4.9-stable review patch.  If anyone has any objections, please let me know.

------------------

[ Upstream commit 3dc7c7badb7502ec3e3aa817a8bdd9e53aa54c52 ]

Before returning -EPERM we should release some resources, as already done
in the other error handling path of the function.

Fixes: d8f9cc328c88 ("IB/mlx4: Mark user MR as writable if actual virtual 
memory is writable")
Signed-off-by: Christophe JAILLET <christophe.jail...@wanadoo.fr>
Signed-off-by: Jason Gunthorpe <j...@mellanox.com>
Signed-off-by: Sasha Levin <sas...@kernel.org>
---
 drivers/infiniband/hw/mlx4/mr.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/drivers/infiniband/hw/mlx4/mr.c b/drivers/infiniband/hw/mlx4/mr.c
index 0d4878efd643..ddd3182138ac 100644
--- a/drivers/infiniband/hw/mlx4/mr.c
+++ b/drivers/infiniband/hw/mlx4/mr.c
@@ -247,8 +247,11 @@ int mlx4_ib_rereg_user_mr(struct ib_mr *mr, int flags,
        }
 
        if (flags & IB_MR_REREG_ACCESS) {
-               if (ib_access_writable(mr_access_flags) && !mmr->umem->writable)
-                       return -EPERM;
+               if (ib_access_writable(mr_access_flags) &&
+                   !mmr->umem->writable) {
+                       err = -EPERM;
+                       goto release_mpt_entry;
+               }
 
                err = mlx4_mr_hw_change_access(dev->dev, *pmpt_entry,
                                               convert_access(mr_access_flags));
-- 
2.17.1



Reply via email to