On Tue, Mar 15, 2022 at 11:26:59PM +0100, Tobias Waldekranz wrote:

On Tue, Mar 15, 2022 at 18:33, Vladimir Oltean [off-list ref] wrote:

quoted

On Tue, Mar 15, 2022 at 01:25:37AM +0100, Tobias Waldekranz wrote:

quoted

If a port joins a bridge that it can't offload, it will fallback to
standalone mode and software bridging. In this case, we never want to
offload any FDB entries to hardware either.

Signed-off-by: Tobias Waldekranz <tobias@waldekranz.com>
---

When you resend, please send this patch separately, unless something
breaks really ugly with your MST series in place.

Sure. I found this while testing the software fallback. It prevents a
segfault in dsa_port_bridge_host_fdb_add, which (rightly, I think)
assumes that dp->bridge is valid. I feel like this should have a Fixes:
tag, but I'm not sure which commit to blame. Any suggestions?

Ok, makes sense. So far, unoffloaded bridge ports meant that the DSA
switch driver didn't have a ->port_bridge_join() implementation.
Presumably that also came along with a missing ->port_fdb_add()
implementation. So probably no NPD for the existing code paths, it is
just your unoffloaded MST support that opens up new possibilities.

Anyway, the dereference of dp->bridge first appeared in commit
c26933639b54 ("net: dsa: request drivers to perform FDB isolation")
which is still just in net-next.

quoted

quoted

 net/dsa/slave.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/net/dsa/slave.c b/net/dsa/slave.c
index a61a7c54af20..647adee97f7f 100644
--- a/net/dsa/slave.c
+++ b/net/dsa/slave.c

@@ -2624,6 +2624,9 @@ static int dsa_slave_fdb_event(struct net_device *dev,
 	if (ctx && ctx != dp)
 		return 0;
 
+	if (!dp->bridge)
+		return 0;
+
 	if (switchdev_fdb_is_dynamically_learned(fdb_info)) {
 		if (dsa_port_offloads_bridge_port(dp, orig_dev))
 			return 0;

-- 
2.25.1