Re: [PATCH 1/4] staging: kpc2000: Unpin partial pinned pages

From: Souptick Joarder
Date: Wed Jun 17 2020 - 13:35:20 EST


On Wed, Jun 17, 2020 at 4:43 PM Dan Carpenter <dan.carpenter@xxxxxxxxxx> wrote:
>
> On Wed, Jun 17, 2020 at 07:57:20AM +0530, Souptick Joarder wrote:
> > There is a bug, when get_user_pages() failed but partially pinned
> > pages are not unpinned. Fixed it.
> >
> > Also, int is more appropriate type for rv. Changed it.
> >
> > Signed-off-by: Souptick Joarder <jrdr.linux@xxxxxxxxx>
> > Cc: John Hubbard <jhubbard@xxxxxxxxxx>
> > Cc: Bharath Vedartham <linux.bhar@xxxxxxxxx>
> > Cc: Dan Carpenter <dan.carpenter@xxxxxxxxxx>
> > ---
> > drivers/staging/kpc2000/kpc_dma/fileops.c | 6 +++++-
> > 1 file changed, 5 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/staging/kpc2000/kpc_dma/fileops.c b/drivers/staging/kpc2000/kpc_dma/fileops.c
> > index 8975346..b136353 100644
> > --- a/drivers/staging/kpc2000/kpc_dma/fileops.c
> > +++ b/drivers/staging/kpc2000/kpc_dma/fileops.c
> > @@ -35,7 +35,7 @@ static int kpc_dma_transfer(struct dev_private_data *priv,
> > unsigned long iov_base, size_t iov_len)
> > {
> > unsigned int i = 0;
> > - long rv = 0;
> > + int rv = 0;
> > struct kpc_dma_device *ldev;
> > struct aio_cb_data *acd;
> > DECLARE_COMPLETION_ONSTACK(done);
> > @@ -193,6 +193,10 @@ static int kpc_dma_transfer(struct dev_private_data *priv,
> > put_page(acd->user_pages[i]);
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> >
> > err_get_user_pages:
> > + if (rv > 0) {
> > + for (i = 0; i < rv; i++)
> > + put_pages(acd->user_pages[i])
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>
> > + }
>
> This isn't a complete fix. "rv" is the negative error code but here we
> are returning a positive value on this path.

In case of error of get_user_pages(), it will return -errno, 0 and 3rd one is
(rv > 0 && rv != acd->page_count). When rv is -errno or 0 there is no need
to call put_pages() in error path. But for 3rd case partially mapped pages
need to unpin.

Correct me if I am missing anything.

>Also it's ugly to have
> same put_page() loop repeated twice.

Yes, I agree, but these are intermediate steps. Patch [4/4] of this series
finally did the same.

>
> It would be better to write it like this:
>
> rv = get_user_pages(iov_base, acd->page_count, FOLL_TOUCH | FOLL_WRITE | FOLL_GET, acd->user_pages, NULL);
> mmap_read_unlock(current->mm); /* release the semaphore */
> if (rv < 0)
> goto free_pages;
> if (rv != acd->page_count) {
> acd->page_count = rv;
> rv = -EFAULT;
> goto put_pages;
> }
>
> ...
>
> put_pages:
> for (i = 0 ; i < acd->page_count ; i++)
> put_pages(acd->user_pages[i]);
> free_pages:
> kfree(acd->user_pages);
>
> regards,
> dan carpenter
>