Re: [PATCH] VFS: Suppress automount on [l]stat, [l]getxattr, etc.

[Ian Kent <raven@themaw.net>]

On Mon, 2011-09-26 at 18:18 -0700, Linus Torvalds wrote:
> On Mon, Sep 26, 2011 at 5:59 PM, Trond Myklebust
> <Trond.Myklebust@netapp.com> wrote:
> >
> > OK. Then how about something along the lines of the following patch
> > (compile tested only)?
> 
> I would seriously suggest aiming for *minimal*.
> 
> So if it already has any of LOOKUP_DIRECTORY, LOOKUP_OPEN,
> LOOKUP_CREATE or LOOKUP_PARENT, don't add LOOKUP_AUTOMOUNT - because
> it's not adding anything.
> 
> We *have* to automount on those flags anyway. They are not like
> LOOKUP_FOLLOW was, where the automount decision was just a random
> (bad) choice.
> 
> From a quick look, pretty much all of the cases you added
> LOOKUP_AUTOMOUNT to were already implicit automount events. So I think
> the two remaining ones are quotactl Q_QUOTAON case, bind mount in
> do_looopback(), and the nfs_follow_remote_path() case.
> 
> Sure, we could add it to the others as some kind of "documentation"
> thing, but I think that should be separate and is debatable to begin
> with.

That was the only question I had when I suggested using LOOKUP_AUTOMOUNT
to restore the previous automounting behavior. The trade off being to
add the flag and simplify the check in follow_automount() or to not add
it to implicit automount cases and retain the check but minimize the
changes to path walk call sites.

Talk about an automount file type was purely a suggestion of something
that would be good to do later along with what would be a lengthy user
space co-ordination and education process, not something I thought was
good to do now.

Making allowances for the change back to the old behavior for those
kernel users that had different semantics at the outset was a second and
separate question and still is.

Ian