Re: [Patch bpf v3 3/4] net: implement ->sock_is_readable() for UDP and AF_UNIX

[John Fastabend <john.fastabend@gmail.com>]

Cong Wang wrote:
> On Thu, Oct 7, 2021 at 1:00 PM Daniel Borkmann <daniel@iogearbox.net> wrote:
> >
> > On 10/2/21 2:37 AM, Cong Wang wrote:
> > > From: Cong Wang <cong.wang@bytedance.com>
> > >
> > > Yucong noticed we can't poll() sockets in sockmap even
> > > when they are the destination sockets of redirections.
> > > This is because we never poll any psock queues in ->poll(),
> > > except for TCP. With ->sock_is_readable() now we can
> > > overwrite >sock_is_readable(), invoke and implement it for
> > > both UDP and AF_UNIX sockets.
> > >
> > > Reported-by: Yucong Sun <sunyucong@gmail.com>
> > > Cc: John Fastabend <john.fastabend@gmail.com>
> > > Cc: Daniel Borkmann <daniel@iogearbox.net>
> > > Cc: Jakub Sitnicki <jakub@cloudflare.com>
> > > Cc: Lorenz Bauer <lmb@cloudflare.com>
> > > Signed-off-by: Cong Wang <cong.wang@bytedance.com>
> > > ---
> > >   net/ipv4/udp.c      | 2 ++
> > >   net/ipv4/udp_bpf.c  | 1 +
> > >   net/unix/af_unix.c  | 4 ++++
> > >   net/unix/unix_bpf.c | 2 ++
> > >   4 files changed, 9 insertions(+)
> > >
> > > diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
> > > index 2a7825a5b842..4a7e15a43a68 100644
> > > --- a/net/ipv4/udp.c
> > > +++ b/net/ipv4/udp.c
> > > @@ -2866,6 +2866,8 @@ __poll_t udp_poll(struct file *file, struct socket *sock, poll_table *wait)
> > >           !(sk->sk_shutdown & RCV_SHUTDOWN) && first_packet_length(sk) == -1)
> > >               mask &= ~(EPOLLIN | EPOLLRDNORM);
> > >
> > > +     if (sk_is_readable(sk))
> > > +             mask |= EPOLLIN | EPOLLRDNORM;
> >
> > udp_poll() has this extra logic around first_packet_length() which drops all bad csum'ed
> > skbs. How does this stand in relation to sk_msg_is_readable()? Is this a concern as well
> > there? Maybe makes sense to elaborate a bit more in the commit message for context / future
> > reference.
> 
> We don't validate UDP checksums on sockmap RX path, so
> it is okay to leave it as it is, but it is worth a comment like
> you suggest. I will add a comment in this code.
> 
> If we really need to validate the checksum, it should be addressed
> in a separate patch(set), not in this one.
> 
> Thanks.

We should validate the checksum before creating the sk_msg so that any parsers running
on top of UDP don't parse a bad checksum payload or pass a bad checksum up to user
space. I thought there would be a check in the read_sock side, but I didn't see it.