This is an interesting performance issue on a Samba ctdb cluster when Samba is configured to use vfs_fileid, /etc/mtab of the system is somewhat large and the workload is eg copying small files to the server.

In strace I noticed a loop over the mtab for every file created, each iteration taking up to 500 ms due to calling stat+statfs on every mtab entry.

Looks like the problem is that we're calling vfs_file_id_from_sbuf() in
open_file_ntcreate() where smb_fname->st is still all zero for newly created files.

Have WIP fix, need bugnumber.