xref: /openbmc/linux/samples/bpf/xdp_adjust_tail_kern.c (revision 71a15258f3c92eb1c4ae98bbfca9459f4723d5d3)
1 /* SPDX-License-Identifier: GPL-2.0
2  * Copyright (c) 2018 Facebook
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 shows how to use bpf_xdp_adjust_tail() by
9  * generating ICMPv4 "packet to big" (unreachable/ df bit set frag needed
10  * to be more preice in case of v4)" where receiving packets bigger then
11  * 600 bytes.
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/icmp.h>
21 #include <bpf/bpf_helpers.h>
22 
23 #define DEFAULT_TTL 64
24 #define MAX_PCKT_SIZE 600
25 #define ICMP_TOOBIG_SIZE 98
26 #define ICMP_TOOBIG_PAYLOAD_SIZE 92
27 
28 /* volatile to prevent compiler optimizations */
29 static volatile __u32 max_pcktsz = MAX_PCKT_SIZE;
30 
31 struct {
32 	__uint(type, BPF_MAP_TYPE_ARRAY);
33 	__type(key, __u32);
34 	__type(value, __u64);
35 	__uint(max_entries, 1);
36 } icmpcnt SEC(".maps");
37 
38 static __always_inline void count_icmp(void)
39 {
40 	u64 key = 0;
41 	u64 *icmp_count;
42 
43 	icmp_count = bpf_map_lookup_elem(&icmpcnt, &key);
44 	if (icmp_count)
45 		*icmp_count += 1;
46 }
47 
48 static __always_inline void swap_mac(void *data, struct ethhdr *orig_eth)
49 {
50 	struct ethhdr *eth;
51 
52 	eth = data;
53 	memcpy(eth->h_source, orig_eth->h_dest, ETH_ALEN);
54 	memcpy(eth->h_dest, orig_eth->h_source, ETH_ALEN);
55 	eth->h_proto = orig_eth->h_proto;
56 }
57 
58 static __always_inline __u16 csum_fold_helper(__u32 csum)
59 {
60 	csum = (csum & 0xffff) + (csum >> 16);
61 	return ~((csum & 0xffff) + (csum >> 16));
62 }
63 
64 static __always_inline void ipv4_csum(void *data_start, int data_size,
65 				      __u32 *csum)
66 {
67 	*csum = bpf_csum_diff(0, 0, data_start, data_size, *csum);
68 	*csum = csum_fold_helper(*csum);
69 }
70 
71 static __always_inline int send_icmp4_too_big(struct xdp_md *xdp)
72 {
73 	int headroom = (int)sizeof(struct iphdr) + (int)sizeof(struct icmphdr);
74 
75 	if (bpf_xdp_adjust_head(xdp, 0 - headroom))
76 		return XDP_DROP;
77 	void *data = (void *)(long)xdp->data;
78 	void *data_end = (void *)(long)xdp->data_end;
79 
80 	if (data + (ICMP_TOOBIG_SIZE + headroom) > data_end)
81 		return XDP_DROP;
82 
83 	struct iphdr *iph, *orig_iph;
84 	struct icmphdr *icmp_hdr;
85 	struct ethhdr *orig_eth;
86 	__u32 csum = 0;
87 	__u64 off = 0;
88 
89 	orig_eth = data + headroom;
90 	swap_mac(data, orig_eth);
91 	off += sizeof(struct ethhdr);
92 	iph = data + off;
93 	off += sizeof(struct iphdr);
94 	icmp_hdr = data + off;
95 	off += sizeof(struct icmphdr);
96 	orig_iph = data + off;
97 	icmp_hdr->type = ICMP_DEST_UNREACH;
98 	icmp_hdr->code = ICMP_FRAG_NEEDED;
99 	icmp_hdr->un.frag.mtu = htons(max_pcktsz - sizeof(struct ethhdr));
100 	icmp_hdr->checksum = 0;
101 	ipv4_csum(icmp_hdr, ICMP_TOOBIG_PAYLOAD_SIZE, &csum);
102 	icmp_hdr->checksum = csum;
103 	iph->ttl = DEFAULT_TTL;
104 	iph->daddr = orig_iph->saddr;
105 	iph->saddr = orig_iph->daddr;
106 	iph->version = 4;
107 	iph->ihl = 5;
108 	iph->protocol = IPPROTO_ICMP;
109 	iph->tos = 0;
110 	iph->tot_len = htons(
111 		ICMP_TOOBIG_SIZE + headroom - sizeof(struct ethhdr));
112 	iph->check = 0;
113 	csum = 0;
114 	ipv4_csum(iph, sizeof(struct iphdr), &csum);
115 	iph->check = csum;
116 	count_icmp();
117 	return XDP_TX;
118 }
119 
120 
121 static __always_inline int handle_ipv4(struct xdp_md *xdp)
122 {
123 	void *data_end = (void *)(long)xdp->data_end;
124 	void *data = (void *)(long)xdp->data;
125 	int pckt_size = data_end - data;
126 	int offset;
127 
128 	if (pckt_size > max(max_pcktsz, ICMP_TOOBIG_SIZE)) {
129 		offset = pckt_size - ICMP_TOOBIG_SIZE;
130 		if (bpf_xdp_adjust_tail(xdp, 0 - offset))
131 			return XDP_PASS;
132 		return send_icmp4_too_big(xdp);
133 	}
134 	return XDP_PASS;
135 }
136 
137 SEC("xdp_icmp")
138 int _xdp_icmp(struct xdp_md *xdp)
139 {
140 	void *data_end = (void *)(long)xdp->data_end;
141 	void *data = (void *)(long)xdp->data;
142 	struct ethhdr *eth = data;
143 	__u16 h_proto;
144 
145 	if (eth + 1 > data_end)
146 		return XDP_DROP;
147 
148 	h_proto = eth->h_proto;
149 
150 	if (h_proto == htons(ETH_P_IP))
151 		return handle_ipv4(xdp);
152 	else
153 		return XDP_PASS;
154 }
155 
156 char _license[] SEC("license") = "GPL";
157