Re: [PATCH 0/9] [RFC v2] safely drop directory dentry on failed revalidate

[Ian Kent <raven@themaw.net>]

On Sun, 2013-09-01 at 08:56 +0800, Ian Kent wrote:
> On Fri, 2013-08-30 at 10:59 +0200, Miklos Szeredi wrote:
> > On Fri, Aug 30, 2013 at 1:44 AM, Ian Kent <raven@themaw.net> wrote:
> > 
> > >> The check in waitq.c above "is" used to validate the need to callback to
> > >> the daemon to request a mount.
> > 
> > Okay.  But then shouldn't the check be  "if (d_mountpoint(dentry)) valid = 0;" ?
> 
> I don't think so.
> 
> But again, may_umount() might do what's needed here too.

Oh, hang on, may_umount() can't be used for the root-less multi-mount
case at all, since there's no vfsmount at the base of the tree the check
can't be restricted to just the tree of subdirs that needs to be
checked. That's been a problem for me for a long time.

> 
> I don't think this is enough because it doesn't cover the case where the
> dentry is not simple_empty() but has no mounts below. That's not a
> normal use case but could happen if the daemon crashed at just the wrong
> time, encountered an error condition that didn't allow it to cleanup
> directories, or a user umounted triggers within the tree before starting
> the daemon.
> 
> > 
> > Thanks,
> > Miklos
>