On Sat, Sep 23, 2017 at 05:39:28PM +0100, Al Viro wrote:
> On Fri, Sep 22, 2017 at 01:18:39AM -0400, Vitaly Mayatskikh wrote:
> > bio_map_user_iov and bio_unmap_user do unbalanced pages refcounting if
> > IO vector has small consecutive buffers belonging to the same page.
> > bio_add_pc_page merges them into one, but the page reference is never
> > dropped.
> >
> > Signed-off-by: Vitaly Mayatskikh <v.mayatskih@xxxxxxxxx>
> >
> > diff --git a/block/bio.c b/block/bio.c
> > index b38e962fa83e..10cd3b6bed27 100644
> > --- a/block/bio.c
> > +++ b/block/bio.c
> > @@ -1383,6 +1383,7 @@ struct bio *bio_map_user_iov(struct request_queue *q,
> > offset = offset_in_page(uaddr);
> > for (j = cur_page; j < page_limit; j++) {
> > unsigned int bytes = PAGE_SIZE - offset;
> > + unsigned short prev_bi_vcnt = bio->bi_vcnt;
> >
> > if (len <= 0)
> > break;
> > @@ -1397,6 +1398,13 @@ struct bio *bio_map_user_iov(struct request_queue *q,
> > bytes)
> > break;
> >
> > + /*
> > + * check if vector was merged with previous
> > + * drop page reference if needed
> > + */
> > + if (bio->bi_vcnt == prev_bi_vcnt)
> > + put_page(pages[j]);
> > +
>
> Except that now you've got double-puts on failure exits ;-/
IOW, the loop on failure exit should go through the bio, like __bio_unmap_user()
does. We *also* need to put everything left unused in pages[], but only from the
last iteration through iov_for_each().
Frankly, I would prefer to reuse the pages[], rather than append to it on each
iteration. Used iov_iter_get_pages_alloc(), actually.
