On Tue, Jul 7, 2020 at 2:42 PM Daniel Borkmann <daniel@xxxxxxxxxxxxx> wrote: > > On 7/7/20 1:42 AM, Andrii Nakryiko wrote: > > On Mon, Jul 6, 2020 at 4:02 PM Stanislav Fomichev <sdf@xxxxxxxxxx> wrote: > >> > >> Implement BPF_CGROUP_INET_SOCK_RELEASE hook that triggers > >> on inet socket release. It triggers only for userspace > >> sockets, the same semantics as existing BPF_CGROUP_INET_SOCK_CREATE. > >> > >> The only questionable part here is the sock->sk check > >> in the inet_release. Looking at the places where we > >> do 'sock->sk = NULL', I don't understand how it can race > >> with inet_release and why the check is there (it's been > >> there since the initial git import). Otherwise, the > >> change itself is pretty simple, we add a BPF hook > >> to the inet_release and avoid calling it for kernel > >> sockets. > >> > >> Signed-off-by: Stanislav Fomichev <sdf@xxxxxxxxxx> > >> --- > >> include/linux/bpf-cgroup.h | 4 ++++ > >> include/uapi/linux/bpf.h | 1 + > >> kernel/bpf/syscall.c | 3 +++ > >> net/core/filter.c | 1 + > >> net/ipv4/af_inet.c | 3 +++ > >> 5 files changed, 12 insertions(+) > >> > > > > Looks good overall, but I have no idea about sock->sk NULL case. > > +1, looks good & very useful hook. For the sock->sk NULL case here's a related > discussion on why it's needed [0]. Thanks for the pointer! I'll resend a v5 with s/sock/sock_create/ you mentioned and will clean up the commit description a bit.
