Tizian-Maxime-Weigt/l3-l4-forwarding-in-xdp
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

main.c aktualisiert

Browse Source
This commit is contained in:
Tizian-Maxime-Weigt
2025-10-28 23:16:31 +00:00
parent 8f988b0a0f
commit ba1c3ba0e6
1 changed files with 190 additions and 224 deletions
Download Patch File Download Diff File Expand all files Collapse all files
main.c
+78 -112
View File
@@ -22,29 +22,31 @@
#endif
#define IPV6_FLOWINFO_MASK bpf_htonl(0x0FFFFFFF)
#define VLAN_MAX_DEPTH 2 /* Support double-tagged VLANs */
#define IPV6_EXT_MAX_CHAIN 6 /* Max IPv6 extension headers to parse */
#define VLAN_MAX_DEPTH 2
#define IPV6_EXT_MAX_CHAIN 6
struct vlan_hdr {
__be16 h_vlan_TCI;
__be16 h_vlan_encapsulated_proto;
};
/* Forwarding ports */
struct {
__uint(type, BPF_MAP_TYPE_DEVMAP);
__uint(key_size, sizeof(int));
__uint(value_size, sizeof(int));
__type(key, __u32);
__type(value, __u32);
__uint(max_entries, 512);
} xdp_l3fwd_ports SEC(".maps");
/* Stats key — identifies a connection or flow */
struct flow_key {
__u8 proto;
__u8 pad[3]; /* alignment */
__u16 vlan_id; /* VLAN ID (0 if untagged) */
__u16 pad2; /* alignment */
__u8 pad[3];
__u16 vlan_id;
__u16 pad2;
union {
__u32 ipv4_src;
__u8 ipv6_src[16];
};
union {
__u32 ipv4_dst;
__u8 ipv6_dst[16];
@@ -54,13 +56,11 @@ struct flow_key {
__u16 dport;
};
/* Stats value — counts packets and bytes */
struct flow_stats {
__u64 packets;
__u64 bytes;
};
/* PERCPU hash map to track stats per flow - NO LOCKING NEEDED */
struct {
__uint(type, BPF_MAP_TYPE_PERCPU_HASH);
__uint(key_size, sizeof(struct flow_key));
@@ -68,7 +68,6 @@ struct {
__uint(max_entries, 65536);
} xdp_flow_stats SEC(".maps");
/* Decrease IPv4 TTL helper */
static __always_inline int ip_decrease_ttl(struct iphdr *iph)
{
__u32 check = (__u32)iph->check;
@@ -77,14 +76,10 @@ static __always_inline int ip_decrease_ttl(struct iphdr *iph)
return --iph->ttl;
}
/* Record stats in the xdp_flow_stats map - OPTIMIZED for PERCPU */
static __always_inline void record_stats(struct xdp_md *ctx, struct flow_key *key, __u64 bytes)
static __always_inline void record_stats(struct flow_key *key, __u64 bytes)
{
struct flow_stats *stats;
stats = bpf_map_lookup_elem(&xdp_flow_stats, key);
struct flow_stats *stats = bpf_map_lookup_elem(&xdp_flow_stats, key);
if (stats) {
/* Direct update - no atomics needed with PERCPU maps */
stats->packets++;
stats->bytes += bytes;
} else {
@@ -92,79 +87,60 @@ static __always_inline void record_stats(struct xdp_md *ctx, struct flow_key *ke
.packets = 1,
.bytes = bytes,
};
/* Use BPF_ANY instead of BPF_NOEXIST for better performance */
bpf_map_update_elem(&xdp_flow_stats, key, &new_stats, BPF_ANY);
}
}
/* Parse VLAN headers and return next protocol and offset */
static __always_inline int parse_vlan(void *data, void *data_end, __u64 *nh_off, __u16 *h_proto, __u16 *vlan_id)
{
struct vlan_hdr {
__be16 h_vlan_TCI;
__be16 h_vlan_encapsulated_proto;
} *vhdr;
int i;
/* Parse up to VLAN_MAX_DEPTH VLAN headers */
struct vlan_hdr *vh;
#pragma unroll
for (i = 0; i < VLAN_MAX_DEPTH; i++) {
for (int i = 0; i < VLAN_MAX_DEPTH; i++) {
if (*h_proto != bpf_htons(ETH_P_8021Q) && *h_proto != bpf_htons(ETH_P_8021AD))
break;
vhdr = data + *nh_off;
if ((void *)(vhdr + 1) > data_end)
vh = (void *)((char *)data + *nh_off);
if ((void *)(vh + 1) > data_end)
return -1;
/* Store the outermost VLAN ID */
