Skip to contentSkip to conversation

mlxsw: spectrum_switchdev: Allow port enslavement to a VLAN-unaware bridge

PATCH In review Jiri Pirko 4 participants 13 messages Feb 20, 2018 – Feb 26, 2018

Subsystems (3)
lore
Share & tools
lore

Jiri Pirko

author
#0 lore raw

PATCH 1/1 mlxsw: spectrum_switchdev: Allow port enslavement to a VLAN-unaware bridge

From: Ido Schimmel <[email protected]>

Up until now we only allowed VLAN devices to be put in a VLAN-unaware bridge, but some users need the ability to enslave physical ports as well.

This is achieved by mapping the port and VID 1 to the bridge's vFID, instead of the port and the VID used by the VLAN device.

The above is valid because as long as the port is not enslaved to a bridge, VID 1 is guaranteed to be configured as PVID and egress untagged.

---
 drivers/net/ethernet/mellanox/mlxsw/spectrum_switchdev.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)
drivers/net/ethernet/mellanox/mlxsw/spectrum_switchdev.c +5 -7
index f9f53af04fe1..917663adf925 100644
@@ -1882,14 +1882,10 @@ mlxsw_sp_bridge_8021d_port_join(struct mlxsw_sp_bridge_device *bridge_device,
 				struct netlink_ext_ack *extack)
 {
 	struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan;
+	struct net_device *dev = bridge_port->dev;
 	u16 vid;
 
-	if (!is_vlan_dev(bridge_port->dev)) {
-		NL_SET_ERR_MSG_MOD(extack, "Only VLAN devices can be enslaved to a VLAN-unaware bridge");
-		return -EINVAL;
-	}
-	vid = vlan_dev_vlan_id(bridge_port->dev);
-
+	vid = is_vlan_dev(dev) ? vlan_dev_vlan_id(dev) : 1;
 	mlxsw_sp_port_vlan = mlxsw_sp_port_vlan_find_by_vid(mlxsw_sp_port, vid);
 	if (WARN_ON(!mlxsw_sp_port_vlan))
 		return -EINVAL;
@@ -1912,8 +1908,10 @@ mlxsw_sp_bridge_8021d_port_leave(struct mlxsw_sp_bridge_device *bridge_device,
 				 struct mlxsw_sp_port *mlxsw_sp_port)
 {
 	struct mlxsw_sp_port_vlan *mlxsw_sp_port_vlan;
-	u16 vid = vlan_dev_vlan_id(bridge_port->dev);
+	struct net_device *dev = bridge_port->dev;
+	u16 vid;
 
+	vid = is_vlan_dev(dev) ? vlan_dev_vlan_id(dev) : 1;
 	mlxsw_sp_port_vlan = mlxsw_sp_port_vlan_find_by_vid(mlxsw_sp_port, vid);
 	if (WARN_ON(!mlxsw_sp_port_vlan))
 		return;
Signature · 1 line
2.14.3
#1 lore raw
Quoted 13 lines · Jiri Pirko

From: Ido Schimmel <[email protected]>

Up until now we only allowed VLAN devices to be put in a VLAN-unaware bridge, but some users need the ability to enslave physical ports as well.

This is achieved by mapping the port and VID 1 to the bridge's vFID, instead of the port and the VID used by the VLAN device.

The above is valid because as long as the port is not enslaved to a bridge, VID 1 is guaranteed to be configured as PVID and egress untagged.

