Sub-second timestamp seems variable, random?

[tomjridge]

Running main.native server, mounting with Ubuntu 16.04.2 kernel 9p client:

 sudo mount -t 9p -o trans=tcp,port=5640,dfltuid=5945,dfltgid=5945 127.0.0.1 /mnt/9p

Consecutive stats are OK, except for access/modify/change times which are seemingly random. For example, two consecutive stats:

$ mnt $ stat 9p/tmp.txt 
  File: 9p/tmp.txt
  Size: 141       	Blocks: 1          IO Block: 8192   regular file
Device: 100085h/1048709d	Inode: 1357078     Links: 1
Access: (0660/-rw-rw----)  Uid: ( 5945/    tr61)   Gid: ( 5945/    tr61)
Access: 2017-09-18 12:09:03.655666165 +0100
Modify: 2017-09-18 12:09:01.655666165 +0100
Change: 2017-09-18 12:09:01.655666165 +0100
 Birth: -

# (h:pc1177) (p:/mnt) (d:/dev/sda2)  
$ mnt $ stat 9p/tmp.txt 
  File: 9p/tmp.txt
  Size: 141       	Blocks: 1          IO Block: 8192   regular file
Device: 100085h/1048709d	Inode: 1357078     Links: 1
Access: (0660/-rw-rw----)  Uid: ( 5945/    tr61)   Gid: ( 5945/    tr61)
Access: 2017-09-18 12:09:03.775758325 +0100
Modify: 2017-09-18 12:09:01.775758325 +0100
Change: 2017-09-18 12:09:01.775758325 +0100
 Birth: -

Intentional behaviour? Or bug?

lofs9p.ml has the following:

  let set_times path atime mtime =
    (* TODO: this can block on Unix.utimes and we shouldn't but
       Lwt_unix does not yet wrap Unix.utimes. *)
    (* NOTE: The behavior of this is slightly wrong as 'any' means
       "don't change" but 0 means "set to now". *)

but we are not setting, only reading (?) so this shouldn't matter.