Re: [RFC] Patch: aliasing bug in blockdev-in-pagecache?

[Andrew Morton <akpm@zip.com.au>]

"Stephen C. Tweedie" wrote:
> 
> ...
> --- fs/buffer.c.~1~     Fri Apr 12 17:59:09 2002
> +++ fs/buffer.c Sat Apr 13 21:09:36 2002
> @@ -1902,9 +1902,14 @@
>         }
> 
>         /* Stage 3: start the IO */
> -       for (i = 0; i < nr; i++)
> -               submit_bh(READ, arr[i]);
> -
> +       for (i = 0; i < nr; i++) {
> +               struct buffer_head * bh = arr[i];
> +               if (buffer_uptodate(bh))
> +                       end_buffer_io_async(bh, 1);
> +               else
> +                       submit_bh(READ, bh);
> +       }
> +
>         return 0;
>  }
> 

Agree.  I have debug checks for this in the 2.5 code and
they do not come out in normal use, so no probs there.

I'm starting to tighten all of this up in the patches
which I'm working on.  Move the debug checks out of
ll_rw_block and into submit_bh.  submit_bh will complain about

- writing non-uptodate buffer
- reading uptodate buffer
- reading dirty buffer
- reading or writing unmapped buffer
- anything else I can think of.

This means that a number of callers need to be changed
to set the bits correctly, which is pointless work.

But really, I think we need that formality - we keep on
making obscure mistakes in this area and the costs are
high.

-