Re: [PATCH v5 15/22] fsnotify: send event with parent/name info to sb/mount/non-dir marks

From: Amir Goldstein <amir73il@gmail.com>
To: Jan Kara <jack@suse.cz>
Cc: linux-fsdevel <linux-fsdevel@vger.kernel.org>
Subject: Re: [PATCH v5 15/22] fsnotify: send event with parent/name info to sb/mount/non-dir marks
Date: Thu, 16 Jul 2020 21:42:20 +0300	[thread overview]
Message-ID: <CAOQ4uxiS2zNkVQZjcErmqq2OSXdfk2_H+gDyRWEAdjzbM+qipg@mail.gmail.com> (raw)
In-Reply-To: <20200716175709.GM5022@quack2.suse.cz>

On Thu, Jul 16, 2020 at 8:57 PM Jan Kara <jack@suse.cz> wrote:
>
> On Thu 16-07-20 20:20:04, Amir Goldstein wrote:
> > On Thu, Jul 16, 2020 at 8:01 PM Jan Kara <jack@suse.cz> wrote:
> > >
> > > On Thu 16-07-20 11:42:23, Amir Goldstein wrote:
> > > > Similar to events "on child" to watching directory, send event "on child"
> > > > with parent/name info if sb/mount/non-dir marks are interested in
> > > > parent/name info.
> > > >
> > > > The FS_EVENT_ON_CHILD flag can be set on sb/mount/non-dir marks to specify
> > > > interest in parent/name info for events on non-directory inodes.
> > > >
> > > > Events on "orphan" children (disconnected dentries) are sent without
> > > > parent/name info.
> > > >
> > > > Events on direcories are send with parent/name info only if the parent
> > > > directory is watching.
> > > >
> > > > Signed-off-by: Amir Goldstein <amir73il@gmail.com>
> > >
> > > Hum, doesn't this break ignore mask handling in
> > > fanotify_group_event_mask()? Because parent's ignore mask will be included
> > > even though parent is added into the iter only to carry the parent info...
> > >
> >
> > Hmm, break ignore mask handling? or fix it?
> >
> > Man page said:
> > "Having these two types of masks permits a mount point or directory to be
> >  marked for receiving events, while at the  same time ignoring events for
> >  specific objects under that mount point or directory."
>
> Right, but presumably that speaks of the case of a mark where the parent
> has FS_EVENT_ON_CHILD set. For case of parent watching events of a child, I
> agree it makes sense to apply ignore masks of both the parent and the child.
>
> > The author did not say what to expect from marking a mount and ignoring
> > a directory.
>
> Yes and I'd expect to apply ignore mask on events for that directory but
> not for events on files in that directory... Even more so because this will
> be currently inconsistent wrt whether the child is dir (parent's ignore mask
> does not apply) or file (parent's ignore mask does apply).
>

Indeed. For that I used this trick in my POC:

        /* Set the mark mask, so fsnotify_parent() will find this mark */
        ovm->fsn_mark.mask = mask | FS_EVENT_ON_CHILD;
        ovm->fsn_mark.ignored_mask = mask;

It's not how users are expected to configure an ignored mask on children
but we can work the ignored mask information into the object mask, like
I already did w.r.t FS_MODIFY and get the same result without the hack.

Thanks,
Amir.

P.S. for whoever is interested, my POC is on ovl-fsnotify branch.
It seems to be working well. I am just trying to get those "ephemeral
exclude marks" to not pin the dir inodes to cache, so that those inodes
could be evicted.