Re: [PATCH net-next v3 1/2] macsec: add support for IFLA_MACSEC_OFFLOAD in macsec_changelink

[Sabrina Dubroca <sd@queasysnail.net>]

2022-12-08, 06:53:18 +0000, Emeel Hakim wrote:
> 
> 
> > -----Original Message-----
> > From: Sabrina Dubroca <sd@queasysnail.net>
> > Sent: Thursday, 8 December 2022 0:04
> > To: Emeel Hakim <ehakim@nvidia.com>
> > Cc: linux-kernel@vger.kernel.org; Raed Salem <raeds@nvidia.com>;
> > davem@davemloft.net; edumazet@google.com; kuba@kernel.org;
> > pabeni@redhat.com; netdev@vger.kernel.org; atenart@kernel.org; jiri@resnulli.us
> > Subject: Re: [PATCH net-next v3 1/2] macsec: add support for
> > IFLA_MACSEC_OFFLOAD in macsec_changelink
> > 
> > External email: Use caution opening links or attachments
> > 
> > 
> > 2022-12-07, 15:52:15 +0000, Emeel Hakim wrote:
> > >
> > >
> > > > -----Original Message-----
> > > > From: Sabrina Dubroca <sd@queasysnail.net>
> > > > Sent: Wednesday, 7 December 2022 17:46
> > > > To: Emeel Hakim <ehakim@nvidia.com>
> > > > Cc: linux-kernel@vger.kernel.org; Raed Salem <raeds@nvidia.com>;
> > > > davem@davemloft.net; edumazet@google.com; kuba@kernel.org;
> > > > pabeni@redhat.com; netdev@vger.kernel.org; atenart@kernel.org;
> > > > jiri@resnulli.us
> > > > Subject: Re: [PATCH net-next v3 1/2] macsec: add support for
> > > > IFLA_MACSEC_OFFLOAD in macsec_changelink
> > > >
> > > > External email: Use caution opening links or attachments
> > > >
> > > >
> > > > 2022-12-07, 12:10:16 +0200, ehakim@nvidia.com wrote:
> > > > [...]
> > > > > +static int macsec_changelink_upd_offload(struct net_device *dev,
> > > > > +struct nlattr *data[]) {
> > > > > +     enum macsec_offload offload;
> > > > > +     struct macsec_dev *macsec;
> > > > > +
> > > > > +     macsec = macsec_priv(dev);
> > > > > +     offload = nla_get_u8(data[IFLA_MACSEC_OFFLOAD]);
> > > >
> > > > All those checks are also present in macsec_upd_offload, why not
> > > > move them into macsec_update_offload as well? (and then you don't
> > > > really need macsec_changelink_upd_offload anymore)
> > > >
> > >
> > > Right, I thought about it , but I realized that those checks are done
> > > before holding the lock in macsec_upd_offload and if I move them to
> > > macsec_update_offload I will hold the lock for a longer time , I want to minimize
> > the time of holding the lock.
> > 
> > Those couple of tests are probably lost in the noise compared to what
> > mdo_add_secy ends up doing. It also looks like a race condition between the
> > "macsec->offload == offload" test in macsec_upd_offload (outside rtnl_lock) and
> > updating macsec->offload via macsec_changelink is possible. (Currently we can
> > only change it with macsec_upd_offload (called under genl_lock) so there's no issue
> > until we add this patch)
> 
> Ack, 
> so getting rid of macsec_changelink_upd_offload and moving the locking inside macsec_update_offload
> should handle this issue

You mean moving rtnl_lock()/unlock inside macsec_update_offload?
changelink is already under rtnl_lock. Just move the checks that you
currently have in macsec_changelink_upd_offload into
macsec_update_offload, and remove them from macsec_upd_offload.

> > 
> > > > > +     if (macsec->offload == offload)
> > > > > +             return 0;
> > > > > +
> > > > > +     /* Check if the offloading mode is supported by the underlying layers */
> > > > > +     if (offload != MACSEC_OFFLOAD_OFF &&
> > > > > +         !macsec_check_offload(offload, macsec))
> > > > > +             return -EOPNOTSUPP;
> > > > > +
> > > > > +     /* Check if the net device is busy. */
> > > > > +     if (netif_running(dev))
> > > > > +             return -EBUSY;
> > > > > +
> > > > > +     return macsec_update_offload(macsec, offload); }
> > > > > +
> > > >
> > > > --
> > > > Sabrina
> > >
> > 
> > --
> > Sabrina
> 

-- 
Sabrina