Re: [PATCH RFC] vfs: add mount umount logs

[Anand Jain <anand.jain@oracle.com>]

Thanks for the comments.

On 05/19/2017 11:01 PM, Theodore Ts'o wrote:
> On Fri, May 19, 2017 at 08:17:55AM +0800, Anand Jain wrote:
>>> XFS already logs its own unmounts.
>>
>>  Nice. as far as I know its only in XFS.
>
> Ext4 logs mounts, but not unmounts.
>
>>>  I prefer to let each filesystem log
>>> its own unmount, because then the mount/unmount messages also have the
>>> same prefix as all other messages coming from that filesystem driver.
>>
>>  Ok. One nitpick though there are other mount types (remount, bind..),
>>  and subsequent mounts of the same FS and FS driver can't log them
>>  effectively.
>
> The other issue is there is a difference between when an unmount is
> started, or when a file system is unmounted in one namespace, and when
> the file system as a whole is unmounted by the low-level because there
> are no longer any references to the file system in any namespace.

> More than once I've gotten a bug report for ext4 that was really
> caused by the fact that some program had forked a daemon process in
> its own namespace, so when the USB thumbdrive was unmounted, it wasn't
> *really* unmounted.  So being able to clearly express that in the logs
> is also a good thing.

  Thanks.

> I think the way to do that is to have one set of logs for when the
> file system is unmounted from one mount namespace (or from a bind
> mount), with perhaps an indication of the number of remaining
> refcounts, and to make this be distinct from the unmount of the
> underlying low file system code, which should be logged by the file
> system code itself.

  Thanks for suggesting. I have added sb->s_active in V2 in the ML.
  And s_active value taken after mount and before umount, so it
  may be a little confusing that the number is not seen as decremented
  when umount is run. Also I didn't want to manipulate it (like -1) at
  the time of logging. Unless if there is a better parameter to track
  though.

Thanks, Anand

> 					- Ted