Re: [PATCH 1/1] [vfs/proc] task_nommu: fix compile failing bug because of spilt file.h

[Al Viro <viro@ZenIV.linux.org.uk>]

On Sun, May 04, 2008 at 11:12:55PM +0800, Bryan Wu wrote:
> diff --git a/fs/proc/task_nommu.c b/fs/proc/task_nommu.c
> index 6cba820..987bc69 100644
> --- a/fs/proc/task_nommu.c
> +++ b/fs/proc/task_nommu.c
> @@ -1,6 +1,7 @@
>  
>  #include <linux/mm.h>
>  #include <linux/file.h>
> +#include <linux/fdtable.h>
>  #include <linux/mount.h>
>  #include <linux/ptrace.h>
>  #include <linux/seq_file.h>

Well...  I see what it's trying to do, but it's a bullshit.  Look: it's from
        if (current->files && atomic_read(&current->files->count) > 1)
                sbytes += kobjsize(current->files);
        else
                bytes += kobjsize(current->files);
Now, what we really have is size of current->files (fixed and not too large)
+ if descriptor table is too large to be embedded,
	size of current->files->fdt
	size of current->files->fdt->fd
	size of current->files->fdt->open_fds/->close_on_exec
And the second term there can be *large*, so if we are really serious about
taking descriptor table footprint into account, we'd better take care of
that too.  I'm not sure that we want to, though - note that we do that only
on nommu targets...

So what's that code really trying to achieve?