From: Jason Gunthorpe jgg@mellanox.com
commit 32ac9e4399b12d3e54d312a0e0e30ed5cd19bd4e upstream.
This lock ordering only happens when bonding is enabled and a certain bonding related event fires. However, since it can happen this is a global restriction on lock ordering.
Teach lockdep about the order directly and unconditionally so bugs here are found quickly.
See https://syzkaller.appspot.com/bug?extid=55de90ab5f44172b0c90
Link: https://lore.kernel.org/r/20200227203651.GA27185@ziepe.ca Signed-off-by: Jason Gunthorpe jgg@mellanox.com Signed-off-by: Greg Kroah-Hartman gregkh@linuxfoundation.org Signed-off-by: Yang Yingliang yangyingliang@huawei.com --- drivers/infiniband/core/cma.c | 13 +++++++++++++ 1 file changed, 13 insertions(+)
diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c index e16872e..d9e3844 100644 --- a/drivers/infiniband/core/cma.c +++ b/drivers/infiniband/core/cma.c @@ -4635,6 +4635,19 @@ static int __init cma_init(void) { int ret;
+ /* + * There is a rare lock ordering dependency in cma_netdev_callback() + * that only happens when bonding is enabled. Teach lockdep that rtnl + * must never be nested under lock so it can find these without having + * to test with bonding. + */ + if (IS_ENABLED(CONFIG_LOCKDEP)) { + rtnl_lock(); + mutex_lock(&lock); + mutex_unlock(&lock); + rtnl_unlock(); + } + cma_wq = alloc_ordered_workqueue("rdma_cm", WQ_MEM_RECLAIM); if (!cma_wq) return -ENOMEM;