Re: [PATCH -mm] Fix lseek on /proc/kcore

[Alexey Dobriyan <adobriyan@sw.ru>]

On Thu, Mar 22, 2007 at 02:14:35PM +0100, Arjan van de Ven wrote:
> On Thu, 2007-03-22 at 12:56 +0300, Alexey Dobriyan wrote:
> > --- a/fs/proc/inode.c
> > +++ b/fs/proc/inode.c
> > @@ -167,8 +167,9 @@ static loff_t proc_reg_llseek(struct fil
> >  	llseek = pde->proc_fops->llseek;
> >  	spin_unlock(&pde->pde_unload_lock);
> >
> > -	if (llseek)
> > -		rv = llseek(file, offset, whence);
> > +	if (!llseek)
> > +		llseek = default_llseek;
> > +	rv = llseek(file, offset, whence);
> >
>
> this has potential impact way outside kcore......
>
> did you audit all proc users to see if they can deal with lseek?

Mainline deals with lseek on proc entries as follows:
* use default_llseek()
* but if proc entry set ->llseek via ->proc_fops, use custom llseek

With introduction of proxying, ->llseek was suddenly set on all proc
entries, so default_llseek() was never used, but -E started to be
returned for all of them that were relying on default_llseek().

So this patch brings proc_reg_llseek() in sync with vfs_llseek().