On Sat, Feb 8, 2020 at 1:48 PM Andres Freund <andres@xxxxxxxxxxx> wrote:
>
> Hi,
>
> On 2020-02-08 08:55:25 -0500, Glauber Costa wrote:
> > - A connect() call is issued (and in the backend I can choose if I use
> > uring or not)
> > - The connection is supposed to take a while to establish.
> > - I call shutdown on the file descriptor
> >
> > If io_uring is not used:
> > - connect() starts by returning EINPROGRESS as expected, and after
> > the shutdown the file descriptor is finally made ready for epoll. I
> > call getsockopt(SOL_SOCKET, SO_ERROR), and see the error (104)
> >
> > if io_uring is used:
> > - if the SQE has the IOSQE_ASYNC flag on, connect() never returns.
>
> That should be easy enough to reproduce without seastar as it sounds
> deterministic - how about modifying liburing's test/connect.c test to
> behave this way?
My plan was to work on that on Monday, but I wanted to get the message
earlier in case it was a known issue or rang an obvious bell. It seems like it's
not, so I'll stick to my plan.
>
> Hm, any chance you set O_NONBLOCK on the fd, before calling the async
> connect?
>
In fact I do the opposite, and I force-remove the O_NONBLOCK flag.
But I actually played around with it while chasing this, and I did, at
some point
set O_NONBLOCK.
This is what the seastar code for connect (without uring) looks like:
// socket is non-block here
pfd->get_file_desc().connect(sa.u.sa, sa.length());
return pfd->writeable().then([pfd]() mutable {
auto err = pfd->get_file_desc().getsockopt<int>(SOL_SOCKET, SO_ERROR);
if (err != 0) {
throw std::system_error(err, std::system_category());
}
return make_ready_future<>();
});
So it essentially issues a nonblock connect, writes for the fd to be
writeable, and then uses getsockopt to figure out what happened.
With io_uring, what I see on an unblocked socket is:
- it returns EINPROGRESS as I would expect
- it is not ever made writeable.
> Wonder if io_connect()
> file_flags = force_nonblock ? O_NONBLOCK : 0;
>
> ret = __sys_connect_file(req->file, &io->connect.address,
> req->connect.addr_len, file_flags);
> if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) {
> fully takes into account that __sys_connect_file
> err = sock->ops->connect(sock, (struct sockaddr *)address, addrlen,
> sock->file->f_flags | file_flags);
> appears to leave O_NONBLOCK set on the file in place, which'd then
> not block in the wq?
>
Isn't not-block the exact opposite of I am seeing ?
If this was really not blocking, I'd expect that to return me
something immediately, even if it was the wrong thing
> Greetings,
>
> Andres Freund
