Re: [RFC/PATCH] revoke/frevoke system calls

[Andrew Morton <akpm@osdl.org>]

On Sat, 22 Jul 2006 09:22:37 +0300 (EEST)
Pekka J Enberg <penberg@cs.Helsinki.FI> wrote:

> On Thu, 20 Jul 2006 15:07:30 +0300 (EEST)
> Pekka J Enberg <penberg@cs.Helsinki.FI> wrote:
> > > This patch implements the revoke(2) and frevoke(2) system calls for all
> > > types of files.
> > >
> > > ...
> > >
> > > -	file = fget_light(fd, &fput_needed);
> > > +	file = fget(fd);
>  
> On Fri, 21 Jul 2006, Andrew Morton wrote:
> > This is sad.
> 
> There are alternatives, playing games with ->f_op, creating fake struct 
> file, and doing IS_REVOKED if-else in the paths, but I think this is by 
> far the simplest way to do it. So in the Andrew scale of sads, how 
> sad is it, exactly?-)

Sad enough.  Certainly worth an if-else to fix.

> On Thu, 20 Jul 2006 15:07:30 +0300 (EEST)
> Pekka J Enberg <penberg@cs.Helsinki.FI> wrote:
> > > +static int revoke_files(struct task_struct *owner, struct inode *inode,
> > > +			struct file *exclude, struct list_head *to_close)
> > > +{
> > > ...
> > > +	spin_lock(&files->file_lock);
> > > ...
> > > +		revoked = kmalloc(sizeof(*revoked), GFP_KERNEL);
>  
> On Fri, 21 Jul 2006, Andrew Morton wrote:
> > This is bad.
> 
> Indeed, I'll come up with a better one as soon as I sort out the mmap 
> takedown issues.
> 

Why is this approach so different from Tigran's, I wonder.

iirc, one of the things we added file.f_mapping for was revokation, but
this patch doesn't use it.  Please ask Al Viro about this.