Do not resolve physical devices for soft offload

[brada4]

Let kernel heuristics take care of offloading decapsulation Packets may still enter flow engine one encapsulation below actual interface subject to heuristics, while exiting it on listed interfaces, in kernel subject to non-flow encapsulation offloads.

Fixes: https://github.com/openwrt/openwrt/issues/13410 Fixes: https://github.com/openwrt/openwrt/issues/10224

This is minimally intrusive, to address issue, likely should be bitmask to selectively enable (forward interfaces|one encap below|virt interfaces|wifi|wifi not unique AP|probably others) or flaggs to +/- exact interface names to list.

Signed-off-by: Andris PE neandris@gmail.com

[jow-]

Ah great, is this confirmed working with PPPoE?

[brada4]

I tested with gre over vlan. Somebody may need to test in ppoee thread....

[brada4]

gre heuristic is to always decap, ppoe checks if mtu is good and no magic bits in protocol are set, but no way to fall 2 encaps apart to disable flow translation completely.

[brada4]

@jow- ? Or it would need to check if (l3dev) and l3dev != dev then l3dev else dev ?

--- a/root/usr/share/ucode/fw4.uc
+++ b/root/usr/share/ucode/fw4.uc
@@ -599,7 +599,7 @@ return {
                        for (let ifc in ifaces.interface) {
                                let net = {
                                        up: ifc.up,
-                                       device: ifc.l3_device ?? ifc.device,
+                                       device: ifc.l3_device,
                                        physdev: ifc.device,
                                        zone: ifc.data?.zone
                                };
@@ -647,7 +647,7 @@ return {
                                                        ...rulespec,

                                                        name: (rulespec.type != 'ipset') ? `ubus:${ifc.interface}[${ifc.proto}] ${rulespec.type || 'rule'} ${n}` : rulespec.name,
-                                                       device: rulespec.device ?? ifc.l3_device ?? ifc.device
+                                                       device: rulespec.device ?? ifc.l3_device
                                                });

                                                n++;

[brada4]

@jow- ok, side effect is issue 10224 is partially addressed, not completely fixed. Should I polish message to less happy?

[brada4]

For the other part of issue is TBD, seems netifd sometimes (pppoe) adds l3 device which is kind of to transfer all rules to, sometimes it has only l3device while GRE tunnel has no routes etc. No idea as of yet regarding needed logic vs netifd emissions

[brada4]

Ill leave it half-way. Fixes: roaming Fixes half pppoe, nothing more.

[brada4]

will repeat with clean tree.