[multicast] adds multicast replication for sidecar-lite

This PR adds multicast packet replication to sidecar-lite's P4 pipeline,
emulating Dendrite's tofino sidecar implementation. Packets arriving
inside geneve with a multicast inner destination are source-filtered,
replicated to external and underlay groups, and tagged via the oxg_mcast
geneve option so downstream hops suppress already-served groups.

## TTL==1 route dropping

Packets with TTL==1 are dropped at the route table level via a
compound key (path_idx, route_ttl_is_1) so the ttl_exceeded action
fires instead of forward, preventing replication of expiring packets.
The control plane programs paired entries for both TTL values.

## Parser

IPv6 multicast scope validation now matches Dendrite's sidecar:
ff01 (interface-local) rejected, ff02 (link-local) forwarded to scrimlet
with hop_limit >= 1, other multicast requires hop_limit > 1.
IPv4 multicast (224.0.0.0/4) TTL==0 and TTL==1 rejected.

## Ingress

Source filtering for geneve-encapsulated multicast is executed via
mcast_source_filter_v4/v6 tables (inner src LPM + inner dst exact).

Replication tables set both grp_a (external) and grp_b (underlay)
unconditionally, while per-packet tag handling suppresses the group
whose stage is already done (0 = external, 1 = underlay, 2 = both).
Non-encapsulated multicast bypasses source filtering.

## Egress

The multicast tag is stamped into geneve options on every replicated
copy. Port reflection filter drops copies where egress == ingress port.
Decap occurs on bifurcated replication, or tag=2 (both groups served),
with inner TTL validation and IPv4 header checksum update (RFC 1624).

Also included:

- A VLAN-aware decap variant that inserts 802.1Q tag with configured
  VLAN ID.
- Multicast dst MAC derivation from IP destination on all copies,
  matching Dendrite's MulticastMacRewrite.
- Per-port src MAC rewritten via the `mcast_src_mac` table.

## Dependencies

p4 crates are currently updated to the `zl/multicast` branch
for multicast replication support in p4rs/softnpu and LHS bit-slice
assignment in x4c (used for multicast MAC derivation).

Author: zeeshanlakhani

Cargo.lock

@@ -6,13 +6,13 @@ members = [
 resolver = "2"
 
 [workspace.dependencies]
-p4rs = { git = "https://github.com/oxidecomputer/p4", branch = "main" }
-p4-macro = { git = "https://github.com/oxidecomputer/p4", branch = "main" }
-p4-test = { package = "tests", git = "https://github.com/oxidecomputer/p4", branch = "main" }
+p4rs = { git = "https://github.com/oxidecomputer/p4", branch = "zl/multicast" }
+p4-macro = { git = "https://github.com/oxidecomputer/p4", branch = "zl/multicast" }
+p4-test = { package = "tests", git = "https://github.com/oxidecomputer/p4", branch = "zl/multicast" }
 usdt = { git = "https://github.com/oxidecomputer/usdt" }
 
 base64 = { version = "0.22" }
-clap = { version = "4.5.54", features = ["derive", "unstable-styles"] }
+clap = { version = "4.5.60", features = ["derive", "unstable-styles"] }
 anstyle = "1.0.13"
 libc = "0.2"
 serde = "1.0"
@@ -22,7 +22,7 @@ errno = "0.3"
 num = { version = "0.4.3", features = ["num-bigint"] }
 macaddr = { git = "https://github.com/luqmana/rust-macaddr", branch = "leading-zeros", version = "1.0.1" }
 bitvec = "1.0"
-colored = "3.0"
+colored = "3.1"
 anyhow = "1"
 pnet = "0.35"
 pnet_macros = "0.35"

Cargo.toml

@@ -1,4 +1,4 @@
-// Copyright 2025 Oxide Computer Company
+// Copyright 2026 Oxide Computer Company
 
 header sidecar_h {
     bit<8> sc_code;
@@ -96,7 +96,7 @@ header geneve_opt_h {
 }
 
 header oxg_opt_multicast_h {
-    bit<2> replication;
+    bit<2> mcast_tag;
     bit<30> reserved;
 }
 

p4/headers.p4

@@ -1,4 +1,4 @@
-// Copyright 2024 Oxide Computer Company
+// Copyright 2026 Oxide Computer Company
 
 parser parse(
     packet_in pkt,
@@ -46,6 +46,30 @@ parser parse(
 
     state ipv6 {
         pkt.extract(hdr.ipv6);
+        if (hdr.ipv6.dst[127:120] == 8w0xff) { transition ipv6_mcast; }
+        transition ipv6_proto;
+    }
+
+    state ipv6_mcast {
+        ingress.is_mcast = true;
+        // Interface-local (ff01) must not be forwarded.
+        if (hdr.ipv6.dst[127:112] == 16w0xff01) { transition reject; }
+
+        // Link-local (ff02) is forwarded to scrimlet.
+        // Allow hop_limit == 1 (link-local scope) but still reject expired
+        // packets.
+        if (hdr.ipv6.dst[127:112] == 16w0xff02) {
+            if (hdr.ipv6.hop_limit == 8w0) { transition reject; }
+            transition ipv6_proto;
+        }
+
+        // Non-link-local multicast requires hop_limit > 1.
+        if (hdr.ipv6.hop_limit == 8w0) { transition reject; }
+        if (hdr.ipv6.hop_limit == 8w1) { transition reject; }
+        transition ipv6_proto;
+    }
+
+    state ipv6_proto {
         if (hdr.ipv6.next_hdr == 8w0xdd) { transition ddm; }
         if (hdr.ipv6.next_hdr == 8w58)   { transition icmp; }
         if (hdr.ipv6.next_hdr == 8w17)   { transition udp; }
@@ -82,6 +106,19 @@ parser parse(
 
     state ipv4 {
         pkt.extract(hdr.ipv4);
+        if (hdr.ipv4.dst[31:28] == 4w0xe) { transition ipv4_mcast; }
+        transition ipv4_proto;
+    }
+
+    state ipv4_mcast {
+        ingress.is_mcast = true;
+        // Multicast with TTL <= 1 must not be forwarded (RFC 1112).
+        if (hdr.ipv4.ttl == 8w0) { transition reject; }
+        if (hdr.ipv4.ttl == 8w1) { transition reject; }
+        transition ipv4_proto;
+    }
+
+    state ipv4_proto {
         if (hdr.ipv4.protocol == 8w17) { transition udp; }
         if (hdr.ipv4.protocol == 8w6)  { transition tcp; }
         if (hdr.ipv4.protocol == 8w1)  { transition icmp; }