if (i == 0)
*vlan_id = bpf_ntohs(vhdr->h_vlan_TCI) & 0x0FFF;
*vlan_id = bpf_ntohs(vh->h_vlan_TCI) & 0x0FFF;
*nh_off += sizeof(*vhdr);
*h_proto = vhdr->h_vlan_encapsulated_proto;
*nh_off += sizeof(*vh);
*h_proto = vh->h_vlan_encapsulated_proto;
}
return 0;
}
/* Skip IPv6 extension headers to find the actual L4 protocol */
static __always_inline int skip_ip6hdrext(void *data, void *data_end, __u64 *nh_off, __u8 next_hdr_type)
static __always_inline int skip_ip6hdrext(void *data, void *data_end, __u64 *nh_off, __u8 next)
{
struct ipv6_opt_hdr {
__u8 nexthdr;
__u8 hdrlen;
} *hdr;
int i;
struct ipv6_opt_hdr *hdr;
#pragma unroll
for (i = 0; i < IPV6_EXT_MAX_CHAIN; i++) {
hdr = data + *nh_off;
for (int i = 0; i < IPV6_EXT_MAX_CHAIN; i++) {
hdr = (void *)((char *)data + *nh_off);
if ((void *)(hdr + 1) > data_end)
return -1;
switch (next_hdr_type) {
switch (next) {
case IPPROTO_HOPOPTS:
case IPPROTO_DSTOPTS:
case IPPROTO_ROUTING:
case IPPROTO_MH:
*nh_off += (hdr->hdrlen + 1) * 8;
next_hdr_type = hdr->nexthdr;
next = hdr->nexthdr;
break;
case IPPROTO_AH:
*nh_off += (hdr->hdrlen + 2) * 4;
next_hdr_type = hdr->nexthdr;
next = hdr->nexthdr;
break;
case IPPROTO_FRAGMENT:
*nh_off += 8;
next_hdr_type = hdr->nexthdr;
next = hdr->nexthdr;
break;
default:
/* Found a header that is not an IPv6 extension header */
return next_hdr_type;
return next;
}
}
return -1;
}
@@ -172,32 +148,25 @@ static __always_inline int xdp_l3fwd_flags(struct xdp_md *ctx, __u32 flags)
{
void *data_end = (void *)(long)ctx->data_end;
void *data = (void *)(long)ctx->data;
struct bpf_fib_lookup fib_params;
struct ethhdr *eth = data;
struct ipv6hdr *ip6h;
struct iphdr *iph;
__u16 h_proto;
__u64 nh_off;
int rc;
__u16 vlan_id = 0;
nh_off = sizeof(*eth);
if (data + nh_off > data_end)
struct ethhdr *eth = data;
__u64 nh_off = sizeof(*eth);
if ((void *)((char *)data + nh_off) > data_end)
return XDP_DROP;
__builtin_memset(&fib_params, 0, sizeof(fib_params));
h_proto = eth->h_proto;
struct bpf_fib_lookup fib_params = {};
__u16 h_proto = eth->h_proto;
__u16 vlan_id = 0;
/* Parse VLAN headers if present */
if (parse_vlan(data, data_end, &nh_off, &h_proto, &vlan_id) < 0)
return XDP_DROP;
struct flow_key key = {};
key.vlan_id = vlan_id;
__u64 bytes = data_end - data;
__u64 bytes = (char *)data_end - (char *)data;
if (h_proto == bpf_htons(ETH_P_IP)) {
iph = data + nh_off;
struct iphdr *iph = (void *)((char *)data + nh_off);
if ((void *)(iph + 1) > data_end)
return XDP_DROP;
@@ -208,19 +177,21 @@ static __always_inline int xdp_l3fwd_flags(struct xdp_md *ctx, __u32 flags)
key.ipv4_src = iph->saddr;
key.ipv4_dst = iph->daddr;
/* FIXED: Proper bounds checking for L4 headers */
__u64 l4_off = nh_off + (iph->ihl * 4);
if (iph->protocol == IPPROTO_TCP) {
struct tcphdr *tcph = data + l4_off;
if ((void *)(tcph + 1) <= data_end) {
key.sport = tcph->source;
key.dport = tcph->dest;
}
} else if (iph->protocol == IPPROTO_UDP) {
struct udphdr *udph = data + l4_off;
if ((void *)(udph + 1) <= data_end) {
key.sport = udph->source;
key.dport = udph->dest;
__u8 ihl = iph->ihl;
if (ihl < 5)
return XDP_DROP;
__u64 l4_off = nh_off + (ihl * 4);
/* Parse L4 ports - check exactly 4 bytes (sport + dport) */
void *l4_hdr = (void *)((char *)data + l4_off);
if ((void *)((char *)l4_hdr + 4) <= data_end) {
if (iph->protocol == IPPROTO_TCP || iph->protocol == IPPROTO_UDP) {
