eBPF Packet Redirect Skill

This skill provides comprehensive guidance for implementing packet redirection and routing in eBPF-based CNFs.

What This Skill Does

Generates code for:

1. Basic packet redirection with bpf_redirect
2. Neighbor-aware redirection with bpf_redirect_neigh
3. Source-based policy routing
4. Map-based routing tables
5. Dynamic route updates from userspace
6. Load balancing across multiple paths
7. Complete CNF router implementations

When to Use

• Building software routers or gateways
• Implementing source-based policy routing
• Creating load balancers
• Handling virtual IPs or anycast addresses
• Building service mesh sidecars
• Solving asymmetric routing problems
• Implementing multi-homing or multiple ISP scenarios
• Creating VPN/tunnel endpoints with custom routing

Redirection Helpers Comparison

bpf_redirect - Basic Redirection

long bpf_redirect(u32 ifindex, u64 flags);

What it does:

• Changes output interface only
• You must handle everything else:
  • MAC address rewriting
  • ARP/NDP resolution
  • Route lookups
  • Next-hop determination

Use when:

• Redirecting within same L2 domain
• You control all network configuration
• Simple interface switching needed

Example:

SEC("xdp")
int xdp_redirect_simple(struct xdp_md *ctx) {
    // Get interface
    struct net_device *eth1 = /* ... */;

    // Simple redirect to eth1
    return bpf_redirect(eth1->ifindex, 0);
}

bpf_redirect_neigh - Neighbor-Aware Redirection (Preferred)

long bpf_redirect_neigh(u32 ifindex, struct bpf_redir_neigh *params,
                        int plen, u64 flags);

What it does:

• Automatic ARP/NDP resolution - Handles neighbor discovery
• Route lookup - Finds next hop automatically
• MAC rewriting - Updates Ethernet headers
• Gateway-aware - Can forward through routers

Advantages:

• ✅ Full routing stack in eBPF
• ✅ Handles all L2/L3 details
• ✅ Production-ready forwarding
• ✅ Works across L3 boundaries

Use when:

• Building routers or gateways (most CNF use cases)
• Need L3 forwarding capability
• Want automatic neighbor resolution
• Forwarding across subnets

Structure:

struct bpf_redir_neigh {
    __u32 nh_family;    // AF_INET or AF_INET6
    union {
        __be32 ipv4_nh;     // IPv4 next hop
        __u32 ipv6_nh[4];   // IPv6 next hop
    };
};

Source-Based Policy Routing

The Problem: Asymmetric Routing

Scenario:

Client (10.0.2.2)
  ↓
Router (10.0.2.1 / 10.0.1.1) [eBPF CNF]
  ↓
Server (10.0.1.2)
  - Default gateway: 192.168.0.1
  - Virtual IP: 192.168.100.5 on lo

Issue:

1. Client → 192.168.100.5 (server's virtual IP)
2. Router forwards to server ✅
3. Server replies, but routing table says "use default gateway"
4. Reply goes to 192.168.0.1 instead of back through router ❌
5. Connection fails - asymmetric routing

eBPF Solution: Attach program to server's egress that redirects based on source IP

Complete CNF Router Example

C Code - eBPF Router (kernel space)

#include <linux/bpf.h>
#include <linux/if_ether.h>
#include <linux/ip.h>
#include <linux/in.h>
#include <linux/pkt_cls.h>
#include <bpf/bpf_helpers.h>
#include <bpf/bpf_endian.h>

// Routing policy record
struct route_policy {
    __u32 interface_id;  // Output interface
    __u32 next_hop;      // Next hop IP (network byte order)
};

// Map: Source IP → Routing Policy
struct {
    __uint(type, BPF_MAP_TYPE_HASH);
    __type(key, __u32);   // Source IPv4 address
    __type(value, struct route_policy);
    __uint(max_entries, 1024);
} policy_routes_v4 SEC(".maps");

