On Mon, Jan 09, 2023 at 02:20:02PM +0200, Ido Schimmel wrote:

On Mon, Jan 09, 2023 at 01:56:53PM +0200, Vladimir Oltean wrote:

quoted

On Mon, Jan 09, 2023 at 01:43:46PM +0200, Ido Schimmel wrote:

quoted

OK, thanks for confirming. Will send a patch later this week if Tobias
won't take care of it by then. First patch will probably be [1] to make
sure we dump the correct MST state to user space. It will also make it
easier to show the problem and validate the fix.

[1]

diff --git a/net/bridge/br.c b/net/bridge/br.c
index 4f5098d33a46..f02a1ad589de 100644
--- a/net/bridge/br.c
+++ b/net/bridge/br.c

@@ -286,7 +286,7 @@ int br_boolopt_get(const struct net_bridge *br, enum br_boolopt_id opt)
 	case BR_BOOLOPT_MCAST_VLAN_SNOOPING:
 		return br_opt_get(br, BROPT_MCAST_VLAN_SNOOPING_ENABLED);
 	case BR_BOOLOPT_MST_ENABLE:
-		return br_opt_get(br, BROPT_MST_ENABLED);
+		return br_mst_is_enabled(br);

Well, this did report the correct MST state despite the incorrect static
branch state, no? The users of br_mst_is_enabled(br) are broken, not
those of br_opt_get(br, BROPT_MST_ENABLED).

I should have said "actual"/"effective" instead of "correct". IMO, it's
better to use the same conditional in the both the data and control
paths to eliminate discrepancies. Without the patch, a user will see
that MST is supposedly enabled when it is actually disabled in the data
path.

The discussion is about to get philosophical, but I don't know if it's
necessary to make a bug more widespread before fixing it..
The br_mst_used is an optimization to avoid calling br_opt_get() when
surely MST is not enabled. There should be no discrepancy between using
and not using it, if the static branch works correctly (not the case here).
I would also expect that consolidation to be part of net-next though.

quoted

Anyway, I see there's a br_mst_is_enabled() and also a br_mst_enabled()?!
One is used in the fast path and the other in the slow path. They should
probably be merged, I guess. They both exist probably because somebody
thought that the "if (!netif_is_bridge_master(dev))" test is redundant
in the fast path.

The single user of br_mst_enabled() (DSA) is not affected by the bug
(only the SW data path is), so I suggest making this consolidation in
net-next after the bug is fixed. OK?

Yes, net-next, sure.