Re: [PATCH] ext4: add generic uevent infrastructure

Re: [PATCH] ext4: add generic uevent infrastructure

[Christoph Hellwig]

On Mon, Nov 18, 2013 at 06:38:40PM +0400, Dmitry Monakhov wrote:
> *Purpose:
> It is reasonable to annaunce fs related events via uevent infrastructure.
> This patch implement only ext4'th part, but IMHO this should be usefull for
> any generic filesystem.

It does indeed look very generic.  How about you try to redo it to sit
at the VFS level?

Also Jan Kara has done quota netlink notifications a while ago, which
fit into the same sort of niche.

Re: [PATCH] ext4: add generic uevent infrastructure

[Dave Chinner]

On Mon, Nov 18, 2013 at 08:20:45AM -0800, Christoph Hellwig wrote:
> On Mon, Nov 18, 2013 at 06:38:40PM +0400, Dmitry Monakhov wrote:
> > *Purpose:
> > It is reasonable to annaunce fs related events via uevent infrastructure.
> > This patch implement only ext4'th part, but IMHO this should be usefull for
> > any generic filesystem.
> 
> It does indeed look very generic.  How about you try to redo it to sit
> at the VFS level?

I certain agree with that. I'd also like to see ENOSPC notifications
as that would obliviate the need for distros like RHEL to ship
systemtap scripts to generate such notifications for admins....

> Also Jan Kara has done quota netlink notifications a while ago, which
> fit into the same sort of niche.

The question I'm asking is whether we really want a new interface
for these events? Shouldn't we really try to use an existing
filesystem event interface for generating these events (e.g.
fanotify) rather than adding yet another disjoint filesystem event
interface to the kernel?

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

Re: [PATCH] ext4: add generic uevent infrastructure

[Dmitry Monakhov]

On Tue, 19 Nov 2013 12:36:21 +1100, Dave Chinner <david@fromorbit.com> wrote:
> On Mon, Nov 18, 2013 at 08:20:45AM -0800, Christoph Hellwig wrote:
> > On Mon, Nov 18, 2013 at 06:38:40PM +0400, Dmitry Monakhov wrote:
> > > *Purpose:
> > > It is reasonable to annaunce fs related events via uevent infrastructure.
> > > This patch implement only ext4'th part, but IMHO this should be usefull for
> > > any generic filesystem.
> > 
> > It does indeed look very generic.  How about you try to redo it to sit
> > at the VFS level?
> 
> I certain agree with that. I'd also like to see ENOSPC notifications
> as that would obliviate the need for distros like RHEL to ship
> systemtap scripts to generate such notifications for admins....
Good point. This may be classified as NON_FATAL_ERROR. Will add this to
next version. But this event type may require some-sort of rate limiting.
> 
> > Also Jan Kara has done quota netlink notifications a while ago, which
> > fit into the same sort of niche.
> 
> The question I'm asking is whether we really want a new interface
> for these events? Shouldn't we really try to use an existing
> filesystem event interface for generating these events (e.g.
> fanotify) rather than adding yet another disjoint filesystem event
> interface to the kernel?
ASAIU fanotify is file-oriented event interface API aimed for antivirus
and index applications. fanotify_event_set = {ACCESS_XXX,OPEN,MODIFY, CLOSE_XXX}
But I want to emmit fs-oriented events (where SB is the object)
fs_event_set = {MOUNT,UMOUNT,FREEZE,UNFREEZE,ERROR,ENOSPC}.
IMHO fs_event_set is closer to {dm,device}_events_set than
fanotify_event_set, that is why I use idea of uevent for devices.
And new fs-events look as intuitive extension of existing dev-event API.
Are you agree?
> 
> Cheers,
> 
> Dave.
> -- 
> Dave Chinner
> david@fromorbit.com
> --
> To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] ext4: add generic uevent infrastructure

[Christoph Hellwig]

On Tue, Nov 19, 2013 at 12:36:21PM +1100, Dave Chinner wrote:
> I certain agree with that. I'd also like to see ENOSPC notifications
> as that would obliviate the need for distros like RHEL to ship
> systemtap scripts to generate such notifications for admins....

The ENOSPC case would be a natural tag on to Jan's quota notification,
and I have a vague memory that someone started implementing it or
at least talked about it.

> 
> > Also Jan Kara has done quota netlink notifications a while ago, which
> > fit into the same sort of niche.
> 
> The question I'm asking is whether we really want a new interface
> for these events? Shouldn't we really try to use an existing
> filesystem event interface for generating these events

Good quetion, and the quota netlink notifications would be the natural
place to tag on at least some of this.

> (e.g.
> fanotify) rather than adding yet another disjoint filesystem event
> interface to the kernel?

It needs to be a per-fs interface, and as Dmitry pointed out fanotify
is a per-file one.  Nevermind thast it's an utterly horrible interface
that shouldn't have been merged and really should be disabled in distro
kernels..

Re: [PATCH] ext4: add generic uevent infrastructure

[Jan Kara]

On Tue 19-11-13 02:25:01, Christoph Hellwig wrote:
> On Tue, Nov 19, 2013 at 12:36:21PM +1100, Dave Chinner wrote:
> > I certain agree with that. I'd also like to see ENOSPC notifications
> > as that would obliviate the need for distros like RHEL to ship
> > systemtap scripts to generate such notifications for admins....
> 
> The ENOSPC case would be a natural tag on to Jan's quota notification,
> and I have a vague memory that someone started implementing it or
> at least talked about it.
  Yes, quota netlink interface is technically very easy to extend to also
provide ENOSPC notifications. It's just that the name of generic netlink
family is 'VFS_DQUOT' so ENOSPC notifications do not fit very well with
that name but it isn't too bad either.

> > > Also Jan Kara has done quota netlink notifications a while ago, which
> > > fit into the same sort of niche.
> > 
> > The question I'm asking is whether we really want a new interface
> > for these events? Shouldn't we really try to use an existing
> > filesystem event interface for generating these events
> 
> Good quetion, and the quota netlink notifications would be the natural
> place to tag on at least some of this.
> 
> > (e.g.
> > fanotify) rather than adding yet another disjoint filesystem event
> > interface to the kernel?
> 
> It needs to be a per-fs interface, and as Dmitry pointed out fanotify
> is a per-file one.
  I agree fanotify is really a bad fit.

								Honza
-- 
Jan Kara <jack@suse.cz>
SUSE Labs, CR