Re: [Bug] io_uring_register_files_update broken

[Victor Stewart <v@nametag.social>]

On Thu, Jul 1, 2021 at 3:51 PM Pavel Begunkov <asml.silence@gmail.com> wrote:
>
> On 6/30/21 10:14 PM, Victor Stewart wrote:
> > i'm fairly confident there is something broken with
> > io_uring_register_files_update,
> > especially the offset parameter.
> >
> > when trying to update a single fd, and getting a successful result of
> > 1, proceeding
> > operations with IOSQE_FIXED_FILE fail with -9. but if i update all of
> > the fds with
> > then my recv operations succeed, but close still fails with -9.
> >
> > on Clear LInux 5.12.13-1050.native
> >
> > here's a diff for liburing send_recv test, to demonstrate this.
> >
> > diff --git a/test/send_recv.c b/test/send_recv.c
> > index 19adbdd..492b591 100644
> > --- a/test/send_recv.c
> > +++ b/test/send_recv.c
> > @@ -27,6 +27,8 @@ static char str[] = "This is a test of send and recv
> > over io_uring!";
> >  #      define io_uring_prep_recv io_uring_prep_read
> >  #endif
> >
> > +static int *fds;
> > +
> >  static int recv_prep(struct io_uring *ring, struct iovec *iov, int *sock,
> >                      int registerfiles)
> >  {
> > @@ -54,17 +56,28 @@ static int recv_prep(struct io_uring *ring, struct
> > iovec *iov, int *sock,
> >                 goto err;
> >         }
> >
> > +       fds = malloc(100 * sizeof(int));
> > +       memset(fds, 0xff, sizeof(int) * 100);
> > +
> >         if (registerfiles) {
> > -               ret = io_uring_register_files(ring, &sockfd, 1);
> > +               ret = io_uring_register_files(ring, fds, 100);
> >                 if (ret) {
> >                         fprintf(stderr, "file reg failed\n");
> >                         goto err;
> >                 }
> > -               use_fd = 0;
> > -       } else {
> > -               use_fd = sockfd;
> > +
> > +               fds[sockfd] = sockfd;
> > +               int result = io_uring_register_files_update(ring,
> > sockfd, fds, 1);
>
> s/fds/&fds[sockfd]/
>
> Does it help? io_uring_register_files_update() doesn't
> apply offset parameter to the array, it's used only as
> an internal index.

i see yes, it works it like this!

io_uring_register_files_update(&ring, fd, &(socketfds[fd]), 1);
io_uring_register_files_update(&ring, fd, &(socketfds[fd] = -1), 1);

and this behavior is clear upon a closer reading of...
https://github.com/axboe/liburing/blob/11f6d56302c177a96d7eb1df86995939a4feb736/test/file-register.c#L80

i guess it's sometimes ambiguous whether int* is requesting an array
or an actual pointer to a single int.

all good now.

>
> > +
> > +               if (result != 1)
> > +               {
> > +                       fprintf(stderr, "file update failed\n");
> > +                       goto err;
> > +               }
> >         }
> >
> > +       use_fd = sockfd;
> > +
> >         sqe = io_uring_get_sqe(ring);
> >         io_uring_prep_recv(sqe, use_fd, iov->iov_base, iov->iov_len, 0);
> >         if (registerfiles)
> >
>
> --
> Pavel Begunkov