[REGRESSION] Select hang with zero sized UDP packets

[Laura Abbott]

Hi,

Fedora received a report[1] of a unit test failing on Ruby when using the
4.7 kernel. This was a test to send a zero sized UDP packet. With the
4.7 kernel, the test now timing out on a select instead of completing.
The reduced ruby test is
  def test_udp_recvfrom_nonblock
    u1 = UDPSocket.new
    u2 = UDPSocket.new
    u1.bind("127.0.0.1", 0)
    u2.send("", 0, u1.getsockname)
    IO.select [u1]  # test gets stuck here
  ensure
    u1.close if u1
    u2.close if u2
  end


which roughly corresponds to this in C

int main()
{
        int fd1, fd2;
        struct sockaddr_in addr1;
        unsigned int len1;
        int ret;
        fd_set rfds;

        fd1 = socket(AF_INET, SOCK_DGRAM|SOCK_CLOEXEC, IPPROTO_UDP);
        fd2 = socket(AF_INET, SOCK_DGRAM|SOCK_CLOEXEC, IPPROTO_UDP);

        if (fd1 < 0 || fd2 < 0) {
                printf("socket fail");
                exit(1);
        }

        len1 = sizeof(addr1);

        memset(&addr1, 0, sizeof(addr1));
        addr1.sin_family = AF_INET;
        addr1.sin_addr.s_addr = inet_addr("127.0.0.1");
        addr1.sin_port = htons(0);
        ret = bind(fd1, (struct sockaddr *)&addr1, len1);
        if (ret < 0) {
                printf("fu %d\n", errno);
                exit(1);
        }

        ret = getsockname(fd1, (struct sockaddr *)&addr1, &len1);
        if (ret < 0) {
                printf("getsockname failed %d\n", errno);
                exit(1);
        }
        ret = sendto(fd2, "", 0, 0, (struct sockaddr *)&addr1, len1);
        if (ret < 0) {
                printf("sendto failed %d\n", errno);
                exit(1);
        }

        FD_ZERO(&rfds);
        FD_SET(fd1, &rfds);
        // hang here
        select(fd1+1, &rfds, NULL, NULL, NULL);
}


Bisection showed

commit e6afc8ace6dd5cef5e812f26c72579da8806f5ac
Author: samanthakumar <samanthaku...@google.com>
Date:   Tue Apr 5 12:41:15 2016 -0400

    udp: remove headers from UDP packets before queueing
    
    Remove UDP transport headers before queueing packets for reception.
    This change simplifies a follow-up patch to add MSG_PEEK support.
    
    Signed-off-by: Sam Kumar <samanthaku...@google.com>
    Signed-off-by: Willem de Bruijn <will...@google.com>
    Signed-off-by: David S. Miller <da...@davemloft.net>


As the offending commit. The issue is still reproducible on master
as of this morning and I don't see anything explicitly tagged in
net-next as fixing this.

Any ideas?

Thanks,
Laura

[1] https://bugzilla.redhat.com/show_bug.cgi?id=1365940