feat(xdp): IPv6 parity, monotonic ban expiry, per-protocol L4 minimum

IPv6 parity:
- blacklist6_map (LPM trie, 20-byte key) and is_blacklisted6(); the
  native_v6 path now passes the port gate and the L4 checks instead of
  returning XDP_PASS early.
- ban_cidr() covers V4 and V6; gateway bans by /32 or /128; subnet monitor
  bans v6 /64 prefixes; ban_ip() removed (dead code is a bug).
- v6 TCP feeds prefix stats through prefix_key_from_v6 (subnet detector
  now sees IPv6); UDP policy=drop and detect_tcp_bypass apply to v6.
- tests/xdp_ban_v6.rs and scripts/xdp_ipv6_smoke.sh (real netns/verifier).

Bug fixes (both families):
- Ban expiry compared SystemTime epoch-ns against bpf_ktime_get_ns(), so
  bans never expired. Userspace now writes CLOCK_MONOTONIC via
  xdp::monotonic_ns(); regression in tests/xdp_ban_clock.rs.
- The L4 gate required sizeof(struct tcphdr) for every protocol, so short
  UDP datagrams bypassed the port gate and the blacklist. The minimum is
  now per-protocol (UDP 8 bytes); smoke covers it.
This commit is contained in:
Loki 2026-11-15 23:08:05 +01:00
parent a90f0da226
commit b64c329125
Signed by: boba
GPG key ID: 6CDE87BC9A778D97
12 changed files with 420 additions and 36 deletions

View file

@ -50,6 +50,12 @@ struct lpm_key {
__u32 ip;
};
// ── LPM key for IPv6 blacklist (format of bpf_lpm_trie_key, 20 bytes) ──
struct lpm_key6 {
__u32 prefixlen;
__u8 addr[16];
};
// ── Ringbuf event (userspace receives these) ──
enum event_type {
EVENT_BAN = 0,

View file

@ -11,6 +11,16 @@ struct {
__uint(map_flags, BPF_F_NO_PREALLOC);
} blacklist_map SEC(".maps");
// 🔗 IPv6 blacklist (LPM_TRIE, 128-bit prefix keys)
// Same value/expiry contract as blacklist_map
struct {
__uint(type, BPF_MAP_TYPE_LPM_TRIE);
__uint(max_entries, 100000);
__type(key, struct lpm_key6);
__type(value, __u64); // ban expiry (ktime_ns)
__uint(map_flags, BPF_F_NO_PREALLOC);
} blacklist6_map SEC(".maps");
// 🔗 Whitelist (LPM_TRIE for CIDR) — checked before any filter
struct {
__uint(type, BPF_MAP_TYPE_LPM_TRIE);

View file

@ -72,6 +72,22 @@ static __always_inline __u8 is_blacklisted(__u32 src_ip, __u64 now)
return 0;
}
// ── IPv6 blacklist lookup with expiry; removes expired entries ──
// Key layout mirrors bpf_lpm_trie_key: prefixlen (host order) + 16 addr bytes.
static __always_inline __u8 is_blacklisted6(const __u8 addr[16], __u64 now)
{
struct lpm_key6 key = {};
key.prefixlen = 128;
__builtin_memcpy(key.addr, addr, 16);
__u64 *ban_until = bpf_map_lookup_elem(&blacklist6_map, &key);
if (!ban_until)
return 0;
if (*ban_until > now)
return 1;
bpf_map_delete_elem(&blacklist6_map, &key); // expired — clean up
return 0;
}
static __always_inline void ban_ip(__u32 src_ip, __u64 until)
{
struct lpm_key key = { .prefixlen = 32, .ip = src_ip };
@ -127,6 +143,8 @@ int rampart_universal_filter(struct xdp_md *ctx)
__u32 dst_ip = 0;
__u8 l4_proto = 0;
__u8 is_ipv6 = 0;
__u8 native_v6 = 0;
__u8 src6[16];
if (eth->h_proto == bpf_htons(ETH_P_IP)) {
struct iphdr *ip = (void *)(eth + 1);
@ -164,15 +182,19 @@ int rampart_universal_filter(struct xdp_md *ctx)
ip6->daddr.in6_u.u6_addr32[0] != 0 ||
ip6->daddr.in6_u.u6_addr32[1] != 0 ||
ip6->daddr.in6_u.u6_addr32[2] != bpf_htonl(0xFFFF)) {
// Non-mapped IPv6 — pass for now (not supported)
return XDP_PASS;
// Native IPv6 — no 32-bit representable address, handled separately
// after the generic L4/port gates; must NOT touch v4 maps.
native_v6 = 1;
is_ipv6 = 1;
l4_proto = ip6->nexthdr;
__builtin_memcpy(src6, &ip6->saddr, 16);
} else {
// FIX: source address must come from saddr, never daddr
src_ip = ip6->saddr.in6_u.u6_addr32[3];
dst_ip = ip6->daddr.in6_u.u6_addr32[3];
l4_proto = ip6->nexthdr;
is_ipv6 = 1;
}
// FIX: source address must come from saddr, never daddr
src_ip = ip6->saddr.in6_u.u6_addr32[3];
dst_ip = ip6->daddr.in6_u.u6_addr32[3];
l4_proto = ip6->nexthdr;
is_ipv6 = 1;
}
// ── Parse L4 header ──
@ -189,7 +211,13 @@ int rampart_universal_filter(struct xdp_md *ctx)
ip_hdr_len = ip->ihl * 4;
}
if ((void *)l4 + sizeof(struct tcphdr) > data_end)
// Minimum L4 header per protocol: a short UDP datagram used to fall through
// to XDP_PASS because the check was sized for TCP, letting it bypass the
// blacklist entirely. IPv6 native fragmentation is already filtered above
// (nexthdr must be TCP/UDP), so a short UDP header here is a real datagram.
__u32 min_l4 = (l4_proto == IPPROTO_TCP) ? sizeof(struct tcphdr)
: sizeof(struct udphdr);
if ((void *)l4 + min_l4 > data_end)
return XDP_PASS;
// ── Port range check (protected services) ──
@ -209,6 +237,35 @@ int rampart_universal_filter(struct xdp_md *ctx)
__u64 now = bpf_ktime_get_ns();
// ── Native IPv6: blacklist + cheap L4 policy (no stateful v6 yet) ──
if (native_v6) {
if (is_blacklisted6(src6, now)) {
inc_blacklist();
inc_drop();
return XDP_DROP;
}
if (l4_proto == IPPROTO_UDP) {
if (G_UDP_POLICY == 1) { // drop
inc_drop();
return XDP_DROP;
}
inc_pass();
return XDP_PASS; // policy 2 (rate-limit) needs a 128-bit key — not supported for v6
}
// TCP: forged-flag combos are L4-generic, no state needed
if (detect_tcp_bypass((struct tcphdr *)l4)) {
inc_drop();
return XDP_DROP;
}
// Per-prefix stats (/64) — feeds the subnet detector for IPv6
struct prefix_key pkey6;
prefix_key_from_v6(&pkey6, src6);
struct tcphdr *tcp6 = (struct tcphdr *)l4;
update_prefix_stats(&pkey6, (tcp6->syn && !tcp6->ack) ? 1 : 0, now);
inc_pass();
return XDP_PASS;
}
// ── Whitelist check (CIDR) — before any filtering ──
struct lpm_key wl_key = { .prefixlen = 32, .ip = src_ip };
if (bpf_map_lookup_elem(&whitelist_map, &wl_key)) {