nfsd: fix failure to set eof in readdir in some situations
authorJ. Bruce Fields <>
Thu, 30 Oct 2008 17:48:33 +0000 (13:48 -0400)
committerJ. Bruce Fields <>
Thu, 30 Oct 2008 21:16:49 +0000 (17:16 -0400)
Before 14f7dd632011bb89c035722edd6ea0d90ca6b078 "[PATCH] Copy XFS
readdir hack into nfsd code", readdir_cd->err was reset to eof before
each call to vfs_readdir; afterwards, it is set only once.  Similarly,
c002a6c7977320f95b5edede5ce4e0eeecf291ff "[PATCH] Optimise NFS readdir
hack slightly", can cause us to exit without nfserr_eof set.  Fix this.

This ensures the "eof" bit is set when needed in readdir replies.  (The
particular case I saw was an nfsv4 readdir of an empty directory, which
returned with no entries (the protocol requires "." and ".." to be
filtered out), but with eof unset.)

Cc: David Woodhouse <>
Cc: Al Viro <>
Signed-off-by: J. Bruce Fields <>

index 0bc56f6..848a03e 100644 (file)
@@ -1912,6 +1912,7 @@ static int nfsd_buffered_readdir(struct file *file, filldir_t func,
                        de = (struct buffered_dirent *)((char *)de + reclen);
                offset = vfs_llseek(file, 0, SEEK_CUR);
+               cdp->err = nfserr_eof;
                if (!buf.full)