// Statistics
struct {
    __uint(type, BPF_MAP_TYPE_ARRAY);
    __type(key, __u32);
    __type(value, __u64);
    __uint(max_entries, 3);
} stats SEC(".maps");

#define STAT_TOTAL_PACKETS   0
#define STAT_POLICY_MATCHES  1
#define STAT_REDIRECTS       2

static __always_inline void update_stat(__u32 key) {
    __u64 *counter = bpf_map_lookup_elem(&stats, &key);
    if (counter)
        __sync_fetch_and_add(counter, 1);
}

SEC("tc")
int policy_router(struct __sk_buff *skb) {
    void *data = (void *)(long)skb->data;
    void *data_end = (void *)(long)skb->data_end;

    update_stat(STAT_TOTAL_PACKETS);

    // Bounds check for Ethernet header
    if (data + sizeof(struct ethhdr) > data_end)
        return TC_ACT_OK;

    struct ethhdr *eth = data;

    // Only handle IPv4
    if (bpf_ntohs(eth->h_proto) != ETH_P_IP)
        return TC_ACT_OK;

    // Bounds check for IP header
    if (data + sizeof(struct ethhdr) + sizeof(struct iphdr) > data_end)
        return TC_ACT_OK;

    struct iphdr *iph = data + sizeof(struct ethhdr);

    // Look up source-based policy
    __u32 src_key = iph->saddr;  // Already in network byte order
    struct route_policy *policy = bpf_map_lookup_elem(&policy_routes_v4, &src_key);

    if (!policy) {
        // No policy match, use normal routing
        return TC_ACT_OK;
    }

    update_stat(STAT_POLICY_MATCHES);

    // Apply policy routing with neighbor-aware redirect
    struct bpf_redir_neigh nh = {
        .nh_family = AF_INET,
        .ipv4_nh = policy->next_hop,  // Already in network byte order
    };

    long ret = bpf_redirect_neigh(policy->interface_id, &nh, sizeof(nh), 0);

    if (ret == TC_ACT_REDIRECT) {
        update_stat(STAT_REDIRECTS);

        // Debug logging
        bpf_printk("Policy redirect: src=%pI4 iface=%d nexthop=%pI4",
                   &iph->saddr, policy->interface_id, &policy->next_hop);
    }

    return ret;
}

char _license[] SEC("license") = "GPL";

Go Code - Policy Configuration (userspace)

package main

import (
    "encoding/binary"
    "fmt"
    "log"
    "net"
    "os"
    "os/signal"
    "syscall"
    "time"

    "github.com/cilium/ebpf"
    "github.com/cilium/ebpf/link"
)

//go:generate go run github.com/cilium/ebpf/cmd/bpf2go -type route_policy PolicyRouter policy_router.c

type RoutingPolicy struct {
    SourceIP  net.IP
    Interface string
    NextHop   net.IP
}

func main() {
    // Load eBPF program
    spec, err := LoadPolicyRouter()
    if err != nil {
        log.Fatalf("loading spec: %v", err)
    }

    objs := &PolicyRouterObjects{}
    if err := spec.LoadAndAssign(objs, nil); err != nil {
        log.Fatalf("loading objects: %v", err)
    }
    defer objs.Close()

    // Attach to default gateway interface (egress)
    iface, err := net.InterfaceByName("eth0")
    if err != nil {
        log.Fatalf("finding interface: %v", err)
    }

    l, err := link.AttachTCX(link.TCXOptions{
        Program:   objs.PolicyRouter,
        Attach:    ebpf.AttachTCXEgress,
        Interface: iface.Index,
    })
    if err != nil {
        log.Fatalf("attaching program: %v", err)
    }
    defer l.Close()

    log.Printf("Policy router attached to %s egress", iface.Name)

