gfs2: Don't clear SGID when inheriting ACLs
authorJan Kara <jack@suse.cz>
Wed, 19 Jul 2017 15:56:42 +0000 (10:56 -0500)
committerBob Peterson <rpeterso@redhat.com>
Wed, 19 Jul 2017 15:58:54 +0000 (10:58 -0500)
When new directory 'DIR1' is created in a directory 'DIR0' with SGID bit
set, DIR1 is expected to have SGID bit set (and owning group equal to
the owning group of 'DIR0'). However when 'DIR0' also has some default
ACLs that 'DIR1' inherits, setting these ACLs will result in SGID bit on
'DIR1' to get cleared if user is not member of the owning group.

Fix the problem by moving posix_acl_update_mode() out of
__gfs2_set_acl() into gfs2_set_acl(). That way the function will not be
called when inheriting ACLs which is what we want as it prevents SGID
bit clearing and the mode has been properly set by posix_acl_create()
anyway.

Fixes: 073931017b49d9458aa351605b43a7e34598caef
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Bob Peterson <rpeterso@redhat.com>
fs/gfs2/acl.c

index 2524807ee0703643867e7e04c803d10b1aab9182..a6d962323790c71795b2c1c1f0a8c5660ec286c1 100644 (file)
@@ -86,19 +86,6 @@ int __gfs2_set_acl(struct inode *inode, struct posix_acl *acl, int type)
        char *data;
        const char *name = gfs2_acl_name(type);
 
-       if (acl && acl->a_count > GFS2_ACL_MAX_ENTRIES(GFS2_SB(inode)))
-               return -E2BIG;
-
-       if (type == ACL_TYPE_ACCESS) {
-               umode_t mode = inode->i_mode;
-
-               error = posix_acl_update_mode(inode, &inode->i_mode, &acl);
-               if (error)
-                       return error;
-               if (mode != inode->i_mode)
-                       mark_inode_dirty(inode);
-       }
-
        if (acl) {
                len = posix_acl_to_xattr(&init_user_ns, acl, NULL, 0);
                if (len == 0)
@@ -130,6 +117,9 @@ int gfs2_set_acl(struct inode *inode, struct posix_acl *acl, int type)
        bool need_unlock = false;
        int ret;
 
+       if (acl && acl->a_count > GFS2_ACL_MAX_ENTRIES(GFS2_SB(inode)))
+               return -E2BIG;
+
        ret = gfs2_rsqa_alloc(ip);
        if (ret)
                return ret;
@@ -140,7 +130,18 @@ int gfs2_set_acl(struct inode *inode, struct posix_acl *acl, int type)
                        return ret;
                need_unlock = true;
        }
+       if (type == ACL_TYPE_ACCESS && acl) {
+               umode_t mode = inode->i_mode;
+
+               ret = posix_acl_update_mode(inode, &inode->i_mode, &acl);
+               if (ret)
+                       goto unlock;
+               if (mode != inode->i_mode)
+                       mark_inode_dirty(inode);
+       }
+
        ret = __gfs2_set_acl(inode, acl, type);
+unlock:
        if (need_unlock)
                gfs2_glock_dq_uninit(&gh);
        return ret;