Re: [bug report][bisected] kmemleak in rdma_core observed during blktests nvme/rdma use siw

[Jason Gunthorpe <jgg@nvidia.com>]

On Wed, May 08, 2024 at 05:31:00PM +0200, Zhu Yanjun wrote:
> 在 2024/5/8 15:08, Yi Zhang 写道:
> > So bisect shows it was introduced with below commit, please help check
> > and fix it, thanks.
> > 
> > commit f8ef1be816bf9a0c406c696368c2264a9597a994
> > Author: Chuck Lever <chuck.lever@oracle.com>
> > Date:   Mon Jul 17 11:12:32 2023 -0400
> > 
> >      RDMA/cma: Avoid GID lookups on iWARP devices
> 
> Not sure if the following can fix this problem or not.
> Please let me know the test result.
> Thanks a lot.
> 
> diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c
> index 1e2cd7c8716e..901e6c40d560 100644
> --- a/drivers/infiniband/core/cma.c
> +++ b/drivers/infiniband/core/cma.c
> @@ -715,9 +715,13 @@ cma_validate_port(struct ib_device *device, u32 port,
>                 rcu_read_lock();
>                 ndev = rcu_dereference(sgid_attr->ndev);
>                 if (!net_eq(dev_net(ndev), dev_addr->net) ||
> -                   ndev->ifindex != bound_if_index)
> +                   ndev->ifindex != bound_if_index) {
> +                       rdma_put_gid_attr(sgid_attr);
>                         sgid_attr = ERR_PTR(-ENODEV);
> +               }
>                 rcu_read_unlock();
> +               if (!IS_ERR(sgid_attr))
> +                       rdma_put_gid_attr(sgid_attr);
>                 goto out;
>         }

That does look needed regardless!

Chuck?

Jason