    // Configure routing policies
    policies := []RoutingPolicy{
        {
            SourceIP:  net.ParseIP("192.168.100.5"),
            Interface: "eth1",
            NextHop:   net.ParseIP("10.0.1.1"),
        },
        // Add more policies as needed
    }

    for _, policy := range policies {
        if err := addRoutingPolicy(objs.PolicyRoutesV4, policy); err != nil {
            log.Fatalf("adding policy: %v", err)
        }
        log.Printf("Added policy: %s via %s (next hop %s)",
            policy.SourceIP, policy.Interface, policy.NextHop)
    }

    // Monitor statistics
    go monitorStats(objs.Stats)

    // Wait for signal
    sig := make(chan os.Signal, 1)
    signal.Notify(sig, syscall.SIGINT, syscall.SIGTERM)
    <-sig

    log.Println("Shutting down policy router...")
}

func addRoutingPolicy(m *ebpf.Map, policy RoutingPolicy) error {
    // Get interface index
    iface, err := net.InterfaceByName(policy.Interface)
    if err != nil {
        return fmt.Errorf("interface %s not found: %w", policy.Interface, err)
    }

    // Convert source IP to key (network byte order)
    srcIP := policy.SourceIP.To4()
    if srcIP == nil {
        return fmt.Errorf("invalid IPv4 address: %s", policy.SourceIP)
    }
    key := binary.BigEndian.Uint32(srcIP)

    // Convert next hop to network byte order
    nextHopIP := policy.NextHop.To4()
    if nextHopIP == nil {
        return fmt.Errorf("invalid next hop IPv4 address: %s", policy.NextHop)
    }
    nextHop := binary.BigEndian.Uint32(nextHopIP)

    // Create policy record
    value := PolicyRouterRoutePolicy{
        InterfaceId: uint32(iface.Index),
        NextHop:     nextHop,
    }

    // Insert into map
    if err := m.Put(&key, &value); err != nil {
        return fmt.Errorf("map insert failed: %w", err)
    }

    return nil
}

func removeRoutingPolicy(m *ebpf.Map, sourceIP net.IP) error {
    srcIP := sourceIP.To4()
    if srcIP == nil {
        return fmt.Errorf("invalid IPv4 address: %s", sourceIP)
    }
    key := binary.BigEndian.Uint32(srcIP)

    if err := m.Delete(&key); err != nil {
        return fmt.Errorf("map delete failed: %w", err)
    }

    return nil
}

func monitorStats(m *ebpf.Map) {
    ticker := time.NewTicker(5 * time.Second)
    defer ticker.Stop()

    for range ticker.C {
        var (
            key   uint32
            value uint64
        )

        // Read statistics
        stats := make(map[string]uint64)

        key = 0 // STAT_TOTAL_PACKETS
        if err := m.Lookup(&key, &value); err == nil {
            stats["total_packets"] = value
        }

        key = 1 // STAT_POLICY_MATCHES
        if err := m.Lookup(&key, &value); err == nil {
            stats["policy_matches"] = value
        }

        key = 2 // STAT_REDIRECTS
        if err := m.Lookup(&key, &value); err == nil {
            stats["redirects"] = value
        }

        log.Printf("Stats: total=%d matches=%d redirects=%d",
            stats["total_packets"], stats["policy_matches"], stats["redirects"])
    }
}

Use Cases

1. Virtual IP / Anycast Handling

// Server has 192.168.100.5 on loopback
// Route traffic from this IP via specific interface
policy := RoutingPolicy{
    SourceIP:  net.ParseIP("192.168.100.5"),
    Interface: "eth1",
    NextHop:   net.ParseIP("10.0.1.1"),
}

2. Multi-Homing / Multiple ISPs

// Route customer A traffic via ISP 1
addRoutingPolicy(map, RoutingPolicy{
    SourceIP:  net.ParseIP("10.10.1.0"), // Customer A subnet
    Interface: "isp1",
    NextHop:   net.ParseIP("203.0.113.1"),
})

// Route customer B traffic via ISP 2
addRoutingPolicy(map, RoutingPolicy{
    SourceIP:  net.ParseIP("10.10.2.0"), // Customer B subnet
    Interface: "isp2",
    NextHop:   net.ParseIP("198.51.100.1"),
})

