted
+ * with close-to-open semantics
+ */
+void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
+{
+	struct inode *inode;
+	struct nfs_server *server;
+
+	if (!(ctx->mode & FMODE_WRITE))
+		return;
+	if (!is_sync)
+		return;
+	inode = ctx->path.dentry->d_inode;
+	if (!list_empty(&NFS_I(inode)->open_files))
+		return;
+	server = NFS_SERVER(inode);
+	if (server->flags & NFS_MOUNT_NOCTO)
+		return;
+	nfs_revalidate_inode(server, inode);
+}
+
 static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
 {
 	struct nfs_open_context *ctx;
@@ -567,24 +593,15 @@ struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
 	return ctx;
 }
 
-static void __put_nfs_open_context(struct nfs_open_context *ctx, int wait)
+static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
 {
-	struct inode *inode;
+	struct inode *inode = ctx->path.dentry->d_inode;
 
-	if (ctx == NULL)
-		return;
-
-	inode = ctx->path.dentry->d_inode;
 	if (!atomic_dec_and_lock(&ctx->count, &inode->i_lock))
 		return;
 	list_del(&ctx->list);
 	spin_unlock(&inode->i_lock);
-	if (ctx->state != NULL) {
-		if (wait)
-			nfs4_close_sync(&ctx->path, ctx->state, ctx->mode);
-		else
-			nfs4_close_state(&ctx->path, ctx->state, ctx->mode);
-	}
+	NFS_PROTO(inode)->close_context(ctx, is_sync);
 	if (ctx->cred != NULL)
 		put_rpccred(ctx->cred);
 	path_put(&ctx->path);
diff --git a/fs/nfs/internal.h b/fs/nfs/internal.h
index a55e69a..2041f68 100644
--- a/fs/nfs/internal.h
+++ b/fs/nfs/internal.h
@@ -152,6 +152,9 @@ extern __be32 *nfs4_decode_dirent(__be32 *p, struct nfs_entry *entry, int plus);
 extern struct rpc_procinfo nfs4_procedures[];
 #endif
 
+/* proc.c */
+void nfs_close_context(struct nfs_open_context *ctx, int is_sync);
+
 /* dir.c */
 extern int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask);
 
diff --git a/fs/nfs/nfs3proc.c b/fs/nfs/nfs3proc.c
index c55be7a..b82fe68 100644
--- a/fs/nfs/nfs3proc.c
+++ b/fs/nfs/nfs3proc.c
@@ -834,4 +834,5 @@ const struct nfs_rpc_ops nfs_v3_clientops = {
 	.commit_done	= nfs3_commit_done,
 	.lock		= nfs3_proc_lock,
 	.clear_acl_cache = nfs3_forget_cached_acls,
+	.close_context	= nfs_close_context,
 };
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index 95f171e..97baccc 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -1572,6 +1572,15 @@ out_drop:
 	return 0;
 }
 
+void nfs4_close_context(struct nfs_open_context *ctx, int is_sync)
+{
+	if (ctx->state == NULL)
+		return;
+	if (is_sync)
+		nfs4_close_sync(&ctx->path, ctx->state, ctx->mode);
+	else
+		nfs4_close_state(&ctx->path, ctx->state, ctx->mode);
+}
 
 static int _nfs4_server_capabilities(struct nfs_server *server, struct nfs_fh *fhandle)
 {
@@ -3776,6 +3785,7 @@ const struct nfs_rpc_ops nfs_v4_clientops = {
 	.commit_done	= nfs4_commit_done,
 	.lock		= nfs4_proc_lock,
 	.clear_acl_cache = nfs4_zap_acl_attr,
+	.close_context  = nfs4_close_context,
 };
 
 /*
diff --git a/fs/nfs/proc.c b/fs/nfs/proc.c
index 1934652..7be72d9 100644
--- a/fs/nfs/proc.c
+++ b/fs/nfs/proc.c
@@ -663,4 +663,5 @@ const struct nfs_rpc_ops nfs_v2_clientops = {
 	.commit_setup	= nfs_proc_commit_setup,
 	.lock		= nfs_proc_lock,
 	.lock_check_bounds = nfs_lock_check_bounds,
+	.close_context	= nfs_close_context,
 };
diff --git a/include/linux/nfs_xdr.h b/include/linux/nfs_xdr.h
index 0691b9c..9708e78 100644
--- a/include/linux/nfs_xdr.h
+++ b/include/linux/nfs_xdr.h
@@ -868,6 +868,7 @@ struct nfs_rpc_ops {
 	int	(*lock)(struct file *, int, struct file_lock *);
 	int	(*lock_check_bounds)(const struct file_lock *);
 	void	(*clear_acl_cache)(struct inode *);
+	void	(*close_context)(struct nfs_open_context *ctx, int);
 };
 
 /*