Re: [PATCH v6 01/30] iov_iter: Add ITER_XARRAY

[Matthew Wilcox <willy@infradead.org>]

On Fri, Apr 09, 2021 at 10:09:05AM +0100, David Howells wrote:
> Al Viro <viro@zeniv.linux.org.uk> wrote:
> 
> > > +#define iterate_all_kinds(i, n, v, I, B, K, X) {		\
> > 
> > Do you have any users that would pass different B and X?
> > 
> > > @@ -1440,7 +1665,7 @@ ssize_t iov_iter_get_pages_alloc(struct iov_iter *i,
> > >  		return v.bv_len;
> > >  	}),({
> > >  		return -EFAULT;
> > > -	})
> > > +	}), 0
> > 
> > Correction - users that might get that flavour.  This one explicitly checks
> > for xarray and doesn't get to iterate_... in that case.
> 
> This is the case for iterate_all_kinds(), but not for iterate_and_advance().
> 
> See _copy_mc_to_iter() for example: that can return directly out of the middle
> of the loop, so the X variant must drop the rcu_read_lock(), but the B variant
> doesn't need to.  You also can't just use break to get out as the X variant
> has a loop within a loop to handle iteration over the subelements of a THP.

"Why does it need a loop? bvecs can contain multi-page vectors"
"memcpy_from_page can't handle that"
"doesn't that mean that iterating over a bvec is already broken?"
"yes"