Re: [Qemu-devel] [PATCH 1/3] char: fix vhost-user socket full

[Marc-André Lureau]

Hi

On Thu, Dec 3, 2015 at 10:53 AM, Didier Pallard
<address@hidden> wrote:
> unix_send_msgfds is used by vhost-user control socket. qemu_chr_fe_write_all
> is used to send a message and retries as long as EAGAIN errno is set,
> but write_msgfds buffer is freed after first EAGAIN failure, causing
> message to be sent without proper fds attachment.
>
> In case unix_send_msgfds is called through qemu_chr_fe_write, it will be
> user responsability to resend message as is or to free write_msgfds
> using set_msgfds(0)
>
> Signed-off-by: Didier Pallard <address@hidden>
> Reviewed-by: Thibaut Collet <address@hidden>
> ---
>  qemu-char.c | 10 ++++++++++
>  1 file changed, 10 insertions(+)
>
> diff --git a/qemu-char.c b/qemu-char.c
> index 5448b0f..26d5f2e 100644
> --- a/qemu-char.c
> +++ b/qemu-char.c
> @@ -2614,6 +2614,16 @@ static int unix_send_msgfds(CharDriverState *chr, 
> const uint8_t *buf, int len)
>          r = sendmsg(s->fd, &msgh, 0);
>      } while (r < 0 && errno == EINTR);
>
> +    /* Ancillary data are not sent if no byte is written
> +     * so don't free msgfds buffer if return value is EAGAIN
> +     * If called from qemu_chr_fe_write_all retry will come soon
> +     * If called from qemu_chr_fe_write, it is the user responsibility
> +     * to resend message or free fds using set_msgfds(0)
> +     */
> +    if (r < 0 && errno == EAGAIN) {
> +        return r;
> +    }
> +

This looks reasonable to me. However, I don't know what happens with
partial write of ancillary data. Hopefully it's all or nothing.
Apparently, reading unix_stream_sendmsg() in kernel shows that as long
as a few bytes have been sent, the ancillary data is sent. So it looks
like it still does the right thing in case of a partial write.

Reviewed-by: Marc-André Lureau <address@hidden>

>      /* free the written msgfds, no matter what */
>      if (s->write_msgfds_num) {
>          g_free(s->write_msgfds);
> --
> 2.1.4
>
>



-- 
Marc-André Lureau