From: Miklos Szeredi <mszeredi@suse.cz>

Currently callers of path_permission() either pass zero or
nameidata->flags as the flags argument.  Passing lookup flags to
filesystems is completely unecessary, only the "intent" flags are
interesting.

More specifically nfs uses LOOKUP_ACCESS and LOOKUP_OPEN flags, while
fuse uses LOOKUP_ACCESS and LOOKUP_CHDIR flags.

So clean up path_permission() calls to just pass these flags.

In case of LOOKUP_CHDIR and LOOKUP_ACCESS the lookup routines need not
be passed these flags at all, they are only needed for the permission
checks.

Also remove the nameidata argument of may_create().  NFS doesn't need
LOOKUP_OPEN in this case, since it handles the creation checks on the
parent directory specially anyway.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
---
 fs/exec.c          |    4 ++--
 fs/inotify_user.c  |    2 +-
 fs/namei.c         |   25 +++++++++++--------------
 fs/open.c          |   15 +++++++--------
 fs/utimes.c        |    2 +-
 net/unix/af_unix.c |    2 +-
 6 files changed, 23 insertions(+), 27 deletions(-)

Index: linux-2.6/fs/exec.c
===================================================================
--- linux-2.6.orig/fs/exec.c	2008-05-21 18:15:02.000000000 +0200
+++ linux-2.6/fs/exec.c	2008-05-21 18:30:34.000000000 +0200
@@ -116,7 +116,7 @@ asmlinkage long sys_uselib(const char __
 	if (!S_ISREG(nd.path.dentry->d_inode->i_mode))
 		goto exit;
 
-	error = path_permission(&nd.path, MAY_READ | MAY_EXEC, nd.flags);
+	error = path_permission(&nd.path, MAY_READ | MAY_EXEC, 0);
 	if (error)
 		goto exit;
 
@@ -664,7 +664,7 @@ struct file *open_exec(const char *name)
 		struct inode *inode = nd.path.dentry->d_inode;
 		file = ERR_PTR(-EACCES);
 		if (S_ISREG(inode->i_mode)) {
-			int err = path_permission(&nd.path, MAY_EXEC, nd.flags);
+			int err = path_permission(&nd.path, MAY_EXEC, 0);
 			file = ERR_PTR(err);
 			if (!err) {
 				file = nameidata_to_filp(&nd,
Index: ...
[ message continues ]