Quoted 6 lines
---
 drivers/net/ethernet/mellanox/mlxsw/spectrum_switchdev.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)

Maybe I am missing something in the setup, but I am not getting host-to-host connectivity. I booted a switch with this patch, configured a bridge:

# ip a sh dev br0
44: br0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state
UP group default qlen 1000
    link/ether 7c:fe:90:e8:3a:79 brd ff:ff:ff:ff:ff:ff
    inet6 3000:1000:1000:1000::1/80 scope global
       valid_lft forever preferred_lft forever
    inet6 fe80::7efe:90ff:fee8:3a79/64 scope link
       valid_lft forever preferred_lft forever

Connected ports:

# ip li sh master br0
36: swp1s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
master br0 state UP mode DEFAULT group default qlen 1000
    link/ether 7c:fe:90:e8:3a:7d brd ff:ff:ff:ff:ff:ff
37: swp1s1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
master br0 state UP mode DEFAULT group default qlen 1000
    link/ether 7c:fe:90:e8:3a:7e brd ff:ff:ff:ff:ff:ff
38: swp1s2: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
master br0 state UP mode DEFAULT group default qlen 1000
    link/ether 7c:fe:90:e8:3a:7f brd ff:ff:ff:ff:ff:ff
39: swp1s3: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
master br0 state UP mode DEFAULT group default qlen 1000
    link/ether 7c:fe:90:e8:3a:80 brd ff:ff:ff:ff:ff:ff
40: swp3s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
master br0 state UP mode DEFAULT group default qlen 1000
    link/ether 7c:fe:90:e8:3a:79 brd ff:ff:ff:ff:ff:ff
41: swp3s1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
master br0 state UP mode DEFAULT group default qlen 1000
    link/ether 7c:fe:90:e8:3a:7a brd ff:ff:ff:ff:ff:ff
42: swp3s2: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
master br0 state UP mode DEFAULT group default qlen 1000
    link/ether 7c:fe:90:e8:3a:7b brd ff:ff:ff:ff:ff:ff
43: swp3s3: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
master br0 state UP mode DEFAULT group default qlen 1000
    link/ether 7c:fe:90:e8:3a:7c brd ff:ff:ff:ff:ff:ff

The switch can see the hosts:

# net show lldp

LocalPort Speed Mode RemotePort RemoteHost
Summary
----------- ------- --------- ------------ ------------------------------------- -----------------------
eth0 1G Mgmt swp37
pioneerMS16.mvlab.cumulusnetworks.com IP: 10.0.3.155/22(DHCP)
swp1s0 10G Access/L2 swp1 host1
Untagged: br0
swp1s1 10G Access/L2 swp1 host2
Untagged: br0
swp1s2 10G Access/L2 swp1 host3
Untagged: br0
swp1s3 10G Access/L2 swp1 host4
Untagged: br0
swp3s0 10G Access/L2 swp1 host5
Untagged: br0
swp3s1 10G Access/L2 swp1 host6
Untagged: br0
swp3s2 10G Access/L2 swp1 host7
Untagged: br0
swp3s3 10G Access/L2 swp1 host8
Untagged: br0

and can talk to the hosts:

# ping6 ff02::2%br0
PING ff02::2%br0(ff02::2) 56 data bytes
64 bytes from fe80::7efe:90ff:fee8:3a79: icmp_seq=1 ttl=64 time=0.073 ms
64 bytes from fe80::202:ff:fe00:2: icmp_seq=1 ttl=64 time=0.661 ms (DUP!)
64 bytes from fe80::202:ff:fe00:5: icmp_seq=1 ttl=64 time=0.705 ms (DUP!)
64 bytes from fe80::202:ff:fe00:1: icmp_seq=1 ttl=64 time=0.720 ms (DUP!)
64 bytes from fe80::202:ff:fe00:3: icmp_seq=1 ttl=64 time=0.729 ms (DUP!)
64 bytes from fe80::202:ff:fe00:6: icmp_seq=1 ttl=64 time=0.739 ms (DUP!)
64 bytes from fe80::202:ff:fe00:4: icmp_seq=1 ttl=64 time=0.748 ms (DUP!)
64 bytes from fe80::202:ff:fe00:7: icmp_seq=1 ttl=64 time=0.757 ms (DUP!)
64 bytes from fe80::202:ff:fe00:8: icmp_seq=1 ttl=64 time=0.766 ms (DUP!)

