[PATCH RFC v1 net-next 11/12] bridge: br_vlan_fill_forward_path_mode no _UNTAG_HW for dsa

[Vladimir Oltean]

Keeping the full email body untrimmed for extra context for the newly
added people.

On Mon, Oct 14, 2024 at 09:22:07AM +0300, Nikolay Aleksandrov wrote:
> On 14/10/2024 09:18, Nikolay Aleksandrov wrote:
> > On 13/10/2024 21:55, Eric Woudstra wrote:
> >> In network setup as below:
> >>
> >>              fastpath bypass
> >>  .----------------------------------------.
> >> /                                          \
> >> |                        IP - forwarding    |
> >> |                       /                \  v
> >> |                      /                  wan ...
> >> |                     /
> >> |                     |
> >> |                     |
> >> |                   brlan.1
> >> |                     |
> >> |    +-------------------------------+
> >> |    |           vlan 1              |
> >> |    |                               |
> >> |    |     brlan (vlan-filtering)    |
> >> |    |               +---------------+
> >> |    |               |  DSA-SWITCH   |
> >> |    |    vlan 1     |               |
> >> |    |      to       |               |
> >> |    |   untagged    1     vlan 1    |
> >> |    +---------------+---------------+
> >> .         /                   \
> >>  ----->wlan1                 lan0
> >>        .                       .
> >>        .                       ^
> >>        ^                     vlan 1 tagged packets
> >>      untagged packets
> >>
> >> Now that DEV_PATH_MTK_WDMA is added to nft_dev_path_info() the forward
> >> path is filled also when ending with the mediatek wlan1, info.indev not
> >> NULL now in nft_dev_forward_path(). This results in a direct transmit
> >> instead of a neighbor transmit. This is how it should be, But this fails.
> >>
> >> br_vlan_fill_forward_path_mode() sets DEV_PATH_BR_VLAN_UNTAG_HW when
> >> filling in from brlan.1 towards wlan1. But it should be set to
> >> DEV_PATH_BR_VLAN_UNTAG in this case. Using BR_VLFLAG_ADDED_BY_SWITCHDEV
> >> is not correct. The dsa switchdev adds it as a foreign port.
> >>
> >> Use BR_VLFLAG_TAGGING_BY_SWITCHDEV to make sure DEV_PATH_BR_VLAN_UNTAG is
> >> set when there is a dsa-switch inside the bridge.
> >>
> >> Signed-off-by: Eric Woudstra <ericwouds at gmail.com>
> >> ---
> >>  net/bridge/br_private.h |  1 +
> >>  net/bridge/br_vlan.c    | 18 +++++++++++++++++-
> >>  2 files changed, 18 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/net/bridge/br_private.h b/net/bridge/br_private.h
> >> index 8da7798f9368..7d427214cc7c 100644
> >> --- a/net/bridge/br_private.h
> >> +++ b/net/bridge/br_private.h
> >> @@ -180,6 +180,7 @@ enum {
> >>  	BR_VLFLAG_MCAST_ENABLED = BIT(2),
> >>  	BR_VLFLAG_GLOBAL_MCAST_ENABLED = BIT(3),
> >>  	BR_VLFLAG_NEIGH_SUPPRESS_ENABLED = BIT(4),
> >> +	BR_VLFLAG_TAGGING_BY_SWITCHDEV = BIT(5),
> >>  };
> >>  
> >>  /**
> >> diff --git a/net/bridge/br_vlan.c b/net/bridge/br_vlan.c
> >> index 1830d7d617cd..b7877724b969 100644
> >> --- a/net/bridge/br_vlan.c
> >> +++ b/net/bridge/br_vlan.c
> >> @@ -3,6 +3,7 @@
> >>  #include <linux/netdevice.h>
> >>  #include <linux/rtnetlink.h>
> >>  #include <linux/slab.h>
> >> +#include <net/dsa.h>
> >>  #include <net/switchdev.h>
> >>  
> >>  #include "br_private.h"
> >> @@ -100,6 +101,19 @@ static void __vlan_flags_commit(struct net_bridge_vlan *v, u16 flags)
> >>  	__vlan_flags_update(v, flags, true);
> >>  }
> >>  
> >> +static inline bool br_vlan_tagging_by_switchdev(struct net_bridge *br)
> > 
> > no inline in .c files and also constify br
> > 
> >> +{
> >> +#if IS_ENABLED(CONFIG_NET_DSA)
> >> +	struct net_bridge_port *p;
> >> +
> >> +	list_for_each_entry(p, &br->port_list, list) {
> >> +		if (dsa_user_dev_check(p->dev))
> > 
> > I don't think this can change at runtime, so please keep a counter in
> > the bridge and don't walk the port list on every vlan add.
> > 
> 
> you can use an internal bridge opt (check br_private.h) with a private opt
> that's set when such device is added as a port, no need for a full counter
> obviously

To continue on Nikolay's line of thought...

Can you abstractly describe which functional behavior do you need the
bridge port to perform, rather than "it needs to be a DSA user port"?

switchdev_bridge_port_offload() has a mechanism to inform the bridge
core of extra abilities (like tx_fwd_offload). Perhaps you could modify
the DSA drivers you need to set a similar bit to inform the bridge of
their presence and ability. That would also work when the bridge port is
a LAG over a DSA user port.

Also, please also CC DSA maintainers when you use DSA API outside
net/dsa/ and drivers/net/dsa/. I am in the process of revamping the
public DSA API and would like to be in touch with changes as they are
made.

> >> +			return false;
> >> +	}
> >> +#endif
> >> +	return true;
> >> +}
> >> +
> >>  static int __vlan_vid_add(struct net_device *dev, struct net_bridge *br,
> >>  			  struct net_bridge_vlan *v, u16 flags,
> >>  			  struct netlink_ext_ack *extack)
> >> @@ -113,6 +127,8 @@ static int __vlan_vid_add(struct net_device *dev, struct net_bridge *br,
> >>  	if (err == -EOPNOTSUPP)
> >>  		return vlan_vid_add(dev, br->vlan_proto, v->vid);
> >>  	v->priv_flags |= BR_VLFLAG_ADDED_BY_SWITCHDEV;
> >> +	if (br_vlan_tagging_by_switchdev(br))
> >> +		v->priv_flags |= BR_VLFLAG_TAGGING_BY_SWITCHDEV;
> >>  	return err;
> >>  }
> >>  
> >> @@ -1491,7 +1507,7 @@ int br_vlan_fill_forward_path_mode(struct net_bridge *br,
> >>  
> >>  	if (path->bridge.vlan_mode == DEV_PATH_BR_VLAN_TAG)
> >>  		path->bridge.vlan_mode = DEV_PATH_BR_VLAN_KEEP;
> >> -	else if (v->priv_flags & BR_VLFLAG_ADDED_BY_SWITCHDEV)
> >> +	else if (v->priv_flags & BR_VLFLAG_TAGGING_BY_SWITCHDEV)
> >>  		path->bridge.vlan_mode = DEV_PATH_BR_VLAN_UNTAG_HW;
> >>  	else
> >>  		path->bridge.vlan_mode = DEV_PATH_BR_VLAN_UNTAG;
> >