On Wed, 2023-09-06 at 07:57 -0700, Jakub Kicinski wrote:
> On Wed, 06 Sep 2023 16:50:23 +0300 Eduard Zingerman wrote:
> > diff --git a/kernel/bpf/offload.c b/kernel/bpf/offload.c
> > index 3e4f2ec1af06..302e38bffffa 100644
> > --- a/kernel/bpf/offload.c
> > +++ b/kernel/bpf/offload.c
> > @@ -199,12 +199,11 @@ static int __bpf_prog_dev_bound_init(struct bpf_prog *prog, struct net_device *n
> > offload->netdev = netdev;
> >
> > ondev = bpf_offload_find_netdev(offload->netdev);
> > + if (bpf_prog_is_offloaded(prog->aux) && (!ondev || !ondev->offdev)) {
> > + err = -EINVAL;
> > + goto err_free;
> > + }
> > if (!ondev) {
> > - if (bpf_prog_is_offloaded(prog->aux)) {
> > - err = -EINVAL;
> > - goto err_free;
> > - }
> > -
> > /* When only binding to the device, explicitly
> > * create an entry in the hashtable.
> > */
>
> LGTM, FWIW.
Thanks, I'll wrap it up as a proper patch with a test.
>
> > With the following reasoning: for offloaded programs offload device
> > should exist and it should not be a fake device create in !ondev branch.
> >
> > Stanislav, could you please take a look? I think this is related to commit:
> > 2b3486bc2d23 ("bpf: Introduce device-bound XDP programs")
