On Sat, Sep 21, 2019 at 10:22:10PM +0200, Pavel Machek wrote:
Hi!
commit ed7a01fd3fd77f40b4ef2562b966a5decd8928d2 upstream.
Tracking CM_ID resource is performed in two stages: creation of cm_id and connecting it to the cma_dev. It is needed because rdma-cm protocol exports two separate user-visible calls rdma_create_id and rdma_accept.
...
Mainline says this needs additional fix, fe9bc1644918aa1d, see below.
--- a/drivers/infiniband/core/restrack.c +++ b/drivers/infiniband/core/restrack.c @@ -209,7 +209,7 @@ void rdma_restrack_del(struct rdma_restr struct ib_device *dev; if (!res->valid)
return;
goto out;dev = res_to_dev(res); if (!dev)
# return;
This test does return, does it need to go through 'goto out', too? (I see it should not happen, but...)
@@ -222,8 +222,10 @@ void rdma_restrack_del(struct rdma_restr down_write(&dev->res.rwsem); hash_del(&res->node); res->valid = false;
- up_write(&dev->res.rwsem);
+out: if (res->task) put_task_struct(res->task);
- up_write(&dev->res.rwsem);
}
Mainline says res->task = NULL is needed there, see fe9bc1644918aa1d.
Good catch, now queued up, thanks!
greg k-h