Re: [PATCH 0/3] Fix IO error reporting on fsync()

[Andrew Morton <akpm@osdl.org>]

On Mon, 9 Oct 2006 13:40:41 +0200
Jan Kara <jack@suse.cz> wrote:

> > What about putting an address_space* into the buffer_head?  Transfer the
> > EIO state into the address_space within, say, __remove_assoc_queue()?
>   Yes, that's of course possible. But it enlarges each buffer head by 4
> bytes (or 8 on 64-bit arch).

I suspect we could get that back by removing buffer_head.b_bdev.  That's
not a trivial thing to do, but should be feasible.

We can't just do bh->b_page->mapping->host->i_sb->s_bdev because of races
with trunate, plus the general horror of it all.  But I expect that all
callers of submit_bh() have the blockdev* easily available by other means,
so adding a `struct block_device*' argument to submit_bh() would get us
there.