From: NeilBrown <neilb@suse.com>
Subject: VFS: don't test owner for NFS in set_posix_acl()
References: bsc#1103405
Patch-mainline: Never, this is a hack

The owner seen in inode->i_uid should not be used for permission
checking in NFS.  The server should always be asked, either with the
ACCESS request, or by trying the operation.

posix_set_acl() does test inode->i_uid, so can deny permission
when it should be granted.  So disable this for NFS.

Upstream fix will probably add a new flag to inode_permission(), and
use that.

Signed-off-by: NeilBrown <neilb@suse.com>
Acked-by: NeilBrown <neilb@suse.com>

---
 fs/posix_acl.c |   11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

--- a/fs/posix_acl.c
+++ b/fs/posix_acl.c
@@ -22,6 +22,7 @@
 #include <linux/xattr.h>
 #include <linux/export.h>
 #include <linux/user_namespace.h>
+#include <linux/magic.h>
 
 static struct posix_acl **acl_by_type(struct inode *inode, int type)
 {
@@ -864,8 +865,14 @@ set_posix_acl(struct inode *inode, int t
 
 	if (type == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode))
 		return acl ? -EACCES : 0;
-	if (!inode_owner_or_capable(inode))
-		return -EPERM;
+	/* NFS doesn't need an owner check, as the server will
+	 * do that.  The owner check is wrong when the server
+	 * is mapping uids, such as with all_squash (which makes
+	 * everyone an owner of a newly created file).
+	 */
+	if (inode->i_sb->s_magic != NFS_SUPER_MAGIC)
+		if (!inode_owner_or_capable(inode))
+			return -EPERM;
 
 	if (acl) {
 		int ret = posix_acl_valid(inode->i_sb->s_user_ns, acl);