[RFC,net-next,13/22] nexthop: Emit a notification when a single nexthop is replaced

From: Ido Schimmel <idosch@nvidia.com>

The notification is emitted after all the validation checks were
performed, but before the new configuration (i.e., 'struct nh_info') is
pointed at by the old shell (i.e., 'struct nexthop'). This prevents the
need to perform rollback in case the notification is vetoed.

The next patch will also emit a replace notification for all the nexthop
groups in which the nexthop is used.

Signed-off-by: Ido Schimmel <idosch@nvidia.com>
---
 net/ipv4/nexthop.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

On 9/8/20 3:10 AM, Ido Schimmel wrote:
> From: Ido Schimmel <idosch@nvidia.com>
> 
> The notification is emitted after all the validation checks were
> performed, but before the new configuration (i.e., 'struct nh_info') is
> pointed at by the old shell (i.e., 'struct nexthop'). This prevents the
> need to perform rollback in case the notification is vetoed.
> 
> The next patch will also emit a replace notification for all the nexthop
> groups in which the nexthop is used.
> 
> Signed-off-by: Ido Schimmel <idosch@nvidia.com>
> ---
>  net/ipv4/nexthop.c | 10 ++++++++++
>  1 file changed, 10 insertions(+)
> 
> diff --git a/net/ipv4/nexthop.c b/net/ipv4/nexthop.c
> index a60a519a5462..b8a4abc00146 100644
> --- a/net/ipv4/nexthop.c
> +++ b/net/ipv4/nexthop.c
> @@ -1099,12 +1099,22 @@ static int replace_nexthop_single(struct net *net, struct nexthop *old,
>  				  struct netlink_ext_ack *extack)
>  {
>  	struct nh_info *oldi, *newi;
> +	int err;
>  
>  	if (new->is_group) {
>  		NL_SET_ERR_MSG(extack, "Can not replace a nexthop with a nexthop group.");
>  		return -EINVAL;
>  	}
>  
> +	err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, new, extack);
> +	if (err)
> +		return err;
> +
> +	/* Hardware flags were set on 'old' as 'new' is not in the red-black
> +	 * tree. Therefore, inherit the flags from 'old' to 'new'.
> +	 */
> +	new->nh_flags |= old->nh_flags & (RTNH_F_OFFLOAD | RTNH_F_TRAP);

Will that always be true? ie., has h/w seen 'new' and offloaded it yet?
vs the notifier telling hardware about the change, it does its thing and
sets the flags. But I guess that creates a race between the offload and
the new data being available.

> +
>  	oldi = rtnl_dereference(old->nh_info);
>  	newi = rtnl_dereference(new->nh_info);
>  
>

On Tue, Sep 08, 2020 at 09:25:40AM -0600, David Ahern wrote:
> On 9/8/20 3:10 AM, Ido Schimmel wrote:
> > From: Ido Schimmel <idosch@nvidia.com>
> > 
> > The notification is emitted after all the validation checks were
> > performed, but before the new configuration (i.e., 'struct nh_info') is
> > pointed at by the old shell (i.e., 'struct nexthop'). This prevents the
> > need to perform rollback in case the notification is vetoed.
> > 
> > The next patch will also emit a replace notification for all the nexthop
> > groups in which the nexthop is used.
> > 
> > Signed-off-by: Ido Schimmel <idosch@nvidia.com>
> > ---
> >  net/ipv4/nexthop.c | 10 ++++++++++
> >  1 file changed, 10 insertions(+)
> > 
> > diff --git a/net/ipv4/nexthop.c b/net/ipv4/nexthop.c
> > index a60a519a5462..b8a4abc00146 100644
> > --- a/net/ipv4/nexthop.c
> > +++ b/net/ipv4/nexthop.c
> > @@ -1099,12 +1099,22 @@ static int replace_nexthop_single(struct net *net, struct nexthop *old,
> >  				  struct netlink_ext_ack *extack)
> >  {
> >  	struct nh_info *oldi, *newi;
> > +	int err;
> >  
> >  	if (new->is_group) {
> >  		NL_SET_ERR_MSG(extack, "Can not replace a nexthop with a nexthop group.");
> >  		return -EINVAL;
> >  	}
> >  
> > +	err = call_nexthop_notifiers(net, NEXTHOP_EVENT_REPLACE, new, extack);
> > +	if (err)
> > +		return err;
> > +
> > +	/* Hardware flags were set on 'old' as 'new' is not in the red-black
> > +	 * tree. Therefore, inherit the flags from 'old' to 'new'.
> > +	 */
> > +	new->nh_flags |= old->nh_flags & (RTNH_F_OFFLOAD | RTNH_F_TRAP);
> 
> Will that always be true? ie., has h/w seen 'new' and offloaded it yet?

Yes. The chain was converted to a blocking chain, so it is possible to
program the hardware inline.

> vs the notifier telling hardware about the change, it does its thing and
> sets the flags. But I guess that creates a race between the offload and
> the new data being available.
> 
> > +
> >  	oldi = rtnl_dereference(old->nh_info);
> >  	newi = rtnl_dereference(new->nh_info);
> >  
> > 
>