netdev - Re: [PATCH net-next 1/5] ptp: unregister virtual clocks when unregistering physical clock.

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20220131124135.62khcaehujxrlhbf@skbuf>
Date:   Mon, 31 Jan 2022 14:41:35 +0200
From:   Vladimir Oltean <olteanv@...il.com>
To:     Miroslav Lichvar <mlichvar@...hat.com>
Cc:     Vinicius Costa Gomes <vinicius.gomes@...el.com>,
        netdev@...r.kernel.org, Yangbo Lu <yangbo.lu@....com>,
        Yang Yingliang <yangyingliang@...wei.com>,
        Richard Cochran <richardcochran@...il.com>
Subject: Re: [PATCH net-next 1/5] ptp: unregister virtual clocks when
 unregistering physical clock.

Hi,

On Mon, Jan 31, 2022 at 11:07:52AM +0100, Miroslav Lichvar wrote:
> On Thu, Jan 27, 2022 at 03:58:02PM -0800, Vinicius Costa Gomes wrote:
> > Miroslav Lichvar <mlichvar@...hat.com> writes:
> > 
> > > When unregistering a physical clock which has some virtual clocks,
> > > unregister the virtual clocks with it.
> 
> > I am not against this change, but I think this problem was discussed
> > before and the suggestions were to fix it differently:
> > 
> > https://lore.kernel.org/all/20210807144332.szyazdfl42abwzmd@skbuf/
> 
> Is a linked device supposed to be unregistered automatically before
> the parent? The referenced document mentions only suspending
> and resuming, nothing about unregistering.
> 
> I tried
> 
> 	device_link_add(parent, &ptp->dev, DL_FLAG_AUTOREMOVE_CONSUMER);
> 
> and also with no flags specified, but it didn't seem to do anything
> for the vclock. It was still oopsing.
> 
> Any hints?
> 
> -- 
> Miroslav Lichvar
> 

First of all, I was wrong about the device hierarchy created by the PTP
subsystem, and as such, device links won't work.

When you unbind a physical PTP clock, what you actually unbind is the
driver from its parent device (the parent->parent, i.e. what is passed
as "parent" to the parent's ptp_clock_register call).

But since the PTP subsystem doesn't register its devices with the device
hierarchy (of device_register(), it only calls the first half:
device_initialize(), it doesn't call device_add), so the dev->kobj
doesn't get added to the devices_kset. The documentation says:

| The earliest point in time when device links can be added is after
| device_add() has been called for the supplier and device_initialize()
| has been called for the consumer.

Furthermore, PTP devices don't have a driver bound to them, with probe()
and remove() callbacks. The only driver is that of the parent of the
physical PHC. So no driver for the vclock => no hook to call
ptp_vclock_unregister() from, even if a device link from ptp->dev to
parent->parent can be added.

So your solution appears to be the correct one given the structure - call
unregister_vclock() manually.

Secondly, you got the arguments in reverse: the consumer is the first
argument, and that is &ptp->dev, and the supplier is "parent".