On Fri, May 05, 2023 at 02:24:04PM +0800, ye.xingchen@xxxxxxxxxx wrote:
> From: Ye Xingchen <ye.xingchen@xxxxxxxxxx>
>
> convert the fget()/fput() uses to fdget()/fdput().
>
> Signed-off-by: Ye Xingchen <ye.xingchen@xxxxxxxxxx>
> ---
> drivers/vhost/net.c | 10 +++++-----
> 1 file changed, 5 insertions(+), 5 deletions(-)
>
> diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c
> index ae2273196b0c..5b3fe4805182 100644
> --- a/drivers/vhost/net.c
> +++ b/drivers/vhost/net.c
> @@ -1466,17 +1466,17 @@ static struct ptr_ring *get_tap_ptr_ring(struct file *file)
>
> static struct socket *get_tap_socket(int fd)
> {
> - struct file *file = fget(fd);
> + struct fd f = fdget(fd);
> struct socket *sock;
>
> - if (!file)
> + if (!f.file)
> return ERR_PTR(-EBADF);
> - sock = tun_get_socket(file);
> + sock = tun_get_socket(f.file);
> if (!IS_ERR(sock))
> return sock;
> - sock = tap_get_socket(file);
> + sock = tap_get_socket(f.file);
> if (IS_ERR(sock))
> - fput(file);
> + fdput(f);
> return sock;
NAK. For the same reason why the sockfd_lookup() counterpart of that
patch is broken. After your change there's no way for the caller
to tell whether we have bumped the refcount of file in question;
this can't possibly work.