but the hosts can not talk to each other:

cumulus@host3:~$ net show lldp
LocalPort  Speed  Mode          RemoteHost   RemotePort
---------  -----  ------------  -----------  ----------
swp1       10G    Interface/L3  mlx-2700-05  swp1s2
cumulus@host3:~$ ping6 3000:1000:1000:1000::2
PING 3000:1000:1000:1000::2(3000:1000:1000:1000::2) 56 data bytes
^C
--- 3000:1000:1000:1000::2 ping statistics ---
3 packets transmitted, 0 received, 100% packet loss, time 1999ms
#2 lore raw
Quoted 78 lines · David Ahern
Quoted 13 lines · Jiri Pirko

From: Ido Schimmel <[email protected]>

Up until now we only allowed VLAN devices to be put in a VLAN-unaware bridge, but some users need the ability to enslave physical ports as well.

This is achieved by mapping the port and VID 1 to the bridge's vFID, instead of the port and the VID used by the VLAN device.

The above is valid because as long as the port is not enslaved to a bridge, VID 1 is guaranteed to be configured as PVID and egress untagged.

Quoted 6 lines
---
 drivers/net/ethernet/mellanox/mlxsw/spectrum_switchdev.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)

Maybe I am missing something in the setup, but I am not getting host-to-host connectivity. I booted a switch with this patch, configured a bridge:

# ip a sh dev br0
44: br0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state
UP group default qlen 1000
    link/ether 7c:fe:90:e8:3a:79 brd ff:ff:ff:ff:ff:ff
    inet6 3000:1000:1000:1000::1/80 scope global
       valid_lft forever preferred_lft forever
    inet6 fe80::7efe:90ff:fee8:3a79/64 scope link
       valid_lft forever preferred_lft forever

Connected ports:

# ip li sh master br0
36: swp1s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
master br0 state UP mode DEFAULT group default qlen 1000
    link/ether 7c:fe:90:e8:3a:7d brd ff:ff:ff:ff:ff:ff
37: swp1s1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
master br0 state UP mode DEFAULT group default qlen 1000
    link/ether 7c:fe:90:e8:3a:7e brd ff:ff:ff:ff:ff:ff
38: swp1s2: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
master br0 state UP mode DEFAULT group default qlen 1000
    link/ether 7c:fe:90:e8:3a:7f brd ff:ff:ff:ff:ff:ff
39: swp1s3: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
master br0 state UP mode DEFAULT group default qlen 1000
    link/ether 7c:fe:90:e8:3a:80 brd ff:ff:ff:ff:ff:ff
40: swp3s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
master br0 state UP mode DEFAULT group default qlen 1000
    link/ether 7c:fe:90:e8:3a:79 brd ff:ff:ff:ff:ff:ff
41: swp3s1: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
master br0 state UP mode DEFAULT group default qlen 1000
    link/ether 7c:fe:90:e8:3a:7a brd ff:ff:ff:ff:ff:ff
42: swp3s2: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
master br0 state UP mode DEFAULT group default qlen 1000
    link/ether 7c:fe:90:e8:3a:7b brd ff:ff:ff:ff:ff:ff
43: swp3s3: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast
master br0 state UP mode DEFAULT group default qlen 1000
    link/ether 7c:fe:90:e8:3a:7c brd ff:ff:ff:ff:ff:ff

The switch can see the hosts:

# net show lldp

LocalPort Speed Mode RemotePort RemoteHost
Summary
----------- ------- --------- ------------ ------------------------------------- -----------------------
eth0 1G Mgmt swp37
pioneerMS16.mvlab.cumulusnetworks.com IP: 10.0.3.155/22(DHCP)
swp1s0 10G Access/L2 swp1 host1
Untagged: br0
swp1s1 10G Access/L2 swp1 host2
Untagged: br0
swp1s2 10G Access/L2 swp1 host3
Untagged: br0
swp1s3 10G Access/L2 swp1 host4
Untagged: br0

Quoted 8 lines

swp3s0 10G Access/L2 swp1 host5
Untagged: br0
swp3s1 10G Access/L2 swp1 host6
Untagged: br0
swp3s2 10G Access/L2 swp1 host7
Untagged: br0
swp3s3 10G Access/L2 swp1 host8
Untagged: br0

LLDP packets are trapped before L2 forwarding, which can explain why LLDP works but forwarding doesn't.

