On Fri, 12 Jun 2020 at 23:36, Alexei Starovoitov
<alexei.starovoi...@gmail.com> wrote:
>
> On Fri, Jun 12, 2020 at 9:02 AM Lorenz Bauer <l...@cloudflare.com> wrote:
> >
> > Using BPF_PROG_ATTACH on a flow dissector program supports neither flags
> > nor target_fd but accepts any value. Return EINVAL if either are non-zero.
> >
> > Signed-off-by: Lorenz Bauer <l...@cloudflare.com>
> > Fixes: b27f7bb590ba ("flow_dissector: Move out netns_bpf prog callbacks")
> > ---
> > kernel/bpf/net_namespace.c | 3 +++
> > 1 file changed, 3 insertions(+)
> >
> > diff --git a/kernel/bpf/net_namespace.c b/kernel/bpf/net_namespace.c
> > index 78cf061f8179..56133e78ae4f 100644
> > --- a/kernel/bpf/net_namespace.c
> > +++ b/kernel/bpf/net_namespace.c
> > @@ -192,6 +192,9 @@ int netns_bpf_prog_attach(const union bpf_attr *attr,
> > struct bpf_prog *prog)
> > struct net *net;
> > int ret;
> >
> > + if (attr->attach_flags || attr->target_fd)
> > + return -EINVAL;
> > +
>
> In theory it makes sense, but how did you test it?
Not properly it seems, sorry!
> test_progs -t flow
> fails 5 tests.
I spent today digging through this, and the issue is actually more annoying than
I thought. BPF_PROG_DETACH for sockmap and flow_dissector ignores
attach_bpf_fd. The cgroup and lirc2 attach point use this to make sure that the
program being detached is actually what user space expects. We actually have
tests that set attach_bpf_fd for these to attach points, which tells
me that this is
an easy mistake to make.
Unfortunately I can't come up with a good fix that seems backportable:
- Making sockmap and flow_dissector have the same semantics as cgroup
and lirc2 requires a bunch of changes (probably a new function for sockmap)
- Returning EINVAL from BPF_PROG_DETACH if attach_bpf_fd is specified
leads to a lot of churn in selftests
Is it worth just landing these fixes on bpf or bpf-next without
backporting them?
--
Lorenz Bauer | Systems Engineer
6th Floor, County Hall/The Riverside Building, SE1 7PB, UK
www.cloudflare.com
