diff mbox series

RDMA/cma: Teach lockdep about the order of rtnl and lock

Message ID 20200227203651.GA27185@ziepe.ca (mailing list archive)
State Mainlined
Commit 32ac9e4399b12d3e54d312a0e0e30ed5cd19bd4e
Delegated to: Jason Gunthorpe
Headers show
Series RDMA/cma: Teach lockdep about the order of rtnl and lock | expand

Commit Message

Jason Gunthorpe Feb. 27, 2020, 8:36 p.m. UTC
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

Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
---
 drivers/infiniband/core/cma.c | 13 +++++++++++++
 1 file changed, 13 insertions(+)

This will trigger siw to fail on module load, maybe other drivers too.

Comments

Jason Gunthorpe March 10, 2020, 5:27 p.m. UTC | #1
On Thu, Feb 27, 2020 at 04:36:51PM -0400, Jason Gunthorpe wrote:
> 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
> 
> Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
> ---
>  drivers/infiniband/core/cma.c | 13 +++++++++++++
>  1 file changed, 13 insertions(+)

Lets give it a go, applied to for-next

Jason
diff mbox series

Patch

diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c
index 5165158a7aaa7d..c623d54ac9f944 100644
--- a/drivers/infiniband/core/cma.c
+++ b/drivers/infiniband/core/cma.c
@@ -4789,6 +4789,19 @@  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;