3. Service Mesh Sidecar

// Pod has multiple interfaces:
// - Service traffic → service network
// - Management traffic → management network

servicePolicies := []RoutingPolicy{
    {
        SourceIP:  net.ParseIP("10.96.0.10"),  // Service IP
        Interface: "net1",
        NextHop:   net.ParseIP("10.96.0.1"),
    },
    {
        SourceIP:  net.ParseIP("192.168.1.10"), // Management IP
        Interface: "net0",
        NextHop:   net.ParseIP("192.168.1.1"),
    },
}

4. VPN / Tunnel Endpoint

// Traffic from tunnel IPs → tunnel interface
addRoutingPolicy(map, RoutingPolicy{
    SourceIP:  net.ParseIP("172.16.0.0"),  // VPN subnet
    Interface: "tun0",
    NextHop:   net.ParseIP("172.16.0.1"),
})

Advanced Patterns

Load Balancing Across Multiple Paths

#define MAX_BACKENDS 4

struct backend {
    __u32 interface_id;
    __u32 next_hop;
    __u32 weight;  // For weighted load balancing
};

struct {
    __uint(type, BPF_MAP_TYPE_ARRAY);
    __type(key, __u32);
    __type(value, struct backend);
    __uint(max_entries, MAX_BACKENDS);
} backends SEC(".maps");

struct {
    __uint(type, BPF_MAP_TYPE_ARRAY);
    __type(key, __u32);
    __type(value, __u32);
    __uint(max_entries, 1);
} backend_count SEC(".maps");

static __always_inline struct backend *select_backend() {
    __u32 key = 0;
    __u32 *count = bpf_map_lookup_elem(&backend_count, &key);
    if (!count || *count == 0)
        return NULL;

    // Round-robin selection
    __u32 idx = bpf_get_prandom_u32() % (*count);
    return bpf_map_lookup_elem(&backends, &idx);
}

SEC("tc")
int load_balancer(struct __sk_buff *skb) {
    // ... parse packet ...

    struct backend *backend = select_backend();
    if (!backend)
        return TC_ACT_OK;

    struct bpf_redir_neigh nh = {
        .nh_family = AF_INET,
        .ipv4_nh = backend->next_hop,
    };

    return bpf_redirect_neigh(backend->interface_id, &nh, sizeof(nh), 0);
}

Conditional Routing Based on Multiple Criteria

// Route based on source IP + destination port
struct policy_key {
    __u32 src_ip;
    __u16 dst_port;
    __u8 protocol;
    __u8 _pad;
} __attribute__((packed));

struct {
    __uint(type, BPF_MAP_TYPE_HASH);
    __type(key, struct policy_key);
    __type(value, struct route_policy);
    __uint(max_entries, 1024);
} complex_policies SEC(".maps");

SEC("tc")
int complex_router(struct __sk_buff *skb) {
    // ... parse packet to get IP + TCP/UDP ...

    struct policy_key key = {
        .src_ip = iph->saddr,
        .dst_port = tcp->dest,  // or udp->dest
        .protocol = iph->protocol,
    };

    struct route_policy *policy = bpf_map_lookup_elem(&complex_policies, &key);
    // ... redirect based on policy ...
}

Policy Chaining (Fallback Policies)

SEC("tc")
int chained_router(struct __sk_buff *skb) {
    struct iphdr *iph = /* ... parse ... */;
    struct route_policy *policy = NULL;

    // Try source-based policy first
    policy = bpf_map_lookup_elem(&src_policies, &iph->saddr);

    // Fall back to destination-based policy
    if (!policy)
        policy = bpf_map_lookup_elem(&dst_policies, &iph->daddr);

    // Fall back to default route
    if (!policy) {
        __u32 default_key = 0;
        policy = bpf_map_lookup_elem(&default_route, &default_key);
    }

    if (!policy)
        return TC_ACT_OK;