Quoted 3 lines

and can talk to the hosts:

# ping6 ff02::2%br0

Can you try ff02::1 ?

Quoted 19 lines

PING ff02::2%br0(ff02::2) 56 data bytes
64 bytes from fe80::7efe:90ff:fee8:3a79: icmp_seq=1 ttl=64 time=0.073 ms 64 bytes from fe80::202:ff:fe00:2: icmp_seq=1 ttl=64 time=0.661 ms (DUP!) 64 bytes from fe80::202:ff:fe00:5: icmp_seq=1 ttl=64 time=0.705 ms (DUP!) 64 bytes from fe80::202:ff:fe00:1: icmp_seq=1 ttl=64 time=0.720 ms (DUP!) 64 bytes from fe80::202:ff:fe00:3: icmp_seq=1 ttl=64 time=0.729 ms (DUP!) 64 bytes from fe80::202:ff:fe00:6: icmp_seq=1 ttl=64 time=0.739 ms (DUP!) 64 bytes from fe80::202:ff:fe00:4: icmp_seq=1 ttl=64 time=0.748 ms (DUP!) 64 bytes from fe80::202:ff:fe00:7: icmp_seq=1 ttl=64 time=0.757 ms (DUP!) 64 bytes from fe80::202:ff:fe00:8: icmp_seq=1 ttl=64 time=0.766 ms (DUP!)

but the hosts can not talk to each other:

cumulus@host3:~$ net show lldp
LocalPort  Speed  Mode          RemoteHost   RemotePort
---------  -----  ------------  -----------  ----------
swp1       10G    Interface/L3  mlx-2700-05  swp1s2
Quoted 5 lines
cumulus@host3:~$ ping6 3000:1000:1000:1000::2
PING 3000:1000:1000:1000::2(3000:1000:1000:1000::2) 56 data bytes
^C
--- 3000:1000:1000:1000::2 ping statistics ---
3 packets transmitted, 0 received, 100% packet loss, time 1999ms

Can you please try

# brctl showstp br0

To make sure STP state is set to forwarding?

Does it matter if you try IPv4 ping or if vlan_filtering is set 1? Unfortunately, I can't reproduce on my switch.

Thanks for testing, David!

#3 lore raw
Quoted 5 lines · Ido Schimmel
Quoted 3 lines

and can talk to the hosts:

# ping6 ff02::2%br0

Can you try ff02::1 ?

same result.

Quoted 26 lines
Quoted 20 lines

PING ff02::2%br0(ff02::2) 56 data bytes
64 bytes from fe80::7efe:90ff:fee8:3a79: icmp_seq=1 ttl=64 time=0.073 ms 64 bytes from fe80::202:ff:fe00:2: icmp_seq=1 ttl=64 time=0.661 ms (DUP!) 64 bytes from fe80::202:ff:fe00:5: icmp_seq=1 ttl=64 time=0.705 ms (DUP!) 64 bytes from fe80::202:ff:fe00:1: icmp_seq=1 ttl=64 time=0.720 ms (DUP!) 64 bytes from fe80::202:ff:fe00:3: icmp_seq=1 ttl=64 time=0.729 ms (DUP!) 64 bytes from fe80::202:ff:fe00:6: icmp_seq=1 ttl=64 time=0.739 ms (DUP!) 64 bytes from fe80::202:ff:fe00:4: icmp_seq=1 ttl=64 time=0.748 ms (DUP!) 64 bytes from fe80::202:ff:fe00:7: icmp_seq=1 ttl=64 time=0.757 ms (DUP!) 64 bytes from fe80::202:ff:fe00:8: icmp_seq=1 ttl=64 time=0.766 ms (DUP!)

but the hosts can not talk to each other:

cumulus@host3:~$ net show lldp
LocalPort  Speed  Mode          RemoteHost   RemotePort
---------  -----  ------------  -----------  ----------
swp1       10G    Interface/L3  mlx-2700-05  swp1s2
Quoted 5 lines
cumulus@host3:~$ ping6 3000:1000:1000:1000::2
PING 3000:1000:1000:1000::2(3000:1000:1000:1000::2) 56 data bytes
^C
--- 3000:1000:1000:1000::2 ping statistics ---
3 packets transmitted, 0 received, 100% packet loss, time 1999ms
Quoted 5 lines

Can you please try

# brctl showstp br0

To make sure STP state is set to forwarding?

it is.

