On Wed, 2024-04-10 at 14:34 -0700, Martin KaFai Lau wrote:
> On 4/9/24 11:13 PM, Geliang Tang wrote:
> > From: Geliang Tang <tanggeliang@xxxxxxxxxx>
> >
> > Some tests, such as the MPTCP bpf tests, require send_recv_data
> > helper
> > to run in nonblock mode.
> >
> > This patch adds nonblock support for send_recv_data(). Check if it
> > is
> > currently in nonblock mode, and if so, ignore EWOULDBLOCK to
> > continue
> > sending and receiving.
> >
> > Signed-off-by: Geliang Tang <tanggeliang@xxxxxxxxxx>
> > ---
> > tools/testing/selftests/bpf/network_helpers.c | 9 ++++++++-
> > 1 file changed, 8 insertions(+), 1 deletion(-)
> >
> > diff --git a/tools/testing/selftests/bpf/network_helpers.c
> > b/tools/testing/selftests/bpf/network_helpers.c
> > index 137cd18ef3f2..ca16ef2b648e 100644
> > --- a/tools/testing/selftests/bpf/network_helpers.c
> > +++ b/tools/testing/selftests/bpf/network_helpers.c
> > @@ -555,6 +555,7 @@ struct send_recv_arg {
> > static void *send_recv_server(void *arg)
> > {
> > struct send_recv_arg *a = (struct send_recv_arg *)arg;
> > + int flags = fcntl(a->fd, F_GETFL);
> > ssize_t nr_sent = 0, bytes = 0;
> > char batch[1500];
> > int err = 0, fd;
> > @@ -578,6 +579,8 @@ static void *send_recv_server(void *arg)
> > if (nr_sent == -1 && errno == EINTR)
> > continue;
> > if (nr_sent == -1) {
> > + if (flags & O_NONBLOCK && errno ==
> > EWOULDBLOCK)
>
> I still don't see why it needs to be a non blocking IO. mptcp should
> work
> with blocking IO also, no? Does it really need non blocking IO to
> make
> mptcp test work? I would rather stay with blocking IO in selftest as
> much as
> possible for simplicity reason.
>
> I am afraid the root cause of the EAGAIN thread has not been figured
> out yet:
> https://lore.kernel.org/all/b3943f9a8bf595212b00e96ba850bf32893312cc.camel@xxxxxxxxxx/
>
> Lets drop patch 3 until it is understood why mptcp needs EAGAIN or
> non-blocking IO.
> It feels like there is some flakiness and it should be understood and
> avoided.
Hi Martin,
I finally found the root cause of this issue. It is indeed an MPTCP
bug. It took me a long time to debug, and the fix is here:
https://patchwork.kernel.org/project/mptcp/patch/0ccc1c26d27d6ee7be22806a97983d37c6ca548c.1715053270.git.tanggeliang@xxxxxxxxxx/
Thank you for insisting on not accepting these work around patches from
me in the user space, almost hiding a kernel bug.
-Geliang
>
> Other than the comment in patch 2, the first two patches lgtm. Please
> respin with
> the first two patches.
>
> > + continue;
> > err = -errno;
> > break;
> > }
> > @@ -599,6 +602,7 @@ static void *send_recv_server(void *arg)
> >
> > int send_recv_data(int lfd, int fd, uint32_t total_bytes)
> > {
> > + int flags = fcntl(lfd, F_GETFL);
> > ssize_t nr_recv = 0, bytes = 0;
> > struct send_recv_arg arg = {
> > .fd = lfd,
> > @@ -622,8 +626,11 @@ int send_recv_data(int lfd, int fd, uint32_t
> > total_bytes)
> > MIN(total_bytes - bytes,
> > sizeof(batch)), 0);
> > if (nr_recv == -1 && errno == EINTR)
> > continue;
> > - if (nr_recv == -1)
> > + if (nr_recv == -1) {
> > + if (flags & O_NONBLOCK && errno ==
> > EWOULDBLOCK)
> > + continue;
> > break;
> > + }
> > bytes += nr_recv;
> > }
> >
>
>
