2 files changed, 13 insertions, 1 deletions

diff --git a/fs/nfs/inode.c b/fs/nfs/inode.c
index a56add0bc5b7..e8c143d182c4 100644
--- a/fs/nfs/inode.c
+++ b/fs/nfs/inode.c
@@ -840,6 +840,12 @@ int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
  *
  * After an operation that has changed the inode metadata, mark the
  * attribute cache as being invalid, then try to update it.
+ *
+ * NB: if the server didn't return any post op attributes, this
+ * function will force the retrieval of attributes before the next
+ * NFS request.  Thus it should be used only for operations that
+ * are expected to change one or more attributes, to avoid
+ * unnecessary NFS requests and trips through nfs_update_inode().
  */
 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
 {
diff --git a/fs/nfs/write.c b/fs/nfs/write.c
index 38ba5c09af08..c12effb46fe5 100644
--- a/fs/nfs/write.c
+++ b/fs/nfs/write.c
@@ -1253,7 +1253,13 @@ int nfs_writeback_done(struct rpc_task *task, struct nfs_write_data *data)
         dprintk("NFS: %4d nfs_writeback_done (status %d)\n",
                 task->tk_pid, task->tk_status);
 
-        /* Call the NFS version-specific code */
+        /*
+         * ->write_done will attempt to use post-op attributes to detect
+         * conflicting writes by other clients.  A strict interpretation
+         * of close-to-open would allow us to continue caching even if
+         * another writer had changed the file, but some applications
+         * depend on tighter cache coherency when writing.
+         */
         status = NFS_PROTO(data->inode)->write_done(task, data);
         if (status != 0)
                 return status;