Quoted 3 lines

Does it matter if you try IPv4 ping or if vlan_filtering is set 1? Unfortunately, I can't reproduce on my switch.

Bring up the hosts and then reboot the switch. At that point I get no host to host communication. As soon as I flap the port on host1 host3 to host1 starts working.

So it seems to be something about the initial boot state.

#4 lore raw
Quoted 44 lines · David Ahern
Quoted 5 lines · Ido Schimmel
Quoted 3 lines

and can talk to the hosts:

# ping6 ff02::2%br0

Can you try ff02::1 ?

same result.

Quoted 26 lines
Quoted 20 lines

PING ff02::2%br0(ff02::2) 56 data bytes
64 bytes from fe80::7efe:90ff:fee8:3a79: icmp_seq=1 ttl=64 time=0.073 ms 64 bytes from fe80::202:ff:fe00:2: icmp_seq=1 ttl=64 time=0.661 ms (DUP!) 64 bytes from fe80::202:ff:fe00:5: icmp_seq=1 ttl=64 time=0.705 ms (DUP!) 64 bytes from fe80::202:ff:fe00:1: icmp_seq=1 ttl=64 time=0.720 ms (DUP!) 64 bytes from fe80::202:ff:fe00:3: icmp_seq=1 ttl=64 time=0.729 ms (DUP!) 64 bytes from fe80::202:ff:fe00:6: icmp_seq=1 ttl=64 time=0.739 ms (DUP!) 64 bytes from fe80::202:ff:fe00:4: icmp_seq=1 ttl=64 time=0.748 ms (DUP!) 64 bytes from fe80::202:ff:fe00:7: icmp_seq=1 ttl=64 time=0.757 ms (DUP!) 64 bytes from fe80::202:ff:fe00:8: icmp_seq=1 ttl=64 time=0.766 ms (DUP!)

but the hosts can not talk to each other:

cumulus@host3:~$ net show lldp
LocalPort  Speed  Mode          RemoteHost   RemotePort
---------  -----  ------------  -----------  ----------
swp1       10G    Interface/L3  mlx-2700-05  swp1s2
Quoted 5 lines
cumulus@host3:~$ ping6 3000:1000:1000:1000::2
PING 3000:1000:1000:1000::2(3000:1000:1000:1000::2) 56 data bytes
^C
--- 3000:1000:1000:1000::2 ping statistics ---
3 packets transmitted, 0 received, 100% packet loss, time 1999ms
Quoted 5 lines

Can you please try

# brctl showstp br0

To make sure STP state is set to forwarding?

it is.

Quoted 1 line
Quoted 8 lines
Quoted 2 lines

Does it matter if you try IPv4 ping or if vlan_filtering is set 1? Unfortunately, I can't reproduce on my switch.

Bring up the hosts and then reboot the switch. At that point I get no host to host communication. As soon as I flap the port on host1 host3 to host1 starts working.

So it seems to be something about the initial boot state.

You didn't have IPv6 *and* IPv4 ping? I'm asking because it's possible host1 sent an MLD join to the Solicited-node multicast address before the bridge started listening, which means it didn't have a corresponding MDB entry.

Assuming your hosts aren't functioning as multicast routers and sending MLD queries and that you didn't configure them as mrouter ports on the switch, then when host3 sent a neighbour solicitation message to host1's Solicited-node multicast address it wasn't flooded to host3 which prevented ping from passing.

