xref: /openbmc/linux/samples/bpf/xdp_fwd_kern.c (revision 37b9b9a5)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2017-18 David Ahern <dsahern@gmail.com>
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of version 2 of the GNU General Public
6  * License as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11  * General Public License for more details.
12  */
13 #define KBUILD_MODNAME "foo"
14 #include <uapi/linux/bpf.h>
15 #include <linux/in.h>
16 #include <linux/if_ether.h>
17 #include <linux/if_packet.h>
18 #include <linux/if_vlan.h>
19 #include <linux/ip.h>
20 #include <linux/ipv6.h>
21 
22 #include "bpf_helpers.h"
23 
24 #define IPV6_FLOWINFO_MASK              cpu_to_be32(0x0FFFFFFF)
25 
26 struct {
27 	__uint(type, BPF_MAP_TYPE_DEVMAP);
28 	__uint(key_size, sizeof(int));
29 	__uint(value_size, sizeof(int));
30 	__uint(max_entries, 64);
31 } xdp_tx_ports SEC(".maps");
32 
33 /* from include/net/ip.h */
34 static __always_inline int ip_decrease_ttl(struct iphdr *iph)
35 {
36 	u32 check = (__force u32)iph->check;
37 
38 	check += (__force u32)htons(0x0100);
39 	iph->check = (__force __sum16)(check + (check >= 0xFFFF));
40 	return --iph->ttl;
41 }
42 
43 static __always_inline int xdp_fwd_flags(struct xdp_md *ctx, u32 flags)
44 {
45 	void *data_end = (void *)(long)ctx->data_end;
46 	void *data = (void *)(long)ctx->data;
47 	struct bpf_fib_lookup fib_params;
48 	struct ethhdr *eth = data;
49 	struct ipv6hdr *ip6h;
50 	struct iphdr *iph;
51 	u16 h_proto;
52 	u64 nh_off;
53 	int rc;
54 
55 	nh_off = sizeof(*eth);
56 	if (data + nh_off > data_end)
57 		return XDP_DROP;
58 
59 	__builtin_memset(&fib_params, 0, sizeof(fib_params));
60 
61 	h_proto = eth->h_proto;
62 	if (h_proto == htons(ETH_P_IP)) {
63 		iph = data + nh_off;
64 
65 		if (iph + 1 > data_end)
66 			return XDP_DROP;
67 
68 		if (iph->ttl <= 1)
69 			return XDP_PASS;
70 
71 		fib_params.family	= AF_INET;
72 		fib_params.tos		= iph->tos;
73 		fib_params.l4_protocol	= iph->protocol;
74 		fib_params.sport	= 0;
75 		fib_params.dport	= 0;
76 		fib_params.tot_len	= ntohs(iph->tot_len);
77 		fib_params.ipv4_src	= iph->saddr;
78 		fib_params.ipv4_dst	= iph->daddr;
79 	} else if (h_proto == htons(ETH_P_IPV6)) {
80 		struct in6_addr *src = (struct in6_addr *) fib_params.ipv6_src;
81 		struct in6_addr *dst = (struct in6_addr *) fib_params.ipv6_dst;
82 
83 		ip6h = data + nh_off;
84 		if (ip6h + 1 > data_end)
85 			return XDP_DROP;
86 
87 		if (ip6h->hop_limit <= 1)
88 			return XDP_PASS;
89 
90 		fib_params.family	= AF_INET6;
91 		fib_params.flowinfo	= *(__be32 *)ip6h & IPV6_FLOWINFO_MASK;
92 		fib_params.l4_protocol	= ip6h->nexthdr;
93 		fib_params.sport	= 0;
94 		fib_params.dport	= 0;
95 		fib_params.tot_len	= ntohs(ip6h->payload_len);
96 		*src			= ip6h->saddr;
97 		*dst			= ip6h->daddr;
98 	} else {
99 		return XDP_PASS;
100 	}
101 
102 	fib_params.ifindex = ctx->ingress_ifindex;
103 
104 	rc = bpf_fib_lookup(ctx, &fib_params, sizeof(fib_params), flags);
105 	/*
106 	 * Some rc (return codes) from bpf_fib_lookup() are important,
107 	 * to understand how this XDP-prog interacts with network stack.
108 	 *
109 	 * BPF_FIB_LKUP_RET_NO_NEIGH:
110 	 *  Even if route lookup was a success, then the MAC-addresses are also
111 	 *  needed.  This is obtained from arp/neighbour table, but if table is
112 	 *  (still) empty then BPF_FIB_LKUP_RET_NO_NEIGH is returned.  To avoid
113 	 *  doing ARP lookup directly from XDP, then send packet to normal
114 	 *  network stack via XDP_PASS and expect it will do ARP resolution.
115 	 *
116 	 * BPF_FIB_LKUP_RET_FWD_DISABLED:
117 	 *  The bpf_fib_lookup respect sysctl net.ipv{4,6}.conf.all.forwarding
118 	 *  setting, and will return BPF_FIB_LKUP_RET_FWD_DISABLED if not
119 	 *  enabled this on ingress device.
120 	 */
121 	if (rc == BPF_FIB_LKUP_RET_SUCCESS) {
122 		/* Verify egress index has been configured as TX-port.
123 		 * (Note: User can still have inserted an egress ifindex that
124 		 * doesn't support XDP xmit, which will result in packet drops).
125 		 *
126 		 * Note: lookup in devmap supported since 0cdbb4b09a0.
127 		 * If not supported will fail with:
128 		 *  cannot pass map_type 14 into func bpf_map_lookup_elem#1:
129 		 */
130 		if (!bpf_map_lookup_elem(&xdp_tx_ports, &fib_params.ifindex))
131 			return XDP_PASS;
132 
133 		if (h_proto == htons(ETH_P_IP))
134 			ip_decrease_ttl(iph);
135 		else if (h_proto == htons(ETH_P_IPV6))
136 			ip6h->hop_limit--;
137 
138 		memcpy(eth->h_dest, fib_params.dmac, ETH_ALEN);
139 		memcpy(eth->h_source, fib_params.smac, ETH_ALEN);
140 		return bpf_redirect_map(&xdp_tx_ports, fib_params.ifindex, 0);
141 	}
142 
143 	return XDP_PASS;
144 }
145 
146 SEC("xdp_fwd")
147 int xdp_fwd_prog(struct xdp_md *ctx)
148 {
149 	return xdp_fwd_flags(ctx, 0);
150 }
151 
152 SEC("xdp_fwd_direct")
153 int xdp_fwd_direct_prog(struct xdp_md *ctx)
154 {
155 	return xdp_fwd_flags(ctx, BPF_FIB_LOOKUP_DIRECT);
156 }
157 
158 char _license[] SEC("license") = "GPL";
159