samples/bpf/xdp_adjust_tail_kern.c

   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_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 struct bpf_map_def SEC("maps") icmpcnt = {
  29         .type = BPF_MAP_TYPE_ARRAY,
  30         .key_size = sizeof(__u32),
  31         .value_size = sizeof(__u64),
  32         .max_entries = 1,
  33 };
  34
  35 static __always_inline void count_icmp(void)
  36 {
  37         u64 key = 0;
  38         u64 *icmp_count;
  39
  40         icmp_count = bpf_map_lookup_elem(&icmpcnt, &key);
  41         if (icmp_count)
  42                 *icmp_count += 1;
  43 }
  44
  45 static __always_inline void swap_mac(void *data, struct ethhdr *orig_eth)
  46 {
  47         struct ethhdr *eth;
  48
  49         eth = data;
  50         memcpy(eth->h_source, orig_eth->h_dest, ETH_ALEN);
  51         memcpy(eth->h_dest, orig_eth->h_source, ETH_ALEN);
  52         eth->h_proto = orig_eth->h_proto;
  53 }
  54
  55 static __always_inline __u16 csum_fold_helper(__u32 csum)
  56 {
  57         return ~((csum & 0xffff) + (csum >> 16));
  58 }
  59
  60 static __always_inline void ipv4_csum(void *data_start, int data_size,
  61                                       __u32 *csum)
  62 {
  63         *csum = bpf_csum_diff(0, 0, data_start, data_size, *csum);
  64         *csum = csum_fold_helper(*csum);
  65 }
  66
  67 static __always_inline int send_icmp4_too_big(struct xdp_md *xdp)
  68 {
  69         int headroom = (int)sizeof(struct iphdr) + (int)sizeof(struct icmphdr);
  70
  71         if (bpf_xdp_adjust_head(xdp, 0 - headroom))
  72                 return XDP_DROP;
  73         void *data = (void *)(long)xdp->data;
  74         void *data_end = (void *)(long)xdp->data_end;
  75
  76         if (data + (ICMP_TOOBIG_SIZE + headroom) > data_end)
  77                 return XDP_DROP;
  78
  79         struct iphdr *iph, *orig_iph;
  80         struct icmphdr *icmp_hdr;
  81         struct ethhdr *orig_eth;
  82         __u32 csum = 0;
  83         __u64 off = 0;
  84
  85         orig_eth = data + headroom;
  86         swap_mac(data, orig_eth);
  87         off += sizeof(struct ethhdr);
  88         iph = data + off;
  89         off += sizeof(struct iphdr);
  90         icmp_hdr = data + off;
  91         off += sizeof(struct icmphdr);
  92         orig_iph = data + off;
  93         icmp_hdr->type = ICMP_DEST_UNREACH;
  94         icmp_hdr->code = ICMP_FRAG_NEEDED;
  95         icmp_hdr->un.frag.mtu = htons(MAX_PCKT_SIZE-sizeof(struct ethhdr));
  96         icmp_hdr->checksum = 0;
  97         ipv4_csum(icmp_hdr, ICMP_TOOBIG_PAYLOAD_SIZE, &csum);
  98         icmp_hdr->checksum = csum;
  99         iph->ttl = DEFAULT_TTL;
 100         iph->daddr = orig_iph->saddr;
 101         iph->saddr = orig_iph->daddr;
 102         iph->version = 4;
 103         iph->ihl = 5;
 104         iph->protocol = IPPROTO_ICMP;
 105         iph->tos = 0;
 106         iph->tot_len = htons(
 107                 ICMP_TOOBIG_SIZE + headroom - sizeof(struct ethhdr));
 108         iph->check = 0;
 109         csum = 0;
 110         ipv4_csum(iph, sizeof(struct iphdr), &csum);
 111         iph->check = csum;
 112         count_icmp();
 113         return XDP_TX;
 114 }
 115
 116
 117 static __always_inline int handle_ipv4(struct xdp_md *xdp)
 118 {
 119         void *data_end = (void *)(long)xdp->data_end;
 120         void *data = (void *)(long)xdp->data;
 121         int pckt_size = data_end - data;
 122         int offset;
 123
 124         if (pckt_size > MAX_PCKT_SIZE) {
 125                 offset = pckt_size - ICMP_TOOBIG_SIZE;
 126                 if (bpf_xdp_adjust_tail(xdp, 0 - offset))
 127                         return XDP_PASS;
 128                 return send_icmp4_too_big(xdp);
 129         }
 130         return XDP_PASS;
 131 }
 132
 133 SEC("xdp_icmp")
 134 int _xdp_icmp(struct xdp_md *xdp)
 135 {
 136         void *data_end = (void *)(long)xdp->data_end;
 137         void *data = (void *)(long)xdp->data;
 138         struct ethhdr *eth = data;
 139         __u16 h_proto;
 140
 141         if (eth + 1 > data_end)
 142                 return XDP_DROP;
 143
 144         h_proto = eth->h_proto;
 145
 146         if (h_proto == htons(ETH_P_IP))
 147                 return handle_ipv4(xdp);
 148         else
 149                 return XDP_PASS;
 150 }
 151
 152 char _license[] SEC("license") = "GPL";