Re: [patch 1/2] vfs: fix overflow in direct-io subsystem

[Andrew Morton <akpm@linux-foundation.org>]

On Wed, 13 Oct 2010 22:45:57 +0200
Edward Shishkin <edward.shishkin@gmail.com> wrote:

> Fix up overflow (ssize_t->int) in the direct-io subsystem.
> 
> Signed-off-by: Edward Shishkin <edward@redhat.com>
> ---
>  fs/direct-io.c |    2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> --- linux-2.6.36-rc7.orig/fs/direct-io.c
> +++ linux-2.6.36-rc7/fs/direct-io.c
> @@ -218,7 +218,7 @@ static struct page *dio_get_page(struct 
>   * filesystems can use it to hold additional state between get_block calls and
>   * dio_complete.
>   */
> -static int dio_complete(struct dio *dio, loff_t offset, int ret, bool is_async)
> +static ssize_t dio_complete(struct dio *dio, loff_t offset, ssize_t ret, bool is_async)
>  {
>  	ssize_t transferred = 0;
>  

I'd call this a truncation error, not an overflow.  Semantics.

So what's the runtime effect?  The subsystem will go stupid when doing
a single transfer of over 2G on a 64-bit machine?

Can this actually happen in practice?  Has nobody ever done this before
now?

<goes to test it>

<discovers that ubuntu 8.04's dd doesn't support conv=direct>

<stupid thing>