syscall_fs2: fix the chown logic to match modern Unix

diff --git a/Kernel/syscall_fs2.c b/Kernel/syscall_fs2.c
index 0bac3e5..2ee0512 100644 (file)
--- a/Kernel/syscall_fs2.c
+++ b/Kernel/syscall_fs2.c
@@ -245,15 +245,27 @@ arg_t _fchmod(void)
 
 static int chown_op(inoptr ino)
 {
-       if ((ino->c_node.i_uid != udata.u_euid ||
-               (group != udata.u_gid &&  !in_group(group))) && esuper())
-               return (-1);
        if (ino->c_flags & CRDONLY) {
                udata.u_error = EROFS;
                return -1;
        }
-       ino->c_node.i_uid = owner;
-       ino->c_node.i_gid = group;
+       /* Owner change must be superuser rights */
+       if (owner != -1) {
+               if (esuper())
+                       return -1;
+               ino->c_node.i_uid = owner;
+       /* owner = group = -1 is a no-op */
+       } else if (group == -1)
+               return 0;
+       /* Group change */
+       if (group != -1) {
+               /* We must be in the target group (and file owner) */
+               if ((ino->c_node.i_uid != udata.u_euid ||
+                       group != udata.u_egid && !in_group(group)) && esuper())
+                       return -1;
+               ino->c_node.i_gid = group;
+       }
+       ino->c_node.i_mode &= ~(SET_GID|SET_UID);
        setftime(ino, C_TIME);
        return 0;
 }