IB/mad: Fix use after free when destroying MAD agent
authorShay Drory <shayd@mellanox.com>
Sun, 21 Jun 2020 10:47:35 +0000 (13:47 +0300)
committerSasha Levin <sashal@kernel.org>
Tue, 30 Jun 2020 19:36:52 +0000 (15:36 -0400)
commit 116a1b9f1cb769b83e5adff323f977a62b1dcb2e upstream.

Currently, when RMPP MADs are processed while the MAD agent is destroyed,
it could result in use after free of rmpp_recv, as decribed below:

cpu-0 cpu-1
----- -----
ib_mad_recv_done()
 ib_mad_complete_recv()
  ib_process_rmpp_recv_wc()
unregister_mad_agent()
 ib_cancel_rmpp_recvs()
  cancel_delayed_work()
   process_rmpp_data()
    start_rmpp()
     queue_delayed_work(rmpp_recv->cleanup_work)
  destroy_rmpp_recv()
   free_rmpp_recv()
     cleanup_work()[1]
      spin_lock_irqsave(&rmpp_recv->agent->lock) <-- use after free

[1] cleanup_work() == recv_cleanup_handler

Fix it by waiting for the MAD agent reference count becoming zero before
calling to ib_cancel_rmpp_recvs().

Fixes: 9a41e38a467c ("IB/mad: Use IDR for agent IDs")
Link: https://lore.kernel.org/r/20200621104738.54850-2-leon@kernel.org
Signed-off-by: Shay Drory <shayd@mellanox.com>
Reviewed-by: Maor Gottlieb <maorg@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/infiniband/core/mad.c

index 9947d16..a9e00cd 100644 (file)
@@ -639,10 +639,10 @@ static void unregister_mad_agent(struct ib_mad_agent_private *mad_agent_priv)
        xa_erase(&ib_mad_clients, mad_agent_priv->agent.hi_tid);
 
        flush_workqueue(port_priv->wq);
-       ib_cancel_rmpp_recvs(mad_agent_priv);
 
        deref_mad_agent(mad_agent_priv);
        wait_for_completion(&mad_agent_priv->comp);
+       ib_cancel_rmpp_recvs(mad_agent_priv);
 
        ib_mad_agent_security_cleanup(&mad_agent_priv->agent);