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

[Dan Carpenter <dan.carpenter@oracle.com>]

On Wed, Jun 17, 2020 at 11:13:32PM +0530, Souptick Joarder wrote:
> On Wed, Jun 17, 2020 at 4:43 PM Dan Carpenter <dan.carpenter@oracle.com> 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@gmail.com>
> > > Cc: John Hubbard <jhubbard@nvidia.com>
> > > Cc: Bharath Vedartham <linux.bhar@gmail.com>
> > > Cc: Dan Carpenter <dan.carpenter@oracle.com>
> > > ---
> > >  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.
>

   182                  kfree(acd);
   183          }
   184          return rv;
   185  
   186   err_descr_too_many:
   187          unlock_engine(ldev);
   188          dma_unmap_sg(&ldev->pldev->dev, acd->sgt.sgl, acd->sgt.nents, ldev->dir);
   189          sg_free_table(&acd->sgt);
   190   err_dma_map_sg:
   191   err_alloc_sg_table:
   192          for (i = 0 ; i < acd->page_count ; i++)
   193                  put_page(acd->user_pages[i]);
   194  
   195   err_get_user_pages:
   196          if (rv > 0) {
                    ^^^^^^
"rv" is positive.

   197                  for (i = 0; i < rv; i++)
   198                          put_pages(acd->user_pages[i])
   199          }
   200          kfree(acd->user_pages);
   201   err_alloc_userpages:
   202          kfree(acd);
   203          dev_dbg(&priv->ldev->pldev->dev, "%s returning with error %ld\n", __func__, rv);
   204          return rv;
                       ^^
"rv" is still positive but it should be -EFAULT.

   205  }

regards,
dan carpenter