[PATCH] vfs: Allow FS_USERNS_MOUNT filesystems to create inodes with unmapped uid/gids

[Johanna Abrahamsson <johanna@mjao.org>]

Fixes: 036d523641c6 ("vfs: Don't create inodes with a uid or gid unknown
to the vfs")

This fixes the bugzilla bug #183461
(https://bugzilla.kernel.org/show_bug.cgi?id=183461) "mkdir(2) returns
EOVERFLOW on tmpfs in user namespace" that was introduced in the
aforementioned commit.

Since filesystems with the FS_USERNS_MOUNT flag (mainly tmpfs in this
case) are safe to use within user namespaces they should be allowed
to create inodes without checking whether their uid and gid has a
mapping or not.

I learned most of what I know about user namespaces while researching
this bug, so there's a fair chance that I've misunderstood something.
Feedback is greatly appreciated.

Signed-off-by: Johanna Abrahamsson <johanna@mjao.org>
---
 fs/namei.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

diff --git a/fs/namei.c b/fs/namei.c
index 1669c93d..9e45e01 100644
--- a/fs/namei.c
+++ b/fs/namei.c
@@ -2816,10 +2816,12 @@ static inline int may_create(struct inode *dir, struct dentry *child)
 		return -EEXIST;
 	if (IS_DEADDIR(dir))
 		return -ENOENT;
-	s_user_ns = dir->i_sb->s_user_ns;
-	if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
-	    !kgid_has_mapping(s_user_ns, current_fsgid()))
-		return -EOVERFLOW;
+	if (!(dir->i_sb->s_type->fs_flags & FS_USERNS_MOUNT)) {
+		s_user_ns = dir->i_sb->s_user_ns;
+		if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
+		    !kgid_has_mapping(s_user_ns, current_fsgid()))
+			return -EOVERFLOW;
+	}
 	return inode_permission(dir, MAY_WRITE | MAY_EXEC);
 }
 
-- 
2.1.4