Re: [RFC] readdir mess

[Linus Torvalds <torvalds@linux-foundation.org>]

On Tue, 12 Aug 2008, Al Viro wrote:
> 
> Doesn't work well for readdir(2)...

Sure it does.

> > 	error = vfs_readdir(file, filldir, &buf);
> > 	lastdirent = buf.previous;
> > 	if (lastdirent) {
> > 		error = count - buf.count;
> > 		if (put_user(file->f_pos, &lastdirent->d_off))
> > 			error = -EFAULT;
> > 	}
> > 	fput(file);
> > 	return error;
> > 
> >    and we wouldn't need any other logic at all.
> 
> you've just lost e.g. -EIO for getdents().

No I've not.

If we returned a partial result, we _should_ return a partial result.

And if we got EIO on the first entry, we should return EIO.

The _current_ code is crap. It sometimes returns the error (if the 
->readdir() function returned error), and sometimes returns the partial 
result (if the "buf.error" was set).

> Frankly, I'd rather keep ->readdir() instances simpler.  There are far
> more of those, for one thing.  As it is, we only have "stop"/"continue"
> ->readdir() has to care about...

Keeping them simple (and not changing them - always returning zero is what 
the _original_ readdir() thing did!) is why the current situation exists.

So if we keep it that way, then we really *KEEP* it that way. Don't go 
around changing any of the existing rules. Just make sure that the 
callbacks keep on always returning negative or zero (and never positive).

			Linus