RDMA/cma: Fix unbalanced cm_id reference count during address resolve

Below commit missed the AF_IB and loopback code flow in
rdma_resolve_addr().  This leads to an unbalanced cm_id refcount in
cma_work_handler() which puts the refcount which was not incremented prior
to queuing the work.

A call trace is observed with such code flow:

 BUG: unable to handle kernel NULL pointer dereference at (null)
 [<ffffffff96b67e16>] __mutex_lock_slowpath+0x166/0x1d0
 [<ffffffff96b6715f>] mutex_lock+0x1f/0x2f
 [<ffffffffc0beabb5>] cma_work_handler+0x25/0xa0
 [<ffffffff964b9ebf>] process_one_work+0x17f/0x440
 [<ffffffff964baf56>] worker_thread+0x126/0x3c0

Hence, hold the cm_id reference when scheduling the resolve work item.

Fixes: 722c7b2bfead ("RDMA/{cma, core}: Avoid callback on rdma_addr_cancel()")
Link: https://lore.kernel.org/r/20200126142652.104803-2-leon@kernel.org
Signed-off-by: Parav Pandit <parav@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Reviewed-by: Jason Gunthorpe <jgg@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>

diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c
index 55a9afa..72f0321 100644 (file)
--- a/drivers/infiniband/core/cma.c
+++ b/drivers/infiniband/core/cma.c
@@ -3148,6 +3148,7 @@ static int cma_resolve_loopback(struct rdma_id_private *id_priv)
        rdma_addr_get_sgid(&id_priv->id.route.addr.dev_addr, &gid);
        rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, &gid);
 
+       atomic_inc(&id_priv->refcount);
        cma_init_resolve_addr_work(work, id_priv);
        queue_work(cma_wq, &work->work);
        return 0;
@@ -3174,6 +3175,7 @@ static int cma_resolve_ib_addr(struct rdma_id_private *id_priv)
        rdma_addr_set_dgid(&id_priv->id.route.addr.dev_addr, (union ib_gid *)
                &(((struct sockaddr_ib *) &id_priv->id.route.addr.dst_addr)->sib_addr));
 
+       atomic_inc(&id_priv->refcount);
        cma_init_resolve_addr_work(work, id_priv);
        queue_work(cma_wq, &work->work);
        return 0;