This also explains why it started working when you flapped the port on host1, as Linux generates MLD joins in these cases.

You can try to disable snooping:

# ip link set dev br0 type bridge mcast_snooping 0

Just a guess, but worth a try.

Thanks!

#5 lore raw
Quoted 7 lines · Ido Schimmel
Quoted 7 lines
Quoted 2 lines

Does it matter if you try IPv4 ping or if vlan_filtering is set 1? Unfortunately, I can't reproduce on my switch.

Bring up the hosts and then reboot the switch. At that point I get no host to host communication. As soon as I flap the port on host1 host3 to host1 starts working.

Quoted 6 lines
Quoted 1 line

So it seems to be something about the initial boot state.

You didn't have IPv6 *and* IPv4 ping? I'm asking because it's possible host1 sent an MLD join to the Solicited-node multicast address before the bridge started listening, which means it didn't have a corresponding MDB entry.

The sim only configures IPv6, but it is not acting as an mcast router. It's really a dummy setup -- bridge on the switch, ports connected to hosts.

Quoted 7 lines

Assuming your hosts aren't functioning as multicast routers and sending MLD queries and that you didn't configure them as mrouter ports on the switch, then when host3 sent a neighbour solicitation message to host1's Solicited-node multicast address it wasn't flooded to host3 which prevented ping from passing.

Quoted 8 lines

This also explains why it started working when you flapped the port on host1, as Linux generates MLD joins in these cases.

You can try to disable snooping:

# ip link set dev br0 type bridge mcast_snooping 0

Just a guess, but worth a try.

good guess. That change gets it working.

maintainer
#6 lore raw
In reply to David Ahern

From: David Ahern <[email protected]>
Date: Wed, 21 Feb 2018 11:16:35 -0700

Quoted 21 lines · Jiri Pirko
Quoted 13 lines

From: Ido Schimmel <[email protected]>

Up until now we only allowed VLAN devices to be put in a VLAN-unaware bridge, but some users need the ability to enslave physical ports as well.

This is achieved by mapping the port and VID 1 to the bridge's vFID, instead of the port and the VID used by the VLAN device.

The above is valid because as long as the port is not enslaved to a bridge, VID 1 is guaranteed to be configured as PVID and egress untagged.

Quoted 6 lines
---
 drivers/net/ethernet/mellanox/mlxsw/spectrum_switchdev.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)
Quoted 3 lines

Maybe I am missing something in the setup, but I am not getting host-to-host connectivity. I booted a switch with this patch, configured a bridge:

I'm waiting for this discussion to be fully resolved before applying this patch. Just FYI...

#7 lore raw
Quoted 24 lines · David Miller

From: David Ahern <[email protected]>
Date: Wed, 21 Feb 2018 11:16:35 -0700

Quoted 21 lines · Jiri Pirko
Quoted 19 lines

From: Ido Schimmel <[email protected]>

Up until now we only allowed VLAN devices to be put in a VLAN-unaware bridge, but some users need the ability to enslave physical ports as well.

This is achieved by mapping the port and VID 1 to the bridge's vFID, instead of the port and the VID used by the VLAN device.

The above is valid because as long as the port is not enslaved to a bridge, VID 1 is guaranteed to be configured as PVID and egress untagged.

---
 drivers/net/ethernet/mellanox/mlxsw/spectrum_switchdev.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)
Quoted 7 lines
Quoted 3 lines

Maybe I am missing something in the setup, but I am not getting host-to-host connectivity. I booted a switch with this patch, configured a bridge:

I'm waiting for this discussion to be fully resolved before applying this patch. Just FYI...

Ido:

IPv4 works at boot; IPv6 requires the mcast snooping disable. For this vlan-unaware bridges can that be set automatically?

And then, what are the options for lldp?

#8 lore raw
Quoted 4 lines · David Ahern

Ido:

IPv4 works at boot; IPv6 requires the mcast snooping disable. For this vlan-unaware bridges can that be set automatically?

