RE: [PATCH rdma-next v1 14/15] RDMA/core: Allow port_groups to be used with namespaces

[Parav Pandit]

> From: Jason Gunthorpe <jgg@xxxxxxxxxx>
> Sent: Tuesday, June 8, 2021 5:02 AM
> 
> On Mon, Jun 07, 2021 at 01:29:58PM +0000, Parav Pandit wrote:
> >
> >
> > > From: Leon Romanovsky <leon@xxxxxxxxxx>
> > > Sent: Monday, June 7, 2021 1:48 PM
> > >
> > > From: Jason Gunthorpe <jgg@xxxxxxxxxx>
> > >
> > > Now that the port_groups data is being destroyed and managed by the
> > > core code this restriction is no longer needed. All the
> > > ib_port_attrs are compatible with the core's sysfs lifecycle.
> > >
> > > Signed-off-by: Jason Gunthorpe <jgg@xxxxxxxxxx>
> > > Signed-off-by: Leon Romanovsky <leonro@xxxxxxxxxx>
> > > drivers/infiniband/core/device.c | 10 ++++------
> > > drivers/infiniband/core/sysfs.c  | 17 ++++++-----------
> > >  2 files changed, 10 insertions(+), 17 deletions(-)
> > >
> > > diff --git a/drivers/infiniband/core/device.c
> > > b/drivers/infiniband/core/device.c
> > > index 2cbd77933ea5..92f224a97481 100644
> > > +++ b/drivers/infiniband/core/device.c
> > > @@ -1698,13 +1698,11 @@ int ib_device_set_netns_put(struct sk_buff
> > > *skb,
> > >  	}
> > >
> > >  	/*
> > > -	 * Currently supported only for those providers which support
> > > -	 * disassociation and don't do port specific sysfs init. Once a
> > > -	 * port_cleanup infrastructure is implemented, this limitation will be
> > > -	 * removed.
> > > +	 * All the ib_clients, including uverbs, are reset when the
> > > +namespace
> > > is
> > > +	 * changed and this cannot be blocked waiting for userspace to do
> > > +	 * something, so disassociation is mandatory.
> > >  	 */
> > > -	if (!dev->ops.disassociate_ucontext || dev->ops.port_groups ||
> > > -	    ib_devices_shared_netns) {
> 
> So this is OK since we have the clean up now
> 
I didn't review the whole series, but yes if there is sysfs clean routine, it should be ok.

> > > +	if (!dev->ops.disassociate_ucontext || ib_devices_shared_netns) {
> > >  		ret = -EOPNOTSUPP;
> > >  		goto ns_err;
> > >  	}
> > > diff --git a/drivers/infiniband/core/sysfs.c
> > > b/drivers/infiniband/core/sysfs.c index 09a2e1066df0..f42034fcf3d9
> > > 100644
> > > +++ b/drivers/infiniband/core/sysfs.c
> > > @@ -1236,11 +1236,9 @@ static struct ib_port *setup_port(struct
> > > ib_core_device *coredev, int port_num,
> > >  	ret = sysfs_create_groups(&p->kobj, p->groups_list);
> > >  	if (ret)
> > >  		goto err_del;
> > > -	if (is_full_dev) {
> > > -		ret = sysfs_create_groups(&p->kobj, device-
> > > >ops.port_groups);
> > > -		if (ret)
> > > -			goto err_groups;
> > > -	}
> > > +	ret = sysfs_create_groups(&p->kobj, device->ops.port_groups);
> > > +	if (ret)
> > > +		goto err_groups;
> >
> > This will expose counters in all net namespaces in shared mode
> > (default case).  Application running in one net namespace will be able
> > to monitor counters of other net namespace.  This should be avoided.
> 
> And you want this to stay blocked because the port_groups mostly contain
> counters?
Yes.