Re: [RFC][PATCH] vfs: add closefrom(2) syscall

[Zheng Liu <gnehzuil.liu@gmail.com>]

On Tue, Apr 08, 2014 at 10:21:37AM +0200, Mateusz Guzik wrote:
> On Tue, Apr 08, 2014 at 03:12:22PM +0800, Zheng Liu wrote:
> >  
> > +int __close_fds(struct files_struct *files, int lowfd)
> > +{
> > +	struct file *file;
> > +	struct fdtable *fdt;
> > +	int fd;
> > +
> > +	if (lowfd < 0)
> > +		lowfd = 0;
> > +	spin_lock(&files->file_lock);
> > +	fdt = files_fdtable(files);
> > +	if (lowfd >= fdt->max_fds)
> > +		goto out_unlock;
> > +	for (fd = lowfd; fd < fdt->max_fds; fd++) {
> > +		file = fdt->fd[fd];
> > +		if (!file)
> > +			continue;
> > +
> > +		rcu_assign_pointer(fdt->fd[fd], NULL);
> > +		__clear_close_on_exec(fd, fdt);
> > +		__put_unused_fd(files, fd);
> > +		spin_unlock(&files->file_lock);
> > +		filp_close(file, files);
> > +		spin_lock(&files->file_lock);
> > +	}
> > +
> > +out_unlock:
> > +	spin_unlock(&files->file_lock);
> > +	return 0;
> > +}
> > +
> 
> Can't comment on the usefulness of the patch, but I would like to note:
> 
> 1. fdt could be freed after you drop the lock, but you never reload the
> pointer, thus this looks like use-after-free
> 2. most of this looks like __close_fd, maybe some parts could be moved
> to an inline function so that code duplication is reduced?

Ah, yes, my fault.  I will fix them in next version.  Thanks for
pointing it out.

Regards,
                                                - Zheng