Can you please try the following patch?

drivers/net/ethernet/mellanox/mlxsw/spectrum_fid.c +1 -1
index bbd238e50f05..54262af4e98f 100644
@@ -112,11 +112,11 @@ static const int mlxsw_sp_sfgc_bc_packet_types[MLXSW_REG_SFGC_TYPE_MAX] = {
 	[MLXSW_REG_SFGC_TYPE_UNREGISTERED_MULTICAST_NON_IP]	= 1,
 	[MLXSW_REG_SFGC_TYPE_IPV4_LINK_LOCAL]			= 1,
 	[MLXSW_REG_SFGC_TYPE_IPV6_ALL_HOST]			= 1,
+	[MLXSW_REG_SFGC_TYPE_UNREGISTERED_MULTICAST_IPV6]	= 1,
 };
 
 static const int mlxsw_sp_sfgc_mc_packet_types[MLXSW_REG_SFGC_TYPE_MAX] = {
 	[MLXSW_REG_SFGC_TYPE_UNREGISTERED_MULTICAST_IPV4]	= 1,
-	[MLXSW_REG_SFGC_TYPE_UNREGISTERED_MULTICAST_IPV6]	= 1,
 };
 
 static const int *mlxsw_sp_packet_type_sfgc_types[] = {

It should fix your problem.

The real problem that I can then address in net-next is the fact that the Linux bridge tries to be smart and only resorts to flooding unregistered multicast packets in case its querier is disabled and in case it didn't detect any other querier in the network. This isn't currently reflected to underlying drivers. Only mcast snooping on/off.

Anyway, it's not related to the patch in question. You'd get the same behavior with VLAN-aware bridges.

Quoted 1 line

And then, what are the options for lldp?

Didn't understand the question. Can you clarify?

#9 lore raw
Quoted 8 lines · Ido Schimmel
Quoted 4 lines · David Ahern

Ido:

IPv4 works at boot; IPv6 requires the mcast snooping disable. For this vlan-unaware bridges can that be set automatically?

Can you please try the following patch?

...

Quoted 2 lines

It should fix your problem.

it does.

Quoted 7 lines

The real problem that I can then address in net-next is the fact that the Linux bridge tries to be smart and only resorts to flooding unregistered multicast packets in case its querier is disabled and in case it didn't detect any other querier in the network. This isn't currently reflected to underlying drivers. Only mcast snooping on/off.

Quoted 7 lines

Anyway, it's not related to the patch in question. You'd get the same behavior with VLAN-aware bridges.

Quoted 1 line

And then, what are the options for lldp?

Didn't understand the question. Can you clarify?

nm. mental lapse.

#10 lore raw
In reply to David Miller

Hi Dave,

Quoted 2 lines · David Miller

I'm waiting for this discussion to be fully resolved before applying this patch. Just FYI...

I have a fix for the issue David reported, but it is not related to this patch (problem manifests itself with VLAN-aware bridges as well).

maintainer
#11 lore raw

From: Ido Schimmel <[email protected]>
Date: Mon, 26 Feb 2018 08:54:33 +0200

Quoted 8 lines

Hi Dave,

Quoted 2 lines · David Miller

I'm waiting for this discussion to be fully resolved before applying this patch. Just FYI...

I have a fix for the issue David reported, but it is not related to this patch (problem manifests itself with VLAN-aware bridges as well).

Ok, I've applied this patch to net-next then.

Thank you.

Tested-by on v1 1/1
#12 lore raw
In reply to Jiri Pirko
Quoted 13 lines · Jiri Pirko

From: Ido Schimmel <[email protected]>

Up until now we only allowed VLAN devices to be put in a VLAN-unaware bridge, but some users need the ability to enslave physical ports as well.

This is achieved by mapping the port and VID 1 to the bridge's vFID, instead of the port and the VID used by the VLAN device.

The above is valid because as long as the port is not enslaved to a bridge, VID 1 is guaranteed to be configured as PVID and egress untagged.

Quoted 6 lines
---
 drivers/net/ethernet/mellanox/mlxsw/spectrum_switchdev.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)