Re: [PATCH v2 3/3] virtiofs: emit uevents on filesystem events

Stefan Hajnoczi <[email protected]> Fri, 9 Feb 2024 07:15:32 -0500
Newsgroups	dev.linux.lists.virtio-fs,org.kernel.vger.linux-fsdevel
Message-ID	<20240209121532.GC748645@fedora>
On Fri, Feb 09, 2024 at 10:39:04AM +0000, Greg KH wrote:
> On Thu, Feb 08, 2024 at 02:32:11PM -0500, Stefan Hajnoczi wrote:
> > Alyssa Ross <[email protected]> requested that virtiofs notifies userspace
> > when filesytems become available. This can be used to detect when a
> > filesystem with a given tag is hotplugged, for example. uevents allow
> > userspace to detect changes without resorting to polling.
> > 
> > The tag is included as a uevent property so it's easy for userspace to
> > identify the filesystem in question even when the sysfs directory goes
> > away during removal.
> > 
> > Here are example uevents:
> > 
> >   # udevadm monitor -k -p
> > 
> >   KERNEL[111.113221] add      /fs/virtiofs/2 (virtiofs)
> >   ACTION=add
> >   DEVPATH=/fs/virtiofs/2
> >   SUBSYSTEM=virtiofs
> >   TAG=test
> > 
> >   KERNEL[165.527167] remove   /fs/virtiofs/2 (virtiofs)
> >   ACTION=remove
> >   DEVPATH=/fs/virtiofs/2
> >   SUBSYSTEM=virtiofs
> >   TAG=test
> > 
> > Signed-off-by: Stefan Hajnoczi <[email protected]>
> > ---
> >  fs/fuse/virtio_fs.c | 15 +++++++++++++++
> >  1 file changed, 15 insertions(+)
> > 
> > diff --git a/fs/fuse/virtio_fs.c b/fs/fuse/virtio_fs.c
> > index 28e96b7cde00..18a8f531e5d4 100644
> > --- a/fs/fuse/virtio_fs.c
> > +++ b/fs/fuse/virtio_fs.c
> > @@ -270,6 +270,17 @@ static void virtio_fs_start_all_queues(struct virtio_fs *fs)
> >  	}
> >  }
> >  
> > +static void virtio_fs_uevent(struct virtio_fs *fs, enum kobject_action action)
> > +{
> > +	char tag_str[sizeof("TAG=") +
> > +		     sizeof_field(struct virtio_fs_config, tag) + 1];
> > +	char *envp[] = {tag_str, NULL};
> > +
> > +	snprintf(tag_str, sizeof(tag_str), "TAG=%s", fs->tag);
> > +
> > +	kobject_uevent_env(&fs->kobj, action, envp);
> > +}
> > +
> >  /* Add a new instance to the list or return -EEXIST if tag name exists*/
> >  static int virtio_fs_add_instance(struct virtio_device *vdev,
> >  				  struct virtio_fs *fs)
> > @@ -309,6 +320,8 @@ static int virtio_fs_add_instance(struct virtio_device *vdev,
> >  
> >  	mutex_unlock(&virtio_fs_mutex);
> >  
> > +	virtio_fs_uevent(fs, KOBJ_ADD);
> 
> Why do you have to explicitly ask for the event?  Doesn't sysfs trigger
> this for you automatically?  Set the kset uevent callback for this,
> right?

I haven't found a way to get an implicit KOBJ_ADD uevent. device_add()
and other kset_uevent_ops users emit KOBJ_ADD manually too. Grepping for
KOBJ_ADD in fs/sysfs/ and lib/ doesn't produce any useful results
either.

It is possible to eliminate the explicit KOBJ_REMOVE though because
kobject_del() already calls it. I will fix that and switch to
kset_uevent_ops->uevent().

Thanks,
Stefan
signature.asc (application/pgp-signature, 488 B)