Re: [PATCH] fuse: Add filesystem attribute in sysfs control dir.

[Miklos Szeredi <miklos@szeredi.hu>]

On Wed, 24 Aug 2022 at 21:00, lepton <ytht.net@gmail.com> wrote:
>
> On Mon, Aug 22, 2022 at 6:37 AM Miklos Szeredi <miklos@szeredi.hu> wrote:
> >
> > On Wed, 29 Jul 2020 at 01:45, Lepton Wu <ytht.net@gmail.com> wrote:
> > >
> > > With this, user space can have more control to just abort some kind of
> > > fuse connections. Currently, in Android, it will write to abort file
> > > to abort all fuse connections while in some cases, we'd like to keep
> > > some fuse connections. This can help that.
> >
> > You can grep the same info from /proc/self/mountinfo.  Why does that not work?
> Hi Miklos, thanks for this hint. That will work. But the code in user
> space will be more complicated and not straightforward.
> For now, I can see 2 issues with mountinfo:
> 1.  one connection could have multiple entries under
> /proc/self/mountinfo if there are bind mounts,  user space code needs
> to handle that.

Using the first match should be okay, bind mounts will have the same
filesystem type.

> 2.  /proc/self/mountinfo is limited by namespace, so in theory, some
> connection under /sys/fs/fuse/connections  could be missed in it.

True.  OTOH this could be a security issue as well, since the point of
namespaces is to hide information.

> While this isn't an issue
> for the current android code, maybe it could get broken in the future.
>
> Overall, I am feeling my kernel patch is a straightforward and simple
> solution and it's just one more attribute under
> /sys/fs/fuse/connections, may I know if there are any downsides to
> doing this?

Your patch adds an ad-hoc interface for a very specific purpose with
likely no other users.  As long as this information can be gotten in
other ways there's no compelling reason to add a new interface.

Thanks,
Miklos