Re: [PATCH 3/4] fix setuid sometimes wouldn't

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

On Sun, Mar 29, 2009 at 11:48:55PM +0200, Oleg Nesterov wrote:
> On 03/29, Alexey Dobriyan wrote:
> >
> > On Sat, Mar 28, 2009 at 11:21:27PM +0000, Hugh Dickins wrote:
> > >
> > > -static struct fs_struct *get_fs_struct(struct task_struct *task)
> > > +static int get_fs_path(struct task_struct *task, struct path *path, bool root)
> > >  {
> > >  	struct fs_struct *fs;
> > > +	int result = -ENOENT;
> > > +
> > >  	task_lock(task);
> > >  	fs = task->fs;
> > > -	if(fs)
> > > -		atomic_inc(&fs->count);
> > > +	if (fs) {
> > > +		read_lock(&fs->lock);
> > > +		*path = root ? fs->root : fs->pwd;
> > > +		path_get(path);
> > > +		read_unlock(&fs->lock);
> > > +		result = 0;
> > > +	}
> > >  	task_unlock(task);
> > > -	return fs;
> > > +	return result;
> > >  }
> >
> > I think it's better to open-code. "root" parameter is unnatural.
> 
> IMHO, open-coding doesn't improve readability. I am not even talking
> about code size (it doesn't matter of course, the kernel is so tiny ;).
> 
> Perhaps "enum what" is more natural compared to "bool root", but this
> helper is simple enough.
> 
> Personally, I think this patch makes sense even if it didn't fix the
> bug, it cleanups the code and makes it more readable.

Aye.  Applied, with enum, but an anonymous one; no point breeding
tags without a reason.

Another note: chroot_fs_refs() shouldn't bump refcount at all; it should
do replacements, count them and then do path_release(old_root) that many
times - after having gone through all tasks.

I'm taking fs_struct handling into a new file (fs/fs_struct.c, unless
somebody has a better name), will do that in the same patch.