Re: [PATCH net-next v2] ppp: remove ppp->closing check

On Thu, Feb 5, 2026 at 7:40 PM Paolo Abeni <[email protected]> wrote:
>
> On 2/2/26 10:21 AM, Qingfang Deng wrote:
> > The ppp->closing flag is used to test if an interface is closing down.
> > However, when .ndo_uninit() is called (where ppp->closing is set to 1),
> > dev_close() has already brought down the interface, and
> > synchronize_net() guarantees that no pending TX/RX in the network path
> > can take place. Thus, the check in the network path is unnecessary.
> >
> > For file operations - ppp_read(), ppp_write(), and ppp_poll(), can
> > normally still send or receive skbs. ppp_read() and ppp_poll() are safe
> > because ppp_dev_uninit() sets pf->dead before waking them up, causing
> > both to exit cleanly.
>
> Please report the accurate call sequence that would lead to such syscall
> complete cleanly. Also what if ndo_uninit() happens just after the
> user-space has been woken-up?

ppp_read() and ppp_poll() do not care about ppp->closing so their
behaviours remain unchanged. And since ppp_file is refcounted there
isn't a UAF issue.

>
> > ppp_write() does not check pf->dead, but
> > ppp_push() verifies that ppp->channels list is not empty before sending.
> >
> > Remove the ppp->closing check.
>
> This still feel risky to me and it's not clear which would be the
> goal/gain. It this change performance oriented? If so please included
> actual figures.

For both performance and size, but I don't have the figures yet.

>
> Thanks,
>
> Paolo
>