[PATCH 1/2] ksmbd: detect RDMA capable lower devices when bridge and vlan netdev is used

[PATCH 1/2] ksmbd: detect RDMA capable lower devices when bridge and vlan netdev is used

[Namjae Jeon]

If user set bridge interface as actual RDMA-capable NICs are lower devices,
ksmbd can not detect as RDMA capable. This patch can detect the RDMA
capable lower devices from bridge master or VLAN. With this change, ksmbd
can accept both TCP and RDMA connections through the same bridge IP
address, allowing mixed transport operation without requiring separate
interfaces.

Signed-off-by: Namjae Jeon <linkinjeon@kernel.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
---
 fs/smb/server/transport_rdma.c | 20 +++++++++++++++++++-
 1 file changed, 19 insertions(+), 1 deletion(-)

diff --git a/fs/smb/server/transport_rdma.c b/fs/smb/server/transport_rdma.c
index 7d86553fcc7c..4a8aeb1df0cc 100644
--- a/fs/smb/server/transport_rdma.c
+++ b/fs/smb/server/transport_rdma.c
@@ -2606,7 +2606,7 @@ void ksmbd_rdma_destroy(void)
 	}
 }
 
-bool ksmbd_rdma_capable_netdev(struct net_device *netdev)
+static bool ksmbd_find_rdma_capable_netdev(struct net_device *netdev)
 {
 	struct smb_direct_device *smb_dev;
 	int i;
@@ -2648,6 +2648,24 @@ bool ksmbd_rdma_capable_netdev(struct net_device *netdev)
 	return rdma_capable;
 }
 
+bool ksmbd_rdma_capable_netdev(struct net_device *netdev)
+{
+	struct net_device *lower_dev;
+	struct list_head *iter;
+
+	if (ksmbd_find_rdma_capable_netdev(netdev))
+		return true;
+
+	/* check if netdev is bridge or VLAN */
+	if (netif_is_bridge_master(netdev) ||
+	    netdev->priv_flags & IFF_802_1Q_VLAN)
+		netdev_for_each_lower_dev(netdev, lower_dev, iter)
+			if (ksmbd_find_rdma_capable_netdev(lower_dev))
+				return true;
+
+	return false;
+}
+
 static const struct ksmbd_transport_ops ksmbd_smb_direct_transport_ops = {
 	.prepare	= smb_direct_prepare,
 	.disconnect	= smb_direct_disconnect,
-- 
2.25.1

[PATCH 2/2] ksmbd: detect RDMA capable netdevs include IPoIB

[Namjae Jeon]

Current ksmbd_rdma_capable_netdev fails to mark certain RDMA-capable
inerfaces such as IPoIB as RDMA capable after reverting GUID matching code
due to layer violation.
This patch check the ARPHRD_INFINIBAND type safely identifies an IPoIB
interface without introducing a layer violation, ensuring RDMA
functionality is correctly enabled for these interfaces.

Signed-off-by: Namjae Jeon <linkinjeon@kernel.org>
Signed-off-by: Steve French <stfrench@microsoft.com>
---
 fs/smb/server/transport_rdma.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/fs/smb/server/transport_rdma.c b/fs/smb/server/transport_rdma.c
index 4a8aeb1df0cc..5d3b48e77012 100644
--- a/fs/smb/server/transport_rdma.c
+++ b/fs/smb/server/transport_rdma.c
@@ -2663,6 +2663,10 @@ bool ksmbd_rdma_capable_netdev(struct net_device *netdev)
 			if (ksmbd_find_rdma_capable_netdev(lower_dev))
 				return true;
 
+	/* check if netdev is IPoIB safely without layer violation */
+	if (netdev->type == ARPHRD_INFINIBAND)
+		return true;
+
 	return false;
 }
 
-- 
2.25.1