__u16 *ports = l4_hdr;
key.sport = ports[0];
key.dport = ports[1];
fib_params.sport = ports[0];
fib_params.dport = ports[1];
}
}
@@ -232,7 +203,7 @@ static __always_inline int xdp_l3fwd_flags(struct xdp_md *ctx, __u32 flags)
fib_params.ipv4_dst = iph->daddr;
} else if (h_proto == bpf_htons(ETH_P_IPV6)) {
ip6h = data + nh_off;
struct ipv6hdr *ip6h = (void *)((char *)data + nh_off);
if ((void *)(ip6h + 1) > data_end)
return XDP_DROP;
@@ -242,32 +213,28 @@ static __always_inline int xdp_l3fwd_flags(struct xdp_md *ctx, __u32 flags)
__builtin_memcpy(key.ipv6_src, &ip6h->saddr, 16);
__builtin_memcpy(key.ipv6_dst, &ip6h->daddr, 16);
/* FIXED: Parse IPv6 extension headers to get actual L4 protocol */
__u64 l4_off = nh_off + sizeof(*ip6h);
int l4_proto = skip_ip6hdrext(data, data_end, &l4_off, ip6h->nexthdr);
if (l4_proto < 0)
l4_proto = ip6h->nexthdr; /* Fallback if parsing fails */
l4_proto = ip6h->nexthdr;
key.proto = l4_proto;
/* FIXED: Proper bounds checking for L4 headers */
if (l4_proto == IPPROTO_TCP) {
struct tcphdr *tcph = data + l4_off;
if ((void *)(tcph + 1) <= data_end) {
key.sport = tcph->source;
key.dport = tcph->dest;
}
} else if (l4_proto == IPPROTO_UDP) {
struct udphdr *udph = data + l4_off;
if ((void *)(udph + 1) <= data_end) {
key.sport = udph->source;
key.dport = udph->dest;
/* Parse L4 ports - check exactly 4 bytes */
void *l4_hdr = (void *)((char *)data + l4_off);
if ((void *)((char *)l4_hdr + 4) <= data_end) {
if (l4_proto == IPPROTO_TCP || l4_proto == IPPROTO_UDP) {
__u16 *ports = l4_hdr;
key.sport = ports[0];
key.dport = ports[1];
fib_params.sport = ports[0];
fib_params.dport = ports[1];
}
}
fib_params.family = AF_INET6;
fib_params.flowinfo = *(__be32 *)ip6h & IPV6_FLOWINFO_MASK;
__be32 flow = *(__be32 *)ip6h & IPV6_FLOWINFO_MASK;
fib_params.flowinfo = flow;
fib_params.l4_protocol = l4_proto;
fib_params.tot_len = bpf_ntohs(ip6h->payload_len);
__builtin_memcpy(fib_params.ipv6_src, &ip6h->saddr, 16);
@@ -277,36 +244,35 @@ static __always_inline int xdp_l3fwd_flags(struct xdp_md *ctx, __u32 flags)
}
fib_params.ifindex = ctx->ingress_ifindex;
rc = bpf_fib_lookup(ctx, &fib_params, sizeof(fib_params), flags);
if (rc == BPF_FIB_LKUP_RET_SUCCESS) {
if (!bpf_map_lookup_elem(&xdp_l3fwd_ports, &fib_params.ifindex))
return XDP_PASS;
int rc = bpf_fib_lookup(ctx, &fib_params, sizeof(fib_params), flags);
if (rc == 0) {
record_stats(&key, bytes);
/* Record stats AFTER fib lookup and port validation */
record_stats(ctx, &key, bytes);
/* Update TTL/hop limit and MAC addresses */
if (h_proto == bpf_htons(ETH_P_IP))
if (h_proto == bpf_htons(ETH_P_IP)) {
struct iphdr *iph = (void *)((char *)data + nh_off);
ip_decrease_ttl(iph);
else if (h_proto == bpf_htons(ETH_P_IPV6))
} else if (h_proto == bpf_htons(ETH_P_IPV6)) {
struct ipv6hdr *ip6h = (void *)((char *)data + nh_off);
ip6h->hop_limit--;
}
__builtin_memcpy(eth->h_dest, fib_params.dmac, ETH_ALEN);
__builtin_memcpy(eth->h_source, fib_params.smac, ETH_ALEN);
return bpf_redirect_map(&xdp_l3fwd_ports, fib_params.ifindex, 0);
return bpf_redirect_map(&xdp_l3fwd_ports, fib_params.ifindex, XDP_PASS);
}
return XDP_PASS;
}
SEC("xdp_l3fwd")
SEC("xdp")
int xdp_l3fwd_prog(struct xdp_md *ctx)
{
return xdp_l3fwd_flags(ctx, 0);
}
SEC("xdp_l3fwd_direct")
SEC("xdp")
int xdp_l3fwd_direct_prog(struct xdp_md *ctx)
{
return xdp_l3fwd_flags(ctx, BPF_FIB_LOOKUP_DIRECT);