On Wed, May 19, 2021 at 4:36 PM Cong Wang <xiyou.wangcong@xxxxxxxxx> wrote:
>
> On Wed, May 19, 2021 at 2:56 PM John Fastabend <john.fastabend@xxxxxxxxx> wrote:
> >
> > Cong Wang wrote:
> > > From: Cong Wang <cong.wang@xxxxxxxxxxxxx>
> > >
> > > We use non-blocking sockets for testing sockmap redirections,
> > > and got some random EAGAIN errors from UDP tests.
> > >
> > > There is no guarantee the packet would be immediately available
> > > to receive as soon as it is sent out, even on the local host.
> > > For UDP, this is especially true because it does not lock the
> > > sock during BH (unlike the TCP path). This is probably why we
> > > only saw this error in UDP cases.
> > >
> > > No matter how hard we try to make the queue empty check accurate,
> > > it is always possible for recvmsg() to beat ->sk_data_ready().
> > > Therefore, we should just retry in case of EAGAIN.
> > >
> > > Fixes: d6378af615275 ("selftests/bpf: Add a test case for udp sockmap")
> > > Reported-by: Jiang Wang <jiang.wang@xxxxxxxxxxxxx>
> > > Cc: John Fastabend <john.fastabend@xxxxxxxxx>
> > > Cc: Daniel Borkmann <daniel@xxxxxxxxxxxxx>
> > > Cc: Jakub Sitnicki <jakub@xxxxxxxxxxxxxx>
> > > Cc: Lorenz Bauer <lmb@xxxxxxxxxxxxxx>
> > > Signed-off-by: Cong Wang <cong.wang@xxxxxxxxxxxxx>
> > > ---
> > > tools/testing/selftests/bpf/prog_tests/sockmap_listen.c | 6 +++++-
> > > 1 file changed, 5 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/tools/testing/selftests/bpf/prog_tests/sockmap_listen.c b/tools/testing/selftests/bpf/prog_tests/sockmap_listen.c
> > > index 648d9ae898d2..b1ed182c4720 100644
> > > --- a/tools/testing/selftests/bpf/prog_tests/sockmap_listen.c
> > > +++ b/tools/testing/selftests/bpf/prog_tests/sockmap_listen.c
> > > @@ -1686,9 +1686,13 @@ static void udp_redir_to_connected(int family, int sotype, int sock_mapfd,
> > > if (pass != 1)
> > > FAIL("%s: want pass count 1, have %d", log_prefix, pass);
> > >
> > > +again:
> > > n = read(mode == REDIR_INGRESS ? p0 : c0, &b, 1);
> > > - if (n < 0)
> > > + if (n < 0) {
> > > + if (errno == EAGAIN)
> > > + goto again;
> > > FAIL_ERRNO("%s: read", log_prefix);
> >
> > Needs a counter and abort logic we don't want to loop forever in the
> > case the packet is lost.
>
> It should not be lost because selftests must be self-contained,
> if the selftests could not even predict whether its own packet is
> lost or not, we would have a much bigger trouble than just this
> infinite loop.
Bugs do happen though, so if you can detect some error condition
instead of having an infinite loop, then do it.
>
> Thanks.