    // Apply policy
    struct bpf_redir_neigh nh = {
        .nh_family = AF_INET,
        .ipv4_nh = policy->next_hop,
    };

    return bpf_redirect_neigh(policy->interface_id, &nh, sizeof(nh), 0);
}

IPv6 Support

// IPv6 routing policy
struct route_policy_v6 {
    __u32 interface_id;
    __u32 next_hop[4];  // IPv6 address (16 bytes)
};

struct {
    __uint(type, BPF_MAP_TYPE_HASH);
    __type(key, struct in6_addr);  // IPv6 source address
    __type(value, struct route_policy_v6);
    __uint(max_entries, 1024);
} policy_routes_v6 SEC(".maps");

SEC("tc")
int policy_router_v6(struct __sk_buff *skb) {
    // ... parse IPv6 packet ...

    struct ipv6hdr *ip6h = /* ... */;

    struct route_policy_v6 *policy = bpf_map_lookup_elem(&policy_routes_v6, &ip6h->saddr);
    if (!policy)
        return TC_ACT_OK;

    struct bpf_redir_neigh nh = {
        .nh_family = AF_INET6,
    };
    __builtin_memcpy(nh.ipv6_nh, policy->next_hop, sizeof(nh.ipv6_nh));

    return bpf_redirect_neigh(policy->interface_id, &nh, sizeof(nh), 0);
}

Return Value Handling

TC Return Values for Redirection

// Success
TC_ACT_REDIRECT  // Successfully redirected

// Errors
TC_ACT_OK        // Continue normal processing
TC_ACT_SHOT      // Drop packet

Error Handling Pattern

long ret = bpf_redirect_neigh(ifindex, &nh, sizeof(nh), 0);

if (ret != TC_ACT_REDIRECT) {
    // Log failure
    bpf_printk("Redirect failed: ifindex=%d ret=%ld", ifindex, ret);

    // Increment error counter
    __u32 key = STAT_REDIRECT_ERRORS;
    __u64 *counter = bpf_map_lookup_elem(&stats, &key);
    if (counter)
        __sync_fetch_and_add(counter, 1);

    // Fall back to normal routing
    return TC_ACT_OK;
}

return ret;

Best Practices

1. Always validate interfaces exist before adding policies
2. Use network byte order for IP addresses in maps
3. Add statistics to monitor policy effectiveness
4. Implement fallback to normal routing if no policy matches
5. Log redirections during development (remove in production)
6. Bounds check before accessing packet data
7. Use bpf_redirect_neigh for L3 forwarding (handles ARP/MAC automatically)
8. Test with real network topologies (not just localhost)
9. Monitor for redirect failures and investigate root causes
10. Document your policies in userspace code comments

Common Pitfalls

• Forgetting to attach to egress (not ingress) for source-based routing
• Using host byte order instead of network byte order in maps
• Not handling the case where policy lookup fails
• Redirecting without checking if interface is up
• Not validating that next hop is reachable
• Forgetting IPv6 support if dual-stack
• Not implementing statistics/monitoring

Debugging

Enable Debug Logging

// In eBPF program
bpf_printk("Redirect: src=%pI4 dst=%pI4 iface=%d",
           &iph->saddr, &iph->daddr, policy->interface_id);

View Logs

# Using tc
sudo tc exec bpf dbg

# Or using bpftool
sudo bpftool prog tracelog

# Or directly
sudo cat /sys/kernel/debug/tracing/trace_pipe

Check Redirect Stats

# View eBPF program stats
bpftool prog show

# Check map contents
bpftool map dump name policy_routes_v4

Verify Interface and Routes

# Check interface is up
ip link show eth1

# Verify next hop is reachable
ip route get 10.0.1.1

Integration with Other Skills

• ebpf-packet-parser: Use to extract flow information before routing decision
• ebpf-map-handler: Manage routing policy maps
• ebpf-attach-hook: Attach router to TC egress
• cnf-networking: Set up test topology with netkit/veth
• ebpf-test-harness: Validate routing with multi-container tests